Fundamental Statistics For The Behavioral Sciences v.1.9 PDF

-1-
Fundamental Statistics for the Behavioral Sciences (v. 1.9)
Copyright 2013 Bryan R. Burnham
This work is licensed under the Creative Commons Attribution-NonCommercial 3.0

Unported License. To view a copy of this license, visit
http://creativecommons.org/licenses/by-nc/3.0/.
Individuals are free to download and/or modify this book for non-commercial use so long
as I am credited. No commercial use of this work is allowed.
-2-
About the Author
Dr. Bryan R. Burnham received his B.A. from Utica

College of Syracuse University ('02) and his M.A. ('04)
and Ph.D. ('07) in cognitive psychology from the
University at Albany (SUNY). He continues to live and
work on the planet Earth and has a tendency of not
tucking in his shirt for class. Other than that, since he is
probably your statistics professor, ask him a question to
find out more. He won't bite!
You'll be surprised at how much Dr. Burnham looks like

his Simpson's picture, though he started growing a beard.
-3-
Acknowledgments
I would like to acknowledge my statistics and research methodology professors who

taught me more than I could ever possibly fit into a book, so I fit some of it into a smaller
book. Steve Specht, Ph.D. (Utica College), Bruce Dudek, Ph.D. (SUNY Albany), Jim
Neely, Ph.D. (SUNY Albany), and Tram Neill, Ph.D. (SUNY Albany). Without your
guidance and instruction I would still think the sum of squares was related to finding the
area of a cube.
-4-
Contents
Foreword .................................................................................................................................................... 11
Chapter 1: Introduction to Statistics ....................................................................................................... 12
1.1 You’ve Decided to take Statistics ..................................................................................................... 12
1.2 What, When, Where, How, and Why of Statistics ............................................................................ 13
1.3 Statistics as a Language................................................................................................................... 15
1.4 Two Types of Statistics ..................................................................................................................... 16
1.5 Variables ........................................................................................................................................... 17
1.6 Values vs. Names of Variables ......................................................................................................... 18
1.7 Quantity vs. Quality ........................................................................................................................... 19
1.8 Hypotheses and Predictions ............................................................................................................. 20
1.9 Populations and Samples ................................................................................................................. 20
1.10 Sampling ......................................................................................................................................... 22
1.11 Random Selection .......................................................................................................................... 22
1.12 Random Assignment ...................................................................................................................... 24
1.13 “Law” of Large Numbers ................................................................................................................. 25
1.14 Representative of Whom? .............................................................................................................. 26
1.15 Research Designs and Statistics .................................................................................................... 27
1.16 Computers and Statistics ................................................................................................................ 28
CH 1 Homework Questions .................................................................................................................... 29
Chapter 2: Math Review............................................................................................................................ 33
2.1 Why Review Math? Didn’t you say Statistics was a ‘Language’ Class? .......................................... 33
2.2 Whole, Rational, Irrational, and Real Numbers ................................................................................ 33
2.3 Basic Mathematical Operators.......................................................................................................... 34
2.4 Order of Operations .......................................................................................................................... 36
2.5 Rounding Values .............................................................................................................................. 36
2.6 Proportions, Probabilities and Percentages ..................................................................................... 37
2.7 Summation Notation (Σ).................................................................................................................... 38
2.8 Scales of Measurement .................................................................................................................... 39
2.9 Continuous vs. Discrete Data ........................................................................................................... 41
2.10 Real vs. Apparent Limits of Continuous Data ................................................................................. 41
Chapter 3: Frequency Distributions and Graphing ............................................................................... 46
3.1 Frequency Distributions for Quantitative Variables .......................................................................... 46
3.2 Frequency Distributions for Qualitative Variables............................................................................. 50
3.3 Outliers in Frequency Distributions ................................................................................................... 51
3.4 “A Picture is worth a Thousand Words” ............................................................................................ 51
3.5 Graphs for Frequency Distributions of Quantitative Variables ......................................................... 53
3.6 Graphs for Frequency Distributions of Qualitative Variables ............................................................ 55
-5-
3.7 Graphing Relative Frequencies, Percentages, Cumulative Frequencies and Cumulative Relative
Frequencies ............................................................................................................................................ 56
3.8 Misuse of Graphs .............................................................................................................................. 57
3.9 Probability Distributions .................................................................................................................... 57
3.10 Empirical vs. Theoretical Distributions ............................................................................................ 60
Chapter 4: Rankings in a Distribution ..................................................................................................... 66
4.1 Percentiles and Percent Ranks ........................................................................................................ 66
4.2 Calculating Percentile Ranks ............................................................................................................ 67
4.3 Calculating Percentiles ..................................................................................................................... 68
Chapter 5: Central Tendency ................................................................................................................... 71
5.1 Measures of Central Tendency ........................................................................................................ 71
5.2 The Mode: The Most Frequent Score ............................................................................................... 72
5.3 The Median: The “Middle” Score ...................................................................................................... 72
5.5 Quartiles............................................................................................................................................ 74
5.6 The Mean: Arithmetic Average ......................................................................................................... 76
5.8 Central Tendency and Scales of Measurement ............................................................................... 77
5.8 The Mean Median and Mode in Normal and Skewed Distributions ................................................. 78
5.9 Box Plots ........................................................................................................................................... 79
Chapter 6: Variability ................................................................................................................................ 84
6.1 What is Variability? Why is Measured? ............................................................................................ 84
6.2 Range ............................................................................................................................................... 85
6.3 Average Deviation (AD) .................................................................................................................... 85
6.4 Sum of Squares ................................................................................................................................ 86
2
6.5 Sample Variance (s ) ........................................................................................................................ 87
6.6 Sample Standard Deviation (s) ......................................................................................................... 88
6.7 The Population Formulas for Variability ............................................................................................ 89
6.8 Scaling Scores .................................................................................................................................. 90
Chapter 7: Standard Scores and Normal Distribution .......................................................................... 95
7.1 Comparing Scores across Distributions ........................................................................................... 95
7.2 Standard (z) Scores .......................................................................................................................... 96
7.3 Characteristics of Standard (z) Scores ............................................................................................. 96
7.4 Using the Mean, Standard Deviation and z-Scores to find Raw Scores .......................................... 97
7.5 Standard Scores and the Normal Distribution .................................................................................. 98
7.6 Probabilities and Proportions in the Normal Distribution ................................................................ 100
CH 7 Homework Questions .................................................................................................................. 103
-6-
Chapter 8: Estimation and Sampling Distributions ............................................................................. 105
8.1 Using Samples to Estimating the of Population.............................................................................. 105
8.2 Sample Mean is an Unbiased Estimator of the Population Mean .................................................. 105
8.3 Sample Variance and Sample Standard Deviation as Biased Estimators ..................................... 106
8.4 Degrees of Freedom ....................................................................................................................... 108
8.5 Sampling Distribution of the Mean .................................................................................................. 109
8.6 Confidence Intervals around a Sample Mean ................................................................................ 111
8.7 Confidence Intervals with Unknown Parameters ............................................................................ 113
8.8 Displaying Confidence Intervals in Graphs ..................................................................................... 114
Chapter 9: Probability ............................................................................................................................. 118
9.1 The Monty Hall Paradox ................................................................................................................. 118
9.2 Some Basic Ideas: Event Algebra .................................................................................................. 118
9.3 Probability Axioms and Things About Probability that are True ..................................................... 120
9.4 Counting Rules: Permutations and Combination ........................................................................... 121
9.5 Probability and Sampling from a Population................................................................................... 125
9.6 Probabilities Based on Multiple Variables ...................................................................................... 126
9.6.1 Simple Probabilities ................................................................................................................. 126
9.6.2 Conditional Probabilities .......................................................................................................... 127
9.6.3 Joint Probabilities .................................................................................................................... 128
9.6.4 Addition Rule of Probability ..................................................................................................... 129
9.6.5 Multiplication Rules of Probability............................................................................................ 130
9.6 Bayes’ Theorem .............................................................................................................................. 132
9.7 Back to Monty Hall .......................................................................................................................... 133
Chapter 10: Binomial Probability and Estimation ............................................................................... 140
10.1 Binomial Probability Expression ................................................................................................... 140
10.2 More on Binomial Probability Distribution ..................................................................................... 144
10.3 Binomial Approximation to the Normal Distribution ...................................................................... 145
CH 10 Homework Questions ................................................................................................................ 146
Chapter 11: Introduction to Hypothesis Testing ................................................................................. 148
11.1 Inferential Statistics and Hypothesis Testing ................................................................................ 148
11.2 Null Hypothesis Significance Testing (NHST) .............................................................................. 149
11.3 Example with Binomial Expression ............................................................................................... 150
11.4 The z-Test ..................................................................................................................................... 151
11.5 Reporting the z-Test in the Literature ........................................................................................... 154
11.6 Type I vs. Type II Errors ............................................................................................................... 155
11.7 Statistical Power ........................................................................................................................... 155
11.8 Limits of z-Test ............................................................................................................................. 156
-7-
11.9 The “Classic” NHST Procedure .................................................................................................... 156
Chapter 12: One-Sample t-Test ............................................................................................................. 160
12.1 Where and Why the one Sample t-Test Exists ............................................................................. 160
12.2 One Sample t-Test........................................................................................................................ 160
12.3 t-Distribution Approximation to the Normal Distribution ................................................................ 162
12.4 Reporting One Sample t-Test in Literature ................................................................................... 162
Chapter 13: Bivariate Designs ............................................................................................................... 165
13.1 What is a Bivariate Design?.......................................................................................................... 166
13.2 Issues in Experimental Designs ................................................................................................... 167
13.3 Parametric vs. Non-Parametric Statistical Tests .......................................................................... 169
13.4 Between Groups Designs vs. Within-Groups Designs ................................................................. 169
13.5 Matched Pairs Designs ................................................................................................................. 170
Chapter 14: Independent Groups t-Test ............................................................................................... 171
14.1 Uses and Requirements ............................................................................................................... 172
14.2 Mean Differences and Variance between Independent Populations ........................................... 172
14.3 Pooled Variance and the Estimated Standard Error of the Difference ......................................... 174
14.4 Hypotheses in the Independent Groups Design ........................................................................... 174
14.5 Degrees of Freedom in the Independent Groups t-Test............................................................... 175
14.6 Example of Independent Groups t-Test........................................................................................ 175
14.6 Reporting in the literature ............................................................................................................. 178
14.6 Confidence Intervals around the Mean Difference ....................................................................... 178
14.7 Effect Size: Eta Squared............................................................................................................... 179
14.8 Effect Size: Cohen’s d .................................................................................................................. 180
Chapter 15: Correlation .......................................................................................................................... 187
15.1 Relationships between Variables ................................................................................................. 188
15.2 Scatterplots ................................................................................................................................... 189
15.3 Pearson Product-Moment Correlation Coefficient ........................................................................ 191
15.4 Calculating the Pearson Correlation (r) ........................................................................................ 192
15.5 Proportion of Explained Variance and Residual (Unexplained) Variance .................................... 195
15.6 Characteristics of the Pearson Correlation ................................................................................... 195
15.7 Hypotheses and Sampling Distributions of the Pearson Correlation ........................................... 196
15.8 Determining Statistical Significance of the Pearson Correlation .................................................. 197
15.9 Determining Significance when ρ ≠ 0 ........................................................................................... 198
15.10 Power and Effect size for Pearson Correlation........................................................................... 199
-8-
15.11 The Spearman Rank-Order Correlation (rS) ............................................................................... 200
15.12 Determining Significance of Spearman Rank Order Correlation (rS).......................................... 202
Chapter 16: Regression .......................................................................................................................... 208
16.1 Prediction ...................................................................................................................................... 208
16.2 Regression Equation .................................................................................................................... 208
16.3 Calculating the Slope Coefficient (b0) ........................................................................................... 210
16.4 Calculating the Intercept Coefficient (b0) ...................................................................................... 211
16.5 Plotting a Regression Line ............................................................................................................ 211
16.6 Residual Variance and Standard Error of the Estimate ................................................................ 212
16.7 Homoscedascity ........................................................................................................................... 214
16.8 Making Inferences about Regression Equations and Coefficients ............................................... 214
Chapter 17: Correlated Samples t-Test................................................................................................. 221
17.1 Recap of t-Tests ........................................................................................................................... 221
17.2 What is the Correlated (Dependent) Samples t-Test? ................................................................. 222
17.3 Standard Error of the Difference Between Correlated Samples................................................... 222
17.4 Hypotheses and Steps in the Correlated Groups t-Test............................................................... 223
17.5 Numerical Example ....................................................................................................................... 224
17.6 Effect Size Measures .................................................................................................................... 226
Chapter 18: Introduction to Analysis of Variance ............................................................................... 230
18.1 Limitations of t-Tests..................................................................................................................... 231
18.2 What is ANOVA? .......................................................................................................................... 232
18.3 F-Statistic and Sources of Variance ............................................................................................. 232
18.4 Hypotheses with ANOVA .............................................................................................................. 234
18.5 Experimental Designs in ANOVA ................................................................................................. 235
18.6 Subscripts and Nomenclature....................................................................................................... 235
Chapter 19: Oneway Between Subjects ANOVA ................................................................................. 238
19.1 What is The 'Oneway' Design? ..................................................................................................... 239
19.2 Numerical Example ....................................................................................................................... 240
19.3 Post Hoc Testing .......................................................................................................................... 243
19.4 Effect Size and Explained Variance ............................................................................................. 245
19.5 Statistical Power for Oneway ANOVA .......................................................................................... 246
Chapter 20: Factorial ANOVA ................................................................................................................ 251
20.1 What are a Factorial Design and Factorial ANOVA? ................................................................... 251
-9-
20.2 What do Main Effects and Interactions Look Like? ...................................................................... 252
20.3 Hypotheses in Factorial Designs .................................................................................................. 254
20.4 The Source Table ......................................................................................................................... 255
20.5 Numerical Example of 2 x 2 ANOVA ............................................................................................ 256
20.6 Effect Size and Explained Variance ............................................................................................. 262
Chapter 21: Chi-Square Analyses ......................................................................................................... 267
21.1 When, Why, Where is Chi Square Used? .................................................................................... 267
21.2 Assumptions of the Null Hypothesis ............................................................................................. 267
21.3 Chi Square with One Independent Variable ................................................................................. 268
21.4 Goodness of Fit Test .................................................................................................................... 270
21.5 Chi Square with Two Independent Variables ............................................................................... 272
21.6 Strength and Power of Chi Square ............................................................................................... 275
21.7 Special Case of the 2x2 Chi-Square ............................................................................................ 276
Chapter 22: Power Analysis and Replication ....................................................................................... 284
22.1 Review of Effect sizes and Power ................................................................................................ 284
22.2 Preparing for Replication .............................................................................................................. 285
22.3 Effect Size Computations ............................................................................................................. 286
22.4. Examples ..................................................................................................................................... 294
Appendix A .............................................................................................................................................. 300
Table 1: Standard Normal (z) Distribution Table .................................................................................. 300
Table 2: Probabilities Under the t-Distribution ...................................................................................... 307
Table 3: Probabilities Under the F-Distribution ..................................................................................... 311
Table 4: Studentized Range Statistic (qk) ............................................................................................. 321
2
Table 5: Critical Values in the Chi-Square (χ ) Distribution .................................................................. 323
Table 6: Fisher’s Transformation of Pearson r to r' (ρ to ρ') ................................................................ 325
Table 7: Factorials of Integers 0 – 50 ................................................................................................... 326
Appendix B: Measuring Skew ................................................................................................................ 327
Appendix C: Calculating Power ............................................................................................................. 329
Appendix D: Post Hoc Testing .............................................................................................................. 331
Appendix E: Answers to Homework Questions .................................................................................. 337
References ............................................................................................................................................... 349
-10-
Foreword
This book is designed as an introduction to statistics for undergraduates enrolled in
various scientific disciplines, but is aimed more toward students in the behavioral
sciences. However, because psychological theories can be dull, which would add to the
conceptual-ness of an already conceptual and generally content-dry course, many of
the examples in this book are related to contemporary society. Read, relax and enjoy!
This book should be considered a work that is always in progress. I realize there are
undoubtedly some mistakes, so please feel free to send me comments. Indeed,
comments help, as I want to make this the best possible recourse for students of
statistics. I update the content of this book occasionally (last update: June-July 2012),
but that does not mean that errors are not fixed as needed. If you have a comment, feel
free to leave me some advice.
Why did I write this book? Don't I have enough to do? I believe in free and open
information sharing, which is why I like science, scholarship, and academics:
information flows freely. A few years ago, I realized that the statistics textbooks that I
would use for my undergraduate course were not getting the job done. They were
confusing my students and I ended up having to correct errors, or had to write small
chapters to address topics that were not addressed in the textbook. This is not to say
that those textbooks were horrible; far from it. But, it was frustrating to have my students
purchase a $175-$225 textbook when I knew that the information in these textbooks
was not what my students needed. I decided that since I had a wealth of knowledge
regarding statistics, I could put that information down on paper (well...html actually), and
provide this free to my students. Indeed, I hope that in the coming semesters I can do
the same for my other classes. Read, relax and enjoy!
"Educate and inform the whole mass of the people. Enable them to see
that it is their interest to preserve peace and order, and they will preserve
them....They are the only sure reliance for the preservation of our liberty." -
-Thomas Jefferson, 1787
-11-
Back to TOC
Chapter 1: Introduction to Statistics
1.1 You’ve Decided to take Statistics

At the time of my revising this book (June 2012), the US Congress enjoys a 17% approval rating
(www.gallup.com), President Obama enjoys a 45% approval rating and a 48% disapproval rating
(www.gallup.com), and the US Supreme Court has a 44% approval rating (Liptak, 2012). How well is your
government listening to you?
th
On the eve of the 200 anniversary of Charles Darwin’s birth, surveys showed only 4 in 10 Americans
believe Evolution, while 2 in 10 reject Evolution in favor of a biblical interpretation of the origin of human
life on earth (e.g., creationism, intelligent design, evangelicals), and 4 in 10 are unsure. Although this
trend is influenced by education level (Newport, 2009), this is odd given there is an abundance of
empirical evidence in favor of Evolution and only a book created by man in favor of Creationism. (I’m not
trying to insult anyone, just making an empirical point.)
At the end of the 2012 Major League Baseball season, the New York Yankees had a winning percentage
of 58.6 (proportion of games won = .586), which allowed them to win the American League’s Eastern
Division.
Contrary to popular myth, there is no relationship between the size of a man’s shoe/foot and penis length
(Shah & Christopher, 2002).
What do these have in common? They involve statistics. If statistics had not been invented we could
complain about congress, the president, and the Supreme Court, but could not describe by how much we
approve or disapprove of their performances. If we did not have statistics we could not determine how
many people do not believe in science. If we did not have statistics we could not determine how much
better one sports team performed relative to other sports teams, to determine playoff standings. Hence,
no team could ever be league champion (sort of like youth t-Ball). Finally, without statistics, we could not
empirically disconfirm popular myths about the relationship between the sizes of various body parts.
Thus, we owe a lot to statistics.
Understanding statistics, where they are used, why they are used, what they are supposed to represent,
how they are to be interpreted, how to properly use them, and how they are relevant to your life is
important. Why? Because statistics are used just about everywhere! Second, taking a statistics course
allows you to understand facts and figures presented in the media or in other courses. Finally,
understanding statistics allows you to be just a little bit more skeptical about information presented to you.
If you do not know how to question information, it is unlikely that you ever will. And you’ll be boring.
-12-
Back to TOC
A lot of students are scared of statistics and research-methodology based courses, because they hear
horror stories from former students. Here are a few statements that I have adapted from conversations
with former students:
“The class is horrible, boring, and stupid!”

“You cannot get an A in the class, ever.”
“The material is terrible, dry and evil.”
“Please kill me.”
“Kill me.”
“I want mommy!”
I may have exaggerated those last few, but I have reduced students to tears a few times (just kidding).
Let me digress though some of the comments from above:
First, it's true a professor can be boring and make something more difficult than it should be. But,
ultimately your education is what you make of it. If you don’t like something, speak up! This isn’t
high school. This is college and it’s supposed to be hard. Get over it! If you think something is
horrible, boring, and stupid, perhaps this course or your major is not right for you (just saying).
Second, I’ll admit that statistics is a bit dry at times and the material can be boring. (Please note
that I think there is a correlation between material that makes you think mathematically and
boring-ness.) This is why most examples in class and this book involve real life and occasionally
involve elves, wolverines, Dr. Evil, zombies, the unpopular media, poking fun at the ruling class,
and Peter Griffin. I’ve gotta’ have fun with my job too! But, make up your own examples if you
don’t like mine. What works for me may not work for you and that’s okay. I’ll work with you. Make
it fun. It’s just statistics!
Third, it is very possible to do well in this course. Any student that has received a grade of C- or
less from me in statistics has done so by their own hand. That is, they did not seek my help or the
help of others...and their grade suffered! Surprise surprise! Such students never asked questions
in class or sought me out for assistance. Remember, I am your guide: use me! By the way, don’t
worry about your grade: Worry about learning and comprehending the material; your grade will
fall into place...hopefully. 
Fourth, the material in statistics is different, but it is doable and you can earn a good grade. A lot
of students come into to statistics with preconceptions that it is an impossible course. Nothing is
impossible...making the Kessel Run in less than 12 parsecs. Erase that nugget of misinformation
from your brain immediately! Delete, delete, delete! Statistics is not impossible, or else nobody
would ever pass! Like I said, students who do poorly in statistics are those who do not get help.
Here is a quote to remember for all courses: “If you think you need help, you do need help!”
Finally, statistics is not stupid. You may think it is stupid if you get a poor grade and you can
blame me all you want. But take a moment to think: If you took the time to get help and had
received a good grade, wouldn’t you think it was fun and not stupid? Anyway, welcome to
statistics, I’m glad that you’re here!
1.2 What, When, Where, How, and Why of Statistics

The first thing to understand is what are statistics and where are they used? Below, I provide brief
descriptions of the various characteristics of statistics.
What are Statistics? Statistics are procedures that can be used to combine, organize, and
summarize data. Scientists collect data to generate and test theories and hypotheses, but the
data are initially unorganized and “raw.” Statistics allow one to organize raw data into something
-13-
Back to TOC
condensed and more meaningful. Statistical methods also allow one to examine what is occurring
in their data by identifying and quantifying relationships. Finally, statistical methods ultimately
allow one to determine whether a theory or hypothesis is supported by the data.
When are Statistics Used? Statistics are used whenever you are trying to answer some
question that requires collecting and summarizing data. Whether it is a psychological study on
political attitudes and personality or a newspaper survey of how much readers like a politician;
whenever you have a question that requires a summary of data you will use statistics.
For example: I hypothesize that age is related to

a
political attitude (liberal-conservative); particularly, Age Average Political Attitude
older individuals tend to be more conservative in 15 3.0
political attitudes. One way to address this hypothesis
would be to survey people, ask for their age and ask 20 3.5
for their political attitude on a liberal-conservative
25 5.0
scale. Based on the collected data I can determine
whether there is a relationship between age and 30 5.5
political attitude. A set of fictitious data that lists the
age and political attitude of several people is 35 6.0
presented to the right. You can see that older 40 6.5
individuals tend to rate their political attitude as more
conservative than younger individuals. Although these 45 8.0
data appear to support my hypothesis the only way to 55 9.0
know whether a relationship exists is by using
statistics to determine whether the apparent 60 9.5
relationship is ‘statistically significant’. a
’Political Attitude’ was measured on an 11-point
scale, from 1 (liberal) to 11 (conservative).
Where are Statistics Found? Statistics are used not
only by scientists; all professions use statistics. Businesses, teachers, schools, sports,
government, the military, the justice system, religions, and social organizations use statistics in
their own ways.
For example, at the end of the 2007 Major League Baseball season the batting average (AVG) of
Derek Jeter was .322, which means he hit the ball and made it to safely to base about 32.2% of
the times he went up to bat. This value is a type of statistic that players, teams, sports writers,
and fans use to gauge how well a player is performing.
As another example, at the time of my revising this book, President Obama is enjoying a 45%
approval rating with a margin of error of ±3%. This survey, performed by the Gallup Poll, included
1,500 adult participants. People use these numbers, based off of a small sample of participants to
infer what percentage of all Americans would approve or disapprove of President Obama’s job
performance. Indeed, statistics generally uses date from a small group, or sample, and makes
inferences about what would be found in a larger group, or population.
How do Statistics work? This is a bit of a difficult question to answer, because there are so
many procedures and analyses. Generally, there are three main “tasks” of statistics:
1. Collecting data.
2. Consolidate, organizing, and exploring data.
3. Drawing conclusions and making decisions based on the data.
Why do I need Statistics? There are a lot of reasons, here are a few:
1. You need to know statistics in order to understand most scientific research. Indeed, most
research methods that you encounter in this course and in Research Methods (PSYC
-14-
Back to TOC
330) go hand-in-hand with statistics. If you have a research
question but cannot find a method to analyze the data, it is not
worth doing the research. In research methods it is likely your
professor will discuss a lot of data analysis procedures in depth.
Indeed, a research methods course is like 'Statistics 2'.
2. Statistics are almost always used to explain topics in upper-level

psychology courses. This is especially true for empirical courses
like Sensation & Perceptions, Cognition, Conditioning & Learning, Social Psychology,
and any course in Neuroscience. If you start taking those courses, but do not know what
‘statistically significant’ means (a topic we will cover later), it is unlikely you will
understand the content.
3. Statistics also allows you to understand information presented in the media. You always
hear percentages and proportions of people that agree or disagree with some topic. For
example, in President Obama’s approval ratings from above, I mentioned that the margin
of error was ±3% in that survey. What does this mean? If you know something about
statistics you will be able to understand. Also, knowing statistics allows you to judge for
yourself whether something in the popular media makes sense or not. The bottom line is
that knowing statistics allows you to understand things going on around you.
1.3 Statistics as a Language

Dr. Burnham’s First Law of Statistics:
Statistics is a language course more than it is a mathematics course.
Dr. Burnham’s Second Law of Statistics:
Repeat the First Law.
Dr. Burnham’s Third Law of Statistics:
Don’t be a dumbass.
I know what you are thinking, what the hell is Dr. Burnham talking about? After all, the chapters are filled
with number-like things and a lot of funny equation-looking-function doohickeys, which look similar to the
alphabet soup you ate last night. How can statistics be a language course; how is it not a mathematics
course? (Note that you do not get GE-Humanities credit for a course in statistics, even though I consider
it a language course.) Statistics involves numbers and involves using formulas to solve statistical
problems, this is true. But, there is more to statistics than just numbers and formulas; and this is what you
should focus on in this course. The concepts, content, and language of statistics is key to understanding
the material in this course and beyond. This course is designed to teach you to ‘read’ statistics; thus,
you’re learning a new language.
Why is learning statistics like learning a language? Statistics is best understood at a conceptual level just
like any language. When you learning a new language you do not focus solely on the new words in
isolation, rather, you relate words from the new language to the meaning of related words from your
native language. The mathematics used in statistics is easy. All you will be doing is adding, subtracting,
multiplying, and dividing. The most difficult thing to do is take square roots, which is done with a
calculator. The math is simple! Thus, you need to focus on the concepts and relate those to ideas you
already have: do not focus on the numbers and equations!
-15-
Back to TOC
The difficult part to statistics is that you will use the mathematical concepts that you should already know
in new and different ways. This is where students get into trouble. Students tend to focus solely on each
term in each formula, but fail to integrate concepts to comprehend the overall meaning statistical
procedures. For example, when you read a word (e.g., corruption) and try to retrieve the meaning of that
word from your mental dictionary, you don’t focus on each letter in the word (c-o-r-r-u-p-t-i-o-n) to retrieve
meaning. Instead, you focus on the entire word to comprehend its meaning. The same is true in statistics.
Don’t focus on the minuscule pieces of formulas and on the numbers; focus on the language and on the
big picture. If you do this, the little pieces and the numbers fall into place.
Another reason why I say learning statistics is like learning a language is that many of the steps involved
with completing a statistical procedure reflect their “theoretical definitions.” For example, there is
something called the sum of squares, which is a measure of how alike scores are in data. The sum of
squares is defined as the sum of the squared deviation scores from a mean. I know you have no clue
what I am talking about yet, but here is the symbolic formula for the sum of squares:

 XX 2
See if you can find the theoretical definition from above in symbolic formula. That is, think of the definition
I gave above and see if you can locate the sum of the squared deviations from the mean. Most statistics
are like this. If you can understand the underlying meaning and language, the math is easy!
Finally, because a lot of you will likely be reading reports, articles, and textbooks that include various
statistics, you need to ‘learn the language’. That is, when you see SD = 12.558 seconds, you need to
know how to interpret this. Thus, there is a language to statistics that you will learn to comprehend.
1.4 Two Types of Statistics

Statistical procedures are classified into one of two categories: descriptive statistics and inferential
statistics. Descriptive statistics are designed to convert raw data it into a representative value that can
be used to describe (represent) the entire set of data. Descriptive statistics are like a member of the
House of Representatives for a congressional district and the raw data are like that congress member’s
constituents. That is, within congressional districts voters elect a person to represent their interests and
their district in the Congress. The same is true for a set of data. Thus, descriptive statistics are single
values that represent something about an entire data set.
For example, take the following set of data { 3, 4, 6, 7, 3, 4, 5, 8 }. You might take the average of these
scores (Avg. = 5) and use that as the representative value for the data set. As another example, you
might have the following set of data { 7, 7, 7, 7, 7, 7, 7, 7 } and could use 7 as a representative value,
because 7 is the only number in that data set. Thus, descriptive statistics summarize a set of data.
To make judgments and decisions about a set of data, to look for relationships, and to determine the
“statistical meaningfulness” of relationships, you use inferential statistics. Inferential statistics analyze
relationships in a set of data and determine whether they are statistically significant; that is, present and
unlikely due to random factors. They are the tests and analyses that can be performed on a set of data to
allow inferences to be made about that set of data.
Alcohol Consumption Average Attractiveness
Rating
For example: Say that I want to examine whether No Alcohol 6
the perceived attractiveness of member of the 2 oz. Alcohol 6.5
opposite sex increases as one consumes more 4 oz. Alcohol 7
alcohol. I collect data from groups of males and 6 oz. Alcohol 7.5
females. Each group consumes a different level of 8 oz. Alcohol 8
alcohol, but all groups rate the attractiveness of a 10 oz. Alcohol 10
picture of the member of the opposite sex on a 12 oz. Alcohol 10
-16-
Back to TOC
scale from 1 (unattractive) to 10 (attractive)). The hypothetical data are presented in the table to the right.
The average attractiveness ratings should be treated as descriptive statistics, because each value is the
average of the attractiveness ratings provided by individuals in each alcohol group. You can see that as
alcohol consumption increased attractiveness ratings tended to increase. Thus, there appears to be a
relationship between alcohol consumption and perceived attractiveness of a member of the opposite sex.
Inferential statistics would allow me to determine whether this relationship is meaningful, that is,
statistically significant; a topic we will return to later.
1.5 Variables
In any area of research you work with variables, or anything that is measurable and can change. For
example, blood pressure is a variable, because it can vary both across people and across time for a
person. Sex (gender) is a variable, because it differs across people. Political attitude is a variable,
because it can differ across individuals. Variables are often labeled with an uppercase letter such as X or
Y, especially if the variable is something that is measured. For example, you may be measuring political
attitudes and for sake of brevity use X to denote the variable ‘Political Attitude’. The only requirement for
something to be a variable is that it must have at least two levels. If something can only have one level, it
is a constant. Constants are anything that remains stable and does not change. For example, a person’s
sex (their being biologically male or female) is constant, because sex cannot change naturally. Numbers
are constants, because 5 means the same thing over time and across situations. Although letters are
usually used to designate variables a letter can be designated a constant if you set that letter equal to a
number (X = 5).
In psychology and most of the other social and behavioral sciences, the variables that are being studied
reflect a hypothetical constructs; that is, something that we believe exists and explains behavior, but is
not directly observable. For example, memory is a hypothetical construct because we cannot actually see
memory in that we cannot see the things that a person remembers, but we think that memory exists,
because information that a person retains will influence behavior.
You need to have an operational definition of the construct that you want to measure. Operational
definitions are clear and concise descriptions of the procedures being used to measure a variable.
Because most of the constructs that psychologists study (and most other scientists as well) are not
directly observable, we must formulate/define an indirect method of measuring these constructs. Thus,
using memory as an example construct, one must have a clear definition stating exactly how memory is
going to be measured. This measurement of memory must be directly observable. Thus, operational
definitions describe how an unobservable construct is measured through some observable variable.
For example, we might operationally define memory as the number of words recalled from a list of 100
words studied 24 hours earlier. We will not dwell too long on operational definitions; just note that when I
say that we are measuring something unobservable like memory, I mean we are measuring something
observable that we have operationally defined as memory.
There are many types of variables, but for statistics, we will be focusing mainly on two: A dependent
variable is anything that is being measured, that is, any behavior or response that can be quantified.
Dependent variables are the operational definition for a hypothetical construct. If we define memory as
the number of words recalled from a list that was studied 24 hours earlier, then the number of words
recalled is a dependent variable. Alternatively, if we measure political attitude by using a rating scale from
1 (extremely liberal) to 11 (extremely conservative), then the ratings that people give on the scale is a
dependent variable. Thus, the construct (memory and political attitude) is not the dependent variable; the
thing that we are directly measuring is the dependent variable.
An independent variable is anything that changes between situations. Independent variables can be
something that naturally differs between groups or situations, for example, sex (male vs. female) differs
between people. Alternatively, independent variables can be something that can be directly manipulated
-17-
Back to TOC
or changed across groups or situations. For example, giving one group of people an anti-anxiety drug and
giving another group an inert placebo is manipulating the independent variable drug condition between
these two groups. Independent variables that are directly manipulated are often called true independent
variables. If a dependent variable is a function of the independent variable, a researcher can be relatively
confident that the change in the dependent variable was related to that change in the independent
variable.
For example, I am interested in the influence of noise on studying. I assign one group of 10 students to
study between 7:00PM to 8:00PM each night while I play loud music, and I assign a second group of 10
students to study during that same time, but I play no music. Thus, I have manipulated noise level
between the groups. To examine the influence of noise on studying I could compare exam scores on the
studied material between groups. If the manipulation of noise had influenced studying (as measured by
the exam scores), then I might expect the results to the right.
You can see that the change in the independent variable (noisy
Noisy Quiet
versus quiet) was associated with a change in the dependent
variable (% Correct). Before I can conclude that the difference in 70% Correct 85% Correct
the exam grades was due to the difference in the noisy versus quite conditions, I must use inferential
statistics. Why? Because it is possible that this difference happened by coincidence. Inferential statistics
test whether the likelihood of this difference in the dependent variable is unlikely due to coincidence, and
hence, is likely due to something else.
Extraneous variables are unrelated to the dependent variable and independent variable and are
generally allowed to randomly vary. An example of an extraneous variable might be a person’s age. It is
unlikely that every subject in a study will have the exact same birthday. Thus, age is allowed to vary
randomly unless it is a variable that should be controlled. In those cases extraneous variables must be
controlled. A control variable is anything that is extraneous to the study, but remains constant across the
levels of independent variables. For example, studies of attention often include only right-handed
individuals; hence, handedness is controlled.
Extraneous variables are not a problem, unless they become confounding variables. A confounding
variable is unrelated to the study (extraneous), but for whatever reason the levels of a confounding
variable change as the levels of the independent variable change. From the example above, say that
students in the noisy condition had an average GPA of 2.00, while the average GPA for students in the
quiet condition was 3.67. In this case, the manipulation of noisy versus quiet studying conditions was
confounded with GPA. (See Table at right) The difference in GPA across levels of the independent
variable is a problem, because there is an alternative explanation
for the observed difference in exam grades between the noisy Noisy Quiet
and quiet conditions. It could be that the higher exam grade in the Exam Grade 70% 85%
quiet condition was actually due to that quite group having
smarter students. Average GPA 2.00 3.67
Unfortunately, there is no way to tell whether the change in exam grades was due to the noise
manipulation or due to the difference in GPA. The only thing that can be done in such a situation is rerun
the study but hold GPA constant across the levels of the independent variable. Of course, this situation
can be completely avoided by careful preparation and planning of a study.
1.6 Values vs. Names of Variables

There are many types and uses of variables, and you need to be very careful when you refer to a variable
so that you don’t refer to it in terms of its potential values. That is, don’t confuse a variable with one of the
possible levels or values the variable can take. For example, a study on college students may require
subjects to identify themselves by college class and each subject would respond they are a freshman,
sophomore, junior, or senior. These choices are the values or levels of the variable. Thus, senior is not a
-18-
Back to TOC
variable as it does not vary from person to person; however, college class is a variable, as it does vary
from person to person.
This is an odd concept, because people can be seniors or not be seniors, so ‘senior’ seems as though it
should be a variable. What differs across people is not ‘senior’ it is ‘being a senior or not,’ which is just
another way of saying ‘college class.’ As a word of warning, be mindful of this when you are referring to
variables vs. levels. If you refer to a level of a variable when you should refer to the variable itself, you are
wrong!
1.7 Quantity vs. Quality

A main distinction among variables is between quantitative variables and qualitative variables. A
quantitative variable is measured as a number and to which mathematical operations can be
meaningfully applied and the outcomes of operations would make sense. Thus, a quantitative variable is
one where levels of the variable are numerical and can be used to express differences in amounts or
ratios. Examples include height, age, GPA, temperature, salary, length, etc.
A qualitative variable (categorical variable) is any variable that is not quantitative and a measurement
takes on one of several categories. Thus, when measured, a qualitative variable will have no numerical
meaning attached to the levels, such that data collected from a qualitative variable cannot express
numerical differences or ratios, but can express differences in types. Examples include sex (male vs.
female; not ‘how many times per week,’ which is quantitative), hair color, college major, college class,
race, political orientation, national affiliation, etc.
Importantly, qualitative variables may be disguised as a quantitative variable. One example is in digital
data files (e.g., Excel, SPSS, R, Minitab), where the levels of the variable sex may be entered as 1 for
male and 0 for female, or vice versa. In this case, the variable sex is still qualitative, because sex is not
actually measured as a number, but rather is measured as a category. Assigning numbers to categories
is dummy-coding and is done mainly because it is easier to work with and to manipulate numbers in
data files than it is to manipulate categories.
Another example of a qualitative variable being disguised as a quantitative variable is a Likert scale,
which is a rating scale where numbers are assigned to different levels of agreement or likability. For
example, here is a question that one of my colleagues and I ask in most studies that we conduct on
political orientations:
Overall, what is your general political attitude?

-5 -4 -3 -2 -1 0 +1 +2 +3 +4 +5
extremely very moderately somewhat slightly totally slightly somewhat moderately very extremely
liberal liberal liberal liberal liberal neutral conservative conservative conservative conservative conservative
Subjects are asked to circle the number coinciding with their level of liberalism-conservativism. But, the
numbers are meaningless, because I could use the following scale with different numeric values, but
where the categorical levels are the same:
1 2 3 4 5 6 7 8 9 10 11
Thus, people are responding to how liberal-neutral-conservative is their political orientation, which is
qualitative, not quantitative. Indeed, say two subjects (A and B) are measure on this scale. Subject A
selects 2 (very liberal) and subject B selects 10 (very conservative). Taking the average of these two
measurements I get a value of 6 [(2 + 10)/2 = 6], which is associated with the “totally neutral” category.
Were these two individuals totally neutral in their political orientation? Hardly! But based on a quantitative
analysis of qualitative data I would erroneously make this conclusion. Thus, be careful when assessing
-19-
Back to TOC
and interpreting Likert-scale data. These scales are used throughout psychology and the results can be
misleading.
The qualitative vs. quantitative issue is VERY important in the use of inferential and descriptive statistics,
so you should watch for these terms throughout the book. Indeed, the types of variables (qualitative vs.
quantitative) used dictates the type of inferential statistic and types of descriptive statistics you would use
to analyze collected data.
1.8 Hypotheses and Predictions

Statistics are used to evaluate research questions. Specifically, when a researcher has a question s/he
wants answered, s/he designs a study, collects data, and then applies appropriate statistical procedures
to determine whether the data support, do not support the question. Thus, everything centers on the
research question in a study, which we call a hypothesis, which generates testable predictions.
Before I describe the two hypotheses used in research, let me set up a very hypothetical scenario: Say
that I am a researcher on Zombie Island and I am studying the eating habits of zombies. (I told you I
would use zombies as examples!) After much field work, I hypothesize zombies who listen to The Cure
will eat more brains than zombies who listen to Aerosmith (Come on...Robert Smith of The Cure looks like
a zombie!) I have a group of zombies listen to The Cure for one month while another group of zombies
listens to Aerosmith for that same month. I survey each zombie in each group asking each zombie the
number of brains they ate over that one month period. I then compare the average number of brains
eaten in each group to determine whether my hypothesis is supported.
There are two hypotheses researchers generate a null hypothesis (H0) and alternate hypothesis (H1).
Each hypothesis includes a specific prediction regarding what will be found when data are collected. The
null hypothesis states that your research question will not be supported by the data. Thus, it would predict
no difference in the dependent variable. The alternate hypothesis predicts that your research question will
be supported by the data. Thus, it would predict a difference in the dependent variable.
Using the zombie example, my null hypothesis would predict no difference in the number of brains eaten
between zombies listening to The Cure and zombies listening to Aerosmith; and the alternate hypothesis
would predict a difference in the number of brains eaten between zombies listening to The Cure and
zombies listening to Aerosmith. Both hypotheses make opposing predictions: The null hypothesis predicts
no difference in the dependent variable and the alternate hypothesis predicts a difference in the
dependent variable.
1.9 Populations and Samples

Any scientist has a target group of interest s/he is studying. For example, if you are interested in studying
the behavioral consequences of having seasonal affective disorder your target group would be people
who have seasonal affective disorder. On the other hand if you are interested in studying the
effectiveness of a new statistics teaching method your target group would be statistics students. If you are
interested in studying the use of a new hip-checking technique for playing hockey your target group of
interest would be all people that play hockey, or people who play in a league where hip-checking is OK.
A population includes all members of a group and the number of people in that population is usually
labeled with an uppercase N. The population is based on the type of subjects you wish to study. In the
first example the population would be all people with seasonal affective disorder; in the second example
the population is all statistics students; and in the third example the population is all hockey players.
Importantly, you want your population to be specific. To use a rather extreme example, if you were
studying depression your population should be people with depression; it would be inappropriate to have
your population be people with anxiety.
-20-
Back to TOC
You make inferences about a population based on collected data; hence, you want to be sure you make
inferences about the correct population. For example, you may conduct research on depression, but the
people with whom you conduct the research were for whatever reasons males between the ages of 20
and 25 with mild depression. Thus, you cannot make inferences about all people with any form of
depression; rather, you can make inferences about all males between the ages of 20-25 with mild
depression. Thus, it is critical to have your subjects reflect your intended target population.
Any statistical measurement based on a population is a population parameter. The average depression
of all males with mild depression would be a population parameter. The batting average of all hitters in
Major League Baseball would be a population parameter. Determining the number of brains that all
zombies on Zombie Island eat is a population parameter. Generally, the symbols used to denote
parameters are Greek symbols (uppercase and lowercase), such as σ (sigma), ρ (rho), μ (myu), and ω
(omega). I have listed some of the more common symbols and their meanings in the table below.
Population Sample Population

Term Measure of... Symbol Symbol Estimate Symbol
Mean Arithmetic average μ X or M X or M
Size Number of subjects N n n
Sum of Squares Total variability SS SS SS
Variance Average variability σ
2
s
2
ŝ 2
Average distance between a
Standard Deviation
value and the mean
σ s ŝ
Strength of relationship
Correlation ρ r r
between two variables
It is impractical and often impossible to conduct research using a population, because most populations
are very large, or are theoretically infinite, and cannot be measured. For example, some researchers,
especially those conducting cognitive research, do so with the intention of making inferences about
‘humans’; thus, the population for hose researchers is every living human being! It would be difficult to
conduct research on that large a scale.
Instead of conducting research using populations, researchers rely on samples from a population, which
are small and representative groups selected from the target population. The size of a sample is denoted
n, and they are used out of convenience. Samples are often readily available and data are much easier to
collect and to work with. As long as the sample is representative of the population, a researcher can use
the data from a sample to make inferences about what would be found in the population. That is, as long
as individual differences from a population are represented in a sample, the researcher can make
conclusions about what would be found in the population. Thus, samples are more convenient because
they are more available and can be used to estimate what would be observed had the entire population
been studied.
Any statistic that is based on data from a sample is a sample statistic. For example, the batting
average of a randomly selected group of 100 Major League Baseball players is a sample statistic. The
number of brains eaten by a randomly-chosen group of 10 zombies from Zombie Island is a sample
statistic. Generally the symbols used to label statistics are English letters (usually lowercase), such r, and
t. Some of the more common symbols and their meanings are listed in the table above.
Importantly, when you use sample data to estimate a population parameter, it is customary to include a ^
(hat) above the symbol for the sample statistic. Please be aware of this! Sample statistics and population
estimates from sample data mean the same thing, but the two values are calculated slightly differently.
Sample statistics measure something within a sample, and population estimates from sample data
estimate a parameter that should be found in a population. If you want to describe only the sample, use a
sample statistic, but if you want to use a sample to make inferences about the population then use a
population estimate.
-21-
Back to TOC
1.10 Sampling
The main issue with sampling is whether the sample selected from a population is representative of that
population; that is, whether the subjects in the sample accurately reflect the subjects of the population.
The best way to ensure a representative sample is through random sampling. This involves selecting
people from the population to be a member of your sample at random and without bias. For example, if I
were to select a random sample of zombies to make up a sample of zombies on Zombie Island, I might
put the entire population of zombies’ names into a hat and randomly select 10 names. These 10 zombies
would then make up my sample.
The downside to sampling is that occasionally, and even with random sampling, your sample may be
biased and non-representative of the population. This occurs when you get too many individuals from a
certain sub-group within a population into your sample. For example, you get too many males or too
many females in a sample. Or, you get too many people with a high IQ in a study. Unfortunately, it is
difficult to detect this, and really the only way to measure the people in a sample on any relevant variable,
to determine whether those variables may have contributed to any observed outcome.
The other issue with sampling is whether you should sample with replacement or sample without
replacement. When you sample with replacement, after you select someone for your sample you place
that person back into the population and they can be selected again. The plus side is that because the
population will always have the same number of individuals, the chance of being selected is always the
same. The downside is that the same person could be selected more than once. In sampling without
replacement, after someone is selected from a population and placed into a sample, they are not returned
to the population and cannot be re-selected. The plus side is, of course, that the same individual cannot
be selected more than once. The downside is that after every selection from the population the size of the
population decreases by one person. Thus, the chance of being selected increases as more individuals
are selected. This only becomes a problem if you have a small population. If the population is large,
which is usually the case it is of little concern.
1.11 Random Selection

I have a research study that is ready to go and I need to collect human data. Let’s say the study is
examining differences in attention between video game players and non-video game players. The study
involves a survey to assess the amount and types of video games people play and a computerized task
to assess attention. For reasons I won’t go into here, I have determined that I need n = 50 subjects for
this study. Lo and behold, I am teaching two sections of statistics and each section has 25 students!
Sweet! I can just recruit all of the students in these two sections for my study (subjects would be properly
compensated of course).
What’s the problem with my using the 50 statistics students in my two classes for the sample in my study?
The problem is this sample is one of convenience and is probably not representative of my intended
population. Actually, in this example I have two intended populations: (1) people who play video games,
and (2) people who do not play video games. It is entirely possible and likely that I would find some video
game players and non-players among these 50 statistics students. But how well do they represent the
general population of video gamers and non-gamers? It’s tough to say.
So, the main problem is that the sample that I use (i.e., the 50 statistics students) is not representative of
my intended populations. When creating a sample the most important thing to remember is that the
sample must be representative of the intended population. This means that individual differences among
subjects in the population are represented, proportionally, in your sample. What I mean by ‘represented
proportionally’ is that large sub-groups in a population have more subjects in a sample and smaller sub-
groups have fewer subjects in a sample.
To use another example about need for a sample to be representative, let’s say that I run a study that
examines vitamin C intake and intelligence. Assume that I teach at MIT (Massachusetts Institute of
-22-
Back to TOC
Technology) and run the study there. What’s the problem with using a sample of MIT students for this
study? MIT has a lot of highly intelligent individuals and very few individuals of low intelligence. Thus, the
sample would not be representative of all intelligences levels. The same is true for the 50 statistics
students. Perhaps statistics students have good attention spans or perhaps they all play a lot of Tetris or
Bomber-Man. This could obviously skew the results. Thus, when selecting subjects for a sample out of
convenience (i.e., convenience sampling) there is a low probability that your sample will be
representative of your intended population.
Aside from representativeness, there is a second problem with using the 50 statistics students for the
sample in my video gaming and attention study, which is related to why this sample would not be
representative of the population. By using these subjects out of convenience I have introduced selection
bias into my sampling procedure. This bias stems from the fact that I allowed only certain individuals into
my study (e.g., college-educated, statistics students, aged 18-22, etc.). By using a sample out of
convenience, you allow individuals with only certain characteristics into the sample, which produced a
naturally biased sample, that is, one that is not representative.
The solution is random sampling, which is simply selecting individuals for a sample in a random,
unbiased order. Think of random sampling as putting everyone’s from a population name into a hat and
then randomly drawing n names, randomly, from the hat to create a sample. Instead of putting names into
a hat and drawing names at random, modern researchers generally do something a little more
technologically advanced. Modern random sampling makes use of random number generators, which
function exactly how their name implies: they generate random numbers. The random numbers that are
generated represent those individuals who are selected for a sample.
For example, say I have the following population of red and blue letters X and O and I need to create a
sample of n = 10 from this population. Assume there are equal numbers of red and blue Xs and Os in the
population (N = 10 for each group). Here’s a graphic representation of this population:
X X X X X X X X X X
O O O O O O O O O O
X X X X X X X X X X
O O O O O O O O O O
To engage in random sampling, the first step is to assign each potential subject a number. In this case, I’ll
just use 1 – 40, which I have placed as subscripts below:
Population
X1 X2 X3 X4 X5 X6 X7 X8 X9 X10
O11 O12 O13 O14 O15 O16 O17 O18 O19 O20
X21 X22 X23 X24 X25 X26 X27 X28 X29 X30
O31 O32 O33 O34 O35 O36 O37 O38 O39 O40
Next, using a random number generator (http://www.randomizer.org/), I generate a set of 10 random

numbers {1, 3, 10, 16, 18, 21, 22, 34, 38, and 40}. I generated 10 random numbers, because I need a
sample of n = 10. Subjects with these numbers will be placed into my sample:
Sample
X1 X3 X10 O16 O18 X22 X22 O34 O38 O40
-23-
Back to TOC
The point of random selection is that each unique sub-group from within a population is represented
proportionally in your sample. In this case, it would mean having an equal number of red Xs, red Os, blue
Xs and blue Os in the sample because there are equal numbers of these four groups in the population.
Obviously, with four groups and a sample size of n = 10 this cannot happen. But, you can see that there
are approximately equal numbers of red Xs, red Os, blue Xs and blue Os in the sample. Thus, the sample
is fairly representative of the population.
Does random selection guarantee that a sample will be representative? Not at all! It is possible that a
random number generator will give the following set of ten numbers {1, 2, 3, 4, 5, 6, 7, 8, 9, and 10}. This
is just as likely to happen as the set above. The point of random assignment is to increase the likelihood
that individual differences in the population are distributed proportionally in a sample. If random selection
is effective, larger groups in a population will have more members in a sample and smaller groups within
a population will have fewer members in a sample.
Is there a way to know whether random selection was effective? The only way to know for certain is to
assess individual differences among subjects and show that pertinent individual differences are
represented equally in a sample. However, this is difficult to do, as it requires measuring people on lots
and lots of factors, and is time consuming.
1.12 Random Assignment

In a particular population, there are happy people (), frowny people (), and stoic/non-emotional people
(). Through random selection, you obtain the following sample of n = 12 people, which contains ten
stoics and two frowny people:
     
     
You need to create k = 2 two groups from the n = 12 subjects in your sample. You could simply split the
group in half, where the subjects on the left are placed into Group A and the subjects on the right are
placed into Group B:
Group A Group B
     
     
What is the obvious problem? Group B has the two frowny individuals from the sample and Group A has
only stoic individuals. (Group A must be super fun!) There is no guarantee this would happen if you
divided your group of n = 12 in half like this, but it is a possibility.
What’s the problem? Say that you manipulate some independent variable between the groups and
measure some dependent variable in each group. For example, individuals in each group are timed for
long it takes to solve a puzzle, but for Group A the puzzle is easy and for Group B the puzzle is hard. The
average time for individuals in Group A to solve the easy puzzle is 2 minutes and the average time for the
individuals in Group B to solve the hard puzzle is 4 minutes.
On the surface (at the level of the average completion times), it appears solving hard puzzles requires
more time than solving easy puzzles. But beneath the surface lies the problem: The two frowny
individuals make group B fundamentally different from Group A, and any difference in completion times
between the groups could be attributed to this fundamental difference in the makeup of the two groups.
Perhaps frowny people take an exceedingly long time to solve puzzles, regardless of a puzzle being easy
or hard, which made the overall average completion time longer than would be expected.
What is the solution? First, it is extremely unlikely you would know for certain that you have two frowny
individuals and 10 stoic individuals in your sample; and that’s the catch. That is, the only way to know the
uniqueness of individuals in a sample is to assess them to determine their personality, mood, blood type,
-24-
Back to TOC
favorite Kafka quote, etc. But, researchers don’t do this, because we don’t have the time or the need, that
is, unless frowny-ness is relevant to your research study, there is no reason to asses a person’s being
frowny, happy, or stoic. And the same is true for any other individual difference that you can think of.
The solution to the problem above is random assignment, which is placing subjects from a sample,
randomly, into different groups. Random assignment is sort of like a second act of random selection,
where everyone in the sample is assigned a number and then numbers/subjects are randomly selected to
be placed into Group A or Group B. For example, say that the n = 12 individuals from above are assigned
the following numbers:
=1 =2 =3 =4 =5 =6

=7 =8 =9  = 10  = 11  = 12
Using a random number generator (http://www.randomizer.org/), I generate a set of six random numbers
{2, 4, 5, 7, 11, and 12}. Individuals with these numbers will be placed into Group A and the other subjects
will be Group B:
Group A Group B
=2 =4 =5 =1 =3 =6
 = 7  = 11  = 12 =8 =9  = 10
Now, you have one frowny individual in each group, hence, there is an equal distribution of frowny-ness
and stoic-ness between groups. The point of random assignment is to increase the chance that individual
differences in your sample are distributed equally between the groups created from the sample. If random
assignment is effective, then any measurable difference between the two groups is likely due to the
manipulation of an independent variable and not due to a priori differences between the groups. However,
there are two important points to make:
(1) Random assignment does not ensure individual differences in a sample will be distributed equally
between groups; random assignment only increases the likelihood of this happening. For example,
another set of random numbers might be {2, 4, 5, 8, 10, and 11}. In this case, both frowny individuals
(underlined numbers) would be in Group A. Thus, there is no guarantee random assignment will always
work.
(2) There are theoretically an infinite number of individual differences across people; it’s what makes
everybody, usually, pretty cool. We cannot account for every single one of them; hence, it is likely there
will be subtle differences between groups based on individual characteristics. This is not meant to make
random assignment seem futile or unnecessary; the best chance we have of equally distributing all of
those individual differences, known and unknown, is by randomly assigning people to groups.
1.13 “Law” of Large Numbers

In later chapters I discuss such topics as effect size,
statistical power, directionality, and alpha level. All of
these are relevant to determining the appropriate
sample size that should be used for a given research
study, but right now it is premature to address this. So,
please keep in mind this section is going to be
discussed in very general terms.
There is something to be said for strength in numbers;

this is why The Borg is such a relentless force in the
Star Trek universe. Statistically speaking larger sample
sizes are generally better than smaller ones (strength
in numbers). This issue will come up repeatedly
-25-
Back to TOC
throughout this book and the class, so please keep this in mind. Most of the reasons for larger sample
sizes are quantitative in nature and we’ll deal with those as they arise. Below, I lay out a simple case of
why larger samples tend to be more representative of a population (strength in numbers).
Say I have the following population (N = 30) of frowny, happy, and stoic people:
           
           
     
Most people in this population are stoics and only a few people are happy or frowny. We call these
individuals who occur infrequently in a population ‘unique’ or outliers. The outliers, though part of the
population and should be part of the sample, are definitely not the most representative of the overall
population. Hence, you do not want many of them in a sample.
Say that I randomly select a sample of n = 5 for a study (assume I used the random number generator).
Here’s the sample that I end up with:
    
The problem? Two of the three happy individuals made it into my sample, no frowny people did, and three
stoic people did. Thus, the happy group is overrepresented and the frowny group is unrepresented. Thus,
the overall sample is not representative. What can I do? Select more people for the sample. Assume that
I select an additional five individuals, in addition to the original five, for a sample size of n = 10:
Original Five Additional Five

         
You can see that I now have one frowny individual; hence, that group is represented in the population.
Assume I choose another ten individuals, in addition to the n = 10 above, to create a sample of n = 20:
Original Ten
         
Additional Ten
         
You can now see that I have two frowny individuals, two happy individuals, and a lot of stoic individuals.
Thus, both small groups are represented equally in the sample, and have a smaller representation that
the stoics, which was the largest group in the population. In this case, with the larger number of
individuals, my sample is more representative of the population (strength in numbers).
The basic idea is that as you increase the sample size, you capture more and more of the individual
differences in the population, which actually makes your sample more and more representative of the
population. With a small sample, if you get one or two outliers, the sample is less representative than if
you had a larger sample. Thus, strive for larger samples; resistance is futile!
1.14 Representative of Whom?

One issue with sampling and human data collection is what population a sample actually represents. That
is, even if you engage in appropriate random selection and random assignment so that a sample should
be representative of a particular population; the question remains, does the sample represent the
intended population, some other population, or a sub-group within the intended population?
This is especially relevant given that most psychological research is conducted at colleges and
universities, using college students as research subjects. Thus, while an investigator may use appropriate
-26-
Back to TOC
sampling techniques to ensure that subjects were selected for a sample without bias, the sample itself
may never be truly representative of a population outside of the college student population.
Stated differently, assume the intended population for an investigator is “people,” that is, normal, run of
the mill people. If the investigator’s sample consists of only college students, then one may question how
well this sample represents “people.” This is because college students have characteristics that may not
be reflective of everyone in the human race. College students are generally of higher intelligence
(especially at academically rigorous schools such as MIT and Williams College), college students often
have a slight socioeconomic advantage (especially at expensive, prestigious schools such as Yale and
Harvard), college students are of the same age group (18 – 23), etc. Thus, a sample of college students
may not be representative of “all people,” rather, a sample of college students may be representative of,
well, other college students.
Even then, because the general makeup of students differs across colleges and universities, a sample of
college students at a particular school may be representative of college students only at that particular
school. Specifically, students attending The University of Albany (SUNY-Albany) are going to differ from
students attending Boston College, who will differ from students attending Yale and MIT. Thus, a
particular sample of college students may not represent “all college students,” but may represent “college
students from school ‘X’.”
Even more reductionist, the sample of college students at a particular school may represent only those
students enrolled in Introduction to Psychology, or only those students who sign up for research studies.
This exercise can go on and on. Thus, it becomes critical that interpretations of data collected within a
population not extend beyond the actual population that is being represented by a sample.
1.15 Research Designs and Statistics

I feel it is worthwhile to address the general research designs that you will encounter in this book. I
present here a very general overview of the various research designs that can be used in the sciences;
these would be substantially elaborated on in a research methods course.
Research designs differ along a number of dimensions, but one of the main differences between designs
is whether independent variables are being used. Descriptive statistics are used in any research design,
and this is to measure the hypothetical construct and quantify a response/behavior. Inferential statistics
are used in most research design, and use the sample data collected to make generalizations about what
would be found in a population. All designs should make use of random-sampling, where each subject in
a population has an equal chance of being selected for the sample, unless there is a reason for non-
random sampling. Finally, the choice of research design depends, mainly, on the research question.
Correlational Designs are used when the research question is asking whether there is a statistical
association between two quantitative dependent variables. Correlational designs do not include
independent variables, only dependent variables. Because no independent variables is being
manipulated, you cannot make cause-and-effect relationships. All that you are doing in a correlational
design is taking the scores from two or more dependent variables and seeing if they are statistically
associated. For example: ‘Is there a statistical association between SAT scores and freshman-year
GPA?’ is a research question that could be answered with a correlational design. You could randomly
select a sample of freshmen students who have taken the SAT. You would obtain their SAT scores and
their GPA at end of freshmen year. Through one of several statistical procedures, you could see if the
SAT scores are related to the GPAs; that is, did students who scores higher on the SATs tend receive
higher GPAs?
Experimental designs are used when the research question is whether there is a causal-link between
two variables and you want to examine the direct influence of an independent variable on a dependent
variable. Specifically, if you want to say that changes in one variable are causing changes in another
variable, you are going to use an experimental design. In an experimental design, a researcher directly
-27-
Back to TOC
manipulates one or more independent variables and measures at least one dependent variable. The aim
is to see if the manipulation of the independent variable is associated with changes in the dependent
variable. If so, then the researcher may be able to conclude that the changes to the independent variable
caused the changes to the dependent variable. An independent variable can be manipulated in one of
two ways:
Between-subjects where each subject is randomly assigned to one group and is tested under only one
level of the independent variable. For example, a researcher may be interested in whether a new
antidepressant drug is effective in alleviating depression. The researcher could randomly assign subjects
to take either the drug, or an inert placebo; thus, the researcher has manipulated whether a subject takes
the drug or a placebo (independent variable). The researcher would measure depression and compare
depression levels between these two groups.
Within-subjects where the same group of subjects is tested under each level of the independent
variable; hence, each subject is tested several times. For example, a researcher may be interested in
whether background noise affects test taking. The researcher could randomly select a sample of students
and have them take some standardized test in a noisy condition, and then test these same subjects in a
quite condition. Thus, noise level has been manipulated (independent variable) and each subject is
expose to each level of this independent variable. The researcher would then compare test scores
between conditions.
Quasi experimental designs are used when the research question is whether there is a difference in
performance for some dependent variable between two or more naturally occurring or preexisting groups
of subjects. Thus, these groups differ along some independent variable, but the independent variable is
not being manipulated; the subjects already fall into a specific group. For example, sex (male versus
female) is an independent variable that cannot be manipulated, because a subject naturally falls into one
of the two levels. Because the independent variable is not being manipulated, a researcher cannot
establish cause-effect relationships; rather, the researcher can only establish group differences.
For example, a researcher may want to know whether males differ in their spatial reasoning skills from
females. The researcher would randomly select a group of males and a group of females, and then test
each male subject’s spatial reasoning skills and test each female subject’s spatial reasoning skills. The
performance between these two groups would then be compared.
1.16 Computers and Statistics

There are a variety of computer programs that allows you perform statistical analyses and procedures on
a set of data (e.g., SPSS/PASW, Minitab, SAS, R, MATLAB, and even Microsoft Excel). Indeed, all of the
procedures that we will cover in this course can be performed using any one of these software, albeit
much more quickly and efficiently than by hand. So why the heck are you going to be learning to calculate
various statistical parameters and perform various statistical procedures by hand? Why not just do it all by
computer?
The simple explanation is that you need to first form a foundational knowledge of what the various
parameters and terms that we will encounter mean, what they represent, what they measure, and there
what come from. I can easily show you how to perform something called a ‘between-subjects factorial
analysis of variance’ using several of the software programs mentioned above; however, if you do not
know anything about this procedure, and what it means to ‘analyze variance’, then you will have no idea
how to interpret the results of such an analysis done by way of one of these programs.
I will be teaching you the nuts and bolts of various procedures by having you perform analyses by hand
and by calculating parameters by hand. This will allow you to see where things like ‘variance’ come from,
what the various measures of ‘variance’ reflect, and when each is to be used. For certain topics, we’ll
then go over how to have a software program calculate these parameters for us. Thus, you must
understand the conceptual framework first, before you can hope to interpret computer-generated output.
-28-
Back to TOC
CH 1 Homework Questions
1. Identify each of the following as a variable or a constant. Explain the reasons for your choices.
a. The number of days in a week.

b. People’s attitudes toward abortion.
c. The country of birth of presidents of the United States.
d. The value of a number divided by itself.
e. The total number of runs scored in baseball games.
f. The number of days in each of the twelve months in a year.
2. Identify each of the following as a variable or a constant. Explain the reasons for your choices.
a. An individual's attitude toward abortion at a specific point in time.

b. The number of days in February.
c. Peoples' opinion of the death penalty.
d. Grade Point Average (GPA).
e. The number of hairs on someone's head.
f. The time it takes to complete a homework assignment.
g. A student's semester GPA at the end of the semester.
h. The number 12
3. Identify each of the following as a qualitative or a quantitative variable:
a. age
b. religion
c. yearly income
d. weight
e. gender
f. eye color
g. college major
h. political party
i. temperature
4. Identify each of the following as a qualitative or a quantitative variable.
a. a person's name
b. goals scored by a hockey team
c. length of a rope
d. shoe color
e. movie titles
f. duration of a movie
g. number of licks to get to the center of a Tootise Roll pop
h. University attended
i. brain activity as measured via EEG
j. camera price
5. Define each of the following, in your own words.
a. Population
b. Sample
c. Null hypothesis
d. Alternate hypothesis
-29-
Back to TOC
6. Define each of the following, in your own words.
a. Independent variable
b. Dependent variable
c. Extraneous variable
d. Confounding variable
e. Control variable
7. How is a sample used to make conclusions about the population?
8. What is the main difference between an experimental design and a correlational design?
9. Answer each of the following, in your own words.
a. What are descriptive statistics used for?

b. What are inferential statistics used for?
10. Answer each of the following, in your own words.
a. What is the difference between a statistic and a parameter?

b. What types of symbols are, typically, used for statistics and parameters?
11. Distinguish between a dependent variable and an independent variable.
12. Distinguish between a between-subjects independent variable and a within-subjects independent

variable.
For Exercises 13 – 16, identify the independent variable and the dependent variable in the scenario, and
indicate whether each variable is quantitative or qualitative.
13. An antidepressant drug has been shown to have positive results on alleviating depression. To
examine the drug’s effectiveness on reducing depression, Dr. Leary administers different dosages of the
drug (0-mg, 10-mg, 20-mg, 30-mg, 40-mg, and 50-mg) to six different groups, and then measures the
amount of brain activity in each individual; where brain activity is a measure of depression.
Independent variable:
Dependent variable:
14. In a classic study on aggression, Eron (1963) studied the relationship between the amount of violent
television shows watched by young children and the amount of aggressive behavior they showed toward
peers. Eron questioned the parents of over 800 children about their child’s television viewing habits; and
Eron created a four-point scale to measure preference for violent television. Aggressive behavior was
measured by collecting ratings of each child by two or more children who knew the child in the study.
Ratings ranged from 0 - 32, with higher scores indicating more aggressive behavior.
Independent Variable:
Dependent Variable:
15. Rosenthal and Fode (1963) examined the effect of experimenter bias on the ability of rats to learn to
navigate a maze. Subjects were told that they were going to teach a rat to learn to navigate a maze. One
group of subjects was told that their rats were genetically smart, “maze bright” rats that should show
learning during the first day and performance should increase. A second group of subjects was told their
rats were genetically dumb, “maze-dull” rats that should show very little evidence of maze learning.
Subjects trained their rats to enter one of two sections of a maze, by rewarding the rat when they entered
-30-
Back to TOC
the correct section. Subjects measured the number of times the rat entered the correct section of the
maze each day.
Dependent Variable:
16. Shepard and Metzler (1971) studied the ability of humans to ‘mentally rotate’ the
image of an object. Subjects were shown two pictures, like those to the right, and
then had to decide whether the two objects were the same shape, or were mirror
images of the same shape. The angle of rotation separating the two objects varied
from 0° to 180° in 15° increments. Subjects were assessed on how quickly they
could correctly respond whether the objects were the same, or were mirror images.
Dependent Variable:
17. What is the main purpose of random selection?
18. What is the main purpose of random assignment? When is it used?
19. Give two examples of a population and a sample of that population. For each of your examples, be
sure that the sample is related to the population.
20. You have a population that includes 1000 individuals and each member of this population is assigned
a number from 1 to 1000. Using the website http://www.randomizer.org/, select a random sample of n =
20 individuals and write the numbers on your answer sheet.
21. Repeat the random selection exercise from #20 and write the numbers of the selected individuals on
your answer sheet. How many of the same individuals were selected to participate in both samples?
22. If the population in #20 and #21 included only 100 individuals, would you be more or less likely to
randomly select the same person twice?
23. Use the following scenario to answer a – f below: Dr. Evil has given up evil to start a potato chip
company that will distribute potato chips to undergraduates in the United States. Dr. Evil wants to know
what potato chip flavor undergraduates like best. He asks 100 undergraduates to taste each of four
potato chip flavors (1) BBQ, (2) Plain, (3) Ranch, and (4) Super-Evil. Each student gives a rating on a
scale of 1 to 10 on how much they like each flavor, with 10 indicating they really like the flavor. The
average ratings for each of the four chip flavors are 4.5 for BBQ, 6.7 for Plain, 2.5 for Ranch and 9.5 for
Super-Evil.
a. The population in this scenario is...

b. The sample in this scenario is...
c. “Fred” rates the Super-Evil Flavored chips a 10. This rating is a...
d. The average ratings in the scenario is(are) a...
24. Use the following scenario to answer a – e below: Dr. Logan is interested in the relationship between
playing violent video games and aggression in adolescent males. He randomly selects 1000 adolescent
males from around New York city and administers a survey for the types of video games that each
adolescent male likes to play. He then obtains measures of each adolescent male's aggression by asking
each adolescent male's teacher to provide a "toughness rating" on a scale from 1 to 10. The average
toughness rating for these adolescent males is 4.56.
a. The population is...

b. The sample is the...
-31-
Back to TOC
c. Say that "Timmy," one of the adolescent males, scores a 7 on the toughness rating. This value of
7 is a...
d. The average toughness rating is a...
e. Is the types of video games that each adolescent male likes to play a qualitative or quantitative
variable?
25. Use the following scenario to answer a – g below: Dr. Smith wants to determine whether playing
music during class causes statistics students to pay more attention. He has three stats classes with 40
students/class. For one semester he teaches his morning class while playing Snoop Dogg, in his
afternoon class he plays Dio and in his night class he plays no music (a control). During each class
meeting, Dr. Smith counts the number of students that fall asleep. He finds an average of 6 students per
meeting fall asleep in his morning class, 2 students per meeting fall asleep in his afternoon class and 5
students per meeting fall asleep in the night class.
a. The population in this scenario is...

b. The sample in this scenario is...
c. In one of Dr. Smith’s night classes, he counts 14 students that fall asleep. This number is...
d. The average of 6 students that fall asleep in the morning class is...
e. The use of different music in each class is a...
f. Counting the number of students that fall asleep uses what kind of measurement scale?
g. Before Dr. Smith can draw any conclusions about the effect of music on statistics student’s
paying attention in class, Dr. Smith will need to use...
26. The town of Petoria (population = 75000) is holding a local election for mayor between candidates Bill
Hicks and Peter Swanson. The local newspaper conducts a survey by asking readers to indicate their
preference for Bill Hicks or Peter Swanson, by sending in a ballot provided in the paper with their
preference indicated. One week later the newspaper reported it received 2500 ballots, of which 2000
favored Bill Hicks. The next day, newspaper printed a story claiming that Bill Hicks would win a landslide
in the upcoming election. Based on this information, is the newspaper’s sample representative of Petoria?
Why or why not? What implication does this have for the newspaper’s prediction about the election?
-32-
Back to TOC
Chapter 2: Math Review
2.1 Why Review Math? Didn’t you
say Statistics was a ‘Language’ Class?
I stand by what I said that statistics is a language
course more than it is a math course. But like any
language that uses symbols (letters, phonemes,
morphemes, and words) and rules for combining
those symbols (syntax) to form meaningful phrases,
so too does statistics. Unlike the natural languages
that use letters as symbols and syntax as the rules
to create meaningful phrases; statistics uses
numbers as symbols and mathematical operators
as rules to create meaning from raw data. So, just
like grammar lessons that you sat through in grade
school to refresh your mind about how to properly
use the English/American language, it’s appropriate
to have a little math refresher here.
I should take a moment to mention something that I

have noticed from teaching statistics. For several
semesters I gave a math skills assessment test on
the first day of class (I may or may not have done
this for the current semester). The test included
questions on algebra, order of operations, and use of mathematical operators. The purpose of the test
was for me and for students to assess mathematical knowledge. The test did not count toward the
students’ grades, but rather, was a tool that I used to show what mathematical skills were essential for
success in statistics and what skills they needed to improve.
What I found was a strong positive correlation between scores on the assessment test and final course
grades. Specifically, students with higher grades on the assessment test tended to be those students who
had higher course grades; a correlation that did not vary based on major, sex, semester, or previous
mathematical courses. Hence, it is imperative that if you feel like you have deficient mathematical skills to
get help from me, CTLE, or a tutor. Remember, statistics is more of a language course, but the language
is based on fundamental mathematics. Math is your friend!
2.2 Whole, Rational, Irrational, and Real Numbers

Types of values you encounter in data take many different forms. Whole numbers, or integers, whole
integers, and counting numbers, are any number not in decimal or fractional form. For example, 0, 1, 2,
3, and 4 are whole numbers. Real numbers are any values that can be created and thought up. Real
numbers include, but are not limited to, positive values, negative values, whole numbers, rational
th
numbers, irrational numbers, n roots (e.g., square roots, cubic roots, etc) and π. Rational numbers are
real numbers that can be expressed in ratio (fractional) form. For example, 2.5 can be expressed as 5/2,
or 3 can be expressed as 9/3. Irrational numbers are real numbers that cannot be expressed in ratio
(fractional) form, because its exact value cannot be determined. For example, π cannot be defined as a
ratio, because its exact value is unknown.
In statistics, most of the data encountered takes the form of rational numbers. Some values are whole
numbers, some are not. Raw data often starts out as whole numbers, but a statistic based on that data
may make the statistic look like a rational real number. For example, the number of students in your
statistics class, which is probably about 30 students, is always going a whole number, because you
cannot have fractions of a person. However, the average number of students in all of my statistics classes
-33-
Back to TOC
is approximately 27.5 students. This value is perfectly fine: It simply means that the average number of
students in my statistics classes is between 27 and 28 students. Just keep in mind that you should always
use the statistical values that you calculate; do not arbitrarily round up or down.
2.3 Basic Mathematical Operators

An operator is a symbol that tells you to perform some task or operation on a set of information. In
mathematics, there are several operators that are used in statistical functions. The + symbol is the
addition operator, which tells you to combine the values on each side of the symbol. For example: 1 + 1
= 2 and 1 + 2 = 3. Adding the same values always results in the same outcome regardless of the order in
which the values are added. For example:
9 + 6 + 5 = 20 is the same as 5 + 6 + 9 = 20 is the same as 9 + 5 + 6 = 20
Remember to take the sign (positive or negative) of each value into account (this is an area that lots of
students get into trouble with). For example:
-1 + -2 + -3 = -6
Hence, adding a negative value is the same as subtracting a positive value:
4 + -9 = 4 – 9 = -5
The – symbol is the subtraction operator and tells you to take remove the value on the right of the
operator from the value on the left of the operator. For example 2 – 1 = 1 and 1 – 1 = 0. Remember,
unlike addition the outcome of subtracting numbers depends on which number is being subtracted. Thus,
rearrangement of the values changes the outcome:
9–6=3 but 6 – 9 = -3
Also remember that subtracting a negative is the same as adding a positive:
4 - -9 = 4 + 9 = 13
The x symbol is the multiplication operator and tells you to increase the value on the left of the operator
by that many times the value on the right of the operator. For example 2 x 2 = 4 and 2 x 3 = 6; hence, 2 x
2 is telling you to add two 2s together and 2 x 3 is telling you to add three 2s together. Like addition,
changing the order of the values will not change the outcome in multiplication:
3 x 4 = 12 is the same as 4 x 3 = 12
Several symbols can be used for the multiplication operator. The other symbols are * and ()(). For
example:
5 x 3 = 15 is the same as 5 * 3 = 15 is the same as (5)(3) = 15
You will come across cases where two variables are denoted by letters and written side by side (e.g.,
XY). This form indicates that the two values should be multiplied. For example, if we let X = 5 and Y = 4:
XY = (5)(4) = 5 * 4 = 5 x 4 = 20
It is rare that x and * symbols are used; most formulas use ()() or simply have two variables listed side by
side, as in XY.
-34-
Back to TOC
The / symbol is the division operator and tells you to decrease the value on the left of the operator by
that many times the value on the right of the operator. For example 4/2 = 2 and 6/2 = 3. Like subtraction,
changing the order of values changes the outcome:
10/2 = 5 but 2/10 = 1/5 or .20
There are also several symbols that can be used for division. The other symbols are ÷ and —, with the
latter including a value above and below the horizontal line. For example:
9
9/3 = 3 is the same as 9÷3=3 is the same as 3
3
Whenever 0 is in the numerator (value being divided), if the value in the denominator (value doing the
dividing) is anything but 0, the result is 0. For example 0/2 = 0. But, whenever 0 is in the denominator, the
result is undefined. For example, 2/0 = undefined.
It is important to remember that for both multiplication and division you must take the sign of each value
into account. In multiplication, if you have an odd number of negative values your outcome will be
negative. For example (2)(-3) = -6 and (2)(-3)(2) = -12. But, if you have an even number of negative
values your outcome will be positive. For example (-2)(-3) = 6 and (-2)(-3)(2) = 12.
Because there are only ever two values in a division operation (numerator and denominator) if one is
negative the outcome is negative, but if both values are negative the outcome is positive. Thus, 4/-2 = -2
and -4/2 = -2, but -4/-2 = 2 Take care when dealing with the signs of values; I cannot tell you how many
times students make this seemingly simple mistake.
The √ symbol is the square root or radical operator. Taking the square root of a number (the number
under the radical sign) means finding the value that multiplied by itself will result in the value under the
radical sign. For example, √4 = 2, because 2 x 2 is equal to 4. Finding the square root by hand is
cumbersome so this is why calculators have the √ symbol. If you need to find the square root simply use
your calculator. If you cannot find the √ symbol on a calculator, you can raise a value to a power of 0.5.
0.5
For example 4 = 2.
One important thing to remember about square roots: You cannot take the square root of a negative
number. Well...you can, but that’s for another class. If you ever need to take a square root of some value,
but find that you are taking the square root of a negative value, you have done something incorrectly. Go
back and check your work! Trust me; there will be no negative values under the radical sign.
Related to roots of numbers is the exponent, which is presented as superscript to the right of a value
n n
(e.g., 2 ), where value of the exponent tells you to multiply the value to its left by itself that many times.
For example:
5
2 = 2 x 2 x 2 x 2 x 2 = 32
For most of this statistics course you only need to square values, that is, multiply a value by itself exactly
once. Remember, squaring a negative value is the same thing as multiplying two negative values, which
2
will result in a positive. For example, for -2 the resulting value is not -4, it is actually 4. This is so,
2
because -2 = (-2)(-2), which is equal to 4. This is another common mistake that statistics students make,
so be careful.
The ! symbol is the factorial operator, which tells you to multiply the value to the left of the factorial
symbol by itself and every whole integer less than it, including one, but not including zero. For example:
5! = 5 x 4 x 3 x 2 x 1 = 120
-35-
Back to TOC
If you have negative value in front of the factorial (e.g., -5! ) then each value you multiply is negative (-5! =
-5 x -4 x -3 x -2 x -1 = -120), but you won’t need to worry about this for statistics. Importantly, 1! = 1 and 0!
= 1 as well. While 1! = 1 is an easy concept to grasp, 0! = 1 is not. Here’s
(http://www.adonald.btinternet.co.uk/Factor/Zero.html) a great simple explanation for why 0! = 1. Some
calculators have a factorial operator built in, others do not. Table 7 in the Statistics Booklet (and Appendix
A) lists factorial values from 0 through 50.
2.4 Order of Operations

So now you know the mathematical operators used in statistical procedures, but what happens if there
are several operators in the same procedure? Believe me, this is very much the case in the procedures
you encounter, so learning the correct order of operators is critical.
The order of operations is the order in which you perform each individual mathematical operation. You
probably heard the phrase “Please Excuse My Dear Aunt Sally” (PEMDAS) somewhere around the fourth
grade. The first letter of each word references a different mathematical operation (E for exponent, A for
addition, etc.). The order of the words in the phrase indicates the order in which each operation should be
used in a procedure. A complete description of what to do at each step is shown in the table below.
Word in Phrase Operator What to Do?

For any values within parentheses,
or grouping symbols, perform all
Please Parentheses
the operations below before
moving on
Raise any value with an exponent
Excuse Exponent
to that power
Perform any and all multiplication
My Multiplication
operations
Perform any and all division
Dear Division
operations
Perform any and all addition
Aunt Addition
operations
Perform any and all subtraction
Sally Subtraction
operations
For example:
2 + [32(6-2) – 6/2](2) = ?
2 + [32(4) – 6/2](2) = ?
2 + [9(4) – 6/2](2) = ?
2 + [36 – 6/2](2) = ?
2 + [36 – 3](2) = ?
2 + [33](2) = ?
2 + 66 = 68
Obviously, if an operation is not required skip it and go to the next step. It may be easy in some
expressions to perform several steps at once, but be careful! Most mistakes I see students make involve
performing too many operations simultaneously.
2.5 Rounding Values

In statistics, you want your answers and results of any statistical analysis to be as accurate as possible,
so it is best to round to several decimal places. I always tell students unless you are told differently round
-36-
Back to TOC
your answers to the third decimal place (thousandths place: 0.001). Also, do not round only your final
answer; round to three places at each step! This maintains a high degree of accuracy in your final
answer. Thus, my rule for homework and for examinations is simple:
Unless noted round to three places at each step in any procedure.
How do you round a value? For the most part rounding in statistics is no different than what you have
probably done for years, but there is a critical caveat that must be followed.
Assuming you are rounding to the third decimal place: Look at the number in the fourth decimal place
(ten-thousandths place). If the value in the fourth decimal place is greater than 5, increase the value in
the third decimal place by 1, and then drop everything that follows the third decimal place. For example:
2.3478 rounds to 2.348, because the value in the fourth decimal place (8) is greater than 5.
If the value in the fourth decimal place is less than or equal to 4, leave the value in the third decimal place
alone and drop everything the follows the third decimal place. For example, 2.3473 rounds to 2.347
because the value in the fourth decimal place (3) is less than or equal to 4.
What should you do when the value in the fourth decimal place is equal to 5? You were probably told to
always round up, which is not totally correct. If you always round up on 5 you are using five possible
values as a basis for rounding up (5, 6, 7, 8, 9), but only four values for leaving a value alone (i.e. 1, 2, 3,
4). Zero, does not count, because if there is a zero in the fourth decimal place then there is nothing there
to worry about for rounding purposes. Why is always rounding up on a 5 problematic? Because you bias
your data toward larger values! In statistics, you want everything to be nice and even.
What do you do? If the number in the fourth decimal place is equal to 5, you must use the number in the
third decimal place (the number you are rounding) to determine whether you round up, or leave that value
alone. If the number in the fourth decimal place is 5 and the number in the third decimal place is even (0,
2, 4, 6, or 8), leave the number in the third decimal place alone and drop the values after that number. For
example: 2.3465 rounds to 2.346 and 2.3445 rounds to 2.344. (Remember, even through 0 reflects the
absence of a value, it is treated as an even digit.) If the number in the fourth decimal place is 5 and the
number in the third decimal place is odd (1, 3, 5, 7, or 9), increase the value in the third decimal place by
1 and drop the values after that number. For example: 2.3475 rounds to 2.348 and 2.3435 rounds to
2.344. This reduces bias in the data.
2.6 Proportions, Probabilities and Percentages

Statistics, especially inferential statistics, are based on probabilities; hence, it is good to review basic
probabilities and proportions. Although probabilities and proportions are calculated the same way, they
mean different things.
Say that you have a bag of 25 marbles (10 red, 5 blue, 5 white and 5 green). The proportion of red
marbles in the bag is 10/25 = 0.400. The proportion of green marbles in the bag is 5/25 = 0.200
Proportions describe the relative frequency of some event out of all possible events. In contrast, a
probability is used to assess the relative likelihood of observing some event in the future. For example,
the probability of selecting a blue marble is 5/25 = 0.200 and the probability of selecting a red marble is
10/25 = 0.400. In short, the distinction between probability and proportion semantics: proportions talk
about collected data and probabilities use data to make inferences about future events.
To convert a probability or proportion to a percentage, multiply the proportion/percentage by 100 and put
a % after the value. For example, the percentage of red marbles in the bag is equal to the proportion of
red marbles (0.400) multiplied by 100 (i.e., 40%). The percent chance of selecting a green marble is
equal to the probability of selecting a green marble (0.200) multiplied by 100 (i.e., 20%).
-37-
Back to TOC
2.7 Summation Notation (Σ)
Most of you have probably seen the symbol Σ, which is Greek uppercase sigma. In mathematical terms,
this symbol is the summation operator and means to sum up all of the values to the right of Σ. As often is
the case, the symbol for a variable (X) is usually presented to the right of the Σ, which means you need to
sum up every value associated with that variable. For example, ΣX means to sum up each value that
belongs to the variable X. Also, Σ is a grouping symbol, so you should do any mathematical work and
operations to the right of Σ before summing up any values.
In the table to the right, five

Professor 2005 Salary (Y) 2006 Salary (X)
professors who teach at
Whatsamatta-University are
Dr. Java 39,000 41,000
listed with their 2005 and 2006
salaries. We’ll use this table of Dr. Spock 43,500 46,000
data to perform some
operations using Σ. Dr. Evil 48,000 51,000
Dr. Griffin 52,500 56,000
Dr. Who 57,000 62,000
ΣX will be equal to all of the 2005 salaries: ΣX = 39,000 + 43,500 + 48,000 + 52,500 + 57,000 = 240,000
ΣY is equal to the sum of all 2006 salaries: ΣX = 41,000 + 46,000 + 51,000 + 56,000 + 62,000 = 256,000
Σ Dr. Evil is equal to the sum of all of Dr. Evil’s salaries: Σ Dr. Evil = 48,000 + 51,000 = 99,000
Summation notation is often written with a subscript and a superscript, such as:
K
X
k 2
The letter in the superscript (K) refers to which group you should be working with, because occasionally
there are several groups denoted by different letters (K, J, M), but a dependent variable (X) is common to
each group. Thus, we could have called the 2006 salaries K and the 2005 salaries J. In this case, if we
have a K in the superscript, it tells us to use group K, which is the 2006 salary data. The subscript refers
to which value within that data set that you should start adding with. For example, the subscript k = 2 says
that you should start adding with whichever professor is associated with case number 2, which in this
case would be Dr. Spock. Thus:
K
 X  46000  51000  56000  62000  215000
k 2
You don’t need to worry too much about subscripts and superscripts now, just note that an uppercase
letter (e.g., X) usually indicates a variable name, or some group name and the lowercase version of that
same letter is used to indicate a specific instance (case) of that variable.
One thing you need to worry about is what to do if you have a set of operations to the right of Σ. As I
mentioned, Σ is a grouping symbol, just like a set of parentheses, and you must complete everything to
the right of Σ before summing values. For example, say that we have the following set of data for
-38-
Back to TOC
variables X and Y. The X and Y values may be a set of scores from two
different variables collected over six people. X Y
10 5
2 9 5
We want to find ΣX . What would you do? Because the summation
symbol acts like a grouping symbol you have to do the work to the right 8 4
2 6 4
of the summation symbol first and then add the values. Thus, for ΣX you
2 2 4 3
find the X values before adding anything, because ΣX is telling you to
2 3 2
add up all of the X values. To do this you take the square of each value
associated with X and then add those squared values. Thus:
2 2 2 2 2 2 2
∑X = 10 + 9 + 8 + 6 + 4 + 3 = 100 + 81 + 64 + 36 + 16 + 9 = 306
2
I find that creating a new column in the data set for the new values first is easiest. That is, to find ΣX ,
2 2 2
create an X column, then find each X value, and finally add the X values.
Here are some other examples:

2 2 2
(ΣX) = (10 + 9 + 8 + 6 + 4 + 3) = (40) = 1600
Σ(X - Y) = (10 - 5) + (9 - 5) + (8 - 4) + (6 - 4) + (4 - 3) + (3 - 2) = 5 + 4 + 4 + 2 + 1 + 1 = 17
2 2 2 2 2 2 2 2 2 2 2 2 2
Σ(X - Y) = (10 - 5) + (9 - 5) + (8 - 4) + (6 - 4) + (4 - 3) + (3 - 2) = 5 + 4 + 4 + 2 + 1 + 1 ...
... 25 + 16 + 16 + 4 + 1 + 1 = 63
ΣXY = (10*5) + (9*5) + (8*4) + (4*3) + (3*2) = 50 + 45 + 32 + 24 + 12 + 6 = 169
2.8 Scales of Measurement

Data takes many forms, but all data are measurable on one of four scales. Each scale is associated with
a different type of data and each scale has different characteristics. I start with the least complex scale
and move to the complex scales. Scales of increasing complexity have the qualities of the scales below,
plus something else.
Data measured on a nominal scale are qualitative, that is, categorical. The levels, groups, or events on a
nominal scale have no numerical values or quantities associated with them and there is no specific order
to the entries on a nominal scale. That is, no entry is quantitatively better or worse than any other entry.
For example, the table below lists the three divisions from the eastern conference of the National Hockey
League (NHL) and the five teams within each division. If I measure a team’s division, this is a variable
that is on a nominal scale.
Northeast Division Atlantic Division Southeast Division

Ottawa Senators New Jersey Devils Carolina Hurricanes
Montreal Canadians New York Rangers Atlanta Thrashers
Buffalo Sabers New York Islanders Florida Panthers
Toronto Maple Leafs Philadelphia Flyers Tampa Bay Lightening
Boston Bruins Pittsburgh Penguins Washington Capitals
The ordinal scale, or rank-order scale, includes entries that have structure. The order of entries is listed
in a way that you can determine whether each entry is better/worse or greater/less than other entries. An
ordinal scale is like a nominal scale because the entries are categorical, but an ordinal scale also has
some order to the entries. The order of entries is based on some relative measure, but the measure itself
is meaningless, because there is no known value associated with each entry. That is, the value of each
entry and the difference in value between any two entries is unknown. The only thing known is whether
something is bigger/smaller or better/worse in relation to others entries.
-39-
Back to TOC
For example, in the table below lists the same data as in the nominal scale table above; however, I listed
the teams within each division in the order in which they finished the 2010-2011 season:
Place Northeast Division Place Atlantic Division Place Southeast Division

1st Boston Bruins 1st Philadelphia Flyers 1st Washington Capitals
2nd Montreal Canadians 2nd Pittsburgh Penguins 2nd Tampa Bay Lightening
3rd Buffalo Sabers 3rd New York Rangers 3rd Carolina Hurricanes
4th Toronto Maple Leaves 4th New Jersey Devils 4th Atlanta Thrashers
5th Ottawa Senators 5th New York Islanders 5th Florida Panthers
You can see the order to the teams within each division based on their place of finish. However, the
ordinal scale does not tell us anything about magnitude of each entry; hence, we cannot know by how
much better or how much worse any one team did than other teams. An ordinal scale provides
information about how events (teams) performed/placed relative to the other events (teams).
An interval scale is similar to an ordinal scale because the data are ordered, but unlike an ordinal scale
an interval scale has a numerical values rather than simply a ranked-order. The interval between adjacent
values is equal. For example, the difference between the values 2 and 3 is 1 on an interval scale and the
difference between the values of 4 and 5 is also 1. With an ordinal scale the difference is unknown.
The table below lists the five teams from the northeast
division by what is known as the plus/minus rating, which Northeast Division Team Plus/Minus Rating
has to do with goals scored. The plus/minus rating is on Boston Bruins 58
an interval scale, that is, varies in increments of 1. This is Buffalo Sabers 9
so even though there are large differences between Montreal Canadians 1
adjacent teams. In reality, the actual values of the Toronto Maple Leaves -22
plus/minus rating occur in equal increments of 1 so the Ottawa Senators -52
difference between Buffalo’s plus/minus rating of 9 and a plus/minus rating of 10 is equal to the difference
between the plus/minus rating of -22 for Toronto and a plus/minus rating of -23.
An interval scale does not have a true zero point, that is, a value at which the measured variable has no
magnitude. A true value of zero indicates that there is nothing of that variable present to be measured. On
an interval scale a value of 0 can be obtained, but the 0 is actually meaningless a value of 0 means that
what you are measuring is present. Hence, there can be negative values on an interval scale and those
negative values make sense. This indicates that a value of 0 on an interval scale still has magnitude,
because it is greater than any negative value. Also, because there is no true zero point, you cannot form
ratios and cannot tell how many times greater-than/less-than one value is from another. In order to form
ratios you need to have a true zero point, because there can be no such thing as ‘negative ratios’.
The ratio scale is identical to an interval scale except the variable has a true zero point, because
negative values are not possible. With a true zero point a value of 0 indicates the absence of any
magnitude. I should note that the true zero point does not have to show up in the data and is unlikely to
ever show up. For example, any measurement of time (days, seconds, hours, milliseconds) is on a ratio
scale, because you cannot have negative time.
Staying with the hockey example, in the table to the right I listed the five
Northeast Division Team Wins
teams from the northeast division in the order of the number of wins
Boston Bruins 46
that each team had at the end of the 2010-2011 season. The variable
Montreal Canadians 44
wins is measured on a ratio scale, because you can have zero wins,
but you cannot have a negative number wins. Buffalo Sabers 43
Toronto Maple Leaves 37
Ottawa Senators 32
What scale of measurement of psychological constructs fall on? Although the actual measures of
psychological constructs may be on ratio scales (e.g., reaction times are ratio, accuracies are ratio); most
constructs are actually measured on ordinal scales and nominal scales. For example, take an IQ tests as
-40-
Back to TOC
a measure of intelligence. Most IQ tests are standardized to have an average score of about 100 with a
theoretical range of 0 - 200-ish. This appears to be a ratio scale, because you certainly cannot have
‘negative intelligence’. But, take two IQ scores, 50 and 100. Is the individual with the score of 100 twice
(2x) more intelligent than the individual with the score of 50? Absolutely not! All that can be said is that
the individual score 50 points higher on the IQ test. Hence, IQ scores would seem to be on an interval
scale. Also, recall from Chapter 1, a Likert scale (see below) appears to be on an interval scale because
of the numerical values, but is actually on an ordinal scale, because people are choosing between
different categories placed relative to each other. Thus, there can be a difference between the scale in
which a variable can be interpreted, and the actual scale by which a variable is measured.
-5 -4 -3 -2 -1 0 +1 +2 +3 +4 +5
2.9 Continuous vs. Discrete Data

For quantitative data that has been measured on an interval or a ratio scale, the data can be continuous
or discrete. With continuous data there are theoretically an infinite number of possible values between
any two measurable points. That is, data are assumed to continue infinitely between any two points. With
continuous data a measurement that contains a decimal or is a fraction of a whole numbers makes
sense. For example, between 1 second and 2 seconds there can be 1.02 seconds, 1.156 seconds,
1.5432678 seconds, etc. Each of those values makes sense because each value can exit.
In contrast, with discrete data, there are a finite, or fixed, number of values that can exist between any
two measurable points. With discrete data decimals and fractions of measurable numbers do not make
sense. For example, one row of desks in a classroom may contain 2 students and a second row may
contain 4 students. The only possible measurement for people between 2 students and 4 students is 3
students, because you cannot have a measurement of say 2.5 people or 3.75 people. Fractions of people
do not exist. This is not to say that you cannot have an average of 4.5 people sitting in each row of a
classroom! With discrete data, individual measurements cannot be in fraction or decimal form, but the
average is a different story. Indeed, an average of 4.5 people sitting in each row simply indicates that
there are, on average, 4 or 5 people sitting in each row.
2.10 Real vs. Apparent Limits of Continuous Data

With continuous data any measurement is only as good as the device making the measurement and as
good as the person who is taking the measurement. Thus, when we say that someone’s reaction time to
push a button was 1.2 seconds, we do not mean exactly 1.2 seconds; rather, we mean approximately 1.2
seconds. Similarly, when I say that it took me 5 years to complete graduate school to receive my Ph.D., I
do not mean 5 years exactly; I mean 5 years approximately.
If you are rounding numbers, you are taking away some of the accuracy of the measurements. For
example, if you are rounding to the tenths place and you measure someone’s actual reaction time to be
1.56, you would round that to 1.6. Thus, you have introduced inaccuracy. Thus, when we say that we
have measured a person’s reaction time to be 1.6 seconds, we may actually mean that the reaction time
was 1.56 seconds or 1.61 seconds, or even 1.599999 seconds.
To account for this approximation, we place real limits around values and consider any value that lies
within these real limits to be associated with the listed value. Thus, real limits are like a boundary around
a reported value that includes similar values that we equate with that reported value. Any value within that
boundary will be listed as the number around which the boundary is based.
Real limits of any number are defined as the values that fall one-half unit of measure above that number
and one-half unit below that number. For example, the real limits of 19 are 18.5 and 19.5. The smaller
-41-
Back to TOC
value is the lower real limit (LRL) and the larger value is the upper real limit (URL). Similarly, the real
limits around 1.2 are 1.25 and 1.15. To find the real limits of any number, follow these steps:
Examples
Step What to Do 9 9.0 9.00
(1) Determine the lowest unit of measure Ones Tenths Hundredths
Take one unit of that lowest unit of

(2) 1 0.1 0.01
measure
(3) Divide that one unit in half 1/2 = 0.5 0.1/2 = 0.05 0.001/2 = 0.005
Add the value in step 3 to the original URL = 9 + 0.5 URL = 9.0 + 0.05 URL = 9.00 + 0.005
(4)
number to get the upper real limit URL = 9.5 URL = 9.05 URL = 9.005
Subtract the value in step 3 from the LRL = 9 - 0.5 LRL = 9.0 - 0.05 LRL = 9.00 - 0.005
(5)
original number to get the lower real limit LRL = 8.5 LRL = 8.95 LRL = 8.995
-42-
Back to TOC
1. Use the following data from eight individuals who were measured on variables X and Y, to calculate the
requested sums, below.
i X Y
A 2 4
B 2 1
C 5 3
D 6 7
E 1 2
F 3 4
G 1 2
H 1 7
2 2
a. ΣX b. Σ(X - Y) c. ΣX d. Σ(X)
2 3
e. (ΣX) f. Σ(Y - X) g. ΣXY h. ΣX - ΣY
2. Use the following data from eight individuals who were measured on variables X and Y, to calculate the
requested sums, below.
i X Y
A 5 3
B 2 9
C 3 8
D 5 7
E 7 6
F 8 5
G 8 4
H 2 3
2
a. ΣX b. ΣY c. ΣXY d. ΣX
2 2
e. (ΣX) f. (ΣXY) g. (ΣX)(ΣY) h. Σ(X - 3)
2
i. Σ(X - 2)(Y - 3) j. Σ(Y - 3)
3. Round each number to three decimal places using the procedures discussed in class and in the book.
a. 1.2345 b. 2.4236 c. 17.16742

d. 311.4205 e. 3.6795 f. 12.3456
g. 66.7295 h. 54.147356
4. Round each of the following values to three (3) places:
a. 23.6578 b. 420.3109 c. 0.6715687

d. 1.596578 e. 999.099501 f. 1.009356
5. Determine the upper real limit and the lower real limit of each following number.
a. 12.6 b. 22.58 c. 10.0 d. 49
6. Determine the upper real limit and the lower real limit of each following number.
a. 9 b. 10.15 c. 0.10009 d. 1.236

7. Use the following data for four individuals to perform the requested calculations and summations. Keep
all of the digits shown, and do not round any of the intermediate calculations, until rounding the final
answers to two decimal places. Next, round each original score to two decimal places and repeat the
-43-
Back to TOC
calculations, rounding all intermediate calculations to two decimal places. Compare the two sets of
results. What accounts for the difference between them?
Individual X
1 2.8753
2 2.2252
3 3.1158
4 3.4016
2 2
a. ΣX b. (ΣX) c. ΣX
8. Consider the following data for five individuals:
Individual X
1 3.8753
2 4.2660
3 4.1156
4 3.4954
5 4.2061
Perform the following calculations on the given scores, keeping the digits shown until rounding the final
answer to two decimal places. Then repeat the calculations, rounding the scores and all intermediate
calculations to two decimal places. Compare the two sets of results. What accounts for the difference
between them?
2 2
9. Identify whether each measure is a nominal measure, ordinal measure, interval measure, or ratio
measure. Explain the reasons for your choices.
a. A cognitive scientist measures the time to solve a puzzle in seconds.

b. A historian groups the books on his shelf based on category.
c. A professor lists the publications on his curriculum vita in alphabetical order based on last names
of the authors.
d. Students are asked to rate their political attitude on a scale from -5 (liberal) to 5 (conservative).
e. The morning weather usually reports the temperature in degrees Fahrenheit.
f. A professor counts the number of students that come to his office each week.
10. Indicate whether each measure is a nominal measure, ordinal measure, interval measure, or ratio
a. inches on a yardstick b. Social Security numbers

c. dollars as a measure of income d. order of finish in a car race
e. intelligence test scores
11. Indicate whether each measure is a nominal measure, ordinal measure, interval measure, or ratio
a. The speed of a slap-shop made by a hockey player.

b. The movie titles in your DVD collection.
c. First through third place in a pie-tasting contest.
d. Temperature measured in degrees-kelvin.
e. Temperature measured in degrees Celsius.
f. The number of baseball cards in a collection.
g. A rating given the following question on the scale below.
-44-
Back to TOC
How much do you agree with the republican party?
Strongly Disagree Somewhat Neither Agree Somewhat Agree Strongly Agree
Disagree Disagree or Disagree Agree
h. Letter grades: A, A-, B, B+.
i. Your name.
12. Express the following statements in summation notation for N = 7, using the subscripts N and i with Σ.
a. X1 + X2 + X3 + X4 + X5 + X6 + X7 b. X2 + X3 + X4 + X5 + X6
c. X 12 + X 22 + X 32 + X 24 + X 52 + X 62 + X 72 
d. X1  X 2  X 3  X 4  X 5  X 6  X 7
2 2 2 2 2 2

2 2
13. Indicate whether each of the following variables is discrete or continuous.
a. Height b. gross domestic product

c. happiness d. grains of sand in a sandbox
14. Identify each of the following as a qualitative or a quantitative variable:
a. weight b. religion c. income

d. age e. gender f. eye color
15. Consider the following data for four individuals:
Individual X
A 1.25
B 2.25
C 3.50
D 2.15
Perform the following calculations on the given scores, being sure to round to three places at each step
as needed.
2 2
16. Consider the following data for four individuals:
Individual X
A 11.265
B 10.585
C 13.255
D 12.265
Perform the following calculations on the given scores, being sure to round to three places at each step
as needed.
2 2
-45-
Chapter 3: Frequency Distributions and Graphing
3.1 Frequency Distributions for Quantitative Variables

When you collect data it will initially be unorganized. One of the basic uses of statistics is to organize raw
data. The graph above provides an example of this chapter, that is, the consolidation of raw data into a
simple, usable device; in this case a graph showing the relationship between membership in unions and
the percentage of income going to the middle class over time. The data presented in the graph are
percentages (on the y-axis), which are based on frequencies, or counts, of individuals falling into certain
categories. This chapter is concerned with taking unorganized raw pieces of information and organizing
them into something meaningful.
A frequency distribution is an organized set of data that takes the form of a table or a graph. In tabled
form each value or entry in is listed along with the number of times that value, or entry, appeared in the
data. As a graph, each value is presented as a function of the number of times that value occurred in the
data. Frequency distributions simply consolidate a data set into something smaller and manageable.
Generally, frequency distribution table, or simply frequency table, refers to a frequency distribution in
tabled form; whereas frequency distribution refers to one in graphed form.
For example, let’s say I ask n = 50 students the following question and have each student rate their
response to the question on a scale from 1 to 11. The question is, ‘Overall, what is your general political
attitude?’ The rating scale is below:
1 2 3 4 5 6 7 8 9 10 11
I obtain the following scores:
5 6 8 9 6 5 7 6 5 6
5 4 5 5 6 7 6 8 2 3
4 6 8 11 6 7 5 3 2 6
6 8 7 9 5 6 11 3 2 5
6 7 9 5 6 7 8 9 8 3
-46-
Back to TOC
In its present form these data are unorganized and it is impossible to draw any conclusions or make any
inferences about political attitudes among the students. Is there anything wrong with the data being
unorganized? No, the data are fine, but reading the data and trying to understand them is difficult in their
currently unorganized format. We need to organize the data.
The simplest method for organizing raw data is to create a frequency distribution table. In a frequency
table each possible value in the range of scores in the data set is listed with its frequency of occurrence.
That is, each value in the data set is listed with the number of times it was recorded, across all of the
subjects. Creating a frequency distribution is easy, but some rules must be followed to preserve clarity.
The most important thing to remember when constructing any frequency distribution or any graph is to
make the table, or graph, easy to understand.
To construct a frequency distribution table, list each value in the range of possible scores from high to low
in one column by placing the highest value at the top of this column and working your way down to the
lowest value. The name at the top of this column is usually denoted X, or the name of the variable
(“Political Attitude”).
You can skip missing values in the range, or not skip them; this is really a judgment call. For example, a
rating of 10 was not recorded in the data above, but because it is in the range between the highest
possible score (11) and the lowest possible score (2), you might want to include it in a frequency
distribution. One reason for including values that were not actually obtained, but were within the range of
values, is to not make people wonder about your data; you want a frequency distribution to be easy to
understand. In the frequency distribution table, I included the values 1 and 10 which were not recorded in
the data above.
Count the number of times each value occurred in the data; this is the value’s frequency. List the
frequency of each value in a new column labeled ‘f’ that is positioned to the right of the column with the
values (see table below). It is good to list the total frequency (nT or just n) at the bottom of this column.
Political Attitude f Except for some more information to be added in a moment, that’s
11 2 it! You’ve created a frequency distribution and organized a set of
10 0 raw data. You can see from this frequency distribution table, that
9 4 most students rated their overall political attitude a 5 (slightly
8 6 liberal) or 6 (totally neutral). You could make a supposition that the
7 6 student population, or institution, from which these students came
6 13 is not overly liberal or conservative, but is more centrist. Without
5 10 the data being organized in this manner, such an inference is much
4 2 more difficult to make.
3 4
2 3
1 0
nT = 50
There is a limitation to list only the absolute frequencies, as these don't tell you anything about the rest
of the distribution. That is, the absolute frequency does not tell you how frequent a value occurred relative
to the rest of the distribution. For example, if I told you that the frequency of people reporting a political
attitude of 9 (moderately conservative) is four people, but gave no other information, what does this say
about a value of 9 with respect to the distribution? Nothing! Knowing four students self-reported having a
moderately conservative political attitude does not tell you anything about whether this is a high frequency
or a low frequency relative to the total frequency. A frequency of 4 would be high if there were only n = 10
scores, but would be low if there were n = 10,000 scores. But if you do not know the total number of
scores in the distribution (nT), then you cannot determine whether it is a high or a low frequency and what
impact the value has on the distribution.
-47-
Back to TOC
We need a standard measurement that tells us of the impact of each value relative to the distribution,
such that higher values indicate a higher impact and lower values indicate a lesser impact. The measure
that accomplishes this is relative frequency (rf) or the proportion of each value relative to all scores in
the distribution. The relative frequency of a value is calculated by dividing the absolute frequency of a
value by the total frequency (nT) and then listing that resulting value in a column to the right of the
frequency column:
Political Attitude f rf You do not include the quotients in the relative

11 2 2/50 = .04 frequency column; only include the proportions.
10 0 0/50 = .00 What’s nice about using relative frequency is that
9 4 4/50 = .08 proportions range from 0 to 1 so they provide a
8 6 6/50 = .12 standard measure of impact that a value has
7 6 6/50 = .12 relative to the distribution. That is, how much
6 13 13/50 = .26 impact does a value have on a distribution? The
5 10 10/50 = .20 higher the relative frequency the greater impact
4 2 2/50 = .04 that value has on the distribution. A similar
3 4 4/50 = .08 measure that accomplishes the same goal is
2 3 3/50 = .06 the relative percentage (%) of a value. The
1 0 0/50 = .00 relative percentage is the relative frequency
nT = 50 multiplied by 100. Hence, it is simply the
proportion of a value converted into a percentage:
Political Attitude f rf %
11 2 .04 .04 x 100 = 4%
10 0 .00 .00 x 100 = 0%
9 4 .08 .08 x 100 = 8%
8 6 .12 .12 x 100 = 12%
7 6 .12 .12 x 100 = 12%
6 13 .26 .26 x 100 = 26%
5 10 .20 .20 x 100 = 20%
4 2 .04 .04 x 100 = 4%
3 4 .08 .08 x 100 = 8%
2 3 .06 .06 x 100 = 6%
1 0 .00 .00 x 100 = 0%
nT = 50
Frequency distributions are also useful for determining the ‘place’ or ‘rank’ of a value relative to other
values. Specifically, frequency distributions can be used to determine the number, proportion, or
percentage of scores greater than, less than, greater than or equal to, or less than or equal to a value. To
do this we must add several other pieces of information to the frequency distribution.
The cumulative frequency (cf) is the number of scores that are less than or equal to a value. Thus, it is
the frequency of a particular value added to the frequency of all values that are less than that value. For
example, take the political attitude rating of 4 (somewhat liberal). The cumulative frequency of a rating of
4 is its frequency (f = 2) added to the frequency for the values that are less than a rating of 4, which are
the ratings of 3 (f = 4), 2 (f = 3), and 1 (f = 0). This results in a cumulative frequency of 9 for the rating of
4. This indicates there are 9 scores that are less than or equal to 4. Stated differently, 9 students have a
political attitude that is ‘extremely liberal’ to ‘somewhat liberal’.
To determine the cumulative frequency for each value in a frequency distribution always start with
the smallest value and work through the larger values. First, determine the cumulative frequency of the
value 1 by adding its frequency (f = 0) to the frequency of all values that are less than 1. Because there
are no values less than 1 the cumulative frequency of 1 is cf = 0. (Note that the cumulative frequency of
the smallest value will always be equal to the absolute frequency of the smallest value.)
-48-
Back to TOC
Next, find the cumulative frequency for the next largest value (2) by adding its frequency (f = 3) to the
frequency of all values that are less than less than 2. The frequency for all of the values that are less than
2 is the cumulative frequency of the value that is immediately less than 2. Thus, by adding the absolute
frequency of 2 to the cumulative frequency of the value 1 (cf = 0), we have the cumulative frequency up to
an including 2 (cf = 3). Continue this for each increasing value until you have filled the cf column.
Importantly, the cf for your largest value should be equal to nT, because the cumulative frequency of the
largest value in a distribution will always be greater than or equal to all of the scores. If you do not find
this, go back and check your work:
Political Attitude f rf % cf
11 2 .04 4% 2 + 48 = 50
10 0 .00 0% 0 + 48 = 48
9 4 .08 8% 4 + 44 = 48
8 6 .12 12% 6 + 38 = 44
7 6 .12 12% 6 + 32 = 38
6 13 .26 26% 13 + 19 = 32
5 10 .20 20% 1 + 9 = 19
4 2 .04 4% 2+7=9
3 4 .08 8% 4+3=7
2 3 .06 6% 3+0=3
1 0 .00 0% 0+0=0
nT = 50
Like the absolute frequency, the cumulative frequency does not say anything about how much impact a
value has with respect to the distribution. That is, cf says nothing about what proportion of scores are less
than or equal to a certain value. To determine this you need the cumulative relative frequency (crf) of a
value, which is obtained by dividing the cumulative frequency by the total frequency (n T). In the table
below, each value in the crf column is the proportion of scores in the distribution that are less than or
equal to the value in the Political Attitude column. You can also convert the cumulative relative frequency
values into cumulative percentages (c%) by multiplying each crf by 100 and adding the percent sign. In
the table below, I have added the cumulative relative frequencies and the cumulative percentages. Again,
you do not have to include the mathematical operations in the crf and c% columns:
Political Attitude f rf % cf crf c%

11 2 .04 4% 50 50/50 = 1.00 1.00 x 100 = 100%
10 0 .00 0% 48 48/50 = .96 .96 x 100 = 96%
9 4 .08 8% 48 48/50 = .96 .96 x 100 = 96%
8 6 .12 12% 44 44/50 = .88 .88 x 100 = 88%
7 6 .12 12% 38 38/50 = .76 .76 x 100 = 76%
6 13 .26 26% 32 32/50 = .64 .64 x 100 = 64%
5 10 .20 20% 19 19/50 = .38 .38 x 100 = 38%
4 2 .04 4% 9 9/50 = .18 .18 x 100 = 18%
3 4 .08 8% 7 7/50 = .14 .14 x 100 = 14%
2 3 .06 6% 3 3/50 = .06 .06 x 100 = 6%
1 0 .00 0% 0 0/50 = .00 .00 x 100 = 0%
nT = 50
The table above, is a complete frequency distribution table that (1) organizes raw data; (2) provides
information about the proportion each value contributes to a distribution, and (3) provides information
about the numbers and proportions of scores less than or equal to a value. Such a distribution provides a
means for someone to assess the importance of the data and to discuss the data and possible make
some inferences.
-49-
Back to TOC
Now that this frequency distribution table has been established for the political attitude data from earlier,
we can start to answer some general questions about this data. The examples that I list below, are just
several of questions that could be answered from a frequency distribution table.
Question: How many students rated their political attitude as a 5 (slightly liberal) or less?
Answer: 19--This comes for the cumulative frequency of 5.
Question: How many students rather their political attitude as being less than a 5 (slightly liberal)?
Answer: 9--This comes from the cumulative frequency of 4, which is the immediately less than 5.
Question: What proportion of students rated their political attitude as a 9 (moderately

conservative)?
Answer: .08--This comes from the relative frequency of 9.
Question: What proportion of students rated their political attitude as an 8 (somewhat

conservative) or less?
Answer: .88--This comes from the cumulative relative frequency of 8.
Question: How many students rated their political attitude as greater than 7 (slightly
conservative)?
Answer: 12--This can be found by adding the absolute frequencies of the values greater than 7
(i.e., 8, 9, 10, and 11); or, by subtracting the cumulative frequency of 7 (crf = 38) from the total
frequency (nT = 50).
Question: How many students rated their political attitude as greater than or equal to 7 (slightly
conservative)?
Answer: 18--This can be found by adding the absolute frequencies of the values that are greater
than or equal to 7 (i.e., 7, 8, 9, 10, and 11); or, by subtracting the cumulative frequency of 6 (crf =
32) from the total frequency (nT = 50).
3.2 Frequency Distributions for Qualitative Variables

The frequency distribution table created above was for data that came from a quantitative variable, that is,
the numerical ratings the 50 students gave for their political attitudes. It may be the case that the data
from a measured variable is qualitative, that is, from a nominal scale. For example, a professor may want
to know want to know numbers of the college majors within a particular class, or perhaps the numbers of
freshmen, sophomores, juniors, and seniors within a class. The data on which these frequency
distributions would be based comes from nominal scales, because you are measuring people based on
categories in which they fall along some qualitative variable.
For example, say that I want to know the numbers of each college major that make up a particular
psychology class with 30 students. I simply ask each student for his/her primary major and list each major
with the number of students in that major:
College Class f Each entry (value) under college class is qualitative and the data
Psychology 15 come from a nominal scale. When creating a frequency distribution
Neuroscience 5 for qualitative data most steps are identical to those when creating a
English 2 frequency distribution with quantitative data. The only difference
Biology 4 between frequency distributions for quantitative data and qualitative
History 4 data is you do not include cumulative frequency, cumulative relative
nT = 30 frequency, and cumulative percentages for qualitative data. These
three concepts are useful only when discussing quantitative data.
-50-
Back to TOC
Frequency distributions for qualitative data can include the relative frequencies and the relative
percentages for each entry. I have listed these in the table below:
College Class f rf %
Psychology 15 .500 50%
Neuroscience 5 .167 16.7%
English 2 .067 .6.7%
Biology 4 .133 13.3%
History 4 .133 13.3%
nT = 30
3.3 Outliers in Frequency Distributions

One of the many uses of frequency distributions is the identification of outliers in a data set. An outlier is
any measurement that has a value that is very different from the other measured value and has a
frequency much lower than all other values. For example, say that I measure the age, in years, of each
subject in a research study. The frequency distribution is below, to the right:
As you can see in the frequency distribution, most of the 80 subjects Age f
ranged in age from 18 to 21 years old, which you would expect if 45 1
this study was conducted on a college campus as is most 21 15
psychological research. However, there is one individual who is 45 20 20
years old; an age that is over twice as old as the next oldest 19 25
subjects. This individual is an outlier based on their age. 18 19
Outliers can be problematic in statistics as their characteristics and performance in tasks might not
represent the majority of the sample; hence, outliers are often eliminated during data analysis.
Outliers can also be identified when data come from a qualitative variable, such as college major. For
example, say I conduct a survey on students’ opinions of campus events and I obtain the following
distribution of majors in my sample:
College Class f In this case, the two philosophy majors would be considered outliers
Psychology 50 and their data may be removed, as their responses may not be in
Philosophy 2 line with the majority of the sample.
English 45
Biology 90
History 45
3.4 “A Picture is worth a Thousand Words”

I’m sure you have heard this saying, but did you ever take a
minute to think about what it means? Basically, a picture
can take the place of the 1000 or so words you would need
to describe something. Not that verbally describing
something is bad, it’s just that understanding something
through seeing it is often a lot easier than having it
explained. Also, it is often easier to show someone what
you mean, than having to explain what you mean. Meaning
can often get “lost in translation.”
-51-
Back to TOC
The same is true in statistics: A picture of data is probably worth more than a thousand words when
describing data statistically, where we try to explain data through the use of graphs. As you are well
aware a graph is simply a chart of data and most of you have probably made graphs for other classes, so
some of this will be a refresher. However, there are some rules that need to be followed when creating
graphs. Indeed, as a reviewer for scientific journals, I cannot say how upsetting it is to get a beautifully
written manuscript with terrible graphs.
The only rule for graphing is to present your data in the simplest and most meaningful way possible. What
does this mean?
Don’t make over-complicated graphs

Include only what is necessary
Include enough information to make your graph easily interpretable
You do not want people thinking too much about what your graph means. You want people to be using a
graph to assist in their interpretation of your data; you do not want them trying to figure out how to read
your graph. Too much ‘stuff’ in a graph only makes it confusing, but not enough stuff will make it difficult
to interpret. For example, say that we wanted to plot a graph of the number of goals scored by each team
in the Northeast division of the National Hockey League for the 2011-2012 regular season. There are five
teams in the division and the numbers of goals each team scored are display in the graph below:
300
250
Frequency (no.)
200
150
100
50
0
Boston Ottawa Buffalo Toronto Montreal
Bruins Senators Sabers Maple Leafs Canadians
NHL Northeast Division Team
The graph is simple by displaying each team on the horizontal axis and the dependent variable on the
vertical axis. From the graph it is easy to see which teams scores more goals than the others.
In any graph, the dependent variable is always displayed on the y-axis (ordinate); that is, the vertical
axis. In the example graph above, the dependent variable is Total Goals Scored and the values of this
dependent variable are presented on the vertical axis. The increasing height on the ordinate indicates
increasing values of the dependent variable. The x-axis (abscissa) is the horizontal axis, and can reflect
an independent variable or another dependent variable. In this case, the variable Northeast Division
Team is a qualitative (nominal) dependent variable, with five different categories (teams). The title given
to each axis should be short and descriptive so as to accurately reflect the variable presented on that
axis. The values levels on the y-axis are up to you, but you generally do not want to have too great a
range of values or too small of a range. This can lead to a very misleading graph.
-52-
Back to TOC
3.5 Graphs for Frequency Distributions of Quantitative Variables
When data are measured on a quantitative variable one of three different graphs are generally used to
display the data: histograms, polygons, and line plots. There are some different rules for when each of
these three graphs are used, which I mention below, this this is not as important as displaying data in a
clear and simple manner. For each of the following examples, I use the frequency distribution of the
Political Attitude data form Section 3.1, which I have reproduced below:
Political Attitude f rf % cf crf c%

11 2 .04 4% 50 1.00 100%
10 0 .00 0% 48 .96 96%
9 4 .08 8% 48 .96 96%
8 6 .12 12% 44 .88 88%
7 6 .12 12% 38 .76 76%
6 13 .26 26% 32 .64 64%
5 10 .20 20% 19 .38 38%
4 2 .04 4% 9 .18 18%
3 4 .08 8% 7 .14 14%
2 3 .06 6% 3 .06 6%
1 0 .00 0% 0 .00 0%
People use the terms histogram and bar graph interchangeably, but they are not the same. Bar graphs
are used when the variable represented on the abscissa is qualitative (nominal), whereas histograms are
used when the variable on the abscissa is quantitative. The difference between bar graphs and
histograms is also aesthetic. In bar graphs the columns/bars above each category on the abscissa do not
touch, but in histograms the columns/bars do touch. The point of having the bars touch is to represent a
quantitative but discrete difference between entries on the abscissa. Each bar is centered on its
corresponding value on the abscissa and the width of each bar is supposed to represent the real limits
around the value. The height of the bar represents the magnitude of the dependent variable measure.
To create a histogram set up the ordinate and the abscissa by giving each axis a proper title. Next, label
the entries on each axis. Next, extend a column up to a point above each value on the abscissa that
corresponds to the frequency of that value on the abscissa. Again, make sure that the bars touch.
14
12
Frequency (no.)
10
0
1 2 3 4 5 6 7 8 9 10 11
Political Attitude (Rating)
It is important to place the smallest values on the left side of the abscissa nearest the ordinate. The point
where the abscissa and the ordinate meet, called the origin, represents the smallest values. This way, as
distance increases from the origin on either axis, it indicates greater and greater values.
-53-
Back to TOC
A polygon uses the height of lines between data points to represent the magnitude of a dependent
variable, instead of columns as in bar graphs and histograms. Polygons are used when the variable on
the abscissa is quantitative and is general used when the variable is continuous, but this is more of a
minor point. Remember from Chapter 2 that continuous data is where there are, theoretically, an infinite
number of possible measurements between any two points; hence, fractional measurement values are
possible. The use of a line to connect data points in a polygon graph is supposed to represent continuity
between the data points, that is, a quantitatively continuous variable. The graph below presents he
Political Orientation data as a polygon:
14
You’ll notice that I added a score of
12 12 to the abscissa and gave it a
frequency of zero. This is because
frequency polygons are always
Frequency (no.)
10
closed on the abscissa. That is, a
8 polygon should always have the
line extend from the lowest value
6
with a frequency down to the
4 abscissa. In this case, the rating of
1 had a frequency of zero, so the
2 left side of the graph was already
closed, but the rating of 11 had a
0 frequency of 2, so an additional
1 2 3 4 5 6 7 8 9 10 11 12 value above 11 and with a
Political Attitude (Rating) frequency of was needed in order
to “close” the polygon.
Similar to the polygon is the line plot, which is simply a frequency polygon that is not closed:
14
12
Frequency (no.)
10
0
1 2 3 4 5 6 7 8 9 10 11
Line plots and histograms can also be used to present data from several different groups within the same
graph. That is, if you have several different groups of people that were measured on the same variable,
these groups can be plotted within the same graph.
For example, say that I measure the time, in minutes, that it takes students to complete an exam. When I
exam the completion times, I plot the data by sophomores and juniors, which are the two college classes
in this particular course.
-54-
Back to TOC
Here is a line plot showing the frequencies of 10
sophomore and juniors that completed the
exam at different times. 8
Frequency (no.)
6
0
39 40 41 42 43 44 45 46 47 48 49 50
Sophom ores Juniors

10
Here is a histogram showing the frequencies
of sophomore and juniors that completed the 8
Frequency (no.)
exam at different times.
6
0
39 40 41 42 43 44 45 46 47 48 49 50
Sophom ores Juniors
3.6 Graphs for Frequency Distributions of Qualitative Variables

If the variable being measured and represented on the abscissa is qualitative and on a nominal scale, a
bar graph is used to display data. Bar graphs and histograms look nearly identical; however, the
difference is in the details. Histograms are used when the variable on the abscissa is quantitative and bar
graphs are used when the variable on the abscissa is qualitative. Also, when using a histogram the
columns/bars touch, but with bar graphs the columns/bars do not touch. This is supposed to indicate a
qualitative (categorical) difference between levels of the independent variable. An example of a bar graph
of the data in Section 3.2 (college major) is below:
16
14
12
Frequency (no.)
10
8
6
4
2
0
Psychology Neuroscience English Biology History
College Major
-55-
Back to TOC
3.7 Graphing Relative Frequencies, Percentages, Cumulative Frequencies and
Cumulative Relative Frequencies
Relative frequencies, percentages, cumulative frequencies and cumulative relative frequencies can also
be presented in graphic form. The only difference is the variable on the y-axis. Graphing relative
frequencies, percentages, cumulative frequencies, or cumulative relative frequencies depends on what
data needs to be presented. Below are examples of each type of graph, presented as a line plot from the
data in Section 3.1.
0.3
Relative Frequency (p)
0.25
0.2
0.15
0.1
0.05
0
1 2 3 4 5 6 7 8 9 10 11
60
Cumulative Frequency (no.)
50
40
30
20
10
0
1 2 3 4 5 6 7 8 9 10 11
1
Cumulative Relative Frequency
0.9
0.8
0.7
0.6
(p)
0.5
0.4
0.3
0.2
0.1
0
1 2 3 4 5 6 7 8 9 10 11
-56-
Back to TOC
3.8 Misuse of Graphs
It is important to never misuse graphs, that is, don’t use graphs to make a difference look bigger than it
actually is or smaller than it is. This is one way people tend to lie with statistics. A classic example is when
a researcher presents a restricted range of values on the y-axis to make a difference between levels on
the x-axis look bigger than it really is. Specifically, a person may present only values right around the
range of their groups or conditions being measured in order to make any difference between the
conditions look larger than it really is.
For example, we have two types of coffee, Brand-X and Brand-Z. Brand-X is the best-selling coffee in the
US and Brand-Z is a competitor. The company that produces Brand-Z does a blind taste test and counts
the number of people that prefer Brand-X, and the number of people that prefer Brand-Z. The frequencies
of people that like Brand-X and Brand-Z (out of 100 total people) are f = 48 and f = 52, respectively. The
difference is very small and the correct way to present the data in a graph is below:
53
The difference in the heights of the two columns is
52 not easy to see, which is good, because there is
Frequency (no.)
51 virtually no difference between the numbers of

people who prefer Brand-X to Brand-Z. Below, is
50
an example of how this data should not be
49 presented graphically (but often is):
48
47
Brand X Brand Z
Preferred Brand
100
In this case the small difference in frequencies 90
appears to be quite large, because the range on the 80
Frequency (no.)
ordinate has been restricted to values around the 70

obtained frequencies. Don’t do this! It is okay to 60
50
restrict the range of values on the ordinate
40
to some degree, but you should never blow a small 30
difference out of proportion. Best advice: If it is a 20
small difference, leave it that way! 10
0
Brand X Brand Z
Preferred Brand
3.9 Probability Distributions
Many graphs display probability as a function of some other variable; that is, probability is represented on
the y-axis and the levels of some independent variable or dependent variable are represented on the x-
axis. A probability distribution is like the frequency distributions covered above, where the levels of one
variable (e.g., Political Attitude) are listed with the frequency of each level.
But a probability distribution is a graph of the probability as a function of those levels. Thus, a probability
distribution is a graph that shows the probability on the y-axis as a function of all possible scores, or
categories, of some variable on the x-axis. A probability distribution for a discrete variable is easy to
create, because with a discrete variable, you should be able to determine the probability of being in one
level of that variable compared to another. For example, say we wanted to construct a probability
distribution for the independent variable sex (male/female) at the University of Scranton, based on fall
2011 data. That is, a probability distribution of full-time male and female undergraduate students at the
-57-
Back to TOC
University of Scranton. On the x-axis, we will plot the levels of the independent variable Sex and the
probability (proportion) of male and female University of Scranton students is plotted on the y-axis.
This graph can be used to approximate 1

the probability of selecting a male versus
a female student attending the University
0.8
of Scranton. It simply lists the probability
on the y-axis as a function of Sex on the
Probability
x-axis. 0.6
The graph to the right shows a probability 0.4

distribution for a discrete, nominal variable
on the x-axis. But what if the variable on 0.2
the x-axis was a continuous variable, like
time? Technically speaking, a probability 0
distribution for a continuous variable is not Male Fem ale
possible, because we could not determine
the probability of every possible Sex
measurement, because the number of
measurements is theoretically infinite.
Instead, a probability distribution for a continuous variable is estimated by plotting a probability density
function, which is basically a function that approximates what a probability density function would look
like for a set of parameters from a continuous variable.

X  2
One such distribution, the normal distribution, is a type of
f X  
probability distribution that is estimated by means of a
1 2 2
e
probability density function. The probability density function
for a normal distribution is: 2 2
Don’t worry, I’m not going to make you memorize this function or use it; we simply do not have enough
time to explain all of the terms and parameters in this function. The normal distribution is a theoretical
probability distribution that is based off of this mathematical formula. All theoretical probability
distributions are based off of mathematical assumptions and formulas, like that above.
The graph to the right includes

several plots of normal
distributions that were created
with the function above (don’t
worry about what the various
parameters mean right now).
The y-axis is the probability and
the x-axis includes values along
some continuous variable.
We will cover the normal

distribution in greater detail
later, but you should note a few
things about the normal
distribution. First, it is assumed
that the total probability (area)
under the curve is equal to 1.00.
The height of the curve at any
point on the x-axis represents
that probability of observing that
particular value on the x-axis:
-58-
Back to TOC
The greater the height of the curve at any point, the more probable that value will be observed. As can
also be seen, in the normal distribution the most probable values tend to be at the center of the
distribution and the least probable values tend to lie in the tails, or ends, of the distribution. This is related
to the concept of central tendency, which will discuss i Chapter 5. Finally, the normal distribution is
perfectly symmetrical around its midpoint, so that the probability of being above that midpoint is .5 and the
probability of being below that midpoint is .5.
It is not always the case that a set of values is normally distributed; however, we often make a normality
assumption with statistics. The normality assumption states that a distribution of collect scores is
normally distributed. However, the normal distribution is theoretical and is based on mathematical
assumptions. Probability distributions that are based on actually collected data are known as empirical
probability distributions and often differ from
normality. If the underlying distribution of scores is not
normally distributed, it can throw off the results of a
study. As long as a distribution of scores does not
deviate too much from normality there is no problem.
When a distribution is not normal it can take on many
different shapes and I have listed some of these below.
A negatively skewed distribution occurs when the

hump lies to the right of center and the prominent tail lies
to the left. This occurs when extremely small values of
low probability occur in the data:
A positively skewed distribution occurs when the hump

of the distribution lies to the left of center and the
prominent tail lies to the right. This occurs when
extremely large values of low probability occur in the
data:
A multi-modal distribution occurs when there are two or

more humps in the distribution. This usually occurs when
you accidentally sample from two different populations:
Finally, a rectangular distribution occurs when each

score in a distribution is equally frequent and the shape
of the distribution resembles a rectangle:
The latter two distributions rarely occur in data. The

distributions we’ll be concerned with most are the normal
distribution, positively skewed distributions, and
negatively skewed distributions.
-59-
Back to TOC
3.10 Empirical vs. Theoretical Distributions
One final and important topic to mention is the difference between theoretical distributions and empirical
distributions. Empirical distributions are based on actual measurements of a variable. The graphical
distributions plotted in Sections 3.1-3.7 would be empirical distributions as those graphs were based on
actual data. Thus, empirical probability distributions express empirically observed relationships between
variables. In contrast, theoretical distributions express relationships between variables mathematically
based on a set of mathematical assumptions. The normal distribution discussed above is a theoretical
probability distribution as it is not based on any collected data; rather, it is based on mathematical logic.
For example, below is a theoretical normal distribution with a mean of 100 and an empirical distribution
derived from the theoretical distribution. Notice that the shapes are similar, but not identical.
Theoretical distribution:
Empirical distribution:
-60-
Back to TOC
1. Use the following data to answer the questions, below: A professor recorded grades from a 5-point quiz
below:
5 1 0 2 1
3 4 5 0 1
2 5 2 3 4
3 2 3 5 3
5 3 2 4 3
a. Create a frequency distribution with the scores provided above. Be sure to include the absolute
(raw) frequencies, the relative f frequencies, the cumulative frequencies, the cumulative relative
frequencies, and the cumulative percentages.
b. What proportion of students had a score of 3?
c. What proportion of students had a score of 3 or less?
d. What proportion of students had a score less than 3?
e. What proportion of students had a greater than 3?
f. What proportion of students had a score of 3 or more?
g. How many students had a score of 4?
h. How many students had a score of 4 or less?
i. How many students had a score of 4 or more?
j. How many students had a score less than 4?
k. What is the probability that a randomly selected student has a score of 5?
l. What is the probability that a randomly selected student has a score of 5 or of 4?
2. Use the following data to answer the questions, below: A professor keeps records for how many times
each of his 20 students come to his office hours each semester. The number of days each of these 20
students came to his offices hours are:
7 7 6 4 3
6 3 7 6 6
4 6 6 6 7
5 6 8 7 5
a. Create a frequency distribution with the data provided above. Be sure to include the absolute
(raw) frequencies, the relative f frequencies, the cumulative frequencies, the cumulative relative
b. What proportion of students came to the office 8 times?
c. What proportion of students came to the office less than 8 times?
d. What proportion of students came to the office 8 times or less?
e. What proportion of students came to the office more than 8 times?
f. What proportion of students came to the office 5 times or 6 times?
g. What proportion of students came to the office 6 times or more?
h. What proportion of students came to the office more than 6 times?
i. How many students did not come the office 4 times?
j. How many students came the office 4 times or less?
k. How many students came the office less than 5 times?
-61-
Back to TOC
3. Use the following data to answer the questions, below: Dr. Evil owns a potato-chip company. He keeps
records of how many days each of his employees were sick during the previous year. The number of
days each employee was sick are:
8 7 8 9 6 5 0 1 2 3 3 4 5
6 8 6 4 6 0 9 2 3 4 7 6 8
5 4 0 1 2 3 5 6 6 4 5 3 2
0
a. Construct a frequency distribution of these scores. Be sure to include the absolute (raw)
frequencies, the relative f frequencies, the cumulative frequencies, the cumulative relative
b. What proportion of employees were sick for exactly 6 days?
c. What proportion of employees were sick for 6 days or fewer?
d. What proportion of employees were sick for 6 days or more?
e. What is the probability an individual was sick for exactly 4 days?
f. What is the probability an was sick for 1 day or 0 days?
4. A college class is normally made up of many different college majors. You want to know which majors
are taking an ‘Introduction to Basket weaving’ course. You ask each student in the class what their
primary major is, and you find that the students in the class are Biology majors (B), English majors (E),
Psychology majors (P), or Undeclared majors (U). The data for the students are as follows:
U U U U U E U B U P P E U U P B U
U E U B U U E U U B U U E U E P U
U E E P U U U P B U U U B U U E
Construct a frequency table that contains absolute frequencies, relative frequencies, and percentages of
the data above.
5. Use the following data to answer the questions, below: In a recent study, the numbers of students from
each of the following majors participated:
Biology = 13 Exercise Science = 12 Communication = 7

Undecided = 9 Counseling = 4 Occupational Therapy = 3
Education = 11 Computer Science = 2 Business = 8
History = 2 Philosophy = 2 Nursing = 4
Psychology = 2 Criminal Justice = 6
a. Construct a frequency distribution of these score and include the relative frequencies (rf) and
percentages.
b. What proportion of students were Education Majors?
c. What proportion of students were Undecided Majors?
d. What is the probability that a randomly selected student is a History Major?
e. What is the probability a randomly selected student is a Counseling Major?
f. What is the probability a randomly selected student is a Psychology Major or a Computer Science
Major?
6. What is a probability distribution? Why is the nature of probability distributions for qualitative and
discrete variables different from that of continuous variables?
7. What is the difference between an empirical and a theoretical distribution?
-62-
Back to TOC
8. A researcher surveyed 2000 individuals on their attitudes toward setting a national speed limit on
interstate highways. One of the questions asked the individuals to indicate whether they thought that the
national speed limit should be set at 65 mph. Responses could range from 1 (“definitely should not be set
to 65 mph”) to 4 (“definitely should be set to 65 mph”). Using the percentages below, compute the
absolute (raw) frequencies, relative frequencies, cumulative frequencies, and cumulative relative
frequencies for the set of responses below:
Response %
4 35.0 %
3 15.0 %
2 20.0 %
1 30.0 %
9 The table below lists the numbers of majors enrolled in a Fundamentals of Psychology class. Using this
frequency distribution table, create a bar graph (be sure to use labels).
Major f
Undeclared 31
Communications 6
English 5
Education 11
Biology 7
Criminal Justice 3
Political Science 2
Business 3
Occupational
7
Therapy
Psychology 8
10. Use the following data to answer the questions, below: A survey asked each participant to rate how
much they agreed with the Democratic Party on a scale ranging from -3 to +3. The frequencies of each
rating given for N = 177 participants are listed below.
Rating -3 -2 -1 0 +1 +2 +3
Frequency 5 19 18 43 39 49 4
a. What type of data is “agreement with the Democratic party” being measured on?
b. What type of a graph should you use to plot these frequency data?
c. Construct the appropriate graph of the data above.
11. Here are frequencies of Freshmen, Sophomores, Juniors, and Seniors enrolled in a Statistics class:
Freshmen = 4 Sophomores = 19 Juniors = 5 Seniors = 3
a. What type of measurement scale is "class" measured on?

b. What type of graph (bar graph, line graph, or histogram) should you create to display these data?
c. Construct the appropriate graph of the data above.
-63-
Back to TOC
Use the frequency distribution table below to complete exercises 11 – 15:
Score f rf cf crf
5 5 .2 25 1.00
4 3 .12 20 .8
3 7 .28 17 .68
2 5 .2 10 .4
1 3 .12 5 .2
0 2 .08 2 .08
11. Create a histogram for the absolute (raw) frequencies.
12. Create a line plot for the absolute (raw) frequencies.
13. Create a line plot for the relative frequencies. Is the shape similar or different to the graph in #12?
14. Create a line plot for the cumulative frequencies above.
15. Create a polygon for the cumulative relative frequencies above.
16. A principal in a small school measured the intelligence of fifth-grade students in her school. The
intelligence test scores for those students were as follows:
129 99 98 113 103 128 102 110 80 105

93 98 109 109 100 111 106 96 108 90
104 94 92 119 127 89 95 92 105 108
83 100 107 106 101 118 84 119 105 111
118 106 122 120 102 117 103 117 103 88
Draw a stem and leaf plot for the set of scores. Rotate the plot so the tens and hundreds digits appear at
the base rather than on the right.
17. The IQ (intelligence) scores of thirty-three students from the same third grade classroom are
presented. One set is for boys and the other set is for girls. Create a back-to-back stem-and-leaf plot so
that you can visually compare the IQ scores of the boys with the IQ scores of the girls.
Boys: 113 113 122 116 122 134 134 105 122 126 129 133
122 133 108 120 144 104
Girls: 113 102 119 114 108 111 122 113 125 118 96 112
124 122 115
-64-
Back to TOC
18. The following histogram shows the GRE Verbal scores for 200 students entering a local university.
Use this graph to answer the questions that follow. (The numbers above each bar are the frequencies.)
50
44
45
40
35
35
31
Frequency
30
25
19 20
20 18
15 12
10
10
5
5 3 2
0 1
0
150 151 152 153 154 155 156 157 158 159 160 161 162
GRE Verbal Score
a. How many of the students had scores between 152 and 156 (inclusive)?
b. How many students had scores less than 155?
c. How many students had scores greater than 160?
d. Do you agree with the following statement? “The distribution looks bell-shaped, with one outlier.”
19. Here is a histogram for a set of test scores from a 10-item makeup quiz given to a group of students
who were absent on the day the quiz was given. Use the graph to answer the questions that follow.
Frequency
Makeup Quiz Score
a. What percentage of students scored higher than 4?

b. What percentage of students scored a 3?
c. What percentage of student scored less than 3?
-65-
Back to TOC
Chapter 4: Rankings in a Distribution
4.1 Percentiles and Percent Ranks

Frequency distributions are used to organize raw data and determine the relative location of a particular
value in a distribution; that is, what proportion or percentage of scores in a distribution are below or above
a value. Frequency distributions can also be used to determine the exact value at which a certain
percentage of scores are below or above a value.
The percent rank or percentile rank (PR) is the percentage of scores that occur at or below a certain
value in a distribution. That is, the percentage of scores that are less than or equal to a value (X). The
percent rank is similar to the cumulative percentage and cumulative relative frequency of a value in a
distribution. Similarly, a percentile (XP) is the value that a certain percentage of scores (%) are less than
or equal to. A percentile is the value associated with a certain percentile rank.
You’ve probably heard these terms before when discussing how well you performed on a standardized
test such as the SATs or ACTs. For example, when you get your SAT score you are also told your
percentile rank. The percentile rank that was given was the percentage of test-takers who you scored
better than or equal to. Here’s a portion of the raw SAT scores and percentile ranks in each of the three
1
SAT sub-tests:
Score Critical Reading Mathematics Writing

560 69 63 72
550 66 61 69
540 63 58 66
530 60 55 63
In the portion of the table above the raw scores in the left column would be considered percentiles and
the value under each subtest column is the approximate percentile ranks associated with each score. For
example, if you took the SATs and received a score of 540 on the Mathematics test your approximate
percentile rank is 58%. This means that your score is greater than or equal to about 58% of all the other
test-takers. Similarly, if you scored 560 on the Writing test then you scored greater than or equal to about
72% of all test-takers. Thus, percentile ranks tell you how well a score is relative to the entire distribution.
1
http://professionals.collegeboard.com/profdownload/SAT-Percentile-Ranks-2009.pdf
-66-
Back to TOC
Alternatively, if you wanted to know approximately what score on the Critical Reading test is greater than
or equal to 66% of all other scores, you look up a percentile rank of 66% under the Critical Reading
column, then find the value in the Score column (550); this would be the approximate score that is greater
than or equal to 66% of all Critical Reading scores. Similarly, if you wanted to know what score is greater
than or equal to 65% of all Mathematics test scores you would locate the percentile rank in the
Mathematics column that is immediately greater than 65%, which is 66%. The score that is associated
with that percentile rank (540) will be the percentile that is approximately greater than or equal to 66% of
all other Mathematics scores.
I have been saying that the percentiles and percentile ranks above are approximate, because, in practice,
you must calculate the percentile or percentile ranks in which you are interested. In the following sections,
we cover how to calculate more exact percentiles and percentile ranks from a frequency distribution.
4.2 Calculating Percentile Ranks

For the following examples of calculating percentiles and percent ranks, we’ll use the frequency
distribution for quantitative data from Chapter 3, which I have reproduced below:
Political Attitude (X) f rf % cf crf c%

11 2 .04 4% 50 1.00 100%
10 0 .00 0% 48 .96 96%
9 4 .08 8% 48 .96 96%
8 6 .12 12% 44 .88 88%
7 6 .12 12% 38 .76 76%
6 13 .26 26% 32 .64 64%
5 10 .20 20% 19 .38 38%
4 2 .04 4% 9 .18 18%
3 4 .08 8% 7 .14 14%
2 3 .06 6% 3 .06 6%
1 0 .00 0% 0 .00 0%
Recall, the percentile rank for a value is the percentage of scores that a value is greater than or equal to.
The formula for determining the percentile rank for a particular value is the following:
 n  n w / i X  L 
PR   L 100
 N 
We will go over how to find each of the terms for this expression in a moment, but we first need a value
for which to determine a percentile rank. Say that we want to find the percentile rank for a Political
Attitude value of X = 8. Locate that value in the frequency distribution table (I highlighted the row for that
value in yellow) and make this your reference point. The values for each term in the percentile rank
formula above are based off this reference point.
The following is a description of each term in the formula above, along with where to find the value in the
frequency distribution table:
nL is the number of scores that are less than the value for which we are calculating the percentile
rank. That is, it is the cumulative frequency (cf) of the value for X = 7. In this case, the cumulative
frequency of X = 7 is 38; hence, nL = 38.
nW is the number of scores associated with the value for which we are calculating the percentile
rank. That is, it is the absolute frequency (f) of the value X = 8. In this case the frequency of X = 8
is 6; hence nW = 6.
-67-
Back to TOC
i is how many different values are associated with the value for which we are calculating the
percentile rank. In our example, there is only one value associated with X = 8. In some
distributions i can be greater than 1; but we will not be covering those distributions. In this
example, i = 1.
L is the lower real limit (LRL) of the value for which we are calculating the percentile rank.
Remember, the lower real limit of any value is one-half unit below our measured value. in this
example, the lower real limit of 8 is L = 7.5.
X is simply the value for which we are calculating the percentile rank (X = 8)
N is the total number of scores in the distribution, which in this case is 50.
 38  6 / 18  7.5 
PR    100
 50 
Once you have located each value in the table, simply plug them into
 38  60.5 
the formula and solve for the percentile rank of X = 8. PR    100
 50 
Thus, X = 8 is greater than or equal to exactly 92% of all other scores
 38  3 
in the set of data. Or, another way to state this is that 92% of the PR    100
individuals surveyed self-reported a political attitude that was at most  50 
‘somewhat conservative’.
 41 
PR    100
 50 
PR  0.92100
As another example, what is the percentile rank for a score of X = 4?
PR  92%
 7  2 / 14  3.5 
PR    100
4.3 Calculating Percentiles  50 
 7  20.5 
Recall, a percentile is the value that is greater than or equal to a PR    100
 50 
certain percentage of scores in a distribution. Below, I reproduced the
 7  1
 100
frequency distribution table from above. Let’s say that we want to PR  
determine the percentile that is associated with a percentile rank of  50 
35%; that is, what value is greater than or equal to 35% of the scores
 8 
in this distribution? PR    100
 50 
Political Attitude f rf % cf crf c% PR  0.16100
(X) PR  16%
11 2 .04 4% 50 1.00 100%
10 0 .00 0% 48 .96 96%
9 4 .08 8% 48 .96 96%
8 6 .12 12% 44 .88 88%
7 6 .12 12% 38 .76 76%
6 13 .26 26% 32 .64 64%
5 10 .20 20% 19 .38 38%
4 2 .04 4% 9 .18 18%
3 4 .08 8% 7 .14 14%
2 3 .06 6% 3 .06 6%
1 0 .00 0% 0 .00 0%
Like calculating percentile ranks in section 4.2, we need to determine a reference point in the frequency
distribution table. This reference point is based on where the percentile rank (35%) we are interested in
falls within the cumulative percentage column (c%). As you will notice, 35% does not appear in the
-68-
Back to TOC
cumulative percentage column. What you do is locate the cumulative percentage in that column that is
greater than the percentile rank we are looking for. In this case, the cumulative percentage that is
immediately greater than 35% is 38%. This is the reference point, which I highlighted in green in the table.
You can also determine where a reference point will be by using the cumulative relative frequencies.
Simply convert your percentile rank (35%) into a proportion (35%/100% = .35) and locate the cumulative
relative frequency that is immediately greater than your value. In this example, that cumulative relative
frequency is .38. This will be your reference point, which is the same reference point that we just found.
The formula for determining the percentile rank for a particular value is the following:
Each of the terms in this formula are the same as those found in the formula for calculating percentile
ranks in Section 4.2. The only new term is the P in the numerator. P is the percentile rank that we are
interested in (35%), expressed as a proportion. That is, given we are interested in calculating the
percentile that is greater than or equal to 35% of the scores, P = .35.
 50.25  9 
X P  4.5   1
I have filled in the remaining values from the table above, into the
 10 
expression below. Before going through the calculations below, I
encourage you to find where each of the values came from in the  12.5  9 
frequency distribution table above.
X P  4.5   1
 10 
 3 .5 
X P  4.5   1
Thus, a score of X = 4.85 is greater than or equal to 35% of all of the  10 
scores in this distribution. Or, the percentile that is associated with 35% of X P  4.5  0.35
the scores is 4.85.
X P  4.85
-69-
Back to TOC
Use the following information to complete Exercises 1 – 3 : A statistics professor gave a 5-point quiz to
the 50 students in his class. Scores on the quiz could range from 0 to 5: The following frequency analysis
resulted:
Quiz Score f rf cf crf c%
5 4 .08 50 1.00 100%
4 10 .20 46 .96 96%
3 14 .28 36 .72 72%
2 10 .20 22 .44 44%
1 8 .16 12 .24 24%
0 4 .08 4 .08 8%
1. Compute the values that define the following percentiles:

th th th th th th
a. 25 b. 50 c. 55 d. 75 e. 80 f. 99
2. What is the interquartile range of the data in #1?
3. Compute the exact percentile ranks that correspond to the following scores:
a. 2 b. 3 c. 4 d. 1
Use the following information to complete Exercises 3 – 5: A political scientist took a pool of the political
attitudes of the students in one of his classes. Students were asked to rate, on a scale from 1 to 11,
“What is your overall political attitude?”, where 1 = extremely liberal and 11 = extremely conservative. The
following frequency analysis resulted:
Political Attitude Score f rf cf crf

11 1 .015 67 1.000
10 3 .045 66 .985
9 14 .209 63 .940
8 6 .090 49 .731
7 2 .030 43 .642
6 10 .149 41 .612
5 9 .134 31 .463
4 3 .045 22 .328
3 11 .164 19 .284
2 7 .104 8 .119
1 1 .015 1 .015
4. Compute the values that define the following percentiles:

th th th th
a. 25 b. 50 c. 57 d. 75
5. What is the interquartile range of the data in #4?
6. Compute the exact percentile ranks that correspond to the following scores:
a. 3 b. 5 c. 7 d. 9
-70-
Back to TOC
Chapter 5: Central Tendency
5.1 Measures of Central Tendency
Okay, so you collected data, created a
frequency distribution, made a graph...now
what? The statistical procedures described
in the previous chapters were all about
organizing data into something easier to
understand. But even after organizing data it
is still complex. It would be nice to have a
single value that could be used to represent
the entire data set, that is, a single value to
represent the whole distribution. A
descriptive statistic is a value that
describes or represents a set of data. Which
value from a distribution should be used as
the representative and why is any one value
in the data set better than another value?
There are at least three characteristics you
look for in a descriptive statistic to represent
a set of data.
First, a good descriptive statistic should be similar to many of the scores in a distribution; hence, its value
should have a high frequency. Second, it should balance the distribution so some scores are greater than
it and some scores are less than it; neither the greater-than or less-than scores are overrepresented.
Finally, it should take the individual values from the distribution into account so no value is left out. So
what value do we use as a descriptive statistic?
To answer this, examine the hypothetical

normal distribution of Family Guy Trivia
Scores to the right. Remember, in a normal
distribution the most frequent (most likely)
scores cluster near the center of the
distribution and the less frequent (less likely)
scores fall into the tails. Central tendency
simply means that most scores in a normally
distributed set of data tend to cluster in
the center of a distribution. Thus, the central
tendency area is associated with most of the
scores in a set of data; about 68% of the
scores to be exact.
Recall from above, one characteristic of a good descriptive statistic is that it should be similar to many
scores in a distribution. Thus, values near the center of a distribution make good descriptive statistics,
because those values are similar to many of the values in the distribution. Also, because a normal
distribution is symmetrical around its center with half of the scores above the center and the other half of
the scores below the center, a score from this central tendency area should roughly equally represent the
larger values and the smaller values of the distribution. Thus, measures of central tendency are useful
for representing a set of data, because measures of central tendency describe where most of a
distribution lies along some continuum.
-71-
Back to TOC
5.2 The Mode: The Most Frequent Score
This section uses the following set of quiz scores (X):
10 10 10 9 9 9 9 8 8 8 8 8 7 7 7 7 7 7 6 6 6 6 6 6 6
5 5 5 5 5 5 5 5 5 4 4 4 4 4 4 4 4 3 3 3 3 3 2 2 1
The mode (Mo) is the most frequently occurring score in a distribution. If you were to determine the
frequency of each value in the data set above, you would find that X = 5 has the greatest frequency with f
= 9 scores; hence, the mode is X = 5. Of course, the mode is even easier to determine if you create a
frequency distribution of the data set. In this case, you look for the greatest frequency value in the
frequency (f) column and look-up the value in the X column that corresponds to that greatest frequency.
The mode is easy to determine but has limitations. Specifically, there can be more than one mode in a
distribution and even if there is only one, it still may not be a good representative relative to other values.
Recall from Chapter 3, a theoretical normal distribution has one hump; hence, a normal distribution has
one mode. However, an empirical distribution can be multi-modal, meaning there is more than one hump
(i.e., more than one mode) in the distribution.
For example, in the data below there are two most frequent scores, X = 9 and X = 4, both with frequency
of f = 6. In this case, each value, 9 and 4, is a mode. But, neither one is a better mode than the other:
10 10 9 9 9 9 9 9 8 8 7 7 6 6 5 5 4 4 4 4 4 4 3 3 2 2 1 1
Another problem with the mode is every score can be a mode. This occurs when you have a rectangular
distribution and each score in the distribution is equally frequent. For example, the data in the set below
has one of each value, which is problematic, because we want a single measure of central tendency.
10 9 8 7 6 5 4 3 2 1 0
Another problem is that the mode might be at one end of a distribution, not at the center. In the data
below, the mode of 10 represents the larger values quite well but not the smaller values.
10 10 10 10 10 10 10 9 8 7 6 5 4 3 2 1 0
Finally, even if the scores in a set of data are normally distributed it is possible that the most frequent
score has a frequency that is only one or two counts greater than the second most frequent score. In the
first example above, the mode was X = 5, because its frequency of nine scores was greatest. However,
the value X = 4 had a frequency of eight, which is only one less than the frequency of nine for X = 5. Why
is 5 a better mode than 4?
5.3 The Median: The “Middle” Score

The median (Md) is the middle score of a distribution, because 50% of the scores are greater than a
th
median and 50% of the scores are less than a median. Thus, the median is the 50 percentile. This is
what generally makes the median a better measure of central tendency than the mode: the median
balances a distribution such that equal numbers of scores are greater than and less than the median.
To find the median value of a distribution you must determine the median’s location and then determine
the median value at that location. Determining median location and value differs whether you have an
even or an odd number of scores. For example, in the set of scores below there are n = 11 scores; hence,
an odd number of scores:
10 10 9 7 7 6 5 4 3 2 2
-72-
Back to TOC
When you have an odd number of scores, use the following equation to identify the median location:
N 1
Md 
2
The N is the number of scores in the data set (11). Solving for the median location, we get:
N  1 11  1 12
Md     6th
2 2 2
This indicates that the median is located at the sixth ordinal position in the distribution. To determine the
median value, first make sure the scores are rank-ordered from smallest to largest. Next, count off six
ordinal positions starting with the smallest value. Once you find the sixth position, the value at that
position is the median’s value. In this case, the median value is X = 6. Note that the fact that the median
th
location (6 ) and median value (X = 6) are the same is coincidence:
X: 2 2 3 4 5 6 7 7 9 10 10
Ordinal Position: 1st 2nd 3rd 4th 5th 6th 7th 8th 9th 10th 11th
If there is an even number of scores, you must find two locations that surround the median, then identify
the values at those two locations, and take the average of those values. The following data an even
number of scores:
20 19 18 16 15 14 12 11 11 11 10 9
Use the following to determine the two locations that surround the median:
N N2
Md  and Md 
2 2
Again, N is the number of scores in the data set (12). Solving for the median locations we have:
N 12 N  2 12  2 14
Md    6th and Md     7 th
2 2 2 2 2
This indicates that the median is between the values at sixth and seventh ordinal positions. To determine
the median value make sure the scores are rank-ordered from smallest value to largest value. Next,
starting with the smallest value, count off six positions and seven positions. Once you locate the sixth
position and seventh position take the average of the values at those two positions. This will be the
median value:
X: 9 10 11 11 11 12 14 15 16 18 19 20
Ordinal Position: 1st 2nd 3rd 4th 5th 6th 7th 8th 9th 10th 11th 20th
In this example, X = 12 is at the sixth position and X = 14 is at the seventh position. The average of 12
and 14 is (14+12)/ = 13. Thus, the median value is X = 13.
-73-
Back to TOC
As good as the median is there is a drawback. The median does not take into account the actual values
of the scores in a set of data. For example, take the following two sets of scores, each with n = 5 scores:
Set 1: 1 4 5 7 10
Set 2: 4 4 5 6 6
In each set the median is equal to 5; hence, the median value is unaffected by the values greater and
less than its value. In order to account for the individual values in a distribution, which makes is a
desirable characteristic of a good measure of central tendency; we must calculate the mean of a
distribution, which we will get to shortly.
Although the median does not take individual values into account, it is frequently reported as a measure
of central tendency when data sets are incomplete, or data are severely skewed; a point that we will
2
return to later. One example of the median being used instead of the mean is median household income.
The median is reported as a measure of central tendency for household income, because there is
extreme positive skew in household income data; that is, a low frequency of extremely high incomes,
which pull the mean up.
5.5 Quartiles
The median divides a distribution in half. However, a researcher may want to divide a distribution into
smaller parts, perhaps thirds, quarters, fifths, sixths, etc. Quartiles split a distribution into fourths, or
st
quarters (25%) of the distribution. There are actually three quartiles in a distribution: The 1 quartile (Q1)
st
separates the lower 25% of the scores from the upper 75% of the scores. Thus, the 1 quartile is the
nd
median of the lower 50% of the scores in a distribution. The 2 quartile (Q2) is the median and separates
rd
the lower 50% of the scores from the upper 50% of the scores. The 3 quartile (Q3) separates the upper
rd
25% of the scores from the lower 75% of the scores. Thus, the 3 quartile is the median of the upper 50%
of the scores in a distribution.
Determining the quartile value is much like determining the median. First, determine the locations of the
first and third quartiles. Second, locate values at those locations. Also, like when determining the median,
there are different procedures for determining the quartiles based on your having an even versus an odd
number of scores.
st
For the 1 quartile (Q1), if you have an odd number of scores, use the following:
N 1
Q1 
4
st
For the 1 quartile (Q1), if you have an even number of scores, use the following:
N2
Q1 
4
rd
For the 3 quartile (Q3), if you have an odd number of scores, use the following:
3N  3
Q3 
4
2
http://www.google.com/url?q=https%3A%2F%2Fwww.efanniemae.com%2Fsf%2Frefmaterials%2Fhudmedinc%2
F&sa=D&sntz=1&usg=AFQjCNGAJBgG_VyRmwKDw6n6NrmMrryRLQ
-74-
Back to TOC
rd
For the 3 quartile (Q3), if you have an even number of scores, use the following:
3N  2
Q3 
4
One thing to note before examples are presented is that the value you calculate from each expression
above provides you with the location of the first or third quartile. This location will be a whole number (2,
3, 4, 5, etc.), which indicates that the value at that location is the first or third quartile. But, the location
value can also end in a .5 (1.5, 2.5, 3.5, 4.5, 5.5, etc.), which indicates that the quartile is the average of
two values. For example, if you calculate the first quartile location and it comes out as Q1 = 2.5; this is
nd rd
telling you that the first quartile is between the 2 and 3 locations; hence, it is the average of the values
nd rd
at the 2 and 3 positions.
For example, the following set of n = 9 scores include an odd number of scores:
7 7 8 9 10 11 11 14 15
The locations of the first and third quartiles are:
N  1 9  1 10
Q1     2.5th Position
4 4 4
3N  3 3* 9  3 27  3 30
Q3      7.5th Position
4 4 4 4
Below, each score is listed with the ordinal position of each score from the data set above.
st nd rd th th th th th th
Position : 1 2 3 4 5 6 7 8 9
Score: 7 7 8 9 10 11 11 14 15
nd rd
The Q1 = 2.5 means the first quartile is between the 2 and 3 ordinal positions; hence, it is the average
of the values at the second and third ordinal positions. The value at the second ordinal position is 7 and
the value at the third ordinal position is 8. Hence the value at first quartile is (7 + 8)/2 = 7.5.
nd th
The Q3 = 7.5 means that the third quartile is between the 7 and 8 ordinal positions; hence, it is the
average of the values at the seventh and the eighth ordinal positions. The value at the seventh ordinal
position is 11 and the value at the eight ordinal position is 14. Hence the value at third quartile is (11
+14)/2 = 12.5.
The difference between the third quartile and the first quartile (Q3 - Q1) is the interquartile range, which
in this case is equal to, 12.5 - 7.5 = 5. The interquartile range is the middle 50% of the scores in a set of
data.
As a second example, the following set of n = 12 scores include an even number of scores:
6 7 7 8 9 10 10 11 11 14
-75-
Back to TOC
The locations of the first and third quartiles are:
N  2 10  2 12
Q1     3rd Position
4 4 4
3N  2 3* 10  2 30  2 32
Q1      8th Position
4 4 4 4
Below, each score is listed with the ordinal position of each score from the data set above.
st nd rd th th th th th th th
Position : 1 2 3 4 5 6 7 8 9 10
Score: 6 7 7 8 9 10 10 11 11 14
rd
The Q1 = 3 means that the first quartile is at the 3 ordinal position. The value at that position is 7; hence
th
Q1 = 7. Similarly, the Q3 = 8 means that the third quartile is at the 8 ordinal positions. The value at the
eight ordinal position is 11; hence Q3 = 11. Thus, the interquartile range is Q3 - Q1 = 11 - 7 = 4.
5.6 The Mean: Arithmetic Average

The mean is the arithmetic average of all the scores in a distribution. The mean is the most-often used
measure of central tendency, because it evenly balances a distribution so both the large and small values
are equally represented by the mean and also takes into account all individual values. It is this second
point that is the main advantage of the mean: it accounts for individual scores in a data set.
To calculate the mean, first sum (add together) all of the scores in a distribution (ΣX). Then, divide that
sum by the total number of scores in the distribution (n). The expressions for calculating the mean from a
sample and from a population are (in APA format, an italicized M is used to represent the mean):
X X
Sample: X  Population:  
n N
Both expressions are functionally equivalent, that is, each requires you to sum the scores in a distribution
and divide that sum by the number of scores.
For example, we want to calculate the mean from the sample of n = 11 scores from the median example
in Section 5.4. The scores in that set of data were:
2 2 3 4 5 6 7 7 9 10 10
To calculate the mean, determine the sum of the values (ΣX = 65). Next, divide that value by the total
number of scores in the distribution (n = 11). Thus:
X 65
X   5.909
n 11
This value (M = 5.909) is close to the median value that was calculated earlier (Md = 6). The difference
between the mean and the median reflects the mean’s taking into account the individual values. To
-76-
Back to TOC
emphasize that the mean takes into account all of the individual values in a set of data: Recall the two
sets of n = 5 scores from Section 5.4 that had the same median location and value:
Set 1: 1 4 5 7 10 Md = 5
Set 2: 4 4 5 6 6 Md = 5
Calculating the mean for each set of data we have:
27 25
Set 1: X   5.4 Set 2: X   5.5
5 5
You can see why the mean is a more accurate measure of central tendency, as it takes the individual
scores into account; the median does not.
The mean is the mathematical center or balancing point of a distribution. This means that the positive
and the negative deviations between the individual scores and the mean will cancel each other out. For
example, by subtracting the mean from each score in Set 2 from above and summing those values, we
get:
X 4 4 5 6 6
X 5 5 5 5 5
X  X  -1 -1 0 1 1 
 XX 0 
The negative deviations (-1 + -1 = -2) and the positive deviations (1 + 1 = 2) cancel other out (-2 + 2 = 0).
Thus, the mean evenly balances large scores and small scores in the distribution. Indeed, this will be true
in any distribution (within rounding error of course).
The sample mean is the preferred measure of central tendency, because it is the best unbiased
estimator of the population mean (μ). This is related to the fact that positive and negative differences
between the sample’s mean and the individual values in that sample will cancel out (the mean is a
“balancing point”). Indeed, as long as the sample was randomly selected from and is representative of the
population, the values that would be expected to be obtained from the population should be represented
in the sample.
There is one problem with the mean and ironically it has to do with its taking individual values into
account. Although this is a good characteristic, by taking individual values into account the mean can be
influenced by extremely large or extremely small values (outliers). If a few extremely large or extremely
small values occur in a data set they can pull the mean in that direction and away from the center of the
distribution. Specifically, extremely small values will pull the mean down and extremely large values will
pull the mean up. This only occurs in skewed distributions. In such cases, it is good to use the median as
a measure of central tendency, because although it does not take individual values into account the
median is not influenced by individual values; hence, it cannot be influenced by extreme scores.
5.8 Central Tendency and Scales of Measurement

One important point is which measure of central tendency should be used based on the measurement
scale (nominal, ordinal, interval, or ratio) of the variable being measured, and if the data is quantitative
whether the values on the scale are continuous or discrete.
When a variable is measured on a nominal scale, that is, when your data is qualitative the mode is the
best measure of central tendency. For example, if I measured the sex of each subject in a research study,
-77-
Back to TOC
I might have 5 males and 15 females. The mode is the 15 females, that is, not just the number 15, but
also the group. I cannot calculate the “median sex” of the “mean sex” when I am counting people.
When a variable is measured on an ordinal scale, any one of the three measures of central tendency
could be used although the median may be more appropriate. This is because the individual values on an
ordinal scale are meaningless and are relative only to the placement of other values.
Finally, when a variable is measured on an interval, or a ratio scale, any one of the three measures of
central tendency is appropriate; however, the mean is generally preferred. This is because the mean
accounts for every score in the distribution.
However, one issue with using the mean is whether the variable is discrete or continuous. If a variable is
continuous, then rounding the mean to any number of decimal places is perfectly appropriate, because
theoretically data from a continuous variable can be any value. For example, if I was measuring time in
seconds, a mean of M = 1.25 seconds or of M = 6.8923 seconds both make sense.
In contrast, if a variable is discrete, the rounding may be a little less appropriate, because fractions of
values are not possible on discrete scales of measurement. For example, if I calculate the average
number of students in all statistics classes that I have even taught, it could out to be about M = 28.75, but,
I cannot have 0.75 of a person! This means that the average number of students in my statistics classes
is between 28 and 29 students per class. When data are discrete, it is customary to round the mean to
the nearest whole integer. Thus, M = 28.75 would round to M = 29.
5.8 The Mean Median and Mode in Normal and Skewed Distributions
The positions of the mean, median and mode are affected by whether a distribution is normally distributed
or skewed; and if skewed, whether the skew is positive or negative. If a distribution is normal, the value of
the mean is equal to the value of median and the values of the mean and median are equal to the value
of the mode. In a perfectly normal distribution, the data is symmetrical around the most frequent value at
the center of the distribution. Because the distribution is perfectly symmetrical around the center, 50% of
the scores must lie above the center point, which is also the mode, and the other 50% of the scores must
lie below. Thus, the mode and the median must be equal. Finally, because the distribution is perfectly
symmetrical the differences between the mean and the values larger than the mean must cancel out all of
the differences between the mean and the values less than the mean, such that the mean is also at the
exact center (i.e., the median and the mode). Thus, the mean must be equal to the median, which must
also be equal to the mode.
When the distribution is negatively skewed such that the

tail of the distribution is to the left and the hump is to the
right, the mean is less than the median which is less than
the mode. The mode is always the value at the hump of the
distribution; hence, in terms of the three measures of central
tendency it will have the largest value. The mean is
smallest, because the few extremely small scores will pull
th
the mean down. The median is always the value at the 50
percentile; hence, it will always lie between the median and
the mode.
-78-
Back to TOC
When a distribution is positively skewed such that the tail
of the distribution is to the right and the hump is to the left,
the mean is greater than the median which is greater than
the mode. Again, the mode is always the value at the
hump; hence, in terms of the three measures of central
tendency it will have the smallest value. The mean will have
the largest value, because the few extremely large scores
will pull the mean up. The median is always the value at the
th
50 percentile; hence, it will always lie between the median
and the mode.
Remember, because the median is always the

th
50 percentile and splits the distribution in half: the median
is better to use as the measure of central tendency when the distribution is severely positively skewed or
severely negatively skewed. When is a distribution skewed and when should the median be used instead
of the mean? Appendix B describes methods for measuring the skew of a distribution. However, I would
recommend reading this only after you have read through chapters 7 (variability) and 8 (standard scores),
because measuring skew requires competency in those topics. I should also note that it is possible for the
median and the mode to be equal, but the mean to be greater than or less than the mode/median value.
This is still a skewed distribution! Skew has to do with the pull of the individual values on the mean, so if
the mean differs from the median/mode, the distribution is skewed.
5.9 Box Plots

A box plot (or box-and-whisker plot) is a method of presenting the dispersion of scores in a set of data by
using five pieces of data from that distribution/: the minimum value, the first quartile, the median, the third
quartile, and the maximum value. Thus a box plot displays the location of the inter-quartile range (Q3 -
Q1), relative the highest and lowest values, and the median. Box plots can alco include the mean, but this
is optional. Below, is an example of a of the general format of a box plot:
The ‘appropriate scale’ on the y-axis is simply the dependent variable. In the graph, the rectangular box
represents the location inter-quartile range within the overall range of values recoded on the dependent
th
variable. The lower edge of the box represents where the 25 percentile (Q1) occurs, and the upper edge
th
represents where the 75 percentile (Q3) occurs. The horizontal line within the box represents the median
position and the little plus sign (+) represents the location of the mean, but the mean optional. The end
-79-
Back to TOC
points of the two lines that extend from the box represent where the minimum and maximum values occur
in the distribution.
For example, let’s say that we have two classes who took the same final exam, and I want to compare the
dispersion of the scores in each class with box plots. The relevant statistics for each class are:
Statistic Class A Class B
Minimum Value 49 50
Q1 60 58
Median 64 65
Q3 73 69
Maximum Value 82 75
The box plots are below, presented in the same graph. The red lines represent the median:
The box plots show that the size of the inter-quartile range is about the same in each class, but it is
shifted down slightly in class B. The medians are about the same for each class, and the overall range of
scores is slightly less in Class B then in Class A. Thus, the box plots are a nice way to quickly examine
the dispersion of the scores in a distribution.
If the median line is equidistant from the edges of the box (i.e., equidistant from Q1 and Q3), then the
data is not likely skewed. But, as can be seen in the box plots above, the median line is not equidistant
from the edges, which suggests that the data are skewed. If the median line is closer to the bottom edge
of the box (closer to Q1), this suggests positive skew, as can be seen in Class A. If the median line is
closer to the upper edge of the box (closer to Q3), this suggests negative skew, as can be seen in Class
B. Here’s a little more on box plots: http://www.netmba.com/statistics/plot/box/
-80-
Back to TOC
1. Identify and define the three measures of central tendency in your own words.
2. When will the mode most likely be used as a measure of central tendency?
3. Compute the median for the following scores: 7, 5, 10, 5, 5
4. Compute the median for the following scores: 4, 1, 6, 2, 11, 4
5. What does a measure of central tendency indicate regarding a distribution?
6. What does it mean that the mean is the mathematical center of a distribution?
7. Compute the mode, median, and mean for the following scores { 0, -1, 2, 1, -2, 0, -1, 0, -1, 0, 1, -1, 1, -
2, 0, -1, 2, -1, 0, -1 }
8. Students were asked to rate their general political attitude on a scale from 1 (Extremely Liberal) to 11
(Extremely Conservative). The ratings from those students are displayed below.
1 2 2 2 3 3 3 3 3 3 4 5 5 5 5 5 6 6 6 6 6 7 8 8 8 9 9 9 9 9 9 9 10 11
a. What is the mode?

b. What is (are) the median's location(s)?
c. What is the median's value?
d. What is the location of the first quartile (Q1)?
e. What is the location of the third quartile (Q3)?
f. What is the value of the first quartile?
g. What is the value of the third quartile?
h. What is the inter-quartile range?
i. What is the sum of the scores equal to?
i. What is the sample mean equal to?
j. Is this distribution skewed, and if so, in what direction?
9. Construct a box plot for the data in #8. Include a ‘+’ where the mean would be located.
10. Use the data below to answer the questions that follow. Here are the numbers of wins the New York
Yankees for each of the last 10 seasons:
Year Wins
2012 95
2011 97
2010 95
2009 103
2008 89
2007 94
2006 97
2005 95
2004 101
2003 101
a. Determine the mode.

b. Determine the median value.
c. Calculate the sample mean.
d. Is the sample of wins positively skewed, negatively skewed, or unskewed? How do you know?
-81-
Back to TOC
Yankees for each of the 10 seasons that preceded those in #10.
Year Wins
2002 103
2001 95
2000 87
1999 98
1998 114
1997 96
1996 92
1995 79
1993 88
1992 76

12. Compute the mean for the following five scores: 10, 11, 12, 13, 14. Now, generate a new set of five
scores by adding a constant of 3 to each original score. Compute the mean for the new scores. Do the
same for another set of five scores generated by subtracting a constant of 10 from each original score.
Do the same for another set of five scores generated by multiplying each original score by a constant of
2. What are the effects of adding a constant to each score, subtracting a constant from each score, and
multiplying each score by a constant in a set of scores?
13. Use the data below to answer the questions that follow.
8 9 7 10 6 9 8 9 5 9

c. Calculate the sample mean
14. Using the data in #13, add a constant of 5 to each value and then answer each of the following
questions.

15. Suppose you measured the mean time it took ten children to solve a problem and found it to be 45.28
seconds. You later discovered, however, that your timing device (a watch) was 3 seconds too slow for
each child. What was the real mean score?
16. Compute the mean for the following five scores: 20, 50, 30, 10, 40. Now generate a new set of five
scores by multiplying each original score by 3. Compute the mean for the new scores. Do the same for
another set of five scores generated by dividing each original score by a constant of 10. What are the
effects on the mean of multiplying or dividing each score in a set of scored by a constant?
17. In your own words, what are the advantages and disadvantages of using the mean a measure of
central tendency? The median? The mode?
-82-
Back to TOC
18. Which measure of central tendency is influenced by extreme scores (outliers)? Which is not
influences by outliers?
19. Take a look at these two sets of scores. In which case is the mean a poorer measure of central
tendency? Why?
Set A Set B
65 65 78 79
66 200 75 78
64 65 76 76
68 65 80 77
70 64 82 75
20. Given a set of scores where the mode is 9, the median is 15 and the mean is 21, are these scores
skewed? If so, how?
21. If the mode were 23, the mean was 20 and the median were 22, would these scored be skewed If so,
how?
22. If the mode were 20, the median were 20, and the mean were 20, would the scores be skewed? If so
how?
23. A teacher wants to change seating arrangement of students in her class in the hope it will increase
the number of comments made by students. She decides to see how many comments students make
with the current seating arrangement. A record of the number of comments made by 8 of her students
during one class period is shown below. She wants to summarize this data by computing the typical
(mean) number of comments made that day. Explain how reporting the mean number of comments could
be misleading.
Student Initials
Number of AA RF AG JG CK NK JL AW
Comments 0 5 2 22 3 2 1 2
24. Explain how it could happen that if one person moves from city A to city B, it is possible for the
average (mean) IQ in both cities to increase.
25. A test to measure aggressive tendencies was given to a group of teenage boys who were members of
a street gang. The test is scored from 10 to 60, with a high score indicating more aggression. The
histogram represents the results for these 28 boys. Which do you think will be larger, the mean or the
median score? Or do you think they will be similar? Explain.
8
6
Frequency
0
12.5 17.5 22.5 27.5 32.5 37.5 42.5 47.5
Agressiveness Score
-83-
Back to TOC
Chapter 6: Variability
6.1 What is Variability? Why is
Measured?
Look at the picture to the right. The variety of
colors, shapes, sizes, flavors, and—importantly—
spiciness of those chili peppers makes me hungry.
Variability is simply variety (differences) among
items. Variability is all around us all the time. There
is variability in the eye color, hair color, height, the
weight, and sex among students in your classes.
There is probably variability across time in the
background noise of where you are right now.
There is even variability in the number of hairs on
your head across the days of the week (we gain
some, we lose some). This is variability, which is critical for understanding statistics.
Statistically, if measures of central tendency (mean, median, and mode) measure where a distribution
tends to fall along a continuum, then variability measures the dispersion and similarity among the scores
in a distribution. Thus, measures of variability tell us something about the variety of the scores.
Why do we measure variability? Recall, that the mean takes into account all scores in a distribution. Even
if you know where the mean lies, you still do not know anything about the actual scores. That is, if I told
you that the mean of a set of n = 10 scores on a quiz was M = 7, but I told you nothing else, could you tell
me anything about those 10 scores, that is, how alike or different those scores were? The answer is no.
Even though the mean accounts for all of the scores in a distribution, knowing the mean does not tell you
anything about the individual scores.
For example, if the mean for a set of ten scores on a statistics quiz is 7, those ten scores could be:
7 + 7 + 7 + 7 + 7 + 7 + 7 + 7 + 7 + 7 = 70/10 = 7, or
6 + 6 + 6 + 6 + 6 + 8 + 8 + 8 + 8 + 8 = 70/10 = 7, or
1 + 1 + 5 + 5 + 9 + 9 + 10 + 10+ 10+ 10 = 70/10 = 7,
...or any other combination where 10 scores adds to 70 (assuming the range is 0 – 10). Thus, even
though the mean takes individual scores into account, it does not measure or account for the variability of
the scores. To know anything about the individual scores in a data set we need a measure of variability.
Statistically, variability measures how alike or how different individual scores are in a set of data. If a
measure of variability is small (approaching zero), it indicates low variability, whereas if a measure of
variability is large (approaching infinity), it indicates large variability. A measure of zero indicates no
variability among scores and all of the scores have the same value (i.e., a rectangular distribution).
There are several measures of variability. In the following sections I start with the least complex measure
and move to the more complex measures. Each successive measure has all the characteristics as the
less-complex measures plus something else. Measures of variability include the range, average
deviation, sum of squares, the variance, and the standard deviation.
-84-
Back to TOC
6.2 Range
For each of the following sections I will calculate each measure of variability on both of the following sets
of n = 10 scores. Note that in both sets the sum of scores and the mean are identical:
Set I: 2 2 2 4 5 5 6 6 8 10 ∑X = 50 X5
Set II: 4 4 5 5 5 5 5 5 6 6 ∑X = 50 X5
The range is the largest value minus the smallest value. For Set I the range is 10 – 2 = 8, and the range
for Set II is 6 – 4 = 2.
The range provides information only about the overall dispersion of scores, that is, the area a distribution
covers. A large range indicates a large spread of scores compared to a small range; however, the range
does not say anything about individual values. For example, the following sets of n = 5 scores have the
same range (range = 9), but different underlying distributions:
Set A: 1 1 1 1 10 Set B: 1 2 4 5 10
Clearly there are more differences among scores in Set B than Set A, but the range does not account for
this variability. What we need is some way to measure the variability among the scores.
6.3 Average Deviation (AD)

The best way to measure variability among scores is to compare each score to some fixed value and
examine the deviations between the scores and that fixed value. What this does is measure the general
variability among raw scores.
The question is what fixed value should the individual scores be compared? One option is to pick one of
the scores from the data and use that as the fixed value, but the question then becomes which score to
use, which leads to more questions and more confusion. Another choice is to use one of the measures of
central tendency, each of which represents a fixed value (a foxed point) in a distribution. The question is
which measure of central tendency do we use? I’ll make it simple and state that the mean is the best
option, because the mean takes into account all of the individual scores in a data set and because we are
trying to now measure the variability among those scores.
Calculating the average deviation requires each score to be mean centered. This involves subtracting
the mean from each score in the distribution. Recall, the mean for each set of data above is 5. The tables
below derives the deviations scores for data Sets I and II, below:
Set 1 Set 2
X X  X  X X  X 
10 5 6 1
8 3 6 1
6 1 5 0
6 1 5 0
5 0 5 0
5 0 5 0
4 -1 5 0
2 -3 5 0
2 -3 4 -1
2 -3 4 -1
-85-
Back to TOC
The difference scores, or deviation scores, “capture” the variability among the scores relative to the
mean. Thus, by mean centering each score we have begun to measure variability. Of course, the
deviation scores are not a single measure of variability so we need to do something with the deviation
scores.
The average deviation score is the sum of the deviation scores divided by n:
AD 
X  X 
n
For Set I, the sum of the deviation scores is 0; thus, the average deviation will be 0. For Set II, the sum of
the deviation scores is also 0; thus, the average deviation will also be 0. This will always be the case!
That is, the sum of the deviation scores will always be equal to zero; thus, the average deviation score
will also always be equal zero. Thus, although mean centering each score captures variability among the
scores in the data, when you sum the deviation scores the outcome will always be zero. This is because
the mean is the balancing point of the distribution and positive deviations from the mean will cancel out
negative deviations from the mean and vice versa. What do we do?
To measure variability we need to capture the variability among individual scores, which we accomplished
by mean centering the scores. But, we need to make sure that adding the mean centered values does not
result in sum of zero (unless there is zero variability). The best way to do this would be to get rid of the
negative deviation scores, which cancel out the positive scores. How do we do this? Remember from the
math review chapter, squaring a negative value results in a positive value. Thus, the solution to getting rid
of negative deviations is to square each deviation score including the positive deviation scores. This
converts negative deviations into positive values and leave the positive deviations positive. Doing this is
the first step to calculating the sum of squares.
6.4 Sum of Squares

The sum of squares (SS) is defined as the sum of the squared deviation scores from the mean. It is
good to keep this in mind, because as you will see, the way in which the sum of squares is calculated
reflects this definition. (Remember, statistics is a language course!!) Here is the formula for the sum of
squares:
SS   X  X 
2
The sum of squares measures the total variation in a set of data; that is, the summed variability among all
the scores. Note that the sum of squares can be equal to zero if there is no variability and all the scores
are equal (rectangular distribution); but, the sum of squares can never be negative. If you ever calculate a
negative sum of squares go back and check work; it is mathematically impossible to obtain a negative
value for sum of squares. Set 1 Set 2
Calculating the sum of squares is X X  X  X  X 
2
X X  X  X  X  2
simple. After mean centering 10 5 25 6 1 1

each score, simply square each 8 3 9 6 1 1
deviation, and then add those 6 1 1 5 0 0
squared deviations. This is done 6 1 1 5 0 0
for Sets I and II to the right. 5 0 0 5 0 0
5 0 0 5 0 0
The sum of squares for Set 1 is 4 -1 1 5 0 0
SS = 64 and for Set 2 is SS = 4. 2 -3 9 5 0 0
Larger sum of squares indicates 2 -3 9 4 -1 1
there is more variability among 2 -3 9 4 -1 1
the scores in Set I. Notice that the SS = 64 SS = 4
-86-
Back to TOC
definition of the sum of squares is reflected in the formula. Recall, the definition of the sum of squares is
the sum of the squared deviations from a mean. The procedure above had you find the sum (Σ) of the
2
squared ( ) deviations from a mean X  X  . Like I said, statistics is a language course!
Sum of squares measures the total variation among scores in a distribution; thus, sum of squares does
not measure average variability, which is what we really want. That is, we want a measure of variability
that takes into account bot the variation and number of scores in a distribution; hence, we want a
measure of average variability of scores in a distribution, which is the variance.
6.5 Sample Variance (s2)

2
Sample variance (s ) is the average sum of the squared deviation scores from a mean, or simple the
average sum of squares. Sample variance measures the average variability among the scores in a
distribution; that is, by how much do scores in a distribution differ on average? The formula for sample
variance is:
s2 

 XX 
2
Calculating sample variance is simple. First, calculate the sum of squares as we did in Section 6.4, then
divide the sum of squares by the number of scores in the data set (n). Hence, once you calculate the sum
of squares, finding the sample variance is as simple as dividing SS by n:
SS
s2 
n
From the data in Section 6.4 we have:
Set I: SS = 64, n = 10 Set II: SS = 4, n = 10
Thus, sample variance for each data set is:
SS 64 SS 4
Set I: s 2    6.4 Set II: s 2    0.4
n 10 n 10
Each value reflects the average variability between the scores in
each set of data. Larger values indicate greater variability. Indeed, X X  X  X  X  2
you can see this by comparing Sets I and II; there is more variability 2
10 cm 5 cm 25 cm
and larger deviations between the scores in Set I compared to Set 2
8 cm 3 cm 9 cm
II. 2
6 cm 1 cm 1 cm
2
6 cm 1 cm 1 cm
The one issue with sample variance is it is in squared units, not the 2
5 cm 0 cm 0 cm
original unit of measurement. Anytime you square a value you 2
5 cm 0 cm 0 cm
change the original unit of measure into squared measurement 2
4 cm -1 cm 1 cm
units. This is more easily understood if you think of the scores in Set 2
2 cm -3 cm 9 cm
I as measurements of length in centimeters (cm). If the original 2
2 cm -3 cm 9 cm
measurements in Sets 1 and 2 were of length in centimeters, then 2
2 2 cm -3 cm 9 cm
when the values were squared the measurement units are now cm ,
-87-
Back to TOC
2
which represents measurements of area. In the table below, I have included cm and cm extensions for
each datum in Set 1 to exemplify this problem.
What is the solution to the issue of squaring both the value and the unit of measure; how do we get the
measurement units back from a squared unit into the original unit of measurement? If you guessed take
the square root of the variance, you are correct. We call this the standard deviation.
6.6 Sample Standard Deviation (s)

Once sample variance has been determined, calculating the sample standard deviation (s) is simple.
The standard deviation is the positive square root of the variance. Taking the square root of the sample
variance, which is in squared units, means that standard deviation is in the original unit of measurement.
The sample standard deviation is defined as the square root of the average sum of squared deviations
from a mean; or simply, the square root of the variance. The formula for the sample standard deviation is:
s

 XX 
2
This is equivalent to:

SS
s
n
This is equivalent to:

s  s2
In each formula, all you are doing is taking the square root of the sample variance. From the data in
Sections 6.4 and 6.5, we have:
2 2
Set I: SS = 64, n = 10, s = 6.4 Set II: SS = 4, n = 10, s = 0.4
The sample standard deviations in Sets I and II are:
s  s 2  6.4  2.53 s  s 2  0.4  0.632
In APA format, instead of s the standard deviation is labeled SD. Thus, for Set I SD = 2.53 and for Set II
SD = 0.4.
The standard deviation is measures the average deviation from the mean of a distribution; that is, the
average difference between the scores that make up the distribution and the mean of that distribution. For
example, in Set I, a score is expected to deviate from the mean by 2.530 and a score is expected to
deviate from the mean by 0.632 in Set II.
This does not mean you will find scores in the data set that deviate from the mean by exactly these
amounts; the standard deviation is an average value. Because the standard deviation measures the
variability of scores in the original unit of measure it is our best measure of variability. Note that the
standard deviation takes into account all the scores in a distribution, because it is based on the variance
and the sum of squares.
-88-
Back to TOC
To recap Sections 6.3 to this point; when calculating measure of variability:
Calculate the mean of a data set

Mean center each score to create deviation scores
Square each deviation score
Sum the squared deviation scores to give you the sum of squares (total variation)
Divide sum of squares by number of scores to get sample variance (average variation)
Take the square root of variance to get the standard deviation (average deviation from the mean)
The following sections discuss alternative methods for calculating measures of variability and how each
measure of variability is calculated in a population, rather than a sample.
6.7 The Population Formulas for Variability

Calculating sum of squares in a population is no different than calculating sum of squares in a sample.
The only thing that differs is the symbols used. The formula for sum of squares in a population is:
SS   X  
2
This is functionally equivalent to the formula for the sample sum of squares in Section 6.4, the only
difference is that here you subtract a population mean (μ) from each score. Thus, the only difference
between these formulas is working with sample data versus population data.
2
The formulas for the population variance (σ , ‘little sigma squared’) and the population standard
deviation (σ, ‘little-sigma’) are functionally equivalent to the formulas for calculating the sample variance
2
(s ) and the sample standard deviation (s), respectively.
The formulas for population variance are:
 X   
2
SS
2  2 
N N
The formulas for population standard deviation are:
 X   
2
SS
    2
N N
The important point to remember is that the concepts sum of squares, variance, and standard deviation
do not change whether you are working with a sample or with a population; and the same is true for
measures of central tendency. Calculating a measure of variability in a sample is just the same as
calculating a measure of central tendency in a population. The procedures do not change, just the
symbols. Too many times I have seen students get hung up on the symbols used in population vs.
samples and think that the different formula measure different things. In one sense, such students are
right: One formula measures a population parameter and the other measures a sample statistic; however,
the concepts of variance, standard deviation, mean, median, etc., and the methods for calculating them
are the same in a sample and a population. The point is to not get caught up in the different symbols.
Concentrate on what the concepts mean, the language of statistics, and what each measure reflects, and
you’ll be fine.
-89-
Back to TOC
6.8 Scaling Scores
There may come a time when for whatever reason you need to change each score in a data set by a
constant; that is, you may have to add, subtract, multiply, or divide each score by a certain amount. When
such a need arises, as long as you perform the same operation to each score it is called scaling. (There
are many more scaling methods than I will briefly describe, below.)
There are few interesting things that changing each score in the data set by a constant will do to the
original mean and standard deviation. That is, if you change each score in a data set by a constant and in
the same manner, you will not have to completely recalculate the mean and standard deviation.
If you add or subtract a constant to each score in a set of data, the mean of that scaled set of data will
increase or decrease by that constant, respectively, from the original mean, while the standard deviation
does not change. For example, if you add X = 2 to each score from Set I (see table below), the original
mean of 5 will increases by 2 (new mean = 7). If you subtract 2 from each score the original mean
decreases by 2 to 3. In both cases, the standard deviation does not change.
If you multiply each score by a constant, the new mean is equal to the original mean multiplied by that
constant and the new standard deviation will be equal to the original standard deviation multiplied by that
constant. (The resulting variance is equal to the original variance multiplied by the square of that
constant.) For example if you multiply each score from Set I by 2, the original mean (5) is now equal to 10
2
and the original standard deviation (2.53) increases to 5.06. (The original variance, s = 6.4, increases to
2 2
s = 25.6 (6.4 * 2 = 6.4 * 4 = 25.6).
Finally, if you divide each score in a data set by a constant, the new mean will be equal to the original
mean divided by that constant, and the new standard deviation will be equal to the original standard
deviation divided by that constant. (The resulting variance will be equal to the original variance divided by
the square of that constant.) For example if you divide each score from Set I by 2, the original mean (5) is
now equal to 2.5, and the original standard deviation (2.53) is now equal to 1.265. (The original variance
2
is now 6.4/2 = 6.4/4 = 1.6). You can check all of this using the original data from Set I and the scaled Set
I scores below if you want. (See table on next page.)
X X+2 X-2 X*2 X/2

10 12 8 20 5
8 10 6 16 4
6 8 4 12 3
6 8 4 12 3
5 7 3 10 2.5
5 7 3 10 2.5
4 6 2 8 2
2 4 0 4 1
2 4 0 4 1
2 4 0 4 1
∑X = 50 ∑X = 70 ∑X = 30 ∑X = 100 ∑X = 25
X5 X7 X3 X  10 X  2.5
2 2 2 2 2
s = 6.4 s = 6.4 s = 6.4 s = 25.6 s = 1.60
s = 2.530 s = 2.530 s = 2.530 s = 5.060 s = 1.265
-90-
Back to TOC
1. Identify and define the three main measures of variability:
2. What is the problem with using the sum of squares as a measure of variability, that is, why is variance
a better measure?
3. What is the standard deviation our best measure of variability?
4. If the variance of a set of scores is 10000.00 what is the standard deviation?
5. If the standard deviation of a set of scores is 9.00 what is the variance?
6. Without calculating it, what must the variance of the following scores equal: 4, 4, 4, 4? What must the
standard deviation equal? What must the sum of squares equal? Why?
7. The following scores were obtained on a quiz: { 3 5 5 6 7 7 7 8 8 9 }
a. Calculate the mean of the scores.

b. Calculate the sum of squares.
c. Calculate the sample variance.
d. Calculate the standard deviation.
8. The following scores were obtained on another quiz { 3 4 4 6 7 7 8 8 9 9}
a. Calculate the mean of the scores.

d. Calculate the standard deviation.
9. For the sets of data in exercises 7 and 8, what descriptive statistics are the same (if any) and which are
different (if any)? What does this tell you about the measures of central tendency and the measures of
variance?
Yankees for each of the last 10 seasons (the data come from problem #10 in Chapter 5):
Year Wins
2012 95
2011 97
2010 95
2009 103
2008 89
2007 94
2006 97
2005 95
2004 101
2003 101
a. Calculate the sum of squares.

b. Calculate the sample variance.
c. Calculate the sample standard deviation
d. Calculate the estimate of the population variance.
e. Calculate the estimate of the population standard deviation.
-91-
Back to TOC
Yankees for each of the 10 seasons that preceded those in #10 (the data come from problem #11 in
Chapter 5):
Year Wins
2002 103
2001 95
2000 87
1999 98
1998 114
1997 96
1996 92
1995 79
1993 88
1992 76

c. Calculate the sample standard deviation.
d. Calculate the estimate of the population variance.
e. Calculate the estimate of the population standard deviation.
13. Use the data below to answer the questions that follow (the data are from problem #13 in Chapter 5).
8 9 7 10 6 9 8 9 5 9

14. Using the data in #13, add a constant of 5 to each value and then answer each of the following
questions (the data are from problem #14 in Chapter 5).

15. Using the data in #13, multiple each value by a constant of 2 and then answer each of the following
questions.
a. Calculate the sample mean.

d. Calculate the sample standard deviation.
16. Compute the variance and standard deviation for the following eight scores: { 6, 10, 8, 8, 10, 8, 6, 8 }.
Now, generate a new set of five scores by adding a constant of 3 to each original score. Compute the
variance and standard deviation for the new scores. What are the effects of adding a constant on the
variance and on the standard deviation?
17. How does the standard deviation help you interpret a set of data?
-92-
Back to TOC
Use the following to answer #18 – 21: A political scientist studied how satisfied people were with the
President’s job performance and Congress’ job performance. Each person was given a job performance
satisfaction test on which scores ranged from 1 to 7, with higher scores indicating more satisfaction. The
scores were:
Individual President’s Job Performance Congress Job Performance

A 2 1
B 5 3
C 6 3
D 7 5
E 5 3
18. Compute the mean of the satisfaction ratings for the President’s and Congress’ job performance.
19. Compute the standard deviation for the satisfaction ratings for the President’s and Congress’ job
performance.
Individual President’s Job Performance Congress Job Performance

2 2
X (X – M) (X – M) X (X – M) (X – M)
A 2 -3 9 1 -2 4
B 5 0 0 3 0 0
C 6 1 1 3 0 0
D 7 2 4 5 2 4
E 5 0 0 3 0 0
20. Based on the results, whose job performance are people more satisfied with? Why?
21. Based on the results, which job performance ratings are more consistent? Why?
22. Suppose you measured the weights of 100 people and found a mean of 150 with standard deviation
11. Then you learned that your scale was 2 pounds to heavy. What were the correct mean and standard
deviation?
23. How accurate are people ate estimating length? In an experiment, people are presented with a length
that was 25-cm in length for five seconds and asked to estimate the length of the rope. Ten people gave
the following estimates (in cm):
16 17 33 31 25 22 30 34 20 32
Compute the mean and standard deviation for these data. How accurate are the estimated considering
the mean score across all participants? How does the standard deviation help to interpret the mean?
24. Following up on #23, in a second experiment, people are presented with the same 25-cm piece of
rope, but for one minute before making their estimates of the ropes length. Ten people gave the following
estimates (in cm):
23 27 25 23 26 24 26 22 23 26
Compute the mean and standard deviation for these data. How accurate are the estimated considering
the mean score across all participants? How does the standard deviation help to interpret the mean?
-93-
Back to TOC
25. An organizational psychologist studied how satisfied employees were in two different companies. All
employees were given a job satisfaction test on which scores could range from 1 to 7, with higher scores
indicating greater satisfaction. The scores were as follows:
Company A: 4 6 5 4 3 2
Company B 4 4 4 4 4 4
Compute the mean and standard deviation for each company. Based on the results, compare employee
satisfaction in the two companies. How do the standard deviations help to interpret the means?
26. Suppose scores on a math achievement test approximate a normal distribution with M = 500, min =
350, and max = 650. Estimate the standard deviation of this distribution.
27. Two students who took a statistics class received the following Quiz scores (each Quiz is out of 100
points): Student A: 60, 90, 80, 60, 80 and Student B: 40, 100, 100, 40, 90. If you had an upcoming
statistics test, who would you have rather had as a study partner, A or B? Support your answer.
28. Consider the two histograms displayed below. The histogram labeled X1 has a mean of 54 and the
histogram labeled X2 has a mean of 53. Please indicate which one has a larger standard deviation and
WHY that histogram has the larger standard deviation. (It is not necessary to do any calculations.)
X1 X2
6 6
5 5
4 4
Frequency
Frequency
3 3
2 2
1 1
0 0
53 54 55 56 57 58 45 50 55 60 65 70
Score Score
-94-
Back to TOC
Chapter 7: Standard Scores and Normal Distribution
7.1 Comparing Scores across Distributions
The picture to the right does a great job of introducing the
concept of a standard score and why comparing across
distributions is tricky. The gentleman (if he can be called
such) was under the impression the numbers that the lady
gave him meant something other than what she meant the
numbers to imply. Thus, what the gentleman interpreted as a
‘42’ was not what the lady meant as ‘42’. This cartoon
demonstrates one of the important applications of statistics:
when comparing data across different distributions, you
cannot rely on raw scores. Rather, raw data must be
converted into a “standard language” that is understandable
in any and all distributions. By converting raw data into some
standard lingo, interpretations of raw data will not get lost in
translation.
To understand why comparing raw scores across

distributions is problematic; assume that you have a friend,
Stu, who thinks he is smarter than everybody. You both take
th
the same 19 Century American Literature course from the same professor, in the same semester, but in
different course sections. You have a test in that course on the same day. On the test, Stu obtains 85 out
of 100 possible points and you obtain 80 out of the possible 100 points. Of course, Stu starts bragging
and boasting about his 19th Century American Literature intelligence. But who really did better on this
test?
It is true that your raw score was less than Stu’s raw score, but what happens when you take additional,
distribution information into account? Say that each test was on the same material, but the version of
each test was slightly different. Indeed, maybe the classes themselves were a little different, because
there are different students in each section and the professor may change lessons plans between those
two sections. So, assume your class’ mean grade on the test was M = 75 and, the Stu’s class mean was
an M = 80. Now who did better?
When you take into account the mean of each class, it seems you and Stu performed equally, because
your scores were each 5 points above your respective class average. However, because the means of
each class are different, this may suggest that the underlying distributions in each class are different! This
suggests that a 5 point difference in one class may be different than a 5 point difference in the other
class. That is, the 5 point difference must be considered relative to the class distribution.
So, who did better, you or Stu? Well, you cannot compare your raw scores, because the different means
suggest different distributions. What you need is a “standard measure” that can be used to compare the
raw scores across the two class distribution.
Fortunately, we already know of such a measure: the standard deviation. Remember, the standard
deviation is the average deviation between a score and the mean of a distribution. Although the value of
the standard deviation will change depending on the data in the distribution, a standard deviation means
the same thing in any distribution. That is one standard deviation away from the mean (1 SD or 1σ)
means the same thing regardless of the underlying distribution. Similarly, 2 SD 2σ) indicates that a score
is two standard deviations from the mean regardless of the underlying distribution; 1.5 SD (1.5σ) means
that a score is one and one-half standard deviations from the mean; etc. Thus, the standard deviation can
be used to measure the distance between any score and the mean of a distribution and that distance can
be used to compare raw scores across different distributions.
-95-
Back to TOC
Basically, if you measure the number of standard deviations between your raw score and the mean of
your class, you can compare that distance to the number of standard deviations between Stu’s raw test
score and his class’ mean. The question is how do you do this? The measure of the number of standard
deviations between any raw score and the mean of a distribution is a standard score, or z-Score.
7.2 Standard (z) Scores

A standard score (or z-Score) measures the number of standard deviations between a raw score and
the mean in standard deviations. All that is required to calculate the standard score for a given raw score
is the mean of the distribution, the standard deviation of that distribution, and a raw score. The formula for
calculating a standard score in a population is:
X
z

The formula for calculating a standard score in a sample is:
XX
z
s
Each formula has the same symbol (z) and both formulas are functionally equivalent: you subtract the
mean of a distribution from a raw score and then divide that difference by the standard deviation of the
distribution. The ’z’ comes from a standard scores when it is calculated from a raw score in a normal
distribution.
Let’s calculate the standard scores for you and Stu. Assume that the standard deviation for your class is s
= 1, and the standard deviation for Stu’s class is s = 5. Remember, your score was X = 80 and Stu’s
score was X = 85. Finally, your class' mean was 75, and Stewart’s class mean was 80. Thus, we have:
X  X 80  75 5 X  X 85  80 5
You: z    5 Stu: z    1
s 1 1 s 5 5
The difference between each raw score and the mean of each class is identical (Difference = 5), but Stu’s
standard score is less than yours (z = 1 vs. z = 5). The reason has to do with the variability (measured by
the standard deviation) of each distribution of the test scores in each class. The standard deviation was
smaller in your class than in Stu’s class. The smaller standard deviation suggests there was less
variability in your class compared to Stu’s class; and less variability indicates that the scores in your class
were, on average, closer to the mean compared to Stu’s class. Stated differently, the higher standard
deviation in Stu’s class suggests a greater dispersion of scores around the mean of that class. Because
most scores in your class were very similar to the mean, deviations from your class’ mean are unlikely. In
contrast, because there is a lot of variability in Stu’s class, deviations from the mean are not surprising.
Standard scores can be used to compare test performance across the two class distributions. Because
your score is farther away from the mean than Stu’s score, in standard deviations, you did better on the
test. This is so regardless of the different distributions. That is, standard scores provide a way to compare
across different distributions. Because raw scores can mean different things in different distributions of
data, it is more appropriate to convert raw scores into standard (z) scores and then compare the standard
scores to make inferences about where a score in one distribution lies in another.
7.3 Characteristics of Standard (z) Scores

There are several noteworthy characteristics of standard scores. First, standard scores can be used to
compare performance across distributions, as was done in the example in the preceding section. Again,
-96-
Back to TOC
when comparing across distributions, unless the mean and standard deviation are identical across
distributions or if the raw scores come from the same distribution, it is never good to compare raw scores.
Instead, you should be comparing raw scores.
Second, larger standard scores indicate larger deviations (distances) from the mean. Thus, the larger the
standard score, the farther the associated raw score is from the mean.
Third, the definition of a standard score is the distance between a raw score and its mean, in standard
deviations. Thus, your z-Score from Section 7.2 of z = 5 indicates that your raw score of 80 is five
standard deviations above your class mean of 75. Stu’s z-score of z = 1 indicates that his raw score of 85
is one standard deviation above his class’ mean of 80. Similarly, a z-Score of z = -1.5 indicates the
associated raw score is one and one-half standard deviations below the mean.
Fourth, standard scores can be positive or negative. Positive z-Scores indicate the associated raw score
was greater than the mean and negative z-Scores indicate the associated raw score was less than the
mean. Importantly, the sign (+/-) does not indicate anything about the magnitude of the z-Score: A z-
Score of -2 is not smaller in magnitude than a z-Score of +1. Rather, the magnitude of z = -2 is twice the
magnitude of z = +1, because z = -2 is two standard deviations away from the mean and z = +1 is only
one standard deviation away from the mean. Thus, the sign (+/-) of a z-Score indicates the direction a raw
score lies from the mean; the absolute value indicates magnitude.
Finally, if you calculate the mean, variance and standard deviation of a distribution of z-Scores that were
calculated from a set of raw scores, the mean of the z-distribution will be zero and both the variance and
standard deviation will be equal to one. Proof is provided in the data set below:
X X  X  X  X 2
z z  z  z  z 2
4 1 1 1.118 1.118 1.25

4 1 1 1.118 1.118 1.25
3 0 0 0 0.000 0
2 -1 1 -1.118 -1.118 1.25
2 -1 1 -1.118 -1.118 1.25
∑X=15 SS = 4 ∑z = 0 SSZ = 5
X3
2
s = 0.8 z0 s 2z  1
s = 0.894 sz  1
7.4 Using the Mean, Standard Deviation and z-Scores to find Raw Scores
So you can use the standard deviation to determine the distance between a raw score and the mean of a
distribution, but what if you want to determine the raw score that is a certain number of standard
th
deviations above or below the mean? That is, say we have a distribution of 19 Century American
Literature test scores with a mean of 70 and standard deviation of 10. Say that any test score that is 2.5
standard deviations above the mean earns an A; thus, the minimum score for an A is 2.5 standard
deviations above the mean. We want to know the raw score that is exactly 2.5 SD above the mean. How
can we do this?
First, remember a z-Score measures the number of standard deviations that a raw score is above/below a
mean. In the example above we are looking for the raw score associated with z = 2.5. Second, to
calculate this raw score we use one of the two following equations, both of which are used to determine
the raw score associated with a z-Score given that we know the mean and standard deviation (s or σ). If
calculating a raw score from a z-score in a population use:
X    z
-97-
Back to TOC
If calculating a raw score from a z-score in a sample use: X  X  zs
Both formulas are functionally equivalent, because both formulas determine a raw score (X) by adding the
mean of a distribution to the product of a z-score and standard deviation of that distribution. The only
difference is that for one formula you are using sample data and the other you are using population data
(Remember, don't get bogged down in symbols!)
From the example above, what is the minimum raw score that is needed for an A in this course, that is,
what raw score is 2.5 standard deviations above the mean? To determine that raw score insert the mean
(70), standard deviation (10), and z-Score (2.5) into either formula from above:
X  X  zs
X  70  2.510
X  70  25
X  95
Thus, a raw score of 95 is exactly 2.5 standard deviations above the mean for this distribution. Thus, if
the minimum for an A is 2.5 standard deviations above the mean, then you need a test grade of at least
95 to get an A on this test.
What about the raw score that lies 1.25 standard deviations below the mean (z = -1.25)? Maybe this is
the minimum grade that is needed to pass the course. Determining this raw score is the same as before,
you just need to be sure to take the negative z-score into account:
X  X  zs
X  70   1.2510
X  70  12.5
X  57.5
Thus, a raw score of 57.5 is exactly 1.25 standard deviations below the mean of 70.
7.5 Standard Scores and the Normal Distribution

When you have a population or sufficiently large sample of normally distributed data, a certain percentage
(proportion) of scores will fall between the mean and a given number of standard deviations from the
mean. Specifically, between the mean (μ) and one standard deviation from the mean, the proportion of
scores in a normal distribution is approximately 0.3413; and this is true whether you are one standard
deviation above the mean or below the mean. Between one standard deviation and two standard
deviations from the mean the proportion of scores in a normal distribution is approximately 0.1359; and
this is true whether you are above the mean or below the mean. Also, the proportion of scores that lie
between one standard deviation below the mean and one standard deviation above the mean is
approximately 0.6826; hence, most scores in a normally distributed set of data fall right around the mean.
These approximate proportions (or probabilities) can be seen in the graph of the normal distribution
below. I should note that this is a prototypical normal distribution that is centers on some mean and is
symmetrical with half of the scores lying above the mean and half lying below the mean. Remember, in a
normal distribution, most scores lie at the "hump" in the middle, and fewer scores lie in the tails.
-98-
Back to TOC
The height of the curve above any point in the x-axis reflects two interrelated ideas: (1) the probability of
observing a score at a particular location along the x-axis; (2) the proportion of scores in the distribution
that are associated with a score at a particular location along the x-axis. The higher the height of the
curve, the more probable that score is likely to be observed and the greater the proportion of the
distribution is made up of that score. As you can see in the graph above, as the distance between a score
on the x-axis and the mean increases, the probability of that score being observed decreases; that is, the
height of the curve decreases as the distance between the mean and a score increases. Thus, scores
that extremely deviate from the mean are less likely to be observed.
Along the x-axis, marks are indicated for the number of Standard Deviations from The Mean, from four
standard deviations below the mean (-4σ) to four standard deviations above the mean (+4σ). Also, marks
are indicated for the Z Scores relative to the mean from z = -4.0 to z = +4.0. (Don’t worry about what the
Cumulative % and the T Scores mean.) As you can see, as you move farther and farther away from the
mean in standard deviations, or z-Scores, the probability of bring that far away become less and less.
This fact about the normal distribution introduces two key points:
First, because the proportions/probabilities in the figure above are associated with specific numbers of
standard deviations from the mean, and because standard deviations are a measure that can be used in
any normally distributed set of data, these proportions/probabilities would be found in any normally
distributed set of data. For example, IQ scores are normally distributed with μ = 100 and σ = 16. This
means that the proportion of scores between an IQ score of 84 (-1σ) and 116 (+1σ) is approximately
0.6826. If we assume scores on the Beck Depression Inventory (BDI) are normally distributed with μ = 15
and σ = 5, then the proportion of scores between a BDI score of 10 (-1σ) and 20 (+1σ) is also
approximately 0.6826. Thus, regardless of the data that is being collected, as long as the data is normally
distributed, these proportions/probabilities will be found.
-99-
Back to TOC
Second, by taking into account the mean and standard deviation of a distribution, we can calculate a z-
score for any raw score and we can use the information about the normal distribution above to determine
proportions and probabilities associated with that z-score.
7.6 Probabilities and Proportions in the Normal Distribution

Table 1 in Appendix A presents the is the Standard Normal (z) Table, which can be used to look up
proportions of scores above or below z-Score within a normal distribution. A portion of Table 1 is
presented below. There are five columns in Table 1: The first column lists z-Scores that increase from
0.00 to 4.00, mainly in increments of 0.01, and Columns 2 through 5 list proportions of scores under the
normal distribution curve that are associated with a particular z-Score. We'll get to these shortly.
Column 1 Column 2 Column 3 Column 4 Column 5

Area between Area in either Area in Area between μ
+z to -z one tail both tails and + z or -z
p (x ≥ +z) or p (0 < x ≤ +z) or

z p (-z ≤ x ≥ +z) p (x ≤ -z or x ≥ +z)
p (x ≤ -z) p (0 > x ≥ -z)
1.15 0.7499 0.1251 0.2501 0.3749
1.16 0.7540 0.1230 0.2460 0.3770
1.17 0.7580 0.1210 0.2420 0.3790
1.18 0.7620 0.1190 0.2380 0.3810
1.19 0.7660 0.1170 0.2340 0.3830
1.20 0.7699 0.1151 0.2301 0.3849
1.21 0.7737 0.1131 0.2263 0.3869
1.22 0.7775 0.1112 0.2225 0.3888
1.23 0.7813 0.1093 0.2187 0.3907
1.24 0.7850 0.1075 0.2150 0.3925
1.25 0.7887 0.1056 0.2113 0.3944
Notice there are no negative z-Scores. This is because the sign (+/-) of a z-Score does not tell you
anything its magnitude; the sign only tells you the direction of the z-score relative to the mean. Thus, z =
1.25 is equal in magnitude to z = -1.25 and proportions under the normal distribution curve that are
associated with a positive z-Score will be the same as the proportions under the normal distribution curve
that are associated with a negative z-Score of the same absolute value. If you have a negative z-Score
just look up a positive z-Score of the same absolute value.
Before going into detail about what information each column provides, let me start with an example. Say
th
we have a normally distributed set of 19 Century American Literature test scores with μ = 80 and σ =
10. We calculate the z-Score for the raw score of X = 92:
X   92  80 12
z    1.2
 10 10
We want to know the proportion of scores that lie between z = 1.2 and the mean of the distribution.
Another way to conceptualize this question is that we want to know the probability of obtaining a score
that is between the mean and z = 1.20. Thus, we want to know p(μ ≤ X ≤ 92).
The same section of Table 1 above is reproduced below. To determine this probability, first look up the z-
Score of z = 1.20 in Column 1, which is highlighted in yellow:
-100-
Back to TOC
Area between Area in either Area in Area between μ
+z to -z one tail both tails and + z or -z
p (x ≥ +z) or p (0 < x ≤ +z) or

z p (-z ≤ x ≥ +z) p (x ≤ -z or x ≥ +z)
p (x ≤ -z) p (0 > x ≥ -z)
1.15 0.7499 0.1251 0.2501 0.3749
1.16 0.7540 0.1230 0.2460 0.3770
1.17 0.7580 0.1210 0.2420 0.3790
1.18 0.7620 0.1190 0.2380 0.3810
1.19 0.7660 0.1170 0.2340 0.3830
1.20 0.7699 0.1151 0.2301 0.3849
1.21 0.7737 0.1131 0.2263 0.3869
1.22 0.7775 0.1112 0.2225 0.3888
1.23 0.7813 0.1093 0.2187 0.3907
1.24 0.7850 0.1075 0.2150 0.3925
1.25 0.7887 0.1056 0.2113 0.3944
Column 5 lists the proportion of scores in a normal distribution that fall between the mean a z-Score.
Thus, the proportion of scores between the mean and z = 1.2 in is 0.3849. This would be the same
proportion of scores that lie between the mean and z = -1.2, because the proportion of scores between
the mean and a positive z-Score is equal to the proportion of scores between the mean and a negative z-
Score of the same magnitude.
We can also use this table to determine the proportion of scores beyond a given z-Score, that is, the
proportion of scores that fall in the tail of a normally distributed set of data. Another way to conceptualize
this question is what is the probability of having score greater than z = 1.2? Thus, we want to know p(1.2
≤ X). This proportion comes from Column 3 in the table above. Thus, for z = 1.2, the proportion of scores
that fall in the tail 0.1151. Note, that the proportion of scores in the tail for z = 1.2 is equal to the
proportion of scores in the tail for z = -1.2. Importantly, if a z-Score is positive then this proportion of
scores in the tail represents the proportion of scores in the distribution that are greater than the raw score;
if a z-Score is negative, then this proportion of scores in the tail represents the proportion of scores in the
distribution that are less than the raw score.
Note that the proportion of scores that fall between the mean and a z-Score (Column 5) to the proportion
of scores that lie beyond that z-Score and into the tail of the distribution (Column 3) add to 0.5000. Thus,
the proportion of scores above the mean in a normal distribution is always 0.5000 and the proportion of
scores below the mean in a normal distribution is always 0.5000. This is so, because in a normal
distribution the mean is equal to the median and there are always 50% of the scores above the median
and 50% of the scores below the median.
What do columns 2 and 4 in Table 1 represent? Notice that the value in Column 2 is twice the value in
Column 5 and the value in Column 4 is twice the value in Column 3. Column 2 provides the proportion of
scores that fall between the positive and negative z-Score of the same absolute value. That is, for z = 1.2
Column 2 tells us the proportion of scores between z = 1.2 and z = -1.2 [p(68 ≤ X ≤ 92)] is equal to
0.7699. Column 4 provides the proportion of scores that fall outside the z-Score in the tails of the
distribution. Thus, for z = 1.2, Column 4 tells us that the proportion of scores in both tails that fall outside
of z = ±1.20 (0.2301).
You can also use the Table 1 to determine the probability of obtaining a score that is greater than or less
th
than a given raw score. For example, with the same normally distributed set of 19 Century American
Literature test scores (μ = 80; σ = 10); say that we now want to determine the probability of obtaining a
score that is less than or equal to a raw score of 75. The z-Score for 75 in this distribution is:
X   75  80  5
z    0.5
 10 10
-101-
Back to TOC
Look up z = 0.50 in the Table 1. (Remember, there are no negative z-Values in the table.) Because we
are interested in the probability of obtaining a score that is less than or equal to 75, you use the value in
Column 3, which tells us that the probability of obtaining a score of 75 or less is p (X ≤ 75) = 0.3085.
What about the probability of obtaining a score of 75 or more? In this case you use the same z-Score of z
= -0.50 (z = .50 in the Table 1). The question is what value do you use from Table 1 to answer this
question? Unfortunately, there is no single proportion that provides you the needed information. What you
need to do is find the probability of a score between the mean and z = .50, which is 0.1915 from Column
5 in Table 1 and then add 0.5000. Adding 0.5000 takes into account the upper half of the distribution.
That is, the probability of obtaining a score of 75 or more is equal to the probability of obtaining a score
between that raw score and the mean added to the probability of obtaining a score in the other half of the
normal distribution (above the mean). Thus, the probability of obtaining a raw score of 75 or greater is
p(75 ≤ X) = 0.6915.
How about the probability of obtaining a score that is 90 or less? Again, calculate the z-score, which is z =
1.00. You need to find the total probability below z = 1.00, which is equal to the area between the mean of
the normal distribution and z = 1.00 (0.3413 from Column 5) added to the total area below the mean
(0.500). Thus, the probability of a score below 90 is p(X ≤ 90) = 0.3413 + 0.5000 = 0.8413
What if you wanted to calculate the probability of obtaining a score between two values? There are two
ways to do this: (1) When both values fall on the same side of the mean, and (2) when one values falls
above the mean and the other value falls below the mean.
th
From the same distribution of 19 Century American Literature test scores with μ = 80 and σ = 10; say
that we want to know the probability of obtaining a score between 65 and 85. First, calculate the z-Scores
for each raw score, which are z = -1.50 for 65 and z = 0.50 for 85.
Next, look up the probability between each z-Score and the mean in Column 5, which is 0.4332 for z = -
1.50 and 0.1915 for z = 0.50. Because each score falls on a different side of the mean (one z-Score is
positive the other is negative), add these two probabilities. The probability of obtaining a score between
65 and 85 is p(65 ≤ X ≤ 85) = 0.4332 + 0.1915 = 0.6247.
Now let’s say that we want to determine the probability of obtaining a score between 85 and 90, both of
which are on the same side of the mean. First, calculate the z-Scores for each raw score, which are z =
0.50 for 85 and z = 1.00 for 90.
Next, look up the probability between each z-Score and the mean in Column 5, which is 0.1915 for z =
0.50 and 0.3413 for z = 1.00. Because both values are on the same side of the mean, subtract the
smaller probability from the larger probability, which will leave you with the probability between 85 and 90.
Thus, the probability of obtaining a score between 85 and 90 is p(85 ≤ X ≤ 90) = 0.3413 - 0.1915 =
0.1498.
The same procedure would be carried out is both values were less than the mean. For example, say we
want to calculate the probability of obtaining a score between 65 and 75, both of which are less than μ =
80. First, calculate the z-Scores for each raw score, which are z = -1.50 for 65 and z = -0.50 for 75. Next,
look up the probability between each z-Score and the mean in Column 5, which is 0.4332 for z = -1.50
and 0.1915 for z = -0.50. Because both values are less than the mean, subtract the smaller probability
from the larger probability. Thus, the probability of obtaining a score between 65 and 75 is p(65 ≤ X ≤ 75)
= 0.4332 - 0.1915 = 0.2417.
Important: When calculating the probability between two scores that are on the same side of the mean,
you will always subtract the smaller probability from the larger probability. But, when calculating the
probability between two scores that are on different sides of the mean, add the two probabilities together.
-102-
Back to TOC
1. Given a distribution with a mean of 15.565 and a standard deviation of 4.255, compute the standard
score equivalents of the following scores:
a. X = 16.569 b. X = 10.895 c. X = 0 d. X = 15.565

e. X = 11.255 f. X = 20.525
2. Given a distribution with a mean of -5.000 and a standard deviation of 2.500, compute the standard
score equivalents of the following scores:
a. X = -6.730 b. X = 2.950 c. X = -2.500 d. X = 0

e. X = -6.550 f. X = 0.850
3. Given a distribution with a mean of 10.000 and a standard deviation of 3.000, compute the raw score
equivalents of the following standard scores:
a. z = 2.250 b. z = -1.000 c. z = -0.750 d. z = 0
4. If a person got a raw grade of 65 on a psychology test. Assuming that the raw grades will be curved
based on the class' performance, which of the following class distributions would provide the most
favorable interpretation of this raw grade? Why?
a. X  55, s  10 b. X  55, s  5 c. X  60, s  2.5 d. X  60, s  1.5
5. For the following distribution, convert a score of 40 to a standard score { 50, 45, 40, 40, 45, 50 }.
6. For the following distribution, convert a score of 50 to a standard score: 60, 55, 50, 50, 55, 60.
7. What are the values of the mean and the standard deviation for any set of standard scores?
8. What is a z-Score?
9. What proportion of z-scores in a normal distribution are:
a. -1.96 or more b. -1.96 or less c. 1.65 or less

d. 1.65 or more e. 1.38 or less f. -1.38 or more
g. between 0 and 2.00 h. between -2.00 and 0 i. between .40 and 3.01
j. between -3.01 and .40 k. between -1.24 and +1.24
10. Suppose IQ scores in a population are normally distributed with μ = 100.00 and σ = 15.00. What
proportions of individuals have IQ scores of:
a. 100 or higher b. 100 or less c. between 110 and 120

d. 95 or less e. 95 or higher f. Between 90 and 110
11. Given a set of normally distributed scores with a mean of 100.00 and a standard deviation of 15.00,
what score corresponds to a z-score of:
a. 2.75 b. 0 c. -2.00 d. -1.50 e. 1.25

f. 0.70
12. Given a normal distribution with a mean of 100.000 and a standard deviation of 15.00, what score
would:
-103-
Back to TOC
a. 33% of the cases be greater than or equal to?
b. 5% of the cases be greater than or equal to?
c. 2.5% of the cases be greater than or equal to?
d. 2.5% of the cases be less than or equal to?
13. Suppose that the income of a small town has a mean of $40,000.00 with a standard deviation of
$10,000.00, what score would:
a. 33% of the cases be greater than or equal to?

b. 15% of the cases be greater than or equal to?
c. 2.5% of the cases be greater than or equal to?
d. 1.0% of the cases be greater than or equal to?
e. 0.1% of the cases be greater than or equal to?
14. Dr. Bob Ross is a humanist psychologist that wants to make people happy. He has developed a
device called the Happy o'meter, which is used to measure the overall happiness of an individual. Scores
on the Happy o'meter are normally distributed with µ = 25.250 and σ = 5.150 Use these population
parameters to calculate the standard (z) score for raw score (X) below:
a. X = 28.350
b. X = 24.195
c. X = 31.565
d. X = 33.000
e. X = 20.250
f. X = 21.555
g. X = 35.255
h. X = 45.200
15 In a normally distributed population of scores with µ = 50 and σ = 15, determine the proportion (not
percentage) indicated in each question below. Please note that you will have to calculate the standard (z)
score for each raw score first. Also, be sure to leave your proportions to four places (ten-thousandths
place).
a. p (X > 55)
b. p (X < 75)
c. p (X < 25)
d. p (X > 35)
e. p (X > 45)
f. p (30 < X < 70)
g. p (55 < X < 65)
h. p (20 < X < 40)
16. A major form of identification in criminal investigations is fingerprints. Fingerprints vary on many
dimensions, one of which is called the ridgecount. Suppose that you know that the ridgecounts of human
beings follow a normal distribution with a mean of 165.00 and a standard deviation of 10.00. Suppose that
a set of fingerprints was found at the scene of a crime and it was determined that the ridgecount was at
least 200 (the exact value being in question because of smudging). Finally, suppose that a particular
suspect has a ridgecount of 225. What should you conclude and why?
-104-
Back to TOC
Chapter 8: Estimation and Sampling Distributions
8.1 Using Samples to Estimating the of Population

The picture above is intended to represent the concept of parameter estimation, which is simply using
sample data to estimate parameters in a population. Most research follows this idea: we select a sample
from some population and the data from the sample are used to make inferences about what we would
find in the population. Remember, a population is a collection of all individuals or cases that are
associated with some trait. Populations are theoretically infinitely large; hence, it is impossible to know
most parameters with absolute certainty. Researchers use sample data to estimate parameters and infer
what would be found in the population.
But, how good of an estimator are sample statistics? How accurately does the sample mean, variance,
and standard deviation estimate the true population mean, variance, and standard deviation? In this
chapter we go through how the sample mean is the unbiased estimator of the population mean (μ), but
2
the sample variance (s ) and sample standard deviation (s) are biased estimators of the population
2
variance (σ ) and standard deviation (σ), respectively, and how such bias is corrected.
8.2 Sample Mean is an Unbiased Estimator of the Population Mean

Assume you are a statistics professor and one of your a classes is populated with N = 25 students. On a
quiz, which has a range of 0 – 10 points, you obtain the scores in the table below. Below the scores is the
2
population (class) mean (μ), variance (σ ), and standard deviation (σ). If we consider this class to be a
population, these are parameters. But assume you do not actually know these parameters.
10 10 9 9 8 8 7 7 6 6 6 6 5
5 5 5 4 4 4 3 3 2 1 1 1
2
N = 25 μ = 5.4 σ = 7.04 σ = 2.636
You meet with other statistics teachers to compare grades. Remember, you do not know the population
parameters from your quiz data (maybe you do not have time to calculate parameters for 25 students), so
2
you randomly sample n = 5 scores and calculate the sample mean, sample variance (s ), and sample
standard deviation (s). This sample contains the following scores { 10, 8, 5, 5, 3 }.
-105-
Back to TOC
The sample mean of the five scores is 6.2, which is not equal to the population mean of 5.4. The
difference between the sample mean (6.2) and the population mean (5.4) is 0.8 and is called sampling
error. Sampling error results from a sample not perfectly reflecting the population. That is, any time you
select a sample from a population you are not selecting all of the scores. Sampling error is always
present when you are working with sample data. This is true even if a sample mean is equal to the true
population mean. The reason is simple: when working with a sample you are not working with the
population, so there is necessarily some degree of error in your statistic relative to a parameter. The only
way a sample can perfectly represent a population is if the sample included the entire population, which in
that case you would have the population!
Say you randomly select another sample of n = 5 scores from the population of N = 25 quiz scores { 9, 6,
5, 4, 2 }. This second sample's mean is 5.2 and the sampling error -0.2 (5.3 - 5.4); thus, this sample's
mean is less than the population mean. If you were to continue with this exercise and generate more
samples from this population, you would find that some sample means are greater than μ, some are less
than μ, and some are equal to μ. Importantly, this indicates is that sample means will never be
consistently greater than or consistently less than the population mean; thus the sample mean is an
unbiased estimator of μ. That the sample mean is an unbiased estimator of μ simply means the sample
mean will not always be greater than or always be less than μ; thus, it is not going to consistently be on
one side of μ or the other.
8.3 Sample Variance and Sample Standard Deviation as Biased Estimators

If the sample mean is an unbiased estimator of μ, are the sample variance and sample standard deviation
unbiased estimators of the population variance and population standard deviation?
2
In the first sample from Section 8.2 the sample variance is s = 6.16 and the standard deviation is s =
2
2.482. From the second sample the sample variance is s = 6.36 and the standard deviation is s = 2.522.
2
In both samples the variance and standard deviation were less than the population variance (σ = 7.04)
and population standard deviation (σ = 2.636). If you continued generating samples from the N = 25 quiz
scores, you would find the sample variance and sample standard deviation are consistently smaller than
2
σ and σ. This means the sample variance and sample standard deviation are biased estimators,
because they consistently underestimate the population variance and the population standard deviation.
2 2
That is, s and s will almost always be smaller than σ and σ. This is bad! Bias is bad. Don’t be biased.
2
How does one correct for the underestimation of σ and σ? When using a sample to estimate the
population variance and population standard deviation, you divide the sum of squares (SS) form your
sample data by n – 1, rather than by n. This will produces an estimate of the population variance:
ˆs 2 
 XX 
2
n 1
This can also be written as:
SS
ˆs 2 
n 1
The “hat” (^) indicates that the value is an estimate of the population parameter. From Section 8.2, the
first sample had a sum of squares (not shown) equal to SS = 30.8, so the estimate of the population
variance is:
SS 30.8
ˆs 2    7.7
n 1 5 1
-106-
Back to TOC
Using the same formula on data from the second sample, which had a sum of squares (not shown) of SS
= 31.8, the variance estimate is:
SS 31.8
ˆs 2    7.95
n 1 5 1
2
In both samples the variance estimate is slightly larger than the population variance (σ = 7.04). This is
actually OK! Surprisingly, for reasons beyond the scope of this book, you would rather overestimate
variability than underestimate variability. Just note that if you used larger sample sizes to estimate the
variance, the variance estimates should be much more accurate estimators of the population variance.
In both a population and a sample, the standard deviation is defined as the square root of the variance.
Hence, it follows that the estimate of the population standard deviation is equal to the positive square
root of the estimate of the population variance:
ˆs 
 XX 
2
n 1
This can also be written as:

SS
ˆs 
n 1
and as:
ˆs  ˆs 2
From samples 1 and 2 above, we have:
Sample 1: ˆs  ˆs 2  7.7  2.775 Sample 2: ˆs  ˆs 2  7.95  2.82
Like the estimated variance, the estimated standard deviations from samples 1 and 2 are slightly larger
than the population standard deviation (σ = 2.636). Again, this is OK, because you would rather
overestimate variability than underestimate variability.
Importantly, if measuring the variability within a sample or within a population, use the sample formulas
and the population formulas, respectively. That is, calculate sum of squares then divide by n (or N) to
obtain variance, and then take the square root of the variance to obtain the standard deviation.
The only time SS is divided by n – 1 is when sample data is being used to estimate the population
variance and standard deviation. This point is made more explicit in the table below, which lists the
formulas you would use based on the type of variance measure that you are looking for:
-107-
Back to TOC
Measure of Variability Population Sample Estimate Population
SS   X   SS   X  X  SS   X  X 
2 2 2
Sum of Squares
 
2  X   
2
s 
2 
 XX 
2
ˆs 
2  XX  2
n n n 1
Variance ----- or ----- ----- or ----- ----- or -----
SS SS
2  s2  SS
N n ˆs 2 
n 1

 X   
2
s

 XX  2
ˆs 

 XX 
2
N n n 1
----- or ----- ----- or ----- ----- or -----
Standard Deviation SS SS SS
 s ˆs 
N n n 1
----- or ----- ----- or ----- ----- or -----
  2
s s 2
ˆs  ˆs 2
There are three important points regarding the table above. First, do not get bogged down in the symbols
and nomenclature; focus on the concepts. Second and related to the first point, each of the three
measures of variability mean the same thing whether you are working with a population, a sample, or are
using a sample to estimate parameters. That is, sum of squares measures the total variability, variance
measures the average variability, and standard deviation is the average difference between a raw score
and the mean. The only thing that differs between statistics, parameters, and parameter estimates is the
reference group. Third and finally, you should note that the only difference in calculations across these
measures of variability is when calculating the variance estimate where you divide SS by n – 1 instead of
by n.
8.4 Degrees of Freedom

Two questions that may arise are where does the n – 1 in the denominator of the variance estimate come
from and what is it called? The n – 1 correction is the degrees of freedom (df) in a sample. The concept
of degrees of freedom is not the easiest thing to describe, so please focus on this section and read it over
a few times (we’ll actually be coming back to this concept during inferential statistics).
Degrees of freedom are the number of scores that can vary independently in a sample. That is, they are
the number of scores in a sample that can take on theoretically any possible value, with the last score
being completely dependent on the other n – 1 scores. This probably still makes no sense, so how about
an example:
Say you have a sample with n = 5 scores and you know the mean is M = 10. Four of the five scores (i.e.,
n – 1) could take on any value, but for the mean to be M = 10, the value of that last (fifth) score depends
on the values of the other four (n – 1) scores. Thus, four scores (n – 1) can vary independently from each
other, but one score depends on the others. Still puzzled? Try this:
Say that four of the five scores are known to be 8, 9, 11, and 12. If the mean is M = 10, what must that
fifth score be in order for the mean to be 10? The four scores sum to 40; hence, to obtain a mean of 10
-108-
Back to TOC
based on five scores the sum of the five values must be 50: the fifth score must be equal to 10. Still not
sure? Let’s try another example:
Say you know that four of the scores are 1, 2, 2, and 5. What must the fifth score be for the mean to be
10? Again, the sum of five scores must be equal to 50 in order to have a mean of 10. Because the sum of
these four scores is only 10, the fifth score must be equal to 40 (50 – 10 = 40).
In any set of sample data, exactly n – 1 scores can vary independently of each other and theoretically
take on any conceivable value for that variable. But exactly one score is completely dependent on the
values of the other n – 1 scores. Thus, degrees of freedom in a data set are the number of scores that are
allowed to vary independently with the last score being completely dependent on the others.
In calculating estimates of the population variance and standard deviation the degrees of freedom
determine the accuracy of those population estimates. That is, as degrees of freedom increase the
accuracy estimating the population parameters also increases. This means that larger samples generate
better estimates of the population.
8.5 Sampling Distribution of the Mean

The sample mean will be greater than or less than the population mean due to sampling error. For
example, from the population of N = 25 quiz scores in Section 8.2, μ = 5.4, the first sample's mean was M
= 6.2 and the second sample's mean was M = 5.2. Both sample means were generated from a random n
= 5 scores from the population. If we continued sampling, with replacement, from the population of N = 25
scores so that we eventually created every possible sample of 5 scores, we would have a collection of all
possible samples from the population. If we calculated the mean in each of those samples, we would end
up with a collection of all possible sample means from the population, that is, a sampling distribution of
the mean.
A sampling distribution of the mean is a collection of all sample means of size n selected from a
population. For any population and for any given sample size, the sampling distribution of the means has
the following characteristics:
The sampling distribution of the mean is normally distributed with the most frequently obtained (most
probable) sample means being similar to the population mean. Most sample means will be similar to the
population mean.
The least frequently obtained sample means are those that are discrepant from the population mean
The mean of the sampling distribution of the mean is centered on the population mean. That is, the mean
of the sampling distribution of the mean is equal to the population mean (μ).
The standard deviation of the sampling distribution of the mean is equal to σ/√n, so long as all of the
samples are the same size (n) and as long as n < N.
Points 3 and 4 form the basis of the central limit theorem, which states that In any population with a
mean (μ) and standard deviation (σ); the sampling distribution of the mean for size n will have a mean
equal to μ and a standard deviation equal to σ/√n. The sampling distribution of the mean will approach a
normal distribution as n approaches N.
The standard deviation of the sampling distribution of means is called the standard error of the mean
(SEM). This value is a measure of the average distance/deviation between the population mean (μ) and a
sample mean of size n:

X 
n
-109-
Back to TOC
Because population parameters are often unknown, the standard error of the mean must be estimated
from sample data using the estimated standard deviation. The formula for the estimated standard error
of the mean is:
ˆs
ˆs X 
n
This formula divides the estimated standard deviation by the square root of the sample size, and is our
best estimate of the true standard error when the population standard deviation is unknown.
To demonstrate how the standard error of the mean is the standard deviation of the sampling distribution
of the mean, consider a population with N = 4 scores {1, 2, 2, 3 }. The population has mean μ = 2 and
standard deviation σ = 0.707.
I randomly select, with replacement, every possible sample of n = 2 scores from this population and
calculate the mean of each of those samples. This is demonstrated in the table below. The X 1 and X2
values are the values in each sample of n = 2 and the table shows the calculation of the sum of squares
(each sample mean minus the mean of sampling distribution [i.e., 2]):
Sample X1 X2 X X    X X   
X
2
A 1 1 1 -1 1
B 1 2 1.5 -0.5 0.25
C 1 2 1.5 -0.5 0.25
D 1 3 2 0 0
E 2 1 1.5 -0.5 0.25
F 2 2 2 0 0
G 2 2 2 0 0
H 2 3 2.5 0.5 0.25
I 2 1 1.5 -0.5 0.25
J 2 2 2 0 0
K 2 2 2 0 0
L 2 3 2.5 0.5 0.25
M 3 1 2 0 0
N 3 2 2.5 0.5 0.25
O 3 2 2.5 0.5 0.25
P 3 3 3 1 1
N = 16  X  32 SS = 4
X  2
The mean of the sampling distribution of the mean is 2, which is equal to the population mean for the N =
4 scores (μ = 2). The standard deviation of these N = 16 sample means is:
SS 4
X    0.25  0.5
N 16
If we calculate the standard error of the mean using the population standard deviation (σ = 0.707) and the
sample size of each sample (n = 2), we get:
 0.707 0.707
X     0.5
n 2 1.414
-110-
Back to TOC
This is equal to the standard deviation of the sampling distribution of the mean (0.5). Thus, the standard
error of the mean is the standard deviation of the sampling distribution of the mean. Remember, that a
standard deviation is a measure of the average difference between a score and a mean. Thus, the
standard error of the mean is a measure of the average deviation between a population mean (μ) and a
sample mean of size n.
The standard error of the mean is influenced by both the population standard deviation (σ) and the
sample size (n). Holding sample size constant, greater population variance (larger σ) results in a larger a
standard error of the mean. Holding population variance (σ) constant larger sample sizes result in smaller
standard error. That the standard error of the mean is influenced by the sample size demonstrates an
important characteristic of the central limit theorem: as n approaches N the sampling distribution of the
mean approaches a normal distribution. Another way to state this is that as n gets larger the sampling
distribution of the mean becomes more “normal.” Or, more succinctly, larger sample sizes produce better
approximations of the population. This means that by using a larger sample size, the population estimates
that come from a sample will more accurately estimate the parameters.
The standard error of the mean will decrease as sample size increases. Thus, as sample size is
increases and the standard error decreases and sample means will be on average closer estimators of μ,
which reflects more accuracy in the sample estimating the population.
8.6 Confidence Intervals around a Sample Mean

Okay, you’ve collected sample data, calculated the mean, which is your best estimate of μ, and you have
estimated population variance and standard deviation. Your sample mean may be different from the
population mean due to sampling error; so it would be nice to know with what degree of certainty the
sample mean reflects the population mean. To do this, one must estimate confidence intervals around
the sample mean.
Confidence intervals are areas around the sample mean (not around the population mean) where we are
pretty sure the population mean exists. In practice, a confidence interval is a range of values around a
sample statistic that has a certain likelihood of including an unknown population parameter. You can
generate a confidence interval any sample statistic, but we will generate a range around a sample mean
that has a certain likelihood of including the population mean.
Researchers generally calculate a 95% confidence interval (95% CI) or 99% confidence interval (99%
CI) around the sample mean. The percentage can be defined in two ways: First, it is the percentage of
scores from a population that surround a sample mean. That is, the area of the distribution that is covered
by the confidence interval. Second, it is the percent chance or likelihood that the population mean falls
within the confidence interval. The larger the CI (the greater the percentage) the more certain we can be
that the population mean falls within that confidence interval and the greater confidence we can have that
our sample mean reflects the population.
Confidence intervals can be calculated with known population standard deviation or unknown population
standard deviation. We’ll start with the situation where you know the population standard deviation. For
these examples, I will use the statistics quiz examples form Section 8.2. In those examples, μ = 5.4 and σ
= 2.636, and ach sample included n = 5 scores {sample 1: 10, 8, 5, 5, 3; sample 2: 9, 6, 5, 4, 2 }. The
mean for sample 1 was M = 6.2 and the mean for sample 2 was M = 5.2. We’ll calculate the 95%
confidence interval around each sample mean.
To calculate a confidence interval around a sample mean when σ is known, first, calculate the standard
error of the mean. From the quiz score data in Section 8.2, the population standard deviation was σ =
2.636 and each sample size was n = 2. Thus, the standard error of the mean for each sample is 2.636/√5
= 1.179.
-111-
Back to TOC
Next, locate the z-Score associated with the confidence interval you are calculating. This z-Score (called
zα), will be based on the size of the confidence interval you are calculating (i.e., 95%CI, 99% CI, etc.).
What value do you use for zα? Remember, we want to estimate a confidence interval around the sample
mean that includes 95% of the values in the population. Stated differently, we want to calculate an interval
around each sample mean that includes all but the most extreme, or distant, 5% of the scores (100% -
95% = 5%) in the population. This 5%, which as a proportion is .05, is known as the alpha level (α). The
alpha level is the probability that the population mean will not fall within the confidence interval around the
sample mean; thus, it is the likelihood that the sample mean is not related to the population mean. We’ll
discuss the alpha-level in more detail in later chapters.
The value for zα is found by looking up the proportion associated with α (.05) in Column 4 of Table 1 in
Appendix A. That is, for whatever confidence interval you are estimating (99%, 95%, etc.), subtract that
percentage from 100% to get α as a percentage (1% or 5%) and then convert that percentage into a
proportion (.01 and .05). Locate that proportion in Column 4 of Table 1 and use the z-Score associated
with that proportion for zα. When you look up α = .05 in Column 4 of Table 1, you should find a z-Score
equal to 1.96. This is zα for the 95% confidence interval.
Now that you have the mean for each sample (6.2 and 5.2), the standard error of the mean (1.179) and z α
= 1.96, you can calculate the confidence interval around the sample means, by using the following
expression. Note that this expression can be used for generating a confidence interval of any size:
CI  X  z   X
Plugging in the sample mean, zα, and the standard error of the mean, we get:
CI  X  z   X CI  X  z   X
Sample 1:
CI  6.2  1.961.179 Sample 2:
CI  5.2  1.961.179
CI  6.2  2.311 CI  5.2  2.311
The confidence interval (± 2.311) around each sample mean is the same, because the standard error of
the mean and the zα value are the same in each case. If the sample sizes differed or of we calculated the
99% confidence interval around one mean, this would not occur.
We can use the confidence intervals to determine whether the sample mean accurately reflects the
population mean. By adding and subtracting the confidence interval from the sample mean, you end up
with an upper confidence limit and lower confidence limit. For the two samples above, we have:
Sample 1: Lower Confidence Limit: 6.2 + 2.311 = 8.511

Upper Confidence Limit: 6.2 - 2.311 = 3.889
Sample-2: Lower Confidence Limit: 5.2 + 2.311 = 7.511

These confidence limits represent the boundaries of the 95% around each sample mean. Stated
differently, we can be 95% certain that the population mean falls between these limits around the sample
mean. In both cases, the population mean (μ = 5.4) does fall within the confidence interval (between the
limits). Thus, we conclude that both sample means are good estimators of the population mean; that is,
each sample accurately reflects the population. If, the population mean had fallen outside of that range
we would conclude that the sample means are not good estimators of the population mean and that the
samples do not accurately reflect the population. In short, if μ is found to lie within the confidence interval,
the sample is a good representative of the population. If μ is found to lie outside the confidence interval,
the sample is a poor representative of the population.
-112-
Back to TOC
8.7 Confidence Intervals with Unknown Parameters
What if the population standard deviation is unknown? In this case you cannot calculate the true standard
error of the mean; however, you can estimate the standard error of the mean and use it to estimate the
true confidence interval around the same mean. Recall, the estimated standard error of the mean is:
ˆs
ˆs X 
n
What about zα? Because we do not know the population parameters, using z-scores, which are
associated with a normally distributed data, is inappropriate, because we do not know anything about the
population. Thus, we need some value to estimate zα. This estimate of zα, which we call tα, comes from
the t-Distribution, which is presented in Table 2 in Appendix A. A portion of the t-Tables is below, which
is described in detail in later chapters.
Table 2: Probabilities Under the t-Distribution

df
t 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
2.69 .2266 .1149 .0744 .0547 .0433 .0361 .0311 .0275 .0248 .0227 .0210 .0197 .0185 .0176 .0168
2.70 .2258 .1142 .0738 .0541 .0428 .0356 .0306 .0271 .0244 .0223 .0207 .0193 .0182 .0173 .0165
2.71 .2250 .1135 .0732 .0535 .0423 .0351 .0302 .0267 .0240 .0219 .0203 .0190 .0179 .0169 .0161
2.72 .2243 .1128 .0725 .0530 .0418 .0346 .0298 .0262 .0236 .0216 .0199 .0186 .0175 .0166 .0158
2.73 .2235 .1121 .0719 .0524 .0413 .0342 .0293 .0258 .0232 .0212 .0196 .0183 .0172 .0163 .0155
2.74 .2228 .1114 .0713 .0519 .0408 .0337 .0289 .0254 .0228 .0208 .0192 .0179 .0169 .0160 .0152
2.75 .2220 .1107 .0707 .0514 .0403 .0333 .0285 .0251 .0225 .0205 .0189 .0176 .0165 .0156 .0149
2.76 .2213 .1100 .0702 .0508 .0398 .0329 .0281 .0247 .0221 .0201 .0186 .0173 .0162 .0153 .0146
2.77 .2206 .1094 .0696 .0503 .0394 .0324 .0277 .0243 .0218 .0198 .0182 .0170 .0159 .0150 .0143
2.78 .2198 .1087 .0690 .0498 .0389 .0320 .0273 .0239 .0214 .0195 .0179 .0167 .0156 .0147 .0140
2.79 .2191 .1080 .0684 .0493 .0384 .0316 .0269 .0236 .0211 .0191 .0176 .0163 .0153 .0145 .0137
2.80 .2184 .1074 .0679 .0488 .0380 .0312 .0265 .0232 .0207 .0188 .0173 .0160 .0150 .0142 .0135
The t-Tables are similar to the z-Tables: The leftmost column lists t-Values and the values in the table are
probabilities under the t-Distribution. The column headings list degrees of freedom (df) values; hence,
each column is associated with a different number of degrees of freedom in a sample. Each sample from
Section 8.2 included n = 5 scores, hence, the degrees of freedom in each sample was df = 4. I highlighted
this column in yellow in the table above. Once you locate the correct column, scroll down that column until
you find the probability associated with the alpha level (.05) or in this case the probability immediately
smaller than alpha (.0498). When you reach that location, the value in the leftmost column (2.78) will be
tα. The expression for estimating the confidence interval with unknown population parameters is:
CI  X  t ˆs X
The is identical to the formula in Section 8.6, except that in place of the standard error of the mean, the
estimated standard error of the mean is being used, and in place of zα, tα is used. The estimated standard
deviations for samples 1 and 2 in Section 8.2 are 2.775 and 2.820, respectively. Hence, the estimated
standard error of the mean for samples 1 and 2 are:
ŝ 2.775 2.775 ŝ 2.82 2.82

Sample 1: ŝ X   
2.236
 1.241 Sample 2: ŝ X   
2.236
 1.261
n 5 n 5
Thus, the estimated 95% confidence intervals around each sample mean are:
-113-
Back to TOC
CI  X  t ŝ X CI  X  t ŝ X
Sample 1:
CI  6.2  2.781.247  Sample 2:
CI  6.2  2.781.261
CI  6.2  3.467 CI  6.2  3.506
And the confidence limits are:


When estimating the confidence intervals as we have done here, because the population parameters are
unknown, all that can be said is that there is a 95% chance that the population mean is contained within
the confidence interval. Or, we are 95% certain that the sample mean accurately reflects the population
mean.
8.8 Displaying Confidence Intervals in Graphs

It is good practice to include confidence intervals around means presented in graphs. These error bars
are drawn extending up from mean to the upper confidence limit and extending down from a mean to the
lower confidence limit. For example, the graph to the right depicts the means from samples 1 and 2 from
the earlier sections. The height of each bar represents the mean of each sample and the black line
extending above the mean and below the mean represents the end points (confidence limits) of the 95%
confidence interval around the sample means. This is a common way to graphically present where the
most probable location where the population mean will appear.
There are very simple ways to do

this in Excel and other graphing
programs. Importantly, APA
formatting suggests including
confidence intervals in all graphs
that present numerical data, so it is
good to become familiar with how
to create them.
-114-
Back to TOC
1. What is sampling error? What is sampling error due to? How can we represent the amount of sampling
error that is present in a statistic?
2. What is an unbiased estimator? What is a biased estimator? Which sample statistics are unbiased and
which are biased?
3. Why is the sample mean the “best estimate” of a population mean?
4. Why is the sum of squares divided by n - 1 rather than by n when computing the variance estimate?
5. What are degrees of freedom?
6. What is the sampling distribution of the mean?
7. What characteristics of a sampling distribution of the mean are addressed by the central limit theorem?
8. What will the mean of a sampling distribution of the mean always equal? Why?
9. What is the standard error of the mean? What information does it convey? How is it calculated?
10. What is the difference between the standard error of the mean and a standard deviation of a set of
raw scores?
11. In cases where the population standard deviation is unknown, or does not exist, the standard of the
mean cannot be calculated. What is used in its place, and how is it calculated?
12. What are the two factors that influence the size of the standard error of the mean, and how do they do
so?
13. Say that a sampling distribution of the mean has a standard error of the mean that is equal to 0. What
does this indicate about the means of the samples drawn from the population?
14. Say that a sampling distribution of the mean has a standard error of the mean that is equal to 0. What
does this say about the variability of scores in the population (σ)?
15. Compute the variance and the standard deviation for the following scores: { 5, 4, 2, 6, 3, 7, 3, 1, 4, 5 }.
Next, estimate the variance and estimate the standard deviation in the population, and then compare the
two sets of results.
16. Compute the variance and the standard deviation for the following scores: 6, 5, 3, 7, 4, 8, 4, 2, 5, 6, 3,
6, 5, 4, 7, 5. Estimate the variance and the standard deviation in the population, and compare the two
sets of results.
17. Compute the variance and the standard deviation for the following scores: { 7, 6, 8, 9, 7, 6, 7, 8, 4, 8 }.
Next, estimate the variance and estimate the standard deviation in the population, and then compare the
two sets of results.
18. The following scores below represent a sample of working memory capacity scores taken from a
group of research subjects in a study on memory
62 73 73 53 74 55 66 49
64 74 75 65 68 75 64
-115-
Back to TOC
d. Calculate the sample standard deviation.
e. Calculate the estimated population variance.
f. Calculate the estimated population standard deviation.
19. Say that the population standard deviation for working memory capacity scores is σ = 8. Calculate the
standard error of the mean based on the sample size from #18.
20. Calculate the 95% confidence interval around the sample mean in exercise #18, using the standard
error of the mean from #19.
21. Calculate the 99% confidence interval around the sample mean in exercise #18, using the standard
error of the mean from #19.
22. Use the following to answer the questions below. The "attentional control scale" is a device used for
measuring individual differences in a person's ability to control and maintain their focus of attention and
thinking. Scores on the attentional control scale range from 15 to 60, with higher scores reflecting better
control over attention. Assume that scores on the attentional control scale are normally distributed with µ
= 37.5 and σ = 10. You administer the attentional control scale to a random sample of twelve psychology
majors and obtain the scores listed below. Use them to answer the following questions:
26 28 47 21 38 27 38 18 36 22 26 45

c. Calculate the estimated population variance.
d. Calculate the estimated population standard deviation.
e. Calculate the standard error of the mean, based on the population standard deviation.
f. Calculate the 95% confidence interval around the sample mean.
g. Calculate the estimated standard error of the mean, based on the estimated standard deviation.
h. Using the estimated standard error of the mean, calculate the 95% confidence interval around the
sample mean.
23. Use the following to answer the questions below. Short term memory is the storage area of memory
that maintains information in an active state just long enough for that information to be moved and stored
in long term memory. The capacity of short term memory is generally µ = 7 pieces of information and σ =
2. A short researcher tests the short term memory for a sample of five neuroscience majors and obtains
the following short term memory scores. Use this information to answer the questions below:
6 8 9 8 9

c. Calculate the estimated population variance.
d. Calculate the estimated population standard deviation.
e. Calculate the standard error of the mean, based on the population standard deviation.
f. Calculate the 95% confidence interval around the sample mean.
g. Calculate the estimated standard error of the mean, based on the estimated standard deviation.
h. Using the estimated standard error of the mean, calculate the 95% confidence interval around the
sample mean.
24. What two things does the size of the confidence interval reflect?
-116-
Back to TOC
25. A sample of n = 20 is drawn from each of two populations. For population A, μ = 8.00 and σ = 4.00;
and for population B, μ = 9.00 and σ = 6.00. Which sample mean is likely a better estimate of its
population mean? Why?
26. Compute the estimated standard error of the mean for the data in Exercise 15.
27. Compute the estimated standard error of the mean for the data in Exercise 16.
28. Compare the estimated standard error of the mean calculated in Exercises 26 and 27. Which sample
mean is probably a better estimate of its population mean? Why?
29. A population has the following parameters: μ = 500 and σ = 100. A sample of n = 100 subjects is
randomly selected, and has M = 520. Calculate the 95% confidence interval around the sample mean. Is
the sample a good representative of the population?
30. Calculate the 99% confidence interval based on the information in #29. Is the sample now a good
representative of the population?
31. Say that the sample size from #29 and 30 was increased to n = 1000. Recalculate the 99%
confidence interval based on the information in #29 and 30. Is the sample now a good representative of
the population?
-117-
Back to TOC
Chapter 9: Probability
9.1 The Monty Hall Paradox
I’ve never understood the reason, but humans hate probabilities. Indeed, one thing I have found as an
instructor of statistics (and I do not mean to be condescending) is students are generally bad at
computing, understanding, and evaluating probabilities. Math is power...math is power! I think a lot of this
comes from students not taking time to evaluate information given to them, and instead calculating the
first probability value that comes to mind. You should always evaluate the situation and conceptualize
what needs to be calculated before putting pencil to your paper. The following is a good example of
people failing to take information into account when estimating probabilities.
Most of you are too young to remember the game show “Let’s Make a
Deal” with host, Monty Hall. The gist of the show was the winner of the
main round would go to the prize round, where the contestant is
presented with three doors: Door 1, Door 2, and Door 3. Behind one of
the doors was a fabulous prize, such as a new car, money, or a vacation.
Behind the other two doors were gag prizes, such as goats.
Monty Hall tells the contestant to pick one of the three doors. Now, the
probability of correctly selecting the door with the prize on this first
selection is 1/3, because there are three doors and only one of door hides a prize. For this scenario
assume Door 1 was initially selected by the contestant. Monty Hall then says something like, 'I’ll tell you
what I’m going do. I’ll open one of the two remaining doors, you can then decide to stay with your original
choice or switch to the other door'. Say Monty Hall opens Door 2, revealing a goat.
The Monty Hall Paradox (Monty Hall Problem) boils down to whether you should stay with your original
choice (Door 1), or switch to the unopened door (Door 3). That is, are you more likely to win by staying
with your original choice, are you more likely to win by switching to the other door, or does it make no
difference whether you stay or switch?
When presented with this problem most people assume it doesn't matter whether you stay or switch, Most
people think the probability of winning by staying with your original choice is 0.5 and the probability of
winning by switching is 0.5, because there are only two doors that can be opened and the prize is behind
exactly one door. But the actual probability of winning by staying is 1/3 and the probability of winning by
switching is 2/3! Don’t believe me? We’ll go over the proof of this later, but for now chew on this: The
reason people assume the probability of winning by staying and by switching are equal is because people
neglect the fact that Monty Hall knows where the prize is! When Monty Hall opens a door he provides
some information about the prize location. This does not mean you will always win by switching, just that
it is more likely. This is why I say, when calculating and estimating probabilities visualize and think about
what you need to calculate. Take into account all available information before making any probabilistic
decisions. Indeed, this is a big issue for inferential statistics, which are based entirely on probabilities.
9.2 Some Basic Ideas: Event Algebra

As a student of the behavioral sciences you should love probability, because you will use them a lot.
Probabilities and probability theory form the base of inferential testing, so if you want to do research you
better get familiar with probability. Before getting into more technical aspects of calculating probabilities
we need to go through a few basic ideas of where probability comes from.
A simple random experiment, or trial, is a procedure or operation performed where the outcome is
unknown in advance. That is, prior to collecting data or performing any operation that leads on to collect
data, the actual data that is collected is unknown. The collection of all possible outcomes of a simple
random experiment is the sample space (not, this is not the same thing as a “sample”). I’ll use the
-118-
Back to TOC
uppercase S to denote the sample space. Here are few examples of simple random experiments and
associated sample spaces:
Simple Random Experiment Sample Space

1 Toss a coin exactly once S = {H, T}
2 Toss a coin twice S = {(H, H), (H, T), (T, H), (T, T)}
3 Rolling a six-sided die exactly once S = {1, 2, 3, 4, 5, 6}
4 Rolling a six-sided die exactly twice S = {(1, 2), (1, 2), (1, 3),...,(6, 5), (6, 6)}
5 Count the number of hairs on a cat S = {1, 2, 3,...,N}, where N = total
number of hairs
6 Select a card from a deck of playing cards S = {Ace of Spaces...}
An event is a unique outcome in the sample space of a simple random experiment. Thus, it is a subset of
the sample space. For example, when flipping a coin exactly once, the sample space includes two events
that can occur: Heads and Tails. Here are some examples of events from the table above:
Simple Random Experiment Event

1 Toss a coin exactly once Head = {H}
2 Toss a coin twice Exactly one tail = {(H, T), (T, H)}
3 Rolling a six-sided die exactly once Outcome is even = {2, 4, 5}
4 Rolling a six-sided die exactly twice Sum is > 10 = {(5, 6), (6, 5), (6, 6)}
5 Count the number of hairs on a cat Number of hairs is greater than 1000 =
{1001, 1002, 1003,...,N}
6 Select a card from a deck of playing cards Card is a King = {King of Hearts, King of
Clubs, King of Diamonds, King of
Spades}
Event algebra is a method of expressing relationships among events and among combinations of events.
It is a symbolic way to express relationships among events and among combinations of events from the
sample space created in a random experiment. Below, I provide the symbolic representations of the more
important relationships between events along with Venn diagrams to provide a graph representation
The union of two events, A and B, (symbolically A U B) is the event

that consist of all outcomes that are associated with event A or event B
or both. Thus, in a sample space S, you have events A and B; the
union of those two events is the event that favors both events A and B
or only event A or only event B. (See diagram at right, the shaded area
represents all outcomes in the union of A and B). Here are some
examples of the union of two events A and B:
A B AUB
1 Sum of two dice is multiple of two = {2, 4, 6, 8, 10, Sum of two dice is multiple of four = {4, 8, {2, 4, 6, 8, 10, 12}
12} 12}
2 Selected card is a two { two of spades, two of Selected card is a heart = {two of hearts, {two of spades, two of
diamonds, two of clubs, two of hearts} three of hearts,...ace of hearts} diamonds, two of clubs,
two of hearts, three of
hearts,...,ace of hearts }
-119-
Back to TOC
The intersection of two events, A and B, (symbolically A ∩ B) is the
event that consist of all outcomes that are associated with event A and
event B. Thus, in a sample space S, you have events A and B; the
intersection of those two events is the event that favors both events A
and B. (See diagram at right.
From Example 1 above, the intersection is A ∩ B = {4, 8, 12}, and the

intersection from Example 2 is A ∩ B = {two of hearts}. That is, in
Example 1, the values 4, 8, and 12 are the only outcomes that are both
a multiple of two (event A) and a multiple of four (event B). For
Example 2, the two of hearts is the only outcome that is a two (event
A) and a heart (event B).
C
The complement of an event (symbolically A ) is the event that
consists of all outcomes that are not associated with event A. From
C
Example 1, the complement of event A is A = {1, 3, 5, 7, 9, 11}, and
C
the complement of event B is B = {1, 2, 3, 5, 6, 7, 9, 10, 11}. From
C
example 2, the complement of event A is A = {all cards that are not
C
two}, and the complement of event B is B = {all chards that are not
hearts}.
If events A and B have no outcomes in common (symbolically A ∩ B =

Φ, where Φ = “impossible”), then events A and B are mutually
exclusive events. Thus, when two events are mutually exclusive the
intersection of those two events is not possible. For example, if event A
is selecting a biological male and event B is selecting a biological
female then A ∩ B = Φ. As another example, if event A is select a club
and even B is select a heart, then A ∩ B = Φ. In Examples 1 and 2
above, events A and B are not mutually exclusive, because the
intersection of events A and B is possible in both examples.
An event is included in another events (symbolically A ⊂ B) if each

and every outcome in event A is also in event B. This is similar to, but
not the same as the intersection of A and B, because for the
intersection of A and B, not all outcomes in A necessarily have to be in
event B. For example, if event A is that the sum of two dice is a
multiple of six and event B is the sum of two dice is a multiple of three,
then event A must be included in event B (A ⊂ B), because all
multiples of six will be multiples of three.
9.3 Probability Axioms and Things About Probability that are True
The probability for a simple random experiment is a function P that assigns a real number, P(E), to each
event E in a sample space S, and satisfies the following three axioms:
P(E) ≥ 0
P(S) = 1
P(A U B) = P(A) + P(B) whenever events A and B are mutually exclusive
-120-
Back to TOC
These three axioms form the basis of all probability theory. Simply, axiom 1 states that the probability of
an event must be greater than or equal to one; axiom 2 states that the probability of observing something
in the sample space is 1 (i.e., the probability that you will observe at least one event is equal to 1); and
axiom 3 states that the probability of observing at least one of two mutually exclusive events if the sum of
their respective probabilities.
The following are some additional results that can be derived using event algebra and the three axioms of
probability:
C
For any event, P(A ) = 1 – P(A)
For any event, P(A) ≤ 1
For all events A and B, P(A U B) = P(A) + P(B) – P(A ∩ B), whether A and B are mutually
exclusive or not
C
For all events A and B, P(A U B) = P(A ∩ B) + P(A ∩ B ), whether A and B are mutually exclusive
or not
To calculating the probability of an event, A, you must identify the number of possible outcomes that favor
event A and divide that by the number of possible outcomes where event A can occur, that is, the sample
space (S). Hence:
PA  
number of outcomes in A A

number of outcomes in S S
For example, we want to calculate the probability of selecting a Jack of Clubs from a deck of cards. In a
standard deck of 52 cards there is only one jack of clubs, so A = 1 and the number of outcomes where
event A can occur is S = 52. Thus, the probability of selecting a jack of clubs is 1/52 = .019.
What about the probability of selecting any king? If event A = any king and there are four possible kings in
any deck of cards {king of clubs, king of spades, king of diamonds, king of hearts}, A = 4 and S = 52; then
the probability of selecting any king is 4/52 = .077.
Thus, calculating a probability is as simple as dividing one value by another. The tricky part is figuring out
what value should be divided by what; and this is where attention to detail is important.
9.4 Counting Rules: Permutations and Combination

It is often the case that several operations or procedures will be applied in a simple random experiment.
Here is a simple rule to follow: If an operation O consists of a sequence of operations O1, followed by
O2,..., followed by Or, and O1 can result in any one of n1 outcomes, O2 can result in any one of n2
outcomes after performing O1, and ... Or can result in any one of nr outcomes after performing O1, O2,...,
Or-1, then the operation O can result in any one of n1n2...nr outcomes. Basically, if an operation )simple
random experiment) consists of many operations and each operation can result in one of several possible
outcomes, then for the simple random experiment any one of a number of outcomes will result.
For example, in the behavioral sciences it is often the case that researchers have a certain number of
conditions that can be presented to subjects and the researcher may want to know all possible
permutations that can be presented to subjects. That is, how many orders in which the conditions can be
presented. A permutation is an ordered arrangement of distinct items. The total number of permutation of
n distinct items is
n!
-121-
Back to TOC
That is, “n factorial.” The factorial (!) operator, tells you to multiply the number (n) by all numbers less than
n not including zero. That is, 4! is shorthand for writing 4 X 3 X 2 X 1 = 24. Thus, whenever you see this
symbol you should expand the expression preceding it and multiply the given number by all values below
it, but not zero. Remember 1! = 1 and 0! = 0. There is a factorial table in the statistical packets (Table 7).
For example:
2! = 2 x 1 = 2
3! = 3 x 2 x 1 = 6
4! = 4 x 3 x 2 x 1 = 24
Using a more concrete example, say an investigator is conducting a taste-preference survey for different
types of scotch (wine is for sissies). The investigator has three types of scotch that can be presented (A:
18 year old single malt; B: 12 year old single malt; C: 12 year cask strength). The researcher wants to
know all of the permutations that are possible for the three scotches.
Permutation Scotch 1 Scotch 2 Scotch 3

1 A B C
2 A C B
3 B A C
4 B C A
5 C A B
6 C B A
However, a researcher may want to present only a subset (r) of all the possible conditions or outcomes.
In this case the researcher may want to know how many permutations are possible from all of the
possible subsets out of the total number of conditions. The number of permutations of r items out of n
items is:
n!
n Pr 
n  r !
For example, the number of different permutations for r = 3 conditions out of n = 10 total conditions is:
10!
10 P3 
10  3!
10  9  8  7  6  5  4  3  2  1
10 P3 
7  6  5  4  3  2 1
6328800
10 P3 
5040
10 P3  760
3!
Using the scotch example above, the researcher might want to know the number of 3 P2 
permutations of two scotches out of the three scotches, which is depicted below: 3  2!
3  2 1
Permutation Scotch 1 Scotch 2 3 P2 
1
1 A B
6
3 P2 
2 B A
3 A C 1
4 C A
5 B C 3 P2  6
6 C B
-122-
Back to TOC
Alternatively, a researcher may simply want to know how many possible combinations of subsets are
possible from some larger number of conditions. While permutations are the total number of possible
orders in which a set of conditions can be presented, combinations are the total number of sets that can
be created from some larger number of conditions, but where order of conditions is not important.
Say we have n = 10 conditions and we want to determine the number of combinations and permutations
possible for subsets of r = 3 conditions. That is, we have 10 conditions that we can present to subjects,
but we only want to present 3 to each subject; so we want to get an idea of how many subjects are
needed. The formula for calculating the number of combinations is:
n!
n Cr 
n  r !r!
From the example above, the number of different combinations of r = 3 conditions out of n = 10
conditions:
10!
10 C 3 
10  3!3!
10  9  8  7  6  5  4  3  2  1
10 C 3 
7  6  5  4  3  2 13  2 1
6328800
10 C 3 
50406
6328800
10 C 3 
30240
C
10 3  120
Using the scotch example, assume that the researcher wants to know the number of 3!
3 C2 
combinations of two scotches out of three scotches. Which is depicted in the table
below:
3  1!2!
3  2 1
3 C2 
Combination Scotch 1 Scotch 2 12 1
#1 A B 6
#2 A C 3 C2 
2
#3 B C
3 C2  3
We’ll now explore calculating probabilities using the ideas of permutations and combinations from above.
Example 1: What is the probability of getting exactly one club and exactly one jack in two randomly dealt
cards; assuming order does not matter?
Event A = one club and one jack
There are 13 clubs and 13 jacks, and the number of unordered combinations of clubs and jacks is
13 x 13 = 169
52!
52 C 2 
Sample Space, S = all unordered two card hands
52  2!2!
Thus, the sample space is all combinations of two cards out of a deck of 52  51
52 cards (I have simplified the math): 52 C 2 
2
Thus, P(A) = 169/1326 = 0.127 52 C 2  1326
-123-
Back to TOC
Example 2: What is the probability that a five-card poker hand contains exactly three clubs?
Event A = three clubs and two non-clubs
In this case, you need to account for the combinations of three clubs of the 13 total clubs and also
account for the combinations of two cards out of the 39 non-clubs. Thus:
For the clubs: For the non-clubs:

13! 39!
13 C 3  39 C 2 
13  33! 39  22!
13  12  11 39  38
13 C 3  39 C 2 
3  2 1 2 1
13 C 3  286 39 C 2  741
Sample Space, S = all five card hands 52!

52 C 5 
Thus, the sample space is all combinations of five cards out of 52  5!5!
a deck of 52 cards (I have simplified the math): 52  51 50  49  48
52 C 5 
5  4  3  2 1
Thus, P(A) = (286 * 741)/2598960 = 0.082 52 C 5  2598960
Example 3: What is the probability of getting five cards from the same suit?
Event A = all five cards are from the same suit
In this case, we are interested in the number of five card combinations from each of four 13-card
suits. The number of five card combinations from each suits:
13!
13 C 5 
13  5!5!
13  12  11  10  9
13 C 5 
5  4  3  2 1
13 C 5  1287
There are four suits, so the number of five card combinations from the four suits is: 4 x 1287 =
5148. That is, there are 1287 five-card combinations of hearts, diamonds, spades, and clubs
(1287 + 1287 +1287 +1287 = 5148).
Sample Space, S = all unordered five card hands 52!

52 C 5 
Thus, the sample space is all combinations of five cards out of
52  5!5!
a deck of 52 cards (I have simplified the math): 52  51 50  49  48
52 C 5 
5  4  3  2 1
Thus, P(A) = 5148/2598960 = 0.002
52 C 5  2598960
-124-
Back to TOC
9.5 Probability and Sampling from a Population
One issue with calculating probability is how the selecting of samples from a population can change the
probability of future selections. As discussed earlier, when selecting a sample you typically engage in
random sampling, which literally means selecting cases from a population at random and without bias.
Random sampling is good, because such a procedure should result in a sample that contains the
individual differences that are present in the population. If so, you have an unbiased random sample.
One issue is whether you sample with replacement or sample without replacement. Sampling with
replacement occurs when after a case is selected it is returned (replaced) to the population. Sampling
without replacement occurs when after a case is selected it is not returned to the population. There are
pros and cons of each method.
When sampling with replacement the population size (N) never changes, so the probability of an event
never changes. But, because you are returning cases to the population you run the risk of selecting the
same case more than once. Sampling without replacement corrects for the possibility of selecting a case
multiple times, because after a case is selected it is not returned to the population. But, because after
each selection the population size (N) will decrease one, the probability of unselected events being
selected increases over time. Because most population are large (on the order of millions), altering the
number of cases that can be selected should not make much of a difference; thus, most behavioral
scientists use sampling without replacement.
As examples of sampling with and without replacement, I’ll use the statistics class example above
with N = 50 students. This class includes 25 psychology majors, 10 neuroscience majors, 10
communications majors and 5 counseling majors. Say we select a student at random from the class. The
initial probabilities of selecting a student from each major are listed in the table below:
Major f(Major)/N p(Major)

Psychology 25/50 .500
Neuroscience 10/50 .200
Communications 10/50 .200
Counseling 5/50 .100
After selecting a student (regardless of major), the student was returned to the class (sampling with
replacement), and N will still be 50. Let's say that say we select a Communications major, but the student
was not returned to the class (sampling without replacement). Because N now equals 49 the probability
that students from the other majors will be selected changes, which can be seen in the table below.
Notice the probability of selecting each major increased, except for that of Communications majors,
because that major was decreased by one student:

Say that we select a second student, a psychology major, and do not return that student to the class.
Because N now equals 48 the new probabilities for a student from each major being selected can be
seen in the table below. The point is that changing the underlying population size, just like changing
information in a situation (e.g., Monty Hall problem), alter probabilities of selecting events:
-125-
Back to TOC
9.6 Probabilities Based on Multiple Variables

Assume a researcher is interested in the association between political orientation and sexual behavior.
The researcher surveys all incoming freshmen at a university (N = 1000 freshmen) on two variables: (1)
political attitude {liberal and conservative} and (2) whether the student has engaged in intercourse {yes
and no}. Thus, the freshmen can be classified into one of four groups: (1) liberals who have had
intercourse, (2) liberals who have not had intercourse, (3) conservatives who have had intercourse, and
(4) conservatives who have not had intercourse. For argument’s sake, assume all freshmen are 18 years
old and that there are equal numbers of males and females in each group.
When you have a combination of variables where frequencies are measured you have a contingency
table of that data (see below). A contingency table is used to record the relationship between two or more
variables. Thus, to examine the relationship between political attitude and sexual behavior the
frequencies of students in each of those four groups will be examined in a tabled form, like that below:
Political Attitude
Previously Engaged in Intercourse? Liberal Conservative Totals
Yes 500 100 600
No 200 200 400
Totals 700 300 N = 1000
Each of the four groups created by combining the two variables mentioned above is referred to as a cell.
The value in each cell are the frequencies of students belonging to that combination of variables; thus,
500 liberal freshmen reported having had intercourse, 100 conservative freshmen reported having had
intercourse, and 200 liberal freshmen as well as 200 conservative freshmen reported not having had
intercourse. Values in the last column and the bottom row (Totals) are the sum of the cell frequencies for
that row or column; and these are the marginal frequencies.
Using this contingency table probabilities of events, combinations of two events, and probabilities that are
conditional on (dependent on) some other event can be calculated. When going through each of the
following probabilities, keep in mind that for each of the following probabilities, you must identify the
number of observations favoring the event of interest (‘event A’), and the total number of relevant
observations where that event can occur
9.6.1 Simple Probabilities Say the researcher wants to determine the probability of selecting a liberal
student from among all incoming freshmen. He needs to identify the number of liberal students and the
total number of students that could be a liberal student, which is the total frequency (N). If we call
selecting a liberal student event A, then the number of liberal students is A = 700 and the total number of
students that could have been a liberal student is the total number of freshmen, N = 1000. Thus, the
probability of selecting a liberal is:
PLiberal  
number of liberals 700
  0.7
N 1000
Likewise, the probability of selecting a conservative student is the total number of conservative students (f
= 300) divided by the total number of students (N = 1000):
-126-
Back to TOC
PConservative 
300
 0.3
1000
These are simple probabilities, which are the probability of selecting a single event or level of a
variable. That is, in both cases, the researcher calculated the probability of one political orientation
without taking into account the variable ‘previously engaged in intercourse?’
Notice that the two probabilities add to 1.0. When this occurs it means that the outcomes for this variable
(political attitude) are exhaustive. That is, all of the possible outcomes of the variable have selected or
identified (the researcher has determined the probabilities for all possible events for the variable political
attitude). Also, because a student can either be a liberal or can be a conservative, but cannot be a
political conservative and a political liberal; this means that the two outcomes for Political Attitude are
mutually exclusive. That the levels of a variable are mutually exclusive means that an case (any
freshman student) can belong to only one level of a variable, but cannot belong to more than one level.
The variable previously engaged in intercourse is also mutually exclusive, because a student can have
previously engaged in sexual intercourse, or not; that is, a student cannot have both engaged in and not
engaged in intercourse.
9.6.2 Conditional Probabilities Conditional probabilities can be used to assess whether there is a
relationship between the variables. A conditional probability is the probability of selecting some event (A)
given some other event (B) is selected, that is, that some other condition is satisfied. A conditional
probability is written as:
number of outcomes favoring A and B A  B

PA | B  
number of outcomes favoring B
P(A|B) is read the probability of event A given event B. In the Venn

diagram to the right, the hatched area is the conditional probability of
selecting A given you have selected B, and the B area is shaded to
signify that we are interested in outcomes only in event B.
From the contingency table above, say the researcher wants to

calculate the probability of selecting a student that has engaged in
intercourse given the student is also a political conservative. Stated
differently, if a selected student is a conservative what is the probability s/he has engaged in intercourse?
In this case, event A is having engaged in intercourse and event B is students who identify themselves as
conservative. Why is this so? This question is asking the probability of selecting a conservative student
who has had intercourse, out of only those students who report being politically conservative. Thus, event
B (the denominator above) needs to be the number of cases that can be a conservative student who has
engaged in intercourse, which is all of the conservative students. Event A, the event we are slightly more
interested in, then, must be students who have reported having intercourse.
From the contingency table, the number of conservative students is B = 300. This value comes from the
marginal frequencies. Out of these 300 conservative students, 100 reported to have engaged in sexual
intercourse; that is, A ∩ B = 100. This value comes from the cell frequencies. Thus, the conditional
probability that a student has engaged in intercourse given the student is a conservative:
Yes  Conservati ve
PYes | Conservative 
Conservati ve
100
 0.333
300
-127-
Back to TOC
Hence, there is about at 33.3 percent chance of selecting a student who has had sexual intercourse, if
you have also selected a conservative student. As another example, let’s say that the researcher is now
interested in the probability of selecting a liberal student given we have selected a student who has not
had intercourse. In this example, event A would be ‘liberal students’, and event B would be all students
who have reported not having had intercourse. Thus, the number of liberal freshmen students who have
not had intercourse is A ∩ B = 200, and the total number of students who have not had intercourse if B =
400. The probability of selecting a liberal student given we have already selected a student who has not
engaged in intercourse is:
PLiberal | No  
200
 0.5
400
Thus, there is about a 50 percent chance that the researcher has selected a liberal freshmen student if he
has selected a student who has not engaged in intercourse.
Conditional probabilities can be used to determine whether there is a relationship between the two events
(and variables), that is, whether the two events are independent or not. If two events are independent it
means that selecting one event does not depend on the other event. In contrast, if two events are not
independent, that is, there is a relationship between the events then selecting one event does depend on
the other. Two events are independent if the following holds:
P(A) = p(A|B)
That is, if the probability of event A is equal to the conditional probability of event A given event B, then
the two events A and B are independent and selecting event A does not depend on selecting B. Two
events are not independent, and there is a relationship between them, if the following holds:
P(A) ≠ p(A|B)
To determine whether having previously engaged in intercourse (event A) is independent of being a
conservative (event B), the first conditional probability that was calculated above must be compared to
the probability of having engaging in intercourse, which is:
PYes  
Yes 600
  0.6
N 1000
From above, we know that,
100
 0.333
300
Because P(Yes) ≠ P(Yes | Conservative) some relationship must exist between having engaged in
intercourse and being a conservative; that is, having had intercourse is influenced by being a
conservative (versus being a liberal). Thus, selecting a freshmen student who has engaged in sexual
intercourse does depend on selecting a politically conservative freshmen student.
9.6.3 Joint Probabilities If you wish to determine the probability of a combination of two events out of
all events you must calculate a joint probability. A joint probability is simply the probability of selecting
tow events, A and B, together. A joint probability is conceptualized as:
AB
PA  B 
N
-128-
Back to TOC
Note that P(A ∩ B) is read the probability of event A and event B. From the contingency table above, say
that the researcher wants to calculate the probability of selecting a student that has not engaged in
intercourse (event A) and is a politically liberal student (event B). The number of students who have not
engaged in intercourse and are liberals is A ∩ B = 200 and N = 1000, thus, the joint probability of
selecting a student that has not had intercourse and is a liberal is:
No  Liberal
PNo  Liberal  
N
PNo  Liberal  
200
 0.2
1000
A joint probability is a cell frequency divided by the total frequency (N). This provides you with the
probability of selecting combination of events, thus, the probability of selecting a student who has not had
intercourse and is a politically liberal student is p = 0.2.
9.6.4 Addition Rule of Probability The addition rule of probability is used to determine the probability
of at least one of two different events, or both events, occurring. That is, what is the probability of only
event A, only event B, or both events A and B occurring? The addition rule of probability is written as:
P(A U B) = p(A) + p(B) - p(A ∩ B)

Thus, to find the probability of event A or event B (or both) occurring, you add the simple probability of
event A to the simple probability of event B, and subtract the joint probability of both events A and B
occurring. Note that p (A ∩ B) will be zero if the two events, A and B, are mutually exclusive.
Say that the researcher wants to calculate the probability of selecting a student who has engaged in
intercourse (event A) or is politically liberal (event B), or had intercourse and is liberal. The probability of
selecting a student who has engaged in intercourse is 600/1000 = 0.6. The probability of selecting a
student who is a political liberal is 700/1000 = 0.7. The joint probability of selecting a student who has had
intercourse and is a political liberal is 500/1000 = 0.5. Thus, the probability of selecting a student who has
engaged in intercourse or a liberal student is:
P(Had Intercourse U Liberal) = 0.6 + 0.7 - 0.5 = 0.8
Thus, there is an 80 percent chance of selecting at least a student how has had intercourse or is a
politically liberal student.
You subtract the joint probability of events A and B because the two events A and B may not be mutually
exclusive. If events A and B are not mutually exclusive, which is the case here because you can have
liberal students who have had intercourse, then the sum of the probabilities of events A and B might be
greater than 1, which would also be the case here. Remember, probabilities can range from between 0 to
1 only. To correct for this you must factor out the joint probability of both events. But again, if events A
and B are mutually exclusive, then p(A ∩ B) will be equal to 0, because being mutually exclusive means
that events A and B do not occur together; hence, the frequency of events A and B occurring together will
be 0.
-129-
Back to TOC
9.6.5 Multiplication Rules of Probability The multiplication rule is used to calculate the probability
of selecting two events, A and B, like the joint probability. In some ways the multiplication rule is just
another manifestation of finding joint probabilities, but can also be used to express relationships or lack
thereof between events. There are two ways that the multiplication rule can be applied: (1) when events
are independent, and (2) when events are not to be independent.
If two events, A and B, are independent and there is no relationship between the events, such that P(A) =
P(A|B), then the following multiplication rule can be applied to events A and B, to determine the
probability of observing both events:
P(A ∩ B) = p(A)p(B)
Thus, if two events, A and B, have no known relationship between them; to find the probability of both
events occurring [p(A , B)], simply multiply the simple probability of event A by the simple probability of
event B. As an example, say that we want to know the probability of rolling a ‘2’ on a six-sided die (event
A) and flipping a ‘tails’ on a two sided coin (event B). The joint probability of these two events can be
found by applying the multiplication rule for independent events:
P(2 ∩ tails) = P(2)P(tails)

P(2 ∩ tails) = (1/6)(1/2)
P(2 ∩ tails) = (0.167)(0.5)
P(2 ∩ tails) = .084
As another example, what is the probability of flipping tails on a two-sided coin (event A) and then flipping
heads on the same coin (event B). If the outcome of the second flip does not depend on the first flip we
can answer this by applying the multiplication rule for independent events:
P(tails ∩ heads) = P(tails)P(heads)

P(tails ∩ heads) = (1/2)(1/2)
P(tails ∩ heads) = (.5)(.5)
P(tails ∩ heads) = .25
Using the contingency table from above, what is the probability that the researcher would select a
politically conservative student, return that student to the population, and select another politically
conservative student? Because the researcher sampled with replacement, N does not change, hence we
can answer this by applying the multiplication rule for independent events:
P(conservative ∩ conservative) = P(conservative)P(conservative)

P(conservative ∩ conservative) =(300/1000)(300/1000)
P(conservative ∩ conservative) =(0.3)(0.3)
P(conservative ∩ conservative) =0.09
Note that this works only when sampling with replacement. If sampling without replacement, then the
probability of selecting the second conservative student does depend on selecting the first student. In this
case, the two events are not independent and you would have to apply the multiplication rule for non-
independent events, which we discus next.
-130-
Back to TOC
If two events A and B are not independent and the events have some relationship, such that P(A) ≠ P(A |
B), then the following multiplication rule can be applied to events A and B, to determine the probability of
observing both events:
P(A ∩ B) = P(B)P(A|B)
The probability of observing two non-independent events A and B is equal to the probability of event B
multiplied by the conditional probability of event A given event B. Also, note that:
P(A ∩ B) = P(A)P(B|A)
Thus,
P(B)P(A|B) = P(A)P(B|A)
From the contingency table, say the researcher wants to calculate the probability of selecting a student
who had intercourse (event A) who is politically conservative (event B). Essentially, this is asking for the
joint probability of selecting freshmen student who has intercourse and is a conservative. This question is,
of course easily answered by dividing the number of conservative freshmen who have not had intercourse
(100) by the total frequency (1000); thus, p(Yes , Conservative ) = 100/1000 = 0.1. But the multiplication
rule for non-independent events is another way to answer the question.
From the contingency table, the probability of selecting a conservative freshmen student is
P(Conservative) 300/1000 = 0.3 and the conditional probability of having had intercourse (event A) given
the student is conservative (event B) is P(Yes | Conservative) = 100/300 = 0.333. Using these two
probabilities and the multiplication rule we can determine the probability of selecting a freshmen student
who has not had intercourse and is a Conservative:
P(Yes ∩ Conservative) = P(Conservative)P(Yes | Conservative)

P(Yes ∩ Conservative) =(0.3)(0.333)
P(Yes ∩ Conservative) = 0.1
Note that this is equal to the joint probability calculated above; that is, P(Yes ∩ Conservative ) = 100/1000
= 0.1.
Now say that the researcher is interested in the probability of selecting a politically conservative student,
not returning that student to the population, and then selecting another politically conservative student?
Because the researcher sampled without replacement, N does change and the probability of selecting the
second conservative student depends on selecting the first conservative student. We can answer this by
applying the multiplication rule for non-independent events. But first, here is the contingency table prior to
selecting anyone:
Political Attitude
Yes 500 100 600
No 200 200 400
Totals 700 300 N = 1000
If the researcher selects a conservative student and does not replace that student to population, the
resulting contingency table would look like this:
Political Attitude
Yes 500 ?? (99 or 100) 600
No 200 ?? (199 or 200) 400
Totals 700 299 N = 999
-131-
Back to TOC
Notice, that the total number of conservative students is now 299. Since we do not know whether the
conservative student who was selected had intercourse or not, I have simply indicated that there may be
the original number of those students or one less. But that point is not important here.
P(conservative 1 ∩ conservative 2) = P(conservative 2)P(conservative 2 | conservative 1)

P(conservative 1 ∩ conservative 2) = (300/1000)(299/999)
P(conservative 1 ∩ conservative 2) = (0.3)(0.299)
P(conservative 1 ∩ conservative 2) = 0.0987
Note that this does round to 0.09, but I wanted to show that there is a difference between sampling with
and sampling without replacement.
9.6 Bayes’ Theorem

From the preceding section, the multiplication rule for non-independent events can be algebraically
rearranged to solve for the conditional probability of event A given event B, using the joint probability of
events A and B divided by the probability of event B. Thus,
P(A ∩ B) = P(B)P(A | B)
This can be rearranged into:
PA  B
PA | B 
PB
Thus, the conditional probability of selecting event A given you have selected event B, is equal to the joint
probability of selecting both events A and B divided by the probability of selecting only event B. In cases
where you do not know P(A ∩ B), you can apply the multiplication rule for non-independent events. That
is, we know that:
P(A ∩ B) = P(A)P(B | A)
By plugging this produce into the numerator from our rearranged expression above, we get:
PA PB | A 
PA | B 
PB
This expression is one form of Bayes’ Theorem, which is used for computing the conditional probability
of an event A given event B. It is another way of calculating a conditional probability of an event when you
know something about the other events.
For example, say that we want to calculate the probability of selecting a student who has had intercourse
(event A) given they are a conservative student (event B); thus, we want to know p(Yes | Conservative),
which from earlier was 100/300 = 0.333. We can also find this probability using Bayes’ Theorem:
PYes PConservative | Yes 

PConservative
-132-
Back to TOC
I have reproduced the contingency table below, so it will be easier to see where the various probabilities
are coming from.
Political Attitude
Yes 500 100 600
No 200 200 400
Totals 700 300 N = 1000
First, we need to know the probability of selecting a student who has had intercourse, which is p(Yes) =
600/1000 = 0.6. Next, we need to know the conditional probability of selecting a conservative student
given they have engaged in intercourse, which is p(Conservative | Yes) = 100/600 = 0.167. Finally, we
need the probability of selecting a conservative student, which is p(Conservative) = 300/1000 = 0.3.
Plugging these values into Bayes’ Theorem, we get:
PYes PConservative | Yes 

PConservative
0.60.167 
0.3
0.1
0.3
PYes | Conservative  0.333
This was the conditional probability calculated earlier. The important point to remember is that you can
use relationships between variables and events to calculate probabilities of observing specific outcomes
from the combinations of events. Consequently, you can use the simple, conditional, and joint
probabilities in such ways to determine if relationships between variables exist. Indeed, this is just one of
the many uses of Bayes’ Theorem, which we will likely cover in more detail during lecture.
9.7 Back to Monty Hall

Now that you know a little about probability and understand the most important thing when calculating a
probability is to take information into account, let’s go back to the Monty Hall problem. Remember, after
you make your initial selection and Monty Hall opens a door exposing a goat the probability of winning by
staying with your original choice of door is 1/3 and the probability of winning through switching to the
other unopened door is 2/3. Why...
The table below displays the Monty Hall problem. That is, there are three doors: Door 1, Door 3, and Door
3. A prize is hidden behind one door (assume Door 1) and goats are hidden behind the other two
doors. The probability that you choose any one of the three doors is 1/3, thus the probability of winning on
your first selection is 1/3. The probabilities change only after Monty Hall opens a door.
You Pick Result Result

Door 1 Door 2 Door 3 p(Scenario)
↓ of Switch? of Stay?
Door 1 You MH opens Lose Win 1/6
Door 1 You MH opens Lose Win 1/6
Door 2 You MH opens Win Lose 1/3
Door 3 MH opens You Win Lose 1/3
Say you initially choose Door 1, which is the door that hides the prize. In this case Monty Hall can open
Door 2 or Door 3 to expose a goat and the probability that Monty Hall opens either door is 1/2. Hence,
-133-
Back to TOC
from the 1/3 chance you select Door 1, there is 1/6 chance Monty Hall will open Door 2 and a 1/6 chance
he will open Door 3. This 1/6 comes from (1/3)(1/2) = 1/6. In either case, if you stay with Door 1 (the door
with the prize) you win if you stay and you lose if you switch. Thus, the probability of winning by staying
with you first choice is 1/3, if you have indeed selected the correct door on the first try.
On the other hand, say you initially pick Door 2, which hides a goat. In this case, because the initial pick
was a door with a goat Monty Hall is forced to open the door that hides the other goat. Remember, Monty
Hall knows which door hides the prize! Thus his opening a door gives you information about where the
prize might be hidden. From the 1/3 chance you select Door 2, the probability that Monty Hall will open
Door 3 to expose the other goat 1.0. This is because, if you have selected a door with a goat, Monty Hall
cannot open up the prize door and he cannot open your door, he must open up the only other door with a
goat. In this case if you switch you win, but if you stay you lose.
Finally, say that you initially pick Door 3, which also hides a goat. Again, because your initial pick was a
door with a goat Monty Hall is forced to open Door 2, which hides the other goat. From the 1/3 chance
you select Door 3, the probability that Monty Hall opens Door 2 is 1.0. In this case if you switch you win,
but if you stay you lose.
In the end, there is a 1/3 chance that you correctly select the door with the prize as your first choice. In
this case, if you stay you win and if you switch you lose. There is a 2/3 chance that you will initially select
a door with a goat. In this case, if you stay you lose and if you switch you win. Thus, there is a 2/3 chance
of winning by switching and a 1/3 chance of winning by staying. This, in this non-intuitive scenario, it is
beneficial to switch your initial choice, because this maximizes your chance of winning the prize. Indeed,
if you look at the statistics from the game show “Let’s Make a Deal”, people won after a switch about 2/3
of the time, and people won from staying about 1/3 of the time.
The point is that information about events, whether subtle as in the case of Monty Hall or more overt,
provides information that can seriously alter a probability structure; and this information should always be
taken into account.
-134-
Back to TOC
1. What is the difference between sampling with replacement versus sampling without replacement?
2. For each of the following, assume you are selecting cards from a standard deck of 52 playing cards.
a. What is the probability that a randomly selected card is a jack?

b. What is the probability that a randomly selected card is a king or a jack?
c. What is the probability that a randomly selected card is a four or an ace?
d. What is the probability that a randomly selected card is an eight or a heart?
e. What is the probability that a randomly selected card is a seven and a queen?
f. What is the probability that a randomly selected card is a diamond and a six?
g. What is the probability that a randomly selected card is a diamond, a king, or a spade?
3. Use this information to answer the questions that follow: Jack has a bag of magic beans. His bag has
contains 5 red beans, 2 blue beans, 10 green beans, 7 black beans, and 6 sparkled beans. Based on this
information, answer the following questions, and be sure to incorporate jack's prior actions selections into
your answer for each successive question.
a. Jack reaches into his bag, selects a bean, looks at it, and replaces it. What is the probability this
was a blue bean?
b. Jack reaches back into his bag, selects a bean, looks at it, and eats it. What is the probability this
was a red bean?
c. Assume that the bean that jack selected in b was green. Jack reaches back into his bag, selects
a bean, looks at it, and then replaces it. What is the probability that jack selected a black bean?
d. Jack reaches back into his bag, selects a bean, looks at it, and eats it. What is the probability this
was a green bean?
e. Assume that the bean that jack selected in d was green. Jack reaches into his bag and pulls out a
bean, looks at it, and eats it. What is the probability that this bean was a sparkled bean or a black
bean?
Use this information to answer Exercises 4 – 18: A political scientist interviewed 500 professors at
university to study the relationship between area of scholarship and attitudes toward embryonic stem cell
research. The following contingency table was observed:
Attitude toward Stem Cell Research

Area of Scholarship Favors Opposes No Opinion Totals:
Natural Sciences 90 20 10 120
Behavioral Sciences 60 20 30 110
Theology 30 110 10 150
Humanities 20 50 50 120
Totals: 200 200 100 nT = 500
4. Are levels of the independent variable ‘Attitude toward stem cell research’ mutually exclusive?
5. What is the probability that a randomly selected individual favors stem cell research?
6. What is the probability that a randomly selected individual has no opinion on stem cell research?
7. What is the probability that a randomly selected individual studies the natural sciences?
8. What is the probability than a randomly selected individual studies theology?
9. What is the probability than a randomly selected individual studies the behavioral sciences and has no
opinion on stem cell research?
-135-
Back to TOC
10. What is the probability that a randomly selected individual favors stem cell research given the
individual is a theologian?
11. What is the probability that a randomly selected individual is a theologian given the individual favors
stem cell research?
12. Why are the probabilities in 10 and 11 different?
13. Are being a natural scientist and opposing stem cell research independent? Why or why not?
14. What is the probability that a randomly selected individual studies the behavioral sciences and also
favors stem cell research?
15. What is the probability that a randomly selected individual is a theologian and opposes stem cell
research?
16. What is the probability that a randomly selected individual studies in the humanities or favors stem
cell research?
17. What is the probability that a randomly selected individual is a behavioral scientist or has no opinion
on stem cell research?
18. What is the probability that a randomly selected individual favors stem cell research or opposes stem
cell research?
Use the following information to complete Exercises 19 – 32: A political scientist interviewed 500 people
to study the relationship between political party identification and attitudes toward public sector
unionization. The following contingency table was observed:
Attitude Toward Public Sector

Unionization
Political Party
Favors Opposes Totals
identification
Democrat 160 40 200
Republican 40 160 200
Independent 90 10 100
Totals 290 210 N = 500
19. What is the probability than an individual favors public sector unionization?
20. What is the probability than an individual opposes public sector unionization?
21. What is the probability than an individual is a Democrat?
22. What is the probability than an individual is a Republican?
23. What is the probability than an individual is an Independent?
24. What is the probability that an individual favors public sector unionization given the individual is a
Democrat?
25. What is the probability that an individual is a Democrat given that the individual favors public sector
unionization?
26. Are being a Democrat and opposing public sector unionization independent? Why or why not?
-136-
Back to TOC
27. What is the probability that an individual is a Republican who favors public sector unionization?
28. What is the probability that an individual is a Republican who opposes public sector unionization?
29. What is the probability that an individual opposes public sector unionization given they are a
Republican?
30. What is the probability that an individual is an Independent who opposes public sector unionization?
31. What is the probability that an individual is a Republican or favors public sector unionization?
32. What is the probability that an individual is an Independent or opposes public sector unionization?
Use the following information to complete Exercises 33 – 41: I asked n = 1000 students the following
questions: (a) are you an underclassman (freshman/sophomore) or are you an upperclassman
(junior/senior); and (b) do you own the latest Tim McGraw album. Use this contingency table to answer
the following questions:
“Own the Latest Tim McGraw
Album?”
College Year Yes No
Underclassman 300 200
Upperclassman 400 100
33. Is the variable “Own the Latest Tim McGraw album” mutually exclusive? Why or why not?
34. What is the probability that a randomly selected student owns the Tim McGraw album?
35. What is the probability that a randomly selected student is an underclassmen?
36. What is the probability that a randomly selected student owns the Tim McGraw album given that
he/she is an upperclassman?
37. What is the probability that a randomly selected student is an underclassman given that he/she owns
the Tim McGraw album?
38. Is the probability that a randomly selected student owns the Tim McGraw album independent of being
an upperclassman? Why or why not?
39. What is the probability that a randomly selected student owns the Tim McGraw album and is an
upperclassman?
40. What is the probability that a randomly selected student is an underclassman, or owns the Tim
McGraw album, or both?
41. What is the probability that a randomly selected student is an underclassman or is an

upperclassman?
42. Two events, A and B, are mutually exclusive. The probability of event A is 0.200 and the probability of
event B is 0.500, what is the probability of at least one of the two events occurring?
43. The probability of event B is 0.410, and the probability of event A given event B is 0.850. What is the
probability of both events A and B occurring?
44. Two events, A and B, are independent. The probability of event A is 0.35 and the probability of event
B is 0.480. What is the probability of both event A and event B occurring?
-137-
Back to TOC
45. What is the probability of flipping a heads on an unbiased, two-sided coin and then rolling a ‘6’ on an
unbiased, six-sided die?
46. What is the probability of flipping a heads on an unbiased, two-sided coin three times in a row?
47. The probability of some event A is 0.800, the probability of some other event B is 0.700, and the
probability of event B given event A is 0.200. Use Bayes Theorem to solve for the probability of even A
given event B.
48. The probability of some event C is 0.060, the probability of some other event D is 0.950, and the
probability of event D given event C is 0.500. Use Bayes Theorem to solve for the probability of even C
given event D.
49. The probability of some event C is 0.060, the probability of some other event D is 0.950, and the
probability of event D given event C is 0.500. Use Bayes Theorem to solve for the probability of even C
given event D.
50. A patient goes to see his doctor complaining of blurred vision and thinks he has dreaded yellow fickle-
berry disease (DYFBD). Based on the patient’s history, the doctor knows the probability of this patient
having DYFBD is 0.100. The doctor orders a test, which comes back positive for DYFBD with a probability
of 0.300. The doctor also knows that if the patient does have DYFBD, the probability of the test being
positive is 0.900. The test comes back positive for DYFBD. Using Bayes Theorem, what is the probability
that the patient has DYFBD, given the positive test result?
51. A patient goes to see his doctor complaining of blurred vision and thinks he has dreaded yellow fickle-
berry disease (DYFBD). Based on the patient’s history, the doctor knows the probability of this patient
having DYFBD is 0.200. The doctor orders a test, which comes back positive for DYFBD with a probability
of 0.300. The doctor also knows that if the patient does have DYFBD, the probability of the test being
positive is 0.900. The test comes back positive for DYFBD. Using Bayes Theorem, what is the probability
that the patient has DYFBD, given the positive test result?
52. Compute the following permutations and combinations:
a. 5P2 b. 5C2 c. 6P2 d. 4C4 e. 4P3
f. 6C3 g. 4P4 h. 3C2 i. 3P2
53. You conduct a study that involves showing participants six clips. You are concerned that the order in
which you present the movie clips could affect the outcome, so you decide to present them to each
person in a different order. In how many ordered sequences can the movie clips be presented?
54. From #53, you decide that you can only show three movie clips to each subject. How many different
ordered sequences of three movie clips of six movie clips can be presented?
55. You are interested in the effects of five independent variables on a dependent variable, but, you can
study only two independent variables. How many combinations of two variables could you study?
56. From #55, how many combinations of three independent variables are possible?
57. A company has developed ten hot sauces, but only wants to market the three best-tasting sauces. To
identify the best-tasting sauces, the company has people taste the hot sauces and give ratings. But, the
company only wants to have each person taste four of the ten sauces. How many combinations of four
hot sauces can be administered across people?
58. A team of three people from a widget company to assess the widget needs of a client is formed from
a group of 3 managers, 10 analysts, and 15 technicians.
-138-
Back to TOC
a. What is the probability that the team is composed of only analysts?
b. What is the probability that the team is composed of only technicians?
c. What is the probability that the team is composed of only managers?
d. What is the probability that the team is composed of two managers and the third person is
either an analyst or a technician?
e. What is the probability that the team is composed of two analysts and one technician?
-139-
Back to TOC
Chapter 10: Binomial Probability and Estimation
The guy in the comic illustrates the main point of this chapter: determining
the probability of a particular outcome of a random experiment when only two
events are possible over a series of trials. You may be thinking, ‘wait, with
only two events possible, assuming each even is equally probably, isn’t the
probability of either one event just 0.5?’. The answer to your question is,
basically, ‘yes’. When you have two possible outcomes on any trial (e.g.,
heads and tails), and each outcome has an equal probability, then the
probability of observing either outcome on any one trial is 1/2 = 0.5. But, what
this chapter addresses is the probability of a certain number of events over a
certain number of trials (a random experiment).
10.1 Binomial Probability Expression

First, a few terms are necessary. Remember, that a simple random experiment is any procedure where
the outcome is unknown. A binomial experiment is a type of simple random experiment, where only two
mutually exclusive outcomes are possible on any one trial. Those two outcomes are loosely called a
success and a failure. Such trials where only one of two mutually exclusive outcomes is possible are
called Bernoulli trials. For example, flipping a coin is a Bernoulli trial, because only a heads or a tails is
possible. Heads could be defined as a “success” and tails could be defined as the “failure.” Rolling a six-
sided die could be a Bernoulli trial is a roll of “six” was defined as a success and a roll of any other
number was a failure. A person with cancer taking a new experimental type of chemotherapy is a
Bernoulli trial, where the patient being cured is a success and the patient not being cured is a failure.
The binomial probability then is the probability of observing a certain number of successes (r) over a
certain number of independent Bernoulli trials (n). The binomial probability distribution is the
theoretical probability distribution of all numbers of possible successes over a certain number of Bernoulli
trials. Below are three examples of binomial probability distributions (for n = 2 trials and n = 4 trials) where
the probability of success on any on trial is p = .50:
-140-
We’ll work more with the binomial distribution later. Just for now note that it is a theoretical probability
distribution that displays the probability of each number of successful outcomes, from 0 – n, in a binomial
experiment, assuming a certain probability of success on each trial.
The binomial probability expression is used to calculate the probability of obtaining a certain number of
successful outcomes over a certain number of trials. This section deals with this expression.
Consider the following example: You have an unweighted (“fair”) six-sided die that when rolled should
result in one dot, two dots, three dots, four dots, five dots, or six dots on the face side with equal
probability. That is, the probability of observing any number of dots on any one roll of the die is p = 1/6 =
0.167. You roll the die and you try to guess how many dots will be showing. Because the die is
unweighted the probability you will be correct (“success”) on any one roll is p(Correct) = 0.167, and the
probability you will be incorrect on any one roll of the die is p(Incorrect) = 1 - 0.167 = 0.833.
You roll the die ten times, guessing the number of dots before each roll. What is the probability you
correctly predict the number of dots showing on not zero of the ten rolls? How about the probability of
being correct on one out of the ten rolls? How about the probability of being correct on two out of ten
rolls? Many believe the probability of being correct a certain number times out of some larger number of
th
trials is equal to the probability of being correct on any one trial raised to r power. That is, is p(correct on
r
r trials) = p This is not correct for two reasons:
First, this can be used only to obtain the probability of a certain number of outcomes in a row. That is, if
we were interested in the probability of rolling three dots on a die five times in a row, the probability of
5
rolling five threes in a row is 0.167 = (.167)(.167)(.167)(.167)(.167) = 0.000129. But we want to know the
probability of correctly guessing the face of the die on a certain number of trials out of ten; and those
correctly-guessed trials could be sequential or spaced out.
Second, this does not take into account the total trials (rolls of die). That is, we want to know the
probability of a certain number of successful guesses of the die roll out of a certain number of trials. But,
r
p assumes that the number of successes we are interested in is equal to the total number of trials. Of
r
course, we might want to know the probability of five successes out of 10 trials; p cannot help up
calculate that probability.
Thus, simply raising the probability of being correct on any one  n!  r n r

pr | n, p    p q
 n  r !r!
trial to a certain power will not suffice. Instead, we use the
binomial probability expression, which is used to calculate the
probability of r successful outcomes over n trials, given the
probability of success on any one trial is p:
In the binomial probability expression, r is the number of successful outcomes over the n total Bernoulli
trials. The p is the probability of being successful on any one trial; and q is the probability of failure (of
-141-
Back to TOC
being unsuccessful) on any one trial, where q = 1 – p. Notice that the term inside of the brackets is the
formula for determining the number of combinations of a subset of events out of a total number of events.
From the example above: let’s say we want to know the probability of correctly predicting five rolls out of
ten rolls of a die; hence r = 5 and n = 10. The probability of successfully predicting a roll of the die on any
one trial is p = 1/6 = 0.167. This is so, because there are six sides of the die and the probability that you
correctly guess the side that is showing after the roll, by chance, is one in six. Thus, the probability of
failing to predict the roll on any one trial is q = 1 – 0.167 = 0.833. Plugging these values into the binomial
probability expression from above, we have:
 10! 
p5 | 10, 0.167     0.167 5 0.83310 5
 10  5!5!
 3628800 
p5 | 10, 0.167     0.0001290.401074
 120120 
p5 | 10, 0.167   2520.0001290.401074
p5 | 10, 0.167   0.013
The probability of correctly predicting five of ten rolls of a die is 0.013, or the percent chance of
accomplishing this feat is about 1.3%.
Let’s say we now want to know the probability of being correct on r = 6 out of the n = 10 trials. We have:
 10! 
p6 | 10, 0.167     0.167 6 0.83310  6
 10  6!6!
 3628800 
p6 | 10, 0.167     0.0000220.481482
 24 720 
p6 | 10, 0.167   2100.0000220.481482
p6 | 10, 0.167   0.002
Notice that as the number of successes increases, the binomial probability decreases. This is because as
the number of successful outcomes out of a finite number of trials increases, the likelihood of observing
that number of successes will necessarily decrease. The table below shows the probability of being
correct for each of the r possible successes out of n trials. That is, given ten rolls of a die, you can
successfully predict the die roll on 0, 1, 2, 3,...,9, or 10 of those rolls:
Number of Correct Probability Cumulative Probability of being

Predictions (r) Probability correct r times or more
10 .00000002 1.00000000 .00000002
9 .0000008 .99999998 .00000086
8 .00002 .99999914 .0000197
7 .0002 .99998 .00027
6 .0022 .9997 .0025
5 .0131 .9975 .0156
4 .0546 .9844 .0702
3 .1555 .9298 .2257
2 .2909 .7743 .5166
1 .3225 .4834 .8391
-142-
Back to TOC
0 .1609 .1609 1.0000
In the table above, the values in the Probability column are the probabilities associated with observing
exactly that number of successful outcomes out of ten. The values in the Cumulative Probability column
are the probabilities associated with observing that many successful outcomes or less. For example, the
probability of correctly predicting r = 4 rolls of the die or less is 0.9844. The values in the last column
(Probability of being correct r times or more) are the probabilities associated observing that many
successful outcomes or more. For example, the probability of correctly predicting r = 8 rolls of the die or
more is 0.0000197.
The binomial expression can be used in hypothesis evaluation. Hypothesis testing in statistics is all
about determining whether the probability of observing some outcome. Generally, if the probability of
observing something is very low then it is unlikely a random event and something must have caused that
outcome.
For example, in the table above you can see that the probability of correctly predicting eight rolls of the
die out of 10 rolls is p = 0.00002, which is a very low probability. Intuitively, one would assume it is highly
unlikely that a person would correctly predict the roll of a die that many times if the person was just
guessing. Because the probability of such an occurrence is so low, if someone does accomplish this feat,
one might conclude that the person has precognition regarding the outcome of a roll of a die and is able
3
to predict how many dots will be showing on the roll. (Check out Daryl Bem’s work on precognition)
As another example of how the binomial expression can be sued in in hypothesis testing, let’s say we
have a coin and we are trying to determine whether the coin is weighted (biased) or unweighted
(unbiased). Our initial hypothesis should be that the coin is unbiased and will equally often show heads
and tails. The reason for this initial hypothesis is that we have no evidence to suggest that the coin is
biased, and because the coin can only show one of two sides when flipped (heads of trials), the
probability of showing heads should be about ½ and the probability of showing tails should be about ½.
To determine whether the coin is based, we decide to perform a random experiment by flipping the coin
ten times We consider a flip of heads to be a successful outcome (we could also do this for tails), so we
count the number of heads that show up over the ten flips. Under our initial hypothesis, if we assume the
coin is unbiased, the probability of heads on any one flip of the coin is p(heads) = 0.5; thus, if we flip the
coin n = 10 times we should expect, by chance, 10 x 0.5 = 5 heads. That is, if the coin is unbiased and
the probability of heads on any one flip is the same as the probability of a tails on any one flip, then we
should expect an equal number of heads and tails over the after 10 flips of the coin.
Let’s say we end up tossing six heads. Does this mean that the coin is biased toward showing heads?
Maybe, but how likely is this outcome of six heads? This is where the binomial probability expression
comes in. If we calculate the probability of observing r = 6 heads over n = 10 flips of the coin (p = 0.5 and
q = 0.5), we find that:
 10! 
p6 | 10, 0.5    0.55 0.510  5
 10  6 !6!
 3628800 
p6 | 10, 0.5    0.031250.03125
 24 720 
p6 | 10, 0.5  2100.031250.03125
p6 | 10, 0.5  0.205078
3
Some websites on precognition, which does not exist: http://dbem.ws/
http://psychsciencenotes.blogspot.com/2010/11/brief-note-daryl-bem-and-precognition.html
-143-
Back to TOC
There is roughly a 20% chance of tossing six heads over 10 flips of the coin, if the coin is unbiased. Thus,
there is pretty good chance of observing this outcome, even if the coin is unbiased. As such, we decide to
stick with our original hypothesis that the coin is unbiased, because there is not substantial evidence to
suggest that the coin is biased.
Generally, an outcome to be considered statistically significant enough for us to change our original
hypothesis, the probability of the outcome must be equal to 0.05 or less. That is, if the binomial probability
comes out to be 0.05 or less, then the probability of that outcome is so low that the initial hypothesis is not
likely correct, so we should change it. But, this is not the case here.
What if we had thrown nine heads? Would we reject our initial hypothesis that the coin is unbiased and
change our hypothesis to the coin is biased? In this case, the probability of r = 9 heads over n = 10 flips if
the probability of heads on any one trial is .5 is p = .009766, which is quite low. In this case, because the
probability of this outcome is so unlikely if the coin was unbiased, we would reject that hypothesis and
conclude the coin is actually biased.
10.2 More on Binomial Probability Distribution
The graph to the right plots the binomial probability of

observing r successful outcomes out of n trials, and
the cumulative binomial probability r successful
outcomes or less out of n trials. The plots are based
on the die-rolling example in the previous section. The
abscissa is the number of successful predictions of a
die roll from r values of zero to 10 and the ordinate is
the probability of each value of r. This is an example of
the binomial probability distribution (blue line) and a
cumulative binomial probability distribution (red line)
for n = 10, r = 0 to 10, and p(success) = 0.167.
Importantly, as you increase the size of n the binomial

probability distribution approximates a normal
distribution. You can see this in the blue line in the
graph to the right, which presents the binomial
probability distribution for all values of r from 0 to n =
100, and p(success) = 0.167. Notice the peak of the
binomial probability distribution (blue line) is shifted to
the left of the center on the abscissa, even though the
shape of the distribution is “normal.” On close
inspection you might notice that the peak is at the
value that you would get by multiplying the probability
of successfully predicting the roll of a die on any one
trial [p(success) = 0.167] by the total number of trials
(n = 100). That is, the peak is near the value 0.167 x
100 = 16.7. This is because with a sufficiently large n
the binomial probability distribution will approximate a
normal distribution with a mean that is equal to p x n; a
topic I will turn to in the next section.
-144-
Back to TOC
10.3 Binomial Approximation to the Normal Distribution
With a sufficiently large n (Bernoulli trials) the binomial probability distribution will approximate a normal
distribution. This binomial approximation of the normal distribution has a mean equal to:
  np
which is the expected number of successful outcomes expected by chance. In the expression, n is the
number of Bernoulli trials and p is the probability of a successful outcome on any one trial. The variance
and the standard deviation of the binomial distribution are:
 2  npq   npq
In each expression, n is the total number of trials, p is the probability of success on any one trial, and q is
the probability of failure on any one trial. For example, say that we roll an unweighted die n = 1,000 times.
We would expect the mean, variance and standard deviation to be:
  (1000)(0.167)  167
 2  (1000)(0.167)(0.833)  139.111
  (1000)(0.167)(0.833)  139.111  11.794
Because we know the mean (μ) and standard deviation (σ) of this binomial distribution, we can calculate
a z-score for any conceivable value of r from 0 to n and then use that z-score to determine different
probabilities associated with r. For example, using the mean and standard deviation from above, say we
want to know the probability of successfully predicting 200 rolls or more rolls of the die. We first calculate
the z-Score for r = 200:
X   200  167 33
z    2.798
 11.794 11.794
Next, we look up this z-score in the standard normal tables (Table 1 in the statistics packet and Appendix
A). Once we find the z-score, we use the values in Column 3 to determine the probability of being correct
on 200 rolls of the die or more, which you will find is p = .0026. This probability can be used in hypothesis
testing.
For example, say we were trying to determine whether a person has precognitive abilities (extrasensory
perception, ESP), and we test their precognitive abilities by having them guess the rolls of a die. Our
initial hypothesis is the individual does not have precognition, because there is no reason to believe the
individual does have precognitive abilities. To determine whether the individual has precognitive abilities
we roll the unweighted die 1000 times and have the individual predict each roll. The individual is correct
on 190 of the rolls. As a criterion, we will assume that if the probability of being correct this many times or
more is less than .05, we will conclude that the individual has precognitive abilities. That is, if the
probability of correctly predicting 190 of 1000 rolls of the die is .05 or less, we will consider this number to
be statistically significant and we will change our original hypothesis.
First, we calculate the z-score that is associated with r = 190 successful predictions:
X   190  167 23
z    1.950
 11.794 11.794
-145-
Back to TOC
Like before, look up this z-score in the standard normal tables. Once we find the z-Score we use Column
3 to determine the probability of being correct on 190 rolls of the die or more, which we find is p = 0.0516,
which is just slightly greater than our criterion probability of .05. Thus, because 0.0516 > 0.05, we will
conclude that correctly predicting 190 rolls of a die out of 1000 is not significantly different from predicting
the expected (mean) number of rolls at 167. Therefore, we will not change our original hypothesis and we
conclude that there is insufficient evidence to indicate this individual has precognitive abilities.
For another example, say we want to know the cumulative probability of successfully predicting between
zero and 175 rolls of the die, that is, of being correct on at most 175 rolls. First, calculate the z-Score for r
= 175:
X   175  167 8
z    0.678
 11.794 11.794
Look up this z-Score in the standard normal tables and use the value in Column 5 to determine the
probability of being correct between μ = 167 and r = 175 rolls of the die, which you will find is p = 0.2517.
(This is the probability between the mean and the z-score.) Finally, add that probability to 0.5, which will
give you the cumulative probability of correctly predicting at most r = 175 rolls of the die: p = .2517 +
.5000 = .7517.
1. A student is taking an eight-question true-false test. If the student randomly guesses on each question,
what is the probability that this student:
a. gets eight correct b. gets six correct

c. gets four correct d. gets one correct
2. You have an unbiased six-sided die that you roll ten times, and record the number of dots on the face
side. You consider a roll of thee dots to be a ‘success’ and any other number of dots to be a ‘failure’.
What is the probability of rolling three dots:
a. eight times b. two times

c. four times d. five times
3. Using the same die and information in Exercise 2, say that you roll the die 10,000 times and count the
number of 3's (successes). What is the expected (mean) number of 3's that you will roll after 10,000 rolls
of the die? What is the variance and standard deviation of the binomial distribution based on n = 10,000
rolls?
4. From Exercise 3, say that you roll a three 1750 times. What is the z-Score of this number of 3’s? What
is the probability of rolling this many three's or more?
5. Calculate the mean, variance and the standard deviation for a binomial distribution with n = 150 trials
and the probability of success on any trial is 0.600.
6. Using the information in Exercise 5, determine the probability of:
a. 100 successes or more:

b. 110 successes or more:
c. 84 successes or fewer:
7. Calculate the mean, variance and standard deviation of a binomial distribution with n = 200 and p =
0.800.
-146-
Back to TOC
8. Using the information in Exercise 7, determine the probability of:
a. 175 successes or more:

b. 140 successes or fewer:
9. A student takes a multiple-choice quiz of ten items. Each item has four response options. The student
gets five items correct. Using an alpha level of α = .05 as a criterion, did the student perform at an ‘above
chance’ level?
10. From #9, what if the student had gotten six items correct. Using an alpha level of α = .05 as a
criterion, did the student perform at an ‘above chance’ level?
11. Dr. Claws'n'paws is a biology professor who teaches a lab in biologic evolution. The lab includes 12
students: 8 female students and four male students. One would assume that the probability that male and
female students enroll in this lab is p = .50 Using this information calculate the binomial probability of
having this many female students enroll in the lab. When you are calculating your answer, be sure to do
any calculations out to as many decimal places as necessary to keep your answer non-zero. Is the
probability of having this many females in the lab likely, or unlikely, due to chance?
12. The final exam in your Cyberporn and Society class (look it up, it exists!) contains 25 questions, each
with four answer to choose from. Assume that you did not study for this exam and you end up selecting
one of the four answer choices to each question at random. You need to get 60% of the 25 questions
correct in order to pass the exam. Use this information to answer the following:
a. By chance, what is the probability of selecting the correct answer on any one question?
b. What is the minimum number of questions that you need to answer correctly to pass the exam?
13. The possible letter grades that you can obtain in any class are (from high to low): A, A-, B+, B, B-, C+,
C, C-, D+, D, and F. Both psychology and neuroscience students must take PSYC 330 (Research
Methods), which requires a grade of C or higher in PSYC 210 (Statistics). Hence, a grade of C or higher
should be considered a "success" and a grade of C- and lower should be considered a "failure" for being
admitted into PSYC 330. Let's say that I go a little crazy one year and start awarding course grades to my
PSYC 210 students at random (i.e., I pull a student's grade out of a hollow glass head).
a. For any one student, what is the probability that I will select a grade that is high enough to get
into PSYC 330?
b. Say that I teach 150 students in PSYC 210 in an academic year. How many students should I
expect will get a grade that is acceptable for PSYC 330?
c. Calculate the variance of the binomial probability distribution.
d. Calculate the standard deviation of the binomial probability distribution.
e. What is the probability of this many students receiving a grade of C or higher?
14. I have a Canadian Toonie (a $2 CD coin). Honestly, I have flipped one of these and it has landed on
its edge. Assume that the probability of showing a heads is 0.49, the probability of a tails is also 0.49;
hence, the probability of landing on its edge is 0.02. You consider a flip of the Toonie landing on its edge
to be a "success."
a. If I flip this Toonie 10,000 times, what is the expected number of times the Toonie should land on
its edge?
b. Calculate the binomial probability variance.
c. Calculate binomial probability standard deviation.
d. Say the Toonie lands on its edge 180 times. What is the probability that the Toonie lands on its
edge 180 times out of these 10000 flips?
-147-
Back to TOC
15. Twenty percent of individuals who seek psychotherapy will recover from their symptoms irrespective
of whether they receive treatment (a phenomenon called spontaneous recovery). A researcher finds that
a particular type of psychotherapy is successful with 30 out of 100 clients. Using an alpha level of .05 as a
criterion, what should she conclude about the effectiveness of this psychotherapeutic approach?
Chapter 11: Introduction to Hypothesis Testing
11.1 Inferential Statistics and Hypothesis Testing

Earlier chapters dealt with mainly descriptive statistics and probability, whereas the remaining chapters
will deal mainly with inferential statistics and hypothesis testing. Inferential statistics are tests and
analyses conducted on sample data, the outcomes of which can be used to make inferences about what
would likely be found in a population. The picture of the scale above is to emphasize that hypothesis
testing is all about weighing evidence, and whether the weight of the evidence is sufficient enough to be
considered statistically significant.
For example, say we provide the herbal supplement ginko-baloba to a sample of college freshmen and
give a placebo (an inert pill) to another sample of freshmen. We find that the freshmen students who were
-148-
Back to TOC
taking ginko-baloba have a higher mean GPA at the end of their freshman year compared to the students
who were taking the placebo. Thus, there appears to be a relationship between taking ginko-baloba
versus taking a placebo and GPA. To assess this apparent relationship, we would use inferential statistics
to determine whether difference in GPA between students taking ginko-baloba and students taking the
placebo is statistically significant. Simply put, two things may be numerically different, but it does not
mean that this difference is statistically different.
Inferential statistics boil down to a test-statistic, which is a ratio of the size of a relationship, or effect, to
some measurement of random sampling error. Thus, any test statistic can be summarized as:
effect
test statistic 
error
We generally want the ratio of effect to error to be large, which would indicate the observed effect or
relationship is greater than random sampling error. The question addressed in this and following chapters
is how large the ratio needs to be for a relationship to be statistically significant.
But first, remember that in any study one must explicitly state the predictions of their null and alternate
hypotheses. Recall, the null hypothesis (H0) predicts the expected relationship will not be observed and
the alternate hypothesis (H1) predicts the expected relationship will be observed. From the ginko-baloba
example above, one might predict that taking ginko-baloba, which some believe affects learning and
memory, will affect student GPAs. In this case the null hypothesis would predict that students taking
ginko-baloba will have GPAs that are different than the GPAs of students taking a placebo; and the
alternate hypothesis would predict that students who are taking ginko-baloba will have GPAs that do differ
from those of students who are taking a placebo.
When stating hypotheses, symbols are used to state the null and alternate hypotheses, and they should
be stated in terms of their populations. For example, in the ginko-baloba example above the null and
alternate hypotheses would be written symbolically as (be sure to use subscripts):
H0: μGinko = μPlacebo
H1: μGinko ≠ μPlacebo

Both hypotheses are symbolically stating the predictions described in the preceding paragraphs. That is,
the null hypotheses states that taking ginko-baloba will not lead to a change in GPA relative to taking a
placebo; or, there will be no difference in GPA between students who are taking ginko-baloba and
students who are taking the placebo. In contrast, the alternative hypothesis is stating that there will be a
difference in GPA between ginko-baloba and taking a placebo. The point is that when stating hypotheses,
you should write them symbolically rather than verbally, because meanings of symbolic statements are
universal, whereas words could be misinterpreted.
You may wonder why population symbols are used rather than sample symbols. Remember, inferential
statistics use sample data to make inferences about what should be found in the population from which
the sample came; thus, hypotheses should reflect the inferences, which is population symbols are used.
11.2 Null Hypothesis Significance Testing

(NHST)
The logic to null hypothesis significance testing (NHST) may
seem a little strange, convoluted, and downright flawed at
first. Indeed, there are some scientists who are critical of
hypothesis testing and believe it should be banned, a point I
-149-
Back to TOC
agree with to a certain extent. Null hypothesis significance testing is a tool for making inferences from
known sample statistics about unknown population parameters; and .when used properly, NHST is very
effective. Unfortunately, some people do not take into account many of the parameters of NHST and
decide a relationship exists when the relationship is weak. So, remember that this chapter is all about
how to use this tool properly, so it’s sort of like learning how to properly use a reciprocating saw by not
shopping off a finger.
In NHST, the null hypothesis, not the alternate hypothesis, is being assessed. This seems strange,
because the alternate hypothesis is usually what one wants to confirm; hence, it would seem logical to
test the alternate hypothesis. But if this was so, then a researcher would be seeking evidence only to
confirm predictions and, thus, may become biased in evidence-gathering. In scientific reasoning you want
to disconfirm objections to your prediction; thus, one should seek to disconfirm their null hypothesis. If you
can disconfirm the null, you can use the alternate in its place (I know this already sound's weird...bear
with me).
In NHST, you start by assuming the null hypothesis is true in the population. That is, without any evidence
to the contrary you assume there should be no relationship between the variables under study. This is
synonymous with in the previous chapter where we started out by assuming that a coin was unbiased,
because we had no reason to believe that it was.
You then collect data and apply inferential statistics to determine the probability of observing the
outcome, while assuming the null hypothesis is true. The inferential statistics are used to tell us p(Data |
H0 = True), that is, the probability of observing our data if the null hypothesis was true. Critically, p(Data |
H0 = True) does not tell you the probability that the null is true; it sells us the probability of observing our
data, under certain assumptions. If the probability of observing the data given that the null is true is very
low, then one rejects the null hypothesis. If p(Data | H 0 = True) is low, this means that obtaining the data
you did is unlikely, if the null hypothesis was true. Stated differently, if the probability of observing data is
very low (assuming the null is true), then the null is unlikely correct and we reject the null hypothesis.
Once the null hypothesis is rejected, we accept the alternate hypothesis.
But at what point is p(Data | H0 = True) low enough that the null hypothesis can be rejected? How low
does the probability of an outcome have to be before we can say that the null hypothesis is unlikely to be
true? The accepted level of statistical significance is p = .05 or less, which is called your alpha-level
(α) or, more commonly, the p-value. That is, assuming the null hypothesis is true, if the probability of
observing an outcome is p = .05, or less, then it is unlikely the null is true and the null is rejected.
Researchers use different alpha levels for different reasons, but the generally accepted level of statistical
significance is α = .05 or less. Important: The alpha level is not the probability that the null hypothesis is
true; it is the probability of observing data given the null hypothesis is true. The alpha level says nothing
about the null hypothesis itself.
11.3 Example with Binomial Expression

Recall, the binomial probability expression is used to calculate the probability of observing some specified
number of successful outcomes (r) out of some number of trials (n). As I show below, the binomial
expression can be used in hypothesis testing.
Say we have a six-sided die and we want to determine whether it is weighted in such a way that four dots
appear on the face side more often than the other sides. The null hypothesis in this example predicts that
the number of fours rolled will be no different than other numbers of dots rolled, that is, the number of
fours that are actually rolled will be equal to the number of rolls of fours that are expected by chance. If
the die is rolled ten times and the assumed probability of rolling four dots is p = 0.167, then we would
expect (0.167)(10) = 1.67 rolls of four-dots, by chance. The alternate hypothesis predicts that the number
of fours rolled will exceed the number that is expected by chance. The alternate hypothesis predicts that
the number of rolls is greater than 1.67, because we are trying to see if the coin is biased to show the
four-dot side, which would mean the four-dot side would snow more often. Thus, the hypotheses are:
-150-
Back to TOC
H0: μ4 = 1.67
H1: μ4 > 1.67
We roll the die 10 times and record whether each roll results in a four (success, p = 0.167) or not
(failure, q = 0.833). Say the four dots side shows up six times. In this case, the number of rolled fours and
the number of rolled fours that is expected by chance  10! 
(1.67) differ numerically. But, what we want to address is p6 | 10,0.167   
6 106
 0.167 0.833
whether the six rolls of four-dots and the expected  10  6 !6!
number of four-dots (1.67) differ statistically. Using the  3628800 
binomial expression, we calculate the binomial probability p6 | 10,0.167    24720  0.0000220.481482
 
of rolling the four-dot side six times:
p6 | 10,0.167   2100.0000220.481482
p6 | 10,0.167   0.002
The probability of rolling four-dots six times is 0.002. This
is less than p = .05, which is the conventional alpha-level. In this case, we can say that the number of
four-dots sides rolled is statistically greater than the number of four dot sides that were expected by
chance. This is a statistically significant outcome and the null hypothesis can be rejected and the
alternate hypothesis can be accepted. In layman’s terms, we would conclude that that die is biased
toward rolling four-dots.
11.4 The z-Test

The z-Test compares a sample mean to a population mean (μ) to determine whether the difference
between means is statistically significant. To use the z-Test, a sample must be drawn from a population
with a known mean (μ) and known standard deviation (σ). The z-Test calculates a z-Score for the sample
mean with respect to the population mean. The greater the difference between the sample mean and the
population mean, the larger the z-Score will be and the less likely the sample mean will be associated
with the population mean.
Staying with the ginko-baloba example, let’s say we predict that taking ginko-baloba will have “some
effect” on freshman GPA. We randomly sample n = 25 freshmen who are taking ginko-baloba. At the end
of the students' freshman year we ask for their GPA and we find this sample has a mean GPA of 3.20
(about a B+). Let's say we also know that the mean GPA of all freshmen (including the 25 students in the
sample) at this particular university is μ = 2.80 (about a B-), with a standard deviation of σ = 1.0. Because
we know μ and σ, we can use the z-Test to determine whether the difference between the sample mean
and μ is statistically significant.
The first step in the z-Test, which is the same for any inferential test, is the state your hypotheses. But
first, I must distinguish between directional (one-tailed) and non-directional (two-tailed) hypotheses,
which affects how the null and alternate hypotheses are stated. Directionality is related to the alternate
hypothesis: A directional alternate hypothesis predicts the sample mean will be, specifically, greater than
or less than the population mean. Hence, the sample mean is predicted to lie in a specific direction
relative to μ, that is, in one of the two tails of the distribution around a mean. In contrast, a non-directional
alternate hypothesis predicts the sample mean will be different from the population mean. Hence, the
sample mean could be greater-than or less-than μ, it doesn't matter; that is, the sample mean could fall
into either one of the two tails of the distribution around a mean. For exposition purposes, I go through
this example using both types of hypotheses simultaneously.
From the ginko-baloba example above, a directional alternate hypothesis might predict that students who
are taking ginko-baloba will have a mean GPA greater than the mean GPA of all freshmen students (μ). A
non-directional alternate hypothesis would state that the mean GPA of students taking ginko-baloba will
be different from the mean of all freshmen. Symbolically, these hypotheses are:
Directional: H0: μGinko = 2.80 H1: μGinko > 2.80
-151-
Back to TOC
Non-directional: H0: μGinko = 2.80 H1: μGinko ≠ 2.80
Before continuing I must say something about the alpha level and directionality. Remember, the alpha-
level is the probability associated with statistical significance, that is, the probability of observing an
outcome assuming the null is true. In this example we’ll use the conventional level of significance by
choosing an alpha-level of α = .05. First, the alpha-level is a probability; it’s the probability associated with
statistical significance. Second, this probability is an area under a distribution, just like we covered in
Chapter 7. Specifically, the alpha-level is an area under the tail end of a distribution. So far so good, the
alpha level is just a probability under a distribution.
Here’s where things get confusing. If you have a directional alternate hypothesis, which is also called a
one-tailed hypothesis test, the alpha-level is an area in only one of the two tails of the distribution. But, if
you have a non-directional alternate hypothesis, which is also called a two-tailed hypothesis test, the
alpha-level is an area under both tails.
For example, assume you predict a sample mean will fall in the upper tail of a distribution. In this case,
the alpha-level will be associated with an area under the right tail of the normal distribution:
Retain H0 ← Decision → Reject H0

Probability
The probability of being in this area is equal to your alpha-level (α = .05 in this example) and this area is
called the critical region. If the sample mean falls in this area we reject the null hypothesis and accept
the alternate hypothesis and state there is a statistically significant difference between the sample mean
and the population mean.
But what about a non-directional alternate hypothesis? In this case, the alpha-level remains the same
value (α = .05 in our example), but this probability is split between the two tails, that is, the alpha-level is
divided by two (α/2 = .05/2 = .025) and this probability is assigned to critical regions located in each tail
(see below). The probability of being in one critical region is .025, but the probability of falling in either
critical region is .05. It is more difficult to be in a critical region for a non-directional test (the areas are
smaller). This means a sample mean must be farther from the population mean for it to be significantly
different from the population mean.
-152-
Back to TOC
Reject H0 ← Decision → Retain H0 Retain H0 ← Decision → Reject H0
Probability
α/2 α/2
These concepts are important in determining statistical significance, because you must take directionality
into account when assessing the outcome of an inferential test. Specifically, the probability associated
with your test-statistic must be less than your chosen alpha level (.05) in order to claim there is a
statistically significant difference between a sample mean and a population mean.
The z-Test calculates the z-Score of a sample mean with respect to a population mean. The z-Test is
simply the difference between a sample mean and its population mean divided by the standard error of
the sampling distribution of the mean (standard error of the mean):
X 
z
X
From our ginko-baloba example, the sample mean was 3.20, the sample size was n = 25. The population
mean was μ = 2.8, and the population standard deviation was σ = 1.0. The standard error of the mean in
this example is:
 1.0 1.0
X     0.2
n 25 5
Plugging the sample mean, the populations mean and standard error of the mean into the z-Test, we
obtain:
X   3.20  2.80 0.40
z   2
X 0.2 0.2
This value (2) is the obtained value (zO) or test statistic, and tells us that the sample mean (3.20) is two
standard errors above the population mean of 2.80. This test statistic value is what we use to determine
whether the difference is statistically significant, and whether we should reject the null hypothesis.
Next, we determine whether the outcome of the z-Test (zO) is statistically significant. To do this, look up
the obtained value of z = 2 in the z-Tables, a portion of which is reproduced, below. We need to
determine the probability associated with z = 2. (We’ll do this for both a directional and a non-directional
alternate hypothesis.) For a directional hypothesis, look up the probability in Column 3 associated with
your zO. This is the p-value that is associated with your test-statistic (zO = 2). If this p-value (.0228) is less
-153-
than your alpha-level (.05) then then the difference between the sample mean and the population mean is
statistically significant and the null hypothesis is rejected and the alternate hypothesis is accepted.

1.95 0.9488 0.0256 0.0512 0.4744
1.96 0.9500 0.0250 0.0500 0.4750
1.97 0.9512 0.0244 0.0488 0.4756
1.98 0.9523 0.0239 0.0477 0.4761
1.99 0.9534 0.0233 0.0466 0.4767
2.00 0.9545 0.0228 0.0455 0.4772
2.01 0.9556 0.0222 0.0444 0.4778
2.02 0.9566 0.0217 0.0434 0.4783
2.03 0.9576 0.0212 0.0424 0.4788
2.04 0.9586 0.0207 0.0414 0.4793
2.05 0.9596 0.0202 0.0404 0.4798
For a non-directional hypothesis, look up the probability in Column 4 associated with your zO. This is the
p-value that is associated with your test-statistic (zO = 2). If this p-value (.0455) is less than half your
alpha-level (.05/2 = .025) then then the difference between the sample mean and the population mean is
statistically significant and the null hypothesis can be rejected and the alternate hypothesis can be
accepted. (As a rule of thumb: Whenever you have a statistically significant outcome, you reject the null
hypothesis and accept the alternate hypothesis. Whenever you have a non-significant outcome you
retain the null hypothesis only and do nothing with the alternate hypothesis. It is a little more complicated
than this, but we’ll cover it in more depth in class.)
In this example, the p-values for both the directional (.0228) and non-directional (.0455) alternate
hypotheses were less than the alpha level of .05. Thus, for either alternate hypothesis there was a
significant difference between the sample mean and the population mean; hence, we can reject the null
hypothesis and accept the alternate hypothesis. In layman's terms for both the directional and non-
directional hypotheses, we conclude that students who are taking ginko-balboa have a significantly higher
mean GPA at the end of their freshman year compared to the mean population GPA of all freshmen.
11.5 Reporting the z-Test in the Literature

When writing up the results of a z-Test in a manuscript for publication, several parameters need to be
reported to convey just enough information to the reader: (1) the sample mean, (2) the population mean
or the value that the sample mean is being compared to, (3) the obtained z-Value, (4) the standard error
of the mean, and (5) the p-value. There are variants to exactly how the outcome of a z-Test (or any
inferential test) is up depending on the journal. But, here is how you would report just the results of the z-
4
Test: z = 2.00, SEM = .20, p = .0228 (two-tailed). The SEM refers to the standard error of the mean.
Below, I present an example of how the results of the non-directional z-Test reported in Section 11.4
th
would be written up in a manuscript (this is based on APA 6 edition style):
Results
In the present study, n = 25 university freshmen who were taking ginko-baloba every day during
their freshmen year were randomly selected from all students identified as taking ginko-baloba. Each
student's GPA was recorded at the end of the freshman year, and this sample was found to have a GPA
of M = 3.20. The mean GPA of all university freshmen at the end of the same academic year was μ =
4
There are several ways of displaying the alpha level. This is one way of displaying the results assuming
a non-directional alternate hypothesis.
-154-
2.80 (σ = 1.0). A z-Test comparing two means resulted in a statistically significant difference between the
sample mean and population mean, z = 2.00, SEM = .20, p = .0228 (two-tailed). Thus, students taking
ginko-baloba had a significantly larger GPA than all other freshmen.
11.6 Type I vs. Type II Errors

Is hypothesis testing perfect? Is the outcome of a study going to be found every time a study is
conducted? Sadly, the answer is no. It is entirely possible to make a mistake in hypothesis testing,
because null NHST is based on probabilities, which are not absolute. Remember, the chose α level is the
probability of observing the outcome that you did if the null hypothesis is true. Thus, α = .05 means that
there is a very small chance that you would observe the outcome with the null being true, but there is still
a small chance you would observe this data even with the null being true. In short, because hypothesis
testing is based on probabilities of being correct or incorrect, there is always going to be some probability
that your decision can be wrong.
In hypothesis testing, if the test statistic is greater than or equal to the critical value, the null hypothesis is
rejected; otherwise, the null hypothesis is retained. Thus, there are two possible decisions that can be
made about the null hypothesis, regardless of whether the null is actually true or false. Independent of
your statistical decision, in reality (in the population) the null hypothesis may be true or may not be true.
Remember, in hypothesis testing you use sample data to make inferences about what you would expect
to find in a population. It is possible that you conclude a sample mean is statistically different from a
population mean, but in reality there is no difference. Thus, if you able to test an entire population you
may find that the null hypothesis is really true or is really false. Thus, there are four combinations of your
two statistical decisions regarding the null hypothesis and two possible realities regarding the null
hypothesis. These are illustrated in the table below:
H0 is Actually True in H0 is Actually False in

Population Population
Type I Error Correct Decision
Decide to Reject H0
p=α p = (1 - β) = Power
Correct Decision Type II Error
Decide to Retain H0
p=1-α p=β
From the table, you can see that there are two possible situations that lead to a correct decision and two
possible combinations that lead to decision errors. I’ll focus on the errors: A Type I Error is committed
when the null hypothesis is rejected, but in reality, the null hypothesis is true and should be retained. The
probability of committing this error is equal to the selected α level. To decrease the chance of making a
Type I Error, a smaller α can be selected (e.g., .01, .001, .0001, etc.) and used as a basis for
retaining/rejecting the null. A Type II Error, is committed when you fail to reject the null hypothesis (retain
the null hypothesis), but in reality the null hypothesis is false and should be rejected. The probability of
making a Type II Error is equal to something called beta (β). The probability of not making a Type II Error,
that is, correctly rejecting a false null hypothesis, is equal to 1 – β. This is the power of a statistical test.
Generally, you want to have high power in your inferential test; that is, a high probability of correctly
rejecting a false null hypothesis. Typically, Power should be .80 of more (up to 1.0).
11.7 Statistical Power

The power of a statistical test is the probability of correctly rejecting a false null hypothesis. There are
methods for calculating the value for β and power, which are covered in Appendix C and in later chapters,
but for now we address how statistical power is influenced by several factors.
-155-
Back to TOC
First, smaller values of the α-level decrease statistical power. Specifically, as the value of α decreases the
chance of making a Type I Error decreases, but so too does statistical power. This is because smaller
alpha-levels are associated with larger critical values; thus, holding everything constant, a statistically
significant difference between means is less likely to be detected as the critical value increases.
The second factor that influences power is sample size (n). Holding everything constant, as sample size
increases the power of a statistical test increases. Indeed, this is why many researchers prefer larger
samples to smaller samples. As you increased the size of a sample, the sample mean becomes a better
estimator of the population mean and the standard error of the mean decreases. Thus, as you increase
the sample size, you decrease the standard error of the mean and increase the chance of detecting a
significant result, because the test statistic should increase. If you increase the sample size you decrease
the standard error of the mean, which is the denominator in the z-Test; hence, you increase the obtained
z-Value.
Third, the effect size, that is, the mean difference between the population mean and the sample mean in
the numerator also influences power. Generally, larger effect sizes are associated with more statistical
power. Holding everything constant, as you increase the effect size you increase the obtained z-Value
and you increase the chance of detecting a significant difference.
The fourth factor that influences power is error; that is, the standard deviation (σ). Larger standard
deviations (more variance) lead to less power. This can be seen in the re-written z-Test above. Holding
everything else constant, as the standard deviation increases the denominator will increase and z O will
decrease; thus, you are less likely to detect a significant result. Hence, larger standard deviations (more
error) results in less statistical power. Of course, it is unlikely that the population standard deviation will
change that much, so there is little to worry about here. Indeed, the three other factors that affect power
are more or less under your control.
11.8 Limits of z-Test

There is a major limitation to the z-Test, and that is it can be used only if both μ and σ are known, which is
unlikely. In cases where the population mean is known, but the population standard deviation is not, if a
sample is sufficiently large, it is acceptable to use the estimated population standard deviation in place of
the true population standard deviation to calculate an estimated standard error of the mean. However,
there is no 'real' criterion for how large a sample must be before it is “sufficiently” large. Another limitation
is that the z-Test only allows a comparison of a sample mean to a known population mean. In many
cases you might have two sample means that you want to compare. With the z-Test this is not technically
possible (though I have seen several books try to get around this). Thus, the z-Test is hampered, which is
what statisticians developed something called the t-Test, which we turn to in the next several chapters.
11.9 The “Classic” NHST Procedure
There is an alternative method for determining statistical significance to the method that was just
described in the preceding sections. This is the “classic” method that is taught in nearly every
undergraduate statistics book and assesses statistical significance by comparing the obtained test
statistic to a pre-determined critical value. This section goes through this alternative approach, for
historical value. The first step in this procedure is the same as before, that is, state the hypotheses:
Directional: H0: μGinko = 2.80 H1: μGinko > 2.80

Non-directional: H0: μGinko = 2.80 H1: μGinko ≠ 2.80
Once the hypotheses are stated you determine what outcome is considered statistically significant, that
is, what is the smallest test statistic that must be obtained for the outcome to be considered “statistically
-156-
Back to TOC
significant?” To do this you select an alpha-level and determine the necessary critical values for
rejecting the null hypothesis. In this example we’ll use the conventional alpha level of α = .05.The critical
values are the boundaries of the critical regions, that is, it is the point on the x-axis where the decision line
falls for whether to reject the null hypothesis or retain the null hypothesis:
Retain H0 ← Decision → Reject H0

Probability
For directional hypotheses there is one critical value and one critical region and for non-directional
hypotheses there are two critical values and two critical regions. Because the critical regions are areas
associated with the selected alpha-level, critical values must be z-scores associated with the alpha-levels.
The critical values are found in the standard normal table. Importantly, the critical z-Value (zα) is
determined by the alpha level (α) and whether you have a directional or a non-directional hypothesis.
From our example, to find the critical value for a non-directional hypothesis: lookup the alpha-level (.05) in
Column 4 and use the z-Score in Column 1 as the critical value. The critical z-Value for a non-directional
test with α = .05 is zα = ±1.96. The plus/minus is necessary, because the sample mean could be greater-
than or less than the population mean. To find the critical z-Value for a directional hypothesis: lookup the
alpha-level (.05) in Column 3 and use the z-Score from Column 1 as the critical value. From Table 1, p =
.05 does not occur in column 3 and the two closest probabilities are .0505 and .0495. The rule of thumb is
use the smaller of the probabilities. Thus the critical z-Value for a directional test with α = .05 is zα =
+1.65. The plus/minus is not necessary for a directional test; however, because the sample mean was
predicted to be greater than the population mean, you must place a plus sign in front of the critical value.
If the sample mean was predicted to be less than the population mean, which means you are predicting a
negative outcome in the z-Test, then you must place a negative sign in front of the critical value. The next
step is to conduct the inferential test. From the earlier sections, here are the standard error of the mean
and the test statistic:  X  0.2 and z  2 .
To determine whether this outcome is statistically significant, compare the test statistic (z = 2) to the
critical value of zα = ±1.96. If the absolute value of the test statistic is greater than or equal to the absolute
value of the critical value, the difference between the sample mean and population mean is statistically
significant. Because zα for the non-directional hypothesis in the ginko-baloba example is zα = ±1.96, the
test statistic of z = 2 is greater than the critical value and difference between the sample mean and the
population mean is statistically significant. Because zα for a directional hypothesis was zα = +1.65, the test
statistic of z = 2 is greater than the critical value; thus, the difference between the sample mean and
population mean is statistically significant by the directional hypothesis.
In both cases, because the test statistic was greater than the critical value, the difference between the
sample mean and the population mean is statistically significant. In this case, the null hypothesis can be
rejected and the alternate hypothesis can be accepted.
In future chapters, I do not cover this alternate method of assessing statistical significance.
-157-
Back to TOC
1. Define each of the following:
a. null hypothesis
b. alternate hypothesis
c. critical value
d. rejection region
e. test statistic
f. alpha level
2. A high school administration is interested in whether their students save more or less than $100 per
month toward college. Translate this question into a null hypothesis and an alternate hypothesis.
3. Why can we never accept the null hypothesis as being true based on statistical tests?
4. Use the following information to answer the questions that follow: H0: μ = 5, H1: μ ≠ 5, X  8 , σ = 15,
and n = 25.
a. Calculate the standard error of the mean.

b. Calculate the test-statistic for the sample mean.
c. What is the p-value for the test-statistic?
d. Using an alpha level of α = .05 test the viability of the null hypothesis.
5. Use the following information to answer the questions that follow: H0: μ = 6, H1: μ ≠ 6, X  4.50 , σ =
3, and n = 36.

6. Use the following information to answer the questions that follow: H0: μ = 6, H1: μ < 6, X  4.50 , σ =
5, and n = 49.

7. Use the following to answer the questions below: Scores on IQ tests are normally distributed with μ =
100 and σ = 15. I am interested in whether University of Scranton Psychology majors have a mean IQ
greater than the general population. I randomly select n = 25 psychology majors and find they have a
sample mean IQ of 104.
a. Expressed in terms of μ, what are the null and alternate hypotheses?

b. Calculate the standard error of the mean.
c. What is the obtained z-Score for the sample mean of 104?
d. What is the p-value associated with this test statistic?
e. Using an alpha level of .05, what decisions should you make regarding the hypotheses? What
should you conclude about the mean IQ of University of Scranton psychology majors compared
to the general population?
8. Use the following to answer the questions below: You are a faculty member at East Buzzkill University
and you believe students at your institution poorer reading comprehension skills than other college
-158-
Back to TOC
students. You test you hypothesis by comparing SAT-Verbal scores of students attending East Buzzkill
University to the mean SAT-Verbal score of all students taking the SATs. Scores on the SAT-Verbal test
are normally distributed with μ = 500 and σ = 75. You sample n = 100 East Buzzkill University students
and ask for their SAT-Verbal scores. The sample has a mean of mean of 480.

c. What is the obtained z-Score for the sample mean of 480?
e. Using an alpha level of .05, what decisions should you make regarding the hypotheses? What
should you conclude about the SAT verbal scores and reading comprehension abilities of East
Buzzkill University students?
9. Use the following to answer the questions below: The Diagnostic Algebra Test (DAT) is a test given to
all incoming freshmen to test their general knowledge of algebra. Scores on the DAT are normally
distributed with μ = 13 and σ = 3. You believe students entering as freshmen Neuroscience majors score
better than average on the DAT, so you randomly sample the DAT scores from n = 10 freshmen
Neuroscience major. You fund this sample mas a mean of 13.8.

c. What is the obtained z-Score for the sample mean of 13.8?

f. Using an alpha level of .01, what decisions should you make regarding the hypotheses? What
should you conclude about the mean DAT score of incoming Neuroscience majors?
e. Say you increase to n = 100. Recalculate the standard error of the mean.
f. Based on this standard error, recalculate the test-statistic.
g. What is the p-value associated with this test statistic?
h. Based on this new information and using the same alpha level, would your conclusion made in ‘f’
change?
10. Define each of the following.

a. Type I error
b. Type II error
c. alpha-level
d. beta
e. power
11. What is the relationship between alpha and the probability of a Type I error. What is the reason for
this relationship?
12. What is the relationship between power and the probability of a Type II error? What is the reason for
this relationship?
13. What is the relationship between alpha and power? What is the reason for this relationship?
14. What effects does sample size have on the power of a statistical test?
15. Under what circumstance should a directional rather than a non-directional test be used? Why? Under
what circumstance should a non-directional rather than a directional test be used? Why?
16. What effect does sample size have on the power of a statistical test? Why?
-159-
Back to TOC
Chapter 12: One-Sample t-Test
12.1 Where and Why the one Sample t-Test Exists
The preceding chapter covered basics of hypothesis testing by introducing the z-Test. The end of the
chapter addressed a major limitation with the z-Test, that is, it can be only used when μ and σ are known.
In cases where μ and σ are unknown, the one-sample t-Test is used.
Procedurally, the one-sample t-Test and z-Test are identical. Incidentally, the procedures in NHST are
basically the same regardless of the test used. In both the z-Test and the one-sample t-Test a sample
mean is compared to a fixed value that represents a population mean. The difference is that in the z-Test
the value to which the sample mean is compared is an actual population mean, whereas in the one-
sample t-Test the value to which the sample mean is compared is assumed to represent a population
mean. For both tests, the difference between the sample mean and the population mean is divided by
sampling error. For the z-Test this sampling error was measured by the standard error of the mean,
whereas in the one-sample t-Test, because the population standard deviation is unknown, the standard
error is estimated, by dividing the estimated standard deviation by the square root of the sample size:
ŝ
ŝ X 
n
This estimated standard error of the mean was introduced earlier and is the denominator of the one-
sample t-Test. Notice that calculating the estimated standard error of the mean is similar to calculating the
true standard error of the mean; the only difference being that in the former you use the estimated
standard deviation and in the latter you use the actual population standard deviation.
12.2 One Sample t-Test

Say we are media researchers and that we are conducting research on television viewing habits. We
want to know (a) how much time students at a Rhode Island university spend watching Family Guy per
week, and (b) whether the amount of time spent watching Family Guy per week is significantly different
from watching Family Guy for 10 hours per week. This 10 hours is our assumption of how much time
students spend watching television per week, but it is not a population parameter.
A sample of Rhode Island college students is selected and each student is asked for how many hours per
week they spend watching Family Guy. The mean number hours these students spend watching Family
Guy per week will be calculated and compared to 10 hours per week, which is the assumed population
mean. At first glance, this seems much like the z-Test, because a sample mean is being compared to a
value. However, in this case, it is assumed that the 10 hours per week is the average number of hours
students spend watching television. We do not know this for sure; hence, there is no population data so
we cannot know μ and σ.
We collect data from n = 50 students at several Rhode Island universities and find the mean number of
hours spent watching Family Guy is 11 hours per week, with an estimated standard deviation of 3.. We
set our alpha-level to α = .05.
The steps in performing the one-sample t-Test are identical to those of the z-Test and we begin by stating
the hypotheses (as before, I will write out and interpret both a directional and a non-directional version of
the hypothesis):
H0: μRI Students = 10 or H0: μRI Students = μ10

H0: μRI Students > 10 or H0: μRI Students > μ10
-160-
Back to TOC
Next, calculate the test statistic in the one-sample t-Test. Recall, the sample mean for number of
hours/week Family Guy was watched was 11 hours/week and that this mean value was being compared
to a predicted mean of 10. Also, the estimated standard deviation from the sample was ̂ . The
formula for the one-sample t-test is:
X 
t
ŝ X
Because the sample mean (11) and the predicted population mean (10) for our example were already
given above, all that we need is the estimated standard error of the mean in the denominator. The
estimated standard error of the mean in our example is:
ŝ 3 3
ŝ X     0.424
n 50 7.071
This value is the denominator in the one-sample t-Test. The outcome of the one-sample t-Test is:
X   11  10 1
t    2.358
ŝ X 0.424 0.424
This is the obtained t-Value andis the test statistic used to determine whether the difference between the
sample mean and the population mean is statistically significant. To determine the statistical significance
of this result, we need to use t-Tables, which is Table 2 in Appendix A (Probabilities Under the t-
Distribution). Using this table is nearly identical to using Table 1, the standard normal tables, to determine
statistical significance; however, when looking up the probability associated with your test statistic you
also need to take into account the degrees of freedom in the sample. From our example, the test statistic
(2.358) rounds to 2.36 and the degrees of freedom in the sample (50 – 1 = 49) is closest to the colum
associated with df = 40. (Note, you should never use a df value greater than your actual degrees of
freedom.) The colum associated with 40 degrees of freedom and the row associated with t = 236 are
highlighted in the table below.
Df
t 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 22 24 26 28 30 40 60 100 200 1000
2.30 .2611 .1481 .1050 .0829 .0698 .0611 .0550 .0505 .0470 .0443 .0420 .0402 .0387 .0374 .0362 .0352 .0344 .0336 .0330 .0323 .0313 .0304 .0297 .0291 .0286 .0267 .0249 .0235 .0225 .0217
2.31 .2601 .1471 .1040 .0820 .0689 .0603 .0542 .0497 .0462 .0435 .0413 .0395 .0379 .0366 .0355 .0346 .0337 .0330 .0323 .0317 .0307 .0298 .0291 .0285 .0279 .0261 .0243 .0229 .0219 .0211
2.32 .2591 .1461 .1031 .0811 .0681 .0594 .0534 .0489 .0455 .0428 .0406 .0388 .0372 .0360 .0348 .0339 .0330 .0323 .0316 .0310 .0300 .0292 .0285 .0279 .0273 .0255 .0238 .0224 .0213 .0205
2.33 .2581 .1451 .1021 .0803 .0672 .0586 .0526 .0482 .0447 .0421 .0399 .0381 .0366 .0353 .0342 .0332 .0324 .0316 .0310 .0304 .0294 .0285 .0278 .0272 .0267 .0249 .0232 .0218 .0208 .0200
2.34 .2571 .1442 .1012 .0794 .0664 .0578 .0518 .0474 .0440 .0413 .0392 .0374 .0359 .0346 .0335 .0326 .0317 .0310 .0304 .0298 .0288 .0279 .0272 .0266 .0261 .0244 .0226 .0213 .0203 .0195
2.35 .2561 .1432 .1003 .0785 .0656 .0571 .0511 .0467 .0433 .0406 .0385 .0367 .0352 .0340 .0329 .0319 .0311 .0304 .0297 .0292 .0282 .0273 .0266 .0261 .0255 .0238 .0221 .0207 .0197 .0190
2.36 .2552 .1422 .0994 .0777 .0648 .0563 .0503 .0460 .0426 .0400 .0378 .0361 .0346 .0333 .0322 .0313 .0305 .0298 .0291 .0285 .0276 .0267 .0261 .0255 .0250 .0232 .0215 .0202 .0192 .0185
2.37 .2542 .1413 .0985 .0768 .0640 .0555 .0496 .0452 .0419 .0393 .0372 .0354 .0339 .0327 .0316 .0307 .0299 .0292 .0285 .0280 .0270 .0262 .0255 .0249 .0244 .0227 .0210 .0197 .0187 .0180
2.38 .2532 .1403 .0976 .0760 .0632 .0548 .0489 .0445 .0412 .0386 .0365 .0348 .0333 .0321 .0310 .0301 .0293 .0286 .0279 .0274 .0264 .0256 .0249 .0244 .0239 .0222 .0205 .0192 .0183 .0175
2.39 .2523 .1394 .0967 .0752 .0624 .0540 .0482 .0439 .0406 .0380 .0359 .0341 .0327 .0315 .0304 .0295 .0287 .0280 .0274 .0268 .0258 .0251 .0244 .0238 .0233 .0216 .0200 .0187 .0178 .0170
From the table, you can see that this test statistic is associated with a p-value of .0232. Because this
value is less than our chosen alpha level (.05) we can say the sample mean is significantly different from
10 hours per week, that is, there is a statistically significant difference between the sample mean and the
assumed population mean of 10. Because the difference is statistically significant, we reject the null
hypothesis and accept the alternate hypothesis. In layman’s terms, because the difference is statistically
significant, we conclude that the amount of time that Rhode Island college students spend watching
Family Guy per week is significantly greater than 10 hours of television viewing per week.
-161-
Important: All the probabilities in Table 2 are associated with non-directional (two-tailed) alpha-levels. If
you have a non-directional alternate hypothesis you can look up the p-value and compare it directly to
your chosen alpha level, as we just did. However, if you have a directional (one-tailed) alternate
hypothesis, after you look up the p-value in Table 2 you must divide that p-value in half in order to assess
statistical significance. In this case, the p-value .0232, which is associated with non-directional alternate
hypotheses, is equal to 0.0232/2 = .0116, which is associated with directional alternate hypotheses.
12.3 t-Distribution Approximation to the Normal Distribution

In earlier chapters I stated that larger sample sizes are preferred over small sample sizes, because the
larger a sample, the better it represents a population. Indeed, as you increase sample size the sampling
distribution of means approximates a normal distribution. This can be seen in t-Distributions
(distributions of critical t-values) for various degrees of freedom. For the t-Distribution, when the degrees
of freedom are small, the distribution is flatter and spread out. As sample size and degrees of freedom
increase the shape of the t-Distribution approximates the shape of a normal distribution.
Theoretically, when degrees of freedom reach infinity the shape of the t-Distribution will be equal to the
shape of the normal distribution. What this means is that for vary large sample sizes it is okay to use the
z-Test instead of the one-sample t-Test, because the t-Test results will be nearly identical to the outcome
of the z-Test. Also, because the t-Distribution approaches normality as the degrees of freedom increases,
this argues for using larger sample sizes in research.
12.4 Reporting One Sample t-Test in Literature

Just like when reporting the outcome of a z-Test in the literature there are several parameters that need
to be reported with the one-sample t-Test: (1) the sample mean, (2) value that the sample mean is
compared, (3) the obtained t-Value, (4) the estimated standard error of the mean, and (5) the p-value. To
report just the results of the t-test, you would write out the following: t(49) = 2.358, SEM = 0.424, p =
.0232 (two tailed). Here, the 49 in parentheses are the degrees of freedom, the SEM stands for the
standard error of the mean.
Below, I present a generic example of how the results of the Family Guy example t-Test would be written
up in a manuscript:
Results
-162-
Back to TOC
Fifty students from Rhode Island universities were asked to report how much time they spent
watching Family Guy for one week. The mean number of hours spent watching Family Guy per week was
found to be M = 11 hours/week (SD = 3). This sample mean was compared to 10 hours/week, which was
assumed to be the amount of television that college students watch in a given week. A one-sample t-Test
yielded a statistically significant difference between the sample mean and the assumed null value of
10.00, t = 2.358, SEM = 0.424, p = .0232 (two tailed). Thus, students watch significantly more Family
Guy, per week, than the average amount of television viewing that students are assumed to watch per
week.
-163-
Back to TOC
1. In what situations would the one sample t-test be used instead of the z-test?
2. State the critical value(s) of t for a one-sample t test for an alpha level of .05 and of .01 under each of
the following conditions:
a. H0: μ = 3, H1: μ ≠ 3, n = 30
b. H0: μ = 3, H1: μ > 3, n = 30
c. H0: μ = 3, H1: μ ≠ 3, n = 50
d. H0: μ = 3, H1: μ < 3, n = 50
e. H0: μ = 3, H1: μ ≠ 3, n = 10
f. H0: μ = 3, H1: μ > 3, n = 10
3. Find the p-value for a one sample t-test under each of the following conditions:
a. H0: μ1 = μ2, H1: μ1 ≠ μ2, n = 20, t = 2.00

b. H0: μ1 = μ2, H1: μ1 ≠ μ2, n = 32, t = 2.50
c. H0: μ1 = μ2, H1: μ1 < μ2, n = 20, t = 2.22
d. H0: μ1 = μ2, H1: μ1 < μ2, n = 10, t = 3.10
e. H0: μ1 = μ2, H1: μ1 ≠ μ2, n = 110, t = 2.00
f. H0: μ1 = μ2, H1: μ1 < μ2, n = 50, t = 2.10
For Exercises 4 – 7, calculate the estimated standard error of the mean, the test-statistic for the given
sample mean, find the p-value, and test the viability of the hypotheses using α = .05.
4. Test the variability of the following hypotheses, H0: μ = 100, H1: μ ≠ 100; with a sample mean of
X  102 (n = 100; ŝ  10 ).
5. Test the variability of the following hypotheses, H0: μ = 100, H1: μ < 100; with a sample mean of
X  97 (n = 10; ŝ  10 ).
6. Test the variability of the following hypotheses, H0: μ = 100, H1: μ < 100; with a sample mean of
X  97 (n = 10; ŝ  3 ).
7. Test the variability of the following hypotheses, H0: μ = 100, H1: μ ≠ 100; with a sample mean of X  95
(n = 25; ŝ  12 ).
8. Use the following information to answer the questions that follow: The administration of a university
wants to know whether the short term memory capacity of college students differs from the short term
memory capacity of seven “items” of information. The short term memory capacity for each 16 students is
measured. The mean short term memory capacity is found to be X  8 items with ŝ 2  1.6 .

b. Calculate the estimated standard error of the mean.
c. What is the obtained t-Value for the sample mean?
d. What is the p-value for this test statistic?
e. Using an alpha level of .01, what decisions should you make about the null and alternate
hypotheses? What should you conclude about the short term memory capacity of college
students?
9. Use the following information to answer the questions that follow: A sample of n = 10 9th grades at
James Woods High School can do an average of 11.5 pull-ups (chin-ups) in 30 seconds, with an
-164-
Back to TOC
estimated population standard deviation of ̂ . The US Department of Health and Human
Services suggests 9th grades should do a minimum of 9 pull-ups in 30 seconds. Is this sample of 9th
grades able to do significantly more pull-ups than the number recommended by the US Department of
Health and Human Services? Using this information, answer the following:

b. Calculate the estimated standard error of the mean.
c. What is the obtained t-Value for the sample mean?
d. What is the p-value for this test statistic?
e. Using an alpha level of .05, what decisions should you make about the null and alternate
students?
10. Use the following information to answer the questions that follow: You want to know whether the
average daily temperature for January in Utica, NY differs from 0°C. You measure the daily temperature
in Utica, NY for ten random days in January. Here are the recorded temperatures:
-3°C 2°C 0°C 0°C -1°C -3°C -5°C -2°C -2°C -1°C
a. Based on this information, what are the null and alternate hypotheses?
b. Calculate the average daily temperature from these measurements.
c. Calculate the estimated standard deviation.
f. Calculate the estimated standard error of the mean.
g. What is the obtained t-Value for the sample mean?
h. What is the p-value for this test statistic?
i. Using an alpha level of .05, what decisions should you make about the null and alternate
students?
11. As the degrees of freedom increase, what happens to the t-distribution?
12. Using a word processor (e.g., MS-Word, OpenOffice), write out the results of the t-tests in APA format
for #s 4 – 7. Each result should appear on a separate line.
Chapter 13: Bivariate Designs
-165-
Back to TOC
13.1 What is a Bivariate Design?
The comic to the right illustrates a basic assumption in
bivariate designs: that two variables systematically covary
with one another and covariation of other, extraneous
variables is controlled. In the comic, the naive researcher
5
failed to control for the astrological sign of the mice. In
this case there is an unintended confounding variable in
the design, which must be corrected before drawing any
conclusions about relationships between the variables
under study. But, what is a bivariate design?
In both the z-Test and one-sample t-Test, a sample mean

was compared to a value that represented a population
mean; the tests were used to determine whether the
difference was statistically significant. These tests tell you
only whether a sample mean differs from some fixed
value; neither test can determine whether a relationship is
exists between variables. Indeed, it is rare that scientists
determine only whether a mean is different from some
value. Scientists usually want to test the significance of a
relationship between variables.
For example, a researcher might manipulate an independent variable by having some students take
ginko-baloba and other students take a placebo for one academic year, and then measure each student's
GPA at the end of the academic year. That is, a researcher could randomly select n = 100 students from
the same population and randomly assign n = 50 of the students to take a ginko-baloba every day for one
school year, and the other n = 50 students take a placebo every day for one school year. At the end of the
school year grades are collected and compared between the groups (see table below). In this example, a
researcher would be examining the relationship between an independent variable (ginko-baloba versus
placebo) and dependent variable (GPA). Whenever a researcher examines the relationship between two
variables the researcher has a bivariate design and the researcher is examining a bivariate
relationship.
Ginko-Baloba Placebo
Mean GPA: 3.30 3.00
Bivariate designs can also examine relationships between two dependent variables. An example of
bivariate relationship between two dependent variables, height and weight, is below, where you can see
that as height increases, weight tends to increase:
Individual Height Weight

A 65” 150 lbs
B 66” 155 lbs
C 68” 160 lbs
D 70” 168 lbs
E 72” 175 lbs
It should be noted that the only way that you can determine whether changes in one variable were
potentially caused by changes in another variable is by manipulating an independent variable, as in the
ginko-baloba vs. placebo example. Assessing the relationship between two dependent variables, as in
the height vs. weight example, can only indicate whether the dependent variables are statistically
associated.
5
In reality, mice do not have astrological signs.
-166-
Back to TOC
Bivariate designs that examine the relationship between an independent variable and dependent variable
are experimental designs or quasi-experimental designs. The difference between experimental
designs and quasi-experimental designs is in an experimental design an independent variable is
manipulated by a researcher (e.g., drug versus placebo); whereas in a quasi-experimental design the
independent variable is observed to change between groups, that is, the levels of the independent
variable naturally designate different groups (e.g., male versus female). Bivariate designs that examine
the relationship between two or more dependent variables are correlational designs, for example,
examining the relationship between height and weight, both of which are measurable variables. For each
type of design there are different statistical analyses to measure the relationship between the variables
and to determine whether the relationship is statistically significant.
13.2 Issues in Experimental Designs

An experimental design is one where an independent variable is manipulated and the objective is to
examine this independent variables’ influence on some dependent variable. For example, the preceding
section described a scenario where some students were given ginko-baloba and other students were
given a placebo, and GPAs of all students were then compared between the two groups. Hence, the
researcher manipulated whether students took ginko-baloba or the placebo.
In an experimental design, the group given the treatment condition that is generally of more interest (i.e.,
the ginko-baloba group in the example above) is the experimental (treatment) group. The group given
the non-treatment condition that is of lesser interest (i.e., the placebo group in the example above) is the
control group. The control group is important, because the behavior of subjects in this group should be
normal and unaffected by the treatment level[s] of the independent variable. Behavior and performance of
subjects in the experimental group should be different than normal, and this difference should be due to
the treatment. Thus, by comparing performance between the control group and the experimental group,
one can determine whether the treatment had any effect on the dependent variable. Under the right
circumstances a researcher could conclude that taking ginko-baloba caused an observed change in GPA.
Thus, experiments are used to establish cause-effect relationships between an independent variable and
a dependent variable.
But what if the independent variable was not manipulated and the levels of the independent variable were
observed to differ between groups (quasi-experimental design)? For example, if a researcher wanted to
compare GPAs between males and female students; the levels of the independent variable sex are
already differs between groups. In most case, the analysis would be exactly the same as that for a true
experimental design. That is, the inferential tests that are used to analyze bivariate relationships between
an independent variable and dependent variable are the same whether the independent variable was
manipulated or not. In quasi-experimental designs, however, you cannot make causal statements about
any difference observed between the groups. For example, if you were to find a difference in GPA
between males and females, you cannot conclude that being male/female caused the difference in GPA;
all that you can state is there is a difference in GPA between the sexes.
There are a number of issues surrounding the use of experimental designs. Those most relevant to
statistics will be described here:
First, in experimental designs a researcher wants to ensure that the groups being compared are identical
except for the level of the independent variable. That is, the only difference between the experimental
group and control group should be the level of the independent variable. Stated differently, a researcher
wants to ensure that subjects in an experimental group have the same characteristics as the subjects in a
control group. If all things are equal between the groups, except for the independent variable, any
difference in the dependent variable between the groups is most likely due to the independent variable
and not to some extraneous variable.
-167-
Back to TOC
The easiest way to ensure subjects in each group are identical is to place subjects into each group
through a process of random assignment, where subjects are placed into groups without any discernible
order. Random assignment attempts to eliminate bias toward having certain types of subject in a group;
thus, random assignment attempts to equally distribute individual differences between groups. Of course,
the only way to know whether random assignment was effective is to measure people on variables other
than the dependent variable that may be related to the study. For example, in the ginko-baloba example
from the preceding section, you may measure in both the experimental and the control groups each
subjects’ high school SAT, ACT and IQ. This way, you can examine whether general intelligence levels
are equal between the two groups.
In experimental designs you generally want to control as many extraneous variables as possible.
Extraneous variables are naturally occurring, random variables that are not directly related to a study
(e.g., slight changes in temperature, time, lighting, age). Most extraneous variables have no effect on
performance or on the outcome of a study; however, it is possible that some variables could exert an
influence and a researcher may want to control these variables so they are constant across levels of an
independent variable. For example, in the ginko-baloba example the amount of time that students study
would certainly vary. A researcher may want to set a minimum and a maximum time that students can
study each week to exert experimental control, that is, hold this extraneous variable constant. Exerting
experimental control allows a researcher to conclude with greater confidence that an observed difference
in a dependent variable between groups was due to the independent variable, and not some extraneous
variable.
It is possible that an extraneous variable may change as the levels of the independent variable change. In
the ginko-baloba example, perhaps all the subjects in the ginko-baloba group were psychology majors
and all subjects in the placebo group were economics majors. The researcher probably did not intend for
this to happen and may not even realize it, but because college major varied with the levels of the
independent variable, any observed difference in GPA between the control group and the experimental
group could be due to the difference in college major. More specifically, say the group taking ginko-
baloba (psychology majors) has a mean GPA of 3.30 and the group taking the placebo (economics
majors) has a mean GPA of 3.00. What is this difference in GPA due to? If you answered it could be due
to ginko-baloba or to the difference in majors’, or answered ‘it cannot be determined, you’re right!
Because two things changed as GPA changed you cannot tell whether the difference in GPA can be
attributed to giving students ginko-baloba versus a placebo or to the difference in college major. It could
be that the psychology major is easy in this particular university or that the economic major is really hard.
In such cases where some extraneous variable changes with the independent variable it is a
confounding variable. Unfortunately, these are difficult to identify and the only correction is to rerun the
study without the confounding factor.
Finally, and perhaps most important with an experimental design, is the issue of causality. Remember, if
a change in a dependent variable is observed after manipulating an independent variable it may be
concluded that changes in the independent variable caused the observed changes in the dependent
variable. For causality to be established, three things must occur:
 There must be a manipulation of an independent variable.

 The manipulation of the independent variable must precede an observed change in a dependent
variable. This goes without saying. If changes in the dependent variable occur before a
manipulation, then changes in the dependent variable may be natural. This is referred to as
temporal precedence.
 A systematic relationship between the manipulated independent variable and changes in the
dependent variable must be observed. This also goes without saying. If there is no change in the
dependent variable as the independent variable is manipulated, then there is no relationship
between the independent and dependent variable.
It is also important to note that an observed change in the dependent variable is caused by changes in
the independent variable in the research design, not the statistical test! You can take two groups (children
without depression and adults with depression), give them some test (happiness test) and run a statistical
-168-
Back to TOC
analysis. There is nothing saying that you cannot do this, and let’s say you find happiness scores of the
children are higher (happier) than adults. Does that mean being a child without depression caused more
happiness? Of course not! Nothing was manipulated! The point is that causality can only be established if
an independent variable is manipulated, which comes through the research design, not the statistical test.
Statistical tests cannot tell you whether one thing causes another; rather, the tests can only tell you if a
relationship is present and is statistically significant.
13.3 Parametric vs. Non-Parametric Statistical Tests

The statistical tests discussed in the following chapters can be classified into one of the following two
categories: parametric statistics and non-parametric statistics. Most inferential statistics are
parametric, meaning that the tests make strict assumptions about population parameters and population
distributions, usually normal distributions. Parametric tests also make use of means and variances. Two
assumptions of parametric statistical tests are the normality assumption and the homogeneity of
variance assumption. The normality assumption means that an inferential statistic assumes the sample
data came from a population that was normally distributed; hence, the sample data should be roughly
normally distributed. Homogeneity of variance means that an inferential statistic assumes the variance is
equal in each different group or condition, such that σ21 = σ22 =,...,= σ2k.
Parametric tests, usually, assume the dependent variable was measured on a ratio or interval scale;
ordinal data can be analyzed with parametric tests if the data approximates an interval scale. Nominal
scale data generally cannot be analyzed with parametric tests. Examples of parametric tests include the
independent groups t-Test and the correlated samples t-Test, which we discuss in the next two chapters.
Non-parametric statistics do not assume the data come from any normally distributed population and
there are few, if any, assumptions about a target population. Non-parametric tests do not make use of
means and variances; rather, the focus is on the differences between distributions of scores in a sample.
Generally, the data comes from nominal or ordinal scales. Some non-parametric tests include the
Wilcoxon Rank Sum Test; Mann-Whiney U-Test and Chi-Square, some of which we will cover later.
13.4 Between Groups Designs vs. Within-Groups Designs

There are two ways an independent variable can be manipulated: between-groups/subjects or within-
groups/subjects. In a between-groups design (independent groups design) each subject is randomly
assigned to only one level of the independent variable; thus, each group is independent of the other
groups. One of the most important factors in independent groups study is whether assignment of subjects
to the conditions/groups was random.
In a within-groups design (correlated-samples design, dependent groups design, or repeated

measures design), one group of subjects is exposed to each level of an independent variable; thus,
subjects are tested multiple times. Because each subject is tested in each condition there is no need for
random assignment, because there is only one group. But selection into that one group should still be
random. Importantly, the order in which conditions are presented should be random, or counterbalanced
across subjects so each permutation of conditions is presented across subjects.
Whether a between-groups design or within-groups design is used depends mainly on the research
question; though most scientists prefer within-subjects designs, because the statistical tests are more
powerful. Between-groups designs are good if you are conducting a study where you want each subject
to be exposed to only one level of the independent variable. For example, in drug studies, because the
physiological effects of taking a drug can be long-lasing most researchers do not like to use within-
subjects designs, because the effects of one condition can influence performance in a later conditions. In
contrast, within-groups designs are good if you are examining performance on some task, but under a
variety of similar conditions. In this case, you may want to see how the same people react to different
changes in the independent variable.
-169-
Back to TOC
There are advantages and disadvantages to a within-groups design and between-groups design that are
listed in the table below. What is interesting is that the advantage for one type of a design is usually
associated to a disadvantage with the other type of design:
Between-Subjects Design Within-Subjects Design

Advantages 1. Statistically easier to analyze 1. Requires fewer subjects
2. Each group is independent of the other groups 2. Less costly and less time consuming
3. No need to worry about carry-over effects 3. Statistically more powerful
Disadvantages 1. Time consuming and generally more costly 1. Carry-over and practice effects
2. Requires more subjects 2. More difficult to analyze by hand
3. Less statistically powerful
Carry over effects occur when something you are exposed to influences performance later. Specifically,
the influence of a level of an independent variable “carries over” and influences performance at a later
point in time. Practice effects mean that performance gets better over time (with practice). Thus,
facilitation in performance later in an experiment may not be due to the particular condition a person is
exposed to; rather, it may be due to performance enhancement through practice.
13.5 Matched Pairs Designs

There is a method not too many researchers take advantage of called a matched-pairs design. This is a
between-groups experimental design that is analyzed as a within-subjects design. More succinctly, an
independent variable is manipulated between-groups, but data are analyzed as though the manipulation
was within-subjects. There are two main reasons for this: (1) Within-subjects analyses are more powerful,
or (2) the independent variable cannot be manipulated within-subjects, but the independent variable is
something that can vary within-subjects (e.g., age). In a matched-pairs design the independent variable is
manipulated between-subjects. Subjects in each group are ranked in order of overall performance on the
dependent variable or performance on another variable. This way the subject with the best performance
in the experimental group is “matched” to the subject with the best performance in the control group. The
logic being that if this were a within-subjects design an individual with the best performance in the
experimental group would likely have the best performance in the control group. Once subjects are
“matched,” the data is analyzed using the appropriate within-subject analyses.
To use a concrete example, consider the ginko-baloba example. In that study, whether subjects took the
ginko-baloba or placebo was manipulated between subjects. A researcher could turn this into a matched-
pairs design by rank-ordering the subjects in each group by GPA, from high to low. Once the subjects are
ranked the data are then analyzed with a within-groups statistical test.
It is important to note that a researcher usually needs to have some theoretical reason for conducting a
matched-pairs design other than a within-subjects design being statistically more powerful. One example
is if a researcher wants to study the effect of an independent variable that varied within-subjects, but
cannot be manipulated. Such an example would be age: Age changes within a person, but age cannot be
manipulated. If a person wants to tests the effects of aging on spatial memory a researcher could sample
a group of young people and a group of old people and test spatial memory. A matched-pairs design
could be used to contrast spatial memory of young subjects compared to their old subject counterparts.
I should also note that there is no specialized statistical test for a matched-pairs design: A matched-pairs
design comes from the experimental design itself, not the statistical analysis. Matched-pairs designs use
the exact same analyses as within-subjects designs.
-170-
Back to TOC
1. Describe the differences between experimental research designs and quasi-experimental
(observational) research designs.
For each of the studies described in Exercises 2 - 4, identify the independent variable, the dependent
variable, and indicate whether the independent variable is between subjects or within subjects.
2. Shepard and Metzler (1972) studied the ability of people to form a visual image of an
object in the mind and then mentally rotate the object. Each subject was presented with
a display containing two objects side by size, and the angle of rotation between the
objects was varied from 0° to 180°, in 15° increments. The response time to decide if
the objects were the same or were mirror images of each other was recorded.
3. Harvath (1943) studied the influence of environmental noise on problem-solving. Subejcts were
randomly assigned to a noise condition where a buzzer was played in the background, or a quiet
condition that did not contain the buzzer. Subjects in each group were given 30 problems, and the
number of correct solutions was determined for each subject.
4. Rosenthal and Fode (1963) examined the effect of experimenter bias on the ability of rats to learn to
navigate a maze. Subjects were told that they were going to teach a rat to learn to navigate a maze. One
group of subjects was told that their rats were genetically smart, “maze bright” rats that should show
learning during the first day and performance should increase. A second group of subjects was told their
rats were genetically dumb, “maze-dull” rats that should show very little evidence of maze learning.
Subjects trained their rats to enter one of two sections of a maze, by rewarding the rat when they entered
the correct section. Subjects measured the number of times the rat entered the correct section of the
maze each day.
5. What is a control group and what is the purpose of including a control group in a research design?
6. For is random assignment? Why is it important?
7. What are confounding variables? How can they be controlled?
8. What are the advantages of within-subjects designs compared with between-subjects designs? What is
a potential problem?
9. How do parametric and nonparametric statistics differ?
10. What is the normality assumption? What is the homogeneity of variance assumption?
Chapter 14: Independent Groups t-Test
-171-
Back to TOC
14.1 Uses and Requirements
From preceding chapters, recall that both the z-test and the ne-sample t-Test compare a sample mean to
a value representing a population mean. For both tests, there was no independent variable manipulated,
rather, subjects in a sample were selected and that sample mean was compared to a population mean.
The limitation of the z-Test and one-sample t-Test is that neither can be used to compare performance
between levels of an independent variable. To do this you must use a bivariate test. This chapter covers
examples of where an independent variable is manipulated between groups of individuals. The type of
statistical analysis used in such cases is an independent-groups t-Test. As you will see, the procedures,
assumptions, and test are not all that much different here than in the one sample t-Test.
The independent groups t-Test is used when:
1. The dependent variable is quantitative in nature

2. The independent variable is qualitative in nature, that is, the levels represent different categories.
3. The independent variable has two and only two levels.
4. The independent variable is manipulated between-subjects.
14.2 Mean Differences and Variance between Independent Populations

The independent groups t-Test is used to determine whether two sample means, which are assumed to
have been drawn from two different populations, are statistically different. Before getting into the
independent groups t-Test, I introduce assumptions about the population distribution that each sample is
assumed to have come from and the sampling distributions of the means of those populations.
In an independent groups design, there are different samples of subjects. It is assumed that the subjects
in each sample are identical except for the level of the independent variable that each sample is exposed.
Subjects in each sample are assumed to come from a different population, that is, subjects in “Sample A”
are assumed to come from “Population A” and subjects in “Sample B” are assumed to come from
“Population B.” These populations are assumed to differ only in their association to a level of an
independent variable. In short, it is assumed there are two independent samples, each coming from its
own independent population, which differ by the level of an independent variable. Thus, any difference
found between the sample means should also exist between population means; and any difference
between the populations means must be due to the difference in the independent variable.
You can visualize this in the figure to the

right, which presents two population
distributions. Subjects in each population
are assumed to be equal except for the
level of an independent variable. The
distance between the means of the
populations is assumed to be due to a
difference in the independent variable.
Thus, the overlap in the distributions
reflects the effect the independent variable
on a dependent variable: The more
overlap, the less of an effect the
independent variable has.
Recall, that if you select all possible samples of a specific size (n) from a population and calculate the
mean of each of sample, you end up with a sampling distribution of the mean. This sampling distribution
of the mean approximates a normal distribution as n approaches infinity, as per the central limit theorem.
The same is true if you have two populations and you select every possible sample of n scores from
-172-
Back to TOC
Population A and every possible sample of n scores from Population B: You end up with a sampling
distribution of means for Population A and a sampling distributions of means for population B. Each of
these sampling distributions has a mean equal to μ of its population and has a standard deviation equal to
σ√n; the standard error of the mean. If we assume the sample mean is the best approximation to μ, then
any difference between sample means should reflect the true difference between population means.
Thus, any difference in the sample means can be through of as the effect of the independent variable.
Recall that test statistics are ratios of some measure of effect to some measure of error. The difference
between sample means can be thought of as the effect of the independent variable. So what is the error?
Maybe we could use the standard error? Nice guess, but the standard error of the mean for each
population represents only one population. Because we have two populations we need to account for the
error variance in both populations, or the variance in the difference between the populations.
Consider this: If you calculated every mean of sample of size (n) from a population you end up with a
sampling distribution of means, right? If you calculate the difference between every possible pair of
sample means, you would end up with a sampling distribution of mean differences between
independent groups. That is, if you selected every possible sample of n scores from Population A
and every possible sample of n scores from Population B, then took the difference between every
possible pair of sample means you end up with a single distribution. This single distribution would be a
distribution of mean differences between samples of size n.
Say that you select a sample from Population A and find it has a sample mean of 10, and you select a
sample from Population B and find it has a sample mean of 11. The mean difference is 10 – 11 = -1. If
you continue to do this for every possible pair of samples drawn from Population A and Population B, you
will find that some differences are greater than zero and some differences are less than zero, but overall
the positive differences would cancel out the negative differences. Also, after calculating the differences
between sample means, you would find that the resulting sampling distributing of mean differences has a
mean difference equal to the true difference between population means (μ A - μB), which is normally
assumed to be zero under the null hypothesis.
The standard deviation of this sampling distributing of mean differences is equal to the standard error of
the mean difference between independent groups:
12  22
X X  
1 2 N1 N 2
Each subscript (1 and 2) refers to a different population or level of the independent variable. The standard
error of the difference basically combines the variance from each independent population (level of an
independent variable) into one measure. Conceptually, the standard error of the difference is the average
difference between sample means of a given size (n) from each population. Note that the standard error
of the difference is actually just the summed standard error of the mean for each distribution:
12  22 12  22
X  X   X  X  
1 2 N1 N2 or 1 2 N1 N2
However, this formula is only useful when several conditions are met:
2. The population variances are known, which is almost never the case.
-173-
Back to TOC
3. The variances in each population are equal, which is almost never the case. Nonetheless, in most
circumstances you assume the variances are equal, as per homogeneity of variance.
Because we almost never know the true population variances, you must estimate the standard error of
the mean difference between independent groups.
14.3 Pooled Variance and the Estimated Standard Error of the Difference
Calculating the estimated standard error of the difference is a two-step process. First, you calculate the
pooled variance estimate, which is the combined average estimated variance for both populations
based on samples of size n1 and n2:
ŝ 2Pooled 
n1  1ŝ12  n 2  1ŝ 22
n1  n 2  2
The subscripts 1 and 2 refer to the two different samples (groups). It does not matter which sample is
labeled 1 and which is labeled 2, so long as you are consistent. The pooled variance estimate is the
average estimated variability between the two independent samples. Because the product of an
estimated population variance from a sample and the degrees of freedom of that sample is equal to the
sum of squares for a sample, the pooled variance estimate can also be calculated as follows:
SS1  SS2
ŝ 2Pooled 
n1  n 2  2
The pooled variance estimate is then used in calculating the estimated standard error of the
difference between independent means, or what I often refer to as “Bob”:
ŝ 2Pooled ŝ 2Pooled
ŝ X  X  
1 2 n1 n2
This is the error term for the independent group t-Test. It represents the estimated average difference
between sample means of size n1 and n2 that were selected from independent populations.
14.4 Hypotheses in the Independent Groups Design

Recall from the preceding section that the mean of the sampling distribution of the mean differences
between independent samples will have a mean equal to the difference between the population
means (μA - μB). Usually, it is assumed that this difference between the population means is equal to
zero, which would indicate there is no effect of some independent variable (i.e., μ A - μB = 0) . Any
deviation from zero (the population difference) between the two sample means is assumed to be due to
the independent variable. Nonetheless, under the null hypothesis the difference between means from two
independent groups is generally expected to be zero:
H0: μ1 - μ2 = 0 or H0: μ1 = μ2
In short, if under the null hypothesis the two population means are expected to be equal, then the
alternative hypothesis is:
H1: μ1 - μ2 ≠ 0 or H1: μ1 ≠ μ2
-174-
Back to TOC
Thus, under the alternative hypothesis the difference between the two population means is not expected
to be zero.
The null and alternate hypotheses above reflect non-directional (two-tailed) hypotheses; that is, the
alternate hypothesis states that there would be some difference between the means, but specifically
whether the mean of Population 1 will be greater/less than the mean of Population 2. It is also possible to
generate directional (one-tailed) hypotheses for an independent group t-Test. For example, say I predict
the mean of Population 1 will be greater than the mean of Population 2. The null and alternative
hypotheses are:
H0: μ1 - μ2 = 0 or H0: μ1 = μ2
H1: μ1 - μ2 > 0 or H1: μ1 > μ2

The alternate hypothesis is predicting the mean of Population 1 will be greater than the mean of
Population 2; that is, the difference between the means of Populations 1 and 2 will be greater than zero.
14.5 Degrees of Freedom in the Independent Groups t-Test

Why did I make a section dedicated to degrees of freedom? Because, to put it bluntly, this is the one thing
I specifically cover in every statistics class and ~50% of students still manage to get it wrong. Recall that
degrees of freedom are equal to n – 1 in a sample. In an independent groups design you have two
samples. Each group has its own n – 1 degrees of freedom. For example, say that sample A has n = 10
people and sample B has n = 12 people. In sample A, df = 10 – 1 = 9 and in sample B, df = 12 – 1 = 11.
When dealing with an independent groups t-Test, you need to account for the total degrees of freedom
(dfT). Most students assume the total degrees of freedom are equal to the total number of subjects across
both samples minus one. Thus, most students guess that the degrees of freedom in this example would
be df = nT – 1 = 22 – 1 = 21, because there are nT = 22 people. But, in this case, you have only accounted
for 21 of the 20 total degrees of freedom between the samples!
Why are there 20 degrees of freedom and not 21? Remember, degrees of freedom are equal to n – 1 in a
sample. Because there are two samples, we need to account for the degrees of freedom in each. Thus,
there are 9 degrees of freedom in sample A and 11 degrees of freedom in sample B, so the total degrees
of freedom is df = 9 + 11 = 20. An easier way to calculate the degrees of freedom in an independent
groups design is df = nT – 2 where nT is the total number of subjects tested across both groups (22).
Hence, dfT = 22 – 2 = 20.
14.6 Example of Independent Groups t-Test

Many colleges and universities require students to take a Freshman Seminar course, where students are
acclimated to college life and taught study skills and time-management skills. Let’s say at Faber College,
freshmen are required to take such a course, and this course has always covered basic study skills. One
year, a psychologist who conducts research in learning develops a new study technique where students
acquire study skills working in groups, rather than studying alone. This new system should increase GPA
relative to covering only basic study skills.
The researcher randomly selects n = 10 freshmen and puts them into a traditional seminar course (Basic
Study,) and randomly selects n = 10 different freshmen and puts them into the new seminar course
(Group Study). All students complete this course and at the end of their freshman year the GPAs from all
20 students are collected and the mean GPA is compared between groups. The researcher predicts that
the mean GPA in the Group Study condition will be greater than the mean GPA in the Basic Study
condition (directional hypothesis); thus, the hypotheses are:
H0: μGroup = μBasic H1: μGroup > μBasic
-175-
Back to TOC
The data (GPAs) for both groups at the end of the Basic Study Group Study
semester are presented in the table below. You
can see there are n = 10 different students in each Student GPA Student GPA
condition and each GPA is measured to the Bob 2.000 Stew 3.300
nearest thousandth. Rob 2.300 Roger 3.300
Tom 2.300 Phil 3.300
Before performing the t-Test, we first need to Judy 2.700 Jen 4.000
determine our critical t-Value. To determine tα you Mary 3.000 Christine 3.700
need the alpha level (α), whether you have a Pete 3.700 Melissa 2.000
directional or non-directional hypothesis, and the Hans 2.700 George 2.000
total degrees of freedom Pat 3.000 Gina 2.300
Floyd 3.300 Tim 2.700
Marge 2.000 Tony 2.700
Above, I stated that the researcher was predicting that the Group Study Condition should have a mean
that is greater than the mean in the Basic Study Condition; thus, the researcher has a directional (one-
tailed) hypothesis. This is important, because directionality will need to be taken into account when
looking up the pi-value. For this example, assume the psychologist selects an alpha-level of α = .01
The steps for conducting the independent groups t-Test, which are going to be similar to those for the
correlated samples t-test in the next chapter, are as follows:
1. Determine the mean and sum of squares for each sample

2. Using the sums of squares, calculate the estimated variance
3. Estimate the standard error of the mean difference between independent groups
4. Conduct the independent groups t-Test
5. Determine the significance of the t-Test and make decisions about hypotheses.
Steps 1 – 3 have been alluded to earlier and the formulas displayed in previous sections, so they will not
be elaborated on here. First, we need to calculate the means and sums of squares of each group. This is
done in the table below, using the data from above:
Group Study Basic Study

Student GPA (X1) X 1
 X1  X 1
 X1 
2
Student GPA (X2) X 2
 X2  X 2
 X2 
2
Stew 3.300 0.37 0.137 Bob 2.000 -0.700 0.490

Roger 3.300 0.37 0.137 Rob 2.300 -0.400 0.160
Phil 3.300 0.37 0.137 Tom 2.300 -0.400 0.160
Jen 4.000 1.07 1.145 Judy 2.700 0 0
Christine 3.700 0.77 0.593 Mary 3.0000 0.300 0.090
Melissa 2.000 -0.93 0.865 Pete 3.700 1.000 0.000
George 2.000 -0.93 0.865 Hans 2.700 0 0
Gina 2.300 -0.63 0.397 Pat 3.000 0.300 0.090
Tim 2.700 -0.23 0.053 Floyd 3.300 0.600 0.360
Tony 2.700 -0.23 0.053 Marge 2.000 -0.700 0.490
ΣX1 = 29.3 SS1 = 4.382 ΣX2 = 20 SS2 = 2.84

X1  2.93 X 2  2.7
Using the sums of squares, we first calculate the pooled variance estimate:
-176-
Back to TOC
SS1  SS2
ŝ 2Pooled 
n1  n 2  2
4.382  2.84
ŝ 2Pooled 
10  10  2
7.222
ŝ 2Pooled 
18
ŝ 2Pooled  0.401
Now that we have the pooled variance estimate we can calculate the estimated standard error of the
difference between independent groups (“Bob”):
0.401 0.401
ŝ X  X  
1 2 10 10
ŝ X  X  0.040  0.040
1 2
ŝ X  X  0.08 0
1 2
ŝ X  X  0.283
1 2
This value (0.283) is the estimated standard error of the difference for the independent groups t-Test; that
is, it is the estimated average mean difference between population the means for the Basic Study
condition and Group Study condition.
The next step is to perform the independent group t-Test and calculate our obtained t-Value:
t
X1  X2   1  2 
ŝ X  X
1 2
The numerator includes the actual difference between the two sample means and the hypothesized
difference between the two population means. The difference between the two population means is
usually assumed to be zero under the null hypothesis. In such cases, the (μ1 - μ2) will be equal to zero
and can be dropped from the t-test:
X1  X 2
t
ŝ X  X
1 2
The next step is inserting the sample means and performing the t-Test. This is very important: If you have
a non-directional (two-tailed) hypothesis test it does not matter which sample mean that you subtract from
the other, because you are predicting “some difference” and that difference could be positive or negative.
In contrast, when you have a directional (one-tailed) hypothesis, it does matter which sample mean that
you subtract from the other. Follow these rules if you have a directional (one-tailed) hypothesis: (1) If you
predict one sample mean to be greater than the other sample mean, then you are predicting a positive
difference between sample means and a positive t-Value (test statistic). The mean that you are
predicting to be greater than the other should be the sample mean with the subscript '1'; that is, the mean
that comes first in the numerator. (2) If you predict one sample mean to be less than the other sample
mean, and then you are predicting a negative difference between sample means and a negative t-Value
(test statistic). The mean that you are predicting to be smaller than the other should be the sample mean
-177-
Back to TOC
with the subscript '1'; that is, the mean that comes first in the numerator. Please note that for both of these
rules, it does not matter whether this is what you find in the means; the placement of the sample means
into the t-Test is based on what you predict, not what you observe.
Substituting in our sample means and the estimated standard error of the mean difference calculated
earlier, we have:
X1  X2 2.93  2.7 0.23
t    0.813
ŝ X  X 0.283 0.283
1 2
This obtained t-Value (t = 0.813) is our test-statistic that we need to look up in Table 2, the t-Tables. In
this example, there were 18 degrees of freedom and the test statistic (0.813) rounds to 0.81. Using these
two values in Table 2, we find that this test statistic is associated with a two-tailed p-value of .4341, which
wen divided by two gives a one-tailed p-value of .2170. Because this p-value is greater than the alphal
level (.01), we conclude there is not enough evidence to claim that there is a statistically significant
difference in GPA between the Basic Study condition and the Group Study condition. That is, the mean
difference between the Basic Study condition and the Group Study condition is not statistically significant.
Therefore we retain the null hypothesis and make no decision regarding the alternate hypothesis. In
layman’s terms, we conclude that exposing freshmen students to a new form of Group Study technique
resulted in a small non-significant increase in end of year GPAs compared to students taking a Basic
Study skills course.
14.6 Reporting in the literature

Several parameters need to be reported with the independent groups t-Test: (a) either both sample
means or the mean difference between the sample means, (b) the obtained t-Vlue, (c) the estimated
standard error of the mean difference between independent samples, (d) the total degrees of freedom,
and (e) the p-value. Below, I present a generic example of how the results from Section 14.5 (Note, that
in APA format, an Italicized M is use as a symbol for the sample mean.):
Results
Twenty Faber College students were randomly assigned to a Basic Study condition or a Group
Study condition, with n = 10 students per group. At the end of the students’ freshman year, the GPA of
each student was recorded and the mean GPA in the Basic Study condition was compared to the mean
GPA in the Group Study condition. An independent group t-Test revealed a non-significant difference
between the mean of the Basic Study condition (M = 2.70) and the Group Study condition (M = 2.93),
t(18) = 0.813, SE = 0.283, p = .2170 (one-tailed). Thus, students learning to study as a group had a small
and non-significant increase on GPA.
14.6 Confidence Intervals around the Mean Difference
-178-
Back to TOC
Confidence intervals can be calculated the difference between the sample means. In an independent
groups t-Test most researchers calculate confidence interval around each sample mean. To do this, you
estimate the standard error of the mean for each sample and using the critical t-Value based on the
degrees of freedom for each sample multiply that estimated standard error by tα to get the confidence
interval around each sample mean. By doing this you can see if each sample mean lies within the
confidence interval of the other sample mean.
Others like to calculate confidence interval around the mean difference between the sample means. To
do this, you use the standard error of the difference and tα from the independent groups t-Test. Thus, for
the example from Section 14.5, we have (note, because α = .01, this would be the 99% confidence
interval around the mean difference):
CI  X1  X 2   ŝ X  X t 
1 2
CI  2.93  2.7   0.2832.88
CI  0.23  0.815
Recall that the null hypothesis predicts that the difference between means will be equal to zero. If zero
falls within this confidence interval we can assume that the mean difference is not significantly different
from the expected mean difference of zero, which is the case in the example here.
14.7 Effect Size: Eta Squared

One question is how much strong an effect the independent variable had on the dependent variable. That
is, from the example above, is a mean difference of 0.230 a large or a small? To know how much of an
effect the independent variable had on the dependent variable, one must calculate a measure of effect
2
size. One measure of effect size measure called eta-squared (η ), which is the proportion of variance in
the dependent variable that is attributable to the effect of the independent variable. The eta-squared
effect size is the ratio of total treatment variance (the effect of the independent variable) to total variance;
thus:
SSEffect
2 
SSTotal
The total sum of squares (SSTotal) is the total variation across all of the scores in a set of data. That is,
combined over both levels of an independent variable, what is the total variability among all of the
scores? To find the total sum of squares, you need to calculate something called the grand mean,, which
is is the sum of all scores in a set of data divided by the total number of scores. In the example from
Section 14.5 the sum of all the scores (GPAs from both the Basic Study and Group Study conditions) was
ΣX = 56.300, and the total number of subjects was n = 20. The grand mean is G = 56.3/20 = 2.815. Note,
this is the same value you would get by adding the mean of each condition and dividing by two.
The sum of squares total is found by subtracting the grand mean from each score in the data, squaring
the differences, and adding the squared differences. This is done in the table below.
The sum of squares effect (SSEffect) is calculated by subtracting the grand mean from the sample mean
that is associated with each individual. For example, in our example from section 14.5, the student Stew
was in the Group Study condition, and that condition had a mean of 2.93. The grand mean (2.815) is
subtracted from the mean of Stew’s group (2.93). This results in a treatment effect of 0.115 for Stew.
This is done for each individual in Stew’s group, and the same is done for each individual in the Basic
Study condition, with the grand mean being subtracted from the mean for that other group (2.7). Thus,
Bob and all of the others in the Basic Study condition have a treatment effect of 2.700 - 2.815 = -0.115.
-179-
Back to TOC
In the table, each student is listed in the leftmost column, and I have then listed which study group
condition each student was in. Each student’s GPA is then listed, followed by the mean (M) of that
2
student’s group. The (X - G) and the (X - G) columns show the calculation of the sum of squares total.
2
Finally, the (M - G) and the (M - G) columns show the calculation of the sum of squares effect.
Calculating SSTotal Calculating SSEffect

Student Group GPA (X) M (X - G) (X - G)2 (M - G) (M - G)2
Stew Group Study 3.300 2.93 0.485 0.235 0.115 0.013
Roger Group Study 3.300 2.93 0.485 0.235 0.115 0.013
Phil Group Study 3.300 2.93 0.485 0.235 0.115 0.013
Jen Group Study 4.000 2.93 1.185 1.404 0.115 0.013
Christine Group Study 3.700 2.93 0.885 0.783 0.115 0.013
Melissa Group Study 2.000 2.93 -0.815 0.664 0.115 0.013
George Group Study 2.000 2.93 -0.815 0.664 0.115 0.013
Gina Group Study 2.300 2.93 -0.515 0.265 0.115 0.013
Tim Group Study 2.700 2.93 -0.115 0.013 0.115 0.013
Tony Group Study 2.700 2.93 -0.115 0.013 0.115 0.013
Bob Basic Study 2.000 2.70 -0.815 0.664 -0.115 0.013
Rob Basic Study 2.300 2.70 -0.515 0.265 -0.115 0.013
Tom Basic Study 2.300 2.70 -0.515 0.265 -0.115 0.013
Judy Basic Study 2.700 2.70 -0.115 0.013 -0.115 0.013
Mary Basic Study 3.000 2.70 0.185 0.034 -0.115 0.013
Pete Basic Study 3.700 2.70 0.885 0.783 -0.115 0.013
Hans Basic Study 2.700 2.70 -0.115 0.013 -0.115 0.013
Pat Basic Study 3.000 2.70 0.185 0.034 -0.115 0.013
Floyd Basic Study 3.300 2.70 0.485 0.235 -0.115 0.013
Marge Basic Study 2.000 2.70 -0.815 0.664 -0.115 0.013
SSTotal = 7.486 SSEffect = 0.264
Using the SSTotal and SSEffect values, the eta-squared effect size is:
SSEffect 0.264
2    0.035
SSTotal 7.486
Thus, about 3.5% of the variance in the dependent variable (GPA) can be accounted for by the effect of
the independent variable (Study Condition). The proportion of variance that cannot be explained is 100%
- 3.5% = 96.5%. Thus, about 96.5% of the variance cannot be explained by the effect of the independent
variable.
Of course, this process would take some time, and there is a simpler method for calculating eta-squared,
which makes use of total degrees of freedom and your obtained t-value:
t2
 
2
t 2  df
Using the total degrees of freedom and obtained t-value from earlier, eta-squared is:
0.8182 0.661 0.661

 
2
   0.035
0.8182  18 0.661  18 18.661
14.8 Effect Size: Cohen’s d
-180-
Back to TOC
Another popular index of effect size is Cohen’s d. There are several methods for calculating Cohen’s d,
but the most appropriate is to divide the difference in the sample means by the estimated pooled
standard deviation, which is just the positive square root of the pooled variance estimate.
What is nice about Cohen’s d is that it provides a standardized measure that can be used to compare
across different studies. Specifically, Cohen’s d is a measure of the standardized difference between
sample means; that is, it is a distance between two sample means in standard deviations.
From the example in Section 14.5, the estimated pooled standard deviation is:
ŝPooled  ŝ 2Pooled  0.401  0.633

Cohen’s d is:
X1  X 2 2.93  2.7 0.23

d    0.363
ŝ Pooled 0.633 0.633
Thus, the two sample means are separated by only 0.363 standard deviations.
Cohen also provides useful labels to describe how ‘big’ or how ‘strong’ of a relationship the effect size
indicates. The table below reports the minimum Cohen’s d and eta-squared values that correspond to
‘weak’, ‘moderate’, and ‘strong’ effect sizes (also called ‘small’, ‘medium’, and ‘large’ effect sizes):
2
Effect Size Cohen’s d eta-squared (η )
“Small” (“Weak”) .20 .01
“Medium” (“Moderate”) .50 .06
“Large” (“Strong”) .80 .14
2
Based on our eta-squared value (η = 0.035) and or Cohen’s d value (d = 0.363), the strength of the effect
of the independent variable in the example from section 14.5 would represent a ‘weak’ or ‘small’ effect of
taking ginko-baloba versus a placebo on GPAs.

Recall, from earlier chapters, statistical power (1 – β) is the probability of correctly rejecting a false null
hypothesis. You want this probability to be high so that a statistically significant result likely reflects a true
significant difference, that is, the null is probably false. The conventional level of statistical power for an
inferential test is .80 or greater (note that power has a limit of .999999....). There are a number of
methods that can be used to determine the power of an inferential test. Below, I show you how to
estimate the power in an inferential test from a program called G*Power (Faul, Erdfelder, Lang &
Buchner, 2007). Following that I show you how to determine the number of subjects needed for a study
based on a desired level of power. In case I did not mention in class, you should also read the Power
Packet on the course website.
A power analysis takes one of two main forms. A post hoc power analysis is performed after data
collection and after an inferential test has been conducted on the collected data to determine the
achieved power of a result. The achieved power is the probability that the results of your inferential test
would correctly reject a false null hypothesis. Obviously, if you want to reject the null, you want to have
high achieved power (generally .80 or above). An a priori power analysis is performed prior to any data
-181-
Back to TOC
collection or data analysis to determine the sample size needed to arrive at a level of desired power.
Researchers—should—start a project by selecting a level of desired power they want to have after data is
collected and inferential tests are performed. That is, say you want to achieve power of .80 after an
independent group t-Test is performed on a set of data. Based on this desired power and several other
parameters you perform a power analysis to determine how many subjects are needed to have that
amount power. This is important, because sample size is related to power and sample size is something
that a researcher has control over. Below, I list the parameters that are associated with statistical power
and their general relationship with power:
1. Effect size: The larger the effect size the more power is in an inferential test.
2. Alpha level: The smaller the alpha level, the less power is in an inferential test.
3. Directionality: Non-directional hypotheses are less powerful than directional hypotheses.
4. Sample size and power: Typically, larger sample sizes have more power.
Importantly, if strength (effect size) of the true relationship between variables is small, it will be difficult to
detect the relationship. In this case, larger sample sizes are better and will increase statistical power. In
contrast, if the strength (effect size) of the true relationship is strong, it will be easy to detect the
relationship, hence, sample sizes do not have to be as large to have sufficient statistical power.
Using the example from Section 14.5, we can calculate the achieved power of the t-Test. Recall, that the
number of subjects per group was n = 10, α = .01, we had a directional (one-tailed) hypothesis, and effect
size was d = .363. To determine the achieved power, we’ll use G*Power 3, which should be on most of
computers. If not, go to http://www.psycho.uni-duesseldorf.de/abteilungen/aap/gpower3/ and download
G*Power 3 for free!
Once you open G*Power 3 (and hopefully looked through the Power Packet on the course website):
under “Test family” be sure “t
tests” is chosen; under
“Statistical test” choose
“Means: Difference between
two independent means (two
groups)”; and under “Type of
Power Analysis” choose “Post
hoc: Compute achieved
power.” These choices change
depending on the type of
inferential test you use and the
type of power analysis you
perform (a priori or post hoc).
Next, you'll need to enter the α-
level, sample sizes, and effect
size from the t-test. The alpha
level was α = .01, we had a
directional hypothesis, and d =
.363.
After clicking Calculate you are

provided with the achieved
power, that is, the probability of
correctly rejecting a false null
hypothesis. In this case, Power
= .057787, which is extremely
low. Given we have such low
power we cannot conclude with
much certainty that the null
hypothesis is truly false; we
-182-
Back to TOC
need more power, which requires more subjects. What this could mean is our result is actually a Type I
error!
G*Power can also be used to perform an a priori power analysis to determine how many subjects are
needed based on a desired level of power. For example, say I plan to run a study that will compare two
levels of an independent variable: A1 vs. A2. Based on previous research I know the effect size is
generally d = .50 when these two levels of the independent variable are compared. I plan to use α = .01
and my hypothesis will be non-directional (two-tailed). I want to correctly reject the null hypothesis with
Power = 0.80, so I need to use the appropriate number of subjects.
Open G*Power 3. In this

example I am running an a
priori power analysis, so
under “Test family” be sure “t
tests” is chosen, under
“Statistical test” choose
“Means: Difference between
two independent means (two
groups)”; and under “Type of
Power Analysis” choose “A
priori: compute required
sample size.” Next, you'll
need to enter the α-level,
expected effect size, desired
power, and choose “two” for
Tail(s). Leave the “Allocation
ratio N1/N2” set at 1.
After clicking Calculate, you

are provided with the sample
size for each group (n = 64
for each group), the total
sample size (n = 128), critical
t-Value (t = 1.979), and the
degrees of freedom (df =
126). To correctly reject the
null hypothesis in this
planned study with probability
of .80, I need 128 subjects.
What if the expected effect

size was d = .80 instead of d
= .50? How many subjects would I need, if everything else remained the same? In this case, the number
of subjects is n =26 per group for a total sample size of n = 52. Remember, larger expected effect sizes
will be easier to detect so fewer subjects are needed.
-183-
Back to TOC
1. When would the independent groups t-test be used to analyze a bivariate relationship?
2. What will the mean of a sampling distribution of the difference between two independent means be
equal to?
3. Determine the p-value for an independent groups t-test under each of the following conditions:
a. H0: μ1 = μ2, H1: μ1 ≠ μ2, n1 = 8, n2 = 8, t = 3.00

b. H0: μ1 = μ2, H1: μ1 ≠ μ2, n1 = 10, n2 = 10, t = 2.50
c. H0: μ1 = μ2, H1: μ1 > μ2, n1 = 9, n2 = 9, t = 2.20
d. H0: μ1 = μ2, H1: μ1 < μ2, n1 = 25, n2 = 25, t = 1.75
e. H0: μ1 = μ2, H1: μ1 ≠ μ2, n1 = 20, n2 = 20, t = 2.10
f. H0: μ1 = μ2, H1: μ1 > μ2, n1 = 5, n2 = 5, t = 3.10
4. Determine the critical t-value(s) for an independent groups t-test for an alpha level of .05 and .01 under
each of the following conditions:
a. H0: μ1 = μ2, H1: μ1 ≠ μ2

n1 = 8, n2 = 8
b. H0: μ1 = μ2, H1: μ1 > μ2
n1 = 8, n2 = 8
c. H0: μ1 = μ2, H1: μ1 ≠ μ2
n1 = 20, n2 = 15
d. H0: μ1 = μ2, H1: μ1 > μ2
n1 = 20, n2 = 15
e. H0: μ1 = μ2, H1: μ1 ≠ μ2
n1 = 8, n2 = 17
f. H0: μ1 = μ2, H1: μ1 < μ2
n1 = 8, n2 = 17
5. Given the following information, compute the estimated pooled variance and the estimated standard
error of the mean difference between independent groups for n1 = 10, n2 = 13, ŝ 2  5.575 , ŝ 2  4.235 .
6. Given the following information, compute the estimated pooled variance and the estimated standard
error of the mean difference between independent groups for n1 = 50, n2 = 50, SS1 = 40000, SS2 = 60000.
7. Use the following information to answer the questions, below: The following information comes from
two samples that were randomly selected from their respective populations:
Sample A Sample B
nA = 100 nB = 100
X A  120 X B  130
SSA = 1025 SSA = 2050
a. Compute the estimated pooled variance.

b. Compute the estimated standard error of the mean difference between independent groups.
c. Compute the test-statistic (t-value) between the means (assume a non-directional hypothesis).
d. Assuming a non-directional alternate hypothesis what is the p-value?
e. What decision should you make about the difference between the two means?
8. Use the following information to answer the questions, below: The following information comes from
two samples that were randomly selected from their respective populations:
-184-
Back to TOC
Sample A Sample B
nA = 5 nB = 8
X A  15 X B  12
SSA = 125 SSA = 135
a. Compute the estimated pooled variance.

b. Compute the estimated standard error of the mean difference between independent groups.
c. Compute the test-statistic (t-value) between the means (assume a non-directional hypothesis).
d. Assuming a directional alternate hypothesis what is the p-value?
e. What decision should you make about the difference between the two means?
9. Use the following information to answer the questions, below: A professor teaches two sections of the
same course and wants to know whether posting audio recordings of his lectures to his website has any
effect on student grades. The professor posts audio recordings for one class, and for the other class he
posts a written summary of his lecture. Below are the data for both conditions:
Audio Recordings No Audio Recordings

n = 25 n = 25
X  3.35 X  2.55
SS = 45 SSA = 45

b. Compute the estimated pooled variance
c. Compute the estimated standard error of the difference between independent groups.
d. Compute the test-statistic (t-value) between the means.
e. What is the p-value?
f. Using an alpha level of .05, what conclusions should you draw about the null and the alternate
hypotheses?
10. Use the information from Exercise 9 to answer the following questions:
2
a. Calculate the eta-squared (η ). Does this represent a small, medium, or large effect?
b. Calculate the estimated pooled standard deviation.
c. Calculate Cohen's d. What does the value tell you about the separation of the two means?
d. Using G*Power, approximately how much statistical power did this t-Test have?
11. Use the following information to answer the questions, below:
Group A Group B
n = 49 n = 49
X  10 X  10.7
SS = 100 SSA = 120
a. Compute the estimated pooled variance

b. Compute the estimated standard error of the difference between independent groups.
c. Compute the test-statistic (t-value) between the means.
d. What is the p-value (assuming a directional hypothesis)?
e. Using an alpha level of .05, what conclusions should you draw about the hypotheses?
-185-
Back to TOC
2
c. Calculate Cohen's d. What does the value of Cohen's d tell you about the separation of the two
means?
13. Use the following information to answer the questions, below: A researcher studied the influence of
marijuana on reaction times by having five subjects eat a marijuana laced brownie until a certain level of
“stoned-ness” was achieved. Another group ate non-marijuana laced brownies (placebo). All subjects
then completed a speeded task. The reaction times (in seconds) are presented in the table:
Marijuana Placebo
3.00 1.00
3.50 0.50
3.00 1.00
2.00 1.00
2.50 1.50
a. Calculate the means for each group and the sum of squares for each group.
b. Calculate the estimated pooled variance.
c. Calculate the estimated standard error of the difference between independent groups.
f. Compute the test-statistic (t-value) between the means.
g. What is the p-value (assuming a non-directional hypothesis)?
h. Using an alpha level of .05, what conclusions should you draw about the null and the alternate
hypotheses?
2
c. Calculate Cohen's d. What does the value of Cohen's d tell you about the separation of the two
means?
15. Use the following information to answer the questions, below: Dr. Jack Daniels examined the effect of
alcohol consumption on reaction time. One group of eight subjects consumed alcohol until a certain level
of intoxication was reached (determined by physiological measure). A second group of eight subjects was
not given any alcohol, but rather, was given a placebo to drink. All participants then completed a reaction
time task. The reaction times (in seconds) for each of the two groups are below. Use this information to
answer the questions.
Alcohol Placebo
2.500 1.950
2.250 1.750
2.350 1.350
2.500 2.000
2.600 2.150
2.650 1.750
2.600 1.900
2.550 1.150
a. Assuming the hypothesis is that consuming alcohol will slow reaction times, in terms of μ, what
are the null and alternate hypotheses?
b. Compute the mean reaction time for each group.
c. Compute the sum of squares for each group.
d. Compute the estimated pooled variance.
e. Compute the estimated standard error of the difference between independent groups.
-186-
Back to TOC
f. Compute the test-statistic (t-value) between the means.
g. What is the p-value?
16. For each of the following situations, use G*Power 3 to find the total number of subjects that would be
needed to achieve the desired level of Power. (Assume equal group sizes N2/N1 = 1)
a. d = .50; α = .05 (non-directional); Power = .80

b. d = .10; α = .05 (directional); Power = .95
c. d = .80; α = .01 (directional); Power = .85
d. d = .25; α = .01 (non-directional); Power = .90
17. For each of the following situations, use G*Power 3 to find the amount of Power, based on the
parameters given. (Assume equal group sizes N2/N1 = 1)
a. d = .25; α = .05 (non-directional); n = 76

b. d = .55; α = .01 (non-directional); n = 180
c. d = .75; α = .05 (directional); n = 30
d. d = .45; α = .01 (directional); n = 200
18. Using a word processor (e.g., MS-Word, OpenOffice), write out the results of the t-test performed in
#7 in APA format.
#8 in APA format.
#9 in APA format.
#13 in APA format.\
Chapter 15: Correlation
-187-
Back to TOC
15.1 Relationships between Variables
Statistics are not used only to describe
central tendency and variability for a single
variable. Rather, statistics are also used to
describe the relationships between two
dependent variables or between
independent variables and dependent
variables. This is the one of the main
applications of statistics: To determine
whether one variable is statistically
associated to another.
The comic to the right provides a good

illustration of correlation. A correlation is
the statistical relationship or association
between two dependent variables. Simply,
a correlation exists when changes in one
dependent variable are statistically
associated with systematic changes in
another dependent variable. Hence, a
correlation is a bivariate relationship.
This chapter introduces methods for
measuring and describing the strength of
the relationship between two quantitative
and dependent variables.
Several examples of relationships between variables are shown in the table below. Each example lists
two variables, X and Y, each with five scores. Think of each x-y pair as coming from a single person. Can
you identify which examples show a relationship between variables X and Y and which examples do not?
Remember, a relationship exists when changes in one variable are associated with systematic changes in
another variable.
Example 1 Example 2 Example 3 Example 4 Example 5

X Y X Y X Y X Y X Y
2 1 7 1 3 2 2 3 3 2
4 2 6 3 5 4 3 4 4 2
6 3 5 5 7 6 6 3 5 2
8 4 4 7 9 4 4 2 6 2
10 5 3 9 11 2 3 2 7 2
Let’s go through each example and see why there is or isn’t a relationship:
Example 1 shows a relationship between X and Y, because as values of X increase the values of Y
increase. Thus, for every change in X there is a change in Y. I should note that you can also look at a
relationship in the opposite direction too. That is, for every change/increase in Y there is a corresponding
change/increase in X.
Example 2 also shows a relationship between X and Y, because as values of X decrease the values of Y
increase. Thus, even though the scores for the two variables are heading in opposite directions (inverse
relationship), there is still a systematic change in Y that corresponds to the changes in X.
Example 3 shows a relationship. Can you see it? Notice that as X increases from 3 to 5 to 7, Y also
increases from 2 to 4 to 6. But, then the values of Y begin to decrease from 6 to 4 to 2 as X continues to
increase. This is also a relationship, because as the values of X increase there is a systematic change in
Y, but then Y begins to decrease. This is a curvilinear relationship.
-188-
Back to TOC
Examples 4 and 5 do not show relationships between X and Y. In Example 4, the data are apparently
random and there is no consistency in the changes between X and Y. Thus, there is no relationship. In
Example 5, as X increases Y does not change, indicating that there are no systematic changes.
15.2 Scatterplots
Determining whether a relationship is present between two dependent variables can be difficult looking at
x-y data pairs, because the data is rarely ever orderly. To see if a relationship is present between
variables most people begin by creating a scatterplot, which is a graph of the x-y pairs for two variables.
For example, say we measure n = 20 people on the following variables: (1) Family Guy Watching: the
number of Family Guy episodes a person watches per week, and (2) Intelligence, as measured by an IQ
test. The hypothetical data are presented below:
Person Family Guy (X) Intelligence (y) You can see from the data that as the number of Family Guy
A 1 95 (X) episodes watched per week increases so does Intelligence
B 1 96
C 2 96
scores (Y).
D 2 97
E 3 99 In a scatterplot, each x-y data point is plotted without any lines
F 3 98 or bars connecting the data points. Specifically, for the first x-y
G 4 99
H 4 100
pair (Person A: x = 1, y = 95), you locate the position on the
I 5 100 abscissa for X = 1 and locate the position on the ordinate for Y
J 5 102 = 95. Plot a point at that position in the graph and do the same
K 6 102 for the remaining x-y pairs.
L 6 103
M 6 102
N 7 103 After you have done this for each x-y pair you have a
O 7 104 scatterplot, which can be seen in the figure below.
P 8 105
Q 8 107
R 9 106
S 9 108
T 9 107
You can see from the scatterplot that as the X values increase so does the Y values. Thus, more Family
Guy viewing corresponds to higher Intelligence; that is, watching Family Guy and intelligence seem to be
associated. It is good to draw a best fit line in a scatterplot that shows the trend in the data. Best fit lines
-189-
Back to TOC
should not connect the dots; rather, they just need to show approximately where the relationship is
heading.
Scatterplot are used to display whether a relationship between two variables is positive linear, negative
linear, curvilinear, or absent. A linear relationship means that the relationship between variables
appears to occur in a straight line; that is, as values of one variable increase the values of a second
variable change in a consistent manner. A curvilinear relationship exists when there is a change or a
“bend” in the relationship between variables. For example, as the values of one variable increase the
values of the other variable also increase, but at some point the values of the second variable decrease
A positive linear relationship is observed

when the values of both variables have a trend
that occurs in the same direction. That is, as the
values of one variable increase the values of the
second variable also increase. A positive linear
relationship can be seen in Example 1 from
Section 15.1, which is reproduced in the
scatterplot to the right.
A negative linear relationship is observed

when the values of variables have trends that
occur in opposite directions (inverse
relationship). That is, as the values of one
variable increase the values of the other variable
tend to decrease. A negative linear relationship
can be seen in Example 2 from Section 15.1,
which is reproduced in the scatterplot to the
right.
There are many forms of a curvilinear

relationship, but generally, such a relationship
exists whenever there is a change in the
relationship between variables. In Example 3
from Section 15.1 as the values of X increased
the Y values increased, but then began to
decrease. This was a curvilinear relationship.
Thus, there is a change in the relationship,
which produces a bend (curve) in the scatterplot
to the right.
A relationship is absent whenever there is no
-190-
Back to TOC
systematic change between variables. For example, if the data points in Examples 4 and 5 from Section
15.1 demonstrate cases where no relationship is present. These examples are plotted in the scatterplots
below.
15.3 Pearson Product-Moment Correlation Coefficient

The statistic most commonly used to measures the relationship between quantitative variables is the
Pearson Produce-Moment Correlation Coefficient or, more succinctly, the Pearson correlation (r or rXY).
The Pearson correlation is used to measure the strength and direction of a linear relationship between
two quantitative variables. The Pearson correlation can be used to measure the relationship between
variables if the following conditions are met:
The variables must be quantitative; variables cannot be categorical (nominal) or ordinal. Hence, the
Pearson correlation can be used only if the values of each variable come from an interval or ratio scale.
There is another type of correlation, the Spearman rank-order correlation, which is used to measure the
correlation between variables when at least one variable was measured on an ordinal scale. Chi square
analyses can be used if the data comes from nominal scales.
Each variable must produce a wide range of its potential values. This is necessary to “capture” the
relationship between the variables. If a limited or restricted range of potential values are measured you
may not observe the true relationship. For example, say you want to measure the relationship between
freshman GPA and SATs. You collect data on n = 100 freshmen. You find the range GPAs is 1.00 to
3.80, which is nearly the full range of potential values, but you find that the range of the SATs collected is
limited to 1000 to 1200 (the range of potential SAT scores is 600 – 2400). In this case you might not find
a relationship between SATs and GPA, because of the restricted range of the SAT scores.
The relationship is not curvilinear: The Pearson correlation is designed to measure the direction and
degree to which two variables are in a linear relationship. If the relationship between two variables is
known to be curvilinear, the Pearson correlation cannot be used.
As mentioned, the Pearson correlation measures the degree to which the relationship between two
variables is linear; that is, measures the degree of linearity between variables. The Pearson correlation
has a range between -1.00 to +1.00. The closer to -1.00 or +1.00, the more linear the relationship is
between the variables. If the value of the Pearson correlation is found to be equal to -1.00 or +1.00, this
indicates that there is a perfect linear relationship between variables.
The absolute value of the Pearson correlation is the strength of the relationship between the variables,
that is, the degree to which the variables are in a linear relationship. The sign (+/-) of the Pearson
correlation tells you whether the relationship is positive-linear or negative-linear; the sign says nothing
-191-
Back to TOC
about the strength of the linear relationship. A perfect Pearson correlation is equal to +1.00 or -1.00. If
such a linear relationship is obtained it is possible to predict a value of Y given that we have a value of X,
and vice versa. However, it is rarely the case that the relationship between two variables is perfectly
linear; thus, it will rarely be the case that you will obtain a correlation equal to +1.00 or -1.00. In reality,
variation of X and Y decreases the strength of linear relationship between variables and the Pearson
correlation will deviate from +1.00 or -1.00.
A zero correlation is a case in which the Pearson correlation is equal to zero (r = 0). In this case there is
no relationship between variables; they are completely independent. In the next section, I discuss how to
calculate the Pearson correlation both in terms of its definition and also computationally.
15.4 Calculating the Pearson Correlation (r)

Generally, the Pearson correlation is used to assess the relationship between variables when:
 Both dependent variables are quantitative and the variables approximate interval or ratio data
 The two dependent variables are measured on the same individuals
 The observations on each variable are between-subjects in nature
The Pearson correlation is defined as standardized covariance between two quantitative variables. What
the heck does this mean? Recall from in Chapter 7 (standard scores) that when calculate a z-Score, you
divide the difference between a raw score and the mean of a distribution by the standard deviation of the
distribution. This means that the difference between the raw score and the mean was standardized. The
Pearson correlation is calculated by doing something similar. A measure referred to as covariance is
divided by the product of two standard deviations; hence, this measure of covariance is standardized. So
what is covariance?
Before introducing covariance, let me introduce a set of data that will be used to calculate the Pearson
correlation. Below is a set of data for n = 10 people. In this hypothetical set of data, assume that I
measured the age (X) of these ten people and measured the number of books that each of these 10
people read per month (Y). I measured these two variables for these 10 people to see if there is any
relationship between a person’s age and book reading behavior:
Name Age (X) Books Read Recall that variance is the average variability among scores for
Per Month (Y) a single variable. You can examine the scores for the variable
Rob 22 8 age (X) and for the variable Books Read per Month (Y) in the
Sheri 20 4 table to the left and see that those scores vary; this variance.
J. 22 4
Sean 21 3 Covariance is the average covariation of scores between
John 35 6 variables, that is, the average amount by which two variables
Riggs 32 7 are changing together. The difference between variance and
Blasko 38 4 covariance is that covariance is the average variation between
Vincent 40 6 scores from two variables; whereas variance is the average
George 20 3 variation among scores of a single variable.
Bela 40 5
How is covariance calculated? Recall the formula for the estimated population variance:
ŝ 
2 
 XX 2
n 1
-192-
Back to TOC
The formula for [estimated] covariance is similar:
cov xy 

 XX YY  
n 1
To calculate covariance you divide the numerator by n – 1, rather than by n, because we are
estimating the covariance in a population from sample data. Covariance is formally defined as the
average sum of the cross produces between two variables. The sum of the cross products (SCP) is the
sum of the products of the mean centered scores from each variable; and this is the numerator in the
formula above. That is, the formula for the sum of the cross products is:

SCP   X  X Y  Y  
Calculating sum of cross products is similar to calculating sum of squares and, indeed, the sum of the
cross products is conceptually similar to sum of squares. That is, the sum of squares is the total variation
within a set of scores; whereas the sum of the cross products is the total variation between two sets of
scores. The sum of the cross products for the data above is calculated in the table below from the data
presented earlier:
X  X  Y  Y  X  XY  Y

Books Read
Name Age (X) Per Month (Y)
Rob 22 8 -7 3 -21
Sheri 20 4 -9 -1 9
J. 22 4 -7 -1 7
Sean 21 3 -8 -2 16
John 35 6 6 1 6
Riggs 32 7 3 2 6
Blasko 38 4 9 -1 -9
Vincent 40 6 11 1 11
George 20 3 -9 -2 18
Bela 40 5 11 0 0
∑X = 290 ∑Y = 50 SCP = 43
X  29 Y5
As you can see in the table above, the first step to calculating the sum of the cross products is to
calculate the mean of each variable and then subtract the mean of each variable from each score. These
first two steps are identical to those for calculating the sum of squares. Next, the mean-centered
difference scores for each variable are multiplied to create cross products. Finally, the products are
summed to obtain the sum of the cross products (SCP), which in this example is SCP = 43. Importantly,
unlike sum of squares, which is always positive, the sum of cross products can be positive or negative. If
SCP is positive the covariance will be positive and the relationship between the variables is positive-
linear. If SCP is negative the covariance will be negative and the relationship between the variables is
negative-linear. If SCP = 0, it indicates a zero relationship.
The next step is to calculate the covariance. Conceptually, all that you need to calculate covariance is to
divide SCP from above by n – 1:
cov xy 

 XX YY

SCP

43  4.778
n 1 n  1 10  1
-193-
Back to TOC
Let’s stop for a moment and recap what we have done: We wanted to calculate the Pearson correlation
between Age and the number of Books Read per Month. The Pearson correlation is defined as
standardized covariance between two variables. Covariance is the average covariation between two
variables, that is, average sum of the cross products. From the data presented above, the sum of the
cross products was found to be SCP = 43. The covariance was calculated by dividing 43 by 10 – 1 to get
covxy = 4.778.
The next step to calculate a Pearson correlation is to standardize the covariance. The formula for the
Pearson correlation is:
cov xy
r
ŝ X ŝ Y
The numerator is covariance (covxy = 4.778). The denominator is the product of the estimated standard
deviation of each variable. Thus, we must calculate the estimated standard deviation for each variable,
which is started by calculating the sum of squares for each variable. This is shown in the table below:
Name Age (X)

Books Read
Per Month (Y) X  X  Y  Y  X  XY  Y X  X  Y  Y 
2 2
Rob 22 8 -7 3 -21 49 9
Sheri 20 4 -9 -1 9 81 1
J. 22 4 -7 -1 7 49 1
Sean 21 3 -8 -2 16 64 4
John 35 6 6 1 6 36 1
Riggs 32 7 3 2 6 9 4
Blasko 38 4 9 -1 -9 81 1
Vincent 40 6 11 1 11 121 1
George 20 3 -9 -2 18 36 4
Bela 40 5 11 0 0 121 0
∑X = ∑Y = 50 SCP = 43 SSX = SSY = 26
290 647
X  29 Y5
The estimated of the standard deviation for each variable are:
SSX 647
ŝ X    71.889  8.479
n 1 10  1
SSY 26
ŝ Y    2.889  1.7
n 1 10  1
Now, we have all three pieces needed to calculate the Pearson correlation:
cov xy 4.778 4.778

r    0.331
ŝ X ŝ Y 8.4791.7 14.414
It is customary to report the Pearson correlation rounded to two decimal places in APA format (don’t ask
me why, I don’t make up the rules). Thus, our Pearson correlation would be reported as r = .33 in a
research report.
-194-
Back to TOC
One obvious question is what a correlation coefficient of this size indicates. Generally speaking, the larger
the absolute value of the Pearson correlation the better. That is, the closer to +1.00 or to -1.00 (the farther
away from zero), the stronger the linear relationship between the variables. Nonetheless, a correlation of r
= .33 is generally large in the behavioral sciences, because there is so much variation in behavior. Later
in the chapter, we’ll discuss the effect size of the Pearson correlation.
15.5 Proportion of Explained Variance and Residual (Unexplained) Variance

2
The coefficient of determination (r ) is calculated by squaring the Pearson correlation. Thus, for our
2 2
Age/Book Reading relationship above, the coefficient of determination is r = 0.331 = 0.11. This value is
the proportion of variance that is accounted for (explained) in the relationship between two variables.
Specifically, the coefficient of determination is the proportion of x-y pairs in which the values of x and y
co-vary in the direction indicated by the Pearson correlation. In the case of a positive linear relationship,
the coefficient of determination is the proportion of X scores that increase with the Y scores. In the case
of a negative linear relationship, the coefficient of determination is the proportion of X scores that
2
decrease as Y increases. You can think of r as the proportion of scores that are correctly predicted by
the relationship. Importantly, the coefficient of determination can never be negative and has a range of 0
to 1, because it is a proportion; bigger values are better.
The residual variance has no symbol, but it is calculated by simply subtracting the coefficient of
2 2
determination (r ) from 1; thus, residual variance = 1 – r . Residual variance is the proportion of variance
between two variables that is not accounted for in the relationship. Thus, it is the proportion of X-Y scores
that do not co-vary together in the direction indicated by the relationship. From the example used here,
residual variance = 1 - 0.11 = 0.89. You can consider this to be the error in the relationship between two
variables; that is, this is the proportion of variance that is not accounted for, or predicted.
15.6 Characteristics of the Pearson Correlation

There are several important characteristics you should know about the Pearson correlation (as well as the
Spearman correlation discussed later).
First and most important is the fact that

having a non-zero correlation between two
variables does not mean that one variable
caused the variation in the other variable.
Correlation does not mean causation!
From the example in the preceding
sections, the Pearson correlation of r =
.331 tells us there is a positive linear
relationship between age and number of
books read per month. The Pearson
correlation does not say anything about
age causing a person to read more. All the
Pearson correlation says is that the two
dependent variables were statistically
associated. The only way to determine
whether the changes in one variable are
causing the changes in another variable is
by manipulating an independent variable
and conducting an experiment.
Second, and related to the preceding point, there is often a misuse and abuse of correlations in the
media. Specifically, a relationship between variables can exist, but there may be no causal explanation.
That is, there may be a correlation present but manipulations of one variable (in an experimental design)
may not have produced the changes in the other variable. For example, there is a known positive
-195-
Back to TOC
correlation between exposure to violent television and the amount of aggression in children, where more
violent television exposure is associated with more aggressive behavior in children. It is impossible to tell
whether one variable causes the changes in other. It could be that watching violent television causes a
child to act aggressive; or it could be that aggressive children like to watch violent television. What is
more is that there could be a third variable that mediates the relationship; that is, the relationship between
violent television and aggression may depend on the presence of some other factor. For example,
perhaps this relationship between violent television and aggression exists only in children with conduct
disorders (childhood equivalent of antisocial personality disorder) and is absent in children without the
disorder. The point is that one should never over-interpret a correlation.
Third, it is occasionally true that a relationship is found by two seemingly unrelated variables; hence, the
relationship is spurious (random). The relationship exists but really does not make any sense and occurs
without theory; or there is some very simple explanation for the relationship. For example, that there is a
strong positive correlation between numbers of churches and numbers of prostitutes: As the number of
prostitutes increases the number of churches increases. Most right-wing Christians (no offense to any of
you out there) take this to mean that people purchasing the company of prostitutes are likely to seek god
and need a church, or that the prostitutes need church to repentant sins. I’m sorry to burst Pat
Robertson’s bubble, but nothing could be further from the truth. The relationship exists because of
population density. For example, what areas are naturally going to have more prostitutes and more
churches? A city! And what areas are likely to have fewer churches and fewer prostitutes? Small/rural
towns! Thus, the relationship between churches and prostitutes exists, but for no real theoretical or
meaningful reason: It exists due to natural population density.
Finally, the restricted range problem alluded to earlier occurs when the range of collected scores for some
variable represents only a small proportion of all possible scores for that variable. For example, in the
memory and cognition literature there is something called ‘working memory’ which can be measured with
something called the OSPAN test, which ranges from 0 to 1. If you measured the correlation between
OSPAN and GPA, but your OSPAN scores were only in the range from 0.50 to 0.60, you have a
restricted range of all of the possible OSPAN scores. Thus, you may not detect the true relationship
between OSPAN and GPA, only part of the relationship. This is a big problem, because you may find a
significant linear relationship if you have a restricted range, when in reality, the relationship between the
two variables is curvilinear. Because of your restricted range, you could not detect that relationship.
15.7 Hypotheses and Sampling Distributions of the Pearson Correlation

Earlier sections of this chapter addressed how a statistical association between two variables could be
measured with the Pearson correlation. This section addresses how we determine whether a correlation
between variables is statistically significant.
Determining statistical significance of a Pearson correlation depends on whether the correlation under the
null hypothesis is assumed to be zero or to be non-zero value. When the correlation under the null
hypothesis is assumed to be zero you use a type of t-Test to assess the significance of the Pearson
correlation. When, under the null hypothesis, the Pearson correlation is assumed to be a value that is
different than zero, you use Fisher’s z-Test to assess statistical significance. Before we cover these tests,
we will discuss hypotheses in inferential testing of correlations.
Recall, the value of a correlation has a range of -1.00 to +1.00 and a zero-correlation (r = 0) indicates no
association between variables. Generally, the null hypothesis predicts no relationship between the
variables. The symbol for the Pearson correlation in a population is the Greek lowercase rho (ρ). Thus,
the null and alternate hypotheses predict that:
H0: ρ = 0 H1: ρ ≠ 0
-196-
Back to TOC
This is a non-directional alternate hypothesis for a Pearson correlation, because the alternate is stating
that the Pearson correlation will be something other than zero. But, the alternate hypothesis is not saying
whether the correlation will be positive-linear or negative-linear.
It is also possible to generate directional alternate hypotheses for the Pearson correlation. If the alternate
hypothesis predicts the Pearson correlation will be positive-linear the null and alternate hypotheses are:
H0: ρ = 0 H1: ρ > 0

If the alternate hypothesis predicts the Pearson correlation will be negative-linear, the null and alternate
hypotheses are:
H0: ρ = 0 H1: ρ < 0
15.8 Determining Statistical Significance of the Pearson Correlation
Say you are interested in whether the Pearson correlation between age and book reading behavior from
the earlier sections is statistically significant. You are not sure whether there is a positive or negative
correlation between these variables; thus, you decide to use a non-directional alternate hypothesis:
H0: ρAge,Books = 0 H1: ρAge,Books ≠ 0

Recall, from that example, n = 10 and r - 0.331. To determine whether this correlation is statistically
significant, we use the following t-Test:
r
t
1  r /n  2
2
Before going through the calculations, note this t-test is used only when ρ is predicted to be zero under
the null hypothesis. First, we’ll select an alpha of α = .05 for a non-directional alternate hypothesis.
Second, for the Pearson correlation degrees of freedom are equal to n – 2, because we need to account
for the degrees of freedom in each dependent variable. In the present example, df = 10 – 2 = 8.
Plugging r and n into the t-Test expression from above, the obtained value is solved below:
0.331
t
1  0.331 /10  2
2
0.331
t
1  0.110 /8
2
0.331
t
0.89 / 8
0.331
t
0.111
0.331
t
0.333
t  0.934
This is the test statistic that we use to assess the statistical significance of the Pearson correlation. To
determine whether this correlation is statistically significant, use Table 2 in Appendix A (t-Tables) by
looking up the value of the test statistic (use t = 1.00) and degrees of freedom in the correlation (df = 8) to
-197-
Back to TOC
find the p-value associated with this correlation, which is p = .3466. Because this p-value is greater than
the selected alpha level (.05), we conclude the Pearson correlation (r = 0.331) is not statistically
significant. Thus, we retain the null hypothesis and make no decision about the alternate hypothesis. In
layman’s terms, we conclude there is insufficient evidence to identify a statistically significant relationship
between age and reading.
15.9 Determining Significance when ρ ≠ 0

In cases where ρ is expected to be something other than zero under the null hypothesis, the t-Test for the
Pearson correlation is inappropriate. This is because the sampling distribution of Pearson correlation
coefficients is skewed when ρ ≠ 0. To overcome this problem, you must make use of the logarithmic
transformation of r into r’ (r-prime) and the formula for converting any given r into r’ is:
r  .50loge 1  r   loge 1  r 
In the formula, r is the Pearson correlation and loge is the natural logarithm of a number. You could use
this formula to determine r’ or you could use Fisher’s z-Transformation Index, which is Table 6 in
Appendix A. To use this table, look up the value of your Pearson correlation value (r) in the r(ρ) columns.
The value to the right of this value in the r’(ρ’) column is the corresponding r’ value. Importantly, r’ values
are normally distributed, so we do not have to worry about skew.
To test whether a Pearson correlation (r) is statistically different from some non-zero population
correlation parameter (ρ), You calculate a z-Score for the difference between r’ and the Fisher’s
transformed value of ρ (ρ') under the null hypothesis, using the standard error of r’ as the denominator.
This is Fisher's z-Test:
r  
z
r
The standard error of r’, which is the error term (denominator) in the formula above, is calculated from:
1
r  
n 3
For example, say the true (population) correlation between studying and sexual activity is ρ = .20. We
want to know whether the correlation between studying and sexual behavior for a sample of students
from a particular university differs from ρ = .20. Hence, we have a non-directional alternate hypothesis:
H0: ρStudying,Sex = 0.20 H1: ρStudying,Sex ≠ 0.20

We randomly select n = 50 students from this particular university and survey each student to assess the
amount of time each student spends studying and the amount of sexual activity each student engages.
After measuring studying time and sexual activity for these 50 students we find r = 0.500.
To determine the statistical significance of this Pearson correlation, first transform the predicted Pearson
correlation under the null hypothesis (ρ = .20) into ρ’ using Table 6 in Appendix A. Fisher’s transformed
value for ρ = .20 is ρ’ = .203. Next, transform the obtained Pearson correlation (r =.500) into an r' value
using the same table. Fisher’s transformed r’ for r = .50 is r’ = .549.
1 1 1 1
Next, calculate the standard error of r’: r      0.146
n 3 50  3 47 6.856
r   0.549  0.203 0.346
The obtained z-Score is: z    2.37
r  0.146 0.146
-198-
Back to TOC
Lookup the test statistic (z = 2.37) in column 1 of Table 1 in Appendix A (z-Tables) and locate the
probability in column 4 of that row, which is p = .0178. We use the probability in column 4 (not column 3),
because this is a non-directional alternate hypothesis. Because the p-value (.0178) is less than the
selected alpha level we conclude the sample correlation between studying and sexual activity (r = 0.500)
is significantly different from the population correlation of ρ = .20. Or more generally, this sample
correlation between studying and sexual activity is statistically significant.
15.10 Power and Effect size for Pearson Correlation

The effect size of the Pearson
Effect Size Pearson’s r Coefficient of
correlation is the absolute value of the 2
Determination (r )
Pearson correlation. That is, the value
of the Pearson correlation as a range “Small” (“Weak”) .10 .01
from -1.00 to +1.00, where values “Medium” (“Moderate”) .30 .09
closer to |1.00| indicate a stronger liner “Large” (“Strong”) .50 .25
relationship. Cohen provides useful labels to describe how strong a relationship is based on the size of a
2
Pearson correlation. The table reports the minimum r and r values that correspond to ‘weak’, ‘moderate’,
and ‘strong’ effect sizes for the Pearson correlation (also called ‘small’, ‘medium’, and ‘large’ effect sizes).
Like with the independent groups t-test in Chapter 14 you can use effect size of the Pearson correlation
and sample sizes to determine the achieved statistical power (1 - β) of the correlation. G*Power (Faul, et
al., 2007) provides a useful tool for performing power analyses on the Pearson correlation. We’ll perform
a power analysis on the correlation between age and books read per month (r = .331).
Open G*Power 3 and under “Test family,” select “t tests” is chosen. Under “Statistical test” choose
“Correlation: point biserial method”,
and under “Type of Power Analysis”
choose “Post hoc: Compute achieved
power.” Next, you'll need to enter the
α-level, sample sizes, and effect size
(r) from the correlation test above.
The alpha level was α = .05, we had a
non-directional hypothesis, n = 10,
and r = .331. After clicking calculate,
G*Power tells you that the statistical
power is .165293, which is extremely
low. As in Chapter 18, the
conventional level of statistical power
is .80 or greater.
Recall from Section 15.7 that the

significance test on the person
correlation was non-significant.
Because our power is so low we
cannot conclude the relationship
between age and books read per
month this is truly non-significant.
Stated differently, our non-significant
correlation might be a Type II error.
To determine whether this is a truly
non-significant relationship or a Type
II error we need to increase the
sample size. If we increase the
sample size and find the correlation is
still non-significant, we can then
-199-
Back to TOC
conclude the relationship is non-existent.
We can also use G*Power in a

priori power analyses of the
Pearson correlation to determine
the number of subjects needed to
end up with a desired level of
power. For example, assume am
planning to collect data from a
sample of subjects on two
dependent variables. I know from
prior studies the correlation
between the variables is generally
“moderate” and near r = .55. For my
inferential test I am planning to use
α = .01 and with a non-directional
alternate hypothesis.
To determine the appropriate

sample size, open G*Power 3 and
under “Test family” be sure “t tests”
is chosen, under “Statistical test”
choose “Correlation: Point biserial
method”; and under “Type of Power
Analysis” choose “A priori: compute
required sample size.” Next, you'll
need to enter the α-level, expected
effect size, desired power, and
choose “two” for Tail(s). For this
example, assume I want Power =
.80.

provided with the total sample size needed to achieve this amount of Power, which is n = 31 in this
example. What if I wanted more Power, say, .95? Entering .95 into the Power areas, I am now told I need
n = 45 subjects. Thus, the more power you want the more subjects you will need.
15.11 The Spearman Rank-Order Correlation (rS)

Recall that in order to use the Pearson correlation the data must come from two quantitative variables
that were measured on an interval or ratio scale. However, it might be the case that the data from one or
both of the variables is not from an interval scale or ratio scale, but from an ordinal scale. When data from
one or both of your variables was measured on an ordinal scale, you must use the Spearman rank-order
correlation (rS) to measure the relationship between the variables.
The Spearman rank-order correlation, like the Pearson correlation, measures the degree to which two
variables are related. In the case of the Spearman correlation, the relationship reflects the degree to
which the rank-ordering of the values for each variable agree or disagree. That is, because the data from
one or both of the variables comes from an ordinal scale, which is the rank-ordering of entries for a
variable, the Spearman correlation measures the degree to which the rank orderings for each variable are
consistent. Generally, the Spearman correlation is used to assess the relationship between variables
when:
 Both dependent variables are quantitative and at least one variable approximate an ordinal scale
 The two dependent variables are measured on the same individuals
 The observations on each variable are between-subjects in nature
-200-
Back to TOC
Say we are college football fans and want to Team Place Points/Game
measure the relationship between order several Springfield Isotopes 2 33
teams finish at the end of a season (1 , 2 , 3 ...) Shelbyville Shelbyvillians 5 25
st nd rd
and the average number of points each team Quahog Clams 3 30

scored per game. The variable Place of Finish is Pawtucket Patriots 8 15
st
ordinal, because a team finishes in 1 place, 2
nd Evergreen Geoducks 1 35
place, etc, but the variable Points per Game is Faber Mongols 4 31
ratio, because a team can score zero or more. PCU Whopping Cranes 7 17
This is okay because the Spearman correlation is Mongol-University 6 27
used when at least one variable is on an ordinal Deathworms
scale; the other variable can be interval, ordinal, or ratio.
Although it is okay to measure the relationship

between an ordinal variable and a ratio/interval Team Place Points/Game
variable with the Spearman correlation, you have Springfield Isotopes 2 2
to first convert any ratio or interval values into Shelbyville Shelbyvillians 5 6
ordinal values. That is, look at the data for the Quahog Clams 3 4
Points/Game variable and decide which value is Pawtucket Patriots 8 8
largest; that value will be assigned an ordinal Evergreen Geoducks 1 1
value of 1. Next, decide which value is second Faber Mongols 4 3
largest; that value will be assigned an ordinal PCU Whopping Cranes 7 7
value of 2. And so on. The table to the right shows Mongol-University
6 5
what you have after converting the Points/Game Deathworms
values into rank-ordered values on an ordinal scale:
There is one condition that would preclude you from using the Spearman correlation at this point. For
both variables, the ordinal rankings on either one variable cannot be duplicated; that is, each rank-order
value for each variable should occur exactly once. For example, if two teams had scored an average of
20 points per game, they would obviously have the same ordinal ranking. If this occurs, you cannot use
the Spearman correlation. But, because this is not the case here, we do not have to worry about it! The
formula for the Spearman correlation is:
6 D 2
rs  1 

N N2 1 
In the formula above, N is the number of cases, or pairs, evaluated. In in this example, N = 8, because
2
there are eight teams being evaluated. The value ΣD is the sum of the squared differences in rankings
between the variables. To compute this term you subtract the ordinal value for the variable on the right
from the corresponding ordinal value from the variable on the left, then square those difference scores,
2
and then add the squared differences. This is demonstrated in the D and D columns in the table below:
2
Team Place Points/Game D D
Springfield Isotopes 2 2 0 0
Shelbyville Shelbyvillians 5 6 -1 1
Quahog Clams 3 4 -1 1
Pawtucket Patriots 8 8 0 0
Evergreen Geoducks 1 1 0 0
Faber Mongols 4 3 1 1
PCU Whopping Cranes 7 7 0 0
Mongol-University Deathworms 6 5 1 1
2
N=8 ∑D = 4
2
Once you have the D value, you can insert that and the value of N into the formula above to calculate the
Spearman correlation. This is done below:
-201-
Back to TOC
6 D 2 64 24 24 24
rs  1 

N N 12
  1

8 8 1
2
  1
864  1
 1
863
 1
504
 1  0.048  0.952
As was the case with the Pearson correlation, it is customary to report the Spearman rank-order
correlation to two decimal places. Thus, the correlation between Place of Finish and point Per Game is
rS = .95.
The larger the Spearman correlation, the greater is the consistency in the ranking (ordinal) values
between the two variables. Importantly, positive values of the Spearman correlation indicate agreement in
the rankings; that is, the rankings are highly similar for both variables. In contrast, a negative Spearman
correlation indicates disagreement in the rankings; that is, the rankings are very different and are in
opposite directions.
15.12 Determining Significance of Spearman Rank Order Correlation (rS)

The statistical significance of the Spearman correlation can also be determined with a type of t-Test:
n2
t  rS
1  r2
The obtained t-value is:
n2 82 6 6
t  rS  0.952  0.952  0.952  ...
1  rS2 1  0.952 2 1  0.906 0.094
...t  0.952 63.83  0.9527.989  7.606
To determine the statistical significance of the Spearman correlation, we need to locate a p-value in Table
t (t-tables), because this test is a variant of the t-test. Assume we are using a non-directional alternate
hypothesis with an alpha level of α = .05. As was the case with the Pearson correlation, the degrees of
freedom in the Spearmen correlation are df = n – 2, where n is the number of ordered pairs.
Because the t-values in the t-tables do not go above 4.00, but the test statistic is equal to t = 7.607, we’ll
use t = 4.00. Looking up this test statistic value and df = 8, we find a p-value equal to p = .0039, which is
less than the chosen alpha level (.05); hence, we conclude the Spearman correlation between place of
finish and tem points per game (rS = 0.952) is statistically significant.
-202-
Back to TOC
1. Draw a scatterplot for the following data:
Individual X Y
1 10 9
2 8 7
3 6 5
4 9 8
5 10 8
6 8 8
7 5 6
2. Draw a scatterplot for two variables that are negatively correlated.
3. Draw a scatterplot for two variables that are negatively correlated.
4. What does the magnitude of a correlation coefficient tell you about the relationship between two
variables?
5. What does the sign of a correlation coefficient tell you about the relationship between two variables?
6. Under what conditions is Pearson correlations typically used to analyze a bivariate relationship?
7. What is the coefficient of determination? What does it tell you? What is the coefficient of alienation?
What does it tell you?
8. State the critical t-Values for a test of the Pearson correlation under each of the following conditions:
a. H0: ρ = 0, H1: ρ ≠ 0, α = .05, n = 30
b. H0: ρ = 0, H1: ρ > 0, α = .05, n = 30
c. H0: ρ = 0, H1: ρ ≠ 0, α = .01, n = 22
d. H0: ρ = 0, H1: ρ < 0, α = .01, n = 22
e. H0: ρ = 0, H1: ρ ≠ 0, α = .05, n = 16
f. H0: ρ = 0, H1: ρ ≠ 0, α = .01, n = 45
9. Locate the p-values for a test of the Pearson correlation under each of the following conditions:
a. H0: ρ = 0, H1: ρ ≠ 0, n = 20, t = 2.50
b. H0: ρ = 0, H1: ρ > 0, n = 20, t = 2.50
c. H0: ρ = 0, H1: ρ ≠ 0, n = 36, t = 2.10
d. H0: ρ = 0, H1: ρ < 0, n = 36, t = 2.10
10. Use the following data to answer the questions, below.
Individual X Y
1 10 7
2 8 8
3 10 8
4 8 7
5 7 9
6 9 7
7 11 5
8 8 6
9 10 6
10 9 7
a. Calculate the sum of squares for each variable and the sum of the cross products.
-203-
Back to TOC
b. Calculate the estimated standard deviations for each variable.
c. Calculate the covariance.
d. Calculate the Pearson correlation.
e. Compute the coefficient of determination and the coefficient of alienation. What does the
coefficient of determination tell you about the relationship?
f. Calculate the t-value for the Pearson correlation.
g. Assuming a non-directional hypothesis, what is the p-value for this correlation?
h. Using an alpha level of α = .05, is the Pearson correlation statistically significant?
i. Using G*Power, how much power was there to detect the relationship?
11. Use the following data to answer the questions, below.
Individual X Y
1 8 7
2 5 5
3 8 7
4 5 5
5 6 6
6 7 5
7 6 6
8 7 5
9 3 4
10 4 2
11 3 4
12 4 4
a. Calculate the sum of squares for each variable and the sum of the cross products.
b. Calculate the estimated standard deviations for each variable.
c. Calculate the covariance.
d. Compute the Pearson correlation.
e. Compute the coefficient of determination and the coefficient of alienation. What does the
f. Calculate the t-value for the Pearson correlation.
g. Assuming a non-directional hypothesis, what is the p-value for this correlation?
h. Using an alpha level of α = .05, is the Pearson correlation statistically significant?
i. Using G*Power, how much power was there to detect the relationship?
12. Use the following to answer the questions below. Dr. Vader wants to measure the relationship
between fear of the darkside of the Force and conformity. He randomly selects (i.e., threatens with Force
Choke Hold) individuals from around the Galactic Empire. Each individual rates how much they fear the
darkside of the force (X) on a scale of 1 – 10, where higher scores indicate more fear of the darkside, and
each individual is measured on their willingness to conform to the Empire's demands (Y) on a scale of 1 -
10, where higher scores indicate more willingness to conform. The data from this sample appear below:
Individual X Y
Dan Solo 1 3
Flubacca 2 1
Duke Cloudwalker 7 9
Lando Griffin 5 6
Princess Vespa 7 9
Dark Helmet 1 2
President Scroob 6 8
Jabba is Nuts 5 7
Oil Can Notopen 2 1
Mr. Spot 4 5
-204-
Back to TOC
a. In terms of ρ, state the hypotheses.
b. Calculate the mean of each variable.
c. Calculate the sum of squares for each variable and the sum of the cross products.
d. Calculate the estimated standard deviations for each variable.
e. Calculate the covariance.
f. Compute the Pearson correlation.
g. Compute the coefficient of determination and the coefficient of alienation. What does the
h. Calculate the t-value for the Pearson correlation.
i. Assuming a non-directional hypothesis, what is the p-value for this correlation?
j. Using an alpha level of α = .05, is the Pearson correlation statistically significant?
13. Use the following to answer the questions below. Recently, researchers examined the relationship
between political attitude (liberal/conservative) and perceptual sensitivity (ability to detect small changes).
One hundred sixty five students rated their political attitude on a scale of 1 – 11 and completed a
perceptual sensitivity task. Sensitivity was measured by a metric called d' (“d-prime”), with larger values
indicating greater sensitivity to detecting change. The researchers found a Pearson correlation between
political attitude and d' of r = -.248.
a. Assuming the researchers made no prediction about the direction of the relationship, in terms of
ρ, what are the null and alternate hypotheses?
b. Calculate the t-value for the Pearson correlation.
c. What is the p-value for this correlation?
d. Using an alpha level of α = .05, is the Pearson correlation statistically significant? What should
these researchers conclude about the relationship?
e. Does the Pearson correlation represent a small, medium, or large relationship? Using G*Power,
how much statistical Power did the researcher have for rejecting the null hypothesis?
14. Use the following to answer the questions below. Researchers examined the relationship between
how much a person agrees with the Republican Party and authoritarian personality. They asked 177
students to rate how much they agreed with the Republican Party on a scale from 1 to 7 and gave them a
test that assessed the authoritarian nature of each subject’s personality, with larger scores indicating a
more authoritarian personality. The researchers found a Pearson correlation between agreement with the
Republican Party and the authoritarian personality test scores of r = .402. They assumed that the more
one agrees with the Republican Party the higher their score would be on the authoritarian personality test
a. In terms of ρ, what are the null and alternate hypotheses?

e. Does the Pearson correlation represent a small, medium, or large relationship? Using G*Power,
how much statistical Power did the researcher have for rejecting the null hypothesis?
15. Use the following to answer the questions below. Dr. Tomsurfrend, an educational psychologist, is
interested in the relationship between poor time management and grades among high school seniors. He
predicts that the more time a person spends on the social networking site Facebook, the lower their
grades will be. He asks 50 high school seniors at Whitesboro Senior high school how much time they
spend, per day, on the Facebook and also obtains the each student's GPA from the previous semester.
The correlation is r = -.45 Using this information, answer the following questions:

-205-
Back to TOC
16. Use the following to answer the questions below. Dr. Hefner is trying to answer a long-debated
question: 'Is shoe size statistically associated with penis length?' Dr. Hefner samples twenty males of the
approximately same age, height, weight, and health. He measures each man's shoe size and, using a
yardstick, measures each man's penis length (isn't it surprising that only twenty males signed up for this
study?) The correlation between shoe size and penis length is +.08.

17. Use the following data to answer the questions below. Below are the rankings of the top ten students
at a school, as determined by the dean and the president of the school.
Student Dean President

Student A 10 10
Student B 4 6
Student C 2 8
Student D 3 3
Student E 9 7
Student F 8 9
Student G 1 4
Student H 6 2
Student I 5 1
Student J 7 5
2 2
a. Calculate the D and D scores. What is the value of ∑D and of ∑D ?
b. Calculate the Spearman correlation.
c. Calculate the t-value for the Spearman correlation.
d. What is the p-value for this correlation?
e. Using an alpha level of α = .05, is the correlation statistically significant? What should the Dean
and the President conclude about their rankings?
For Exercises 18 – 20, hypothetical values of r, ρ, and n are given. (a) Use Fisher's z-Transformation to
determine the values of r' and ρ'. (b) Calculate σr'. (c) Calculate the test statistic (z-test) for the correlation.
and (d) Determine the p-value. (e) Determine whether the value of r' is significant. Assume all hypotheses
are non-directional (two-tailed) with α = .05.
18. r = -.25, ρ = -.10, n = 30
19. r = .86, ρ = .25, n = 20
20. c. r = .35, ρ = -.15, n = 10, α = .01
21. In a recent national poll, all registered voters were asked how much they agreed with the theory of
evolution on a scale of 1 (strongly disagree) to 11 (strongly agree) and were also asked their overall
political attitude on a scale of 1 (extremely liberal) to 11 (extremely conservative). The correlation was ρ =
-.50, suggesting the more politically conservative someone is the less they tend to agree with the theory
of evolution. Dr. Denis Leary is interested in whether this correlation is stable in various states, because
some states are much more politically liberal than others and some are much more politically
conservative than others. Dr. Leary samples 100 residents from Vermont who are registered voters and
obtains their ratings to each of the two questions above. He finds that the correlation between agreement
with the theory of evolution and political attitude is r = -.10 Use this information answer the following
(assume α = .01 for all questions):
-206-
Back to TOC
a. Dr. Leary wants to know whether the correlation obtained from his sample of Vermont voters is
greater than the national correlation. In terms of ρ, state the hypotheses.
b. Determine the Fisher's Transformation of the r and ρ values.
c. Calculate σr' for the sample of Vermont voters.
d. Calculate the z-Score of r'.
f. Is the difference between r and ρ statistically significant?
22. For each of the following situations, use G*Power to find the total number of subjects that would be
needed to achieve the desired level of Power.
a. r = .50; α = .05 (non-directional); Power = .80

b. r = .10; α = .05 (directional); Power = .95
c. r = .80; α = .01 (directional); Power = .85
d. r = .25; α = .01 (non-directional); Power = .90
e. r = -.70; α = .01 (directional); Power = .85
f. r = -.25; α = .05 (directional); Power = .90
23. For each of the following situations, use G*Power, find the amount of achieved Power based on the
parameters given.
a. r = .25; α = .05 (non-directional); n = 45

b. r = .55; α = .01 (non-directional); n = 110
c. r = -.35; α = .05 (directional); n = 30
r = -.45; α = .01 (directional); n = 25
-207-
Back to TOC
Chapter 16: Regression
16.1 Prediction
The last chapter addressed measuring the relationship (correlation)
between variables. Once a relationship between variables has been
established, wouldn't it be nice to find a way to predict values of one
variable for given a value of another? After all, if you found a
statistical relationship between two variables and relationships
indicate systematic changes in one variable as a function of another
variable, you should be able to make predictions.
For example, consider the data to the right. You can see that for every increase in X by 1,
X Y
there is a corresponding increase in Y by 2, suggesting a positive linear relationship. Say
2 10
that we have a value of X = 8, which does not appear in the table. What value of Y would
3 12
you predict X = 8 to be associated with? If the relationship between X and Y stays the same
4 14
and every increase in X by 1 is associated with a corresponding increase in Y by 2, you
5 16
should predict that when X = 8, Y = 22. How about X = 9? Again, if the relationship does not
6 18
change, then when X = 9, Y = 24. How about when we know that Y = 8, what should X be
7 20
equal to? Because we know the relationship between X and Y to be positive and linear and
for every change of 2 in Y there is a change in X by 1; thus, when Y = 8, X = 1.
This chapter deals with linear regression, which is used to model the linear relationship between two
variables, so values of one variable can be predicted from another variable. Specifically, given that we
have two variables in a correlation, linear regression attempts to fit a linear model to that relationship. If
you look to the cartoon at the top of the page, you can see a scatterplot depicting a positive linear
relationship. The relationship is not perfectly linear and the individual in the cartoon is trying to plot a best
fit line through the data points, which shows the approximate trend in the data. Linear regression does is
attempts to fit a linear model, that is, a mathematical way of explaining a relationship between variables,
to data. Thus, linear regression is used to determine the best fitting line to a set of data in a scatterplot.
Univariate regression uses one variable to model changes in another variable, and multivariate
regression (multiple regression) uses several variables to model changes in another variable. In
univariate regression there are two variables: a regressed variable (predicted variable), which is the
variable whose values we are predicting, and a regressor variable (predictor variable), which is the
being used to model changes in the other. In multivariate regression there is one regressed variable and
two or more regressor variables.
The first thing to note is that because relationships are never perfect, prediction will never be perfect, that
is. Rather, because behavior is variable and dependent variables used to measure behavior capture that
variability, the relationship will be affected by that variability. Nonetheless, the regression procedures are
designed in such a way that the error in prediction is minimized through something called the least-
squares criterion; a topic we will return to later.
16.2 Regression Equation

In regression, you use the values of one variable to predict values of a second variable. In this procedure
you are calculating a regression equation of a linear model, which takes this generic form:
Y  b0  b1X
-208-
Back to TOC
You should be familiar with this equation, because it is the equation for a straight line, which takes the
mathematical form:
y = mx + b
In a linear equation, values of variable y are being modeled (predicted) from the function mx + b. In the
function, the m is a constant that represents the change in y for every one-unit change in variable x, the b
is a constant that is expected when x = 0, and x is a variable that is plugged into the equation to give the
predicted value of y. Thus, the linear equation is modeling changes in variable y based on changes in
variable x. For example, say we have the following linear equation, which is based on the data in the table
below:
y = 2x + 10
The value 2, which was the constant m in the equation above, is the x 2x + 10 y
predicted change in y for every one-unit change in x. That is, if x increases 2 2(2) + 10 14
by 1, y is predicted to change by +2. The value 10, which was the constant b 3 2(3) + 10 16
in the equation above, is the value of y when x = 0. Given this function, we 4 2(4) + 10 18
could plug in values of the variable x to calculate values of y. In the table to
6 2(6) + 10 22
the right, for each given x, I have the corresponding y values calculated from
the function above. 8 2(8) + 10 26
9 2(9) + 10 28
You can see that for every one-unit increase in x (from 2-3 or from 8-9), 11 2(11) + 10 32
there is a corresponding two-unit increase in y (from 14-16 and 26-28, 12 2(12) + 10 34
respectively). Thus, the value of m in the linear equation (b 1 in the regression 13 2(13) + 10 36
equation) can be thought of as the slope of the linear equation. The slope of
14 2(14) + 10 38
the linear equation can be better understood by plotting all of the x-y data
points in a scatterplot and then drawing a line through them:
40
36
32
28
24
20
y
16
12
8
4
0
0 2 4 6 8 10 12 14 16
x
You can see that there is a straight line linking all of the x-y data points, and that the slope (m = 2) shows
a two-unit increase in y for every one-unit increase in x. Indeed, if you were to take any value of y from
this graph (e.g., Y = 28) and call this y1 and call its corresponding x value x 1 (X = 9), and take any other y-
value (Y = 26) and call this y2 and its corresponding x value x2 (X = 8), and then substitute those values
into the following equation you would find the outcome to be equal to the slope (m) of the linear equation:
y1  y 2 28  26 2
m   2
x1  x 2 98 1
Also, if you were to extend this straight line so that it passed through the ordinate you would find that it
intersects at y = 10, which is the value of the constant b in the linear equation (b0 in the regression
equation). This value can be called the y-intercept. The y-intercept is the value of y when x = 0.
-209-
Back to TOC
Why did I take the time to give a brief background of linear equation when this chapter is about linear
regression? Because linear regression does the exact same thing as the linear equation, it models a
linear relationship between two variables, that is, it estimates the best-fit linear relationship between two
quantitative variables.
In the regression equation, Y' (‘y-prime’) is the predicted value of Y for a value of X; b0 is the expected
value of Y when X is equal to 0; b1 is the predicted change in Y for every one-unit change in X (slope);
and X is the regressor variable that is used to model the changes in Y. The following sections show how
to calculate the slope (b1) and the intercept (b0) coefficients of the linear regression equation.
16.3 Calculating the Slope Coefficient (b0)

The most important thing to keep in mind
when calculating the slope and intercept in Name Age (X) Books per Month (Y)
a regression equation is which variable you Rob 22 8
are regressing on the other, that is, which Sheri 20 4
variable is being used to predict values of J. 22 4
the other? The data from Chapter 15 are Sean 21 3
presented to the right, where the correlation John 35 6
was calculated between age (X) and the Riggs 32 7
number of books a person reads per month Blasko 38 4
(Y). The descriptive statistics for each Vincent 40 6
variable as are the summary statistics. George 20 3
Bela 40 5
Let’s say we want to fit a linear regression
equation to these data to model the number ∑X = 290 ∑Y = 50
of books a person reads per month based
on age. That is, we want to generate linear X  29 Y5
regression equation that can predict the SCP = 43 SSX = 647 SSY = 26
number of books a person reads per month
from a person’s age.
covXY = 4.778 ŝ 2X  71.889 ŝ 2Y  2.889
r = 0.331 ŝ X  8.479 ŝ Y  1.7
Luckily, most of the terms needed to calculate the slope and intercept coefficients were calculated in
Chapter 15, but this brings up an important point about linear regression: When calculating the slope and
intercept values, you must calculate the slope first, because you need its value to calculate the intercept.
Thus, start by calculating the slope regressing Y on X. The slope equation is:
 ŝ 
b1  r Y 
 ŝ X 
This is the estimated standard deviation in the dependent variable Y divided by the estimated standard
deviation of the regressor (X) variable, weighted by the correlation between the variables. Remember, the
slope coefficient is the change in Y for every one unit change in X; thus, whichever variable you are
predicting change in goes in the numerator and the regressor variable goes in the denominator. Here are
two other equations for the slope coefficient, each will provide you with roughly the same value; however,
I will be using the one from above:
SCP cov xy
b1  or b1 
SSX ŝ 2X
-210-
Back to TOC
 ŝ   1.7 
Solving for the slope, we have: b1  r Y   0.331   0.3310.2  0.066
 ŝ X   8.479 
Thus, Y (books read per month) is predicted to increase by 0.066 books for every one-unit increase in
age (X). That is for every increase in a person’s age by one year the number of books read per month is
predicted to increase by 0.066 books. For example, if a person is 20 years old and that person reads 4
books per month, a 21 year old is predicted to read 4.066 books per month.
16.4 Calculating the Intercept Coefficient (b0)

Once you have calculated the slope you can calculate the intercept (b0). The equation is actually a variant
of the regression equation that has been rewritten to solve for a:
b0  Y  b0 X
In this equation, you must determine the mean of Y, which is the mean of the regressed/predicted
variable, and the mean of X, which is the mean of the regressor/predictor variable. From the data in the
table above, X  29 and Y  5 . Plugging those values into the equation we get:
b 0  5  0.06629
b 0  5  1.914
b 0  3.086
This value (b0 = 3.086) is the intercept in the regression equation, or more formally, this is the predicted
value of Y when X = 0. Now that you have solved for the slope and intercept you can write out the
regression equation by inserting the values of b0 and b1 into the generic regression equation from before:
Y  b 0  b1X
Y  3.086  0.066X
You can use this equation to solve for predicted values of Y by plugging in any possible value of X and
solving for Y', which we will do in the following section.
16.5 Plotting a Regression Line

The regression equation is the model of a straight line through the data points for two variables. Hence,
we can plot the function (regression equation) in a scatterplot with the data. There are two ways you can
do this: (1) Select a small value of X to compute a predicted value of Y and select a large value of X to
predict a second value of Y. Plot both X-Y' data points in the scatterplot and join them by a straight line.
(2) Compute each Y' value corresponding to each known X-value, plot each X-Y' data point in the
scatterplot and draw a line through all of them.
In the table below, I calculated the predicted value (Y') for each X-value. Notice there are differences
between the actual y-values and the predicted y’-values. This is because the regression equation
is predicting values of Y. Because the relationship between X and Y is not perfectly linear (as indicated by
r = 0.331), prediction will not be perfect:
-211-
Back to TOC
Name Age (X) Books (Y) Y' = 3.086 + 0.066X Y’
Rob 22 8 3.086 + .066(22) 4.538
Sheri 20 4 3.086 + .066(20) 4.406
J. 22 4 3.086 + .066(22) 4.538
Sean 21 3 3.086 + .066(21) 4.472
John 35 6 3.086 + .066(35) 5.396
Riggs 32 7 3.086 + .066(32) 5.198
Blasko 38 4 3.086 + .066(38) 5.594
Vincent 40 6 3.086 + .066(40) 5.726
George 20 3 3.086 + .066(20) 4.406
Bela 40 5 3.086 + .066(40) 5.726
Plotting the X - Y' data points in a scatterplot 9

with the actual X-Y data points, we get the graph 8
to the right. In the scatterplot, the actual X-Y
Books / Month (Y)

7
data points are plotted in blue and the X-Y' data 6
points are plotted in red with a line passing 5
through them. This is the ‘best fit’ linear 4 Y'=3.086+0.066X
relationship between age (X) and number of
3
books read per month (Y). Again, notice that the
2
Y' values differ from the actual Y values. Again,
1
with an imperfect linear relationship, this is to be
expected. The differences between the Y values 0
and predicted Y' values is residual variance, 18 20 22 24 26 28 30 32 34 36 38 40 42 44
Age (X)
which we will turn to in the next section.
16.6 Residual Variance and Standard Error of the Estimate

Before getting into how residual variance is calculated let me explain a little about what residual variance
is and where it comes from. The total variability in the scores of the predicted variable (Y) is simply the
sum of squares for that variable (SSY). This is composed of two sources of variance: variability due to
random sampling error and variability due to regression. Specifically:
Total Variability = Regression Variability + Error Variability

If total variability in the predicted variability is the sum of squares for that variable (SS Y), then the total
error variability must be a sum of squares for error, which we call sum of squares residual (SSResid).
Also, the regression variability must be a sum of squares due to regression, which we call sum of
squares regression (SSReg). Thus, total variability is equal to sum of squares residual added to the sum
of squares regression:
SSY = SSReg + SSResid

Computationally, this is:
 2 
 Y  Y   Y  Y   Y  Y
2
2
Sum of squares-regression (SSReg) is the total variance in the predicted scores (Y’) that can be explained
in the linear relationship between X and Y. You can see this in the first term after the equal sign in the
equation above: The sum of squares regression is the total variability between the predicted values (Y’)
and the mean of the actual y-values (Ȳ). Hence, sum of squares residual is measuring how close the
predicted scores (Y’) are to the actual mean. The sum of squares residual (SSResid) is the total variance in
-212-
Back to TOC
Y that cannot be explained in the linear relationship between X and Y; thus, it is unexplained variance.
You can see this in the second term after the equal sign in the equation above: The sum of squares
residual is the variability between the predicted y-values (Y’) and the actual y-values (Y). Hence, sum of
squares residual is measuring how closely the predicted scores (Y’) are to the actual scores (Y), where
any deviation must be due to random, unexplainable error.
The components to the regression equation (b0 and b1) are calculated in such a way that residual
variance is minimized in what is called the least-squares criterion. What we want to know is how much
error there is in predicting Y from X; that is, how much unexplained variation there is in the predicted
values of Y.
To calculate sum of squares regression, use the following formula:
SSReg = r2SSY
To calculate sum of squares residual, use the following formula:
SSResid = (1 - r2)SSY
From data in the preceding sections, the sum of squares for the number of books a person read per
month was SSY = 26 and r = 0.331. Plugging these values into the equations above, we have:
SSReg = r2SSY SSResid = (1 - r2)SSY

SSReg = (0.3312)26 SSResid = (1 - 0.3312)26
SSReg = (0.11)26 SSResid = (1 - 0.11)26
SSReg = 2.86 SSResid = (0.890)26
SSResid = 23.14
These values are summed, or total variance values. What we want is the average residual variance. The
formula for average residual variance is found by taking the average of the sum of squares residual:
ŝ 2Y  
 Y  Y
2

 
1  r 2 SSY SSRe sid

n2 n2 n2
You can see that residual variance is the sum of squares residual divided by degrees of freedom, which is
n – 2 in this case, because there are two dependent variables. Note that each of these formulas are the
same way of symbolically writing ‘the average sum of squares residual’; that is, the numerator in each
case is just the sum of square residual. Solving for the average residual variance, we have:
23.14
ŝ 2Y   2.892
10  2
This value is the variation in Y that cannot be explained in the linear relationship. If we wanted to know
the average deviation between each actual Y value and the predicted value of Y we would take the
square-root of residual variance, which will give us the standard error of the estimate. That is, the
standard error of the estimate is simply the square root of the average residual variance that was just
calculated:
ŝ Y 
 Y  Y
2

1  r SS
2
Y 
SSRe sid
 ŝ 2Y
n2 n2 n2
-213-
Back to TOC
In each variant of this formula above, the term under the radical sign is just the formula for average
residual variance. Conceptually, the standard error of the estimate is the square root of residual variance.
The standard error of the estimate is functionally equivalent to a standard deviation. Remember, the
standard deviation is the average difference between a score and the mean of a distribution; thus, the
standard error of the estimate is the average deviation between the actual y values and the predicted
values:
ŝ Y  ŝ 2Y  2.892  1.7
Thus, for this set of data, the predicted values (Y') are estimated to deviate from the actual values (Y) by
1.7 units (books read per month).
16.7 Homoscedascity
In order for residual variance and the standard error of the estimate to accurately measure the error in
prediction, we must make an assumption of homoscedascity. Homoscedascity is seen when the actual Y-
values vary to the same degree at each level of X; that is, the variance in Y for every value of X is
equivalent. Heteroscedascity occurs when the variance in Y is not equal at every level of X.
Homoscedascity assumes that the values of Y are normally distributed at each value of X; hence, there is
equal variance in Y at each value of X. If the data is heteroscedastic, it means the estimate of the residual
variance and the standard error of the estimate does not accurately measure the variability between the
actual Y-value and the predicted values.
16.8 Making Inferences about Regression Equations and Coefficients

There are tests of statistical significance for regression coefficients (intercept and slope) and for the entire
regression equation. There are several things that can be significant in any regression equation: First, the
entire regression equation can be a significant model of a linear relationship. Second, the slope (b1) can
be a significant predictor for changes in Y. Finally, the intercept (b0) can be significant (which is usually
assumed to be zero under the null). Instead of burrowing into the various assumptions and population
distributions about the slope, intercept and regression equation, I’ll focus more on how significance is
determined for the regression equation and components that we calculated in earlier sections.
Recall that the regression equation that predicted number of books read per month (Y) from age (X) was:
Y  3.086  0.066X
We will evaluate the statistical significance of the entire regression equation, the slope, and the intercept.
We’ll start with the entire regression equation, because some of that information will be necessary for
evaluating the significance of the slope and intercept coefficients.
We can conclude the regression equation is a significant predictor of the number of books read per month
from age if a significant proportion of the variance in Y (books read per month) can be explained by
variation in X (age). Recall, that the total variation in the predicted variable (SSY), which we’ll call the sum
of squares total (SSTotal) is a combination of the total variance that can be explained in the relationship
between Y and X (SSReg) and the total error, or residual variance, that cannot be explained (SSResid).
Recall that the sum of squares regression (SSReg) and is calculated by summing the squared deviations
between the predicted values (Y’) and the mean of Y:

SSRe g   Y  Y 2
-214-
Back to TOC
This can be more easily found from the following computational formula:
SSReg = r2SSY
The sum of squares residual (SSResid) and is calculated by summing the squared deviations between the
actual values (Y) and the predicted values (Y’):
SS Re sid   Y  Y
2
This can be more easily found from the following computational formula:
SSResid = (1 - r2)SSY
Of course the total variance is just the sum of squares total (SSTotal) and is calculated by summing the
squared deviations between the actual values (Y) and the mean of the Y values. Thus:
SSTotal   Y  Y 
2
The total variance in the predicted variable is equal to (can be partitioned into) total residual variation and
the total variation due to regression:
 2 
 Y  Y   Y  Y   Y  Y
2
2
or
SSTotal = SSReg + SSResid
These three sources of variance were obtained in earlier sections and chapters:
SSTotal = SSY = 26 SSReg = 2.86 SSResid = 23.14
To reiterate: SSTotal is the total variation in Y; SSReg is the variation in Y that can be accounted for by
variation in X; and SSResid is the error variation that cannot be accounted for in the relationship between Y
and X.
Now that we have the three necessary sources of variance, we can analyze whether the variance that
can be accounted for in the relationship between number of books red per month and age (SSReg) is
significantly greater than the variance that cannot be accounted for in this relationship (SSResid). Thus, we
are going to analyze variance in this relationship. Recall that all inferential statistics boil down to a ratio of
some amount effect (or explained variance) to some amount of random error (unexplained variance).
Thus, we can determine the significance of the regression equation by forming a ratio of the variance due
to regression over the residual variance:
 2Re gression
test statistic 
 2Re sidual
-215-
Back to TOC
As you will learn about more in Chapters 19-21, this ratio, which is a ratio of variance over variance, forms
the basis of the analysis of variance test, which is called the F-Test. However, we cannot place the sums
of squares regression and residual into this test statistic, because we need to take into account the
degrees of freedom associated with each source of variance. That is, we need to divide first each sum of
squares above by their associated degrees of freedom to get the ‘mean variance’ of each source of
variance. This ‘mean variance’ is called a mean square (MS).
The degrees of freedom associated with the regression variance (dfReg) are always equal to k – 1, where
k is the number of coefficients in the regression equation. In our example, there are k = 2 coefficients; the
slope (b1) and the intercept (b0). Hence, in this example, dfReg = 2 – 1 = 1. This degrees of freedom is also
known as the degrees of freedom due to the effect of a variable (df Effect) and also the degrees of freedom
between groups (dfB). The degrees of freedom associated with the residual variance (df Resid) are equal to
n – 2; where n is the number of subjects in the data set. In this example, there are n = 10 subjects, so the
degrees of freedom associated with residual variance are dfResid = 10 – 2 = 8. This degrees of freedom is
also known as the degrees of freedom due to error (dfErrort) and the degrees of freedom within groups
(dfW).The total degrees of freedom (dfTotal) are equal to n – 1; thus, 10 – 1 = 9. Note that dfTotal = dfReg +
dfResid (9 = 1 + 8). The degrees of freedom regression and the degrees of freedom residual are used to
determine the statistical significance of the regression equation.
Now we can calculate the mean squares. To simplify, a mean square is just a sum of squares (total
variability) divided by its associated degrees of freedom; thus:
SS
MS 
df
The mean square for regression is the sum of squares regression divided by the degrees of freedom for
regression and the mean square for residual variance is the sum of squares residual divided by the
degrees of freedom residual. We only need to calculate the mean square for the regression variance and
residual variance, which I have done below:
SSRe g 2.86
MSRe g    2.86
df Re g 1
SSRe sid 23.14
MSRe sid    2.892
df Re sid 8
2
Note that the mean square residual (MSResid) is the average residual variance ( ŝ Y ) in predicting Y from X,
which we calculated in Section 16.6. Hence, the square root of the mean square residual will give you
the standard error of the estimate ( ŝ Y ), which was also calculated in section 16.6. This is important,
because the standard error of the estimate will be used later.
Recall, the test statistic for determining significance of the regression equation is the regression
(explained) variance over the residual (error) variance. This test statistic is called the F-Ratio:
MSRe g
F
MSRe sid
Thus, from our data we have:
2.86
F  0.989
2.892
-216-
Back to TOC
This is the test statistic that will be used to determine the statistical significance of the regression
equation. To determine significance we need to look-up a p-value in Table 3 in Appendix A, which are the
probabilities associated with areas under the F-Distribution (more on this in the following chapters). There
are actually six tables associated Table 3, Table 3-A through Table 3-F, with each table being associated
with a different number of degrees of freedom due to regression (dfReg). In this example, we have df Reg =
1, so we’ll use Table 3-A, which is associated with dfB = 1. A portion of Table 3-A is reproduced, below.
Table 3-A: dfB = 1
Degrees of Freedom Error (dfW)
F 2 3 4 5 6 7 8 9 10 12 14 16 18 20 22 24 26 28 30 40 45 50 60 80 100 250 1000
1.00 0.4226 0.3910 0.3739 0.3632 0.3559 0.3506 0.3466 0.3434 0.3409 0.3370 0.3343 0.3322 0.3306 0.3293 0.3282 0.3273 0.3265 0.3259 0.3253 0.3233 0.3227 0.3221 0.3213 0.3203 0.3197 0.3183 0.3176
1.10 0.4043 0.3713 0.3535 0.3423 0.3347 0.3291 0.3249 0.3216 0.3190 0.3149 0.3120 0.3099 0.3081 0.3068 0.3057 0.3047 0.3039 0.3032 0.3026 0.3006 0.2999 0.2993 0.2985 0.2974 0.2968 0.2953 0.2945
1.20 0.3876 0.3534 0.3349 0.3233 0.3153 0.3096 0.3052 0.3018 0.2990 0.2948 0.2918 0.2895 0.2878 0.2863 0.2852 0.2842 0.2834 0.2827 0.2820 0.2799 0.2792 0.2786 0.2777 0.2766 0.2760 0.2744 0.2736
1.30 0.3724 0.3370 0.3178 0.3059 0.2977 0.2917 0.2872 0.2836 0.2808 0.2765 0.2733 0.2710 0.2692 0.2677 0.2665 0.2655 0.2646 0.2639 0.2632 0.2610 0.2602 0.2596 0.2587 0.2576 0.2569 0.2553 0.2545
1.40 0.3583 0.3219 0.3022 0.2899 0.2815 0.2753 0.2707 0.2670 0.2641 0.2596 0.2564 0.2540 0.2521 0.2506 0.2494 0.2483 0.2474 0.2467 0.2460 0.2437 0.2429 0.2423 0.2414 0.2402 0.2395 0.2378 0.2370
1.50 0.3453 0.3081 0.2879 0.2752 0.2666 0.2603 0.2555 0.2518 0.2487 0.2442 0.2409 0.2384 0.2365 0.2349 0.2336 0.2326 0.2317 0.2309 0.2302 0.2278 0.2270 0.2264 0.2255 0.2243 0.2235 0.2218 0.2210
1.60 0.3333 0.2952 0.2746 0.2617 0.2528 0.2464 0.2415 0.2377 0.2346 0.2299 0.2266 0.2240 0.2220 0.2204 0.2191 0.2180 0.2171 0.2163 0.2156 0.2132 0.2124 0.2118 0.2108 0.2096 0.2088 0.2071 0.2062
1.70 0.3222 0.2833 0.2623 0.2491 0.2401 0.2335 0.2286 0.2246 0.2215 0.2167 0.2133 0.2107 0.2087 0.2071 0.2058 0.2047 0.2037 0.2029 0.2022 0.1997 0.1989 0.1983 0.1973 0.1960 0.1953 0.1935 0.1926
1.80 0.3118 0.2722 0.2508 0.2374 0.2283 0.2216 0.2165 0.2126 0.2094 0.2046 0.2011 0.1984 0.1964 0.1947 0.1934 0.1923 0.1913 0.1905 0.1898 0.1873 0.1864 0.1858 0.1848 0.1835 0.1828 0.1809 0.1800
1.90 0.3020 0.2619 0.2402 0.2266 0.2173 0.2105 0.2054 0.2014 0.1981 0.1932 0.1897 0.1870 0.1850 0.1833 0.1819 0.1808 0.1798 0.1790 0.1783 0.1757 0.1749 0.1742 0.1732 0.1719 0.1712 0.1693 0.1684
2.00 0.2929 0.2522 0.2302 0.2164 0.2070 0.2002 0.1950 0.1909 0.1877 0.1827 0.1792 0.1765 0.1744 0.1727 0.1713 0.1701 0.1692 0.1683 0.1676 0.1650 0.1642 0.1635 0.1625 0.1612 0.1604 0.1585 0.1576
Assume we select an alpha level of α = .05. First, note that there is no such thing as a directional or non-
directional hypothesis in the F-Distribution; hence, once we find the p-value associated with the test
statistic (0.989) we always compare it directly to the selected alpha level. To determine the statistical
significance of the regression equation lookup the value of the F-Ratio (0.989, which rounds to 1.00 in
Table 3-A) in the leftmost column (highlighted in yellow) and lookup the value for the degrees of freedom
residual (dfResid = 8) in the column headings (highlighted in yellow). The p-value associated with this F-
Ratio is 0.3466. This value is greater than the selected alpha level (.05), so we conclude the regression
equation is a not significant predictor of books red per month. That is, this linear regression equation
based on age does not account for a significant amount of variance in the number of books read per
month based on age.
Although the regression equation was been found to be non-significant, you could evaluate whether the
slope (b1) and intercept (b0) are statistically significant. Note that the interpretations of the significance of
these coefficients can be misleading if the regression equation is not significant, as was the case here.
Nonetheless, below, I show how to determine statistical significance of these regression coefficients.
For both the slope and the intercept, we will conduct a t-Test. Each t-Test is the difference between the
calculated regression coefficient and its expected value under the null hypothesis, over a standard error.
The expected value of each regression component under the null hypothesis, denoted β 1 for the slope
and β0 for the intercept, are usually expected to be zero, but they do not have to be. The standard error
for the slope, SE(b1), is:
SEb1  
ŝ Y
SSX
The standard error for the slope, SE(b0), is:
SEb 0   ŝ Y 
1 X

n SSX
In both cases, ŝ Y  is the standard error of the estimate, SS X is the sum of squares for the regressor
variable (age), and X is the mean of the regressor variable (age). Remember, the standard error of the
estimate can be found by taking the square root of the mean square residual (MS Resid):
-217-
ŝ Y  MSRe sid  2.892  1.7
Both t-Tests take the same general form. For the slope:
b1  1
t
SEb1 
And for the intercept:
b 0  0
t
SEb 0 
From the data in the example, the standard error of the slope and intercept are (SSX comes from earlier):
SEb 0   1.7
1 29

SEb1  
1.7 10 647
647 SEb 0   1.7 0.1  0.117
SEb1  
1.7 SEb 0   1.7 0.217
SEb 0   1.70.466
25.436
SEb1   0.067
SEb 0   0.792
Plugging in the coefficients and the standard errors into the t-Test formulas, we get (the slope and
intercept coefficient values come from the regression equation earlier):
b1  1 b 0  0
t t
SEb1  SEb 0 
0.066  0 3.086  0
t t
0.067 0.792
t  0.985 t  3.896
To evaluate the statistical significance of each regression coefficient, look up the t-values in Table 2 using
df = 8. For the slope (t = 0.985, using t = 1.00 in Table 2), the p-value is p = .3466, indicating the slope is
not statistically significant. This means is that there is not a significant increase in the number of books
red per month based on an increase in a person’s age. For the intercept (t = 3.896, using t = 3.89 in Table
2) the p-value is p = .0202, which is less than our chosen alpha level (.05), indicating the intercept if
statistically significant. This means the intercept is significantly different from zero and that this sample
reads a significant number of books per month, compared to not reading any books per month.
1. What is the linear regression line (model)?
-218-
Back to TOC
2. Describe what information is conveyed by the slope coefficient.
3. Describe what information is conveyed by the intercept coefficient.
4. What is the generic form of the linear regression model?
5. Assume you have a perfect linear relationship between variables X and Y and the relationship has a
slope of 2.5. By how many units does Y change if X changes by 1 Unit? If X changes by 3 units? If X
changes by 8 units?
6. What information is provided by the standard error of the estimate? What measure of variability is the
standard error of the estimate similar to?
7. Use the following to answer the questions below. A researcher collected data on variables X any Y
from each of ten subjects. The data for each of these ten subjects are below, as are the summary
statistics and Pearson correlation between the two variables:
i X Y X5 SSX = 90 ŝ 2X  10 ŝ X  3.162

A 10 0
B 9 2 Y4 SSY = 82 ŝ 2Y  9.111 ŝ Y  3.018
C 7 1 SCP = -79 covxy = -8.778 r = -0.920
D 6 2
E 6 2
F 4 4
G 3 6
H 3 8
I 2 7
J 0 8
a. Calculate the slope coefficient of a linear equation regressing Y on X (predicts Y from X).
b. Calculate the intercept coefficient for a linear equation regressing Y on X.
c. State the regression equation based on the coefficients just calculated.
d. Calculate the predicted values of Y (Y’) for each of the following individuals.
Individual B
Individual H
Individual E
e. Calculate the sum of squares residual and the sum of squares regression (using the
computational methods).
f. Calculate the standard error of the estimate. What does this value tell you about the predicted
values of Y?
8. Use the following to answer the questions below: A researcher collected data on variables X any Y
from each of ten subjects. The data for each of these ten subjects are below, as are the summary
statistics and Pearson correlation between the two variables:
-219-
Individual X Y X  5 SSX = 50 ŝ 2X  5.556 ŝ X  2.357
A 3 7
B 8 9 Y  6 SSY = 60 ŝ Y  6.667
2
ŝ Y  2.582
C 3 3 SCP = 10 covxy = 10 r = 0.182
D 2 8
E 6 8
F 6 9
G 8 6
H 5 4
I 7 2
J 2 4
a. Calculate the slope coefficient of a linear equation regressing Y on X (predicts Y from X).
b. Calculate the intercept coefficient for a linear equation regressing Y on X.
c. State the regression equation based on the coefficients just calculated.
d. Calculate the predicted values of Y (Y’) for each of the following individuals.
Individual A
Individual G
Individual J
e. Calculate the sum of squares residual and the sum of squares regression (using the
computational methods).
f. Calculate the standard error of the estimate. What does this value tell you about the predicted
values of Y?
9. Based on the standard errors of the estimated calculated in Exercises 7 and 8, which regression
equation is a better predictor of variable Y?
10. Use the following to answer the questions below: A professor has noticed that students who lack
basic math skills do not perform well in his statistics course. The professor creates a math pretest that
has a range of zero to ten. The test is given at the first class meeting to determine whether each student
has the math skills necessary for the statistics course (n = 8). To evaluate the validity of his pretest, at the
end of the semester the professor regresses student course grade points (Y) on the pretest scores (X).
The regression equation and statistics appear in the table below.
Pretest (X) Grade Points(Y) Y’ = 0.884 + 0.326X
X5 Y  2.675 r = 0.744
SSX = 82 SSY = 17.715 cov = 26.7
a. Calculate the sum of squares regression and the sum of squares residual.
b. Determine the degrees of freedom regression, degrees of freedom residual, and degrees of
freedom total.
c. Calculate the mean square residual and the mean square regression
d. Calculate the F-Ratio.
e. Determine the p-value for the F-Ratio.
f. Using α = .05, is the regression equation statistically significant? Why?
g. Calculate the standard error of the estimate from the average residual variance (MSResidual).
h. Calculate the standard error of the slope SE(b1) and standard error of the intercept SE(b0).
i. Calculate the t-Value for the slope (b1). What is the p-value?
j. Calculate the t-Value for the intercept (b0). What is the p-value?
11. Use the following to answer the questions below: Using the data from Chapter 15 Exercise 12, Dr.
Vader wants to predict a person's willingness to conform (Y) from their self-reported fear of the darkside
of the force (X). Recall, that each of n = 10 individuals rated how much they feared the darkside of the
force (X) on a scale of 1 – 10, where higher scores indicate more fear of the darkside, and was also
measured on their willingness to conform to the Empire's demands (Y) on a scale of 1 - 10, where higher
-220-
Back to TOC
scores indicate more willingness to conform. The regression equation and statistics appear in the table
below.
Fear of Darkside (X) Conformity (Y) Y’ = -2.1 + 1.22X

X4 Y5 r = 0.941
SSX = 50 SSY = 84 cov = 6.778
a. Calculate the sum of squares regression and the sum of squares residual.
b. Determine the degrees of freedom regression, degrees of freedom residual, and degrees of
freedom total.
c. Calculate the mean square residual and the mean square regression
e. Determine the p-value for the F-Ratio.
f. Using α = .05, is the regression equation statistically significant? Why?
g. Calculate the standard error of the estimate from the average residual variance (MSResidual).
h. Calculate the standard error of the slope SE(b1) and standard error of the intercept SE(b0).
i. Calculate the t-Value for the slope (b1). What is the p-value?
Calculate the t-Value for the intercept (b0). What is the p-value?
Chapter 17: Correlated Samples t-Test

17.1 Recap of t-Tests
-221-
Back to TOC
There are several different t-Tests that you can use, and the type of t-Test you should use depends on
your research design:
 A one-sample t-Test is used when you have a single sample and are comparing its mean to some
value that is assumed to represent a population mean.
 An independent-groups t-Test is used when you have two samples, each representing a different
and independent group of subjects. The groups differ along one independent variable with two
levels, and you compare the mean of those two groups (between-subjects design).
A correlated-samples, or dependent-samples t-Test, which is being introduced here, is used when you
have one sample of subjects who are tested several times but under different conditions. That is, each
subject is measured on the same dependent variable, but under different levels of an independent
variable. You compare performance of the subjects between the different levels of this independent
variable (with-subjects design).
17.2 What is the Correlated (Dependent) Samples t-Test?

A correlated-groups t-Test is preferred over the independent groups t-Test. The preference is based on
the fact that a correlated-group t-Test, as well as all inferential tests designed to analyze within-subjects
designs, is more statistically powerful then the independent-groups t-Test. This means that you are more
likely to produce a statistically significant result and reject the null hypothesis with a correlated-groups t-
Test than an independent-groups t-Test.
The reason for this increased power is not based on the mean difference between conditions; rather, it is
based on the error term in the t-Test (denominator). If you were to take any set of data that includes two
levels of an independent variable, the mean difference between those two levels will be the same
regardless of whether you use an independent-groups t-Test or a correlated-groups t-Test. However, the
error term will always be smaller for a correlated-groups t-Test than for an independent-groups t-Test.
This is what makes the correlated-groups t-Test (and all within-subject analyses) more powerful. Because
it has a smaller error term, the obtained t-Value will always be larger. The only exception is when there is
no correlation between the levels of the independent variable. In this limited case, the correlated samples
t-test is as powerful as the independent groups t-test.
The correlated samples t-Test is used when:
 The dependent variable is quantitative in nature

 The independent variable is qualitative in nature, that is, the levels represent different categories.
 The independent variable has two and only two levels.
 The independent variable is manipulated within-subjects.
17.3 Standard Error of the Difference Between Correlated Samples

The error term in the correlated-samples t-Test is smaller than the error term in the independent-groups t-
Test, because after the variance from both levels of an independent variable has been pooled, the
covariance between levels is factored out. That is, a correlated-groups t-Test analyzes data between two
conditions that are assumed to be correlated, that is, responses are assumed to co-vary between
conditions. This covariance is present in each condition. If the variance of each condition were combined,
as in the independent-groups t-Test, the error term would be inflated, because of this covariance. To
compensate, the covariance is factored out, which reduces the error term.
Here’s what I mean: Here's the expression (for population data) for the error term for an independent-
groups t-test:
-222-
Back to TOC
12  22
X X  
1 2 N1 N 2
The variance from each condition (subscripts 1 and 2) include the variance in that condition and any co-
variability with the other condition. In the independent group t-test, because the conditions are assumed
to be independent, adding them together is not an issue. But, if the conditions are assumed to be
correlated, this co-variability is being added to itself as the variances are summed. Hence, the co-
variability has to be factored out, for the error term for a correlated samples t-test, as is shown below:
12  22
X X    2  cov12 
1 2 N1 N 2
Which is the same as:
12  22    
X X    2  12  1  2 
1 2 N1 N 2  N1 N 2 
Thus, the error tern for the correlated samples t-test added the variance for each condition, and then
subtracts out the co-variability between those conditions, which reduced the error term,. Notice, that the
first ‘half’ of the error term for the correlated samples t-test is the same as the entire error term for the
independent groups t-test; the only difference it the factoring out of the covariance.
Of course, in practice, we rarely ever know the population parameters. Hence, we must estimate the
population variances, correlations, and standard deviations from our sample data. Here’s the estimated
standard error of the difference between correlated sample means, or what I’ll occasionally call
“Pete”, which is the error term for the correlated-groups t-Test:
ŝ12 ŝ 22  ŝ  ŝ 
ŝ X  X    2r  1  2 
1 2 n1 n 2  n  n 
 1  2 
The first term under the radical-sign is essentially the estimated pooled variance for the two conditions
being compared. The term after the minus-sign is the co-variance between the two conditions that are
being compared. Subtracting this second term reduces the size of the overall error term. Note that r is the
Pearson correlation between the two conditions.
17.4 Hypotheses and Steps in the Correlated Groups t-Test

The correlated samples-t-Test follows a similar set of procedures as the independent groups t-Test:
 State your hypotheses, determine critical value.

 Find the mean and estimated variance for each condition X1  X 2
 Pool the variance and calculate a standard error term. t
 Perform the t-Test, and the the expression for the correlated-groups t-test is: ŝ X  X
1 2
Both the correlated samples t-tests and the independent groups t-test are a ratio of the mean difference
between conditions over an error term. Please note that the error terms are symbolically written the same
-223-
Back to TOC
for both t-Tests; however, the error terms are calculated differently. Also, the hypotheses in the correlated
samples t-Test are stated the same way as with the independent-groups t-Test.
For a non-directional alternate hypothesis, either of the following can be used:
H0: μ1 - μ2 = 0 or H0: μ1 = μ2
H1: μ1 - μ2 ≠ 0 or H1: μ1 ≠ μ2
For a directional alternate hypothesis, either of the following can be used:
H0: μ1 - μ2 = 0 or H0: μ1 = μ2
H1: μ1 - μ2 > 0 or H1: μ1 > μ2
17.5 Numerical Example
Say that a researcher is studying the effect of environmental noise on a person’s ability to solve statistics
problems. The researcher randomly samples n = 10 students and exposes them to a noisy condition for
10 minutes and then to a quiet condition for 10 minutes. While each student is in each condition, they are
given a set of statistics problems to solve. The dependent variable is the number of mistakes made on the
statistics problems during the ten minutes. Here, the researcher is testing a non-directional hypothesis,
because she wants to know if there is any effect of noise on performance (errors); thus:
H0: μNoise = μQuiet

H1: μNoise ≠ μQuiet
This is a non-directional hypothesis; assume we select an alpha level of α = .01 so assess the statistical
significance of the outcome. Here are the hypothetical data with all relevant calculations for the means,
sums of squares, and sum of cross products:
Noisy Condition Quiet Condition

Student XN X N  X N  X N
 XN 
2
XQ XQ  XQ  X Q  XQ 
2
X N

 XN XQ  XQ 
A 9 1 1 6 0 0 0
B 9 1 1 7 1 1 1
C 6 -2 4 7 1 1 -2
D 7 -1 1 5 -1 1 1
E 6 -2 4 4 -2 4 4
F 7 -1 1 4 -2 4 2
G 9 1 1 6 0 0 0
H 11 3 9 9 3 9 9
I 7 -1 1 5 -1 1 1
J 9 1 1 7 1 1 1
XN  8 SSN = 24 XN  6 SSQ = 22 SCP = 17
To calculate the estimated standard error of the difference between correlated sample means (“Paul”),
you must first estimate the population variance and standard deviation for each condition, and then
calculate the correlation between the conditions. The estimated variances are:
SSN 24
ŝ2N    2.667 and ŝN  ŝ2N  2.667  1.633
n  1 10  1
The estimated standard deviations are:
-224-
Back to TOC
SSQ 22
2
ŝQ    2.444 and ŝQ 
2
ŝQ  2.444  1.563
n  1 10  1
Recall, the Pearson correlation (r) is calculated by dividing covariance (covxy) between two conditions by
the product of the estimated standard deviations for each condition. We already have the estimated
standard deviations for our two conditions above; and the the covariance is the sum of the cross products
(SCP) divided by n - 1:
SCP 17
cov xy    1.889
n  1 10  1
The Pearson correlation is:
cov xy 1.889 1.889

r    0.740
ŝN ŝQ 1.6331.563 2.552
Now that we have calculated the estimated standard deviations, estimated variances, and the Pearson
correlations, we can calculate the estimated standard error of the difference between correlated sample
means:
ŝ12 ŝ 22  ŝ  ŝ 
ŝ X  X    2r  1  2 
1 2 n1 n 2  n  n 
 1  2 
 1.633  1.563 
 20.74
2.667 2.444
ŝ X  X    
1 2  
10 10  10  10 
 1.633  1.563 
ŝ X  X  0.267  0.244  20.74  
1 2  3.162  3.162 
ŝ X  X  0.267  0.244  20.740.5160.494
1 2
ŝ X  X  0.511  0.377
1 2
ŝ X  X  0.134
1 2
ŝ X  X  0.366
1 2
This value ( ̂ ̅ ̅ ) is the estimated average difference between sample means

X1  X 2
of size n = 10, drawn from the noisy condition and the quiet condition. This is the value t
that is plugged into the denominator of the correlated samples t-test; while the means of ŝ X  X
the Noise and Quiet conditions are plugged into the numerator. Because we have a
1 2
non-directional alternate hypothesis, it does not matter which mean is first in the 86
numerator. t
0.366
With n = 10 subjects there are 10 – 1 = 9 degrees of freedom. Unlike the independent- 2
groups t-Test you do not count each condition (noise and quiet) as a different sample; t
rather, in the correlated-groups t-Test, n is the total number of subjects. To assess the
0.366
statistical significance of this outcome, we use the t-Tables in Appendix A (Table 2), t  5.464
using t = 4.00 in place of the test statistic (5.464) and df = 9. The p-value is p = .0031,
which is less than the select alpha level (.01); hence, there is a statistically significant difference between
-225-
Back to TOC
the two means and we can reject the null hypothesis and accept the alternative hypothesis. In layman’s
terms, students made significantly more errors in the Noisy condition than in the Quiet condition; hence,
Noise seems to distract students and lead to errors in solving statistics problems.
17.6 Effect Size Measures

Effect size measures can also be calculated for the correlated samples t-test; and these mean the same
as those measures used for the independent groups t-Test. One measure of effect size measure called
2
eta-squared (η ), which is the proportion of variance in the dependent variable that is attributable to the
effect of the independent variable:
t2
2 
t 2  df
Using the total degrees of freedom and obtained t-value from earlier, eta-squared is:
5.4642 29.855 29.855

2     0.768
5.4642  9 29.855  9 38.855
Thus, about 76.8% of the variance in the dependent variable (Errors) can be accounted for by the effect
of the independent variable (Noise versus Quiet).
Another popular index of effect size is Cohen’s d. Cohen’s d is a measure of the distance between two
sample means in standard deviations, which for the correlated samples t-test is best estimated from a
computational formula:
d  t 21  r  / n 
Using the sample size (n), the obtained t-value (5.464), and correlation (.74) from earlier, Cohen’s d is
1.246. Thus, two sample means (8 and 6) are separated by 1.246 standard deviations.
d  5.464 21  .74 / 10

d  5.464 2.26  / 10
d  5.464 .52 / 10
d  5.464 .052
d  5.464.228
d  1.246

As in Chapters 15 and 16 for the independent groups t-test and Pearson correlation, G*Power can be
used to determine for both post hoc and a priori power analyses for the correlated samples t-Test
-226-
Back to TOC
To perform a post hoc power analysis,
open G*Power 3 and, under “Test
family,” select “t tests.” Under
“Statistical test” choose “Means:
Difference between two means
(matched pairs)”, and under “Type of
Power Analysis” choose “Post hoc:
Compute achieved power.” Next,
you'll need to enter the α-level,
sample size, and effect size (d) from
the t-test above. The alpha level was
α = .05, we had a non-directional
hypothesis, n = 10, and d = 2.443.
After clicking calculate, G*Power tells
you that the statistical power is
.999993, which is extremely high,
which is good. Thus, there is a high
probability that we are correctly
rejecting the null hypothesis if it is
indeed false
We can also use G*Power in a priori

power analyses for the correlated
samples t-test to determine the
number of subjects needed to end up
with a desired level of power. For
example, assume am planning to
collect data from a sample of subjects
in a within subjects design. I know
from prior studies that the effect size
for such a manipulation is generally
“moderate” and near d = .60. For my
inferential test I am planning to use α
= .05 and with a directional alternate
hypothesis, and I want Power = .90.
To determine the appropriate sample

size, open G*Power 3 and under “Test
family” be sure “t tests” is chosen,
under “Statistical test” choose “
Means: Difference between two
means (matched pairs)”; and under
“Type of Power Analysis” choose “A
priori: compute required sample size.”
Next, you'll need to enter the α-level,
expected effect size, desired power,
and choose “one” for Tail(s). For this
example, assume I want Power = .90.

provided with the total sample size
needed to achieve this amount of
Power, which is n = 26 in this
example.
-227-
Back to TOC
1. When is the correlated samples t-test normally used?
2. State the critical value of t for a correlated samples t-test under each of the following conditions (α =
.05):
a. H0: μ1 = μ2, H1: μ1 ≠ μ2, n = 11
b. H0: μ1 = μ2, H1: μ1 > μ2, n = 11
c. H0: μ1 = μ2, H1: μ1 ≠ μ2, n = 16
d. H0: μ1 = μ2, H1: μ1 < μ2, n = 16
3. Find the p-value for a correlated samples t-test under each of the following conditions:
a. H0: μ1 = μ2, H1: μ1 ≠ μ2, n = 10, t = 2.50
b. H0: μ1 = μ2, H1: μ1 > μ2, n = 10, t = 2.60
c. H0: μ1 = μ2, H1: μ1 ≠ μ2, n = 25, t =2.20
d. H0: μ1 = μ2, H1: μ1 < μ2, n = 25, t =2.80
4. Why is the correlated samples t-test more powerful than a corresponding independent groups t-test?
5. When will the correlated samples t-test not be more powerful than a corresponding independent groups
t-test? Why?
6. Given the following information, compute the estimated standard error of the difference between
correlated sample means: n = 15, ŝ12  5.575 , ŝ 22  4.235 , r = 0.797.
7. Given the following information compute the estimated standard error of the difference between
correlated sample means: n = 50, ŝ12  12.125 , ŝ 22  25 , r = 0.384.
8. Use the following to answer the questions below. The following scores resulted on a memory test that
was administered to five subjects in two different conditions (A and B):
Subject A B
1 15 9
2 4 4
3 11 8
4 10 4
5 20 15
a. Calculate the means for each condition, sum of squares for each condition, and the sum of the
cross products between conditions
b. Calculate the estimated variances and estimated standard deviations for each condition, and
calculate the covariance between conditions.
c. Calculate the Pearson correlation.
d. Calculate the estimated standard error of the difference between correlated sample means.
e. Assuming a non-directional alternate hypothesis, perform a correlated samples t-test to compare
the means between conditions A and B.
f. What is the p-value for this t-test?
g. Assuming α = .05, is there a statistically significant difference between the means?
h. Compute eta squared. Does the observed value represent a small, medium, or large effect?
i. Compute the value of Cohen’s d, using the computational formula. Does the observed value
represent a small, medium, or large effect?
j. Using G*Power, how much statistical power was there to reject the null hypothesis?
9. Use the following to answer the questions below. A professor teaches a research methods lecture,
which requires knowledge of statistics. The professor wants to examine whether his Statistics Refresher
-228-
Back to TOC
Lecture increases his students’ understanding of the statistical concepts used throughout the course. To
address this question, the professor gives a statistics knowledge test on the first day of class (pretest),
and gives a similar test in class following his Statistics Refresher Lecture (posttest). Performance on each
of the two tests is compared to examine whether students did better on the second test. The relevant data
are presented in the table below.
Pretest n = 10 Posttest
X Pr etest  13 X Posttest  14
SSPretest = 120 SSPosttest = 268
r = 0.446
a. Expressed in terms of μ, what are the null and alternate hypotheses for this example?
b. Calculate the estimated variances and estimated standard deviations for each condition.
c. Calculate the estimated standard error of the difference between correlated sample means.
d. Perform a correlated samples t-test to compare the means between the Pretest and Posttest
conditions.
e. What is the p-value for this t-test?
f. Assuming α = .05, is there a statistically significant difference between the means? What
decisions would we make regarding the hypotheses?
g. Compute the value of eta squared. Dues the observed value represent a small, medium, or large
effect?
h. Compute the value of Cohen’s d, using the computational formula. Does the observed value
represent a small, medium, or large effect?
i. Using G*Power, how much statistical power was there to reject the null hypothesis?
10. Use the following to answer the questions below. The cafeteria wants to know which leading brand of
cola is preferred by students. They conduct a taste test study on campus by randomly sampling ten
undergraduates during dinner. Each undergraduate drinks a sample of Pepsi and a sample of Coke and
rates how much they like each brand of cola on a scale of 1 - 5 with 1 being "don't like" and 5 being "really
like." The relevant statistics are below; use them to answer the following questions.
Pepsi Coke
M = 3.4 M = 3.9
SS = 14.4 SS = 10.9
r = 0.032
a. Expressed in terms of μ, what are the null and alternate hypotheses for this example?
b. Calculate the estimated variances and estimated standard deviations for each condition.
c. Calculate the estimated standard error of the difference between correlated sample means.
d. Perform a correlated samples t-test to compare the means between the Pretest and Posttest
conditions.
e. What is the p-value for this t-test?
f. Assuming α = .05, is there a statistically significant difference between the means? What
decisions would we make regarding the hypotheses?
a. d = .50; α = .05 (non-directional); Power = .80
b. d = .10; α = .05 (directional); Power = .95
c. d = .80; α = .01 (directional); Power = .85
d. d = .25; α = .01 (non-directional); Power = .90
parameters given.
a. d = .25; α = .05 (non-directional); n = 26
-229-
Back to TOC
b. d = .55; α = .01 (non-directional); n = 40
c. d = .75; α = .05 (directional); n = 30
d. d = .45; α = .01 (directional); n = 20
Chapter 18: Introduction to Analysis of Variance
-230-
Back to TOC
18.1 Limitations of t-Tests
Earlier chapters introduced t-Tests that were used to
assess the statistical significance of the difference in the
means between two levels of an independent variable. The
independent-groups t-Test was used to compare sample
means that differed by levels of a between-subjects
independent variable, and the correlated-samples t-Test
was used to compare sample means that differed by levels
of a within-subjects independent variable.
The t-Test is used often, because most researchers want

to determine which levels of an independent variable are
statistically different from each other. However, t-Tests
have a limit, because you can compare only two levels of
an independent variable. The t-Test cannot be used to
compare three or more levels of an independent variable
simultaneously. For example, in a study that examines the
ability of Drug-X to alleviate anxiety, a researcher may
want to examine the effect of taking Drug-X on anxiety
compared to taking a placebo and taking nothing. Thus,
there are three levels of the independent variable Drug
Condition: (1) Drug, (2) Placebo, (3) Nothing/Control. How
could you analyze these data?
To analyze the data, the researcher could conduct several t-Tests. With three levels of the independent
variable Drug Condition there are three pairs of the conditions that can be compared using t-Tests:
Pair 1 = Drug vs. Placebo

Pair 2 = Drug vs. Control
Pair 3 = Placebo vs. Control
But, there is a problem in doing this, which is one of the least apparent topics in statistics. Whenever you
perform an inferential test on a set of data, there is a probability of incorrectly rejecting a true null
hypothesis, that is, making a Type I error. The probability of making a Type I error is equal to the chosen
or achieved alpha level (α). Hence, if you set α = .05, the probability of making a Type I error is .05. In
short, there is always some chance you will be wrong in your statistical decisions (remember, the
decisions are based on probabilities).
Importantly, the probability of making a Type I error is additive for all inferential tests performed on a set
of data. Specifically, every time you perform a t-Test on the same set of data you increase the probability
of making a Type I error by the alpha-level for each test. If you set alpha to α = .05 for each t-Test and
you perform one t-Test on a set of data, then the probability you will make a Type I error is .05. But if you
conduct a second t-Test on some part of that same set of data, then the probability you made a Type I
error on either test, or both tests, is now .05 + .05 = .10. This is because the probability of making a Type
I error was .05 for each test. If you perform a third t-Test, then the probability you make at least one Type
I error is .15,...,and so on. Thus, for every test you perform on the same data, you increase the chance of
making a Type I error by α. This is called Type I error inflation and must be avoided.
Because of Type I error inflation it can be unwise to conduct multiple t-Tests on the same set of data.
There is an easy way to ensure you are unlikely to make a Type I error over multiple t-Tests, which is
accomplished by adjusting the alpha level for all t-Tests to be performed. This is the Bonferroni
correction. In the Bonferroni correction you select an alpha-level that you would like to be at after
conducting all necessary t-Tests called the Familywise Type I error rate (αFW), which should be αFW =
.05 or less. Then divide the chosen Familywise error rate by the number of tests to be performed. If you
-231-
Back to TOC
plan to conduct three t-Tests and you want αFW = .05, divide .05 by 3 to get .0167. You use this adjusted
alpha-level for each t-Test, that is, instead of using α = .05 for each t-Test you use α = .0167. After your
three t-Tests, your overall alpha level will be α = .05.
Unfortunately, the Bonferroni correction is very conservative and you may fail to reject a null hypothesis
that in other situations you normally would reject. Thus, the Bonferroni is not well-liked. Appendix D
discusses the Bonferroni correction as well as several other procedures that are useful for conducting
multiple t-tests.
So multiple t-Tests are not good a good option when assessing more than two levels of an independent
variable...what do you use? We need a test that can simultaneously analyze several levels of an
independent variable. Luckily, there is an Analysis of Variance (ANOVA) procedure that allows one to
simultaneously analyze data belonging to two or more levels of an independent variable and also analyze
data from several independent variables; thus, this is a very applicable statistical procedure.
18.2 What is ANOVA?

Analysis of variance (ANOVA) does just what its name implies; it analyzes variance. ANOVA is
preferred over t-Tests in situations where you have more than two levels of an independent variable,
because you do not need to run several ANOVAs to analyze the data; ANOVA requires just one test.
What this means is there is no Type I error inflation when you must compare more than two levels of an
independent variable. But, just be aware that you can still produce Type I error inflation if you conducted
multiple ANOVAs or other tests on the same set of data.
ANOVA can also be used to analyze data when a research design has several independent variables.
Recall that with t-Tests you can compare two means drawn from two different samples that are assumed
to differ only by the levels of a single independent variable. This is also a limitation of t-Tests: The two
conditions being compared must differ along one independent variable. With ANOVA you can analyze
data from several independent variables simultaneously. For example, in the Drug example in Section
18.1 a researcher could manipulate whether people take Drug-X, Placebo, or Nothing and also
manipulate whether people go to a Relaxation Training class or not. If you “cross” the three levels of the
Drug Condition independent variable with the two levels of the Therapy independent variable, you get six
combinations, which are represented in the table below:
Drug Condition
Drug-X Placebo Control
Yes X X X
Relaxation Training?
No X X X
Research designs like these with two or more independent variables are factorial designs, and the
analyses are known as factorial analysis of variance. These will be discussed in Chapter 21. In Chapter
20, the oneway analysis of variance is discussed, which analyzes data from conditions that differ by the
levels of a single independent variable.
18.3 F-Statistic and Sources of Variance

ANOVA is a procedure for analyzing variance, but what ‘variance’ is being analyzed and where does it
come from? I’ll use the Drug-X example from above to explain below.
Consider a case where we manipulate whether different groups of people take Drug-X, a Placebo, or
Nothing (Control group). Say there are n = 10 people in each group and each person’s level of anxiety is
measured at the beginning and end of the study and the difference in anxiety between the beginning and
-232-
Back to TOC
end is the dependent variable. Below are hypothetical mean anxiety difference scores in each of the three
groups (assume that the anxiety difference scores have a range from 0 to 60, with higher values
indicating a greater decrease in anxiety):
Drug Condition
Drug-X Placebo Control
n = 10 n = 10 n = 10
X  30 X5 X4
Ignoring the group to which each subject belongs, if the variance in the anxiety scores was calculated
across all thirty people you would end up with a measure of total variation. That is, all variation in anxiety
due to differences among the subjects in the study as well any variation in anxiety across the levels of the
independent variable Drug Condition. This total variation includes variation from two sources, that is,
between group variance plus the within group variance:
 Total
2
  2Between   2Within
Between-groups variance is the variation in the dependent variable (anxiety difference scores) between
the levels of the independent variable. Thus, between groups variance is seen when there are differences
in the means across the levels of the independent variable. Between groups variance is assumed to
come from an influence of the independent variable and to sampling error. That is, differences between
means in the table above could reflect an effect of the independent variable, but could also be due to
random sampling error. Thus, between group variance is composed of variance due to an effect of the
independent variable and to variance due to sampling error:
 2Between   2Effect   2Error
Within-groups variance is the variation in the dependent variable (anxiety difference scores) that cannot
be attributed to the independent variable. Specifically, it is the variability among scores within each level
of the independent variable and summed over all of the levels of the independent variable. Because
within-group variance is occurring within levels of the independent variable it does not reflect an influence
of the independent variable. Rather, within group variance only reflects random uncontrollable sampling
error among subjects. Thus, within group variance is composed of only variance due to sampling error:
 2Within   2Error
Remember, between-group variance reflects an influence of the independent variable on the dependent
variable and sampling error, whereas within-group variance reflects only sampling error. If one could
show that variation between groups is greater than the random variation within groups (sampling error), it
could reflect a statistically significant influence of the independent variable on the dependent variable.
How do you determine whether the variability due to the effect of the independent variable is greater than
the random sampling error? Divide the between group variance by the within group variance, which forms
a variance ratio called the F-Ratio:
 2Between
F
 2Within
Between group variance reflects both the effect of the independent variable and random error; whereas
within group variance reflects only sampling error; thus:
-233-
Back to TOC
F
 Effect   Error
2 2

 2Error
By dividing a value that represents the combination of effect and error by error alone, the F-Ratio reflects
the variability due to the independent variable. Importantly, the expected value of the F-Ratio given the
null hypothesis is true and there is no effect of the independent variable is equal to 1 [i.e., E(F | H0 = True)
= 1]. Hence, if there is no influence of the independent variable on the dependent variable, the F-Ratio
should be a ratio of sampling error over sampling error; and dividing a value by itself yields a value of 1.
Thus, the expected F-Value under the null is one and F-Ratio can never be equal to zero, or negative. In
practice, because we are basing our analyses on sample data we need to estimate the between group
variance and the within groups variance. These estimated variances are called mean squares (MS) and
are obtained by dividing a sum of squares by a degree of freedom value:
SS
MS 
df
Because we have estimate two sources of variance (between groups and within groups), we calculate a
mean square within groups (MSW) and a mean square between groups (MSB):
SSB SSW
MSB  MSW 
df B df W
The mean square within groups and mean square between groups form the F-Ratio that we will calculate
in Chapter 19:
MSB
F
MSW
Thus, instead of relying on population
parameters for the F-Ratio, we estimate the
parameters from sample data, which we will go
over in detail in the following chapters.
18.4 Hypotheses with ANOVA

Hypotheses with ANOVA can get cumbersome
to write especially with more than one
independent variable. This is why most
researchers set all of the means equal to each
other under the null hypothesis and set all of the
means as unequal to each other under the
alternative hypothesis. Using the example from
above the hypotheses are:
H0: μDrug-X = μPlacebo = μControl

H1: μDrug-X ≠ μPlacebo ≠ μControl
The null hypothesis states that all three means
are predicted to equal one another (no
difference between means); and the alternate
hypothesis states that there will be some difference between the means, that is, some influence of the
-234-
Back to TOC
independent variable. Note that the alternative hypothesis is not stating which pairs of means are
predicted to be different; the difference could be between the Drug-X and Placebo groups, between the
Drug-X and Control groups, between the Placebo and Control groups, or between all groups.
18.5 Experimental Designs in ANOVA

What types of research designs can be analyzed with an analysis of variance? There are ANOVA
procedures for analyzing between-subjects experimental designs, ANOVA procedures for analyzing
within-subjects experimental designs, and procedures analyzing crossed or mixed-model designs. These
latter research designs include at least one variable manipulated between-subjects and at least one
variable manipulated within-subjects. Such designs are mixed in the sense that that include a mixture of
within-subject and between-subject independent variables.
18.6 Subscripts and Nomenclature

I have resisted using subscripts and superscripts in the statistical formulas to this point, because they
generally make things more confusing than they need to be, but for ANOVA they are necessary. Before
moving onto actual ANOVA procedures I want to address a little bit of the nomenclature you will
encounter in Chapters 19 and 20.
The formula below is used for obtaining something called the within-group sums of squares, which is a
measure of the total within group variation in a set of data:
 2
K n
SSW    Xik  Xk
k 1i 1
Yes, there two summation symbols and no, this is not a mistake. This formula is really not all that different
from the sums of squares formula used throughout the book. The superscript 'K' refers to the levels of an
independent variable that are being compared. In the example above there are three levels of the
th
independent variable Drug condition. The subscript ‘k’ refers to the k level of that independent variable.
The superscript 'n' refers to the number of subjects included in each level of the independent variable. In
th
the example above, there were n = 10 subjects in each level. The subscript 'i' refers to the i person in
that level of the independent variable. For example, if k = 1 and i = 1, as above, this refers to the first
rd
subject in the first level of the independent variable. If k = 2 and i = 3, this refers to the 3 subject in the
nd
2 level of the independent variable. And so on.
Consider first the portion of the formula in the parentheses. The X ik refers to the dependent variable score
from individual 'i' from level ‘k’, that is, the score of some individual who belongs to some level of the
independent variable. The X̄k value refers to the mean of the level from which that individual (Xik) came.
For example, if you have Drug, Placebo, and Control groups as the three levels of an independent
th
variable and there are n = 10 subjects in each group: Xik may refer to the score of the 4 subject from the
th
Placebo group, and X̄k would refer to the mean of the Placebo group. Or, Xik could be the 9 subject in the
Control group, and X̄k would refer be the mean of the Control group. Hence, the part in the parentheses is
simply telling you to calculate the difference between an individual’s dependent variable score and the
mean of his/her group.
The summation symbol with the superscript n tells you to do this for each individual in a particular level of
the independent variable. That is, calculate the difference between each individual’s score and the mean
of that group, square those differences, and sum the squared differences. The summation with the
superscript K tells you to do this for each level of the independent variable. That is, after you find the sum
of squares for one level, calculate the sum of squares for each other level. Thus, all that this equation
above is telling you to do is find the deviation between an individual’s score and the mean of that
individual’s group, square that difference, sum the squared differences, and do this for each subject in
-235-
Back to TOC
each group. The subscripts and nomenclature look crazy at first, but the best think to do is to identity what
‘i’, ‘n’, ‘k’, and any other nomenclature stand for. Once you have that figured out, start within the
parentheses and work your way out from there.
-236-
Back to TOC
1. What does the alternate hypothesis ask for a one-way between-subjects analysis of variance?
2. What is the difference between between-group variability and within-group variability?
3. What does between-group variability reflect? Where does it come from?
4. What does within-group variability reflect? Where does it come from?
5. When will the F ratio approach 1.00? When will the F ratio be greater than 1.00?
6. What is the difference between the sums of squares total, the sums of squares between-groups, and
the sums of squares within-groups? How are they related?
7. Consider the scores in an experiment involving four levels of an independent variable. Without
computing anything, what value must the sums of squares within-groups equal? Why?
A B C D
10 12 9 4
10 12 9 4
10 12 9 4
10 12 9 4
10 12 9 4
8. State the critical value of F-Value for a oneway ANOVA under each of the following conditions. Note
the n values are the number of subjects in each group (they are not nT):
a. k = 3, n = 7, α = .01
b. k = 5, n =10, α = .01
c. k = 4, n = 5, α = .05
d. k = 6, n = 4, α = .05
e. k = 3, n = 10, α = .01
f. k = 2, n = 12, α = .05
g. k = 5, n = 15, α = .01
h. k = 4, n = 5, α = .05
9. Determine the p-values for a oneway ANOVA under each of the following conditions. Note the n values
are the number of subjects in each group (they are not nT):
a. k = 2, n = 10, F = 2.50
b. k = 3, n = 10, F = 2.50
c. k = 2, n = 8, F = 5.20
d. k = 4, n = 8, F = 5.00
e. k = 3, n = 6, F = 4.20
f. k = 3, n = 12, F = 4.20
10. State the null and alternate hypotheses for a oneway ANOVA with 5 levels of the independent
variable:
11. Insert the missing entries in the summary table for a one-way analysis of variance having four levels
of the independent variable.
Source SS df MS F
Between 3 18 3.5
Within
Total 23
12. Insert the missing entries in the summary table for a one-way analysis of variance having three levels
of the independent variable and n = 20.
-237-
Back to TOC
Source SS df MS F
Between 25
Within 140
Total 190 99
13. Insert the missing entries in the summary table for a one-way analysis of variance having three levels
Source SS df MS F
Between 2 10
Within 4
Total 128
14. Insert the missing entries in the summary table for a one-way analysis of variance having seven levels
Source SS df MS F
Between 6 5125
Within 2500 25
Total
15. Assume you have an independent variable, A, with three levels (A1, A2, A3). If you set α = .05, what is
the Familywise Type I error rate after running all possible t-tests?
16. Assume you have an independent variable, A, with four levels (A1, A2, A3, A4). If you set α = .01, what
is the Familywise Type I error rate after running all possible t-tests?
17. Assume you have an independent variable, A, with four levels (A1, A2, A3, A4). If you set α = .03, what
is the Familywise Type I error rate after running all possible t-tests?
18. Assume you have an independent variable, A, with four levels (A1, A2, A3, A4). If you want your
Familywise Type I error rate to be .05 after running all possible t-tests, what must the alpha level be set to
for each test?
19. Assume you have an independent variable, A, with three levels (A1, A2, A3). If you want your
Familywise Type I error rate to be .02 after running all possible t-tests, what must the alpha level be set to
for each test?
Chapter 19: Oneway Between Subjects ANOVA
-238-
Back to TOC
19.1 What is The 'Oneway' Design?
The simplest research design that can utilize analysis of
variance includes one independent variable with at least two
levels. In such designs it matters not whether you use a t-
Test or ANOVA to analyze your data. Indeed, if you were to
perform an ANOVA on such data and then perform a t-Test
on the same data (just for fun of course!) you would find that
2
F=t .
An ANOVA that is conducted on data from a research design

that includes only one independent variable is known a
oneway ANOVA, where oneway refers to there being only
one independent variable that can influence the dependent
variable. This chapter examines how to properly conduct
a oneway between-subjects ANOVA, that is, use ANOVA
procedures to analyze the influence of a single independent
variable that was manipulated between-subjects.
Generally, the oneway ANOVA is used to assess the relationship between variables when
 The dependent variables is quantitative

 The independent variable is qualitative in nature
 The independent variable has two or more levels
 The independent variable is manipulated between-subjects
For example, say an educational psychologist is interested in the influence of testing location on test
performance. Specifically, does the location where you study for a test and take a test have an influence
on test performance? The researcher samples n = 12 students and has each student study a list of 20
words. One week later each student’s memory for the words is tested by having each student recall as
many studied words as possible. The manipulation is the testing location.
One group of n = 3 students is tested in the same room where the words were studied (Same condition);
a second group of n = 3 students is tested in a different room, but with the same characteristics as the
original room (Different condition); a third group of n = 3 students is tested in a basement (Basement
condition); and a fourth group of n = 3 students is tested in the basement, but is presented with a picture
of the room where the words were studied (Picture condition). Below, the number of words correctly
recalled by each student is presented for each group:
Condition
Same Different Basement Picture
Subject XS Subject XD Subject XB Subject XP
A 8 D 5 G 3 J 7
B 10 E 7 H 5 K 8
C 6 F 6 I 4 L 6
∑XS = 24 ∑XD = 18 ∑XB = 12 ∑XP = 21
XS  8 XD  6 XB  4 XP  7
The sum of the scores and the mean in each group is included in the table. We will analyze these data
using a between-subjects oneway ANOVA. The prediction is that the location in which the recall test was
taken will have some effect on test performance. Thus, the null and alternate hypotheses are below, and
we’ll select an alpha level of α = .05:
H0: μS = μD = μB = μP H1: μS ≠ μD ≠ μB ≠ μP
-239-
Back to TOC
Section 19.2 covers how the sums of squares, degrees of freedom, mean squares, and F-Ratio are
calculated. Remember, there are three sources of variance: between-group variance, within-group
variance, and total variance. Hence, we need to calculate three sums of squares and three degrees of
freedom values. We then need to find two mean squares and the F-Ratio.
19.2 Numerical Example

Before beginning, it is good to construct an ANOVA
summary table (see example at right), in which you
Variance Source SS df MS F
list each degrees of freedom (df), sum of squares Between ?? ?? ?? ??
(SS), mean square (MS), and the F-Ratio, as each Within ?? ?? ??
value is calculated. This is a nice and neat way to Total ?? ??
store your ANOVA information, in one location. We'll Note: Cells marked with ?? must be calculated; the
start by determining the degrees of freedom for each empty cells are not needed
source of variance.
Between-subjects degrees of freedom (dfB) are equal to k – 1, where k is the number of levels of the
independent variable. In this case there are four levels of the independent variable (Same, Different,
Basement, and Picture); thus, k = 4, and dfB = 4 – 1 = 3.
Within subjects degrees of freedom (dfW) are equal to nT – k, where nT is the total number of subjects.
In this case there are nT = 12 subjects, so dfW = 12 – 4 = 8. Another way to obtain the degrees of freedom
within subjects is by multiplying the degrees of freedom in each level of the independent variable (n – 1)
by the number of levels of the independent variable (k), such that df W = k(n - 1). In this example, there are
n = 3 subjects in each level of the independent variable; thus, df W = 4(3 - 1) = 8. But, this works only when
there are an equal number of subjects in each level of the independent variable.
Total degrees of freedom (dfT) are equal to nT – 1. I this example, with twelve subjects, df T = 12 – 1 =
11. Notice that dfT = dfB + dfW (11 = 3 + 8). This is true in any oneway ANOVA.
(The degrees of freedom between groups and degrees of freedom within groups are used to determine
the statistical significance of the outcome of the F-Ratio.)
Next, calculate the sum of squares for each of the three sources of variance. We’ll start with the sum of
squares associated with the total variance, which is the sum of squares total (SST). The sum of squares
total is the total variation over all of the scores in the data set, and is found by comparing each individual
value to the grand mean of the data set. The grand mean ( X or G) is the mean of all the scores in the
data set; that is, the total sum of all 12 scores divided by nT. The grand mean is:
 X 75
X   6.25
n T 12
You can also calculate the grand mean by summing the means across the levels of the independent
variable and dividing by k. Indeed, the grand mean can be thought of as the mean of means:
 X 25
X   6.25
k 4
K n 2
SST     Xik  X 
k 1 i 1 
The formula for the sum of squares total is:
-240-
Back to TOC
To find the sum of squares total: first subtract the grand
mean from each score in the data set (Xik). Next, square Subject Xik X Xik  X X
ik 
2
each the differences. Finally, sum the squared A 8 1.750 3.062

differences. This is demonstrated in the table to the B 10 3.750 11.062
right. As you can see, calculating the total sum of C 6 -0.250 0.062
squares is no different than calculating a sum of squares D 5 -1.250 1.562
for any group of scores. E 7 0.750 0.562
F 6 -0.250 0.062
Next, calculate sum of squares within groups (SSW). G 3 -3.250 10.562
The formula for within-group sum of squares is: H 5 -1.250 1.562
I 4 -2.250 5.062
 
K n 2 J 7 0.750 0.562
SSW    Xik  X k K 8 1.750 3.062
k 1 i 1 L 6 -0.250 0.062
SST = 40.244
This look similar to the formula for SST, except that the grand mean has been replaced by the mean of
each group (X̄k). When calculating the sum of squares within groups you are actually finding the sum of
squares for each level of the independent variable and then adding the sums of squares across all levels
of the independent variable. Specifically, first you calculate the difference between an individual’s
score (Xik) and the mean of the level of the independent variable that individual belongs (X̄k). You do this
for each individual, being careful to subtract the correct mean from each individual. Next, square these
differences and add them to get SSW . This is demonstrated in the table below, where the (X̄k) values for
each individual are included as a reference:
Subject Xik Xk X  Xk 
ik X
 Xk  ik
2
A 8 8 8-8=0 0
B 10 8 10 - 8 = 2 4
C 6 8 6 - 8 = -2 4
D 5 6 5 - 6 = -1 1
E 7 6 7-6=1 1
F 6 6 6-6=0 0
G 3 4 3 - 4 = -1 1
H 5 4 5-4=1 1
I 4 4 4-4=0 0
J 7 7 7-7=0 0
K 8 7 8-7=1 1
L 6 7 6 - 7 = -1 1
SSW = 14
The sum of squares within groups represents the summed variability within each level of the independent
variable and summed across all levels. Specifically, it is the sum of squares for each level of the
independent variable, added together.
K 2
SSB   n k  Xk  X 
k 1  
Finally, calculate the sum of squares between groups (SSB):
Back to TOC
This formula is a bit different from the previous two, but is just another sum of squares. The part of the
formula in parentheses tells you to subtract the grand mean from the mean associated with a level of the
independent variable. You next square this mean difference and then multiply that squared mean
difference by the sample size for that level of the independent variable (nk). Once this is done for each
group, add the products together to get SSB. Remember, nk is equal to 3 for each level of the independent
variable in this example. This is demonstrated in the table below:
-241-
Group Xk X k X  X k X 2

n k Xk  X 2
Same 8 1.750 3.062 9.186

Different 6 -0.250 0.062 0.186
Basement 4 -2.250 5.0620 15.186
Picture 7 0.750 0.562 1.686
SSB = 26.244
This value, (SSB = 26.244) is a measure of the total variability between the means of the levels of the
independent variable.
We now have SST, SSW , and SSB. Note that SST = SSW + SSB (40.244 = 14 + 26.244). In practice, you
could actually calculate two of the three sums of squares values and use those two to obtain the third
value.
Next, we must calculate the necessary mean squares (MS) that are then used to calculate the F-Ratio;
specifically, we must calculate a mean-square between groups (MSB) and a mean-square within
groups (MSW). Calculating each mean square simply involves dividing a sum of squares by associated
degrees of freedom. Remember, the mean square is the average variability for a source of variance.
Mean-square between groups is:

SSB 26.244
MSB    8.748
df B 3
Mean square within groups is:

SSW 14
MSW    1.75
df W 8
The F-Ratio is calculated by dividing the mean square between groups (effect) by the mean square within
groups (error):
MSB 8.748
F  5
MSW 1.75
The table to the right is the complete ANOVA Variance Source SS df MS F

summary table is listed with each of the
Between 26.244 3 8.748 5
values we just calculated. The value of F = 5
is the obtained F-Ratio, which is used to Within 14 8 1.75
determine whether the analysis of variance Total 40.244 11
was statistically significant.
To assess the significance of the F-Ratio you look-up a p-value in Table 3 in Appendix A. Table 3 (there
are six of them) lists the probabilities associated with areas under the F-Distribution. Each of the six
tables 3 are associated with a different number of degrees of freedom between groups (df B), ranging from
dfB = 1 through dfB = 10 (see upper left corner of each table). Each table is structured the same, with F-
Ratio values in the leftmost column (F) and degrees of freedom error/within subjects (df W) values in each
column heading. In the example there are dfB = 3, so we’ll use Table 3-C, a portion which is reproduced,
below.
F 2 3 4 5 6 7 8 9 10 12 14 16 18 20 22 24 26 28 30 40 45 50 60 80 100 250 1000
4.50 0.1872 0.1242 0.0901 0.0695 0.0558 0.0464 0.0395 0.0343 0.0303 0.0246 0.0207 0.0180 0.0159 0.0144 0.0131 0.0122 0.0113 0.0107 0.0101 0.0082 0.0076 0.0072 0.0065 0.0057 0.0053 0.0043 0.0038
4.60 0.1837 0.1210 0.0873 0.0668 0.0535 0.0442 0.0375 0.0324 0.0285 0.0230 0.0193 0.0166 0.0147 0.0132 0.0120 0.0111 0.0103 0.0097 0.0092 0.0074 0.0068 0.0064 0.0058 0.0051 0.0047 0.0037 0.0033
-242-
Back to TOC
4.70 0.1804 0.1180 0.0845 0.0644 0.0512 0.0421 0.0356 0.0307 0.0269 0.0215 0.0180 0.0154 0.0136 0.0122 0.0111 0.0102 0.0094 0.0088 0.0083 0.0067 0.0061 0.0057 0.0052 0.0045 0.0041 0.0033 0.0029
4.80 0.1772 0.1150 0.0819 0.0620 0.0491 0.0402 0.0338 0.0290 0.0254 0.0202 0.0167 0.0143 0.0125 0.0112 0.0101 0.0093 0.0086 0.0080 0.0076 0.0060 0.0055 0.0051 0.0046 0.0040 0.0036 0.0029 0.0025
4.90 0.1741 0.1122 0.0794 0.0598 0.0471 0.0384 0.0321 0.0275 0.0239 0.0189 0.0156 0.0133 0.0116 0.0103 0.0093 0.0085 0.0079 0.0073 0.0069 0.0054 0.0050 0.0046 0.0041 0.0035 0.0032 0.0025 0.0022
5.00 0.1712 0.1096 0.0770 0.0577 0.0452 0.0367 0.0306 0.0261 0.0226 0.0178 0.0146 0.0124 0.0107 0.0095 0.0086 0.0078 0.0072 0.0067 0.0063 0.0049 0.0045 0.0041 0.0037 0.0031 0.0028 0.0022 0.0019
5.10 0.1683 0.1070 0.0747 0.0557 0.0434 0.0351 0.0291 0.0247 0.0214 0.0167 0.0136 0.0115 0.0099 0.0088 0.0079 0.0072 0.0066 0.0061 0.0057 0.0044 0.0040 0.0037 0.0033 0.0028 0.0025 0.0019 0.0017
5.20 0.1655 0.1045 0.0726 0.0537 0.0417 0.0335 0.0277 0.0234 0.0202 0.0157 0.0127 0.0107 0.0092 0.0081 0.0072 0.0066 0.0060 0.0056 0.0052 0.0040 0.0036 0.0033 0.0029 0.0025 0.0022 0.0017 0.0014
5.30 0.1628 0.1021 0.0705 0.0519 0.0401 0.0321 0.0264 0.0223 0.0191 0.0147 0.0119 0.0099 0.0085 0.0075 0.0067 0.0060 0.0055 0.0051 0.0047 0.0036 0.0032 0.0030 0.0026 0.0022 0.0020 0.0015 0.0013
5.40 0.1602 0.0998 0.0685 0.0502 0.0385 0.0307 0.0252 0.0211 0.0181 0.0139 0.0111 0.0093 0.0079 0.0069 0.0061 0.0055 0.0050 0.0046 0.0043 0.0032 0.0029 0.0027 0.0023 0.0020 0.0017 0.0013 0.0011
5.50 0.1577 0.0976 0.0666 0.0485 0.0371 0.0294 0.0240 0.0201 0.0171 0.0131 0.0104 0.0086 0.0073 0.0064 0.0057 0.0051 0.0046 0.0042 0.0039 0.0029 0.0026 0.0024 0.0021 0.0017 0.0015 0.0011 0.0010
To locate the p-value associated with the obtained F-Ratio (5.00), locate the dfW = 8 value in the column
headings (highlighted in yellow, above) and scroll down that column until you come to the row associated
with F = 5.00 (highlighted in yellow, above). The p-value associated with this F-Ratio is .0306, which is
less than the selected alpha level of .05. This means that the outcome of the ANOVA is statistically
significant. In this case, we reject the null hypothesis and accept the alternate hypothesis. This significant
difference in an oneway ANOVA is known as a main effect, which means there is a significant difference
among the means along independent variable.
19.3 Post Hoc Testing

A statistically significant ANOVA tells us that the independent variable had some influence on the
dependent variable, but does not tell you where. That is, a statistically significant ANOVA tells us that at
least two of the means are significantly different, but does not tell us which means are significantly
different. To determine which means are significantly different from each other we need to run a post-hoc
test. Post-hoc tests are used to determine which means are significantly different from each other. It is
important to remember that post-hoc testing should be conducted only if an ANOVA was significant; if the
ANOVA is not statistically significant you should never do post-hoc testing, because any results from
those post-hoc tests would be meaningless.
A common post-hoc test is Tukey’s Honestly Significant Difference (HSD). The value calculated in the
Tukey's test is a common difference (CD) that must be found between means to conclude the difference
between means is statistically significant. That is, we’ll calculate the Tukey’s HSD value and then
compare each pair of means to find each mean difference. If the difference between any two means
exceeds the Tukey’s HSD value, then that difference is considered to be statistically significant. The
formula for Tukey’s HSD is:
MSW
HSD  q
nk
In the formula, MSW is the mean-square within groups that was calculated in the ANOVA above (MSW =
1.750), and nk is the number of subjects per level of the independent variable; that is, the number of
subjects in any one level of the independent variable (nk = 3). Note that nk must be equal across the
levels of the independent variable to use this version of the Tukey’s test.
The q is the Studentized range statistic and is obtained from Table 4 in Appendix A. To find the value
for q, in the column headings, look up the number of levels of the independent variable that were being
compared in the ANOVA, which in the example was k = 4. Next, move down that column until you come
to the rows associated with dfW from the ANOVA, which was dfW = 8 in the example above. You will see
two values, 4.53 and 6.19. Use the upper value (q = 4.53) if α = .05 in the ANOVA, and use the lower
value (q = 6.20) if α = .01 in the ANOVA. Because α =.05 in the ANOVA above, we’ll use the q = 4.53.
Tukey’s HSD for this example is:
 4.53 0.583  4.530.764  3.461

MSW 1.75
HSD  q  4.53
nk 3
-243-
This value (3.461) is the minimum difference needed for a difference between means to be considered
significantly significant. Looking at that the means for the four groups in the example we have:
Same condition: XS  8
Different condition: XD  6
Basement condition: XB  4
Picture condition: XP  7
With HSD = 3.461, the only two means that are statistically significant from each other is the difference
between the Same condition and Basement condition, which has a mean difference of 4. Note that when
comparing means, you are just looking for the absolute difference between the means.
Unfortunately, this version of the Tukey’s HSD test cannot be used when you have unequal sample sizes
across the levels of the independent variable. If you have unequal sample sizes across the level of the
independent variable, you can use Fisher’s Least Significant Difference (LSD) test as a post-hoc test
following a statistically significant ANOVA test.
The logic of Fisher’s LSD is the same as Tukey’s HSD test. You first calculate an LSD value that
represents the minimum differences that is needed between a pair of means for those means to be
considered statistically significant. The catch is that with the Fisher’s LSD test, you have to calculate a
new LSD value for each pair of means, due to unequal sample sizes across levels of the independent
variable. But, if you have equal sample sizes across the levels of your independent variable, you only
need to calculate one LSD value. Hence, in our example in this chapter, we need to calculate only one
Fisher’s LSD value. Fisher’s LSD formula is:
 1 1 
LSD  t  MSW   
 n1 n 2 
The MSW is the mean square within subjects from the ANOVA. The n1 and n2 values are the sample sizes
of the two groups that are being compared in the Fisher's LSD test. In the present example, because all
of the sample sizes are equal to nk = 3, n1 = n2 = 3, so we only need to calculate on LSD value. But, if we
had two samples with unequal sizes, then n1 ≠ n2 and we might have to
calculate several Fisher's LSD values. The tα value is the critical t-Value  1 1 
that would be used if comparing these two groups with an independent LSD  t  MS W  n  n 
groups t-Test. This t-Value is obtained by looking up a critical t-Value for a  1 2
non-directional hypothesis in Table 2 in Appendix A or in the statistics 1 1

packet, using the alpha level from the ANOVA, and the degrees of LSD  2.78 1.75 3  3 
freedom for comparing these two levels of the independent variable (df =
n1 + n2 – 2). In this example α = .05 from the ANOVA above and df = 3 + 3 LSD  2.78 1.750.333  0.333
– 2 = 4. Thus, tα = 2.78. Using this information, the value of Fisher’s LSD LSD  2.78 1.750.666 
is:
LSD  2.78 1.166
LSD  2.781.08
LSD  3.002
Use this Fisher’s LSD value to determine whether the means from the example above are significantly
different from each other. Again, you would normally calculate a different Fisher’s LSD any time you are
comparing two levels of the independent variable that have different sample sizes, but because we have
equal sample sizes we only need to calculate the one LSD value. Also, note that Fisher’s LSD = 2.998 is
smaller than the Tukey’s HSD = 3.461 that we calculated above. This will almost always be the case,
-244-
Back to TOC
which makes Fisher's LSD a more powerful post-hoc test than Tukey's HSD. In the present example, the
mean difference between the Same group ( X S  8 ) and the Basement group ( X F  4 ) is still statistically
significant, because the mean difference (4) is greater than the LSD value. But now, the mean difference
between the Picture ( X P  7 ) group and the Basement group ( X B  4 ) is also significant. Hence, you are
more likely to say two groups are significantly different with Fisher’s than with Tukey’s, which is why LSD
is more powerful.
19.4 Effect Size and Explained Variance

2
Like other inferential statistics you can also calculate the effect size (η ) of the significant main effect; that
is, the proportion of the total variance that is explained in the relationship between the independent
variable and dependent variable. In an analysis of variance the total variance is equal to SST. The
relationship between the independent variable and dependent variable. The effect of the independent
variable on the dependent variable is the variation between groups (SS B). If you divide the between-
groups variation (SSB) by total variation (SST) you are left with a value that is equal to the proportion of
variance explained in the relationship between the independent variable and dependent variable. Thus,
from the example above, effect size is:
SSB 26.244
2    0.652
SST 40.244
About 65% of the variance in the dependent variable is attributable to the independent variable. Another
popular measure of effect size is Cohen’s f, which is similar to Cohen’s d from earlier chapters. Cohen’s f
can be conceptualized as the standard deviation of the ‘effect’ over the standard deviation of the ‘error’:
 Effect
f
 Error
However, it is often not clear how the standard deviation of the ‘effect’ and the standard deviation of the
‘error’ can be calculated. Thus, most researchers opt for calculating Cohen’s f from the eta-squared
value:
2
f
1  2
2
The eta-squared value if the one we calculated above (η = 0.652). Thus, in this example, Cohen’s f is:
2 0.652 0.652
f    1.874  1.369
1  2 1  0.652 0.348
Cohen also provides useful labels to describe how ‘big’ or how ‘strong’ of a relationship the effect size
indicates. The table below reports the minimum Cohen’s d and eta-squared values that correspond to
‘weak’, ‘moderate’, and ‘strong’ effect sizes (also called ‘small’, ‘medium’, and ‘large’ effect sizes):
2
Effect Size Cohen’s f eta-squared (η )
“Small” (“Weak”) .10 .01
-245-
Back to TOC
“Medium” (“Moderate”) .25 .06
“Large” (“Strong”) .50 .14
19.5 Statistical Power for Oneway ANOVA

Recall, from earlier chapters, statistical power (1 - β) is the probability of correctly rejecting a false null
hypothesis. You want this probability to be high, so that a statistically significant result most likely reflects
a true significant difference; or to show that a non-significant difference is most likely not significant. Thus,
you generally want the statistical power of an inferential test to be .80 or greater.
Using the ANOVA example from earlier sections, we can determine our achieved power in the analysis of
variance for correctly rejecting a false null hypothesis. Recall, that the total number of subjects in the data
set was n = 12, the alpha level was α = .05, and the effect size was f = 1.369.
To perform a post hoc power analysis, open G*Power 3 and, under “Test family,” select “F tests.” Under
“Statistical test” choose “ANOVA: Fixed effects, omnibus, one-way”, and under “Type of Power Analysis”
choose “Post hoc: Compute achieved power.” Next, you'll need to enter the α-level, total sample size,
effect size and the number of groups from the t-test above. The alpha level was α = .05, n = 12, f = 1.369,
and number of groups was three (k = 3).
After clicking calculate, G*Power tells you that the statistical power is .9521915, which is high and is
good. Thus, there is a high probability that we are correctly rejecting the null hypothesis if it is indeed
false.
-246-
Back to TOC
We can also use G*Power in a priori
power analyses for a oneway ANOVA
to determine the number of subjects
needed to end up with a desired level
of power. For example, assume I am
planning to collect data from a sample
of subjects in a between subjects
design that has five levels of the
independent variable (k = 5). I know
from prior studies that the effect size
for such a manipulation is generally
“small” and near f = .15. For my
inferential test I am planning to use α
= .05 and I want Power = .85.
To determine the appropriate sample

size, open G*Power 3 and under
“Test family” be sure “F tests” is
chosen, under “Statistical test”
choose “ANOVA: Fixed effects,
omnibus, one-way”; and under “Type
of Power Analysis” choose “A priori:
compute required sample size.” Next,
you'll need to enter the α-level,
expected effect size, desired power,
and number of groups.
After clicking Calculate you are provided with the total sample size needed to achieve this amount of
Power, which is n = 605 in this example.
-247-
Back to TOC
1. Use the following information to answer the questions below. A researcher studied the relationship
between task difficulty and performance. Twenty subjects worked on an identical task, but five subjects
were told that the task was of easy, five were told the task moderately difficult five were told that the task
highly difficult, and five were given no information. Scores range from 0 to 10, where higher values
indicate better task performance. The data follow:
Easy Moderate High No Info

8 6 2 5
7 6 3 4
5 4 2 5
8 3 5 6
7 6 3 5
a. Calculate the mean for each condition and the grand mean.
b. Determine the degrees of freedom total, the degrees of freedom between groups, the degrees of
freedom within groups.
c. Calculate the mean within groups, and the mean square between groups.
f. Assuming α = .05, is the result of the analysis of variance statistically significant? What decisions
do we make with the null and alternate hypotheses?
g. Analyze the nature of the relationship between supposed task difficulty and task performance
using the Tukey HSD test. That is, calculate Tukey’s HSD value and then use that value to
determine which means are significantly different.
h. Compute the value of eta-squared and then Cohen’s f. Does the observed Cohen’s f value
represent a small, medium or large effect?
i. Use Cohen’s f and G*Power 3, determine the amount of statistical power to correctly reject the
null hypothesis.
2. Use the following information to answer the questions below. An investigator tested the relationship
between drug intake and anxiety. Eighteen subjects who had been diagnosed with anxiety were given a
different level of an antianxiety drug (100 mg/day, 200 mg/day, or 300 ms/day). Each subject had the
same baseline level of anxiety, and each subject’s level of anxiety was re-measured after three months
taking the drug. Scores range from 0 to 10, with higher values indicating more anxiety. The data follow:
100 mg/day 200 mg/day 300 mg/day

8 6 2
5 7 1
6 5 0
8 6 2
9 5 2
9 7 2
c. Calculate the sum of squares total, the sum of squares within groups, and the sum of square
between groups.
d. Calculate the mean within groups, and the mean square between groups.
e. Calculate the F-Ratio.
f. What is the p-value?
-248-
g. Assuming α = .01, is the result of the analysis of variance statistically significant? What decisions
h. Analyze the nature of the relationship between supposed task difficulty and task performance
i. Compute the value of eta-squared and then Cohen’s f. Does the observed Cohen’s f value
j. Using Cohen’s f and G*Power 3, determine the amount of statistical power to correctly reject the
null hypothesis.
3. For each of the following situations, find the total number of subjects that would be needed to achieve
the desired level of Power.
a. k = 3; f = .30; α = .05; Power = .80

b. k = 3; f = .30; α = .05; Power = .95
c. k = 4; f = .20; α = .01; Power = .80
d. k = 4; f = .20; α = .05; Power = .80
4. For each of the following situations, find the amount of Power, based on the parameters given.
a. k = 2; f = .25; α = .05; n = 75
b. k = 2; f = .40; α = .01; n = 180
c. k = 4; f = .45; α = .05; n = 30
d. k = 5; f = .35; α = .01; n = 190
5. Use the following information to answer the questions below. The bystander effect is the phenomenon
where an individual is less likely to help a person when others are around. I examine the bystander effect
among Boy Scouts. I randomly select 24 Boy Scouts and assign each to one of four groups. In each
group the scout is seated at a table with my assistant, but the scout believes my assistant is another
subject. While the scout and my assistant complete paperwork, my assistant falls out of his chair and I
record the time (in seconds) it takes the scout to get help. I vary the number of people seated at the table.
In Group A only the one scout and my assistant are present (one bystander); in Group B the scout, my
assistant, and one other person are present (two bystanders); in Group C the scout, my assistant, and
three others are present (four bystanders); and in Group D the scout, my assistant, and seven others are
present (eight bystanders). The time, in seconds, it took each scout to get help are below.
Group A Group B Group C Group D

(1 Bystander) (2 Bystanders) (4 Bystanders) (8 Bystanders)
7 7 7 14
8 8 8 11
6 12 7 15
5 11 12 16
5 8 11 17
11 8 9 17
c. Calculate the sum of squares total, the sum of squares within groups, and the sum of square
between groups.
d. Calculate the mean within groups, and the mean square between groups.
e. Calculate the F-Ratio.
f. What is the p-value?
-249-
g. Assuming α = .01, is the result of the analysis of variance statistically significant? What decisions
h. Analyze the nature of the relationship between supposed task difficulty and task performance
i. Compute the value of eta-squared and then Cohen’s f. Does the observed Cohen’s f value
j. Using Cohen’s f and G*Power 3, determine the amount of statistical power to correctly reject the
null hypothesis.
-250-
Back to TOC
Chapter 20: Factorial ANOVA
20.1 What are a Factorial Design and Factorial ANOVA?
From earlier chapters, recall that a factorial design includes
more than one independent variable, and the levels of each
independent variable are ‘crossed’ or ‘combined’ with the levels
of the other independent variables. A simple example of this
concept is presented in the picture to the right. Think of the bottle
of pills (Pharmaceutical consumption) and the bottles and cans of
beer (Alcohol consumption) as two different independent
variables that a person may or may not take. Taken alone, each
independent variable has its own (independent) influence on the
person doing the consuming: both the pills and the alcohol cause
drowsiness. But, when the pills and the alcohol are consumed
together (independent variables are combined), something new
happens: the consumer dies. Hence, when they are combined,
the independent effects of taking pills alone and drinking alcohol
alone interact to produce a new influence on an individual: death!
This is what a factorial design is all about (without the killing of
subjects...usually). Factorial experimental designs combine the
influence of two or more independent variables to examine their
combined (or ‘joint’) influence on a dependent variable. Indeed, the purpose of factorial designs to t
examine whether there is a combined influence of impendent variables, or not. To analyze factorial
designs, we use factorial ANOVA.
Generally, factorial, between-subjects ANOVA is used to assess the relationship between variables when
 The dependent variables is quantitative

 Both of the independent variables are qualitative in nature
 Both of the independent variables are manipulated between-subjects
 Each independent variable has two or more levels
 Each level of each independent variable is combined with each level of the second independent
variable
To generalize the example from above, say that I have two independent Variable J
variables, ‘J’ and ‘K,’ and each independent variable has two levels (J1, J1 J2
J2, K1, and K2). If each level of ‘J’ is crossed/combined with each level of
‘K’, I have a factorial design. In this case, because there are two levels Variable K K1 J1K1 J2K1
of J (J1 and J2) and two levels of K (K1 and K2) there are 2 (levels of J) x K2 J1K2 J2K2
2 (levels of K) = 4 total combinations/conditions, which can be seen in
the table below. This type of a design with two independent variables each with two levels is called 2 by 2
factorial design and is the simplest factorial design. This type of factorial design will be analyzed using
ANOVA in this chapter.
With a factorial design, several effects may be statistically significant in the ANOVA performed on the
data. First, there can be a significant main effect of each independent variable. A main effect occurs
when the mean difference between the levels of an independent variable is statistically significant. For
example, a main effect of ‘J’ would indicate that the difference between the mean of J1 ( J1 ) and the mean
of J2 ( J 2 ) is statistically significant. In the case of a 2 x 2 factorial design, which has two independent
variables, there are two potential main effects: A main effect of J and a main effect of K.
An interaction between variables can be present. An interaction occurs between the independent
variables, not the levels of the independent variables. An interaction is found when the influence of one
-251-
Back to TOC
independent variable on the dependent variable changes across the levels of another independent
variable. States differently, the effect of one independent variable on the dependent variable depends on
the levels other independent variables. For example, the difference between J1 and J2 may differ across
the levels of K.
The table to the right illustrates what an interaction may look Variable J
like if the difference between J1 and J2 changes across J1 J2 Difference
levels of K. Specifically, the mean difference between J1 and
J2 at level K1 is only a difference of 5; but the difference Variable K K1 90 85 5
between J1 and J2 at level K2 is a difference of 40. Thus, the K2 50 10 40
influence of ‘J’ on the dependent variable is changing across
levels of K. This is an interaction.
The number of potential interactions depends on the number of independent variables in the factorial
design. In the case of a 2 by 2 factorial design, there is the potential for only one interaction. If there were
three independent variables (e.g., J, K, L) then there are four potential interactions: (1) between J and K;
(2) between J and L; (3) between K and L; and (4) between J, K, and L. Thus, the number of potential
interactions depends on the number of combinations of variables, including combinations of more than
two independent variables.
20.2 What do Main Effects and Interactions Look Like?

What do main effects and interactions look like in a 2 x 2 design? That is, what do significant main effects
and interactions look like? There are many potential outcome patterns of data, but there are only eight
possible outcomes based on the statistical significance of the main effects and an interaction. IN the
tables and graphs below, I have displayed hypothetical examples of what these eight outcomes will look
like. Take some time to go through these. For each of these outcomes, assume that the measurement
scale is 0 - 10. The means for each level of ‘J’ and for each level of ‘K’ are also given, as are the
differences between J1 and J2 to help show any potential interaction.
Outcome 1: No Main effects, No Interaction. Outcom e 1

10
9
Variable J 8
7
6 J1
J1 J2 Kk Difference
DV
5
4 J2
K1 5 5 5 0 3
Variable K 2
K2 5 5 5 0 1
0
Jj 5 5 K1
K
K2
Outcome 2: Main effect of K only. Outcom e 2

10
9
Variable J 8
7
6 J1
J1 J2 Kk Difference
DV
5
4 J2
K1 1 1 1 0 3
Variable K 2
1
K2 9 9 9 0 0
K1 K2
Jj 5 5 K
-252-
Back to TOC
Outcome 3: Main effect of J only Outcom e 3
10
9
Variable J 8
7
J1 J2 Kk Difference 6 J1
DV
5
4 J2
K1 9 1 5 8 3
Variable K 2
K2 9 1 5 8 1
0
Jj 9 1 K1 K2
K
Outcome 4: Main effects of J and K. Outcom e 4

10
9
Variable J 8
7
6 J1
J1 J2 Kk Difference
DV
5
4 J2
K1 5 1 3 4 3
Variable K 2
1
K2 9 5 7 4 0
K1 K2
Jj 7 3 K
Outcome 5: Main effects of J and K with Outcom e 5

interaction. 10
9
Variable J 8
7
6 J1
J1 J2 Kk Difference
DV
5
4 J2
K1 1 1 1 0 3
Variable K 2
1
K2 9 1 5 8 0
K1 K2
Jj 5 1 K
Outcome 6: Main effect of K with interaction. Outcom e 6

10
Variable 9
J 8
7
DV
5
4 J2
K1 1 5 3 -4 3
Variable K 2
1
K2 9 5 7 4 0
K1 K2
Jj 5 5 K
-253-
Back to TOC
Outcome 7: Main effect of J with interaction. Outcom e 7
10
Variable 9
J 8
7
DV
5
4 J2
K1 5 5 5 0 3
Variable K 2
K2 9 1 5 8 1
0
Jj 7 3 K1 K2
K
Outcome 8: Main effects, with interaction. Outcom e 8

10
9
Variable J 8
7
6 J1
J1 J2 Kk Difference
DV
5
4 J2
K1 1 9 5 -8 3
Variable K 2
1
K2 9 1 5 8 0
K1 K2
Jj 5 5 K
20.3 Hypotheses in Factorial Designs

With a factorial design you have a hypothesis for each main effect and each interaction. The null
hypothesis for a main effect is that means between the levels of an independent variable will be equal;
whereas the alternate hypothesis for a main effect states that the means will not be equal. Thus, for the
examples using Variables J and K above, the null and alternate hypotheses:
H0: μJ1 = μJ2 H0: μK1 = μK2

H1: μJ1 ≠ μJ2 H1: μK1 ≠ μK2
For an interaction, the hypotheses are a little trickier. First, you must decide which independent variable is
your focal variable, that is, the variable you believe will produce a direct change in performance on the
dependent variable. The other variable will be your moderator variable, which is the variable you believe
will influence the relationship between the independent variable and the dependent variable. Say that ‘K’
is the moderator variable and ‘J’ is the focal variable. Thus, we would be interested in the difference
between J1 and J2 at each level of K. If the interaction is not significant, the difference between J1 and J2
will be equal at each level of K. If the interaction is significant, the difference between J1 and J2 will differ
at each level of K. Thus, the null hypothesis for the interaction can be stated as:
H0: (μJ1K1 - μJ2K1) = (μJ1K2 - μJ2K2)

The term before the equal sign is the mean difference between J1 and J2 at level K1, and the term after
the equal sign is the difference between J1 and J2 at level K2. The alternate hypothesis is:
H0: (μJ1K1 - μJ2K1) ≠ (μJ1K2 - μJ2K2)
-254-
Back to TOC
Thus, the alternate hypothesis is that the difference between the cell means J1 and J2 will not be equal
across levels of independent variable K.
20.4 The Source Table

In a factorial design, because there is more than one potentially statistically significant effect, we must
calculate an F-Ratio for each main effect and also the interaction. For the most part there is little
difference in the procedures for conducting a factorial ANOVA from the procedures used to conduct a
oneway ANOVA in chapter 19. It is good to create a source table to list each sums of squares, degrees of
freedom, and mean square that will be used in your F-Ratios. A blank source table for a factorial
ANAOVA is below:
Between Groups ?? ?? ?? ??
Variable ‘J’ ?? ?? ?? ??
Variable ‘K’ ?? ?? ?? ??
Interaction (J x K) ?? ?? ?? ??
Within Groups (error) ?? ??
Total Variance ?? ??
You can see that several df’s, several SS’s, several MS’s and several F’s need to be found. I will start
with the df’s and then simply show the formulas for calculating the SS’s, which will be used in the next
section with a numerical example.
The total degrees of freedom (dfT) are equal to nT – 1, where nT is the total number of subjects in the data.
Between group degrees of freedom (df B) are equal to (j)(k) – 1, where j is the number of levels of variable
J and k is the number of levels of variable K. Degrees of freedom for Variable J (df J) are equal to j – 1,
and degrees of freedom for Variable K (dfK) are equal to k – 1. Degrees of freedom for the interaction
(dfJxK) are equal to (j – 1)(k – 1). Finally, within-group degrees of freedom (dfW) are equal to nT – (j)(k).
The sums of squares for each source of variance are:
Sum of squares total (SST):

J K n
SST     Xijk  X
j1k 1 i 1
 
2
Sum of squares within-groups (SSW):

J K n
SSW     Xijk  X jk  2
j1k 1 i 1
Sum of squares between-groups (SSB):

JK

SSB   n jk X jk  X
jk1

2
Sum of squares for the main effect of J (SSJ):

J

SSJ   n j X j  X
j1

2
Sum of squares for the main effect of K (SSK):

K
SSK   n k X k  X
k 1
 
2
Sum of squares for the interaction (SSJxK):

J
SSJ x K    n jk
K
k 1 k 1
X jk  X j  X k  X 
2
-255-
Back to TOC
Importantly, note that that SST = SSB + SSW and that dfT = dfB + dfW. Also, note that SSB = SSJ + SSK +
SSJ x K and that dfB = dfJ + dfK + dfJ x K. Also, the mean square for each source of variance is simply the
ratio of the sum of squares for that source of variance divided by the degrees of freedom for that source
of variance. Finally, the F-Ratio for each main effect and the interaction is calculated by dividing the mean
square for that effect by the mean-square within-subjects, which is always the error term for each F-Ratio.
20.5 Numerical Example of 2 x 2 ANOVA

For this example we’ll expand on the ANOVA example from Chapter 20, where the location in which a
person tried to recall a list of words, relative to where they studied those words, was manipulated
between subjects. A researcher manipulates the location where a person studies and where a person is
tested so that the locations are the Same (J1) or Different (J2). For example, a person studies a list of
words in a classroom and is then tested on their memory for those words when they are in the same
classroom, or in a different classroom. We’ll call this independent variable the Location factor (J). The
researcher also manipulates the time of day the material is studied and tested. The time of day can be the
Same (K1) or Different (K2). We’ll call this the Time factor (K). For this example, we’ll use an alpha level of
α = .05 to assess the statistical significance of each main effect and the interaction.
Thus, there are two independent variables (Location and Time) that are crossed to create four conditions:
(1) Same-Location/Same-Time (J1K1); (2) Same-Location/Different-Time (J1K2); (3) Different-
Location/Same-Time (J2K1); and (4) Different-Location/Different-Time (J2K2). Because we know from
chapter 19 that the location where a person’s memory was tested relative to where they studied
influenced memory, we will assume the focal variable is Location and the moderator variable is Time. The
table below lists the memory scores (number of words correctly recalled) and the summary statistics for
the n = 5 subjects in each group:
Location (J)
Same (J1) Different (J2) Totals
9 5
7 7
10 3 nK1 = 10
Same (K1) 8 4 ∑XK1 = 65
6 6 X K1  6.5
∑XJ1K1 = 40 ∑XJ2K1 = 25
X J1K1  8 X J 2 K1  5
Time (K) 6 2
7 4
9 0 nK2 = 10
Different (K2) 7 2 ∑XK2 = 45
6 2 X K 2  4.5
∑XJ1K2 = 35 ∑XJ2K2 = 10
X J1K 2  7 X J 2K 2  2
nJ1 = 10 nJ2 = 10 nT = 20
Totals ∑XJ1 = 75 ∑XJ2 = 35 ∑X = 110
X J1  7.5 X J 2  3.5 X  5.5
-256-
Back to TOC
First, we'll calculate sum of squares total (SST). The Sum of Squares Total (SST)
formula for the sum of squares total (see below) has you
subtract the grand mean from each individual subject’s Group Xijk 
X ijk  X X ijk  X
2
  
score, square that difference, and the add up those J1K1 9 9 - 5.5 = 3.5 12.25
squared differences. Hence, you are simply calculating J1K1 7 7 - 5.5 =1.5 2.25
the sum of squares over all of the subjects in a set of J1K1 10 10 - 5.5 = 4.5 20.25
data. In the table to the right, each individual’s score is J1K1 8 8 - 5.5 = 2.5 6.25
listed in the second column (the J x K group each subject J1K1 6 6 - 5.5 = .5 0.25
came from is listed in the first column). The grand mean is J1K2 6 6 - 5.5 = .5 0.25
then subtracted from each score in the third column, then J1K2 7 7 - 5.5 = 1.5 2.25
that difference is squared in the fourth column. The J1K2 9 9 - 5.5 = 3.5 12.25
summation of the squared differences is SST at the J1K2 7 7 - 5.5 = 1.5 2.25
bottom (SST = 139). This is the total variation across all n J1K2 6 6 - 5.5 = .5 0.25
= 20 subjects across all four groups. J2K1 5 5 - 5.5 = -.5 0.25
J2K1 7 7 - 5.5 = 1.5 2.25
J K n

SST     Xijk  X
j1k 1 i 1

2 J2K1
J2K1
3
4
3 - 5.5 = -2.5
4 - 5.5 = -1.5
6.25
2.25
J2K1 6 6 - 5.5 = .5 0.25
J2K2 2 2 - 5.5 = -3.5 12.25
J2K2 4 4 - 5.5 = -1.5 2.25
J2K2 0 0 - 5.5 = -5.5 30.25
J2K2 2 2 - 5.5 = -3.5 12.25
J2K2 2 2 - 5.5 =-3.5 12.25
SST = 139
Next, we'll calculate sum of squares within-

Sum of Squares Within Groups (SSW)
X  X jk 
groups (SSW ). The formula (see below) has you
find the difference between an individual Group Xijk X jk ijk X ijk  X jk 
2
subject’s score (Xijk) and the mean of that

J1K1 9 8 9–8=1 1
subject's group ( X jk ). You do this for each
J1K1 7 8 7 – 8 = -1 1
subject’s score, and then square each J1K1 10 8 10 – 8 = 2 4
difference. Finally add the squared differences J1K1 8 8 8–8=0 0
to get SSW . In the table to the right, the second J1K1 6 8 6 – 8 = -2 4
column lists each subject’s score and the third J1K2 6 7 6 – 7 = -1 1
column lists the mean of the group for that J1K2 7 7 7–7=0 0
subject. The fourth Column lists the difference J1K2 9 7 9–7=2 4
between each subject’s score and the mean of J1K2 7 7 7–7=0 0
that subject’s group. The fourth column is the J1K2 6 7 6 – 7 = -1 1
squared differences and at the bottom of that J2K1 5 5 5–5=0 0
column the value of SSW is shown to be 34. This J2K1 7 5 7–5=2 4
value is the total within group variation and is J2K1 3 5 3 – 5 = -2 4
equal to the “total error variance”. J2K1 4 5 4 – 5 = -1 1
J2K1 6 5 6–5=1 1
 2
J K n J2K2 2 2 2–2=0 0
SSW     Xijk  X jk J2K2 4 2 4–2=2 4
j1k 1 i 1 J2K2 0 2 0 – 2 = -2 4
J2K2 2 2 2–2=0 0
J2K2 2 2 2–2=0 0
SSW = 34
-257-
Back to TOC
Next, we'll calculate the total between-group sum of squares (SSB) and the sum of squares for the main
effect of each independent variable (SSJ and SSK), and the sum of squares for the interaction effect (SSJ x
K). The formula for SSB below has you find the difference between each group (cell) mean and the grand
mean and then square the differences. Then, you multiply the squared difference by the number of
subjects in the group (cell), which is n = 5 in this example. Finally, add the products to give you SSB.
JK
SSB   n jk X jk  X
jk1
 
2
The table below lists each J x K group mean from the example above. The third column shows the
difference between each cell mean and the grand mean, the fourth column shows the square of that
difference, and the fifth column shows the product of that squared difference with the sample size of each
group. The value at the bottom (SSB = 105), is the total variance between the four groups.
Sum of Squares Between Groups (SSB)

Group X jk X jk X  X jk X 
2

n jk X jk  X 
2
J1K1 8 8 – 5.5 =2.5 6.25 5(6.25) = 31.25

J1K2 7 7 – 5.5 =1.5 2.25 5(2.25) = 11.25
J2K1 5 5 – 5.5 = -0.5 0.25 5(0.25) = 1.25
J2K2 2 2 – 5.5 =-3.5 12.25 5(12.25) = 61.25
SSB = 105
Next we’ll calculate the sum of squares for the main effect of Location (SSJ). This procedure will be
identical to that for finding the um of squares for the main effect of Time (SSK); the only difference will be
that instead of the mean of each level of J (for SSK), we’ll use the mean for each level of K for SSK. The
formula for SSJ, is:
J
SSJ   n j X j  X
j1
  2
The formula has you find the mean difference between the grand mean and the mean of each level of the
independent variable Location, which we have labeled J. The mean of each level of J is shown in the
second column of the table below. The third column shows the calculation of the mean difference for each
level of ‘J’. The fourth column shows the squared mean difference between the grand mean and each
level of J. You next multiply the squared difference by the number of subjects associated with that level of
‘J’, which will be n = 10 for this example. Finally, you add up those products to get SSJ = 80 for this
example:
Sum of Squares for Main Effect of J (SSJ)

Group Xj X j X  X j X 2

n j Xj  X 2
J1 7.5 7.5 – 5.5 = 2 4 10(4) = 40

J2 3.5 3.5 – 5.5 = -2 4 10(4) = 40
SSJ = 80
-258-
Back to TOC
Next, calculate the sum of squares for the Time factor (SSK). The formula for SSK is functionally
equivalent to that for SSJ:
K
SSK   n k X k  X
k 1
 
2
Like finding SSJ, you first find the mean difference between the grand mean and the mean of each level of
the independent variable Time (K1 and K2), which is shown in the third column in the table below. Next,
square the mean difference and then multiply the squared mean difference by the number of subjects in
that level of K, which is n = 10 in this example. Finally, add the products to get SSK = 80:
Sum of Squares for Main Effect of K (SSK)

Group Xk X k X  X k X 
2

n k Xk  X 2
K1 6.5 6.5 – 5.5 = 1 1 10(1) = 10

K2 4.5 4.5 – 5.5 = -1 1 10(1) = 10
SSK = 20
Finally, we'll calculate the sum of squares for the interaction (SSJ x K), which has the formula:
J K

SSJ x K    n jk X jk  X j  X k  X
k 1 k 1

2
The formula has you take each J x K group mean (each cell mean) and subtract from that value the mean
for the level of variable J associated with the group. For example, group J1K1 has a mean of 8. This group
is part of level J1, which has a mean of 7.5; hence, you subtract that mean (7.5) from 8. Next, subtract the
mean for the level of variable K associated with the group. In the case, group J 1K1 is part of level K1,
which has a mean of 6.5. Next, add the grand mean (5.5) to the value.
In the table below, the mean of each J x K group is listed with the mean of each level of J and K that that
particular group is part of. The value in the fourth column is the value that is obtained after subtracting
the mean of the levels of variable J and variable K that are associated with that particular group, and then
adding the grand mean. The fifth column is the squared value of column 4; and the sixth column is the
product of column 5 and the sample size associated with the J x K group, which is always be n = 5 in this
example. Adding the products in the sixth column gives you SSJ x K = 5 in this example:
Sum of Squares for Interaction (SSJ x K)

Group X jk X j X k X jk  X j  X k  X  X jk  X j  X k  X 
2

n jk X jk  X j  Xk  X 
2
J1K1 8 7.5 6.5 8 - 7.5 - 6.5 + 5.5 = -0.5 0.250 5(.25) = 1.250
J1K2 7 7.5 4.5 7 - 7.5 - 4.5 + 5.5 = 0.5 0.250 5(.25) = 1.250
J2K1 5 3.5 6.5 5 - 3.5 - 6.5 + 5.5 = 0.5 0.250 5(.25) = 1.250
J2K2 2 3.5 4.5 2 - 3.5 - 4.5 + 5.5 = -0.5 0.250 5(.25) = 1.250
SSJxK = 5
Remember that SSB = SSJ + SSK + SSJ x K. Indeed, you can see from the sums of squares that we just
calculated that 105 = 80 + 20 + 5. Also remember that SST = SSB + SSW , and indeed you can see that
139 = 105 + 34.
-259-
Back to TOC
Now that we have the sums of squares and the degrees of freedom for each source of variance, the next
step is to calculate each of the mean squares. Remember, the mean square is simply the sum of
squares for a source of variance divided by the degrees of freedom for that course of variance:
SS
MS 
df
The mean square between-groups (MSB) is not absolutely necessary to calculate, but the mean squares
for both independent variables and the interaction term are necessary. However, strictly speaking, you
should determine whether there is significant between group variability before determining where that
significant influence is coming from.
The ANOVA source table on the top of the next page lists the mean square for the main effects of each
independent variables (MSJ and MSK), the interaction (MSJ x K), and between groups. Remember, all that
you do to calculate the mean squares for each source of variance is divide the sum of squares by its
associated degrees of freedom value.
The degrees of freedoms are:
Degrees of freedom total: dfT = nT - 1 = 20 - 1 = 19

Degrees of freedom within: dfW = nT - jk = 20 - (2)(2) = 20 - 4 = 16
Degrees of freedom between: dfB = jk - 1 = (2)(2) - 1 = 4 - 1 = 3
Degrees of freedom for Location: dfJ = j - 1 = 2 - 1 = 1
Degrees of freedom for Time: dfK = k - 1 = 2 - 1 = 1
Degrees of freedom for Location x Time: dfJ x K = (j - 1)(k - 1) = (2 - 1) (2 - 1) = 1
From the sums of squares and the mean squares, here are the mean squares:
Between Groups 105 3 105/3 = 35 ??
Location (J) 80 1 80/1 = 80 ??
Time (K) 20 1 20/1 = 20 ??
Location x Time (J x K) 5 1 5/1 = 5 ??
Within Groups (error) 34 16 34/16 = 2.125
Total Variance 139 19
The final step is to calculate the necessary F-Ratios for the main effects and for the interaction. Each F-
Ratio is found by dividing the mean square associated with each effect by MS W, which is the the error
term for an F-Ratio.
We have three effects that can be significant (main effect of Location, main effect of Time, and a Location
by Time interaction); thus, we need three F-Value. The F-Ratio for the between-groups source of variance
is not absolutely necessary, but I have included it below for completeness. The F-Ratios are listed in the
ANOVA summary table below:
-260-
Back to TOC
Between Groups 105 3 35 35/2.125 = 16.470
Location (J) 80 1 80 80/2.125 = 37.647
Time (K) 20 1 20 20/2.125 = 9.412
Location x Time (J x K) 5 1 5 5/2.125 = 2.353
Within Groups (error) 34 16 2.125
Total Variance 139 19
To determine whether each main effect and the interaction is statistically significant, we must look up a
separate p-value for each F-Ratio in Table 3 in Appendix A. Because df B = 1, we’ll use Table 3-A, and
because the F-Ratio for the Location factor is greater than 10, we’ll use F = 10 to assess the statistical
significance of the main effect of location. (You do not assess the statistical significance of the overall
between groups variance.) The table below lists the p-values for each main effect and the interaction:
Variance Source SS df MS F p
Location (J) 80 1 80 37.647 .0060
Time (K) 20 1 20 9.412 .0074
Location x Time (J x K) 5 1 5 2.353 .1448
Within Groups (error) 34 16 2.125 ---
Total Variance 139 19 ---
The p-value for each main effect is less than the selected alpha level (.05), so both main effects are
statistically significant. However, the p-value for the interaction (.1448) is greater than the alpha level, so
the interaction is not statistically significant. What does it mean? A significant main effect suggests there
is a significant mean difference between at least two levels for the independent variable. In the present
example, because each variable has only two levels the significant main effect suggests that the mean
difference between the levels of each independent variable is significant. Thus, for the variable ‘Location’
the mean of the ‘Same’ condition (7.5) and the ‘Different’ condition (3.5) significantly differ; and for the
variable ‘Time’ the mean of the ‘Same’ condition (6.5) and the ‘Different’ condition (4.5) significantly differ.
You can see these main effects (and the lack of an interaction), in the graph of the data below:
7
Mean # Words Recalled
5 Same Location (J1)

Different Location (J2)
4
0
Same Time (K1) Different Time (K2)
Time
-261-
20.6 Effect Size and Explained Variance
You can also find the effect size for each main effect and the interaction. Each of these values, below,
represents the proportion of variance explained in the relationship between the independent variable, or
both independent variables in the interaction, with the dependent variable.
SSJ 80
2J    0.576
SST 139
SSK 20
2K    0.144
SST 139
SSJK 5
2JK    0.036
SST 139
-262-
Back to TOC
1. What is a factorial experimental design? What result can be observed in a factorial design that cannot
be observed in a single-factor experimental design?
2. Differentiate between a main effect and an interaction.
3. How many groups are in a 3 x 3 factorial design? How many independent variables?
4. How many groups are in a 3 x 2 x 5 factorial design? How many independent variables?
5. How many groups are in a 2 x 2 x 2 x 2 factorial design? How many independent variables?
6. For each of the following sets of population means, indicate whether there is a main effect of factor A, a
main effect of factor B, and/or an interaction effect:
a. B1 B2 Main Effect of A:
A1 4.00 5.00 Main Effect of B:
A2 4.00 5.00 Interaction Effect:
b. B1 B2 Main Effect of A:
A3 7.00 7.00
c. B1 B2 Main Effect of A:
d. B1 B2 Main Effect of A:
Use the following information to answer #7 – 9: For each of the following sets of population means,
indicate whether there is a main effect of factor A, a main effect of factor B, and/or an interaction effect.
7. Dr. Kruger has developed an herbal supplement that when taken before an exam should reduce
anxiety and improve exam performance. Dr. Kruger pretests 100 students on an test that is designed to
measure 'test anxiety' (Score Range 10-90). He then randomly assigns half of the students to an
"Experimental" group that receives a daily dosage of the supplement and the other half of the students to
a "Control" group that does not receive the supplement. After one month Dr. Kruger retests all the
students on a the anxiety test (posttest). The means are:
Group Pretest Posttest

Experimental 75 40
Control 70 65
a. Is there a main effect of Pretest-Posttest?

b. Is there a main effect of Group?
c. Is there an interaction?
-263-
Back to TOC
8. Dr. Keenan believes that consuming LSD and listening to Pink Floyd helps theoretical physicists
understand the concept "matter is simply energy condensed to a slow steady vibration." To examine this
issue Dr. Keenan samples 200 theoretical physicist and randomly assigns each physicist to one of four
groups created by combining two variables. These four groups differ in whether the theoretical physicists
(a) consume LSD or not, and (b) listen to Pink Floyd or not. In each group, Dr. Keenan explains his
concept and has each theoretical physicist take a test (Score Range: 0-50) that assesses how well they
understand the concept "matter is simply energy condensed to a slow steady vibration." The means of
each group are:
LSD Listening to Pink Floyd Not Listening to Pink Floyd

Consuming 48 38
Not Consuming 35 10
a. Is there a main effect of Listening to Pink Floyd?

b. Is there a main effect of Consuming LSD?
9. Dr. Feely believes that cognitive behavioral therapy works just as good at alleviating an individual's
depression as drug therapy. To examine this, he randomly samples individuals that have just been
diagnosed with moderate depression and have the same score on the Beck Depression Inventory (Score
Range: 0-63), but have not started receiving any treatment. He then randomly assigns each individual to
one of four groups that differ in whether the individuals (a) receive cognitive behavioral therapy or
standard therapy, and (b) receive an antidepressant or not. After two months, he re-measures each
individual's level of depression using the Beck Depression Inventory. The mean Beck Depression
Inventory scores are:
Antidepressant Cognitive Behavioral Therapy Standard Therapy

Taking 10 30
Not Taking 30 50
a. Is there a main effect of Cognitive Behavioral Therapy vs. Standard Therapy?

b. Is there a main effect of Taking the Antidepressant vs. Not Taking the Antidepressant?
10. Below are several incomplete ANOVA summary tables. Fill in the missing terms based on the
information given. Assume n is equal for each group:
a. Source of Variance SS df MS F
Between-Groups 60 20
Main Effect of A 3
Main Effect of B 7
Interaction (A x B) 50
Within Groups 144 36 --
Total -- --
b. Source of Variance SS df MS F
Between-Groups 3 136.667
Main Effect of A 200 200
Main Effect of B 1
Within Groups 10 --
Total 1370 99 -- --
-264-
Back to TOC
c. Source of Variance SS df MS F
Between-Groups 200 3
Main Effect of A
Main Effect of B 50
Within Groups 10 --
Total 47 -- --
11. Below are several incomplete ANOVA summary tables. Fill in the missing terms based on the
information given. Assume n is equal for each group:
Between 39 3.9
Main Effect of X 100 1
Main Effect of Y 2
Interaction (X x Y) 1 1.5
Within Groups 360 36 10 ---
Total --- ---
Between 151
Variable X 0.04
Variable Y 50
Interaction 4
Within 76 25 ---
Total 2051 79 --- ---
12. Determine the p-value for the F-Ratios for each main effect and the interaction in #10 a – c.
a. Main Effect of A : Main Effect of B: Interaction:

b. Main Effect of A : Main Effect of B: Interaction:
c. Main Effect of A : Main Effect of B: Interaction:
13. Determine the p-value for the F-Ratios for each main effect and the interaction in #11 a – b.
a. Main Effect of X : Main Effect of Y: Interaction:

b. Main Effect of X : Main Effect of Y: Interaction:
-265-
Back to TOC
14. Use the following to answer the questions that follow: An investigator tested the relationship between
perceived task difficulty and time limits on performance. Twenty students worked on the same verbal
analogy task, but half the students were told that the task was of low difficulty, while the other half were
told that that the task was of high difficulty. Half of the students in each of these two groups were given
five minutes to complete the task (time limit), and the other half of the students in each difficulty level
group were given unlimited time (no limit). Thus, there were four independent groups. Scores on the task
could range from 0 to 10, with higher values indicating better task performance. The data are below:
Task Difficulty (J)

Low (J1) High (J2)
9 6
8 7
No Limit (K1) 10 5
7 8
6 4
Time Limit (K)
7 2
6 3
Limit (K2) 4 4
8 3
5 3
a. Calculate the mean for each of the four groups, then calculate the mean for each level of each
independent variable, and then calculate the grand mean.
b. Calculate each degrees of freedom value.
c. Calculate the sum of squares between groups, the sum of squares within groups and the sum of
squares total. Then calculate the sum of squares for each main effect and for the interaction.
d. Calculate each mean square.
e. Calculate each F-Ratio.
f. Determine the p-value for each main effect and the interaction.
g. Based on the p-values and assuming α = .05, what decisions should be made with respect to
each main effect and the interaction? What should the investigator conclude with respect to the
relationship between task difficulty, time limits, and task performance?
15. Use the following to answer the questions that follow: Scores (0 - 10) were obtained from each of four
groups that were created by combining the levels of two independent variables. The data are below:
Variable J
J1 J2
9 1
K1 8 3
10 2
Variable K
3 9
K2 2 8
4 7
a. Calculate the mean for each of the four J x K groups. Calculate the mean for each level of each
independent variable, and then calculate the grand mean.
b. Calculate each degrees of freedom value.
c. Calculate the sum of squares between groups, the sum of squares within groups and the sum of
squares total. Then calculate the sum of squares for each main effect and for the interaction.
d. Calculate each mean square.
e. Calculate each F-Ratio.
f. Determine the p-value for each main effect and the interaction.
g. Based on each p-value and assuming α = .05, what decision should be made with respect to
each main effect and the interaction?
-266-
Back to TOC
Chapter 21: Chi-Square Analyses
21.1 When, Why, Where is Chi Square Used?
2
Chi-square (χ ) analyses are used to examine
whether observed data fits with expected data. More
specifically, does an observed frequency distribution
‘fit’ with an expected frequency distribution. As a
simple example, look at the picture to the right. It is a
snapshot of the television network C-Span at about
the end of a US Senate vote on the Economic
Stabilization act. Assuming 100 US Senators can
vote, and assuming they can vote ‘Yea’ or Nay on
the Act: how many Senators would vote Yea and
vote Nay, by chance? If Senators voted Yea and
Nay randomly, you would expect, by chance, 50
Yea’s and 50 Nay’s. This is the expected frequency
distribution of Yea’s and Nay’s. Clearly though, there
is a difference in the actual (observed) number of
Yea’s (74) and Nay’s (25), in the picture above. What a chi-square analysis does is determine whether
the observed frequency distribution is significantly different from the expected frequency distribution.
As a simpler, and non-political example: by chance, you would expect that after flipping a coin 10 times,
you will observe five heads and five tails. What if you observed nine heads and one tail, would this
outcome differ from what is expected by chance? Does this mean that the coin is biased? The chi-square
test can help answer this, that is, whether the observed frequencies of heads and tails in the coin-flipping
example differ from the frequencies of heads and tails that would be expected to be observed by chance.
In the behavioral sciences, chi-square analyses are often used to evaluate relationships between
independent variables that are on a nominal scale and when the collected data are frequencies. Thus,
chi-square analyses can be used to evaluate whether frequencies that were observed in a study differ
from frequencies that one expects to obtain in a study. For example, a chi-square test can be used to
assess whether the numbers of freshmen, sophomores, juniors, and seniors that are enrolled in a course
differ from the numbers of each college class that you would expect to be taking the class. For example,
in my spring 2011 Cognitive Psychology class, I had 4 freshmen, 15 sophomores, 7 juniors and 8 seniors
enrolled. Is this distribution of college classes different from what I should expect? Why are there not
equal numbers of students from each college class? A chi-square test can be used to help evaluate this.
If I find that the observed frequencies differ from what is expected, I might conclude that there is
something about this course that influences students of a certain college class to take the course.
Generally, chi-square is used to assess the relationship between variables when
The variable, or variables, are qualitative in nature (nominal scale)

If there are two variables, the same subjects have been measured on each variable
The observations on the variable, or variables, are between subjects
Each level of each independent variable is combined with each level of the second independent variable
21.2 Assumptions of the Null Hypothesis

Under the null hypothesis, one would assume that a set of observed frequencies (fo) will be equal to the
expected frequencies (fe). That is, if I expect to observe five freshmen in a class, then I should observe
five freshmen in that class. Thus, the null hypothesis predicts no difference between the observed
-267-
Back to TOC
frequencies and the expected frequencies, such that the expected chi-square value of under the null
hypothesis is zero. Hence, the null hypothesis is:
H0: f0 = fe
The alternate hypothesis predicts that the observed and the expected frequencies will not be equal:
H1: f0 ≠ fe
There is no such thing as directionality of the alternate hypothesis for chi-square analyses.
21.3 Chi Square with One Independent Variable

Before discussing the chi-square test I must describe several assumptions. First, chi-square tests assume
that the dependent variable is frequency data; hence, numbers of observed cases in a given category.
Second the levels of each independent variable come from a nominal scale. In the Cognitive Psychology
example above, the frequencies of freshmen, sophomores, juniors, and seniors represent groups that
differ along a nominal variable (college class). Third, levels of each independent variable must be
mutually exclusive; that is, a person, or whatever is being tallied, can belong to only one level for each
independent variable.
The example below will deal with a chi-square analysis for a single independent variable. Note that the
independent variables in each example in this chapter are not manipulated; rather, the independent
variable represents groups that differ along some naturally occurring dimension. We will use the chi-
square test to determine whether the frequencies observed across the categories of one variable differ
from what are expected. Because there is only one variable, this type of a design is a oneway design;
thus, the test is a oneway chi-square.
Say I am interested in the distribution of college majors in my Sensation & Perception class. I examine my
class’ roster to determine the major for each of the n = 39 students in my class. I find that there are 17
psychology majors, 3 counseling and human service majors, 3 neuroscience majors, 6 occupational
therapy majors, 4 communications majors, and 6 students falling into some ‘other’ major (history, English,
etc.). Thus, I classified the students into six mutually exclusive groups based on major. Hence, I have one
independent variable (College Major), with six levels. The observed frequencies for each major are:
Major
Counseling & Occupational
Psychology Neuroscience Communications Other
Human Services Therapy
fo = 17 fo = 3 fo = 3 fo = 6 fo = 4 fo = 6
Remember, chi-square analyses will compare a distribution of observed frequencies (fo) to a distribution
of expected frequencies (fe). The table above lists the observed frequencies, so the expected frequencies
need to be determined.
I have n = 39 students in this class and there are six different majors. If choice of college major did not
determine whether a student took this course, I would expect there equal numbers of students in each of
the six observed majors. Because there are k = 6 majors (‘k’ is the number levels of the independent
variable), I would expect there to be 39/6 = 6.5 students in each college major that are in this course. The
-268-
Back to TOC
formula for figuring out the expected frequencies for each group in a one-way chi-square can be
expressed as:
n
fe  T
k
Where nT is the total number of subjects, and ‘k’ is the total number of levels of the independent variable.
In a one-way chi-square, fe will be the same for each group. The table below lists the observed and
expected frequencies for each group:
Major
Counseling & Occupational
Psychology Neuroscience Communications Other
Human Services Therapy
fo = 17 fo = 3 fo = 3 fo = 6 fo = 4 fo = 6
fe = 6.5 fe = 6.5 fe = 6.5 fe = 6.5 fe = 6.5 fe = 6.5
Once the distribution of expected frequencies has been determined, a chi-square analysis can be
conducted. The formula for the chi-square analysis compares each observed frequency to each expected
frequency for each level of the independent variable. Remember, if there is no difference between the
expected and observed frequencies, then the value of the chi-square should be equal to 0. The the
farther the chi-square value is from zero the more likely there is a significant difference between the
observed and the expected frequencies. The formula for the chi-square test is:
 f  f 2 
   o e 
2
 fe 
 
First, you first subtract the expected frequency for a level of the independent variable from the observed
frequency for that level. You then square the difference and then divide the squared difference by the
expected frequency, before finally adding the quotients. This is done in the table below:
2 2
Group fo fe (fo - fe) (fo - fe) (fo - fe) /fe
Psychology 17 6.5 10.5 110.25 16.962
Counseling & Human Services 3 6.5 -3.5 12.25 1.885
Neuroscience 3 6.5 -3.5 12.25 1.885
Occupational Therapy 6 6.5 -0.5 0.25 0.038
Communications 4 6.5 -2.5 6.25 0.962
Other 6 6.5 -0.5 0.25 0.038
2  21.770
To assess the statistical significance of the outcome of a chi-square test, you first calculate the degrees of
freedom, which are equal to df = k – 1, where k is the number of levels of the independent variable. In this
case there are six categories, so df = 6 – 1 = 5. Assume we select an alpha-level of α = .05. Probabilities
Back to TOC
under the chi-square distribution are presented in Table 5 in Appendix A, a portion of which appears on
the next page. Table 5 is set up similarly to the t-Tables (Table 2), where the test statistic values are listed
in the leftmost column and the degree of freedom values are listed in the column headings.
-269-
In Table 5 locate the column associated with df = 5 in the column headings (highlighted in yellow) and
scroll down that column until you locate the row associated with the obtained chi-square value (21.77) or
the closest chi-square value that is less than the observed value (21.50, highlighted in yellow). IN this
2
example, the p-value associated with χ = 21.77 is p = .0007, which is less than the selected alpha level
of .05; hence, the null hypothesis can be rejected and the alternate hypothesis accepted. I conclude there
is a significant difference between the distribution of observed frequencies from the distribution of
expected frequencies of college majors in my Sensation & Perception class. This means is that there may
be some relationship between this course and the types students, based on college major, who take the
course. Specifically, certain majors may be more likely to take this particular course.
Degrees of Freedom (df)
χ2 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 25 30
20.00 0.0000 0.0000 0.0002 0.0005 0.0012 0.0028 0.0056 0.0103 0.0179 0.0293 0.0453 0.0671 0.0952 0.1301 0.1719 0.2202 0.2742 0.3328 0.3946 0.4579 0.7468 0.9165
20.50 0.0000 0.0000 0.0001 0.0004 0.0010 0.0023 0.0046 0.0086 0.0151 0.0249 0.0389 0.0582 0.0834 0.1151 0.1536 0.1985 0.2495 0.3054 0.3651 0.4271 0.7201 0.9029
21.00 0.0000 0.0000 0.0001 0.0003 0.0008 0.0018 0.0038 0.0071 0.0127 0.0211 0.0334 0.0504 0.0729 0.1016 0.1368 0.1785 0.2263 0.2794 0.3368 0.3971 0.6926 0.8879
21.50 0.0000 0.0000 0.0001 0.0003 0.0007 0.0015 0.0031 0.0059 0.0106 0.0179 0.0285 0.0435 0.0636 0.0895 0.1216 0.1601 0.2047 0.2549 0.3098 0.3682 0.6644 0.8716
22.00 0.0000 0.0000 0.0001 0.0002 0.0005 0.0012 0.0025 0.0049 0.0089 0.0151 0.0244 0.0375 0.0554 0.0786 0.1078 0.1432 0.1847 0.2320 0.2843 0.3405 0.6357 0.8540
22.50 0.0000 0.0000 0.0001 0.0002 0.0004 0.0010 0.0021 0.0041 0.0074 0.0128 0.0208 0.0323 0.0481 0.0689 0.0953 0.1278 0.1662 0.2105 0.2601 0.3140 0.6067 0.8352
23.00 0.0000 0.0000 0.0000 0.0001 0.0003 0.0008 0.0017 0.0034 0.0062 0.0107 0.0177 0.0277 0.0417 0.0603 0.0841 0.1137 0.1493 0.1906 0.2373 0.2888 0.5776 0.8153
23.50 0.0000 0.0000 0.0000 0.0001 0.0003 0.0006 0.0014 0.0028 0.0052 0.0090 0.0150 0.0238 0.0361 0.0526 0.0741 0.1010 0.1337 0.1721 0.2160 0.2649 0.5484 0.7942
24.00 0.0000 0.0000 0.0000 0.0001 0.0002 0.0005 0.0011 0.0023 0.0043 0.0076 0.0127 0.0203 0.0311 0.0458 0.0651 0.0895 0.1194 0.1550 0.1962 0.2424 0.5194 0.7720
24.50 0.0000 0.0000 0.0000 0.0001 0.0002 0.0004 0.0009 0.0019 0.0036 0.0064 0.0108 0.0174 0.0268 0.0398 0.0571 0.0791 0.1065 0.1393 0.1777 0.2212 0.4907 0.7489
25.00 0.0000 0.0000 0.0000 0.0001 0.0001 0.0003 0.0008 0.0016 0.0030 0.0053 0.0091 0.0148 0.0231 0.0346 0.0499 0.0698 0.0947 0.1249 0.1605 0.2014 0.4624 0.7250
One thing that you will notice is that there are no means to compare across levels of the independent
variable. Indeed, this means that the chi-square analyses are a type of non-parametric statistic. So,
how does one perform a post-hoc analysis once the chi-square analysis has established a significant
difference between the observed and expected frequencies? There are a number of post-hoc analyses
that can be performed, but they are beyond the scope of this chapter. Instead, I will describe how you can
go about interpreting the results of the chi-square analysis.
You can describe the difference between the observed frequencies and the expected frequencies by
2
examining the (fo - fe) /fe values and the fo - fe values for each level of the independent variable in the table
2
above. Specifically, because the (fo - fe) /fe values are summed to give you the chi-square statistic, larger
2
(fo - fe) /fe values will contribute more to the rejection of the null hypothesis. What you do, then, is look for
2
the largest values in the (fo - fe) /fe column, and then use the corresponding fo - fe values to determine how
that level of the independent variable contributed to the significant relationship between the independent
variable and the observed frequencies
2
The largest (fo – fe) /fe value in that column is for Psychology majors (16.962), while the other values are
all quite small (Range = 0.038 – 1.885). Examination of the corresponding fo – fe value indicates that
psychology majors are more likely to take this class than expected (fo – fe = 10.5). Although much smaller,
2
the (fo – fe) /fe value for Neuroscience and Counseling & Human Services (1.885) also indicate a
contribution to the rejection of the null hypothesis. Examination of the corresponding fo - fe value indicates
that both Neuroscience and Counseling & Human Services majors are less likely to take this class than
expected (fo – fe = -3.5). The point is that you should examining the differences between what is expected
and what is observed after finding a significant chi-square statistic, in order to decipher the relationship.
21.4 Goodness of Fit Test

The chi-square goodness of fit test is similar to the oneway chi-square test from Section 21.3. The
goodness of fit test is used to determine whether the distribution of groups in a sample is the same as the
distribution of those groups in a population. That is, we know something about how many people, or the
relative frequencies of people, within various groups in a population. We use the goodness of fit test to
determine whether, in a sample, there is a significant deviation of those expected frequencies in the
population. Stated differently, the goodness of fit test is used to determine whether the relative
frequencies (proportions) of groups in a population are proportional in a sample.
-270-
For example, the relative frequencies (proportions) of females and males living in the United States is
about p(Male) = .49 and p(Female) = .51. Say that I need a representative sample of n = 100 subjects to
take survey. If I want this sample to represent the United States’ population of males and females, I need
to have 49 males and 51 females, based on the proportions so males and females in the population (i.e.,
nMales = .49 x 100 = 49; and nFemales = .51 x 100 = 51). What if I end up with 55 females and 45 males? Is
this a significant deviation from the numbers that I should expect to find in my sample, based on the
population parameters? The chi-square goodness of fit test can answer this.
As an example, it is estimated that the proportion of left-handed people in the United States is p(Left
Handed) = .10, and the proportion of right-handed people is p(Right Handed) = .90. We want to know
whether these proportions of left-handed and right-handed people hold for all of the individuals who have
held the office of President of the United States. That is, over all of the Presidents of the United States,
does the proportion of left-handed Presidents and right-handed Presidents deviates from the population
proportions?
First, we need to know how many people has been the President of the United States, and then how
many were left-handed and how many were right-handed. In all, there have been n = 44 Presidents of the
United States, including our current President Barack Obama. Of these n = 44 Presidents, there have
been eight left-handed Presidents (Barack Obama, Bill Clinton, George Bush Sr., Ronald Reagan, Gerald
Ford, Harry Truman, Herbert Hoover, and James Garfield. Each of the other 36 Presidents of the United
States was right handed.
Next, we determine the expected frequencies (f e) of left-handed and right-handed Presidents, based on
the population proportions of left- and right-handed people and the total number of Presidents (44). The
expected frequencies for the goodness of fit test can be found by the following equation:
fe = (p)(n)
In this equation, p is the relative frequency (proportion) associated with group ‘i’ in the population. In this
example, the p value will be the proportion of left-handed people in the United States (.10) and the
proportion of right-handed people in the United States (.90). The n is the number of subjects in the
sample. In this case the sample size is the n = 44 Presidents of the United States. Thus, to obtain the
expected frequencies, simply multiply the sample size (n) by the relative frequency for each group (p):
Expected Number of Left-Handed Presidents: fe = (.10)(44) = 4.4

Expected Number of Right-Handed Presidents: fe = (.90)(44) = 39.6
Next, we determine the critical chi-square value. As with the oneway chi-square analysis, the degrees of
freedom in the goodness of fit test is equal to k - 1, where ‘k’ is the number of groups in the sample. In our
example, we have two groups in the sample: left-handed and right-handed Presidents. Thus, df = 2 – 1 =
1 and I will select an alpha level of α = .05. The formula for calculating the chi-square statistic is the same
as that used in the oneway chi-square in Section 21.3:
 f  f 2 
2    o e 
 fe 
 
As before, you subtract the expected frequency for group from the observed frequency for that group.
You then square the difference and then divide the squared difference by the expected frequency for that
group, before finally adding the quotients. This is done in the table below:
-271-
Back to TOC
2 2
Left-Handed Presidents 8 4.4 -3.6 12.96 2.945
Right-Handed Presidents 36 39.6 3.6 12.96 0.327
 2  3.272
2
From Table 5, the p-value associated with χ = 3.272 is p = .0736, which is greater than the selected
alpha-level (.05); hence, we retain the null hypothesis and make no decision on the alternate hypothesis.
Thus, the observed frequencies of left-handed and right-handed Presidents of the United States do not
significantly deviate from the relative frequencies of left-handed and right-handed people in the population
of the United States. Surprisingly, this is a good thing! It means that the groups within a population are not
over or underrepresented in your sample.
21.5 Chi Square with Two Independent Variables

Chi-square analyses can also be used to examine whether a distribution of observed frequencies differs
from a distribution of expected frequencies among the combinations of two (or more) independent
variables. For example, in the previous section, say I counted the number of males and the number of
females in each of the six college majors. I would have two nominal independent variables, College
Major, and Sex, with a total of 6 (College Major) x 2 (Sex) = 12 mutually exclusive groups. I could count
the number of students belonging to each of these 12 groups, and then determine the expected number
of students who should belong to each of these groups, in order to perform an chi-square analysis.
When you have two nominal independent variables and the dependent variable is frequency, you perform
a factorial chi-square to analyze the frequency data. This type of chi-square analysis will test for the
presence of a statistically significant relationship between the two independent variables. For example,
from the preceding paragraph, if a chi-square analysis performed on that set of observed frequencies was
found to be statistically significant, it suggests that there is a relationship between a student's sex and
their choice of college major that leads a student to enroll in Sensation & Perception.
When examining the relationship between two nominal independent variables and when the dependent
variable is frequencies, you start by setting up a contingency table that lists all of the observed
frequencies for each of the combinations created by the independent variables. Recall from Chapter 10
(Probability), a contingency table lists two independent variables, one represented by the rows and the
other represented by the columns in the table. The table lists the observed frequency associated with
each combination of the two independent variables. From such a contingency table, we will calculate the
expected frequency for each combination of variables and then perform a chi-square analysis. Using the
sex and major example from above, the contingency table might look the one below:
Major
Counseling & Occupational Row
Sex Psychology Neuroscience Communications Other
Human Services Therapy Totals
Male fo = 4 fo = 0 fo = 2 fo = 0 fo = 2 fo = 3 11
Female fo = 13 fo = 3 fo = 1 fo = 6 fo = 2 fo = 3 28
Column
17 3 3 6 4 6 nT = 39
Totals
The row totals and the column totals are marginal frequencies. It is important to list these in the
contingency table, because they are used to calculate the expected frequencies for each Sex by Major
combination. A factorial chi-square can take one of two forms; however, both use exactly the same
-272-
Back to TOC
procedures and have the same assumptions. The only difference is how the frequency data is collected
and how results are interpreted.
A chi-square test for independence is where the marginal frequencies vary independently, that is, the
marginal frequencies for both independent variables are not fixed, and are unknown prior to collecting
data. A chi-square test for independence is used to determine whether there is a relationship between the
independent variables. The example above is a chi-square test for independence, because there is no
way to know the marginal frequencies for either independent variable before the course had started.
A chi-square test for homogeneity of variance is where the marginal frequencies of only
one independent variable are determined before data is collected (marginal frequencies for both
independent variables cannot be fixed). For example, I may be interested in the choice of college major
for male and female students that get their coffee at Starbucks. I could wait outside the Starbucks in the
student center and ask the first 50 males and the first 50 females about their choice of major. In this
example, I decided on the marginal frequencies of males and females before collecting data; that is, I
decide to collect data from only 50 males and only 50 females. In this case, the marginal frequencies of
males and females are fixed and cannot vary. A chi-square test for homogeneity of variance evaluates
whether the difference between the observed frequencies and expected frequencies are equal
(homogeneous) for males and females. For present purposes I cover only the chi-square test for
independence. Just note that the testing procedures are exactly the same, the only difference is how the
data is being collected and interpreted.
A factorial chi-square analysis has the same assumptions as the oneway chi-square: (1) The dependent
variable is frequency; (2) the independent variables are on a nominal scale, and (3) the levels of the
independent variables are mutually exclusive. A factorial chi-square has one additional assumption: the
frequency of each group is at least five. Unfortunately, the frequencies of several groups in the sex and
major example are less than five, thus, the example violates this important assumption.
A good example of where the chi-square test could be used to assess frequency data is congressional
voting records from the US House of Representatives. Specifically, chi-square tests can examine the
relationship between political party and voting (“Yea” and “Nay”) on a particular bill. Such a situation has
two independent variables, Political Party and Voting Decision, and the dependent is the frequency of
responses in each combination of political party and voting decision.
One particular bill was House of Representatives Bill HR6, the Energy Independence and Security Act
th
which passed in the House on December 18 , 2007. The bill set to “reduce our Nation's dependency on
foreign oil by investing in clean, renewable, alternative energy resources, promoting new energy
technologies, developing greater efficiency, and creating a Strategic Energy Efficiency and Renewable
Reserve to invest in alternative energy, and for other purposes.” Below, is a contingency table that lists
the frequency for each combination of political party (Democrats and Republicans) and Decision on the
bill (Yea, Nay, Abstain):
Political Party
Voting Decision Democrats Republicans Row Totals (RMF)
Yea f0 = 219 f0 = 95 314
Nay f0 = 4 f0 = 96 100
Abstain f0 = 9 f0 = 10 19
Column Totals (CMF) 232 201 nT = 433
Note that nT should be equal to the sum of the row marginal frequencies ( 314 + 100 + 19 = 433) and also
be equal to the sum of the column marginal frequencies (232 + 201 = 433). I admit that the observed
frequency of Democrats voting “Nay” is less than five; hence, the assumption that each cell frequency is
at least five is violated, but we'll ignore that for now, because 4 is close to 5.
-273-
Back to TOC
The first step is to determine the critical chi-square value. Degrees of freedom in a factorial chi-square are
equal to df = (c – 1)(r – 1), where 'c' is the number of columns in the contingency table (number of levels
for the column independent variable) and 'r' is the number of rows in the contingency table (number of
levels for the row independent variable). Here, Political Party (column variable) has c = 2 two levels and
Voting Decision (row variable) has r = 3 levels. Thus, there are (2 – 1)(3 – 1) = (1)(2) = 2 degrees of
freedom and I will use an alpha-level of α = .05.
The next step is to calculate the expected frequency for each combination of the independent variables.
In the case of the factorial chi-square analysis, a single fe that will be equal for all cells will not suffice;
rather, a different fe value is necessary for each cell. Each fe value will be based on the row marginal
frequency (RMF) and column marginal frequency (CMF) associated with a particular cell (group). The
expected frequency of any cell can be calculated using the following equation:
fe 
CMFRMF
nT
RMF and CMF refer to the row marginal frequency and column marginal frequency associated with a
particular group, respectively. For example, the observed frequency of Democrats voting Yea is f o = 219.
The row marginal frequency that is associated with that cell in the table is RMF = 314, and the column
marginal frequency associated with that cell is CMF = 232. That is, to figure out the expected frequency of
Democrats voting Yea, you locate that cell in the contingency table, and then use the CMF of that cell’s
column and the RMF of that cell’s row in the formula above. Thus, using the equation above, the
expected frequency for Democrats voting Yea is:
fe 
CMFRMF  232314  72848  168.240
nT 433 433
If the independent variables Political Party and Voting Decision were unrelated, we would expect about
168 Democrats to vote Yea on this bill. The calculations for each additional expected frequency for each
cell in this example are calculated below:
Democrats-Nay: fe 
232100  23200  53.580
433 433
Democrats-Abstain: fe 
23219  4408  10.180
433 433
Republicans-Yea: fe 
201314  63114  145.760
433 433
Republicans-Nay: fe 
201100  20100  46.420
433 433
Republicans-Abstain: fe 
20119  3819  8.820
433 433
In the table below, the expected frequencies are listed with the observed frequency for each combination
of the two independent variables. Although the marginal frequencies are not listed, note that within
rounding error the expected frequencies in a row sum to the row marginal frequency and the expected
frequencies in a column sum to the column marginal frequency:
-274-
Back to TOC
Political Party
Voting Decision Democrats Republicans
fo = 219 fo = 95
Yea
fe = 168.240 fe = 145.760
fo = 4 fo = 96
Nay
fe = 53.580 fe = 46.420
fo = 9 fo = 10
Abstain
fe = 10.180 fe = 8.820
Once the expected frequencies have been obtained the next step is to perform the chi-square test: The
formula for doing so and the procedure is identical to the chi-square test for a one-way design. With a
factorial chi square, each expected frequency is compared to each observed frequency. The chi-square
procedure is demonstrated in the table below:
2 2
Democrats/Yea 219 168.240 50.760 2576.578 15.315
Democrats/Nay 4 53.580 -49.580 2458.176 45.879
Democrats/Abstain 9 10.180 -1.180 1.392 0.137
Republican/Yea 95 145.760 -50.760 2576.578 17.677
Republican/Nay 96 46.420 49.580 2458.176 52.955
Republican/Abstain 10 8.820 1.180 1.392 0.158
 2  132.121
2
From Table 5, the p-value associated with χ = 132.121 (use 40.00 in Table 5) is p < .0001, which is less
than the selected alpha-level (.05); hence the null hypothesis is rejected and the alternate hypothesis is
accepted. Thus, we would conclude that for this particular House Bill there is a significant relationship
between a congresspersons political party and voting decision. As before, we can describe the
2
relationship between the independent variables by examining first the (fo - fe) /fe values and then the fo - fe
2
values, for each group. The two largest (fo - fe) /fe values are Democrats Voting Nay (45.879) and
Republicans Voting Nay (52.955). Examining the fo - fe values, Democrats were much less likely to vote
Nay on this bill (-49.580); whereas Republicans were much more likely to vote Nay on the Bill (49.850).
2
The next two largest (fo - fe) /fe values are Democrats Voting Yea (15.315) and Republicans Voting Yea
(17.677). Examining the fo - fe values, Democrats were much more likely to vote Yea on this bill (50.760);
whereas Republicans were much less likely to vote Yea on the Bill (-50.760).
21.6 Strength and Power of Chi Square

The strength of the relationship between the two nominal variables that were analyzed with a chi-square
test can be estimated by Cramer’s Contingency Coefficient (C). This value is similar to eta-squared
2
(η ), the proportion of variance explained in a relationship between variables. Cramer’s C is used only
when you have a two-way chi-square and the number of levels of at least one variable is greater than
two. Cramer’s C is calculated from:
2
C
n T  2
The chi-square values in the expression are the obtained chi-square statistics. The strength of the
relationship for the example in the preceding section is:
-275-
Back to TOC
2 132.121 132.121
C    0.234  0.484
n T  2 433  132.121 565.121
Thus, the amount of variance that is explained in the relationship between political party and voting
decision on this Bill is about 48.4%.
A more standardized measure of effect size is Cohen’s w. which can be calculated from Cramer’s C:
C2
w
1  C2
In this example, Cohen’s w is equal to:
0.484 2 0.234 0.234

w    0.305  0.552
1  0.484 2 1  0.234 0.766
Cohen provides descriptive labels for this effect size:
Effect Size Cohen’s w

“Small” .10
“Medium” .30
“Large” .50
In this case, we have a “large” effect size. Cramer’s w is also useful for determining the statistical power
2
that was observed in a chi-square test. Using G*Power 3, under Test Family select “χ tests,” and under
Type of Power Analysis select “Post hoc:...”. Enter the information from the chi-square test into the
appropriate areas under Input Parameters (w = .0552, α = .05, Total sample size = 433, df = 2). G*Power
tells us that Power is equal to 1.000, which simply means that the probability that we correctly rejected a
false null hypothesis is greater than 0.999.
21.7 Special Case of the 2x2 Chi-Square

The chi-square test is not terribly difficult to perform; however, it can be time consuming. There is a
special case where a computational formula for the chi-square test can be used. When you have a
factorial design and there are only two independent variables and each independent variable has only
two-levels, you can use the following computational formula, which is described below:
n ad  bc2
2 
a  bc  d a  cb  d 
Note, that this can be used only if you have a 2 x 2 design; if you have anything larger than a 2x2 design,
you must use the procedures discussed earlier. The letters, a – d, in the formula refer to specific cells
within a 2 x 2 contingency table that is formed between the two independent variables. The cells
associated with each letter are listed in the table below. The ‘n’ is total frequency:
-276-
Back to TOC
Independent Variable “A”
Independent Variable “B” A1 A2
B1 a b
B2 c d
As an example, say that we count the number of male republicans, male democrats, female republicans,
and female democrats in a political science course and obtain the frequencies in the table below:
Political Party
Sex Democratic Republican
Males fo = 30 fo = 30
Females fo = 20 fo = 10
In this example there are n = 100 students taking the course and there are two independent variables
(Sex and Political Party), each with two levels. This is a 2 x 2 factorial design. You could use the
procedures that were developed in section 21.5 to perform a chi-square analysis on these data, or, you
could use the computational formula. Plugging in the values associated with each cell in the example
above into the 2 x 2 chi-square expression, we have:
10030  10  30  202
2 
30  3020  1030  2030  10
Once you have all of the values plugged into the equation the rest is basically just appropriate application
of the order of operations:
10030  10  30  20 2
2 
30  3020  1030  2030  10
100300  600 2
2 
60305040
100 300 2
2 
60305040
10090000 
2 
60305040
9000000
2 
3600000
 2  2.5
Because there are two rows and two columns in the 2 x 2 design there is only 1 degree of freedom. From
2
Table 5, the p-value associated with χ = 2.5 is p = .1138, which is greater than the selected alpha-level
(.05); hence the null hypothesis is retained and we conclude there is not enough evidence to suggest a
relationship between sex and political party among students taking a political science course.
-277-
Back to TOC
Effect size (strength of the relationship) can also be measured for a 2 x 2 chi square, but instead of using
Cramer’s C when the design is 2 x 2 the phi-coefficient (Φ) should be used. The phi-coefficient is
calculated by taking the square root of the quotient of the obtained chi-square over the total frequency:
2 2.5
   0.025  0.158
n 100
Incidentally, the phi coefficient is equal to Cohen’s w; hence, you can use this to determine statistical
power by inputting this, df = 1, α = .01, and n = 100 into G*Power. In this case, power equal to only 0.352;
hence, we need additional data before we can conclude that there is no statistical; association
-278-
Back to TOC
1. When is a chi-square test typically used to analyze a bivariate relationship?
2. What is the difference between the chi-square test of independence and the chi-square test of
homogeneity?
3. Describe the difference between the observed frequencies and the expected frequencies.
2
4. State the critical value of χ for a chi-square test for an alpha level of α = .05 and also of α = .01 under
each of the following conditions:
a. k=2
b. k=3
c. k=4
d. r = 2, c = 3
e. r = 3, c = 3
f. r = 2, c = 4
g. r = 4, c = 4
5. Determine the p-value for each of the following given scenarios:

2
a. k = 2, χ = 4.50
2
b. k = 2, χ = 2.60
2
c. k = 6, χ = 11.00
2
d. k = 6, χ = 9.00
2
e. r = 2, c = 2, χ = 3.50
2
f. r = 3, c = 3, χ = 12.20
2
g. r = 2, c = 4, χ = 6.40
6. Use the following to answer the questions below: A psychology professor is interested in the numbers
of students in each college class (freshmen, sophomore, junior, senior) enrolled in his Fundamentals of
Psychology lecture and wants to know whether there is significant difference across the numbers of
students in each a college class. The professor records the class of each student who is enrolled in his
Fundamentals of Psychology course. The data follows.
College Class
Freshmen Sophomore Junior Senior
fo = 162 fo = 39 fo = 6 fo = 5
a. Calculate the expected frequency of each college class.

b. Perform a chi-square test on the data.
c. What is the p-value for this test statistic?
d. Assuming α = .05, is there a statistically significant difference across the four college classes, that
is, is there is difference between the observed and expected frequencies? What decisions are
made with the hypotheses?
2
e. Use the (fo - fe) /fe and (fo - fe) values to discern the significant result.
7. Use the following to answer the questions below: You are interested in the mating habits of baboons.
You find a baboon troop with eight males of breeding age and eighty females of breeding age. You
assume each male has equal mating rights to the females; that is, each male baboon will mate with an
equal number of female baboons. You observe the baboon troop during mating season find that each
male baboon mates with the following number of female baboons:
-279-
Back to TOC
Male 1 Male 2 Male 3 Male 4 Male 5 Male 6 Male 7 Male 8
1 5 0 1 3 65 2 3
a. Based on this information, how many female baboons is each male expected to mate with?
b. Perform a chi-square test on the data.
d. Assuming α = .05, is there a statistically significant difference in the number of female baboons
that mate with each male baboon? What decisions are made with the hypotheses?
8. Use the following to answer the questions below: The data from a recent US Congressional vote are
displayed below by the political party of the congressional representative and his/her vote. Use this data
to answer the questions that follow.
Decision
Political Party Yea Nay Abstain Totals
Democratic fo = 140 fo = 30 fo = 10 180
Republican fo = 20 fo = 225 fo = 10 255
Totals 160 255 20 nT = 435
a. Calculate the expected frequency of each cell.

b. Perform the chi-square test on these data.
d. Assuming α = .05, is there a statistically significant difference between the observed and the
expected frequencies? What decisions are made with the hypotheses?
2
f. Calculate Cramer's Contingency Coefficient.
g. Calculate Cohen’s w from Cramer’s Contingency Coefficient
h. Using G*Power 3, how much statistical power was there to reject the null hypothesis?
9. Use the following to answer the questions below: The table displays the observed frequencies of
political party affiliation and residential area for a sample of individuals.
Residential Area
Party
Affiliation City Suburbs Country Totals
Democrat 120 80 60 260
Republican 40 80 20 140
Independent 40 40 20 100
Totals 200 200 100 nT = 500

b. Perform the chi-square test on these data
2
f. Calculate Cramer's C statistic.
g. Calculate Cohen’s w from Cramer’s Contingency Coefficient.
h. Using G*Power 3, how much statistical power was there to reject the null hypothesis?
-280-
Back to TOC
10. Use the following to answer the questions below: There is a stereotype that university professors tend
to be atheist or agnostic. This stereotype is even more pronounced if the professor has an appointment in
the sciences (e.g., biology, chemistry, physics, psychology) compared to the humanities (e.g., history,
philosophy, theology, English). To examine whether there is a relationship between religions view and
appointment of a professor, 250 professors are asked whether their appointment is in the sciences, the
humanities, or some other field and whether they are Religious, Agnostic, or Atheist. The following
frequencies are observed:
Field Religious Agnostic Atheist

Sciences 25 25 75
Humanities 25 25 50
Other 15 5 5

b. Perform the chi-square test on these data
e. Calculate Cramer's C statistic.
f. Calculate Cohen’s w from Cramer’s Contingency Coefficient.
g. Using G*Power 3, how much statistical power was there to reject the null hypothesis?
11. Use the following to answer the questions below: Suppose the relative frequencies of students
belonging to each of three colleges at a particular university are as follows:
College rf
Natural Sciences .20
Business School .25
Social Sciences .35
Arts and Humanities .20
Furthermore, suppose that a general education course of 100 students yields the following observed
frequencies:
College fo
Natural Sciences 40
Business School 20
Social Sciences 20
Arts and Humanities 20
a. Determine the expected frequencies for each college.

b. Perform a chi-square goodness of fit test on these data.
2
12. Use the following to answer the questions below: Suppose that the relative frequencies of male and
female students at a particular high school are as follows:
College rf
Male .47
Female .53
-281-
Back to TOC
Suppose that a course of 50 students yields the following observed frequencies:
College fo
Male 26
Female 24
a. Determine the expected frequencies for each college.

b. Perform a chi-square goodness of fit test on these data.
13. Use the following to answer the questions below: A political science teacher is examining the
relationship between political party affiliation and college class among the students in one of his lectures.
The following contingency table was created:
Political Party
Affiliation
Sex DemocraticRepublican
Underclassmen 10 20
Upperclassmen 25 5
a. Using the computational formula for a 2x2 contingency table, calculate the obtained value of the
chi-square statistic.
b. What is the p-value for this test statistic?
c. Assuming α = .05, is there a statistically significant difference between the observed and the
d. Calculate the phi coefficient.
14. I want to know whether there is a relationship between sex (males/female) and getting caught text-
messaging in class. For one academic year I count the number of males and females caught text-
messaging in my classes. I also count the numbers of males and females from those classes that I did
not catch text-messaging (i.e., who were not likely text-messaging). The frequencies of males and
females caught text-messaging and not caught text-messaging in class for this academic year are below.
Use this information to answer the following questions:
Caught Text Messaging

Sex Yes No
Males 15 55
Females 30 80
a. Using the computational formula for a 2x2 contingency table, calculate the obtained value of the
chi-square statistic.
b. What is the p-value for this test statistic?
c. Assuming α = .05, is there a statistically significant difference between the observed and the
d. Calculate the phi coefficient.
-282-
Back to TOC
a. w = .10, α = .05, Power = .80, df = 2

b. w = .20, α = .05, Power = .80, df = 2
c. w = .10, α = .05, Power = .95, df = 2
d. w = .20, α = .05, Power = .80, df = 4
parameters given.
a. w = .30, α = .01, n = 100, df = 2

b. w = .30, α = .01, n = 100, df = 4
c. w = .40, α = .01, n = 200, df = 2
d. w = .40, α = .05, n = 50, df = 2
-283-
Back to TOC
Chapter 22: Power Analysis and Replication
22.1 Review of Effect sizes and Power
Throughout earlier chapters, I addressed issues related to effect size and power, as well as some of the
factors that influence power, such as sample size, alpha level, directionality, and type of inferential test.
To recap, effect size is the strength or size of the relationship being examined and statistical power is the
probability of correctly rejecting a false null hypothesis (i.e., of correctly detecting a statistically significant
result). Furthermore, effect size and power are related, in that, small effect sizes require more statistical
power to detect whereas larger effect sizes do not require as much power to detect. Finally, remember
that sample size is also related to power, where larger sample sizes generally increase the power to
detect a true significant effect.
In earlier chapters I mentioned that power can be used in, mainly, one of two ways: (1) After an inferential
test is performed on a set of data power can be calculated based on observed results. This is a post-hoc
power analysis and is used mainly to estimate the observed power of an inferential test and to ensure
the probability of correctly rejecting a null hypothesis meets some accepted criterion, which is generally
.80 or greater. (2) Before beginning data collection a power analysis can be performed to determine
needed sample sizes. This is an a priori power analysis and is used during the planning phase of a
study to determine the number of subjects needed to achieve a certain level of statistical power based on
a predicted effect size, a chosen alpha level, and the planned statistical test.
Some would argue that conducting a priori power analyses is more important than conducting post hoc
power analyses, because by determining the number of subjects needed for a study based on a set of
criteria you ensure that you are not collecting too many or too few subjects. Thus, the statistical decision
that is made should be legitimate given the amount of collected data. In contrast, a post hoc power
analyses is generally performed only to ensure that a set of results can stand up to statistical scrutiny.
Thus, they may be used, somewhat inappropriately, to justify adding more subjects to a study if statistical
significance is not achieved. Nonetheless, knowing how much power was obtained in a statistical analysis
is still important and I recommend reporting measures of power, or at least effect size, along with the
results of inferential statistics.
A priori power analyses are also important when planning replication studies, which are studies that
either fully replicate or partially replicate the methods of a previously conducted and possibly published
study. There are two main classes of replication studies: full replications and partial replications. Full
replications mimic the methods, stimuli, and procedures of a previous study in every detail, thus, full
replications are instances where a study is basically performed again. Partial replications mimic certain
aspects of a previous study, but change certain elements, with the intent to extend the results beyond the
previous study.
The reasons for partial replication are easy to identify, that is, to extend previous results; but what is the
purpose of a full replication? That is, why would a researcher want to conduct someone else’s study?
Indeed, this is a valid question, because journals are less likely to publish replication studies that produce
the same results as an original study. Thus, replications not only take time and energy, but are less likely
to bet published, so why do people do them?
The simple answer is that replication is confirmation that an observed result is real and is not some
statistical fluke. Remember, when you claim that a result is statistically significant, say with an alpha level
of .01 and a statistical power of .80, you are saying that there is less than a 1% chance of detecting your
result if the null hypothesis was true and an 80% chance that you have correctly rejected a false null
hypothesis. Thus, there is still a chance that your result could be obtained even if the null was true (Type I
error). Remember, statistics are all based on probabilities, so there is always a chance of being wrong.
But, if the results of a study are replicated, this makes it more likely that the observed result in the original
study was real and was not just a fluke. Perhaps two brief examples will help to elucidate the importance
of replication.
-284-
Back to TOC
Noted Cornell social psychologist, Daryl Bem, published a highly
controversial paper in the Journal of Personality and Social Psychology,
which he claimed to show evidence of the psi phenomenon
(extrasensory perception, ESP; Bem, 2011). Such a claim is great, as
many people doubt the existence of the psi phenomenon. To his credit,
Bem conducted a priori power analyses based on an expected effect
size, his chosen statistical test, and the desires and conventional level
of statistical power (.80). Thus, procedurally Bem did most things
correct and he claimed to show evidence for the psi phenomenon in
nine separate experiments that in all involved roughly 1000 subjects.
Because of the controversy surrounding the psi phenomenon, a number
of researchers attempted to replicate Bem’s study and, potentially,
evidence of the psi phenomenon. But, after a number of published
replication studies (e.g., Ritchie, Wiseman & French, 2012) and
mentions of replication studies by other researchers, the evidence of psi is becoming less and less.
Specifically, each replication study has failed to replicate Bem’s results, suggesting that the original
6
results may be a fluke (Type I error).
As a second example, at the time of my writing this (July

th
5 , 2012), The European Organization of Nuclear
Research (CERN) has just announced the likely
discovery of the Higgs hoson; a subatomic particle that
interacts with certain other subatomic particles to create
matter and give that matter mass. The Higgs boson was
predicted in the standard model of physics in the 1960’s
and, until CERN’s announcement, had only been
theoretical. Although this is a monumental discovery,
before concluding that the data collected by CERN is
evidence of the Higgs boson additional data must be
collected; thus, the study and the results must be
replicated. That is, it is possible, though highly unlikely,
that the evidence for the Higgs boson was just a fluke. The evidence must be replicated again and again
to show that the boson was actually discovered.
In short, replication is important to science for verifying or disconfirming new discoveries. Importantly,
when preparing for replications, statistical power analyses must be conducted to ensure that the original
study is repeated in as many was as possible, which includes using appropriate sample sizes.
22.2 Preparing for Replication

As implied in the previous section, there is more to conducting a replication study than making sure the
same procedures, equipment, materials, and population are being used; hence, replication is more than
just methodological. To properly conduct a replication study you must also ensure that the study will have
sufficient statistical power to detect an effect of the size or greater that was reported in the original study.
Thus, an a priori power analysis must be performed in order to determine the appropriate number of
subjects to test in the replication study, so that the eventual statistical decision will be justified based on
the data that is collected. Thus, a priori power analyses are all about determining the appropriate sample
size to use.
6
For interesting commentary on this issue, see Chris French’s blog at The Guardian:
http://www.guardian.co.uk/science/2012/mar/15/precognition-studies-curse-failed-replications
-285-
Back to TOC
Before getting into conducting the actual power analysis, let me briefly mention the steps involved with a
replication study:
1. Identify a study to be replicated. People do not replicate a study simply to replicate, and not all
studies need to be replicated. Rather, a study that reports a novel discovery (e.g., psi
phenomenon, Higgs boson) is more likely a candidate for replication.
2. Design the methods of the replication study. This comes from the methods of the original study
and is beyond the scope of this book. But, this basically involved using the same stimuli,
procedures, equipment, etc. that were used in the original study.
3. Perform a power analysis to determine the appropriate sample size. More on this below.
4. Begin the study.
From earlier chapters, recall that there are many power analysis tools available on the Internet. One in
particular is G*Power 3, which was introduced in earlier chapters and will be used here as well. In order to
perform an a priori power analysis to determine appropriate sample size, you need to know several
values from the original study that you are planning to replicate. Specifically, to properly replicate a study
you must know, from the original experiment:
1. The type of inferential test

2. The alpha level used (if this value is unknown or if the achieved alpha-level was reported for each
inferential test, then it is OK to use .05 or .01)
3. Whether the inferential test was one-tailed or two-tailed (if applicable)
2
4. The obtained effect size in the test (i.e., Cohen’s d, Cohen’s f, Cohen’s w, η ).
Most of these values can be obtained fairly easy from the Results sections of a published manuscript; if
not then contacting the corresponding author may be needed. But, some authors can be reluctant to
provide certain information, if they do not feel replication of their study is warranted. (people have egos.)
However, one issue is that many studies do not report effect size measures with the results of an
inferential test. For example, one may report the result of an ANOVA as F(3, 36) = 5.00, MSE = 10, p <
.05. This provides information about the statistical significance of the ANOVA, but does not provide any
2
immediate and direct information about the effect size. Specifically, Cohen’s f or η were not reported. If
2
the results of the ANOVA had included Cohen’s f or η , then these values could easily be used in
G*Power to determine the appropriate sample size for a replication study. But, in cases where effect size
measures are not reported with an inferential test, the effect size must first be estimated from the
available information, before the power analysis can be conducted. The next section describes some
methods for doing so.
22.3 Effect Size Computations

As mentioned in the preceding section, one problem in conducting a priori power analyses for replication
studies is determining the effect size in a published study, because authors often fail to report effect size
2
measurements (Cohen’s d, Cohen’s f, Cohen’s w, η ) with the results of inferential statistical tests. Thus,
2
the F statistic for ANOVA is reported, but a corresponding value for η or Cohen’s f is not. In such cases
the appropriate effect size measure must be calculated and estimated based on the available information
in the article. Luckily, there are several straightforward methods for calculating the effect size from the
information that is almost always reported in any inferential statistic.
This section covers some of the “tricks” for computing effect size measurements from available
information in Results sections. The one thing to keep in mind is that when you need to calculate an effect
size for the results of a published study, identify all of the information that is available in that study first
and then choose the best method for calculating the appropriate effect size.
-286-
Back to TOC
Independent Groups t-Tests
Assume that a researcher compared two independent groups (A 1 and A2) with n = 10 subjects per group
using an independent groups t-Test and obtained the following results: MA1 = 10, SDA1 = 3; MA2 = 14,
SDA2 = 4; t(18) = 2.53, SE = 1.58, p < .05 (two-tails). Recall, that the standard measure of effect size,
which is also used in G*Power, is Cohen’s d, which is defined as the number of standard deviations that
separate two means and is calculated thus:
̅ ̅
| |
̂
Although it is almost certainly the case that the mean of each condition, or at least the mean difference
between the two conditions, will be reported in a published article, it is almost never the case that the
pooled standard deviation ( ̂ ) is reported. Thus, the pooled standard deviation may need to be
estimated from available information. Recall that the pooled standard deviation is simply the square root
of the pooled variance: ̂ √ ̂ , and the pooled variance can be found from this formula:
( )̂ ( )̂
̂
Thus, the pooled standard deviation is:
( )̂ ( )̂
̂ √
From the information in the example above, we know that n1 and n2 are both 10, that is, the sample size
for each condition was 10. The ̂ and ̂ values are simply the variances for each condition. Although we
do not know the variance for each condition, we do know the standard deviation for each condition (SDA1
= 3 and SDA2 = 4), and because we know that the variance is the square of the standard deviation, we
can easily determine the variance of the two conditions: ̂ and ̂
Thus, if we take the known sample sizes and variances from each group we can plug those values into
the formula above to solve for the pooled standard deviation:
( ) ( )
̂ √ √ √
Using this value and the means of the two conditions (MA1 = 10 and MA2 = 14), Cohen’s d is equal to:
| |
-287-
Back to TOC
This value could then be used in a power analysis to determine the appropriate sample size to use in a
replication study. Incidentally, G*Power has a built in feature for determining the effect size based on the
means and standard deviations of conditions. With G*Power open and either “Post hoc” or “A priori”
selected under Type of power analysis, click the Determine => button, and a window will open to the right:
In the window that opens you can enter the mean and standard deviation of each condition (under n1 =
n2) and click the Calculate button to obtain the effect size, which you can then transfer to the main
7
window by clicking the ‘Calculate and transfer to main window’ button.
It may be the case that the standard deviations are not reported, but the standard error of the mean is
reported. Recall the standard error of the mean is equal to the standard deviation divided by the square
root of the sample size. In the example above the standard errors are SEA1 = 3/√10 = 0.949 and SEA2 =
4/√10 = 1.265. If instead of the standard deviation the standard error of the mean is reported, you can
calculate the standard deviation from the following formula:
̂ √
7
This example here is for an independent groups t-test; however, G*Power’s ability to calculate the effect
size in this manner is also available for the correlated samples t-test, ANOVA, chi-square, and other
inferential statistical tests.
-288-
Back to TOC
Once the standard deviations have been calculated from the standard errors, you can use enter that
information into G*Power to determine the effect size.
Nonetheless, it if often the case that the standard deviations are not reported with condition means. In
such cases, the above procedures will not work, because the standard error, the variance, or the
standard deviation is needed to determine the effect size. In those situations, there are computational
formulas that can be used to estimate the effect size from a t-test directly, that is, from the t-statistic and
the sample sizes.
In cases where the t-statistic and the sample size of each condition is known, the following formula can
be used to estimate Cohen’s d:
√( )( )
Here, tI is the value of the independent-groups t-test, which was equal to 2.53 in this example.
Importantly, this formula cannot be used for correlated-samples t-tests, as these formulas overestimate
the effect size for correlated samples tests. I provide some tops for calculating power of correlated
samples t-tests later.
Plugging in the t-value and samples sizes into this formula, we get:
√( )( )
Although this is slightly larger than the d = 1.13 values obtained above and through G*Power, it is quite
close. In cases where the sample size of each independent group is not provided, you can use the
following formula:
√( ⁄ ) or ( ⁄ )
where nk is the sample size, per group. In our example, we have nk = 10 per group, thus
√( ⁄ )
which is equal to the d = 1.13 from above. In cases where the sample sizes are unknown and may not be
equivalent in size, simply divide the total sample size by two and use that as nk.
Correlated Samples t-Tests
The above examples were formulas for calculating effect size in an independent groups t-test. There is
some controversy about how to calculate effect size in a correlated samples t-test. The following formula
cannot be used, because it overestimates the true effect size:
√( ⁄ )
-289-
Back to TOC
The following formula is designed to correct for the overestimation (Dunlap, Cortina, Vaslow & Burke,
1996):
√[ ( )⁄ ] or [ ( )⁄ ]
where tC is the value of the correlated samples t-test, r is the Pearson correlation between conditions, and
n is the total sample size. Unfortunately, the Pearson correlation between conditions is almost never
reported with the correlated samples t-test; hence, this formula proves to be largely useless.
Recall though that Cohen’s d (dz in the correlated samples t-test) is the difference between two means
divided by the pooled standard deviation. The pooled standard deviation can also be through of as the
standard deviation of the difference ( ̂ ). In the correlated samples t-test, the test statistic is, technically
speaking, equal to the mean difference (Md) divided by the standard error of the difference (SEd):
The standard error of the difference is equal to the standard deviation of the difference divided by the
square root of the sample size:
̂
√
Importantly, the standard error of the difference is the term that is almost always reported with the results
of a correlated-samples t-test. Also, if you know the standard error of the difference and the sample size,
you can calculate the standard deviation of the difference from the following formula:
̂ √
For example, say a researcher compared two means (B1 and B2) in a correlated samples t-test and
obtained the following results: MB1 = 6; MB2 = 9; t(9) = 2.50, SE = 1.20, p < .05 (two-tailed). The SE value
(1.20) is the standard error of the difference (SEd in the formulas above), which can be used to calculate
the standard deviation of the difference:
̂ √
Once you know the standard deviation of the difference, you can use that to calculate Cohen’s d, by using
̂ for the pooled standard deviation:
| |
-290-
Back to TOC
You can also enter this information into the pop-out window in G*Power to obtain the effect size value
(Cohen’s dz):
Analysis of Variance
The following is mostly applicable to finding effect size in oneway between subjects ANOVA, but the
techniques could also be used to calculate effect size for other ANOVA techniques. As with t-tests, it is
often the case that the results of an F-test for ANOVA include the F-statistic, the degrees of freedom, the
2
p-value, and the mean square for the error term, but not a measure of effect size (Cohen’s f, η ) and,
hence, must be estimated from available information.
2
Recall that the proportion of explained variance, η is found from the formula:
And the effect size, Cohen’s f is found from the formula:
-291-
Back to TOC
√
2
Thus, except for alternative examples of Cohen’s f below, to obtain f you need to first calculate η . But,
the sums of squares are almost never reported in published articles; hence, the formula above is useful
2
only when one has first-hand knowledge of the data. But η can also be thought of, more correctly, as the
following:
Where SSEffect is the sum of squares for the effect you are interested in and SS Effect is the sum of squares
for the error term. But again, you need to know the sums of squares values, which is rarely the case.
However, with some mathematical rearrangement and manipulation, this formula can be rewritten as:
( )
( )
where dfEffect is the degrees of freedom between-groups (i.e., for the effect of the independent variable),
dfEffect is the degrees of freedom within subjects (i.e., for the error term), and f is the F-statistic obtained in
the ANOVA.
For example, say that a researcher runs a oneway Source of Variance SS df MS F

between subjects ANOVA on a set of data with k = 4 Between 150 3 50 5
levels of an independent variable and n = 10 subjects
Within 360 36 10
per level. Here is a hypothetical ANOVA summary
table of the results. Total 510 39
2
From the original formula for η , the proportion of explained variance is:
2
Using the modified formula from above, η is:
( ) ( )
( ) ( )
Whichever formula is used, the resulting value can be used in the formula for Cohen’s f and then used in
G*Power. In this example, Cohen’s f is:
√ √
-292-
Back to TOC
An alternative method for calculating Cohen’s f, this time directly from the F-statistic and the degrees of
freedom values comes from this formula, provided in Rosenthal and Rosnow (2008), which defines the F-
statistic as the effect size multiplied by the radio of the degrees of freedom within groups (error) to the
degrees of freedom in the effect (between groups):
2
Importantly, the term with η is simply the term under the numerator of Cohen’s f, thus, it is equivalent to
the square of Cohen’s f:
Thus,
2
Rearranging to solve for f , we have:
Taking the square root of each side, we have a formula that we can use to directly calculate Cohen’s f:
Using the ANOVA summary table from above and this formula we find that Cohen’s f is:
which is the same value that was obtained above. This Cohen’s f value would then be entered into
G*Power to be used in a power analysis.
Chi-Square Analyses
Although, like ANOVA and t-tests, it is rare for authors to report effect size measures with a chi-square
test Cohen’s w, Cramer’s Correlation Coefficient [C], Φ), calculating these values from given information
is relatively easy.
Recall, that Cohen’s w is computed using the following formula:
Where C is Cramer’s Correlation Coefficient, which is computed from the following formula:
-293-
Back to TOC
√
Thus, to calculate C, you need to know the chi-square statistic and the total number of subjects n), which
are almost always reported in the results of an analysis using chi-square. For example, say an author
2
reports the following results of a chi-square test of independence: χ (2, N = 1000) = 3.55, p < .05.
Cramer’s C is equal to:
From Cramer’s C, you can now calculate Cohen’s w:
This Cohen’s w value would then be entered into G*Power to be used in a power analysis.
22.4. Examples
This section will provide several examples for properly conducting a power analysis based on a published
study. Note that the procedures of a power analysis is basically the same, that is, estimate the effect size
of the published study, choose an alpha level and desired level of power, etc.; the only thing that differs is
how the effect size is estimated.
Bem (2011) reported an experiment entitled “Retroactive Facilitation of Recall II” (Experiment 9), in which
he observed greater “recall” of words prior to the words being studied. That is, subjects wrote down a
greater percentage of words that they would see in the future compared to words they would not see in
the future. The mean performance score of 4.21% was compared to 0% using a one-sample t-test, t(49) =
2.96, p = .002 (one-tail), d = 0.42. Thus, in this case, the effect size was provided.
Say we want to replicate Bem’s (2011) Experiment 9. We know that Cohen’s d = .42, we know that we
need to run a directional (on-tailed) one-sample t-test. We choose an alpha level of .01 and we want
Power = .90. Using G*Power, we enter these and find that we need a sample size of n = 77 to achieve
this amount of power (.90) based on a Cohen’s d of .42 and an alpha level of .01:
-294-
Back to TOC
As a second example, say that a researcher examined the difference in SAT math scores between males
and females and found that males had significantly greater SAT math scores (M = 670) than females (M =
640), t(98) = 2.10, SE = 500, p < .05 (two-tails). In this case, the researcher had to use an independent
groups t-test, because the two groups are naturally independent (assume the group sizes were equal,
that is, n = 50 for both male and females). Because we do not know the standard deviations or the
standard errors, we must estimate Cohen’s d using the following formula:
√( ⁄ )
Thus,
√( ⁄ )
-295-
Back to TOC
From the information given, we know that in order to replicate this study we need to use a non-directional,
independent groups t-test. We decide to use an alpha level of .01 and we want Power = .90. Using
G*Power, we find that we need n = 342 subjects (171 per group):
As a final example, assume that a researcher conducts a study comparing three levels of orange juice
consumption to examine orange juice consumption on number of colds over a five-year period. The
researcher recruits n = 300 subjects, with 100 subjects per level of the independent variable. The data are
analyzed using a oneway between subjects ANOVA and the results are, F(2, 297) = 10.50, MSE =
5000.00, p < .05. Because no effect size measures were provided by the author, Cohen’s f must be
estimated from the result of the F-test:
-296-
Back to TOC
√ √
From the information given, we know that in order to replicate this study we need to use a oneway
between subjects ANOVA with three groups. We decide to use an alpha level of .05 and we want Power
= .80. Using G*Power, we find that we need n = 141 subjects (47 per group):
-297-
Back to TOC
For #1 – 10, use the available information to calculate the effect size.
1. Independent groups t-test: t(28) = 2.00, SE = 10.00, p < .01 (one-tailed)
2. Correlated samples t-test: t(9) = 5.00, SE = 4.50, p < .05 (one-tailed)
3. Independent groups t-test: M1 = 15, SD1 = 5, n1 = 20; M2 = 18, SD2 = 9, n2 = 20
4. Pearson correlation: r(98) = .66, p < .01 (two-tailed)
5. Oneway between subjects ANOVA: F(1, 29) = 3.50, MSE = 100.00, p < .05
6. Oneway between subjects ANOVA: F(4, 95) = 4.25, MSE = 2500.00, p < .05
7. Correlated samples t-test: t(14) = 3.65, r = .95, p < .01
8. Independent groups t-test: t(8) = 4.50, SE = 2.50, p < .05 (two-tailed)

2
9. Chi-square test of independence: χ (1, N = 100) = 5.25, p < .05
2
10. Chi-square test of independence: χ (4, N = 500) = 6.50, p < .01
For #11 – 21, use the available information to calculate the needed sample sizes.
11. Independent groups t-test, Cohen’s d = .25, p = .01 (two-tails), Power = .95
12. Independent groups t-test, Cohen’s d = .10, p = .05 (one-tail), Power = .80
13. Correlated samples t-test, Cohen’s dz = .50, p = .001 (two-tails), Power = .99
14. Correlated samples t-test, Cohen’s dz = .15, p = .005 (one-tail), Power = .85
15. Oneway between subjects ANOVA, Cohen’s f = .05, p = .005, k = 4, Power = .80
18. Pearson correlation, r = .20, p = .0001 (one-tail), Power = .95
19. Pearson correlation, r = .40, p = .05 (two-tails), Power = .80
20. Chi-square test for independence, Cohen’s w = .20, p = .05, df = 2, Power = .85
21. Chi-square test for independence, Cohen’s w = .30, p = .001, df = 4, Power = .95
-298-
Back to TOC
For each of the following sets of results, calculate the appropriate effect size and then determine the
number of subjects needed in a replication study with α = .05 and Power = .80, and a two-tailed test
where appropriate.
22. Carter, Ferfuson and Hassin (2011) found that participants who were presented with a picture of the
American flag reported a greater likelihood to vote for John McCain (M = 0.072, SD = 0.47) than people
who were not presented with the American flag (M = -0.070, SD = 0.48), t(181) = 2.02, p = .04.
23. Kay. Moscovitch and Laurin (2010) had subjects unscramble each of 16 five-word sets. For half of the
participants some of these words were related to randomness (“chance”, “random”) and for the other half
the words were not related to randomness. Using a oneway, they found that subjects who were presented
with the randomness-related words reported a stronger beliefs in the existence in supernatural sources of
control than subjects who were not presented with randomness-related words, F(1,33) = 4.47, p < .05.
24. Tybur, Bryan, Magnan and Caldwell-Hooper (2011) observed that a group of subjects who were
exposed to a noxious smelling compound (“Liquid ASS”) reported a greater intention to use condoms
than did a group of subjects who were not exposed to the noxious compound, t(97) = 2.06, p < .05.
-299-
Back to TOC
Appendix A
Table 1: Standard Normal (z) Distribution Table
p (x ≥ +z) or p (0 < x ≤ +z) or

z p (-z ≤ x ≥ +z) p (x ≤ -z or x ≥ +z)
p (x ≤ -z) p (0 > x ≥ -z)
0.00 0.0000 0.5000 1.0000 0.0000
0.01 0.0080 0.4960 0.9920 0.0040
0.02 0.0160 0.4920 0.9840 0.0080
0.03 0.0239 0.4880 0.9761 0.0120
0.04 0.0319 0.4840 0.9681 0.0160
0.05 0.0399 0.4801 0.9601 0.0199
0.06 0.0478 0.4761 0.9522 0.0239
0.07 0.0558 0.4721 0.9442 0.0279
0.08 0.0638 0.4681 0.9362 0.0319
0.09 0.0717 0.4641 0.9283 0.0359
0.10 0.0797 0.4602 0.9203 0.0398
0.11 0.0876 0.4562 0.9124 0.0438
0.12 0.0955 0.4522 0.9045 0.0478
0.13 0.1034 0.4483 0.8966 0.0517
0.14 0.1113 0.4443 0.8887 0.0557
0.15 0.1192 0.4404 0.8808 0.0596
0.16 0.1271 0.4364 0.8729 0.0636
0.17 0.1350 0.4325 0.8650 0.0675
0.18 0.1428 0.4286 0.8572 0.0714
0.19 0.1507 0.4247 0.8493 0.0753
0.20 0.1585 0.4207 0.8415 0.0793
0.21 0.1663 0.4168 0.8337 0.0832
0.22 0.1741 0.4129 0.8259 0.0871
0.23 0.1819 0.4090 0.8181 0.0910
0.24 0.1897 0.4052 0.8103 0.0948
0.25 0.1974 0.4013 0.8026 0.0987
0.26 0.2051 0.3974 0.7949 0.1026
0.27 0.2128 0.3936 0.7872 0.1064
0.28 0.2205 0.3897 0.7795 0.1103
0.29 0.2282 0.3859 0.7718 0.1141
0.30 0.2358 0.3821 0.7642 0.1179
0.31 0.2434 0.3783 0.7566 0.1217
0.32 0.2510 0.3745 0.7490 0.1255
0.33 0.2586 0.3707 0.7414 0.1293
0.34 0.2661 0.3669 0.7339 0.1331
0.35 0.2737 0.3632 0.7263 0.1368
0.36 0.2812 0.3594 0.7188 0.1406
-300-
Back to TOC
0.37 0.2886 0.3557 0.7114 0.1443
0.38 0.2961 0.3520 0.7039 0.1480
0.39 0.3035 0.3483 0.6965 0.1517
0.40 0.3108 0.3446 0.6892 0.1554
0.41 0.3182 0.3409 0.6818 0.1591
0.42 0.3255 0.3372 0.6745 0.1628
0.43 0.3328 0.3300 0.6599 0.1700
0.44 0.3401 0.3264 0.6527 0.1736
0.45 0.3473 0.3264 0.6527 0.1736
0.46 0.3545 0.3228 0.6455 0.1772
0.47 0.3616 0.3192 0.6384 0.1808
0.48 0.3688 0.3156 0.6312 0.1844
0.49 0.3759 0.3121 0.6241 0.1879
0.50 0.3829 0.3085 0.6171 0.1915
0.51 0.3899 0.3050 0.6101 0.1950
0.52 0.3969 0.3015 0.6031 0.1985
0.53 0.4039 0.2981 0.5961 0.2019
0.54 0.4108 0.2946 0.5892 0.2054
0.55 0.4177 0.2912 0.5823 0.2088
0.56 0.4245 0.2877 0.5755 0.2123
0.57 0.4313 0.2843 0.5687 0.2157
0.58 0.4381 0.2810 0.5619 0.2190
0.59 0.4448 0.2776 0.5552 0.2224
0.60 0.4515 0.2743 0.5485 0.2257
0.61 0.4581 0.2709 0.5419 0.2291
0.62 0.4647 0.2676 0.5353 0.2324
0.63 0.4713 0.2643 0.5287 0.2357
0.64 0.4778 0.2611 0.5222 0.2389
0.65 0.4843 0.2578 0.5157 0.2422
0.66 0.4907 0.2546 0.5093 0.2454
0.67 0.4971 0.2514 0.5029 0.2486
0.68 0.5035 0.2483 0.4965 0.2517
0.69 0.5098 0.2451 0.4902 0.2549
0.70 0.5161 0.2420 0.4839 0.2580
0.71 0.5223 0.2389 0.4777 0.2611
0.72 0.5285 0.2358 0.4715 0.2642
0.73 0.5346 0.2327 0.4654 0.2673
0.74 0.5407 0.2296 0.4593 0.2704
0.75 0.5467 0.2266 0.4533 0.2734
0.76 0.5527 0.2236 0.4473 0.2764
0.77 0.5587 0.2206 0.4413 0.2794
0.78 0.5646 0.2177 0.4354 0.2823
0.79 0.5705 0.2148 0.4295 0.2852
0.80 0.5763 0.2119 0.4237 0.2881
0.81 0.5821 0.2090 0.4179 0.2910
0.82 0.5878 0.2061 0.4122 0.2939
0.83 0.5935 0.2033 0.4065 0.2967
0.84 0.5991 0.2005 0.4009 0.2995
0.85 0.6047 0.1977 0.3953 0.3023
0.86 0.6102 0.1949 0.3898 0.3051
0.87 0.6157 0.1922 0.3843 0.3078
0.88 0.6211 0.1894 0.3789 0.3106
0.89 0.6265 0.1867 0.3735 0.3133
0.90 0.6319 0.1841 0.3681 0.3159
0.91 0.6372 0.1814 0.3628 0.3186
0.92 0.6424 0.1788 0.3576 0.3212
0.93 0.6476 0.1762 0.3524 0.3238
-301-
Back to TOC
0.94 0.6528 0.1736 0.3472 0.3264
0.95 0.6579 0.1711 0.3421 0.3289
0.96 0.6629 0.1685 0.3371 0.3315
0.97 0.6680 0.1660 0.3320 0.3340
0.98 0.6729 0.1635 0.3271 0.3365
0.99 0.6778 0.1611 0.3222 0.3389
1.00 0.6827 0.1587 0.3174 0.3413
1.01 0.6875 0.1562 0.3124 0.3438
1.02 0.6923 0.1539 0.3077 0.3461
1.03 0.6970 0.1515 0.3030 0.3485
1.04 0.7017 0.1492 0.2983 0.3508
1.05 0.7063 0.1469 0.2937 0.3531
1.06 0.7109 0.1446 0.2891 0.3554
1.07 0.7154 0.1423 0.2846 0.3577
1.08 0.7199 0.1401 0.2801 0.3599
1.09 0.7243 0.1379 0.2757 0.3621
1.10 0.7287 0.1357 0.2713 0.3643
1.11 0.7330 0.1335 0.2670 0.3665
1.12 0.7373 0.1314 0.2627 0.3686
1.13 0.7415 0.1292 0.2585 0.3708
1.14 0.7457 0.1271 0.2543 0.3729
1.15 0.7499 0.1251 0.2501 0.3749
1.16 0.7540 0.1230 0.2460 0.3770
1.17 0.7580 0.1210 0.2420 0.3790
1.18 0.7620 0.1190 0.2380 0.3810
1.19 0.7660 0.1170 0.2340 0.3830
1.20 0.7699 0.1151 0.2301 0.3849
1.21 0.7737 0.1131 0.2263 0.3869
1.22 0.7775 0.1112 0.2225 0.3888
1.23 0.7813 0.1093 0.2187 0.3907
1.24 0.7850 0.1075 0.2150 0.3925
1.25 0.7887 0.1056 0.2113 0.3944
1.26 0.7923 0.1038 0.2077 0.3962
1.27 0.7959 0.1020 0.2041 0.3980
1.28 0.7995 0.1003 0.2005 0.3997
1.29 0.8029 0.0985 0.1971 0.4015
1.30 0.8064 0.0968 0.1936 0.4032
1.31 0.8098 0.0951 0.1902 0.4049
1.32 0.8132 0.0934 0.1868 0.4066
1.33 0.8165 0.0918 0.1835 0.4082
1.34 0.8198 0.0901 0.1802 0.4099
1.35 0.8230 0.0885 0.1770 0.4115
1.36 0.8262 0.0869 0.1738 0.4131
1.37 0.8293 0.0853 0.1707 0.4147
1.38 0.8324 0.0838 0.1676 0.4162
1.39 0.8355 0.0823 0.1645 0.4177
1.40 0.8385 0.0808 0.1615 0.4192
1.41 0.8415 0.0793 0.1585 0.4207
1.42 0.8444 0.0778 0.1556 0.4222
1.43 0.8473 0.0764 0.1527 0.4236
1.44 0.8501 0.0749 0.1499 0.4251
1.45 0.8529 0.0735 0.1471 0.4265
1.46 0.8557 0.0721 0.1443 0.4279
1.47 0.8584 0.0708 0.1416 0.4292
1.48 0.8611 0.0694 0.1389 0.4306
1.49 0.8638 0.0681 0.1362 0.4319
1.50 0.8664 0.0668 0.1336 0.4332
-302-
Back to TOC
1.51 0.8690 0.0655 0.1310 0.4345
1.52 0.8715 0.0643 0.1285 0.4357
1.53 0.8740 0.0630 0.1260 0.4370
1.54 0.8764 0.0618 0.1236 0.4382
1.55 0.8789 0.0606 0.1211 0.4394
1.56 0.8812 0.0594 0.1188 0.4406
1.57 0.8836 0.0582 0.1164 0.4418
1.58 0.8859 0.0571 0.1141 0.4429
1.59 0.8882 0.0559 0.1118 0.4441
1.60 0.8904 0.0548 0.1096 0.4452
1.61 0.8926 0.0537 0.1074 0.4463
1.62 0.8948 0.0526 0.1052 0.4474
1.63 0.8969 0.0516 0.1031 0.4484
1.64 0.8990 0.0505 0.1010 0.4495
1.65 0.9011 0.0495 0.0989 0.4505
1.66 0.9031 0.0485 0.0969 0.4515
1.67 0.9051 0.0475 0.0949 0.4525
1.68 0.9070 0.0465 0.0930 0.4535
1.69 0.9090 0.0455 0.0910 0.4545
1.70 0.9109 0.0446 0.0891 0.4554
1.71 0.9127 0.0436 0.0873 0.4564
1.72 0.9146 0.0427 0.0854 0.4573
1.73 0.9164 0.0418 0.0836 0.4582
1.74 0.9181 0.0409 0.0819 0.4591
1.75 0.9199 0.0401 0.0801 0.4599
1.76 0.9216 0.0392 0.0784 0.4608
1.77 0.9233 0.0384 0.0767 0.4616
1.78 0.9249 0.0375 0.0751 0.4625
1.79 0.9265 0.0367 0.0735 0.4633
1.80 0.9281 0.0359 0.0719 0.4641
1.81 0.9297 0.0351 0.0703 0.4649
1.82 0.9312 0.0344 0.0688 0.4656
1.83 0.9328 0.0336 0.0672 0.4664
1.84 0.9342 0.0329 0.0658 0.4671
1.85 0.9357 0.0322 0.0643 0.4678
1.86 0.9371 0.0314 0.0629 0.4686
1.87 0.9385 0.0307 0.0615 0.4693
1.88 0.9399 0.0301 0.0601 0.4699
1.89 0.9412 0.0294 0.0588 0.4706
1.90 0.9426 0.0287 0.0574 0.4713
1.91 0.9439 0.0281 0.0561 0.4719
1.92 0.9451 0.0274 0.0549 0.4726
1.93 0.9464 0.0268 0.0536 0.4732
1.94 0.9476 0.0262 0.0524 0.4738
1.95 0.9488 0.0256 0.0512 0.4744
1.96 0.9500 0.0250 0.0500 0.4750
1.97 0.9512 0.0244 0.0488 0.4756
1.98 0.9523 0.0239 0.0477 0.4761
1.99 0.9534 0.0233 0.0466 0.4767
2.00 0.9545 0.0228 0.0455 0.4772
2.01 0.9556 0.0222 0.0444 0.4778
2.02 0.9566 0.0217 0.0434 0.4783
2.03 0.9576 0.0212 0.0424 0.4788
2.04 0.9586 0.0207 0.0414 0.4793
2.05 0.9596 0.0202 0.0404 0.4798
2.06 0.9606 0.0197 0.0394 0.4803
2.07 0.9615 0.0192 0.0385 0.4808
-303-
Back to TOC
2.08 0.9625 0.0188 0.0375 0.4812
2.09 0.9634 0.0183 0.0366 0.4817
2.10 0.9643 0.0179 0.0357 0.4821
2.11 0.9651 0.0174 0.0349 0.4826
2.12 0.9660 0.0170 0.0340 0.4830
2.13 0.9668 0.0166 0.0332 0.4834
2.14 0.9676 0.0162 0.0324 0.4838
2.15 0.9684 0.0158 0.0316 0.4842
2.16 0.9692 0.0154 0.0308 0.4846
2.17 0.9700 0.0150 0.0300 0.4850
2.18 0.9707 0.0146 0.0293 0.4854
2.19 0.9715 0.0143 0.0285 0.4857
2.20 0.9722 0.0139 0.0278 0.4861
2.21 0.9729 0.0136 0.0271 0.4864
2.22 0.9736 0.0132 0.0264 0.4868
2.23 0.9743 0.0129 0.0257 0.4871
2.24 0.9749 0.0125 0.0251 0.4875
2.25 0.9756 0.0122 0.0244 0.4878
2.26 0.9762 0.0119 0.0238 0.4881
2.27 0.9768 0.0116 0.0232 0.4884
2.28 0.9774 0.0113 0.0226 0.4887
2.29 0.9780 0.0110 0.0220 0.4890
2.30 0.9786 0.0107 0.0214 0.4893
2.31 0.9791 0.0104 0.0209 0.4896
2.32 0.9797 0.0102 0.0203 0.4898
2.33 0.9802 0.0099 0.0198 0.4901
2.34 0.9807 0.0096 0.0193 0.4904
2.35 0.9812 0.0094 0.0188 0.4906
2.36 0.9817 0.0091 0.0183 0.4909
2.37 0.9822 0.0089 0.0178 0.4911
2.38 0.9827 0.0087 0.0173 0.4913
2.39 0.9832 0.0084 0.0168 0.4916
2.40 0.9836 0.0082 0.0164 0.4918
2.41 0.9840 0.0080 0.0160 0.4920
2.42 0.9845 0.0078 0.0155 0.4922
2.43 0.9849 0.0075 0.0151 0.4925
2.44 0.9853 0.0073 0.0147 0.4927
2.45 0.9857 0.0071 0.0143 0.4929
2.46 0.9861 0.0069 0.0139 0.4931
2.47 0.9865 0.0068 0.0135 0.4932
2.48 0.9869 0.0066 0.0131 0.4934
2.49 0.9872 0.0064 0.0128 0.4936
2.50 0.9876 0.0062 0.0124 0.4938
2.51 0.9879 0.0060 0.0121 0.4940
2.52 0.9883 0.0059 0.0117 0.4941
2.53 0.9886 0.0057 0.0114 0.4943
2.54 0.9889 0.0055 0.0111 0.4945
2.55 0.9892 0.0054 0.0108 0.4946
2.56 0.9895 0.0052 0.0105 0.4948
2.57 0.9898 0.0051 0.0102 0.4949
2.58 0.9901 0.0049 0.0099 0.4951
2.59 0.9904 0.0048 0.0096 0.4952
2.60 0.9907 0.0047 0.0093 0.4953
2.61 0.9909 0.0045 0.0091 0.4955
2.62 0.9912 0.0044 0.0088 0.4956
2.63 0.9915 0.0043 0.0085 0.4957
2.64 0.9917 0.0041 0.0083 0.4959
-304-
Back to TOC
2.65 0.9920 0.0040 0.0080 0.4960
2.66 0.9922 0.0039 0.0078 0.4961
2.67 0.9924 0.0038 0.0076 0.4962
2.68 0.9926 0.0037 0.0074 0.4963
2.69 0.9929 0.0036 0.0071 0.4964
2.70 0.9931 0.0035 0.0069 0.4965
2.71 0.9933 0.0034 0.0067 0.4966
2.72 0.9935 0.0033 0.0065 0.4967
2.73 0.9937 0.0032 0.0063 0.4968
2.74 0.9939 0.0031 0.0061 0.4969
2.75 0.9940 0.0030 0.0060 0.4970
2.76 0.9942 0.0029 0.0058 0.4971
2.77 0.9944 0.0028 0.0056 0.4972
2.78 0.9946 0.0027 0.0054 0.4973
2.79 0.9947 0.0026 0.0053 0.4974
2.80 0.9949 0.0026 0.0051 0.4974
2.81 0.9950 0.0025 0.0050 0.4975
2.82 0.9952 0.0024 0.0048 0.4976
2.83 0.9953 0.0023 0.0047 0.4977
2.84 0.9955 0.0023 0.0045 0.4977
2.85 0.9956 0.0022 0.0044 0.4978
2.86 0.9958 0.0021 0.0042 0.4979
2.87 0.9959 0.0021 0.0041 0.4979
2.88 0.9960 0.0020 0.0040 0.4980
2.89 0.9961 0.0019 0.0039 0.4981
2.90 0.9963 0.0019 0.0037 0.4981
2.91 0.9964 0.0018 0.0036 0.4982
2.92 0.9965 0.0018 0.0035 0.4982
2.93 0.9966 0.0017 0.0034 0.4983
2.94 0.9967 0.0016 0.0033 0.4984
2.95 0.9968 0.0016 0.0032 0.4984
2.96 0.9969 0.0015 0.0031 0.4985
2.97 0.9970 0.0015 0.0030 0.4985
2.98 0.9971 0.0014 0.0029 0.4986
2.99 0.9972 0.0014 0.0028 0.4986
3.00 0.9973 0.0013 0.0027 0.4987
3.01 0.9974 0.0013 0.0026 0.4987
3.02 0.9975 0.0013 0.0025 0.4987
3.03 0.9976 0.0012 0.0024 0.4988
3.04 0.9976 0.0012 0.0024 0.4988
3.05 0.9977 0.0011 0.0023 0.4989
3.06 0.9978 0.0011 0.0022 0.4989
3.07 0.9979 0.0011 0.0021 0.4989
3.08 0.9979 0.0010 0.0021 0.4990
3.09 0.9980 0.0010 0.0020 0.4990
3.10 0.9981 0.0010 0.0019 0.4990
3.11 0.9981 0.0009 0.0019 0.4991
3.12 0.9982 0.0009 0.0018 0.4991
3.13 0.9983 0.0009 0.0017 0.4991
3.14 0.9983 0.0008 0.0017 0.4992
3.15 0.9984 0.0008 0.0016 0.4992
3.16 0.9984 0.0008 0.0016 0.4992
3.17 0.9985 0.0008 0.0015 0.4992
3.18 0.9985 0.0007 0.0015 0.4993
3.19 0.9986 0.0007 0.0014 0.4993
3.20 0.9986 0.0007 0.0014 0.4993
3.21 0.9987 0.0007 0.0013 0.4993
-305-
Back to TOC
3.22 0.9987 0.0006 0.0013 0.4994
3.23 0.9988 0.0006 0.0012 0.4994
3.24 0.9988 0.0006 0.0012 0.4994
3.25 0.9988 0.0006 0.0012 0.4994
3.30 0.9990 0.0005 0.0010 0.4995
3.35 0.9992 0.0004 0.0008 0.4996
3.50 0.9995 0.0002 0.0005 0.4998
3.60 0.9997 0.0002 0.0003 0.4998
3.70 0.9998 0.0001 0.0002 0.4999
3.80 0.9999 0.0001 0.0001 0.4999
3.90 0.9999 0.0000 0.0001 0.5000
4.00 0.9999 0.0000 0.0001 0.5000
-306-
Back to TOC
Table 2: Probabilities Under the t-Distribution
Note: The probabilities are associated with non-directional, two-tailed hypothesis-tests. For directional,
one-tailed hypothesis tests divide the probability obtained from the table in half.
df
t 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 22 24 26 28 30 40 60 100 200 1000
0.001.0001.0001.0001.0001.0001.0001.0001.0001.0001.0001.0001.0001.0001.0001.0001.0001.0001.0001.0001.0001.0001.0001.0001.0001.0001.0001.0001.0001.0001.000
0.20.8743.8600.8543.8512.8494.8481.8472.8465.8459.8455.8451.8448.8446.8444.8442.8440.8439.8437.8436.8435.8433.8432.8430.8429.8428.8425.8422.8419.8417.8415
0.40.7578.7278.7159.7096.7057.7030.7011.6996.6985.6976.6968.6962.6957.6952.6948.6944.6941.6939.6936.6934.6930.6927.6924.6922.6920.6913.6906.6900.6896.6892
0.60.6560.6094.5908.5808.5747.5705.5674.5651.5633.5619.5607.5597.5588.5581.5575.5569.5564.5560.5556.5552.5546.5541.5537.5533.5530.5519.5508.5499.5492.5486
0.80.5704.5076.4822.4685.4600.4542.4500.4468.4443.4423.4406.4393.4381.4371.4362.4354.4347.4341.4336.4331.4323.4316.4310.4304.4300.4284.4269.4256.4247.4239
1.00.5000.4226.3910.3739.3632.3559.3506.3466.3434.3409.3388.3370.3356.3343.3332.3322.3313.3306.3299.3293.3282.3273.3265.3259.3253.3233.3213.3197.3185.3176
1.10.4697.3860.3517.3331.3215.3135.3077.3033.2999.2971.2948.2929.2913.2899.2887.2876.2867.2858.2851.2844.2832.2822.2814.2807.2801.2779.2757.2740.2727.2716
1.20.4423.3530.3163.2964.2839.2754.2692.2645.2608.2578.2553.2533.2516.2501.2487.2476.2466.2457.2449.2442.2429.2419.2410.2402.2395.2372.2349.2330.2316.2304
1.30.4174.3232.2845.2635.2503.2413.2348.2298.2259.2228.2202.2180.2162.2146.2132.2120.2110.2100.2092.2084.2071.2059.2050.2042.2035.2010.1986.1966.1951.1939
1.40.3949.2965.2560.2341.2204.2110.2042.1991.1950.1918.1891.1868.1849.1833.1819.1806.1795.1785.1776.1768.1755.1743.1733.1725.1718.1692.1667.1646.1631.1618
1.50.3743.2724.2306.2080.1939.1843.1773.1720.1679.1645.1618.1595.1575.1558.1544.1531.1520.1510.1500.1492.1478.1467.1457.1448.1441.1415.1389.1368.1352.1339
1.51.3724.2701.2282.2056.1914.1818.1748.1695.1653.1620.1592.1569.1550.1533.1518.1505.1494.1484.1475.1467.1453.1441.1431.1422.1415.1389.1363.1342.1326.1314
1.52.3705.2679.2258.2031.1890.1793.1723.1670.1628.1595.1567.1544.1525.1508.1493.1480.1469.1459.1450.1442.1428.1416.1406.1397.1390.1364.1338.1317.1301.1288
1.53.3685.2657.2235.2008.1866.1769.1699.1645.1604.1570.1543.1519.1500.1483.1468.1455.1444.1434.1425.1417.1403.1391.1381.1372.1365.1339.1313.1292.1276.1263
1.54.3666.2635.2212.1984.1842.1745.1675.1621.1579.1546.1518.1495.1475.1459.1444.1431.1420.1410.1400.1392.1378.1366.1356.1348.1340.1314.1288.1267.1251.1239
1.55.3648.2613.2189.1961.1818.1721.1651.1597.1556.1522.1494.1471.1451.1434.1420.1407.1396.1385.1376.1368.1354.1342.1332.1324.1316.1290.1264.1243.1227.1215
1.56.3629.2591.2167.1938.1795.1698.1627.1574.1532.1498.1470.1447.1428.1411.1396.1383.1372.1362.1353.1344.1330.1319.1309.1300.1292.1266.1240.1219.1203.1191
1.57.3611.2570.2144.1915.1772.1675.1604.1551.1509.1475.1447.1424.1404.1387.1373.1360.1348.1338.1329.1321.1307.1295.1285.1276.1269.1243.1217.1196.1180.1167
1.58.3592.2549.2122.1893.1749.1652.1581.1528.1486.1452.1424.1401.1381.1364.1350.1337.1325.1315.1306.1298.1284.1272.1262.1253.1246.1220.1194.1173.1157.1144
1.59.3574.2528.2101.1870.1727.1629.1559.1505.1463.1429.1401.1378.1358.1342.1327.1314.1303.1292.1283.1275.1261.1249.1239.1231.1223.1197.1171.1150.1134.1122
1.60.3556.2507.2079.1848.1705.1607.1536.1483.1441.1407.1379.1356.1336.1319.1304.1292.1280.1270.1261.1253.1239.1227.1217.1208.1201.1175.1149.1128.1112.1099
1.61.3538.2487.2058.1827.1683.1585.1514.1461.1419.1385.1357.1334.1314.1297.1282.1269.1258.1248.1239.1231.1217.1205.1195.1186.1179.1153.1126.1106.1090.1077
1.62.3521.2467.2037.1805.1662.1564.1493.1439.1397.1363.1335.1312.1292.1275.1261.1248.1236.1226.1217.1209.1195.1183.1173.1164.1157.1131.1105.1084.1068.1055
1.63.3503.2447.2016.1784.1640.1542.1471.1417.1375.1342.1314.1291.1271.1254.1239.1226.1215.1205.1196.1188.1173.1162.1152.1143.1136.1110.1083.1062.1047.1034
1.64.3486.2427.1995.1763.1619.1521.1450.1396.1354.1320.1293.1269.1250.1233.1218.1205.1194.1184.1175.1166.1152.1140.1130.1122.1115.1088.1062.1041.1026.1013
1.65.3469.2407.1975.1743.1599.1500.1429.1376.1333.1300.1272.1249.1229.1212.1197.1184.1173.1163.1154.1146.1131.1120.1110.1101.1094.1068.1042.1021.1005.0993
1.66.3452.2388.1955.1723.1578.1480.1409.1355.1313.1279.1251.1228.1208.1191.1177.1164.1152.1142.1133.1125.1111.1099.1089.1081.1073.1047.1021.1000.0985.0972
1.67.3435.2369.1935.1702.1558.1460.1388.1335.1293.1259.1231.1208.1188.1171.1156.1144.1132.1122.1113.1105.1091.1079.1069.1061.1053.1027.1001.0980.0965.0952
1.68.3418.2350.1915.1683.1538.1440.1368.1315.1273.1239.1211.1188.1168.1151.1137.1124.1112.1102.1093.1085.1071.1059.1049.1041.1033.1008.0982.0961.0945.0933
1.69.3401.2331.1896.1663.1518.1420.1349.1295.1253.1219.1191.1168.1149.1132.1117.1104.1093.1083.1074.1066.1052.1040.1030.1021.1014.0988.0962.0941.0926.0913
1.70.3385.2312.1877.1644.1499.1400.1329.1276.1233.1200.1172.1149.1129.1112.1098.1085.1074.1063.1054.1046.1032.1021.1011.1002.0995.0969.0943.0922.0907.0894
1.71.3369.2294.1858.1624.1480.1381.1310.1256.1214.1181.1153.1130.1110.1093.1079.1066.1055.1044.1035.1027.1013.1002.0992.0983.0976.0950.0924.0904.0888.0876
1.72.3353.2276.1839.1606.1461.1362.1291.1237.1195.1162.1134.1111.1091.1075.1060.1047.1036.1026.1017.1009.0995.0983.0973.0965.0957.0932.0906.0885.0870.0857
1.73.3337.2258.1821.1587.1442.1344.1273.1219.1177.1143.1115.1092.1073.1056.1041.1029.1017.1007.0998.0990.0976.0965.0955.0946.0939.0913.0888.0867.0852.0839
1.74.3321.2240.1802.1568.1423.1325.1254.1200.1159.1125.1097.1074.1055.1038.1023.1011.0999.0989.0980.0972.0958.0947.0937.0928.0921.0895.0870.0849.0834.0822
1.75.3305.2222.1784.1550.1405.1307.1236.1182.1140.1107.1079.1056.1037.1020.1005.0993.0981.0971.0963.0954.0941.0929.0919.0911.0903.0878.0852.0832.0817.0804
1.76.3289.2205.1766.1532.1387.1289.1218.1164.1123.1089.1061.1038.1019.1002.0988.0975.0964.0954.0945.0937.0923.0911.0902.0893.0886.0861.0835.0815.0799.0787
1.77.3274.2187.1749.1514.1369.1271.1200.1147.1105.1072.1044.1021.1002.0985.0970.0958.0947.0937.0928.0920.0906.0894.0885.0876.0869.0843.0818.0798.0783.0770
1.78.3259.2170.1731.1497.1352.1254.1183.1130.1088.1054.1027.1004.0984.0968.0953.0941.0930.0920.0911.0903.0889.0877.0868.0859.0852.0827.0801.0781.0766.0754
1.79.3243.2153.1714.1480.1335.1237.1166.1112.1071.1037.1010.0987.0968.0951.0937.0924.0913.0903.0894.0886.0872.0861.0851.0843.0836.0810.0785.0765.0750.0738
1.80.3228.2137.1697.1462.1318.1220.1149.1096.1054.1021.0993.0970.0951.0934.0920.0907.0896.0886.0878.0870.0856.0844.0835.0826.0819.0794.0769.0749.0734.0722
1.81.3213.2120.1680.1445.1301.1203.1132.1079.1037.1004.0977.0954.0935.0918.0904.0891.0880.0870.0861.0853.0840.0828.0819.0810.0803.0778.0753.0733.0718.0706
1.82.3199.2104.1663.1429.1284.1186.1116.1063.1021.0988.0961.0938.0919.0902.0888.0875.0864.0854.0846.0838.0824.0813.0803.0795.0788.0762.0737.0718.0703.0691
1.83.3184.2087.1647.1412.1268.1170.1099.1046.1005.0972.0945.0922.0903.0886.0872.0859.0848.0839.0830.0822.0808.0797.0787.0779.0772.0747.0722.0702.0687.0675
1.84.3169.2071.1630.1396.1251.1154.1083.1030.0989.0956.0929.0906.0887.0871.0856.0844.0833.0823.0814.0807.0793.0782.0772.0764.0757.0732.0707.0687.0673.0661
1.85.3155.2055.1614.1380.1235.1138.1068.1015.0974.0940.0913.0891.0872.0855.0841.0829.0818.0808.0799.0791.0778.0767.0757.0749.0742.0717.0692.0673.0658.0646
1.86.3140.2040.1598.1364.1220.1122.1052.0999.0958.0925.0898.0876.0857.0840.0826.0814.0803.0793.0784.0777.0763.0752.0742.0734.0727.0703.0678.0658.0644.0632
1.87.3126.2024.1583.1348.1204.1107.1037.0984.0943.0910.0883.0861.0842.0825.0811.0799.0788.0778.0770.0762.0749.0737.0728.0720.0713.0688.0664.0644.0629.0618
1.88.3112.2009.1567.1333.1189.1092.1022.0969.0928.0895.0868.0846.0827.0811.0797.0784.0774.0764.0755.0748.0734.0723.0714.0706.0699.0674.0650.0630.0616.0604
1.89.3098.1993.1552.1318.1174.1077.1007.0954.0913.0881.0854.0831.0813.0796.0782.0770.0759.0750.0741.0733.0720.0709.0700.0692.0685.0660.0636.0617.0602.0590
1.90.3084.1978.1536.1302.1159.1062.0992.0940.0899.0866.0840.0817.0798.0782.0768.0756.0745.0736.0727.0719.0706.0695.0686.0678.0671.0647.0622.0603.0589.0577
1.91.3071.1963.1521.1287.1144.1047.0978.0925.0885.0852.0825.0803.0784.0768.0754.0742.0732.0722.0713.0706.0693.0682.0672.0664.0657.0633.0609.0590.0576.0564
1.92.3057.1948.1506.1273.1129.1033.0963.0911.0871.0838.0812.0789.0771.0755.0741.0729.0718.0708.0700.0692.0679.0668.0659.0651.0644.0620.0596.0577.0563.0551
1.93.3043.1934.1492.1258.1115.1018.0949.0897.0857.0824.0798.0776.0757.0741.0727.0715.0705.0695.0687.0679.0666.0655.0646.0638.0631.0607.0583.0564.0550.0539
1.94.3030.1919.1477.1244.1101.1004.0935.0883.0843.0811.0784.0762.0744.0728.0714.0702.0692.0682.0674.0666.0653.0642.0633.0625.0618.0595.0571.0552.0538.0527
1.95.3017.1905.1463.1230.1087.0991.0922.0870.0830.0797.0771.0749.0731.0715.0701.0689.0679.0669.0661.0653.0640.0630.0620.0613.0606.0582.0559.0540.0526.0515
1.96.3003.1891.1449.1216.1073.0977.0908.0857.0816.0784.0758.0736.0718.0702.0688.0677.0666.0657.0648.0641.0628.0617.0608.0600.0593.0570.0546.0528.0514.0503
1.97.2990.1876.1434.1202.1059.0964.0895.0843.0803.0771.0745.0724.0705.0689.0676.0664.0654.0644.0636.0628.0616.0605.0596.0588.0581.0558.0535.0516.0502.0491
-307-
Back to TOC
df
t 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 22 24 26 28 30 40 60 100 200 1000

1.98.2977.1863.1421.1188.1046.0950.0882.0830.0791.0759.0733.0711.0693.0677.0664.0652.0641.0632.0624.0616.0603.0593.0584.0576.0569.0546.0523.0505.0491.0480
1.99.2964.1849.1407.1175.1033.0937.0869.0818.0778.0746.0720.0699.0680.0665.0651.0640.0629.0620.0612.0604.0592.0581.0572.0564.0558.0535.0512.0493.0480.0469
2.00.2952.1835.1393.1161.1019.0924.0856.0805.0766.0734.0708.0687.0668.0653.0639.0628.0617.0608.0600.0593.0580.0569.0560.0553.0546.0523.0500.0482.0469.0458
2.01.2939.1821.1380.1148.1006.0912.0844.0793.0753.0722.0696.0675.0657.0641.0628.0616.0606.0597.0588.0581.0569.0558.0549.0542.0535.0512.0489.0471.0458.0447
2.02.2926.1808.1367.1135.0994.0899.0831.0781.0741.0710.0684.0663.0645.0629.0616.0605.0594.0585.0577.0570.0557.0547.0538.0530.0524.0501.0479.0461.0447.0436
2.03.2914.1795.1353.1122.0981.0887.0819.0769.0729.0698.0673.0651.0633.0618.0605.0593.0583.0574.0566.0559.0546.0536.0527.0520.0513.0490.0468.0450.0437.0426
2.04.2902.1782.1340.1109.0969.0874.0807.0757.0718.0686.0661.0640.0622.0607.0594.0582.0572.0563.0555.0548.0535.0525.0516.0509.0502.0480.0458.0440.0427.0416
2.05.2889.1769.1328.1097.0956.0862.0795.0745.0706.0675.0650.0629.0611.0596.0583.0571.0561.0552.0544.0537.0525.0514.0506.0498.0492.0470.0447.0430.0417.0406
2.06.2877.1756.1315.1085.0944.0851.0784.0734.0695.0664.0639.0618.0600.0585.0572.0561.0550.0542.0534.0527.0514.0504.0495.0488.0482.0459.0437.0420.0407.0397
2.07.2865.1743.1302.1072.0932.0839.0772.0722.0684.0653.0628.0607.0589.0574.0561.0550.0540.0531.0523.0516.0504.0494.0485.0478.0472.0450.0428.0410.0397.0387
2.08.2853.1730.1290.1060.0921.0827.0761.0711.0673.0642.0617.0596.0579.0564.0551.0540.0530.0521.0513.0506.0494.0484.0475.0468.0462.0440.0418.0401.0388.0378
2.09.2841.1718.1278.1048.0909.0816.0750.0700.0662.0631.0606.0586.0568.0553.0541.0529.0520.0511.0503.0496.0484.0474.0465.0458.0452.0430.0409.0392.0379.0369
2.10.2829.1705.1266.1037.0898.0805.0739.0689.0651.0621.0596.0575.0558.0543.0531.0519.0510.0501.0493.0486.0474.0464.0456.0449.0442.0421.0399.0382.0370.0360
2.11.2818.1693.1254.1025.0886.0794.0728.0679.0641.0610.0586.0565.0548.0533.0521.0510.0500.0491.0483.0477.0465.0455.0446.0439.0433.0412.0390.0374.0361.0351
2.12.2806.1681.1242.1013.0875.0783.0717.0668.0630.0600.0576.0555.0538.0524.0511.0500.0490.0482.0474.0467.0455.0445.0437.0430.0424.0403.0382.0365.0352.0343
2.13.2794.1669.1230.1002.0864.0772.0707.0658.0620.0590.0566.0546.0528.0514.0501.0490.0481.0472.0465.0458.0446.0436.0428.0421.0415.0394.0373.0356.0344.0334
2.14.2783.1657.1219.0991.0853.0762.0696.0648.0610.0580.0556.0536.0519.0504.0492.0481.0472.0463.0455.0449.0437.0427.0419.0412.0406.0385.0364.0348.0336.0326
2.15.2772.1645.1207.0980.0842.0751.0686.0638.0600.0571.0546.0526.0510.0495.0483.0472.0462.0454.0446.0440.0428.0418.0410.0403.0397.0377.0356.0340.0328.0318
2.16.2760.1634.1196.0969.0832.0741.0676.0628.0591.0561.0537.0517.0500.0486.0474.0463.0453.0445.0438.0431.0419.0410.0402.0395.0389.0368.0348.0332.0320.0310
2.17.2749.1622.1185.0958.0821.0731.0666.0618.0581.0552.0528.0508.0491.0477.0465.0454.0445.0436.0429.0422.0411.0401.0393.0386.0381.0360.0340.0324.0312.0302
2.18.2738.1611.1173.0947.0811.0721.0656.0609.0572.0542.0519.0499.0482.0468.0456.0445.0436.0428.0420.0414.0402.0393.0385.0378.0372.0352.0332.0316.0304.0295
2.19.2727.1599.1163.0937.0801.0711.0647.0599.0563.0533.0510.0490.0474.0460.0447.0437.0428.0419.0412.0405.0394.0385.0377.0370.0364.0344.0324.0308.0297.0288
2.20.2716.1588.1152.0927.0791.0701.0637.0590.0553.0524.0501.0481.0465.0451.0439.0428.0419.0411.0404.0397.0386.0377.0369.0362.0356.0336.0317.0301.0290.0280
2.21.2705.1577.1141.0916.0781.0691.0628.0581.0544.0516.0492.0473.0457.0443.0431.0420.0411.0403.0396.0389.0378.0369.0361.0354.0349.0329.0309.0294.0282.0273
2.22.2694.1566.1130.0906.0771.0682.0619.0572.0536.0507.0484.0464.0448.0434.0422.0412.0403.0395.0388.0381.0370.0361.0353.0347.0341.0322.0302.0287.0275.0266
2.23.2684.1555.1120.0896.0762.0673.0610.0563.0527.0498.0475.0456.0440.0426.0414.0404.0395.0387.0380.0374.0363.0354.0346.0339.0334.0314.0295.0280.0269.0260
2.24.2673.1544.1110.0886.0752.0663.0601.0554.0518.0490.0467.0448.0432.0418.0407.0396.0387.0379.0372.0366.0355.0346.0339.0332.0327.0307.0288.0273.0262.0253
2.25.2662.1534.1099.0876.0743.0654.0592.0546.0510.0482.0459.0440.0424.0411.0399.0389.0380.0372.0365.0359.0348.0339.0331.0325.0319.0300.0281.0266.0255.0247
2.26.2652.1523.1089.0867.0734.0645.0583.0537.0502.0474.0451.0432.0416.0403.0391.0381.0372.0365.0358.0351.0341.0332.0324.0318.0312.0293.0275.0260.0249.0240
2.27.2642.1512.1079.0857.0724.0637.0575.0529.0494.0466.0443.0424.0409.0395.0384.0374.0365.0357.0350.0344.0334.0325.0317.0311.0306.0287.0268.0254.0243.0234
2.28.2631.1502.1069.0848.0715.0628.0566.0521.0486.0458.0435.0417.0401.0388.0377.0367.0358.0350.0343.0337.0327.0318.0311.0304.0299.0280.0262.0247.0237.0228
2.29.2621.1492.1060.0839.0706.0619.0558.0513.0478.0450.0428.0409.0394.0381.0369.0359.0351.0343.0336.0330.0320.0311.0304.0298.0292.0274.0256.0241.0231.0222
2.30.2611.1481.1050.0829.0698.0611.0550.0505.0470.0443.0420.0402.0387.0374.0362.0352.0344.0336.0330.0323.0313.0304.0297.0291.0286.0267.0249.0235.0225.0217
2.31.2601.1471.1040.0820.0689.0603.0542.0497.0462.0435.0413.0395.0379.0366.0355.0346.0337.0330.0323.0317.0307.0298.0291.0285.0279.0261.0243.0229.0219.0211
2.32.2591.1461.1031.0811.0681.0594.0534.0489.0455.0428.0406.0388.0372.0360.0348.0339.0330.0323.0316.0310.0300.0292.0285.0279.0273.0255.0238.0224.0213.0205
2.33.2581.1451.1021.0803.0672.0586.0526.0482.0447.0421.0399.0381.0366.0353.0342.0332.0324.0316.0310.0304.0294.0285.0278.0272.0267.0249.0232.0218.0208.0200
2.34.2571.1442.1012.0794.0664.0578.0518.0474.0440.0413.0392.0374.0359.0346.0335.0326.0317.0310.0304.0298.0288.0279.0272.0266.0261.0244.0226.0213.0203.0195
2.35.2561.1432.1003.0785.0656.0571.0511.0467.0433.0406.0385.0367.0352.0340.0329.0319.0311.0304.0297.0292.0282.0273.0266.0261.0255.0238.0221.0207.0197.0190
2.36.2552.1422.0994.0777.0648.0563.0503.0460.0426.0400.0378.0361.0346.0333.0322.0313.0305.0298.0291.0285.0276.0267.0261.0255.0250.0232.0215.0202.0192.0185
2.37.2542.1413.0985.0768.0640.0555.0496.0452.0419.0393.0372.0354.0339.0327.0316.0307.0299.0292.0285.0280.0270.0262.0255.0249.0244.0227.0210.0197.0187.0180
2.38.2532.1403.0976.0760.0632.0548.0489.0445.0412.0386.0365.0348.0333.0321.0310.0301.0293.0286.0279.0274.0264.0256.0249.0244.0239.0222.0205.0192.0183.0175
2.39.2523.1394.0967.0752.0624.0540.0482.0439.0406.0380.0359.0341.0327.0315.0304.0295.0287.0280.0274.0268.0258.0251.0244.0238.0233.0216.0200.0187.0178.0170
2.40.2513.1385.0959.0744.0616.0533.0475.0432.0399.0373.0352.0335.0321.0309.0298.0289.0281.0274.0268.0262.0253.0245.0239.0233.0228.0211.0195.0182.0173.0166
2.41.2504.1375.0950.0736.0609.0526.0468.0425.0392.0367.0346.0329.0315.0303.0292.0283.0276.0269.0263.0257.0248.0240.0233.0228.0223.0206.0190.0178.0169.0161
2.42.2495.1366.0942.0728.0601.0519.0461.0419.0386.0361.0340.0323.0309.0297.0287.0278.0270.0263.0257.0252.0242.0235.0228.0223.0218.0202.0186.0173.0164.0157
2.43.2485.1357.0933.0720.0594.0512.0454.0412.0380.0355.0334.0317.0303.0291.0281.0272.0265.0258.0252.0246.0237.0229.0223.0218.0213.0197.0181.0169.0160.0153
2.44.2476.1348.0925.0712.0587.0505.0448.0406.0374.0348.0328.0312.0298.0286.0276.0267.0259.0253.0247.0241.0232.0224.0218.0213.0208.0192.0177.0164.0156.0149
2.45.2467.1339.0917.0704.0579.0498.0441.0399.0368.0343.0322.0306.0292.0280.0270.0262.0254.0247.0241.0236.0227.0220.0213.0208.0203.0188.0172.0160.0151.0145
2.46.2458.1331.0909.0697.0572.0491.0435.0393.0362.0337.0317.0300.0287.0275.0265.0257.0249.0242.0236.0231.0222.0215.0209.0203.0199.0183.0168.0156.0147.0141
2.47.2449.1322.0901.0689.0565.0485.0428.0387.0356.0331.0311.0295.0281.0270.0260.0251.0244.0237.0232.0226.0217.0210.0204.0199.0194.0179.0164.0152.0143.0137
2.48.2440.1313.0893.0682.0558.0478.0422.0381.0350.0325.0306.0290.0276.0265.0255.0246.0239.0233.0227.0222.0213.0205.0199.0194.0190.0174.0160.0148.0140.0133
2.49.2431.1305.0885.0675.0552.0472.0416.0375.0344.0320.0300.0284.0271.0260.0250.0242.0234.0228.0222.0217.0208.0201.0195.0190.0185.0170.0156.0144.0136.0129
2.50.2422.1296.0877.0668.0545.0465.0410.0369.0339.0314.0295.0279.0266.0255.0245.0237.0229.0223.0217.0212.0204.0197.0191.0186.0181.0166.0152.0140.0132.0126
2.51.2414.1288.0869.0661.0538.0459.0404.0364.0333.0309.0290.0274.0261.0250.0240.0232.0225.0218.0213.0208.0199.0192.0186.0181.0177.0162.0148.0137.0129.0122
2.52.2405.1279.0862.0654.0532.0453.0398.0358.0328.0304.0285.0269.0256.0245.0236.0227.0220.0214.0208.0203.0195.0188.0182.0177.0173.0158.0144.0133.0125.0119
2.53.2396.1271.0854.0647.0525.0447.0392.0353.0322.0299.0280.0264.0251.0240.0231.0223.0216.0210.0204.0199.0191.0184.0178.0173.0169.0154.0140.0130.0122.0116
2.54.2388.1263.0847.0640.0519.0441.0387.0347.0317.0294.0275.0259.0247.0236.0226.0218.0211.0205.0200.0195.0187.0180.0174.0169.0165.0151.0137.0126.0118.0112
2.55.2379.1255.0839.0633.0513.0435.0381.0342.0312.0289.0270.0255.0242.0231.0222.0214.0207.0201.0196.0191.0183.0176.0170.0165.0161.0147.0133.0123.0115.0109
2.56.2371.1247.0832.0626.0506.0429.0376.0336.0307.0284.0265.0250.0237.0227.0218.0210.0203.0197.0191.0187.0179.0172.0166.0162.0157.0143.0130.0120.0112.0106
2.57.2362.1239.0825.0620.0500.0423.0370.0331.0302.0279.0261.0245.0233.0222.0213.0206.0199.0193.0187.0183.0175.0168.0163.0158.0154.0140.0127.0116.0109.0103
2.58.2354.1231.0818.0613.0494.0418.0365.0326.0297.0274.0256.0241.0229.0218.0209.0201.0195.0189.0184.0179.0171.0164.0159.0154.0150.0137.0123.0113.0106.0100
2.59.2346.1223.0811.0607.0488.0412.0359.0321.0292.0269.0251.0237.0224.0214.0205.0197.0191.0185.0180.0175.0167.0161.0155.0151.0147.0133.0120.0110.0103.0097
2.60.2338.1215.0804.0600.0482.0407.0354.0316.0287.0265.0247.0232.0220.0210.0201.0193.0187.0181.0176.0171.0163.0157.0152.0147.0143.0130.0117.0107.0100.0095
2.61.2329.1208.0797.0594.0477.0401.0349.0311.0283.0260.0243.0228.0216.0206.0197.0190.0183.0177.0172.0168.0160.0154.0148.0144.0140.0127.0114.0104.0097.0092
2.62.2321.1200.0790.0588.0471.0396.0344.0307.0278.0256.0238.0224.0212.0202.0193.0186.0179.0174.0168.0164.0156.0150.0145.0140.0137.0124.0111.0102.0095.0089
-308-
Back to TOC
df
t 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 22 24 26 28 30 40 60 100 200 1000

2.63.2313.1193.0783.0582.0465.0391.0339.0302.0274.0252.0234.0220.0208.0198.0189.0182.0176.0170.0165.0160.0153.0147.0142.0137.0133.0121.0108.0099.0092.0087
2.64.2305.1185.0777.0576.0460.0385.0334.0297.0269.0247.0230.0216.0204.0194.0186.0178.0172.0166.0161.0157.0150.0143.0138.0134.0130.0118.0106.0096.0089.0084
2.65.2297.1178.0770.0570.0454.0380.0329.0293.0265.0243.0226.0212.0200.0190.0182.0175.0168.0163.0158.0154.0146.0140.0135.0131.0127.0115.0103.0094.0087.0082
2.66.2289.1170.0763.0564.0449.0375.0325.0288.0260.0239.0222.0208.0196.0187.0178.0171.0165.0159.0155.0150.0143.0137.0132.0128.0124.0112.0100.0091.0084.0079
2.67.2281.1163.0757.0558.0443.0370.0320.0284.0256.0235.0218.0204.0193.0183.0175.0168.0162.0156.0151.0147.0140.0134.0129.0125.0121.0109.0097.0089.0082.0077
2.68.2274.1156.0750.0552.0438.0365.0315.0279.0252.0231.0214.0200.0189.0179.0171.0164.0158.0153.0148.0144.0137.0131.0126.0122.0118.0106.0095.0086.0080.0075
2.69.2266.1149.0744.0547.0433.0361.0311.0275.0248.0227.0210.0197.0185.0176.0168.0161.0155.0150.0145.0141.0134.0128.0123.0119.0116.0104.0092.0084.0077.0073
2.70.2258.1142.0738.0541.0428.0356.0306.0271.0244.0223.0207.0193.0182.0173.0165.0158.0152.0147.0142.0138.0131.0125.0120.0116.0113.0101.0090.0081.0075.0071
2.71.2250.1135.0732.0535.0423.0351.0302.0267.0240.0219.0203.0190.0179.0169.0161.0155.0149.0143.0139.0135.0128.0122.0118.0114.0110.0099.0088.0079.0073.0068
2.72.2243.1128.0725.0530.0418.0346.0298.0262.0236.0216.0199.0186.0175.0166.0158.0151.0146.0140.0136.0132.0125.0119.0115.0111.0108.0096.0085.0077.0071.0066
2.73.2235.1121.0719.0524.0413.0342.0293.0258.0232.0212.0196.0183.0172.0163.0155.0148.0143.0137.0133.0129.0122.0117.0112.0108.0105.0094.0083.0075.0069.0064
2.74.2228.1114.0713.0519.0408.0337.0289.0254.0228.0208.0192.0179.0169.0160.0152.0145.0140.0135.0130.0126.0120.0114.0110.0106.0102.0091.0081.0073.0067.0063
2.75.2220.1107.0707.0514.0403.0333.0285.0251.0225.0205.0189.0176.0165.0156.0149.0142.0137.0132.0127.0123.0117.0111.0107.0103.0100.0089.0079.0071.0065.0061
2.76.2213.1100.0702.0508.0398.0329.0281.0247.0221.0201.0186.0173.0162.0153.0146.0139.0134.0129.0125.0121.0114.0109.0105.0101.0098.0087.0077.0069.0063.0059
2.77.2206.1094.0696.0503.0394.0324.0277.0243.0218.0198.0182.0170.0159.0150.0143.0137.0131.0126.0122.0118.0112.0106.0102.0098.0095.0085.0074.0067.0061.0057
2.78.2198.1087.0690.0498.0389.0320.0273.0239.0214.0195.0179.0167.0156.0147.0140.0134.0128.0124.0119.0116.0109.0104.0100.0096.0093.0082.0073.0065.0060.0055
2.79.2191.1080.0684.0493.0384.0316.0269.0236.0211.0191.0176.0163.0153.0145.0137.0131.0126.0121.0117.0113.0107.0102.0097.0094.0091.0080.0071.0063.0058.0054
2.80.2184.1074.0679.0488.0380.0312.0265.0232.0207.0188.0173.0160.0150.0142.0135.0128.0123.0118.0114.0111.0104.0099.0095.0092.0089.0078.0069.0061.0056.0052
2.81.2177.1067.0673.0483.0375.0308.0261.0228.0204.0185.0170.0157.0147.0139.0132.0126.0120.0116.0112.0108.0102.0097.0093.0089.0086.0076.0067.0060.0054.0051
2.82.2169.1061.0667.0478.0371.0304.0258.0225.0200.0182.0167.0155.0145.0136.0129.0123.0118.0113.0109.0106.0100.0095.0091.0087.0084.0074.0065.0058.0053.0049
2.83.2162.1055.0662.0473.0367.0300.0254.0222.0197.0179.0164.0152.0142.0134.0127.0121.0116.0111.0107.0103.0097.0093.0089.0085.0082.0072.0063.0056.0051.0047
2.84.2155.1048.0656.0469.0362.0296.0250.0218.0194.0175.0161.0149.0139.0131.0124.0118.0113.0109.0105.0101.0095.0090.0086.0083.0080.0071.0062.0055.0050.0046
2.85.2148.1042.0651.0464.0358.0292.0247.0215.0191.0173.0158.0146.0137.0128.0122.0116.0111.0106.0102.0099.0093.0088.0084.0081.0078.0069.0060.0053.0048.0045
2.86.2141.1036.0646.0459.0354.0288.0243.0212.0188.0170.0155.0144.0134.0126.0119.0113.0108.0104.0100.0097.0091.0086.0082.0079.0076.0067.0058.0052.0047.0043
2.87.2134.1030.0641.0455.0350.0284.0240.0208.0185.0167.0152.0141.0131.0124.0117.0111.0106.0102.0098.0095.0089.0084.0081.0077.0075.0065.0057.0050.0045.0042
2.88.2128.1024.0635.0450.0346.0281.0237.0205.0182.0164.0150.0138.0129.0121.0114.0109.0104.0100.0096.0093.0087.0082.0079.0075.0073.0064.0055.0049.0044.0041
2.89.2121.1018.0630.0446.0342.0277.0233.0202.0179.0161.0147.0136.0126.0119.0112.0107.0102.0098.0094.0091.0085.0080.0077.0074.0071.0062.0054.0047.0043.0039
2.90.2114.1012.0625.0441.0338.0273.0230.0199.0176.0158.0144.0133.0124.0116.0110.0104.0100.0095.0092.0089.0083.0079.0075.0072.0069.0060.0052.0046.0041.0038
2.91.2107.1006.0620.0437.0334.0270.0227.0196.0173.0156.0142.0131.0122.0114.0108.0102.0098.0093.0090.0087.0081.0077.0073.0070.0068.0059.0051.0045.0040.0037
2.92.2101.1000.0615.0432.0330.0266.0223.0193.0170.0153.0139.0128.0119.0112.0106.0100.0095.0091.0088.0085.0079.0075.0071.0068.0066.0057.0049.0043.0039.0036
2.93.2094.0994.0610.0428.0326.0263.0220.0190.0168.0150.0137.0126.0117.0110.0103.0098.0093.0089.0086.0083.0078.0073.0070.0067.0064.0056.0048.0042.0038.0035
2.94.2087.0988.0605.0424.0323.0259.0217.0187.0165.0148.0134.0124.0115.0108.0101.0096.0092.0088.0084.0081.0076.0072.0068.0065.0063.0054.0047.0041.0037.0034
2.95.2081.0983.0600.0420.0319.0256.0214.0184.0162.0145.0132.0121.0113.0105.0099.0094.0090.0086.0082.0079.0074.0070.0066.0064.0061.0053.0045.0040.0036.0033
2.96.2074.0977.0595.0416.0315.0253.0211.0181.0160.0143.0130.0119.0111.0103.0097.0092.0088.0084.0080.0077.0072.0068.0065.0062.0060.0052.0044.0038.0034.0031
2.97.2068.0971.0591.0411.0312.0250.0208.0179.0157.0140.0127.0117.0108.0101.0095.0090.0086.0082.0079.0076.0071.0067.0063.0060.0058.0050.0043.0037.0033.0030
2.98.2061.0966.0586.0407.0308.0246.0205.0176.0154.0138.0125.0115.0106.0099.0093.0088.0084.0080.0077.0074.0069.0065.0062.0059.0057.0049.0042.0036.0032.0030
2.99.2055.0960.0581.0403.0304.0243.0202.0173.0152.0136.0123.0113.0104.0097.0092.0087.0082.0079.0075.0072.0067.0064.0060.0058.0055.0048.0040.0035.0031.0029
3.00.2048.0955.0577.0399.0301.0240.0199.0171.0150.0133.0121.0111.0102.0096.0090.0085.0081.0077.0074.0071.0066.0062.0059.0056.0054.0046.0039.0034.0030.0028
3.01.2042.0949.0572.0396.0298.0237.0197.0168.0147.0131.0119.0109.0100.0094.0088.0083.0079.0075.0072.0069.0064.0061.0057.0055.0053.0045.0038.0033.0029.0027
3.02.2036.0944.0568.0392.0294.0234.0194.0166.0145.0129.0117.0107.0099.0092.0086.0081.0077.0074.0070.0068.0063.0059.0056.0053.0051.0044.0037.0032.0029.0026
3.03.2029.0938.0563.0388.0291.0231.0191.0163.0142.0127.0114.0105.0097.0090.0084.0080.0076.0072.0069.0066.0062.0058.0055.0052.0050.0043.0036.0031.0028.0025
3.04.2023.0933.0559.0384.0288.0228.0188.0161.0140.0125.0112.0103.0095.0088.0083.0078.0074.0070.0067.0065.0060.0056.0053.0051.0049.0042.0035.0030.0027.0024
3.05.2017.0928.0554.0380.0284.0225.0186.0158.0138.0123.0110.0101.0093.0086.0081.0076.0072.0069.0066.0063.0059.0055.0052.0050.0048.0040.0034.0029.0026.0023
3.06.2011.0923.0550.0377.0281.0222.0183.0156.0136.0120.0109.0099.0091.0085.0079.0075.0071.0067.0064.0062.0057.0054.0051.0048.0046.0039.0033.0028.0025.0023
3.07.2005.0917.0546.0373.0278.0219.0181.0153.0134.0118.0107.0097.0089.0083.0078.0073.0069.0066.0063.0060.0056.0053.0050.0047.0045.0038.0032.0028.0024.0022
3.08.1999.0912.0541.0369.0275.0217.0178.0151.0131.0116.0105.0095.0088.0081.0076.0072.0068.0065.0062.0059.0055.0051.0048.0046.0044.0037.0031.0027.0024.0021
3.09.1993.0907.0537.0366.0272.0214.0176.0149.0129.0114.0103.0094.0086.0080.0075.0070.0066.0063.0060.0058.0053.0050.0047.0045.0043.0036.0030.0026.0023.0021
3.10.1987.0902.0533.0362.0269.0211.0173.0147.0127.0113.0101.0092.0084.0078.0073.0069.0065.0062.0059.0056.0052.0049.0046.0044.0042.0035.0029.0025.0022.0020
3.11.1981.0897.0529.0359.0266.0208.0171.0144.0125.0111.0099.0090.0083.0077.0072.0067.0064.0060.0058.0055.0051.0048.0045.0043.0041.0034.0029.0024.0021.0019
3.12.1975.0892.0525.0355.0263.0206.0168.0142.0123.0109.0097.0089.0081.0075.0070.0066.0062.0059.0056.0054.0050.0047.0044.0042.0040.0033.0028.0024.0021.0019
3.13.1969.0887.0521.0352.0260.0203.0166.0140.0121.0107.0096.0087.0080.0074.0069.0065.0061.0058.0055.0053.0049.0045.0043.0041.0039.0033.0027.0023.0020.0018
3.14.1963.0882.0517.0348.0257.0201.0164.0138.0119.0105.0094.0085.0078.0072.0067.0063.0060.0057.0054.0052.0048.0044.0042.0040.0038.0032.0026.0022.0019.0017
3.15.1957.0877.0513.0345.0254.0198.0162.0136.0117.0103.0092.0084.0077.0071.0066.0062.0058.0055.0053.0050.0046.0043.0041.0039.0037.0031.0025.0022.0019.0017
3.16.1951.0872.0509.0342.0251.0196.0159.0134.0116.0102.0091.0082.0075.0070.0065.0061.0057.0054.0052.0049.0045.0042.0040.0038.0036.0030.0025.0021.0018.0016
3.17.1945.0868.0505.0339.0248.0193.0157.0132.0114.0100.0089.0081.0074.0068.0063.0059.0056.0053.0050.0048.0044.0041.0039.0037.0035.0029.0024.0020.0018.0016
3.18.1940.0863.0501.0335.0245.0191.0155.0130.0112.0098.0088.0079.0072.0067.0062.0058.0055.0052.0049.0047.0043.0040.0038.0036.0034.0028.0023.0020.0017.0015
3.19.1934.0858.0497.0332.0243.0188.0153.0128.0110.0097.0086.0078.0071.0065.0061.0057.0054.0051.0048.0046.0042.0039.0037.0035.0033.0028.0023.0019.0017.0015
3.20.1928.0853.0493.0329.0240.0186.0151.0126.0108.0095.0085.0076.0070.0064.0060.0056.0052.0050.0047.0045.0041.0038.0036.0034.0032.0027.0022.0018.0016.0014
3.21.1923.0849.0490.0326.0237.0184.0149.0124.0107.0093.0083.0075.0068.0063.0058.0055.0051.0049.0046.0044.0040.0037.0035.0033.0032.0026.0021.0018.0015.0014
3.22.1917.0844.0486.0323.0235.0181.0147.0122.0105.0092.0082.0074.0067.0062.0057.0053.0050.0047.0045.0043.0039.0037.0034.0032.0031.0025.0021.0017.0015.0013
3.23.1911.0840.0482.0320.0232.0179.0145.0121.0103.0090.0080.0072.0066.0060.0056.0052.0049.0046.0044.0042.0038.0036.0033.0032.0030.0025.0020.0017.0014.0013
3.24.1906.0835.0479.0317.0229.0177.0143.0119.0102.0089.0079.0071.0065.0059.0055.0051.0048.0045.0043.0041.0038.0035.0033.0031.0029.0024.0020.0016.0014.0012
3.25.1900.0831.0475.0314.0227.0175.0141.0117.0100.0087.0077.0070.0063.0058.0054.0050.0047.0044.0042.0040.0037.0034.0032.0030.0028.0023.0019.0016.0014.0012
3.26.1895.0826.0471.0311.0224.0172.0139.0115.0098.0086.0076.0068.0062.0057.0053.0049.0046.0043.0041.0039.0036.0033.0031.0029.0028.0023.0018.0015.0013.0012
3.27.1889.0822.0468.0308.0222.0170.0137.0114.0097.0084.0075.0067.0061.0056.0052.0048.0045.0043.0040.0038.0035.0032.0030.0029.0027.0022.0018.0015.0013.0011
-309-
Back to TOC
df
t 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 22 24 26 28 30 40 60 100 200 1000

3.28.1884.0817.0464.0305.0220.0168.0135.0112.0095.0083.0073.0066.0060.0055.0051.0047.0044.0042.0039.0037.0034.0032.0030.0028.0026.0022.0017.0014.0012.0011
3.29.1879.0813.0461.0302.0217.0166.0133.0110.0094.0082.0072.0065.0059.0054.0050.0046.0043.0041.0039.0037.0033.0031.0029.0027.0026.0021.0017.0014.0012.0010
3.30.1873.0808.0457.0299.0215.0164.0131.0109.0092.0080.0071.0063.0057.0053.0049.0045.0042.0040.0038.0036.0033.0030.0028.0026.0025.0020.0016.0013.0011.0010
3.31.1868.0804.0454.0297.0212.0162.0129.0107.0091.0079.0070.0062.0056.0052.0048.0044.0041.0039.0037.0035.0032.0029.0027.0026.0024.0020.0016.0013.0011.0010
3.32.1863.0800.0451.0294.0210.0160.0128.0105.0089.0077.0068.0061.0055.0051.0047.0043.0041.0038.0036.0034.0031.0029.0027.0025.0024.0019.0015.0013.0011.0009
3.33.1857.0796.0447.0291.0208.0158.0126.0104.0088.0076.0067.0060.0054.0050.0046.0042.0040.0037.0035.0033.0030.0028.0026.0024.0023.0019.0015.0012.0010.0009
3.34.1852.0791.0444.0288.0206.0156.0124.0102.0087.0075.0066.0059.0053.0049.0045.0042.0039.0036.0034.0033.0030.0027.0025.0024.0023.0018.0014.0012.0010.0009
3.35.1847.0787.0441.0286.0203.0154.0123.0101.0085.0074.0065.0058.0052.0048.0044.0041.0038.0036.0034.0032.0029.0027.0025.0023.0022.0018.0014.0011.0010.0008
3.36.1842.0783.0437.0283.0201.0152.0121.0099.0084.0072.0064.0057.0051.0047.0043.0040.0037.0035.0033.0031.0028.0026.0024.0023.0021.0017.0014.0011.0009.0008
3.37.1836.0779.0434.0280.0199.0150.0119.0098.0083.0071.0063.0056.0050.0046.0042.0039.0036.0034.0032.0030.0028.0025.0024.0022.0021.0017.0013.0011.0009.0008
3.38.1831.0775.0431.0278.0197.0149.0118.0096.0081.0070.0061.0055.0049.0045.0041.0038.0036.0033.0031.0030.0027.0025.0023.0022.0020.0016.0013.0010.0009.0008
3.39.1826.0771.0428.0275.0195.0147.0116.0095.0080.0069.0060.0054.0048.0044.0040.0037.0035.0033.0031.0029.0026.0024.0022.0021.0020.0016.0012.0010.0008.0007
3.40.1821.0767.0425.0273.0193.0145.0114.0094.0079.0068.0059.0053.0047.0043.0040.0037.0034.0032.0030.0028.0026.0024.0022.0020.0019.0015.0012.0010.0008.0007
3.41.1816.0763.0422.0270.0190.0143.0113.0092.0077.0067.0058.0052.0047.0042.0039.0036.0033.0031.0029.0028.0025.0023.0021.0020.0019.0015.0012.0009.0008.0007
3.42.1811.0759.0418.0268.0188.0141.0111.0091.0076.0065.0057.0051.0046.0041.0038.0035.0033.0031.0029.0027.0025.0022.0021.0019.0018.0015.0011.0009.0008.0007
3.43.1806.0755.0415.0265.0186.0140.0110.0090.0075.0064.0056.0050.0045.0041.0037.0034.0032.0030.0028.0027.0024.0022.0020.0019.0018.0014.0011.0009.0007.0006
3.44.1801.0751.0412.0263.0184.0138.0108.0088.0074.0063.0055.0049.0044.0040.0036.0034.0031.0029.0027.0026.0023.0021.0020.0018.0017.0014.0011.0008.0007.0006
3.45.1796.0747.0409.0261.0182.0136.0107.0087.0073.0062.0054.0048.0043.0039.0036.0033.0031.0029.0027.0025.0023.0021.0019.0018.0017.0013.0010.0008.0007.0006
3.46.1791.0743.0406.0258.0180.0135.0105.0086.0072.0061.0053.0047.0042.0038.0035.0032.0030.0028.0026.0025.0022.0020.0019.0017.0016.0013.0010.0008.0007.0006
3.47.1786.0740.0403.0256.0178.0133.0104.0084.0070.0060.0052.0046.0041.0038.0034.0032.0029.0027.0026.0024.0022.0020.0018.0017.0016.0013.0010.0008.0006.0005
3.48.1781.0736.0401.0254.0177.0131.0103.0083.0069.0059.0051.0045.0041.0037.0034.0031.0029.0027.0025.0024.0021.0019.0018.0017.0016.0012.0009.0007.0006.0005
3.49.1777.0732.0398.0251.0175.0130.0101.0082.0068.0058.0051.0045.0040.0036.0033.0030.0028.0026.0025.0023.0021.0019.0017.0016.0015.0012.0009.0007.0006.0005
3.50.1772.0728.0395.0249.0173.0128.0100.0081.0067.0057.0050.0044.0039.0035.0032.0030.0027.0026.0024.0023.0020.0018.0017.0016.0015.0012.0009.0007.0006.0005
3.60.1725.0692.0368.0228.0155.0114.0087.0070.0057.0048.0042.0036.0032.0029.0026.0024.0022.0020.0019.0018.0016.0014.0013.0012.0011.0009.0006.0005.0004.0003
3.70.1680.0659.0343.0208.0140.0101.0077.0060.0049.0041.0035.0030.0027.0024.0021.0019.0018.0016.0015.0014.0013.0011.0010.0009.0009.0006.0005.0004.0003.0002
3.80.1638.0628.0320.0191.0126.0090.0067.0052.0042.0035.0029.0025.0022.0020.0017.0016.0014.0013.0012.0011.0010.0009.0008.0007.0007.0005.0003.0002.0002.0002
3.90.1598.0599.0299.0175.0114.0080.0059.0045.0036.0030.0025.0021.0018.0016.0014.0013.0012.0010.0010.0009.0008.0007.0006.0005.0005.0004.0002.0002.0001.0001
4.00.1560.0572.0280.0161.0103.0071.0052.0039.0031.0025.0021.0018.0015.0013.0012.0010.0009.0008.0008.0007.0006.0005.0005.0004.0004.0003.0002.0001.0001.0001
-310-
Back to TOC
Table 3: Probabilities Under the F-Distribution
Table 3-A: dfB = 1
F 2 3 4 5 6 7 8 9 10 12 14 16 18 20 22 24 26 28 30 40 45 50 60 80 100 250 1000
1.00 0.4226 0.3910 0.3739 0.3632 0.3559 0.3506 0.3466 0.3434 0.3409 0.3370 0.3343 0.3322 0.3306 0.3293 0.3282 0.3273 0.3265 0.3259 0.3253 0.3233 0.3227 0.3221 0.3213 0.3203 0.3197 0.3183 0.3176
1.10 0.4043 0.3713 0.3535 0.3423 0.3347 0.3291 0.3249 0.3216 0.3190 0.3149 0.3120 0.3099 0.3081 0.3068 0.3057 0.3047 0.3039 0.3032 0.3026 0.3006 0.2999 0.2993 0.2985 0.2974 0.2968 0.2953 0.2945
1.20 0.3876 0.3534 0.3349 0.3233 0.3153 0.3096 0.3052 0.3018 0.2990 0.2948 0.2918 0.2895 0.2878 0.2863 0.2852 0.2842 0.2834 0.2827 0.2820 0.2799 0.2792 0.2786 0.2777 0.2766 0.2760 0.2744 0.2736
1.30 0.3724 0.3370 0.3178 0.3059 0.2977 0.2917 0.2872 0.2836 0.2808 0.2765 0.2733 0.2710 0.2692 0.2677 0.2665 0.2655 0.2646 0.2639 0.2632 0.2610 0.2602 0.2596 0.2587 0.2576 0.2569 0.2553 0.2545
1.40 0.3583 0.3219 0.3022 0.2899 0.2815 0.2753 0.2707 0.2670 0.2641 0.2596 0.2564 0.2540 0.2521 0.2506 0.2494 0.2483 0.2474 0.2467 0.2460 0.2437 0.2429 0.2423 0.2414 0.2402 0.2395 0.2378 0.2370
1.50 0.3453 0.3081 0.2879 0.2752 0.2666 0.2603 0.2555 0.2518 0.2487 0.2442 0.2409 0.2384 0.2365 0.2349 0.2336 0.2326 0.2317 0.2309 0.2302 0.2278 0.2270 0.2264 0.2255 0.2243 0.2235 0.2218 0.2210
1.60 0.3333 0.2952 0.2746 0.2617 0.2528 0.2464 0.2415 0.2377 0.2346 0.2299 0.2266 0.2240 0.2220 0.2204 0.2191 0.2180 0.2171 0.2163 0.2156 0.2132 0.2124 0.2118 0.2108 0.2096 0.2088 0.2071 0.2062
1.70 0.3222 0.2833 0.2623 0.2491 0.2401 0.2335 0.2286 0.2246 0.2215 0.2167 0.2133 0.2107 0.2087 0.2071 0.2058 0.2047 0.2037 0.2029 0.2022 0.1997 0.1989 0.1983 0.1973 0.1960 0.1953 0.1935 0.1926
1.80 0.3118 0.2722 0.2508 0.2374 0.2283 0.2216 0.2165 0.2126 0.2094 0.2046 0.2011 0.1984 0.1964 0.1947 0.1934 0.1923 0.1913 0.1905 0.1898 0.1873 0.1864 0.1858 0.1848 0.1835 0.1828 0.1809 0.1800
1.90 0.3020 0.2619 0.2402 0.2266 0.2173 0.2105 0.2054 0.2014 0.1981 0.1932 0.1897 0.1870 0.1850 0.1833 0.1819 0.1808 0.1798 0.1790 0.1783 0.1757 0.1749 0.1742 0.1732 0.1719 0.1712 0.1693 0.1684
2.00 0.2929 0.2522 0.2302 0.2164 0.2070 0.2002 0.1950 0.1909 0.1877 0.1827 0.1792 0.1765 0.1744 0.1727 0.1713 0.1701 0.1692 0.1683 0.1676 0.1650 0.1642 0.1635 0.1625 0.1612 0.1604 0.1585 0.1576
2.05 0.2885 0.2476 0.2255 0.2116 0.2022 0.1953 0.1901 0.1860 0.1827 0.1777 0.1742 0.1715 0.1693 0.1676 0.1663 0.1651 0.1641 0.1633 0.1625 0.1600 0.1591 0.1584 0.1574 0.1561 0.1553 0.1535 0.1525
2.10 0.2843 0.2432 0.2209 0.2070 0.1975 0.1906 0.1853 0.1812 0.1779 0.1729 0.1693 0.1666 0.1645 0.1628 0.1614 0.1602 0.1593 0.1584 0.1577 0.1551 0.1542 0.1535 0.1525 0.1512 0.1504 0.1486 0.1476
2.15 0.2802 0.2388 0.2165 0.2025 0.1929 0.1860 0.1807 0.1766 0.1733 0.1683 0.1647 0.1619 0.1598 0.1581 0.1567 0.1556 0.1546 0.1537 0.1530 0.1504 0.1495 0.1488 0.1478 0.1465 0.1457 0.1438 0.1429
2.20 0.2763 0.2346 0.2122 0.1981 0.1885 0.1816 0.1763 0.1722 0.1688 0.1638 0.1602 0.1574 0.1553 0.1536 0.1522 0.1510 0.1500 0.1492 0.1484 0.1458 0.1450 0.1443 0.1432 0.1419 0.1412 0.1393 0.1383
2.25 0.2724 0.2306 0.2080 0.1939 0.1843 0.1773 0.1720 0.1679 0.1645 0.1595 0.1558 0.1531 0.1510 0.1492 0.1478 0.1467 0.1457 0.1448 0.1441 0.1415 0.1406 0.1399 0.1389 0.1376 0.1368 0.1349 0.1339
2.30 0.2686 0.2266 0.2040 0.1898 0.1802 0.1732 0.1678 0.1637 0.1603 0.1553 0.1516 0.1489 0.1467 0.1450 0.1436 0.1424 0.1414 0.1406 0.1398 0.1372 0.1364 0.1357 0.1346 0.1333 0.1325 0.1306 0.1297
2.35 0.2650 0.2228 0.2001 0.1859 0.1762 0.1691 0.1638 0.1596 0.1563 0.1512 0.1476 0.1448 0.1427 0.1409 0.1395 0.1384 0.1374 0.1365 0.1358 0.1332 0.1323 0.1316 0.1305 0.1292 0.1284 0.1265 0.1256
2.40 0.2615 0.2191 0.1963 0.1820 0.1723 0.1653 0.1599 0.1557 0.1524 0.1473 0.1436 0.1409 0.1387 0.1370 0.1356 0.1344 0.1334 0.1326 0.1318 0.1292 0.1283 0.1276 0.1266 0.1253 0.1245 0.1226 0.1217
2.45 0.2580 0.2155 0.1926 0.1783 0.1686 0.1615 0.1562 0.1520 0.1486 0.1435 0.1398 0.1371 0.1349 0.1332 0.1318 0.1306 0.1296 0.1288 0.1280 0.1254 0.1245 0.1238 0.1228 0.1215 0.1207 0.1188 0.1178
2.50 0.2546 0.2120 0.1890 0.1747 0.1649 0.1579 0.1525 0.1483 0.1449 0.1398 0.1362 0.1334 0.1313 0.1295 0.1281 0.1269 0.1259 0.1251 0.1243 0.1217 0.1209 0.1202 0.1191 0.1178 0.1170 0.1151 0.1142
2.55 0.2514 0.2086 0.1855 0.1712 0.1614 0.1543 0.1490 0.1448 0.1414 0.1363 0.1326 0.1299 0.1277 0.1260 0.1246 0.1234 0.1224 0.1215 0.1208 0.1182 0.1173 0.1166 0.1155 0.1142 0.1134 0.1116 0.1106
2.60 0.2482 0.2053 0.1822 0.1678 0.1580 0.1509 0.1455 0.1413 0.1379 0.1328 0.1292 0.1264 0.1243 0.1225 0.1211 0.1199 0.1189 0.1181 0.1173 0.1147 0.1139 0.1132 0.1121 0.1108 0.1100 0.1081 0.1072
2.65 0.2451 0.2020 0.1789 0.1645 0.1547 0.1476 0.1422 0.1380 0.1346 0.1295 0.1258 0.1231 0.1209 0.1192 0.1178 0.1166 0.1156 0.1148 0.1140 0.1114 0.1105 0.1098 0.1088 0.1075 0.1067 0.1048 0.1039
2.70 0.2421 0.1989 0.1757 0.1613 0.1515 0.1443 0.1390 0.1348 0.1314 0.1263 0.1226 0.1199 0.1177 0.1160 0.1146 0.1134 0.1124 0.1115 0.1108 0.1082 0.1073 0.1066 0.1056 0.1043 0.1035 0.1016 0.1007
2.75 0.2391 0.1958 0.1726 0.1582 0.1483 0.1412 0.1358 0.1316 0.1282 0.1231 0.1195 0.1167 0.1146 0.1129 0.1114 0.1103 0.1093 0.1084 0.1077 0.1051 0.1042 0.1035 0.1025 0.1012 0.1004 0.0985 0.0976
2.80 0.2362 0.1929 0.1696 0.1551 0.1453 0.1382 0.1328 0.1286 0.1252 0.1201 0.1165 0.1137 0.1116 0.1098 0.1084 0.1073 0.1063 0.1054 0.1047 0.1021 0.1012 0.1005 0.0995 0.0982 0.0974 0.0955 0.0946
2.85 0.2334 0.1900 0.1666 0.1522 0.1423 0.1352 0.1299 0.1256 0.1223 0.1172 0.1135 0.1108 0.1086 0.1069 0.1055 0.1043 0.1033 0.1025 0.1017 0.0992 0.0983 0.0976 0.0966 0.0953 0.0945 0.0926 0.0917
2.90 0.2307 0.1871 0.1638 0.1493 0.1395 0.1324 0.1270 0.1228 0.1194 0.1143 0.1107 0.1079 0.1058 0.1041 0.1027 0.1015 0.1005 0.0997 0.0989 0.0963 0.0955 0.0948 0.0938 0.0925 0.0917 0.0898 0.0889
2.95 0.2280 0.1844 0.1610 0.1465 0.1367 0.1296 0.1242 0.1200 0.1166 0.1115 0.1079 0.1052 0.1030 0.1013 0.0999 0.0988 0.0978 0.0969 0.0962 0.0936 0.0928 0.0921 0.0910 0.0897 0.0890 0.0871 0.0862
3.00 0.2254 0.1817 0.1583 0.1438 0.1340 0.1269 0.1215 0.1173 0.1139 0.1089 0.1052 0.1025 0.1004 0.0987 0.0973 0.0961 0.0951 0.0943 0.0935 0.0910 0.0901 0.0894 0.0884 0.0871 0.0863 0.0845 0.0836
3.05 0.2229 0.1791 0.1557 0.1412 0.1313 0.1242 0.1189 0.1147 0.1113 0.1063 0.1026 0.0999 0.0978 0.0961 0.0947 0.0935 0.0925 0.0917 0.0910 0.0884 0.0876 0.0869 0.0859 0.0846 0.0838 0.0820 0.0810
3.10 0.2204 0.1765 0.1531 0.1386 0.1288 0.1217 0.1163 0.1121 0.1088 0.1037 0.1001 0.0974 0.0953 0.0936 0.0922 0.0910 0.0901 0.0892 0.0885 0.0859 0.0851 0.0844 0.0834 0.0821 0.0813 0.0795 0.0786
3.15 0.2179 0.1740 0.1506 0.1361 0.1263 0.1192 0.1138 0.1097 0.1063 0.1013 0.0977 0.0950 0.0928 0.0912 0.0898 0.0886 0.0876 0.0868 0.0861 0.0835 0.0827 0.0820 0.0810 0.0797 0.0790 0.0771 0.0762
3.20 0.2155 0.1716 0.1481 0.1337 0.1238 0.1168 0.1114 0.1073 0.1039 0.0989 0.0953 0.0926 0.0905 0.0888 0.0874 0.0863 0.0853 0.0845 0.0837 0.0812 0.0804 0.0797 0.0787 0.0774 0.0767 0.0748 0.0739
3.25 0.2132 0.1692 0.1458 0.1313 0.1215 0.1144 0.1091 0.1049 0.1016 0.0966 0.0930 0.0903 0.0882 0.0865 0.0851 0.0840 0.0830 0.0822 0.0815 0.0790 0.0781 0.0775 0.0764 0.0752 0.0744 0.0726 0.0717
3.30 0.2109 0.1669 0.1434 0.1290 0.1192 0.1121 0.1068 0.1027 0.0993 0.0943 0.0907 0.0881 0.0860 0.0843 0.0829 0.0818 0.0808 0.0800 0.0793 0.0768 0.0759 0.0753 0.0743 0.0730 0.0723 0.0705 0.0696
3.35 0.2087 0.1646 0.1412 0.1267 0.1169 0.1099 0.1046 0.1004 0.0971 0.0921 0.0886 0.0859 0.0838 0.0821 0.0808 0.0797 0.0787 0.0779 0.0772 0.0747 0.0738 0.0732 0.0722 0.0709 0.0702 0.0684 0.0675
3.40 0.2065 0.1624 0.1390 0.1245 0.1148 0.1077 0.1024 0.0983 0.0950 0.0900 0.0865 0.0838 0.0817 0.0801 0.0787 0.0776 0.0766 0.0758 0.0751 0.0726 0.0718 0.0711 0.0701 0.0689 0.0682 0.0664 0.0655
3.45 0.2044 0.1602 0.1368 0.1224 0.1126 0.1056 0.1003 0.0962 0.0929 0.0880 0.0844 0.0818 0.0797 0.0780 0.0767 0.0756 0.0746 0.0738 0.0731 0.0706 0.0698 0.0691 0.0682 0.0669 0.0662 0.0644 0.0635
3.50 0.2023 0.1581 0.1347 0.1203 0.1106 0.1036 0.0983 0.0942 0.0909 0.0859 0.0824 0.0798 0.0777 0.0761 0.0747 0.0736 0.0727 0.0719 0.0712 0.0687 0.0679 0.0672 0.0662 0.0650 0.0643 0.0625 0.0617
3.55 0.2002 0.1561 0.1326 0.1182 0.1085 0.1015 0.0963 0.0922 0.0889 0.0840 0.0805 0.0778 0.0758 0.0742 0.0728 0.0717 0.0708 0.0700 0.0693 0.0668 0.0660 0.0654 0.0644 0.0632 0.0625 0.0607 0.0598
3.60 0.1982 0.1540 0.1306 0.1163 0.1066 0.0996 0.0943 0.0903 0.0870 0.0821 0.0786 0.0760 0.0739 0.0723 0.0710 0.0699 0.0689 0.0681 0.0674 0.0650 0.0642 0.0636 0.0626 0.0614 0.0607 0.0589 0.0581
3.65 0.1962 0.1521 0.1287 0.1143 0.1046 0.0977 0.0925 0.0884 0.0851 0.0803 0.0768 0.0742 0.0721 0.0705 0.0692 0.0681 0.0672 0.0664 0.0657 0.0633 0.0625 0.0618 0.0609 0.0597 0.0589 0.0572 0.0564
3.70 0.1943 0.1501 0.1268 0.1124 0.1028 0.0958 0.0906 0.0866 0.0833 0.0785 0.0750 0.0724 0.0704 0.0688 0.0674 0.0664 0.0654 0.0646 0.0639 0.0616 0.0608 0.0601 0.0592 0.0580 0.0573 0.0555 0.0547
3.75 0.1924 0.1482 0.1249 0.1106 0.1009 0.0940 0.0888 0.0848 0.0816 0.0767 0.0733 0.0707 0.0687 0.0671 0.0658 0.0647 0.0637 0.0630 0.0623 0.0599 0.0591 0.0585 0.0575 0.0563 0.0556 0.0539 0.0531
3.80 0.1906 0.1464 0.1230 0.1088 0.0991 0.0923 0.0871 0.0831 0.0798 0.0750 0.0716 0.0690 0.0670 0.0654 0.0641 0.0630 0.0621 0.0613 0.0607 0.0583 0.0575 0.0569 0.0559 0.0548 0.0541 0.0524 0.0515
3.85 0.1888 0.1445 0.1213 0.1070 0.0974 0.0905 0.0854 0.0814 0.0782 0.0734 0.0699 0.0674 0.0654 0.0638 0.0625 0.0614 0.0605 0.0598 0.0591 0.0567 0.0559 0.0553 0.0544 0.0532 0.0525 0.0509 0.0500
3.90 0.1870 0.1428 0.1195 0.1053 0.0957 0.0889 0.0837 0.0797 0.0765 0.0717 0.0683 0.0658 0.0638 0.0623 0.0610 0.0599 0.0590 0.0582 0.0576 0.0552 0.0544 0.0538 0.0529 0.0517 0.0510 0.0494 0.0486
3.95 0.1852 0.1410 0.1178 0.1036 0.0940 0.0872 0.0821 0.0781 0.0749 0.0702 0.0668 0.0643 0.0623 0.0607 0.0595 0.0584 0.0575 0.0567 0.0561 0.0537 0.0530 0.0524 0.0514 0.0503 0.0496 0.0480 0.0471
4.00 0.1835 0.1393 0.1161 0.1019 0.0924 0.0856 0.0805 0.0766 0.0734 0.0687 0.0653 0.0628 0.0608 0.0593 0.0580 0.0569 0.0560 0.0553 0.0546 0.0523 0.0516 0.0509 0.0500 0.0489 0.0482 0.0466 0.0458
4.10 0.1802 0.1360 0.1129 0.0988 0.0893 0.0825 0.0775 0.0735 0.0704 0.0657 0.0624 0.0599 0.0580 0.0564 0.0552 0.0541 0.0533 0.0525 0.0519 0.0496 0.0488 0.0482 0.0473 0.0462 0.0455 0.0439 0.0431
4.20 0.1769 0.1328 0.1098 0.0957 0.0863 0.0796 0.0746 0.0707 0.0676 0.0629 0.0596 0.0572 0.0553 0.0538 0.0525 0.0515 0.0506 0.0499 0.0493 0.0470 0.0463 0.0457 0.0448 0.0437 0.0430 0.0415 0.0407
4.30 0.1738 0.1298 0.1068 0.0928 0.0835 0.0768 0.0718 0.0680 0.0649 0.0603 0.0570 0.0546 0.0527 0.0512 0.0500 0.0490 0.0482 0.0474 0.0468 0.0446 0.0439 0.0433 0.0424 0.0413 0.0407 0.0391 0.0384
4.40 0.1708 0.1269 0.1039 0.0900 0.0807 0.0741 0.0692 0.0654 0.0623 0.0578 0.0546 0.0522 0.0503 0.0488 0.0476 0.0467 0.0458 0.0451 0.0445 0.0423 0.0416 0.0410 0.0402 0.0391 0.0385 0.0369 0.0362
4.50 0.1679 0.1240 0.1012 0.0874 0.0781 0.0716 0.0667 0.0629 0.0599 0.0554 0.0522 0.0499 0.0480 0.0466 0.0454 0.0444 0.0436 0.0429 0.0423 0.0401 0.0394 0.0389 0.0380 0.0370 0.0364 0.0349 0.0341
4.60 0.1652 0.1213 0.0986 0.0848 0.0757 0.0691 0.0643 0.0606 0.0576 0.0531 0.0500 0.0477 0.0459 0.0444 0.0433 0.0423 0.0415 0.0408 0.0402 0.0381 0.0374 0.0369 0.0360 0.0350 0.0344 0.0329 0.0322
4.70 0.1624 0.1187 0.0960 0.0824 0.0733 0.0668 0.0620 0.0583 0.0554 0.0510 0.0479 0.0456 0.0438 0.0424 0.0413 0.0403 0.0395 0.0388 0.0382 0.0362 0.0355 0.0349 0.0341 0.0331 0.0325 0.0311 0.0304
4.80 0.1598 0.1162 0.0936 0.0800 0.0710 0.0646 0.0598 0.0562 0.0533 0.0489 0.0459 0.0436 0.0419 0.0405 0.0393 0.0384 0.0376 0.0369 0.0364 0.0343 0.0337 0.0331 0.0324 0.0314 0.0308 0.0294 0.0287
4.90 0.1573 0.1137 0.0913 0.0778 0.0688 0.0625 0.0578 0.0541 0.0512 0.0470 0.0440 0.0417 0.0400 0.0386 0.0375 0.0366 0.0358 0.0352 0.0346 0.0326 0.0320 0.0314 0.0307 0.0297 0.0291 0.0278 0.0271
-311-
Back to TOC
F 2 3 4 5 6 7 8 9 10 12 14 16 18 20 22 24 26 28 30 40 45 50 60 80 100 250 1000
5.00 0.1548 0.1114 0.0890 0.0756 0.0667 0.0604 0.0558 0.0522 0.0493 0.0451 0.0421 0.0399 0.0382 0.0369 0.0358 0.0349 0.0341 0.0335 0.0329 0.0310 0.0304 0.0298 0.0291 0.0281 0.0276 0.0262 0.0256
5.10 0.1525 0.1091 0.0868 0.0735 0.0647 0.0585 0.0539 0.0503 0.0475 0.0433 0.0404 0.0382 0.0366 0.0353 0.0342 0.0333 0.0325 0.0319 0.0314 0.0295 0.0288 0.0283 0.0276 0.0266 0.0261 0.0248 0.0241
5.20 0.1502 0.1069 0.0848 0.0715 0.0628 0.0566 0.0520 0.0485 0.0458 0.0417 0.0388 0.0366 0.0350 0.0337 0.0326 0.0318 0.0310 0.0304 0.0299 0.0280 0.0274 0.0269 0.0262 0.0252 0.0247 0.0234 0.0228
5.30 0.1479 0.1048 0.0827 0.0696 0.0609 0.0548 0.0503 0.0468 0.0441 0.0400 0.0372 0.0351 0.0335 0.0322 0.0312 0.0303 0.0296 0.0290 0.0284 0.0266 0.0260 0.0255 0.0248 0.0239 0.0234 0.0221 0.0215
5.40 0.1458 0.1027 0.0808 0.0677 0.0591 0.0531 0.0486 0.0452 0.0425 0.0385 0.0357 0.0336 0.0320 0.0308 0.0298 0.0289 0.0282 0.0276 0.0271 0.0253 0.0247 0.0242 0.0235 0.0227 0.0222 0.0209 0.0203
5.50 0.1437 0.1007 0.0789 0.0660 0.0574 0.0514 0.0470 0.0436 0.0410 0.0370 0.0343 0.0322 0.0307 0.0294 0.0284 0.0276 0.0269 0.0263 0.0258 0.0241 0.0235 0.0230 0.0223 0.0215 0.0210 0.0198 0.0192
5.60 0.1416 0.0988 0.0771 0.0642 0.0558 0.0499 0.0455 0.0422 0.0395 0.0356 0.0329 0.0309 0.0294 0.0282 0.0272 0.0264 0.0257 0.0251 0.0246 0.0229 0.0223 0.0219 0.0212 0.0204 0.0199 0.0187 0.0182
5.70 0.1396 0.0970 0.0754 0.0626 0.0542 0.0483 0.0440 0.0407 0.0381 0.0343 0.0316 0.0296 0.0281 0.0269 0.0260 0.0252 0.0245 0.0240 0.0235 0.0218 0.0212 0.0208 0.0201 0.0193 0.0188 0.0177 0.0171
5.80 0.1377 0.0952 0.0737 0.0610 0.0527 0.0469 0.0426 0.0394 0.0368 0.0330 0.0304 0.0284 0.0270 0.0258 0.0248 0.0241 0.0234 0.0229 0.0224 0.0207 0.0202 0.0198 0.0191 0.0183 0.0179 0.0168 0.0162
5.90 0.1358 0.0934 0.0721 0.0595 0.0512 0.0455 0.0413 0.0380 0.0355 0.0318 0.0292 0.0273 0.0258 0.0247 0.0238 0.0230 0.0224 0.0218 0.0213 0.0197 0.0192 0.0188 0.0182 0.0174 0.0169 0.0158 0.0153
6.00 0.1340 0.0917 0.0705 0.0580 0.0498 0.0441 0.0400 0.0368 0.0343 0.0306 0.0281 0.0262 0.0248 0.0236 0.0227 0.0220 0.0214 0.0208 0.0204 0.0188 0.0183 0.0179 0.0172 0.0165 0.0160 0.0150 0.0145
6.10 0.1322 0.0901 0.0690 0.0565 0.0485 0.0428 0.0387 0.0356 0.0331 0.0295 0.0270 0.0252 0.0237 0.0226 0.0218 0.0210 0.0204 0.0199 0.0194 0.0179 0.0174 0.0170 0.0164 0.0156 0.0152 0.0142 0.0137
6.20 0.1305 0.0885 0.0675 0.0552 0.0472 0.0416 0.0375 0.0344 0.0320 0.0284 0.0260 0.0242 0.0228 0.0217 0.0208 0.0201 0.0195 0.0190 0.0185 0.0170 0.0165 0.0161 0.0156 0.0148 0.0144 0.0134 0.0129
6.30 0.1288 0.0869 0.0661 0.0538 0.0459 0.0404 0.0364 0.0333 0.0309 0.0274 0.0250 0.0232 0.0218 0.0208 0.0199 0.0192 0.0186 0.0181 0.0177 0.0162 0.0157 0.0154 0.0148 0.0141 0.0137 0.0127 0.0122
6.40 0.1271 0.0854 0.0647 0.0525 0.0447 0.0392 0.0353 0.0322 0.0299 0.0264 0.0240 0.0223 0.0210 0.0199 0.0191 0.0184 0.0178 0.0173 0.0169 0.0155 0.0150 0.0146 0.0141 0.0134 0.0130 0.0120 0.0116
6.50 0.1255 0.0840 0.0633 0.0513 0.0435 0.0381 0.0342 0.0312 0.0289 0.0255 0.0231 0.0214 0.0201 0.0191 0.0183 0.0176 0.0170 0.0165 0.0161 0.0147 0.0143 0.0139 0.0134 0.0127 0.0123 0.0114 0.0109
6.60 0.1240 0.0826 0.0620 0.0501 0.0424 0.0371 0.0332 0.0302 0.0279 0.0246 0.0223 0.0206 0.0193 0.0183 0.0175 0.0168 0.0163 0.0158 0.0154 0.0140 0.0136 0.0132 0.0127 0.0121 0.0117 0.0108 0.0103
6.70 0.1224 0.0812 0.0608 0.0489 0.0413 0.0360 0.0322 0.0293 0.0270 0.0237 0.0215 0.0198 0.0185 0.0176 0.0168 0.0161 0.0156 0.0151 0.0147 0.0134 0.0129 0.0126 0.0121 0.0114 0.0111 0.0102 0.0098
6.80 0.1210 0.0798 0.0596 0.0478 0.0402 0.0350 0.0312 0.0284 0.0261 0.0229 0.0207 0.0190 0.0178 0.0168 0.0161 0.0154 0.0149 0.0145 0.0141 0.0128 0.0123 0.0120 0.0115 0.0109 0.0105 0.0097 0.0093
6.90 0.1195 0.0785 0.0584 0.0467 0.0392 0.0341 0.0303 0.0275 0.0253 0.0221 0.0199 0.0183 0.0171 0.0162 0.0154 0.0148 0.0143 0.0138 0.0134 0.0122 0.0117 0.0114 0.0109 0.0103 0.0100 0.0092 0.0088
7.00 0.1181 0.0773 0.0572 0.0457 0.0382 0.0331 0.0294 0.0267 0.0245 0.0213 0.0192 0.0176 0.0164 0.0155 0.0148 0.0142 0.0136 0.0132 0.0129 0.0116 0.0112 0.0109 0.0104 0.0098 0.0095 0.0087 0.0083
7.20 0.1153 0.0748 0.0550 0.0436 0.0364 0.0314 0.0278 0.0251 0.0230 0.0199 0.0178 0.0163 0.0152 0.0143 0.0136 0.0130 0.0125 0.0121 0.0117 0.0106 0.0102 0.0099 0.0094 0.0089 0.0085 0.0078 0.0074
7.40 0.1127 0.0725 0.0530 0.0418 0.0346 0.0298 0.0262 0.0236 0.0215 0.0186 0.0166 0.0151 0.0140 0.0132 0.0125 0.0119 0.0115 0.0111 0.0107 0.0096 0.0092 0.0089 0.0085 0.0080 0.0077 0.0070 0.0066
7.60 0.1102 0.0703 0.0510 0.0400 0.0330 0.0282 0.0248 0.0222 0.0202 0.0174 0.0154 0.0140 0.0130 0.0122 0.0115 0.0110 0.0105 0.0102 0.0098 0.0088 0.0084 0.0081 0.0077 0.0072 0.0069 0.0063 0.0059
7.80 0.1079 0.0683 0.0492 0.0383 0.0315 0.0268 0.0235 0.0210 0.0190 0.0163 0.0144 0.0130 0.0120 0.0112 0.0106 0.0101 0.0097 0.0093 0.0090 0.0080 0.0076 0.0074 0.0070 0.0065 0.0063 0.0056 0.0053
8.00 0.1056 0.0663 0.0474 0.0367 0.0300 0.0255 0.0222 0.0198 0.0179 0.0152 0.0134 0.0121 0.0111 0.0104 0.0098 0.0093 0.0089 0.0085 0.0083 0.0073 0.0070 0.0067 0.0064 0.0059 0.0057 0.0051 0.0048
8.20 0.1034 0.0644 0.0458 0.0353 0.0287 0.0242 0.0210 0.0187 0.0169 0.0143 0.0125 0.0113 0.0103 0.0096 0.0090 0.0086 0.0082 0.0078 0.0076 0.0066 0.0063 0.0061 0.0058 0.0053 0.0051 0.0045 0.0043
8.40 0.1013 0.0626 0.0442 0.0339 0.0274 0.0230 0.0199 0.0176 0.0159 0.0134 0.0117 0.0105 0.0096 0.0089 0.0083 0.0079 0.0075 0.0072 0.0069 0.0061 0.0058 0.0056 0.0052 0.0048 0.0046 0.0041 0.0038
8.60 0.0993 0.0609 0.0427 0.0325 0.0262 0.0219 0.0189 0.0167 0.0150 0.0125 0.0109 0.0098 0.0089 0.0082 0.0077 0.0073 0.0069 0.0066 0.0064 0.0055 0.0053 0.0051 0.0048 0.0044 0.0042 0.0037 0.0034
8.80 0.0973 0.0592 0.0413 0.0313 0.0251 0.0209 0.0180 0.0158 0.0141 0.0118 0.0102 0.0091 0.0083 0.0076 0.0071 0.0067 0.0064 0.0061 0.0059 0.0051 0.0048 0.0046 0.0043 0.0040 0.0038 0.0033 0.0031
9.00 0.0955 0.0577 0.0399 0.0301 0.0240 0.0199 0.0171 0.0150 0.0133 0.0111 0.0096 0.0085 0.0077 0.0071 0.0066 0.0062 0.0059 0.0056 0.0054 0.0046 0.0044 0.0042 0.0039 0.0036 0.0034 0.0030 0.0028
9.20 0.0937 0.0562 0.0387 0.0290 0.0230 0.0190 0.0162 0.0142 0.0126 0.0104 0.0089 0.0079 0.0072 0.0066 0.0061 0.0057 0.0054 0.0052 0.0050 0.0042 0.0040 0.0038 0.0036 0.0033 0.0031 0.0027 0.0025
9.40 0.0919 0.0547 0.0374 0.0279 0.0221 0.0182 0.0154 0.0134 0.0119 0.0098 0.0084 0.0074 0.0067 0.0061 0.0057 0.0053 0.0050 0.0048 0.0046 0.0039 0.0037 0.0035 0.0033 0.0030 0.0028 0.0024 0.0022
9.60 0.0903 0.0534 0.0363 0.0269 0.0212 0.0174 0.0147 0.0128 0.0113 0.0092 0.0079 0.0069 0.0062 0.0057 0.0052 0.0049 0.0046 0.0044 0.0042 0.0036 0.0033 0.0032 0.0030 0.0027 0.0025 0.0022 0.0020
9.80 0.0887 0.0520 0.0352 0.0259 0.0203 0.0166 0.0140 0.0121 0.0107 0.0087 0.0074 0.0065 0.0058 0.0053 0.0049 0.0045 0.0043 0.0041 0.0039 0.0033 0.0031 0.0029 0.0027 0.0024 0.0023 0.0020 0.0018
10.00 0.0871 0.0508 0.0341 0.0250 0.0195 0.0159 0.0133 0.0115 0.0101 0.0082 0.0069 0.0060 0.0054 0.0049 0.0045 0.0042 0.0040 0.0037 0.0036 0.0030 0.0028 0.0027 0.0025 0.0022 0.0021 0.0018 0.0016
Table 3-B: dfB = 2

F 2 3 4 5 6 7 8 9 10 12 14 16 18 20 22 24 26 28 30 40 45 50 60 80 100 250 1000

1.00 0.5000 0.4648 0.4444 0.4312 0.4219 0.4149 0.4096 0.4053 0.4019 0.3966 0.3927 0.3897 0.3874 0.3855 0.3840 0.3827 0.3816 0.3806 0.3798 0.3769 0.3759 0.3751 0.3739 0.3724 0.3715 0.3693 0.3682
1.10 0.4762 0.4382 0.4162 0.4019 0.3918 0.3842 0.3784 0.3738 0.3700 0.3642 0.3600 0.3568 0.3542 0.3522 0.3505 0.3491 0.3479 0.3468 0.3459 0.3427 0.3417 0.3408 0.3395 0.3379 0.3369 0.3345 0.3333
1.20 0.4545 0.4141 0.3906 0.3753 0.3644 0.3564 0.3501 0.3452 0.3411 0.3349 0.3304 0.3269 0.3242 0.3220 0.3202 0.3186 0.3173 0.3162 0.3152 0.3118 0.3106 0.3097 0.3083 0.3066 0.3055 0.3029 0.3016
1.30 0.4348 0.3921 0.3673 0.3511 0.3396 0.3311 0.3244 0.3192 0.3149 0.3083 0.3035 0.2998 0.2969 0.2946 0.2927 0.2910 0.2897 0.2885 0.2874 0.2838 0.2826 0.2816 0.2801 0.2782 0.2771 0.2744 0.2730
1.40 0.4167 0.3720 0.3460 0.3290 0.3170 0.3080 0.3011 0.2955 0.2910 0.2841 0.2791 0.2752 0.2722 0.2697 0.2677 0.2660 0.2646 0.2633 0.2622 0.2584 0.2571 0.2561 0.2545 0.2526 0.2514 0.2485 0.2471
1.50 0.4000 0.3536 0.3265 0.3088 0.2963 0.2870 0.2798 0.2740 0.2693 0.2621 0.2569 0.2529 0.2497 0.2472 0.2451 0.2433 0.2418 0.2405 0.2394 0.2354 0.2341 0.2330 0.2314 0.2293 0.2281 0.2251 0.2236
1.60 0.3846 0.3366 0.3086 0.2903 0.2774 0.2678 0.2603 0.2544 0.2495 0.2421 0.2367 0.2326 0.2293 0.2267 0.2245 0.2227 0.2211 0.2198 0.2186 0.2145 0.2132 0.2121 0.2104 0.2083 0.2070 0.2040 0.2024
1.70 0.3704 0.3209 0.2922 0.2734 0.2601 0.2502 0.2425 0.2364 0.2315 0.2239 0.2183 0.2141 0.2107 0.2080 0.2058 0.2040 0.2024 0.2010 0.1998 0.1956 0.1942 0.1931 0.1914 0.1892 0.1879 0.1848 0.1832
1.80 0.3571 0.3065 0.2770 0.2577 0.2441 0.2340 0.2262 0.2200 0.2149 0.2072 0.2015 0.1972 0.1938 0.1911 0.1888 0.1869 0.1853 0.1839 0.1827 0.1784 0.1770 0.1758 0.1741 0.1719 0.1706 0.1674 0.1658
1.90 0.3448 0.2930 0.2630 0.2433 0.2295 0.2192 0.2113 0.2049 0.1998 0.1919 0.1862 0.1818 0.1784 0.1756 0.1733 0.1714 0.1698 0.1684 0.1671 0.1628 0.1614 0.1602 0.1585 0.1563 0.1549 0.1517 0.1501
2.00 0.3333 0.2806 0.2500 0.2300 0.2160 0.2056 0.1975 0.1911 0.1859 0.1780 0.1722 0.1678 0.1643 0.1615 0.1592 0.1573 0.1556 0.1542 0.1530 0.1486 0.1472 0.1460 0.1443 0.1420 0.1407 0.1375 0.1359
2.05 0.3279 0.2747 0.2439 0.2238 0.2096 0.1992 0.1911 0.1847 0.1794 0.1714 0.1656 0.1612 0.1577 0.1549 0.1526 0.1507 0.1490 0.1476 0.1464 0.1420 0.1406 0.1394 0.1377 0.1354 0.1341 0.1309 0.1293
2.10 0.3226 0.2690 0.2380 0.2177 0.2035 0.1930 0.1849 0.1784 0.1732 0.1652 0.1594 0.1549 0.1515 0.1486 0.1463 0.1444 0.1427 0.1413 0.1401 0.1358 0.1343 0.1331 0.1314 0.1292 0.1278 0.1246 0.1230
2.15 0.3175 0.2634 0.2323 0.2119 0.1977 0.1871 0.1790 0.1725 0.1672 0.1592 0.1534 0.1489 0.1454 0.1426 0.1403 0.1384 0.1367 0.1353 0.1341 0.1298 0.1283 0.1271 0.1254 0.1232 0.1218 0.1186 0.1170
2.20 0.3125 0.2581 0.2268 0.2064 0.1920 0.1814 0.1732 0.1668 0.1615 0.1535 0.1476 0.1432 0.1397 0.1369 0.1346 0.1327 0.1310 0.1296 0.1284 0.1240 0.1226 0.1214 0.1197 0.1175 0.1161 0.1129 0.1113
2.25 0.3077 0.2530 0.2215 0.2010 0.1866 0.1760 0.1678 0.1613 0.1560 0.1480 0.1421 0.1377 0.1342 0.1314 0.1291 0.1272 0.1255 0.1241 0.1229 0.1186 0.1171 0.1160 0.1142 0.1120 0.1107 0.1075 0.1059
2.30 0.3030 0.2480 0.2163 0.1958 0.1814 0.1707 0.1625 0.1560 0.1507 0.1427 0.1369 0.1324 0.1290 0.1262 0.1239 0.1219 0.1203 0.1189 0.1177 0.1134 0.1119 0.1108 0.1090 0.1069 0.1055 0.1024 0.1008
2.35 0.2985 0.2432 0.2114 0.1908 0.1763 0.1657 0.1575 0.1510 0.1457 0.1377 0.1318 0.1274 0.1239 0.1212 0.1189 0.1169 0.1153 0.1139 0.1127 0.1084 0.1070 0.1058 0.1041 0.1019 0.1006 0.0975 0.0959
2.40 0.2941 0.2385 0.2066 0.1859 0.1715 0.1608 0.1526 0.1461 0.1408 0.1328 0.1270 0.1226 0.1191 0.1164 0.1141 0.1122 0.1105 0.1091 0.1079 0.1037 0.1022 0.1011 0.0994 0.0972 0.0959 0.0928 0.0912
2.45 0.2899 0.2340 0.2020 0.1813 0.1668 0.1561 0.1479 0.1414 0.1362 0.1282 0.1224 0.1180 0.1145 0.1118 0.1095 0.1076 0.1060 0.1046 0.1034 0.0991 0.0977 0.0966 0.0949 0.0927 0.0915 0.0884 0.0868
2.50 0.2857 0.2296 0.1975 0.1768 0.1623 0.1516 0.1434 0.1369 0.1317 0.1237 0.1179 0.1136 0.1101 0.1074 0.1051 0.1032 0.1016 0.1002 0.0990 0.0948 0.0934 0.0923 0.0906 0.0885 0.0872 0.0841 0.0826
2.55 0.2817 0.2254 0.1932 0.1724 0.1579 0.1473 0.1391 0.1326 0.1274 0.1194 0.1137 0.1093 0.1059 0.1032 0.1009 0.0990 0.0974 0.0961 0.0949 0.0907 0.0893 0.0882 0.0865 0.0844 0.0832 0.0801 0.0786
2.60 0.2778 0.2213 0.1890 0.1682 0.1537 0.1431 0.1349 0.1285 0.1232 0.1153 0.1096 0.1053 0.1019 0.0992 0.0969 0.0950 0.0935 0.0921 0.0909 0.0868 0.0854 0.0843 0.0826 0.0805 0.0793 0.0763 0.0748
2.65 0.2740 0.2173 0.1850 0.1642 0.1497 0.1391 0.1309 0.1245 0.1193 0.1114 0.1057 0.1014 0.0980 0.0953 0.0931 0.0912 0.0897 0.0883 0.0871 0.0830 0.0817 0.0806 0.0789 0.0768 0.0756 0.0726 0.0711
-312-
Back to TOC
F 2 3 4 5 6 7 8 9 10 12 14 16 18 20 22 24 26 28 30 40 45 50 60 80 100 250 1000
2.70 0.2703 0.2134 0.1811 0.1603 0.1458 0.1352 0.1270 0.1206 0.1155 0.1076 0.1019 0.0976 0.0943 0.0916 0.0894 0.0876 0.0860 0.0847 0.0835 0.0794 0.0781 0.0770 0.0754 0.0733 0.0721 0.0692 0.0677
2.75 0.2667 0.2097 0.1773 0.1565 0.1420 0.1314 0.1233 0.1169 0.1118 0.1040 0.0983 0.0941 0.0907 0.0881 0.0859 0.0841 0.0825 0.0812 0.0801 0.0760 0.0747 0.0736 0.0720 0.0700 0.0688 0.0659 0.0644
2.80 0.2632 0.2060 0.1736 0.1528 0.1384 0.1278 0.1197 0.1134 0.1082 0.1005 0.0949 0.0906 0.0873 0.0847 0.0825 0.0807 0.0792 0.0779 0.0767 0.0728 0.0714 0.0704 0.0688 0.0668 0.0656 0.0627 0.0613
2.85 0.2597 0.2025 0.1700 0.1493 0.1349 0.1243 0.1163 0.1099 0.1048 0.0971 0.0915 0.0874 0.0841 0.0815 0.0793 0.0775 0.0760 0.0747 0.0736 0.0696 0.0683 0.0673 0.0657 0.0637 0.0626 0.0597 0.0583
2.90 0.2564 0.1990 0.1666 0.1458 0.1315 0.1210 0.1129 0.1066 0.1016 0.0939 0.0884 0.0842 0.0810 0.0784 0.0762 0.0745 0.0730 0.0717 0.0706 0.0667 0.0654 0.0643 0.0628 0.0608 0.0597 0.0569 0.0555
2.95 0.2532 0.1957 0.1632 0.1425 0.1282 0.1177 0.1097 0.1035 0.0984 0.0908 0.0853 0.0812 0.0780 0.0754 0.0733 0.0715 0.0700 0.0688 0.0677 0.0638 0.0625 0.0615 0.0600 0.0581 0.0569 0.0542 0.0528
3.00 0.2500 0.1925 0.1600 0.1393 0.1250 0.1146 0.1066 0.1004 0.0954 0.0878 0.0824 0.0783 0.0751 0.0725 0.0705 0.0687 0.0673 0.0660 0.0649 0.0611 0.0598 0.0588 0.0573 0.0554 0.0543 0.0516 0.0502
3.05 0.2469 0.1893 0.1568 0.1362 0.1219 0.1115 0.1036 0.0974 0.0924 0.0849 0.0795 0.0755 0.0723 0.0698 0.0677 0.0660 0.0646 0.0633 0.0623 0.0585 0.0573 0.0563 0.0548 0.0529 0.0518 0.0491 0.0478
3.10 0.2439 0.1862 0.1538 0.1332 0.1190 0.1086 0.1007 0.0946 0.0896 0.0822 0.0768 0.0728 0.0697 0.0672 0.0651 0.0635 0.0620 0.0608 0.0597 0.0560 0.0548 0.0538 0.0523 0.0505 0.0494 0.0468 0.0455
3.15 0.2410 0.1832 0.1508 0.1302 0.1161 0.1058 0.0980 0.0918 0.0869 0.0795 0.0742 0.0702 0.0671 0.0647 0.0627 0.0610 0.0596 0.0584 0.0573 0.0537 0.0524 0.0515 0.0500 0.0482 0.0471 0.0446 0.0433
3.20 0.2381 0.1803 0.1479 0.1274 0.1133 0.1030 0.0953 0.0892 0.0843 0.0769 0.0717 0.0678 0.0647 0.0623 0.0603 0.0586 0.0572 0.0560 0.0550 0.0514 0.0502 0.0492 0.0478 0.0460 0.0450 0.0424 0.0412
3.25 0.2353 0.1775 0.1451 0.1246 0.1106 0.1004 0.0927 0.0866 0.0818 0.0745 0.0693 0.0654 0.0624 0.0600 0.0580 0.0564 0.0550 0.0538 0.0528 0.0492 0.0480 0.0471 0.0457 0.0439 0.0429 0.0404 0.0392
3.30 0.2326 0.1747 0.1424 0.1220 0.1080 0.0978 0.0901 0.0841 0.0793 0.0721 0.0670 0.0631 0.0601 0.0577 0.0558 0.0542 0.0528 0.0517 0.0507 0.0471 0.0460 0.0451 0.0437 0.0420 0.0409 0.0385 0.0373
3.35 0.2299 0.1720 0.1398 0.1194 0.1054 0.0954 0.0877 0.0818 0.0770 0.0698 0.0647 0.0609 0.0580 0.0556 0.0537 0.0521 0.0508 0.0496 0.0486 0.0452 0.0440 0.0431 0.0418 0.0401 0.0391 0.0367 0.0355
3.40 0.2273 0.1694 0.1372 0.1169 0.1030 0.0930 0.0854 0.0795 0.0747 0.0676 0.0626 0.0588 0.0559 0.0536 0.0517 0.0501 0.0488 0.0477 0.0467 0.0433 0.0422 0.0413 0.0399 0.0383 0.0373 0.0349 0.0338
3.45 0.2247 0.1668 0.1347 0.1144 0.1006 0.0906 0.0831 0.0772 0.0725 0.0655 0.0605 0.0568 0.0539 0.0516 0.0497 0.0482 0.0469 0.0458 0.0448 0.0415 0.0404 0.0395 0.0382 0.0365 0.0356 0.0333 0.0321
3.50 0.2222 0.1643 0.1322 0.1121 0.0983 0.0884 0.0809 0.0751 0.0704 0.0635 0.0585 0.0548 0.0520 0.0497 0.0479 0.0464 0.0451 0.0440 0.0430 0.0397 0.0387 0.0378 0.0365 0.0349 0.0339 0.0317 0.0306
3.55 0.2198 0.1619 0.1299 0.1098 0.0961 0.0862 0.0788 0.0730 0.0684 0.0615 0.0566 0.0530 0.0502 0.0479 0.0461 0.0446 0.0433 0.0423 0.0413 0.0381 0.0370 0.0362 0.0349 0.0333 0.0324 0.0302 0.0291
3.60 0.2174 0.1595 0.1276 0.1075 0.0939 0.0841 0.0767 0.0710 0.0664 0.0596 0.0548 0.0512 0.0484 0.0462 0.0444 0.0429 0.0417 0.0406 0.0397 0.0365 0.0355 0.0346 0.0334 0.0318 0.0309 0.0287 0.0277
3.65 0.2151 0.1572 0.1253 0.1054 0.0918 0.0821 0.0747 0.0691 0.0645 0.0578 0.0530 0.0494 0.0467 0.0445 0.0428 0.0413 0.0401 0.0390 0.0381 0.0350 0.0340 0.0331 0.0319 0.0304 0.0295 0.0274 0.0263
3.70 0.2128 0.1549 0.1231 0.1032 0.0898 0.0801 0.0728 0.0672 0.0627 0.0560 0.0513 0.0478 0.0451 0.0429 0.0412 0.0398 0.0385 0.0375 0.0366 0.0335 0.0325 0.0317 0.0305 0.0290 0.0282 0.0261 0.0251
3.75 0.2105 0.1527 0.1210 0.1012 0.0878 0.0782 0.0710 0.0654 0.0609 0.0543 0.0496 0.0462 0.0435 0.0414 0.0397 0.0383 0.0371 0.0361 0.0352 0.0322 0.0312 0.0304 0.0292 0.0278 0.0269 0.0249 0.0238
3.80 0.2083 0.1506 0.1189 0.0992 0.0859 0.0763 0.0692 0.0636 0.0592 0.0527 0.0481 0.0446 0.0420 0.0399 0.0382 0.0368 0.0357 0.0347 0.0338 0.0308 0.0299 0.0291 0.0279 0.0265 0.0257 0.0237 0.0227
3.85 0.2062 0.1485 0.1169 0.0973 0.0840 0.0745 0.0674 0.0619 0.0576 0.0511 0.0465 0.0431 0.0406 0.0385 0.0368 0.0355 0.0343 0.0333 0.0325 0.0296 0.0286 0.0279 0.0267 0.0253 0.0245 0.0226 0.0216
3.90 0.2041 0.1464 0.1149 0.0954 0.0822 0.0728 0.0657 0.0603 0.0560 0.0496 0.0451 0.0417 0.0392 0.0371 0.0355 0.0342 0.0330 0.0321 0.0312 0.0284 0.0274 0.0267 0.0256 0.0242 0.0234 0.0215 0.0206
3.95 0.2020 0.1444 0.1130 0.0935 0.0804 0.0711 0.0641 0.0587 0.0544 0.0481 0.0436 0.0403 0.0378 0.0358 0.0342 0.0329 0.0318 0.0308 0.0300 0.0272 0.0263 0.0255 0.0245 0.0231 0.0223 0.0205 0.0196
4.00 0.2000 0.1424 0.1111 0.0917 0.0787 0.0694 0.0625 0.0572 0.0529 0.0467 0.0423 0.0390 0.0365 0.0346 0.0330 0.0317 0.0306 0.0296 0.0288 0.0261 0.0252 0.0245 0.0234 0.0221 0.0213 0.0195 0.0186
4.10 0.1961 0.1386 0.1075 0.0883 0.0754 0.0663 0.0595 0.0542 0.0501 0.0440 0.0397 0.0365 0.0341 0.0322 0.0307 0.0294 0.0283 0.0274 0.0267 0.0240 0.0231 0.0224 0.0214 0.0202 0.0194 0.0177 0.0169
4.20 0.1923 0.1350 0.1041 0.0850 0.0723 0.0633 0.0566 0.0515 0.0474 0.0414 0.0373 0.0342 0.0318 0.0300 0.0285 0.0273 0.0263 0.0254 0.0247 0.0221 0.0213 0.0206 0.0196 0.0184 0.0177 0.0161 0.0153
4.30 0.1887 0.1315 0.1008 0.0820 0.0694 0.0605 0.0539 0.0489 0.0449 0.0391 0.0350 0.0320 0.0298 0.0280 0.0265 0.0253 0.0244 0.0235 0.0228 0.0203 0.0195 0.0189 0.0180 0.0168 0.0162 0.0146 0.0138
4.40 0.1852 0.1282 0.0977 0.0790 0.0666 0.0579 0.0514 0.0465 0.0426 0.0369 0.0329 0.0300 0.0278 0.0261 0.0247 0.0236 0.0226 0.0218 0.0211 0.0187 0.0180 0.0174 0.0165 0.0154 0.0147 0.0132 0.0125
4.50 0.1818 0.1250 0.0947 0.0762 0.0640 0.0554 0.0490 0.0442 0.0404 0.0348 0.0310 0.0281 0.0260 0.0243 0.0230 0.0219 0.0210 0.0202 0.0195 0.0173 0.0165 0.0160 0.0151 0.0141 0.0134 0.0120 0.0113
4.60 0.1786 0.1219 0.0918 0.0736 0.0615 0.0530 0.0468 0.0421 0.0383 0.0329 0.0291 0.0264 0.0243 0.0227 0.0214 0.0204 0.0195 0.0187 0.0181 0.0159 0.0152 0.0147 0.0138 0.0129 0.0123 0.0109 0.0103
4.70 0.1754 0.1190 0.0891 0.0710 0.0591 0.0508 0.0447 0.0400 0.0364 0.0311 0.0274 0.0248 0.0228 0.0212 0.0200 0.0189 0.0181 0.0174 0.0168 0.0147 0.0140 0.0135 0.0127 0.0118 0.0112 0.0099 0.0093
4.80 0.1724 0.1162 0.0865 0.0686 0.0569 0.0487 0.0427 0.0381 0.0346 0.0294 0.0259 0.0233 0.0213 0.0198 0.0186 0.0176 0.0168 0.0161 0.0155 0.0135 0.0129 0.0124 0.0116 0.0107 0.0102 0.0090 0.0084
4.90 0.1695 0.1135 0.0840 0.0663 0.0548 0.0467 0.0408 0.0363 0.0329 0.0278 0.0244 0.0219 0.0200 0.0185 0.0174 0.0164 0.0156 0.0150 0.0144 0.0125 0.0119 0.0114 0.0107 0.0098 0.0093 0.0082 0.0076
5.00 0.1667 0.1109 0.0816 0.0642 0.0527 0.0448 0.0390 0.0346 0.0313 0.0263 0.0230 0.0206 0.0188 0.0173 0.0162 0.0153 0.0145 0.0139 0.0134 0.0115 0.0109 0.0105 0.0098 0.0090 0.0085 0.0074 0.0069
5.10 0.1639 0.1083 0.0793 0.0621 0.0508 0.0430 0.0373 0.0331 0.0297 0.0249 0.0217 0.0193 0.0176 0.0162 0.0151 0.0143 0.0135 0.0129 0.0124 0.0106 0.0101 0.0096 0.0090 0.0082 0.0078 0.0067 0.0063
5.20 0.1613 0.1059 0.0772 0.0601 0.0490 0.0413 0.0357 0.0315 0.0283 0.0236 0.0205 0.0182 0.0165 0.0152 0.0141 0.0133 0.0126 0.0120 0.0115 0.0098 0.0093 0.0089 0.0083 0.0075 0.0071 0.0061 0.0057
5.30 0.1587 0.1036 0.0751 0.0582 0.0472 0.0397 0.0342 0.0301 0.0270 0.0224 0.0193 0.0171 0.0155 0.0142 0.0132 0.0124 0.0117 0.0112 0.0107 0.0091 0.0086 0.0082 0.0076 0.0069 0.0065 0.0056 0.0051
5.40 0.1563 0.1014 0.0730 0.0563 0.0456 0.0381 0.0328 0.0288 0.0257 0.0213 0.0183 0.0161 0.0146 0.0133 0.0124 0.0116 0.0109 0.0104 0.0099 0.0084 0.0079 0.0075 0.0070 0.0063 0.0059 0.0051 0.0046
5.50 0.1538 0.0992 0.0711 0.0546 0.0440 0.0367 0.0314 0.0275 0.0245 0.0202 0.0173 0.0152 0.0137 0.0125 0.0116 0.0108 0.0102 0.0097 0.0092 0.0078 0.0073 0.0069 0.0064 0.0058 0.0054 0.0046 0.0042
5.60 0.1515 0.0971 0.0693 0.0529 0.0424 0.0353 0.0301 0.0263 0.0234 0.0191 0.0163 0.0143 0.0129 0.0117 0.0108 0.0101 0.0095 0.0090 0.0086 0.0072 0.0067 0.0064 0.0059 0.0053 0.0050 0.0042 0.0038
5.70 0.1493 0.0951 0.0675 0.0513 0.0410 0.0340 0.0289 0.0252 0.0223 0.0182 0.0155 0.0135 0.0121 0.0110 0.0101 0.0094 0.0089 0.0084 0.0080 0.0066 0.0062 0.0059 0.0054 0.0049 0.0045 0.0038 0.0035
5.80 0.1471 0.0931 0.0657 0.0498 0.0396 0.0327 0.0278 0.0241 0.0213 0.0173 0.0146 0.0128 0.0114 0.0103 0.0095 0.0088 0.0083 0.0078 0.0074 0.0061 0.0057 0.0054 0.0050 0.0044 0.0041 0.0034 0.0031
5.90 0.1449 0.0913 0.0641 0.0483 0.0383 0.0315 0.0267 0.0231 0.0203 0.0164 0.0139 0.0120 0.0107 0.0097 0.0089 0.0082 0.0077 0.0073 0.0069 0.0057 0.0053 0.0050 0.0046 0.0041 0.0038 0.0031 0.0028
6.00 0.1429 0.0894 0.0625 0.0469 0.0370 0.0304 0.0256 0.0221 0.0194 0.0156 0.0131 0.0114 0.0101 0.0091 0.0083 0.0077 0.0072 0.0068 0.0064 0.0053 0.0049 0.0046 0.0042 0.0037 0.0035 0.0028 0.0026
6.10 0.1408 0.0877 0.0610 0.0456 0.0358 0.0293 0.0246 0.0212 0.0185 0.0149 0.0124 0.0107 0.0095 0.0085 0.0078 0.0072 0.0067 0.0063 0.0060 0.0049 0.0045 0.0043 0.0039 0.0034 0.0032 0.0026 0.0023
6.20 0.1389 0.0860 0.0595 0.0443 0.0347 0.0282 0.0237 0.0203 0.0177 0.0141 0.0118 0.0101 0.0089 0.0080 0.0073 0.0068 0.0063 0.0059 0.0056 0.0045 0.0042 0.0039 0.0036 0.0031 0.0029 0.0024 0.0021
6.30 0.1370 0.0843 0.0581 0.0430 0.0336 0.0272 0.0227 0.0195 0.0170 0.0135 0.0112 0.0096 0.0084 0.0076 0.0069 0.0063 0.0059 0.0055 0.0052 0.0042 0.0039 0.0036 0.0033 0.0029 0.0026 0.0021 0.0019
6.40 0.1351 0.0827 0.0567 0.0418 0.0325 0.0263 0.0219 0.0187 0.0162 0.0128 0.0106 0.0091 0.0080 0.0071 0.0064 0.0059 0.0055 0.0051 0.0048 0.0039 0.0036 0.0034 0.0030 0.0026 0.0024 0.0019 0.0017
6.50 0.1333 0.0812 0.0554 0.0407 0.0315 0.0254 0.0211 0.0179 0.0155 0.0122 0.0101 0.0086 0.0075 0.0067 0.0061 0.0055 0.0051 0.0048 0.0045 0.0036 0.0033 0.0031 0.0028 0.0024 0.0022 0.0018 0.0016
6.60 0.1316 0.0797 0.0541 0.0396 0.0305 0.0245 0.0203 0.0172 0.0149 0.0117 0.0096 0.0081 0.0071 0.0063 0.0057 0.0052 0.0048 0.0045 0.0042 0.0033 0.0031 0.0029 0.0026 0.0022 0.0020 0.0016 0.0014
6.70 0.1299 0.0782 0.0528 0.0385 0.0296 0.0237 0.0195 0.0165 0.0143 0.0111 0.0091 0.0077 0.0067 0.0059 0.0053 0.0049 0.0045 0.0042 0.0039 0.0031 0.0028 0.0026 0.0024 0.0020 0.0019 0.0015 0.0013
6.80 0.1282 0.0768 0.0517 0.0375 0.0287 0.0229 0.0188 0.0159 0.0137 0.0106 0.0086 0.0073 0.0063 0.0056 0.0050 0.0046 0.0042 0.0039 0.0037 0.0029 0.0026 0.0024 0.0022 0.0019 0.0017 0.0013 0.0012
6.90 0.1266 0.0755 0.0505 0.0365 0.0278 0.0221 0.0181 0.0153 0.0131 0.0101 0.0082 0.0069 0.0060 0.0053 0.0047 0.0043 0.0039 0.0037 0.0034 0.0027 0.0024 0.0023 0.0020 0.0017 0.0016 0.0012 0.0011
7.00 0.1250 0.0741 0.0494 0.0355 0.0270 0.0214 0.0175 0.0147 0.0126 0.0097 0.0078 0.0065 0.0056 0.0050 0.0044 0.0040 0.0037 0.0034 0.0032 0.0025 0.0023 0.0021 0.0019 0.0016 0.0014 0.0011 0.0010
7.20 0.1220 0.0716 0.0473 0.0337 0.0254 0.0200 0.0163 0.0136 0.0116 0.0088 0.0071 0.0059 0.0050 0.0044 0.0039 0.0036 0.0032 0.0030 0.0028 0.0021 0.0019 0.0018 0.0016 0.0013 0.0012 0.0009 0.0008
7.40 0.1190 0.0692 0.0453 0.0320 0.0240 0.0188 0.0152 0.0126 0.0107 0.0081 0.0064 0.0053 0.0045 0.0039 0.0035 0.0031 0.0029 0.0026 0.0024 0.0018 0.0017 0.0015 0.0013 0.0011 0.0010 0.0008 0.0006
7.60 0.1163 0.0669 0.0434 0.0305 0.0227 0.0176 0.0141 0.0117 0.0098 0.0074 0.0058 0.0048 0.0040 0.0035 0.0031 0.0028 0.0025 0.0023 0.0021 0.0016 0.0014 0.0013 0.0011 0.0010 0.0008 0.0006 0.0005
7.80 0.1136 0.0648 0.0416 0.0290 0.0214 0.0165 0.0132 0.0108 0.0091 0.0068 0.0053 0.0043 0.0036 0.0031 0.0028 0.0025 0.0022 0.0020 0.0019 0.0014 0.0012 0.0011 0.0010 0.0008 0.0007 0.0005 0.0004
8.00 0.1111 0.0627 0.0400 0.0277 0.0203 0.0156 0.0123 0.0101 0.0084 0.0062 0.0048 0.0039 0.0033 0.0028 0.0024 0.0022 0.0020 0.0018 0.0016 0.0012 0.0011 0.0010 0.0008 0.0007 0.0006 0.0004 0.0004
8.20 0.1087 0.0608 0.0384 0.0264 0.0192 0.0146 0.0116 0.0094 0.0078 0.0057 0.0044 0.0035 0.0029 0.0025 0.0022 0.0019 0.0017 0.0016 0.0014 0.0010 0.0009 0.0008 0.0007 0.0006 0.0005 0.0004 0.0003
8.40 0.1064 0.0590 0.0370 0.0252 0.0182 0.0138 0.0108 0.0087 0.0072 0.0052 0.0040 0.0032 0.0026 0.0022 0.0019 0.0017 0.0015 0.0014 0.0013 0.0009 0.0008 0.0007 0.0006 0.0005 0.0004 0.0003 0.0002
8.60 0.1042 0.0572 0.0356 0.0241 0.0173 0.0130 0.0102 0.0082 0.0067 0.0048 0.0037 0.0029 0.0024 0.0020 0.0017 0.0015 0.0014 0.0012 0.0011 0.0008 0.0007 0.0006 0.0005 0.0004 0.0004 0.0002 0.0002
-313-
Back to TOC
F 2 3 4 5 6 7 8 9 10 12 14 16 18 20 22 24 26 28 30 40 45 50 60 80 100 250 1000
8.80 0.1020 0.0556 0.0343 0.0230 0.0164 0.0123 0.0095 0.0076 0.0062 0.0044 0.0034 0.0026 0.0022 0.0018 0.0016 0.0014 0.0012 0.0011 0.0010 0.0007 0.0006 0.0005 0.0004 0.0004 0.0003 0.0002 0.0002
9.00 0.1000 0.0540 0.0331 0.0220 0.0156 0.0116 0.0090 0.0071 0.0058 0.0041 0.0031 0.0024 0.0020 0.0016 0.0014 0.0012 0.0011 0.0010 0.0009 0.0006 0.0005 0.0005 0.0004 0.0003 0.0003 0.0002 0.0001
9.20 0.0980 0.0525 0.0319 0.0211 0.0149 0.0110 0.0084 0.0067 0.0054 0.0038 0.0028 0.0022 0.0018 0.0015 0.0012 0.0011 0.0010 0.0008 0.0008 0.0005 0.0004 0.0004 0.0003 0.0003 0.0002 0.0001 0.0001
9.40 0.0962 0.0511 0.0308 0.0202 0.0142 0.0104 0.0079 0.0063 0.0050 0.0035 0.0026 0.0020 0.0016 0.0013 0.0011 0.0010 0.0008 0.0008 0.0007 0.0005 0.0004 0.0003 0.0003 0.0002 0.0002 0.0001 0.0001
9.60 0.0943 0.0497 0.0297 0.0194 0.0135 0.0099 0.0075 0.0059 0.0047 0.0032 0.0024 0.0018 0.0015 0.0012 0.0010 0.0009 0.0008 0.0007 0.0006 0.0004 0.0003 0.0003 0.0002 0.0002 0.0002 0.0001 0.0001
9.80 0.0926 0.0484 0.0287 0.0186 0.0129 0.0093 0.0071 0.0055 0.0044 0.0030 0.0022 0.0017 0.0013 0.0011 0.0009 0.0008 0.0007 0.0006 0.0005 0.0003 0.0003 0.0003 0.0002 0.0002 0.0001 0.0001 0.0001
10.00 0.0909 0.0471 0.0278 0.0179 0.0123 0.0089 0.0067 0.0052 0.0041 0.0028 0.0020 0.0015 0.0012 0.0010 0.0008 0.0007 0.0006 0.0005 0.0005 0.0003 0.0003 0.0002 0.0002 0.0001 0.0001 0.0001 0.0001
Table 3-C: dfB = 3

F 2 3 4 5 6 7 8 9 10 12 14 16 18 20 22 24 26 28 30 40 45 50 60 80 100 250 1000
1.00 0.5352 0.5000 0.4789 0.4649 0.4547 0.4471 0.4411 0.4363 0.4323 0.4262 0.4217 0.4182 0.4155 0.4133 0.4114 0.4098 0.4085 0.4074 0.4064 0.4028 0.4016 0.4006 0.3992 0.3973 0.3962 0.3935 0.3921
1.10 0.5087 0.4697 0.4462 0.4305 0.4191 0.4105 0.4038 0.3984 0.3939 0.3870 0.3819 0.3779 0.3748 0.3723 0.3702 0.3684 0.3669 0.3656 0.3644 0.3604 0.3590 0.3579 0.3562 0.3541 0.3528 0.3497 0.3482
1.20 0.4846 0.4422 0.4166 0.3993 0.3869 0.3774 0.3700 0.3641 0.3591 0.3515 0.3459 0.3415 0.3381 0.3353 0.3330 0.3310 0.3293 0.3279 0.3266 0.3221 0.3206 0.3194 0.3175 0.3152 0.3138 0.3104 0.3086
1.30 0.4626 0.4172 0.3897 0.3711 0.3577 0.3475 0.3395 0.3331 0.3277 0.3195 0.3134 0.3087 0.3050 0.3020 0.2994 0.2973 0.2955 0.2940 0.2926 0.2877 0.2861 0.2848 0.2828 0.2802 0.2787 0.2750 0.2731
1.40 0.4424 0.3944 0.3652 0.3455 0.3312 0.3204 0.3119 0.3050 0.2994 0.2906 0.2842 0.2792 0.2752 0.2720 0.2693 0.2671 0.2651 0.2635 0.2620 0.2569 0.2551 0.2537 0.2516 0.2489 0.2473 0.2433 0.2413
1.50 0.4240 0.3735 0.3429 0.3222 0.3072 0.2958 0.2869 0.2797 0.2738 0.2646 0.2578 0.2526 0.2484 0.2451 0.2423 0.2399 0.2379 0.2361 0.2346 0.2292 0.2274 0.2259 0.2237 0.2209 0.2192 0.2151 0.2130
1.60 0.4069 0.3544 0.3225 0.3009 0.2853 0.2735 0.2642 0.2568 0.2506 0.2411 0.2341 0.2286 0.2244 0.2209 0.2180 0.2155 0.2134 0.2117 0.2101 0.2045 0.2026 0.2011 0.1988 0.1959 0.1942 0.1899 0.1878
1.70 0.3912 0.3368 0.3038 0.2815 0.2654 0.2532 0.2437 0.2360 0.2297 0.2199 0.2127 0.2071 0.2027 0.1991 0.1962 0.1937 0.1916 0.1897 0.1881 0.1824 0.1805 0.1790 0.1766 0.1737 0.1719 0.1675 0.1654
1.80 0.3766 0.3206 0.2867 0.2638 0.2473 0.2348 0.2250 0.2172 0.2107 0.2008 0.1934 0.1878 0.1833 0.1796 0.1766 0.1741 0.1720 0.1701 0.1685 0.1627 0.1608 0.1592 0.1568 0.1539 0.1521 0.1477 0.1455
1.90 0.3631 0.3056 0.2709 0.2475 0.2307 0.2180 0.2081 0.2001 0.1936 0.1835 0.1760 0.1703 0.1658 0.1621 0.1591 0.1566 0.1544 0.1525 0.1509 0.1451 0.1432 0.1416 0.1392 0.1363 0.1345 0.1301 0.1279
2.00 0.3505 0.2918 0.2564 0.2326 0.2156 0.2027 0.1927 0.1846 0.1780 0.1678 0.1604 0.1546 0.1501 0.1464 0.1434 0.1409 0.1387 0.1368 0.1352 0.1294 0.1275 0.1259 0.1236 0.1206 0.1188 0.1145 0.1123
2.05 0.3445 0.2852 0.2496 0.2256 0.2085 0.1955 0.1855 0.1774 0.1708 0.1606 0.1531 0.1474 0.1429 0.1392 0.1362 0.1336 0.1315 0.1296 0.1280 0.1223 0.1203 0.1188 0.1164 0.1135 0.1117 0.1074 0.1053
2.10 0.3387 0.2789 0.2430 0.2189 0.2017 0.1887 0.1786 0.1705 0.1639 0.1537 0.1462 0.1405 0.1360 0.1323 0.1293 0.1268 0.1246 0.1228 0.1212 0.1155 0.1135 0.1120 0.1097 0.1068 0.1050 0.1007 0.0986
2.15 0.3331 0.2729 0.2367 0.2125 0.1952 0.1822 0.1720 0.1639 0.1573 0.1471 0.1397 0.1340 0.1295 0.1258 0.1228 0.1203 0.1182 0.1163 0.1147 0.1091 0.1072 0.1056 0.1033 0.1004 0.0987 0.0945 0.0924
2.20 0.3277 0.2670 0.2306 0.2063 0.1889 0.1759 0.1658 0.1577 0.1510 0.1409 0.1334 0.1278 0.1233 0.1197 0.1167 0.1142 0.1121 0.1102 0.1086 0.1030 0.1011 0.0996 0.0973 0.0945 0.0927 0.0886 0.0865
2.25 0.3224 0.2613 0.2247 0.2004 0.1829 0.1699 0.1598 0.1517 0.1451 0.1349 0.1275 0.1219 0.1174 0.1138 0.1109 0.1084 0.1063 0.1045 0.1029 0.0973 0.0954 0.0939 0.0917 0.0889 0.0872 0.0831 0.0810
2.30 0.3174 0.2559 0.2191 0.1946 0.1772 0.1642 0.1540 0.1460 0.1394 0.1293 0.1219 0.1163 0.1118 0.1083 0.1053 0.1029 0.1008 0.0990 0.0974 0.0919 0.0901 0.0886 0.0864 0.0836 0.0819 0.0779 0.0758
2.35 0.3124 0.2506 0.2137 0.1892 0.1717 0.1587 0.1486 0.1405 0.1339 0.1239 0.1165 0.1109 0.1066 0.1030 0.1001 0.0977 0.0956 0.0938 0.0923 0.0868 0.0850 0.0836 0.0814 0.0786 0.0770 0.0730 0.0710
2.40 0.3077 0.2455 0.2084 0.1839 0.1664 0.1534 0.1433 0.1353 0.1287 0.1187 0.1114 0.1059 0.1016 0.0980 0.0952 0.0928 0.0907 0.0890 0.0874 0.0821 0.0803 0.0788 0.0767 0.0739 0.0723 0.0684 0.0664
2.45 0.3030 0.2405 0.2034 0.1788 0.1614 0.1483 0.1383 0.1303 0.1238 0.1138 0.1066 0.1011 0.0968 0.0933 0.0905 0.0881 0.0861 0.0843 0.0828 0.0775 0.0758 0.0743 0.0722 0.0695 0.0679 0.0641 0.0622
2.50 0.2985 0.2358 0.1985 0.1739 0.1565 0.1435 0.1335 0.1255 0.1190 0.1092 0.1020 0.0965 0.0923 0.0888 0.0860 0.0837 0.0817 0.0800 0.0785 0.0733 0.0715 0.0701 0.0680 0.0654 0.0638 0.0601 0.0582
2.55 0.2942 0.2311 0.1938 0.1692 0.1518 0.1389 0.1289 0.1210 0.1145 0.1047 0.0976 0.0922 0.0880 0.0846 0.0818 0.0795 0.0775 0.0758 0.0744 0.0692 0.0675 0.0661 0.0641 0.0615 0.0600 0.0563 0.0544
2.60 0.2899 0.2267 0.1893 0.1647 0.1473 0.1344 0.1245 0.1166 0.1102 0.1005 0.0934 0.0881 0.0839 0.0806 0.0778 0.0755 0.0736 0.0719 0.0705 0.0654 0.0637 0.0624 0.0604 0.0578 0.0563 0.0527 0.0509
2.65 0.2858 0.2223 0.1849 0.1603 0.1430 0.1301 0.1202 0.1124 0.1061 0.0964 0.0894 0.0842 0.0800 0.0767 0.0740 0.0718 0.0699 0.0682 0.0668 0.0618 0.0602 0.0589 0.0569 0.0544 0.0529 0.0494 0.0476
2.70 0.2818 0.2181 0.1807 0.1561 0.1388 0.1260 0.1162 0.1084 0.1021 0.0925 0.0856 0.0804 0.0764 0.0731 0.0704 0.0682 0.0663 0.0647 0.0633 0.0584 0.0568 0.0555 0.0536 0.0512 0.0497 0.0463 0.0446
2.75 0.2779 0.2141 0.1766 0.1521 0.1348 0.1221 0.1123 0.1046 0.0983 0.0889 0.0820 0.0769 0.0729 0.0697 0.0670 0.0648 0.0630 0.0614 0.0600 0.0552 0.0537 0.0524 0.0505 0.0481 0.0467 0.0433 0.0417
2.80 0.2741 0.2101 0.1727 0.1482 0.1310 0.1183 0.1086 0.1009 0.0947 0.0853 0.0786 0.0735 0.0695 0.0664 0.0638 0.0616 0.0598 0.0583 0.0569 0.0522 0.0507 0.0494 0.0476 0.0453 0.0439 0.0406 0.0390
2.85 0.2704 0.2063 0.1688 0.1444 0.1273 0.1147 0.1050 0.0974 0.0913 0.0820 0.0753 0.0703 0.0664 0.0633 0.0607 0.0586 0.0568 0.0553 0.0540 0.0494 0.0478 0.0466 0.0448 0.0426 0.0412 0.0380 0.0364
2.90 0.2668 0.2026 0.1651 0.1408 0.1237 0.1112 0.1016 0.0940 0.0880 0.0788 0.0722 0.0672 0.0634 0.0603 0.0578 0.0557 0.0540 0.0525 0.0512 0.0467 0.0452 0.0440 0.0422 0.0400 0.0387 0.0356 0.0341
2.95 0.2633 0.1990 0.1616 0.1372 0.1203 0.1078 0.0983 0.0908 0.0848 0.0757 0.0692 0.0643 0.0605 0.0575 0.0551 0.0530 0.0513 0.0498 0.0486 0.0441 0.0427 0.0415 0.0398 0.0376 0.0364 0.0333 0.0318
3.00 0.2599 0.1955 0.1581 0.1339 0.1170 0.1046 0.0951 0.0877 0.0817 0.0728 0.0663 0.0615 0.0578 0.0549 0.0524 0.0504 0.0488 0.0473 0.0461 0.0417 0.0403 0.0392 0.0375 0.0354 0.0342 0.0312 0.0298
3.05 0.2566 0.1921 0.1547 0.1306 0.1138 0.1015 0.0921 0.0847 0.0788 0.0700 0.0636 0.0589 0.0552 0.0523 0.0500 0.0480 0.0463 0.0449 0.0437 0.0395 0.0381 0.0370 0.0353 0.0333 0.0321 0.0292 0.0278
3.10 0.2534 0.1888 0.1515 0.1274 0.1107 0.0985 0.0892 0.0819 0.0761 0.0673 0.0610 0.0564 0.0528 0.0499 0.0476 0.0457 0.0440 0.0427 0.0415 0.0373 0.0360 0.0349 0.0333 0.0313 0.0301 0.0274 0.0260
3.15 0.2502 0.1856 0.1483 0.1243 0.1077 0.0956 0.0864 0.0792 0.0734 0.0647 0.0586 0.0540 0.0504 0.0476 0.0454 0.0435 0.0419 0.0405 0.0394 0.0353 0.0340 0.0330 0.0314 0.0295 0.0283 0.0256 0.0243
3.20 0.2471 0.1825 0.1453 0.1214 0.1048 0.0928 0.0837 0.0765 0.0708 0.0623 0.0562 0.0517 0.0482 0.0455 0.0432 0.0414 0.0398 0.0385 0.0374 0.0334 0.0321 0.0311 0.0296 0.0277 0.0266 0.0240 0.0227
3.25 0.2441 0.1794 0.1423 0.1185 0.1021 0.0901 0.0811 0.0740 0.0684 0.0599 0.0540 0.0495 0.0461 0.0434 0.0412 0.0394 0.0379 0.0366 0.0355 0.0316 0.0304 0.0294 0.0279 0.0261 0.0250 0.0225 0.0212
3.30 0.2412 0.1765 0.1395 0.1157 0.0994 0.0875 0.0786 0.0716 0.0660 0.0577 0.0518 0.0474 0.0441 0.0414 0.0393 0.0375 0.0360 0.0348 0.0337 0.0299 0.0287 0.0277 0.0263 0.0245 0.0235 0.0210 0.0198
3.35 0.2383 0.1736 0.1367 0.1131 0.0968 0.0851 0.0762 0.0693 0.0638 0.0555 0.0498 0.0455 0.0422 0.0396 0.0375 0.0357 0.0343 0.0330 0.0320 0.0283 0.0271 0.0262 0.0248 0.0231 0.0221 0.0197 0.0185
3.40 0.2355 0.1708 0.1340 0.1105 0.0943 0.0827 0.0739 0.0670 0.0616 0.0535 0.0478 0.0436 0.0403 0.0378 0.0357 0.0340 0.0326 0.0314 0.0304 0.0268 0.0256 0.0247 0.0234 0.0217 0.0207 0.0184 0.0173
3.45 0.2328 0.1681 0.1313 0.1080 0.0919 0.0803 0.0716 0.0649 0.0595 0.0515 0.0459 0.0418 0.0386 0.0361 0.0341 0.0324 0.0310 0.0299 0.0288 0.0254 0.0242 0.0233 0.0220 0.0204 0.0195 0.0172 0.0162
3.50 0.2301 0.1655 0.1288 0.1055 0.0896 0.0781 0.0695 0.0628 0.0575 0.0496 0.0441 0.0401 0.0369 0.0345 0.0325 0.0309 0.0295 0.0284 0.0274 0.0240 0.0229 0.0220 0.0208 0.0192 0.0183 0.0161 0.0151
3.55 0.2275 0.1629 0.1263 0.1032 0.0873 0.0760 0.0674 0.0608 0.0556 0.0478 0.0424 0.0384 0.0354 0.0330 0.0310 0.0294 0.0281 0.0270 0.0260 0.0227 0.0217 0.0208 0.0196 0.0181 0.0172 0.0151 0.0141
3.60 0.2250 0.1604 0.1239 0.1009 0.0852 0.0739 0.0654 0.0589 0.0537 0.0461 0.0408 0.0368 0.0339 0.0315 0.0296 0.0281 0.0268 0.0257 0.0247 0.0215 0.0205 0.0197 0.0185 0.0170 0.0161 0.0141 0.0132
3.65 0.2225 0.1579 0.1216 0.0987 0.0831 0.0719 0.0635 0.0571 0.0520 0.0444 0.0392 0.0353 0.0324 0.0301 0.0283 0.0268 0.0255 0.0244 0.0235 0.0204 0.0194 0.0186 0.0174 0.0160 0.0152 0.0132 0.0123
3.70 0.2200 0.1555 0.1193 0.0965 0.0810 0.0700 0.0617 0.0553 0.0503 0.0429 0.0377 0.0339 0.0310 0.0288 0.0270 0.0255 0.0243 0.0232 0.0223 0.0193 0.0183 0.0176 0.0164 0.0150 0.0142 0.0124 0.0115
3.75 0.2176 0.1532 0.1171 0.0944 0.0791 0.0681 0.0599 0.0536 0.0486 0.0413 0.0363 0.0326 0.0297 0.0275 0.0258 0.0243 0.0231 0.0221 0.0212 0.0183 0.0173 0.0166 0.0155 0.0142 0.0134 0.0116 0.0107
3.80 0.2153 0.1509 0.1149 0.0924 0.0772 0.0663 0.0582 0.0520 0.0471 0.0399 0.0349 0.0312 0.0285 0.0263 0.0246 0.0232 0.0220 0.0210 0.0202 0.0173 0.0164 0.0157 0.0146 0.0133 0.0126 0.0108 0.0100
3.85 0.2130 0.1487 0.1129 0.0904 0.0753 0.0645 0.0565 0.0504 0.0456 0.0385 0.0336 0.0300 0.0273 0.0252 0.0235 0.0221 0.0210 0.0200 0.0192 0.0164 0.0155 0.0148 0.0138 0.0125 0.0118 0.0101 0.0094
3.90 0.2108 0.1466 0.1108 0.0885 0.0735 0.0628 0.0549 0.0489 0.0441 0.0371 0.0323 0.0288 0.0262 0.0241 0.0225 0.0211 0.0200 0.0191 0.0183 0.0155 0.0147 0.0140 0.0130 0.0118 0.0111 0.0095 0.0087
3.95 0.2086 0.1444 0.1088 0.0867 0.0718 0.0612 0.0534 0.0474 0.0427 0.0358 0.0311 0.0277 0.0251 0.0231 0.0215 0.0201 0.0191 0.0181 0.0174 0.0147 0.0139 0.0132 0.0123 0.0111 0.0104 0.0089 0.0082
4.00 0.2064 0.1424 0.1069 0.0849 0.0701 0.0596 0.0519 0.0460 0.0413 0.0346 0.0299 0.0266 0.0240 0.0221 0.0205 0.0192 0.0182 0.0173 0.0165 0.0140 0.0131 0.0125 0.0116 0.0104 0.0098 0.0083 0.0076
4.10 0.2023 0.1384 0.1032 0.0814 0.0669 0.0566 0.0491 0.0433 0.0388 0.0323 0.0278 0.0245 0.0221 0.0202 0.0187 0.0175 0.0165 0.0157 0.0150 0.0125 0.0118 0.0112 0.0103 0.0093 0.0087 0.0073 0.0066
4.20 0.1983 0.1346 0.0997 0.0782 0.0639 0.0538 0.0464 0.0408 0.0364 0.0301 0.0258 0.0227 0.0204 0.0186 0.0171 0.0160 0.0150 0.0142 0.0135 0.0113 0.0106 0.0100 0.0092 0.0082 0.0076 0.0064 0.0058
-314-
Back to TOC
4.30 0.1944 0.1310 0.0963 0.0751 0.0610 0.0512 0.0440 0.0385 0.0342 0.0281 0.0240 0.0210 0.0187 0.0170 0.0157 0.0146 0.0137 0.0129 0.0123 0.0101 0.0095 0.0089 0.0082 0.0073 0.0068 0.0056 0.0050
4.40 0.1907 0.1275 0.0932 0.0722 0.0584 0.0487 0.0417 0.0363 0.0322 0.0263 0.0223 0.0194 0.0173 0.0156 0.0143 0.0133 0.0124 0.0117 0.0111 0.0091 0.0085 0.0080 0.0073 0.0064 0.0060 0.0049 0.0044
4.50 0.1872 0.1242 0.0901 0.0695 0.0558 0.0464 0.0395 0.0343 0.0303 0.0246 0.0207 0.0180 0.0159 0.0144 0.0131 0.0122 0.0113 0.0107 0.0101 0.0082 0.0076 0.0072 0.0065 0.0057 0.0053 0.0043 0.0038
4.60 0.1837 0.1210 0.0873 0.0668 0.0535 0.0442 0.0375 0.0324 0.0285 0.0230 0.0193 0.0166 0.0147 0.0132 0.0120 0.0111 0.0103 0.0097 0.0092 0.0074 0.0068 0.0064 0.0058 0.0051 0.0047 0.0037 0.0033
4.70 0.1804 0.1180 0.0845 0.0644 0.0512 0.0421 0.0356 0.0307 0.0269 0.0215 0.0180 0.0154 0.0136 0.0122 0.0111 0.0102 0.0094 0.0088 0.0083 0.0067 0.0061 0.0057 0.0052 0.0045 0.0041 0.0033 0.0029
4.80 0.1772 0.1150 0.0819 0.0620 0.0491 0.0402 0.0338 0.0290 0.0254 0.0202 0.0167 0.0143 0.0125 0.0112 0.0101 0.0093 0.0086 0.0080 0.0076 0.0060 0.0055 0.0051 0.0046 0.0040 0.0036 0.0029 0.0025
4.90 0.1741 0.1122 0.0794 0.0598 0.0471 0.0384 0.0321 0.0275 0.0239 0.0189 0.0156 0.0133 0.0116 0.0103 0.0093 0.0085 0.0079 0.0073 0.0069 0.0054 0.0050 0.0046 0.0041 0.0035 0.0032 0.0025 0.0022
5.00 0.1712 0.1096 0.0770 0.0577 0.0452 0.0367 0.0306 0.0261 0.0226 0.0178 0.0146 0.0124 0.0107 0.0095 0.0086 0.0078 0.0072 0.0067 0.0063 0.0049 0.0045 0.0041 0.0037 0.0031 0.0028 0.0022 0.0019
5.10 0.1683 0.1070 0.0747 0.0557 0.0434 0.0351 0.0291 0.0247 0.0214 0.0167 0.0136 0.0115 0.0099 0.0088 0.0079 0.0072 0.0066 0.0061 0.0057 0.0044 0.0040 0.0037 0.0033 0.0028 0.0025 0.0019 0.0017
5.20 0.1655 0.1045 0.0726 0.0537 0.0417 0.0335 0.0277 0.0234 0.0202 0.0157 0.0127 0.0107 0.0092 0.0081 0.0072 0.0066 0.0060 0.0056 0.0052 0.0040 0.0036 0.0033 0.0029 0.0025 0.0022 0.0017 0.0014
5.30 0.1628 0.1021 0.0705 0.0519 0.0401 0.0321 0.0264 0.0223 0.0191 0.0147 0.0119 0.0099 0.0085 0.0075 0.0067 0.0060 0.0055 0.0051 0.0047 0.0036 0.0032 0.0030 0.0026 0.0022 0.0020 0.0015 0.0013
5.40 0.1602 0.0998 0.0685 0.0502 0.0385 0.0307 0.0252 0.0211 0.0181 0.0139 0.0111 0.0093 0.0079 0.0069 0.0061 0.0055 0.0050 0.0046 0.0043 0.0032 0.0029 0.0027 0.0023 0.0020 0.0017 0.0013 0.0011
5.50 0.1577 0.0976 0.0666 0.0485 0.0371 0.0294 0.0240 0.0201 0.0171 0.0131 0.0104 0.0086 0.0073 0.0064 0.0057 0.0051 0.0046 0.0042 0.0039 0.0029 0.0026 0.0024 0.0021 0.0017 0.0015 0.0011 0.0010
5.60 0.1553 0.0954 0.0647 0.0469 0.0357 0.0282 0.0229 0.0191 0.0162 0.0123 0.0098 0.0080 0.0068 0.0059 0.0052 0.0047 0.0042 0.0039 0.0036 0.0027 0.0024 0.0022 0.0019 0.0015 0.0014 0.0010 0.0008
5.70 0.1529 0.0934 0.0630 0.0454 0.0344 0.0270 0.0219 0.0182 0.0154 0.0116 0.0092 0.0075 0.0063 0.0055 0.0048 0.0043 0.0039 0.0035 0.0033 0.0024 0.0021 0.0019 0.0017 0.0014 0.0012 0.0009 0.0007
5.80 0.1506 0.0914 0.0613 0.0439 0.0331 0.0259 0.0209 0.0173 0.0146 0.0109 0.0086 0.0070 0.0059 0.0051 0.0044 0.0040 0.0036 0.0032 0.0030 0.0022 0.0019 0.0018 0.0015 0.0012 0.0011 0.0008 0.0006
5.90 0.1484 0.0895 0.0596 0.0426 0.0319 0.0249 0.0200 0.0165 0.0139 0.0103 0.0081 0.0065 0.0055 0.0047 0.0041 0.0036 0.0033 0.0030 0.0027 0.0020 0.0017 0.0016 0.0013 0.0011 0.0009 0.0007 0.0005
6.00 0.1462 0.0876 0.0581 0.0412 0.0308 0.0239 0.0191 0.0157 0.0132 0.0097 0.0076 0.0061 0.0051 0.0044 0.0038 0.0034 0.0030 0.0027 0.0025 0.0018 0.0016 0.0014 0.0012 0.0010 0.0008 0.0006 0.0005
6.10 0.1441 0.0858 0.0566 0.0400 0.0297 0.0230 0.0183 0.0150 0.0125 0.0092 0.0071 0.0057 0.0047 0.0040 0.0035 0.0031 0.0028 0.0025 0.0023 0.0016 0.0014 0.0013 0.0011 0.0009 0.0007 0.0005 0.0004
6.20 0.1420 0.0841 0.0551 0.0388 0.0287 0.0221 0.0175 0.0143 0.0119 0.0087 0.0067 0.0054 0.0044 0.0037 0.0032 0.0029 0.0025 0.0023 0.0021 0.0015 0.0013 0.0012 0.0010 0.0008 0.0007 0.0004 0.0004
6.30 0.1400 0.0824 0.0538 0.0376 0.0277 0.0212 0.0168 0.0136 0.0113 0.0082 0.0063 0.0050 0.0041 0.0035 0.0030 0.0026 0.0023 0.0021 0.0019 0.0013 0.0012 0.0010 0.0009 0.0007 0.0006 0.0004 0.0003
6.40 0.1381 0.0808 0.0524 0.0365 0.0267 0.0204 0.0161 0.0130 0.0108 0.0078 0.0059 0.0047 0.0038 0.0032 0.0028 0.0024 0.0022 0.0019 0.0018 0.0012 0.0011 0.0009 0.0008 0.0006 0.0005 0.0003 0.0003
6.50 0.1362 0.0793 0.0511 0.0354 0.0258 0.0197 0.0154 0.0124 0.0103 0.0074 0.0056 0.0044 0.0036 0.0030 0.0026 0.0022 0.0020 0.0018 0.0016 0.0011 0.0010 0.0008 0.0007 0.0005 0.0005 0.0003 0.0002
6.60 0.1344 0.0777 0.0499 0.0344 0.0250 0.0189 0.0148 0.0119 0.0098 0.0070 0.0052 0.0041 0.0034 0.0028 0.0024 0.0021 0.0018 0.0016 0.0015 0.0010 0.0009 0.0008 0.0006 0.0005 0.0004 0.0003 0.0002
6.70 0.1326 0.0763 0.0487 0.0334 0.0242 0.0182 0.0142 0.0114 0.0093 0.0066 0.0049 0.0039 0.0031 0.0026 0.0022 0.0019 0.0017 0.0015 0.0014 0.0009 0.0008 0.0007 0.0006 0.0004 0.0004 0.0002 0.0002
6.80 0.1309 0.0749 0.0475 0.0325 0.0234 0.0176 0.0136 0.0109 0.0089 0.0063 0.0047 0.0036 0.0029 0.0024 0.0021 0.0018 0.0016 0.0014 0.0012 0.0008 0.0007 0.0006 0.0005 0.0004 0.0003 0.0002 0.0002
6.90 0.1292 0.0735 0.0464 0.0316 0.0226 0.0169 0.0131 0.0104 0.0085 0.0059 0.0044 0.0034 0.0027 0.0023 0.0019 0.0016 0.0014 0.0013 0.0011 0.0007 0.0006 0.0006 0.0005 0.0003 0.0003 0.0002 0.0001
7.00 0.1276 0.0721 0.0454 0.0307 0.0219 0.0163 0.0126 0.0100 0.0081 0.0056 0.0041 0.0032 0.0026 0.0021 0.0018 0.0015 0.0013 0.0012 0.0010 0.0007 0.0006 0.0005 0.0004 0.0003 0.0003 0.0002 0.0001
7.20 0.1244 0.0696 0.0433 0.0290 0.0206 0.0152 0.0116 0.0091 0.0074 0.0051 0.0037 0.0028 0.0022 0.0018 0.0015 0.0013 0.0011 0.0010 0.0009 0.0006 0.0005 0.0004 0.0003 0.0002 0.0002 0.0001 0.0001
7.40 0.1214 0.0672 0.0414 0.0275 0.0193 0.0142 0.0108 0.0084 0.0067 0.0046 0.0033 0.0025 0.0020 0.0016 0.0013 0.0011 0.0010 0.0008 0.0007 0.0005 0.0004 0.0003 0.0003 0.0002 0.0002 0.0001 0.0001
7.60 0.1185 0.0649 0.0396 0.0261 0.0182 0.0132 0.0100 0.0077 0.0062 0.0041 0.0030 0.0022 0.0017 0.0014 0.0011 0.0010 0.0008 0.0007 0.0006 0.0004 0.0003 0.0003 0.0002 0.0002 0.0001 0.0001 0.0000
7.80 0.1158 0.0628 0.0380 0.0248 0.0171 0.0124 0.0092 0.0071 0.0056 0.0037 0.0027 0.0020 0.0015 0.0012 0.0010 0.0008 0.0007 0.0006 0.0005 0.0003 0.0003 0.0002 0.0002 0.0001 0.0001 0.0001 0.0000
8.00 0.1131 0.0607 0.0364 0.0235 0.0161 0.0116 0.0086 0.0066 0.0052 0.0034 0.0024 0.0018 0.0013 0.0011 0.0009 0.0007 0.0006 0.0005 0.0005 0.0003 0.0002 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000
8.20 0.1106 0.0588 0.0349 0.0224 0.0152 0.0108 0.0080 0.0061 0.0048 0.0031 0.0021 0.0016 0.0012 0.0009 0.0008 0.0006 0.0005 0.0005 0.0004 0.0002 0.0002 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000
8.40 0.1082 0.0570 0.0335 0.0213 0.0144 0.0102 0.0075 0.0056 0.0044 0.0028 0.0019 0.0014 0.0011 0.0008 0.0007 0.0005 0.0005 0.0004 0.0003 0.0002 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000
8.60 0.1059 0.0553 0.0322 0.0203 0.0136 0.0095 0.0070 0.0052 0.0040 0.0026 0.0017 0.0012 0.0009 0.0007 0.0006 0.0005 0.0004 0.0003 0.0003 0.0002 0.0001 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000
8.80 0.1038 0.0536 0.0310 0.0194 0.0129 0.0090 0.0065 0.0048 0.0037 0.0023 0.0016 0.0011 0.0008 0.0006 0.0005 0.0004 0.0003 0.0003 0.0002 0.0001 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000
9.00 0.1016 0.0520 0.0298 0.0185 0.0122 0.0084 0.0061 0.0045 0.0034 0.0021 0.0014 0.0010 0.0007 0.0006 0.0004 0.0004 0.0003 0.0002 0.0002 0.0001 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000
9.20 0.0996 0.0506 0.0287 0.0177 0.0116 0.0080 0.0057 0.0042 0.0032 0.0020 0.0013 0.0009 0.0007 0.0005 0.0004 0.0003 0.0003 0.0002 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000
9.40 0.0977 0.0491 0.0277 0.0169 0.0110 0.0075 0.0053 0.0039 0.0029 0.0018 0.0012 0.0008 0.0006 0.0004 0.0003 0.0003 0.0002 0.0002 0.0002 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
9.60 0.0958 0.0478 0.0267 0.0162 0.0105 0.0071 0.0050 0.0036 0.0027 0.0016 0.0011 0.0007 0.0005 0.0004 0.0003 0.0002 0.0002 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
9.80 0.0940 0.0465 0.0258 0.0155 0.0099 0.0067 0.0047 0.0034 0.0025 0.0015 0.0010 0.0007 0.0005 0.0003 0.0003 0.0002 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
10.00 0.0923 0.0452 0.0249 0.0149 0.0095 0.0063 0.0044 0.0032 0.0024 0.0014 0.0009 0.0006 0.0004 0.0003 0.0002 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
Table 3-D: dfB = 4

F 2 3 4 5 6 7 8 9 10 12 14 16 18 20 22 24 26 28 30 40 45 50 60 80 100 250 1000
1.00 0.5556 0.5211 0.5000 0.4857 0.4752 0.4672 0.4609 0.4558 0.4516 0.4449 0.4400 0.4362 0.4332 0.4307 0.4286 0.4269 0.4254 0.4241 0.4229 0.4189 0.4175 0.4164 0.4147 0.4126 0.4113 0.4082 0.4065
1.10 0.5273 0.4883 0.4643 0.4478 0.4357 0.4265 0.4192 0.4132 0.4083 0.4005 0.3947 0.3903 0.3867 0.3838 0.3813 0.3793 0.3775 0.3760 0.3746 0.3699 0.3682 0.3669 0.3649 0.3624 0.3609 0.3571 0.3552
1.20 0.5017 0.4587 0.4320 0.4136 0.4001 0.3897 0.3815 0.3748 0.3692 0.3605 0.3540 0.3489 0.3449 0.3416 0.3388 0.3365 0.3345 0.3327 0.3312 0.3258 0.3240 0.3225 0.3202 0.3173 0.3156 0.3113 0.3092
1.30 0.4784 0.4318 0.4027 0.3827 0.3679 0.3565 0.3475 0.3402 0.3341 0.3245 0.3174 0.3118 0.3074 0.3037 0.3007 0.2982 0.2960 0.2941 0.2924 0.2865 0.2844 0.2828 0.2803 0.2772 0.2753 0.2706 0.2682
1.40 0.4571 0.4073 0.3762 0.3546 0.3388 0.3266 0.3169 0.3091 0.3025 0.2923 0.2846 0.2786 0.2739 0.2700 0.2668 0.2640 0.2617 0.2596 0.2578 0.2515 0.2493 0.2476 0.2449 0.2415 0.2395 0.2345 0.2319
1.50 0.4375 0.3849 0.3520 0.3292 0.3125 0.2996 0.2894 0.2811 0.2742 0.2634 0.2553 0.2490 0.2440 0.2399 0.2365 0.2336 0.2312 0.2290 0.2272 0.2205 0.2182 0.2164 0.2136 0.2101 0.2079 0.2027 0.2000
1.60 0.4195 0.3644 0.3300 0.3062 0.2887 0.2753 0.2646 0.2559 0.2488 0.2375 0.2291 0.2226 0.2175 0.2132 0.2097 0.2067 0.2042 0.2020 0.2000 0.1931 0.1908 0.1889 0.1860 0.1824 0.1802 0.1748 0.1721
1.70 0.4029 0.3457 0.3099 0.2853 0.2671 0.2533 0.2423 0.2333 0.2259 0.2144 0.2058 0.1991 0.1938 0.1895 0.1859 0.1829 0.1803 0.1780 0.1761 0.1690 0.1667 0.1648 0.1619 0.1582 0.1559 0.1505 0.1478
1.80 0.3875 0.3284 0.2915 0.2662 0.2476 0.2334 0.2221 0.2130 0.2055 0.1937 0.1850 0.1782 0.1729 0.1685 0.1648 0.1618 0.1592 0.1569 0.1549 0.1479 0.1455 0.1436 0.1407 0.1370 0.1348 0.1294 0.1266
1.90 0.3733 0.3125 0.2747 0.2488 0.2298 0.2154 0.2039 0.1947 0.1871 0.1752 0.1664 0.1596 0.1542 0.1499 0.1462 0.1432 0.1405 0.1383 0.1363 0.1293 0.1269 0.1250 0.1222 0.1185 0.1163 0.1110 0.1083
2.00 0.3600 0.2978 0.2593 0.2329 0.2137 0.1990 0.1875 0.1782 0.1705 0.1586 0.1498 0.1431 0.1377 0.1333 0.1297 0.1267 0.1241 0.1219 0.1199 0.1130 0.1107 0.1088 0.1060 0.1024 0.1003 0.0951 0.0925
2.05 0.3537 0.2909 0.2520 0.2254 0.2061 0.1914 0.1799 0.1706 0.1629 0.1510 0.1422 0.1355 0.1301 0.1258 0.1222 0.1192 0.1167 0.1144 0.1125 0.1057 0.1034 0.1015 0.0987 0.0952 0.0931 0.0880 0.0854
2.10 0.3476 0.2842 0.2450 0.2183 0.1989 0.1842 0.1726 0.1633 0.1556 0.1438 0.1350 0.1283 0.1230 0.1187 0.1152 0.1122 0.1096 0.1075 0.1055 0.0988 0.0965 0.0947 0.0919 0.0885 0.0864 0.0814 0.0788
2.15 0.3418 0.2778 0.2384 0.2115 0.1921 0.1773 0.1657 0.1564 0.1488 0.1369 0.1283 0.1216 0.1163 0.1121 0.1085 0.1056 0.1031 0.1009 0.0990 0.0924 0.0901 0.0883 0.0856 0.0822 0.0802 0.0752 0.0727
2.20 0.3361 0.2716 0.2319 0.2050 0.1855 0.1707 0.1592 0.1499 0.1422 0.1305 0.1218 0.1152 0.1100 0.1058 0.1023 0.0994 0.0969 0.0948 0.0929 0.0863 0.0841 0.0824 0.0797 0.0764 0.0744 0.0695 0.0671
2.25 0.3306 0.2656 0.2258 0.1987 0.1792 0.1644 0.1529 0.1436 0.1360 0.1243 0.1158 0.1092 0.1040 0.0999 0.0964 0.0936 0.0911 0.0890 0.0872 0.0807 0.0785 0.0768 0.0742 0.0709 0.0690 0.0642 0.0619
2.30 0.3253 0.2599 0.2198 0.1927 0.1732 0.1584 0.1469 0.1377 0.1301 0.1185 0.1100 0.1035 0.0984 0.0943 0.0909 0.0881 0.0857 0.0836 0.0818 0.0755 0.0733 0.0716 0.0691 0.0659 0.0640 0.0594 0.0571
2.35 0.3201 0.2543 0.2141 0.1870 0.1674 0.1527 0.1412 0.1321 0.1246 0.1130 0.1046 0.0982 0.0931 0.0891 0.0857 0.0829 0.0806 0.0785 0.0768 0.0705 0.0685 0.0668 0.0643 0.0612 0.0593 0.0548 0.0526
2.40 0.3151 0.2490 0.2086 0.1815 0.1619 0.1473 0.1358 0.1267 0.1192 0.1078 0.0995 0.0931 0.0882 0.0842 0.0809 0.0781 0.0758 0.0738 0.0720 0.0659 0.0639 0.0623 0.0598 0.0568 0.0550 0.0506 0.0485
-315-
Back to TOC
F 2 3 4 5 6 7 8 9 10 12 14 16 18 20 22 24 26 28 30 40 45 50 60 80 100 250 1000
2.45 0.3103 0.2438 0.2033 0.1762 0.1567 0.1420 0.1307 0.1216 0.1142 0.1029 0.0946 0.0884 0.0835 0.0795 0.0763 0.0736 0.0713 0.0693 0.0676 0.0617 0.0597 0.0581 0.0557 0.0527 0.0510 0.0467 0.0446
2.50 0.3056 0.2388 0.1983 0.1711 0.1516 0.1370 0.1257 0.1167 0.1094 0.0982 0.0900 0.0839 0.0790 0.0751 0.0720 0.0693 0.0671 0.0651 0.0635 0.0576 0.0557 0.0542 0.0518 0.0490 0.0472 0.0431 0.0411
2.55 0.3010 0.2339 0.1933 0.1662 0.1468 0.1323 0.1210 0.1121 0.1048 0.0937 0.0857 0.0796 0.0748 0.0710 0.0679 0.0653 0.0631 0.0612 0.0596 0.0539 0.0520 0.0505 0.0483 0.0455 0.0438 0.0398 0.0378
2.60 0.2966 0.2293 0.1886 0.1615 0.1421 0.1277 0.1165 0.1077 0.1005 0.0895 0.0816 0.0756 0.0709 0.0672 0.0641 0.0615 0.0594 0.0576 0.0560 0.0504 0.0486 0.0471 0.0449 0.0422 0.0406 0.0367 0.0348
2.65 0.2923 0.2247 0.1841 0.1570 0.1377 0.1233 0.1122 0.1034 0.0963 0.0855 0.0777 0.0718 0.0672 0.0635 0.0605 0.0580 0.0559 0.0541 0.0525 0.0471 0.0453 0.0439 0.0418 0.0392 0.0376 0.0339 0.0321
2.70 0.2881 0.2203 0.1797 0.1526 0.1334 0.1191 0.1081 0.0994 0.0924 0.0817 0.0740 0.0682 0.0637 0.0601 0.0571 0.0547 0.0526 0.0509 0.0494 0.0441 0.0423 0.0410 0.0389 0.0363 0.0348 0.0312 0.0295
2.75 0.2840 0.2161 0.1754 0.1484 0.1293 0.1151 0.1042 0.0956 0.0886 0.0781 0.0705 0.0648 0.0604 0.0568 0.0540 0.0516 0.0496 0.0478 0.0464 0.0412 0.0395 0.0382 0.0362 0.0337 0.0323 0.0288 0.0271
2.80 0.2801 0.2120 0.1713 0.1444 0.1254 0.1113 0.1005 0.0919 0.0850 0.0746 0.0672 0.0616 0.0572 0.0538 0.0510 0.0486 0.0467 0.0450 0.0436 0.0386 0.0369 0.0356 0.0337 0.0313 0.0299 0.0266 0.0249
2.85 0.2762 0.2080 0.1673 0.1405 0.1216 0.1076 0.0969 0.0884 0.0816 0.0714 0.0640 0.0585 0.0543 0.0509 0.0481 0.0459 0.0440 0.0423 0.0409 0.0361 0.0345 0.0332 0.0314 0.0290 0.0277 0.0245 0.0229
2.90 0.2725 0.2042 0.1635 0.1368 0.1180 0.1041 0.0935 0.0851 0.0784 0.0683 0.0610 0.0556 0.0515 0.0482 0.0455 0.0433 0.0414 0.0398 0.0385 0.0337 0.0322 0.0310 0.0292 0.0270 0.0256 0.0226 0.0211
2.95 0.2689 0.2004 0.1598 0.1332 0.1145 0.1007 0.0902 0.0819 0.0753 0.0653 0.0582 0.0529 0.0488 0.0456 0.0430 0.0408 0.0390 0.0375 0.0361 0.0316 0.0301 0.0289 0.0272 0.0250 0.0238 0.0208 0.0194
3.00 0.2653 0.1968 0.1563 0.1297 0.1111 0.0974 0.0870 0.0789 0.0723 0.0625 0.0555 0.0503 0.0464 0.0432 0.0406 0.0385 0.0368 0.0353 0.0340 0.0295 0.0281 0.0270 0.0253 0.0232 0.0220 0.0192 0.0178
3.05 0.2619 0.1932 0.1528 0.1264 0.1079 0.0943 0.0840 0.0760 0.0695 0.0598 0.0530 0.0479 0.0440 0.0409 0.0384 0.0364 0.0347 0.0332 0.0319 0.0277 0.0263 0.0252 0.0235 0.0215 0.0204 0.0177 0.0164
3.10 0.2585 0.1898 0.1494 0.1231 0.1048 0.0913 0.0812 0.0732 0.0668 0.0573 0.0506 0.0456 0.0418 0.0388 0.0363 0.0343 0.0327 0.0313 0.0300 0.0259 0.0245 0.0235 0.0219 0.0200 0.0189 0.0163 0.0150
3.15 0.2552 0.1865 0.1462 0.1200 0.1018 0.0885 0.0784 0.0705 0.0642 0.0549 0.0483 0.0434 0.0397 0.0367 0.0344 0.0324 0.0308 0.0294 0.0282 0.0242 0.0229 0.0219 0.0204 0.0185 0.0175 0.0150 0.0138
3.20 0.2520 0.1833 0.1431 0.1170 0.0989 0.0857 0.0757 0.0680 0.0618 0.0526 0.0461 0.0413 0.0377 0.0348 0.0325 0.0306 0.0291 0.0277 0.0266 0.0227 0.0214 0.0204 0.0190 0.0172 0.0162 0.0138 0.0127
3.25 0.2489 0.1801 0.1400 0.1141 0.0961 0.0831 0.0732 0.0655 0.0594 0.0504 0.0440 0.0394 0.0358 0.0330 0.0308 0.0289 0.0274 0.0261 0.0250 0.0212 0.0200 0.0191 0.0177 0.0160 0.0150 0.0127 0.0116
3.30 0.2458 0.1771 0.1371 0.1113 0.0935 0.0805 0.0708 0.0632 0.0572 0.0483 0.0421 0.0375 0.0340 0.0313 0.0291 0.0273 0.0259 0.0246 0.0235 0.0199 0.0187 0.0178 0.0164 0.0148 0.0139 0.0117 0.0107
3.35 0.2429 0.1741 0.1342 0.1086 0.0909 0.0781 0.0684 0.0610 0.0551 0.0463 0.0402 0.0358 0.0324 0.0297 0.0276 0.0258 0.0244 0.0232 0.0221 0.0186 0.0175 0.0166 0.0153 0.0138 0.0128 0.0108 0.0098
3.40 0.2400 0.1712 0.1315 0.1060 0.0884 0.0757 0.0662 0.0588 0.0530 0.0444 0.0384 0.0341 0.0308 0.0282 0.0261 0.0244 0.0230 0.0218 0.0208 0.0174 0.0163 0.0155 0.0143 0.0128 0.0119 0.0099 0.0090
3.45 0.2371 0.1684 0.1288 0.1034 0.0860 0.0735 0.0641 0.0568 0.0511 0.0426 0.0368 0.0325 0.0293 0.0267 0.0247 0.0231 0.0217 0.0206 0.0196 0.0163 0.0153 0.0145 0.0133 0.0118 0.0110 0.0091 0.0082
3.50 0.2344 0.1657 0.1262 0.1010 0.0837 0.0713 0.0620 0.0548 0.0492 0.0409 0.0352 0.0310 0.0278 0.0254 0.0234 0.0218 0.0205 0.0194 0.0185 0.0153 0.0143 0.0135 0.0124 0.0110 0.0102 0.0084 0.0076
3.55 0.2317 0.1631 0.1237 0.0986 0.0815 0.0692 0.0600 0.0530 0.0474 0.0393 0.0336 0.0296 0.0265 0.0241 0.0222 0.0207 0.0194 0.0183 0.0174 0.0143 0.0134 0.0126 0.0115 0.0102 0.0094 0.0077 0.0069
3.60 0.2290 0.1605 0.1212 0.0963 0.0793 0.0672 0.0581 0.0512 0.0457 0.0377 0.0322 0.0282 0.0252 0.0229 0.0210 0.0195 0.0183 0.0173 0.0164 0.0134 0.0125 0.0118 0.0107 0.0095 0.0087 0.0071 0.0064
3.65 0.2264 0.1579 0.1189 0.0941 0.0772 0.0652 0.0563 0.0494 0.0440 0.0362 0.0308 0.0269 0.0240 0.0217 0.0199 0.0185 0.0173 0.0163 0.0154 0.0126 0.0117 0.0110 0.0100 0.0088 0.0081 0.0066 0.0058
3.70 0.2239 0.1555 0.1165 0.0919 0.0752 0.0633 0.0545 0.0478 0.0425 0.0348 0.0295 0.0257 0.0228 0.0206 0.0189 0.0175 0.0163 0.0154 0.0145 0.0118 0.0109 0.0103 0.0093 0.0082 0.0075 0.0060 0.0054
3.75 0.2215 0.1531 0.1143 0.0898 0.0733 0.0615 0.0528 0.0462 0.0410 0.0334 0.0282 0.0245 0.0218 0.0196 0.0179 0.0166 0.0154 0.0145 0.0137 0.0111 0.0102 0.0096 0.0087 0.0076 0.0069 0.0055 0.0049
3.80 0.2190 0.1508 0.1121 0.0878 0.0714 0.0598 0.0512 0.0446 0.0395 0.0321 0.0270 0.0234 0.0207 0.0186 0.0170 0.0157 0.0146 0.0137 0.0129 0.0104 0.0096 0.0090 0.0081 0.0070 0.0064 0.0051 0.0045
3.85 0.2167 0.1485 0.1100 0.0859 0.0696 0.0581 0.0496 0.0432 0.0381 0.0309 0.0259 0.0224 0.0197 0.0177 0.0161 0.0148 0.0138 0.0129 0.0122 0.0097 0.0090 0.0084 0.0075 0.0065 0.0060 0.0047 0.0041
3.90 0.2144 0.1463 0.1079 0.0840 0.0679 0.0565 0.0481 0.0418 0.0368 0.0297 0.0248 0.0214 0.0188 0.0169 0.0153 0.0141 0.0130 0.0122 0.0115 0.0091 0.0084 0.0078 0.0070 0.0061 0.0055 0.0043 0.0038
3.95 0.2121 0.1441 0.1059 0.0821 0.0662 0.0549 0.0467 0.0404 0.0355 0.0285 0.0238 0.0204 0.0179 0.0160 0.0145 0.0133 0.0123 0.0115 0.0108 0.0086 0.0079 0.0073 0.0065 0.0056 0.0051 0.0040 0.0035
4.00 0.2099 0.1420 0.1040 0.0804 0.0645 0.0534 0.0453 0.0391 0.0343 0.0274 0.0228 0.0195 0.0171 0.0152 0.0138 0.0126 0.0117 0.0109 0.0102 0.0080 0.0074 0.0068 0.0061 0.0052 0.0047 0.0037 0.0032
4.10 0.2056 0.1379 0.1003 0.0769 0.0614 0.0506 0.0426 0.0366 0.0320 0.0254 0.0210 0.0179 0.0156 0.0138 0.0124 0.0113 0.0104 0.0097 0.0091 0.0071 0.0064 0.0060 0.0053 0.0045 0.0041 0.0031 0.0027
4.20 0.2014 0.1340 0.0967 0.0737 0.0585 0.0479 0.0402 0.0344 0.0299 0.0235 0.0193 0.0163 0.0142 0.0125 0.0112 0.0102 0.0094 0.0087 0.0081 0.0062 0.0057 0.0052 0.0046 0.0039 0.0035 0.0026 0.0022
4.30 0.1975 0.1303 0.0934 0.0707 0.0558 0.0454 0.0379 0.0323 0.0280 0.0218 0.0178 0.0150 0.0129 0.0113 0.0101 0.0092 0.0084 0.0078 0.0072 0.0055 0.0050 0.0046 0.0040 0.0033 0.0030 0.0022 0.0019
4.40 0.1937 0.1268 0.0902 0.0678 0.0532 0.0431 0.0358 0.0303 0.0262 0.0203 0.0164 0.0137 0.0118 0.0103 0.0092 0.0083 0.0075 0.0069 0.0064 0.0048 0.0044 0.0040 0.0035 0.0029 0.0026 0.0019 0.0016
4.50 0.1900 0.1234 0.0872 0.0651 0.0508 0.0409 0.0338 0.0285 0.0245 0.0188 0.0151 0.0126 0.0107 0.0094 0.0083 0.0074 0.0068 0.0062 0.0057 0.0043 0.0038 0.0035 0.0030 0.0025 0.0022 0.0016 0.0013
4.60 0.1865 0.1202 0.0843 0.0626 0.0485 0.0388 0.0319 0.0268 0.0229 0.0175 0.0140 0.0116 0.0098 0.0085 0.0075 0.0067 0.0061 0.0056 0.0051 0.0038 0.0034 0.0031 0.0026 0.0021 0.0019 0.0013 0.0011
4.70 0.1831 0.1171 0.0815 0.0602 0.0464 0.0369 0.0302 0.0253 0.0215 0.0163 0.0129 0.0106 0.0090 0.0077 0.0068 0.0061 0.0055 0.0050 0.0046 0.0033 0.0030 0.0027 0.0023 0.0019 0.0016 0.0011 0.0009
4.80 0.1798 0.1141 0.0789 0.0579 0.0443 0.0351 0.0286 0.0238 0.0202 0.0152 0.0120 0.0098 0.0082 0.0070 0.0062 0.0055 0.0049 0.0045 0.0041 0.0029 0.0026 0.0024 0.0020 0.0016 0.0014 0.0010 0.0008
4.90 0.1766 0.1113 0.0764 0.0557 0.0424 0.0335 0.0271 0.0225 0.0190 0.0142 0.0111 0.0090 0.0075 0.0064 0.0056 0.0049 0.0044 0.0040 0.0037 0.0026 0.0023 0.0021 0.0017 0.0014 0.0012 0.0008 0.0006
5.00 0.1736 0.1086 0.0741 0.0537 0.0406 0.0319 0.0257 0.0212 0.0178 0.0132 0.0103 0.0083 0.0069 0.0059 0.0051 0.0045 0.0040 0.0036 0.0033 0.0023 0.0020 0.0018 0.0015 0.0012 0.0010 0.0007 0.0005
5.10 0.1706 0.1059 0.0718 0.0517 0.0390 0.0304 0.0244 0.0200 0.0168 0.0123 0.0095 0.0076 0.0063 0.0053 0.0046 0.0041 0.0036 0.0032 0.0030 0.0020 0.0018 0.0016 0.0013 0.0010 0.0009 0.0006 0.0005
5.20 0.1677 0.1034 0.0697 0.0499 0.0373 0.0290 0.0232 0.0189 0.0158 0.0115 0.0088 0.0071 0.0058 0.0049 0.0042 0.0037 0.0033 0.0029 0.0026 0.0018 0.0016 0.0014 0.0012 0.0009 0.0008 0.0005 0.0004
5.30 0.1650 0.1010 0.0676 0.0481 0.0358 0.0277 0.0220 0.0179 0.0149 0.0108 0.0082 0.0065 0.0053 0.0045 0.0038 0.0033 0.0029 0.0026 0.0024 0.0016 0.0014 0.0012 0.0010 0.0008 0.0007 0.0004 0.0003
5.40 0.1623 0.0987 0.0656 0.0464 0.0344 0.0264 0.0209 0.0169 0.0140 0.0101 0.0076 0.0060 0.0049 0.0041 0.0035 0.0030 0.0027 0.0024 0.0021 0.0014 0.0012 0.0011 0.0009 0.0007 0.0006 0.0003 0.0003
5.50 0.1597 0.0964 0.0637 0.0448 0.0330 0.0253 0.0199 0.0161 0.0132 0.0094 0.0071 0.0056 0.0045 0.0037 0.0032 0.0027 0.0024 0.0021 0.0019 0.0013 0.0011 0.0009 0.0008 0.0006 0.0005 0.0003 0.0002
5.60 0.1572 0.0943 0.0619 0.0433 0.0317 0.0241 0.0189 0.0152 0.0125 0.0088 0.0066 0.0051 0.0041 0.0034 0.0029 0.0025 0.0022 0.0019 0.0017 0.0011 0.0010 0.0008 0.0007 0.0005 0.0004 0.0002 0.0002
5.70 0.1548 0.0922 0.0602 0.0418 0.0305 0.0231 0.0180 0.0144 0.0118 0.0083 0.0062 0.0048 0.0038 0.0031 0.0026 0.0023 0.0020 0.0017 0.0016 0.0010 0.0008 0.0007 0.0006 0.0004 0.0004 0.0002 0.0002
5.80 0.1524 0.0902 0.0585 0.0405 0.0294 0.0221 0.0172 0.0137 0.0111 0.0078 0.0057 0.0044 0.0035 0.0029 0.0024 0.0021 0.0018 0.0016 0.0014 0.0009 0.0007 0.0006 0.0005 0.0004 0.0003 0.0002 0.0001
5.90 0.1501 0.0883 0.0569 0.0391 0.0283 0.0212 0.0164 0.0130 0.0105 0.0073 0.0054 0.0041 0.0032 0.0026 0.0022 0.0019 0.0016 0.0014 0.0013 0.0008 0.0007 0.0006 0.0005 0.0003 0.0003 0.0001 0.0001
6.00 0.1479 0.0864 0.0554 0.0379 0.0272 0.0203 0.0156 0.0123 0.0100 0.0069 0.0050 0.0038 0.0030 0.0024 0.0020 0.0017 0.0015 0.0013 0.0011 0.0007 0.0006 0.0005 0.0004 0.0003 0.0002 0.0001 0.0001
6.10 0.1458 0.0846 0.0539 0.0367 0.0262 0.0195 0.0149 0.0117 0.0094 0.0064 0.0047 0.0035 0.0028 0.0022 0.0018 0.0016 0.0013 0.0012 0.0010 0.0006 0.0005 0.0004 0.0003 0.0002 0.0002 0.0001 0.0001
6.20 0.1437 0.0829 0.0525 0.0355 0.0253 0.0187 0.0142 0.0112 0.0089 0.0061 0.0044 0.0033 0.0026 0.0021 0.0017 0.0014 0.0012 0.0011 0.0009 0.0006 0.0005 0.0004 0.0003 0.0002 0.0002 0.0001 0.0001
6.30 0.1417 0.0812 0.0512 0.0344 0.0243 0.0179 0.0136 0.0106 0.0085 0.0057 0.0041 0.0030 0.0024 0.0019 0.0016 0.0013 0.0011 0.0010 0.0008 0.0005 0.0004 0.0003 0.0003 0.0002 0.0001 0.0001 0.0001
6.40 0.1397 0.0796 0.0498 0.0334 0.0235 0.0172 0.0130 0.0101 0.0080 0.0054 0.0038 0.0028 0.0022 0.0017 0.0014 0.0012 0.0010 0.0009 0.0008 0.0004 0.0004 0.0003 0.0002 0.0002 0.0001 0.0001 0.0000
6.50 0.1378 0.0780 0.0486 0.0324 0.0227 0.0165 0.0124 0.0096 0.0076 0.0051 0.0036 0.0026 0.0020 0.0016 0.0013 0.0011 0.0009 0.0008 0.0007 0.0004 0.0003 0.0003 0.0002 0.0001 0.0001 0.0001 0.0000
6.60 0.1359 0.0765 0.0474 0.0314 0.0219 0.0159 0.0119 0.0092 0.0072 0.0048 0.0033 0.0025 0.0019 0.0015 0.0012 0.0010 0.0008 0.0007 0.0006 0.0004 0.0003 0.0002 0.0002 0.0001 0.0001 0.0000 0.0000
6.70 0.1341 0.0750 0.0462 0.0305 0.0211 0.0153 0.0114 0.0087 0.0069 0.0045 0.0031 0.0023 0.0017 0.0014 0.0011 0.0009 0.0008 0.0006 0.0006 0.0003 0.0003 0.0002 0.0002 0.0001 0.0001 0.0000 0.0000
6.80 0.1323 0.0736 0.0451 0.0296 0.0204 0.0147 0.0109 0.0083 0.0065 0.0042 0.0029 0.0021 0.0016 0.0013 0.0010 0.0008 0.0007 0.0006 0.0005 0.0003 0.0002 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000
6.90 0.1306 0.0722 0.0440 0.0287 0.0197 0.0141 0.0105 0.0080 0.0062 0.0040 0.0028 0.0020 0.0015 0.0012 0.0009 0.0008 0.0006 0.0005 0.0005 0.0003 0.0002 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000
7.00 0.1289 0.0709 0.0430 0.0279 0.0191 0.0136 0.0100 0.0076 0.0059 0.0038 0.0026 0.0019 0.0014 0.0011 0.0009 0.0007 0.0006 0.0005 0.0004 0.0002 0.0002 0.0001 0.0001 0.0001 0.0001 0.0000 0.0000
7.20 0.1257 0.0683 0.0410 0.0263 0.0179 0.0126 0.0092 0.0069 0.0054 0.0034 0.0023 0.0016 0.0012 0.0009 0.0007 0.0006 0.0005 0.0004 0.0003 0.0002 0.0001 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000
7.40 0.1226 0.0659 0.0391 0.0249 0.0167 0.0117 0.0085 0.0064 0.0049 0.0030 0.0020 0.0014 0.0010 0.0008 0.0006 0.0005 0.0004 0.0003 0.0003 0.0001 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000
7.60 0.1196 0.0637 0.0374 0.0236 0.0157 0.0109 0.0078 0.0058 0.0044 0.0027 0.0018 0.0012 0.0009 0.0007 0.0005 0.0004 0.0003 0.0003 0.0002 0.0001 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000
-316-
Back to TOC
F 2 3 4 5 6 7 8 9 10 12 14 16 18 20 22 24 26 28 30 40 45 50 60 80 100 250 1000
7.80 0.1169 0.0615 0.0358 0.0224 0.0148 0.0102 0.0073 0.0053 0.0040 0.0025 0.0016 0.0011 0.0008 0.0006 0.0005 0.0004 0.0003 0.0002 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000
8.00 0.1142 0.0595 0.0343 0.0212 0.0139 0.0095 0.0067 0.0049 0.0037 0.0022 0.0014 0.0010 0.0007 0.0005 0.0004 0.0003 0.0002 0.0002 0.0002 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
8.20 0.1116 0.0576 0.0329 0.0202 0.0131 0.0089 0.0062 0.0045 0.0034 0.0020 0.0013 0.0009 0.0006 0.0004 0.0003 0.0003 0.0002 0.0002 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
8.40 0.1092 0.0558 0.0315 0.0192 0.0123 0.0083 0.0058 0.0042 0.0031 0.0018 0.0011 0.0008 0.0005 0.0004 0.0003 0.0002 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
8.60 0.1069 0.0541 0.0303 0.0183 0.0116 0.0078 0.0054 0.0038 0.0028 0.0016 0.0010 0.0007 0.0005 0.0003 0.0002 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
8.80 0.1046 0.0524 0.0291 0.0174 0.0110 0.0073 0.0050 0.0036 0.0026 0.0015 0.0009 0.0006 0.0004 0.0003 0.0002 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
9.00 0.1025 0.0509 0.0280 0.0166 0.0104 0.0068 0.0047 0.0033 0.0024 0.0013 0.0008 0.0005 0.0004 0.0003 0.0002 0.0001 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
9.20 0.1004 0.0494 0.0270 0.0158 0.0099 0.0064 0.0044 0.0031 0.0022 0.0012 0.0007 0.0005 0.0003 0.0002 0.0002 0.0001 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
9.40 0.0985 0.0480 0.0260 0.0151 0.0093 0.0061 0.0041 0.0028 0.0020 0.0011 0.0007 0.0004 0.0003 0.0002 0.0001 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
9.60 0.0966 0.0466 0.0250 0.0145 0.0089 0.0057 0.0038 0.0026 0.0019 0.0010 0.0006 0.0004 0.0002 0.0002 0.0001 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
9.80 0.0947 0.0454 0.0241 0.0138 0.0084 0.0054 0.0036 0.0024 0.0017 0.0009 0.0005 0.0003 0.0002 0.0001 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
10.00 0.0930 0.0441 0.0233 0.0133 0.0080 0.0051 0.0033 0.0023 0.0016 0.0008 0.0005 0.0003 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
Table 3-E: dfB = 5

F 2 3 4 5 6 7 8 9 10 12 14 16 18 20 22 24 26 28 30 40 45 50 60 80 100 250 1000
1.00 0.5688 0.5351 0.5143 0.5000 0.4894 0.4813 0.4748 0.4695 0.4651 0.4582 0.4530 0.4489 0.4457 0.4430 0.4408 0.4389 0.4373 0.4359 0.4346 0.4302 0.4287 0.4275 0.4256 0.4233 0.4218 0.4183 0.4165
1.10 0.5395 0.5008 0.4765 0.4596 0.4471 0.4374 0.4297 0.4233 0.4180 0.4097 0.4034 0.3984 0.3945 0.3913 0.3885 0.3862 0.3843 0.3825 0.3810 0.3756 0.3738 0.3723 0.3700 0.3671 0.3653 0.3609 0.3587
1.20 0.5129 0.4696 0.4423 0.4232 0.4089 0.3978 0.3890 0.3817 0.3756 0.3660 0.3588 0.3531 0.3485 0.3448 0.3417 0.3390 0.3367 0.3347 0.3330 0.3267 0.3246 0.3228 0.3202 0.3168 0.3147 0.3097 0.3071
1.30 0.4886 0.4414 0.4113 0.3902 0.3745 0.3622 0.3524 0.3444 0.3376 0.3270 0.3189 0.3126 0.3076 0.3034 0.2999 0.2970 0.2944 0.2922 0.2903 0.2833 0.2809 0.2790 0.2761 0.2723 0.2700 0.2644 0.2615
1.40 0.4665 0.4157 0.3833 0.3605 0.3435 0.3302 0.3196 0.3109 0.3036 0.2921 0.2834 0.2767 0.2712 0.2667 0.2629 0.2598 0.2570 0.2546 0.2526 0.2451 0.2425 0.2404 0.2373 0.2332 0.2308 0.2248 0.2217
1.50 0.4462 0.3923 0.3578 0.3336 0.3155 0.3015 0.2902 0.2810 0.2733 0.2611 0.2519 0.2448 0.2390 0.2343 0.2303 0.2270 0.2241 0.2216 0.2194 0.2115 0.2089 0.2067 0.2034 0.1992 0.1966 0.1903 0.1871
1.60 0.4276 0.3709 0.3347 0.3093 0.2904 0.2757 0.2639 0.2543 0.2462 0.2336 0.2240 0.2166 0.2106 0.2057 0.2017 0.1982 0.1952 0.1927 0.1904 0.1823 0.1796 0.1773 0.1739 0.1696 0.1670 0.1606 0.1573
1.70 0.4104 0.3513 0.3137 0.2873 0.2676 0.2524 0.2403 0.2304 0.2221 0.2091 0.1993 0.1917 0.1856 0.1807 0.1765 0.1730 0.1700 0.1674 0.1651 0.1569 0.1541 0.1519 0.1485 0.1442 0.1415 0.1351 0.1318
1.80 0.3945 0.3333 0.2944 0.2673 0.2471 0.2315 0.2191 0.2090 0.2006 0.1873 0.1775 0.1698 0.1637 0.1587 0.1545 0.1510 0.1480 0.1453 0.1431 0.1349 0.1321 0.1299 0.1265 0.1223 0.1197 0.1133 0.1101
1.90 0.3798 0.3168 0.2769 0.2491 0.2285 0.2127 0.2001 0.1898 0.1813 0.1681 0.1581 0.1505 0.1444 0.1394 0.1352 0.1318 0.1288 0.1262 0.1239 0.1159 0.1132 0.1110 0.1077 0.1035 0.1010 0.0948 0.0917
2.00 0.3661 0.3015 0.2608 0.2325 0.2117 0.1957 0.1830 0.1727 0.1642 0.1509 0.1411 0.1335 0.1274 0.1225 0.1184 0.1150 0.1121 0.1095 0.1073 0.0995 0.0969 0.0948 0.0915 0.0875 0.0851 0.0792 0.0762
2.05 0.3596 0.2944 0.2533 0.2248 0.2038 0.1878 0.1751 0.1648 0.1563 0.1431 0.1333 0.1257 0.1197 0.1149 0.1108 0.1075 0.1046 0.1021 0.0999 0.0922 0.0896 0.0875 0.0844 0.0804 0.0780 0.0723 0.0694
2.10 0.3533 0.2875 0.2460 0.2174 0.1964 0.1803 0.1676 0.1573 0.1489 0.1357 0.1260 0.1185 0.1125 0.1077 0.1038 0.1004 0.0976 0.0951 0.0930 0.0854 0.0829 0.0808 0.0778 0.0739 0.0716 0.0660 0.0632
2.15 0.3473 0.2808 0.2391 0.2103 0.1893 0.1732 0.1605 0.1503 0.1418 0.1287 0.1191 0.1117 0.1058 0.1011 0.0971 0.0938 0.0910 0.0886 0.0865 0.0791 0.0766 0.0746 0.0716 0.0679 0.0656 0.0602 0.0575
2.20 0.3414 0.2744 0.2325 0.2036 0.1825 0.1664 0.1538 0.1436 0.1352 0.1222 0.1126 0.1053 0.0995 0.0948 0.0909 0.0877 0.0850 0.0826 0.0805 0.0733 0.0709 0.0689 0.0660 0.0623 0.0601 0.0549 0.0523
2.25 0.3357 0.2682 0.2261 0.1971 0.1760 0.1600 0.1474 0.1372 0.1289 0.1160 0.1065 0.0993 0.0936 0.0890 0.0852 0.0820 0.0793 0.0770 0.0750 0.0679 0.0655 0.0636 0.0608 0.0572 0.0551 0.0500 0.0475
2.30 0.3303 0.2623 0.2199 0.1909 0.1698 0.1538 0.1413 0.1312 0.1229 0.1101 0.1008 0.0936 0.0880 0.0835 0.0798 0.0767 0.0740 0.0717 0.0698 0.0629 0.0606 0.0587 0.0560 0.0525 0.0505 0.0456 0.0431
2.35 0.3249 0.2566 0.2141 0.1850 0.1639 0.1480 0.1355 0.1254 0.1172 0.1046 0.0954 0.0883 0.0828 0.0784 0.0747 0.0717 0.0691 0.0669 0.0650 0.0582 0.0560 0.0542 0.0515 0.0482 0.0462 0.0415 0.0392
2.40 0.3198 0.2510 0.2084 0.1793 0.1583 0.1424 0.1300 0.1200 0.1119 0.0994 0.0903 0.0834 0.0780 0.0736 0.0700 0.0670 0.0645 0.0624 0.0605 0.0539 0.0518 0.0500 0.0474 0.0442 0.0423 0.0378 0.0355
2.45 0.3148 0.2457 0.2030 0.1739 0.1529 0.1371 0.1247 0.1149 0.1068 0.0945 0.0855 0.0787 0.0734 0.0691 0.0656 0.0627 0.0602 0.0581 0.0563 0.0500 0.0479 0.0462 0.0437 0.0406 0.0387 0.0344 0.0322
2.50 0.3100 0.2405 0.1977 0.1687 0.1478 0.1320 0.1198 0.1100 0.1020 0.0898 0.0810 0.0743 0.0691 0.0649 0.0615 0.0587 0.0563 0.0542 0.0524 0.0463 0.0442 0.0426 0.0402 0.0372 0.0354 0.0313 0.0292
2.55 0.3053 0.2356 0.1927 0.1637 0.1428 0.1272 0.1150 0.1053 0.0974 0.0854 0.0767 0.0702 0.0651 0.0610 0.0577 0.0549 0.0526 0.0506 0.0488 0.0429 0.0409 0.0393 0.0370 0.0341 0.0324 0.0284 0.0265
2.60 0.3008 0.2307 0.1878 0.1589 0.1381 0.1225 0.1105 0.1009 0.0931 0.0813 0.0727 0.0663 0.0613 0.0573 0.0541 0.0514 0.0491 0.0472 0.0455 0.0397 0.0378 0.0363 0.0341 0.0313 0.0297 0.0258 0.0240
2.65 0.2963 0.2261 0.1831 0.1543 0.1336 0.1181 0.1062 0.0967 0.0890 0.0774 0.0690 0.0627 0.0578 0.0539 0.0507 0.0481 0.0459 0.0440 0.0424 0.0368 0.0350 0.0335 0.0314 0.0287 0.0271 0.0235 0.0217
2.70 0.2920 0.2216 0.1786 0.1498 0.1293 0.1139 0.1021 0.0927 0.0851 0.0736 0.0654 0.0592 0.0545 0.0507 0.0476 0.0450 0.0429 0.0411 0.0395 0.0341 0.0323 0.0309 0.0289 0.0263 0.0248 0.0213 0.0197
2.75 0.2879 0.2172 0.1743 0.1456 0.1251 0.1099 0.0982 0.0889 0.0814 0.0701 0.0621 0.0560 0.0514 0.0477 0.0447 0.0422 0.0401 0.0383 0.0368 0.0316 0.0299 0.0285 0.0266 0.0241 0.0227 0.0194 0.0178
2.80 0.2838 0.2130 0.1701 0.1415 0.1211 0.1060 0.0945 0.0853 0.0779 0.0668 0.0589 0.0530 0.0484 0.0448 0.0419 0.0395 0.0375 0.0358 0.0343 0.0293 0.0276 0.0263 0.0244 0.0221 0.0208 0.0176 0.0161
2.85 0.2799 0.2089 0.1661 0.1375 0.1173 0.1024 0.0909 0.0819 0.0746 0.0637 0.0559 0.0501 0.0457 0.0422 0.0394 0.0370 0.0351 0.0334 0.0320 0.0271 0.0256 0.0243 0.0225 0.0203 0.0190 0.0160 0.0146
2.90 0.2760 0.2050 0.1622 0.1337 0.1137 0.0988 0.0875 0.0786 0.0714 0.0607 0.0531 0.0474 0.0431 0.0397 0.0370 0.0347 0.0328 0.0312 0.0298 0.0251 0.0236 0.0224 0.0207 0.0186 0.0173 0.0145 0.0132
2.95 0.2723 0.2011 0.1584 0.1301 0.1102 0.0955 0.0843 0.0755 0.0684 0.0579 0.0504 0.0449 0.0407 0.0374 0.0347 0.0325 0.0307 0.0291 0.0278 0.0233 0.0219 0.0207 0.0190 0.0170 0.0159 0.0132 0.0119
3.00 0.2687 0.1974 0.1547 0.1266 0.1068 0.0922 0.0812 0.0725 0.0656 0.0552 0.0479 0.0425 0.0384 0.0352 0.0326 0.0305 0.0287 0.0272 0.0259 0.0216 0.0202 0.0191 0.0175 0.0156 0.0145 0.0119 0.0107
3.05 0.2651 0.1938 0.1512 0.1232 0.1036 0.0892 0.0782 0.0697 0.0628 0.0527 0.0455 0.0403 0.0363 0.0332 0.0306 0.0286 0.0269 0.0254 0.0242 0.0200 0.0187 0.0177 0.0161 0.0143 0.0132 0.0108 0.0097
3.10 0.2617 0.1903 0.1478 0.1199 0.1005 0.0862 0.0754 0.0670 0.0602 0.0503 0.0433 0.0382 0.0343 0.0312 0.0288 0.0268 0.0252 0.0238 0.0226 0.0186 0.0173 0.0163 0.0149 0.0131 0.0121 0.0098 0.0088
3.15 0.2583 0.1869 0.1445 0.1168 0.0975 0.0834 0.0727 0.0644 0.0578 0.0480 0.0412 0.0362 0.0324 0.0294 0.0271 0.0252 0.0236 0.0222 0.0211 0.0172 0.0160 0.0151 0.0137 0.0120 0.0111 0.0089 0.0079
3.20 0.2551 0.1836 0.1413 0.1138 0.0946 0.0806 0.0701 0.0619 0.0554 0.0458 0.0392 0.0343 0.0306 0.0278 0.0255 0.0236 0.0221 0.0208 0.0197 0.0160 0.0148 0.0139 0.0126 0.0110 0.0101 0.0081 0.0071
3.25 0.2519 0.1804 0.1383 0.1109 0.0918 0.0780 0.0676 0.0596 0.0532 0.0438 0.0373 0.0325 0.0290 0.0262 0.0240 0.0222 0.0207 0.0194 0.0184 0.0148 0.0137 0.0128 0.0116 0.0101 0.0092 0.0073 0.0064
3.30 0.2488 0.1773 0.1353 0.1080 0.0892 0.0755 0.0653 0.0573 0.0511 0.0418 0.0355 0.0309 0.0274 0.0247 0.0225 0.0208 0.0194 0.0182 0.0171 0.0138 0.0127 0.0119 0.0107 0.0092 0.0084 0.0066 0.0058
3.35 0.2457 0.1743 0.1324 0.1053 0.0866 0.0731 0.0630 0.0552 0.0490 0.0400 0.0338 0.0293 0.0259 0.0233 0.0212 0.0195 0.0182 0.0170 0.0160 0.0128 0.0117 0.0110 0.0098 0.0085 0.0077 0.0060 0.0052
3.40 0.2428 0.1713 0.1296 0.1027 0.0842 0.0708 0.0608 0.0531 0.0471 0.0382 0.0322 0.0278 0.0245 0.0220 0.0200 0.0183 0.0170 0.0159 0.0150 0.0118 0.0109 0.0101 0.0090 0.0078 0.0070 0.0054 0.0047
3.45 0.2399 0.1685 0.1269 0.1001 0.0818 0.0686 0.0588 0.0512 0.0452 0.0366 0.0306 0.0264 0.0232 0.0207 0.0188 0.0172 0.0160 0.0149 0.0140 0.0110 0.0101 0.0094 0.0083 0.0071 0.0064 0.0049 0.0043
3.50 0.2370 0.1657 0.1243 0.0977 0.0795 0.0665 0.0568 0.0493 0.0435 0.0350 0.0292 0.0251 0.0220 0.0196 0.0177 0.0162 0.0150 0.0139 0.0131 0.0102 0.0093 0.0086 0.0077 0.0065 0.0059 0.0045 0.0038
3.55 0.2343 0.1630 0.1217 0.0953 0.0773 0.0644 0.0549 0.0475 0.0418 0.0335 0.0278 0.0238 0.0208 0.0185 0.0167 0.0152 0.0140 0.0130 0.0122 0.0095 0.0086 0.0080 0.0071 0.0060 0.0054 0.0040 0.0035
3.60 0.2316 0.1604 0.1193 0.0930 0.0752 0.0625 0.0530 0.0458 0.0402 0.0321 0.0265 0.0226 0.0197 0.0175 0.0157 0.0143 0.0132 0.0122 0.0114 0.0088 0.0080 0.0074 0.0065 0.0055 0.0049 0.0037 0.0031
3.65 0.2289 0.1578 0.1169 0.0908 0.0732 0.0606 0.0513 0.0442 0.0387 0.0307 0.0253 0.0215 0.0187 0.0165 0.0148 0.0135 0.0124 0.0114 0.0107 0.0082 0.0074 0.0068 0.0060 0.0050 0.0045 0.0033 0.0028
3.70 0.2263 0.1553 0.1145 0.0887 0.0712 0.0588 0.0496 0.0426 0.0372 0.0294 0.0242 0.0204 0.0177 0.0156 0.0140 0.0127 0.0116 0.0107 0.0100 0.0076 0.0069 0.0063 0.0055 0.0046 0.0041 0.0030 0.0025
3.75 0.2238 0.1529 0.1123 0.0866 0.0693 0.0570 0.0480 0.0411 0.0358 0.0282 0.0231 0.0194 0.0168 0.0148 0.0132 0.0119 0.0109 0.0101 0.0093 0.0071 0.0064 0.0058 0.0051 0.0042 0.0037 0.0027 0.0023
3.80 0.2214 0.1505 0.1101 0.0846 0.0674 0.0553 0.0464 0.0397 0.0345 0.0270 0.0220 0.0185 0.0159 0.0139 0.0124 0.0112 0.0102 0.0094 0.0087 0.0066 0.0059 0.0054 0.0047 0.0039 0.0034 0.0025 0.0020
3.85 0.2189 0.1482 0.1080 0.0827 0.0657 0.0537 0.0449 0.0383 0.0332 0.0259 0.0210 0.0176 0.0151 0.0132 0.0117 0.0106 0.0096 0.0088 0.0082 0.0061 0.0055 0.0050 0.0043 0.0035 0.0031 0.0022 0.0018
-317-
Back to TOC
F 2 3 4 5 6 7 8 9 10 12 14 16 18 20 22 24 26 28 30 40 45 50 60 80 100 250 1000
3.90 0.2166 0.1459 0.1059 0.0808 0.0640 0.0521 0.0435 0.0370 0.0320 0.0248 0.0201 0.0167 0.0143 0.0125 0.0111 0.0099 0.0090 0.0083 0.0077 0.0057 0.0051 0.0046 0.0040 0.0032 0.0029 0.0020 0.0017
3.95 0.2143 0.1438 0.1039 0.0790 0.0623 0.0506 0.0421 0.0357 0.0308 0.0238 0.0192 0.0159 0.0136 0.0118 0.0104 0.0094 0.0085 0.0078 0.0072 0.0053 0.0047 0.0043 0.0037 0.0030 0.0026 0.0018 0.0015
4.00 0.2120 0.1416 0.1019 0.0772 0.0607 0.0492 0.0408 0.0345 0.0297 0.0228 0.0183 0.0152 0.0129 0.0112 0.0099 0.0088 0.0080 0.0073 0.0067 0.0049 0.0044 0.0040 0.0034 0.0027 0.0024 0.0017 0.0013
4.10 0.2076 0.1375 0.0982 0.0738 0.0577 0.0464 0.0383 0.0322 0.0276 0.0210 0.0167 0.0138 0.0116 0.0100 0.0088 0.0078 0.0071 0.0064 0.0059 0.0042 0.0038 0.0034 0.0029 0.0023 0.0020 0.0014 0.0011
4.20 0.2034 0.1336 0.0946 0.0706 0.0548 0.0439 0.0360 0.0301 0.0257 0.0194 0.0153 0.0125 0.0105 0.0090 0.0079 0.0070 0.0062 0.0057 0.0052 0.0037 0.0032 0.0029 0.0024 0.0019 0.0017 0.0011 0.0009
4.30 0.1994 0.1298 0.0913 0.0677 0.0522 0.0415 0.0339 0.0282 0.0239 0.0179 0.0140 0.0114 0.0095 0.0081 0.0070 0.0062 0.0055 0.0050 0.0046 0.0032 0.0028 0.0025 0.0021 0.0016 0.0014 0.0009 0.0007
4.40 0.1955 0.1262 0.0881 0.0649 0.0497 0.0393 0.0319 0.0264 0.0223 0.0165 0.0129 0.0104 0.0086 0.0073 0.0063 0.0055 0.0049 0.0044 0.0040 0.0028 0.0024 0.0021 0.0018 0.0014 0.0012 0.0007 0.0006
4.50 0.1918 0.1228 0.0851 0.0622 0.0474 0.0372 0.0300 0.0248 0.0208 0.0153 0.0118 0.0094 0.0078 0.0066 0.0056 0.0049 0.0044 0.0039 0.0035 0.0024 0.0021 0.0018 0.0015 0.0012 0.0010 0.0006 0.0005
4.60 0.1882 0.1195 0.0822 0.0597 0.0452 0.0353 0.0283 0.0232 0.0194 0.0142 0.0108 0.0086 0.0070 0.0059 0.0050 0.0044 0.0039 0.0035 0.0031 0.0021 0.0018 0.0016 0.0013 0.0010 0.0008 0.0005 0.0004
4.70 0.1847 0.1164 0.0795 0.0573 0.0431 0.0335 0.0267 0.0218 0.0181 0.0131 0.0100 0.0079 0.0064 0.0053 0.0045 0.0039 0.0034 0.0031 0.0027 0.0018 0.0015 0.0014 0.0011 0.0008 0.0007 0.0004 0.0003
4.80 0.1814 0.1134 0.0769 0.0551 0.0412 0.0318 0.0252 0.0205 0.0170 0.0122 0.0092 0.0072 0.0058 0.0048 0.0041 0.0035 0.0031 0.0027 0.0024 0.0016 0.0013 0.0012 0.0009 0.0007 0.0006 0.0003 0.0002
4.90 0.1781 0.1106 0.0744 0.0530 0.0394 0.0302 0.0239 0.0193 0.0159 0.0113 0.0084 0.0066 0.0053 0.0043 0.0037 0.0031 0.0027 0.0024 0.0021 0.0014 0.0012 0.0010 0.0008 0.0006 0.0005 0.0003 0.0002
5.00 0.1750 0.1078 0.0721 0.0510 0.0376 0.0287 0.0226 0.0181 0.0149 0.0105 0.0078 0.0060 0.0048 0.0039 0.0033 0.0028 0.0024 0.0021 0.0019 0.0012 0.0010 0.0009 0.0007 0.0005 0.0004 0.0002 0.0002
5.10 0.1720 0.1052 0.0698 0.0491 0.0360 0.0274 0.0214 0.0171 0.0139 0.0097 0.0072 0.0055 0.0044 0.0036 0.0030 0.0025 0.0022 0.0019 0.0017 0.0010 0.0009 0.0007 0.0006 0.0004 0.0003 0.0002 0.0001
5.20 0.1691 0.1026 0.0677 0.0473 0.0345 0.0261 0.0202 0.0161 0.0131 0.0091 0.0066 0.0051 0.0040 0.0032 0.0027 0.0023 0.0019 0.0017 0.0015 0.0009 0.0008 0.0006 0.0005 0.0003 0.0003 0.0001 0.0001
5.30 0.1663 0.1002 0.0656 0.0456 0.0330 0.0248 0.0192 0.0152 0.0123 0.0084 0.0061 0.0046 0.0036 0.0029 0.0024 0.0020 0.0017 0.0015 0.0013 0.0008 0.0007 0.0006 0.0004 0.0003 0.0002 0.0001 0.0001
5.40 0.1636 0.0979 0.0637 0.0439 0.0317 0.0237 0.0182 0.0144 0.0115 0.0079 0.0057 0.0043 0.0033 0.0027 0.0022 0.0018 0.0016 0.0013 0.0012 0.0007 0.0006 0.0005 0.0004 0.0002 0.0002 0.0001 0.0001
5.50 0.1610 0.0956 0.0618 0.0424 0.0304 0.0226 0.0173 0.0136 0.0109 0.0073 0.0052 0.0039 0.0030 0.0024 0.0020 0.0016 0.0014 0.0012 0.0010 0.0006 0.0005 0.0004 0.0003 0.0002 0.0002 0.0001 0.0001
5.60 0.1584 0.0935 0.0600 0.0409 0.0292 0.0216 0.0164 0.0128 0.0102 0.0069 0.0049 0.0036 0.0028 0.0022 0.0018 0.0015 0.0012 0.0011 0.0009 0.0005 0.0004 0.0004 0.0003 0.0002 0.0001 0.0001 0.0000
5.70 0.1560 0.0914 0.0583 0.0395 0.0280 0.0206 0.0156 0.0121 0.0096 0.0064 0.0045 0.0033 0.0025 0.0020 0.0016 0.0013 0.0011 0.0010 0.0008 0.0005 0.0004 0.0003 0.0002 0.0002 0.0001 0.0001 0.0000
5.80 0.1536 0.0894 0.0567 0.0382 0.0269 0.0197 0.0148 0.0115 0.0091 0.0060 0.0042 0.0031 0.0023 0.0018 0.0015 0.0012 0.0010 0.0009 0.0007 0.0004 0.0003 0.0003 0.0002 0.0001 0.0001 0.0000 0.0000
5.90 0.1513 0.0874 0.0551 0.0369 0.0259 0.0188 0.0141 0.0109 0.0086 0.0056 0.0039 0.0028 0.0021 0.0017 0.0013 0.0011 0.0009 0.0008 0.0007 0.0004 0.0003 0.0002 0.0002 0.0001 0.0001 0.0000 0.0000
6.00 0.1490 0.0856 0.0536 0.0357 0.0249 0.0180 0.0135 0.0103 0.0081 0.0052 0.0036 0.0026 0.0020 0.0015 0.0012 0.0010 0.0008 0.0007 0.0006 0.0003 0.0002 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000
6.10 0.1468 0.0838 0.0521 0.0345 0.0239 0.0172 0.0128 0.0098 0.0076 0.0049 0.0034 0.0024 0.0018 0.0014 0.0011 0.0009 0.0007 0.0006 0.0005 0.0003 0.0002 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000
6.20 0.1447 0.0820 0.0508 0.0334 0.0230 0.0165 0.0122 0.0093 0.0072 0.0046 0.0031 0.0022 0.0016 0.0013 0.0010 0.0008 0.0007 0.0005 0.0005 0.0002 0.0002 0.0002 0.0001 0.0001 0.0000 0.0000 0.0000
6.30 0.1426 0.0804 0.0494 0.0323 0.0222 0.0158 0.0117 0.0088 0.0068 0.0043 0.0029 0.0021 0.0015 0.0012 0.0009 0.0007 0.0006 0.0005 0.0004 0.0002 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000
6.40 0.1406 0.0787 0.0481 0.0313 0.0214 0.0152 0.0111 0.0084 0.0064 0.0040 0.0027 0.0019 0.0014 0.0011 0.0008 0.0007 0.0005 0.0004 0.0004 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000
6.50 0.1387 0.0772 0.0469 0.0304 0.0206 0.0146 0.0106 0.0080 0.0061 0.0038 0.0025 0.0018 0.0013 0.0010 0.0007 0.0006 0.0005 0.0004 0.0003 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000
6.60 0.1368 0.0756 0.0457 0.0294 0.0199 0.0140 0.0101 0.0076 0.0058 0.0036 0.0024 0.0016 0.0012 0.0009 0.0007 0.0005 0.0004 0.0004 0.0003 0.0001 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000
6.70 0.1350 0.0742 0.0446 0.0285 0.0192 0.0134 0.0097 0.0072 0.0055 0.0034 0.0022 0.0015 0.0011 0.0008 0.0006 0.0005 0.0004 0.0003 0.0003 0.0001 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000
6.80 0.1332 0.0728 0.0435 0.0277 0.0185 0.0129 0.0093 0.0069 0.0052 0.0032 0.0021 0.0014 0.0010 0.0007 0.0006 0.0004 0.0004 0.0003 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000
6.90 0.1314 0.0714 0.0424 0.0269 0.0179 0.0124 0.0089 0.0065 0.0049 0.0030 0.0019 0.0013 0.0009 0.0007 0.0005 0.0004 0.0003 0.0003 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000
7.00 0.1297 0.0700 0.0414 0.0261 0.0173 0.0119 0.0085 0.0062 0.0047 0.0028 0.0018 0.0012 0.0009 0.0006 0.0005 0.0004 0.0003 0.0002 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000
7.20 0.1264 0.0675 0.0395 0.0246 0.0161 0.0110 0.0078 0.0057 0.0042 0.0025 0.0016 0.0011 0.0007 0.0005 0.0004 0.0003 0.0002 0.0002 0.0002 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
7.40 0.1233 0.0651 0.0377 0.0233 0.0151 0.0102 0.0072 0.0052 0.0038 0.0022 0.0014 0.0009 0.0006 0.0005 0.0003 0.0003 0.0002 0.0002 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
7.60 0.1204 0.0628 0.0360 0.0220 0.0142 0.0095 0.0066 0.0047 0.0035 0.0020 0.0012 0.0008 0.0005 0.0004 0.0003 0.0002 0.0002 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
7.80 0.1175 0.0607 0.0344 0.0208 0.0133 0.0088 0.0061 0.0043 0.0031 0.0018 0.0011 0.0007 0.0005 0.0003 0.0002 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
8.00 0.1148 0.0587 0.0329 0.0198 0.0125 0.0082 0.0056 0.0040 0.0029 0.0016 0.0010 0.0006 0.0004 0.0003 0.0002 0.0001 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
8.20 0.1123 0.0568 0.0315 0.0187 0.0117 0.0077 0.0052 0.0036 0.0026 0.0014 0.0008 0.0005 0.0003 0.0002 0.0002 0.0001 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
8.40 0.1098 0.0550 0.0303 0.0178 0.0111 0.0072 0.0048 0.0033 0.0024 0.0013 0.0008 0.0005 0.0003 0.0002 0.0001 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
8.60 0.1074 0.0533 0.0290 0.0169 0.0104 0.0067 0.0045 0.0031 0.0022 0.0012 0.0007 0.0004 0.0003 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
8.80 0.1052 0.0516 0.0279 0.0161 0.0098 0.0063 0.0042 0.0028 0.0020 0.0010 0.0006 0.0004 0.0002 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
9.00 0.1030 0.0501 0.0268 0.0154 0.0093 0.0059 0.0039 0.0026 0.0018 0.0009 0.0005 0.0003 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
9.20 0.1009 0.0486 0.0258 0.0147 0.0088 0.0055 0.0036 0.0024 0.0017 0.0009 0.0005 0.0003 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
9.40 0.0989 0.0472 0.0248 0.0140 0.0083 0.0052 0.0034 0.0022 0.0015 0.0008 0.0004 0.0002 0.0002 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
9.60 0.0970 0.0459 0.0239 0.0134 0.0079 0.0049 0.0031 0.0021 0.0014 0.0007 0.0004 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
9.80 0.0952 0.0446 0.0231 0.0128 0.0075 0.0046 0.0029 0.0019 0.0013 0.0006 0.0003 0.0002 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
10.00 0.0934 0.0434 0.0223 0.0122 0.0071 0.0043 0.0027 0.0018 0.0012 0.0006 0.0003 0.0002 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
Table 3-F: dfB = 10

F 2 3 4 5 6 7 8 9 10 12 14 16 18 20 22 24 26 28 30 40 45 50 60 80 100 250 1000

1.00 0.5981 0.5677 0.5484 0.5349 0.5247 0.5166 0.5101 0.5046 0.5000 0.4926 0.4869 0.4823 0.4786 0.4755 0.4729 0.4706 0.4687 0.4669 0.4654 0.4599 0.4579 0.4564 0.4539 0.4508 0.4488 0.4439 0.4414
1.10 0.5662 0.5292 0.5051 0.4878 0.4745 0.4638 0.4551 0.4478 0.4416 0.4315 0.4236 0.4173 0.4121 0.4078 0.4041 0.4009 0.3981 0.3957 0.3935 0.3856 0.3828 0.3806 0.3770 0.3725 0.3697 0.3625 0.3588
1.20 0.5373 0.4944 0.4661 0.4454 0.4294 0.4165 0.4059 0.3970 0.3894 0.3770 0.3673 0.3595 0.3530 0.3477 0.3431 0.3391 0.3357 0.3327 0.3300 0.3201 0.3167 0.3138 0.3094 0.3038 0.3002 0.2913 0.2867
1.30 0.5111 0.4630 0.4309 0.4073 0.3890 0.3743 0.3621 0.3519 0.3431 0.3288 0.3177 0.3087 0.3013 0.2951 0.2899 0.2853 0.2814 0.2779 0.2749 0.2636 0.2596 0.2564 0.2514 0.2449 0.2409 0.2308 0.2255
1.40 0.4871 0.4345 0.3991 0.3731 0.3529 0.3367 0.3233 0.3120 0.3023 0.2866 0.2744 0.2646 0.2565 0.2498 0.2441 0.2391 0.2348 0.2311 0.2277 0.2156 0.2113 0.2078 0.2025 0.1955 0.1912 0.1805 0.1749
1.50 0.4652 0.4086 0.3705 0.3425 0.3208 0.3033 0.2890 0.2769 0.2666 0.2499 0.2369 0.2265 0.2180 0.2109 0.2049 0.1998 0.1953 0.1914 0.1879 0.1753 0.1709 0.1674 0.1619 0.1548 0.1504 0.1396 0.1340
1.60 0.4451 0.3850 0.3446 0.3150 0.2921 0.2737 0.2586 0.2460 0.2353 0.2179 0.2045 0.1938 0.1851 0.1779 0.1718 0.1666 0.1620 0.1581 0.1546 0.1420 0.1376 0.1340 0.1286 0.1217 0.1174 0.1069 0.1015
1.70 0.4266 0.3635 0.3212 0.2903 0.2665 0.2475 0.2319 0.2189 0.2079 0.1902 0.1766 0.1659 0.1572 0.1499 0.1439 0.1387 0.1342 0.1303 0.1269 0.1146 0.1103 0.1069 0.1017 0.0950 0.0910 0.0811 0.0761
1.80 0.4095 0.3439 0.3000 0.2681 0.2436 0.2241 0.2083 0.1951 0.1840 0.1662 0.1527 0.1420 0.1334 0.1263 0.1204 0.1154 0.1110 0.1073 0.1040 0.0922 0.0882 0.0849 0.0800 0.0738 0.0701 0.0610 0.0565
1.90 0.3937 0.3259 0.2808 0.2481 0.2231 0.2034 0.1874 0.1742 0.1631 0.1455 0.1321 0.1217 0.1133 0.1065 0.1008 0.0959 0.0918 0.0882 0.0851 0.0740 0.0703 0.0673 0.0628 0.0571 0.0537 0.0456 0.0416
2.00 0.3791 0.3094 0.2632 0.2300 0.2048 0.1850 0.1690 0.1558 0.1448 0.1275 0.1145 0.1044 0.0963 0.0898 0.0843 0.0798 0.0759 0.0725 0.0696 0.0594 0.0559 0.0532 0.0491 0.0440 0.0410 0.0339 0.0304
2.05 0.3721 0.3016 0.2550 0.2216 0.1963 0.1765 0.1606 0.1475 0.1366 0.1194 0.1066 0.0967 0.0889 0.0825 0.0772 0.0728 0.0690 0.0658 0.0630 0.0531 0.0499 0.0473 0.0434 0.0386 0.0357 0.0291 0.0259
-318-
Back to TOC
F 2 3 4 5 6 7 8 9 10 12 14 16 18 20 22 24 26 28 30 40 45 50 60 80 100 250 1000
2.10 0.3655 0.2942 0.2472 0.2136 0.1883 0.1685 0.1526 0.1397 0.1289 0.1119 0.0994 0.0897 0.0820 0.0758 0.0707 0.0664 0.0627 0.0596 0.0569 0.0475 0.0444 0.0420 0.0383 0.0338 0.0312 0.0250 0.0221
2.15 0.3590 0.2870 0.2397 0.2060 0.1807 0.1610 0.1452 0.1323 0.1216 0.1050 0.0926 0.0831 0.0757 0.0696 0.0647 0.0606 0.0571 0.0541 0.0515 0.0425 0.0396 0.0373 0.0338 0.0296 0.0271 0.0214 0.0187
2.20 0.3528 0.2801 0.2326 0.1988 0.1735 0.1538 0.1382 0.1254 0.1149 0.0985 0.0864 0.0771 0.0698 0.0640 0.0592 0.0553 0.0519 0.0490 0.0466 0.0380 0.0353 0.0331 0.0298 0.0259 0.0236 0.0184 0.0159
2.25 0.3467 0.2735 0.2257 0.1919 0.1666 0.1471 0.1315 0.1189 0.1085 0.0924 0.0806 0.0716 0.0645 0.0589 0.0543 0.0504 0.0472 0.0445 0.0421 0.0340 0.0314 0.0293 0.0263 0.0226 0.0205 0.0157 0.0135
2.30 0.3409 0.2671 0.2192 0.1853 0.1601 0.1407 0.1253 0.1128 0.1026 0.0867 0.0752 0.0664 0.0596 0.0541 0.0497 0.0460 0.0430 0.0403 0.0381 0.0304 0.0280 0.0260 0.0232 0.0198 0.0178 0.0134 0.0114
2.35 0.3353 0.2610 0.2129 0.1790 0.1539 0.1346 0.1194 0.1071 0.0970 0.0815 0.0702 0.0617 0.0551 0.0498 0.0456 0.0420 0.0391 0.0366 0.0345 0.0272 0.0249 0.0231 0.0204 0.0173 0.0155 0.0114 0.0096
2.40 0.3298 0.2551 0.2069 0.1730 0.1480 0.1289 0.1138 0.1017 0.0918 0.0766 0.0656 0.0573 0.0509 0.0459 0.0418 0.0384 0.0356 0.0332 0.0312 0.0243 0.0222 0.0205 0.0180 0.0151 0.0134 0.0098 0.0081
2.45 0.3245 0.2494 0.2011 0.1673 0.1424 0.1235 0.1086 0.0966 0.0869 0.0720 0.0613 0.0533 0.0471 0.0422 0.0383 0.0351 0.0324 0.0302 0.0282 0.0218 0.0197 0.0182 0.0159 0.0132 0.0116 0.0083 0.0068
2.50 0.3194 0.2439 0.1955 0.1618 0.1371 0.1183 0.1036 0.0918 0.0823 0.0677 0.0573 0.0495 0.0436 0.0389 0.0351 0.0321 0.0295 0.0274 0.0256 0.0195 0.0176 0.0161 0.0140 0.0115 0.0101 0.0071 0.0058
2.55 0.3145 0.2386 0.1902 0.1566 0.1320 0.1134 0.0989 0.0873 0.0779 0.0637 0.0536 0.0461 0.0403 0.0359 0.0323 0.0293 0.0269 0.0249 0.0231 0.0174 0.0156 0.0143 0.0123 0.0100 0.0087 0.0060 0.0048
2.60 0.3096 0.2335 0.1851 0.1516 0.1272 0.1088 0.0945 0.0831 0.0739 0.0600 0.0501 0.0429 0.0374 0.0331 0.0296 0.0268 0.0245 0.0226 0.0210 0.0156 0.0139 0.0127 0.0108 0.0087 0.0076 0.0051 0.0041
2.65 0.3050 0.2286 0.1802 0.1468 0.1226 0.1044 0.0903 0.0791 0.0701 0.0565 0.0469 0.0399 0.0346 0.0305 0.0272 0.0245 0.0224 0.0205 0.0190 0.0140 0.0124 0.0112 0.0095 0.0076 0.0066 0.0043 0.0034
2.70 0.3004 0.2238 0.1754 0.1422 0.1182 0.1002 0.0863 0.0753 0.0665 0.0532 0.0440 0.0372 0.0321 0.0281 0.0250 0.0225 0.0204 0.0187 0.0172 0.0125 0.0111 0.0100 0.0084 0.0066 0.0057 0.0037 0.0029
2.75 0.2960 0.2192 0.1709 0.1379 0.1140 0.0962 0.0825 0.0717 0.0631 0.0502 0.0412 0.0347 0.0298 0.0260 0.0230 0.0206 0.0186 0.0170 0.0156 0.0112 0.0098 0.0088 0.0074 0.0058 0.0049 0.0031 0.0024
2.80 0.2918 0.2148 0.1665 0.1336 0.1100 0.0925 0.0790 0.0684 0.0599 0.0473 0.0386 0.0323 0.0276 0.0240 0.0211 0.0188 0.0170 0.0154 0.0142 0.0100 0.0088 0.0078 0.0065 0.0050 0.0043 0.0026 0.0020
2.85 0.2876 0.2105 0.1623 0.1296 0.1062 0.0889 0.0756 0.0652 0.0569 0.0447 0.0362 0.0301 0.0256 0.0221 0.0194 0.0173 0.0155 0.0141 0.0128 0.0090 0.0078 0.0069 0.0057 0.0044 0.0037 0.0022 0.0017
2.90 0.2836 0.2063 0.1583 0.1258 0.1026 0.0854 0.0724 0.0622 0.0541 0.0422 0.0340 0.0281 0.0238 0.0205 0.0179 0.0158 0.0142 0.0128 0.0117 0.0080 0.0070 0.0062 0.0051 0.0038 0.0032 0.0019 0.0014
2.95 0.2796 0.2023 0.1543 0.1220 0.0991 0.0822 0.0693 0.0593 0.0514 0.0398 0.0319 0.0263 0.0221 0.0189 0.0165 0.0145 0.0129 0.0116 0.0106 0.0072 0.0062 0.0055 0.0045 0.0033 0.0027 0.0016 0.0012
3.00 0.2758 0.1984 0.1506 0.1185 0.0958 0.0791 0.0665 0.0567 0.0489 0.0376 0.0300 0.0245 0.0205 0.0175 0.0152 0.0133 0.0118 0.0106 0.0096 0.0065 0.0055 0.0049 0.0039 0.0029 0.0024 0.0014 0.0010
3.05 0.2721 0.1946 0.1469 0.1151 0.0926 0.0761 0.0637 0.0541 0.0466 0.0356 0.0282 0.0229 0.0191 0.0162 0.0140 0.0122 0.0108 0.0097 0.0087 0.0058 0.0049 0.0043 0.0035 0.0025 0.0021 0.0011 0.0008
3.10 0.2685 0.1909 0.1434 0.1118 0.0895 0.0733 0.0611 0.0517 0.0443 0.0337 0.0265 0.0214 0.0178 0.0150 0.0129 0.0112 0.0099 0.0088 0.0079 0.0052 0.0044 0.0038 0.0030 0.0022 0.0018 0.0010 0.0007
3.15 0.2649 0.1874 0.1400 0.1086 0.0866 0.0706 0.0586 0.0494 0.0422 0.0318 0.0249 0.0200 0.0165 0.0139 0.0119 0.0103 0.0091 0.0080 0.0072 0.0047 0.0039 0.0034 0.0027 0.0019 0.0015 0.0008 0.0006
3.20 0.2615 0.1839 0.1368 0.1056 0.0838 0.0681 0.0563 0.0473 0.0402 0.0301 0.0234 0.0188 0.0154 0.0129 0.0110 0.0095 0.0083 0.0073 0.0066 0.0042 0.0035 0.0030 0.0024 0.0017 0.0013 0.0007 0.0005
3.25 0.2581 0.1806 0.1336 0.1027 0.0812 0.0656 0.0540 0.0452 0.0384 0.0285 0.0221 0.0176 0.0143 0.0119 0.0101 0.0087 0.0076 0.0067 0.0060 0.0037 0.0031 0.0027 0.0021 0.0015 0.0011 0.0006 0.0004
3.30 0.2549 0.1774 0.1306 0.0999 0.0786 0.0633 0.0519 0.0433 0.0366 0.0270 0.0208 0.0164 0.0134 0.0111 0.0093 0.0080 0.0070 0.0061 0.0054 0.0034 0.0028 0.0024 0.0018 0.0013 0.0010 0.0005 0.0003
3.35 0.2517 0.1742 0.1276 0.0972 0.0761 0.0610 0.0499 0.0414 0.0349 0.0256 0.0196 0.0154 0.0124 0.0103 0.0086 0.0074 0.0064 0.0056 0.0049 0.0030 0.0025 0.0021 0.0016 0.0011 0.0009 0.0004 0.0003
3.40 0.2486 0.1712 0.1248 0.0945 0.0738 0.0589 0.0480 0.0397 0.0333 0.0243 0.0184 0.0144 0.0116 0.0095 0.0080 0.0068 0.0059 0.0051 0.0045 0.0027 0.0022 0.0019 0.0014 0.0010 0.0007 0.0003 0.0002
3.45 0.2455 0.1682 0.1220 0.0920 0.0715 0.0569 0.0461 0.0380 0.0318 0.0231 0.0174 0.0135 0.0108 0.0089 0.0074 0.0063 0.0054 0.0047 0.0041 0.0024 0.0020 0.0017 0.0013 0.0008 0.0006 0.0003 0.0002
3.50 0.2426 0.1653 0.1193 0.0896 0.0693 0.0549 0.0444 0.0364 0.0304 0.0219 0.0164 0.0127 0.0101 0.0082 0.0068 0.0058 0.0049 0.0043 0.0037 0.0022 0.0018 0.0015 0.0011 0.0007 0.0006 0.0002 0.0001
3.55 0.2397 0.1625 0.1168 0.0873 0.0672 0.0530 0.0427 0.0349 0.0290 0.0208 0.0155 0.0119 0.0094 0.0077 0.0063 0.0053 0.0045 0.0039 0.0034 0.0020 0.0016 0.0013 0.0010 0.0006 0.0005 0.0002 0.0001
3.60 0.2369 0.1598 0.1143 0.0850 0.0652 0.0513 0.0411 0.0335 0.0277 0.0197 0.0146 0.0112 0.0088 0.0071 0.0059 0.0049 0.0042 0.0036 0.0031 0.0018 0.0014 0.0012 0.0009 0.0006 0.0004 0.0002 0.0001
3.65 0.2341 0.1571 0.1118 0.0829 0.0633 0.0495 0.0396 0.0322 0.0265 0.0187 0.0138 0.0105 0.0082 0.0066 0.0054 0.0045 0.0038 0.0033 0.0028 0.0016 0.0013 0.0011 0.0008 0.0005 0.0004 0.0001 0.0001
3.70 0.2314 0.1546 0.1095 0.0808 0.0614 0.0479 0.0381 0.0309 0.0254 0.0178 0.0130 0.0099 0.0077 0.0062 0.0050 0.0042 0.0035 0.0030 0.0026 0.0014 0.0011 0.0009 0.0007 0.0004 0.0003 0.0001 0.0001
3.75 0.2288 0.1520 0.1072 0.0788 0.0597 0.0463 0.0367 0.0296 0.0243 0.0169 0.0123 0.0093 0.0072 0.0057 0.0047 0.0038 0.0032 0.0028 0.0024 0.0013 0.0010 0.0008 0.0006 0.0004 0.0003 0.0001 0.0001
3.80 0.2262 0.1496 0.1050 0.0768 0.0579 0.0448 0.0354 0.0285 0.0232 0.0161 0.0117 0.0087 0.0068 0.0053 0.0043 0.0036 0.0030 0.0025 0.0022 0.0012 0.0009 0.0007 0.0005 0.0003 0.0002 0.0001 0.0000
3.85 0.2237 0.1472 0.1029 0.0749 0.0563 0.0434 0.0341 0.0274 0.0223 0.0153 0.0110 0.0082 0.0063 0.0050 0.0040 0.0033 0.0027 0.0023 0.0020 0.0011 0.0008 0.0007 0.0005 0.0003 0.0002 0.0001 0.0000
3.90 0.2212 0.1449 0.1008 0.0731 0.0547 0.0420 0.0329 0.0263 0.0213 0.0146 0.0104 0.0077 0.0059 0.0046 0.0037 0.0030 0.0025 0.0021 0.0018 0.0009 0.0007 0.0006 0.0004 0.0002 0.0002 0.0001 0.0000
3.95 0.2188 0.1426 0.0988 0.0713 0.0532 0.0407 0.0318 0.0253 0.0204 0.0139 0.0099 0.0073 0.0055 0.0043 0.0035 0.0028 0.0023 0.0019 0.0017 0.0009 0.0007 0.0005 0.0004 0.0002 0.0001 0.0001 0.0000
4.00 0.2165 0.1404 0.0968 0.0696 0.0517 0.0394 0.0307 0.0243 0.0196 0.0132 0.0094 0.0069 0.0052 0.0040 0.0032 0.0026 0.0021 0.0018 0.0015 0.0008 0.0006 0.0005 0.0003 0.0002 0.0001 0.0000 0.0000
4.10 0.2119 0.1362 0.0931 0.0664 0.0489 0.0370 0.0286 0.0225 0.0180 0.0120 0.0084 0.0061 0.0046 0.0035 0.0028 0.0022 0.0018 0.0015 0.0013 0.0006 0.0005 0.0004 0.0003 0.0001 0.0001 0.0000 0.0000
4.20 0.2075 0.1322 0.0896 0.0634 0.0463 0.0348 0.0267 0.0209 0.0166 0.0109 0.0076 0.0054 0.0040 0.0031 0.0024 0.0019 0.0016 0.0013 0.0011 0.0005 0.0004 0.0003 0.0002 0.0001 0.0001 0.0000 0.0000
4.30 0.2033 0.1284 0.0862 0.0605 0.0439 0.0327 0.0249 0.0194 0.0153 0.0100 0.0068 0.0048 0.0036 0.0027 0.0021 0.0016 0.0013 0.0011 0.0009 0.0004 0.0003 0.0002 0.0002 0.0001 0.0001 0.0000 0.0000
4.40 0.1993 0.1247 0.0831 0.0578 0.0416 0.0308 0.0233 0.0180 0.0141 0.0091 0.0061 0.0043 0.0031 0.0024 0.0018 0.0014 0.0011 0.0009 0.0008 0.0003 0.0003 0.0002 0.0001 0.0001 0.0000 0.0000 0.0000
4.50 0.1954 0.1212 0.0801 0.0553 0.0395 0.0290 0.0218 0.0167 0.0130 0.0083 0.0055 0.0039 0.0028 0.0021 0.0016 0.0012 0.0010 0.0008 0.0006 0.0003 0.0002 0.0002 0.0001 0.0000 0.0000 0.0000 0.0000
4.60 0.1917 0.1179 0.0773 0.0530 0.0376 0.0274 0.0205 0.0156 0.0121 0.0076 0.0050 0.0035 0.0025 0.0018 0.0014 0.0011 0.0008 0.0007 0.0005 0.0002 0.0002 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000
4.70 0.1881 0.1147 0.0746 0.0507 0.0357 0.0259 0.0192 0.0145 0.0112 0.0069 0.0045 0.0031 0.0022 0.0016 0.0012 0.0009 0.0007 0.0006 0.0005 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000
4.80 0.1846 0.1117 0.0721 0.0487 0.0340 0.0245 0.0180 0.0135 0.0104 0.0064 0.0041 0.0028 0.0019 0.0014 0.0010 0.0008 0.0006 0.0005 0.0004 0.0002 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000
4.90 0.1813 0.1088 0.0696 0.0467 0.0324 0.0231 0.0169 0.0127 0.0096 0.0058 0.0037 0.0025 0.0017 0.0012 0.0009 0.0007 0.0005 0.0004 0.0003 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000
5.00 0.1781 0.1060 0.0673 0.0448 0.0309 0.0219 0.0159 0.0118 0.0090 0.0054 0.0034 0.0022 0.0015 0.0011 0.0008 0.0006 0.0005 0.0004 0.0003 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000
5.10 0.1750 0.1033 0.0652 0.0430 0.0295 0.0208 0.0150 0.0111 0.0083 0.0049 0.0031 0.0020 0.0014 0.0010 0.0007 0.0005 0.0004 0.0003 0.0002 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
5.20 0.1720 0.1008 0.0631 0.0414 0.0281 0.0197 0.0141 0.0104 0.0078 0.0045 0.0028 0.0018 0.0012 0.0009 0.0006 0.0005 0.0003 0.0003 0.0002 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
5.30 0.1691 0.0983 0.0611 0.0398 0.0269 0.0187 0.0133 0.0097 0.0072 0.0042 0.0026 0.0017 0.0011 0.0008 0.0005 0.0004 0.0003 0.0002 0.0002 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
5.40 0.1663 0.0959 0.0592 0.0383 0.0257 0.0178 0.0126 0.0091 0.0067 0.0039 0.0023 0.0015 0.0010 0.0007 0.0005 0.0003 0.0003 0.0002 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
5.50 0.1636 0.0937 0.0574 0.0369 0.0246 0.0169 0.0119 0.0086 0.0063 0.0036 0.0021 0.0014 0.0009 0.0006 0.0004 0.0003 0.0002 0.0002 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
5.60 0.1609 0.0915 0.0557 0.0355 0.0235 0.0161 0.0112 0.0081 0.0059 0.0033 0.0020 0.0012 0.0008 0.0005 0.0004 0.0003 0.0002 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
5.70 0.1584 0.0894 0.0540 0.0342 0.0225 0.0153 0.0106 0.0076 0.0055 0.0031 0.0018 0.0011 0.0007 0.0005 0.0003 0.0002 0.0002 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
5.80 0.1559 0.0874 0.0524 0.0330 0.0216 0.0146 0.0101 0.0071 0.0052 0.0028 0.0017 0.0010 0.0006 0.0004 0.0003 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
5.90 0.1535 0.0855 0.0509 0.0319 0.0207 0.0139 0.0095 0.0067 0.0048 0.0026 0.0015 0.0009 0.0006 0.0004 0.0003 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
6.00 0.1512 0.0836 0.0495 0.0308 0.0199 0.0132 0.0091 0.0063 0.0045 0.0024 0.0014 0.0008 0.0005 0.0003 0.0002 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
6.10 0.1490 0.0818 0.0481 0.0297 0.0191 0.0126 0.0086 0.0060 0.0043 0.0023 0.0013 0.0008 0.0005 0.0003 0.0002 0.0001 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
6.20 0.1468 0.0800 0.0468 0.0287 0.0183 0.0121 0.0082 0.0056 0.0040 0.0021 0.0012 0.0007 0.0004 0.0003 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
6.30 0.1447 0.0784 0.0455 0.0278 0.0176 0.0115 0.0078 0.0053 0.0038 0.0020 0.0011 0.0006 0.0004 0.0002 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
6.40 0.1426 0.0767 0.0443 0.0269 0.0169 0.0110 0.0074 0.0050 0.0035 0.0018 0.0010 0.0006 0.0004 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
6.50 0.1406 0.0752 0.0431 0.0260 0.0163 0.0105 0.0070 0.0048 0.0033 0.0017 0.0009 0.0005 0.0003 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
6.60 0.1387 0.0737 0.0420 0.0252 0.0157 0.0101 0.0067 0.0045 0.0031 0.0016 0.0009 0.0005 0.0003 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
-319-
Back to TOC
F 2 3 4 5 6 7 8 9 10 12 14 16 18 20 22 24 26 28 30 40 45 50 60 80 100 250 1000
6.70 0.1368 0.0722 0.0409 0.0244 0.0151 0.0097 0.0064 0.0043 0.0030 0.0015 0.0008 0.0004 0.0003 0.0002 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
6.80 0.1349 0.0708 0.0399 0.0236 0.0145 0.0093 0.0061 0.0041 0.0028 0.0014 0.0007 0.0004 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
6.90 0.1331 0.0694 0.0388 0.0229 0.0140 0.0089 0.0058 0.0039 0.0026 0.0013 0.0007 0.0004 0.0002 0.0001 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
7.00 0.1314 0.0681 0.0379 0.0222 0.0135 0.0085 0.0055 0.0037 0.0025 0.0012 0.0006 0.0003 0.0002 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
7.20 0.1280 0.0655 0.0361 0.0209 0.0126 0.0078 0.0050 0.0033 0.0022 0.0011 0.0005 0.0003 0.0002 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
7.40 0.1248 0.0632 0.0344 0.0197 0.0117 0.0072 0.0046 0.0030 0.0020 0.0009 0.0005 0.0002 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
7.60 0.1218 0.0609 0.0328 0.0186 0.0110 0.0067 0.0042 0.0027 0.0018 0.0008 0.0004 0.0002 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
7.80 0.1189 0.0588 0.0313 0.0175 0.0102 0.0062 0.0039 0.0025 0.0016 0.0007 0.0004 0.0002 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
8.00 0.1161 0.0568 0.0299 0.0166 0.0096 0.0057 0.0035 0.0022 0.0014 0.0006 0.0003 0.0002 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
8.20 0.1135 0.0550 0.0286 0.0157 0.0090 0.0053 0.0033 0.0020 0.0013 0.0006 0.0003 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
8.40 0.1110 0.0532 0.0274 0.0149 0.0085 0.0050 0.0030 0.0019 0.0012 0.0005 0.0002 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
8.60 0.1086 0.0515 0.0263 0.0142 0.0080 0.0046 0.0028 0.0017 0.0011 0.0005 0.0002 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
8.80 0.1063 0.0499 0.0252 0.0135 0.0075 0.0043 0.0026 0.0016 0.0010 0.0004 0.0002 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
9.00 0.1041 0.0484 0.0242 0.0128 0.0071 0.0040 0.0024 0.0014 0.0009 0.0004 0.0002 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
9.20 0.1020 0.0469 0.0233 0.0122 0.0067 0.0038 0.0022 0.0013 0.0008 0.0003 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
9.40 0.0999 0.0456 0.0224 0.0116 0.0063 0.0035 0.0020 0.0012 0.0007 0.0003 0.0001 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
9.60 0.0980 0.0442 0.0216 0.0111 0.0060 0.0033 0.0019 0.0011 0.0007 0.0003 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
9.80 0.0961 0.0430 0.0208 0.0106 0.0056 0.0031 0.0018 0.0010 0.0006 0.0002 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
10.00 0.0943 0.0418 0.0200 0.0101 0.0053 0.0029 0.0017 0.0010 0.0006 0.0002 0.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
-320-
Back to TOC
Table 4: Studentized Range Statistic (qk)
For a One-Way ANOVA, or a comparison of the means from a main effect, the value of k is the number
of means in the factor. To compare the means from an interaction, find the appropriate design
(number of cell means) in the table below and obtain the adjusted value of k. Then use adjusted k to
find the value of qk
Values of Adjusted k
Design of Study - or - Number of Cell Means Adjusted Value of k

2x2 4 3
2x3 6 5
2x4 8 6
3x3 9 7
3x4 12 8
4x4 16 10
4x5 20 12
k (Number of Groups Being Compared)

df
Within α 2 3 4 5 6 7 8 9 10 11 12
Groups
1 0.05 18 27 32.8 37.1 40.4 43.1 45.4 47.4 49.1 50.6 52
0.01 90 135 164 186 202 216 227 237 246 253 260
2 0.05 6.09 8.3 9.8 10.9 11.7 12.4 13 13.5 14 14.4 14.7
0.01 14 19 22.3 24.7 26.6 28.2 29.5 30.7 31.7 32.6 33.4
3 0.05 4.5 5.91 6.82 7.5 8.04 8.48 8.85 9.18 9.46 9.72 9.95
0.01 8.26 10.6 12.2 13.3 14.2 15 15.6 16.2 16.7 17.1 17.5
4 0.05 3.93 5.04 5.76 6.29 6.71 7.05 7.35 7.6 7.83 8.03 8.21
0.01 6.51 8.12 9.17 9.96 10.6 11.1 11.5 11.9 12.3 12.6 21.8
5 0.05 3.64 4.6 5.22 5.67 6.03 6.33 6.58 6.8 6.99 7.17 7.32
0.01 5.7 6.97 7.8 8.42 8.91 9.32 9.67 9.97 10.2 10.5 10.7
6 0.05 3.46 4.34 4.9 5.31 5.63 5.89 6.12 6.32 6.49 6.65 6.79
0.01 5.24 6.33 7.03 7.56 7.97 8.32 8.61 8.87 9.1 9.3 9.49
7 0.05 3.34 4.16 4.69 5.06 5.36 5.61 5.82 6 6.16 6.3 6.43
0.01 4.95 5.92 6.54 7.01 7.37 7.68 7.94 8.17 8.37 8.55 8.71
8 0.05 3.26 4.04 4.53 4.89 5.17 5.4 5.6 5.77 5.92 6.05 6.18
0.01 4.74 5.63 6.2 6.63 6.96 7.24 7.47 7.68 7.87 8.03 8.18
9 0.05 3.2 3.95 4.42 4.76 5.02 5.25 5.43 5.6 5.74 5.87 5.98
0.01 4.6 5.43 5.96 6.35 6.66 6.91 7.13 7.32 7.49 7.65 7.78
10 0.05 3.15 3.88 4.33 4.65 4.91 5.12 5.3 5.46 5.6 5.72 5.83
0.01 4.48 5.27 5.77 6.14 6.43 6.67 6.87 7.05 7.21 7.36 7.48
11 0.05 3.11 3.82 4.26 4.57 4.82 5.03 5.2 5.35 5.49 5.61 5.71
0.01 4.39 5.14 5.26 5.97 6.25 6.48 6.67 6.84 6.99 7.13 7.26
12 0.05 3.08 3.77 4.2 4.51 4.75 4.95 5.12 5.27 5.4 5.51 5.62
0.01 4.32 5.04 5.5 5.84 6.1 6.32 6.51 6.67 6.81 6.94 7.06
-321-
Back to TOC
k (Number of Groups Being Compared)
df
Within α 2 3 4 5 6 7 8 9 10 11 12
Groups
13 0.05 3.06 3.73 4.15 4.45 4.69 4.88 5.05 5.19 5.32 5.43 5.53
0.01 4.26 4.96 5.4 5.73 5.98 6.19 6.37 6.53 6.67 6.79 6.9
14 0.05 3.03 3.7 4.11 4.41 4.64 4.83 4.99 5.13 5.25 5.36 5.46
0.01 4.21 4.89 5.32 5.63 5.88 6.08 6.26 6.41 6.54 6.66 6.77
15 0.05 3.01 3.67 4.08 4.37 4.59 4.78 4.94 5.08 5.2 5.31 5.4
0.01 4.17 4.84 5.25 5.56 5.8 5.99 6.16 6.31 6.44 6.55 6.66
16 0.05 3 3.65 4.05 4.33 4.56 4.74 4.9 5.03 5.15 5.26 5.35
0.01 4.13 4.78 5.19 5.49 5.72 5.92 6.08 6.22 6.35 6.46 6.56
17 0.05 2.98 3.63 4.02 4.3 4.52 4.7 4.86 4.99 5.11 5.21 5.31
0.01 4.1 4.74 5.14 5.43 5.66 5.85 6.01 6.15 6.27 6.38 6.48
18 0.05 2.97 3.61 4 4.28 4.49 4.67 4.82 4.96 5.07 5.17 5.27
0.01 4.07 4.7 5.09 5.38 5.6 5.79 5.94 6.08 6.2 6.31 6.41
19 0.05 2.96 3.59 3.98 4.25 4.47 4.65 4.79 4.92 5.04 5.14 5.23
0.01 4.05 4.67 5.05 5.33 5.55 5.73 5.89 6.02 6.14 6.25 6.34
20 0.05 2.95 3.58 3.96 4.23 4.45 4.62 4.77 4.9 5.01 5.11 5.2
0.01 4.02 4.64 5.02 5.29 5.51 5.69 5.84 5.97 6.09 6.19 6.29
24 0.05 2.92 3.53 3.9 4.17 4.37 4.54 4.68 4.81 4.92 5.01 5.1
0.01 3.96 4.54 4.91 5.17 5.37 5.54 5.69 5.81 5.92 6.02 6.11
30 0.05 2.89 3.49 3.84 4.1 4.3 4.46 4.6 4.72 4.83 4.92 5
0.01 3.89 4.45 4.8 5.05 5.24 5.4 5.54 5.56 5.76 5.85 5.93
40 0.05 2.86 3.44 3.79 4.04 4.23 4.39 4.52 4.63 4.74 4.82 4.91
0.01 3.82 4.37 4.7 4.93 5.11 5.27 5.39 5.5 5.6 5.69 5.77
60 0.05 2.83 3.4 3.74 3.98 4.16 4.31 4.44 4.55 4.65 4.73 4.81
0.01 3.76 4.28 4.6 4.82 4.99 5.13 5.25 5.36 5.45 5.53 5.6
120 0.05 2.8 3.36 3.69 3.92 4.1 4.24 4.36 4.48 4.56 4.64 4.72
0.01 3.7 4.2 4.5 4.71 4.87 5.01 5.12 5.21 5.3 5.38 5.44
∞ 0.05 2.77 3.31 3.63 3.86 4.03 4.17 4.29 4.39 4.47 4.55 4.62
0.01 3.64 4.12 4.4 4.6 4.76 4.88 4.99 5.08 5.16 5.23 5.29
-322-
Back to TOC
Table 5: Critical Values in the Chi-Square (χ2) Distribution
Degrees of Freedom (df)
χ2 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 25 30
0.00 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
0.10 0.7518 0.9512 0.9918 0.9988 0.9998 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
0.20 0.6547 0.9048 0.9776 0.9953 0.9991 0.9998 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
0.30 0.5839 0.8607 0.9600 0.9898 0.9976 0.9995 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
0.40 0.5271 0.8187 0.9402 0.9825 0.9953 0.9989 0.9997 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
0.50 0.4795 0.7788 0.9189 0.9735 0.9921 0.9978 0.9994 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
0.60 0.4386 0.7408 0.8964 0.9631 0.9880 0.9964 0.9990 0.9997 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
0.70 0.4028 0.7047 0.8732 0.9513 0.9830 0.9945 0.9983 0.9995 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
0.80 0.3711 0.6703 0.8495 0.9384 0.9770 0.9921 0.9974 0.9992 0.9998 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
0.90 0.3428 0.6376 0.8254 0.9246 0.9702 0.9891 0.9963 0.9988 0.9996 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
1.00 0.3173 0.6065 0.8013 0.9098 0.9626 0.9856 0.9948 0.9982 0.9994 0.9998 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
1.10 0.2943 0.5769 0.7771 0.8943 0.9541 0.9815 0.9931 0.9975 0.9992 0.9997 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
1.20 0.2733 0.5488 0.7530 0.8781 0.9449 0.9769 0.9909 0.9966 0.9988 0.9996 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
1.30 0.2542 0.5220 0.7291 0.8614 0.9349 0.9717 0.9884 0.9956 0.9984 0.9994 0.9998 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
1.40 0.2367 0.4966 0.7055 0.8442 0.9243 0.9659 0.9856 0.9942 0.9978 0.9992 0.9997 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
1.50 0.2207 0.4724 0.6823 0.8266 0.9131 0.9595 0.9823 0.9927 0.9971 0.9989 0.9996 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
1.60 0.2059 0.4493 0.6594 0.8088 0.9012 0.9526 0.9786 0.9909 0.9963 0.9986 0.9995 0.9998 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
1.70 0.1923 0.4274 0.6369 0.7907 0.8889 0.9451 0.9746 0.9889 0.9954 0.9982 0.9993 0.9997 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
1.80 0.1797 0.4066 0.6149 0.7725 0.8761 0.9371 0.9701 0.9865 0.9942 0.9977 0.9991 0.9997 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
1.90 0.1681 0.3867 0.5934 0.7541 0.8628 0.9287 0.9652 0.9839 0.9930 0.9971 0.9988 0.9995 0.9998 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
2.00 0.1573 0.3679 0.5724 0.7358 0.8491 0.9197 0.9598 0.9810 0.9915 0.9963 0.9985 0.9994 0.9998 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
2.10 0.1473 0.3499 0.5519 0.7174 0.8351 0.9103 0.9541 0.9778 0.9898 0.9955 0.9981 0.9992 0.9997 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
2.20 0.1380 0.3329 0.5319 0.6990 0.8208 0.9004 0.9479 0.9743 0.9879 0.9946 0.9977 0.9990 0.9996 0.9999 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
2.30 0.1294 0.3166 0.5125 0.6808 0.8063 0.8901 0.9414 0.9704 0.9858 0.9935 0.9971 0.9988 0.9995 0.9998 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
2.40 0.1213 0.3012 0.4936 0.6626 0.7915 0.8795 0.9344 0.9662 0.9835 0.9923 0.9965 0.9985 0.9994 0.9997 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
2.50 0.1138 0.2865 0.4753 0.6446 0.7765 0.8685 0.9271 0.9617 0.9809 0.9909 0.9958 0.9982 0.9992 0.9997 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
2.60 0.1069 0.2725 0.4575 0.6268 0.7614 0.8571 0.9194 0.9569 0.9781 0.9893 0.9950 0.9978 0.9990 0.9996 0.9998 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
2.70 0.1003 0.2592 0.4402 0.6092 0.7461 0.8454 0.9113 0.9518 0.9750 0.9876 0.9941 0.9973 0.9988 0.9995 0.9998 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
2.80 0.0943 0.2466 0.4235 0.5918 0.7308 0.8335 0.9029 0.9463 0.9717 0.9857 0.9931 0.9968 0.9986 0.9994 0.9997 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
2.90 0.0886 0.2346 0.4073 0.5747 0.7154 0.8213 0.8941 0.9405 0.9681 0.9837 0.9920 0.9962 0.9983 0.9992 0.9997 0.9999 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000
3.00 0.0833 0.2231 0.3916 0.5578 0.7000 0.8088 0.8850 0.9344 0.9643 0.9814 0.9907 0.9955 0.9979 0.9991 0.9996 0.9998 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000
3.10 0.0783 0.2122 0.3765 0.5412 0.6846 0.7962 0.8756 0.9279 0.9602 0.9790 0.9893 0.9948 0.9975 0.9989 0.9995 0.9998 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000
3.20 0.0736 0.2019 0.3618 0.5249 0.6692 0.7834 0.8659 0.9212 0.9558 0.9763 0.9878 0.9940 0.9971 0.9987 0.9994 0.9997 0.9999 1.0000 1.0000 1.0000 1.0000 1.0000
3.30 0.0693 0.1920 0.3476 0.5089 0.6538 0.7704 0.8559 0.9141 0.9512 0.9735 0.9861 0.9930 0.9966 0.9984 0.9993 0.9997 0.9999 0.9999 1.0000 1.0000 1.0000 1.0000
3.40 0.0652 0.1827 0.3340 0.4932 0.6386 0.7572 0.8457 0.9068 0.9463 0.9704 0.9843 0.9920 0.9961 0.9981 0.9991 0.9996 0.9998 0.9999 1.0000 1.0000 1.0000 1.0000
3.50 0.0614 0.1738 0.3208 0.4779 0.6234 0.7440 0.8352 0.8992 0.9411 0.9671 0.9823 0.9909 0.9954 0.9978 0.9990 0.9995 0.9998 0.9999 1.0000 1.0000 1.0000 1.0000
3.60 0.0578 0.1653 0.3080 0.4628 0.6083 0.7306 0.8245 0.8913 0.9357 0.9636 0.9802 0.9896 0.9948 0.9974 0.9988 0.9994 0.9997 0.9999 1.0000 1.0000 1.0000 1.0000
3.70 0.0544 0.1572 0.2957 0.4481 0.5934 0.7172 0.8136 0.8831 0.9300 0.9599 0.9779 0.9883 0.9940 0.9970 0.9986 0.9993 0.9997 0.9999 0.9999 1.0000 1.0000 1.0000
3.80 0.0513 0.1496 0.2839 0.4337 0.5786 0.7037 0.8025 0.8747 0.9241 0.9559 0.9754 0.9868 0.9931 0.9966 0.9983 0.9992 0.9996 0.9998 0.9999 1.0000 1.0000 1.0000
3.90 0.0483 0.1423 0.2725 0.4197 0.5639 0.6902 0.7912 0.8660 0.9179 0.9517 0.9728 0.9852 0.9922 0.9960 0.9980 0.9991 0.9996 0.9998 0.9999 1.0000 1.0000 1.0000
4.00 0.0455 0.1353 0.2615 0.4060 0.5494 0.6767 0.7798 0.8571 0.9114 0.9473 0.9699 0.9834 0.9912 0.9955 0.9977 0.9989 0.9995 0.9998 0.9999 1.0000 1.0000 1.0000
4.10 0.0429 0.1287 0.2509 0.3926 0.5351 0.6631 0.7682 0.8480 0.9047 0.9427 0.9669 0.9816 0.9901 0.9948 0.9974 0.9987 0.9994 0.9997 0.9999 0.9999 1.0000 1.0000
4.20 0.0404 0.1225 0.2407 0.3796 0.5210 0.6496 0.7565 0.8386 0.8978 0.9379 0.9637 0.9796 0.9889 0.9941 0.9970 0.9985 0.9993 0.9997 0.9998 0.9999 1.0000 1.0000
4.30 0.0381 0.1165 0.2308 0.3669 0.5071 0.6361 0.7446 0.8291 0.8906 0.9328 0.9603 0.9774 0.9876 0.9934 0.9966 0.9983 0.9992 0.9996 0.9998 0.9999 1.0000 1.0000
4.40 0.0359 0.1108 0.2214 0.3546 0.4934 0.6227 0.7327 0.8194 0.8832 0.9275 0.9567 0.9751 0.9861 0.9925 0.9961 0.9980 0.9990 0.9995 0.9998 0.9999 1.0000 1.0000
4.50 0.0339 0.1054 0.2123 0.3425 0.4799 0.6093 0.7207 0.8094 0.8755 0.9220 0.9529 0.9726 0.9846 0.9916 0.9956 0.9977 0.9989 0.9994 0.9997 0.9999 1.0000 1.0000
4.60 0.0320 0.1003 0.2035 0.3309 0.4666 0.5960 0.7086 0.7993 0.8677 0.9162 0.9490 0.9700 0.9830 0.9906 0.9950 0.9974 0.9987 0.9994 0.9997 0.9999 1.0000 1.0000
4.70 0.0302 0.0954 0.1951 0.3195 0.4536 0.5828 0.6965 0.7891 0.8596 0.9103 0.9448 0.9673 0.9812 0.9896 0.9944 0.9971 0.9985 0.9993 0.9996 0.9998 1.0000 1.0000
4.80 0.0285 0.0907 0.1870 0.3084 0.4408 0.5697 0.6844 0.7787 0.8514 0.9041 0.9405 0.9643 0.9793 0.9884 0.9937 0.9967 0.9983 0.9991 0.9996 0.9998 1.0000 1.0000
4.90 0.0269 0.0863 0.1793 0.2977 0.4282 0.5567 0.6722 0.7682 0.8429 0.8978 0.9359 0.9612 0.9773 0.9872 0.9929 0.9962 0.9980 0.9990 0.9995 0.9998 1.0000 1.0000
5.00 0.0253 0.0821 0.1718 0.2873 0.4159 0.5438 0.6600 0.7576 0.8343 0.8912 0.9312 0.9580 0.9752 0.9858 0.9921 0.9958 0.9978 0.9989 0.9994 0.9997 1.0000 1.0000
5.20 0.0226 0.0743 0.1577 0.2674 0.3920 0.5184 0.6356 0.7360 0.8165 0.8774 0.9211 0.9510 0.9705 0.9828 0.9903 0.9947 0.9971 0.9985 0.9992 0.9996 1.0000 1.0000
5.40 0.0201 0.0672 0.1447 0.2487 0.3690 0.4936 0.6113 0.7141 0.7981 0.8629 0.9103 0.9433 0.9653 0.9794 0.9882 0.9934 0.9964 0.9981 0.9990 0.9995 1.0000 1.0000
5.60 0.0180 0.0608 0.1328 0.2311 0.3471 0.4695 0.5872 0.6919 0.7792 0.8477 0.8987 0.9349 0.9595 0.9756 0.9857 0.9919 0.9955 0.9976 0.9987 0.9993 1.0000 1.0000
5.80 0.0160 0.0550 0.1218 0.2146 0.3262 0.4460 0.5633 0.6696 0.7598 0.8318 0.8864 0.9258 0.9531 0.9713 0.9829 0.9901 0.9944 0.9969 0.9984 0.9991 1.0000 1.0000
6.00 0.0143 0.0498 0.1116 0.1991 0.3062 0.4232 0.5397 0.6472 0.7399 0.8153 0.8734 0.9161 0.9462 0.9665 0.9797 0.9881 0.9932 0.9962 0.9979 0.9989 1.0000 1.0000
6.20 0.0128 0.0450 0.1023 0.1847 0.2872 0.4012 0.5166 0.6248 0.7197 0.7982 0.8597 0.9057 0.9386 0.9612 0.9762 0.9858 0.9917 0.9953 0.9974 0.9986 1.0000 1.0000
6.40 0.0114 0.0408 0.0937 0.1712 0.2692 0.3799 0.4939 0.6025 0.6993 0.7806 0.8454 0.8946 0.9304 0.9554 0.9722 0.9832 0.9901 0.9943 0.9968 0.9982 0.9999 1.0000
6.60 0.0102 0.0369 0.0858 0.1586 0.2521 0.3594 0.4717 0.5803 0.6787 0.7626 0.8305 0.8829 0.9216 0.9490 0.9678 0.9802 0.9882 0.9931 0.9961 0.9978 0.9999 1.0000
6.80 0.0091 0.0334 0.0786 0.1468 0.2359 0.3397 0.4500 0.5584 0.6579 0.7442 0.8150 0.8705 0.9122 0.9421 0.9630 0.9769 0.9860 0.9917 0.9952 0.9973 0.9999 1.0000
7.00 0.0082 0.0302 0.0719 0.1359 0.2206 0.3208 0.4289 0.5366 0.6371 0.7254 0.7991 0.8576 0.9022 0.9347 0.9576 0.9733 0.9835 0.9901 0.9942 0.9967 0.9999 1.0000
7.20 0.0073 0.0273 0.0658 0.1257 0.2062 0.3027 0.4084 0.5152 0.6163 0.7064 0.7827 0.8441 0.8916 0.9267 0.9519 0.9692 0.9808 0.9883 0.9931 0.9960 0.9998 1.0000
-323-
Back to TOC
χ2 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 25 30
7.40 0.0065 0.0247 0.0602 0.1162 0.1926 0.2854 0.3885 0.4942 0.5955 0.6872 0.7658 0.8301 0.8804 0.9182 0.9456 0.9648 0.9778 0.9863 0.9918 0.9952 0.9998 1.0000
7.60 0.0058 0.0224 0.0550 0.1074 0.1797 0.2689 0.3692 0.4735 0.5749 0.6678 0.7486 0.8156 0.8686 0.9091 0.9388 0.9599 0.9744 0.9840 0.9903 0.9942 0.9997 1.0000
7.80 0.0052 0.0202 0.0503 0.0992 0.1676 0.2531 0.3506 0.4532 0.5544 0.6484 0.7311 0.8006 0.8564 0.8995 0.9316 0.9546 0.9706 0.9815 0.9886 0.9931 0.9996 1.0000
8.00 0.0047 0.0183 0.0460 0.0916 0.1562 0.2381 0.3326 0.4335 0.5341 0.6288 0.7133 0.7851 0.8436 0.8893 0.9238 0.9489 0.9665 0.9786 0.9867 0.9919 0.9995 1.0000
8.20 0.0042 0.0166 0.0421 0.0845 0.1456 0.2238 0.3153 0.4142 0.5141 0.6093 0.6953 0.7693 0.8303 0.8786 0.9155 0.9427 0.9621 0.9755 0.9845 0.9905 0.9994 1.0000
8.40 0.0038 0.0150 0.0384 0.0780 0.1355 0.2102 0.2986 0.3954 0.4944 0.5898 0.6771 0.7531 0.8166 0.8675 0.9067 0.9361 0.9572 0.9721 0.9822 0.9889 0.9992 1.0000
8.60 0.0034 0.0136 0.0351 0.0719 0.1261 0.1974 0.2827 0.3772 0.4750 0.5704 0.6588 0.7367 0.8024 0.8558 0.8975 0.9290 0.9520 0.9683 0.9795 0.9871 0.9990 1.0000
8.80 0.0030 0.0123 0.0321 0.0663 0.1173 0.1851 0.2673 0.3594 0.4559 0.5512 0.6403 0.7199 0.7879 0.8436 0.8877 0.9214 0.9463 0.9642 0.9767 0.9851 0.9988 0.9999
9.00 0.0027 0.0111 0.0293 0.0611 0.1091 0.1736 0.2527 0.3423 0.4373 0.5321 0.6219 0.7029 0.7729 0.8311 0.8775 0.9134 0.9403 0.9597 0.9735 0.9829 0.9986 0.9999
9.20 0.0024 0.0101 0.0267 0.0563 0.1013 0.1626 0.2386 0.3257 0.4190 0.5132 0.6034 0.6858 0.7577 0.8180 0.8668 0.9049 0.9338 0.9549 0.9700 0.9805 0.9983 0.9999
9.40 0.0022 0.0091 0.0244 0.0518 0.0941 0.1523 0.2252 0.3097 0.4012 0.4946 0.5850 0.6684 0.7421 0.8046 0.8557 0.8960 0.9269 0.9497 0.9662 0.9778 0.9980 0.9999
9.60 0.0019 0.0082 0.0223 0.0477 0.0874 0.1425 0.2124 0.2942 0.3838 0.4763 0.5667 0.6510 0.7263 0.7908 0.8441 0.8867 0.9195 0.9442 0.9621 0.9749 0.9976 0.9999
9.80 0.0017 0.0074 0.0203 0.0439 0.0811 0.1333 0.2002 0.2793 0.3669 0.4582 0.5485 0.6335 0.7102 0.7767 0.8321 0.8769 0.9118 0.9382 0.9577 0.9717 0.9972 0.9998
10.00 0.0016 0.0067 0.0186 0.0404 0.0752 0.1247 0.1886 0.2650 0.3505 0.4405 0.5304 0.6160 0.6939 0.7622 0.8197 0.8666 0.9036 0.9319 0.9529 0.9682 0.9967 0.9998
10.50 0.0012 0.0052 0.0148 0.0328 0.0622 0.1051 0.1620 0.2317 0.3115 0.3978 0.4860 0.5722 0.6526 0.7248 0.7872 0.8392 0.8813 0.9144 0.9395 0.9582 0.9951 0.9996
11.00 0.0009 0.0041 0.0117 0.0266 0.0514 0.0884 0.1386 0.2017 0.2757 0.3575 0.4433 0.5289 0.6108 0.6860 0.7526 0.8095 0.8566 0.8944 0.9238 0.9462 0.9929 0.9994
11.50 0.0007 0.0032 0.0093 0.0215 0.0423 0.0741 0.1182 0.1749 0.2430 0.3199 0.4024 0.4866 0.5690 0.6464 0.7164 0.7776 0.8294 0.8719 0.9059 0.9322 0.9902 0.9991
12.00 0.0005 0.0025 0.0074 0.0174 0.0348 0.0620 0.1006 0.1512 0.2133 0.2851 0.3636 0.4457 0.5276 0.6063 0.6790 0.7440 0.8001 0.8472 0.8856 0.9161 0.9866 0.9986
12.50 0.0004 0.0019 0.0059 0.0140 0.0285 0.0517 0.0853 0.1303 0.1866 0.2530 0.3273 0.4064 0.4871 0.5662 0.6409 0.7089 0.7690 0.8204 0.8632 0.8978 0.9821 0.9979
13.00 0.0003 0.0015 0.0046 0.0113 0.0234 0.0430 0.0721 0.1118 0.1626 0.2237 0.2933 0.3690 0.4478 0.5265 0.6023 0.6728 0.7362 0.7916 0.8386 0.8774 0.9765 0.9970
13.50 0.0002 0.0012 0.0037 0.0091 0.0191 0.0357 0.0608 0.0958 0.1413 0.1970 0.2619 0.3338 0.4100 0.4876 0.5637 0.6359 0.7021 0.7611 0.8120 0.8549 0.9698 0.9958
14.00 0.0002 0.0009 0.0029 0.0073 0.0156 0.0296 0.0512 0.0818 0.1223 0.1730 0.2330 0.3007 0.3738 0.4497 0.5255 0.5987 0.6671 0.7291 0.7837 0.8305 0.9617 0.9943
14.50 0.0001 0.0007 0.0023 0.0059 0.0127 0.0245 0.0430 0.0696 0.1056 0.1514 0.2065 0.2699 0.3396 0.4132 0.4880 0.5615 0.6314 0.6960 0.7538 0.8043 0.9523 0.9923
15.00 0.0001 0.0006 0.0018 0.0047 0.0104 0.0203 0.0360 0.0591 0.0909 0.1321 0.1825 0.2414 0.3074 0.3782 0.4514 0.5246 0.5955 0.6620 0.7226 0.7764 0.9414 0.9897
15.50 0.0001 0.0004 0.0014 0.0038 0.0084 0.0167 0.0301 0.0501 0.0781 0.1149 0.1607 0.2152 0.2772 0.3449 0.4160 0.4884 0.5595 0.6274 0.6903 0.7471 0.9289 0.9866
16.00 0.0001 0.0003 0.0011 0.0030 0.0068 0.0138 0.0251 0.0424 0.0669 0.0996 0.1411 0.1912 0.2491 0.3134 0.3821 0.4530 0.5238 0.5925 0.6573 0.7166 0.9148 0.9827
16.50 0.0000 0.0003 0.0009 0.0024 0.0056 0.0113 0.0209 0.0358 0.0571 0.0862 0.1236 0.1694 0.2232 0.2838 0.3496 0.4186 0.4887 0.5577 0.6237 0.6852 0.8991 0.9781
17.00 0.0000 0.0002 0.0007 0.0019 0.0045 0.0093 0.0174 0.0301 0.0487 0.0744 0.1079 0.1496 0.1993 0.2562 0.3189 0.3856 0.4544 0.5231 0.5899 0.6530 0.8818 0.9726
17.50 0.0000 0.0002 0.0006 0.0015 0.0036 0.0076 0.0144 0.0253 0.0414 0.0640 0.0939 0.1317 0.1774 0.2305 0.2899 0.3540 0.4210 0.4890 0.5560 0.6203 0.8629 0.9661
18.00 0.0000 0.0001 0.0004 0.0012 0.0029 0.0062 0.0120 0.0212 0.0352 0.0550 0.0816 0.1157 0.1575 0.2068 0.2627 0.3239 0.3888 0.4557 0.5224 0.5874 0.8424 0.9585
18.50 0.0000 0.0001 0.0003 0.0010 0.0024 0.0051 0.0099 0.0178 0.0298 0.0471 0.0707 0.1013 0.1394 0.1849 0.2373 0.2954 0.3580 0.4232 0.4893 0.5545 0.8204 0.9499
19.00 0.0000 0.0001 0.0003 0.0008 0.0019 0.0042 0.0082 0.0149 0.0252 0.0403 0.0611 0.0885 0.1231 0.1649 0.2137 0.2687 0.3285 0.3918 0.4568 0.5218 0.7971 0.9400
19.50 0.0000 0.0001 0.0002 0.0006 0.0016 0.0034 0.0068 0.0124 0.0213 0.0344 0.0527 0.0772 0.1084 0.1467 0.1920 0.2436 0.3006 0.3617 0.4252 0.4896 0.7725 0.9289
20.00 0.0000 0.0000 0.0002 0.0005 0.0012 0.0028 0.0056 0.0103 0.0179 0.0293 0.0453 0.0671 0.0952 0.1301 0.1719 0.2202 0.2742 0.3328 0.3946 0.4579 0.7468 0.9165
20.50 0.0000 0.0000 0.0001 0.0004 0.0010 0.0023 0.0046 0.0086 0.0151 0.0249 0.0389 0.0582 0.0834 0.1151 0.1536 0.1985 0.2495 0.3054 0.3651 0.4271 0.7201 0.9029
21.00 0.0000 0.0000 0.0001 0.0003 0.0008 0.0018 0.0038 0.0071 0.0127 0.0211 0.0334 0.0504 0.0729 0.1016 0.1368 0.1785 0.2263 0.2794 0.3368 0.3971 0.6926 0.8879
21.50 0.0000 0.0000 0.0001 0.0003 0.0007 0.0015 0.0031 0.0059 0.0106 0.0179 0.0285 0.0435 0.0636 0.0895 0.1216 0.1601 0.2047 0.2549 0.3098 0.3682 0.6644 0.8716
22.00 0.0000 0.0000 0.0001 0.0002 0.0005 0.0012 0.0025 0.0049 0.0089 0.0151 0.0244 0.0375 0.0554 0.0786 0.1078 0.1432 0.1847 0.2320 0.2843 0.3405 0.6357 0.8540
22.50 0.0000 0.0000 0.0001 0.0002 0.0004 0.0010 0.0021 0.0041 0.0074 0.0128 0.0208 0.0323 0.0481 0.0689 0.0953 0.1278 0.1662 0.2105 0.2601 0.3140 0.6067 0.8352
23.00 0.0000 0.0000 0.0000 0.0001 0.0003 0.0008 0.0017 0.0034 0.0062 0.0107 0.0177 0.0277 0.0417 0.0603 0.0841 0.1137 0.1493 0.1906 0.2373 0.2888 0.5776 0.8153
23.50 0.0000 0.0000 0.0000 0.0001 0.0003 0.0006 0.0014 0.0028 0.0052 0.0090 0.0150 0.0238 0.0361 0.0526 0.0741 0.1010 0.1337 0.1721 0.2160 0.2649 0.5484 0.7942
24.00 0.0000 0.0000 0.0000 0.0001 0.0002 0.0005 0.0011 0.0023 0.0043 0.0076 0.0127 0.0203 0.0311 0.0458 0.0651 0.0895 0.1194 0.1550 0.1962 0.2424 0.5194 0.7720
24.50 0.0000 0.0000 0.0000 0.0001 0.0002 0.0004 0.0009 0.0019 0.0036 0.0064 0.0108 0.0174 0.0268 0.0398 0.0571 0.0791 0.1065 0.1393 0.1777 0.2212 0.4907 0.7489
25.00 0.0000 0.0000 0.0000 0.0001 0.0001 0.0003 0.0008 0.0016 0.0030 0.0053 0.0091 0.0148 0.0231 0.0346 0.0499 0.0698 0.0947 0.1249 0.1605 0.2014 0.4624 0.7250
26.00 0.0000 0.0000 0.0000 0.0000 0.0001 0.0002 0.0005 0.0011 0.0020 0.0037 0.0065 0.0107 0.0170 0.0259 0.0380 0.0540 0.0745 0.0998 0.1302 0.1658 0.4076 0.6751
27.00 0.0000 0.0000 0.0000 0.0000 0.0001 0.0001 0.0003 0.0007 0.0014 0.0026 0.0046 0.0077 0.0124 0.0193 0.0287 0.0415 0.0581 0.0790 0.1047 0.1353 0.3559 0.6233
28.00 0.0000 0.0000 0.0000 0.0000 0.0000 0.0001 0.0002 0.0005 0.0010 0.0018 0.0032 0.0055 0.0090 0.0142 0.0216 0.0316 0.0449 0.0621 0.0834 0.1094 0.3079 0.5704
29.00 0.0000 0.0000 0.0000 0.0000 0.0000 0.0001 0.0001 0.0003 0.0006 0.0012 0.0023 0.0039 0.0065 0.0105 0.0161 0.0239 0.0345 0.0484 0.0660 0.0878 0.2639 0.5176
30.00 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0001 0.0002 0.0004 0.0009 0.0016 0.0028 0.0047 0.0076 0.0119 0.0180 0.0263 0.0374 0.0518 0.0699 0.2243 0.4657
31.00 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0001 0.0001 0.0003 0.0006 0.0011 0.0020 0.0034 0.0055 0.0088 0.0135 0.0200 0.0288 0.0404 0.0552 0.1890 0.4154
32.00 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0001 0.0002 0.0004 0.0008 0.0014 0.0024 0.0040 0.0064 0.0100 0.0150 0.0220 0.0313 0.0433 0.1580 0.3675
33.00 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0001 0.0001 0.0003 0.0005 0.0010 0.0017 0.0029 0.0047 0.0074 0.0113 0.0167 0.0240 0.0337 0.1311 0.3225
34.00 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0001 0.0002 0.0004 0.0007 0.0012 0.0021 0.0034 0.0054 0.0084 0.0126 0.0184 0.0261 0.1079 0.2808
35.00 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0001 0.0001 0.0002 0.0005 0.0008 0.0015 0.0025 0.0040 0.0062 0.0095 0.0140 0.0201 0.0882 0.2426
36.00 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0001 0.0002 0.0003 0.0006 0.0010 0.0018 0.0029 0.0046 0.0071 0.0106 0.0154 0.0716 0.2081
37.00 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0001 0.0001 0.0002 0.0004 0.0007 0.0013 0.0021 0.0034 0.0052 0.0079 0.0117 0.0577 0.1771
38.00 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0001 0.0002 0.0003 0.0005 0.0009 0.0015 0.0025 0.0039 0.0059 0.0089 0.0463 0.1497
39.00 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0001 0.0001 0.0002 0.0004 0.0006 0.0011 0.0018 0.0028 0.0044 0.0067 0.0368 0.1257
40.00 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0001 0.0001 0.0003 0.0005 0.0008 0.0013 0.0021 0.0033 0.0050 0.0292 0.1049
-324-
Back to TOC
Table 6: Fisher’s Transformation of Pearson r to r' (ρ to ρ')
r (ρ) r' (ρ') r (ρ) r' (ρ') r (ρ) r' (ρ') r (ρ) r' (ρ') r (ρ) r' (ρ')
0.000 0.000 0.200 0.203 0.400 0.424 0.600 0.693 0.800 1.099
0.005 0.005 0.205 0.208 0.405 0.430 0.605 0.701 0.805 1.113
0.010 0.010 0.210 0.213 0.410 0.436 0.610 0.709 0.810 1.127
0.015 0.015 0.215 0.218 0.415 0.442 0.615 0.717 0.815 1.142
0.020 0.020 0.220 0.224 0.420 0.448 0.620 0.725 0.820 1.157
0.025 0.025 0.225 0.229 0.425 0.454 0.625 0.733 0.825 1.172
0.030 0.030 0.230 0.234 0.430 0.460 0.630 0.741 0.830 1.188
0.035 0.035 0.235 0.239 0.435 0.466 0.635 0.750 0.835 1.204
0.040 0.040 0.240 0.245 0.440 0.472 0.640 0.758 0.840 1.221
0.045 0.045 0.245 0.250 0.445 0.478 0.645 0.767 0.845 1.238
0.050 0.050 0.250 0.255 0.450 0.485 0.650 0.775 0.850 1.256
0.055 0.055 0.255 0.261 0.455 0.491 0.655 0.784 0.855 1.274
0.060 0.060 0.260 0.266 0.460 0.497 0.660 0.793 0.860 1.293
0.065 0.065 0.265 0.271 0.465 0.504 0.665 0.802 0.865 1.313
0.070 0.070 0.270 0.277 0.470 0.510 0.670 0.811 0.870 1.333
0.075 0.075 0.275 0.282 0.475 0.517 0.675 0.820 0.875 1.354
0.080 0.080 0.280 0.288 0.480 0.523 0.680 0.829 0.880 1.376
0.085 0.085 0.285 0.293 0.485 0.530 0.685 0.838 0.885 1.398
0.090 0.090 0.290 0.299 0.490 0.536 0.690 0.848 0.890 1.422
0.095 0.095 0.295 0.304 0.495 0.543 0.695 0.858 0.895 1.447
0.100 0.100 0.300 0.310 0.500 0.549 0.700 0.867 0.900 1.472
0.105 0.105 0.305 0.315 0.505 0.556 0.705 0.877 0.905 1.499
0.110 0.110 0.310 0.321 0.510 0.563 0.710 0.887 0.910 1.528
0.115 0.116 0.315 0.326 0.515 0.570 0.715 0.897 0.915 1.557
0.120 0.121 0.320 0.332 0.520 0.576 0.720 0.908 0.920 1.589
0.125 0.126 0.325 0.337 0.525 0.583 0.725 0.918 0.925 1.623
0.130 0.131 0.330 0.343 0.530 0.590 0.730 0.929 0.930 1.658
0.135 0.136 0.335 0.348 0.535 0.597 0.735 0.940 0.935 1.697
0.140 0.141 0.340 0.354 0.540 0.604 0.740 0.950 0.940 1.738
0.145 0.146 0.345 0.360 0.545 0.611 0.745 0.962 0.945 1.783
0.150 0.151 0.350 0.365 0.550 0.618 0.750 0.973 0.950 1.832
0.155 0.156 0.355 0.371 0.555 0.626 0.755 0.984 0.955 1.886
0.160 0.161 0.360 0.377 0.560 0.633 0.760 0.996 0.960 1.946
0.165 0.167 0.365 0.383 0.565 0.640 0.765 1.008 0.965 2.014
0.170 0.172 0.370 0.388 0.570 0.648 0.770 1.020 0.970 2.092
0.175 0.177 0.375 0.394 0.575 0.655 0.775 1.033 0.975 2.185
0.180 0.182 0.380 0.400 0.580 0.662 0.780 1.045 0.980 2.298
0.185 0.187 0.385 0.406 0.585 0.670 0.785 1.058 0.985 2.443
0.190 0.192 0.390 0.412 0.590 0.678 0.790 1.071 0.990 2.647
0.195 0.198 0.395 0.418 0.595 0.685 0.795 1.085 0.995 2.994
-325-
Back to TOC
Table 7: Factorials of Integers 0 – 50
n n! n n!
0 1 26 4.03E+026
1 1 27 1.089E+28
2 2 28 3.049E+29
3 6 29 8.842E+30
4 24 30 2.653E+32
5 120 31 8.223E+33
6 720 32 2.631E+35
7 5040 33 8.683E+36
8 40320 34 2.952E+38
9 362880 35 1.033E+40
10 3628800 36 3.72E+41
11 39916800 37 1.376E+43
12 479001600 38 5.23E+44
13 6.227E+09 39 2.04E+46
14 8.718E+10 40 8.159E+47
15 1.308E+12 41 3.345E+49
16 2.092E+13 42 1.405E+51
17 3.557E+14 43 6.042E+52
18 6.402E+15 44 2.658E+54
19 1.216E+17 45 1.196E+56
20 2.433E+18 46 5.503E+57
21 5.109E+19 47 2.586E+59
22 1.124E+21 48 1.241E+61
23 2.585E+22 49 6.083E+62
24 6.204E+23 50 3.041E+64
25 1.551E+25
-326-
Back to TOC
Appendix B: Measuring Skew
Scientists who use statistics often assume that a set of data is normally distributed, but is this assumption
ever violated, and how can you tell? That is, how can you tell whether a distribution of scores is normally
distributed, or skewed? Indeed, it is good to know whether there is large and significant skew to your
distribution; if so, a transformation of the data to normalize the data may be necessary. This is so,
because nearly all inferential statistical tests assume that the data is normally distributed; hence,
violations of normality can lead to serious problems when interpreting data.
All formulae that have been proposed to measure skew are based on the assumption that the difference
between the mean and the mode will be zero if the distribution is normal. Early measures of skew
subtracted the mode (Mo) from the mean and then standardize that difference by dividing the difference
by the standard deviation of the distribution:
X  Mode
skew 
s
If the mean is less than the mode the resulting value will be negative and the distribution is negatively
skewed. If the mean is greater than the mode the resulting value will be positive and the distribution is
positively skewed.
The more common formula for measuring skew, which is the third moment around a mean, is:
3
 n  XX
skew      
 n  1n  2  ˆs 
You must use the estimated standard deviation, because you need to determine if the skew is significant.
Say the following set of n = 8 scores are collected and the skew must be determined. The table lists all of
steps necessary for calculating the skew:
XX
3
XX
X X  X  X  X  2
   
 ˆs   ˆs 
9 3.5 12.25 1.464 3.140
9 3.5 12.25 1.464 3.140
6 0.5 0.25 .209 0.009
5 -0.5 0.25 -.209 -0.009
5 -0.5 0.25 -.209 -0.009
4 -1.5 2.25 -.628 -0.248
3 -2.5 6.25 -1.046 -1.144
3 -2.5 6.25 -1.046 -1.144
3
∑X = 44 XX
X  5.5

 XX 2
 44     3.735
 ˆs 
-327-
Back to TOC
SS 40
The estimated standard deviation is: ˆs    5.714  2.39
n 1 8 1
Solving for the skew, we have:

3
 n  XX
skew     
 n  1n  2  ˆs 
 
 3.735
8
skew  
 8  18  2 
8
skew    3.735
 42 
skew  0.193.735
skew  0.71
This value is positive, which indicates that the distribution is positively skewed. The range of this measure
of skew is between only -1.00 and +1.00; hence, it is easily used to determine the magnitude of any skew.
Any value greater than ± 0.3 is considered severe skew; thus, in the example above, the skew of 0.71
indicates a severely positively skewed distribution.
But this value provides only a measure of the distribution’s skew. It would be better to know if the skew of
the distribution is so large that it deviates significantly from what is expected in the population, which is no
skew. To determine whether the skew of the distribution is significant, you first calculate the estimated
standard error of skew, which is estimated in the following equation:
6
ˆs skew 
n
6 6
From this example, the estimated standard error of skew is: ˆs skew    0.75  0.866
n 8
skew
The statistical significance test for distribution skew is a type of z-score: z skew 
ˆs skew
skew 0.71
Thus, in this example, the z-score of skew is: z skew    0.82
ˆs skew 0.866
To determine whether the skew is statistically significant, locate that z-score in the z-Tables. Look at the
probability of obtaining that z-score in one tail (Column 3). If the probability is .05 or less, then the skew is
statistically significant. You always use a one-tail test rather than a two-tail test, because skew can go in
one direction (one tail) only. Also, note that the magnitude of the estimated standard error of skew will
decrease as n increases, making it more likely that you will find significant skew with larger sample size.
In effect, having larger sample sizes actually punishes you, which is counter-intuitive, because larger
sample sizes are supposed to better approximate the population. However, in the case of skew, it is
assumed that larger sample sizes will create better approximates toward a normal distribution; that is, as
the sample size increases, the skew should decrease. If this is the case, then as the sample size
increases and the estimated standard error of skew decreases, so too should the magnitude of skew. The
result will be less likelihood of finding significant skew.
-328-
Back to TOC
Appendix C: Calculating Power
From the chapters on inferential testing, recall that all decisions with respect to rejecting or failing to reject
a null hypothesis are based on probabilities; hence, there is always a degree of uncertainty that your
decision about the null hypothesis was correct. More succinctly, your decision about the null could be
wrong!
The alpha-level that you choose (α = .05, or less) is the probability that you are willing to make a Type I
error. The probability of making a Type II error (β) is the probability you incorrectly retain the null
hypothesis when it is actually false; that is, when you fail to reject the null when it is actually false. Finally,
1 - β is the statistical power that you have in your inferential test for correctly rejecting a false null
hypothesis. The two overlapping distributions in the figure below demonstrate where α, β, 1 - α, and 1 - β
occur:
Think of these as two distributions of scores measured from the same variable, where μ A represents a
distribution of scores where the null hypothesis is true, and μ B represents a distribution of scores where
the null hypothesis is false (and the alternate hypothesis is true). The point where the distributions overlap
(the decision line) is the criterion on which you base your decision about a statistical test (the critical
value/critical region). If there is sufficient evidence the null hypothesis is false, such that the evidence is
greater than your criterion, you reject the null hypothesis. In contrast, if there is insufficient evidence, then
you retain the null hypothesis. Basically, the decision whether to reject or retain the null comes down to
your criterion, which is a critical value related to the α-level.
From the figure above, you can see a relationship between α, β, 1 - α, and 1 - β. The μA distribution is
associated with both α and 1 - α, and the μB distribution is associated with β and 1 - β (power). It is
important to remember that p[β + (1 - β)] = 1.00 and that p[α + (1 - α)] = 1.00; thus, [β + (1 - β)] = [α + (1 -
α)]. Based on this, we can use any selected α level to determine the statistical power of an inferential test.
-329-
Back to TOC
Computing the power of a statistical test makes use of z-scores. To calculate power you must know (i) the
value of μ under the null hypothesis (the value to which you are comparing a sample mean), (ii) the value
of the sample mean, (iii) the standard error of the mean, and (iv) the alpha-level. For the present
example, assume that the alpha-level is .05 and we are using a two-tailed (non-directional) hypothesis.
Also, assume that μ = 100, X  105 , and  X  2.5
Calculating the power you have in a statistical test takes several steps that require calculating z-scores,
looking up probabilities associated with z-scores from the z-tables (Appendix A, Table 1), and calculating
raw scores from z-scores.
First, using the z-Tables, find the z-score related to the chosen α-level (zα). You will find that zα = 1.96.
Second, determine the value of a sample mean that would be associated with that zα given μ, zα, and the
standard error of the mean. That is, what is the minimum mean value necessary to conclude that a
sample mean is significantly different from a population mean? In the figure above, this is the value that is
located at the decision line. This value is calculated as:
X    zX
X   100  1.962.5
X   100  4.9
X   104.9
Because both distributions lie along the same continuum, 104.9 would be found in both distributions;
hence, it can be used as a reference point by both distributions.
Third, calculate the z-Score for the difference between 104.9 and the sample mean, which is labeled μ B in
the figure above:
X   X 104.9  105  0.1

z    0.04
X 2.5 2.5
This is the z-score that is associated with your decision criterion in the alternate distribution; that is, the
point where both distributions meet in the figure above. Look up this value in the z-Tables. When you find
this z-score, determine the probability of obtaining a z-score less than or equal to that value from Column
3. This will be the probability of observing a z-Score in the lower tail of the alternate distribution; thus, it is
β, the probability of making a Type II error. In this example, β = .4840.
Finally, to determine the power of the test subtract β in the preceding step from 1.0; thus, the Power to
detect a statistically significant result is 1 - β. In this example, the power (probability) to detect a
statistically significant difference and to correctly reject the null hypothesis is 1.0 - .4840 = .5160. This is
the probability that you will correctly reject the null hypothesis and accept the alternative hypothesis. This
value ranges on a continuum from 0 to 1.00, with higher values indicating more statistical power.
-330-
Back to TOC
Appendix D: Post Hoc Testing
In a research design with more than two groups/samples/conditions for any independent variable you
need to make comparisons between those groups to determine which are significantly different from the
others, contingent on an analysis of variance being significant. For example, assume that you conduct an
experiment on the effects of noise while studying to see if different noise levels interfere with a student’s
ability to study and do well on tests. You include five groups of students. Each student in each group
studies a list of words for 15-minutes and one hour later they attempt to recall as many words as possible
from memory. The groups differ with respect to the level of noise each student in a group is exposed to
while studying. Group-1 studies in sound isolation; Group-2 studies with 20 dBL (decibels) of white noise;
Group-3 studies with 40 dBL; Group-4 studies with 60 dBL; and Group-5 studies with 80 dBL.
You run a one-way analysis of variance with Noise as the independent variable with five levels and
“Percent or Studied Words Correctly Recalled” (Recall) as the dependent variable. Assume that the F-
Ratio from the ANOVA was statistically significant; hence, there is at least one significant difference
between two of these five groups. The question is, how do you determine which groups are significantly
different from each other?
You could use Tukey’s HSD, Fisher’s LSD, or like most people you could conduct multiple t-Tests. What
makes using multiple t-Tests a problem is that the more comparisons you make between groups in a set
of data, the greater your chance of making a Type I error becomes. That is, as the number of
comparisons increase the chance of incorrectly rejecting the null hypothesis for any one of those
comparisons (when it should be retained) also increases.
In the example above there are five groups that you can compare, for a total of ten possible comparisons
that can be made (Group-1 vs. Group-2, Group-1 vs. Group-3, Group-1 vs. Group-4, Group-1 vs. Group-
5, Group-2 vs. Group-3, Group-2 vs. Group-4, Group-2 vs. Group-5, Group-3 vs. Group-4, Group-3 vs.
Group-5, Group-4 vs. Group-5). If α = .05 for each of the ten comparisons and if you use a t-Tests for
each comparison, then what is the chance that at least one Type I error is made? If you think it is p = .05,
you’re wrong. Let me explain...
First remember that the selected alpha-level is the probability that one is willing to live with for making a
Type I error on any test. If one t-Test is conducted on these data, then the probability of making a Type I
error is p = .05. If a second t-Test is performed using all or part of the same data, the the probability of
making a Type I error on that second t-Test is also p = .05. Thus, the probability of making a Type I error
on at least one of those two t-Tests is approximately p = .05 + .05 = .10. If a third t-Test is performed on
these data, then the probability of making at least one Type I error on is now at about p = .05 + .05 + .05
= .15. Basically, the probability of making a Type I error inflates for every additional t-Test performed on
the same data. The probability of making at least one Type I error over multiple comparisons is known as
Familywise Error Rate (αFW) and is calculated as follows:
-331-
Back to TOC
αFW = 1 - (1 - αi)c
Where c is the number of comparisons (number of t-Tests performed), and αi is the alpha level selected
for each individual t-Test, which will usually be the same for each t-Test and the conventional level is αi =
.05. In our example, the Familywise Error Rate after all 10 comparisons are made would be:
αFW = 1 - (1 - .05)10
αFW = 1 - (.95)10
αFW = 1 - (0.599)
αFW = 0.401
Thus, the probability of making at least one Type I error somewhere among these ten t-Tests is p = .401,
which is eight times greater than the probability at which you want to keep the probability of making a
Type I error (α = .05)! The question now is how to control for inflation of the Type I error rate when making
multiple comparisons. Several methods are available and are described below. But first, Table 1, below,
lists all ten comparisons that can be made in the hypothetical example above, complete with mean recall
of the two groups in each comparison and the outcomes of each t-test (these results are purely
hypothetical).
Table 1
a
Comparison Mean 1 Mean 2 n df t p
Isolation vs. 20-dBL .635 .705 20 18 1.980 .0632
Isolation vs. 40-dBL .635 .730 20 18 3.981 .0009
Isolation vs. 60-dBL .635 .660 20 18 0.901 .3794
Isolation vs. 80-dBL .635 .545 20 18 2.886 .0098
20-dBL vs. 40-dBL .705 .730 20 18 0.777 .4474
20-dBL vs. 60-dBL .705 .660 20 18 1.280 .2168
20-dBL vs. 80-dBL .705 .545 20 18 4.218 .0005
40-dBL vs. 60-dBL .730 .660 20 18 2.970 .0082
40-dBL vs. 80-dBL .730 .545 20 18 6.718 .0001
60-dBL vs. 80-dBL .660 .545 20 18 3.715 .0016
a
Exact p-value of t-Test (from SPSS output file).
Note that if you set αi to p = .05 for each t-Test and to not care about Type I error inflation, then any test
with an exact p-value that is less than .05 would be declared statistically significant. But don't ever do
that!
Bonferroni Correction
The Bonferroni correction for multiple comparisons is the simplest method of dealing with Type I error
rate inflation. First, decide what you want the Familywise error rate (α FW ) to be after all of you
comparisons are made, which is typically αFW = .05, or less. (For all of these methods assume you want
αFW = .05) Next, determine how many comparisons (c) are going to be made. In the example above ten
comparisons are made (note that not every comparison has to be made). The Bonferroni correction
simply divides αFW by the number of comparisons. This gives you an alpha-level (αi) to use in assessing
whether each individual t-Test is significant. Thus, for the example above:
 FW .05
i    .005
c 10
This alpha-level (αi = .005) that you will use to determine whether each individual t-Test is statistically
significant. That is, rather than using α = .05 to determine the significance of each individual t-Test, αi =
-332-
.005 will be used. What is nice about this is that even after all ten comparisons, the Familywise error rate
is less than .05:
αFW = 1 - (1 - .005)10
αFW = 1 - (.995)10
αFW = 1 - (.951)
αFW = .049
Table 2 (later) lists whether each comparison is significant using the Bonferroni correction. Unfortunately,
the Bonferroni correction is quite conservative and many outcomes that would normally be declared
statistically significant are not, because of the very low alpha-level required for statistical significance.
Indeed, the difference between the Isolation Group and the 80-dBL group has an exact p-Value of .0098,
which would be significant under normal circumstances, but because .0098 > .005 it must be concluded
that this differences is not statistically reliable. Because of this conservative criterion for significance, the
Bonferroni correction is often criticized and rarely used when many comparisons are made.
Modified Bonferroni Correction

An alternative method to the Bonferroni is the Modified Bonferroni correction. Functionally, this
procedure does the same thing as the simple Bonferroni correction above; that is, using a desired α FW
and a desired number of comparisons (c), you calculate an alpha-level (αi) that is used when assessing
the statistical significance of each individual t-Test. The Modified Bonferroni correction is as follows:
αi = 1 - (1 - αFW)1/c
αi = 1 - (1 - .05)1/10
αi = 1 - (.95)1/10
αi = 1 - .9949
αi = .0051
Notice that αi (.0051) is slightly larger than the αi value calculated using the original Bonferroni correction
(.005), making the Modified Bonferroni slightly less conservative, though not by much. Indeed, if you look
at Table 2, you have gained nothing from using the Modified Bonferroni. Thus, both the Modified
Bonferroni and the Bonferroni procedures are criticized for their conservative significance criterion.
Table 2 Result of
Result of Modified
Bonferroni Bonferroni
Comparison Mean 1 Mean 2 n df t p
Isolation vs. 20-dBL .635 .705 20 18 1.980 .0632 ns ns
Isolation vs. 40-dBL .635 .730 20 18 3.981 .0009 sig sig
20-dBL vs. 40-dBL .705 .730 20 18 0.777 .4474 ns ns
20-dBL vs. 60-dBL .705 .660 20 18 1.280 .2168 ns ns
20-dBL vs. 80-dBL .705 .545 20 18 4.218 .0005 sig sig
40-dBL vs. 60-dBL .730 .660 20 18 2.970 .0082 ns ns
40-dBL vs. 80-dBL .730 .545 20 18 6.718 .0001 sig sig
60-dBL vs. 80-dBL .660 .545 20 18 3.715 .0016 sig sig
Sequentially Rejective Procedure (SRP)

A criticism of the Bonferroni and Modified Bonferroni has been that neither takes into account very small
and “not-even-remotely-close-to-being-significant" differences. In the example above, the exact p-value of
the comparison between the 20-dBL and 40-dBL groups (p = .4474) is not even close to .05! In such
cases, the difference between groups is so small that it will never be statistically significant, even at α =
-333-
Back to TOC
.05. This is an important point, because if these differences that are not significant at α = .05, then these
differences will never be significant and there is no chance of making a Type I error on them! So, the
question has been why these “never going to be significant" differences should be allowed to inflate the
Type I error rate. Put differently, should t-Tests that are never going to be statistically significant on their
own inflate Type I error rate? The answer, of course, is no.
There are two methods to “factor out” the influence of these “never going to be significant" differences so
they do not inflate Type I error rate. I won’t get into the theory behind them, just the procedures. Both
procedures allow you to change αi depending on what ordinal number comparison you are making (e.g.,
th
the c out of c comparisons). Thus, αi will not be the same for every comparison: It changes as the
number of comparisons you make increases.
The Sequentially Rejective Procedure (SRP), which as its name implies, involves rejecting or failing to
reject the null hypothesis for individual t-Tests one at a time and in a sequential order. This procedure
starts out using a conservative (small) alpha-level (αi) to assess the largest difference (largest obtained t-
value). As more comparisons are made the alpha-level (αi) becomes less conservative.
To use SRP: first perform all of the t-Test desired and obtain exact p-values for each test (from SPSS or
another statistical software). Second, rank-order the comparisons based on the exact p-values of the t-
Tests, from smallest to largest, which is shown in Table 3 below. Third, begin calculating the critical αi for
each individual t-Test. Table 3 lists a column ‘i’, which refers to the rank-ordered number of each
comparison, where i = 1 is the first comparison, i = 2 is the second, etc. The alpha-level (αi) for each
comparison in SRP is calculated from the following expression:
αi = αFW/(c - i + 1)
Where αFW is your chosen Familywise error rate, ‘c’ is the total number of comparisons being made, and
‘i’ is the ordinal comparison of the current t-Test. Table 3 lists the value of (c - i + 1) for each comparison
and the corresponding alpha-level (αi) for each comparison.
Result of
Result of Result
Table 3 Modified
Bonferroni of SRP
Bonferroni
i Comparison p c-i+1 αi
1 40-dBL vs. 80-dBL .0001 10 .0050 sig sig sig
3 Isolation vs. 40-dBL .0009 8 .0062 sig sig sig
5 40-dBL vs. 60-dBL .0082 6 .0083 ns ns sig
6 Isolation vs. 80-dBL .0098 5 .0100 ns ns sig
7 Isolation vs. 20-dBL .0632 4 .0125 ns ns ns-STOP
8 20-dBL vs. 60-dBL .2168 3 .0167 ns ns ---
9 Isolation vs. 60-dBL .3794 2 .0250 ns ns ---
10 20-dBL vs. 40-dBL .4474 1 .0500 ns ns ---
The final step of SRP is to determine whether each t-Test is statistically significant. Start with the
comparison that has the smallest exact p-value (40-dBL vs. 80-dBL, p = .0001) and calculate the alpha-
level for the first comparison, which is αi = .0050. Compare this αi value to the exact p-value of that t-Test.
If αi > p-value the difference is statistically significant and you go on to the next comparison; that is, the t-
Test with the next largest exact p-value. Recalculate αi for this next t-Test, which is αi =.0056, and
compare that new αi to the exact p-value of that next t-Test. If that difference is significant, repeat the
procedure.
-334-
Back to TOC
Basically, with SRP, you continue calculating a new αi for each successive comparison until you come to
a t-Test that is found to be non-significant. In this example, the first t-Test that would be declared not
statistically significant using the SRP procedure is the Isolation vs. 20-dBL (i = 7, where .0632 > .0125).
When this first non-significant t-Test is found, all t-Tests that would come after that t-Test cannot possibly
be significant and you stop evaluating t-Tests. That is, once an exact p-value is found to be non-
significant, all exact p-values that are greater than that one are necessarily non-significant.
SRP it is less conservative and more powerful than the Bonferroni and Modified Bonferroni. Indeed,
notice that Table 3 shows how two comparisons that are significant by SRP were not significant by the
Bonferroni and Modified Bonferroni corrections. However, SRP is not without fault. The problem is that it
penalizes large “highly-significant” differences. For example, in Table 3 the first two or three comparisons
have very small p-Values corresponding to large t-Values. These comparisons are going to be significant
in almost any comparison; hence, there is, at best a remote chance that rejecting the null hypothesis for
those t-Tests would result in a Type I error. Thus, by starting at the top with the smallest exact p-values,
SRP penalizes you for these large and highly significant differences, just like the Bonferroni and Modified
Bonferroni penalize non-significant comparisons. This is where the false discovery rate procedure comes.
False Discovery Rate

False Discovery Rate (FDR) is by far the best method for determining the statistically significance of
individual t-Tests after a statistically significant analysis of variance; that is to say that it is the least
conservative and powerful post-hoc analysis procedure. The procedure is similar to SRP where a new α i
is calculated for each t-Test that is being evaluated. Indeed, on the surface FDR is SRP in reverse;
however, you are much more likely to call a t-test significant using FDR then using SRP.
The first step in FDR is to calculate all necessary t-Tests and to identify the exact p-value of each test,
just like with SRP. Second, like with SRP in FDR the t-Tests must be rank-ordered by exact p-values.
This is shown in Table 4, below. At this point, nothing is different from SRP.
Next a new and different αi is calculated for each t-Test that is compared; however, using FDR the
calculation of αi for each t-Test is different than when using SRP. Also, with FDR the first α i is calculated
for the t-Test with the largest exact p-value, rather than for the smallest exact p-value as in SRP. Hence,
with FDR you start at the bottom of the rank-ordered comparisons and work up. The αi for each
successive t-test is calculated in the equation:
αi = (i/c)αFW
Where αFW is the Familywise error rate you want to keep, c is the total number of comparisons that you
can make, and i is the comparison currently being made. Table 4 shows the value of (i/c) for each
comparison and the corresponding αi.
Result of
Result of Result Result
Table 4 Modified
Bonferroni of SRP of FDR
Bonferroni
i Comparison p (i/c) αi
1 40-dBL vs. 80-dBL .0001 0.100 .0050 sig sig sig sig
3 Isolation vs. 40-dBL .0009 0.300 .0150 sig sig sig sig
5 40-dBL vs. 60-dBL .0082 0.500 .0250 ns ns sig sig
6 Isolation vs. 80-dBL .0098 0.600 .0300 ns ns sig sig & UP
7 Isolation vs. 20-dBL .0632 0.700 .0350 ns ns ns-STOP ns
8 20-dBL vs. 60-dBL .2168 0.800 .0400 ns ns ns ns
9 Isolation vs. 60-dBL .3794 0.900 .0450 ns ns ns ns
10 20-dBL vs. 40-dBL .4474 1.000 .0500 ns ns ns ns
-335-
Back to TOC
The final step in FDR is to determine whether each t-Test is statistically significant by comparing each αi
with the exact p-value of each comparison. Start with the t-Test with largest exact p-value (20-dBL vs. 40-
dBL, p = .4474) and calculate αi, which is .0500. Compare that αi to the exact p-value. If αi < p-value, then
the comparison is not statistically significant and the t-Test with the next smallest exact p-value is
compared to its calculated αi which is .0450. If that comparison is not statistically significant repeat the
same procedure. That is, continue to calculate αi for each successive t-Test until you reach one that is
significant. In this example, the first comparison that would be labeled significant is 40-dBL vs. 20-dBL (i =
6, where .0098 < .03). Once the first significant t-Test is found, it can be concluded that all t-Tests above
that point are necessarily significant.
What’s nice about FDR is that with the exception of the i = 1 comparison and the i = 10 comparison (or
whatever the last comparison number is), αi for each comparison are a little larger in FDR compared to
SRP. Thus, FDR has more statistical power than SRP and you are more likely to reject the null
hypothesis for any comparison.
One final comment to note about both the SRP and the FDR procedure: you do not have to calculate the
αi for each comparison as I did in Tables 3 and 4. You only need to do this until you identify the first non-
significant comparison, as in the SRP, or identify the first significant comparison, as in the FDR
procedure. Once that comparison has been identified, you’re done.
-336-
Back to TOC
Appendix E: Answers to Homework Questions
Note: Answers that require verbal explanation or that must be presented in your own words are not
provided.
Chapter 1
1. a. Constant b. Variable c. Constant d. Constant e. Variable f. Variable
3. a. quantitative b. qualitative c. quantitative d. quantitative e. qualitative f. qualitative

g. qualitative h. qualitative i. quantitative
11. A dependent variable is measured and an independent variable differs between situations and can be
naturally occurring or manipulated.
13. Independent variable: Dosage of drug, quantitative

Dependent variable: Brain activity, quantitative
15. Independent Variable: Whether subjects believed a rat was maze-bright or maze-dull, qualitative
Dependent Variable: Number of times the rat entered the correct section of the maze, quantitative
23. a. All undergraduates in the US

b. The 100 undergraduates who rated the potato-chips
c. Measurement, or raw data, or datum (any one is acceptable)
d. Descriptive statistic or a sample statistic (either is acceptable)
25. a. All statistic students.

b. The students in Dr. Smith's three statistics classes.
c. Measurement, or raw data, or datum (any one is acceptable)
d. Descriptive statistic or a sample statistic (either is acceptable)
e. Independent variable.
f. Ratio
g. Inferential statistics
Chapter 2
1. a. 21 b. -9 c. 81 d. 81 e. 441 f. 219 g. 90 h. -9
3. a. 1.234 b. 2.424 c. 17.167 d. 311.420 e. 3.680 f. 12.346

g. 66.730 h. 54.147
5. a. URL = 12.65, LRL = 12.55 b. URL = 22.585, LRL = 22.575

c. URL = 10.05, LRL = 9.95 d. URL = 49.5, LRL = 48.5
7. Original scores: Rounded scores:

a. 11.62 a. 11.62
b. 134.98 b. 135.02
c. 34.50 c. 34.51
9. a. Ratio b. Nominal c. Ordinal d. Interval e. Interval f. Ratio
11. a. Ratio b. Nominal c. Ordinal d. Ratio e. Interval f. Ratio

g. Ordinal h. Ordinal i. Nominal
13. a. continuous b. discrete c. continuous d. discrete
15. a. 9.15 b. 83.722 c. 23.496
-337-
Back to TOC
Chapter 3
1. a. X f rf cf crf c% b. .28 c. .68 d. .4 e. .32
5 5 .2 25 1.00 100% f. .6 g. 3 h. 20 i. 8
4 3 .12 20 .8 80% j. 17 k. .2 l. .32
3 7 .28 17 .68 68%
2 5 .2 10 .4 40%
1 3 .12 5 .2 20%
0 2 .08 2 .08 8%
3. a Number of Days Sick f rf % cf crf b. .175 c. .800 d. .375

9 2 .050 5% 40 1.000 e. .125 f. .050 + .100 = .150
8 4 .100 10% 38 .950
7 2 .050 5% 34 .850
6 7 .175 17.5% 32 .800
5 5 .125 12.5% 25 .625
4 5 .125 12.5% 20 .500
3 5 .125 12.5% 15 .375
2 4 .100 10% 10 .250
1 2 .050 5% 6 .150
0 4 .100 10% 4 .100
5. a. Major f rf % b. .129 c. .106 d. .024

Biology 13 .153 15.3% e. .047 f. .024 + .024 = .048
Exercise Science 12 .141 14.1%
Communication 7 .082 8.2%
Undecided 9 .106 10.6%
Counseling 4 .047 4.7%
Occupational Therapy 3 .035 3.5%
Education 11 .129 12.9%
Computer Science 2 .024 2.4%
Business 8 .094 9.4%
History 2 .024 2.4%
Philosophy 2 .024 2.4%
Nursing 4 .047 4.7%
Psychology 2 .024 2.4%
Criminal Justice 6 .070 7.0%
9.
Frequency
11. a. Nominal scale b. Bar Graph c. (See graph below)
-338-
Back to TOC
Back to TOC
17.
15.
13.
Cumulative Frequency Frequency Frequency
-339-
Score
Score
9
9
8
8
7
6
6
6
5
5
9 5 9
8 4 9
8 3 8
6 3 8
5 3 7
9 4 2 7 9
8 3 2 3 8
4 2 1 1 7
3 2 0 1 2
0 0 0 0 0
8 9 10 11 12
19. a. 149 b. 109 c. 1 d. Yes, agree.
Chapter 4
1. a. 1.55 b. 2.714 c. 2.893 d. 3.65 e. 3.9 f. 5.375
3. a. 34% b. 58% c. 92% d. 16%
5. IQR = 5.294
Chapter 5
1. Mode, median, and mean
3. 5
7. Mode = -1 Median = 0 Mean = -0.2
9.
11
10
9
8
7
6
5
4
3
2
1
11. a. No single mode. b. 93.5 c. 92.8 d. Negatively skewed
13. a. 9 b. 8.5 c. 9 d. Negatively skewed
-340-
Back to TOC
15. 48.28
19. The mean is a poorer measure of central tendency for Set A
21. Negatively skewed
Chapter 6
1. Sum of Squares, Variance, Standard Deviation,
5. 49
7. a. 6.5 b. 28.5 c. 2.85 d. 1.688
9. The sample mean and range are the same. The sums of squares, sample variance, and sample
standard deviation differ.
11. a. 1125.6 b. 112.56 c. 10.609 d.125.067 e. 11.183
13. a. 22 b. 2.2 c. 1.483
15. a. 16 b. 88 c. 8.8 d. 2.966
19. President = 1.673 Congress = 1.265
21. Ratings are more consistent for Congress
23. Mean = 26 SS = 424 Variance = 42.4 Standard Deviation = 6.512
25. Company A Mean: 4 Company A Standard Deviation: 1.414

Company B Mean: 4 Company B Standard Deviation: 0
27. Student A
Chapter 7
1. a. 0.240 b. -1.098 c. -3.658 d. 0 e. -1.013 f. 1.166
3. a. 16.75 b. 7 c. 7.75 d. 10
5. -1.225
7. Mean = 0, Standard Deviation = 1
9. a. 0.9750 b. 0.0250 c. 0.9505 d. 0.0495 e. 0.9162 f. 0.9162

g. 0.4772 h. 0.4772 i. 0.3433 j. 0.6541 k. 0.7850
11. a. 141.250 b. 100 c. 70 d. 77.5 e. 118.75 f. 110.5
13. a. 34300 b. 42900 c. 49600 d. 53300 e. 60800
15 a. 0.3707 b. 0.9525 c. 0.0475 d. 0.8413 e. 0.6293 f. 0.8164

g. 0.2120 h. 0.2286
Chapter 8
-341-
Back to TOC
1. Difference between a sample statistic and a population parameter.
3. Sample mean is unbiased.
5. Number of scores that can vary in a sample.
7. Mean, standard deviation, and the shape of the sampling distribution.
9. The standard deviation of a sampling distribution of the mean.
11. The estimated standard error of the mean
13. The means are equal to the population mean.
15. ̂ ̂
17. ̂ ̂
19. 2.006
21.
23. a. 8 b. 6 c. 1.5 d. 1.225 e. 0.894

f. g. 0.548 h.
25. Sample from population A will have a mean that is a better estimate of the population mean.
27. 0.447
29. ̅ Not a good representative.
31. ̅ Not a good representative.
Chapter 9
1. Sampling with replacement means returning an individual to a population and sampling without
replacement means an individual is not returned to a population.
3. a. 0.067 b. 0.167 c. 0.241 d. 0.310 e. 0.464
5. 0.400
7. 0.240
9. 0.060
11. 0.150
13. 0.100
15. 0.220
17. 0.360
19. 0.580
-342-
Back to TOC
21. 0.400
23. 0.200
25. 0.552
27. 0.080
29. 0.800
31. 0.900
33. Yes
35. 0.500
37. 0.428
39. 0.400
41. 1.000
43. 0.348
45. 0.084
47. 0.228
49. 0.032
51. 0.600
53. 720
55. 10
57. 210
Chapter 10
1. a. 0.004 b. 0.112 c. 0.269 d. 0.032
3. Mean = 1670, Variance = 1391.110, Standard deviation = 37.298
5. Mean = 90, Variance = 36, Standard deviation = 6
7. Mean = 160, Variance = 32, Standard deviation = 5.657
9. p = 0.060, the student did not perform better than chance.
11. p = .123
13. a. 0.636 b. 54.4 c. 34.726 d. 5.893 e. z = 2.478, p = .0066
15. Mean = 20, Standard deviation = 4, z = 2.5, p = .0062
-343-
Back to TOC
Chapter 11
3. Due to sampling error
5. a. 0.5 b. -3 c. .0027 d. Significant, Reject H0, accept H1.
7. a. H0: μ = 100 H1: μ > 100 b. 3 c. 1.333 d. .0918

e. Not significant, retain H0, no decision on H1.
9. a. H0: μ = 13 H1: μ > 13 b. 0.949 c. 0.823 d. .2061

e. Not significant, retain H0, no decision on H1. f. 0.3 g. 2.667
h. .0038 i. Yes. Significant, Reject H0, accept H1.
11. P(Type 1 Error) = α
13. As alpha increases, power decreases.
Chapter 12
1. When the parameters are unknown.
3. a. .0600 b. .0181 c. .0194

d..00635 (.0064) e. .0482 f. .02105 (.0210)
5. ̂ ̅ p = .22215 (.2222) Not significant, retain H0, no decision on H1.
7. ̂ ̅ p = .0484 Significant, reject H0, accept H1.
9. a. H0: μ = 9 H1: μ > 9 b. 1 c. 2.5 d. p. .01695 (.0170)

e. Significant, reject H0, accept H1.
Chapter 13
1. Experimental strategy involves manipulating an independent variable and the observational strategy
involves measuring values that naturally exist in research participants.
3. IV: Noise condition (between-subjects). DV: Number of correct solutions.
5. A group not exposed to an experimental manipulation.
7. Variables that change as the levels of the independent variable change.
Chapter 14
3. a. .0096 b. .0223 c. .0214 d. .0439 e. .0442 f. 00735 (.0074)
5. ̂ ̂̅ ̅
7. a. 15.53 b. 0.577 c. -17.953 d. .0001 e. Significant.
9. a. H0: μAudio = μNo Audio H1: μAudio ≠ μNo Audio b. 1.875 c. 0.387 d. 2.067
e. 0.450 f. Significant. Reject H0, accept H1.
11. a. 2.292 b. 0.307 c. -2.280 d. .0131

e. Significant. Reject H0, accept H1.
13. a. MM = 2.8, MP = 1, SSM = 1.3, SSP = 0.5 b. 0.225 c. 0.3

d. 6 e. .0039 f. Significant. Reject H0, accept H1.
-344-
Back to TOC
15. a. H0: μAlcholo = μPlacebo H1: μAlcohol > μPlacebo b. MAlcohol = 2.5 MPlacebo = 1.75
c. SSAlcohol = 0.128 SSPlacebo = 0.804 d. 0.066
e. 0.126 f. 5.952 g. .00065 (.0006)
17. a..189 b. .860 c. .640 d. .798
19. t(8) = 1.09, SE = 2.74, p = .1694 (two-tails)
21. t(8) = 6.00, SE = 0.30, p < .0039 (two-tails)
Chapter 15
1.
Y
5. Direction
9. a. .0186 b. .0093 c. .0442 d. .0221
11. a. SSX = 35 SSY = 22 SCP = 22 b. ̂ ̂ c. 2

2 2
d. 0.793 e. r = 0.629 1-r =0.371 f. 4.13 g. .0025\
h. Significant. Reject H0, accept H1. i. 0.981
13. a. H0: ρ = 0, H1: ρ ≠ 0 b. -3.221 c. .0013

d. Significant. Reject H0, accept H1. e. 0.751
15. a. H0: ρ = 0, H1: ρ < 0 b. -3.462 c. .00065 (.0006)

d. Significant. Reject H0, accept H1.
2
17. a. ∑D = 0 ∑D = 90 b. 0.455 c. 1.445
d. .1991 e. Not significant. Retain H0, no decision on H1.
19. a. r' = 1.293, ρ' = 0.255 b. 0.242 c. 4.289

d. .0001 e. Significant.
21. a. H0: ρ = -.50, H1: ρ > -.50 b. r' = -0.100, ρ' = -0.549 c. 0.102
d. 4.402 e. Significant. Reject H0, accept H1.
23. a. 0.395 b. 0.999 d. 0.637 e. 0.518
Chapter 16
3. Number of units Y is equal to when X = 0.
-345-
Back to TOC
5. If X changes by 1 unit, Y will change by 2.5 units. If X changes by 3 units, Y will change by 7.5 units. If
X changes by 8 unit, Y will change by 20 units.
7. a. -0.878 b. 8.390 c. Y’ = 8.39 – 0.878X d. B = 0.488, H = 5.756, E = 3.122

e. SS Regression = 69.372, SSResidual = 12.628 f. 1.256
9. The regression equation in Exercise 7 is a better predictor.
11. a. SS Regression = 74.34, SSResidual = 9.66 b. dfRegression = 1, dfResidual = 8, dfTotal = 9

c. MSRegression = 74.34, MSResidual = 1.208 d. 61.54
e. .0133 f. Yes, significant.
g. 1.099 h. SE(b1) = 0.155, SE(b0) = 0.712
i. t = 7.871, p = .0039 j. t = -2.949, p = .0184
Chapter 17
3. a. .0339 b. .01435 (or .0144) c. .0377 d. .00495 (or .0050)
5. When the correlation is 0.
7. 0.689
9. a. H0: μPre = μPost; H1: μPre < μPost

b. ̂ , ̂ , ̂ , ̂
c. 1.591
d. 0.628
e. .28165 (or .2816)
f. Not significant Retain H0 and make no decision on H1.
g. .042, small
h. .209, small
i. .05
11. a. 34 b. 1084 c. 21 d. 242
Chapter 18
1. The population means are not equal.
3. Differences between means.
7. Zero (0)
9. a. .1313 b. .1016 c. .0388 d. .0036 e. .0258 f. .0135
11. SSB = 54, SSW = 102.86, SST = 156.86, dfW = 20, MSW = 5.143
13. SSB = 80, SSW = 48, dfW = 12, dfT = 14, MSB =40
15. 0.15
17. 018
19. 0. 007
-346-
Back to TOC
Chapter 19
1. a. MEasy = 7, MMod = 5, MHigh = 3, MNo = 5, G = 5
b. dfT = 19, dfB = 3, dfW = 16
c. SSW = 22, SST = 62, SSB = 40
d. MSB = 13.333, MSW = 1.375
e. 9.697
f. .0007
g. Significant. Reject H0 and accept H1.
h. HSD = 2.122, Significant difference between Low Difficulty and High Difficulty.
2
i. η = 0.645, f = 1.348. Large effect.
j. .997
3. a. 111 b. 177 c. 396 d. 280
5. a. MA = 7, MB = 9, MC = 9, MD = 15, G = 10
b. dfT = 23, dfB = 3, dfW = 21
c. SSW = 94, SST = 310, SSB = 216
d. MSB = 72, MSW = 4.476
e. 16.086
f. .0003
g. Significant. Reject H0 and accept H1.
h. HSD = 4.337, Significant difference between A and D, B and D, C and D.
2
i. η = 0.697, f = 1.516. Large effect.
j. .999
Chapter 20
1. Includes two or more independent variables
3. 9 groups, 2 independent variables
5. 16 groups; 4 independent variables
7. a. Yes b. Yes c. Yes
9. a. Yes b. Yes c. No
11.
Between 117 3 39 3.9
Main Effect of X 100 1 100 10
Main Effect of Y 2 1 2 0.5
Interaction (X x Y) 15 1 15 1.5
Within Groups 360 36 10 ---
Total 477 39 --- ---
Between 151 3 50.333 2.013
Variable X 1 1 1 0.04
Variable Y 50 1 50 2
Interaction 100 1 100 4
Within 1900 76 25 ---
Total 2051 79 --- ---
13. a. Main Effect of X: p = .0036 Main Effect of Y: p = .3253 Interaction: p = .2302

b. Main Effect of X: p = .3213 Main Effect of Y: p = .1625 Interaction: p = .500
-347-
Back to TOC
15. a. MJ1/K1 = 9, MJ1/K2 = 3, MJ2/k1 = 2, MJ2/K2 = 8, MJ1 = 6, MJ2 = 5, MK1 = 5.5, MK2 = 5.5, G = 5.5
b. dfT = 11, dfW = 8, dfB = 3, dfJ = 1, dfK = 1, dfJxK = = 1
c. SST = 119, SSW = 8, SSB = 111, SSJ = 3, SSK = 0, SSJxK = 108,
d. MSB = 37, MSWithin = 1, MSJ = 3, MSK = 0, MSInteraction = 108
e. FJ = 3, FK = 0, FInteraction = 108
f. pJ = .1215, pK = .3466, pInteraction = .0133
g. Neither main effect is statistically significant; however, the interaction is statistically significant.
Chapter 21
3. Observed frequencies are the actual numbers. Expected frequencies are expected under a null
hypothesis.
5. a. .0339 b. .1069 c. .0514 d. .1091

e. .0614 f. .0174 g. .0937
7. a. 10 b. 347.4 c. .0000 (or < .0001) d. Significant. Reject H0, accept H1.
9. a. Democrat-City = 104, Democrat-Suburbs = 104, Democrat-Country = 52,

Republican-City = 56, Republican-Suburbs = 56, Republican-Country = 23,
Independent-City = 40, Independent-Suburbs = 40, Independent-Country = 20
b. 26.374 c. .0000 (or < .0001) d. Significant. Reject H0, accept H1.
e. 0.224 f. 0.230 g. 0.698
11. a. Natural Sciences = 20, Business School = 25, Social Sciences = 35, Arts and Humanities = 20
b. 27.478 c. .0000 (or < .0001) d. Significant. Reject H0, accept H1.
13. a. 15.425 b. .0001 c. Significant. Reject H0, accept H1. d. 0.507
15. a. 964 b. 241 c. 1545 d. 299
Chapter 22
1. d = 0.73
3. ̂ , ̂ , ̂ , d = 0.412
5. f = 0.345 or
7. d = 0.307
9. C = 0.224, w = 0.230
11. 1144
13. 132
15. 6940
17. 180
19. 44
21. 353
23. f = 0.366, n = 62
-348-
Back to TOC
References
Bem, D. J. (2011). Feeling the Future: Experimental evidence for anomalous retroactive influences on
cognition and affect. Journal of Personality and Social Psychology, 100, 407-425.
Carter, T. J., Ferguson, M. J., & Hassin, R. R. (2011). A single exposure to the American flag shifts
support toward Reublicanism up to 8 months later. Psychological Science, 22, 1011-1018.
Cohen, J. (1992). A power primer. Psychological Bulletin, 112, 155-159.
Dunlap, W. P., Cortina, J. M., Vaslow, J. B., & Burke, M. J. (1996). Meta-analysis of experiments with
matched groups or repeated measures designs. Psychological Methods, 1, 170-177.
French, C. (2012, March 15). Precognition studies and the curse of the failed replications. The Guardian.
Retrieved from http://www.guardian.co.uk/science/2012/mar/15/precognition-studies-curse-failed-
replications
Kay, A. C., Moscovitch, D. A., & Laurin, K. (2010). Randomness, attributions of arousal, and belief in God.
Psychological Science, 21, 216-218.
Liptak, A. (2012, June 7). Approval rating for justices hits just 44% in new poll. The New York Times.
[Website] Retrieved from http://www.nytimes.com/2012/06/08/us/politics/44-percent-of-
americans-approve-of-supreme-court-in-new-poll.html?pagewanted=all
Newport, F. (2009, February 11). On Darwin’s birthday, only 4 in 10 believe in evolution. Retrieved from
http://www.gallup.com/poll/114544/Darwin-Birthday-Believe-Evolution.aspx
Ritchie SJ, Wiseman R, French CC (2012) Failing the Future: Three Unsuccessful Attempts to Replicate
Bem's ‘Retroactive Facilitation of Recall’ Effect. PLoS ONE 7, e33423.
doi:10.1371/journal.pone.0033423
Rosnow, R. L., & Rosenthal, R. (1996). Computing contrasts, effect sizes, and counternulls on other
people’s published data: General procedures for research consumers. Psychological Methods, 1,
331-340.
Rosnow, R. L., Rosenthal, R., & Rubin, D. B. (2000). Contrasts and correlations in effect-size estimation.
Psychological Science, 11, 446-453.
Shah, J., & Christopher, N. (2002), Can shoe size predict penile length? BJU International, 90, 586–587.
doi: 10.1046/j.1464-410X.2002.02974.x
Tybur, J. M., Bryan, A. D., Mangan, R. E., & Caldwell Hooper, A. E. (2011). Smells like safe sex: Olfactory
pathogen primes increase intentions to use condoms. Psychological Science, 22, 478-480.
The Following were used as References for Statistical Formulas

nd
Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2 Ed.). Hillsdale, NJ: Lawrence
Erlbaum.
th
Hays, W. L. (1994). Statistics (5 Ed.). Belmont, CA: Thompson-Wadsworth.
th
Heiman, G. W. (2006). Basic statistics for the behavioral sciences (5 Ed.).Boston, Houghton Mifflin.
rd
Keppel, G. (1991). Design and analysis: A researcher’s handbook (3 Ed.). Upper Saddle River, NJ:
Prentice Hall.
-349-
Back to TOC
th
Jaccard, J., & Becker, M. A. (2010). Statistics for the behavioral sciences (5 Ed.). Belmont, CA:
Wadsworth Cengage.
nd
Myers, J. L., & Well, A. D. (2003). Research design and statistical analysis (2 Ed.). Mahwah, NJ:
Lawrence Erlbaum.
rd
Rosenthal, R., & Rosnow, R. L. (2008). Essentials of behavioral research: Methods and data analysis (3
Ed.). Boston: McGraw-Hill.
Timhane, A. C., & Dunlop, D. D. (2000). Statistics and data analysis: From the elementary to the
intermediate. Upper Saddle River, NJ: Prentice Hall.
-350-
Back to TOC

Fundamental Statistics For The Behavioral Sciences v.1.9 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Fundamental Statistics For The Behavioral Sciences v.1.9 PDF

Uploaded by

Copyright:

Available Formats

-1-

Fundamental Statistics for the Behavioral Sciences (v. 1.9)

Copyright 2013 Bryan R. Burnham

This work is licensed under the Creative Commons Attribution-NonCommercial 3.0

Dr. Bryan R. Burnham received his B.A. from Utica

You'll be surprised at how much Dr. Burnham looks like

I would like to acknowledge my statistics and research methodology professors who

1.1 You’ve Decided to take Statistics

“The class is horrible, boring, and stupid!”

1.2 What, When, Where, How, and Why of Statistics

For example: I hypothesize that age is related to

2. Statistics are almost always used to explain topics in upper-level

1.3 Statistics as a Language

Statistics is a language course more than it is a mathematics course.

Dr. Burnham’s Second Law of Statistics:

Repeat the First Law.

Dr. Burnham’s Third Law of Statistics:

1.4 Two Types of Statistics

1.6 Values vs. Names of Variables

1.7 Quantity vs. Quality

Overall, what is your general political attitude?

1.8 Hypotheses and Predictions

1.9 Populations and Samples

Population Sample Population

1.11 Random Selection

Next, using a random number generator (http://www.randomizer.org/), I generate a set of 10 random

1.12 Random Assignment

=1 =2 =3 =4 =5 =6

1.13 “Law” of Large Numbers

There is something to be said for strength in numbers;

Original Five Additional Five

1.14 Representative of Whom?

1.15 Research Designs and Statistics

1.16 Computers and Statistics

a. The number of days in a week.

a. An individual's attitude toward abortion at a specific point in time.

3. Identify each of the following as a qualitative or a quantitative variable:

4. Identify each of the following as a qualitative or a quantitative variable.

5. Define each of the following, in your own words.

7. How is a sample used to make conclusions about the population?

9. Answer each of the following, in your own words.

a. What are descriptive statistics used for?

10. Answer each of the following, in your own words.

a. What is the difference between a statistic and a parameter?

11. Distinguish between a dependent variable and an independent variable.

12. Distinguish between a between-subjects independent variable and a within-subjects independent

17. What is the main purpose of random selection?

18. What is the main purpose of random assignment? When is it used?

a. The population in this scenario is...

a. The population is...

a. The population in this scenario is...

I should take a moment to mention something that I

2.2 Whole, Rational, Irrational, and Real Numbers

2.3 Basic Mathematical Operators

9 + 6 + 5 = 20 is the same as 5 + 6 + 9 = 20 is the same as 9 + 5 + 6 = 20

Hence, adding a negative value is the same as subtracting a positive value:

Also remember that subtracting a negative is the same as adding a positive:

5 x 3 = 15 is the same as 5 * 3 = 15 is the same as (5)(3) = 15

10/2 = 5 but 2/10 = 1/5 or .20

2.4 Order of Operations

Word in Phrase Operator What to Do?

2.5 Rounding Values

Unless noted round to three places at each step in any procedure.

ΣXY = (105) + (95) + (84) + (43) + (3*2) = 50 + 45 + 32 + 24 + 12 + 6 = 169