Professional Documents
Culture Documents
Subject Psychology: Paper No. 2 Quantitative Methods Module No. 27-Analysis of Co-Variance (ANCOVA)
Subject Psychology: Paper No. 2 Quantitative Methods Module No. 27-Analysis of Co-Variance (ANCOVA)
Subject Psychology: Paper No. 2 Quantitative Methods Module No. 27-Analysis of Co-Variance (ANCOVA)
Subject PSYCHOLOGY
TABLE OF CONTENTS
1. Learning Outcomes
2. Introduction
3. The concept of ANCOVA
3.1 Defining ANCOVA
3.2 Use of ANCOVA
3.3 Process of ANCOVA
3.4 Choosing Covariates
3.4 Assumptions of ANCOVA
4. ANCOVA and its applications
5. Comparing ANOVA and ANCOVA
6. Using IBM SPSS20 for ANCOVA
7. Summary
1. Learning Outcomes
After studying this module, you shall be able to
Comprehend terms like variation, co-variation, confounding and learn their use in
research investigations.
Evaluate the difference between ANOVA and ANCOVA.
Understand the assumptions to be taken care of for ANCOVA.
Apply ANCOVA further study practically.
2. Introduction
The first time Indian audience saw our own national superhero Shaktimaan in 1997, it was
considered a phenomenon. The belief in a superhero reached to such an extent that children
jumped from the roofs of their homes thinking Shaktimaan would save them. At the same time it
was a new sort of entertainment. As children we might remember glued to television to watch his
stunts. Some of us who had watched it earlier if have a look at it today or those who were born
after the Shaktimaan era would experience it to be funny and poor quality technology usage.
Considering newer present generation movies like Krrish 3 or Avengers, we have no doubt
moved on and the present ones seem more attractive. If one is asked to rate the best of all, it
might not be easy considering the first time experience that one had with respect to that particular
age of ours in which we enjoyed this new phenomenon making entry in India. In order to really
understand which one was the best, thus, one will have to take into consideration the
entertainment and fun attributable to one’s age at that point of time. This will provide a purer
measure of how much variance in the entertainment is attributable to the event itself. ANCOVA
provides for a measure to factor in other variables that may be influencing the outcome variable.
The important point to notice here is that many a times while investigating a problem the
researcher tends to study the independent and dependent variable relationship, ignoring the
underlying dependent covariates that may intervene in the relationship. This intervention of
dependent co-variate contaminates the relationship and produces results which may not be pure
and thus, fraught with errors. It is therefore, important that a statistical procedure be developed
that helps in identifying the variance contributed by the covariate, thus, helping the researcher in
arriving at a much purer understanding of the relationship between the dependent and the
independent variables.
The quantitative methods of research focus on establishing and analyzing the relationships
between the independent and dependent variables in the data. The data varies in the degree of
complexity with the types of research designs chosen. The experimental and correlational designs
focus on the idea of analyzing one dependent variable at a time. The data with different levels of
However, this may not be the case in all situations as, many a times several dependent variables
may act in tandem. When such variables are continuous in their impact on the dependent variable
they are referred to as co-variates. The ANCOVA is used to study the differences between the
different levels of an independent variable for the possible effects of the co-variates on the results
Let us take an example, it is hypothesized that mnemonic technique used will have an
effect on the recall. There are many variables like age, gender, training experience,
reasoning ability etc. that may affect the link between the mnemonic technique and recall.
The ANCOVA allows to study the possible effect of these continuous variables on the
results. By studying thrsrs one can get purer results. Which can throw better light on the
relationship between the independent and the dependent variable, so, ANCOVA basically
extends the idea of ANOVA but takes into account the variance contributed by the
covariates that influence the outcome variable. So, when we measure the co-variates and
include them in the analysis it is called ANALYSIS OF CO-VARIANCE or ANCOVA in
short.
3.1 ANCOVA
The regression equation for ANOVA can be extended to include one or more continuous
variables that can predict the outcome or dependent variable. These continuous variables are not a
part of the main experimental manipulation but do impact the dependent variable and are referred
to as covariates. When covariates are included in the analysis of variance, it becomes ANCOVA.
For example, the quality of an individual’s dance performance will not only depend on the
amount of practice but also the age, physical flexibility, interest of the individual and so on. If the
variables apart from duration of practice are considered as covariates and are measured, it
becomes possible to control their influence on dependent variable by including them in the
regression model.
o Reduce within group or error variance by firstly comparing the variability in data that can
be explained by the experiment and variability that cannot be explained by the
experiment. If this unexplained variance (SSR) can be described in terms of other
variables or covariates, error variance gets reduced allowing more accurate
assessment of effect of the independent variable (SSM). Thus, use of 2 or 3
covariates in such studies can be considered, which have relatively low
correlations with each other as each covariate will remove a part of the error
variance from dependent variable.
o ANCOVA helps in eliminating systematic bias, confounds or the unmeasured
variables that vary systematically with experimental manipulation. Systematic
bias implies that groups differ systematically on a key variable that is related to
performance on dependent variable. For instance, if there are treatments involved
in groups, a significant difference on a post test after giving the treatments will be
mixed with initial differences as it would be unclear whether treatments are
making the difference or whether initial differences are simply transferred to
posttest means. For instance, if one is comparing the effect of 4 stress situations
on the blood pressure, the third situation may have been found to be more
significantly stressful than the other 3. At the same time, blood pressure of the
subjects in group 3 under minimal stress is found to be greater than subjects of the
other groups. One now as a researcher needs to find out whether the blood
pressure for group 3 would still be higher if posttest means for all 4 groups were
adjusted in a way to account for initial differences in blood pressure. Later, one
notes that the posttest means are adjusted in a linear manner to what they would
be if all groups started out equally on the covariate i.e. at the grand mean.
The best possible way to eliminate this bias may be random assignment of
subjects and hence, within sampling error, one would be confident that groups
don’t differ systematically on variables. But in many studies, random assignment
may not be possible, so one might partially equate the groups by either matching
on key variables or using ANCOVA to control key variance. ANCOVA reduces
but not eliminates the bias caused by these variables and once identified, the
confounding variable can be measured and entered into the analysis as a
covariate.
Yi` Adjusting
Yi Compute
(Predictable from degrees of
(Each individual ANOVA on
uncontrolled freedom lost
variable Zi) obtained
score) during
(Yi – Yi`)
estimation of
regression
ANOVA SITUATION
Total Variance in task performance
ANCOVA SITUATION
Total Variance in task performance
Applications of ANCOVA:
ANOVA ANCOVA
Tests effect of
Whether common covariate on
mean shared by 3 or dependent
more data groups or variable…
not…
Divides between
Talks about between group variation
group variation in into treatment
general……. effect & covariate
effect
Since the sample is randomly assigned to each treatment conditions, IQ data is not used by the
investigator for the analysis as she doesn’t feel the need because of random sampling. A t-test for
independent samples on achievement is carried out on achievement at the 0.05 level of
significance. The t value is found to be 1.676which according to the t tables is not significant as
the critical value of t is +/- 2.074. But the pooled within correlation of IQ with achievement for
the data is found to be 0.80. this implies that 64% of the variation in achievement test scores is
related to individual differences in IQ.
Here, ANCOVA comes to rescue as it removes that part from the error variance and yields a t
value significant at o.o5 level i.e. t= 2.25.
A researcher wishes to study the effect of substance abuse treatment conditions on the duration of
abstinence when adjustment for the covariate of age is made among the adults. The independent
variable has two substance abuse treatment conditions: standard care and standard care with
contingency management. The sample of 150 adults with problem of substance were randomly
Click Analyze> click> General Linear Model > click > Univariate> click over LDA
under Dependent Variable, Treatment Condition under Fixed Factors, and Age over to
Covariate.
Click on the Plots button > click Treatment Condition to Horizontal Axis > click the Add
button > click Continue.
Click on the Options button > click (overall) and treatment Condition to Display Means
for. Select Compare Main Effects. Select LSD (none).
Click on the boxes for descriptive statistics, Estimates of effect size, observed power, and
Homogeneity tests > the Significance level should be .01> click on Continue> click OK.
Save the generated output as ANCOVA with age results.
7. Summary
The ANCOVA is used to study the differences between the different levels of an
independent variable for the possible effects of the co-variates on the results. When
covariates get included in ANOVA, it becomes ANCOVA.
ANCOVA is a statistical control procedure that yields a purer and accurate evaluation of
the result of an experiment as it reduces biases or errors attributable to factors covarying
with the dependent variable.
ANCOVA as a statistical technique enables to: Reduce within group or error variance by
firstly comparing the variability in data that can be explained by the experiment and
variability. ANCOVA helps in eliminating systematic bias, confounds or the
unmeasured variables that vary systematically with experimental manipulation
The regression equation for ANOVA can be extended to include one or more continuous
variables that can predict the outcome or dependent variable
Its applicability ranges from experimental to observational designs producing more
powerful statistical results.
====================================================================