Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 25

Lecture after Workshop - 29 Jan

Dependent variable – the “problem” you want to explain

Multiple choice exam


- Work in SPSS – with a new data set that we will get in advance, with some new
variables.
- Will be uploaded on Canvas
- Mockup exam – Thursday afternoon, 4-5 hours. Get the result when the test closes.
Will look like the regular exam.
- Will be in classroom, he will check the registration.

Research proposal
- Pass or non pass exam – hard to not pass.
Oral exam
- Will be examined after one another in the group.
- Not super focus on the research design, more use it as an example.
- Example question: Sample – what is the thoughts behind the way you sampled in this
research? What are alternatives? What are the preconditions the we expect from a
sample?  Respect to population, assume that it is a random sample. In some
population, unable to develop a sampling frame (?). Be careful looking into biases in
our sampling. Variation and cluster sampling, sample within a sample.
- Other example: Alternative choices, Given the scales – the items could be similar –
what should you think about that? Understanding of what is the predicted values in a
regression model vs the residuals, what do they signify, what do they mean? Various
kinds of validity tests – the reliability of your scales (Cronbach or p value below 0.05).
- Question a lot based on the last workshop (?).
- 20 minutes – including grade discussion. 5 min presentation, discussion 12 minutes, 3
minutes setting the grade.
- What would we have done better and WHY. Adding to what is already available 
present new information.
The problem (of finding a good problem)
• The tricky “why” question
• Recap: the difference between qualitative and quantitative (variance) studies
• Example: Why are ‘actual’ risks underreported in risk reports

 Causation  the independent variable will cause something to the dependent


variable
 Correlation does not say what is the prediction and the response – there we instead
use logic of the world.

The qualitative approach


• A crude try: simply ask the people reporting
• Or a more sophisticated way: observe their behavior, map their communication networks
and then ask them more elaborate questions (more details) based on your observations

 Why are they not reporting as they are “supposed to” report?

• The point is: the path to answering your “why” question is mediated by what you can
observe through your senses
• You gradually reach your answer through induction (bits and pieces begin to make sense)

The quantitative (variance) approach


• First step: Create a theory though ‘disciplined imagination’ (Weick 1989, AMR 14(4): 516-
31) You try to come up with a plausible answer – abduction
- Based on the premises of “what we know about the world”

• Second step: construct the theory through logical reasoning, defining terms, specifying
relationships etc. Specify/justify your answer based on available knowledge – deduction

• Third step: test/evaluate (a model of) the theory You test if data confirms the hypotheses in
your model - induction

• Alternative theories  reflected in alternative methods  triangulation


 Perceived ability to comply – example: are they to ambitious?

What information do you want and how will results be analyzed?


A typical variance study
- We assume you have a theory/explanation – based on some hypothesis that can
visualized in a model.
- You then ask questions that can confirm or reject your hypothesis.
You should be attentive to two issues:
1. The quality of your questions
2. The way respondents can reply (scales)

Minimum sample: 50 + 8 x number of variables

Double sample size if there are a comparison between different dependent variable
(women/men)

 Do not need to do a questionnaire if we use a preexisting scales


Lecture before workshop - 29 jan
(Missed)

Workshop Theory (for group project) – 24 jan

* Your data should be considered through multiple regression analysis – consider when
creating research question (it should not be “exploratory”)

* See if data behaves as or model predicts


* Deductive logic (?)
* Use both induction (observing empirical events, problem we can be wrong, use data –
inductive way of using the data that we have) deduction (?), abduction (combination, but
also describe how human reason by looking at the things that surprise us and image an
world, making a hypothesis)

My idea:

Research question: Do communicators that work with social media get more easily burnout?
- Problem: social media can have a negative effect on your mental health. If you work
with it a lot it can affect your health. Plus the blurred boundary from personal life
when you are at home on social media, and work research.

H1: Working with sociala media on tour daytime job effects your health.
H2: Communicators that work with social media have a bigger issue seeing the boundary
between personal social and work social.

Theory: Think there is a lot of theory and literature around social media and health effects.

Idea 2 keywords: Political, New parti, Change of perspective, why?


Idea 3 keywords: Student, Part time job, Career, Money
Idea 4 keywords: Political, Strategic Communication action, Voting

Lecture after Workshop – 24 jan

Multivariate analysis 1/4


What can MRA do for us?
• How well a set of variables predict an outcome
• Which variable in a set of variables is the best predictor
• Whether a variable can predict an outcome when controlled for
(Pallant 2010 p. 149)

Multivariate analysis 2/4


Requirements for multiple regression analysis
1. Dependent variable (y) must be quantitative (interval/ratio)
2. Two or more independent variables (x1, x2...) that must be quantitative –
3. - OR binary (dummy variables)
4. Linear relationship between each independent variable and the dependent variable
(linearity)
5. Independent variables cannot be highly correlated (r=.9 and above) (multicollinearity)
6. Residual must have normality, linearity, homoscedasticity (equal or similar variance) and
independence
7. REVISIT PALLANT CH 6!

* To describe as a linear function, the variables must together form somewhat of a line

Lecture before Workshop - 24 jan

Dummy variable
Measure something in the absent of other (?) – will it have an affect on the dependent
variable
- Ex different between male and female
- Use t-test to see if it is significance
- Easy to put in regression analysis, but can only measure absent or not
Next workshop  Regression models
Based on correlation analysis
Large residual – hard to make regression line
Important to look at it graphically

Quantitative variable in three points (ex communicators, manager, coworkers)


How code? Make more dummy variable.  Number of categories minus 1
Don’t take all categories and categories into dummy variables
Remember: measuring something in the absent of something other

Significance
= A significant difference in the regression line we observe
- No slope at all  No affect (no significance)

* Important to algin the scale we use so that it can become comparable


 Standardize the variables

Coefficient
Y= 2x + 5 (2an är koefficiensen)
- Tells us the steepness of the curve  how big the effect of the independent variables
has on the dependent variable

To be able to understand the effect of the independent variables  needs to be standardize


 we will get to know the standardize beta coefficient
* The dummy variables stand for the standardization

The higher the correlation coefficient – the more the model can explain
Also interested in the R-square – saying something about what the variance can explain

What we see in the regression line is if it would be statistically significant (not random or
pure luck)

Some other variable that we did not account for (?)  Cannot use the model as a reliable
explanation

Lecture Workshop 3 – 19 Jan

Outline for scientific project description (research proposal) – 5 pages

 Introduction
- Topic
- Problem statement/research question
- Relevance
- Significance to knowledge/research
- Significance to society/practice
 Literature review
- Previous research
- Highlight important findings and unanswered questions (gaps)
- Relate unanswered questions (and how they inter-connect) to research question(s)
 Methodology
- What is your general approach/research design?
- Which data and how to get them?
- Specific technique(s) used to analyze data?
 Expected results
- Explain how the outcome of your study relates to theory and RQ (restate/sum Up)
 References

* Note: the outline of out project description is almost identical to a standard research paper,
you can change a little bit but should cover these aspects.

* Working with a regression model  need to be able to explain everything in advance


before the step where we get the data.

* Recommended from Henrik: Come up with a model that can be tested with a multiple
regression analysis

Research question= dependent variable, Research questions answer= independent variable


Two main approaches:
 Quantitative (variance) studies: What causes what? (this is the one we will do(?))
 Qualitative (process) studies: How do events happen?

Recommend to read: “Robert K Merton: in science ‘establishing the phenomenon’ is


paramount”

 When finding your problem, ask questions such as:


- For example?
- From whose point of view?
- What is the point of view?
- Avoid pseudo-problems that are not grounded in reality

Three fragments
 Establishing the phenomenon
- Difference between fact and phenomenon
 Specified ignorance
- Building upon what is known as a precondition for identifying what needs to be
known for the specific scientific fields to progress
 Strategic research materials
- Identifying places, materials etc. where you are likely to find empirical facts that
reflects the phenomenon under study
My ideas for Workshop:
- Lack of clear communication from teachers when studying (a Master in Strategic com)
- Problem of reducing screen time – main reasons
- Problem of grocery shopping sustainable when you are a student
- How February and January in Sweden affects study motivation
- Fake news – problem on news channels accurate communication

Idea number 8 – 3 points


Idea number 3 – 2 points
Idea number 9 – 1 point

Lecture after workshop 2 – 19 Jan


Standard error of the mean – measure significance

Bivariate analysis 1/7 (t-test?) - The bivariate analysis allows you to investigate the
relationship between two variables. It is useful to determine whether there is a correlation
between the variables and, if so, how strong the connection is.

Overview
- Studies the relationship between TWO variables.
- Can be a method for analysis in its own right!
- In this course, used as a step towards a multivariate analysis!
1. Qual-qual: cross tables
2. Qual-quant: compare means
3. Quant-quant: correlation
See T-test wishful measurement above

Residual – distance from points to the line

* Consider steepness of the curve – if not steep curve, some positive effect but not so much.
Calculates the effect from or independent variable to dependent variable.

Lecture before workshop 2 – 19 Jan

- Look at what they are measuring

Cronbachs alfa – Use to measure internal consistency (reliability)


- It is most commonly used when you have multiple Likert questions in a
survey/questionnaire that form a scale and you wish to determine if the scale is reliable.

Correlation above 0.7 and maximum 1


Quantiative variables

Residual – the distance between the actual measuring points and where actual regression
line is.
Test for dimensionality – factor analysis – is it more than one dimension (??)

Reliability analysis allows you to study the properties of measurement scales and the items
that compose the scales. The Reliability Analysis procedure calculates a number of commonly
used measures of scale reliability and also provides information about the relationships
between individual items in the scale.
 High Cronbachs alfa close to 1 = good

* Missing variables (learn more about later) – part of factor analysis

Google: “Multiple regression is the prediction of the dependent variable by the independent
variables, while factor analysis is the grouping of variables by their underlying dimensions.”

T- test – empirical alternative to normal distribution


= t-test eller Students t-test är inom statistiken beteckningen på en hypotesprövning där man
vill jämföra om skillnad föreligger mellan två normalfördelade populationer där man inte
känner till det exakta värdet på standardavvikelsen.
T-test - (under Sig. check if the value is below 0,05)  The more under 0.05 the
better/more accurate

Lecture after workshop 1 – 17 Jan

Manipulating variables 1/8


• Reduction of categories
• Indices/indexes
• Reversing scales
• Normalizing variables
• Dummy variables
• Factor analysis
• (for methods, see instructions to workshops) workshop 2 (?)
= Getting a clearer result when comparing the results. Can be done in SPSS by recoding to
new variable.

Indices/indexes 3/8
Sometimes we can ask several questions that has the same underlying overarching question.
For example personality. Or as in the case I found in internet: job-related stress where one
could ask the following questions:
• "When I think about myself and my job, I feel downhearted and blue."
• "When I’m at work, I often get tired for no reason."
• "When I’m at work, I often find myself restless and can’t keep still."
• "When at work, I am more irritable than usual."
 Correlation analysis – is there a correlation and formula? – Will give a more robust
scale/variable in the end.
- Which items are suitable for collapsing into a new scale?

* New variables will have an increase points (5 options + 5 option + 5 options = 15)
* The steepness of the line will depend on the more options. Therefore you need to
normalize the variables (How?) – find a common measurement

Regression model – different regression lines – going from independent variable to


dependent variables

Indices/indexes 4/8
Step 1: Do a reliability-test to see whether variables are suitable to combine into an index 
Cronbachs alpha (above 0,7)

Step 2: Bunch the variables you want to have together

Step 3: Use the created index to run comparisons with another variable. For example will
men or women answer more yes or no on the questions above.
* Answers that are incoherent – maybe just answer to get a reward from participating –
remove that person.

Reversing the scale  done in SPSS  change 1 with 5 and 2 with 4

Normalizing variables 6/8


• Sometimes we want to compare very different variables. Let’s say we have two variables
using different scales. One scale could be Number of years worked and the other scale could
be Current salary.
• In your analysis you want to predict the Future salary of a sample.
• You have two quantitative scales but you have a potential problem in how the variable with
Current salary has a far wider range and much higher values compared to the variable
Number of years worked.
• A solution is to normalize your scales. This will remove the previous scales and produce a
new common scale that is suitable for comparing data.
• There are different methods for this, e.g. Z-Score Normalization.

Dummy variables 7/8


Reduce scores to 1 or 0, present or not.

For example: Instead of having one question asking the level of education we can have three
dummy variables
High education yes/no
Middle education yes/no
Low education yes/no

We need these later when we do more advanced statistical.

- Dummy variable= Want to add qualitative variable or nominal variable into regression
analysis
- Needed for qualitative values that could be expected to change the result.
- Will the present or absence of the color green affect the result?
- Needed when comparing mean values
- Codes to 1 and 0.

Factor analysis 8/8


• Roots in intelligence research
• To find patterns in co-variation between a large amount of variables.
• Will for example several aspects of life and values have the underlying factor of
ideology/political affiliation?

 With factor analysis we can find out which variables that seem to co-variate and see
whether it could be an underlying factor steering these.

* Will hypothesis that what is behind the measurements there will be a land variable that
controlled the other variables.
* Large data – group variables together

What you examen different approaches: Exploratory vs Confirmatory (Utforskande eller


bekäftande)

* Two dimensions – correlation between these

Lecture before workshop 1 – 17 jan

To open file SPSS


- Open Spss
- Press file  open  data  choose your file

Likert scale – How much do you like ice cream? 1-5, 1 hate, 5 love – clear neutral middle.
Concreate statement that you have to agree or disagree with.

Semantic scale – Similar to Likert scale, more open ended question, don’t need to use
“agree” or “disagree”, can be terrible and great

Attitude – Hard to measure. Ask many questions to get the “answer”. Answers correlating
(Future workshop  if it is fit to put the answers “together” and create a common variable)
* Factor analysis – which type of correlation (?)
* Dependent variables (affects common variable), Independent variable (?)
* Research problem – always dependent variable
* Research questions answer – always independent variable

Scale reliability test – see if it makes sense

Analyze – Descriptive statistic – Frequencies

Other analysis types in SPSS:


Compare means – only quantitative variables
Correlation
Regression
Scale – reliability analysis
Dimension Reduction – Factor analysis

Transform (Next workshop)


– Compute variables
- Recode variables (change from quan to qual?)
- Dummy variables

Introduction lecture – 15 jan

5 workshops
- Introduction lecture, then workshop, then finishing lecture about “operation”

* Work with pre-existing data set


- Old data, note that it will have faults in it

* Understand the basics of statistic, while understanding comprehensively why different


research methods of statics are used when (analyzing focus) + critical approach

Literature
- Pallant, Julie (2010). SPSS survival manual: A step by step guide to data analysis using
SPSS. New York: McGraw-Hill. [ISBN 9780335242405, 352 pages]
* Work with like a handbook

- Van de Ven, Andrew. H. (2007). Engaged scholarship: A guide for organizational and
social research. Oxford: Oxford University Press. [ISBN 9780199226306, 352 pages]
* More “reading”

- Field, Andy.(2016). An Adventure in Statistics. The reality enigma. London: Sage. [ISBN
9781446210444, 746 pages]
* More like a novel, “beyond this course” (not important for this course)

Two parallel tracks


Statistics – SPSS
- Intro lecture
- Reading – lectures and tutorials
- Intro lecture to workshop
- Workshop
- 2h written exam (multiple choice, quite easy to get a high grade) 3 ECTS (A-F)

Research design
- Reading – discussing – project work
- Lecture/Workshops
- Feedback seminar
- Supervision
- Written group exam 1 ECTS (think only pass or fail)
- Oral individual exam 3.5 ECTS (A-F)

SPSS – download on canvas

Part 2 – 15 Jan

Different types of variables


Qualitative variables/categorical variables
- Nominal variable. Categories that cannot be ordered. Example: Gender, nationality.
- Ordinal: Categories that can be ordered but there is no meaningful distance between
them. Low, medium, high

Quantitative/continuous variables
- Interval: Scale where the difference between each step is equidistant but 0 does not
refer to an absolute point of zero. Example: Celsius degrees, dates, constructed
scales. Discrete or continuous.
- Ratio: Equidistant scale and absolute zero, such as age, income, etc. Discrete or
continuous.

Continuous = how hot it is, continues spectrum


Discreate= how many trees in the garden, absolute number

* Ask questions to use the scale as it is intended (quantitative or qualitative approach?)


Univariate analysis 1/9
Overview of relevant measures
• Central tendency (mean, median, mode)
• Dispersion (standard deviation)
• Normal distribution
• Standard error

Univariate analysis 3/9


Central tendency: Median

Median value: The middle score of a distribution


When you can’t trust the mean
11222233344
Median = 2
1222233344
Median = (2+3)/2 = 2.5

In the first example the mean is 1+1+2+2+2+2+3+3+3+4+4/11 = 2,45


But what if we have an outlier?

You might also like