How to analyze your data and summarize results

1
How to analyze your data and summarize results

• Note This is not about statistics!

2
Overview

• Planning data analysis
• Running analyses
• 4 Steps
• Some tricks
• Writing a summary

3
First things first

• You have a job to do
• And you want to do the best you can
• Me and my friend, if we were you
• Would start to make a plan..

4
The Master Plan

• Know the theory
• Guidance throughout
• Predictions (keep it broad !)
• Know your audience
• Conventions and standards
• Know your data
• What should you look for
• Associations (correlations, EQS)
• Differences in Means
• Where should you look first
• Contingency plans

5
Executing the plan

• Prepare data
• Data reduction
• Perform targeted analyses
• Iterate, but not indefinitely

6
Executing the plan

• Bad result? Explore!

7
Executing the plan

• Bad result? Explore!
• Encouraging result? Massage!

8
Executing the plan

• Bad result? Explore!
• Encouraging result? Massage!
• Positive result? Polish!

9
Executing the plan

• Bad result? Explore!
• Encouraging result? Massage!
• Positive result? Polish!
• And then? Explore some more!

10
Four steps

• Clean up data check frequencies, recode
• Scale construction / data reduction / aggregation
• Test predictions
• Explore

11
Step 1Cleaning up
Once you have the data in, check that the
frequencies of all variables are OK. For
example, if you have a 7-point scale, you can't
have the value 23. You can also check if you
defined your missing variables correctly.
Here's an example frequency command. PS all
the bits in brackets are where you have to
enter variable names from your system
file!. FREQUENCIES var1 var2 ... /statistics
mean sd min max skewness kurtosis /histogram.
12
Step 1
Now you can recode variables. Some scale
items are "negatively worded". For those items,
you should recode such that 71 and 17 (on a
7-point scale). The simplest way of doing this
is COMPUTE var1 8 - var1.
13
Step 2 Data reduction

• The objective in this step is to reduce the
  number of variables in your dataset to a limited,
  manageable number. There are no clear pointers
  how many you should have, but in general less is
  more.
• Usually you know which items go with which, but
  sometimes you're not so sure. It might be useful
  to do a factor analysis. Note that this is not
  always useful.

14
Step 2

• More useful, and something you always need to do,
  is to test reliability.
• This you do with a-priori scales constructed so
  that items should go together.
• In the output, you should watch inter-item
  correlation as it is the most important indicator
  of scale robustness.
• But you report alpha in your report or paper.

15
Step 2

• Now you can compute scales.
• COMPUTE new variable name here mean( var1,
  var2, ... ).

16
Step 3 Hypothesis testing

• You should know what to do here. If you don't
  then you are likely to end up in Step 4 (or in
  limbo)
• Theory should guide you. If you do not know how
  to execute this step then
• Either you do not know enough about statistics,
  in which case ask someone to help, read a book,
  ask supervisor
• Or you do not know well enough why you did your
  research and what you would expect to find.
  Again ask someone to help, read a book, and/or
  ask supervisor

17
Step 4 Exploratory analyses

• In this step you explore the data to assess what
  else there is. Often, the confirmation of the
  predictions is not the only thing. Often, the
  unexpected is more interesting.
• The challenge in exploration is to stop it from
  becoming a fishing expedition. Explorations are
  not random, but guided by theory. Ideally they
  try and test some of the ideas that your data
  ALSO allows you to explore.

18
Step 4

• Examples
• to check for mediation
• to check if there are different patterns of
  correlation across conditions (this could be a
  sign of moderated mediation--different processes
  occurring in each condition)
• to check for some "opportune" effects (gender
  differences, age effects, alternative IVs)

19
When are you done?

• Normally you cycle through steps 2, 3 and 4
  numerous times
• Exhausting the scope of your data
• Polishing the end result
• Throughout, you continue to revise your SYNTAX
• Doing this pays off. An exact record of the
  procedures used to transform raw data into
  analyses reported in a paper is extremely useful,
  especially when you need to get back into the
  analyses after a while.
• You are done when the end result is able to
  convince your target audience

20
Questions? Comments?
21
Some tips and tricks

• People typically use a very limited set of
  analyses.
• It is useful to compile a library of favourites
• See example in handout
• You can use this as an external memory, and to
  speed up standard procedures
• I will demonstrate this...

22
Some tips and tricks

• Dont be afraid to stop analyzing
• Write a good summary of results and discuss this
  with others
• Continue polishing the data while you are
  writing. Use your results section as an
  opportunity to tell a persuasive story. You will
  discover that you return to your data as you are
  doing this.

23
Writing a summary

• Always follow APA guidelines
• Do not use SPSS output anywhere
• Keep it down to 2 or 3 pages

24
Comparing means

• If you are comparing means, start with tables
  that summarize the key effects
• Use subscripts to indicate which cells are
  significantly different according to simple main
  effects analyses

25
Comparing means

• If necessary summarily report which effects are
  significant and which are not
• Start by reporting main effects, then
  interactions, ending with highest order one.
• Make sure this can easily be scaled up into a
  full results section paragraph!

26
(No Transcript)
27
A summary of a 2 X 2 ANOVA

• Depersonalization manip check
• Depersonalization F(1, 26) 27.53, p lt .001,??2
  .51.
• Group formation F(1, 26) 0.44, ns,??2 .01.
• Interaction F(1, 26) 1.14, ns,??2 .04.

28
This can easily be reworked to a full paragraph
in results section

• The check of the depersonalization manipulation
  indicated that it was successful. Results showed
  a main effect of depersonalization on the
  anonymity check, F(1, 26) 27.53, p lt .001,??2
  .51. In the depersonalized condition, groups
  indicated that they felt anonymous (M 5.69, SD
  1.12) compared with the individuated condition
  (M 3.78, SD 0.80). The group formation main
  effect and interaction were not significant (F's
  lt 1.2).

29
An example from the book
30
Example pic
31
Reporting associations

• Start with a correlation matrix
• Add tables for regressions or structural equation
  models
• Make figures to illustrate the findings

32
Examples from the book
33
Correlation
34
Stepwise multiple regression
35
SEM fit table
36
SEM figure
37
Questions? Comments?