Statistical Sciences 2035 Quiz: Follow-up analyses
Follow-up analyses
(independent from the table above)
- You should only conduct follow-up analysis on effects that have more than 2 levels,
because if you have only 2 levels, then you know between which levels the significant
effect is!
- Example: for rotation angle (which has 4 levels), you should conduct a follow-up analysis
to see which angles indeed differ significantly
- Types of follow-up analyses:
o Pairwise comparisons
o Independent samples t-test
o Paired samples t-test
o WS contrasts
Interaction significant yes/no?
Yes. Then:
- Check simple effects:
o Conduct paired sample t-test
▪ In which we split the Rest effect → 2 t-tests (1 for each level of rest)
▪ Here: it’s basically the same as pairwise comparisons
No. Then:
- Check main effects
o No splitting → 1 paired t-test
o Pairwise comparison: tests the difference between bonus on timepoint 1 and 2
and does the same for Rest
- It doesn’t matter if you take interaction out or not
o The p-values of the main effects won’t change because the effects are
independent → important characteristic of orthogonal designs
▪ This is the same as in a 2-way BS ANOVA
Document Summary
You should only conduct follow-up analysis on effects that have more than 2 levels, because if you have only 2 levels, then you know between which levels the significant effect is! Example: for rotation angle (which has 4 levels), you should conduct a follow-up analysis to see which angles indeed differ significantly. Types of follow-up analyses: pairwise comparisons. Independent samples t-test: paired samples t-test, ws contrasts. Check simple effects: conduct paired sample t-test. In which we split the rest effect 2 t-tests (1 for each level of rest: here: it"s basically the same as pairwise comparisons. Check main effects: no splitting 1 paired t-test, pairwise comparison: tests the difference between bonus on timepoint 1 and 2 and does the same for rest. Allows you to observe which kind of effect you have. How the factors relate with the outcome variable. You can also check the profile plots. Example from spss (rotation angle & same/diff as factors):