I am wondering why you don't (normally) run post-hoc analysis after insignificant interaction effect from a Two-Way ANOVA even if there is a simple main effect?
There may be confusion.
If there is a significant interaction effect, then the post-hoc on the main effects are often not of interest. But the post-hoc on the interaction is of interest.
If there is not a significant interaction effect, but there are significant main effects, then the post-hoc is performed on the significant main effects.
I'm pretty sure that this is standard thinking.
You _do_ normally run post-hoc analysis on the significant main effects.
I never run or interested in running post-doc analysis for insignificant interactions as well. Even you run this, then effect will be seen in the same direction and only your significant main effect will have a role there, I think.
There may be confusion.
If there is a significant interaction effect, then the post-hoc on the main effects are often not of interest. But the post-hoc on the interaction is of interest.
If there is not a significant interaction effect, but there are significant main effects, then the post-hoc is performed on the significant main effects.
I'm pretty sure that this is standard thinking.
Thank you Salvatore and Prabhat. You have both clarified it well for me. I believe I understand what you both are saying. So even if there is no significant interaction effect but there are significant main effects, then I would only do post hocs for those groups within each main effect, correct?
David, for posterity, the answer is yes. Say you have an experiment with two diets and two timepoints
Standard diet and high fat diet
10 wks and 20 wks
If your main effect of diet is significant, and the main effect of time is not significant, and your interaction of diet*time is not significant
Then one would perform a test comparing diet groups
Dear David Im and David Choi,
Wei et al (“Comparisons of treatment means when factors do not interact in two-factorial studies”, Amino Acids, 2012 42(5):2031-5) provide examples of 2-factor studies where it is useful to perform a post-hoc analysis when only one factor and not both factors or their interaction is significant. In the abstract, they wrote "when the two factors do not interact, a common understanding among biologists is that comparisons among treatment means cannot or should not be made. Here, we bring this misconception into the attention of researchers. "
Instead of focusing on p-values, take into consideration effect size comparisons. You'll find theoretical and practical implications in:
Becker, B.J., 1988. Synthesizing standardized mean-change measures. British Journal of Mathematical and Statistical Psychology 41(2), 257-278.
Jasbi, M., Sadeghi Bahmani, D., Karami, G., Omidbeygi, M., Peyravi, M., Panahi, A., Mirzaee, J., Holsboer-Trachsler, E., Brand, S., 2018. Influence of adjuvant mindfulness-based cognitive therapy (MBCT) on symptoms of post-traumatic stress disorder (PTSD) in veterans - results from a randomized control study. Cogn Behav Ther, 1-16.
In SPSS user guide, it says that there is no post hoc test for factorial ANOVA. the data have to be rearranged to perform a One-way ANOVA.
- Badges
- Science topic