Post-hoc multiple comparisons are significant but interaction term is not signicant
I could use some help with interpreting the results of a statistical analysis conducted using generalized linear mixed models fitted with the GLIMMIX procedure of SAS (SAS Institute Inc., Version 9.2). This was a 2x2 factorial study examining disease in 2 classes of animals (male and female) receiving 2 drug treatments. Animals and pen nested within sex and treatment was incorporated in the linear predictor to recognize pen as the experimental unit for these factors. For binomial responses, inference was conducted after checking for absence of overdispersion based on the Pearson Chi-Square/degrees of freedom fit statistic. Model parameters were estimated using Laplace integral approximation to maximum likelihood. Pairwise comparisons were conducted using Bonferroni's method to adjust for multiple comparisons and avoid inflation of Type I error rate.
The p-values for the main effect of sex and treatment and the interaction of sex*treatment were as follows;
Sex Trt Sex* Trt
0.01 0.06 0.45
0.07 0.28 0.36
0.06 0.31 0.21
Even though the interaction term in this case was not significant (P=0.45 and P=0.21), pairwise comparison between treated and untreated animals of the same sex was significant (p=0.06 for one outcome and p = 0.003 for another outcome).
I have reported the results of the pairwise comparisons but I was told that I can't use Bonferroni's test for multiple comparisons if the interaction term is not significant.
Is this correct? I would appreciate any assistance you could provide.
Re: Post-hoc multiple comparisons are significant but interaction term is not signica
I have the same question.
Basically, rats are tested for their ability to recognize an object (familiar or unfamiliar) and receive a drug or vehicle treatment, so I have a 2x2 ANOVA. The unfamilar condition is basically a control condition to account for changes in general activity.
I have a significant effect of recognition, and a significant effect of drug treatment, but no significant interaction. I did some Bonferroni post-hoc tests to sort out the effect I am really interest in: Vehicle familiar vs drug familiar is significant; vehicle unfamiliar vs drug unfamiliar is not. This exactly what we expected, so everything was fine. Except...
I have since been told that I cannot use a Bonferroni comparison without a significant interaction by one person, and that Bonferroni correction eliminates the need for a significant interaction by another person.