Hi everyone: I scored four mutually-exclusive behaviors (e.g. feeding, brooding, standing, preening) on a number of observations of birds in two different treatments, and two different nesting attempts. I'd like to see if these behaviors differ between attempts, treatments or both. Behaviors were all scored as a percentage of total time, treatment and attempt are both factors with two levels each.
I can use a GLM to analyze behaviors e.g.
mod1 <- glm(feeding ~ treatment * attempt)
Which gives me something like this:
Estimate Std. Error df t value Pr(>|t|)
(Intercept) 0.1177 0.5285 26.4473 0.223 0.82549
treatmentW 1.3943 0.8726 26.8958 1.598 0.12176
attempt 1.0762 0.2963 15.9821 3.632 0.00225 **
treatmentW:attempt -1.4901 0.5409 17.6237 -2.755 0.01322 *
My question: Since I'm running a separate model for each behavior, what's the best way to deal with multiple comparisons? Throw the p-vals from all tests into a Bonferroni (Dunn)? Should I include only significant p values, (attempt and treatment:attempt in this example) or all the p values from all tests (treatment, attempt and treatment:attempt)? Should I instead do a bunch of chi-sq tests and then bonferroni them? Or ignore multiple comparisons and just forge ahead?