I have done a series of analyses with random effects in all analyses, one analysis has a binomial error structure, another is Gaussian, and another is zero-altered poisson. In two of these analyses I am examining how interactions between phenotypic traits and environmental conditions influence the probability of dispersal from a population (the binomial) and dispersal distance (Gaussian). The zero-altered model is to look at lifetime reproductive success. I submitted a manuscript on this and was told by one of the reviewers to include post hoc tests for all the interactions in my final model. I have both 2 and 3 way interactions in my models. The 3 way interactions include population (there are 2 of them) and then two other traits that interact differently depending upon the population. In some of my interactions, there are categorical variables (sex and population) and for others, the 2 way interactions include only continuous variables. I am not a fan of post hoc tests except in the case of ANOVAs. I know that one can use post hoc tests with glmer in r. But given that my categorical variables never have more than 2 terms and the rest of the terms are continuous, the comments by the reviewer do not make sense to me. This means that one of the levels is in the intercept term. Maybe I could just change the reference level (the intercept) to have it be the other level of the categorical variable rather than doing a post hoc? Maybe I am way off base, but I wanted to see if anyone has an opinion about the validity of post hoc tests with mixed models, only categorical variables with 2 levels, and binomial or count data.

Any input you have would be appreciated.

Thank you,
Corey