I did a GLMM (corrected for school and administration levels) and analysed the association between food environment and weight status. I corrected for several demographics and other urban variables. As I did a forward selection model with those other potential confounder, do I still need to do a post hoc test or calculate the Bonferroni upfront to avoid type 1 error?
Thereafter, I analysed the adjusted models (for confounder) for interactions. Do I also need to test those with post hoc?
You definitely need an adjustment of some sort, but it isn't clear how we can reliably adjust the results of a stepwise procedure.
Members of this site routinely warn against using stepwise regression for inference. There are a number of problems with the procedure, but they can mostly be summed up by saying all the results are biased (p values, confidence intervals, effect sizes etc).
You're best not to use it and rely on your subject matter knowledge to select which variables are important.
