I'm having a bit of difficulty in deciding what measures I need to take given my situation.
I have about 8 outcome variables and I use a variety of models (depending on the outcome) to investigate whether my grouping variables (3 levels - a, b, and c) differ after adjusting for covariates.
For example, I have 2 models that are Cox-Regression and 6 that are Poisson Regressions. In each analysis I'm concerned with all possible pair wise comparisons in the group covariate (a vs b, a vs c, and c vs b).
My question is how to (if necessary) adjust the P-values in these analyses as the analysis will be submitted for publication. I've worked with multiple comparisons before and am familar with some procedures such as holm's, bonferroni, etc.
However in this situation, I have multiple outcomes too - each of which will have the same 3 comparisons in them.
What's the best strategy to approach this? Does any body have a resource?
So you have reoccurring independent predictors (3) within multiple models with different dependent variables and are conducting pairwise comparisons in all of the models. I am interested to see what others says. I would think if you correct the alpha in each model you would be fine for publication standpoints, but I understand the threat you are wondering about and could imagine corrective purists may argue to possibly further correct. Lets see what people say, I cannot recall ever seeing an example of this where additional corrective action was taken.
Reply With Quote
