I'm working on a problem involving clinical trial data and am trying to account for the different clinics patients were treated at. In papers I've read, authors present p-values for differences in means/proportions (chi-sq, Wilcoxon, etc tests) both normally and then "adjusted for site."
This makes sense in that there is often clustering in outcomes across sites (which I intend to account for in subsequent analyses using mixed effect models), but I'm wondering what kind of method is used in the case of hypothesis testing, as it's never reported what this entails statistically (perhaps I am missing something obvious). Is it simply stratifying by site and pooling the p-values? Or is there some other way to account for this type of clustering when analyzing differences in means/proportions? Thank you!
