To be quite frank, repeated hypothesis tests on ad-hoc groupings is _terrible_ science.