This is the "what if lab rats just get cancer a lot" joke except realDo you have a correlation in your data? Or is one of your sample groups simply more likely to generate *statistics*?