This is something we all need to understand, and to check whether studies use it or not. You can’t just compare differences in outcomes between two populations, you need to compare the difference in differences. This tends to reduce the size of the effect and therefore the publishing value of the paper.

That a painfully high proportion of neuroscience papers fall foul of this is described by Ben Goldacre, but it sounds like it is likely to take place elsewhere too (is it commonly used in pharmaceutical research?).