Most researchers believe Simpson's paradox (where the trend disappears or reverses when going from looking at the whole data set to subgroups or vice versa) is rare but the rate at which it's found and simulation studies say otherwise