Simpson's paradox is a phenomenon where a trend observed in several groups reverses or disappears when they are combined, because of a grouping variable that unbalances the counts. It is a reminder to analyze the data by subgroups before concluding about the whole.