On the frequency of Simpson's paradox and the pitfalls of aggregation

Abstract

Simpson's paradox is well known in statistics and arises in real-life situations dealing with social, life and medical sciences, but may not be that well known in business school communities, and especially not well covered in introductory business statistics courses. The paradox occurs when data from two or more groups is aggregated into a single group. This paper introduces Simpson's paradox in a business context and shows through simulations, under different distributions, that Simpson's paradox is a plausible event though not a very frequent one.