An overall average blends results from different groups. If the groups have different typical values and their shares change, the combined number can move in a direction that none of the individual groups does. This is often called Simpson’s paradox.
Imagine two clinics whose success rates both improve. If many more difficult cases shift to the clinic with the lower rate, the combined success rate can fall. The decline reflects the changing case mix, not a deterioration inside either clinic.
The reverse can happen as well: an overall improvement may come from a larger share of observations entering the higher-performing group even while every subgroup worsens. The aggregate alone cannot reveal which mechanism produced the change.
Break results down by relevant characteristics such as age, region, risk level or product type, but choose groups based on the question rather than searching until a preferred pattern appears. Report group sizes alongside their rates because the weights drive the combined result.
Neither the subgroup nor the total is automatically the correct answer. The total describes the population mix that actually occurred; subgroup comparisons help separate performance from composition. Understanding both prevents a real arithmetic result from receiving the wrong explanation.
Original educational commentary by Breaking Click. This explainer is designed to help readers evaluate information more carefully.



