Definition and the Berkeley Graduate Admissions Example

Simpson's Paradox is a statistical phenomenon in which a trend observed within individual subgroups reverses when the groups are combined. One example comes from UC Berkeley graduate admissions data for fall 1973, where women appeared to have a lower overall acceptance rate than men, even though department-level data showed no clear pattern of bias against women. The reversal occurred because women's applications were concentrated in departments with lower acceptance rates.

The Role of Confounding Variables

At the heart of the paradox lies the confounding variable - a third factor that influences both the apparent cause and the outcome. In the Berkeley case, the choice of department was the confounder. Ignoring it produced a misleading aggregate picture.

Techniques such as stratified analysis, regression adjustment, and propensity score matching help control for confounders. The paradox shows that naive aggregation can lead to wrong conclusions about causal relationships.

Occurrence in Ranking Data

Simpson's Paradox can arise in global rankings. For instance, a country might outperform the world average in every age group for a health metric, yet fall below the world average overall because its population skews toward an age group with lower scores.

Why Aggregation Level Matters

The level at which data is aggregated can flip conclusions. Whether you look at a national ranking, a regional breakdown, or an age-stratified view, your apparent position may change.