Data Analytics: Foundations to Practice · Correlation, Relationships, and Causation
Simpson's Paradox and Aggregation Traps
A pattern can reverse direction entirely between an aggregated view and a properly segmented view of the exact same underlying data, a genuinely counterintuitive phenomenon with a specific name and a specific, well-documented cause.
Simpson's Paradox describes a genuinely counterintuitive situation where a trend that appears clearly in several distinct, separate groups of data reverses direction entirely, or disappears altogether, when those same groups are combined and analyzed together in aggregate, a pattern that can seriously mislead an analysis relying only on an aggregated, whole-population view without ever checking whether that aggregated pattern actually holds up consistently within each meaningful underlying subgroup.
- Simpson's Paradox describes a trend appearing in separate groups that reverses or disappears entirely once those groups are combined into an aggregate.
- This can occur when one treatment outperforms another within every subgroup individually, yet appears worse overall due to uneven subgroup distribution.
- The paradox typically arises from a confounding variable unevenly distributed across compared groups, combined with meaningfully different subgroup sizes.
- The defense is checking whether an aggregate finding holds up when segmented by relevant subgroups; a reversal signals a confounding variable, not a reason to dismiss the segmented result.