Simpson's Paradox & Confounding in Categorical Data

Investigate how case mix can reverse aggregate comparisons even when subgroup results agree. Compute pooled and standardized rates, analyze changing weights and distinguish common causes, mediators and selection effects. Original service and admissions scenarios introduce Simpson reversals and Berkson selection with explicit assumptions. Prerequisites: conditional proportions and weighted averages; causal claims remain carefully qualified. Includes single-select and multiselect practice with worked explanations and misconception-specific feedback.

Further reading and method references:

• Stanford Encyclopedia: Simpson’s paradox