Selection bias occurs when the treated and untreated groups differ systematically in ways that also affect the outcome, making a simple comparison between the two groups a biased estimate of the causal effect. In other words, the "selection" into treatment is correlated with the potential outcomes. Every causal design pattern in this book addresses selection bias through a different identification strategy: [[Design Pattern I - Instrumental Variable (IV)|instrumental variables]] exploit an exogenous source of variation, [[Design Pattern II - Regression Discontinuity (RD)|regression discontinuity]] appeals to local randomization at a cutoff, and so on. The common thread is that none of them relies on the naïve comparison. > [!info]- Last updated: April 12, 2026