This paper by David S. Lee explores the conditions under which causal inferences from a regression-discontinuity (RD) analysis can be as credible as those from a randomized experiment. Specifically, it examines a standard treatment evaluation problem where treatment is assigned based on an observed continuous covariate \( V \) crossing a known threshold \( v_0 \). The paper establishes that if \( V \) has a continuous density function that is continuous at \( v_0 \), then treatment status can be considered as good as randomized in a local neighborhood of \( V = v_0 \). This result is illustrated through an analysis of U.S. House elections, where the inherent uncertainty in the final vote count implies that the party winning is essentially randomized among elections decided by a narrow margin. The evidence supports this prediction, providing "near-experimental" causal estimates of the electoral advantage to incumbency. The paper also discusses the practical implications of these findings, including the ability to test the validity of the RD design by examining the distribution of baseline characteristics on either side of the threshold.This paper by David S. Lee explores the conditions under which causal inferences from a regression-discontinuity (RD) analysis can be as credible as those from a randomized experiment. Specifically, it examines a standard treatment evaluation problem where treatment is assigned based on an observed continuous covariate \( V \) crossing a known threshold \( v_0 \). The paper establishes that if \( V \) has a continuous density function that is continuous at \( v_0 \), then treatment status can be considered as good as randomized in a local neighborhood of \( V = v_0 \). This result is illustrated through an analysis of U.S. House elections, where the inherent uncertainty in the final vote count implies that the party winning is essentially randomized among elections decided by a narrow margin. The evidence supports this prediction, providing "near-experimental" causal estimates of the electoral advantage to incumbency. The paper also discusses the practical implications of these findings, including the ability to test the validity of the RD design by examining the distribution of baseline characteristics on either side of the threshold.