This paper examines the validity of regression-discontinuity (RD) designs in estimating causal effects, particularly in the context of U.S. House elections. It argues that under certain conditions, RD designs can be as credible as randomized experiments. The key idea is that even when treatment status is determined by non-random self-selection, the inherent randomness in the final vote count can create a situation where treatment status near the threshold is effectively randomized. This is illustrated using an analysis of U.S. House elections, where the vote share is influenced by both the candidate's characteristics and random chance. The paper shows that the probability of a candidate winning an election is not solely determined by their qualities but also by random factors, which can lead to a discontinuity in the distribution of pre-determined variables at the threshold. This discontinuity can be used to test the validity of the RD design. The analysis finds that the incumbency advantage is significant, with the probability of a candidate winning the next election being about 0.40 to 0.45 higher than for non-incumbents. The paper also discusses the implications of this finding for the validity of RD designs and the importance of testing the continuity of the density function around the threshold. The results suggest that the RD design can provide reliable estimates of causal effects in the context of U.S. House elections.This paper examines the validity of regression-discontinuity (RD) designs in estimating causal effects, particularly in the context of U.S. House elections. It argues that under certain conditions, RD designs can be as credible as randomized experiments. The key idea is that even when treatment status is determined by non-random self-selection, the inherent randomness in the final vote count can create a situation where treatment status near the threshold is effectively randomized. This is illustrated using an analysis of U.S. House elections, where the vote share is influenced by both the candidate's characteristics and random chance. The paper shows that the probability of a candidate winning an election is not solely determined by their qualities but also by random factors, which can lead to a discontinuity in the distribution of pre-determined variables at the threshold. This discontinuity can be used to test the validity of the RD design. The analysis finds that the incumbency advantage is significant, with the probability of a candidate winning the next election being about 0.40 to 0.45 higher than for non-incumbents. The paper also discusses the implications of this finding for the validity of RD designs and the importance of testing the continuity of the density function around the threshold. The results suggest that the RD design can provide reliable estimates of causal effects in the context of U.S. House elections.