We use data for 198121 candidates and 1351 random election outcomes to estimate the effect of incumbency status on future electoral success. We find no evidence of incumbency advantage using data on randomized elections. In contrast, regression discontinuity design, using optimal bandwidths,...