Presentation Title: Can close election regression discontinuity designs identify effects of winning politician characteristics?
Abstract: Regression discontinuity (RD) designs leveraging close elections are increasingly used to estimate the effects of elected politician characteristics on downstream outcomes. Unlike textbook RD designs, treatment is defined by one predetermined characteristic of election winners (of potentially many) that could affect their victory margin. I prove that such post-treatment conditioning causes RD estimators to capture the effect of both the specific characteristic of interest and all compensating differentials—candidate- or context-level characteristics that ensure winning candidates remain in close races despite being advantaged/disadvantaged by the specified characteristic. Isolating this characteristic’s effect generally requires assuming either that the specified characteristic does not affect candidate vote shares or that no compensating differential affects the outcome. Since theories of voting behavior suggest that neither strong assumption often holds, I further consider whether and how balance tests, covariate adjustment, bounding, mechanism tests, and bundled treatments can mitigate the post-treatment bias afflicting politician characteristic RD designs.