Addressing bias in politician characteristic regression discontinuity designs

Abstract

Politician characteristic regression discontinuity (PCRD) designs are a popular strategy when attempting to casually link a specific trait of an elected politician with a given outcome. However, recent research has revealed that this methodology often fails to retrieve the target causal effect¿a problem also known as the PCRD estimation bias. In this paper, I provide a new econometric framework to address this limitation in applied research. First, I propose a covariate-adjusted local polynomial estimator that corrects for the PCRD estimation bias provided all relevant confounders are observed. I then leverage the statistical properties of this estimator to propose several decompositions of the bias term and discuss their potential applications. Next, I devise a strategy to assess the robustness of the new estimator to omitted confounders that could potentially invalidate results. Finally, I illustrate these methods through an application: a PCRD aimed at evaluating the impact of female leadership during the COVID-19 pandemic.