UCLA

Regression Discontinuity (RD) designs are widely employed for causal inference in political science and economics, particularly in analyzing policy interventions and electoral outcomes. This thesis examines the interpretation and implementation of Politician Characteristic Regression Discontinuity (PCRD), a specialized RD approach that integrates close-election settings with politician characteristics. Despite its increasing adoption, PCRD has led to interpretational ambiguities and methodological challenges. This study aims to clarify these issues and improve the implementation of PCRD for more robust and credible causal inference.

To enhance the reliability of PCRD, this thesis addresses two key challenges: the misinterpretation of causal estimates and methodological limitations in RD estimation. First, it clarifies the distinction between leader-level and district-level PCRD, providing a structured framework to help researchers select the appropriate approach and define a precise estimand. Second, it introduces Gaussian Process Regression Discontinuity (GPrd) as a more flexible and robust alternative to conventional RD estimation methods like \texttt{rdrobust}. GPrd offers superior uncertainty quantification, effectively manages data sparsity near the cutoff, and demonstrates enhanced performance in small-sample settings, making it particularly well-suited for close-election analyses.

An empirical application using close-election RD designs investigates the relationship between partisanship and foreign direct investment (FDI) in the U.S. The findings reveal no consistent evidence that foreign investors favor districts based on the partisanship of the elected leaders, challenging conventional theories on the influence of political affiliation on economic activity. The study also includes rigorous diagnostic tests to assess the validity of RD assumptions. By combining clearer interpretative frameworks with advanced estimation techniques, this thesis offers a practical guide for the effective design and implementation of PCRD, ensuring more reliable causal inference.