Empirical and Computational Methods for Electoral Politics

Abstract

This collection of three essays introduces new computational and empirical methods for data analysis and validation in political science, with specific applications to the practice and study of electoral politics.

In the first chapter, I introduce a set of graphical diagnostics that can be used to improve statistical models of heterogeneous treatment effects. I adapt the uplift curve, which has been used previously to visualize cumulative benefits from targeting treatments, and repurpose it as a tool for model selection to make model-building workflows for predicting treatment effects less ad-hoc. I also present a stack-ranking graphical diagnostic for heterogeneous treatment effect models to help visualize model fit, and I apply both diagnostics to a canonical experiment on voter mobilization using social pressure appeals. The chapter advances a more principled model-building approach for predicting individual-level treatment effects.

The second chapter builds on recent computational and simulation-based methods for studying the effects and consequences of congressional redistricting. As ensembles of simulated redistricting plans become more common in the academic literature and as legal evidence, validating the accuracy and representativeness of those algorithms has become increasingly important. This paper makes two advances --- first, it applies a recently-developed enumeration algorithm to increase the size and complexity of simulation validation datasets, and second, it introduces a new validation diagnostic that can be used to ensure redistricting simulators are accurately representing the target distribution of plans. I then apply the new validation test to two competing redistricting simulation methods.

The third chapter (adapted from work coauthored with Ted Enamorado and Kosuke Imai) is a practical guide to data merging and record linkage using fastLink, a new open-source implementation of the canonical Fellegi-Sunter probabilistic record linkage model. I briefly review fastLink and the computational improvements it implements, and I then walk through several applied data examples using the software that illustrate its preprocessing, merging, and inspection functionalities. Finally, I apply the entire proposed workflow and software to a validation exercise using data on local-level politicians in Rio de Janeiro, Brazil, where I use fastLink to analyze rates of party switching across election cycles.