Model Assisted Causal Inference

These essays develop and implement design-based methods for causal inference, with applications in political science ranging from American political behavior to international political economy. Through wedding known random assignment processes to nonparametric models, the proposed methods facilitate inference in situations where standard econometric or statistical assumptions may falter. In so doing, these essays address important methodological and substantive issues in the discipline. Using original results for design-based estimators, causal effects are estimated for a complex randomized field experiment on the effects of a voter awareness campaign in Northern India. Both direct and spillover effects are estimated in an original multilevel field experiment on anti-conflict efforts conducted in 56 United States middle schools. The habit-formation effects of voting are robustly estimated by using a novel instrumental variables estimator on data from a downstream field experiment on voter turnout. Applying new identification results for Mendelian randomization to an original survey, the causal effects of skin color on attitudes are estimated for a sample of mixed-race and African-American dizygotic twins. Using a conditional randomization test, spillover effects are detected in a field experiment designed to assess the effect of mailings on voter turnout in a United States primary election.