Essays on causal inference in observational studies

This dissertation consists of three essays discussing methods for causal inference and show how they may be applied to estimate the effects of policy interventions in nonexperimental settings. The first essay (coauthored with Jasjeet S. Sekhon) introduces genetic matching, a multivariate matching method that uses a genetic algorithm to optimize the search for a suitable control group. Empirical examples are drawn from Monte Carlo simulations and a classic job training dataset. The second essay explains how the Rubin causal model (Holland 1986) and matching methods can address problems for study design in a complex yet common observational setting: when there are multiple heterogeneous treatments that may be related to prior treatments and observed outcomes. TrEffer (Treatment Effect and Prediction), a German government project pertaining to the evaluation of job training programs, is used as an empirical example. The third essay investigates the impact of United Nations peacekeeping following civil war. King and Zeng (2007) found that prior work on this topic (Doyle and Sambanis 2000) had been based more on indefensible modeling assumptions than on evidence. This essay revisits the Doyle and Sambanis (2000) causal questions and answers them using new matching-based methods. These new methods do not require assumptions that plagued prior work, and they are broadly applicable to many important inferential problems in political science and beyond. When the methods are applied to the Doyle and Sambanis (2000) data, there is a preponderance of evidence to suggest that UN peacekeeping has had a positive effect on peace and democracy in the aftermath of civil war.