Causal Inference for Policy Researchers under Potential Outcomes: The Use of Propensity Score Methods
In public policy settings, the main questions of interest are often about the causal relationship between predictors and outcomes of interest, rather than mere correlations. For example, it is not the correlation between teacher compensation and student achievement or between infrastructure investments in conflict zones and local support of terrorists that is of interest, but the true effect of increasing a teacher’s compensation by a reasonable amount on his or students achievement score or the true effect of increasing infrastructure investments by a fixed amount on how often locals support or engage in terrorist activity.
In this course we study methods for estimating and identifying such causal effects using the potential outcomes framework. We briefly discuss the gold standard technique used in the statistical literature for establishing causal effects, namely randomized clinical trials. We then move on to discuss observational studies with an overview of methods used in that setting and extensively discuss the dominant method of propensity scores in the potential outcomes framework. We debate theoretical and practical issues arising in causal inference as well as examine applications in public policy studies where these or other methods have been employed. We will discuss conditions that are required for credible inference for causal effects in this framework and work on extension or derivation of existing methods to deal with specific issues encountered in public policy research.