Computational Approaches to Identification and Inference for Causal Parameters in Experiments and Quasi-Experiments

This project is aimed at developing new computational approaches to identification and inference for causal parameters in some common experimental and quasi-experimental settings, and applying the resulting approaches to empirical examples in labor economics, development economics, and in medicine. The settings we are looking at include (1) factorial experiments with noncompliance, where alternative treatments are randomly assigned both alone and in combination with each other but where not all subjects assigned to a treatment take-up that treatment; (2) augmented experimental designs, where the experimental design includes additional sequential stages of randomization to help infer the pathways through which the treatment affects the outcome of interest; and (3) quasi-experimental designs with judge/examiner type instruments. In such areas, we are developing computationally feasible methods for inference on parameters that are only partially identified, in other words, can only be bounded given the assumptions we are imposing.

There are two distinct RA positions, one for helping primarily with coding and the more formal aspects of the work, and the other helping primarily with literature review of the relevant empirical literature and working with data for applied examples.