Hello! I am a graduate student of economics at the Pennsylvania State University, primarily interested in issues of causal inference with applications in labor and development.
You can find my CV here.
Job Market Paper
Bounding Treatment Effects in Experimental Studies with Non-Compliance: The Role of Follow Up Surveys ( PDF)
Abstract: In this paper I study the problem of identifying the causal effect of an experimental treatment when the experiment suffers from non-compliance. In particular, I consider the identifying power of information collected from non-complying participants after the completion of the treatment phase. Follow up surveys often ask study participants why they chose not to accept an offer of treatment despite being assigned to it, and answers to such questions offer insights into the decision process by which agents choose to comply with their assigned treatment status. I propose a model that rationalizes an agent's compliance decision. Based on this model, I characterize the set of values for the average treatment effect that are conformable with data observed in a randomized trial. This model and the implied set of identified values for the average treatment effect rely crucially on the availability of follow up surveys that ask agents why they chose to not comply. This underscores the importance of following up with non-complying agents since the model often leads to substantially tighter identified sets for the average treatment effect than what is possible without this information. I apply the proposed model to data from the Job Training Partnership Act Study to estimate identified sets for the average treatment effect for a number of employment outcomes.
Semiparametric Identification of Binary Choice Models with Misreported Outcomes(with Rui Wang , PDF)
Abstract: Our paper characterizes partial identification of a binary choice model when the binary dependent variable is potentially misreported. We propose two different approaches by exploiting different instrumental variables respectively. In the first approach, the instrument is assumed to only affect the true dependent variable but not misreporting probabilities. The second approach uses an instrument that only affects misreporting probabilities monotonically but does not influence the true dependent variable. These approaches neither impose distributional assumptions on unobserved disturbances nor assume parametric models for the misreporting process. We characterize conditional moment inequalities based on the identification results and demonstrate using simulations that the two approaches perform more robustly than the alternative parametric method. We apply the proposed approaches to study educational attainment using data from National Longitudinal Surveys in 1976.
Optimizing Experimental Site Selection for External Validity: Theory and an Application to Mobile Money in South Asia (with Michael Gechter, Keisuke Hirano, Jean Lee, Mahreen Mahmud, Jonathan Morduch, Saravana Ravindran, Abu Shonchoy)
Abstract: A social experiment is typically intended to inform policy decisions for contexts beyond the place and time it is implemented; it is intended to have external validity. In this paper we consider the problem of where to locate a small set of experiments in order to maximize the effectiveness of a derived statistical policy recommendation intended for a much larger set of locations. In our application, we choose six migration corridors from thousands of candidates across South Asia on which to experimentally evaluate a program teaching migrants to send remittances via mobile money, with the goal of deriving policy recommendations for all the candidate sites. We adopt a quasi-Bayesian approach to this design problem by developing a prior specification for the joint distribution of site level average treatment effects based on a microeconometric structural model for migrant remittance decisions and additionally, build in the possibility that there are other sources of heterogeneity across sites that are not captured by the structural model. In ongoing work, we run experiments in the sites selected in this way, and, within our set of chosen locations, evaluate the performance of our methodology in determining the relative informativeness of each site.