Robust Methods for Program Evaluation.

Abstract

My dissertation research focuses on different approaches to conduct robust estimation and inference in the context of program evaluation. In Chapter 1, I look at the effects of teacher and peer characteristics on student achievement in the STAR Project conducted in Tennessee in the late 1980s. It generalizes previous empirical research by allowing unobserved effects to enter the structural function nonseparably. No functional form assumptions are needed for identification. The empirical analysis looks at the effects of class size, teacher experience and gender composition of the classroom on test scores. Findings suggest that nonseparable heterogeneity is an important source of individual-level variation in academic performance. In Chapter 2, a joint work with Matias Cattaneo and Rocio Titiunik, we study robust inference in the context of regression discontinuity (RD) design. Local polynomial estimators are routinely employed to construct confidence intervals for treatment effects. The performance of these confidence intervals in applications, however, may be seriously hampered by their sensitivity to the specific bandwidth employed. Available bandwidth selectors typically yield a large bandwidth, leading to data-driven confidence intervals that may be severely biased. We propose new theory-based, more robust confidence interval estimators for average treatment effects at the cutoff in sharp RD, sharp kink RD, fuzzy RD and fuzzy kink RD designs, constructed using a bias-corrected RD estimator together with a novel standard error estimator. Finally, in Chapter 3, also written jointly with Matias Cattaneo and Rocio Titiunik, we present new results regarding RD plots, which are nowadays widely used in applications, despite its formal properties being unknown: these plots are typically presented employing ad hoc choices of tuning parameters. We formally study the most common RD plot based on an evenly-spaced binning of the data, and propose an optimal data-driven choice for the number of bins. In addition, we introduce an alternative RD plot based on quantile-spaced binning, study its formal properties, and propose the corresponding optimal data-driven choice for the number of bins. The main proposed data-driven selectors employ spacings-based estimators, which are simple and easy to implement in applications because they do not require additional choices of tuning parameters.