|
This dissertation consists of two chapters of in-depth empirical analyses in the domain of public economics, and one chapter of causal inference methodology regarding distributional robustness. The first chapter is titled ``The Effect of Air Pollution on Academic Performances: Evidence from South Korea, " and investigates the causal effect of air pollution on academic performances. While the effect of air pollution on health has been extensively studied, little is known about its effect on education, especially in a causal context. Here, I exploit the unique geography of Korea and a meteorological phenomenon called Asian Dust Storm (ADS) to get exogenous shocks of air pollution carried by the wind from China. Using two stage least squares regression, I find that an increase in particulate matter (PM10) leads to an increase in the share of students who underperform, while its effect on the share of students who overperform is not different from zero. I find similar results for elementary and middle school test outcomes, and find that air pollution disproportionately affects the types of schools associated with low socioeconomic status. Looking at both short term and long term effect of air pollutants, I find that air pollution has both acute and cumulative effect on the academic performances of the students. I explore health as a mechanism through which air pollution affects academic outcomes, and find that the most detrimental effect comes from the most harmful to health pollutants—PM10 and ozone—and do not find any evidence of preemptive absenteeism or mobilization due to air pollution on the school level. The second chapter, co-authored with Mark Duggan, Irena Dushi, and Gina Li, is titled ``The Effects of Changes in Social Security's Delayed Retirement Credit: Evidence from Administrative Data." The delayed retirement credit (DRC) increases monthly OASI (Old Age and Survivors Insurance) benefits for primary beneficiaries who claim after their full retirement age (FRA). For many years, the DRC was set at 3.0 percent per year (0.25 percent monthly). The 1983 amendments to Social Security more than doubled this actuarial adjustment to 8.0 percent per year. These changes were phased in gradually, so that those born in 1924 or earlier retained a 3.0 percent DRC while those born in 1943 or later had an 8.0 percent DRC. In this paper, we use administrative data from the Social Security Administration (SSA) to estimate the effect of this policy change on individual claiming behavior. We focus on the first half of the DRC increase (from 3.0 to 5.5 percent) given changes in other SSA policies that coincided with the later increases. Our findings demonstrate that the increase in the DRC led to a significant increase in delayed claiming of social security benefits and strongly suggest that the effects were larger for those with higher lifetime incomes, who would have a greater financial incentive to delay given their longer life expectancies. The third chapter, co-authored with Hongseok Namkoong, is titled `Robust Causal Inference Under Covariate Shift via Worst-Case Subpopulation Treatment Effects." In this chapter, we propose the worst-case treatment effect (WTE) across all subpopulations of a given size, a conservative notion of topline treatment effect. Compared to the average treatment effect (ATE), whose validity relies on the covariate distribution of collected data, WTE is robust to unanticipated covariate shifts, and positive findings guarantee uniformly valid treatment effects over subpopulations. We develop a semiparametrically efficient estimator for the WTE, leveraging machine learning-based estimates of the heterogeneous treatment effect and propensity score. By virtue of satisfying a key (Neyman) orthogonality property, our estimator enjoys central limit behavior---oracle rates with true nuisance parameters---even when estimates of nuisance parameters converge at slower rates. For both randomized trials and observational studies, we establish a semiparametric efficiency bound, proving that our estimator achieves the optimal asymptotic variance. On real datasets where robustness to covariate shift is of core concern, we illustrate the non-robustness of ATE under even mild distributional shift, and demonstrate that the WTE guards against brittle findings that are invalidated by unanticipated covariate shifts.
|
