My dissertation examines the estimation of treatment effects in presence of plausible data quality limitations related to program participation status. In addition, my dissertation also uses a novel administrative dataset to estimate the impact of air pollution exposure, plausibly widening inequalities in testing conditions, on high-stakes exam performance in Tanzania. My dissertation aims to provide reliable estimates of the impact of the program(s) to help policymakers design cost-effective and potentially welfare-improving interventions. It also informs education and labor market policy on inequalities in high-stakes exam testing conditions due to air pollution exposure, which may add noise to this measure of student ability and lead to suboptimal education and labor market outcomes.

The first chapter proposes a method to consistently estimate the individual and joint treatment effect of overlapping (and exogenous) programs that are plausibly misreported. This chapter provides the asymptotic bias expression of the ordinary least squares (OLS) estimator and shows that it is not possible to determine the direction of the bias a priori. The joint treatment effect may also have an opposite sign to the true effect, which may have dramatic consequences if used to inform policy on whether the programs are complements or substitutes. This chapter then develops a consistent estimator of treatment effects using misclassification probabilities, which may be available through validation studies and other external sources. When misclassification probabilities are unknown, the chapter provides a two-step approach, estimating them in the first step and applying them in the proposed method in the second step. In addition, we present a way to compute the average marginal effects of participating in a single program in this framework with measurement error in multiple binary regressors. Monte Carlo simulations show that the estimator performs well in finite samples and is superior to the naive OLS estimator. Finally, we provide an empirical example, estimating the effect of the Supplemental Nutrition Assistance Program (SNAP) and the Special Supplemental Nutrition Program for Women, Infants, and Children (WIC) on food security and healthy eating using National Household Food Acquisition and Purchase Survey (FoodAPS) data.

The second chapter proposes a method to estimate treatment effects when program participation is plausibly missing. The chapter considers endogenous participation and explores different missing data mechanisms, including missing at random (MAR) and the general case of missing not at random (MNAR). The asymptotic bias expression for complete-case analysis OLS and Instrumental Variables (IV) estimators are provided and discussed. The chapter then proposes a consistent estimator of the treatment effects, the three-step estimator. The first step reframes the missing problem to that of misreporting by assigning missing data to program non-participation status. The second step estimates the true participation status when information regarding participation and missingness is available. The last step uses the predicted participation status to obtain consistent treatment effect estimates. Next, the chapter assesses the performance of this estimator in finite samples through Monte Carlo simulations and compares it with other approaches. An empirical example, estimating the impact of maternal prenatal smoking on birth weight using U.S. Natality data, is provided to illustrate the application of the proposed method in empirical studies.

The third chapter uses novel data on students' performance on national exams administered during secondary schooling in Tanzania to study how air pollution exposure on the day of the exam affects students' performance on these exams. The chapter leverages plausibly exogenous changes in local wind direction in an IV setup to obtain causal effects. IV estimates show that an increase in PM2.5 concentration by 10 µg/m³ on the day a student appears for the exam worsens their performance on the exam by 0.05 standard deviations. These results are robust to a host of falsification checks. The chapter also documents that the effects are pronounced for younger students, girls, students appearing for exams in government-owned examination centers, students in poorer regions, and those at the lower end of the achievement distribution. Further, the chapter provides suggestive evidence that adverse effects of air pollution on exams that test fluid intelligence drive the main results.