Empirical Methods for Low-Quality Data

Loading...
Thumbnail Image

Date

2020-12-08

Authors

Jerman, Michael

Journal Title

Journal ISSN

Volume Title

Publisher

University of Oregon

Abstract

This dissertation presents methods for economic analysis in settings characterized by sparse data. In the first substantive chapter, I show that difference-in-differences estimators can be biased in the presence of treatment externalities. I then develop a model that accounts for these externalities, and estimate the model using data on Indian river pollution. I show that failure to account for treatment externalities can substantially bias estimates toward zero. I find significant reductions in measured pollution levels in the areas downstream of sewage treatment facilities when compared to untreated areas. Next, I propose a universal method for disaggregating count statistics. The method is able to disaggregate regional statistics such as those collected by censuses or surveys. I demonstrate the algorithm by disaggregating Ugandan census counts of population, tabooda (kerosene lamp) usage, people consuming two or more meals per day, and subsistence farms counted at the subcounty level (the smallest administrative unit reported by the census). Out-of-sample validation suggests that the procedure performs similarly for each statistic and that out-of-sample errors are approximately mean zero throughout the distribution. When combined with nighttime light luminosity data, the disaggregated data can describe within-subcounty distributions of income and poverty. I find that this previously unobserved within-subcounty inequality accounts for 39.3\% of aggregate observed inequality. Next I show that the disaggregated census data can be combined with satellite-derived air pollution data to estimate pixel-level estimates of pollution exposure. I find that 22\% of aggregate inequality in air pollution exposure is caused by within-subcounty inequality in exposure. Finally, I analyze the allocation of environmental resources following India's general election of 1996. Electorally competitive cities in the Ganges Basin during this period were more likely to receive funding for pollution abatement from the federal government of India. These same cities were less likely to receive increased water pollution monitoring. The empirical findings are explained by forward-looking policymakers engaging in clientelism. I emphasize the need for dramatically increased water pollution monitoring along India's rivers and streams.

Description

Keywords

Citation