FAS Theses and Dissertations

This dissertation studies how the estimation of first-stage nuisance parameters associated with control and instrumental variables affects second-stage estimation of treatment-effect parameters. Chapter 1 approaches the analysis of experimental data as a mechanism-design problem that acknowledges that researchers choose between estimators according to their own preferences. Specifically, I focus on covariate adjustments, which can increase the precision of a treatment-effect estimate, but open the door to bias when researchers engage in specification searches. I establish that unbiasedness is a requirement on the estimation of the average treatment effect that aligns researchers’ preferences with the minimization of the mean-squared error relative to the truth, and that fixing the bias can yield an optimal restriction in a minimax sense. I then provide a characterization of unbiased treatment-effect estimators as sample-splitting procedures. Chapter 2 gives two examples in which we can improve estimation by shrinking in high-dimensional nuisance parameters while avoiding or even reducing the bias in a low-dimensional target parameter. I first consider shrinkage estimation of the nuisance parameters associated with control variables in a linear model, and show that for at least three control variables the standard least-squares estimator is dominated with respect to variance in the treatment effect even among unbiased estimators when treatment is exogenous. Second, I consider shrinkage in the estimation of first-stage instrumental variable coefficients in a two-stage linear regression model. For at least four instrumental variables, I establish that the standard two-stage least-squares estimator is dominated with respect to bias. Chapter 3 (with Alberto Abadie) considers regression analysis of treatment effects after nearest-neighbor matching on control variables. We show that standard errors that ignore the matching step are not generally valid if the second-step regression model is misspecified. We offer two easily implementable alternatives, (i) clustering the standard errors at the level of the matches, or (ii) a nonparametric block bootstrap procedure, that produce approximations to the distribution of the post-matching estimator that are robust to misspecification, provided that matching is done without replacement.

As growing populations continue to drive demand for water, managers of this fundamental resource face the dual challenge of providing both sufficient and clean supplies. In this dissertation, I undertake two analyses exploring a policy aimed at maintaining sufficiency of supply through conservation, and a third analysis evaluating a regulatory approach to promote water quality. Because lawns comprise a large share of residential water demand, water utilities across the western United States offer subsidies to replace lawns with less water intensive landscape. In my first analysis, I estimate the water savings and property value effects of one such subsidy, the Southern Nevada Water Authority's "Cash-for-Grass" rebate program. Using event studies and panel fixed-effects models, I find that the average conversion reduces baseline water consumption by 21 percent and increases property values by about 1 percent, however I find little evidence of property value spillovers to neighboring properties. I also show that the smallest savings coincide with years in which many conversions took place, suggesting a possible trade-off between program participation and maintaining the effectiveness of individual conversions. I also find that participants with high pre-conversion water demand save more water than participants with lower pre-conversion water demand. Conservation subsidies present attractive alternatives to price-based approaches to water conservation. However, I find that a relatively modest 6 percent price increase may have achieved equivalent savings. Finally, combining my water savings and housing price impact results, I show that the program generates net benefits of $2.00 per square foot of desert landscape converted. My results expand our knowledge of water conservation rebates and, more generally, contribute to our understanding of the long-term dynamics of conservation rebate program savings as well as how heterogeneous participant characteristics affect conservation rebate program performance.
In my second analysis, which is joint work with Sheila Olmstead, we study the effect that an information disclosure policy has on national water quality violations. Since the 1980's, information disclosure policies have grown in popularity as a means by which to promote policy outcomes where direct regulation is a challenge. In this spirit, the 1996 amendments to the Safe Drinking Water Act require water utilities to disclose drinking water violations to their customers in annual water quality reports. We explore the impact of these reports on health-based drinking water quality violations in a differences-in-differences framework using a nationally comprehensive data set of water quality violations and water systems. Our results suggest that reports published in local newspapers or mailed directly to customers may have reduced violations, but we uncover less evidence that posting reports online had any impact. We also show that reductions in violations remain stable over time, and that the effect of the reports appears to be stronger for those water systems serving higher-income counties. Finally, we provide evidence that implies reports induce reductions in microbial contaminants without increasing disinfection byproducts. Our analysis is among the first to explore the long-term impacts of information disclosure policies, and builds on the small but growing literature exploring heterogeneity in the responsiveness to these policies. I return to the Las Vegas Cash-for-Grass program in my third analysis. Following the empirical framework and analysis of Bollinger and Gillingham (2012), I estimate the presence of peer effects in the Cash-for-Grass program. Like Bollinger and Gillingham, I find positive peer effects that grow with time when defining the peer network by a zip code, but unlike these authors, I show that only the cumulative conversions rather than the area of conversions drive these effects. Overall, however, my results largely validate the method of Bollinger and Gillingham. Going beyond their results, I also estimate peer effects within zip codes interacted with deciles of assessed home values. At the zip code-decile, I find peer effects to be even stronger than at the zip code, suggesting that a possible driver of the peer effect works through a desire to maintain competitiveness with other homes in an individual's housing market. I also show that my estimated peer effects are at least an order of magnitude larger than the impact of several targeted marketing campaigns administered by the Southern Nevada Water Authority. To my knowledge, mine is the first analysis to compare a peer effect with the impact of advertising efforts. Overall, my dissertation contributes to environmental economists' understanding of water conservation and water quality policies and aims to improve the management of one of civilizations' most critical resources.

This three-paper dissertation studies education systems where decision-making power and funding is shared across different levels of government, their achievement levels and inequality across socieoconomic groups. The first chapter, “Population Density and Educational Inequality: The Role of Public School Choice and Accountability” relates National Assessment of Educational Progress (NAEP) test differences to levels of population agglomeration. Using detailed individual level-data, I establish that being in a denser area is associated with an increase in the socioeconomic gap of about 1 percent for each 10 percent increase in density levels. This is robust to the use of a geographic regression discontinuity design that leverages discontinuities in density across neighboring commuting zones. It is consistent with a theory whereby density leads to greater school socioeconomic segregation. The findings underscore that conventional democratic accountability mechanisms are insufficient to overcome this segregation. In the second chapter, “The Quality and Access Tradeoff in Decentralization Reform: Evidence from Education in the OECD and Spain” (co-authored with Susana Cordeiro Guerra) we study the effect of decentralization reform on education outcomes. While there are theoretical reasons for bringing the locus of decision-making closer to citizens, our knowledge of the effects of decentralization is limited. We first draw on a panel dataset from OECD countries to capture changes on where different decisions are made and their effects on outcomes. Decentralizing has a broadly negative effect on quality indicators but a positive effect on access indicators. We test the argument that upon decentralizing, some regional governments have incentives to assert their legitimacy and pursue highly visible and short-term policies, such as expanding access. Exploiting the exogenous timing of decentralization in Spain (in 1980-2000), we find that variation in the degree of regional assertiveness is related to the magnitude and sign of the effects of decentralization. The third paper, “The Relation between Local Financing and Education Outcomes: Evidence from US School Districts” explores the effects of locally raised revenue. I hypothesize that a higher local revenue share will be related to better school performance, as citizens will be more demanding and in control of the school budget. I look at test scores from the 2007 NAEP, with three empirical approaches. First, I exploit changes in house prices during the 2000 to 2007 boom. Second, I use variation from state fiscal regimes; and third, I leverage covariation within a geographic area. Consistently, local revenue share is associated with about .3-.7% SD higher scores per additional percentage point of local revenue share. It is also associated with greater inequality by socioeconomic status. I find evidence that locally financed districts differ by spending more on teacher salaries and less on administration and that Tiebout competition amplifies the effect of local financing. I find no evidence of a relation between local revenue and citizen participation in education governance. These findings suggest an important behavioral dimension of local financing that would support its continued use.