UC Santa Cruz

This thesis consists of three projects based on the application of Regression Discontinuity Design (RDD) analysis to questions of public economics. Chapter 1 evaluates the effects of English Language Learner programs on student achievement. There is a significant and well-documented academic achievement gap between native English speakers and English language learners (ELLs) in US public schools. This gap is particularly large in California, the US state with the largest population of ELLs. As a result of being classified as having limited English proficiency, ELLs follow a different track through the primary and secondary public school system in the U.S., according to federal requirements and implemented at the state level. The effect of this differential treatment of non-native English speakers is ambiguous, and the division of students into separate tracks based on English lacks support in existing literature. I use longitudinal microdata from a large northern California school district to estimate the effects of ELL programs on student outcomes. Students in California are categorized as ELLs based on performance on a standardized assessment of English language proficiency, the California English Language Development Test (CELDT). Students classified as are eligible for enrollment in a bilingual programs. For primary school students, this is the Alternative Language Acquisition (ALA) program. I exploit discontinuity in the probability of enrollment in ALA at the threshold CELDT score for English proficiency. I find that schools do not observe the rule for assignment to ALA based on CELDT score.

Chapter 2 evaluates whether consumers respond to information about product quality in the context of automobile safety. Consumer choice over new vehicles is a function of multiple vehicle attributes, including price, fuel efficiency and safety. However, because vehicle safety is often correlated with other characteristics of vehicle quality, estimating consumer preferences for safety over other attributes is empirically difficult. Using a federal program in the United States that provides public safety ratings for new passenger vehicles, I exploit discontinuity in the assignment of 5-star vehicle safety ratings in continuous probability of injury measurements calculated based on crash test performance. I evaluate whether new vehicle models that just miss a star threshold on the National Highway Transportation and Safety Administration's New Car Assessment Program's 5-Star Rating scale experience lower national sales volumes relative to vehicles that just exceed the ratings threshold.

Chapter 3 evaluates the accuracy of the New Car Assessment Program (NCAP) safety ratings in predicting real-world crash outcomes. Regulatory policies such as safety standards for seat belts or airbags are aimed at improving vehicle safety and reducing occupant injury and fatality rates, but may distort driver and occupant behavior. Vehicle safety ratings can provide standardized, transparent and comparative information to buyers. However, the value of any safety rating to consumers depends on the accuracy of the rating regime. In the context of transportation policy, this means real-world loss of life, injury and property damage on a national scale. In the U.S., the National Highway and Transportation Administration (NHTSA) evaluates the safety of all new vehicles sold in the United States and publishes these safety ratings via the New Car Assessment Program (NCAP), a program which is emulated internationally. Using a novel dataset with the continuous underlying running variable, probability of injury, used to calculate the 5-Star NCAP safety rating seen by consumers, I evaluate whether U.S. NCAP safety ratings accurately predict real-world crash outcomes in terms of vehicular damage, personal injury and loss of life. Matching NCAP rating with crash report data from the U.S. Fatality Analysis Reporting System and Texas' Crash Records Information System, I find minimal correlation between NCAP rating and real-world crash outcomes.

In summary, this dissertation applies Regression Discontinuity Design analysis to public education policy in the context of the CELDT and to regulatory and transportation policy in the context of NCAP. Successful implementation of these policy regimes depends on accuracy of classification, whether of students or vehicles. Failures of evaluation and enforcement result in distortionary effects with real-world effects on educational attainment and motor vehicle safety. Regression Discontinuity Design provides a powerful tool for evaluating the true efficacy of public policies.