This paper studies how much discretion colleges should have in admissions. We exploit two features of Chile's centralized college admissions system: programs' admission rules are publicly observed, and the assignment mechanism is known. We leverage a reform that expanded programs' discretion by giving them the option to increase emphasis on within-high school rank in admissions. Programs used this discretion very differently: although most could have increased diversity in socioeconomic and gender enrollment, many chose not to. We estimate a model of students' application choices and graduation outcomes that decomposes graduation into selection and heterogeneous causal effects, and we recover programs' preferences over student composition from their observed rule choices. Student–program match explains about one fifth of the variance in predicted graduation, and both students and programs select on it in their applications and admission rules. Mirroring the descriptive evidence, programs differ widely in how much they value match and socioeconomic diversity. Chile's limited discretion performs well against alternatives: test-only admissions and lotteries reduce access or graduation, while full discretion increases segregation.

One way workers can adapt to adverse shocks after entering the labor market is by returning to school later in life, a choice involving substantial opportunity costs in terms of foregone earnings and lost work experience. This paper uses administrative data on earnings and enrollment histories to estimate a dynamic model quantifying these tradeoffs. I first exploit variation from the Great Recession to show that adult workers respond to industry-level shocks by increasing community college enrollment, with heterogeneity across fields of study and industries playing a key role in driving enrollment. The model unpacks what these patterns imply for policies intended to increase worker earnings, such as targeted tuition subsidies. A key contribution of the framework is that it flexibly quantifies multidimensional skills using a data-driven approach, capturing interactions between fields of study and career pathways. A counterfactual tuition subsidy increases enrollment but results in returns that are negative on average, with substantial heterogeneity. Decomposing these returns emphasizes the importance of on-the-job work experience and aligning education with desired career transitions in the design of retraining policies.

This paper studies how the choice of college and major is related to earnings in a context where selection bias can potentially confound direct measurement. Three decades of ranked college application and assignment data for the population of college students in Chile are linked to a decade of administrative tax data on earnings, enabling the study of how student choices, college, and major interact to affect earnings. Ranked order application lists, biased OLS regressions, and a multitude of local RD estimates are tied together in a model of college major choice and earnings outcomes. Estimates from the model indicate that student outcomes are quite heterogeneous – match effects explain a large share of the variance in earnings, much of which is on observable dimensions. Students display a weak preference for these earnings gains, which is strongest for mean returns and observable match components. The results suggest that uniform policy recommendations may be misleading in the context of college major choice, but that targeted program expansions can significantly improve student outcomes without requiring fundamental changes to the assignment mechanism.

We use differences in water chemistry across cities with historical lead piping to identify the causal effects of waterborne lead exposure on cognitive ability. We use the AFQT scores of the universe of applicants to the US military from 1976 to 1991.  We find that childhood exposure to lead pipes in cities with extreme $pH$ levels lowers AFQT scores by at least 1.3 percentiles.  This reduces the present value of lifetime earnings by about \$14,000.  Under conservative assumptions, a policy that removes lead service lines in cities with extreme $pH$ would be highly cost effective, with a marginal value of public funds (MVPF) of 2.8.