UCLA

Credible causal inference with observational data critically relies on untestable and extra-statistical assumptions. Researchers have developed well-recognized methods for settings in which they believe those assumptions most likely hold. In this dissertation, I first discuss this notion, arguing that credible causal inference do not need to excessively rely on design templates, and that graphical models oer a powerful way of capturing the main features of empirical research as it is. I then oer two applications of causal inference with observational data in the context of educational stratification in Chile. In a first application, I study the effects of attending exclusively vocational versus exclusively general schools during the first two years of secondary education, relying on selection on observables and parallel trends assumptions. I show that vocational schools negatively affect students’ academic performance and educational expectations, even before the tracked curriculum begins. In a second empirical study, I evaluate the impacts of a recently introduced tuition-free college policy in Chile (2016), using a single-group interrupted time series design and difference-in-differences. I show how the reform, intended to diversify higher education and avoid student debt, unexpectedly benefited high school students, lowering dropout rates especially among the most socioeconomically disadvantaged students.