Selected work in progress
with Carmen Vargas
This paper investigates the “brother penalty”, the hypothesis that firstborn daughters experience worse educational outcomes when their younger sibling is male rather than female, in Peru. Using administrative data linking secondary-school records to post-secondary outcomes for 32,834 firstborn daughters in two-child families, I estimate a three-factor structural model of cognitive ability, math self-efficacy, and Spanish self-efficacy separately on a Sister subsample and a Brother subsample, and use a split-sample counterfactual design to quantify how much of the Brother–Sister gap in sequential educational choices is attributable to differences in the factor distributions. The analysis yields three main findings. First, the factor-distribution shifts are domain-specific: cognitive skills and math self-efficacy are negatively affected by the presence of a brother, while Spanish self-efficacy is affected positively (although not significantly), a result that is difficult to reconcile with symmetric resource-competition mechanisms. Second, the brother penalty operates at the intensive margin of institutional choice rather than the extensive margin of educational progression: high-school graduation and college-enrollment rates differ by less than half a percentage point across subsamples, but daughters with brothers are 1.7pp more likely than daughters with sisters to enroll in a public rather than a private institution. Third, the public-college gap exhibits a suppressor pattern: equalizing the factor distributions widens rather than closes the Brother–Sister gap (now +2.9pp), consistent with families steering daughters toward lower-cost public institutions for reasons other than measured ability when the household also contains a son. These findings shed light on the fact that family environment continues to shape gender gaps in education even in settings where formal access is equal.
Working papers
with Wesley T. Morris, Sarah Montana, and David Blazar
Revise and Resubmit, Education Finance and Policy
Recruiting locally connected individuals has gained policy attention under recent “Grow-Your-Own” initiatives, yet evidence linking teacher localness to student achievement and retention is limited. Using statewide Maryland data on teachers’ high school enrollment, postsecondary training, prior school employment, and certification, we examine three dimensions of localness: growing up locally, engaging in local teacher training, and working locally before teaching (e.g., as an instructional assistant). Value-added models with student fixed effects show that certification and licensure pathway, not localness itself, is the strongest predictor of early-career effectiveness. Teachers with formal university-based preparation in Maryland outperform young conditionally certified teachers in math by 0.06 standard deviations—even though conditionally certified teachers are the most likely of any certification group to have grown up in-state. Local upbringing is only modestly and imprecisely associated with achievement gains, while high school teacher-preparation pathways show positive effects comparable to university-based preparation. Former aides, who have strong local ties and are more racially/ethnically diverse than other novices, perform similarly to peers. Retention analyses show a different pattern: local upbringing and local training most strongly predict remaining in the same school or district. Local recruitment complements, but does not substitute for, high-quality preparation.
with Anika Alam and Jing Liu
Despite the rapid adoption of AI-powered learning tools in K–12 classrooms, there is limited evidence about their effectiveness in improving student learning. This paper examines the usage of Goblins—a teacher-mediated multimodal AI math platform that allows students to type, draw, and speak while receiving personalized feedback—and how that usage associates with middle school math achievement. The study took place in a large Arizona school district where Goblins was offered to all middle grade math teachers in the 2025–26 school year. Using a weighted propensity design, we find that Goblins usage is associated with a 0.11 standard deviation increase in MAP math scores within the district where students in the highest usage group demonstrate a growth of 0.31 standard deviations. These results contribute to emerging research that examines the extent to which sustained engagement with AI-powered tutoring can improve math achievement.
with Miguel Benítez-Rueda and Eric Parrado
Revise and Resubmit, World Development
Mobility restrictions have the potential to accelerate automation by increasing constraints on the use of labor. Using a triple-difference empirical strategy, this paper estimates the impact of pandemic-related restrictions on labor market outcomes in Peru, focusing on differences across occupations with varying exposure to automation risk. Our results indicate that mobility restrictions significantly decreased employment rates (-13.5%) and hourly labor earnings (-64%) for workers in highly automatable jobs up to 18 months after the shock. These effects were particularly pronounced among women, small and medium-sized firms, and informal workers, and were present across the skill distribution, with the largest effects among medium-skilled workers. Evidence suggests that occupational mobility from high-risk to low-risk jobs drove recovery afterwards. These patterns are consistent with firms adjusting to disruptions in labor availability by reducing reliance on routine-intensive tasks and highlight how labor supply shocks can interact with automation risk in developing economies.