Research
A Loanly Road to Teaching: Recruiting Talented Students Through Service Scholarships
Hiring skilled professionals for public service jobs in high-need areas is challenging, especially in teaching. A popular solution is service scholarships—college grants that convert into loans if recipients do not meet the agreed requirements to enroll, graduate, and work in targeted schools for a set number of years. Do these programs recruit additional talented students into teaching? I study the impact of Chile’s free-tuition Beca Vocación de Profesor on attracting high-scoring students to teach in publicly funded schools. Using a fuzzy regression discontinuity design and 14 years of national data, I find that the program increased recipients’ probability of teaching in such schools by 18 percentage points, a lasting effect beyond the service agreement. A complier decomposition shows that this impact was driven by increased enrollment, rather than by lower dropout or improved transition into the teaching workforce. Consequently, 95% of recipients did not fulfill the agreement they signed, and the sizable grant turned into unmanageable debt. I discuss the challenge of addressing default—a persistent concern in loan-based aid programs that remains unaddressed in such programs.
Beyond Exclusion: The Role of the Causal Effect of Testing on Attendance on the Day of the Test (Coming Soon!)
With Christopher Neilson and Magdalena Bennett
High-stakes testing is widely used across educational systems, shaping accountability, resource allocation, and school choice. While skewed attendance may undermine these goals, little is known about its impact beyond the exclusion of low performers. Using an event-study design and rich administrative data from Chile, we examine how testing affects student attendance across grades and performance levels. We find highly heterogeneous effects—ranging from 0 to 10 percentage points across the GPA distribution—with larger impacts on younger students, no negative effects on older students, and positive effects for high performers. We rule out the student exemption policy and test anxiety as key drivers, instead pointing to alternative mechanisms: school-level performance incentives, strategic communication, prior absenteeism, and weakened ties between low-performing students and their schools. Finally, using a machine learning attendance prediction model, we propose imputing scores differently depending on whether absences reflect strategic or non-strategic behavior, aiming to better align school and policymaker incentives while avoiding sizable penalties for disadvantaged schools when absences are unrelated to the test.
Who Showed Up LATE? Counting and Characterizing Compliers and Noncompliers (Coming Soon!)
With Pablo Mones
Counting and characterizing compliers and noncompliers in instrumental variables (IV) settings under the local average treatment effect (LATE) framework is widely used to understand generalizability, treatment take-up, and variation in treatment effects across different instruments. Several identification methods have been proposed to characterize compliers; however, we find that when accounting for overlooked identification assumptions, Abadie’s kappa-weighting scheme is more general and relies on fewer assumptions, though it is more challenging to estimate. Building on this result, we extend the kappa-weighting scheme to account for noncompliers and propose a general approach for counting and characterizing both groups. We then explain how this approach relates to two-stage least squares (TSLS)—the most commonly used IV method—which identifies weighted LATE estimates. Finally, we introduce a partial identification strategy to assess how relaxing monotonicity—an increasingly debated assumption—affects the counting and characterization of compliers and noncompliers. Using this approach, we show that the variation in LATE estimates across instruments in Angrist and Evans (1998) may be explained by a small number of defiers, rather than by differences in complier characteristics.