Abstract
This paper relies on a large-scale field experiment in Mexico to measure peer effects while in middle school. Conditional on initial academic performance, we group students following two different models: tracking à la Duflo et al. (2011) and bimodal classrooms à la Carrel et al. (2013). Relative to the randomization of students to classes in the control group, treatment assignment expands the support of peer ability composition, which allows us to estimate different models of peer effects on academic ability. We thus experimentally measure the impact of both group formation strategies and compare it to the effect predicted by these models. We identify important and similar average gains in student performance in both types of allocations. Even though the impact of both models is greater among high scorers, there are no losses among low performers. Gains persist over time, particularly among top performers. Our simulations roughly replicate the reduced form treatment effects, but also highlight that, even under extreme random variation in peer composition, it is risky to rely on observational data to predict the impact of grouping strategies.
Local Organizers: Vincenzo Scrutinio, Alessandro Sforza