Seminars

We develop a unified framework to disentangle school effects and peer effects on student achievement. Identification leverages exogenous tie‐breaker lotteries in centralized school choice systems. To preserve statistical power while accounting for the potential non-linearity of peer effect, we summarize treatments and confounders parsimoniously: the full distribution of peers’ baseline scores is captured by functional principal components (FPCA), and selection on observables is addressed by conditioning on expected peer distributions and school‐specific admission probabilities; lottery–based deviations from these expectations provide instruments for realized peers. Applying the approach to New York City middle schools, we find a non-monotonic and non-linear peer effect: performance varies with the gap between classmates’ and a student’s own baseline scores, exhibiting both a rank effect and positive spillovers in parts of the distribution of peer baseline test scores. When we estimate school effects net of peers, the cross-school dispersion of value added rises by about 20\% relative to conventional VA measures that omit peers, indicating that peer composition compresses VA estimates.