In large urban districts, schools enrolling more white students tend to have higher school performance ratings. We use an instrumental variables strategy leveraging centralized school assignment to identify the drivers of the correlation between racial make-up and ratings. Estimates from Denver and New York City suggest the relationship between widely-reported school performance ratings and white enrollment shares reflects selection bias rather than causal school value-added. In fact, value-added in these two cities is essentially unrelated to white enrollment shares. A simple regression adjustment is shown to yield school ratings that are uncorrelated with race, while predicting causal value-added as well or better than the corresponding unadjusted measures.