Youth who share a school and neighborhood often show similar levels of academic achievement, but some studies find all or most of this correlation is due to sorting (not causation). We analyze the National Education Longitudinal Survey (NELS) in three ways to decompose sorting versus causality: We first control for much richer measures of family background than other analysts have used. We next use characteristics of the students' future high school as an instrument for family background (as future high school quality is correlated with unobserved family background but cannot cause junior high test scores). Finally, we use regression and semi-nonparametric matching methods to look at changes in test score when youth change schools. The results create a collage of evidence that a significant fraction of the correlation is causal. (Contains 4 tables.)