Abstract
Test score measures of teacher quality may not fully capture teachers’ impact on students. We use test score and non–test score measures of student achievement and behavior to estimate multiple dimensions of teacher quality. We find that these two measures of teacher quality are only weakly correlated and that both affect students’ high school performance. A teacher removal simulation that uses both measures improves most long-term student outcomes by over 50%, compared to a policy that uses test scores alone. Our results also show that for high school outcomes the effects of teachers in later grades are larger than those in earlier grades and that performance in core elementary school subjects matters more than that in other subjects.