Instrumental variables (IV )estimatesS chool improvement efforts often focus on inputs to education production, the most important of which is staffing ratios. Parents, teachers, and policymakers look to small classes to boost learning. The question of whether changes in class size have a causal effect on achievement remains controversial, however. Regression estimates often show little gain to class-size reductions, with students in larger classes sometimes appearing to do better (Hanushek 1995). At the same time, a large randomized study, the Tennessee STAR experiment, generated evidence of substantial learning gains in smaller classes (Krueger 1999). An investigation of * Angrist: