for comments and discussions. I especially thank Michael Schwarz for outstanding research assistance. Abstract Classroom education has public good aspects. The technology is such that when one student disrupts the class, learning is reduced for all other students. A disruption model of educational production is presented. It is shown that optimal class size is larger for better behaved students, which helps explain why it is difficult to find class size effects in the data. Additionally, the

more »

... e of discipline is analyzed and applied to differences in performance of Catholic and public schools. An empirical framework is discussed where the importance of sorting students, teacher quality and other factors can be assessed. 3 There is a large literature here. An early empirical paper is Henderson, Mieskowski and Sauvageau [1977]. 4 Hanushek [1998b] reports that expenditures per student more than doubled between 1960 and 1990 at a time when there was no steady trend in test scores or other measures of performance. Other recent papers to examine the relation between expenditures and class size are Card and Krueger [1992] and Betts [1996]. There is evidence that competition both lowers costs and improves school performance. See Hoxby [1996, 1998], McMillan [1999] , and Urquiola [1999]. 2 the class size puzzle. The basic structure begins with the recognition that education in a classroom environment is a public good. But as with most public goods, classroom learning has congestion effects, which are negative externalities created when one student impedes the learning of all other classmates. There is empirical support for this proposition. Peer effects have long been recognized as crucial in education. 3 While hardly novel, to understand peer interaction effects it is necessary to embed the spillovers in a framework where changing the size of a class or its composition has a cost. The primary cost takes the form of teacher salary and infrastructure. The answer to the class size puzzle rests on the realization that class size is a choice variable and the optimal class size varies inversely with the attention span of the students. It is efficient to use fewer teachers and a higher student-teacher ratio when the students are better behaved. But an envelope theorem implies that actual educational output varies directly with the behavior of the student, despite the fact that fewer teacher inputs are used. An implication is that class size matters, but the observed relation of educational output to class size may be small or even positive. 4 The main purpose of the model presented here is to tie together a wide variety of facts and to integrate the literature on class size and student performance. A further goal is to provide a new