The analysis begins with a basic description of the data, e.g., histograms of expulsions and suspensions, sample covariances between test scores and demographic measures. The main body of the paper explores implications of a specific theory of peer effects.

While studies suggest that peer effects have an impact on measures of student performance, the specific mechanism by which they operate remains unclear. One explanation is that disruptive behavior by a student impedes the learning of every other student in a classroom. If so, then systematic efforts made by teachers and educators to reduce disruptive behavior ought to lead to higher test scores, other things being equal. The paper examines an implication of this basic theory. Taking suspension rates and expulsion rates as proxies for discipline, one could attempt to determine the effect of discipline on test scores by means of a simple ordinary least squares regression. The immediate difficulty with such an approach is that there is apt to be correlation between the error term and the regressors. One would expect higher rates of expulsion at schools whose students come from troubled or dysfunctional social environments and have behavior problems because of unobserved factors. These students might perform poorly on tests because of the same unobserved traits that affect the rate of expulsion.

The paper attempts to distinguish between the two sources of correlation, endogenous and exogenous, by constructing a panel of test scores, suspension rates, and expulsion rates for ninth graders in California, school by school, using data from 1998 and 2000. The analysis includes both fixed-effect and between-effects estimates of the coefficients on expulsion and suspension, the assumption being that neighborhood effects for a given school changed little in two years.

The analysis indicates that when one does not control for school-specific unobserved effects, higher rates of expulsion are associated with lower math scores. The paper argues that this is evidence of endogeneity: schools whose students have been unobservably disadvantaged by their local environment exhibit more behavior problems and also lower math scores. In the fixed-effects regression, however, the correlation is positive. Holding constant the school, an increase in expulsion rates between 1998 and 2000 was associated with an increase in math scores. The strictness of discipline policy, then, might have a positive effect on test scores. The result is merely suggestive, however: the analysis was not robust to heteroskedastic specifica