In the US, the federal government plays a relatively minor role in setting school policy, and the separate states are an important source of policy variation that sets the environment faced by local school districts. The variation in state policies plausibly has a significant impact on student achievement. Little is known about the magnitude of such effects, because data limitations have seldom allowed researchers to specify fully the state
policy environment when analyzing school effects. Differences in overall school policies may, however, help to reconcile the contradict0 y findings about the effectiveness of school resource usage that exist. We develop a simple theoretical model demonstrating that the bias induced by omitting relevant state characteristics is greater in state-level analyses than it is in less aggregate studies. Our exploration of aggregation bias using the High School and Beyond data set suggests that aggregation to the state level inflates the coefficients on
school input variables. Moreover, the results do not support the competing hypothesis that aggregation is beneficial because it reduces biases from measurement error. These results are completely consistent with the findings of production function studies where positive school resource effects an achievement are much much likely to be found when estimation involves state-level data.