ABSTRACT
Evaluating the effectiveness of school resources involves making causal inferences about the relationships between school inputs and academic achievement. Researchers must assess achievement outcomes with and without school inputs using administrative data on students, absent experimental manipulation of the resources.
This chapter explores three different statistical methodologies available for causal inference. Cross-sectional regression offers informative insights into potential causation, but rigor is limited by the absence of the time dimension. We cannot assess a fundamental requirement of causation: that outcomes change in response to varying levels of school inputs.
More plausible causal inferences can be made if both outcomes and inputs are measured over time, because we can then test whether a given school resource is related to changes in the achievement from one year to the next. Such value-added models, as used in this study, incorporate change over time.
Finally, another limitation of administrative data is the omitted student characteristics that might affect outcomes. The use of student fixed effects regression models controls for unmeasured student characteristics (like academic ability) that are constant over time. In this way, we can estimate the most plausible causal inferences in the absence of an experimental design.
