School Effects in Chain-Linked Predictive Analytics Models
Robert H. Meyer
The predictive analytics models considered in this memo included school effects for each outcome in each grade-specific equation (although some school effects, such as in the educational model, encompass school effects for multiple grades). We are interested in school effects for at least several reasons. First, schools may differentially contribute to student outcomes, a prime motivation for implementing value-added models. It is arguably important to include school effects in predictions of student outcomes if these effects vary across the different school attended or predicted to be attended by students. Second, it is important to identify which outcomes have the largest school effects, typically captured by the noise-corrected variance of effects. This information can be used to prioritize further investigation into the determinants of positive and negative school effects. Third, information on the magnitude of school effects for each outcome could be helpful in identifying variables that are not measured on the same scale across schools. In other words, large school effects may be evidence that schools measure and report variables on different scales rather than evidence of differences in true school effectiveness.