Decomposition of Estimates in a Layered Value-Added Assessment Model

The layered model of Sanders, et al. (1997) is a value-added assessment model that uses multivariate longitudinal mixed model methodology. This model, whose advantages are reemphasized here, has been criticized for its complexity and lack of transparency. The main purpose of this paper is to provide more transparency for this model by comparing it to three sub-models which provide a decomposition of the layered model estimates. The comparisons make use of simple examples as well as explicit formulas which reveal how the layered model constructs teacher effects from student data. The comparisons reveal how the layered model incorporates past students scores, and future scores when available, to produce estimates with more appealing characteristics than alternative models, particularly in the case when the student population is divided into independent tracks.