Value-added methods & their application to teachers and district decisions about teaching.

摘要

This dissertation project in applied research uses value added models to estimate learning effects in classrooms in one district under two conditions, schools piloting a reorganization and schools not yet under reorganization. The project addresses one of the current impasses in deciding how to take advantage of the variance among teachers provided by the use value-added modeling (VAM) to represent classroom learning effects, against the cold hard truth that a relationship between value-added measures and observable teacher practices, characteristics, or organizational structures based in the literature of teachers and teaching are not well established. Without such a connection, there is no way to say exactly what value-added measures capture or how to begin to suggest a strategy to apply them in practice. One practical way to address the impasse is to move to a different level of analysis, maintaining the valid construction of value added measures at the classroom level, but applying them to questions more properly formed at a district or school level which include teachers as inputs.;Applied to a reform intervention in a low performing district, value-added indicators of classroom learning over the initial years of the intervention suggest modest but meaningful, though inconsistent progress in contributing to student learning outcomes capable of closing learning gaps particularly in the student groups targeted for the intervention. Though program schools and their classrooms remained for the most part low performing, analysis of value added gain scores demonstrated higher than expected growth in program schools and in some cases significantly so. When value added measures of classroom productivity are used to predict student achievement outcomes the following year, as indicators they represent the largest contribution to the explained variance of classroom inputs, but provide slightly different fitting functions depending on the model used. Models predicting gain scores are superior to models predicting status achievement in terms of explained variance and lower standard errors.