First Advisor

Pulos, Steven M.

Second Advisor

Woody, William Douglas

Document Type

Dissertation

Date Created

12-6-2016

Abstract

Value-added models are a class of growth models used in education to assign responsibility for student growth to teachers or schools. For value-added models to be used fairly, sufficient statistical precision is necessary for accurate teacher classification. Previous research indicated precision below practical limits. An alternative approach has been explored in which value-added models are incorporated into composite measures of teacher quality alongside subjective indicators. The aim of the current research was to explore the relative precision of the two evaluative approaches using simulated data. It was found that the composite measure produced fewer classification errors than the stand-alone value-added model. The magnitude of the reduction was largest when sample sizes were small, when composite indicators were highly correlated, and when using conservative alpha levels for hypothesis testing. The magnitude of the difference shrank as an increasing number of aggregated evaluation cycles were incorporated into the evaluation, when composite indicators were poorly correlated, and when less conservative alpha levels were used. Implications for this research are mixed, but there is tentative evidence that there is a tradeoff between precision and resource expenditure when comparing stand-alone value-added versus composite evaluative models.

Keywords

Composite Measure, Monte Carlo, Precision, Value-Added

Extent

168 pages

Local Identifiers

Spencer_unco_0161D_10535

Rights Statement

Copyright belongs to the author.

Share

COinS