A Model-Averaging Approach to Improve the Efficiency of Teacher Value-Added Estimates

• Activity Number: 70
• Type: Topic Contributed
• Date/Time: Sunday, August 4, 2013 : 4:00 PM to 5:50 PM
• Sponsor: Social Statistics Section
• Author(s): Bing Han*+ and Daniel F. McCaffrey and J.R. Lockwood
• Companies: RAND Corporation and ETS and ETS
• Keywords: Model averaging ; Fay-Herriott model ; Small area estimation ; value-added ; teacher evaluation

Abstract

The Fay-Herriot model for small area estimation has been applied to improve the efficiency of teacher value-added estimation. However, constraints such as large amount of missing data in teacher-level characteristics (e.g., qualification, experience, professional developments, prior performance) make it impossible to apply a single best Fay-Herriot model. Researchers are forced to use a set of working models according to the patterns of observed information. The working models are often times misspecified since they are entirely determined by availability of data rather than their goodness of fit to the data. This talk presents a model-averaging approach, which uses a jackknife-based weighting strategy adapted from Hansen and Racine (2012, J. of Econometrics, 167, 38-46) to combine multiple working models to form an efficient model-averaging estimator. This model-averaging approach overcomes the difficulty of comparing working models with different likelihoods, and improves the estimation efficiency for small areas (i.e., teachers with small classrooms). We applied this approach to estimate the value-added of roughly 600 teachers form a large school district.