|
Activity Number:
|
94
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Monday, August 4, 2008 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Social Statistics Section
|
| Abstract - #300874 |
|
Title:
|
Does a Teacher's Value Added Require Data To Be Missing at Random?
|
|
Author(s):
|
Daniel McCaffrey*+ and J. R. Lockwood
|
|
Companies:
|
RAND Corporation and RAND Corporation
|
|
Address:
|
4570 Fifth Avenue, Suite 600, Pittsburgh, PA, 15213-2665,
|
|
Keywords:
|
Bayesian models ; Mixed models ; Teacher effects ; Variable Persistence Model ; Student Achievement
|
|
Abstract:
|
Incomplete data are a common concern for analyses that estimate teacher contributions to learning from longitudinal student achievement data. Students with missing scores are at greater risk for low performance and data might not be missing at random (MAR). We develop models for estimating teacher effects that allow for missing not at random data. We consider a selection model where the number of observed test scores depends on a student's latent general level of achievement and a pattern mixture model that allows the means and covariances of test scores to depend on the student's pattern of observed scores. We fit these models to five years of longitudinal test score data from a large urban school district. These models yield very similar teacher effects as models that assume data are MAR. We discuss likely explanations for the robustness of estimates to assumptions about missing data.
|
