Latent Regression Modeling in Educational Surveys: A Simulation Study from the National Assessment of Educational Progress (NAEP)

• Activity Number: 156
• Type: Topic Contributed
• Date/Time: Monday, August 3, 2009 : 10:30 AM to 12:20 PM
• Sponsor: Social Statistics Section
• Author(s): Xueli Xu and Yue Jia*+ and Lei Ye
• Companies: Educational Testing Service and Educational Testing Service and Educational Testing Service
• Keywords: latent regression modeling ; educational survey ; secondary bias ; item response model

Abstract

The focus of many educational surveys, such as NAEP and TIMSS, is to report the overall and subgroup student proficiencies. The estimation procedure commonly used is a latent regression model, which is based on student responses to a test and background characteristics. In NAEP, for example, more than 200 background variables and interactions are included in the model. The rationale is that the subgroup proficiency estimates could be biased if those particular group membership indicators are not included in the model. However, a saturated model is subject to overfitting, and little is known about to what extent a model with less predictors can be used that are relatively robust to bias. Simulation studies are performed to examine the more parsimonious models. Two factors are considered: the number of variables included in the model and the number of questions tested per student.