Using Big Data Techniques to Understand Correlates of Mathematics Achievement in NAEP

• In Event: 61.075 - NAEP Studies SIG Paper Session: NAEP's Role in Educational Achievement

Abstract

This papers develops a high-dimensional statistical procedure for NAEP. By design, NAEP provides proficiency scores for groups of students as opposed to individual students using a multiple imputation (MI) procedure. The MI technique uses student, teacher, and school administrator responses to 1,000s of background questions to predict latent achievement for imputing individual student scores. Rather than employ 1,000s of variables, principal component analysis is used as a data reduction technique, and some variability in proficiency is lost in subsequent calculations. This paper develops a latent Bayesian Lasso regression (BLR) model to improve the precision of predictions for the MI technique and applies the new technique to the 2011 NAEP mathematics data.

Authors

• Steven Andrew Culpepper, University of Illinois at Urbana-Champaign

• Trevor Park, University of Illinois at Urbana-Champaign