Multiple Imputation and Multidimensionality in Large-Scale Educational Assessments

• Activity Number: 236
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Lauren Harrell*+ and Thomas Belin and Li Cai
• Companies: and University of California, Los Angeles and University of California, Los Angeles/CRESST
• Keywords: multiple imputation ; large-scale assessments ; multidimensional item response theory ; National Assessment of Education Progress

Abstract

For the measurement of population-level achievement, large-scale educational assessments, such as the National Assessment of Education Progress (NAEP), have traditionally relied upon multiple imputation of student scores, drawing five plausible values from the posterior distribution for each content subdomain measured per student. Recent advances in multidimensional item response theory have allowed for more complex models to be investigated from this data with correlated content domains. Posterior distributions of student proficiencies are estimated from the likelihood of item responses under a given item response theory model multiplied by a conditional normal latent regression model as the prior for scale scores. Due to the sparse nature of the assessment data, however, the number of plausible values necessary for adequate coverage of the posterior distribution of scores is examined under both unidimensional and multidimensional item response theory models.