Explore Collateral Information on Math Achievement With Multilevel Item Response Theory

• In Event: Practice and Evaluation in Psychometrics

Abstract

The study demonstrates the use of multilevel item response models (MLIRT) to explore the effect of free/reduced lunch status on math achievement. Results reveal that ⑴ MLIRT models yield larger posterior standard deviations due to accounting for measurement error on math achievement.⑵ MLIRT models show more negative effects of free/reduced lunch status on math achievement than multilevel linear models (MLM) using summed scores as the representation of the math achievement. Given the nonlinear relationship between the latent trait and test responses, informative interpretations could be made at different latent trait levels. ⑶ Moreover, the EAP estimates of item and ability parameters in the MLIRT model were similar to those in the item factor model, but with smaller standard errors.

Author

• Hong Chen, University of Iowa