Noncompensatory Multidimensional Item Response Theory Models: An Application of Assessing Math Ability in English Language Learners

• In Event: 14.046 - Poster Session 1
  In Poster Session: 14.046-6 - Contemporary Models and Their Applications: A Poster Session

Abstract

Accurate estimation of English Language Learners’ (ELLs) math ability may be confounded with linguistically complex word problems. Since high math ability cannot compensate for low English language proficiency (ELP), a non-compensatory model of MIRT is most appropriate; however, the most commonly employed MIRT model is compensatory. This simulation study, therefore, aims at investigating bias in ability estimates resulting from applying the compensatory model onto underlying non-compensatory data. Findings show that examinees like ELLs, being high on math but low on ELP, are systematically underestimated with respect to their high math ability. Such a finding supports the use of non-compensatory models for estimating ELLs’ math ability when linguistically complex items are present.

Authors

• Janine Buchholz, German Institute for International Educational Research

• Joseph A. Rios, Educational Testing Service

• Johannes Hartig, German Institute for International Educational Research