Sample and Model Selection for Item Response Theory Pre-Equating

• In Event: Roundtable Session 18
  In Roundtable Session: Methodological Advances in Equating

Abstract

Previous research on IRT pre-equating suggests that context effects, item locations, and ability levels of the equating samples (Eignor, 1985; Stocking & Eignor, 1986) may affect the pretest and final form item parameter estimates, which in turn may cause discrepancies between the pre-equating and final form equating result. The proposed study will evaluate to what extent the ability difference, item location change, and context effects impact the pre-equating results and whether the 2PL model is less population dependent than the 3PL model. Data from a large scaled standardized test will be used to illustrate the evaluation/comparison results.

Authors

• Nuo Xi, Educational Testing Service

• Hongwen Guo, ETS

• Hyeonjoo J. Oh, ETS

• Christine M. Mills, ETS