Unipolar IRT and Measurement of Vocabulary Knowledge

• In Event: Item Response Theory

Abstract

Unipolar IRT models (e.g., log normal distribution of proficiency, cumulative log-logistic measurement model) use positive latent trait metrics having an absolute zero to represent absence of the construct. We apply unipolar IRT to a vocabulary test and demonstrate how empirically arbitrary IRT modeling choices significantly affect quantification of item/test information.

Authors

• Qi Huang, PURDUE UNIVERSITY

• Daniel Bolt, UNIVERSITY OF WISCONSIN, MADIS