Item-Rest Regressions, Item Response Functions, and the Relation Between Test Forms

Abstract

Levine (1982) used item-rest regressions for the estimation of item parameters, and this relation ship was exploited in this research in the context of vertical equating. Results from a simulation and an empirical dataset were used to demonstrate that item-rest regressions were useful in verifying the relationship between two tests obtained from item parameter estimates. It is shown that in vertical equating designs the Rasch model cannot replicate the relationship between tests at the lower score levels when guessing is present. At higher score lev els, however, the correct transformation function can be estimated, irrespective of the IRT model used.