The Rasch model plus guessing

Abstract

The need for equating is one of the principal reasons for using an item response model rather than the classical model to guide test construction. Use of the Rasch item response model in test construction makes equating especially easy. The Rasch model, however, does not have a guessing parameter, whereas most tests that require equating contain multiple-choice items subject to guessing. This limitation of the Rasch model has led many test developers to use a three-parameter model that accounts for guessing, though at the expense of increased difficulty in equating. Despite the Rasch model's limitations with respect to guessing, some test developers still use it in multiple-choice testing. This article examines the use of the Rasch model in multiple-choice testing, particularly with respect to the problems of goodness of fit and equating. Although goodness of fit turns out to be a problem small enough to ignore, modification of the Rasch model to account for guessing may be necessary to ensure accurate equating.