An Investigation of Procedures for Computerized Adaptive Testing Using Partial Credit Scoring

Publisher Website
Google Scholar
Abstract
The purposes of our investigation were to manipulate systematically various aspects of the computerized adaptive testing (CAT) procedure for partial credit scoring and to determine the effects of the manipulations on the operational characteristics of the CAT. We examined the effects of item-pool size, item-pool information, and stepsizes used along the trait continuum until maximum likelihood estimates could be calculated. We found that item banks consisting of as few as 30 items could be used successfully for partial-credit adaptive testing. The use of a variable stepsize method for trait estimation prior to maximum-likelihood estimation and the use of information functions to select items virtually eliminated the problem of nonconvergence of trait estimation. The implications of these findings for educational applications are also discussed.

This publication has 5 references indexed in Scilit: