A Simulation Study of Computerized Adaptive Testing with a Misspecified Measurement Model Abstract

George Engelhard, Emory University Computerized adaptive testing provides an approach to measurement in the social and behavioral sciences which can be used to individualize and tailor a unique set of test items for each individual. Computerized adaptive testing methodsprovide a means for selecting and administering test items that provide the best information about the examinee's location on a latent variable. The methods used for computerized adaptive testing are basically sequential designs for estimating a quantal response curve which relates the characteristics of test items to the probability of an examinee responding correctly. Two major measurement models, one proposed by Rasch (1960) and another proposed by Birnbaum (1968), can be used to represent this response curve. The effects of using a one-parameter item response mode i (Rasch) when the simulated responses of the examinees are generated by a two-parameter item response model (Birnbaum) were examined in this study. The results of the simulations suggest that the effects of '%nisspecifying" the measurement mode I using the Rasch mode 1 when the simulated examinee responses were generated by the Birnbaum model can be minimized by using a robust estimate of ability based on the principle of Tukey's biweight which was proposed by Mislevy and Bock (1982). Increasing the number of items administered in the computerized adaptive testing session does not appear to be as effective as the use of the robust estimate of ability.