The Studies On Stratified Termination Rules And Item Selection Strategies In Computerized Adaptive Testing

Computerized adaptive testing (CAT) combines item response theory andmodern computer techniques, which creates a new field of psychological andeducational measurement. It has clear advantages over the traditional paper-penciltesting in many aspects including shorter and more accurate tests, more flexible examtime, more topics forms and more friendly in environment. It is subject to more andmore people’s attention and favor of the theoretical research and practical applicationssince it was introduced into the testing field in the early1970’s. However, it alsoraises new issues and challenges. It has important theoretical and practicalsignificance to solve these problems and challenges.There are two topics in the thesis: stratified termination rules and strategies ofitem selection.This paper studies the stratified termination rules in variable length CAT whichhas few studies currently. Based on dichotomous scoring and a-stratified CAT, theauthor proposes a new stratified termination rule and promotes it to polytomousscoring CAT. The Monte Carlo simulations show that the new rule is better than thecurrent stratified termination rules.Item Selection Strategy constitutes the key content of CAT, which not onlymakes tests intelligent but also exerts a direct impact on the quality of test. Based ondichotomous scoring and variable length CAT, the author proposes two new itemselection strategies. The Monte Carlo simulations show that the first topic fixes superbin controlling item exposure rate and the second strategy hits a fabulous balance interms of efficiency, accuracy and exposure control.