Performance of the Modified Continuous a-Stratification Indices in Computerized Adaptive Testing

Abstract

Computerized adaptive testing (CAT) has become increasingly popular for many purposes, including educational assessment in schools, personnel recruitment by institutions, and clinical diagnosis in hospitals. Practically, it is critical that items are not overexposed because exposure can translate into sharing with other examinees, which might threaten the validity of test scores. A popular method for monitoring item exposure is a-stratification. When removing old items from an item bank or adding new items to the item bank, the optimal strata for a-stratification would need to be obtained by having additional simulation studies. It is undesirable for practitioners in high-stakes testing when an item bank needs to be updated frequently. The continuous a-stratification index (CAI) not only avoids partitioning the item bank but also monitors item exposure; however, the CAI still yields a high percentage of overexposed items. Therefore, this study proposed three modified CAI methods for monitoring item exposure and compared their performance in item selection with some CAT item exposure control methods.