A Continuous a -Stratification Index for Item Exposure Control in Computerized Adaptive Testing

Abstract

The method of $a$ -stratification aims to reduce item overexposure in computerized adaptive testing, as items that are administered at very high rates may threaten the validity of test scores. In existing methods of $a$ -stratification, the item bank is partitioned into a fixed number of nonoverlapping strata according to the items’ $a$ , or discrimination, parameters. This article introduces a continuous $a$ -stratification index which incorporates exposure control into the item selection index itself and thus eliminates the need for fixed discrete strata. The new continuous $a$ -stratification index is compared with existing stratification methods via simulation studies in terms of ability estimation bias, mean squared error, and control of item exposure rates.