Adaptive mastery multistage testing using a multidimensional model

Mastery testing is used to classify a student as a master or nonmaster. In this chapter, I consider sequential mastery testing (SMT; see, e.g., Lewis and Sheehan 1990; Wainer 1990; Weiss 1983) based on Bayesian decision theory, where the cost of testing is explicitly taken into account. In SMT, sets of one or more items are administered sequentially. After every administration of an item set (also referred to as a testlet), the cost of continuing testing relative to the cost of expected misclassifications is evaluated. If the cost of continuing testing outweighes the expected loss due to a misclassification, testlet adminstration is stopped. So SMT is designed to maximize the proportion of correct classification decisions, while minimizing the total test length. In a simulation study, Lewis and Sheehan (1990) showed that average test lengths can be reduced by half without sacrificing classification accuracy.