Publications & Resources

An Adaptation of Stochastic Curtailment to Truncate Wald’s SPRT in Computerized Adaptive Testing

Sep 2003

Matthew Finkelman

Computerized adaptive testing (CAT) has been shown to increase efficiency in educational measurement. One common application of CAT is to classify students as either proficient or not proficient in ability. A truncated form of Wald’s sequential probability ratio test (SPRT), in which examination is halted after a prespecified number of questions, has been proposed to provide a diagnosis of proficiency. This article studies the further truncation provided by stochastic curtailment, where an exam is stopped early if completion of the remaining questions would be unlikely to alter the classification of the examinee. In a simulation study presented, the increased truncation is shown to offer substantial improvement in test length with only a slight decrease in accuracy.

Finkelman, M. (2003). An adaptation of stochastic curtailment to truncate Wald’s SPRT in computerized adaptive testing (CSE Report 606). Los Angeles: University of California, Los Angeles, National Center for Research on Evaluation, Standards, and Student Testing (CRESST).