Comparison of Item Selection Algorithm in Computerized Adaptive Testing with a Time Limit

Abstract

This study compared several item selection algorithms in a situation where a time limit exists in terms of estimation accuracy and item bank security. Simulation results show the following: (1) adjusting item information using the expected response time will improve both the estimation accuracy and the item bank security; (2) when item discrimination and expected response time are positively correlated, the mean numbers of the answered items decrease particularly in the algorithms which determine the next item according to the information; (3) under the condition in (2), algorithms that suppress the inflation of overlap rate demonstrate better estimation accuracy and; (4) the root-mean-square error and test information show slightly different results between algorithms.