How big is big enough? Sample size requirements for CAST item parameter estimation

Abstract

Adaptive tests offer the advantages of reduced test length and increased accuracy in ability estimation. However, adaptive tests require large pools of precalibrated items. This study looks at the development of an item pool for 1 type of adaptive administration: the computer-adaptive sequential test. An important issue is the sample size required for adequate estimation of item response theory item parameters. The authors simulated responses of 300, 500, and 1,000 respondents per item, estimated item parameters with the BILOG program, and then evaluated the adequacy of the parameter estimates. The results suggest that sample sizes as small as 300 respondents per item are adequate for estimating ability and classifying examinees as masters or nonmasters. Copyright © 2006, Lawrence Erlbaum Associates, Inc.