The Effects of Item Calibration Sample Size and Item Pool Size on Adaptive Testing

Abstract

A simulation study of the effects of varying the item calibration sample size on varying size item pools was run for the maximum information adap tive test. Items were calibrated on the three-pa rameter logistic model on sample sizes of 500, 1,000, and 2,000. Item pools of 100, 200, or 300 items were developed from the three calibration sample sizes. Fixed-length adaptive tests of 10, 15, 20, 25, 30, and 35 items were given to a different group of 500 simulated subjects for each combina tion of item pool size and calibration sample size. Results indicated that high correlations between ability and estimated ability would be obtained in any testing if a sufficient number of items were ad ministered. The reduction of absolute error of abil ity estimation was found to require at least 200 items calibrated on 2,000 subjects.