Empirical investigation of the stability of irt item-parameters estimation

This study examined the effect of various sample sizes (200, 500, 1000, 5000, 10000, and 20000) and test lengths (15, 30, and 60) on the accuracy of item response theory item- parameters estimation using real test data. Estimates of item parameters were obtained by fitting the three-parameter logistic model. The main findings of this study confirmed those findings in previous studies which used simulated data in that longer tests resulted in more accurate estimates of all item parameters across different sample sizes and across different ability levels, especially at ability levels lower than zero. Item difficulty parameter appeared to be the most sensitive to fluctuations in sample size and test length; whereas, item guessing parameter appeared to be the least sensitive. On the other hand, different samples yielded comparable results in terms of accuracy in estimating the three item parameters. Finally, the minimum requirements for accurate parameters estimation tended to be 500 for sample size and 30 for test length. However, sample sizes as small as 200 can still yield acceptable estimates when combined with test lengths longer than 15.

IRT ,

item difficulty ,

item discrimination ,

item guessing ,

parameters estimation ,