Abstract
In computerized adaptive testing pretest items are presented in conjunction with operational items to renew the item bank. Pretest items are calibrated, and possible differential item functioning (DIF) is analyzed. Some difficulties arise due to the large amount of missing responses, which can be avoided by the use of fixed item parameter calibration (FIPC; Kim, 2006) methods. In this study, we applied the multiple weights updating and multiple EM cycles method, with response imputation (as suggested by Lei, Chen, & Yu, 2006) and without response imputation for non-applied items. The IRT likelihood ratio test (IRT-LRT) was used for DIF detection. The manipulated factors were type of DIF, DIF size, impact size, test length, and sample size. The results showed that the FIPC method is suitable for detecting large-size DIF in large samples. In the presence of impact the use of imputation led to a bias in the effect-size measure of the DIF.