Files
Abstract
Marginal maximum likelihood estimation (MMLE) is a popular item calibration approach used in the targeted testing design, where missing responses commonly exist due to the testing design structure. Previous research studies investigated the performance of single group MMLE, that is MMLE without using students’ background information, and the multiple group MMLE, that is MMLE using students’ background information, in some large-scale targeted testing designs. However, the practical educational settings often imply small-scale testing scenarios, that is the sample size is small or median. This research investigated the performance of these two versions of MMLE under different small-scale targeted testing designs. A series of simulation studies were conducted across a variety of sample sizes, multiple form designs, number of test forms, number of anchor items, and anchor item selection strategies. Our results shed light on the selection of item parameter estimation in different small-scale targeted testing designs.