A mixture cross-classification IRT model for test speededness
MetadataShow full item record
Previous research has shown that under time limits, items near the end of the test appear harder for the speeded examinees than for the non-speeded examinees (Bolt, et al.,2002; Oshima, 1994). Moreover, speeded examinees tend to omit more items near the end of the test (Bolt, et al., 2002; Cohen, et al., 2002). Therefore, certain person and item characteristics related to test speededness may help to explain the differences in estimates of examinees' abilities and item diffculties. The test speededness models proposed thus far have focused on modeling speededness effects rather on attempting to explain them. In addition, the investigation of differential speededness is typically implemented in a two-step procedure. First, the measurement model is used to identify speeded groups, generally as latent classes. Next, a statistical analysis is done to examine characteristics of members of speeded and non- speeded groups. The purpose of this dissertation was to propose a mixture multilevel IRT model with person and item covariates that could be used to detect test speededness effect in paper-and-pencil test. Unlike the regular IRT models which treat persons as random and items as fxed, however, this dissertation treated both as random, making it possible to add item covariates into the model. De Boeck (2008) has shown that treating items as random not only makes sense theoretically, but also is promising for identifying DIF items and for explaining differential item diffculties. A multilevel mixture IRT model was developed in this dissertation for use in detection of speeded and non-speeded latent classes. Covariates are illustrated as being incorporated into the model for use in helping to characterize members of each latent class.