ARC Laboratory Sharing

Differential item functioning (DIF) is a traditional topic of study in psychometrics. Previously, focuses were devoted to study of DIF that occur across manifest groups, which is one form of DIF present in the literature. More recently, researchers turned to investigate another form of DIF that is not due to manifest group difference, i.e. DIF between latent classes, called latent DIF. This article, in particular, focuses on the use of secondary dimension in DIF detection. In other words, it introduced a mixed dimensionality model to explain latent DIF in both statistical and substantive sense.

What is innovative is this study is its treatment of secondary dimension – including it in one of the two latent classes, i.e. the DIF latent class. The computer program LatentGOLD 4.5 was used for model specification by MMLE. The two empirical applications showed the secondary dimension under mixed dimensionality approach proposed in the article can produce a better goodness of fit for both datasets. For example, the results in one application suggest that some less able respondents who made use of an alternative strategy which turned out not to be really helpful. This is an important empirical support to the usefulness of the model since this shows it not only models DIF but also explains DIF. The model also deals with DIF, dimensionality, and local dependence in a common framework, making an unitary and more efficient approach.