Abstract:

The conventional scoring method of forced-choice personality tests produces the ipsative data to which reliability and validity analysis, factor analysis, and analysis of variance (ANOVA) can not apply. In order to overcome these problems, some researchers have come up with item response theory-based (IRT-based) models for the scoring and analysis of forced-choice tests. These models, such as Thurstonian IRT model and Multi-Unidimensional Pairwise-Preference Model (MUPP), can avoid the disadvantages of ipsative data. On one hand, Thurstonian IRT model can be easily used for parameter estimation and it is flexible in model specification, while MUPP has poor expansibility and still needs improvement in parameter estimation method. On the other hand, MUPP has been used to develop forced-choice tests against faking but Thurstonian IRT model is still far from this kind of application. More empirical studies are needed to test the applicability and effectiveness of both models.

Key words: forced-choice tests, ipsative data, Thurstonian IRT model, MUPP