Can the Generalized Graded Unfolding Model Fit Dominance Responses?

Theoretically, the generalized graded unfolding model (GGUM) is more flexible than the generalized partial credit model (GPCM), a dominance model. For item responses generated by the GPCM, the GGUM estimations can generate overlapping item response curves with those from the GPCM over a range of latent trait scores covering almost all of the population. The discrimination and category threshold estimates from the two models are approximately equal. It is necessary to use an informative prior around an extreme location (e.g., 4 for a positive GPCM item) or fix the extreme locations in the GGUM estimation of GPCM items to achieve the desired estimation. The simulation study and the applications on two real datasets support the theoretical claims. Various practical implications are discussed, and suggestions for future research are provided.