This study proposed an approach to investigating dimensionality of the MDS questionnaire and discriminating power of the items by using a latent class multidimensional 2-PL model. This procedure started at the exploration of the latent class (LS), followed by a comparison between the LS model and the multidimensional 2-PL model, an examination of dimensionality, and reductions of items. The authors stated that this approach simultaneously deals with the issues of dimensionality, latent class, and item evaluation of a clinical measure.
However, some statements in this paper are not very clear to me. For example, why not studying the number of latent classes and dimensionality simultaneously? Are there any reasons to explain why the investigation for latent class should be conducted before the examination of dimensionality? Second, it is not clear why the authors chose BIC and CAIC results, ignoring inconsistent results from AIC and AIC3 (listed in Table 2)? Third, Table 4 and Table 5 suggested that the estimated ability parameters of the first class supported the authors’ statements, but those values of the other classes were not very consistent across five domains. Therefore, it is not appropriate to make a strong statement about patients’ characteristics of individual class. Finally, it seems to be uncommon to use discriminating power only for item reduction, ignoring item fit statistics.