DIMENSIONALITY OF THE LATENT STRUCTURE AND ITEM SELECTION
VIA LATENT CLASS MULTIDIMENSIONAL IRT MODELS
F. BARTOLUCCI, G.E. MONTANARI, AND S. PANDOLFI
The former nursing home ranking use unidimensional criteria to classify people，which is a very restricted assumption. If it is violated then the conclusion will be misleading.
The study investigates the dimensionality of the latent structure behind questionnaires about the health conditions of the patients hosted in nursing homes using multidimensional item response theory (IRT) model（2pl）. They also address the issue of item selection on the basis of the discriminating power of the items composing these questionnaires.
The dataset is from ULISSE (“Un Link Informatico sui Servizi Sanitari Esistenti per l’Anziano”—“A computerized network on health care services for older people”),which is about the health status of elderly people.
Steps:
1 Model comparison through model fit evaluation using AIC/AIC3, BIC and CAIC (LC model (Goodman, 1974) and the multidimensional 2p IRT model proposed by Bartolucci (2007).) Then the validation of the model was conducted.
2 A hierarchical clustering algorithm was used for grouping item (Items in the same group are supposed to measure the same dimension and then each group corresponds to a different dimension). The numbers of dimensions are tested.
3 To reduce the number of items on the basis of the estimated discrimination indices.
What is the specific criterion of reducing items (the threshold to be used)?
For the second purpose of reducing the number of items, can we use MCAT？