Dimensionality of the Latent Structure and Item Selection via Latent Class Multidimensional IRT Models
The study consist of four procedures: 1) according to the LC model to determine number of latent classes; 2) in the same number of latent classes of procedure one, to compare fit indices under the multidimensional 2PL model with that under the LC model; 3) evaluating the number of dimensionality under the 2PL model; and 4) deleting items to assess the performance of latent class change. One empirical data was analyzed by aforementioned procedures. Results showed that the number of latent class is 9 according to BIC and CAIC and 5 dimensions with the smallest BIC. Moreover, the performance of latent class change was discussed along with different threshold. It could provide us to evaluate how cost we should pay while the threshold is lenient. The whole study design looks like two stages, first to make sure the number of latent classes, then using the 2PL model to evaluate the number of dimension. I was wondering whether it would be much better if take a unite model (one stage) to determine the number of latent classes and dimensionality simultaneously or not.