A general LC model was used to find out the number of latent class in exploratory way firstly. Secondly, the constrained model--M2PLM (using the same number of class obtained from previous analysis by LC model) was used to fit the data as well. The BIC (or other index) was compared between the two competing model in order to select the ‘better’ model. Then, a step-by-step hierarchical clustering algorithm was used to search for the number of dimension. Finally, the number of dimensions was assessed.
1. LC measures each item on separate dimension, is its parameter estimation stable? If not, the number of the latent class we found out may be inaccurate.
2. Why not just use the mixture M2PLM in one stage rather than two in order to avoid standard error?
3. Finding the number of latent class and number of dimensionality had better being conducted at the same time rather than several stages.
4. I think a serial of simulation studies should be conducted to confirm its efficiency and accuracy of the approach the author came up with before analyzing the real data. Does it perform better than other existing methods?