1. The number of latent class was decided by BIC, AIC & CAIC, usually by BIC. The accuracy and consistency of the index depends on various conditions. It is not desirable to find as many as 9 latent classes.
2. The proposed strategy aimed at assessing the number of dimensions and to reduce the number of the items was unusual because the selection of latent classes, validation of the IRT model and number of dimensions are separated. I don’t know what are the logic in these procedure and whether it is applicable for other dataset.
3. What is manifest probability?