1. In Table 2, the two-class solution with item 15 as an anchor item is actually an over-constrained result.
2. The sample size is limited (310 students only). I'm curious how could the parameter estimates be so precise as illustrated in Table 4.
3. What are the uncovered latent classes? Are they qualitatively or quantitatively different? Apparently the proposed model is too powerful to recognize the nature of differences between classes.
4. There was no student classified into Class 2 at Time 1, and the authors did not give any explanation on this. That is, there were not as many latent classes as they imagined. So eventually a parsimony model could be used to fit data by reducing the number of latent classes.