I am conducting a 3-step LTA for continuous variables a 2 time points (pre-treatment and post-treatment) with a moderating variable (treatment condition). The best fitting LTA model according to the log-likelihood and adj BIC is a 4 class solution.
However, I get a warning message for a non-positive definite product matrix and another stating that a number of logit parameters were fixed to prevent singularity. The fixed parameters are regression coefficients for transitions with empty cells. In total I have 6 (out of 32) transition paths that are empty and 7 with just 1 person transitioning. The model estimation terminated normally.
Can I ignore these warnings or is there something fundamentally problematic about empty cells in LTA? If so, why might this be the best-fitting model, and what can I do to get rid of the warnings?