I am evaluating the trajectory classes of psychological distress in a sample of refugees (n=107) measured across 12 time points.
When extracting trajectory classes BIC, theory, and interpretability point to a 4 or 5 class model. However when applying tech 11 to help determine the ideal number of classes, I am unable to secure a significant VUONG-LO-MENDELL-RUBIN LIKELIHOOD RATIO TEST for any number of classes.
Is this likely due to my relatively small sample size?
If sample size is a limitation for this test, does it still make sense to follow the other indicators listed above? Do you have any suggestions for addressing this issue?
I am having a similar problem as Jonathan Codell described above in my latent profile analyses. Do you happen to have a citation for such (i.e., VLMR being affected by sample size)? I have found Nylund, Asparouhov, and Muthen (2007), though sample size is not clearly specified as the reason for comparatively poor VLMR performance. Thank you!
Hi there - I have a somewhat similar question. I previously ran an LCA and the VLMR, LMR, and BLRT were all significant for the 3-class model. I then reran the analysis with the same sample using cluster, strata, and weight. In the weighted analysis, the VLMR and LMR are no longer significant for the 3 class model or 2 model. Other fit indices indicate that the 3 class model makes the most sense. And these outcomes are similar (not the same) as a similar analysis with a different data set. Is it problematic that these change between the weighted and unweighted analysis? Thanks in advance! Nicky