Jon Heron posted on Monday, February 17, 2014 - 3:57 am
before the release of version 7, Mplus only produced one classification matrix. Back then I wrote a (now obsolete) piece of Stata code to calculate the other matrix (Vermunt's D-matrix) and also the logits so I could implement the three-step method.
I've just dusted off my Stata code to run it on some Monte Carlo output and discovered that it does not work. I cannot calculate the second classification matrix from the first (and vice versa), although I can derive the second matrix from the logits.
Any idea what is going on here? are there a different set of rules for MC output? perhaps one matrix is averaged across draws.
Please send the output from the Monte Carlo analysis pointing out which matrices you are concerned with.
Jon Heron posted on Tuesday, February 18, 2014 - 1:12 am
Thanks Linda, I've just posted some details to support.
I have a further quick question on the topic of Monte Carlo simulations. If the analytical model is identical for the two approaches, would we expect the same answer when performing a one-stage MC compared to a two-stage in which all datasets are exported using REPSAVE?
Either there is good reason for differences - akin to the old fscores/factor-determinacy problem - or I am suffering from class switching when analysing my REPSAVEd data.