Jan Fax posted on Friday, September 19, 2008 - 5:47 am
3 questions how to relate the Mplus ex10.6 output to the parametric ML-LCA approach suggested by Vermunt (2003).
For the cluster-level covariate w, there is an effect of .79 on the factor capturing the between-variation of the within-level latent class intercepts.
Q1: Can this estimate be used to answer a question like "how does w affect the log odds of belonging to (within-level)latent class 1 vs. 2"? (such an interpretation is given in the vermunt-article, example on "educational research", effect of the school-level variable CITO, page 232).
Q2: If so, how?
Q3: Can I estimate intraclass correlations to check whether the lc intercepts differ substantially between groups? How?
where slides 44-49 go through an example and slide 49 discusses the level 2 effect in 3 different ways. The example concerns a random intercept. With a continuous level-2 covariate w you can consider a 1 SD change in w and how much higher or lower the class probabilities are due to this change. Such probabilities can then be converted into odds.