My study is on two latent variables at school level, measured at teacher level. I would like to explore changes over time (three measurement occasions). We expect the two latent variables to be related, and would therefore suggest a multivariate approach.
When I try to model this, I get warnings like: *** ERROR in MODEL command The labels %WITHIN% or %BETWEEN% must be specified before class labels.
I have specified MODEL as follows: ------------------------------------- MODEL: %WITHIN% %OVERALL% sk_in WITH sl_in; %BETWEEN% %OVERALL% sk_sch WITH sl_sch; sk_in#1-sk_in#3 ON sk_sch#1-sk_sch#3; sl_in#1-sl_in#3 ON sl_sch#1-sl_sch#3;
See UG ex 10.5 and 10.7 where you see that your Model sl_in also needs to be followed by %Within% or %Between%.
Apart from the UG chapter 10, examples are also given in the paper on our website under Multilevel Mixture Modeling:
Henry, K. & Muthén, B. (2010). Multilevel latent class analysis: An application of adolescent smoking typologies with individual and contextual predictors. Structural Equation Modeling, 17, 193-215. Click here to view figures and syntax for all models.
Alvin posted on Saturday, September 26, 2015 - 8:57 am
Hi Bengt/Linda, I recently ran a multilevel LTA across two time points amongst a dyadic cohort (no. of clusters/dyads=233) with only within-level covariates. From the results, it seems that the estimates for c1 on c2 (e.g c1#1/c1#2 on c2#1/c2#2) are not significant (does this mean that there is no continuity from t1 to t2?). I got the estimates (and ORs) for covariates (on c2#1, c2#2; c1#1, c1#2). I noticed in the multilevel LTA model transitional probabilities were not given. In the multilevel mixture models webnote (2006, p.8), it is stated that "the probability of switching from class 2 to class 1 is 18%", is this based on conditional probabilities? If I were to derive the posterior probabilities for each possible transitional class and look at these elsewhere (e.g. STATA), would this still account for the clustering/dyadic nature of the data?
Which multilevel mixture models webnote (2006, p.8) is that?
Also check the Henry-Muthen (2010) multilevel LTA article on our website.
Alvin posted on Sunday, September 27, 2015 - 9:28 pm
The notes I was referring to are Asparahov and Muthen (2006) - multilevel mxiture models. Following the multilevel LTA, I ran a multilevel multinomial logisitc model with fixed effects based on the conditonal probablties of transitional classes and the results look fine. I wondered would this be OK? Many thanks
The 18% probability you mention above is computed from the joint distribution for the class variables reported in
FINAL CLASS COUNTS AND PROPORTIONS FOR THE LATENT CLASS PATTERNS BASED ON ESTIMATED POSTERIOR PROBABILITIES
If c1 on c2 is not significant I am afraid that there is no continuity. Double check this with LRT test, i.e., fix c1 on c2@0 and compare the log-likelihoods. The LRT test should be better than the T-test I think.
Alvin posted on Wednesday, September 30, 2015 - 11:36 pm
Many thanks Tihomir. I wondered would it be statistically valid to look at the predictors of the derived transitional probabilities in a mixed effect multilevel logistic model (outside Mplus)? Best, Alvin