ALTA in Mplus PreviousNext
Mplus Discussion > Latent Variable Mixture Modeling >
 Katie Witkiewitz posted on Thursday, September 18, 2008 - 9:07 am
I'm trying to conduct Associative Latent Transition Analysis in Mplus, in which I want to jointly condition class membership of some outcome latent class (C) at time t+1 on class membership in some predictor latent class (D) at t+1, as well as class membership for both the outcome and predictor latent classes at time t.

Here is my attempt:

c2#1 on d2#1 d2#2 d1#1 d1#2 c1#1 c1#2;
c2#2 on d2#1 d2#2 d1#1 d1#2 c1#1 c1#2;

Is that correct?
 Bengt O. Muthen posted on Thursday, September 18, 2008 - 1:38 pm
Yes. Make sure c2 follows d2 in the list of classes = .
 R Carta posted on Friday, March 12, 2010 - 4:19 am
Dear Professors

we are trying to run a similar model, but we also trying to condition class membership on the predictor latent class (D) at time t+1 on class membership for both variables at time t.

The Overall model is:

c2#1 on d2#1 d2#2 d1#1 d1#2 c1#1 c1#2;
c2#2 on d2#1 d2#2 d1#1 d1#2 c1#1 c1#2;

d2#1 on d1#1 d1#2 c1#1 c1#2;
d2#2 on d1#1 d1#2 c1#1 c1#2;

However, when we try this model we get an error message:

ERROR in Model command
Invalid ON Statement: D2#1 ON C1#1
The order of categorical latent variables does not allow for this regression.

Can you please help?
 Linda K. Muthen posted on Friday, March 12, 2010 - 5:52 am
The order of the variables in the CLASSES option determines which regressions can be specified. For example,

CLASSES = c1 (2) c2 (2);

allows the regression of c2 on c1 but not c1 on c2.
 Raghav Ramachandran posted on Friday, May 29, 2015 - 2:44 pm
Dr. Muthen,

I am trying to do an ALTA in MPlus. I need to fix certain latent status patterns to zero in the ALTA since very few people have these patterns as their most likely latent statuses.

I've got two time points for each LTA: c1 and c2 with 4 classes, d1 and d2 with 3 classes. I would like to set all associations with d1#3 and c1#1 to zero but since the last class (class 3) of the d1 variable is the reference class, I get an error saying that "references to slopes of the last class are not allowed". I would also like to set certain associations involving a specific combination of all 4 latent variables to zero.

I tried PARAMETERIZATION = PROBABILITY but that does not allow me to regress a latent variable on more than one variable as needed for an LTA (c2 on c1; d1 on c2 c1; d2 on d1 c2 c1;). Please let me know if you have any ideas on how I can set associations involving the reference class of one of the latent variables to zero. Thanks for your help.

 Bengt O. Muthen posted on Saturday, May 30, 2015 - 8:17 am
The FAQ on our website

LTA with Movers-Stayers

shows how to use logits to specify zero/one probabilities. Although it deals with mover-stayer modeling, that FAQ can be drawn on to get what you want.
 Raghav Ramachandran posted on Thursday, July 02, 2015 - 1:14 pm
Drs. Muthen,

I am trying to do a 3-step ALTA to ensure that latent class formation is not affected by the addition of covariates to my model. I am following the steps for the 3-step LTA with measurement invariance (MPlus Web Note 15: Appendix K-N). I noticed in Appendix K that the LCA models at the two time points are run independently of each other (i.e. no c2 on c1 regression).

I have two different constructs, c and d, measured at two time points resulting in four latent variables: c1, c2, d1, and d2. For step 1, should I run the four LCA models independently of each other but with item-response prob. constrained to be equal across (c1 and c2) and (d1 and d2)? I then plan on repeating Appendices L and M for each latent variable. Finally, in step 3, I will include covariates and latent variable regressions (full, independence, cross-sectional, longitudinal ALTA models). Does this approach account for associations between the c and d constructs or should the 3-step LTAs for each construct be combined in the last step? Thanks for your help.
 Tihomir Asparouhov posted on Thursday, July 02, 2015 - 4:18 pm
In the last step you should include any associations between the c and d constructs.

If you want measurement invariance use the approach featured in Appendix K.
Back to top
Add Your Message Here
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Options: Enable HTML code in message
Automatically activate URLs in message