GMM, Cont distal outcome, time varyin... PreviousNext
Mplus Discussion > Growth Modeling of Longitudinal Data >
Message/Author
 shrihari sridhar posted on Monday, April 23, 2007 - 11:54 am
Hi,

I am trying to estimate this model currently

a) a vector of 3 observed varaibles measured at 5 time points
b) a vector of time invariant covariates influenceing the intercept and slope of the vector in a)
c) a scalar time varying covariate influencing the slope and intercept of of a)
d) a continuous distal outcome measured at time 6
e) a, b and c varying across classes with mixture indicator d)

my questions are

a) i saw an example of the moel with a distal outcome and no time varying covariate in Hix-Small et al 2004. Is it possible to build the model i am suggesting
b) what would be a good reference to a growth mixture model of this nature with a time varying covariate?
c) does the distal continuous outcome pose a problem?
thanks!
hari
 Linda K. Muthen posted on Tuesday, April 24, 2007 - 8:21 am
a. Yes.
b. I don't know of any such paper.
c. No.
 Jungeun Lee posted on Wednesday, April 16, 2008 - 4:07 pm
Hi,

I am hoping to run a growth mixture model which is very similar to the example 8.6 in the Mplus manual. In my model, I'd like to add another distal outcome so that a latent variable 'c' predicts my first distal outcome and then my first distal outcome predicts my final outcome. All observed variables in my model are continuous. Is it possible to run such model in Mplus? If yes, could you point me to examples that I can refer to for a Mplus syntax?
 Linda K. Muthen posted on Wednesday, April 16, 2008 - 5:07 pm
Just remove the CATEGORICAL option from Example 8.6, add the y variable to the USEVARIABLES list, and add y ON u to the MODEL command.
 Anne Chan  posted on Friday, January 22, 2010 - 6:10 am
Hello, I am planning to run a analysis which is exactly like example 8.6 in the Mplus guide, only the distal outcome in my analysis is a continous variable, but not a binary one. May I ask how to do it?
 Linda K. Muthen posted on Friday, January 22, 2010 - 6:14 am
The setup would be identical to Example 8.6 without the CATEGORICAL option.
 Anne Chan  posted on Sunday, January 24, 2010 - 5:46 pm
Thanks. I followed the instruction and got the means for each class. I would like to check if there are any significant differences of the distal outcomes between each pair of classes (not comparing all the classes altogether). May I ask is there a way to do so?
 Linda K. Muthen posted on Monday, January 25, 2010 - 9:04 am
I would use MODEL TEST. See the user's guide for further information.
 Anne Chan  posted on Tuesday, January 26, 2010 - 2:31 am
Thanks, may I ask 2 follow-up questions:

1)Is the model test command used to request Wald chi-square test?

2)Is the Model Test command both applicable to continuous distal outcomes (by comparing the means of the classes) as well as categorical distal outcomes (by comparing the thresholds)?

Thanks a lot!
 Linda K. Muthen posted on Tuesday, January 26, 2010 - 7:37 am
1. Yes.
2. Yes.
 JBP posted on Tuesday, August 31, 2010 - 11:09 am
Hello,
I also want to use the 8.6 example. First question, as my variables are censored, and i don't put Algorithm=integration it's more a LCGA than a GMM that is estimated?
In my model, x has a different status. My aim is to estimate the prediction of U (my distal categorical outcome) by C but controlling for the effect of x to estimate if the prediction of U by C still remains significant after controlling for x. I think i must introduce a regression path: u on x; and a covariance between C and x but I'm not sure how to do that. Would that be a proper specification for the model if there is three latent classes ?
MODEL:
%OVERALL%
i s | y1@0 y2@1 y3@2 y4@3;
u ON x;
c#1 with x;
c#2 with x;

I also want to estimate the same model but with the distal outcome being a count variable. Is it ok to specify: COUNT is U; instead of categorical, will the model run properly.
Many Thanks!
JB
 Bengt O. Muthen posted on Tuesday, August 31, 2010 - 4:31 pm
It is an LCGA if you don't see variances estimated for the growth factors.

I would instead say

c on x.

You can use a count variable u without changing anything else in the model.
 JBP posted on Tuesday, August 31, 2010 - 5:52 pm
Just to confirm the code would be:
u on x;
c on x;
so that the estimate for the path c->u would be controled for the confounding effect of x ?
Thanks for your quick answer!
 Linda K. Muthen posted on Wednesday, September 01, 2010 - 10:13 am
Yes.
 IYH Boon posted on Tuesday, September 09, 2014 - 1:29 pm
I am trying to fit an lcga model that's very similar to the one that JBP describes, above. The idea is to regress a continuous distal outcome, y, onto an indicator of respondents' latent trajectory group, while controlling for a categorical mediating variable, x. I've tried the following Model statement:

Model:
%OVERALL%
i s q | b1 @ 0 b2 @ .1 b3 @ .2 b4 @ .3 ;
y ON x ;

When I run the model, I get the following error: One or more MODEL statements were ignored. These statements may be incorrect or are only supported by ALGORITHM-INTEGRATION.
 Bengt O. Muthen posted on Tuesday, September 09, 2014 - 3:51 pm
You can add

algortihm - integration;

However, "to regress a continuous distal outcome, y, onto an indicator of respondents' latent trajectory group" would seem to call for e.g.

y on b4;

Also having a categorical mediator - I assume between b4 and y - complicates matter. This is because mixtures use ML and ML does not have access to an underlying continuous mediator as is required - see

Muthén, B. & Asparouhov T. (2014). Causal effects in mediation modeling: An introduction with applications to latent variables. Forthcoming in Structural Equation Modeling.

You can try Bayes estimation which does offer that approach - but that is more advanced.
 IYH Boon posted on Wednesday, September 10, 2014 - 11:07 am
Thanks for the quick response; I greatly appreciate it. One follow-up question:

What if my goal was to determine the relationship between latent trajectories (c) and a distal outcome (y) AFTER controlling for an additional set of variables (x). Would the model statement that I used above be correct?
 Bengt O. Muthen posted on Wednesday, September 10, 2014 - 1:01 pm
Yes, but I would probably also add

c on x;

The class-varying intercepts of y would be what you would then focus on.
 JIKIM posted on Wednesday, April 18, 2018 - 5:11 am
Hi
I am trying to do simulation study "GMM with distal outcome". While I was taking Lanza approach (DCON), I got this error.

"" PROBLEMS OCCURRED DURING THE DISTAL VARIABLE ESTIMATION.
THE PROBLEM MAY BE RESOLVED BY INCREASING THE MITERATIONS OPTION.
MORE DETAILED INFORMATION CAN BE OBTAINED BY RUNNING A MODEL
WITH THE AUXILIARY VARIABLE USED AS A CLASS PREDICTOR.""

Actually, I already set the miteration already (50000)

My question is what "more detailed information can be obtained by running a model with the auxiliary variable used as a class predictor".

Lanza approach is take the auxiliary variable(distal outcome) as a class predictor at the first step, then probably "DCON"option already include this, no?
do I need to designate anything?

What I did was
AUXILIARY=(DCON) z;

and also put
[z](1);
z(2);

this in MODEL because in your article it is mentioned this command is needed when the model is with distal outcome.

It would be nice if you leave some advice. Thanks
 Tihomir Asparouhov posted on Wednesday, April 18, 2018 - 4:43 pm
If you are running a simulation study, save the data from the simulation and run the LCA model with Z as a latent class predictor in just one problematic replication. That may give you some more information in where the problem is. That computation is done internally but if you do it separately you may be able to spot the problem, i.e., focus on just that portion of the computation. It maybe the case that the classes are poorly identified or some of the classes are very small.
 JIKIM posted on Thursday, April 19, 2018 - 10:01 pm
Thank you so much for your answer.

I read your article "Auxiliary Variables in Mixture Modeling:
Using the BCH Method in Mplus to Estimate a
Distal Outcome Model and an Arbitrary
Secondary Model"

There, you mentioned simulation conditions as follows:
" We estimate and generate data according to a 4 class LCA model with 8 binary
indicators. The class proportions are as follows: 0.375, 0.25, 0.1875 and 0.1875."


I want to set the same class proportion condition.

[c#1*0.40547];
[c#2*0.69315];
[c#3*0.69315];

Is this right value? Because when I did it I got different proportions...
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: