Longitudinal Growth Mixture Models PreviousNext
Mplus Discussion > Growth Modeling of Longitudinal Data >
 elizabeth coyne crowe posted on Tuesday, April 22, 2008 - 11:31 am

I have been working on a GMM of elementary student vocabulary scores at preschool, first, third, and fifth grade and although the model runs, I am consistenly receiving error messages that indicate that something is wrong with the slope.

I am using W scores which are the Rasch-based scores for a commonly used measure of reading. My stats in Mplus look accurate (means, etc).

I have also tried adding a quadratic trend, but the error remains the same.

My syntax is as follows:

FILE IS picvo_no_out.dat;
FORMAT IS f6.0, f3.0, f3.0, f3.0, f3.0;
CLASSES = c (2);

Any advice would be greatly appreciated.
 Linda K. Muthen posted on Tuesday, April 22, 2008 - 12:35 pm
It is impossible to make any suggestions without seeing the input, data, output, and your license number at support@statmodel.com.
 Sar posted on Tuesday, August 21, 2012 - 2:31 am
Hi, I have been working on estimating a growth mixture model for disordered eating and have developed a four class model that best fits the data.
When I add predictors, the trajectories and number of people in each class changes. I understand that the additional variables helps predict the classes however I plan to publish a paper now looking at trajectories (and outcomes) of disordered eating and a different paper later in the year looking at brain data and another looking at additional predictor variables that I have not yet analyzed. It does not make conceptual sense to put all of the predictors in the one paper however I am worried that when I add extra variables it will continually change the classes. Is it reasonable to fix the classes using the parameters that have been estimated so that when I add predictors it is predicting the original class membership?
Thanks for your help!
 Bengt O. Muthen posted on Tuesday, August 21, 2012 - 7:55 am
The changing of classes when adding covariates is discussed in

Muthén, B. (2004). Latent variable analysis: Growth mixture modeling and related techniques for longitudinal data. In D. Kaplan (ed.), Handbook of quantitative methodology for the social sciences (pp. 345-368). Newbury Park, CA: Sage Publications.

which is on our web site.

If you want the classes to be determined by only the latent class indicators and not also by the relationship between the predictors and the indicators, you have to take another approach. This is the "3-step" approach that is part of the forthcoming Mplus Version 7. See the Hopkins handouts at


Part 2, slides 36 and onwards.
 L. Siemons posted on Wednesday, July 31, 2013 - 8:24 am

I did run a 3 class quadratic GMM analysis and I saved the CPROB to be able to see a person's final class assigned. However, I get many columns in the output of the CPROB and I don't know the meaning of all of them. An example of 1 person:
3.260 2.658 1.749 2.102 2.083 1.000 3.454 -0.884 0.131 3.454 -0.885 0.131 0.000 1.000 0.000 2.000

The first 5 values are the scores that the person had on my output measure: 3.260 2.658 1.749 2.102 2.083

Next is the person ID: 1.000

The last 4 columns are the posterior probabilities of the person for each class (0.000 1.000 and 0.000), followed by te final assigned class membership: 2.000

However, I don;t know the meaning of the 6 columns in between: 3.454 -0.884 0.131 3.454 -0.885 0.131

Can you please explain the meaning of these columns to me? Does it have anything to do with the persons intercept or slope?

Many thanks!
 Linda K. Muthen posted on Wednesday, July 31, 2013 - 9:34 am
See the end of the output where it shows what is saved and the order it is saved.
 Hillary Gorin posted on Tuesday, December 13, 2016 - 4:04 pm
Hello Dr. Muthen,

I have a database containing longitudinal data for over 2000 participants. I am hoping to run a growth mixture model but I know that I have a large amount of missing data, over 50%.

Will it be possible to run this model using five time points with over 50% of the missing data?

 Bengt O. Muthen posted on Tuesday, December 13, 2016 - 6:27 pm
If you have 50% missing already at time 1 I would not recommend doing the growth analysis because it will rely too strongly on model assumptions.
 Hillary Gorin posted on Tuesday, December 13, 2016 - 9:28 pm
Hi Dr. Muthen,

Ok thank you for the information. Do you think imputing data would be helpful? If I imputed data, could I run such an analysis (with confidence in my findings)?

 Linda K. Muthen posted on Wednesday, December 14, 2016 - 6:48 am
The same issues hold for imputing data with that much missingness.
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