ML-GMM with different within-level va... PreviousNext
Mplus Discussion > Growth Modeling of Longitudinal Data >
 Christopher Poile posted on Saturday, February 25, 2012 - 12:05 pm
Hello everyone,

Is it possible to model a multilevel growth mixture model, if there are two groups, and the within-level variables are different between those two groups?

For example, suppose we have an experiment that has a manipulation which changes the observable variables. Within-level data:

Condition 1:
Round 1: OutcomeA, MeasureA, MeasureB
Round 2: OutcomeA, MeasureA, MeasureB
etc... for 5 rounds.

Condition 2:
Round 1: OutcomeA, MeasureA, MeasureB, MeasureC
Round 2: OutcomeA, MeasureA, MeasureB, MeasureC
etc... for 5 rounds.

The between-level variables will be the manipulation, and other time-invariant covariates such as personality. We predict these will affect OutcomeA, and possibly Measures A, B, C.

I would like to use Multilevel Growth Mixture Modeling because I want to compare the two conditions in the same overall model. This way I can test for manipulation differences in the intercept and slope of OutcomeA, and how those intercepts and slopes are affected by the second level variables.

Is this possible? I can't find a relevant example in the Mplus manual. But I think I might be able to do it by using a modification of 10.10 (Mplus manual) and adding KNOWNCLASS, and combining that with 6.18, 6.12, or 6.16. Would this work?

I greatly appreciate any help you could give me with this perplexing problem!
 Linda K. Muthen posted on Saturday, February 25, 2012 - 1:41 pm
The first group should be assigned missing on MeasureC. If you use KNOWNCLASS, this should work. You should be able to add this to any example. You need to use TYPE=MIXTURE, the CLASSES and KNOWNCLASS options, and add %OVERALL% to the MODEL command.
 Christopher Poile posted on Saturday, February 25, 2012 - 4:49 pm
Thank you Linda, that is very helpful!

Is it okay if I build on this problem? I'm trying to understand what is allowable and not in this type of model. In general, do the same assumptions and restrictions about LGC models need to hold? For example, does the time span between OutcomeA's need to be the same across conditions?

To clarify, suppose a round lasts 5 minutes. OutcomeA is a measure of the behaviors person 'A' can do in 5 minutes. In Condition1, person A is allowed to do behaviors every round, and there are 5 total rounds:

Round 1 (5 mins): OutcomeA, MeasureA, MeasureB
Round 2 (5 mins): OutcomeA, MeasureA, MeasureB
...for 5 rounds.

But in Condition2 there are twice as many rounds in total. OutcomeA is measured only 5 times (to match Condition1) but there is a round in between OutcomeA and the other measures:

Condition 2:
Round 1 (5 mins): OutcomeA
Round 2 (5 mins): MeasureA, MeasureB, MeasureC
Round 3 (5 mins): OutcomeA
Round 4 (5 mins): MeasureA, MeasureB, MeasureC
...for 10 rounds.

Can I combine the 10 rounds of Condition2 into 5 rounds to match Condition1? For ex:

Condition 2:
Round 1 (10 mins): OutcomeA, MeasureA, MeasureB, MeasureC
Round 2 (10 mins): OutcomeA, MeasureA, MeasureB, MeasureC
for '5' rounds?

 Linda K. Muthen posted on Sunday, February 26, 2012 - 11:07 am
The time scores can be different as long as the growth shape is the same in both groups, that is, linear, quadratic, etc.
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