Two groups analysis in growth models PreviousNext
Mplus Discussion > Growth Modeling of Longitudinal Data >
Message/Author
 Anonymous posted on Friday, August 27, 2004 - 6:16 pm
Hello. I am doing some analyses using the latent intercepts (i1, i2) and slopes (s1, s2) of two variables to predict some alcohol use outcomes. I used TYPE = RANDOM MEANSTRUCTURE MISSING.

In this framework, I want to know whether the effects of i1, i2, s1, s2 on alcohol use are the same across two groups of adolescents: low & high emotionality.

My initial thoughts are: I could do a two-group analyses systematically constraining the paths from the latent variables (e.g., i1, s1) to the outcomes (e.g., early onset of alcohol use)in the two groups to be the same and then gradually release the constraints. The Chi-square difference test between the constrained and unconstrained models will tell me whether the effects of the latent variables on the outcomes are the same between the two groups.

Is this an acceptable approach? I would appreciate any feedback or suggestion. Thanks a lot.
 bmuthen posted on Friday, August 27, 2004 - 6:22 pm
Yes, this is a good approach. And, probably the easiest way to study such interactions.
 Sylvana Robbers posted on Wednesday, May 14, 2008 - 8:52 am
I am doing growth curve analyses with two groups (GROUPING option) and I would like to test whether the intercepts and the slopes of the two groups are different.
Can I do this by constraining like this:

model group 1:
[i] (1);
model group 2:
[i] (1);

Then, can I compare the BIC value of this constrained model with the original model? Or do I need to look at other fit indices or maybe chi-square? So, how do I test whether two intercepts or slopes are different between groups?

Another question: Is it a problem that one group has N=1358 whereas the other group has N=16038?

Thanks in advance.
 Linda K. Muthen posted on Wednesday, May 14, 2008 - 3:58 pm
You could compare those models using chi-square difference testing or loglikelihood difference testing.

The large difference in the sample sizes may affect the results.
 Sylvana Robbers posted on Thursday, May 15, 2008 - 8:44 am
Thank you.

I read your instructions on this website about chi-square difference testing. Unfortunately in my output I don't see the scaled chi-square value, or the scaling correction factor, probably because I use ML now. However, if I use MLR I get the following error:

THE STANDARD ERRORS FOR H1 ESTIMATED SAMPLE STATISTICS COULD NOT BE COMPUTED. THIS MAY BE DUE TO LOW COVARIANCE COVERAGE. THE ROBUST CHI-SQUARE COULD NOT BE COMPUTED.

With ML, I get:

Chi-Square Test of Model Fit

Value 217.044
Degrees of Freedom 57
P-Value 0.0000

Chi-Square Contributions From Each Group

SING 65.367
TWIN 151.677

Chi-Square Test of Model Fit for the Baseline Model

Value 4927.082
Degrees of Freedom 53
P-Value 0.0000

Loglikelihood

H0 Value -59972.855
H1 Value -59864.333

Also, I obtain CFI/TLI/BIC/AIC/RMSEA.


What would you recommend me to do to test for group differences in i and s?
 Linda K. Muthen posted on Thursday, May 15, 2008 - 4:16 pm
You can use MODEL TEST or do a difference test with ML which does not require a scaling factor.
 Sylvana Robbers posted on Friday, May 16, 2008 - 7:24 am
Thank you for your swift response!
 Sylvana Robbers posted on Tuesday, May 27, 2008 - 7:51 am
I have a following question related to my previous posts. With the analysis I want to perform (see first post), you say that the large difference in group size may affect my results. Do the group sizes need to be exactly the same, or is there a range like maximum 1,5 times larger?
Do you have a reference on multigroup comparisons in growth modeling with chi-square testing?

Thanks in advance for your time.
 Linda K. Muthen posted on Tuesday, May 27, 2008 - 2:13 pm
I don't know exactly what the proportion of the group sizes should be. I don't know of any papers that have examined this.

I think the Bollen book probably covers multiple group comparison using chi-square difference testing.
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: