Effect of multigroup CFA on RMSEA PreviousNext
Mplus Discussion > Confirmatory Factor Analysis >
 rgrove posted on Thursday, August 18, 2011 - 12:01 am
I am wondreing whether anyone can help me to understand the impact that a multigroup analysis has on a RMSEA.

The RMSEA values for a two factor model is 0.00 for my three groups when I look at them individually but this is increased to 0.098 when I conduct the multigroup analysis. All other fit indices are within the specified cut-offs.

Is this an artifact of the increase in the sample size? Or does this mean that the model is not interpretable.

Thanks for your help
 Bengt O. Muthen posted on Thursday, August 18, 2011 - 12:03 am
I assume your multiple-group model imposes some invariance constraints on the parameters, so the worsening RMSEA may be a reflection of that.
 rgrove posted on Thursday, August 18, 2011 - 2:38 am
Thanks for getting back to me. The model is the 1st step in that it is the same model across the three groups allowing the factor loadings to vary. I guess I'm wondering whether I am still able to write this up as a good model considering the large RMSEA value?
 Bengt O. Muthen posted on Thursday, August 18, 2011 - 2:53 am
I don't think RMSEA is typically used with multiple-group models; more with correlational models such as EFA.
 Bengt O. Muthen posted on Thursday, August 18, 2011 - 3:25 am
Sorry, I was thinking SRMR. Why don't you post your full set of fit indices?
 rgrove posted on Thursday, August 18, 2011 - 6:40 am
So for my two factor models across the separate groups (df = 1, free parameters = 13, N = 232, 298, 363) the fit statistics are as follows:
Chi-square = 1.8, 0.2, 0.4, RMSEA = 0.00, CFI = 1.00, TLI = 1.01

And for the multigroup (df = 8, free parameters = 34, N = 893) they are as follows:
Chi-square = 30.8**, RMSEA = 0.098, CFI = 0.984, TLI = 0.964
 Bengt O. Muthen posted on Thursday, August 18, 2011 - 2:12 pm
I don't see how you get 34 parameters when your MG model doesn't impose any equality restrictions across groups. How many groups do you have?

Generally speaking, CFI can be good when your variables have high correlations because then the baseline model fit is very poor and the H0 model looks great in comparison.
 rgrove posted on Wednesday, August 24, 2011 - 9:02 am
I have three groups. Have I done something wrong here?

So does that mean that the CFI is not a good fit index to base my interpretations on?

Thanks again for all your help
 Bengt O. Muthen posted on Wednesday, August 24, 2011 - 6:29 pm
That doesn't sound right. You need to send your outputs to support for us to be able to answer.
 rgrove posted on Sunday, September 18, 2011 - 10:48 pm
Am I right in thinking that the CFI and TLI values are 1 due to the model being just parameterized (i.e. df = 1)? If this is the case is this model still interpretable?

Thanks again for all your help
 Linda K. Muthen posted on Tuesday, September 20, 2011 - 4:22 pm
With only one restriction, it is likely that the model will fit well. It is not just-identified so fit can be assessed.
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