Model fit indices for a logistic regr... PreviousNext
Mplus Discussion > Categorical Data Modeling >
 Nicole Watkins posted on Friday, August 30, 2019 - 7:40 am
I am trying to conduct a multiple group logistic regression, and I am getting weird model fit statistics. Is my model underidentified?

Here is the input:

y1 x1 x2 x3 x4 x5 x6 x7 x8
x9 x78 x79;

GROUPING = x1(0=male 1=female);

ITERATIONS = 100000;

y1 ON x1 x2 x3 x4 x5 x6 x7 x8
x9 x78 x79;

And here is some of the output:


Number of Free Parameters 22

Chi-Square Test of Model Fit

Value 0.000*
Degrees of Freedom 0
P-Value 0.0000

Chi-Square Contribution From Each Group

MALE 0.000
FEMALE 0.000
RMSEA (Root Mean Square Error Of Approximation)

Estimate 0.000
90 Percent C.I. 0.000 0.000
Probability RMSEA <= .05 0.000


CFI 1.000
TLI 1.000
 Bengt O. Muthen posted on Saturday, August 31, 2019 - 5:33 pm
This model is "just-identified" so that no overall model test of fit is available. You can test if there is gender equality. The way your model is now written, you let males and females have different estimates.
 Nicole Watkins posted on Friday, September 06, 2019 - 12:05 pm
Dr. Muthen,
Thank you for your response.
I am interested specifically in the path y1 ON x78. x1-x5 are control variables in the analysis and I do not have any theoretical reason as to why they should be unconstrained. Should I then, in service of having a more parsimonious model, have the paths y1 ON x1-x5 constrained, and then compare the model that also constrains the y1 on x78 path? I disagree with a colleague on this topic. I believe that I should constrain the control variable path but they believe I should leave it unconstrained and only try constraining the path I am interested in and test the difference. Is there a good article/book that would walk through this sort of decision?

Thank you for your insight!
 Bengt O. Muthen posted on Saturday, September 07, 2019 - 4:35 pm
I think you could argue both ways depending on the situation. You want to test gender invariance for one of the paths. The best way to do that is in a model that is close to the correct model so that other sources of misfit don't play in. So if you really think/have evidence that the control variable influences are gender invariant, you should hold them invariant. If you don't, holding them invariant starts the testing of what you are interested in using a model that is too far from the correct one.
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