Fit-Indices are too good PreviousNext
Mplus Discussion > Multilevel Data/Complex Sample >
Message/Author
 Julia Kamischke posted on Thursday, June 09, 2016 - 1:31 am
I'm currently using Mplus 7 (demo version) to perform an Actor Partner Interdependence Mediation Model. The fit indices for my model are good, in fact they are too good.
I'm quite new to Mplus and I don't really know why they are so over the top.

Number of Free Parameters 27

H0 Value -3062.519
H0 Scaling Correction Factor 1.0254
for MLR
H1 Value -3062.519
H1 Scaling Correction Factor 1.0254
for MLR

Akaike (AIC) 6179.039
Bayesian (BIC) 6269.920
Sample-Size Adjusted BIC 6184.364
Value 0.000*
Degrees of Freedom 0
P-Value 0.0000
Scaling Correction Factor 1.0000
for MLR
Estimate 0.000
90 Percent C.I. 0.000 0.000
Probability RMSEA <= .05 0.000

CFI 1.000 TLI 1.000

Chi-Square Test of Model Fit for the Baseline Model
Value 217.002
Degrees of Freedom 14
P-Value 0.0000

SRMR
Value 0.000
 Linda K. Muthen posted on Thursday, June 09, 2016 - 6:19 am
The model has zero degrees of freedom. Model fit cannot be assessed in this case.
 Andrea M Reina Tamayo posted on Wednesday, January 08, 2020 - 9:25 am
I have the same problem with my path models. CFI is exactly 1 and the TLI is greater than 1. SRMR = 0.005, RMSEA = 0.00
However, The chi-square test of model fit indicates a value of 1.523 with 2 degrees of freedom and the number of free parameters is 25.

But does these high values of CFI/TLI indicate that there are problems with my model?
I don't get any errors and the findings are highly in line with the theoretical expectations.

What can I do about this? Do you have any references about TLI greater than 1? or having such good indices?

Thank you in advance!
 Bengt O. Muthen posted on Wednesday, January 08, 2020 - 10:58 am
TLI can be greater than 1 but that should simply be considered as 1, that is, perfect fit according to TLI. This situation of a very well-fitting model can arise when the sample size is low and/or when the correlations are low. Also check on SEMNET.
 Andrea M Reina Tamayo posted on Thursday, January 09, 2020 - 2:51 am
Thank you for your answer Bengt and reference to SEMNET. The sample size was quite ok, 82 clusters with 494 observations. Indeed many of the correlations were low, is that bad?
What I am concerned about is if there is overfit in my model? In this sense, I would be worry about my estimates being of use or value.
Additionally, if I can say that the results can be generalised to the outside population?

Your guidance is appreciated :-)
 Bengt O. Muthen posted on Thursday, January 09, 2020 - 3:58 pm
Having low correlations isn't necessarily bad but it gives less power to reject the model. A chi-square value less than the degrees of freedom is sometimes an indication of overfit.
 Andrea M Reina Tamayo posted on Friday, January 10, 2020 - 4:50 am
I appreciate the response! Good to hear is not bad :-) Do you have citations about this last part you mentioned by any chance?
--> "less power to reject the model. A chi-square value less than the degrees of freedom is sometimes an indication of overfit."
 Bengt O. Muthen posted on Friday, January 10, 2020 - 11:34 am
I can't think of references for this off hand. The power issue is well known and probably has many articles on it and is perhaps also in Bollen's SEM book - ask on SEMNET. The overfitting part is just my own undocumented experience/hunch.
 Andrea M Reina Tamayo posted on Friday, January 17, 2020 - 7:38 am
thank you!
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: