

Model fit indices with count outcome 

Message/Author 

Xiang Liu posted on Friday, May 06, 2016  7:39 am



Hi, I am modeling count outcome using a negative binomial regression on one latent factor and several observed covariates. It seems that besides the AIC and BIC, Mplus doesn't provide any absolute mode fit index. Any suggestions? Thanks! 


Chisquare and related fit statistics are not available for this model. In this case, nested models can be compared using 2 times the loglikelihood difference which is distributed as chisquare. Nonnested models with the same set of dependent variables can be compared using BIC. 

BL posted on Wednesday, September 18, 2019  1:36 pm



I've estimated a model with 1 predictor, 3 mediators, and 2 latent outcome variables (Y1, Y2); Y2 is informed by 4 manifest count variables (nb). I thought chisquare fit would not be generated because the unit of the latent outcome variable is a count, but several chisquare indices were calculated. Are these valid for describing absolute fit? If so, can you clarify about which is most appropriate to report? ChiSquare Test of Model Fit for the Count Outcomes** Pearson ChiSquare Value 2547.717 Degrees of Freedom 7973 PValue 1.0000 Likelihood Ratio ChiSquare Value 1029.021 Degrees of Freedom 7973 PValue 1.0000 ChiSquare Test for MCAR under the Unrestricted Latent Class Indicator Model for the Count Outcomes Pearson ChiSquare Value 25.589 Degrees of Freedom 799 PValue 1.0000 Likelihood Ratio ChiSquare Value 6.937 Degrees of Freedom 799 PValue 1.0000 


No, these are not overall tests of model fit. Such a test has not been developed. 

Back to top 

