Message/Author 

Lili Garrard posted on Thursday, February 11, 2010  1:07 pm



I'm a beginner user with Mplus. I'm running a CFA with n=169, 5 dichotomous variables and 2 factors(WLSMV). The initial CFA analysis shows that the model is not a good fit. ChiSquare Test of Model Fit Value 13.188* Degrees of Freedom 4** PValue 0.0104 CFI 0.987 TLI 0.981 RMSEA 0.117 WRMR 0.909 After I removed a variable that has weaker association with one of the factors. I got better model fit indices except WRMR. From the readings I did, WRMR close to 1 incicates a good model fit. ChiSquare Test of Model Fit Value 0.828* Degrees of Freedom 3** PValue 0.8428 CFI 1.000 TLI 1.007 RMSEA 0.000 WRMR 0.212 May I ask what I shoud do next? Thank you very much. 


WRMR is an experimental fit statistic. I would not be concerned about it if all of the other fit statistics look good. 


I'm also a beginner with using Mplus. I used Mplus to test 2 models: a mediation model and a moderation model, both with the same ordinal outcome measure. In the mediation model I used a bootstrap and in the moderation model MLR. However, I was wondering, since I want to compare the 2 models on fit, which model fit indices I should and could report comparing these 2 models. It seems that I cannot get similar fit indices: the mediation model, with the bootstrap, only reports WRMR, and the moderation, using MLR, only reports BIC and AIC. furthermore, the WRMR in the mediation model is .003. I thought that the WRMR should be < .09 to be considered an adequate fit. However, the above post stated that WRMR should be around 1 to be considered adequate. Is the latter true? 


WRMR is an experimental fit statistic and I would not use it. With bootstrap we don't give fit statistics. We bootstrap only the standard errors. You can run the analysis without BOOTSTRAP and obtain regular fit statistics. When means, variances, and covariances are not sufficient statistics for model estimation, chisquare and related fit statistics are not defined and are not available. 


Thank you for your quick response. I just have one more question, is it correct to use a bootstrap in the final analyses and report these results (estimates and s.e. etc.), however, also report the fit measure e.g. RMSEA for the same model, obtained without using the bootstrap? 


Yes, as long as you don't say the fit statistics are bootstrapped. 


I am using mPlus version 5. I preformed a CFA with a combination of binary and continuous variables; however, the output only provides RMSEA and WRMR. Should I be concerned that chisquared and CFI are not included? Thank you, Here my input. Usevariables are idugrp_r nidfr15d durhr_r strang1 strang2 strang3 strang4 strang5; CATEGORICAL ARE idugrp_r nidfr15d durhr_r; Missing are all (99999); Analysis: ESTIMATOR = WLSMV; iterations = 100000; BOOTSTRAP = 50; Model: F1 BY strang1 strang2 strang3 strang4 strang5 idugrp_r nidfr15d durhr_r; Output: MODINDICES STDYX; 


It seems odd that you get RMSEA but not chisquare. We don't support Version 5. I would suggest upgrading and if you find the same thing sending the output and your license number to support@statmodel.com. 

Back to top 