Witold, this fit index is explained in detail in Ching-Yun Yu's dissertation (p. 17):

https://www.statmodel.com/download/Yudissertation.pdf

WRMR is suitable for models where sample statistics have widely disparate variances and when sample statistics are on different scales such as in models with mean and/or threshold structures. Therefore, WRMR is always reported when data are categorical. In this dissertation, Yu investigates different cut-off values under normality and nonnormality, continuous and binary data.

Yu found out that a cutoff value of 0.95 or 1.0 has acceptable type I error rates.

Regards, Karin

Dear Karin, 

I am comparing fit indices of three alternative, nested models. However, contradicting RMSEA, CFA, TLI, and chi-square 'getting better', WRMR is increasing (up to 1.400). Could this be due to a fact that my predictors and outcome variables are continuous and then WRMR is not the best fit index to look at? 

Regards, Witold

Witold, well, rule-of-thumbs for cut-off criteria are often problematic. I need some more information: Are all your variables in the model continuous? Are the variables normally distributed or not? Are there any categorical variables? Do you have missing data, and if yes, how do you deal with them? Do you have a large number of df or a small number?

I am not sure if I can help you, but I hope to get an idea concerning WRMR's strange behavíor.

Regards, Karin