ML estimation versus WLS estimation PreviousNext
Mplus Discussion > Structural Equation Modeling >
 Giovanni Piumatti posted on Thursday, September 15, 2016 - 3:45 am

I am testing a longitudinal cross-lagged model with both ordinal (with four categories) and continuous variables. My sample size is very big (N > 10,000) and my variables are not normally distributed. Is it ok to use maximum likelihood estimation for such a model?


 Bengt O. Muthen posted on Thursday, September 15, 2016 - 10:53 am
The answer to that seemingly simple question has many layers of complexity. First, what do you mean by ML when you have ordinal and continuous variables? Do you mean treating all variables as continuous? Or do you mean specifying the ordinal ones as categorical? The latter can also be done using ML. Second, in the latter ML analysis the mediating ordinal variables (say time 2 vbles) are treated as continuous when they are used as predictors. WLSMV and Bayes can instead use their continuous latent response variables which might be a better alternative. See also our FAQ:

Estimator choices with categorical outcomes
 Bengt O. Muthen posted on Thursday, September 15, 2016 - 10:55 am
Third, if the ordinal variables don't show strong floor or ceiling effects, you might want to claim that they are continuous - and use ML.
 Talma Claviger posted on Friday, September 07, 2018 - 4:17 am
is it possible in Mplus to analyse a correlation matrix with WLS?

 Bengt O. Muthen posted on Friday, September 07, 2018 - 2:25 pm
No because you need more information for the weight matrix of WLS to get SEs and chi-square. With only a correlation matrix, you can pretend the variables are continuous and use ULS without SEs and chi-2.
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