Edmond Ng posted on Monday, October 10, 2005 - 11:45 am
Dear Linda and Bengt,
Can Mplus fit multilevel structural equation models where responses are non-normal? If so, could you kindly point me to the relevant reference please? Many thanks in advance.
Best wishes, Edmond
bmuthen posted on Monday, October 10, 2005 - 4:46 pm
Yes, you can have non-normal continuous variables using the MLR estimator and also categorical variables using ML. Regarding the latter, references are those for two-level estimation in the multilevel literature.
Edmond Ng posted on Tuesday, October 11, 2005 - 6:34 am
Thank you for your prompt reply. Just a quick follow-up question.
If one takes the TWOLEVEL approach to fit a multilevel SEM in Mplus and the outcome is binary or ordered categorical, what link functions are available to the users? Logit or Probit, or both?
Also it seems under the TWOLEVEL approach, unordered categorical outcome is not allowed. Am I right?
The logit link is available for multilevel models in Mplus when dependent variables are continuous, binary, ordered categorical, or combinations of these variable types. Unordered categorical variables are not allowed for multilevel models in Mplus. They are available for other types of models.