Multilevel categorical data PreviousNext
Mplus Discussion > Multilevel Data/Complex Sample >
 adel powell posted on Saturday, December 17, 2011 - 4:05 pm

I am trying to generate multilevel categorical data. I am trying to study various aspect of that type of data. I am having problems. My data is ordinal but non-normal. I am getting this error:
*** ERROR in MODEL command
Variances for categorical outcomes are not allowed on the within level.
Variance given for: Y1

Can Mplus not do multilevel modeling on categorical non-continuous data?

I was thinking Mplus can treat the data like Logistical multilevel modeling where the within error is estimated by pie/3. I forgot the exact figure?
 Bengt O. Muthen posted on Saturday, December 17, 2011 - 4:57 pm
Yes, Mplus can do this. But with categorical outcomes there is not a free within-level variance parameter so you can't estimate/mention that in the MODEL command, which is what the error message complains about. As you say, with logit link the residual variance is pi-square/3, that is, a fixed quantity, not a free parameter. Mplus uses that fixed quantity implicitly.
 Cecily Na posted on Friday, March 01, 2013 - 8:29 am
Hi Linda and Bengt,
I have a categorical outcome (three levels) in a two-level model. The
model syntax is as following, (A, B are predictors at the within
level, C is the categorical outcome, and D is the predictor at the
between level). Is the following syntax correct? Why is the model
syntax for categorical outcomes the same as for continuous outcomes?


In a multilevel model for continuous outcome, the intercept or slope of the
within level outcome is modeled as random and treated as a latent
factor at the between level. For categorical outcome, what is modeled
as random?
Thanks a lot!
 Linda K. Muthen posted on Friday, March 01, 2013 - 9:53 am
The model you specify above is a random intercept model both for a continuous or categorical outcome. See Example 9.3 where one outcome is continuous and one outcome is categorical.
 Matthew Porter Wilcox posted on Saturday, March 26, 2016 - 11:16 am
the Output for my analysis has also returned:
"Variances for categorical outcomes are not allowed on the within level."

Is there a way to estimate a free parameter for item variances for the within level using ordinal data in version 7?

I am trying to estimate the within-level reliability of a 12 item scale, and need residual variance estimates for each item to do so.
 Linda K. Muthen posted on Saturday, March 26, 2016 - 12:34 pm
Variances of categorical variables are not parameters in the model. See the following paper which would deal with this:

Raykov, T., Dimitrov, D.M. & Asparouhov, T. (2009). Evaluation of scale reliability with binary measures using latent variable modeling. Forthcoming in Structural Equation Modeling.
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