Quasi Complete Separation PreviousNext
Mplus Discussion > Categorical Data Modeling >
 James Algina posted on Tuesday, March 15, 2016 - 11:29 am

Would Mplus check for quasi-complete separation for the following model:

Nominal are Dv;
Dv on Trt Site TxS Z1 TxZ1 Z2 TxZ2 ;

Trt are Site effect codes for dichotomous variables. Z1 and Z2 are effect codes for a factor (in the ANOVA sense) with three levels.

No warning was issued by Mplus.

I also estimated the model in SAS which issued the warning "There is possibly a quasi-complete separation of data points"

There are three Dvs in my data set. This issue only comes up for one of The Dvs. For the others estimates by Mplus and SAS agree.


 Bengt O. Muthen posted on Tuesday, March 15, 2016 - 5:42 pm
No check - but it would seem that would influence the SE calculations.
 James Algina posted on Friday, March 18, 2016 - 7:34 am
Hello Bengt,

Thank you for your comment. I had estimated the model using the complex option because of clustering in the data. The standard errors were not remarkable, although certain effects were somewhat extreme. Dropping complex resulted in identification warnings related to the extreme estimates. So perhaps, for models with categorical outcomes in addition to using the complex option to address clustering, an analysis without the clustering may be useful to help detect quasi-complete separation.

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