proc logistic

[deleted_user]

Hello,

It's me again. 😞
I've done some proc logistic for a while.

I have seen on some outputs (alledgedly) some nice tables with the observed responses on the left and the predicted responses on the right. In each cell, there is the frequency of people taking the corresponding observed and predicted responses.
If the model adjusted the data perfectly, the table should be diagonal.
So it's a good measure of the quality of the model.
How can I do that ?

For the moment, I use the "Association of Predicted Probabilities and Observed Response" with the "Percent Concordant" thing.

Thank you !

[Olivier]

What you are looking to is called "classification table". You should try the CTABLE option (in the MODEL statement options) ; this is not really cross-tabulated, but contains all information you need.
You can also create predicted observations with the OUTPUT statement, then create a format with PROC FORMAT to transform continuous predicted probabilities into binary predictions (e.g. 0-0.5="0" 0.5<-1="1") and use both table and format in a FREQ procedure.