11-07-2011 12:56 PM
I'm wondering if it's possible to calculate anything other than a Wald 95% confidence limits using the surveylogistic procedure. I'm using the Likelihood ratio test to decide whether to retain or remove variables from a model but am reporting the CI's using Wald. Sometimes these aren't consistent,and I'd prefer to report the LRT
Here is my code:
proc surveylogistic data=example;
class a (ref='1');
model z (event=1) = a b c ;
11-07-2011 05:16 PM
Try adding an ALPHA= to the PROC statement with an alpha appropriate for the different confidence limit you would be interested in. The default is ALPHA=.05, so use ALPHA=.10 if you want 90% limits. This should affect all test and limits produced.