Date:         Mon, 18 Jan 2010 05:52:15 -0500
Reply-To:     Peter Flom <peterflomconsulting@mindspring.com>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Peter Flom <peterflomconsulting@MINDSPRING.COM>
Subject:      Re: Question regarding PROC SURVEYLOGISTIC
Comments: To: haiyan chen <haiyan_c@YAHOO.COM>
Content-Type: text/plain; charset=UTF-8

haiyan chen <haiyan_c@YAHOO.COM> wrote

>I am running logistic regression to the complex survey data using PROC SURVEYLOGISTIC and got some strange outputs:
>Â
>My program was as follows:
>Â
>Proc surveylogistic data=one;
>Â stratum statement;
>Â cluster statement;
>Â model response (event='1') = predictor;
>Â weight statement;
>run;
>Â
>The predictor has 5 levels (originally were represented by 5 binary variables: level1, level2, level3, level4, and levelm). Before running the model, I've made the desired reference group, level4, as the highest level of predictor by setting predictor=1*level1+2*level2+3*level3+4*levelm+5*level4;
>Â
>Some strange outputs I got showed that, for example, the odds ratio (95%CI)Â for level2Â was estimated 1.50(1.20, 1.99) with a p-value =0.87!
>Â
>I can't come out with a reasonable explanation of this output. I was wondering if Anybody on this list has ever encountered the situation or know why so?


I don't know why you are getting this odd output, but I do know that, if the predictor is categorical (as yours seems to be) then you should use the CLASS statement.  You can then use PARAM = REF and REF = level4 to set the reference to level 4.  See SAS-STAT manual

http://support.sas.com/documentation/cdl/en/statug/63033/HTML/default/statug_surveylogistic_sect006.htm


for the details

Peter

Peter L. Flom, PhD
Statistical Consultant
Website: http://www DOT statisticalanalysisconsulting DOT com/
Writing; http://www.associatedcontent.com/user/582880/peter_flom.html
Twitter:   @peterflom