Date: Tue, 19 Dec 2000 10:33:35 -0500 "Kilili, Anthony, BMG Direct - Indianapolis" "SAS(r) Discussion" "Kilili, Anthony, BMG Direct - Indianapolis" Re: Predicted Probability and Graph text/plain; charset="iso-8859-1"

James: If you are going to score unknown cases, you may want to save the parameter estimates in a dataset using the outest option as in: Proc logistic data=yourdata outest=betas descending; Model success= x1 x2 x3 x4; Run; Then run the data with unknown cases through proc score as in: Proc Score data=newdata out=scre score=betas type=parms; Var x1 x2 x3 x4; Run; The score= option identifies the location of the parameter estimates, type= option says that we are using parameter estimates for the linear combinations. The out= option creates a dataset (scre) with the scores attached to each case. These scores are actually logits or log(odds of success) (note the descending option in proc logistic, which means we are predicting the probability that success=1). The name of the variable that SAS creates consisting of the logits is likely to be success2 (in version8) or estimate (in version6, need to check this to make sure). To calculate the probability of success (P), we need to solve for P from the logit using a data step as follows: Data new; Set scre; P=1/1+exp(-success2)); Run; I hope this will help point you in the right direction. It is of course advisable to read your friendly manual for more info.

Anthony

-----Original Message----- From: James [mailto:xuj73@HOTMAIL.COM] Sent: Monday, December 18, 2000 4:56 PM To: SAS-L@LISTSERV.UGA.EDU Subject: Predicted Probability and Graph

Is there anyone who knows how to get the predicted probability in logit models? For instance, after I run the proc logistic or catmod, and if I am very interested the probability of success corresponding to specified levels of X variables, how can I get this probability. In addition, is ther anyone who knows how to use these predicted probabilities to get a graph to compare, let's say men's probability of voting with women's across all levels of education with all other x variables held at their means. In addition, if I am interested in the probability of certain specified levels of X variables which is out of the range of the sample data, is there any statement to get that? Thanks a lot.

James Mu Dept of Sociology Indiana University

Back to: Top of message | Previous page | Main SAS-L page