Example 5 for PROC LOGISTIC
/****************************************************************/
/* S A S S A M P L E L I B R A R Y */
/* */
/* NAME: LOGIEX5 */
/* TITLE: Example 5 for PROC LOGISTIC */
/* PRODUCT: STAT */
/* SYSTEM: ALL */
/* KEYS: logistic regression analysis, */
/* binomial response data, */
/* PROCS: LOGISTIC */
/* DATA: */
/* */
/* SUPPORT: Bob Derr */
/* REF: SAS/STAT User's Guide, PROC LOGISTIC chapter */
/* MISC: */
/* */
/****************************************************************/
/*****************************************************************
Example 5. Stratified Sampling
*****************************************************************/
/*
Consider the hypothetical example in Fleiss (1981, pp. 6-7) in
which a test is applied to a sample of 1000 people known to have a
disease and to another sample of 1000 people known not to have the
same disease. In the diseased sample, 950 were tested positively;
in the nondiseased sample, only 10 were tested positively. If the
true disease rate in the population is 1 in 100, you should specify
PEVENT= .01 in order to obtain the correct positive and negative
predictive values for the stratified sampling scheme. Omitting the
PEVENT= option is equivalent to using the overall sample disease
rate (1000/2000 = .5) as the value of the PEVENT= option and
thereby ignoring the stratified sampling.
*/
title 'Example 5. Stratified Sampling';
data Screen;
do Disease='Present','Absent';
do Test=1,0;
input Count @@;
output;
end;
end;
datalines;
950 50
10 990
;
proc logistic data=Screen;
freq Count;
model Disease(event='Present')=Test
/ pevent=.5 .01 ctable pprob=.5;
run;