There are two contexts of the rate computation to consider. If it is a count from a binomial distribution setting, y successes in n trials, you could run:

DATA upv(keep=y tot) upvr(keep=lvl yy); INPUT y tot; OUTPUT upv; lvl=1; yy=y; OUTPUT upvr; lvl=2; yy=tot-y; OUTPUT upvr; cards; 0 58 ; proc print data = upvr; run;

ods select onewayfreqs BinomialCLs ;

proc freq data=upvr; TABLES lvl / cl binomial(all level=1) alpha=0.05 ; weight yy / zero; title "Binomial Confidence intervals"; run; run;

If it is a count, y, based on a exposure variable, tot, then:

proc print data = upv; run;

DATA tst; SET upv; alpha = .05 ; mean = y / tot ; dfLL = 2*y ; chL = cinv( alpha/2 ,dfLL); LL = chL /(2*tot); IF y = 0 then LL=0; dfUL = 2*(y+1); chU = cinv(1-(alpha/2),dfUL); UL = chU /(2*tot); run;

proc print DATA=tst NOObs; var alpha y tot ll mean ul; format ll mean ul 7.4; title "Exact Poisson interval"; run;

Robin High UNMC

From: Vladimir Grechko <vlad.grechko@MAIL.RU> To: SAS-L@LISTSERV.UGA.EDU Date: 11/11/2011 11:37 AM Subject: What is the upper confidence limit if rate = 0 Sent by: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>

People, I have 0 events over a certain observation period. I need to report Mean (LCL, UCL). My Mean = LCL = 0, but an event _could_ happen, so the upper CL is different than 0. What should it be? How can I calculate it? Thank you!