proc freq: lower 95% limit on risk difference greater than upper 95% limit

MRB3855 — Tue, 04 Dec 2018 14:36:05 GMT

Good day: Here is my data (x_data):

Treat outcome count

A x 10
A y 0
B x 0
B y 20

proc freq data=x_data;
title "Agresti-Min (Exact);
tables treat*outcome /riskdiff(cl=(exact));
weight count;
exact riskdiff(method=score2);
run;

The output (attached) gives an Exact 95% Confidence Limit on the difference as 0.7000 to 0.0000. Other methods within proc freq don't do this (they all give 1.000 as an upper limit). I recognize some methods may fall apart when count=0...but, by definition, the lower limit can't be greater than the upper limit. Is this a known error or what?

Re: proc freq: lower 95% limit on risk difference greater than upper 95% limit

MRB3855 — Mon, 17 Dec 2018 16:45:21 GMT

I raised this issue with SAS and they responded as follows:

"R&D have had a closer look at the issue and agree this is a new bug and will be raising a Defect and SAS Usage Note.
The issue will be fixed in a future release of SAS."

As you were 😉

topic Re: proc freq: lower 95% limit on risk difference greater than upper 95% limit in Statistical Procedures

proc freq: lower 95% limit on risk difference greater than upper 95% limit

Re: proc freq: lower 95% limit on risk difference greater than upper 95% limit