Date: Thu, 28 Jan 1999 18:08:24 -0500
Reply-To: HERMANS1 <HERMANS1@WESTAT.COM>
Sender: "SAS(r) Discussion" <SAS-L@UGA.CC.UGA.EDU>
From: HERMANS1 <HERMANS1@WESTAT.COM>
Subject: Re: Exact Test, 3x4 untenable?
Content-Type: text/plain; charset=US-ASCII
The number of permutations that have to be tested in a nXm Fisher's
Exact test depends both on the dimensions of the table and the data
(the number of tables, conditioned by the marginal totals, that have
more extreme results than the observed table). The same holds true of
2X2Xk tables (for the Zelen equivalent to the CMH statistic for 2X2Xk
tables). Check the Cytel Web page for details. Mehta and Patel's
StatExact programs use very efficient algorithms to compute exact
statistics for large tables. (I have no financial connection to
Cytel, but have written SAS macroprograms to compute exact statistics
for 2X2Xk tables and from that have learned to appreciate the value of
their programs.) Cytel sells a SAS add-on for exact statistics.
SAS's PROC FREQ handles only the simpler problems. Sig
______________________________ Question _________________________________
Subject: Exact Test, 3x4 untenable?
Author: Undetermined origin c/o LISTSERV administrator
<owner-LISTSERV@VTVM1.CC.VT.EDU> at Internet-E-Mail
Date: 1/28/99 3:38 AM
After posting about someone else' Exact Test, I came across
a situation in SAS.
Using a 2x4 table, the Exact Measures ran in seconds on a
unix HP machine, or seemed so.
I thought I would add another column for the relatively few "Don't Know"
answers to see what the results would be. Not necessary
and does not seem logical for the analysis, but I was curious.
The cpu time went up to almost 50%, where I was dueling it out
with a matlab job, but when the memory usage climbed to 50 MB
I suspected a problem.
When I looked at the log, I got a severe error:
SEVERE ERROR: SAS shutting down due to signal: 15
(I tried to figure out without documentation if this was
from my killing the job or this occured beforehand.)
NOTE: The SAS System stopped processing this step because of errors.
NOTE: SAS set option OBS=0 and will continue to check statements.
This may cause NOTE: No observations in data set.
Is this an infinite loop of some kind? Wonder why the SAS V612
could not exit. I had to kill the process each time after
snooping around and surmising this was not good. It's a good
thing I did not leave the running for hours and was skeptical.
I gather a 3x4 table is too much for the Exact Test then?
Is it possible that the administrator have kept the unix box
to 50 MB's and 50% cpu usage for each machine so the Exact
Test might run and there is nothing really wrong with SAS
but the constraints placed on it? I noticed a steady climb to
50 MB's each time for memory usage.
To get back to me, put "sundor96" before the @ sign, thanks.