[18822] in s-news-athena
[S] Error Rate Error in SPlus 2000 Discrim
daemon@ATHENA.MIT.EDU (Marc Feldesman)
Mon Aug 2 17:50:09 1999
Message-Id: <4.2.0.58.19990802143801.0098b030@odin.cc.pdx.edu>
Date: Mon, 02 Aug 1999 14:46:06 -0700
To: s-news@wubios.wustl.edu
From: Marc Feldesman <feldesmanm@pdx.edu>
Cc: support@statsci.com
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"; format=flowed
I've been using the discrim() function in Splus 2000. I thought I was
losing my mind as I'd compare the resubstitution confusion matrix and the
cross-validation confusion matrix printed by the summary.discrim function
with comparable results obtained in SAS. Both programs produced identical
confusion matrices (the counts were the same in both programs for both
analyses), but the error rates in SPlus were different from the error rates
in SAS. If I summed the diagonals (correct classifications) in SP2000 and
divided by the total, the error rates equal those in SAS but do not equal
the numbers reported in SPlus 2000.
Command line looks like this:
homtest.disc<-discrim(homtest[,3]~., data=homtest[,6:15],
+family=Canonical("linear"), prior=c("uniform"))
summary(homtest.disc)
Everything is fine until the program prints the "Plug-In Classification
Matrix", and the "Cross-Validation Classification Matrix". The marginal
error rates are correct but the error rates for both matrices are
wrong. They aren't even close to being right.
Dr. Marc R. Feldesman
Professor and Chairman
Anthropology Department
Portland State University
P.O. Box 751
Portland, Oregon 97207
email: feldesmanm@pdx.edu
phone: 503-725-3081
fax: 503-725-3905
-----------------------------------------------------------------------
This message was distributed by s-news@wubios.wustl.edu. To unsubscribe
send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
message: unsubscribe s-news
