[S] Diff logit reg output b/w sas & S

Edward Malthouse (ecm@casbah.acns.nwu.edu)
Mon, 11 May 1998 10:23:50 -0500 (CDT)


I want to thank Alan Zaslavsky, Frank Harrell, Kent Holsinger, and
Matthew Zack for their responses to my question about the difference
between Splus and SAS logistic regression output. The key to
getting the the parameter estimates to agree is to set epsilon=1e-8
in the call to the Splus GLM function. The Wald Chi-Square P-values
in the SAS output cannot be compared to the P-values from the
anova(fit, test="Chisq") output, because the two test different
hypotheses. The anova function gives something similar to
sequential sums of squares. The Wald test can be misleading due to
the Hauck-Donner effect. See the discussion in V&R2 or do a search
of the S-news archives and read Brian Ripley's posting on the topic
from a few months back.

Thanks again for setting me straight on this.

Ed Malthouse

-----------------------------------------------------------------------
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