There was an error in the Fortran code for EXTSIG announced recently
(real versus double precision conflict). On Sun machines this error
caused Fisher's exact p-values greater than one to be printed. The
error apparently doesn't affect the WIN32 code or the MAC code or the
code on several varieties of Unix. The authors do not know why the
code does not always give wrong answers, so it might be prudent to
download the fixed version which is the only one now posted (version
1.3).
Thanks to Roger L Berger <[log in to unmask]> for pointing out this
error. We apologize for any inconvenience this may have caused.
Dr. Berger also made another excellent suggestion that may lead to a
slightly improved algorithm.
For those with questions about the methods used in EXTSIG, we have
posted a pdf (Acrobat reader required) file of a draft of a paper in
progress available from
ftp://odin.mdacc.tmc.edu/pub/docs/exact.pdf
(pointers to this paper are also in the URL shown below).
For those who missed the original message:
EXTSIG performs tests for the difference in two binomial proportions.
After a two by two table is entered, the program reports one- and
two-sided p-values from Fisher's test and the chi-square test and
optionally exact significance levels obtained by the unconditional
method. EXTSIG is available from
http://odin.mdacc.tmc.edu/anonftp/
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|