Where he mentions that the geometric mean is consistent with Fisher's
method, but no more details are given:
"The connection would take too much time to explain here, but it is a great
justification for why this should work really well."
I understant GM is now obsolete and has been replaced in favor of Fisher's,
but I am interested in the derivation of the method (or at least the basic
intuition behind it) and its relationship with Fisher's. I could not find
any explanation in the archives of SpamBayes.
The most similar thing to Robinson's GM I could find is "logarithmic opinion
pool" with weights wi=1/n and constant k=1, but I have no idea whether the
two things are related.
Any explanation about the derivation of the GM method will be greatly
Thanks in advance
P.S. I sent this same question to the [Spambayes - General] list and
somebody suggested me to write here.