[Spambayes] RE: For the bold
Tim Peters
tim.one@comcast.net
Sun, 06 Oct 2002 18:26:08 -0400
[Rob Hooft]
> ...
> BTW: I tried using the direct values hmean and smean in the pickles as
> well. This worked fine immediately for clt2 and gives (as expected) the
> exact same results.
Actually, I wouldn't expect that: the "zscores" aren't a function of smean
or hmean alone, they also depend on each msg's n value (in a way that's
highly dubious in clt1 and clt2, and wholly unjustified in clt3).
To the limited extent that the zscores *may* make sense under clt1 and clt2,
it's the dependence on n that the sense comes from.
If you got the same results by ignoring n completely (which just looking at
hmean or smean does), then that at least confirms that fiddling with n is in
fact of no value in the current pseudo-zscore computation.