Re: [S] reasonable p-values for Fisher exact's test - WAS strange ...

David Ross (ross@math.hawaii.edu)
Tue, 24 Mar 1998 22:33:29 -1000


>Also, note that fisher.test() uses an algorithm which allows R x C
>tables. This isn't required in simple 2 x 2 tables (and it wouldn't be
>too hard to put in a switch for such tables), but this is what gets
>used.
>
>Getting to the point:
>
>This algorithm usually yields answers that differ numerically from the
>exact hypergeometric probability

The S+ algorithm used here is the same as in StatXact (at least, the articles
referenced are by the creators of the latter program). I no longer have
access to the latter program; it would be interesting if someone who does
have it could try to reproduce these strange p-values. (I think the StatXact
implementation is also availabe as an SPSS module.)

If I remember the algorithm correctly (I read the papers 8 years or so ago),
it involves some tree-climbing and lots of saving of partial calculations,
even on a 2x2 matrix, and so there's room for cumulative roundoff errors to
creep in.

- David

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