Fisher exact error

This may no longer be a problem, but I thought I would post it anyway.

I'm still using version 3.0 (running under Windows XP on a Dell Optiplex). The last time I taught my biostatistics course, I encountered a problem with a program that computes the two-tailed p value for a Fisher exact test. The input table was

5 1
1 4

and the p value generated was 0.0909. The correct p value is 0.0801. I made a note to check this and finally got around to doing so. I discovered that rearranging the input format would change the result. Thus, the following table also gave the (same) incorrect result:

1 5
4 1

However, the following two tables gave the correct result:

4 1
1 5

and

1 4
5 1

I finally found the source of the error. The Fisher exact subroutine provided with the "Statistics Graphics Toolkit" returns an incorrect value (0.01299) for the more extreme table shown below:

0 6
5 0

All other rearrangements of this table return the correct value: 0.00216.

Because I have been able to duplicate this problem repeatedly, I can only assume that there is a bug in the Fisher exact subroutine that I am calling.

If this is no longer an issue, good. However, if that subroutine is still being used, then this does need attention.

Comments

Fisher exact

Many thanks for pointing out this discrepancy. Despite the fact that you are using a version that is at least 10 years out-of-date, I suspect the Statistics Toolkit has remained un-changed during that time. I will make sure the author is notified.

Big John