View Article - PDF     Dataset (text)

Title: When to Use Fisher's Exact Test
Author: Keith M. Bower
Publication Source: American Society for Quality, Six Sigma Forum Magazine, Vol. 2, No. 4, August 2003, pp. 35-37

Abstract

Traditional methods used by Six Sigma practitioners to assess differences between items such as operators or machines may produce misleading results if the number of observations obtained for analysis is small. R.A. Fisher's exact test may provide a more appropriate analysis. The test uses the hypergeometric distribution rather than approximate results using a chi-square distribution to calculate the p-value. A sidebar article further explains the calculations for Fisher's exact test.

Notes

This paper was essentially created as a by-product of my needing to learn more about this procedure as it was being added to the software I used for training at that time. In particular, I was interested in assessing the different results that may be obtained when using Karl Pearson's Chi-Square test instead of Fisher's Exact Test, when the requirements hold for Fisher's exact test.

The scenario used for the worked example is rather trivial, though I realized it may be intuitively appealing. The response variable is straightforward and can only feasibly have a binary outcome (correct/incorrect).

As an irrelevant aside, I sent a copy of the published paper to David Letterman, but did not receive a response (maybe it should've been sent to Paul instead?)

To download Adobe Acrobat (for free) click here.