[Skip to Content]
[Skip to Content Landing]
December 8, 1989

Fisher's Inexact Test: Probability Too Serious to Be Left to Statisticians

Author Affiliations

Northwick Park Hospital & Clinical Research Centre Harrow, England

Northwick Park Hospital & Clinical Research Centre Harrow, England

JAMA. 1989;262(22):3129. doi:10.1001/jama.1989.03430220050017

To the Editor.—  I am sure we all applaud the careful study by McKinney et al1 that reviewed published errors in the use of Fisher's Exact Test. However, their case does not go far enough. Physicians are criticized for lack of care in statistics and for not consulting experts when needed, but is the statistical establishment itself above reproach here?McKinney et al do not mention that statisticians advocate more than one way of finding the total probability in Fisher's Exact Test; the discrepancies between these methods can be just as great as that between the one- and twotailed P values. Worse still, the commonly recommended ways of finding the second tail can, in certain cases, give impossible P values. I have enumerated these anomalies and also have offered a solution that avoids them.2,3 Of course, statisticians do not have to accept my solution; if a better one