On Fisher's exact test for detecting gametic disequilibrium between DNA polymorphisms.

Abstract

Nonrandom associations between DNA polymorphisms are commonly tested by Fisher's exact test in spite of it is seriously conservative. Theoretical statistical studies have shown that the exact test with Tocher's correction does not present this problem but Tocher's correction is never used by experimentalists for detecting gametic associations. We have examined the practical consequences of using these two alternative tests for the detection of nonrandom associations in populations. A total of 1566 pairs of RFLPs of eleven gene regions and sixteen populations from previously published human and Drosophila disequilibrium data were examined. The analysis reveals remarkable differences between the two tests for detecting gametic associations. In some gene regions, the exact test with Tocher's correction detects a percentage of significant associations between RFLPS which is twice or three times higher than that detected by the exact test without correction. Therefore, the widely used exact test can be seriously underestimating disequilibrium in populations. In addition, the study shows that the standard chi-square test for independence in 2 x 2 tables detects a similar percentage of significant associations between RFLPs than Tocher's test.