Why corrections for multiple tests should NOT be applied to HWE p-values

Explanation from Jon (posted by Debs) =

The problem is this – Bonferroni (or sequential Bonferroni) corrections are very conservative, which in normal hypothesis testing is what you want to be.

The more tests you do, the more likely it is that one will be significant by chance (we call this a Type 1 error, or a false positive). Therefore, by dividing the normal significance threshold (P < 0.05) by the number of tests (the number of loci examined), you reduce the chances of getting a Type 1 error. However, when testing microsats we don’t want to miss loci with null alleles (which cause departures from HWE) and end up typing them in lots of individuals, when in fact they are rubbish loci. By setting a significance threshold too stringently we end up including too many ‘bad’ loci in downstream applications. Therefore, it is best to use a threshold of P < 0.05 for each locus we test. This is particularly problematic when only typing a small number of individuals (e.g. 24) because a locus with a null at really high frequency still might not be Bonferroni significant. .. Bottom line is that one shouldn’t use Bonferroni corrections for HWE tests (, and 24 individuals are not really enough to estimate allele frequencies with much confidence.)

One Response to Why corrections for multiple tests should NOT be applied to HWE p-values

  1. Thanks a lot this is very very usefull!!
    -Óscar

Leave a Reply