I work with microsatellites and I have been using Bonferroni correction for multiple tests to correct the significance of my results, but I have read that False Discovery Rate (FDR) behaves better under certain situations.
I'm using fdotool package on R to estimate the corrected p-values(FRD). The thing is that sometimes I have only 10 tests to compare (e.g. test of Hardy-Weinberg Equilibria for each population), and the package says that 10 tests are too little.
Should I keep doing Bonferroni? What do you guys do?
Doesn't p.adjust(orig.p.values, "fdr") work (this function is in the standard library "stats")? Why do you need a special package?
I don't see why controlling the *expected* FDR should not work on smaller numbers of tests.
Bonferroni is not a substitute - it controls the FWER not the FDR. These are different things. You should know what you want to control, and than do control this. If you want to control the FWER then use Bonferroni (or better Holm), if you want to control the FDR then use Benjamini-Hochberg (BH). Using BH simply because it is more liberal is a flawed argument.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Statistical Software
- R