I am comparing multiple libraries and would like to apply Bonferroni correction (as you did in Scottish soil case study in Schloss et al., 2004 and more recently in Riviere et al., 2009). Due to to extensive use of Wikipedia, now I almost! understand what the Bonferroni correction is, but I still have no idea how to apply it. Now, my country is playing against England and I am still here. So, can you save a poor girl and help me out?
Don’t know that I should help some one from a country that stole a draw from the States, but here goes:
The basic approach would be to say that I want to have no more than a 5% chance of saying a comparison is significant when it isn’t. Then you say, ok, I’ve got 5 communities and I can do 20 [i.e. 5 * (5-1)] pairwise comparisons (each test has 2 comparisons). So then divide 0.05 by 20 and you get 0.0025. Now you go through your 20 p-values and determine which of those is less than 0.0025. Those are significant at the 0.05 level. Also note, that if the comparison between A and B is significant or B and A is significant then A and B are significantly different.
Hope this helps - go Slovenia!
A single conversation with a wise man is better than ten years of study.
thanks a lot!
p.s. eeeee we lost,…
Please note that the original Bonferroni is really conservative when it comes to many comparisons. In case of 20 pairwise comparisons I would suggest Hochberg adjustment of the Bonferroni.
- Make the pairwise tests
- Arrange the p-values of the tests in decreasing order
- Test the largest p-value at level 0.05 the second largest at level 0.05/2 (=0.025) the third largest at 0.05/3 (=0.0166) and so on…
When you reach the first significant result then the results of that test and of all the tests with lower p- values thereon are significant.
See Quinn and Keough: Experimental design and data analysis for biologist.(2002): chapter 3.4: multiple testing
Agreed. Also, libshuff is really weird and I’d probably avoid it along with the Bonferroni.