Hi all, please see my original query to EvolDir below and (few) replies. If any further suggestions are forthcoming please e-mail me: markus.debruyn@gmail.com Thanks, Mark Dear Colleagues, We would like to test whether 3 pairwise PhiST estimates, and 3 Pi estimates are significantly different from one another (single locus - mtDNA HVR). I understand one method of doing so is by bootstrapping. Is this the preferable method, and does anyone know of a program that would automate these calculations? (Arelquin and FSTAT would appear to require data from at least 4-6 loci). Alternatively, instructions for doing this 'by hand' would be greatly appreciated. Bootstrapping Fst and friends is done over loci, so it should be easy... There are methods for bootstrapping over populations, but I'm not convinced by them in general (O'Hara, R.B. and Merilä, J., 2005. Genetics 171: 1331-1339), and for 2 populations, it'll obviously be difficult to bootstrap. You should be able to construct a parametric bootstrap, by re-sampling the data from a multinomial distribution, with probabilities equal to the observed allele frequencies. This might be worth testing first, though - I'm not sure how much of a bias in induced if you have rare alleles that are not in the sample. Bob O'Hara Bob O'Hara Department of Mathematics and Statistics P.O. Box 68 (Gustaf Hällströmin katu 2b) FIN-00014 University of Helsinki Finland Hi Markus, Regarding your query to evoldir. The reason Fstat etc require multiple loci to bootstrap confidence intervals is that the unit used for bootstrapping is the locus. Thus it is not possible to sample randomly from a distribution of loci unless you have more than one. Moreover, it has been shown that the bootstrap method can be very unreliable when performed across less than five loci. It is possible to bootstrap across other units (e.g. individuals) but you have to ask the question - where is my uncertainty? Is it that the individuals in your sample are a representative random sample from your population; or is it that your locus is a representative sample of the genome? In all likelihood your locus/loci are a very small sample of the entire genome, which is why you want to estimate the Cis. This is the case for most population genetics studies that use a relatively small number of loci to infer neutral genetic diversity (and the process acting upon it) as a sample from a potentially very large genome, hence the tendency to bootstrap across multiple loci. In your case bootstrapping is not an option I'm afraid - you have nothing to bootstrap! There may be some other tests that are more applicable and I hope you will get some answers from your query that will help. I would also say that p values are not the be all and end all. If your tests show large biological differences than you should focus on that. All the best Matt Dr Matthew Oliver Research Fellow School of Biological Sciences University of Aberdeen Zoology building Tillydrone Avenue Aberdeen AB24 2TZ UK markus.debruyn@gmail.com