Hi,

I am currently doing Fst compartmentalization analysis with permutation test. I set permutation as 1000 and I am getting some permutation results of 0.

How do I correct permutation result of 0?
Is there a way to increase decimal places in this analysis?

Thank you! 

Eunok

Hi Eunok,

Sorry, I don't quite understand your question. Can you paste the text output by F_st and point out to the zeros that are suspect?

Sergei

Hi Sergie,

So, I used F_ST.bf from the standard analysis. For distance computation, I chose Full likelihood and then I loaded DNA sequence data. Then I entered regular expressions to define 2 clades. For the standard standard substitution model, I chose CUSTOM so I could type "012345" for GTR. After selecting several parameters, I did permutation test (1000 permutations).

Then, I obtained "zero" for Prob (Random F_ST> Observed F_ST) like below: 

Prob {Random F_ST > Observed F_ST}

Hudson, Slatkin and Maddison : 0
Slatkin                      : 0
Hudson, Boos and Kaplan           : 0
Hudson, S_nn                 : 0

My question is whether the result above is indeed 0. (Or is it because of small number of decimal places? If it is so, how can I increase the number of decimal places?)

I hope my question makes sense.........

Eunok   

Hi Eunok,

OK, I understand. The zero you see is actually a zero (i.e. no decimal places). A general rule of thumb for you to estimate p-values based on permutations when there are exactly 0 / N replicates that beat the observed value, is to say that P <= 1/(N+1) [because in the worst case, the very next replicate might have beat the observed value]. Hence, you can report the p-value as <=1/1001

Sergei

Thank you so much!!!!  :)