RufusAtticus

Basically he's using popgen theory that is for infinite populations and applying it to really small populations.

For infinitely large populations you can say that under constant selection, that if the frequency of an allele is now p=0.1 that it will be p' = 0.1001 next generation. (Assume that p'=w*p, where w=1.001.) This is because there are no stocastic effects in an infinite population. They're a deterministic system. However, when you relax that assumption and allow the population to have finite population sizes, then drift begins to play a part. If the pop size is N=50 and p=0.1, then the number of copies of the allele is now 2Np=10. However, the next generation will be formed by sampling from the previous one. Therefore, it will not be deterministic, but rather the probability that p' is r follows a bionomial distribution where, P{p'==r} = B(100, 0.1001). This of course assumes a constant population size. The deterministic result can be seen from this by looking at what happens to the bionomial distribution as N->+inf. The variance of p' -> 0.

What nickb has done at ARN is completely ignored not only real life, but classical popgen theory. Instead of realizing the stocastic nature of the situation, he has argued that p' = floor(2N*w*p), which in our case means that p'=0.1=p. Using this error he assumes that selection does not work in small populations.

GIGO.