NOTE 19W: We have ignored the initial random establishment of the favorable allele. Given that it will be fixed, its expected frequency is higher than this simple deterministic prediction by a factor 2s; in effect, the initial frequency is 1/4Ns, rather than 1/2N. So, a better prediction would be (1/s)log_e(4Ns). However, this makes little difference to the estimate because the dependence is via a logarithm. See Problem 18.5.