But for the sake of argument, if the mutation rate is assumed to be 2 10^-8 per basepair and, on average, 0.8 basepair differences are observed between the sequences of two individuals at a gene that is 1000 basepairs long, then this would suggest that N is 10000, because the per gene mutation rate is 2 10^-5 so 4N = 4N (2 10^-5) = 0.8, giving N=10000.

Takahata's method is more sophisticated, in the sense that he gets a more accurate estimate by using more information from the data than just the average pairwise difference. But the gist of what he is doing is the same.