This estimate is in-between 1.2 and 2.5x10-8, the two most quoted values for this parameter. It seems like such an important number that I'm wondering if it would be possible to brute force estimate it. Maybe people who have whole genomes of parents-offspring should get together and do the mother of all meta-analyses to pin down this number.

From the paper:
For example, a recent method for estimating population split times from coalescent rates placed the median split of African from non-African populations at 60–80 ky and the split of Native Americans from East Asians at ∼ 20 ky, both assuming a per-generation mutation rate of 1.25 × 10−8 and an average generation interval of 30 years [28]. [...] Using our inferred rate also makes the dates more recent, but only by a factor of about 1.3 rather than 2, i.e., ∼ 46–61 and 15 ky (with some associated uncertainty both from the model and from our estimated rate), neither of which contradicts external evidence.
The first of these estimated splits overlaps the dates of estimated Neandertal admixture by the Ust' Ishim and Kostenki papers. For a variety of reasons that I've repeated ad nauseam, I think that the split of Africans from non-Africans first happened about 100ka with Out-of-Africa-into-Arabia. But, if there was back-migration into Africa, maybe this can be brought down. The 15ky value for the East Asian/Native America split seems too young: it's as late as could plausibly maintained for the colonization of the Americas, but the split of the two must have happened some time before that (because the ancestors of Native Americans and East Asians would have split long before a group of them made the crossing into the Americas).

bioRxiv http://dx.doi.org/10.1101/015560

Calibrating the Human Mutation Rate via Ancestral Recombination Density in Diploid Genomes

Mark Lipson et al.

The human mutation rate is an essential parameter for studying the evolution of our species, interpreting present-day genetic variation, and understanding the incidence of genetic disease. Nevertheless, our current estimates of the rate are uncertain. Classical methods based on sequence divergence have yielded significantly larger values than more recent approaches based on counting de novo mutations in family pedigrees. Here, we propose a new method that uses the fine-scale human recombination map to calibrate the rate of accumulation of mutations. By comparing local heterozygosity levels in diploid genomes to the genetic distance scale over which these levels change, we are able to estimate a long-term mutation rate averaged over hundreds or thousands of generations. We infer a rate of 1.65 +/- 0.10 x 10^(-8) mutations per base per generation, which falls in between phylogenetic and pedigree-based estimates, and we suggest possible mechanisms to reconcile our estimate with previous studies. Our results support intermediate-age divergences among human populations and between humans and other great apes.

Link