Robust inference of historical human generation times

Abstract

Ragsdale and Thornton (2023) raise concerns about our recent estimates of historical human generation times, concluding that our results were "predominantly driven by nonbiological artifacts." While we believe these authors have pointed out several important sources of uncertainty, we show here that their main concerns are either not relevant to our study or support our conclusions as much as they cast doubt on them. In particular, the demographic simulations carried out by Ragsdale and Thornton assume all individuals with recent African ancestry are from West Africa, which is not appropriate for our sample. In contrast to the lack of visual concordance between predictions and data cited by these authors as evidence for a lack of fit, we demonstrate that our model provides a good statistical fit to data on the overall historical mutation spectrum, though one particular mutation type is an outlier. Furthermore, we show that the historical generation times inferred when using alternative methods for estimating the ages of individual alleles are largely in agreement with our results, particularly so when using results from Relate. Importantly, these analyses, as well as recent work from an independent group, confirm the idea that a model built on de novo mutations and applied to polymorphism data provides useful and reliable estimates of generation times in widely distant mammals.