Fig. 2.

Decay of r² with distance. (A) Comparison between our estimator ($r_{÷}^2$) and Rogers and Huff (2009) (RH) under steady-state demography. The $r_{÷}^2$-curve displays the appropriate decay behavior and is invariant to phasing, whereas the RH estimator gives upward biased r², and this general approach is sensitive to phasing. Estimates were computed from 1,000 1 Mb replicate simulations with constant mutation and recombination rates (each $2\times {10}^{-8}$ per base per generation) for n = 50 sampled diploids using msprime (Kelleher et al. 2016). (B) $r_{\text{RH}}^2$ Decay for five populations in 1000 Genomes Project Consortium et al. (2015), including two putatively admixed American populations (MXL and PUR), computed from intergenic regions. (C) $r_{÷}^2$ Decay for the same populations. (D) Decay of ${\sigma }_D^2$ computed as $\sum \widehat{D^2}/\sum \widehat{{\pi }_2}$. The $r_{\text{RH}}^2$ decay curves show excess long-range LD in each population, whereas our estimator qualitatively differentiates between populations.