2.6.2 Population-level estimates of selfing rates
Selfing rates were estimated using three methods [i.e.,s(g2), s(ML) and s(BES) ] from the microsatellite data of 613 individuals of 22 natural populations (see section 2.3). First, selfing rates and associated 95% CI were estimated using the reciprocal of the two-locus heterozygosity disequilibrium (s(g2) ) and a Maximum Likelihood (s(ML) ) estimator, both implemented in RMES (David, Pujol, Viard, & Goudet, 2007)⁠ by running 10000 iterations. Both s(g2) and s(ML) estimate correlation among loci, providing complementary information about identity disequilibrium and selfing rate. Given that estimation of selfing using s(g2) and s(ML) can be biased due to the removal of homozygous loci (David et al.,2007)⁠, we compared these selfing rates with a model-based Bayesian estimation of selfing that makes use of all loci regardless of heterozygosity (BES v0.1.3; Redelings et al., 2015)⁠. Standard deviations were estimated using 10000 bootstrap iterations. To test if population-level estimates of selfing rates are positively correlated with the frequency of homostyles in natural populations, we used generalized additive models where frequency of homostyles was used as independent variable and specifying a Beta distribution for the estimates of selfing rates as implemented in the R package ‘gamlss’ v5.1.6 (Rigby, Stasinopoulos, & Lane,, 2005).