2.7. Relatedness analysis

Relatedness (r) is measured for two individuals on a 0 to 1 scale (0 being unrelated, 1 being identical) based on how much of the genome these two individuals are estimated to share (see Blouin, 2003, for a review of relatedness theory, methods, and studies). As sea turtles typically display natal breeding fidelity (with a margin of error but typically within MUs; Lohmann et al., 2008), we expect a higher probability of relatedness between individuals from within the same inferred cluster than between individuals from different inferred clusters. We therefore examined average pairwise relatedness within inferred structuring using two different algorithms (LRM: Lynch & Ritland, 1999; and QGM: Queller & Goodnight, 1989) over 10,000 iterations in GenAlEx (Peakall & Smouse, 2006). Queller and Goodnight's (QGM; 1989) estimator is a coefficient based only on the estimated identity by descent (IBD; Grafen, 1985). Lynch and Ritland (1999) estimator uses a regression calculation to determine relatedness coefficients for any pair of individuals based on shared IBD alleles, but can perform poorly if few related individuals are sampled, or if loci are too highly polymorphic (Blouin, 2003). Both estimators may also have high variances when few loci (n < 20) are used, but can provide a good estimation of relatedness between groups of individuals (Blouin, 2003; Queller & Goodnight, 1989). GenAlEx tests specifically for significantly high relatedness within groups (i.e., inferred structuring) relative to global relatedness, and we therefore focused on relative values of relatedness and the significance of inferred structuring relatedness rather than on specific thresholds of relatedness.

Further, we explored the potential for inferred structuring to comprise closely related individuals (parent–offspring pairs, full siblings, and half siblings) in COLONY v. 2.0.6.5 (Jones & Wang, 2010). First, to verify that both microsatellite data sets were powerful enough to exclude incorrect parent pairs (in light of the absence of known parents), we calculated P3Exc (the probability of excluding incorrect parent pairs when both parent genotypes are unknown) in GenAlEx for increasing combinations of loci for inferred structuring in Costa Rican and Eastern Tropical Pacific olive ridleys. We ran COLONY with all loci for each data set (n = 6 for Costa Rican data, n = 10 for Eastern Tropical Pacific data) for five “very long” runs of the full‐likelihood method of determining parentage and sibships with “very high” likelihood precision. We allowed for both male and female polygamy, as multiple paternity is commonly documented in sea turtles (see Lee, Schofield, Haughey, Mazaris, & Hays, (2018) for a review), and also allowed for inbreeding. We did not include a sibship prior. We examined the number of parent–offspring, full‐sibling, and half‐sibling pairs within and between inferred structuring groups to determine whether inferred structuring groups contained more parent–offspring or sibling pairs (i.e., family lineages) than pairs with members in two different groups.