Bayesian inference from the conditional genetic stock identification model

Genetic stock identification (GSI) estimates stock proportions and individual assignments through comparison of genetic markers with reference populations. It is used widely in anadromous fisheries to estimate the impact of oceanic harvest on riverine populations. Here, we provide a formal, explicit description of Bayesian inference in the conditional GSI model, documenting an approach that has been widely used in the last 5 years, but not formally described until now. Subsequently, we describe a novel cross-validation method that permits accurate prediction of GSI accuracy when making Bayesian inference from the conditional GSI model. We use cross-validation and simulation of genetic data to confirm the occurrence of a bias in reporting-unit proportions recently reported in Hasselman et al. (2016). Then, we introduce a novel parametric bootstrap approach to reduce this bias, and we demonstrate the efficacy of our correction. Our methods have been implemented as a user-friendly R package, rubias, which makes use of Rcpp for computational efficiency. We predict rubias will be widely useful for GSI of fish populations.