Figure 1.

An example of the difficulty in genetic inference. Cumulative frequency distribution of F_st for data simulated using Simcoal (Excoffier et al. 2000– heavy lines) and analysed using FDist2 (Beaumont and Balding 2004– thin lines). Top panel: Island model with 1 and 10 migrants per generation (solid and dashed lines respectively). Because the data were simulated with the island model, the distribution is well matched by FDist2. Bottom panel: Populations phylogeneticaly related within two lineages at migration/drift equilibrium (solid lines) and recently diverged (dashed lines) 5N_e generations ago. Departure from the Island model greatly increases the stochastic variability of the data relative to assumptions of the analysis. Data were simulated assuming 10 populations each with N_e = 10 000, and the infinite alleles model with μ = 0.5 × 10⁻⁶. The island model assumed equal migration among all populations, whereas the phylogenetically structured model assumed five populations within each lineage, with Nm = 1 among populations within each lineage and restricted (Nm = 0.001) migration between lineages.