Table 3.

ABC analysis

Model	Ncur	Nanc/Ncur	T	µ	p	Marginal density
Priors
Free model	500–50,000	0.01–100	100–31,700	10−3–10−5	0.4–0.9	0.31
Model selection: decline	500–50,000	1–100	100–31,700	10−3–10−5	0.4–0.9	0.24
Model selection: expand	500–50,000	0.01–1	100–31,700	10−3–10−5	0.4–0.9	1.26
Posteriors
Free model	9,200 [900–45,600]	0.90 [0.26–25.84]	1,100 [100–15,900]	5.1e-5 [2.4e-5–1.1e-4]	0.82 [0.75–0.89]
Model selection: expand	10,200 [3800–37,900]	0.47 [0.04–1.00]	1,900 [1100–19,500]	4.8e-5 [2.4e-5–9.8e-5]	0.81 [0.72–0.89]

ABC analysis of D. robustus mtDNA and microsatellite data combined. Priors were incorporated on a log scale with a uniform distribution, but here they are shown on a natural scale for clarity. N_cur is “current” population size, representing effective population size at the last sampling point (602 B.P.). N_anc/N_cur is the size of the “ancient” effective population size at the onset of demographic change relative to N_cur. Numbers are diploid individuals rounded to the nearest 100, converted from haploids as used in ABCtoolbox. T is the time at the onset of the demographic change measured in years relative to the youngest sampling point, rounded to the nearest 100, and converted from generations, as used in ABCtoolbox (generation time assumed to be 10 y for D. robustus). µ is the microsatellite mutation rate per generation, and p is the shape parameter of the geometric distribution of the General Stepwise Model. Marginal densities were compared for the two scenarios in the model selection approach, and the Bayes factor support for the expand model was 5.25 (1.26/0.24), indicating substantial support (Jeffreys; ref. 58). Posterior parameter estimates represented by mode values, and 90% HPDs are shown for the two models with highest marginal density (free and expand models).