TABLE 3.

Comparison of the influence of the approximation on the power of LBF for simple models with different migration schemes

Evidence	Counts (based on , , and LBFHM)
Model	1a			1b			2a			2b
Nm	∞			1250			0.25			0.0025
Approximation	16	4	H	16	4	H	16	4	H	16	4	H
Against M0	0	5	26	0	8	29	70	49	53	100	100	78
Against M1	100	94	73	100	92	71	30	51	47	0	0	22

LBF compared a full model (model M₁ = □⇆▪) with a panmictic population (model M₀ = ). Models used to simulate the data were as follows: 1a, a single population, the sampled individuals split randomly into two sets (); 1b, two populations exchanging many migrants (Nm = 1250); 2a, two populations exchanging a moderate number of migrants (Nm = 0.25); and 2b, two populations with very low migration rate (Nm = 0.0025). The marginal likelihoods used in the LBF were approximated with thermodynamic integration (TI) with 16 and 4 scaler bins and with the harmonic mean (HM₄).