Panel a: filled contour with the distribution of Bayes factors, B_RA and B_AR, over the two-dimensional data space 𝒟. Solid lines connect points with p(𝒟|M_R) = p(𝒟|M_A) (Bayes factor equal to zero). The computations are performed over a grid of points (N_f = 80, N_{⟨P(f)⟩ratio} = 155) over the ranges f ∈ [0.05, 4] and ⟨P(f)⟩_ratio ∈ [0.25, 4.1]. The priors are p(ξ_E)∼𝒢(1.9, 0.3) for M_R and p(ξ_E)∼𝒢(4.3, 0.6); p(R₀)∼𝒢(0.5, 0.02) for M_A. Triangles represent CoMP data. The error model for observed and synthetic average power ratio is the same adopted in the inference analysis, Sect. 3.1. Following Kass & Raftery (1995), the evidence in favour of a model, i, in front of an alternative, j, is inconclusive for values of 2log(B_ij) from 0 to 2; positive from 2 to 6; strong from 6 to 10; and very strong for values above 10. Panel b: marginal likelihood for models M_R and M_A for the data points circled in (a) as a function of average power ratio. The vertical dotted lines are the CoMP average power ratio values.