Illustration of model comparison via evidence and of its associated scatter for a one-dimensional data vector d and two models, M_i with i = 1, 2, each with a single parameter p. The joint distributions Pr(d,p|M_i) are shown in blue and red shades. The projections of these distributions onto the parameter and data axes yield the prior Pr(p|M_i) and the evidence or marginal likelihood, respectively (shown in purple and orange for M₁ and M₂). An experiment produces the observation d, shown as the solid black line. Conditioning on d yields the posterior Pr(p|d,M_i), shown in green and brown for M₁ and M₂. Evaluating the marginal likelihood at d yields Pr(d|M_i), which is used in the model comparison. The dotted horizontal lines mark the 1σ interval of possible alternative realisations of the data given the best-fit parameter p_best (dashed vertical lines) of either model. The blue and red boxes show the resulting 1σ range in possible evidence values under M₁, which has higher evidence in this case (the corresponding ranges for M₂ are shown as hatched areas).