Figure 3:

Probability of selecting the wrong model when using the ordinary BIC for testing M₀ : θ₁ = θ₂ = 0 against $M_u:\theta \in {ℝ}^2$ (dashed line), and the two order-constrained BICs when testing M₀ : θ₁ = θ₂ = 0, M₁ : θ₂ > θ₁ > 0, and M₂ : θ₂ ≱ θ₁ ≱ 0 (dotted and solid line for the non-local and local unit-information prior, respectively) for true effects of (θ₁, θ₂) = (a, 2a), for a = 0, .1, .2, and .4. The sample size on the x-axis is on a logarithmic scale.