Simulations for Case II, training sample size = 100, test sample size = 100.

BVS-SL(κ) represents the Bayes variable selection with belief parameter κ for all edges. Pencred, SSVS, Lasso, EL, SCAD, SSL, and Flasso represent the penalized joint credible regions approach, stochastic search variable selection, L1 penalized regression, and elastic net, the smooth clipped absolute deviation, the spike and slab lasso, and sparse fused lasso respectively. MSPE: out of sample predictive MSE; Pwr(10% FDR) is sensitivity controlling for 90% specificity; MS: estimated model size; FP: false positives, and Cov95 is coverage under 95% predictive intervals. The true model size is 10.

BVS-SL(κ) 表示针对所有边设置置信参数κ的贝叶斯变量选择方法。Pencred、SSVS、Lasso、EL、SCAD、SSL与Flasso分别指代惩罚联合置信区域方法（penalized joint credible regions approach）、随机搜索变量选择（stochastic search variable selection）、L1正则化回归（L1 penalized regression）、弹性网（elastic net）、平滑截断绝对偏差（smooth clipped absolute deviation）、尖峰平板Lasso（spike and slab lasso）以及稀疏融合Lasso（sparse fused lasso）。其中，MSPE指代样本外预测均方误差（out of sample predictive MSE）；Pwr(10% 错误发现率（FDR）) 为控制90%特异性的灵敏度；MS 为估计的模型规模；FP 为假阳性样本；Cov95 为95%预测区间下的覆盖率。真实模型规模为10。