Fig. 5

Power comparison of six methods under two-condition Laplace setting. In each panel, for each specific (r, s) pair, powers of the six methods were computed from 100,000 replicates of two-group samples with samples sizes (40 vs. 40) simulated from ℒaplace(0, 1) and ℒaplace(r, (1 + s)²), respectively. At each (r, s) pair, the power of each method was estimated by the empirical proportion that the method rejected the dual null hypothesis H₀₃ at significance level 0.05. For the SMVT, both the significance level of Welch test and that of the Levene test were set to be \( 1-\sqrt{1-0.05} \) to control overall type I error rate at 0.05. a Dual heterogeneity, b Variance heterogeneity, c Mean heterogeneity