Research Article

Effect-Size Estimation Using Semiparametric Hierarchical Mixture Models in Disease-Association Studies with Neuroimaging Data

Figure 1

Average bias in estimating effect sizes for each of the top 500 voxels across 100 simulations when the sample size n is 50 (left), 100 (center), and 200 (right). Panels (a), (b), and (c) represent scenarios with various degrees of dependency among contiguous voxels specified by the parameter γ of the Ising model when the proportion of disease-associated voxels is 20%.
(a) Strong dependency:
(b) Intermediate dependency:
(c) Week dependency: