Fig. 8.

Comparison of analysis methods with uniform and tapered contrast vectors using type-I error control. (a) Each subplot shows the empirical FPR versus reported $p$-value curves of recognizing the hemoglobin activity within a single ROI using the two types of contrast vector (indicated by color) for data with ROI $\text{radius}=14\mathrm{mm}$ from 2000 iterations of simulations under the conditions where the probe registration information (including head size and registration error) is known or not (indicated by column). The two rows indicate oxy-/deoxyhemoglobin, respectively. In an ideal situation, the empirical FPR equals the model-reported $p$-value, which is represented by the dotted diagonal of each plot. Both methods underestimate the FPR at smaller $p$-values (enlarged and embedded at the corner). (b) Each subplot shows the ROC AUC as a function of ROI radius. Every data point represents the AUC calculated from 2000 simulation iterations against the corresponding ROI radius (in mm) used in the simulation. Every data point represents the empirical FPR calculated from 2000 simulation iterations against the corresponding ROI radius (in mm) used in the simulation. Here, 0.05 is used as the type-I error control (threshold) that is indicated by the dashed line.