Results

Figures 7 and 8 show cluster-thresholded ( $z > 2.3,p<0.01$) group activation for the two motor tasks. Table 9 shows the number of suprathreshold voxels and the maximum z-statistics for the two tasks. Figure 7 shows the results from index finger tapping against rest ([INDEX] dataset). There is a general decrease in z-statistics in potentially activating voxels. This demonstrates the dominance of one of the two possible effects of incorporating first-level variances into the second level estimation process - that is we get an increase in estimated group variance, $\sigma_g$, due to it being constrained to be positive. Figure 8 shows the results of a contrast of sequential finger tapping vs index finger tapping ([SEQUENTIAL] dataset). There is a general increase in z-statistics in potentially activating voxels. This demonstrates the dominance of the other possible effect of incorporating first-level variances into the second level estimation process - that is we get increased efficiency in parameter estimation due to the use of lower-level variance heterogeneity.

Figure 7: Cluster thresholded ( $z > 2.3,p<0.01$) group activation from the [INDEX] dataset. [top] the [OLS] approach, and [bottom] the [HYBRID] approach.

Figure 8: Cluster thresholded ( $z > 2.3,p<0.01$) group activation from the [SEQUENTIAL] dataset. [top] the [OLS] approach, and [bottom] the [HYBRID] approach.

Figure 9: [left] Number of suprathreshold voxels, and [right] Maximum z-statistic, from the [INDEX] and [SEQUENTIAL] FMRI datasets using the two different inference techniques [OLS] and [HYBRID].