On 2015 Feb 25, Gerard Ridgway commented:

In relation to the failure "to replicate all the group differences in GMV reported in previous studies", it would be very helpful to extract GMV for the previously reported coordinates (or clusters) and to report the magnitudes of the group differences on the new data, for example like the plot in Figure 2 does for the significant cluster here. It would be more compelling to support an assertion of an absent effect on such an extracted summary by demonstrating that the confidence interval is reasonably narrow (i.e. confined to trivially small group differences) rather than just that the region is not significant (in this case, after correction for multiple comparisons).

It is also not made clear whether the methods closely match the unreplicated studies. For example, the smoothing of 4mm here seems very low for a VBM study (particularly one using cluster-extent inference), and it would be interesting to know whether previous studies have used similarly low smoothing, or more typical values such as 8mm. As another example, did all previous studies adjust for total intracranial volume? In cases like this, where there is a significant group difference in TIV, I think both the adjusted and unadjusted analyses can be of interest.

On 2015 Feb 25, Gerard Ridgway commented:

In relation to the failure "to replicate all the group differences in GMV reported in previous studies", it would be very helpful to extract GMV for the previously reported coordinates (or clusters) and to report the magnitudes of the group differences on the new data, for example like the plot in Figure 2 does for the significant cluster here. It would be more compelling to support an assertion of an absent effect on such an extracted summary by demonstrating that the confidence interval is reasonably narrow (i.e. confined to trivially small group differences) rather than just that the region is not significant (in this case, after correction for multiple comparisons).

It is also not made clear whether the methods closely match the unreplicated studies. For example, the smoothing of 4mm here seems very low for a VBM study (particularly one using cluster-extent inference), and it would be interesting to know whether previous studies have used similarly low smoothing, or more typical values such as 8mm. As another example, did all previous studies adjust for total intracranial volume? In cases like this, where there is a significant group difference in TIV, I think both the adjusted and unadjusted analyses can be of interest.