Hi James,

Thank you for your insights. Regarding your first point, could you please elaborate? The double-dipping issue was not something I had previously thought of. To clarify, this is how the ROI was defined: activation map for C1 correlation with contrast 1 at the corrected threshold was obtained. Similarly, I obtained the activation map for C2 correlation with contrast 2. Next, I simply used imcalc on the two activation maps to find voxels that were present in both maps. These voxels were the ROI.

As for your second point, C1 and C2 are not significantly correlated.

Finally, in the original manuscript, we did look at the voxels that constituted the non-overlapping part (at the corrected threshold), so basically "voxels with significant correlation A and evidence of no correlation B" and vice versa. Apparently, this wasn't sufficient. That's why I'm unsure what else the reviewer wants with a dysjunction analysis.

Best,

Miranda

On Wed, Apr 10, 2019 at 4:44 PM James Rowe <[log in to unmask]> wrote:

Dear Miranda,

There are several potential problems, but the extent to which they matter depends on the precise conclusions you wish to draw.

1.       How did the choice of ROI depend on the SPM voxel-wise maps? If you chose the ROIs based on the SPMs, then there is a risk of double dipping - if the tests/statistical models applied to the ROIs are similar to the tests/models used to generate the SPMs.

2.       A second problem concerns the analysis of clinical correlations. The NoGo>Go and Go>NoGo contrasts are ‘equal and opposite’.   But what about the clinical measures – are they independent of each other?   Imagine you have two clinical measures, say C1 as a measure of cognitive ability, and C2 as an index of cognitive deficits. They will be negatively correlated. The correlation of C1 with NoGo>Go will therefore be very similar to the correlation of C2 with G>NoGo.   The marked spatial overlap in activations would in this case be inevitable, and uninformative.   So, in your study, you need to consider the relationship between the clinical measures (either explicit or implicit) in order to interpret the overlap meaningfully.

3.       For a disjunction,, be careful not to infer the null hypothesis form the absence of a significant [frequentist] correlation. Absence of evidence is not evidence of absence of an effect.     You could test for voxels with significant correlation A and evidence of no correlation B, but the evidence of no correlation would have to come from a Bayesian analysis, not a frequentist statistical inference (ie not a t-test).

Bw

James

From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On Behalf Of Miranda Elizabeth
Sent: 10 April 2019 21:10
To: [log in to unmask]
Subject: [SPM] Methodological and theoretical issues with contrasts

Hi,

We recently received a revise-and-resubmit for a paper submission. One of the reviewers made a couple of comments regarding our contrasts. I am hoping to get some help from fellow SPM users here so I can adequately address these comments.

In a go/nogo task, we used Go>Nogo contrast to get activation to response (contrast 1), and Nogo>Go to get activation to response inhibition (contrast 2). Next, we performed 2 multiple regressions, each using a different clinical measure from our participants as the independent variable and activation from the contrast as the dependent variable. The two multiple regressions showed significant activations which partially overlapped. We explored this overlapping regions in an ROI analysis afterward. Regarding our choice of contrasts and the overlap, the reviewer said "the response contrast and the response inhibition contrast are the opposite of the same contrast" and listed this as one of the issues. I don't really see a problem with this as response and response inhibition are the opposite actions. I acknowledge that the brain activity for the overlapping region in one contrast is exactly as the other contrast with the sign reversed. Does anyone see a methodological/theoretical issue with this contrast selection?

Since we have overlapping activation, the reviewer also suggested doing a dysjunction analysis, presumably the opposite of the overlap. I'm not sure how to do this dysjunction analysis. I'd appreciate any suggestions you may have.

Thank you!

Miranda