Dear Experts,

We are having some issues with replicating our previous results since upgrading to the current version of SDM from our previous version downloaded in early 2018. Cluster sizes at a threshold of .005 are now significantly smaller than they were on the previous version - in some cases up to 10 times smaller.

Any insight as to why this may be would be greatly appreciated. 

Many Thanks, 

Daniel

Dear Daniel,

The new SDM-Psi method corrects the p maps using the family wise error. You can use a threshold value of 0.05.

Best,

Anton

Dear Anton, 

Many thanks for your reply. Can we just double-check as in the new guidance for thresholding/exploration of results it states:

"As in most neuroimaging software, this dialog conducts a pair of one-tailed tests. To avoid doubling the false positive rate, please halve the statistical significance level (e.g., to 0.025)."

As we do not have a directional prediction about our results we have now adopted the more conservative threshold of 0.025 but perhaps 0.05 FWE as you mention would be apropriate? The outputted .html also reports these results at 'uncorrected' - maybe this is text preserved from the previous versions but is adding to our confusion. 

Best wishes
Jack (and Daniel)

Dear Jack and Daniel,

Are you thresholding a corrected map or an uncorrected map? The corrected maps are the results of running fwe correction (which results in two maps: voxel-wise corrp or tfce corrp). The uncorrected maps are the result of an uncorrected p derivation from the z map, computed at the end of the Mean computation.

Regarding the threshold value to be used, in any neuroimaging study (e.g., SPM, FSL, SDM, etc), while it is true that a pair of one-tailed tests are being conducted and it may be more accurate to use 0.025, it is common to use 0.05 value for thresholding corrected p-maps. Therefore, this remains a choice of the user :)

Best,

Anton

Hi Anton,

Thank you very much for your help. We are using the maps computed at the end of a mean computation and so are uncorrected.

If I can just ask one more question, would we still have the same degree of freedom to choose between a threshold of .05 or .025 when using an uncorrected map or is the threshold generally deemed to need to be on the stricter side?

Many thanks again, 

Daniel