Hello experts,
Is there any precedent in the literature anyone is aware of for reporting the actual smoothness of 2nd level models instead of the smoothing kernel applied during preprocessing? The selection of smoothing kernel size and initial height threshold seems so arbitrary, yet the complex interaction with degrees of freedom and actual signal heavily impacts reported results. I feel like it would be easier to develop some kind of data-driven standard if people were reporting the actual smoothness of their data (and their t/f height threshold in addition to p, given the influence of DOF)... Thoughts?
All the best,
Patrick
|