Currently, average_response does a direct mean of response function ZSH coefficients across subjects. This means that if the overall amplitude of the response function is larger in one subject than in others, the shape of the response function for that subject will have a greater influence on the shape of the group average response function. If equalising the intensity of WM b=0 signal across subjects using dwiintensitynorm before this is done, this could be perceived as a desirable effect: subjects with a smaller RF after dwiintensitynorm would have less influence on the shape of the mean response. However in the current multi-tissue FBA documentation, the dwiintensitynorm step is no longer performed. Hence the scanner reconstruction gain factor could influence which subjects have greater influence on the shape of the group average response.

In this case, it could be preferable to instead normalise each subject's RF to a fixed l=0, average, then re-scale the resulting average response to the mean l=0 across subjects.

What's not written in stone is how this normalisation should occur in a multi-shell context, i.e. whether to normalise independently for each shell or take e.g. a geometric mean of scaling factors across shells for that subject.

Tagged question for a reason. 👂