Dear Lee,
> A simple question. What is the difference when you create your activation
> map with and without global normalisation? What does global normalisation
> actually do to the raw signal?
when you are doing global normalisation SPM will calculate for you an average (across voxels) intensity of each scan, and then divide all the voxels in that scan with that average. This means that after global normalisation all the scans in your time-series will have the same average intensity.
The reason for global normalisation is to remove confounding "global" effects such as e.g. amount of injected radio-tracer in the case of PET, or gain-drifts in the case of fMRI.
The problem with it is that the global intensity is really only the sum of all the local intensities and if the local intensity is changing somewhere as a result of your experiment (i.e. you have an activation somewhere) then the global intensity will change as a consequence of that. Then dividing by that global intensity may potentially decrease the magnitude of your true activations and cause artefactual deactivations.
Because of this, and also because the high-pass filter is quite successful in removing most gain-drifts and other global effects, global normalisation is normally not recommended for fMRI data.
Good luck Jesper
|