When discussing PLS methods of analyzing fMRI data, McIntosh et al. (2004) normalize the data for each trial with respect to the signal at the onset of a trial, rather than using the mean signal in a voxel across the time series (which is often included in AFNI examples) to make it more comparable to an ERP measure (which is always deviated from a local pre-trial baseline). Just wondering if this is common? I realize one can probably normalize the signal in many ways - just curious why AFNI tends to suggest using the mean value across the entire timeseries, rather than a local pre-trial baseline?

From the McIntosh paper:
In constructing the data matrix, the MR signals are normalized within a trial with respect to the signal at the onset of the trial. The normalization maintains the amplitude information of the BOLD signals and is similar to baseline correction in ERP analyses.

thanks, MB