I recently analyzed using an SPMG1 HRF model. I have since reanalyzed it (to deal with the left-right flip issue noted in Glen et al., (2020) and got different group-level results. I examined the output of my afni_proc.py log and noticed that in the new log, the HRF is listed as SPMG1 model, despite not changing the afni version. In looking at the 3dDeconvolve documentation, I note that the SPMG1 HRF model is normalized to a peak value of 1, whereas the SPMG1 HRF model is not. The pattern of results on the subject-wise level are almost identical, but the betas and Thr values differ, and they differ at the second level as I said before.
The duration of the events in the main contrast of interest are the same (3 seconds). Is there any way to know what the scaling is for the SPMG1 HRF model? My assumption is that different normalizations would lead to different betas as a function of the convolving operation. Is this true? And are there situations in which it is valid (including my own) to use SPMG1 over SPMG1? If it makes such a big difference, I might just use the GAM function I’d used in the past. I’d rather not have to reanalyze my data.