Useful Papers - Percent Signal Change
Dale & Buckner (1997), "Selective averaging of rapidly presented individual trials using fMRI," Human Brain Mapping 5, 329-340 PDF
Summary: Once of the first series of event-related fMRI papers, this might have been the first fMRI paper to explore event-related averaging in parallel fashion to ERPs. The authors test the linearity of the addition of HRFs in a visual stimulation paradigm and find that overlapping HRFs sum in a roughly linear fashion - a key to the sort of deconvolution done today.
Bottom line: Event-locked averaging can be done even with overlapping trials, and single events can generate a strong enough response to be analyzed with fMRI
Glover (1999), "Deconvolution of impulse response in event-related BOLD fMRI," NeuroImage 9, 416-429 PDF
Summary: The other side of the linearity issue. For longer stimuli (longer than 4-5 seconds), Glover shows that the HRF does not track the standard boxcar-convolved-with-hrf model at all. A variant of the full FIR model, Wiener deconvolution, is used to show the effectiveness of linear deconvolution with stimuli separated by at least 4 seconds and with a subject-specific deconvolution filter.
Bottom line: Linear deconvolution works - at least up to a point. For longer stimuli, boxcars convolved with an HRF aren't really appropriate.
Ward, AFNI 3dDeconvolve manual (a good reference to skim, particularly early parts on theory) PDF
Summary: A great overview of the theory behind the FIR model and many of the issues influencing construction of those models.
Aguirre et. al (1998), "The variability of human BOLD hemodynamic responses," NeuroImage 8, 360-369 PDF
Summary: We saw this guy earlier... Aguirre et. al tested various sets of 40 subjects tested on various days and in various sessions. They found a good deal of variance accounted for by differences in subjects, and significant differences for many subjects between different scanning days. Within the same day and subject, thought, the HRF seemed relatively stable.
Bottom line: Shows that subject-to-subject and day-to-day variance in HRF can be high, but within a day across runs, the HRF is relatively stable.
Miezin et. al (2000), "Characterizing the hemodynamic response: Effects of presentation rate, sampling procedure, and the possibility of ordering brain activity based on relative timing (Ed. - whew!)," NeuroImage 11, 735-759 PDF
Summary: Ditto on seeing this earlier. Excellent look at the factors influencing the HRF and how stable all its aspects are. Demonstrates HRF remains stable and linear within a subject and certain timing parameters, but outside those, less so.