I recently watched a video in which Rick Reynolds made a case to avoid filtering resting state data at short TRs because of losses of degrees of freedom. He suggested using 3dTproject to get rid of unwanted frequencies (if I understood correctly). However, wouldn’t that result in a loss of degrees of freedom? My sense was it mainly served to make the issue more explicit.
If the suggestion is not to bandpass at all, in the absence of physiological recordings, how could one address the frequencies associated with respiration (for example)? If one collects data at TR=2, then the frequencies could still include physiologically meaningful nonbrain signal (if respirations were slow, for example).