I want to check my understanding of a few things regarding 3dDeconvolve.
First, an IRF is the numerical solution of one of the functions in a convolution integral (the other being the known stimulus input function) that is calculated based on the actual known MR data, estimates of the baseline, linear trend (assuming -polort 1), and noise. Each input stimulus has its own IRF function. IRF isn't equal to the observed signal due to a stimulus; observed signal due to a stimulus is the IRF convolved with the input simulus function (the solution of the convolution integral).
Second, the partial F statistic relates how well the convolved IRF of a particular stimulus accounts for the observed signal. This calculation is based on how well the IRF convolved with the input stimulus function fits to the actual observed signal (minus the baseline & noise).
Third, the full model F statistic is computed using all of the calculated IRFs, the input stimulus functions, and estimates of baseline, trend, and noise to relate how well the data fits the full model.
Fourth, minlag and maxlag specify the duration that a response will last. Thus, if a response to input x appeared 1 TR after application and ended 3 TRs later, setting the minlag to 0 and maxlag to 3 would correctly identify the IRF. while a minlag of 2 and maxlag of 3 would not. A 2,3 setting would cause 3dDeconvolve to find an IRF that begins at a lag of 2 TR (but not before) and ends at least 1 TR later. (Basically, setting contraints on how many TRs previous are summing with the current TR).
Is all (or even any) of this correct? Please instruct me. I've read the whole manual and all of the tutorials, but a lot of it is greek to me.
Thanks!