Hello Jess,
Yes, use 1/400 Hz. Load one of your single-subject models and then double click on the design matrix illustration. SPM will report the corresponding beta values in the Matlab window, one column for each of the regressors. I've plotted the time course of the beta values corresponding to a condition with onsets at 587.5, 827.5 and 987.5 s and different high-pass filters, see attachment. Accidentally, I used a TR of 2 seconds insted of 2.5, and I assumed a length of 750 scans = 1500 s, but this should not matter for illustrative purpose.
Basically, you would roughly expect values of zero as long as there is no stimulation, and then an increase each time one of your stimuli is presented (so, boxcar function convolved with canonical HRF). You will get this only if you don't use any high-pass filter at all (which is not recommended due to noise and so on). As you can see, if you use a high-pass filter of 1/128 s, then the expected time-course looks quite odd (this is visible in the design matrix as well, see the different shades of grey where it should be uniform). This is due to the fact that one would need some slow frequencies < 1/128 s to reconstruct the signal in a proper way. When setting the high-pass filter to 2x max. distance according to Nyquist theorem, the time-course looks more realistic. In your case, there seems to be not that much of a difference between 1/400 and 1/600 Hz, so 1/400 Hz should be ok.
In case your events were shorter, then you would still lose some information with a filter of 1/128 Hz, but the unwanted side effects would be less dramatic.
Hope this helps,
Gabor
|