Hi Rik, Dorian,
I've been looking into this a bit lately because i'd like to set up a design where events won't
have the same duration, and is there a bit of a problem because the blood flow model
created by SPM will be better for some durations than others? I.e., particularly when the
durations are very different, you might get distorted results? E.g. Glover 1999 found
marked discrepancy between blood flow response as predicted from a brief event, and
actual blood flow (though maybe this no longer applies), and Zwart et al 2009 note that in
this setting
<If the GLM design matrix used describes the response to one stimulus type less accurately
than the response to another, the residual after fit will be larger. Since the standard
deviation of this residual is used to compute significance of the fitted amplitude, the
computed t-score will be reduced. This not only reduces detection power, but will often be
erroneously interpreted as decreased activation.> Zwart et al., 2009. p1656
|