Hi Dav,
If you assume that the neural activity of interest is on during the RT period, and that the RT period is 400-550 ms, then by modelling it as 1000 ms you are introducing two sources of error. First, the shape of the BOLD response for a 1000 ms epoch will not be the same as for a 500 ms epoch. Though this effect is likely to be small, the size of the error may be different in the two conditions and depending on your S/N, may be significant. Second, there is a distribution of trial durations, but you are assuming that they are constant i.e. equal to 1000 ms. Although the beta weights will be proportional to the mean, you are ignoring the variance around the mean. This will increase your residual and lower your z-scores.
cheers,
jack
On Tue, 6 Dec 2011 15:01:43 -0800, Dav Clark <[log in to unmask]> wrote:
>Hi,
>
>I'm modeling an experiment in which participants select an action (and immediately execute it) after seeing a cue. The duration of the action selection process is different in two different conditions (i.e., RTs are longer in one condition than in the other). Mean RTs are in the 400-550 msec range.
>
>My hunch is that I should just set the duration in the 3-column EV format to "1" and the beta weights will be sensitive then to a combination of duration and the instantaneous amount of neural processing - i.e., they will be related to the sum of physiological activity over that period of < 1 second. Is there any reason to use a different duration for different responses? Or a shorter or longer fixed duration?
>
>I have a 2 second TR if that's of any relevance here.
>
>Thanks!
>Dav
|