Hi,
It is the case that all "unmodelled" parts of the timeseries are the implicit baseline in FSL.
As for your design, it is possible to model it this way but the contrasts to achieve the mean are more complicated because you have to take account of the relative proportion of button presses to make sure that you are giving the appropriate weight in the calculation of the overall mean. This would then vary for each subject. Alternatively, with the design I suggested you simply have contrasts that are of the form [1 0] or [0 1] to get at the overall mean or the difference between the button presses, and all the relevant weighting by proportion is taken care of within the GLM automatically. So, personally, I would recommend against your current design and go with the one I proposed before.
For trials with no button press, this is most easily dealt with by having a third EV per condition (on top of the mean and button difference EVs) that contains just the trials with no button press. This will then treat those completely separately from the mean and the button difference, as those trials are not typical of what happens when a button is pressed. That would then imply that the overall mean EV excludes these trials, and hence is only the overall mean for the case where a button is pressed, but I think that's normally what you would want to know. If not, then you could incorporate these no-button trials into the overall mean EV instead, and not have an extra EV at all. This is your choice, depending on how you want to treat those trials.
All the best,
Mark
On 3 Oct 2013, at 00:57, Lauren S <[log in to unmask]> wrote:
> Dear Mark-
> Thank you for your response! I hadn't thought to do it that way, and it makes a lot of sense. Since participants very occasionally forgot to press the button, could I enter those trials as 0 in my latter 6 EVs?
>
> Over the past couple of days I created a new design file that also has 12 EVs (well, 13, but I'll get to the null trial in a moment). Instead of having 6 EVs of mean and 6 EVs of difference, I have 6 EVs of button press 1 for each stimulus type and 6 EVs of button press 2 for each stimulus type. I then planned to model the means through contrasts (i.e. by having a contrast that sums the button press 1 EV and button press 2 EV for each stimulus type). Would this get me the desired result as well, or would it be less effective than the method you suggested?
>
> Finally, I was in fact unaware that null trials (fixation cross) shouldn't be modeled. How does it "know" that we are specifically interested in fixation trials as our baseline? Does FSL use any timepoints that are unmodeled as baseline?
>
>
> Thanks so much. This has been very helpful!
> Lauren
|