Dear Ben,
Hello. I would just like to give some of my own experiences with a
similar letter version of the task.
The first question is whether you are interested in comparing
incorrect versus correct trials?
If you are then you will need to make sure that the subjects make
"enough" incorrect responses. "Enough" is based on the estimibility of
the regressor. You do not want to model 1 or 2 trials in a regressor
because it will not be a very good estimate of what is truly happening
and will most likely be capturing noise. If you do not have enough
incorrect regressors, then you can make the task more difficult or
longer.
But if you are not interested in comparing correct and incorrect, or
memory load effects across incorrect trials then I would suggest to
model all incorrect responses as a single regressor. Therefore, this
regressor does not split trials based on the memory load, like you do
for the correct trials. This will increase the regressor's
estimability and captures this variance from the data and account for
it. This same approach can be used for your "time-out/failure to
respond" trials.
But like Don says, do not ignore something that you know is in your
data. It just gets left for the error term in that case.
I hope this helps somewhat,
Jason.
On Mon, Mar 21, 2011 at 5:22 AM, Ben Becker <[log in to unmask]> wrote:
> Dear SPMers,
>
> We used an event-related spatial working memory fMRI paradigm with 3 levels of working memory load. The task is a delayed-matching-to-sample task. First a sample with 1, 3 or 5 points (in 30 possible locations) is shown. Subsequently a probe with one point is shown. Participants have to press the left button if the probe point does not match the location of the previously shown 1, 3, or 5 points. If the probe matches the sample they have got to press the right button.
> My question concerns the first level design-matrix:
>
> The matrix consists of the following regressors:
>
> 1. Baseline
> 2. Correct trials WM Load 1
> 3. Correct trials WM Load 2
> 4. Correct trials WM Load 3
>
>
> My questions:
>
> 1. Would it be valid to introduce two further regressors coding the correct and incorrect trials (for all WM loads together, some participants did not make any mistakes in the WM Load 1 conditions, else it would have been fine to introduce separate regressors for the three WM Load conditions)? My main concern introducing the correct and incorrect regressors is that the regressor coding the correct trials will not be independent from the regressors coding the correct trials for the single WM Load conditions.
>
> 2. What about missed trials? They will not be further analyzed in the imaging data analysis – so do I have to code them in a separate regressor? Or would it be ok to ignore the missed trials?
>
> Thanks & Kind regards,
>
> Ben
>
--
Jason Steffener, Ph.D.
Department of Neurology
Columbia University
http://www.cogneurosci.org/steffener.html
|