Actually, looking at the real numbers, my easy regressors have an
average duration of 1.1s, and my hard regressor has an average
duration of 8.3s (though the median is 4.5s; it's inflated by a few
problems that took a long time to solve). The discrepancy I'm seeing
in varcopes is approximately 20x - 50x, depending on where I'm looking
in the brain, so your explanation seems plausible.
Sadly, though, I'm still confused by why the copes would be different.
Let me take a giant step backward and ask a very simple question:
Say I have a visual checkerboard experiment with two different
durations (3s and 8s). Both create a peak 2.5% signal change in the
raw BOLD data, but the first one lasts for ~3s, and the second for
~8s. My regressors are modeled as a 3s boxcar convolved with the HRF,
and as an 8s boxcar convolved with the HRF.
If the cope is reflecting the "scaling factor" by which you have to
multiply the regressor to re-construct the raw signal data, shouldn't
these copes be equal in order to reconstruct the 2.5% signal change?
Thanks,
Todd
On Wed, Feb 17, 2010 at 3:07 PM, Tim Behrens <[log in to unmask]> wrote:
> My guess is that the regressors for the easy are roughly 3x smaller than for
> the hard (as you have modelled them for 1s instead of 3).
>
> This means that, for the same effect size, the copes will be 3x higher and
> the varcopes 9x higher.
>
> T
>
>
> On 17 Feb 2010, at 18:55, Todd Thompson wrote:
>
>> Hi, all. I apologize if this is a dumb question, but I'm having
>> trouble trouble-shooting something:
>>
>> I have a simple slow event-related design:
>> 1) display a hard problem to the subject -- usually around 3s, but
>> variable from 1s to 15s.
>> 2) let them take their time solving it
>> 3) wait 12s for ITI
>> 4) display an easy problem to the subject
>> 5) let them take their time solving it -- consistently less than 1s
>> 6) wait 12s for ITI
>>
>> (repeat for 600s)
>>
>>
>> When I analyze my data, I set up two regressors (easy/hard), each with
>> a 3-column timing file with the subject's RTs for that problem type as
>> the durations.
>>
>> My contrasts have unexpected stats, and in trying to track down why, I
>> discovered that the values in the varcopes for the easy (faster)
>> events are hugely higher than the varcopes for the hard (slower)
>> events. This is true for every one of six different subjects. Is this
>> expected? If so, can anyone explain it to me? If anything, I was
>> expecting the opposite, since the responses to easy trials are more
>> uniform than the responses to the harder trials. (For what it's worth,
>> there doesn't seem to be a difference in stimulus-correlated motion
>> between my two conditions...)
>>
>> Thanks!
>> Todd
>>
>
|