On Sat, 30 Aug 2008 10:15:16 +0300, John Gelburg
<[log in to unmask]> wrote:
>Thanks for your answers. For example, I have design with 2 conditions:
>picture of scene and name of the scene. Suppose I want to remove
>volumes 2 and 5. So, based on what you suggest my design will look
>like:
>
>id reg_cond1 reg_cond2 dummy_reg1
dummy_reg2
>1 1 0 0 0
>2 0 1 1 0
>3 0 1 0 0
>4 1 0 0 0
>5 1 0 0 1
>6 0 1 0 0
>
>In trials to remove I have "1" in both reg_cond and dummy_reg. As far
>as I understand, the reason that this event is removed is because in
>dummy_reg there is only one event with "1", which makes it's weight in
>GLM modeling much stronger than reg_cond for events, which I remove.
>This is the reason I need separate regreessors for each event to
>remove. Correct?
Remember, as was previously pointed out by the other commenter, you have a
separate regressor for every _volume_ you want to remove. Which may or
may not be the same as an "event".
In terms of the statistical reasoning, in a simple GLM model, having a regressor
with a single 1 and the rest zeros means that the _other_ regressors' weights
will not depend on this observation. Why? Because no matter what those
other weights are, the weight for this dummy regressor can always be set (by
the maths, of course) so the error at that observation is zero. So the value
of the signal at this time point cannot possibly affect the other weights.
(One can imagine that this statement might not be quite true given all the
fancy statistics SPM does, but it's probably good enough.)
One thing that I'm a little concerned about is that you want to exclude
volumes for things like erroneous subject behavior.
If you're excluding volumes for things like scanner artifact or motion artifact,
you know exactly which volumes to exclude. But if you're excluding because
of erroneous behavior, you have to exclude enough to take into account any
hemodynamic response to the neural activity associated with that behavior.
That could be quite a few volumes.
In general, it's always best to discard "bad" data, though of course there are
situations when that can be "too expensive" and the data aren't so "bad" that
they can't be salvaged.
>
>Thanks again.
>
>
>
>
>On 8/30/08, MCLAREN, Donald <[log in to unmask]> wrote:
>> Another alternative is to added N regressors for each bad trial. Where
>> N is the expected length of the response. This essentially removes the
>> effect of the frames from bad trials.
>> Each regressor added should have a 1 for a bad frame. So that n-1
>> frames will have a value of zeros and the bad frame will have a value
>> of 1.
>>
>> On Fri, Aug 29, 2008 at 3:17 PM, <John> <Gelburg>
>> <[log in to unmask]> wrote:
>>> Hi,
>>>
>>> I am creating my ER design with optseq. Suppose some volumes I don't
want
>>> to
>>> include in my analysis (for example, incorrect subject' behavioral
>>> response). Is it possible just to remove the volume and numbers of
>>> consecutive events will be shifted? I would assume this to be problematic
>>> from ER deconvolution point of view. Then, is it's possible, what is the
>>> correct way to do this.
>>>
>>> Thanks!
>>>
>>
>>
>>
>> --
>> Best Regards, Donald McLaren
>> =====================
>> D.G. McLaren
>> University of Wisconsin - Madison
>> Neuroscience Training Program
>> Tel: (773) 406 2464
>> =====================
>> This e-mail contains CONFIDENTIAL INFORMATION which may contain
>> PROTECTED HEALTHCARE INFORMATION and may also be LEGALLY
PRIVILEGED
>> and which is intended only for the use of the individual or entity
>> named above. If the reader of the e-mail is not the intended recipient
>> or the employee or agent responsible for delivering it to the intended
>> recipient, you are hereby notified that you are in possession of
>> confidential and privileged information. Any unauthorized use,
>> disclosure, copying or the taking of any action in reliance on the
>> contents of this information is strictly prohibited and may be
>> unlawful. If you have received this e-mail unintentionally, please
>> immediately notify the sender via telephone at (773) 406 2464 or
>> email.
>>
|