Yes, your understanding is correct - and for each subject it was a
different order of runs. I agree that it is not difficult to model this in
the batch system - which is why I wanted to do it that way. The question
is - does it matter if the sessions aren't modeled int he way they actually
occurred. Donald's email suggests it doesn't matter. The only I can think
of that it could is if there are any assumptions about those regressors -
that the order does matter - e.g., scanner drift that occurs over time would
be accounted for. My thought was if SPM treated these regressors as ordinal
numbers that are related in time it would be better to reflect that in the
order they were entered into the design. If, however, they are treated as
nominal categories that don't actually ahve a relationship in time then it
shouldn't matter.
elise
On Wed, 26 Aug 2009 17:21:02 -0400, Chris Watson
<[log in to unmask]> wrote:
>Perhaps I'm not understanding something here, but are you saying that,
>you modeled every subject as having seen e.g. lists 1-2-3-4-5-6, even if
>they saw, e.g. 6-5-4-3-2-1?
>With SPM8's batch system and even before, it shouldn't be an
>inconvenience to set up different models for different subjects.
>
>Elise Temple wrote:
>> I am having a debate with one of my graduate students. We have 6 runs
>> (sessions) in this experiment. We have 6 different lists with the
>> conditions in different orders in each list. Because of counterbalancing
>> the order the lists is different across subjects (some do list 1 1st, some
>> list 2, etc). While doing single subject 1st level analysis, for pure
>> convenience reasons, she did not model the sessions in the order they
>> actually occurred, but rather based on the list they were doing.
>>
>> My thought is that the session effects not only represent the variance that
>> is different between each session, but should also represent the actual
>> order the subject did the sessions because there is meaning in the order of
>> them. The question is - does the way SPM deals with this regressor take
>> into account sessions as an ordinal value where their number has meaning or
>> does it treat them as nominal so it really doesnt matter what order they are
>> put in?
>>
>> It still seems to me one's model should match as closely as possible what
>> you think actually is going on - but if its purely for aesthetic reasons she
>> probably doesnt have to reanalyze. On the other hand, if the order is
>> meaningful and by shuffling them we are not taking advantage of how
>> effective the estimate of these regressors are then reanalysis is necessary
>> even if no visible differences are seen between the results.
>>
>> BTW: We are using SPM8.
>>
>> best,
>>
>> elise temple
>>
>>
>> Elise Temple, Ph.D.
>> Educational Neuroscience Laboratory
>> http://www.dartmouth.edu/~templelab/
>>
>> Assistant Professor
>> Department of Education
>> Graduate faculty in Psychological & Brain Sciences
>> Dartmouth College
>> Hanover, NH 03755
>>
>>
|