I would average the PET scans and average the covariate. Then run the
regression. You only have 17 subjects, so 17 observations would be a
correct way to model the relationship. The only thing you have done by
sampling each multiple times is increase the accuracy of the estimate
in each subject, but haven't changed the between-subject variance.
Best Regards, Donald McLaren
=================
D.G. McLaren, Ph.D.
Research Fellow, Department of Neurology, Massachusetts General Hospital and
Harvard Medical School
Postdoctoral Research Fellow, GRECC, Bedford VA
Website: http://www.martinos.org/~mclaren
Office: (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.
On Wed, Jul 31, 2013 at 4:03 PM, MCLAREN, Donald
<[log in to unmask]> wrote:
> This will not work as the GLM error term is the within-subject error term.
>
> You need a linear mixed model with multiple error terms - which I
> don't believe is available in SPM. There is the LME model in
> Freesurfer and I believe AFNI has a mixed model with multiple error
> terms as well.
>
>
> On Wed, Jul 31, 2013 at 3:53 PM, Chris Brown
> <[log in to unmask]> wrote:
>> Dear all,
>>
>> I'm uncertain about how to model the following in SPM.
>>
>> I have data from a PET study involving 17 subjects, 2 within-subject conditions, and a covariate of interest that was measured for every subject and each condition. I'm not interested in the condition effect or condition*covariate interactions for this analysis. I just want to know what the main effect is of the covariate across subjects, and would ideally like to model this covariate by including the data from both conditions to improve the power in the model (e.g. using a random intercept model). If I were to simply average the PET data across conditions and perform a regression with the covariate of interest also averaged across conditions, I will lose power by halving the number of data points. I also cannot perform a regression including all 34 data points (17 subjects times 2 conditions) because the 2 observations per subject (i.e. 2 conditions) are not independent, and this will lead to lower P values than is realistic.
>>
>> My understanding of the above problem is that a mixed model might be appropriate. So I'm wondering if the following would work. If I implement a flexible factorial design, I can include factors for 'subject' and 'condition', add the covariate of interest (without interacting with condition), and run a model in which the main effect is 'condition'. When selecting the contrast of interest, if I select '1' for the covariate, does this model the regression I am looking for and also control for the clustering of the observations (i.e. the 2 conditions) effectively? Or do I need to totally different model?
>>
>> Many thanks
>> Chris
|