Dear Roberto
Thank you very much for your response.
>what unbalancedness does to your analysis is increase the
>susceptibility of the test to violations of the normality assumptions.
>This would happen in a contrast involving B and any other condition,
>since B has half the sample size of the others.
Are there any references or studies to understand how much unbalancedness
is practically acceptable or not (especially in neuroimaging analyses)?
>There is one thing you can do to counter this possible effect,
>however; this is using a large smoothing kernel.
So, this is because large smoothing kernel reduces the possibility of
violating normality assumptions?
Many thanks!
Nobu
>Quoting Nobu Sawamoto <[log in to unmask]>:
>
>> Dear SPMers
>>
>> We performed 3 PET scans consisting of task A, B and C in 7 subjects.
>> Since we have 2 PET scan data consisting of task A and C in another
>> 7 subjects, we analyzed those 14 subjects data together with SPM2 and
>> got some results. Previous post says that missing data is no problem
>> for SPM.
>> http://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind04&L=SPM&P=R403916&I=-3
>>
>> Is the above combined analysis (unbalanced design) also valid? If it is
>> valid, I would appreciate it very much if you could let me know any
>> references discussing the validity of the anlaysis.
>>
>> Many thanks!
>>
>> Nobu Sawamoto
>>
>========================================================================
|