< Just a little reminder of my earlier question, as I noticed
similar discussion was raised :) >
(I have n subjects, for each three "different drug" sessions,
and within each session 7 different condition (doses).)
Thanks a lot
Naj
N. Khalili Mahani
Leiden University Medical Center
[log in to unmask]
On Tue, 14 Apr 2009, Najmeh Khalili-Mahani wrote:
> Dear Tom,
> Thanks for the prompt response. I am afraid averaging in my case is not the
> solution; since each of the sessions are acquired under different
> experimental factors (i.e. drugs).
>
> So, for each session, I am interested in differences among the 7 conditions.
> But, i am also interested in differences between sessions. In fact the main
> effects are less important than contrasts. The reason for preferring ANOVA
> over t-test is because I want to control for (as much as possible)
> within-subject correlations between sessions, as well as between conditions.
>
>
> Would you still think t-test is the best way to go?
>
> (If I may brave a "meta" GLM design; how many permutations must I do in
> randomise?)
>
> Thanks a lot,
> Naj
>
>
> On Tue, Apr 14, 2009 at 9:31 PM, Thomas Nichols <[log in to unmask]>wrote:
>
>> Dear Naj,
>>
>> Randomise, in as much as it implements a GLM, should be able to address
>> most of your questions.
>>
>> While you say you have a 'two-way within-subject ANOVA', if one of your
>> factors are session, I won't actually consider that a factor... just average
>> over it with a contrast in Feat (if using fmri) or just average the session
>> results w/ fslmaths. Hence you basically just have a one-way ANOVA with 7
>> conditions.
>>
>> BUT, are you really interested in a 6- or 7-degrees-of-freedom F-test?
>> Testing for *any* differences among all those conditions (6 DF)? Or
>> testing for any non-zero effect in *any* of those conditions (7DF)?
>>
>> Rather, I bet you probably have some specific question to ask between those
>> 6 conditions, and each question can be answered with a (1DF) t-test. If
>> this is the case you don't need anything more fancy than a 1-sample t-test
>> on the appropriate contrast (or computed average difference).
>>
>> Does this help?
>>
>> -Tom
>>
>>
>>
>> On Tue, Apr 14, 2009 at 4:48 PM, Najmeh Khalili M. <
>> [log in to unmask]> wrote:
>>
>>> Hi,
>>>
>>> I am wondering if randomise is suitable for performing a two-way
>>> within-subject ANOVA (3 sessions x 7 condition) to determine main effects,
>>> interaction and post-hoc contrasts?
>>>
>>> I apologize if this question is already answered, but I cannot trace it
>>> back in archives or the tutorial. Any hint is appreciated.
>>>
>>> Best
>>> Naj
>>>
>>>
>>
>> ____________________________________________
>> Thomas Nichols, PhD
>> Director, Modelling & Genetics
>> GlaxoSmithKline Clinical Imaging Centre
>>
>> Senior Research Fellow
>> Oxford University FMRIB Centre
>>
>
|