Dear Anderson,
Thank you so much for your advice. But in the tripled t-test in GLM manual, I don't understand how can we model the condition A, B and C from PE1 and PE2 (A= a+b, B= -a, C= -b).
Excuse me for another question. The aim of our study is 60 subjects with the comparison of before and after treatment, if I carry all in one test, the number of input will gonna be huge (60 * 3* 2 = 360 inputs). I wonder if I can split it into 3 tests of two-sample paired t-test (60 * 2 = 120 inputs) according to each conditions?
Thanks alots,
Kim
On Thu, 5 Mar 2015 09:34:41 +0000, Anderson M. Winkler <[log in to unmask]> wrote:
>Hi Kim,
>
>The design you show isn't right I'm afraid. Use the tripled t-test as
>indicated in the GLM manual.
>
>All the best,
>
>Anderson
>
>
>On 4 March 2015 at 18:38, Kim TRAN <[log in to unmask]> wrote:
>
>> Dear FSL experts,
>>
>> I'm conducting a 2 sample t-test on 20 subjects, each scanned under 3
>> conditions A, B, C. I'm interested in finding the affects of 1 condition vs
>> each others. So 3 EVs are set for each condition. I enter subject 1 with
>> the conditions A, B, C and then subject 2 with the conditions A, B, C, ect.
>> until subject 20 (each input is individual FEAT result at first level) like
>> below:
>>
>> Subject1: 1 0 0
>> Subject1: 0 1 0
>> Subject1: 0 0 1
>>
>> Subject2: 1 0 0
>> Subject2: 0 1 0
>> Subject2: 0 0 1
>> ...
>> Subject20: 0 0 1
>>
>> I was planning to compare these conditions with these contrasts:
>>
>> C1 (A>B): 1 -1 0
>> C2 (A<B): -1 1 0
>> C3 (A>C): 1 0 -1
>> C4 (A<C): -1 0 1
>> C5 (B>C): 0 1 -1
>> C6 (B<C): 0 -1 1
>>
>> Should I split the Evs into 3 groups according to the conditions? Because
>> I got 2 different results for 2 cases of designs, I'm concerned that could
>> the different cross-subject variance change the result? Or the design could
>> change the statistical test from two sample t-test onto triple t-test?
>>
>> Thanks in advance,
>>
>> Kim
>>
>>
>
|