Hi,
This is not quite right.
You only need one design matrix and it is like the one in your step 4 but slightly (and importantly) different.
The difference is that it should have 346 columns, not 347. The first two columns you've used are wrong. You should replace them with one EV that is -1 for the first 345 rows and +1 for the other 345 rows. The rest of the EVs are fine.
If you do this change, and use the same matrix for both first_utils and randomise then you'll be fine.
All the best,
Mark
On 24 Sep 2012, at 05:43, Jiyang Jiang <[log in to unmask]> wrote:
> Hi Mark,
>
> Thanks very much for your reply!
>
> As I am a newbie to design matrix, I want to confirm with you if my understanding is correct.
>
> I have 345 subjects scanned twice (altogether 690 scans), and what I want to do is a paired t-test of the two time points, i.e. what were changed during the follow-up period. What I did was
>
> 1). concatenating the 690 scans in orders using "concat_bvars",
> 2). designing a matrix using GLM with 690 inputs and 1 EV. The first 345 inputs were valued -1, and the last 345 were valued 1. T and F contrasts were both 1.
> 3). using the command "first_utils" to generate a 4D image
> 4). designing a matrix using GLM with 690 inputs and 347 EV's for randomise. The first column was 345 0's and 345 1's in order, and the second column was 345 1's and 345 0's. The third to the 347th columns represented subject 1 to 345, and the 690 rows represented 690 scans. 1 was entered if the scan belonged to that subject, making the last 345 columns two 345*345 identity matrices.
> 5). using the "randomise" command to do the statistical analysis
>
> If there is any problem, could you please point it out?
>
> Thanks very much,
> Jiyang
>
>
|