Dear Andrew,

I have 13 subjects with two repeated measurements. 
My design matrix is:
1 0 (subject1, time1)
1 0 (subject2, time1)
1 0 (subject3, time1)
0 1 (subject1, time2)
0 1 (subject2, time2)
0 1 (subject3, time2)

And my contrast is [-1 1] or [1 -1] for time2-time1 and time1-time2.
I also add the exchange block: [1:13 1:13]

I was wondering, is that corrected?

And I also check the matrix of 'nbs.NBS.test_stat', I feel the t-value matrix from NBS is quite different with the matrix I directly estimated from paired t-test.
Is there anything wrong?

Thanks for your help.

Hi Li-Ming,

The design matrix would be as follows:

 1  1 0 0 (subject 1, time 1)
 1  0 1 0 (s2 t1)
 1  0 0 1 (s3, t1)
-1  1 0 0 (s1, t2)
-1  0 1 0 (s2, t2)
-1  0 0 1 (s3, t2)

The first column is the main effect of time. Columns 2, 3 and 4 model the subject means.

Contrast: [1 0 0 0] or [-1 0 0 0]

Test: t-test

Exchange blocks: [1 2 3 1 2 3]

This contrast will test for an increase/decrease between the two time points.

Andrew


Originally posted by Li-Ming Hsu:

Hi Andrew,

Thanks for your reply.

Li-Ming