Posted by Jian on March 09, 2001 at 09:43:29:
I am a beginner of AFNI, in the frequently asked questions, I got the following information:
(Q52. How can I do a t-test on time series data?
There are two possible ways to do intra-dataset comparisons to get a t-test result. The first is to use program 3dttest. At present, this requires breaking the input dataset into pieces, 1 sub-brick per dataset, since 3dttest can only read 1 sub-brick datasets. You could do this break up with the command line program 3dbucket, used once for each sub-brick to be extracted. This would be ugly, but possible.
The second way is to use the FIM analysis with the ideal set to be equal to 1 for the "on" periods and equal to 0 for the "off" periods. A correlation with such a 0-1 valued ideal function is mathematically equivalent to a t-test between the "on" and "off" time points. The correlation coefficient that is returned can be converted to a t-statistic via the formula)
My question is:
1. if I have serveral fucntional runs (on and off) across sessions and want do a t-test on time series data, I want to determine if 'on' state is different from 'off' states for a group subjects. I can use the second method describe in the frequently asked questions, doing correlation one by one and then use 3dmerge to averge the different functional runs togather, is that resonable? is it possible if I average (use 3dmerge) different runs first and then do correlaton with the average one?
2. if I use 3dttest, since I have only one datasets I have to use -base1, is there a method to figure out the nubmer of base1 using my time series data?
Thanks