 |
 |
Hi Yuan-fang, This isn't right I'm afraid. The [1 -2 1] does not test for a linear increase (or decrease), and will fail to test the hypothesis that is of interest. All the best, Anderson On 7 April 2017 at 13:24, Yuan-fang Zhao <[log in to unmask]> wrote: > Hi Dan, > > Just a rush thought. For a linear increase, say you have 3 conditions A, > B, C, and A > B > C, increase linearly; it is equal to test (A-B) - (B-C) = > 0, which is the same with A + C - 2B; so you can write you contrast vector > accordingly. > > Best > > 2017-04-07 23:26 GMT+08:00 Dan Wolf <[log in to unmask]>: > >> Hi FSL, >> I found this old answer saying that to test a linear trend across three >> groups the correct contrast is -1 0 1. >> That is the same contrast for testing the difference between the two >> groups, which to me means the data in the middle group is simply ignored, >> which I think means that this middle group could be higher or lower than >> both other groups, which is not really within the spirit of seeing if a >> linear trend holds across the three groups. >> Is there any way to test this more stringent "linear trend across three >> groups and the middle group is intermediate" idea with a simple statistical >> test in FSL? >> Thanks, >> Dan >> > > > > -- > > >