On Fri, Jul 13, 2012 at 5:45 PM, Burns, Scott S
<[log in to unmask]> wrote:
> SPMers -
>
> I'll describe my 2nd-level design matrix. I have 15 subjects and I'm bringing up their first level (TaskA - Baseline) contrast images to a one-sample T test. I also have a behavioral test score (collected out of the magnet) for these 15 subjects (we can call it TEST). I've set the covariate (TEST) to have an interaction with the first factor (TaskA - Baseline). Given the first column in the design matrix is the contrast images and the second is my covariate, for the following T contrast vectors I would love input on whether 1) is it valid, and 2) how one would interpret the
results.
(1) The first column of the design matrix is the constant term, your
contrast images are the dependent variable. If you demean your
covariate, then the constant will represent the mean of your contrast
images. If you don't demean your covariate, then the constant is the
intercept of the contrasts as if the covariate was 0.
>
> [1 0]
This contrast is testing if the constant is greater than 0.
>
> [0 1]
This contrast is testing if the slope of the covariate is greater than 0.
>
> [.5 .5]
This contrast is meaningless as there is no way to interpret the
average of the covariate slope and the constant.
>
> [1 -1]
>
> [-1 1]
These two contrasts are also meaningless as there is no way to
interpret the difference between the constant and the covariate slope.
>
> Thank you for any and all help you can provide.
Whenever you are about to create a contrast, start by creating the
null hypothesis (e.g. Is the constant greater than 0?) and then make
sure you can interpret the null hypothesis if you reject it. Then you
can use the null hypothesis to form the contrast vector.
>
>
> Scott Burns
> Neuroimaging Analyst
> Education and Brain Sciences Research Lab
> Kennedy Center
> Vanderbilt University
|