Dear Christidi,
the multiple regression testing for polynomial effects is described in the CAT12 manual (p. 19, 26):
http://www.neuro.uni-jena.de/cat12/CAT12-Manual.pdf
You can use the attached function to orthogonalize your parameters. In the contrast manager you can finally define contrasts for linear effects or quadratic effects or for testing for any linear or quadratic effect. If you have one sample (1st column in your design matrix) and linear (2nd column) and quadratic (3rd column) effects you can use the following contrasts:
T-test
• For positive linear effect: 010
• For positive quadratic effect: 001
• For negative linear effect: 0 -1 0
• For negative quadratic effect: 0 0 -1
F-test
• For any linear effect: 0 1 0
• For any quadratic effect: 0 0 1
• For any linear or quadratic effect (2 rows):
0 1 0
0 0 1
Best,
Christian
On Fri, 23 Mar 2018 10:37:24 +0000, Foteini Christidi <[log in to unmask]> wrote:
>Dear experts,
>
>we would like to check nonlinear effects on vbm for continuous variables (eg age, clinical severity or even cognitive scores) in order to examine whether some regions correlate with the variable of interest on a linear way only, nonlinear only and both linear and nonlinear. In the latter, we would like to check the variances explained by linear and nonlinear terms as well.
>Let's say we want to check it for a cognitive score X.
>The more appproriate way to do that is by regression??? using age,tiv,gender as nuisance variable; and X as regressor (for linear); X^2 (or X^3,etc) as regressor (for nonlinear). By fusing the two resulted maps we can visually check areas with linear and nonlinear effects. Is that correct?
>If we want to regress out the linear effect when we examine the nonlinear one (and vice versa), should we use the X^2 as regressor and the X as nuisance??? or something else?
>
>Thank you in advance
|