Hi,
To disentangle the intercept from the slope (which is what you want to do to look at correlation-like relationships) you should demean both of your covariates and add a column of ones to separately model the mean (intercept). The contrast just has a one associated with the demeaned covariate of interest.
All the best,
Mark
Leighton Barnden <[log in to unmask]> wrote:
Before plunging into 'randomise' to generate nonparametric TFCE stats, can you confirm that my design and contrast files are OK please? My design is a basic 1 sample T test with one covariate of interest (severity) and one nuisance covariate (global level).
design matrix (example for 5 subjects - actually have 25) where column 1 is severity and column 2 is global.
33 700
44 755
28 801
42 707
49 853
contrast matrix (for negative regression with severity)
-1 0
and the command would then be
randomise -i OneSamp4D -o OneSampT -d design.mat -t design.con -n 1000 -T
where design.mat and design.con are text files?
Thanks Leighton
|