Print

---

Hi John,

Please see below:


On 13 February 2017 at 21:48, John anderson <[log in to unmask]> wrote:

> Dear FSL experts,
> I have two groups of subjects and I want to study the difference between
> the groups using the command :
> randomise -i 4Dmerged.nii.gz -vw -m mask.nii.gz -d design.mat -t
> design.con -n 5000 -T
>
> 1. In order to study the difference between the groups without covriates.
> I used in my design the following:
> 1 0
> 1 0
> 1 0
> .
> .
> .
> 0 1
> 0 1
> 0 1
> .
> .
> .
> etc
>
> 2. I want to include age covariate: I demeaned age between the groups and
> included it in a third column in the design.
>
> 3. I want to include age and gender covariates: I included the demeaned
> age in a third column of the design, and I have created a fourth column and
> gave (male=1, female=0).
> Regarding this analysis, are there nay rules to choose (0, 1 ) for
> categorical variables (e.g. male/female). When I gave one for male, and
> zero for female, I found difference between the groups. If I choose Zero
> for male and one for female, the difference disappear. Why this happen? How
> this can be explained?
>

Provided that sex is used only as nuisance (i.e., it is not part of any
contrast, nor involved in any interaction), then coding as 0 vs 1, 1 vs 0,
+1 vs -1, -1 vs +1, pi vs. sqrt(2), etc, makes no difference on the
contrasts being tested.

If, however, you test the effect of sex, then you need to know which was
coded a higher or lower value, so you can tell the direction of the effect.
In that case, the results will change vastly: it's as testing A>B vs. B>A.

All the best,

Anderson



>
> I highly appreciate your valuable comments!
>
> John
>
>