Hi Maria,
> I have one group of subjects either male or female, each with a pych. score.
>
> I would like to show where activation correlates, both positively and
> negatively with the psych score after the effects of gender (or another
> psych. score) have been removed. I'm not sure if I would use a multiple
> regression model since the ancova model only allows one covariate in
> spm99.
Multiple reg should be fine, just code gender as a binary covariate
(male = 0, female = 1 or similar)
> A simple question as well : in a regression model, in my contrast a 1
> means a positive correlation, a -1 means a negative correlation , does
> '0' mean the effects of the covariate are removed?
Basically, yes. I think a clearer way to understand things, is that a
t-contrast (c) specifies a null hypothesis (c'beta=0). If you have a
multiple regression model with covariates/betas for gender, score1,
and score2, then a contrast of [0 0 -1] has the null hypothesis:
0*gender + 0*score1 + (-1)*score2 = 0
or in other words:
-1*score2 = 0
The other variables, gender and score1, are still in the model under
this null hypothesis, which means that their effects are adjusted for
(or "removed") when testing score2 like this.
SPM (all versions, to the best of my knowledge, but certainly SPM5)
treats t-contrasts as right-tailed, so the alternative hypothesis is:
-1*score2 > 0
which means
score2 < 0
so, you can see that in this case [0 0 -1] tests for a significant
negative correlation with score2, after gender and score1 have been
controlled for.
Other contrasts should follow reasonably easily from this.
Best,
Ged
|