Kelly
>
> Hello,
>
> I want to make a multiple regression in SPM5 and I have some doubts:
>
> 1) Do I have to put the variable of interest into the model (option
> “covariate” that appears below “Scans”) and the confounding variable
> outside (option “covariate”)?
>
> 2) Into the option “covariate” that is outside, we have this menu:
> vector, name, interactions [none, with factor 1, with factor 2, with
> factor 3], centering [overall mean, factor 1 mean, factor 2 mean,
> factor 3 mean, no centering, user specified value, as implied by
> ANCOVA, GM]. In which cases should I use the options “with factor” on
> the interactions? When and how can I use the options that appear into
> “centering”?
>
> 3) If I want to make a multiple regression with two groups,
> should I put the group like a covariate? If so, how contrast could I
> put for this variable?
>
1) regressors of interest get into the covariates of the multiple
regression sub-menu ; remove intercept -
2) covariates submenu is used for regressor of no interest (age, total
volume etc) -- note that indeed doesn't matter anymore (see notes in
spm) - you do not have to compute interaction, it's your own choice to
choose how you control confounding factors - for sure centering is a
good idea, personnaly I prefer zscoring all my covariables before
inputting stuff into matlab - ie all regressors (interest or not) have
the same weight
3) yes put this as regressor of interest (see previous e-mail on the
same topic)
Best
Cyril
--
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.
|