Hi Muhammad,
let me try to give an answer here:
In case of the two-sample t-test you do not need a separate group
regressor, because you specify the groups already via the
corresponding design (second-level model; cf. the first do columns of
the resulting design matrix).
There you would define a covariate for every additional factor you like
to control for in the model.
Hope that helps,
-Lucas
> Hi Udo,
>
> Is this true even for 2-sample T-test, where we specify scans for each
> group differently? Do we still need a separate 'group' regressor?
>
> -Muhammad
>
> On Sat, Jun 27, 2009 at 3:19 AM, Udo Dannlowski
> <[log in to unmask] <mailto:[log in to unmask]>> wrote:
>
> Dear Lucas,
>
> for a group comparison, you would have to model a regressor for
> group. If you
> have regressors for "group", "IV", "age", "gender" you'd have to
> set the
> contrast weight either 1 0 0 0 or -1 0 0 0 for a t-test showing
> group1 >
> group2 or group2 > group1 respectively, controlling for IV, age,
> and gender.
>
> Good luck,
>
> Udo
>
> Lucas Eggert schrieb am 2009-06-27:
> > Dear Donald McLaren,
>
> > >You want [1 0 0] which is the positive effect, assuming a T-test,
> > >of your IV on volume.
>
> > >[1 1 1] is the average effect of your IV, age, and gender. This
> > >would be very hard to interpret.
>
> > Ok.
>
> > In case, I would try a two-sample t-test to compare the gray matter
> > volume between two groups, again controlling for the effect of age
> > and gender,
> > would the contrast be defined analogously, i.e. e.g. "1 -1 0 0"? Or
> > would I do this also in a multiple regression model with an
> > additional regressor 'group'? And would the
> > contrast be "1 1 0 0" then (for "group", "IV", "age", "gender")?
>
> > Thank you very much for your help!
>
> > All the best,
> > -Lucas
>
> --
> Dr. med. Dipl.-Psych. Udo Dannlowski
> Cognitive Neuropsychiatry Research Group
> Department of Psychiatry
> University of Münster
> Albert-Schweitzer-Str. 11
> 48149 Münster
> Phone: ++49 251 8356601
> Fax : ++49 251 8356612
> email: [log in to unmask] <mailto:[log in to unmask]>
>
>
|