The outcome or dependent variable must also be mean centered and I doubt that it is. Just add a column of 1s to your design matrix. It won't hurt, it will most likely just make your model fit better.
Jeanette
On Thu, Oct 14, 2010 at 10:31 AM, Kirstie Whitaker <[log in to unmask]> wrote:Yes - I did already mean center all the EVs. I should have said that.
Thank you for all your help - I'd been reading other posts which said that [-1 1] was an interaction, but in retrospect they all had one or two categorical variables. Thank you for clarifying :)
KxOn 14 October 2010 07:26, Jeanette Mumford <[log in to unmask]> wrote:
Just an added note. Is your outcome variable already mean centered (your FA values)? If not, you also need a column of 1's in your design matrix to model the overall mean.
Jeanette
On Tue, Oct 12, 2010 at 8:59 AM, Michael Harms <[log in to unmask]> wrote:Hello,
Your interpretation of Contrasts [1 0] and [-1 0] are correct.
However, the contrasts [-1 1] and [1 -1] are not assessing the
"interaction" (in the statistical sense) between behavior and age, but
rather whether the effect of age on FA is "larger" than behavior ([-1
1]) or vice-versa ([1 -1])
Interaction effects typically involve either only categorical variables,
or whether the effect of a continuous variable differs across groups
(i.e., whether there are differential slopes between groups). The
interaction of two continuous variables would really just be assessing
whether the product of the continuous variables was significant (and to
test that you would need a separate EV).
cheers,
-MH
On Mon, 2010-10-11 at 23:02 -0700, Kirstie Whitaker wrote:
> Good morning!
>
> I was hoping that you could help me understand the meaning of an
> interaction contrast when using randomise for TBSS data.
>
> My two evs are behavior (EV1) and age (EV2). (These are children so
> the age effect on both behavior and FA is positive). Both EVs are
> demeaned.
>
> As far as I understand:
> Contrast [1 0] shows areas which have a positive relationship between
> FA and behavior after correcting for the effects of age on FA.
> Contrast [-1 0] shows areas which have a negative relationship between
> FA and behavior after correcting for the effects of age on FA.
>
> How would I interpret the contrasts [-1 1] and [1 -1]?
>
> Thank you so much for your help!
>
> Kx
>
--
Kirstie Whitaker
Doctoral Candidate
Cognitive Control and Development Laboratory
Helen Wills Neuroscience Institute
University of California at Berkeley
tel: 510 684 2456
web: bungelab.berkeley.edu
blog: http://bungelab.blogspot.com/