I think Martin's question has a lot in common with the one I just
asked a few minutes ago, so I'd like to comment here too.
On 6/22/05, Martin Zalesak <[log in to unmask]> wrote:
> Dear all,
>
> I have two groups: controls and schizophrenics. I used the canonical hrf +
> the two derivatives in my model. I created the following contrasts for the
> difference between A and B for each subject from each group:
>
> [1 0 0 -1 0 0] canonical HRF A-B
> [0 1 0 0 -1 0] temporal derivative A-B
> [0 0 1 0 0 -1] dispersion derivative A-B
>
> I was then able to examine the overall effect for _each group_ separately
> using a one-way ANOVA (without constant term), modelling the three contrast
> images as three groups using the F-contrast:
> [1 0 0
> 0 1 0
> 0 0 1]
I think this step is a mistake (one I was making until yesterday!).
This ANOVA means that you are looking for areas where the HRF, temp
and disp differ from each other, not where they differ from zero. So
if you had a giant effect with positive values for all 3 contrasts,
you wouldn't find it in this analysis. Can anyone confirm and suggest
the proper analysis, because I don't know what it is?
Antonia.
-----------------------------------------------------------------------------------
> and also got the t-contrast for the HRF alone:
> [1 0 0]
>
> What if I want to compare the two groups now? Do I do a one-way ANOVA where
> I enter six groups - one for each contrast for each group and then do an
> F-contrast like:
> [1 0 0 -1 0 0
> 0 1 0 0 -1 0
> 0 0 1 0 0 -1]
> to get the overall effect?
>
> Can I do an equivalent of the t-test above, such as:
> [1 0 0 -1 0 0] to find A>B group 1>2?
>
> Please advise. I understand what to do for one group but I am not sure how
> to set this up correctly for between groups comparisons.
>
> Thank you!
> Martin
>
>
|