Pia Rotshtein wrote:
> In SPM5 there is a new option for conjunction analysis: testing an
> intermediate null hypothesis.
>
> If I understand correctly, in my case I used conjunction of three contrasts.
>
> So my null hypothesis is: no more than 2 contrasts are real
>
> Rejecting it will suggest that all three contrast are likely to be real.
>
[...]
>
> The explanation about the conjunction with the intermediate null,
> mentions that one can only use it for contrast that have same degrees of
> freedom.
Hi Pia,
Did you ever get a reply to this, and/or work it out? My understanding
is as follows:
The "global null" is 0 effects under the null, and if rejected, you
can infer there is at least one effect. The "conjunction null" is that
there are n-1 (2 in your case) effects under the null, and that if
rejected, all n (3) contrasts are significant.
In SPM5, if you have three contrasts and choose "intermediate null",
then this is that there is 1 effect under the null, and rejection
implies 2 or more effects from the three.
After choosing "intermediate null", with more than 3 contrasts (at
least in SPM5) it prompts for numbers between 1 and n-1. I think
choosing n-1 is equivalent to choosing "conjunction null" instead of
intermediate. Any particular number, means that many effects under
null, rejecting the null means more than that number.
So, if you are picking "intermediate" in SPM5, and/or the number under
the null as 1, then you are testing for 2 or 3 out of 3. While if you
choose "conjunction null" or pick the number as 2, then you are
testing for all three.
I'm not certain about your "same DF" question, but I think the
limitation is that all the contrasts must come from the same SPM.mat
-- this ensures that they have the same error/denominator DF. The
numerator DF may be different for different contrasts but I don't
think this matters; and the different group sizes within a
pooled-error ANOVA, won't, as far as I can see, affect the DF. So I
think that should be fine.
Best,
Ged.
|