On Thu, 16 Jun 2005 13:39:35 +0100, Juraj Kukolja <j.kukolja@FZ-
JUELICH.DE> wrote:
>Dear statistics experts:
>
>Basic question:
>If I have conditions A and B and want to compare both with condition C
>(design matrix: [A B C]; each condition with equal trial number) in an
fMRI
>experiment, a common procedure would be to use contrast [1 1 -2].
>What does the "double weighting" of condition C do to it? Does this
>contrast merely multiply beta values by -2, or does it multiply the trial
>number of condition C by -2 and thus artificially reduce the variance? Is
>this a problem for the statistical inference?
>Is there any literature specifically dealing with this topic?
[1 1 -2] is correct. This contrast refers to the null hypothesis that
1*beta_A + 1*beta_B -2*beta_C = 0
or, equivalently, that
average of (beta_A, beta_B) = beta_C
It doesn't do anything to the number of trials and handles variance
correctly; as you say, it merely mutliplies the beta value of C by -2 in
the comparison.
>
>Thanks in advance
>
>
>Juraj
|