Dear Andy,
> I was just about to ask the same question as Eric about the
> calculation of contrast images. Now that it has been answered, my second
> question is - what is the error term for testing the difference between
> contrast images (say, condition A vs. B, within subject, in the simple
> case)? Is the error term (1) a pooled residual sum squares from the GLM or
> is it (2) based on the specific standard error of beta_CondA vs. beta_CondB
> difference, as in a classic regression model testing for differences
> between betas? I hope the question makes sense....
Yes, I think so. The answer is 1) a pooled residual sum of squares from the
GLM. As you imply, these analyses assume that the errors for all the
observations in the model are drawn from a random normal distribution
with the same variance. In the above case this would mean that the errors
for observations in condition A and condition B are from the same (normal)
distribution.
Cheers,
Matthew
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|