Geoffrey Aguirre wrote:
>
> As regards your question, I would think that smoothing the data and
> smoothing the statistical map could not be rendered equivalent in any simple
> way, as smoothing the data actually alters the degrees of freedom of the
> test statistic (due to alterations in the modeled noise power spectrum).
>
> I'm eager to hear what the SPM gurus have to say about this,
I'm eager to hear what they think, too! If I've understood it
correctly, Worsley's scale space search method, described in Human Brain
Mapping 4:74-90, is heavily based on sequential spatial smooths of maps
approximating random Gaussian fields -- Z maps, or SPM{Z}'s. Is
smoothing of Z maps invalid?
If it's okay to spatially smooth Z maps, and if the t distribution
approaches a normal Gaussian as the degrees of freedom goes up, can we
conclude that it's therefore okay to smooth a t-test map (given a large
number of df; see Section 3.4 of Human Brain Mapping 4:58-73)?
I'll check out your paper.
Joe.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|