Dear Agnieszka,
Thanks for your email.
The answer is: correct at the second level only.
When you do eg. a t-test at the first level, SPM will write out a
contrast image eg. con_0005.img (as you know, these con images become
the data for your 2nd level analysis). SPM creates the same image *no
matter what p-value threshold you choose* at the first level.
This is because the con images are just a function of the regression
coefficients.
Best wishes,
Will.
A.A.Reid wrote:
> Dear Will,
>
> I just would like to double check with you the issue of correcting for
> multiple comparisons and choosing the p threshold value in the first and
> second level analyses.
>
> I remember that there was quite a lot of information about correcting for
> multiple comparisons at the SPM5 course, however, little was said about
> doing these corrections in the context of both the first level and the
> second level analyses.
>
> My aim is to compare fMRI data for two groups of participants. I would
> like to look at the activation across the entire brain, so I am assuming
> that I need to correct for multiple comparisons, using e.g. (FWE
> correction). However, I am not at all clear in which analysis it should be
> done. Should one correct for the multiple comparisons at the 1st level of
> the analysis (and set the p threshold to let’s say p<0.05) and do nothing
> (no FWE correction and p=0) at the second level analysis? Or, should one
> select no corrections for multiple comparisons and p threshold = 0 at the
> first level analysis and do e.g. FWE correction at p <0.05 at the second
> level analysis? Or, should one use FWE correction and p<0.05 in both
> analyses, which seems a little bit too conservative? Or, perhaps it does
> not matter in which analysis one does the correction for multiple
> comparisons, as long, as it is done once?
>
> The reason I am asking is that I would like to maximise my chances of
> detecting activation, but at the same time I am aware that it must be done
> statistically correctly.
>
> I would be grateful for a fast response, if at all possible.
>
> Thank you very much for your help. Best wishes,
> Agnieszka
>
>
>
>
--
William D. Penny
Wellcome Trust Centre for Neuroimaging
University College London
12 Queen Square
London WC1N 3BG
Tel: 020 7833 7475
FAX: 020 7813 1420
Email: [log in to unmask]
URL: http://www.fil.ion.ucl.ac.uk/~wpenny/
|