Hi Tom,
Thanks for taking the time to reply.
On Jun 4, 2010, at 1:44 AM, Thomas Nichols wrote:
> Dear Colm,
>
>
> I've been running ANOVA on some VBM in randomise. I'm almost certain I've got the design and contrasts right based on Eugene's email of 27 May last (Re: Correct design setup for ANCOVA with nuisance variable).
>
> So I've got 3 groups L, M, S and they are in the order of the ls command in the GM_mod_merg file.
>
> The (abbreviated to 2 subjects per group) design and contrasts looks like this (I've been following the ANOVA: 1-factor 4-levels example on the webpage):
>
> EV1 EV2 EV3
> 1 1 0
> 1 1 0
> 1 0 1
> 1 0 1
> 1 0 0
> 1 0 0
>
> So assuming EV1 models the last group (S) the EVs are as follows
> EV1: mean
> EV2: L-S
> EV3: M-S
>
> Actually, the interpretation of EV1 is exactly 'S', but as you'll never test it (or the mean) it doesn't matter. Otherwise this is correct.
>
> Contrasts:
> EV1 EV2 EV3 F-test
> C1 L-M 0 1 0 *
> C2 M-S 0 0 1 *
>
> So assuming I've got this right (i'd be grateful if some one could confirm this for me), I then proceed to run randomise using TFCE, threshold (and binarise) the tfce_corrp_fstat1 and tfce_corrp_tstat files at 0.95 and use the resultant masks on to the corresponding fstat and tstat files. Is this correct should I be thresholding the tfce files like this and masking in this fashion? (The exact commands are below).
>
> I'm not sure why you're working so hard. Isn't it sufficient to view the regions of significance in the corrp TFCE image? Or is that you want to view the f-values within the TFCE significant regions? If so, such a masking approach is reasonable.
I was under the impression that once you have the tfce_corrp images from randomise it was necessary to threshold them so that one can say at what P level the correction was performed at. Is that not correct? If this thresholding step is not necessary how does one determine from the results the the P-level of correction for multiple comparisons?
If I could get away without having to do this thresholding step and still be able to state p-level for multiple comparisons correction, I'd be very happy as the regions showing in the tfce_corrp_fstat1 image are great places for this particular cohort of subjects.
Regards,
--
Colm G. Connolly, Ph. D.
Dept of Psychiatry
University of California, San Diego
9500 Gilman Dr. #0855
La Jolla, CA 92093-0855
Tel: +1-858-246-0608
Fax: +1-858-534-9450
|