Hi Ashley,
a very basic question: are you looking at the corrp maps to get the p-values? If so, you should instead look at *1-corrp*, as specified in the randomise manual ("randomise produces (...) sets of P-value images, stored as 1-P for more convenient visualization, as bigger is then "better"). Significant p-values then correspond to corrp>0.95.
If you were really looking at 1-corrp map, then Steve or Tom would have a better clue on what's going wrong...
Cheers,
Gwenaelle
--- En date de : Lun 20.10.08, Ashley DeMarco <[log in to unmask]> a écrit :
> De: Ashley DeMarco <[log in to unmask]>
> Objet: [FSL] TBSS t-statistics
> À: [log in to unmask]
> Date: Lundi 20 Octobre 2008, 18h03
> Hello All,
>
> I have a question about the randomise tool we are running
> and the
> t-statistic results we are getting. We are looking at
> correlations between
> a behavioral measure and FA values for 10 subjects using
> TBSS. We have
> demeaned our behavioral measure, created a design matrix
> and run the following:
>
> randomise -i all_FA_skeletonised -o tbss -m
> mean_FA_skeleton_mask -d
> stats.mat -t stats.con –D -n 5000 --T2 -V
>
> We get t-statistics that are too high for their
> corresponding p-values and
> also, we are observing that the p-values and t-statistics
> don’t properly
> correlate; as p-values decrease (become more significant),
> our t-statistic
> is actually decreasing as well. Any insight or
> recommendations on this
> would be greatly appreciated.
>
> Thanks,
> Ashley
__________________________________________________
Do You Yahoo!?
En finir avec le spam? Yahoo! Mail vous offre la meilleure protection possible contre les messages non sollicités
http://mail.yahoo.fr Yahoo! Mail
|