Hi,
Adding more covariates only helps if they usefully explain variance in
the data (e.g. maybe age helps). The answer to this question will
depend completely on your data and the covariates.
It's not in general acceptable to report uncorrected p-values as a
significant result, as there's then a good chance that the result just
happened by accident. Two obvious ways of possibly 'getting
significance' are a) if you can honestly reduce your search space
(mask) the the areas you are considering then the multiple comparisons
problem is lessed (but it's pretty hard, post-hoc, to honestly
convince yourself that you would have intended to use that mask before
seeing the results) and b) test the other diffusion measures (such as
mean diffusivity and mode) to see if there's an interesting story in
any of them.
Cheers.
On 31 Oct 2008, at 16:05, Yi-Shin Sheu wrote:
> Hello Steve and all the FSL experts,
>
> Regarding to my last post about DTI TBSS analysis (please see
> below), after I double checked the p value for TFCE stats map again,
> I found my max p value is actually 0.76, so my question now becomes
> two-fold.
> 1. Is there any legitimate way to boost up p value? My experience
> is that if I add more covariates, usually it helps.
> 2. If I just can't get the p value boost up to 0.95, how can I
> describe my results more intelligently and objectively? I've seen
> in some DTI paper, they basically just set an arbitrary threshold/
> limit and reported the uncorrected p value and the cluster size, if
> no regions survive after running randomise. If that is what I want
> to do (though I know it is not very legitimate), which randomise
> output I should look at? The TFCE uncorrected p stats
> (tbss_tfce_p_tstat2.nii.gz), or the raw tstats file
> (tbss_tstat2.nii.gz)?
>
> Thank you in advance.
>
> Yi-Shin
>
>
> On 10/30/08, Sheu, Yi-Shin <[log in to unmask]> wrote: Hello
> Steve,
>
> Mmm... I guess easythresh is out when it comes to tbss analysis.
>
> Actually, I want to do both (boost the significance of the result and
> describe them more concisely). However, by looking at the maximum
> value
> of the TFCE enhanced stats map, it is only p<0.38, far from 0.95.
> So if I
> can't boost it to 0.95, how can I describe the results more
> intelligently
> and objectively?
>
> I am only using the existent data and behavioral measure to generate
> some
> preliminary result, so I do not need to be very strict on the
> outcome. I
> just want to present the outcome in a scientific way, and that is
> why I
> need information like cluster size, uncorrected p value, etc...
>
> Any suggestion???
>
> Thank you.
>
> Best,
>
> Yi-Shin
>
>
> > Is your main goal to try to 'boost' the significance of the regions
> > that you found, or to describe them concisely?
>
> > Hi,
> >
> > On 30 Oct 2008, at 14:55, Yi-Shin Sheu wrote:
> >
> >> Dear FSL experts,
> >>
> >> I am doing some preliminary analysis on 42 controls subjects that
> we
> >> have.
> >>
> >> After running randomise on FA skeleton by using the command:
> >>
> >> randomise -i all_FA_skeletonised -o tbss -m mean_FA_skeleton_mask
> -d
> >> design.mat -t
> >> design.con -n 1000 --T2 -V -x
> >>
> >> I didn't get any significant results by looking at
> >> "tbss_tfce_corrp_tstat2.nii.gz" file (the
> >> corrected p stats file), which is expected, since the purpose of my
> >> analysis is basically
> >> exploring.
> >>
> >> So I look at the "tbss_tfce_p_tstat2.nii.gz" file (showing
> >> uncorrected p) in order to get a
> >> sense of what regions might show up. By set the threshold to (1-
> >> p)>0.99, and I did see
> >> two region that really support my hypothesis. But I do not know
> how
> >> to generate a
> >> report that show the cluster size of those regions.
> >
> > You can use the 'cluster' program to produce output images that tell
> > you how large the clusters are. However, this isn't terribly
> useful -
> > this won't give you an 'independent' measure of significance,
> because
> > the TFCE enhancement already takes into account spatial extent.
> >
> >> What I really would need is a command that might be similar to
> >> easythresh, which I can
> >> specify the threshold for "tbss_tfce_p_tstat2.nii.gz" and cluster
> >> size in a sensible way.
> >
> > I'm afraid this would be even dodgier - easythresh inference is
> based
> > on Gaussian random field theory, which a) isn't applicable to the
> > _output_ of TFCE, and b) isn't in any case applicable in the case of
> > TBSS skeletons - sorry!
> >
> > Is your main goal to try to 'boost' the significance of the regions
> > that you found, or to describe them concisely?
> >
> > Cheers.
> >
> >> I did try easythresh on the raw t-stats data, but it seems like
> >> there is a discrepancy
> >> between the generated outputs and what I see on the thresholded
> >> "tbss_tfce_p_tstat2.nii.gz". I am not sure which one makes more
> >> sense to look at. The
> >> command I tried with easythresh is:
> >> easythresh tbss_tstat2.nii.gz mean_FA.nii.gz 3.307 0.001
> >> mean_FA_mask.nii.gz
> >> Easythresh_3.307_0.001_tstat2
> >>
> >> Thank you for any input.
> >>
> >> Best wishes,
> >>
> >> Yi-Shin
> >>
> >
> >
> >
> ---------------------------------------------------------------------------
> > Stephen M. Smith, Professor of Biomedical Engineering
> > Associate Director, Oxford University FMRIB Centre
> >
> > FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
> > +44 (0) 1865 222726 (fax 222717)
> > [log in to unmask] http://www.fmrib.ox.ac.uk/~steve
> >
> ---------------------------------------------------------------------------
> >
> >
>
>
> Sheu, Yi-Shin
> Research Assistant
> McLean Hospital
> Developmental Biopsychiatry Research Program
> Tel: 617-855-2942
> Fax: 617-855-3712
> email: [log in to unmask]
>
>
> The information transmitted in this electronic communication is
> intended only
> for the person or entity to whom it is addressed and may contain
> confidential
> and/or privileged material. Any review, retransmission,
> dissemination or other
> use of or taking of any action in reliance upon this information by
> persons or
> entities other than the intended recipient is prohibited. If you
> received this
> information in error, please contact the Compliance HelpLine at
> 800-856-1983 and
> properly dispose of this information.
>
>
>
---------------------------------------------------------------------------
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director, Oxford University FMRIB Centre
FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
---------------------------------------------------------------------------
|
