Stefano

----Messaggio originale----
Da: greve@nmr.mgh.harvard.edu
Data: 4-ago-2013 20.34
A: <freesurfer@nmr.mgh.harvard.edu>
Ogg: Re: [Freesurfer] Statistical concerns

On 8/1/13 7:56 AM, stdp82@virgilio.it wrote:

Hi list,

I have some questions please.

1- I have carried out Monte Carlo Simulation to assess the difference among three groups (23, 24 and 32 subjects). Main results were confirmed also by FDR, but I prefer to shown the Monte Carlo becouse it better describes my thesis.

Recently I have submitted my paper but a referee asked me "Why use cluster based Monte Carlo methods for correction of multiple comparisons instead of the more conventional vertex-wise FDR correction? I suspect the former is much less conservative than the latter. It would be interesting to know the results for FDR corrected thresholds (p<0.05), and whether the present results would remain statistically 'active'.

Now, I'm looking literature and I'm noting that several recent articles published on top journals used Monte Carlo Simulation (Ehrlich et al., 2013 Schizophr Bull.; Sasamoto et al., 2013 Schizophr Bull).

What do you think about it? How do you advise to reply to the referee? I'd like to show Monte Carlo results.

This is a very surprising request by the reviewer. FDR is hardly "conventional", and the brain imaging community has been using clusterwise correction for 20 years. I'm not even sure what "voxelwise" FDR is. Is that different than the one introduced by Genovese in 2002?

As for how to respond to the reviewer, I'm not sure. You can certainly make a case that doing the clusterwise correction is the "industry standard". You can also site a paper out of the Friston group from a few years back critisizing the Genovese-style FDR. Have you applied the Genovese-style FDR that we offer in FS? If that works out, then there is no need to fight with the reviewer.

2- I should do also correlation analysis. I'm using QDEC as recommanded in guideliness (http://surfer.nmr.mgh.harvard.edu/fswiki/FsTutorial/QdecGroupAnalysis_freeview). In one group, after fsgd creation, I'm performing correlation analysis between thickness and neuropsychiatric test scores (it has been put in covariete windows-it should be dipendent factor). I'm also putting some nuisance factors (they should be dipendent factor in the regression analysis) to exclude the their effect on data.

a- Is this procedure corrected?

b- Is corrected to talk regression analysis in the text of my manuscript?

c- Is corrected whether I explain this correlation analysis as reported in the description performed above?

I'm confused. If you are using qdec, then you put your factors (which are independent) in the qdec table, not in the FSGD file. Otherwise the procedure looks correct and you can talk about "regression analysis" in your ms. I don't understand (c).
doug

Thank you very much,

Stefano

_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer