Dear Dr. Pernet,

Thank you for your reply. 

In fact, in my case, using a CDT 0.001 leads to only 1 cluster which does not survive any correction (Pfwe = 0.98, Qfdr = 0.65, Puncorr = 0.96). 
So, if I have some a priori ROIs that I am interested, is it a valid approach to extract the beta-values in each of them to compare the groups outside SPM?

Thanks again

On Mon, Jul 25, 2016 at 4:49 PM, PERNET Cyril <[log in to unmask]> wrote:

Hi Fernanda 


no difference 1.5T or 3T, CDT 0.005 simply gives to large blobs for cluster RFT to be valid so you cannot use this. Now when you do CDT 0.001, maybe you still have significant effect looking at the FDR corrected values rather than RFT? that's valid too (although more a discovery type of thing)


in all cases, you should plot estimates in ROI you ad before the experiment to see if these is truly no difference or if you just under threshold. Note however that an absence of effect is not a proof of absence, i.e. you cannot conclude there is no effect, only that you did not observe any (maybe there is no effect, maybe you have to much variance within groups, maybe you don't have enough power) - to prove the null you need to do a Bayesian analysis.




--
Dr Cyril Pernet,
Senior Academic Fellow, Neuroimaging Sciences
Centre for Clinical Brain Sciences (CCBS)

The University of Edinburgh
Chancellor's Building, Room GU426D
49 Little France Crescent
Edinburgh EH16 4SB
[log in to unmask]
http://www.sbirc.ed.ac.uk/cyril
http://www.ed.ac.uk/edinburgh-imaging



From: [log in to unmask] <[log in to unmask]> on behalf of Fernanda Palhano <[log in to unmask]>
Sent: 25 July 2016 19:48
To: PERNET Cyril
Subject: Follow-up question after Keep calm and scan on post
 
Dear Dr. Pernet,

I have tried to post a comment on your OHBM blog, but I couldn't (There was an error submitting your comment. Please try again). It was I am writing this email.
After reading the post, I have a follow-up question about this topic. 
If I am not wrong, all data used to run the analyses in Eklund et al, 2016 was acquired on a 3T scanner. Could exist some difference in the previous finds if one is using data from a 1.5T scanner?

In our lab, we are analysing data from depressive patients and health volunteers acquired on a 1.5T GE scanner. The results when comparing patients vs. health volunteers using a RM-ANOVA only appear under a p <0.05 CDT, and most of clusters do not survive to any correction (FWE or FDR). Could we report this results? Or, should we infer that there is no effect at all?

Thank you very much in advance.

Sincerely,
Fernanda Palhano
PHD student at Brain Institute - UFRN/Brazil

The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.