Print

Print
Hi - this is the same issue as when two groups / conditions / etc  are not significantly statistically different from each other, but one just passes thresholding, and the other doesn't.  Such scenarios are an inevitable result of having to apply a final statistical thresholding to any analysis (eg corrp<0.05).    The closer the "successful" test is to *just* passing the thresholding, the more likely is this scenario.

The danger is that one should not make a claim that implies that the two conditions (or in your case halves of the brain) are significantly different from each other, without explicitly testing for that. In the case of two conditions, this means just contrasting them, and seeing if their difference is significant.

Cheers, Steve.




On 24 Jul 2012, at 16:46, Lindsay Walker wrote:

Dear FSL community,

This is not a question specific to a single analysis, but rather a general question.

I have, on more than one occasion, found statistically significant clusters on one side of the brain and not on the other.  However, if I loosen up the requirements for statistical significance (for instance, at p<0.07 instead of p<0.05) I find that the other hemisphere is suddenly involved also.  This is always done on data that is corrected for multiple comparisons.

Personally, with that type of result, I do not believe that there is any real hemisphere-specific difference between my subject groups.

Is there any feeling as to why this occurs?  Any concensus on how this type of finding should be reported in the literature?

I look forward to hearing your thoughts!

Lindsay.



---------------------------------------------------------------------------
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
---------------------------------------------------------------------------