That approach, plus a correction for autocorrelation of time points,
would be accurate for the single subject analyses; however, the group
level statistics are based on the group variance, not variance of the
subject data itself.
For example if everyone had a correlation of .01 between voxel A and
the PCC, the T-statistic would be the same as if the correlation was
.5 in all the subjects. This is because there is no variance between
subjects.
Additionally, the input into the one-sample t-test are the Fisher's
transform of the correlation coefficients to make them normally
distributed. Although, you could convert these back to an r-value to
use your formula -- but the r/t conversion at the second level relates
to the correlation of the regressor and correlations across subjects
and not to the correlations between regions as I said earlier.
Best Regards, Donald McLaren
=================
D.G. McLaren, Ph.D.
Postdoctoral Research Fellow, GRECC, Bedford VA
Research Fellow, Department of Neurology, Massachusetts General
Hospital and Harvard Medical School
Office: (773) 406-2464
=====================
This e-mail contains CONFIDENTIAL INFORMATION which may contain
PROTECTED HEALTHCARE INFORMATION and may also be LEGALLY PRIVILEGED
and which is intended only for the use of the individual or entity
named above. If the reader of the e-mail is not the intended recipient
or the employee or agent responsible for delivering it to the intended
recipient, you are hereby notified that you are in possession of
confidential and privileged information. Any unauthorized use,
disclosure, copying or the taking of any action in reliance on the
contents of this information is strictly prohibited and may be
unlawful. If you have received this e-mail unintentionally, please
immediately notify the sender via telephone at (773) 406-2464 or
email.
On Mon, Sep 13, 2010 at 10:32 AM, michel grothe <[log in to unmask]> wrote:
> Dear Paola,
> functional (and structural) connectivity between ROIs is usually defined by
> a (thresholded) correlation coefficient r. Thus, once you have chosen a
> threshold for r you could calculate a corresponding T-value for your sample
> size based on the formula: T=r*sqrt(df)/sqrt(1-r^2). I have never done that
> myself nor read about it, so this is just an idea.
>
> Best regards,
> Michel
>
>
>
>> Date: Mon, 13 Sep 2010 15:06:31 +0100
>> From: [log in to unmask]
>> Subject: [SPM] how to threshold one sample statistics with very large
>> number of subjects?
>> To: [log in to unmask]
>>
>> Dear mailing list,
>>
>> the problem that I am now facing seems to go in the opposite direction
>> than the one I encounter in most of fmri experiments....
>> I analyzed by means of functional connectivity (with the PCC as seed
>> region) a large number of healthy subjects (N>100) and I would like to do a
>> one-sample statistic on the 100 functional connectivity maps of my subjects.
>> The problem is related to thresholding the results of the one-sample T test.
>> Even if I choose a very low FWE corrected p-value (e.g., p=0.0000001 FWE
>> corrected), I obtain an average functional connectivity map of my 100
>> subjects including most of the entire brain. Only if I arbitrarily choose a
>> large T value (e.g., t=20 or 30) for thresholding, I obtain a nice average
>> functional connectivity map (resembling the correct DMN pattern, as
>> expected). If I do a one-sample t test on a subset of my subjects (e.g., on
>> 30 subjects), I obtain almost the same average functional connectivity map
>> by using a "normal" threshold of p=0.05 FWE-corrected.
>> Do you have some suggestions on how to threshold the one-sample results of
>> my large dataset?
>> Kind regards and thanks for any advice
>> Paola
>
|