Ah, I think I understand now. I believe doing a correlation across voxels such as you described is invalid since they are not independent observations. As you mentioned, because of smoothing you would expect there to be a relationship, and I do not think there is anything you can do to make this comparison valid. What would be ok would be to average across voxels within an ROI and do a correlation across subjects, assuming you have enough power in your sample. Sorry if I am still misunderstanding.

Cheers,
Michael

On Tue, Oct 26, 2010 at 9:10 AM, Goulas Alexandros (PSYCHOLOGY) <[log in to unmask]> wrote:

Dear Michael,

 

Thanks a lot for the quick response. I think that I didn’t describe well enough what I would like to do:

I have connectivity metrics (strength/corr coef with some seeds calculated on resting state) for a set of voxels in a mask, along with their T values from a different data set (where there is a cognitive paradigm). Hence I have two vectors that contain the values for the voxels at hand. I do NOT want to examine which correlations are significant, something that would require correction for multiple comparisons, but to examine if the higher the strength the higher the T values obtained for a voxel. So I think that I only have to adapt the degrees of freedom when I compute the correlation between the variables/values between the vectors mentioned above, right? The scatter plot indicates a clear linear relation but we don’t have as many independent observations as voxels, so I need to use this correction to obtain a meaningful p value.

 

Thanks for the help!

 

Alex

 

From: [log in to unmask] [mailto:[log in to unmask]] On Behalf Of Michael T Rubens
Sent: dinsdag 26 oktober 2010 17:56
To: Goulas Alexandros (PSYCHOLOGY)
Cc: [log in to unmask]
Subject: Re: [SPM] degrees of freedom/level of significance estimation

 

You should consider doing a standard correlation, and then using a cluster extent threshold to correct. I think the vol_corr code you are referring to is for a small volume correction, which would be used after the fact to correct for multiple comparisons as I described rather than running a correlation, but I may be mistaken. I'm more inclined to use stat_threshold by the late great Keith Worsley though (http://www.math.mcgill.ca/keith/fmristat/#thresholding).

Cheers,
Michael

On Tue, Oct 26, 2010 at 8:30 AM, Goulas Alexandros (PSYCHOLOGY) <[log in to unmask]> wrote:

Dear SPM experts,

 

   I would like to know if there is a handy script that calculates the Pearson correlation values that correspond to a specified alpha value, given the estimated smoothness in the data and the voxels at hand, by using Gaussian Random Fields. More specifically, I want to check the significance of the correlation of the values of the voxels within a mask, obtained in 2 different sessions. I found a vol_corr utility by M. Brett, that seems to do the job, but I have problems compiling to routines.

 

Many thanks in advance!

 

Alex

 

 

 




--
Research Associate
Gazzaley Lab
Department of Neurology
University of California, San Francisco




--
Research Associate
Gazzaley Lab
Department of Neurology
University of California, San Francisco