SPM2 Gem 6: Contrast SE Computation

An email from Will Penny on how a contrast’s standard error is computed.

Subject: Re: Beta SE in SPM2?
From: Will Penny <wpenny@FIL.ION.UCL.AC.UK>
Date: Fri, 27 Feb 2004 11:26:35 +0000


> How do I extract beta SE in SPM2?  As far as I can see I can only get
> y, Y, beta, and Bcov...have I missed something?

If you're referring to the Standard Error or the Standard deviation of the
parameter estimates then this isn't stored explicitly.

Its computed on the fly when one makes an inference about a contrast. See the
following lines (153-158) in spm_contrasts:

case 'T'                                  %-Compute SPM{t} image
        cB    = spm_get_data(xCon(ic).Vcon,XYZ);
        l     = spm_get_data(VHp,XYZ);
        VcB   = xCon(ic).c'*SPM.xX.Bcov*xCon(ic).c;
        Z     = cB./sqrt(l*VcB);

So, if you did a contrast xCon(ic).c'=[1 0 0 0] to test the null
hypothesis that the first parameter is zero then cB will be equal to
beta1 and sqrt(l*VcB) will be equal to std(beta1) - the standard
deviation of the first parameter estimate.

Note that l is the residual sum of squares at that voxel which is
stored in the image ResMS.img and accessed via the file handle VHp.

In this way SPM does'nt have to store Standard Deviations/covariances
for every voxel and beta - so saving disk space. But its not that
handy for what you want !

Hope this helps, Will.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s