Dear Nicholas,
You indeed have one term missing in the computation of the t-statistic, see:
https://github.com/spm/spm12/blob/r7219/spm_contrasts.m#L215-L220
Best regards,
Guillaume.
On 11/07/18 04:07, Nicholas Blauch wrote:
> Hello SPM users and experts,
>
> I would like to compute t-contrasts of single beta coefficients to use in multivariate pattern analyses, following Miyaki et. al (2010) showing superior performance decoding with t-values compared to beta coefficients. I'd like to do it on the fly rather than creating spmT images, since this seems wasteful.
>
> I believe the basic operation is to divide the voxelwise beta weight by the sqrt of the ResMS image. I have compared the output of this computation with the output of a formal spmT contrast made with the batch editor. They are related by a constant scale factor (in my test case, 73.55).
>
> My question is: what is the constant scale factor? It should not matter for any subsequent calculations, this is simply so that I can properly say I am dealing with t-values, and for my understanding.
>
> Thank you.
>
> Nick
>
--
Guillaume Flandin, PhD
Wellcome Centre for Human Neuroimaging
University College London
12 Queen Square
London WC1N 3BG
|