Hi Alex,
On 3/31/13 7:37 AM, Alex Kell wrote:
freesurfers,

i have a question about how t-tests are computed in fsfast.  

so t = ces/sqrt(cesvar), but how exactly is cesvar calculated?  

i thought it would be:
cesvar = variance_of_residual *    c*(X' * X)^-1*c'
  where X is the stim matrix and c is the contrast vector and ' denotes transpose
That equation looks right.

two questions:

1. i tried computing cesvar myself, but it is not the equation above (for the one subject i was playing with, cesvar was ~1.5x the above equation).  what is the equation that is used?
Did the FSFAST analysis using temporal whitening (on by default)? Is so, you need to take that into account. If you want to verify your equations, then turn whitening off (--no-whiten).

2. intuitively, the second half of that [ c*(X' * X)^-1*c ] penalizes for correlations in your stimulus matrix, particularly for those relevant to your contrast.  is that a good way to think about it?

Yes
doug

thanks in advance!


alex


_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer