When the residual variance is exactly 0, this is what the matlab code. In the C implementation, if (glm->rvar < FLT_MIN) glm->rvar = FLT_MIN;

which will have similar results.

doug

Lars M. Rimol wrote:

Hi,

We have experimented a little with different ways of computing the p-values in a GLM analysis based on FreeSurfer data.

The way the Matlab reference implementation (fast_fratio.m) handles a situtation where the residuals are exactly zero, seems to be to set those vertices to F = 0 and p = 1. Is this correctly understood? Is this also the case in the underlying C-code?

Thank you!

LMR