Re: calculation of statistics

Hi,

> For simplcitiy sake, let me start with the T-stat.  What does Marsbar divide
> the mean beta/weighted sum beta by to get the t-stat.  It doesn't appear to
> be the standard error of the betas.  From looking at the MarsS structure, it
> seems like it might be extracting info from the ResMS file associated with
> the SPM.mat file of the design, but what exactly and where can i find these
> figures/calulcations in the MarsS structure?

Marsbar gets the residual mean squared error from the ResidualMS
matrix in marsbar's saved version of the SPM.mat ('_mres.mat').  It
then uses this with the design matrix covariance and the contrast to
get the estimated error of the contrast statistic.  It should be
mathematically identical to SPM.  The calculations are - for SPM2
designs and above - in <marsbar>/@mardo_2/private/pr_stat_compute.m.
In there, you'll find these lines for the T statistic:

    con(i,:)   = xCon(ic).c'*betas;
    VcB        = xCon(ic).c'*SPM.xX.Bcov*xCon(ic).c;
    stat(i,:)  = con(i,:)./sqrt(Hp*VcB);

where xCon(ic).c is the contrast matrix, VcB is the variance of the
covariances of the betas, and Hp is the residual mean squared error.

Best,

Matthew

------------------------------------------------------------------------------
Throughout its 18-year history, RSA Conference consistently attracts the
world's best and brightest in the field, creating opportunities for Conference
attendees to learn about information security's most important issues through
interactions with peers, luminaries and emerging and established companies.
http://p.sf.net/sfu/rsaconf-dev2dev