Todd,
My question is are the values contained in the output ResMS.img
files scaled and by what value?
Yes, they are scaled by (1/ effective degrees of freedom) = (1/trRV) =
(1/SPM.xX.X)
I want to retrieve just the raw
residual sum of squares values and I imagine they are in the ResMS
files either multiplied by a scale factor or not. Note: By raw
residual sum of squares I mean ResSS = (Y(actual image values) -
Y(estimated image values)) or i.e. the sum(residuals^2))
The following generates the actual ResSS. From your stats directory run
this code:
----------------------------------------------------------------------------------------------------------
load SPM.mat
trRV = SPM.xX.trRV;
V = spm_vol('ResMS.img');
ResMS = spm_read_vols(V);
ResSS = ResMS .* trRV;
P.fname = 'ResSS.img';
P.mat = V.mat; P.dt = V.dt; P.dim = V.dim;
P.pinfo = [1 0 0]';
P = spm_create_vol(P);
for i=1:P.dim(3)
spm_write_plane( P , ResSS( : , : , i ) , i );
end
-----------------------------------------------------------------------------------
Good luck,
Brian
On 9/20/07, Todd Penney <[log in to unmask]> wrote:
>
> Hello all,
>
> I ran a t-test statistical analysis on some fMRI data and I need
> the ResMS image files as part of my study.
>
> When reading spm_spm.m file to find out how the mean squared error
> (ResMS) is calculated, there is mention of a scale factor in 2 places.
> One of the comments says the ResMS values written to the ResMS.hdr
> files are scaled by the effective degrees of freedom. At another
> comment further down, it mentions that the scale factor ,
> 1/effective-degrees-of-freedom, is written into the [struct].pinfo(1)
> variable.
>
> My question is are the values contained in the output ResMS.img
> files scaled and by what value? I want to retrieve just the raw
> residual sum of squares values and I imagine they are in the ResMS
> files either multiplied by a scale factor or not. Note: By raw
> residual sum of squares I mean ResSS = (Y(actual image values) -
> Y(estimated image values)) or i.e. the sum(residuals^2))
>
> Thank you.
>
> Sincerely,
>
> Todd Penney
>
