Dear Takanori,
> Dear SPMers
>
> I have a question about adjusted sampling error option in realignment procedure.
> I investigated the change in the total variance of image time series by adjusting sampling error.
> First, I performed realign and reslice with and without use of this option,
> and sampled the entire volume of time series.
> The size of each image matrix (time * voxel) is completely the same.
> The total variance of time series was calculated by var(var(image_matrix $B!G (B)).
>
> As a result, Var=1.9650*10^8 in applying the adjusted sampling error,
> but Var=1.2025*10^8 without this option.
> It seems that the total variance increased in the adjustment of sampling error.
> Will this result be acceptable? In what case does such a result arise?
I don't think there is any case where the variance after "adjustment of sampling error" should be
larger than without the adjustment.
I think perhaps the problem is in your estimation of the total variance. I don't quite understand
the $B!G stuff (possibly a Netscape problem) but assuming your data is organised in an m-by-n matrix
Y where m is the number of time-points and n the number of voxels then var(Y) will give you an
1-by-n row-vector with the variance for each voxel. Now what you want (I suspect) is the mean, or
equivalently the sum, of variance (over time) across all voxels, which you get from mean(var(Y)).
Your calculation above (which I assume is equivalent to var(var(Y))) will give you the variance
(across voxels) of the variance across time, which seems like a rather strange entity. Even if your
matrix is organised differently I still think that var(var()) will always give you the wrong
(strange) answer.
Good luck Jesper
|