Dear SPM experts,
I am new to SPM and am trying to match subjects on residual variance after full model fitting. I am basing myself on a paper that used the ResMS.img in SPM2 to do so. I've looked up which information this image contains exactly but because this appears to be different for SPM2, so I just wanted to check if I got it right:
I read somewhere that ResMS.hdr/img contains an image of estimated residual variance (e = Y - Xb --> ResMS = e'e/(n-f), where n = number of scans, f = number of factors) - but understand from Chapter 7 in HBF2 that in reality e'e is divided by trace(RV) to obtain the unbiased estimate of residual variance for complex designs, which is then stored in the ResMS.img.
However, I found in several posts in the SPM mailing list that this ResMS.img is not scaled by the degrees of freedom in SPM2 (contains just e'e).
Is all of this correct and am I right in concluding that in SPM2 the ResMS image does not contain an unbiased estimate of the individual's residual variance but does so in later versions?
thanks,
Tessa
--
Dept. Visual Neuroscience,
UCL Institute of Ophthalmology
11-43 Bath Street
London EC1V 9EL
[log in to unmask]
