Optimal fMRI multi-subject analysis requires that the multiple sources of variation are
modelled. The two principal sources are intrasubject variability, in the form of scan-toscan
(possibly autocorrelated) noise, and intersubject variability, in the form of individual
differences in response magnitude to the same stimulus. In this talk we will review the
history of hierachical modelling in functional neuroimaging and highlight a well-used
shortcut, the summary statistic approach (Holmes and Friston, 1998). We will describe
how early PET modelling neglected the repeated measures issue, and suffered little
from this, and yet how fMRI exposed the issue dramatically. We will review the
motivation for the summary statistic approach and its strengths and weaknesses. While
the approach is unparalleled in its ease of application, its simpler implementations have
several limitations, some well known (e.g., that only a single image per subject can be
considered), and some little-known (e.g., it may make an assumption of homogeneous
intrasubject variance). We will then review standard mixed effects modeling approach
to such data, and then two widely used fMRI methods implimented in FSL3 and SPM2.
Copyright. All Rights Reserved.