Hi,
On 29 Nov 2007, at 20:08, Yoshiko Yamada wrote:
> Dear List,
>
> At the last FSL course, it was recommended that we use a fixed
> effects analysis at the mid level
> (i.e., single-subject multiple-session) when fewer than 10 sessions
> had been collected for each
> subject. I have questions regarding this recommendation.
>
> Our data sets consist of 14 subjects with 2-4 runs each (done in a
> single session). (These data
> were obtained from young children, and some of the runs have been
> excluded from the analysis
> due to excessive head motion.) We are interested in making
> inferences about the population from
> which 14 samples were drawn. Thus, my understanding is that you
> recommend using a fixed
> effects analysis on the single-subject multiple-run level and a
> mixed effects analysis on the
> multiple-subject/group level. (Please correct me if I’m wrong.)
That's right.
> When I compared the outputs of the fixed effects and mixed effects
> analyses for each subject
> (combining multiple runs), I see greater extent of activations, and
> this conforms with what I have
> reading (e.g., Friston et al., 1999), namely that a fixed effects
> analysis is generally more sensitive
> to activations than a random effects analysis.
>
> Subsequently, I performed two group-level analyses using a mixed
> effects analysis (FLAME 1); one
> analysis had mid-level fixed effects stats as the input (FE-ME), and
> the other had mid-level mixed
> effects stats as the input (ME-ME). Contrary to my expectation, some
> significant activations
> present in the ME-ME analysis are less robust in the FE-ME analysis.
>
> Why should using a mixed effects analysis at the mid level be more
> sensitive to activations (at
> least in some areas) than using a mid-level fixed effects analysis?
> If we have a reason to suspect a
> large run-to-run variability, would it be better to use a mixed
> effects analysis at the mid level, or
> do you think that run-to-run variance cannot be estimated well with
> a small number of runs we
> have in our data?
It could go either way. Changing the variance model at the second-
level means that different subject-wise variances are passed up to the
third level. This means that the different subjects will be weighted
(relative to each other) differently: In the 2nd=mixed approach, the
weighting will be largely driven by the subject's cross-session
variability. In the2nd=fixed approach, the weighting will be driven by
the ('average' of) the first-level variabilities.
Therefore it's not quite right to say that one approach leads to more
or less 'sensitive' results at the 3rd level, just that they are
asking different questions. For the reasons outlined in the latest
version of the manuals (and on the course), we think that 2nd=fixed is
the most sensible model, and so would recommend that.
For more details on these issues, see the course lecture notes, and
also the papers:
M.W. Woolrich, T.E.J. Behrens, C.F. Beckmann, M. Jenkinson, and S.M.
Smith.
Multi-level linear modelling for FMRI group analysis using Bayesian
inference.
NeuroImage, 21(4):1732-1747, 2004.
and
C.F. Beckmann, M. Jenkinson, and S.M. Smith.
General multi-level linear modelling for group analysis in FMRI.
NeuroImage, 20:1052-1063, 2003.
Cheers.
> Also, I’d like to have better understanding of how these two
> different types of analyses are
> performed in FEAT, and how the stats of these analyses are carried
> to the higher-level analysis.
> Are there any documents or references on these you can point to me?
>
> Thank you in advance for your advice!
>
> -Yoshiko
---------------------------------------------------------------------------
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director, Oxford University FMRIB Centre
FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
---------------------------------------------------------------------------
|