Tom, Torben, others -
Another brief comment on the issue of standardising results reporting, and
a question:
I think it would help to distinguish: a/ reporting what one’s actually done
in sufficient detail to allow the analysis to be replicated - essential and
already what’s expected, even if we aren’t always sure what’s sufficient,
b/ reporting quality-control of data and analysis, and c/ perhaps allowing
reviewers &/or, later on, interested readers, to see supplementary data of
type a/ or type b/ as a matter of course
My question concerns when & how much we should care about residuals that
don’t wash white. My understanding is roughly:
1 - that in a single-level procedure using a prewhitening approach (e.g.
SPM2) (I think it’s true of a full ‘mixed-effects’ analysis as well as a
fixed effects analysis, but correct me if I’m wrong), anything resulting in
grimy residuals will bias the statistics in favour of over-significance
i.e. false positives. As bias can be quantified as a proportion (of the
variance of a true contrast to that of its estimator, see e.g. Friston et
al 2000, NI), presumably this problem can be quantified for a known effect,
if we wish. 2 - But, in a two-level random effects (‘summary statistic’)
procedure such as the one many of us use in SPM2, incorrect prewhitening at
the 1st level will simply produce less efficient estimators of the true 1st
level betas and therefore add to the second level contrast variance to a
relatively small extent
Is this right? This is the kind of reasoning that leads me to think that
with present procedures including diagnostics of residuals at the first
level is not necessary (which doesn’t mean it isn’t an important quality
control tool, and may not become more important if procedures change).
I’d also like to say that the great thing about this kind of discussion is
that hopefully as well as perhaps producing the guidelines people are
talking about, the community can find out (in a nice polite fashion!) just
what it does and doesn’t agree on...
All the best, Alexa PS It seems my fairy (hairy?) godmother put my name on
the Wiki for me, but I'm pleased to be a part of it!
|