balancing for nuisance effects in cluster-randomization stats

[Eric Maris]

Hi Michael,


I think your question is about experimental design (i.c., control for
confounding variables) and not about statistics. That being said, you could
consider the following:

1. Use a Go-NoGo paradigm and instruct the subject to give the same response
(e.g., NoGo) to the stimulus conditions that you want to compare.
2. Use a blocked Go-NoGo paradigm in which you reverse the stimulus-response
associations between blocks (A-Go and B-NoGo in block 1, B-Go and A-NoGo in
block 2).

Good luck,

Eric Maris


> -----Original Message-----
> From: FieldTrip discussion list [mailto:FIELDTRIP at NIC.SURFNET.NL] On
Behalf Of
> Michael Wibral
> Sent: Tuesday, October 30, 2007 2:30 PM
> To: FIELDTRIP at NIC.SURFNET.NL
> Subject: [FIELDTRIP] balancing for nuisance effects in
cluster-randomization stats
>
> Dear Eric, dear Fieldtrippers,
>
> I have a questions regading the control of certain effects that are
usually
> controlled by balancing over subjects in old fashioned analyses:
> Imagine subjects see stimuli A and B and have to respond with the buttons
C
> (for seeing A) and D (for seeing B). Of course, one could then not
> distinguish between perceptual effects (response to A,B) and motor effcts
> (pressing C,D).
> Usually one would now balance the button presses over subjects such that
one
> (random) half of the subjects gets the inverted instruction (press D when
> seeing A and press C when seeing B). For an experiment with , say, six
> subjects one would get then the following set of correct 'trials':
> set1:
> {1-AC, 2-AC, 3-AC, 4-AD, 5-AD, 6-AD}
> set2:
> {1-BD, 2-BD, 3-BD, 4-BC, 5-BC, 6-BC}
>
> If one now tries to check parametrically whether there is an A versus B
> effect this should work given the balancing has worked.
> In permutation testing using dependend samples the following happens: One
of
> the permutations will be (last three subjects with exchanged conditions):
> permutation set1:
> {1-AC, 2-AC, 3-AC, 4-BC, 5-BC, 6-BC}
> permutation set2:
> {1-BD, 2-BD, 3-BD, 4-AD, 5-AD, 6-AD}
>
> Hence one will get the full C versus D effect in this permutation sample
and
> similar ones in all permutations that are not too far away from it. If the
C
> versus D effect is a large one (as e.g. button presses tend to be) this
will
> definitely dominate the extreme ends of the cluster-t distribution,
killing
> any chance of detecting an A versus B effect. (I assume that
> clusterrandomisation also shouldn't work in this case because the
> prerequsite of exchangeability is violated even when the A/B null
hypothesis
> was true.)
>
> Hence, my question how to design an experiment to control for the
> omnipresent button presses (or motor readiness potentials if one chooses a
> delayed response paradigm)?
>
> Any ideas appreciated,
> Michael Wibral
>
> ----------------------------------
> The aim of this list is to facilitate the discussion between users of the
FieldTrip
> toolbox, to share experiences and to discuss new ideas for MEG and EEG
analysis.
> See also http://listserv.surfnet.nl/archives/fieldtrip.html and
> http://www.ru.nl/fcdonders/fieldtrip.

----------------------------------
The aim of this list is to facilitate the discussion between users of the FieldTrip  toolbox, to share experiences and to discuss new ideas for MEG and EEG analysis. See also http://listserv.surfnet.nl/archives/fieldtrip.html and http://www.ru.nl/fcdonders/fieldtrip.