Dear Dick and dear Sander
Many thanks to you and all the SPMers who gave us some very fruitful
comments to our question about self-paced er-fMRI (Kalina's comments were
particularly interesting too).
* Note that the bias you refer to (differences in reaction
* times between, say, patients and normal controls) will occur regardless of
* the design you choose: the period of interest, i.e. the stimulus-response
* interval, will be the same for fixed-paced and self-paced designs. Only
the
* response-stimulus interval will vary in a fixed-paced design, whereas it
is
* constant in a self-paced design.
Absolutely. However, stimulus response variation can be taken into account
within the analysis (we were interest to see that RT dependent regressors
were already published - many thanks to Kalina for revealing us this paper),
whereas variation due to a different protocol exposure seems to us less easy
to manage.
* Another issue is the difference in total
* performance time in a self-paced design; until now, we dealt with that
* problem by using a fixed number of scans (observations) with a variable
* number of trials depending on the individuals' response times.
Although the problem of a difference of power due to a different number of
observations (scans) can be taken into account by this proposition, the
paradigm still different. Isn't it possible that it could imply different
strategies and hence different processes ?
In fact what puzzle me in choosing this kind of protocol for comparing two
groups, is that you introduce a new bias. Event protocols give the very
interesting opportunity to compare what is the more similar that we can get
between patients and controls : succeeded trials. Thus the classical bias
related to difference in performance is taken into account. I will
acknowledge that this is not always the case, especially when you only have
a force 2 choice response (different performance may lead to the assumption
that there is a different amount of guess responses in succeeded trials).
However, if while taking into account the performance effect, you introduce
another systematic bias between the two populations, inference that you can
get from your study may be damped.
Let me rephrase this differently and imagine that your are the referee (hard
job) of an article where we describe a smaller frontal activation for
schizophrenics vs. controls, using a self paced study, and that patients
either performed less trials or toke more scans to perform the same number
of trial than controls.
What would you think of the following conclusion : hypofrontality is due to
the group effect ?
Wouldn't you raise doubt about this because of a systematic bias in stimulus
presentation (ISI and number of stimuli) ? Hypo-frontality is probably not a
good example because it has been largely demonstrated in this population but
what about an unexpected area ?
A last technical question related to your self-paced event design : how do
you arrange the order of your stimuli (orthogonal ? anti-correlated ?) ? Did
you compute the most appropriate order for the maximal "discriminability"
between your two conditions ? If it is the case, can the timing variations
challenge the discrimination between conditions ? The same question in the
case of ~20% false responses ?
Thanks for your input and thanks for any other
Jack
|