On Wed, 10 Aug 2005 08:32:30 +0300, Cyril Pernet
<[log in to unmask]> wrote:
>Hi Stephen,
>
>>>Hi Sherif,
>>>
>>>>Hello,
>>>>A few years ago I came across the following design:
>>>>ArArArArBrBrBrB
>>>>
>>>>with r being a 'baseline' condition while A and B were 'neutral'
>>>>and 'emotional' conditions, respectively (r=15 sec, A and B = 30 sec
>>>>each). The rationale of the authors was that they wanted to avoid
having
>>>>condition A contaminated by condition B (assuming it could happen if a
>>>>block or blocks of B were to be placed before a block of A during a
given
>>>>run).
>>>>
>>>I don't see anything wrong to separate A and B .. the prblm of blocked
>>>designs is always the same (assuming similar effect along the blocks)
>>>whether A & B are intermixed or not
>>>
>>>
>>
>>I'd have to disagree with that. It's clear there's a significant low-
>>frequency component in the A-B contrast in the given design. The
simplest
>>way to see that is to consider the case of a linearly increasing signal.
>>For simplicity, suppose we've already "regressed out" the mean, so the
>>signal starts out at (say) -1 and ends up at +1. In that case, clearly
>>the estimated value for the parameter corresponding to "r" is zero,
to "B"
>>is positive, and to "A" is negative. So A-B shows up negative, and
>>doesn't appear to be negligible.
>>
>>
>Well I agree with the low frequency as I said in the previous e-mail..
>but for A and B separately I really don't 'see' the prblm. Why, taking
>your example, estimates would be -1 and 1? Given that A and B are
>totally separated within the session it's similar to run two separate
>sessions (but without the pblm of two sessions), one with A and one with
>B (except the transition A r B). The increasing of the signal along the
>session is possible but I don't see really why it should be? is there
>any paper showing that? or is that related to something else?
Right, I agree [as I've communicated off-list] that there's no problem
with low-frequency drift if we view it as
Ar1Ar1Ar1ArBr2Br2Br2B
and consider (A - r1) - (B - r2).
Which raises the question: why can't similar arguments be made against
the claim that it's not good to compare conditions in different sessions
(or runs)?
>>Also, I'm not sure what you mean about the problem of blocked designs.
>>For detecting a signal (though of course not actual estimation of the
>>shape of the response), block designs are superior to event related
>>designs. Their disadvantage is that psychological considerations may
>>disallow them.
>>
>>
>Well, of course, it was about psychological/physiological factors,
>detection power is clearly superior.
>
>Best
> - cyril
>=========================================================================
|