Hello SPMers,
I would appreciate if anyone could advice us on the proper design to
approach this problem. We are interested in looking at sequential effects
in a simple choice reaction time paradigm. We have a two-by-two factorial
design, where the probability and the ISI between two different kinds of
trials (A and B) are varied. Ideally, we would like to have four
conditions: in condition 1 the probability of A and B is the same and the
ISI is short. In condition 2, the probability of A and B is the same but
the ISI is long. In condition 3, the probability of A is 0.61 and that of
B is 0.39 and the ISI is short, and in condition 4 the probability of A is
0.61 and B is 0.39 but the ISI is long. Our contrast of interest would be
(condition1 - condition2) - (condition 3 - condition 4)
Although the block design would be optimal in terms of power, given that
the number of stimuli presented per block should be kept equal, we would
end up with blocks of very different lengths (condition 2 and condition 4
would have to be 4 times longer than condition 1 and condition 3 because
the ISI is 4x), and thus with a different cumulative BOLD signal across
different ISIs. Thus, unless there is a way of adjusting for the
differential degrees of freedom and magnitude of the overall BOLD signal
across ISIs that we are unaware of, a block design wouldn't be suitable
for our study.
We are therefore thinking of using an event-related design but we are
running into the problem of the probability of null events for our
baseline. We know that if the probability of A = probability B then the
probability of null events should be 1/(N+1). However, in two of our
conditions prob A > prob B. We are unsure of what would be the
probability of occurrence of null events in this case. Would there be a
way of including a baseline while mantaining the differential
probability?
Many thanks,
Valeria
|