dear list.
i habe a basic question according the design of a parametric model at first
level. i did not find a sufficient answer in the archives (or i was to
stupid to locate one), so i need help.
my design is a simple approach with two different stimuli, each with 3
different strengths of stimulation, like ABC (first stimulus) and abc
(second stimulus) with greater stimulation from a to b to c.
i modelled a design with 6 colums (a A b B c C). i want to know which brain
areas show activation related to the stimulus strength for each stimulus.
so i assume i have to model simple parametric contrasts.
my first problem is about the usage of the numbers for the regressors in
the design. for the hypothesis of a<b<c do i have to specify something like
[-1 0 0 0 1 0] (as some colleagues of mine favour, which i think must be
wrong) or would [-1 0 -0.5 0 1.5 0] be better.
if the second would be the answer, what about the problem that in this
design the differences between the factors are unequal (a - b =0.5 , b - c
= 2). does this assume a different increase other than true linear in which
every step is equal? is there a better design for a three step design?
second problem is related to the first one. what if i am not sure if i
should expect a true linear increase and not a different one. would
modelling a f-test be better than a t-test?
any idea or link to preexisting answers would be great.
thanks a lot
markus
|