Hello Rachel,
with reg. to the 2nd question: you could subtract the con-images that you
would usually put into your paired design, and set up a monovariate
correlation between these difference images and the difference of your
behavioural difference.
Alternatively, you could use the classical paired t-test design, explore
the respective clusters, extract the statistical values of those clusters
(e. g. by using the VOI function) for each time point, and perform
correlation analysis between the difference values of those clusters and
the difference values of your beh. variable 'offline', i. e. outside SPM
in a statistical programm.
Both approaches may produce different results, as in solution B you do no
assume a linear relationship first, but identify only clusters that make
it to a group difference. Different clusters may emerge from approach A in
which you identify only clusters that show that linear relationship with
the beh. variable. It depends on the question/context which way is the
more suitable one.
with reg. to the 1st question: not sure, if this is necessary - as
randomization should correct for this "order bias". But I may be wrong...
best regards,
Philipp
Max Planck Institute of Psychiatry
NMR Research Group
Kraepelinstr. 2-10
80804 Munich
|