Sorry to bring this issue up again but this is a particular problem for
myself and a few others here - so was wondering if anyone knew a good way
around it.

Basically I'm doing an event-related study that consists of 3 scan runs -
the long and the short of it is that I have 6 different memory response
categories, which means that not all runs will contain all 6 categories. 
Problem is that I need all 6 repsonse categories to conduct a linear trend
analysis.  All of my EVs are essential for my contrasts.  As a result it is
not possible for me to just leave an EV blank in a given run and carry the
contrast forward to the higher level (subject) analysis.

Someone here suggested that it would be ok to model each EV within each run
and then using the 'input cope files' function at the higher level create an
average for each EV with which to make contrasts (at the subject level). 
This in theory overcomes the problem of empty EVs for us - however when
comparing results that are produced by making contrasts at the lower level
and those that are produced by making contrasts at the higher subject level,
clusters and effects sizes are markedly reduced - even to the point that
when averaging across all participants effects can dissappear (see )

Is there anyway of leaving an EV blank and still conducting contrasts at the
lower level or a way of increasing the power of the contrasts made at the
subject level to mirror those made at the lower level?  What is the best way
for us to go?

Any help would be really really appreciated.