Hello Andy:
Yes, the unequal numbers of odd and even half-TR events are responsible for
the differences in the parameter std. devs. However, I'm not sure that
further randomization is necessary or useful. The fact that the input
stimulus functions will always be of finite length means that some structure
is unavoidable. For example, there could be a tendency for the even half-TR
events of condition B to be followed by odd half-TR events of condition D, etc.
So, it's not just the individual parameter std. devs. that you would have
to worry about, but all of the off-diagonal elements of the covariance matrix.
Fortunately, at least for the internally calculated statistics, program
3dDeconvolve "knows" about this structure, and this is taken into account.
So, I would just use a separate randomization for each run, and for each
subject, and not worry about the half-TR randomization. I guess it depends
on what it is that you want to do with the results.
However, if you insist on equal numbers for the half-TR events, you could
try the following: (Warning: I haven't tried this myself)
Divide each of the conditions into two sub-conditions (for odd and even
half-TR events). Also, the remaining null events will now be specifically
coded as a separate condition.
RSFgen \
-nt 450 -num_stimts 9 -seed 957821451 -one_file -prefix myStim \
-nreps 1 16 -nblock 1 2 -nreps 2 16 -nblock 2 2 \ <-- Condition V1
-nreps 3 06 -nblock 3 2 -nreps 4 06 -nblock 4 2 \ <-- Condition I1
-nreps 5 16 -nblock 5 2 -nreps 6 16 -nblock 6 2 \ <-- Condition V8
-nreps 7 06 -nblock 7 2 -nreps 8 06 -nblock 8 2 \ <-- Condition I8
-nreps 9 136 -nblock 9 2 <-- Null events
Transpose the columns of myStim.1D to rows.
For rows 1,3,5,7, replace "1 1" with "1 0".
For rows 2,4,6,8, replace "1 1" with "0 1".
Transpose the rows back to columns.
Use a spreadsheet program to add columns 1 and 2, add columns 3 and 4,
add columns 5 and 6, and add columns 7 and 8. These 4 column sums will now
represent the 4 input stimulus functions.
As mentioned above, this level of randomization is probably unnecessary.
On the other hand, I think that paranoia is a useful psychological trait
for statisticians (and scientists). As I always say: If you think Nature has
stacked the deck, you'd better shuffle the cards.
Doug Ward
BRI/MCW/WI/USA/SOL3/MWG/LGG/VSC/MOG
P.S. I guess you didn't recognize me when I paid a visit over the weekend.
I was wearing a black suit and dark glasses.