Dear Rebecca,
> I am helping someone analyse a set of experiments where the design is
> always a simple ABABAB and there should be 12 scans per block.
> Sometimes A is the active task and B the reference, sometimes vice
> versa. My plan was to specify just the active condition in each
> experiement (whether A or B) and model that. However we have a bit of a
> problem in that, due to a cock-up over dummy scans, 2 scans from the
> first A block are consistently missing and we therefore have unequal
> epoch lengths (A is 10 12 12 rather than 12 12 12). As I understand it,
> the SPM99 conceptualisation of epochs and how to model them depend on
> each epoch of a condition being the same length and we therefore can't
> specify A as things stand. So I'm considering 3 options.
>
> 1. Always specify B rather than A whichever is the active task, since B
> DOES always have consistent length epochs. If B is the active task then
> 1 0 gives us what we want, if A is the active task its -1 0.
>
> 2. Introduce an extra condition C, call the design ABCBCB, have A with
> 10 second epoch, B and C with 12 and then look at the effects of
> interest with contrasts [0.33 -1 0.67] or [-0.33 1 -0.67]
>
> 3. (risking excommunication from the spm fraternity here........) make
> duplicate copies of the first 2 scans we do have in each experiment
> thereby magically creating 12 scans per block.
>
> My question: 1, 2 or 3 (or some fiendishly clever option 4 that has
> not occurred to me)?
I would go for 2 and 'recombine' them, in terms of inference, with the
appropriate contrast. The disadvantage of 1 is that you would not be
able to model adaptation or 'within-epch' changes during the activation
conditions (e.g. mean and exponential decay basis set). However, if
you were simply using a box-car regressor then 1 would be fine. I will
not comment on 3! 4 would be to model the data in an event-related
fashion using trains of stimuli that last for the [variable] duration
of each activation epoch.
With very best wishes - Karl
|