Dear Andy,
can you provide some more information on your design? It sounds a little cryptic ;-) If you think of different trials, each of them associated with a particular task or stimulus configuration, e.g. A and B, and if you're interested in the impact of the preceding task/trial, you could go with four separete regressors
Ar(epeat) = A preceded by A
As(witch) = A preceded by B
Br(epeat) = B preceded by B
Bs(witch) = B preceded by A
For those trials in which there is no preceding trial (e.g. at the beginning of the experiment/block, after longer breaks, null events, errors) you might have to introduce additional regressors like A0 and B0.
Thus, a task/stimulus configuration like A B A A B A B B would result in A0, Bs, As, Ar, Bs, As, Bs, Br. For second-level analysis you would e.g. set up a within-subject 2x2 ANOVA, one factor dealing with task/stimulus (A, B) and the other with trial history (switch, repeat).
Depending on the experiment and expectations (e.g. the more repetitions the less activation) you might also work with just two standard regressors for A and B and a parametric modulator to deal with trial history.
The same should be applicable on experiments with more conditions (visual stimulus preceded by accoustic stimulus = Va, visual stimulus preceded by tactile stimulus = Vt, visual stimulus preceded by visual stimulus = Vv, ...). You just have to make sure you still have enough trials per condition.
Hope this helps,
Helmut
|