Hi Jim,
Two downsides leap to mind when applying a very large polort.
One is the statistical power in your results. Since each run is
modelled separately, a large polort will chew up degrees of freedom
quickly. For 10 runs with a polort of 3, that uses 40 regressors
and degrees of freedom. If the polort were changed to 15, it would
use 160 degrees of freedom. So your statistics would probably be
deflated.
A potentially bigger problem is if you have a block design, or even
a fast event design that can have some events pile up together. A
high-degree polynomial can match many fluctuations in a signal. So
one might stand a resonable chance of having parts of regressors of
interest being modelled by the polynomial "baseline".
Similarly, if there are just a few events of some class in some run,
a high degree baseline has a chance of modelling such a response.
- rick