Yes, either could happen if the polort is too small.
It is easiest to imagine with constant and linear, but
let's go up just one degree. Suppose you have assumed
a linear baseline, but there is a strong U-shape to it
(quadratic, say).
The model fit will probably go right through the middle
of the data. The fit parameters could be garbage, and
the error could be huge (meaning the stats will be ~0).
Ignoring the really bad stats, the beta weights could
be tiny or huge. Maybe betas of 20% signal change will
happen to fit the U shape better. Who knows?
---
Fortunately the baseline does not tend to fluctuate so
wildly, but it helps to make the point. Modelling the
baseline accurately is important.
Of course, with too many parameters one can over-model
the data, also leading to garbage... :)
- rick