Well, I still think that the 200 & 2004 Republican tickets ought to have been Cheney/GWB.
Despite new innovations, polling is less & less accurate, despite increasing the sample size.
For instance (by no means exhaustive):
1. Bradley/Royko effect--the tendency for people polled to just plain lie (Royko) to pollsters, especially when a black (Bradley) candidate is running in a race where the constituency is largely non-black. Usually about 5% of those polled will say they are voting for / voted for the black guy when they do otherwise.
2. Cell phone effect--quick-reaction polls are limited to land-lines. With a great proportion of the population going cell-only, the sample population is skewed. Generally, the younger the person, the more likely they are cell-only. But, the younger they are, the less likely they are to actually vote.
There are other challenges, but those two seem most pertinent at the moment.
Rassmussen decreases the margin of error by using three night's worth of polling, but I do not see if they filter by registered or likely voters (likely being a better measure).
