It turns out that the most important factor in polling is the proper mix of republicans, democrats, and independents.
In the 2008 election:
It is not a coincidence that Rasmussenreports reported before the election that his party identification weightings for every poll were:
Instead of using the results from the prior election, leaving the survey unweighted for party id, or just guessing, Rasmussen did separate polling to determine the accurate number. It could be that this methodology will make polling much more of an accurate science rather than an art, or it could be just luck.