[tobyhill]

It's a geographically, racially, gender, etc, weighted poll. Party affiliation has rarely varied by more than a couple points.
If you know that the 2010 elections throughout the country showed an equal amount of affiliations yet less than two years later the poll shows a wide variable then more than likely the weighted sample is wrong.
It is a joke poll.
A true poll will ask 1 question with 1000 people in each state through 50 states (no questions on party, race, gender, etc), “who will you vote for”? No need to weigh the poll because with 50,000 there will be a very good base sample.

[muawiyah]

Sorry, there's no weighting in these polls ~ they do a nationwide random sample ~ call those people ~ then get their numbers.

What you are describing is a stratified poll with pre-selected individuals of given characteristics and persuasions. Such polls can be done AT IMMENSE COST!