What I'm hearing is that for the voters in the middle who decide elections, party affiliation is highly fluid. If they think they're gonna vote for Romney, they'll say they're Republicans. On a different day, when Obama's looking good, they'll say they're Democrats. That's why we get these big swings - not because of sampling error, but because not everyone's a die-hard party member.
But if party affiliation is as loosely defined as you've described (and it may very well be), then nobody doing a poll has any business even making adjustments for it. You can't go out and do a poll with all of these party-affiliation adjustments in it, project Candidate X as the winner, and then come back after Candidate X loses and say you were wrong because party affiliation is "highly fluid."
Something else to consider here is that most people are completely overlooking the whole concept of a "margin of error" in a poll. That is an absolutely meaningful number, and what it indicates is that any polling margin within that margin of error is basically meaningless. If a candidate is up by 2% in a poll with a 2.5% margin of error, then you can throw the poll out the window because it's not telling you anything. It's not even telling you that the candidate has a slight advantage.
A better indication -- and one that everyone here on FreeRepublic either ignored or tried to explain away -- would be something like an Intrade projection. This is basically an online auction/bidding site that allows people to place bets on all different kinds of things. It's probably a pretty good indicator of how things like an election will go because people are putting their own money on the line and they are placing bets regardless of party affiliation or even preference for a candidate. Nobody has a vested interest in placing a bet on a candidate they expect to lose, and the odds are adjusted accordingly as bets are placed.
In the final months of the election season Intrade was consistently giving a pretty strong indication (65% at first, then growing to 80%+) that Obama was going to win, which meant that behind all the nonsense was an underlying sentiment that polls may not necessarily capture well within their margins of error.