Exit polling flaws:
1) cluster sampling;
2) the tendency of face-to-face interviewers to choose people like themselves to interview, instead of sampling randomly. Add to that the fact that exit pollsters are people willing and able to work at minimum wage for just a day, a cohort that tends to be young and/or poorly educated, and that fact that exit pollsters receive minimal training;
3) exit polling cannot take absentee and early voting into account;
4) post-election polling can’t correct for the natural human tendency to identify with the winner of an election immediately after that election, which is why the same exit polling that had Kerry up hugely in the early afternoon miraculously became 37-37 when post election interviews were tabulated in.
Pollsters used to think that party ID was static, that people registered with a party when they turned 18 and that was the party they identified with for life. Not true. Party ID is, in fact, a very labile concept for the majority of the electorate. Pollsters will tell you that if they poll during the Dem convention, huge numbers will say they’re Dem. Poll during the Pubbie convention a week later, and shazam, everybody’s a Pubbie. Poll right after an unparalleled worldwide financial crisis happening on the watch of an already unpopular Republican President, and suddenly everybody’s a Democrat again.
Today at a rally McCain said that he was 6 points down in the polls. Yesterday his campaign spokesman said the same thing, that their internal polling had him down 6. I’ll go out on a limb here and assume that McCain’s pollster is not in the tank for Obama and oversampling Dems. That -6 puts McCain right about where the good trackers say he is and gives me an invaluable data point from which to analyse the head-to-head polls.
Good exit pollers, in fact, are required by training to sample randomly and NOT to "choose people like themselves." Sure, it happens.
Absentee voting usually benefits Republicans, esp. in battleground sates. OH and FL are notorious for having absentees that benefit Republicans.
#4. You've gotta be kidding, right? Most people are too busy to hang out on Drudge and get freaked out over exit polls (most people don't even know what an exit poll is). They arrive early, vote, then go to work. You've gotta do better than this.
Yes, party ID is pretty static. It's only bad pollsters who see shifts during conventions. In 2000, and 2004, we knew who our Rs were and we knew exactly how many we had to get out to win. Any good pol in any major organization will tell you the same thing.