Monday, September 15, 2008

Exit Polls and Party ID

Frequently when we put out a poll I get e-mails criticizing our party identification breakdown because the exit polls in 2004 said 'x.'

I don't put much stock in the relevance of party id breakdowns in exit polls to what we're finding in our current polling for two reasons, one of which is a fault of the exit polls and one of which is just a function of human nature:

1) Many of the 2004 exit polls were badly off on their calculations of fixed demographics, such as gender, race, and age. If they couldn't get those things, which you can actually go back and look up, right why should they be trusted to have gotten the correct breakdown of non-fixed demographics like party?

2) I believe that how an exit poll respondent just voted for President has a much greater impact on how they fill out their survey than it would in a pre-election survey. For instance, the exit polls almost always find a much higher percentage of independent voters than we do in our polls. I think it is entirely plausible that if you generally consider yourself a Democrat but just voted George W. Bush for President, you might check the independent box instead of the Democratic box. But if I call you in September and you plan to vote for John McCain but to vote Democratic for the state offices that you don't get asked about on an exit poll, you might still tell me you're a Democrat.

There are many more folks who otherwise consider themselves Democrats who vote Republican for President than there are folks who otherwise consider themselves Republican but vote Democratic for President. I think that may lead to an unrealistic tilt in a Republican direction on how party id is calculated for exit polls, and it's why I'm not going to worry about whether the party id ratios we find are in line with 2004 or not.

I think we're getting pretty reasonable overall results while frequently finding a party breakdown that is much more Democratic than past exit polls would indicate. I'm a lot more concerned with the the numbers for the former than the numbers for the latter.

No comments:

Web Statistics