The Art of Politics

Massachussetts Liberal

Polling, party identification, turnout models, and getting it right...

Much has been made of polling lately, with the trailing candidate often questioning methodology, and going so far in fact that the reactionaries at MoveOn.org have assailed the Gallup organization and their methodology.

The biggest reason polling numbers, turnout models, and party identification are so contentious is that the sample sizes in most polls are far too small, and far too scattered, to get an accurate poll that reflects regional differences with national assumptions.

For instance, take a 1000 voter sample size, with an assumption of say 33% Democrat turnout, and 32% Republican. Now, that's only 330 Democrats. Now, how are you going to disperse that across the nation? That's an average of only 6 Democrats per state. And while surely you will get far more Democrats asked in New York than in Iowa - is it not obvious to everyone that Iowa Democrats are a very different breed from New York Democrats? The larger sample in the larger state will flavor the poll toward the larger state's political predilections - simply stated, you'd get far different numbers polling 330 New York Democrats than you would 330 Texas Democrats, and that shows up in the poll results.

Does anyone think a 18 voter cull in Wisconsin gives an accurate representation of that state's political breakdown?

Add in the uncertainty of turnout models, which are very much in question considering the high interest level in this election, the aftermath of the 2000 election, and new voter registration drives, and you get varying expectations that may turn out to be very wrong on election day.

Posted by MEC2 at September 29, 2004 01:18 PM