Election Polls

It seems worth noting what most of the media seems determined to ignore: most of the election polling going on right now is close to meaningless, at least insofar as pundits are trying to use the data. What the polls seem to show is that the electorate is fairly evenly divided. What pundits try to show (depending who you read) is that Bush/Kerry is doing better/worse in a significant way. Most of this is bogus, either because the difference is within the poll's margin of error or due to the simple fact that the election is five months off, and plenty can happen to change voters' perceptions between now and then (including but not limited to greater voter exposure to the Boston Fog Machine, developments in Iraq, other developments in international affairs, increased awareness of the strength of the economic recovery, developments in domestic politics, etc, etc).

But even putting aside those factors, we still should take polls with a grain of salt. In recent elections, polls have had limited use in predicting the outcomes. The SD special election a few days ago ended up a lot closer than most polls were showing in the preceding days, and LA governor's race faced similar problems last year. In 2002, almost every polling organization missed the Republican surge that led to Republicans retaking the Senate after winning several close races. And, perhaps most famously, the major media organizations completely blew it in trying to call the 2000 election (oscillating from "Gore won Florida" to "too close to call" to "Bush won Florida" to "too close to call" in the space of a few hours).

One of the major problems is figuring out who's going to vote in the first place. In an election that isn't terribly close, rough estimates (involving asking pollees if they voted in the last election and the like) can work to estimate likely voters. In a close election, though, this may break down. Both parties are running ever-more-sophisticated get-out-the-vote operations, and voter registration drives are a staple of election years. In years in which divisive and/or explosive factors are in play (Iraq, culture wars of various sorts, fever-swamps Bush-hatred), predicting turnout is probably much harder than in more 'normal' years.

Another problem is reaching people in the first place. Telephone surveys rely on people being A) home, and B) willing to talk to the pollster. In addition, increasing numbers of people are abandoning land lines in favor of mobile phones, while pollsters don't call, which could impact poll results in a statistically significant way.

Finally, it's worth noting that since the Electoral College vote is the one that counts, national polls between Bush/Kerry are rather useless. Polls of swing states matter more.

So the upshot of all of this is that, in my impression, any poll before the election that shows less than a 10-point gap between candidates is probably almost useless for purposes of prognostication. (And at that, some polls that purport to show huge differences may still be unreliable).

[Further note: I'm too lazy to provide links to most of this stuff, though perhaps if someone begged nicely in the comments I could be persuaded to change my mind].