Tuesday, November 04, 2008

Why exit polls should not be trusted

When you've waited years for an election and sufferred through 8 years of the worst President in modern US History (not an easy thing to accomplish when Reagan was also President in the same era), your natural reaction can be to fall victim to confirmation bias: believing something because it confirms your view of the world, giving little or no credence to the quality of the data involved. No matter what the exit polls are telling you today, DO NOT DRINK THAT KOOL AID!

Remember 2004? Exit polls across the country suggested that John Kerry was not only winning, but that he was leading by a sizeable margin. Kerry was going to win! The only problem was that he didn't win. And we got 4 more years of out of control spending, conservative supreme court justices, and awful foreign policies.

The idea behind a political poll is that you ask a small group of people for their opinion and hope that your small group is an adequate representation of the larger group of people that's actually voting. Most poll accuracy problems beyond strange methodological practices (like polling equal numbers of Democrats and Republicans when there aren't equal numbers in the population, for example) are the direct result of bad sampling practices. Exit polls have all of the same problems as regular polls with some truly damning additional limitations.

Exit polls have a selection problem, which is that only people who are willing to share their decision with pollsters get counted. If one candidate has a younger, more enthusiastic base than another candidate, the polls might show that candidate with a huge advantage on exit polls that doesn't reflect what was actually going on. Samples are often relatively small based on the fact that these are done locally at the polls, which means the likelihood of getting a sample that doesn't match the larger electorate is greatly increased. Finally, the popularity of early voting and voting by mail means that a large percentage of the electorate won't be available for exit polling. This year, the early voters tended to be much more Democratic, which means that samples taken at the polls this year could underestimate their numbers substantially.

At the end of the day, we'll know the results of the election. I'll post some exit poll data here as it comes up, only so that we can compare those data with the actual counts when they come in. DO NOT TRUST EXIT POLLS TO TELL YOU WHO IS GOING TO WIN. Go vote and then check back here later.

1 comment:

Lord Bling said...

And let's not forget that in 2000, Gore was up almost seven points in the polls as late as the day before the election. Granted, he won the popular vote, but not by seven points.

That '+/- 3 points' at the bottom of every poll should be widened to '+/- 10 points'. Of course, that would render the polls obsolete ... which wouldn't be all that bad...