Heisenberg's Polling Principle

The problem with polls is...well they're polls. They are attempts to measure an individual's likely voting behavior and in so doing they offer a prediction as to the likely outcome of an election were it to have been run at that moment. The problem is, the very act of trying to measure the likelihood an individual will vote, and who they will vote for, can alter the outcome. Sound familiar? Well its close to the problem physicists face when trying to measure the position and velocity of a particle. Heisenberg described this in his Uncertainty Principle:

Light can be considered as being made up of packets of energy called photons. To measure the position and velocity of any particle, you would first shine a light on it, then detect the reflection. On a macroscopic scale, the effect of photons on an object is insignificant. Unfortunately, on subatomic scales, the photons that hit the subatomic particle will cause it to move significantly, so although the position has been measured accurately, the velocity of the particle will have been altered. By learning the position, you have rendered any information you previously had on the velocity useless. In other words, the observer affects the observed.

The problem faced by pollsters is when they measure and report the intentions of likely voters, they alter the population of likely voters, and sometimes they even alter the votes of the very people they polled. For example, if the election looks to be decisive for one candidate or the other, I'll vote my true leanings (libertarian), but if the election looks close, I may vote strategically (the lesser of two evils). Still other people who are not likely to vote, will become much more likely when the polls report that the vote looks to be close, since they will feel they can influence the outcome. Sometimes this happens in predictable ways, but as Ray Fair notes in response to a reporter suggesting his prediction of Bush winning would reduce participation by Kerry supporters:

It could work the other way. If Kerry supporters see that I have made this big prediction for Bush, more of them could turn out just to prove an economist wrong.

So you can think of polls as attempts to measure two things, one would be your probability of voting, and the other would be your preference in a candidate. It seems unlike the physicists we can't even measure one of those precisely. Unlike Ray Fair some social scientists aren't willing to let the polling chips fall where they may, which is why Krugman recently penned the following:

By the way, why does the Gallup poll, which is influential because of its illustrious history, report a large Bush lead when many other polls show a dead heat? It's mostly because of how Gallup determines "likely voters": the poll shows only a three-point Bush lead among registered voters. And as the Democratic poll expert Ruy Teixeira points out (using data obtained by Steve Soto, a liberal blogger), Gallup's sample of supposedly likely voters contains a much smaller proportion of both minority and young voters than the actual proportions of these voters in the 2000 election.

But the question of who is right is an empirical one. If Krugman were honest, instead of the political hack he has become, he would point out that maybe the proportions of minority and young voters in 2000 where not unusually high, but rather part of a rising trend of increased participation for those groups? Even so, one data point does not a trend make, so the fact that Gallup is using a longer period average to weight participation does not make a conspiracy. Although you wouldn't know it from this full page ad over at Moveon.org. I don't disagree with the assessment that the Gallup polls are over stating Bush's lead, because I do believe we will see a much larger turnout among the youth and minorities, not because I believe it is a right wing conspiracy. The fact is turn out among youth and minorities is very volatile, so only time will tell who is right.

Keywords: Polls, Surveys, BUS230