Politics: Poll Rant (Part 2)

Sorry for the long delay between posts, but I've been pretty busy with after school stuff. In any case, I thought it was high time that I continue on with my rant about polls by expounding on what seems to be the latest hot topic in the blogosphere: Likely voters.

Usually when polls are reported to the media, they have two sets of numbers: Registered voters, which include anyone in the sample who is registered to vote, and likely voters. Some of sloppier news organizations will only report likely voters, so you have to be careful when you look at stories about polls. To give you an idea about how off the likely voter model can be, take a look at the Gallup poll from October 24-26, 2000. It shows a 13 point Bush lead on the popular vote. You'll hopefully recall that Gore actually beat Bush on total number of votes merely two weeks later.

So who are the likely voters? Depends on who you ask. Unfortunately, I have yet to find a single good model that makes sense for this election. The simplest model asks the question, "Are you going to vote?" Unfortunately, many people, especially undecideds respond with "Don't know." Many events between now and the election can change their mind. The worst part is that this is the best model.

The next model commonly used is: "How excited are you to vote?" If the person responds "Very excited," or possibly "Somewhat excited" then they are considered a likely voter. Unfortunately, many people who aren't excited to vote will do so anyway. This question has been shown to have almost no correlation between the answer and the likelihood of voting. The results get even worse when the question becomes "How excited are you to vote for your candidate?" To be quite honest, I have very little excitement for my candidate. On a scale of 1 to 10, where 1 is "I might give him a polite nod if I happened to pass him on the street" and 10 is "I'm changing my name to John F. Kerry and getting a chin and forehead implant," I'm probably about a 3. But I'm an 11 on "I think Bush is harmful enough that he needs to be voted against." So I'm not a likely voter, but I've already sent in my absentee ballot for Kerry.

At the bottom of the barrel are the questions that ask about previous voting behavior. Questions like "Did you vote in the last election?" and "Who did you last cast a ballot for?" fail to recognize new voters and occasional voters. With voter registrations surging, new voters are going to play a huge role. Occasional voters, those who only vote in elections they feel are "very important". are also likely to participate heavily this year.

The moral of this story: If you really want to see how a candidate is doing in a certain state, look at the trend in registered voters. It won't give you the exact numbers that a candidate is at, but it will give you a good picture of the movement towards or away from a given candidate. Likely voter models, however, should be knocked from their perch of statistical science and put into the junkheap of irrelevance.