Gallup: To Illustrate Our Point

Our past two posts dealt with potential errors and biases in various election polls. Just to illustrate how dicey it is to predict what's going to happen this year, let's look at Gallup's national tracking poll.

Gallup is interviewing at least 1000 respondents across the country every day. It then reports, on a daily "tracking" basis, the results of that poll over the last three days. The Gallup poll, unlike some others, does include cell phones, so we take out the "landline effect" we discussed in our last post.

But Gallup can't figure out who's actually going to vote this year. So they're reporting three numbers: the margin among registered voters; the margin among "likely voters" based on Gallup's traditional "likely voter" model; and the margin among "likely voters" based on a modified model.

The difference between the two "likely voter" models is that in the traditional model, Gallup relies both on the respondent's professed intention to vote and on the respondent's past voting history (i.e., did they vote in the last one or two presidential elections), whereas in the modified model, Gallup simply accepts the respondent's professed intention to vote.

So how much difference is there in these numbers? Quite a bit. As of today, Gallup reports the following:

Registered Voters: Obama 52% McCain 42%
Modified Likely Voters: Obama 51% McCain 44%
Traditional Likely Voters: Obama 49% McCain 46%

Given how close the last two elections have been, that's quite a contrast!