|
As election day nears most of the polling numbers being reported are of "likely voters" rather than "registered voters." For example, yesterday you probably read somewhere the Obama-McCain race was a dead heat. That type of headline resulted from an AP-Gf**k poll that showed the race at 44-43. If you dig into that poll you'll find that those are "likely" voter numbers. Among registered voters they show a more reasonable 5-point race. LV polls tend to reduce Democratic votes compared to RV polls because groups that lean D have had lower turnout rates in the past. Therefore, these LV polls have been better predictors of the final margins than the RV polls. But this may not be true this year given the historic nature of the Democratic nominee. Nate Silver has an excellent piece describing two very different types of LV estimates. Basically, the Gallup LV "traditional" method is used by some pollsters and the Gallup LV 2 "expanded" estimate is used by others. The old-school method trims about 5 points off the RV estimate while the new-school method differs by only a point (making RV a decent estimate of LV, this year). The difference is that the old method relies more upon voters' previous behavior while the new method relies more upon what voters say they will do. But we don't merely have to choose which method is better. As I said last week, all pollsters need to be asking their respondents this question: have you voted already? Nate explains the usefulness of this question: Pollsters ought to make certain that they're asking people whether they've already voted. Moreover, they ought to be putting these early voters through their likely voter models as a sanity check. That is, they should be testing to see whether a substantial number of people who have actually voted would in fact have been excluded by their likely voter screens. If the answer to this question is yes, they ought to be asking themselves whether their likely voter models have any basis in reality. This is very straightforward. Either "actual" already-been-to-the-polls voters will look like the LV screens say they should look or they will not. If they don't, the pollster needs to rethink their LV estimates. If the goal is making the most accurate representation of the state of the race, they have the data they need to adjust their models now rather than sometime after the election.
|
STATE BLOGS
