AP Poll Provides Good Lesson in Polling

The AP poll released yesterday showed the race between President Obama and Mitt Romney essentially tied among likely voters. Obama is leading by one point, 47-46, well within the margin of error. I think the partisan split for likely voters is maybe a tad too generous to the Democrats, but its reasonable. The D+4 in their likely voter sample is almost half the advantage Democrats had in 2008. More interesting, though, is that the poll provides results for the three different polling universes; adults, registered voters (rv) and likely voters (lv). Its a very instructional lesson in polling.

I’ve stressed repeatedly that the voter screen and partisan sample have a profound effect on the results in any poll. Polling adults will generally give you results more favorable to Democrats, since many of those who aren’t registered tend to identify with that party. Polls of registered voters cut down on this bias, but because GOP voters are the most likely to vote, there is historically a 2-3 point bias towards Democrats. Likely voter polls remove the inherent bias and are also the most predictive of what will happen on election day.

The other major variable is the partisan split in the poll. The best poll will be the one that most closely matches the electorate on election day. If 32% of voters who show up are Democrats and 30% of those who show are Republicans, we’ll say the election was D+2. In 2008, the electorate was D+7, meaning Democrats had a seven point advantage on election day. In 2004, the electorate was D+0, meaning equal number of Republicans and Democrats voted.

The main problem with most media polls is they reflect the results IF the electorate in November is as or more Democrat as 2008. For a long list of reasons, not the least of which is that GOP voters this year are far more enthusiastic about voting this year, this is a ridiculous assumption. Any poll based on a higher Democrat turnout this year is almost meaningless.

The AP poll shows the strong correlation between the partisan split and the head-to-head numbers.

The Adult sample is D+12. Obama leads this group by 15 points. RV sample is D+11 (way too high) and Obama leads by 10 points. The LV sample is D+4 and Obama leads by just 1.

What will the electorate look like in November? Obviously, with 7 weeks to go there are a lot of factors still unknown to effect turnout. It won’t look anything like 2008 and D+7. My hunch is it will look a lot more like 2004 when it was D/R +0. The closer it is to that number, the more likely it is that Romney wins.