Wednesday, July 18, 2012

Registered and Likely: A Tale of Two Voters

A number of polls have been released in the past few days that continue to show the presidential race to be very close. What's worth discussing at this point is the difference between polls that measure registered voters (RV) and those that have switched over to likely voter (LV) models. There's a terrific article by Mark Blumenthal of HuffPost/Pollster that discusses this issue and he comes to some interesting, and ultimately unsatisfying conclusions. From the article:

The consistent difference between the registered and likely voter samples raises the question: If likely voter screens applied at the end of the campaign nudged the horse race numbers in a more accurate (and more Republican) direction, why not apply such screens now?

The answer, in part

These "measures of engagement and intention to vote are useful as indicators of likely turnout on the aggregate level" at this stage in the campaign," Pew Research associate research director Michael Dimock explained to The Huffington Post in June, "but they are only loosely predictors of whether an individual is or is not likely to vote this November."

"As we get closer to Election Day," he added, "particularly after the conventions and into the debate season -- these indicators become stronger."

With this in mind, let's take a look at some recent polls that use LV models. Purple Strategies released a number of swing state polls showing Obama leading in Ohio (+3), Virginia (+2) and Colorado (+1) and Romney leading in Florida (+3). All of these results are within the margin of error and seem to suggest that Obama does well in states where he polls well with women and lags where men and independents do not support him. The Florida results are also in line with a Mason-Dixon poll that also uses an LV model and shows Obama with a 1 point advantage. Likewise, a new Rasmussen poll of Virginia also shows Obama with a small lead of 47-46%.

The Rasmussen daily tracking poll uses an LV model and shows Romney with a 1 point lead today, within the margin of error.

If it's true that Republicans tend to do better with LV models, then we can assume that Obama is truly ahead in the states where he leads and that Romney will need to move the dials a bit to overtake him.
On the RV side, a new Quinnipiac poll of New Jersey gives Obama a 49%-38% lead while slightly oversampling Republicans. New Jersey is D 33/R 20/I 47 while the poll was D 34/R 24/ I 37.

Presumably, even if Quinnipiac went to an LV screen, Romney would not be within threatening distance of the president. My view is that New Jersey will remain blue this November.

A PPP Poll of Iowa has Obama leading by 48-43% with a D 35/ R 34/ I 30 split, when in reality, Iowa breaks D 31/ R 32/ I 37. PPP, which has a Democratic leaning house effect, has also slightly underpolled Republicans and overpolled Democrats. This leads me to believe that Iowa will probably lean Romney once polls move to an LV model.

At this point in the campaign, using likely voter models do not seem to be diverging a great deal from polls that use registered voters. As we get closer to the campaign's major events, that will change. Romney has yet to choose a running mate, and that will help his numbers. The conventions will also provide major bumps for both candidates, as will the results of the first televised debate. If voters are settling in on one candidate, that will become evident by the middle of September, and as long as respondents tell the truth about their voting habits (what is the percentage of people who lie about such things?) we'll get a fairly accurate view of any election trends.

Right now? Enjoy the summer, baseball and the Olympics.

Register your likely comments at and on Twitter @rigrundfest

No comments:

Post a Comment