Poll watch: Making sense of conflicting polls

The last few days have brought a raft of new polls about the presidential race -- and some sharp conflicts in the data.

One widely followed poll, the Rasmussen survey, had Republican Mitt Romney 1 point ahead on Wednesday, but by Thursday had moved to a three-point lead for President Obama, 50%-47%. Two others, the nightly Gallup tracking poll and a survey for the Associated Press, had 1-point leads for Obama on Wednesday.

But at least six other surveys this week by Fox News; NBC and the Wall St. Journal; the Washington Post; CBS, the New York Times and Quinnipiac University; Pew Research Center; and Marquette University in Wisconsin have results nationally or in key battleground states showing Obama leading by significantly larger margins.

The Pew survey had Romney trailing by 8 points nationally, 51% to 43%. The NBC/Wall St. Journal survey had the gap at 5 points, 50% to 45%. The Fox News poll had Romney behind in three swing states -- down 5 points in Florida, 7 in Ohio and 7 in Virginia. The CBS/New York Times/Quinnipiac poll had a 4-point Romney deficit in Virginia as well as a 6-point gap in Wisconsin. Colorado, in that survey, was nearly tied, with Romney down 1 point, 48% to 47%. The Marquette poll had Romney down 14 points in Wisconsin, 54% to 40%.

PHOTOS: Mitt Romney’s campaign gaffes

Those results all present a roughly similar picture of a race in which Obama has a significant lead -- something quite different from what the Gallup, AP and Rasmussen surveys suggest.

Often when polls differ, the simplest explanation is the truest: Polls are statistical samples, and all samples involve some random variation. In this case, however, that simple explanation doesn’t suffice. The differences are part of a longer-running pattern in which some polls consistently have shown better results for Obama than others.

What’s happening?

First, dismiss the conspiracy theories. Campaigns sometimes release skewed poll results for strategic reasons. But public pollsters have no incentive to do so and plenty of reason not to -- their reputation for getting the result right is their stock in trade.

Even acting in good faith, however, polling operations have different methodologies and assumptions about who will vote.

One difference involves cellphones. Polling analyst Nate Silver noted Wednesday that Obama consistently has appeared stronger in polls that call cellphones -- as well as land lines -- than in surveys that do not. About one-third of U.S. households no longer have a land line and use cellphones exclusively. Those voters tend to be younger, but even when age is taken into account, cell-only voters tend to be more Democratic.

Automated polls legally cannot call cellphones, so pollsters who rely on automated dialing, including Rasmussen, try to adjust their numbers in order to make up for what would otherwise be too few Democrats in their samples. Those adjustments don’t always work. For example, if voters who only use cellphones had suddenly become more enthusiastic about voting, an adjustment that worked five months ago would no longer yield an accurate result. There’s some evidence to suggest that may be happening now as younger voters and minorities become more engaged in the campaign.

INTERACTIVE: Battleground states map

Another issue involves whether or how much to “weight” a sample to account for partisan balance.

Pollsters always adjust their samples to make sure that the distribution of ages, genders and racial groups matches known standards, usually the Census. That way, they make sure they don’t have too many young people or too few Latinos or not enough women in a sample.

Some pollsters go a step further and make a guess about what the ratio of self-identified Democrats to Republicans should be and weight their sample accordingly. Other pollsters think that’s a bad practice. For some voters, party identification can shift from one election -- or sometimes even one month -- to another. Weighting the sample to a preconceived idea of the “right” party balance would mean missing any shift in the electorate, those pollsters say.

Rasmussen adjusts its poll for party identification and currently assumes an electorate which is slightly more Republican than Democratic -- one fairly similar to the voting pattern in the 2004 election that George W. Bush won. If the electorate this year looks like that, then Romney probably will win. But if it doesn’t, Rasmussen’s results will underestimate support for Obama.

Leaving aside the issue of partisan balance, weighting can be done in a variety of ways, and that too can create differences among polls. In the past, some analysts have said that Gallup’s nightly tracking poll includes too few minority voters because of the way Gallup weights its sample. That would lead to underestimating Obama’s vote.

On the other hand, some Republicans argue that polls done by Pew and others overestimate the number of minorities and other Democratic-leaning groups.

There’s no way to know in advance who is making the right assumptions about who will vote and how to come up with a sample that is truly representative. For now, the best solution is to take an average of well respected polls and see how that stands. Averaging tends to cancel out the various disagreements about method. Then, in just under seven weeks, we’ll find out who’s been right.

[For the Record, 7:36 a.m. PST Sept. 20: This post has been updated the include the most recent poll data from the Rasmussen.]

PHOTOS: Paul Ryan’s past

Follow Politics Now on Twitter and Facebook

david.lauter@latimes.com

Twitter: @DavidLauter