Several polls of likely voters released this week show President Barack Obama with a lead over former Massachusetts Gov. Mitt Romney in the race for the White House. But both campaigns should keep something in mind: In recent years, "likely voter" polls have been consistently wrong in predicting victory margins.
Current likely voter methodology -- which asks about past voting as well as intentions for the upcoming election -- tends to undercount minorities, single women and younger voters. Polling all registered voters, as opposed to just those deemed likely to vote, makes far fewer mistakes.
This year, when the electorate is starkly divided on race and age and there's a growing gender gap, those mistakes could be significant. The issue isn't partisan advantage, but the accuracy of the information presented to voters.
In 2008, Gallup realized its likely voter methodology wasn't tracking the increased share of the vote that minority and younger voters would cast. Its likely voter data consistently showed Obama with only a 2 percent to 3 percent lead over Arizona Sen. John McCain. So Gallup's pollsters created what they called a "likely voter for 2008" sample, which showed Obama had closer to a 5 percent to 8 percent lead. Interestingly, this new criteria tracked Gallup's polling data for all registered voters -- and Obama ultimately beat McCain by 7 percent.
In 2010, likely voter polls underestimated the minority vote, especially Hispanics, in key congressional races. In Nevada and Colorado, for instance, incumbent Democratic U.S. Sens. Harry Reid and Michael Bennet won despite end-of-campaign polls of likely voters showing them losing. Hispanic voters turned out in greater numbers -- and were more supportive of Reid and Bennet -- than those polls projected.
In New York, we're used to this. In the 2002 state comptroller's race and the 2005 and 2009 New York City mayoral elections, the winners' ultimate margins werefar closer than the results of Quinnipiac University and Marist College likely voter polls before Election Day. At the close of the 2010 campaign, Quinnipiac and Marist's likely voter polls showed Cuomo heading toward 55 percent support, while Siena's likely voter poll called the attorney general and comptroller races dead heats. Cuomo won with 61 percent, Eric Schneiderman had a 12-point win for attorney general, and Thomas DiNapoli remained comptroller with a 4-point margin.
Marist projected the suburbs casting more votes than New York City, while the Siena College poll projected men outvoting women, and minorities casting a mere 15 percent of the total vote. When the returns came in, the city's share of the total vote beat the suburbs by 6 percent; women cast 53 percent of the vote; and the aggregate minority share was 29 percent.
The increasing number of cellphone-only households and the fact that cellphone customers can keep their phone numbers after they move to a different district or state are posing challenges to polling accuracy. And these challenges are heightened in likely voter samples: Younger, single and minority voters are simply more mobile and less likely to have landline phones.
Polling data have become the main grist for the 24/7 media mill, from blogs to print to broadcast. So it's a problem if these data saturate the public with needlessly inaccurate information. The false perception of a Michael Bloomberg landslide in 2009, for example, thwarted challenger Bill Thompson's ability to generate meaningful coverage and funding in that New York City mayoral race.
It is long past time for the collective amnesia in the media regarding likely voter polling to be replaced with standards for projecting more accurate data to the public. The underlying polling data is sound, but the way it is projected is not.
Since polls of registered voters have been more accurate than likely voter samples, news media outlets should demand that both sets of data be provided (as Pew Research, Marist and the Siena polls, to their credit, do). And the media should give greater reporting weight to the more accurate polls of registered voters -- unless and until likely voter samples reprove their empirical validity.
Pollsters pretending that they alone have a window on who is likely to vote is a fiction that should no longer be presented as a fact.
Bruce N. Gyory is a consultant with Corning Place Communications and an adjunct professor of political science at the University at Albany.