If you picked up a copy of the Washington Post last week, you might have read a story on the debate between the Democratic candidates for governor in Virginia. This article, which focused mostly on differences of style and tactics among Creigh Deeds, Brian Moran and Terry McAuliffe, included a single reference to polling. The candidates, it said, "sought to stamp a final impression in a race where polls show the majority of voters remain undecided."
Which polls? That, the Post won't tell you. "None of the recent polls in the Virginia governor's race," polling director Jon Cohen explained in a column the previous Sunday, "meet our current criteria for reporting polls."
Polling in primary elections may be perilous, but it doesn't have to be impossible.
Cohen offered two primary reasons for shunning more specific coverage. First, the polls were of questionable quality -- two were conducted using an automated methodology, and one was an internal campaign poll. Second, polling in low-turnout primary contests is not for the faint of heart.
Cohen cited the undeniably "monumental task" of trying to identify likely voters. Virginia has held primary elections to select Democratic nominees for less than 10 years, and the turnout has varied widely. Cohen framed the problem exactly right:
Will Democratic turnout be in line with the approximately 115,000 who voted in the party's lieutenant governor primary four years ago? The 150,000 or so voters who cast ballots in the 2006 Democratic primary for U.S. Senate? Or the nearly 1 million voting in last year's battle between Barack Obama and Hillary Rodham Clinton?
Guessing the level of turnout is just the beginning. Even with a swami-like ability to predict turnout, there are still two huge hurdles. First, a turnout comparable to the 2006 primary (150,000 voters) amounts to less than 3 percent of Virginia's eligible adults. Imagine the screening that would be necessary to sample that population.
Second, the profile of voters in those low-low-turnout elections is not static. In 2005, according to one pollster I talked to, the portions of Northern Virginia in the D.C. media market accounted for 34 percent of the Democratic primary voters. A year later, the D.C. market made up 46 percent of the Democratic primary electorate. So predicting the profile of likely voters is as iffy as guessing their number.
As such, Cohen and the Post are right to downplay the specifics of horse race results in the handful of available public polls.
But does it make sense to ignore them altogether? Can it even be done? Tom Jensen, communications director for Public Policy Polling, a North Carolina firm that has conducted and released surveys in Virginia using an automated method, pointed out that being ignored by the Post does nothing to dent the "three or four days worth of explosion in traffic" the company receives on its Web site "every time we've put out a Virginia primary poll." In the Internet era, media outfits like the Post and Hotline no longer control the flow of information flowing to journalists, insiders and political junkies.
Polling in primary elections may be perilous, but it doesn't have to be impossible. The campaign pollsters who conduct these surveys and live to tell about it constantly remind their clients of their limitations. "In primary elections," Republican pollster Bill McInturff wrote in an e-mail, "I tell my clients (repeatedly) how difficult it is to know if you have captured all the moving pieces that can impact the result." He puts greatest emphasis on watching trends in surveys that have been conducted "in a rigorous and consistent way.... The most important thing to look at is whether the lines are going up (good) or down (bad)."
In that spirit, let's take a quick tour through the three public polls released in Virginia last week (all fielded before the Post's endorsement of Deeds last Friday).
The polls conducted by SurveyUSA and Public Policy Polling use an automated methodology in which respondents answer by pressing keys on their touch-tone phones. The poll conducted by Research 2000 and sponsored by DailyKos uses live interviewers.
SurveyUSA and Research 2000 both start with samples of randomly generated telephone numbers designed to cover every working landline telephone line in Virginia. They use screen questions to identify those who say they vote regularly and intend to vote in the June 9 primary.
Survey USA interviewed 2,000 adults to identify 502 likely voters, so its sample represents roughly 25 percent of Virginia adults -- a much bigger slice than is expected to participate on June 9. Research 2000 does not provide sufficient information with its public release to estimate the proportion of adults defined as likely voters.
Public Policy Polling uses voter registration lists to select a sample of households that include registered voters who actually cast ballots in Democratic primaries since 2005. This choice involves a trade-off. The firm probably samples a narrower slice of the potential electorate, but it misses the roughly 30 percent of Virginia voters whose telephone numbers cannot be identified in telephone directories. Also, given the limitations of its automated methodology, it did not attempt to match respondents to the names on the list with actual voting history (so while these pollsters too ask questions that screen for "likely voters," they may end up interviewing Virginians who have no primary voting history but happen to live with voters who do).
The findings? All three show McAuliffe nominally ahead -- his share of the vote varies from 29 percent to 37 percent -- but far from dominant. Two of the three, SurveyUSA and Public Policy Polling, show Deeds in second place (with 26 percent and 20 percent, respectively) and gaining over the last month. The Research 2000 poll shows Moran running ahead of Deeds (22 percent to 13 percent), with Deeds slightly lower than in early April.
All three surveys show considerable evidence of indecision among the voters interviewed. Nearly a third of the voters in the Research 2000 (29 percent) and PPP surveys (31 percent) are completely undecided. SurveyUSA, which typically pushes respondents a bit harder for a choice, shows only 14 percent undecided but reports more than half the likely voters (57 percent) saying that their minds "could still change." A similar follow-up on the PPP survey shows 42 percent say they could still change their minds.
Which Virginia poll is the most reliable? Are any providing an accurate snapshot of the true likely electorate? The truth is, no one knows for sure. No poll can possibly predict the likely electorate with any confidence, and the results we have indicate considerable indecision among those willing to describe themselves as likely voters.
And we still have two weeks remaining, which is an eternity in a race with so many uncertain voters.