There's much to chew on in the firestorm over a recent series of House district surveys sponsored by the liberal Web site Firedoglake and conducted by the firm SurveyUSA, but one question has particular importance this election cycle: Where are all the young voters?
Polls sponsored by candidates, parties or interest groups are not new, and neither are attacks on their accuracy. So the criticism of these district surveys wasn't exactly shocking. Firedoglake has been an outspoken opponent, from the left, of the health care reform legislation pushed by President Obama and the Democrats; its surveys showed Democratic incumbents facing surprisingly strong challenges from their likely Republican opponents and (surprise, surprise) support for Firedoglake's framing of health care reform.
As reported by Politico's Ben Smith, the surveys drew fire from several blogs; an attack memo by the Democratic polling firm Global Strategy Group (which polls for Rep. Tim Bishop, D-N.Y., the subject of one of the surveys); a response by SurveyUSA CEO Jay Leve; a subsequent critical memo authored by Emory political scientist Alan Abramowitz and distributed by a "Democratic source"; and a second vehement response from Leve.
There's plenty to say about all of this, but the small sampling of likely voters ages 18 to 34 deserves special attention.
Before delving into that issue, let's remember that we are able to pick at the demographic composition of SurveyUSA's polls because the pollster chooses to disclose it. Of the various accusations hurled at the firm, the one that rings most hollow is that its surveys featured "incomplete disclosure of demographics." Yes, they could disclose more, but in my experience, SurveyUSA has been among the most transparent and responsive of pollsters.
Back to the numbers. Most political insiders know that younger voters typically turn out in far greater numbers for presidential elections than for midterm contests. But the proportion of 18-to-34-year-olds among the likely voters identified by SurveyUSA was still surprisingly low: just 5 percent in Indiana's 9th District, 3 percent in Ohio's 1st and 1 percent in New York's 1st.
Wondering how those numbers compared to the age composition in a previous off-year election, I turned to a resource rarely tapped by journalists: the databases of registered voters culled from official records and maintained by vendors that work with political campaigns. I contacted Catalist, the company headed by Democratic operative Harold Ickes reputed to have very complete records of "vote history" -- which voters actually cast ballots in recent elections.
At my request, Catalist tabulated the age composition of currently active voters that cast ballots in the 2006 general election, using voter ages as of November 2006. They show 18-to-34-year-olds making up a bigger proportion of the electorate than SurveyUSA's likely voters for 2010: 16 percent in Indiana's 9th District, 17 percent in Ohio's 1st and 10 percent in New York's 1st .
So why such a big age difference?
• Voter registration records are imperfect. Officials regularly purge voters who have moved (and tend to be younger) or who have died (and tend to be older). The potential impact of purging is relatively small, as the Catalist count of voters slightly exceeds the number that cast a ballot for Congress in two districts (Indiana-09 and Ohio-01) and falls slightly short in the other (N.Y.-01). The small differences (roughly 2 percent of the total in each district) are explained mostly by voters who showed up to vote but failed to record a valid choice for Congress in 2006.
• The age composition of this year's midterms may be different. While a helpful guide, the 2006 age statistics may not predict the pattern in 2010. The 2006 elections were a high-water mark for Democrats in terms of their share of the national House vote in recent midterm elections. While the Democratic enthusiasm gap was probably greater in 2008 than 2006, the tide has certainly turned to the Republicans in recent months. If that pattern holds through November, the 2010 electorate will likely be older than in 2006.
• The screen might have been too tight. For these polls, SurveyUSA also relied on samples drawn from lists of registered voters, and its sample included only those households with voters who cast ballots in 2008 and either 2006 or 2002. As the Global Strategy memo points out, that is quite restrictive. It would exclude any new registrants from 2008 (who were disproportionately young). While no two pollsters define "likely voters" in exactly the same way, it is not surprising that this particular definition results in a very old sample.
• Cell-phone-only households were not included. According to the latest federal estimates, nearly two-thirds of adults between 25 and 29 live in households where there is no landline phone (45.8 percent) or cell phones are used for "all or almost all calls" (17.7 percent). So pollsters are having a much harder time reaching younger voters.
In this case, the problem of missing younger voters cannot be solved by simple weighting of the data, mostly because there are too few 18-34-year-olds to "weight up."
Leve has spoken candidly about the cell-phone-only problem recently and shared a similar sentiment with me via e-mail last week. "Lost in the [recent] hurly burly is an opportunity for real reflection," he wrote. Neither of us has seen a publicly released telephone poll for a congressional district that included cell-phone-only households. "Whether one anticipates that in 2010 young voters will turn out in record numbers," he added, "or stay home in record numbers, the problem of how to count those voters is real, and right before us."
It certainly is.