In the race to represent the 2nd Congressional District of Arkansas, the Republican, Tim Griffin, has long been the favorite. His opponent, Joyce Elliott, would be the first African American to represent the state nationally, and analysts have been predicting an uphill battle for her from the start: The district is 70 percent white and encompasses a rural swath of the southern Ozarks. The district went overwhelmingly for John McCain in 2008, despite the presence of the urban county that includes Little Rock, which is the state's largest city and one of the few places that went for Obama.
It's all but certain Elliott will lose. But it was also all but certain she was going to lose the primary in May -- a race she won with 55 percent of the vote in a June runoff. There wasn't any polling in that race, and there isn't much public polling now. The sheer number, nearly 200, of House races thought to be in play this year means the logistics of polling are more complicated than ever, and House races typically draw fewer national polls than Senate races do, anyway. The few polls conducted are problematic, because they're often snapshots involving small sample sizes. But that hasn't stopped big predictions for the House. Which begs the question: How do people know?
A finger in the wind? A crystal ball? The short answer is that analysts rely on a lot of information: internal polls, financial data, long-term trends in the district, and the degree to which the national party committees are invested in the race. But the more complicated answer is that we rely a great deal on a handful of experts, like Charlie Cook and Stuart Rothenberg, to predict outcomes for specific races and, during big election years, to predict big-picture outcomes, too. These analysts rely on better information than talking-head pundits, but, for them, forecasting elections is still as much an art as it is a science. "In some of these races, you just have to take their word for it," says Ruy Teixeira, a political scientist who writes about demographic trends and elections. "It doesn't mean it's wrong; it's just something to put a little asterisk by."
To some extent, national patterns and generic polls -- polls that don't ask voters about specific candidates in specific races but ask, instead, whether voters prefer a particular party over another -- can give clues as to what's happening in districts where there isn't much data. But imposing national narratives on individual races obscures important particulars: like policy changes that affect some congressional districts more than others -- drawing voter opprobrium -- or incumbents who are losing not because they're caught in an anti-incumbent wave but because they deserve to lose and would in any year. At the same time, individual races that match the national trend, like nearly every one involving a Blue Dog Democrat in trouble, reinforce the narrative.
Analysts don't make predictions entirely without data, of course. While there may be a dearth of public polling, other information -- like fundraising activity, the amount of advertising, and historical trends -- is public. These external clues sometimes inspire pollsters to jump on a race. A candidate like Sean Bielat, Barney Frank's Republican challenger in Massachusetts, raised millions from outside conservative groups who hope to unseat the unabashed liberal. That focused attention on a race that would otherwise have seemed a sure thing for Frank, who still has a solid 13-point lead, according to a Boston Globe poll. Similar advertising blitzes indicated that the races involving incumbents Solomon Ortiz, a Democrat from Texas, and Maurice Hinchey, a Democrat from New York, might be in play, putting those races on the national radar screens of analysts -- and funders. Neither race has yet attracted national polling.
Generating that kind of buzz is a big reason that campaigns leak internal data to the forecasters. When a campaign's internal polls are favorable, or show an incumbent with a less comfortable lead than assumed, it's in the campaign's interest to share. But that's reason enough to be suspicious: Campaigns aren't going to release polls unfavorable to their candidates. And internal polls tend to favor their own candidates by as much as 6 percentage points, according to analysis by Nate Silver. Yet official forecasters rely a lot on those kinds of leaks and that kind of inside information.
Everything becomes clearer as Election Day nears, says Tim Sahd, who researches House races for National Journal. There are a few races that the national campaigns are pulling out of now; a good sign they think it's a hopeless cause. But in cases where there's truly a dearth of data, it's difficult to determine motivations, particularly given national and generic polls that show we're in a wave election that is going to be much better for the Republicans than for the Democrats. But how much better doesn't exactly match up with individual race predictions. For House races, the Cook Political Report has Republicans picking up 40 House seats, but the predictions that individual seats will go to Republicans are less than 40. "It would be very interesting if the Democrats wind up losing fewer seats than the expert assessments say they will," Teixeira says. "It would show their micro assessments are pretty good, but their macro assessments got caught up in some sort of narrative."
There's not a lot of information about how that narrative can reinforce itself, but it led John Sides, a political scientist at George Washington University who blogs at The Monkey Cage, to quote a colleague of his who called this election a bubble. "There are lots of people saying the Republicans are going to take over Congress in part because lots of people are saying Republicans are going to take over Congress," he writes. "We know that the wisdom of crowds can be deadly even when people have lots of their own money at stake (read: subprime mortgages); we shouldn't necessarily expect it to be any better just because people's reputations are on the line, should we?"
