Taken individually, few of these polls stick out as outliers. Kansas does stick, and arguably so does Florida, where YouGov gives Rick Scott his biggest lead yet in a nonpartisan poll. And the results in Michigan, North Carolina and Montana's Senate races buck recent trends that had shown improvements for Democrats. In general, though, YouGov's distance from the polling averages is generally plausible.
But when these polls are taken together, it's striking that in almost every case the results are rosier for the GOP than what other polls are suggesting. In seven of the 9 tightest Senate races, Republican candidates are ahead by more than the Huffington Post's Pollster average—in many cases (such as in Michigan and North Carolina) significantly.
Add to that a collection of implausibly tight results in states where Democratic incumbents have otherwise been cruising: Mark Warner's 10 percent lead in Virginia, Dick Durbin's 8 percent lead in Illinois, and Cory Booker's 7 percent lead in New Jersey are all about half of their current average margins.
There's precedent for this: In 2012, the margin of YouGov's final polls favored Republicans by an average 5 points, including large errors in competitive races in Nevada, Wisconsin, Connecticut, and Virginia—all in the GOP's favor.
At the very least, that means that these polls are at the Republican-friendly end of the spectrum of plausibility. But it also does raise some methodological questions, some of which the Times's Nate Cohn addresses.
At the top of the list is the sample's composition. Cohn observes that only 81 percent of Americans use the Internet and that those who don't "tend to be less educated, less affluent and more likely to be Hispanic or over age 65." And there's no doubt that YouGov had trouble reaching nonwhite respondents: Only 6 percent of its Michigan respondents are black (as opposed to 16 percent of the 2012 electorate).
Now, that's only the unweighted share of African-American respondents. YouGov attempted to correct this by weighting, giving smaller samples of under-represented populations larger weight in the final results. Problems, however, remain.
For one, YouGov must guess the electorate's composition, which means that it may underestimate minority voters. (And YouGov made it hard to assess their decisions, since the crosstabs don't specify the samples' weighted composition.) For another, upweighting a group to more than double its original size can wreak havoc on the final results. Small sub-samples can generate bizarre results, which get magnified during the weighting process.
That affected YouGov's measurement of the non-white vote in 2012, and it may have done so again this time. Michigan's two statewide Democratic candidates get just 69 and 57 percent of African-American respondents' vote, for instance. And Dana Houle notes that most of the races in which Democrats saw favorable results in this batch of polling (Alaska, Minnesota, and New Hampshire) are in overwhelmingly white states, i.e., those states in which Democrats are less reliant on harder-to-measure non-white voters.
But what really makes these YouGov polls seem particularly sloppy is that they did not include any notable third-party candidates.
Sure, third-party candidates often poll higher than they end up getting. But how in the world could they not include Eliot Cutler in Maine's Governor race, an independent who took 36 percent in the 2010 election (17 percent more than the Democratic candidate), and when his presence on the ballot this year is the only reason the election is competitive in the first place? Democrat Mike Michaud leads the two-way match-up against Republican Governor LePage by 13 percent, but this poll is utterly meaningless without Cutler.
And how can they not include the Libertarians running in Georgia when a huge question in these races is whether the general election will go to a runoff? How can they not include former Sen. Larry Pressler, whose candidacy is a key factor that could make South Dakota's Senate contest much tighter than expected? These are glaring omissions.
Questions still remain as to how reliable Internet polling can be, and how attentive these national panels tend to be to state-specific details that matter in close elections. And these specific polls, as we've elaborated above, have issues of their own. But they also do give us an indication of what a realistically bad Election Day would look like for Democrats—and it wouldn't be pretty.
Comments are closed on this story.