:format(webp)/cdn.vox-cdn.com/uploads/chorus_image/image/51733919/621764386.0.jpg)
We’ve been avidly following general election polls for months, and now their final verdict is in — Hillary Clinton leads Donald Trump by 3 to 4 points nationally, and leads in enough states to give her the presidency.
The only question, now, is whether they’ll be right.
Now that the final batch of polling is in, Clinton’s national margin is at 3.3 points in the RealClearPolitics average, 3.5 points in FiveThirtyEight’s model, and 4.6 points in HuffPost Pollster’s average.
Of course, the outcome won’t be determined by the popular vote nationally, but rather by the Electoral College. And you can think of the swing states in three main categories:
- Clinton’s firewall: Virginia, Colorado, Wisconsin, Michigan, Pennsylvania, and New Hampshire. These are the states that polls have indicated provide Clinton’s easiest path to victory. Pretty much all of the forecasters and poll averages agree that Clinton is favored in all of her firewall states, though they disagree as to how much.
- The toss-up trio: Florida, North Carolina, and Nevada. These states have generally looked more like total toss-ups in the polling.
- The “lean Trump” swing states: Ohio, Arizona, and Iowa. Polls have tended to show Trump ahead by a few points in these states, on average.
For a sense of the difference in these categories, FiveThirtyEight’s model gives Clinton a 70 percent chance of winning in each of her firewall states, a 54 to 57 percent chance in each toss-up state, and a 30 to 33 percent chance in each lean Trump swing state.
Overall, the various forecasting models — you can check out a roundup at the Upshot — estimate Clinton’s chances of victory between 72 percent and 99 percent. So they disagree on whether she's a reasonably solid favorite or an overwhelming one, but they all view her as likely to win.
The polls might be right. But they also might be off, for these reasons.
Generally, polling averages have proven to be a pretty good overall indicator of where the race will end up. But it's actually not that uncommon for them to be off by a couple of percentage points — national polls underestimated Obama’s margin by 2 to 3 points in 2012, after all. They could be off for any of the following reasons.
Pollsters could be modeling the electorate wrong. As Nate Cohn has written, even when pollsters use the same underlying data, they make a ton of decisions about how to model the electorate that often lead to different overall results. For instance, they could be underestimating Hispanic voter turnout, or sampling predominantly English-speaking Hispanic voters who might have different views than Spanish speakers. Alternatively, they could be underestimating working-class white turnout, or overestimating black voter turnout.
The ground games could play into this — that is, one candidate’s campaign could be better at identifying and turning out voters in the key states.
Undecided voters — there are still 5 to 6 percent of them on average in national polls — could also presumably swing overwhelmingly to one candidate, thus tipping the balance.
Alternatively, the polls could be thrown off by “shy” voters, who tell pollsters they're voting for one candidate or that they're undecided but secretly intend to vote for somebody else. This is the theory behind “shy Trump” voters, though experts say there probably aren’t enough of these to swing the election.
And then there's the prospect of herding. As Nate Silver writes, polls snapped into a startling degree of consensus (showing Clinton with a 3- or 4-point lead) in the final days, when there had been more variety among various pollsters’ results before. This is a known phenomenon that happens near the end of elections, as some pollsters apparently start to fear their results may be too far off from the consensus, and therefore skew them a bit.
Keep in mind, though, that if any of these effects are throwing off polls a bit, we don't know in what direction. They could be overrating Clinton’s margin, but they could also be underrating it. And different effects could cancel out each other somewhat.
