Whatever arguments we’ve had about the polls this week will soon be swamped by the reaction to Monday’s presidential debate. As a rough guide, I’d expect us to have some initial sense of how the debate has moved the numbers by Thursday or Friday based on quick-turnaround polls, and a clearer one by next Sunday or so, when an array of higher-quality polls will begin to report their post-debate results as well.
But in the meantime, let’s take one more step back and ask our usual collection of 10 questions about where the presidential race stands. Think of these as time capsules of a sort; you can find previous editions here (July 15), here (Aug. 15) and here (Sept. 6).
1. Who’s ahead in the polls right now?
Hillary Clinton, but tenuously. There were some semantic debates on Twitter this morning after ABC News and the Washington Post released their new national poll that showed Clinton 2 points ahead of Donald Trump. Did the Post convey the right impression in describing the race as a “virtual dead heat” in its headline?
I might have chosen slightly different vocabulary: “Clinton has razor-thin advantage,” or something like that. But it’s basically correct, at least based on the FiveThirtyEight forecast, to characterize the election as both close and competitive. The ABC News/Post poll is typical of recent national polls, which have Clinton up by about 2 points on average. (That average includes some high-quality polls that have Clinton ahead by as many as 6 points, but also a handful of others that show Trump with a lead.) Meanwhile, in the Electoral College, Clinton is leading in the states she needs to win, but only in those states, and not by all that much. Trump is one string of good polls in Pennsylvania or Colorado or New Hampshire away from erasing that advantage.
To put it another way, a narrow Trump win would not count as a major polling foul-up if the election were held today: It would be within a reasonable range of disagreement among pollsters. A clear Trump win — or for that matter, a Clinton landslide — would be more of a problem for the polls.
2. What’s the degree of uncertainty?
It remains fairly high. This is the point that we really can’t emphasize enough, and it’s why FiveThirtyEight shows somewhat better odds for Trump than most other forecast models. Not all 2-point leads are created equal, and Clinton’s is on the less-safe side, certainly as compared with the roughly 2-point lead that President Obama had over Mitt Romney on the eve of the 2012 election.
Perhaps the most important reason for that is the higher-than-usual number of undecided and third-party voters. Clinton leads Trump roughly 42-40, based on our national polling average; late in the 2012 race, by contrast, Obama led Romney about 48-46. That means about 18 percent of the electorate isn’t yet committed to one of the major-party candidates, as compared with 6 percent late in 2012.1 The number of undecided and third-party voters has a strong historical correlation with both polling volatility and polling error — and in fact, the polls have been considerably more volatile this year than in 2012.
We also have a wider playing field of swing states this year — with states ranging from Georgia to Maine having been competitive at various points of the campaign. Thus, there’s a good chance that the polls (and the polling aggregators) will “call” several states wrong instead of getting 48 or 49 or 50 of them right, as they did in 2008 and 2012. That introduces important assumptions about how the errors between states are related. FiveThirtyEight’s model assumes these errors are somewhat correlated, especially in demographically similar states. If Trump beats his polls in Wisconsin, for example, he’s also likely to do so in Michigan.
Finally, it can be easy to lose sight of the fact that we still have a fairly long way to go. With the party conventions held early this year, everyone’s sense of timing was thrown off, and the 44 remaining days in this campaign are going to feel like an eternity.
It’s important to underscore that this uncertainty cuts in both directions. We give Trump better odds of winning than most other models, but we also assign higher odds to a Clinton landslide.
3. What’s the short-term trend in the polls?
In the very short-term — i.e., what the polls look like as compared with a week ago — it’s not clear. Clinton had appeared to be regaining some ground on Trump, but the polls that came in over the weekend were middling for her, including surveys showing just a 1-point Clinton lead in Colorado and a 2-point lead in Pennsylvania. Whether these changes reflect voters reacting to events in the news cycle or are just random fluctuations is hard to say.
4. What’s the medium-term trend in the polls?
It’s been toward Trump for a long time now. Clinton’s position peaked on Aug. 8 in the polls-plus model and Aug. 14 in polls-only. Trump has slowly and fairly steadily gained ground since then, closing his deficit by about 1 percentage point a week, to narrow Clinton’s lead from about 8 points in mid-August to 2 points now.
5. Which states shape up as most important?
There’s not any one key state, which is part of the reason the election remains uncertain — and exciting. Instead, the various swing states are currently lined up on either side of a gap, with Clinton leading in states representing 272 electoral votes2 and Trump ahead in states totalling 266 electoral votes.
If any of the states just to Clinton’s side of the gap slips toward Trump — Pennsylvania, Colorado, New Hampshire, Wisconsin and Michigan are the most plausible candidates — he’ll pull ahead in the Electoral College. But conversely, Trump leads by less than 1 percentage point in Florida and by barely more than that in North Carolina and if either of those were to fall to Clinton, his electoral math would become very difficult. The same is theoretically true for Ohio, although Trump’s lead has been more consistent there.
6. Does one candidate appear to have an overall edge in the Electoral College, relative to his or her position in the popular vote?
It’s complicated. There could plausibly be an Electoral College-popular vote split in either direction, but our models say that Trump is somewhat more likely to benefit from this.
Right now, Clinton is ahead by 2.0 percentage points in our national popular vote estimate. She’s also ahead by 2.8 percentage points in Colorado, which is currently the tipping-point state — the state that would give her just enough votes to win the Electoral College. That’s a potential advantage for Clinton, but it requires the polls to be pretty much exactly on the mark.
By contrast, Clinton’s position overall in swing state polls has not been especially good, in part because they tend to have a high proportion of white working-class voters — Trump’s best group. In particular, she’s gotten some pretty awful numbers in Ohio, Iowa and Nevada lately, and her position in North Carolina seems to be worsening. If instead of treating Colorado as definitely being the tipping-point state, we instead weight the states by their probability of being the tipping-point state, Clinton’s lead over Trump is 1.2 percentage points in the average swing state, less than her national margin and therefore a potential Electoral College disadvantage for her.
Basically, it’s a question of whether you’d rather have pretty good polling in exactly enough states to win 270 electoral votes, at the cost of pretty bad polling in the swing states overall. Our models say that isn’t a great trade-off for Clinton because having one good path leaves too little room for error. But this calculation is somewhat sensitive to our model’s assumptions. At a minimum, it’s another source of uncertainty.
7. How do the “fundamentals” look?
Non-polling factors such as economic conditions suggest that a race between a “generic” Democrat and a “generic” Republican ought to be close. In that sense, it shouldn’t be hard to see how Trump could win. He either becomes normalized enough that he performs about the same as a generic Republican would, or he significantly underperforms a generic Republican but Clinton’s problems are just as bad.
8. How do FiveThirtyEight’s forecasts compare against prediction markets?
In general, FiveThirtyEight’s polls-plus model has closely tracked betting markets — more closely than any other major forecast (including our own polls-only model). There’s a modest gap between them now. Whereas polls-plus gives Clinton a 58 percent chance and Trump a 42 percent chance, Betfair gives Clinton a 62 percent chance and Trump a 34 percent chance (reserving a 4 percent chance that someone other than Clinton or Trump somehow becomes the next president, a possibility that FiveThirtyEight’s models do not consider).
For a variety of reasons, ranging from the fact that that gap isn’t all that large to that betting markets can plausibly account for some factors that FiveThirtyEight’s models don’t, I’m not sure about whether I think there’s enough edge there that I’d actually advocate laying money down on Trump. I suppose I wish there were a betting market open only to political journalists and commentators, though. My hard-to-prove sense is that they underrate Trump’s chances as compared with both betting markets and our forecasts.
9. What would keep me up late at night if I were Clinton?
You mean, other than the fact that the election keeps getting closer every time I look?
As I put it in July: “I’d be worried that Americans come to view the race as one between two equally terrible choices, instead of Trump being uniquely unacceptable.” In particular, I’d be worried that my brand has irrevocably been tarnished with a reputation for dishonesty. Between Trump’s knack for exploiting this weakness (“Crooked Hillary”), the news media’s tendency to frame events as contributing to my honesty and trust problems, and some left-over hard feelings from the primaries — Clinton has yet to win over many of the millennials who voted for Bernie Sanders — I’m generally losing when polls ask who the more trustworthy candidate is.
In the short term, I’d be worried that the talk of Trump’s “low expectations” at the first debate is a tip-off that the media hivemind might frame a debate tie as a Trump win.
10. What would keep me up late at night if I were Trump?
Trump’s concerns also aren’t all that different from those he had in July.
If I were him, I’d be worried that even at my best moments, I’ve never really pulled ahead of Clinton, instead only drawing to within a point or two of her. I wouldn’t be that worried because it’s not clear how predictive those patterns are. But it certainly wouldn’t thrill me, especially given that the debates make for natural turning points and if the pattern holds, the next turning point will be back toward Clinton. I’d continue to be worried about my ground game, or lack thereof. It’s hard to say exactly how much that’s worth, but underperforming my polls by even half a percentage point in the swing states would make my Electoral College path meaningfully harder.