Donald Trump now trails Hillary Clinton by 6.6 percentage points, according to our polls-only model. And the clock is ticking. We’re 26 days from the election, and by this point in past campaigns, the concrete had basically dried.
Trump could stage a comeback. It’s possible. But it would be basically unprecedented.
Back in August, after the conventions, I wrote an article saying that while Clinton was a heavy favorite — she led Trump by between 6 and 8 percentage points — the race was far from over. We had plenty of historical examples of races shifting by more than 6 points, even after the political conventions. Since 1952, half of the races for which we had polling after the conventions featured shifts of 5 percentage points or more. Some elections, such as 1968, 1976 and 2000, looked like they could be near or total blowouts, but ended up being rather close.
But look at those same elections now. By this point in those campaigns, the polls had largely settled at a level close to where the elections would end up. And that’s true for pretty much all modern campaigns. Here are polling averages for the week preceding Oct. 12 for elections since 1952, along with the final margin1:
In most years, the early-October polls were pretty close to the mark, with a correlation of +0.96 between the polls and the final result. Ten of the 16 elections featured errors of 3 percentage points or less, and in all but three campaigns, the polls were within 5 percentage points of the final outcome. Even in 2012, when Mitt Romney closed his deficit against President Obama after the first debate, the polls at this point still showed Obama leading. At this point in the election cycle, the average error of polls for all elections is just 3.3 percentage points (much lower than the 4.7-point error we found for just after the conventions), and every candidate who’s been ahead in the popular vote in mid-October went on to win the election.
Simply put, there isn’t a precedent for a candidate coming back to win this late in the game after being behind by as much as Trump is now. That’s not to say Trump is dead in the water — polls are not perfectly predictive — but history doesn’t offer much hope for candidates in Trump’s position.
The nearest precedent available — if you squint and cover one eye — is probably 1992. The polling error that year, 8.5 percentage points, was larger than Trump’s current deficit in our polls-only forecast. In mid-October of 1992, Bill Clinton held a double-digit lead over George H.W. Bush. Clinton won, but only by about 6 points.
But the 1992 campaign isn’t a great model for the Trump campaign. First, Bush still lost. Second, 1992 featured a strong independent candidate in Ross Perot, which made polling the race more difficult. Gary Johnson’s support, in contrast, appears to be shrinking, which creates less volatility. And while Bush was hemorrhaging Republican support — as Trump is now — he still ended up losing more Republicans than Bill Clinton did Democrats on Election Day. That won’t work for Trump this year given that there are usually more self-identified Democrats than Republicans.
Trump’s supporters are better off pointing to two other races, 1980 and 1968, though neither of those are particularly encouraging either. In 1980, Ronald Reagan had a small lead over Jimmy Carter at this point in the campaign but ended up crushing Carter by nearly 10 percentage points. So that’s a decent precedent for Trump, right? Not really. The 1980 race shifted a great deal in the final weeks of the campaign, but even if 2016 underwent a similarly sized lurch towards Trump, he’d still lose. Not only that, but Reagan still had a major card to play in that election: the only debate between him and Carter. Reagan’s calm, collected performance there gave him a significant boost in the polls. This year, of course, two debates are already in the books, and Trump has lost both. Trump could obviously do better in the third debate, but it would likely have far less of an impact than the lone Carter-Reagan debate in 1980 did.
Instead, the most encouraging precedent for Trump is probably 1968. In that year, Democrat Hubert Humphrey was down by 5 percentage points and ended up losing by 1 point. Humphrey consolidated a divided Democratic base — just as Trump needs to do now with Republicans. Humphrey was also likely aided by a major October surprise — the halting of bombing in Vietnam by Democratic President Lyndon B. Johnson. That’s not a bad template for Trump — it would be difficult and he would need some outside help, but you can imagine it happening. Still, Trump is losing by a wider margin than Humphrey was, and the October surprises so far in 2016 seem to be working against Trump rather than in his favor.
As with any study of modern elections, we’re limited by sample size. That’s one reason that the FiveThirtyEight models give Trump roughly a 15 percent chance of winning instead of zero. But a Trump comeback would be like nothing we’ve ever seen before.