I was just sitting here thinking that we were in for a relatively newsless, perhaps even anticlimactic, finish to the presidential campaign. Then the news broke that FBI Director James Comey sent a letter to congressional leaders saying that the FBI had “learned of the existence of emails that appear to be pertinent to the investigation” of Hillary Clinton’s personal email server during her time as secretary of state. The FBI will take “appropriate investigative steps” to review the emails, the letter said.
The emails apparently came from electronic devices belonging to Anthony Weiner, the former congressman, and his wife, Huma Abedin, an aide to Clinton, and surfaced as part of an investigation into lewd text messages that Weiner sent to underage women. It isn’t clear that the emails directly implicate Clinton, and the reporting I’ve followed so far suggests that in a legal sense, Comey’s decision to inform Congress may be something done out of an “abundance of caution.” But in a political sense, there’s certainly some downside for Clinton in the appearance of headlines containing the words “FBI,” “investigation” and “email” just 11 days before the election.
We’ll return to the FBI news in a moment, but first, a quick look at where our forecast stands — and I’ll remind you that it is based on polls and won’t reflect any effect from the FBI news until the polls do. We’ve reached the point in the campaign in which there are so many polls coming in — state polls, national polls, tracking polls, one-off polls — that it’s really nice to have a model to sort out all the data. A couple of days ago, the model was beginning to detect tenuous signs that the presidential race was tightening. Now, that seems a bit clearer. Clinton’s lead over Donald Trump is now 5.7 percentage points in our polls-only model, down from 7.1 points on Oct. 17. And Trump’s chances of winning the election have recovered to 18 percent from a low of 12 percent. Trump’s chances in our polls-plus forecast are 21 percent, improved from a low of 15 percent.
Almost all the tightening is happening because Trump’s numbers have improved. Clinton’s share of the vote — about 46 percent in national polls — is still as high as it’s been all campaign. But Trump seems to have brought home some Republicans who were thinking about sitting out the election or voting for a third-party candidate. Libertarian Gary Johnson has fallen to 5 percent in our popular-vote forecast — his lowest point to date. (If Johnson finishes with less than 5 percent, the Libertarian Party would be deprived of federal matching funds for the 2020 election.) The undecided vote is also declining, although it remains high compared with recent elections.
The tightening is modest enough that it’s not apparent in every poll. It isn’t hard to find polls with favorable trend lines for Clinton, in fact. But there are slightly more of them that have favorable trend lines for Trump, at least compared with his mid-October lows.
To take a wider vantage point: If you’d told Clinton a year ago that she would enter the final week of the campaign with a 5- or 6-point lead and roughly an 80 percent chance of winning the presidency, she’d probably have been very pleased with that. At the same time, we haven’t seen the bottom completely fall out for Trump as seemed possible a week or two ago. There have been no major opposition-research dumps on Trump over the last week, and he’s been relatively quiet on the campaign trail, making less news than he usually does. (Obligatory caveat: That could change at any time.)
Instead, the surprise of the day was a negative one for Clinton. Our general view of campaign-related news coverage is that it’s evolved a lot from 2012, when many events were hyped to be “game-changers” and very few turned out to move the polls. In fact, among some of the more empirically minded journalists I follow, the conventional wisdom may have overcompensated too far in the direction of “lolz nothing matters,” downplaying the importance of events that seemed highly likely to move the polls, such as the first presidential debate. In general, making news of any kind has been bad for Clinton and Trump, as periods of more intense coverage have been followed by declines in the polls.
One of those events came July 5, when Comey announced that there would be no criminal charges in the investigation into Clinton’s server but repudiated Clinton for the way she handled email. That announcement brought the story back into the news and preceded a 2-percentage-point drop for Clinton in national polls, although there was no obvious gain for Trump — instead, voters retreated into the undecided column.
If, hypothetically, the same thing were to happen again — Clinton loses 2 points to undecided overnight — her odds of winning the election would decline to 68 percent in our polls-only model. So that’s a significant shift.
But it’s not clear that the situations are comparable, especially if the “investigative steps” have nothing directly to do with Clinton or her email server and won’t be completed until after the election. My hunch (like The Washington Post’s Dave Weigel’s) is that Weiner is such a tragicomic figure, and such a lightning rod for news coverage, that he could insulate Clinton from some of the fallout she might have suffered otherwise. There are also fewer undecided voters now than there were in July, voter choices are more locked in, and many people have already voted — which could lessen the impact.
Even so, the news could get Republican partisans riled up, increasing turnout, and could play into a closing message for Trump about Clinton’s “corruption” — if he’s disciplined enough to sustain one. At a minimum, there’s no upside in the story for Clinton.
To repeat, our forecasts won’t show any impact until and unless the polls do. But in general, FiveThirtyEight’s models are relatively quick about detecting trends from the polls. At the same time, they account for a greater amount of uncertainty than most other models, which results in a better chance for the trailing candidate (in this case, Trump). Our forecast already saw Clinton as less of a sure thing than other models, which variously give her a 92 percent to a 99 percent chance of winning as compared with our roughly 80 percent chance, and that gap could widen in the coming days.