Swing state poll: the FBI letter totally changed the race — but just for one day

With six days before the 2016 presidential election, Hillary Clinton is still ahead in the polls. But many of her supporters are anxious — worried the revelation that the FBI is looking through 650,000 more emails in its investigation into her use of a private email server will change voters’ minds at the last minute, and polls won’t be able to capture the shift until it’s too late.

A new poll of Wisconsin, conducted by Marquette University Law School, should put those worries to rest — or at least reduce them greatly.

That’s not just because the poll shows Clinton beating Donald Trump by 6 points in Wisconsin. It’s because the poll was in the field while the news of the new emails broke — so it offers a window into how breaking news actually changes poll results and shapes the race.

The bottom line is that it’s easy to overstate how much a single news story can change the race. The effects of news, even big news, simply don’t last very long — and a news story is more likely to generate a big swing in polling results than it is in voters’ actual preferences.

Because Clinton’s use of a private email server has been an ongoing news story (and other stories involving email and Clinton allies have gotten lumped in under the heading of “Clinton email scandal”), Marquette’s pollsters were asking voters how concerned they were about it even before the letter from FBI director James Comey was released — and about half of likely voters agreed they were “very concerned.”

The day of Comey’s letter, that spiked to 60 percent. But in the days following that, only 48 percent of voters said they were “very concerned” about Clinton’s emails — if anything, fewer voters than were bothered before the new emails were revealed.

This makes sense if you think about it. At this point in a presidential election, people have generally made up their minds about candidates, and new information is more likely to confirm their beliefs than change them. (Clinton’s “email scandal” was already enough of a problem to concern half the electorate.) Just as importantly, people quickly stop caring about new information, whether it’s a news story or a campaign ad.

In the 2012 election, political scientists John Sides and Lynn Vavreck found that the effectiveness of ads basically “decayed” within five days of the ad’s airing. Ads that aired the day before Election Day were several times more effective than ads aired in the days before that — and more than a week before the election, ads hardly mattered at all. Similarly, they found, news stories that had appeared to change the race ultimately didn’t present permanent shifts in support.

Even though the presidential campaign’s been going on for a year and a half, most of that information really doesn’t matter to the election’s ultimate outcome. That’s true even when the news is late and big.

That said, if anything, the Marquette poll actually overstates the effect the FBI letter had on the presidential race. And other polls that were in the field that Friday probably did, too.

Here’s what Marquette pollsters found in terms of support for Clinton and Trump before, the day of, and after the letter came out:

The poll implies that 12 percent of likely Wisconsin voters decided to support Donald Trump the day the FBI letter came out, and then 8 percent abandoned him in the days after that. Those are huge swings!

But they probably don’t reflect huge changes in how likely Wisconsin voters felt about Donald Trump. More likely, they reflect changes in how willing Trump supporters in Wisconsin were to talk to pollsters.

This is called “survey response bias,” and it’s something that many polling experts — most notably the pollsters at YouGov — believe is having a big effect in the last weeks of the presidential race. They point out that when news breaks that makes one candidate look good, that candidate’s supporters get more enthused about the election; when news breaks that makes a candidate look particularly bad, her supporters are less likely to stay on the phone with a pollster and talk about the race.

If you control for who’s responding to you, the YouGov pollsters show, you get a much steadier picture of the 2016 campaign than the one you get from most polling: Instead of a hugely dynamic race with Clinton opening up double-digit leads at some points and falling even with Trump at others, you get a consistent Clinton lead of a few percentage points. That’s less exciting, but it’s enough to win her the presidency.

It is possible that the FBI letter will be such a huge deal that it really does shift people’s feelings about the campaign — or that something even bigger will break in the days right before the election, catching voters before the effects of a news story wear off. But the Marquette poll is good evidence that the FBI letter wasn’t quite the game changer it appeared to be — and that polls that say it was might need to be taken with a grain of salt.