In the final days of the presidential campaign, we’ve all become very good at forgetting the lessons we learned earlier in the election cycle. “Anything could happen” seems to be the common refrain ― including from Hillary Clinton’s campaign, which is regularly sending fundraising emails with that theme.
The Clinton campaign needs to do that in order to keep excitement up, generate donations and convey the urgency of getting out the vote. But number-crunchers need to look at the numbers and what we really know about election polling.
What we know is this: Public opinion doesn’t swing wildly in the final days before a presidential election without some huge crisis. The FBI’s mishandled email issues certainly aren’t big enough, especially since Clinton’s email woes are nothing new. And previous “October surprises” haven’t moved the mark much, either.
While voter preferences don’t move back and forth like a pendulum at the end of a campaign, polling numbers do. There are lots of reasons for that, some of which we don’t even completely understand.
Polls have many sources of error. They leak in at every point in the survey process ― including how the survey is done; the sample selection; who answers the poll; how the questions are worded; and even how the data are treated after the poll is completed.
But the source of error that’s most pertinent to this week’s debates is a phenomenon called “differential nonresponse.” In the context of election polls, that means when things are going badly for one candidate, their supporters are less likely to answer a poll than when things are going well. So polls that don’t control for changes in who responds will swing based on events ― not because anyone changed their mind, but because people stopped responding to pollsters for a few days.
YouGov documented the partisan differential nonresponse patterns that resulted from the latest FBI email dust-up. The news made very little difference to the poll’s overall bottom line, but the pattern of fewer Democrats responding in the initial period after the scandal is apparent.
There’s no evidence that anyone’s mind changed or that anyone’s likelihood of voting changed ― instead, they just didn’t answer the poll.
So given all this instability in polls, the inevitable question is how I’m so certain reality favors Clinton and not Donald Trump. That’s fairly simple: If this were a truly competitive race, there would be more national polls showing Trump ahead. The Electoral College will decide the election at the state level, of course, but the national polls provide a decent indicator of the trends.
These trends indicate that Trump has closed the race slightly, yes, but ― and I can’t emphasize this enough ― there are no polls showing Trump leading nationally. At least none that meet HuffPost Pollster’s transparency criteria to make it into our charts (which most polls do). Plus, large tracking pollsand panels show relatively little movement in voters’ opinions over time. And in states like Florida and North Carolina that Trump would have to win, Clinton is up in the vast majority of the polls.
All this is to say that when we average the polls out, there’s not a lot of movement around a solid Clinton advantage. And there hasn’t been all year. If there were a substantial risk of Trump actually being ahead, there would be more polls showing him leading.