Election Update: ‘Likely Voters’ Aren’t Helping Trump Much

Want these election updates emailed to you right when they’re published? Sign up here.

FiveThirtyEight’s highest traffic often comes on the day just after major events, like debates or key presidential primaries. Everyone wants to know how those events are going to move the polls. We’re sometimes happy to speculate about that, when we think the answer is obvious enough. But the problem is that there usually isn’t any data that soon, since the polls will take a few days to register any effects.

In fact, it’s rare to see pollsters release results at all in the first day or two after a debate. So you usually end up with a weird mix of polls. The daily tracking polls don’t stop publishing, but they’re an outlier-y set of polls this year that usually aren’t worth getting too worked up about. Someone decided to publish a Tennessee poll today? Hey, that’s cool, I guess. About the only number of any real consequence on Thursday was a Suffolk University poll that showed a tied race in Ohio — not a bad result for Donald Trump, except that he’d been up 3 percentage points in Suffolk’s previous poll of the state.

Therefore, our forecast is largely unchanged, with Hillary Clinton having an 87 percent chance of winning the presidency according to our polls-only model, and 84 percent according to polls plus. To repeat: None of the polls yet reflect any potential effects from the third presidential debate on Wednesday, so our forecast doesn’t account for the debate either.

So let’s use today’s Election Update as an excuse to talk about something else: the effect that likely-voter models are having on the polls. Every pollster has its own method for determining who’s a likely voter, although I generally think of likely-voter models as falling into two categories: “soft” likely-voter screens that simply ask voters whether they’re likely to vote, and “hard” likely-voter screens like the one that Gallup employed that use a longer battery of questions, including voting history. I won’t go into too much more detail on methodology here, but there’s a great primer on likely-voter models at the old Mystery Pollster site if you’re into that kind of thing.

Typically, likely-voter models help Republican candidates, since their voters are older, whiter and have higher socioeconomic status, all of which correlate with higher turnout. (Although, note that the latter variable may not hold with Trump because his support has collapsed with college-educated voters.) But the effect varies from election to election. In the 2010 midterms, there was a severe difference between likely-voter and registered-voter polls, with likely-voter polls being a net of about 6 percentage points better for Republicans. And the likely-voter polls basically turned out to be right: Republicans had a great year, underperforming their polls in a couple of close Senate races but winning even more House seats than expected.

The difference between likely-voter and registered-voter polls tends to be smaller in presidential election years. In fact, it probably should be fairly small since most registered voters — somewhere north of 80 percent, according to the Current Population Survey — turn out to vote. In 2008 — a year of profound Democratic enthusiasm behind Barack Obama — there was almost no difference between registered- and likely-voter polls. In 2012, there was a gap, with Mitt Romney gaining a net of about 2 points in likely-voter surveys. Given that likely-voter polls underestimated Obama’s performance in the end, however, perhaps that was too much.

This year, there also isn’t much of a gap. We keep diligent track of polls like this one that release both registered- and likely-voter numbers, since they’re used to calibrate the likely-voter adjustment that our models use. So here’s a summary of every such poll we’ve found. In the table, the “weight” column indicates how much influence the pollster has on the likely-voter adjustment, which is a function of the number of polls it has conducted and its pollster rating.

A couple of things to point out here. First, while we usually talk about a likely-voter screen as helping the Republican — or, occasionally, as helping the Democrat — technically speaking they usually help both major-party candidates, as both gain votes from the undecided column (and sometimes from third-party candidates). It’s just that the Republican usually gains a little more.

That’s true this year also, but the difference is minor. Trump gains a net of 2.5 percentage points in likely-voter polls, as compared with registered-voter polls. But Clinton gains 1.7 points. So the net gain for Trump is only 0.8 points.

And second, there’s actually quite a bit of agreement among pollsters about the likely- vs. registered-voter gap this year, at least when you look at each firm’s polling in the aggregate. Most show somewhere between a 1-point gap favoring Clinton and a 2-point gap favoring Trump. Individual polls can sometimes show larger gaps, especially if a pollster is using “hard,” Gallup-style likely-voter screens. (Those “hard” screens can sometimes introduce artificial volatility into a poll.) But taken as a whole, the consensus is that there’s at most a modest advantage for Trump in the likely-voter electorate, somewhat less than a Republican candidate would typically receive.

That makes a certain amount of sense. Trump voters may be more enthusiastic, although the data has been a bit mixed on that, and voter enthusiasm can vary with events on the campaign trail. And his voters are certainly older and whiter than Clinton’s. But Trump’s reliance on voters without college degrees — especially men without college degrees — could disadvantage him because they turn out at lesser rates.

Then there’s Trump’s lack of a turnout operation, which may or may not be reflected in polls. If Clinton’s campaign is more successful at persuading low-propensity voters to go to the polls, those voters may make it through the “soft” likely-voter screens but may be incorrectly screened out by “hard” screens that rely on past voting history. The effect of Clinton’s ground game on the polls — and the eventual outcome — is one of the big unanswered questions about the race, but presents some further downside risk for Trump.