Latest Upshot poll shows Trump with a lead in Florida

New York Times News Service

Donald Trump has slowly but surely improved his standing in state and national polls since the final presidential debate.

A New York Times Upshot/Siena poll released Sunday is consistent with that trend: It gives Trump a 4-point lead in Florida, 46 percent to 42 percent, in a four-way race. In our first poll of Florida a month ago, Trump trailed Hillary Clinton by a percentage point.

The survey is Trump’s best recent poll in Florida, and it should be interpreted with caution. In general, it is best to look at an average of polls. Clinton still leads in an average of recent Florida surveys by nearly 3 points.

But the poll is not the only one to show Trump in the lead. A Bloomberg/Selzer poll, which is methodologically similar to the New York Times Upshot/Siena poll, showed Trump with a 2-point edge last week.

Trump, a Republican, has no plausible path to the presidency without Florida’s 29 electoral votes. But his Democratic opponent has many ways to win without the state. Clinton would almost certainly win if she carried North Carolina and Pennsylvania, where recent Upshot/Siena polls have shown her with a comfortable advantage.

The poll was taken before the FBI director, James Comey, informed Congress that the bureau had obtained additional information of potential relevance to an investigation into Clinton’s emails. National polls before Comey’s letter showed Clinton with a 6-point lead, an edge that was somewhat smaller than it was earlier in the month.

The poll paints a much rosier picture for Trump. Across every dimension of the survey, the poll has subtle but good news for his chances. If Trump won the election, it probably would look a lot like this:

Far greater Republican unity: Trump won 86 percent of self-identified Republicans — the highest percentage of that group in any Upshot/Siena survey this year.

He had the support of 84 percent of registered Republicans, up from 72 percent in September and also the highest of any Upshot/Siena survey this year.

Republican Presidential Candidate Donald J. Trump speaks during his rally at the Pennsylvania Farm Show Complex & Expo Center in Harrisburg Thursday. Amanda J. Cain photo

Trump’s consolidation of Republican-leaning voters is a trend in national surveys, and it comes alongside a corresponding decline in the number of supporters for Gary Johnson, the Libertarian candidate, who received just 4 percent of the vote in our survey — the lowest of any Upshot/Siena poll. Republicans have been likelier than Democrats to support Johnson in most of our polls.

Even college-educated white voters, who have been skeptical of Trump nationwide, are showing less skepticism in Florida. He has a lead of 51 percent to 35 percent among those voters in our survey.

Clinton weakness among white working-class Democrats: Trump leads among white voters without a college degree by an impressive margin of 63 percent to 24 percent. He’s so strong that Clinton has just 55 percent of the vote among white registered Democrats without a degree, compared with Trump’s 32 percent.

The combination of Republican unity and a large dissenting vote among registered Democrats is responsible for Trump’s lead.

Clinton actually leads among voters who are unaffiliated with a major party — something that’s been true in all five Upshot surveys: in North Carolina (two surveys), Florida (two) and Pennsylvania (one). In this case, it’s by a 10-point margin.

Our survey is adjusted to have the right number of registered Democrats and Republicans, which is generally a good way to make sure that a sample doesn’t wind up being too Democratic or Republican. Democrats have a 1-point registration edge among likely voters.

But because of this large dissenting vote from white working-class registered Democrats, and the unity of registered Republicans, the Republicans have a 2-point advantage in voters’ party identification in survey responses.

A Clinton challenge with black voters? Clinton has had nearly unanimous support among black voters in Upshot/Siena surveys, but not in this one: She had a lead of 81 percent to 11 percent. It might not seem like a big deal, but the difference between that support and the 90-1 we saw from black voters in Pennsylvania covers about half of Trump’s lead.

One possibility is that this is random sampling error — we’re talking about seven black respondents who support Trump in our survey. But there are at least hints that Trump may be a tad stronger among black voters in Florida than elsewhere. Our first Upshot/Siena poll of Florida in September gave Clinton a lead of 83 percent to 4 percent among black voters, which was her worst performance among black voters in any of The Upshot’s polls up until now. A recent Selzer poll of Florida also gave Clinton a similar lead of 80 percent to 10 percent among black voters.

Clinton also had a challenge with black turnout. Voters who indicated on their voter registration form that they were black made up 12.7 percent of the likely electorate, down from 13.9 percent in 2012. Our North Carolina survey also showed the black share of the electorate dipping by about 1 point lower than 2012 levels. It’s a pattern that’s consistent with the initial early voting data, which shows lower black turnout than in 2012.

Cubans return to Trump: Clinton leads among Hispanic voters by a wide margin of 59 percent to 30 percent in our survey — a tally that’s pretty comparable to most recent Florida polls. But it is better for Trump than our September survey, when Clinton led by a margin of 61 percent to 21 percent.

Here, it is Cuban-American voters who make the difference. In September, Clinton had a lead of 41 percent to 33 percent among Cuban voters, with a huge number undecided or supporting a minor-party candidate. Now Trump leads, 52 percent vs. 42 percent.

Again, these are very small samples. But it’s plausible to imagine Trump recovering among Cuban voters.

A strong showing among Cuban voters also helped Sen. Marco Rubio lead by 9 points, 51 percent to 42 percent, against his Democratic challenger, Patrick Murphy. Murphy led by just 2 percentage points among Hispanic voters, and Rubio led by 69-28 among Cuban voters.

A mystery factor: partisan nonresponse: One of the biggest questions in political survey research is partisan nonresponse — the possibility that Democrats or Republicans are more or less likely to respond to polls.

Most public polls don’t have many ways to deal with it. They weight their surveys to match the demographic composition of adults — say, the right number of white and black voters — but they don’t adjust the number of Democrats or Republicans.

Our surveys are different: As mentioned earlier, they’re adjusted to have the right number of Democrats or Republicans.

In our survey, registered Democrats were much likelier to respond than Republicans. Registered Democrats had an 8-point registration advantage in our unweighted sample, even though it was representative by other measures.

Clinton would have actually led in The Upshot’s survey if it, like most others, didn’t weight by party registration.

One possibility is that the public polls are understating Trump’s support because registered Republicans aren’t answering the telephone.

I would love to write that this was the only poll capturing the “hidden” Trump vote. And I would note that the Selzer/Bloomberg poll of Florida also happens to be weighted by party registration.

But based on my reporting, this is probably not what’s going on. Private pollsters are conducting surveys using similar methods, and they’re not seeing this. And after all, our poll found that registered Republicans were extremely likely to indicate their support for Trump — not exactly the shy Trump voter.

Puzzle: So let me conclude with a puzzle that highlights just how hard survey research can be.

Suppose, for the moment, that Republicans dissatisfied with Trump were less likely to respond to our survey, either by chance or because they were demoralized.

You would end up with a survey with too few Republicans, but those that you did have would be likely to support Trump.

It might not be so easy to adjust for partisan nonresponse.

If you then adjusted your survey to include the right number of registered Republicans, you would wind up with a sample full of Republicans who full-throatedly supported Trump — even though the reason you had too few Republicans was because so many don’t support him.

Is this what’s going on here? Who knows. The point is that any time there’s an adjustment for nonresponse, a pollster is making an assumption that the people being up-weighted are representative of the people who didn’t respond. That’s just an assumption, and it’s not always right.

We’ll find out for sure in eight very long days.