His previous best, the day of the first presidential debate, was 41.5 percent. His crapola performance that night plus a heap of scandal in the following weeks kept him durably below that number for a solid month — until today, when he’s up to 41.6 percent and seemingly still rising. In any other year, for any other major-party nominee, topping out at 42 percent would be known in professional parlance as “hot garbage” but Trump has the good fortune to be running against a candidate who is herself terrible. Hillary briefly cracked 46 percent two weeks ago after the “Access Hollywood” tape surfaced but she’s back to 45 percent now, several points higher than she’s been for most of the race but not enough to make Democrats comfortable. With her lead down to 3.4 points, I think this Harry Enten tweet perfectly captures the suspicion gnawing at the pros right now:
The key ? next few days isn't whether Trump closes the polling gap. It's whether he can get within 3 where a polling error could save him.
— (((Harry Enten))) (@ForecasterEnten) October 30, 2016
It’s not so much that they expect Trump’s surge to carry him past Clinton in the national average — although stay tuned — as that, if he narrows her lead to two points or so on the final weekend, the odds that the polls are wrong by pure chance and that Trump will pull the upset become much higher. Ten days ago, to believe he had a chance of winning, you had to believe that national polls were off by six points, a number outside the margin of error in many surveys. That would be a catastrophic miscalculation. A two-point lead, though, is within the MOE of most surveys, essentially making the race a toss-up. (Assuming a two-point lead with a three-point MOE, a one-point Trump win or five-point Clinton victory would be equally likely as I understand it.) That’s what data nerds are watching for now.
It’s not just Trump’s average that’s peaking either. In the most recent polls, he’s hit a number repeatedly that he’s rarely seen before:
He’s at 45 percent in Rasmussen, Gravis, and the ABC/WaPo tracker. Scroll down through the list of dozens upon dozens of surveys here and you’ll find that the total number of polls before this in which Trump had hit 45 is … one. One time. He did it in a CNN poll in the first week of September. That’s it. Suddenly he’s hitting that mark repeatedly, and nearly all of the data in the three polls I mentioned was collected before the FBI news on Friday. What’s happening, I’d guess, is that a chunk of Republicans who were lukewarm towards Trump, perhaps enough to support Gary Johnson instead, have decided to go all-in to beat the Democrats and swung around behind him after all. The danger of the Comey announcement to Clinton is that, by reminding undecideds and “soft” anti-Trumpers of what they don’t like about her rather than what they don’t like about him, it’ll intensify that trend. (Which is also why you’d expect Democrats to drop whatever’s left in their oppo bag against Trump in the next 72 hours, to turn the “who’s a bigger scumbag?” spotlight back onto Trump.) If she starts losing even a few percentage points among the usually Republican-leaning white college grads who’ve been tilting towards her all year, she’s in deep trouble.
But wait a sec. Let’s pause to remember the shining lesson of 2012, that national polls showing the race as a toss-up can be misleading. In the end we have 50 state elections, not a national election, and Trump’s electoral math remains exceptionally difficult. Hillary’s “blue wall” plus her consistent strength in key battlegrounds like Pennsylvania and New Hampshire plus the surprising tightness of the races in some red states like Arizona and Georgia means that she’s holding the equivalent of a full house. To beat her Trump needs the electoral version of four of a kind. How likely is that? Let’s look at the state polls. The first card in his four-of-a-kind play is Florida, where he trailed by 2.4 points or worse for most of the last three weeks, an ominous margin given how tight the state was in 2012. Clinton was notching leads of three to four points repeatedly in various surveys 10 days ago. But not anymore:
Dead heat. Trump’s come all the way back and leads by a comfortable margin of four points in not one but two polls, including the latest from the New York Times’s data site. Not all of the news there is good for him — check out the surge in Latino registered voters in Florida since 2012 — but the early- and late-voter polling seems to point to a photo finish. He really might be holding this card. (Derek Hunter wonders, not implausibly, whether Marco Rubio’s strong performance in the Senate race there is helping turn out Republicans who might not otherwise show up to vote for Trump.) He might be holding a second card too, namely, Ohio. Trump fell behind there after the “Access Hollywood” tape came out but he recovered two weeks ago and now leads by a point on average after having led there for most of the summer. It’s a toss-up, but if he really is climbing nationally, you’d expect his lead to grow a bit there over the next nine days. If so, that’s two of the four cards he needs for a winning hand.
The third card he needs is North Carolina. Here’s where things get dicey.
Both candidates are at their highest numbers on average in the state all year, but Clinton, not Trump, seems to be rising more steeply. She’s at 47 percent and leads by 2.9 points, essentially tied for her biggest margin of the campaign. Only one recent poll shows her leading by as little as two points whereas two show her up in the 6-7 range. And if you’re counting on the big FBI news to turn this around, remember that North Carolina has lots of early voting. That’s why Hillary chose NC as the site of her big joint rally with Michelle Obama last week. She’s desperate to bank as much of her lead in the state as she can. And according to this estimate by Nate Cohn of The Upshot, she’s doing a good job of it: Follow the last link for Cohn’s methodology but he guesses she’s leading by about 14 points right now in the early vote, a margin that will obviously narrow on Election Day but probably not enough to save Trump unless his post-Comey surge is bigger than most people expect. FiveThirtyEight currently gives Clinton a 64 percent chance of winning there. Early voting may end up saving this election for Democrats.
Let’s pause again and step back to look at the map. If Trump isn’t holding the third card, i.e. North Carolina, what is he up against? We’ll assume that he wins Florida and Ohio, and we’ll further assume that he holds onto Georgia and Utah even though those states are tight right now. (We’ll also assume that he holds onto, er, Alaska, where one very recent poll has Clinton up by four points.) What you’re looking at here is Clinton 263, Trump 228 with the five states in gray as toss-ups:
Give North Carolina to Trump if you like. Now he’s got three of the cards he needs in his four-of-a-kind bid. All Clinton needs to win the election at that point, though, is either Colorado or Arizona. Apart from one week after the first debate, she’s led in Colorado for the entire race. A new CBS poll today has her lead down to three points but Trump hasn’t led a single poll there in more than a month. Arizona is a true toss-up at this point, one which you would expect Trump to hold if he gains two or three more points nationally in the final week of the race, but Clinton is contesting it aggressively. (He leads there by two in the same CBS poll released this morning.) She’s spending $2 million in AZ and has sent A-list surrogates in Michelle Obama and Bernie Sanders to turn local Dems out. Clinton herself will hold a rally there this week. If she pulls the upset in that state, that’s ballgame nationally.
Looking back at the map, then, Trump’s fourth card in his attempt at four-of-a-kind is to either (a) win a clean sweep of Colorado, Arizona, Iowa, and Nevada after also winning North Carolina or (b) shock the world by pulling a blue state that no one’s expecting out of Clinton’s column. Which one? Pennsylvania? That’d be a gamechanger, but Clinton’s lead there is still nearly six points on average and was eight points in the CBS battleground poll this morning. Trump’s led exactly one poll there all year, in early July. New Hampshire? Clinton’s ahead by more than five points there too. Trump’s never led a poll of the state. Wisconsin? She’s ahead by six there and, again, he’s never led a single poll. Four-of-a-kind is a tough hand to put together, huh?
The beauty of the Comey disclosure, though, is that it introduces a degree of uncertainty into the race that could change all of this quickly. Go back again to Harry Enten’s tweet above. Is it possible that the polls are systemically underestimating the number of “hidden Trump voters” out there to the extent that their data will miss the outcome by six points? That … seems unlikely. Is it possible that basic statistical error plus “hidden Trumpers” could account for two or three points, though? Yeah, that seems possible. And if it happens, that fourth card may be in play after all.
Update: Sean Trende of RCP emails to correct me on a point. It’s not the margin of error that data nerds are worried about right now, it’s the error they might be collectively making in estimating turnout — specifically among “hidden Trump voters” and among core Democratic groups like blacks and young adults. They’re confident that they couldn’t guess so wildly wrong about turnout by each of those groups as to get the wrong result when Clinton’s lead is six points nationally. They’re a lot less confident about it if her lead is two points. Quote:
Even with a two-point race, sampling errors (MOE) should cancel each other out, leaving the true value as the average. The odds of dozens of polls being off in one direction even by two points by random chance is perishingly small.
What we’re concerns about is modeling error, where subjective choices made by pollsters move them off the true value. So, for example, if they think the African-American share of the electorate should be 13%, but it is 11%, that’s not MOE stuff or statistical stuff. It’s just the pollster being flat-out wrong about how to weight the raw data. Again, there’s no good way to assign a probability to that.