By Nate Silver – – – – –
There’s no longer any doubt that the party conventions have shifted the presidential election substantially toward Hillary Clinton. She received a larger bounce from her convention than Donald Trump got from his, but Trump has continued to poll so poorly in state and national surveys over the past two days that his problems may be getting worse.
The recent Fox News, Marist College and NBC News/Wall Street Journal national polls show Trump trailing Clinton by 9 to 14 percentage points, margins that would make for the largest general election blowout since 1984 if they held. Clinton’s numbers in those polls are on the high end of what we’ve seen lately — Marist, for instance, has generally had a Clinton-leaning house effect in its polls this year. By contrast, a series of polls released earlier in the week generally put Clinton’s advantage at 5 to 8 percentage points.
The new polls are noteworthy, however, because they postdate the earlier surveys — Marist’s poll was conducted Monday through Wednesday, for instance. That opens up the possibility that the spiral of negative stories for Trump, such as his criticism of the family of a Muslim-American soldier killed in action and his renewed feud with GOP leadership, are deepening his problems above and beyond Clinton’s convention bounce. Not only have Clinton’s numbers risen since the Democratic National Convention, but Trump’s numbers have fallen back into the mid- to high 30s in polls that include third-party candidates. And Trump’s favorability ratings, following modest improvement after his convention, are now about as bad as they’ve ever been.
Meanwhile, polls of Michigan, Pennsylvania and New Hampshire — three swing states with demographics that, in theory, could be friendly to Trump — showed Clinton with leads of 9 percentage points, 11 points and 15 points, respectively. Those are big leads for Clinton, but they shouldn’t be all that surprising: The margins look a lot like the ones by which Barack Obama defeated John McCain in those states in 2008, an election he won by 7.3 percentage points overall. According to our now-cast, Clinton would defeat Trump by a similar margin nationally, 7.9 percentage points, in a hypothetical election held today. Compared with that new, higher baseline for Clinton, a Suffolk poll showing her “only” 4 points ahead of Trump in Florida, which would have looked like an excellent result for her a week ago, is middling.
Overall, the now-cast estimates that Clinton’s electoral vote total, in an election held today, would be similar to the 365 electoral votes that Obama won in 2008. Although she’d be unlikely to carry Indiana, which Obama surprisingly won in 2008, she could make up for it by winning Arizona or Georgia, states that the now-cast has as tossups. Utah might even be competitive in an election held today — and the now-cast thinks that Texas would produce a closer finish than Pennsylvania.
But the real election is still just more than 94 days away. And our forward-looking models, which project the Nov. 8 result instead of evaluating what the polls look like now, are more conservative.
Our polls-plus forecast projects Clinton to win by about 4 percentage points on Nov. 8, meaning a margin more like Obama in 2012 than Obama in 2008. And given the wide uncertainty in forecasting an election three months out, it has Trump with a 26 percent chance of winning. Clinton has nothing to complain about — her 74 percent chance is her highest mark in the polls-plus forecast all year. But the model tweaks her numbers downward for two reasons.
First, it adjusts for potential convention bounces. Although the bounce following the second convention is historically not as misleading as the one following the first convention,1 Clinton’s numbers may still be elevated by a couple of percentage points from the convention afterglow.
Second, polls-plus combines the polls with a “fundamentals” forecast based on an economic index, and the economy is average, suggesting that the election ought to be close. Obviously, there’s a big assumption embedded in there — that Trump is a normal candidate who can take advantage of macroeconomic conditions in the same way that (for instance) John Kasich or Marco Rubio might. Still, American presidential elections have tended to tighten down the stretch run more often than not.
Our polls-only forecast also discounts the recent polls to some degree, projecting Clinton to win by 6 percentage points on Nov. 8 and giving her an 80 percent chance of winning the Electoral College. Polls-only doesn’t use the economic index, nor does it lower Clinton’s numbers because of a potential convention bounce. But it does weight polls taken during the conventions less. Furthermore, it’s deliberately a bit sluggish to update its forecasts because presidential polls are mildly mean-reverting, meaning that gains in the polls are more likely to reverse themselves than to continue unabated.
A model can be too stubborn to update its forecast. Clinton, after blowing a 7-point lead in July, is now in the midst of one of the bigger convention bounces in recent years. This election has produced large swings by historical standards, and the odds ought to have shifted back and forth, in the same way they would if an NFL team forfeited a two-touchdown lead before halftime and then regained it in the third quarter. But you shouldn’t rush to judgment based on two days of polling (admittedly excellent though they were for Clinton) when there are still about 94 days to go. A poll showing Clinton with a 9-point lead three weeks from now would be more meaningful than three more such polls taken tomorrow.
- The problem after the first convention is that voters have heard a one-sided argument, since the other party hasn’t had a chance to respond.
Nate Silver is the founder and editor in chief of FiveThirtyEight.