At FiveThirtyEight, we generally prefer state polls to national polls. So far, though, we haven’t had many of them to work with. If you’re getting dozens of national polls every week, but just a smattering of state-level surveys — and that’s what we’ve been getting — you’re better off inferring what’s going on in the states from the trend in national polls, rather than the other way around.
For example, Hillary Clinton has gone from having roughly a 3 or 4 percentage point lead over Donald Trump in national polls in early July to more like an 8-point lead now. Therefore, we’d expect her to gain perhaps 4 or 5 points in polls of Pennsylvania, Florida, Ohio and other swing states if polls were taken in those states now, compared to the previous versions of those polls conducted a month ago.
On Tuesday, we finally got a bunch of state polls to test the theory — three polls each from Quinnipiac University and Marist College. In fact, the new data mostly confirms our hypothesis, although with some caveats. Clinton gained an average of 4 percentage points across the six surveys. The clearest trend toward Clinton is in Pennsylvania, which is now part of her path of least resistance to 270 electoral votes. Here are the new surveys:
| STATE | POLLSTER | JULY POLL | LATEST POLL | SWING |
|---|---|---|---|---|
| Pennsylvania | Marist | Clinton +8 | Clinton +9 | Clinton +1 |
| Ohio | Marist | Clinton +3 | Clinton +4 | Clinton +1 |
| Iowa | Marist | Tie | Tie | — |
| Pennsylvania | Quinnipiac | Trump +6 | Clinton +9 | Clinton +15 |
| Ohio | Quinnipiac | Trump +1 | Clinton +2 | Clinton +3 |
| Florida | Quinnipiac | Trump +5 | Tie | Clinton +5 |
| Average | Tie | Clinton +4 | Clinton +4 |
Note that these numbers are based on the versions of the polls with third-party candidates included, which is the version FiveThirtyEight’s models use. The head-to-head versions of the Marist and Quinnipiac polls were a bit better for Clinton, perhaps because there are still a fair number of Bernie Sanders supporters who say they’ll vote for Libertarian Gary Johnson or the Green Party’s Jill Stein. In fact, Johnson gained ground in the Marist polls, when third-party candidates typically lose ground after the party conventions.
Overall, these are good but not great numbers for Clinton; as I said, they’re about in line with what you’d expect based on trends in national polls. It’s also not surprising that Clinton improved more in Quinnipiac’s polls than in Marist’s, since Quinnipiac previously had a strongly Republican-leaning “house effect” while Marist had among the more favorable polls for Clinton. Sometimes those house effects diminish as the election draws onward, with polls tending to converge toward one another. Note also that Quinnipiac shifted from surveying registered voters to likely voters, a change that normally helps Republican candidates but that might not this year given Trump’s potential reliance on lower-propensity voters.
The best news for Clinton is in Pennsylvania, where a variety of post-convention surveys, including the two released on Tuesday, show her ahead by around 9 percentage points. Our polls-only model now projects Clinton to win Pennsylvania by 8.5 percentage points, making it just a pinch bluer than the national average. (Polls-only has Clinton winning the national popular vote by 8.0 percentage points.) By contrast, polls-only has Clinton projected to win Ohio, Florida and Iowa by 5 to 6 percentage points, meaning that they lag her national numbers a bit.
Let’s not bury the lead: Clinton is polling really well right now, and if you held an election today, she’d probably win in a landslide, possibly including states such as Georgia and Arizona along with most or all of the traditional swing states.
Still, Clinton doesn’t quite have a firewall of 270 electoral votes. If Trump makes a big comeback, her Electoral College position is decent but not great. Including Pennsylvania, Clinton has 269 electoral votes in states where our polls-only model projects her to win by more than 8 percentage points. This category also includes Wisconsin, Michigan, Colorado and Virginia, states where Clinton can feel fairly confident about her position based on polling and demographics. Note, however, that 269 electoral votes is not enough to win — it’s only enough to tie. Also, the 269 total includes an electoral vote from Maine’s rural 2nd Congressional District (Maine splits its vote by congressional district). Trump led the only poll of the 2nd district, and our model favors Clinton there, but sees a highly competitive race.
So Clinton needs one more state, of any size, in this firewall scenario. Ohio and Florida are candidates, of course, but they’re heavy lifts. Couldn’t Clinton target something a little smaller?
On the basis of demographics, Nevada might seem like a strong Clinton state, but her polling there has been mediocre all year. She also hasn’t polled that well in Iowa, where Marist’s latest poll showed a tied race. New Hampshire, where Clinton had a big lead in the only post-conventions poll, might be the best option, but it can be swingy, meaning that if the election shifted back toward Trump overall, New Hampshire might swing right along with it. North Carolina is a possibility: Clinton’s polling has been no worse there (and no better) than most of the other states I’ve mentioned. North Carolina could really be a tipping-point state this year that makes the difference between winning and losing maps for Clinton, and not just an add-on that helps her run up the score.
So Pennsylvania has helped, and Clinton has a lot of options. But overall, our models have her as being more likely to win the popular vote and lose the Electoral College than the other way around. Of course, neither possibility is all that likely.
Overall, Clinton’s position is strong and may still be improving. Instead of seeing her convention bounce fade, she’s holding onto or even improving her lead over Trump in a variety of national polls. You can find one or two polls showing a rebound toward Trump if you look hard enough, but they’re outweighed by polls showing Clinton gaining, including a few that give her a double-digit national lead.
Our polls-plus model continues to expect a rebound toward Trump, in part because Clinton may still be in the midst of a convention bounce. (It will be another week or two before we can be confident that the convention’s effects have faded.) Still, the polls-plus model gives Clinton a 79 percent chance of winning the Electoral College, her highest figure so far this year. And polls-only, which doesn’t adjust for convention bounces, has Clinton with an 88 percent chance.
For now, those lofty percentages for Clinton are mostly based on her strong standing in national polls — and swing state polls that seem to be consistent with the national data. The volume of state polling typically increases a lot at about this stage of the campaign, however. So before long, those state polls will be driving most of the changes in the outlook.