At the end of our last wave of post-debate battleground polls, there were two state poll results that didn’t seem to fit the rest.
One was Pennsylvania: Kamala Harris led by four percentage points, making it her best result in the battlegrounds. It was our only state poll conducted immediately after the debate, when her supporters might have been especially excited to respond to a poll.
The other was Arizona: Donald J. Trump led by five points, making it his best result among the battlegrounds. Even stranger, it was a huge swing from our previous poll of the state, which Vice President Harris had led by five points.
In both cases, it seemed possible that another New York Times/Philadelphia Inquirer/Siena College poll would yield a significantly different result. With that in mind, we decided to take an additional measure of Arizona and Pennsylvania before our final polls at the end of the month.
The result? Essentially the same as our prior polls.
Ms. Harris leads by four points in Pennsylvania, just as she did immediately after the final debate.
Mr. Trump leads by six points in Arizona, about the same as the five-point lead he held three weeks ago.
That’s not what I expected. The average of other polls continues to show a tighter race in both states, and — unlike in Florida — there isn’t an obvious explanation for why the Times/Inquirer/Siena poll is producing a somewhat different result in these two states.
Nonetheless, the rest of this newsletter will try to make sense of it all. To be blunt, I can’t really make complete sense of it; there’s no conclusive explanation. You may find this unsatisfying — I do too — but it does yield an important insight: With less than four weeks to go, the polls don’t offer a clear answer on who will win.
Arizona
Whether or not today’s polls are off-target, the last two polls of Arizona take a clear position: Mr. Trump is well ahead.
Other polls differ. As of Friday, he led by only one point in the Arizona polling average compiled by The Times, before the Times/Inquirer/Siena poll was added.
The other polls come a lot closer to what I would have expected. Arizona does not easily fit the two overarching explanations for Democratic weakness offered by Times/Siena polling this cycle. It’s not a state, like Florida, where Democrats struggled in the midterm elections. And while the state does have many Hispanic voters, Ms. Harris has a fairly healthy 20-point lead among them.
Instead, the two Times/Inquirer/Siena polls show something we’re not seeing in many places in the country: a real challenge for Ms. Harris among white voters, including white college graduates.
If I had to craft a narrative to explain why Arizona is playing out differently, it would probably center on the unusual Democratic dependence on Republican-leaning voters (think, McCain Republicans and independents). Republicans have a clear advantage in party registration; to win, Democrats need to peel off a sizable chunk of Republicans and win unaffiliated voters by a wide margin.
When Mark Kelly won his Senate race in Arizona in 2022, he won 10 percent of self-identified Republicans and 14 percent of registered Republicans, and he held a wide lead among unaffiliated voters in our final Times/Siena poll. His fellow Democrat Ruben Gallego is currently seeing very similar results in his Senate race against Kari Lake. Mr. Gallego’s seven-point lead is even larger than Mr. Kelly’s was two years ago.
Ms. Harris is not winning over traditionally Republican-leaning voters, at least not in the last two Times/Siena polls. She isn’t winning independents, and she doesn’t draw away more Republicans than Mr. Trump draws Democrats, perhaps because she is not seen as especially moderate.
It’s also possible, of course, that the Times/Siena poll is simply off by a few points. Polls are inherently imprecise, and the Times/Siena poll is not perfect. On balance, our methodological choices have yielded more accurate results than other surveys, but it’s certainly possible Arizona will go the other way this year; this is the kind of case that poll averages are built to handle.
Pennsylvania
Over the last month, we’ve done 14 state or national polls. Of those polls, there are only two where Ms. Harris is running more than two points ahead of President Biden’s performance in the 2020 election: the two polls of Pennsylvania.
While the Times/Inquirer/Siena poll in Pennsylvania doesn’t differ as much from the average of other polls as it does in Arizona, the stakes of a four-point lead for Ms. Harris in Pennsylvania could be quite a bit higher. If Pennsylvania really is her strongest battleground state, it’s a big deal.
Here again, this is not what I would have expected. In fact, it wasn’t that long ago when polls suggested that Pennsylvania seemed as if it might be the weak link for Ms. Harris in her likeliest path to victory. The voter registration numbers have looked very good for Republicans as well.
But the Times/Inquirer/Siena polling isn’t entirely alone in suggesting that Pennsylvania might be a relative strong point for Ms. Harris, at least at the moment. Just last week, Quinnipiac had Ms. Harris ahead by three in Pennsylvania, even as it had her trailing in Michigan and Wisconsin. And a series of district polls — a Susquehanna poll in the 10th District and a Muhlenberg College poll in the Seventh District — both looked good or even great for Ms. Harris.
There are plausible reasons for Ms. Harris’s strength in Pennsylvania relative to Wisconsin and Michigan: It has a more highly educated population; it also may have the smallest share of white evangelical Christians (newly relevant in a year when abortion is a key issue); and it has a smaller Arab American and Muslim population than Michigan does (many of these voters are furious with the Biden administration’s handling of the war in Gaza). Together, that may create the ingredients for Pennsylvania to scoot to the left of Michigan (it was already to the left of Wisconsin in 2020).
Of course, it’s hard to avoid wondering about another possibility: nonresponse bias, which could happen if Democrats were overly eager to pick up the phone. That seemed to be the likeliest interpretation of the last Pennsylvania poll, and there were signs of it in the data.
This time, it’s harder to find. In fact, white registered Democrats were only 4 percent likelier to respond to the survey than white Republicans. (After the first debate, they were 16 percent likelier to respond.) There could still be nonresponse bias if, for instance, we simply get the wrong kind of Democrats or Republicans, but it’s reasonable to expect that Republicans as a whole would be much less likely to respond if Mr. Trump’s supporters were less likely to respond. At the very least, this is very different than it was ahead of the 2020 election, when Democrats responded to the Times/Siena poll in far greater numbers than Republicans.
It’s also worth noting that, historically, Pennsylvania hasn’t been an especially bad state for nonresponse bias. It hasn’t been immune, of course; the polls underestimated Mr. Trump here in both 2016 and 2020. But by most measures, the polls were more accurate in Pennsylvania than in the other relatively white Northern states. So while nonresponse bias can certainly explain a good result for Ms. Harris, it is less obvious that it explains why Pennsylvania looks decent for her compared with Times/Siena results like Trump +17 in Montana, Biden +2 in Wisconsin, Trump +6 in Arizona and Trump +13 in Florida.
More on recalled vote
There is, however, one warning sign in the Pennsylvania data — and it’s one we’ve been talking about a lot lately: “recalled 2020 vote.”
As regular readers may know by now, recall vote is a measure of how respondents say they voted in the last presidential election. Some pollsters weight their polls using recalled vote, essentially adjusting the number of Biden ’20 or Trump ’20 voters in their poll to match the outcome of that election.
Historically, weighting polls by recalled vote increases support for the party that lost the last election (because, among other reasons, people are more likely to say they voted for the winner). But there’s a case the measure is improving, and pollsters have been using it more, particularly as a way to make sure their polls won’t underestimate Mr. Trump yet again.
By recall vote, the Times/Inquirer/Siena poll in Pennsylvania is Biden +10, even though Mr. Biden actually won the state by just one point in 2020. While recall vote may be inaccurate, this is out of line with our other Times/Siena results. Our national poll, for instance, showed Biden +5 on recall vote (actual result, Biden +4.5).
Mr. Trump would lead the Pennsylvania poll if it were weighted on recall vote. And notably, the previously mentioned polls showing good results for Ms. Harris in Pennsylvania — the Quinnipiac poll, and the Muhlenberg and Susquehanna district polls — aren’t weighted on recall vote, either. And, like Times/Inquirer/Siena, the Muhlenberg poll of the Seventh District asked respondents about their recalled vote and found Mr. Biden with a wider lead on recall vote than the actual result (seven points, compared with 0.6 points in the district in 2020).
As I wrote last weekend, the recall vote measure has been extremely inaccurate in the past — so inaccurate that it would have made the polls less accurate in every election in recent memory. But a majority of pollsters are now using it anyway, and it’s entirely possible that it will make their polls look more like the final result in the end. After all, the apparent degree of error on recall vote varies greatly from election to election, and this election contains a variety of twists — namely, that the loser of the last election is running again (and doesn’t concede defeat) and the winner isn’t running again (and in a sense lost a rematch against the loser). Maybe, this time, it will do the trick.
If we had used recall-vote weighting this cycle, the Times/Siena polls would tell a very different story.
How recent Times/Siena polls would have changed
National: Harris+4 —> Harris+4
Wisconsin: Harris+2 —> Trump+1
Michigan: Harris+1 —> Trump+1
Pennsylvania: Harris+4 —> Trump+2
Texas: Trump+6 — > Trump+3
Arizona: Trump+6 —> Trump+6
Georgia: Trump+4 —> Trump+6
North Carolina: Trump+3 —> Trump+6
Florida: Trump+13 —> Trump+7
If the final election results look like these recalled-vote-weighted estimates, it would be something of a nightmare for Democrats. It would also be a world where the Democrats made a big miscalculation by not going harder for victory in Texas. Ted Cruz would be hanging on by a thread in his Senate race if these recall-weighted results were right, even as other Democrats fell behind (including Bob Casey in Pennsylvania, who leads by five points in today’s poll).
There’s one other thing that’s interesting about these numbers for weighted past vote: If they were the final election results, it would mean a pretty good year for Times/Siena accuracy. The average error across the reported Times/Siena polls would be 2.7 points — well below the long-term average (4.3 points) for presidential election polling. The polls would have understated Mr. Trump by only one point on average — also better than usual, and far better than in 2016 or 2020.
Of course, “good” by statistical measures is not necessarily “good enough” for the purpose at hand. The recall-weight-voted polls tell a very different story about this election. Ms. Harris wins in one; in the other, she loses. But this is mostly because the election is so close: With the race essentially tied, even a modestly below-average amount of error can yield an enormously different outcome.
We will find out which set of results comes closer in November. Either way, the significant effect of recall-vote weighting — and the decision of pollsters to use it — illustrates the uncertainty that’s always inherent in polling and ultimately in our understanding of the election with less than four weeks to go.
The post A Mystery Repeats: Harris Up 4 in Pennsylvania, and Trump Up 6 in Arizona appeared first on New York Times.