On Saturday, the New York Times and Siena College released their latest round of swing-state polling on the Presidential race. It showed Kamala Harris leading Donald Trump by five points in Arizona and by two points in North Carolina, while trailing Trump by a point in Nevada and by four points in Georgia. (In 2020, Joe Biden edged out Trump in all of these states except North Carolina.) The cumulative results show a very slight Harris edge. Coupled with the previous set of Times/Siena polls—which had Harris leading Trump by four points across three battleground states in the Rust Belt (Pennsylvania, Wisconsin, and Michigan)—the over-all picture of the race has transformed since President Biden stepped aside from contesting the Democratic nomination in July. Harris is narrowly ahead.
To talk about what it all means, I recently spoke by phone with Nate Cohn, the Times’ chief political analyst who also oversees the paper’s polling. (Full disclosure: Cohn and I worked together at The New Republic, and are friends.) During our conversation, which has been edited for length and clarity, we discussed the surprising ways in which Harris’s coalition appears to differ from Biden’s in 2020, how to think about the Sun Belt versus the Rust Belt, and the prospect of a third straight Presidential election with serious polling error.
You decided to do these polls in two rounds, one in the Rust Belt and then one in the Sun Belt. Why did you make that decision, and do you think it’s helpful for people who are following the election to separate those two areas in their minds?
Well, there’s one practical reason, which is that it is difficult for us to simultaneously and quickly field surveys in seven states. This is a really dynamic race, and so, if we had polled all seven of these simultaneously, we might’ve had to field over ten days, and I think there would’ve been some valid questions about whether the results that we had at the end were still reflective of the race as it is today.
There’s also a substantive reason, which is that the Sun Belt and Rust Belt states, if we can call them that, have been very different this cycle. The Northern battleground states are relatively white, and the polls this cycle have shown Democrats faring relatively well among white voters. As a consequence, even Joe Biden was fairly competitive there. The Sun Belt states, on the other hand, are relatively diverse, and the polls have shown Democrats faring relatively poorly among nonwhite voters this cycle. As a consequence, Donald Trump had a significant lead in the Sun Belt battleground states even as the Rust Belt states remained competitive. So I think there was a pretty good reason to be treating them separately, even beyond the practical reasons.
When you’re looking at these polls, either in the three Rust Belt states [Pennsylvania, Wisconsin, and Michigan] or the four Sun Belt states [Arizona, Georgia, Nevada, and North Carolina], are you looking at the individual states very closely? Or are you grouping them together in your mind and averaging them out?
Both. Obviously individual states are idiosyncratic, and they can go in their own distinct ways. On the other hand, these two groups of states have a lot in common, and they voted fairly similarly four years ago, and, when we put them all together, we have a larger sample size. So those estimates can be more stable, while the individual states can bounce around. When you’re looking at a single poll, it can be a little challenging to figure out whether you’re looking at something that’s genuinely unique about that state or a thing that’s weird about a smaller sample.
If you take the four Sun Belt states together, you get Kamala Harris ahead by four-tenths of a point. That happens to be the exact result of the 2020 election across those four states. So I see a result that in some ways is profoundly unsurprising. It’s almost exactly what we would’ve guessed two years ago, but it also represents a huge change from earlier in the year when Trump had a large lead across these diverse states.
Now that you’ve been able to do several rounds of polls, what do you feel comfortable saying demographically about how this race is different from when Biden was running, but also compared to Biden in 2020?
In May, we had Trump ahead by five points across the battleground states. Right now, we have Harris ahead by two, so it’s a big seven-point swing. And we show outsized gains for Harris among young and nonwhite voters, and women, and we show smaller gains, but still some improvement, for Harris among men and white voters.
Now, what’s interesting is that, compared to 2020, Harris is still slightly underperforming where Biden finished among young, Black, and Hispanic voters, even though she’s doing better than he was three months ago. We have Harris doing a little bit better among white voters and older voters than Biden did in 2020. It’s possible that this just reflects a continuation of a longer-term trend during the last decade toward somewhat less racial polarization. It’s possible this will change as the campaign continues. Maybe some of the white working-class voters who have seen three weeks of great coverage for Harris, and particularly white working-class women, will eventually turn against her as the campaign goes on. It’s a really unique political moment.
Right, there have been so many headlines about Harris picking up support from young voters and Black voters and Hispanic voters, but that’s just compared to how Biden was doing with those groups in the polls. And it is pretty striking when you say that, compared to the actual result of the last Presidential election, she’s actually doing better among white and older voters and worse among all those other groups.
It’s definitely not what we would’ve guessed two or three years ago. If you told me that Harris, a Black and Indian younger woman, was not going to have any material advantage over Biden among Black and Hispanic voters in the 2020 election, that would’ve surprised me. I do think that we need to see this race settle out a bit before we can say why that’s true and whether it’s going to last. I really do think it’s conceivable that Harris is riding an extraordinary wave of momentum that inflates her numbers just a little bit among swing voters across the board, and that might come back to earth. And ultimately there may be some room for her to make additional gains among Black and Latino voters as the campaign goes on.
Your current poll has largish gaps between Arizona and Nevada and then also between North Carolina and Georgia. I think most people assume that those gaps will end up being pretty close on Election Day.
Well, I certainly share that expectation. Arizona and Nevada have a large Latino population. Georgia and North Carolina have large Black populations. So you would expect those two groups of states to move somewhat in tandem. But it’s worth noting that that’s not necessarily what has been happening over the last decade. Democrats have made big gains in Arizona and Georgia while they have made few gains in Nevada and North Carolina. So it is conceivable that these states can move independently. Still, I don’t think our polling should necessarily be taken as a definitive claim that North Carolina is now going to vote to the left of Georgia or Arizona will vote far to the left of Nevada, not by any stretch.
We have the most to say about Georgia and Arizona because we’ve done a lot of polls in those states by now. The previous times that we’ve polled Arizona and Georgia this cycle, they weren’t that much different from the battlegrounds as a whole. But in this poll both those states diverged from where they’ve been in our prior polling. Georgia is now well to the right of the other states. Arizona is now to the left. My instinct is that this is probably just a little bit of random noise. I think that if we polled those two states again tomorrow, there’s a pretty darn good chance we would find Arizona closer to even, and Georgia closer to even.
This is the first time we polled North Carolina this cycle, so we don’t have context to see where Times/Siena usually gets this state with respect to the others. And Nevada is weird for a different reason, which is that our polling to this point has shown Trump faring exceptionally well. I’ve talked to a lot of pollsters who have at prior points in this cycle shown Democrats faring better than we did, and I’ve also talked to some who showed Democrats faring just as bad as we showed them doing in Nevada. Nailing down exactly what’s going on in Nevada has been challenging.
To go back to what you said earlier about Harris still lagging a bit among Black voters, Georgia has a larger share of Black voters than North Carolina does, and so, if there’s been some Democratic slippage in one of those two states, you might expect to see it in Georgia.
That’s reasonable. I don’t think it covers the six-point gap that we find between North Carolina and Georgia. But if in November North Carolina actually does vote a little bit to the left of Georgia, I wouldn’t be surprised by that at all. And not simply because of that demographic argument you made but just because states are a little idiosyncratic and weird things happen. Four years ago, people didn’t expect Georgia was going to vote to the left of North Carolina. It did. We can go through all kinds of examples of cases where states move in ways that are just difficult to explain.
And North Carolina was to the left of Georgia in ’08, ’12, and ’16, I believe.
That’s right. And there are other cases like this. If you put us in a time machine and go back to 2014, I don’t know how you explain how Florida and Arizona have taken such divergent paths. It’s easy to say now, but demographically there’s no great explanation for why these two states full of Latino voters and retirees have turned in opposite directions.
Your poll and some other polls have shown an interesting gap in some of the swing-state Senate races. Basically, most of them show incumbent Democrats, or non-incumbent Democrats like Ruben Gallego, of Arizona, performing better than Harris is. Who are these voters who are going to vote for Trump and for the Democratic candidates for Senate? What can you tell us about that?
It’s a different story from state to state. If I could step out and make a broad statement, I would say that the Democratic candidates tend to fare a little bit better than Harris among white voters, and then they are achieving the usual Democratic strength among young and nonwhite voters. That is interesting to me because it suggests that some of the strength that Trump has built among young and nonwhite voters really does represent his own strength and not necessarily strength that carries over to other Republicans.
We need to wait a little longer to see all this stuff sort out. These Senate campaigns are not active. And we saw in 2022, you may recall, that Mark Kelly started out with a double-digit lead, and John Fetterman started out with a double-digit lead, and the Democrat in Wisconsin Mandela Barnes started out with a significant lead over Ron Johnson. This was before the campaigns really got under way and before the Republicans were able to run attack ads tying Democrats to liberal policies. And I’m not saying that’s what’s going to happen this time, but it is a reminder that these races break later than Presidential contests do.
My assumption has been that Trump would fare a little bit better among nonwhite voters than these generic Republican Senate candidates, and at the same time that some of the Republican Senate candidates would poll a little better with educated white voters than Trump does. But you’re saying there’s not some consistent story like that?
I wouldn’t say so, but that may be how it shakes out in November. I really do think that Senate races develop on a different time line. And, in some of these states, the Republican nominee was only decided recently.
Looking at polling data, it seems that the gap between the Electoral College and the popular vote in 2024 is going to be quite a bit smaller than the more than four percentage points it was in 2020. Do you feel confident making any claims about this? There’s been so few good state polls that it’s just very hard to tell.
I think I felt pretty confident that the gap between the Electoral College and the popular vote in a Biden-Trump matchup was poised to be much smaller than it was four years ago. That was supported by a full year’s worth of state polls, and it was supported by the basic demographic patterns that we saw in both national and state polls. The Harris-Trump matchup is obviously brand new. There isn’t a robust set of polling. We haven’t even done a Times/Siena national poll that shows Harris in the lead yet. I expect that if we did one, by the way, that we would find Harris in the lead, given how well she’s doing in the swing states, but we haven’t done one. So it’s really very early.
Although her relative strength with white voters for a Democrat augurs well because of the bias of the Electoral College, correct?
That is right. If I had to guess, I would certainly guess that Donald Trump’s Electoral College edge with respect to the popular vote is poised to shrink compared to 2020. How much? I think it’s way too soon to say anything more than that.
How are you feeling about polling error these days?
Oh, God.
I know that no pollster can put polling error out of their head. Obviously there was polling error in 2016 and in 2020, in which it favored Biden, which meant that Trump outperformed the polls. What are you looking for under the hood to detect possible error?
There’s no way to anticipate whether the polls are going to be right or wrong ahead of the election. In 2022, there were a lot of reasons to wonder whether the polls were going to be accurate. Our polling showed Democrats doing very well in many of the same states we’re talking about here, like Pennsylvania and Arizona. It was hard to know whether that was real or not. It was. Conversely, in 2020, we showed Democrats doing really well and we had every reason to think it was real at the time. It turned out not to be.
Looking back, you could point to a few signs that maybe indicated that the polls were poised to be off. One of those signs was that Democratic voters were much likelier to respond to polls in 2020 than Republican voters. And that’s based on their party registration, something that we know before we even place a telephone call to them. We use that to make sure we have the right number of Democrats and Republicans, but if more Democrats are responding to polls than Republicans, I think it raises the question about whether simply adjusting the number of Democrats and Republicans is enough. It may be that you’re getting too many hard-core Democrats while the Republicans you are getting include too many Never Trumpers, and so on. That’s certainly what happened in 2020.
So far this cycle, we don’t see that same pattern. The proportion of Democrats and Republicans who respond to our polls is roughly even. I think that’s a somewhat positive sign, but I find it hard to say that that rules out any systematic problem in the polls. It’s worth noting that non-response bias is not the only reason the polls could be wrong. Right now, Harris has had three exceptional weeks of media coverage. Even if you think our polls are too blue for her right now, that may simply be because voters right now haven’t heard the case against her, and, if the election were held tomorrow, maybe a bunch of Republican-leaning, white, working-class voters would shift from the answer they gave us in the survey.
But the non-response-bias question is, in terms of your own polls, the thing that you would look at the most closely just to scan for obvious error?
This is a complicated thing to get right. The thing that was so challenging about non-response bias in 2020 was that, even if a poll had the right number of registered Democrats and the right number of registered Republicans, even if it had the right number of college graduates or people without a college degree, even if it had the right number of old and young people, the right number of white and Black people, it still was wrong because, within every one of those demographic groups, there were too many people who backed Biden and not enough who backed Trump. That is a form of non-response bias that was basically unobserved. There was no indication in the numbers that we had too few Trump voters. By its nature, that is extremely difficult to diagnose in advance.
The reason I look at the response bias by party is because if we stipulate that there are too many Biden voters responding to polls, I wonder whether that ought to increase the response rate among registered Democrats, as most registered Democrats will vote for Biden. And that therefore, if you have a hugely lopsided propensity for Democrats to respond compared to Republicans, what actually could be happening is that Biden voters, regardless of party, are too likely to respond to surveys. And because we can’t control the number of Biden voters in our poll, we might still be prone to have an inaccurate survey result, even if we have the right number of Democrats and Republicans.
Having established that context, the data that we’re looking at is the proportion of Democrats and Republicans who respond to polls, and in 2020 it was lopsided toward Democrats. This time it’s fairly even—so I’m cautiously optimistic that this means that we don’t have a deep, hidden non-response bias. It’s hard to imagine how Biden or Harris voters can be responding in extraordinary numbers compared to Trump voters but not driving the Democratic response rate up compared to the Republican response rate.
I will say that when I saw in your current poll that the state that was the best for Trump was Georgia, which I believe had almost no polling error last time, I did wince a tiny bit.
Yeah, it’s a fair point. It’s funny because in a lot of ways, compared to our other polls, it looks like the outlier. And so you might be inclined to assume that that’s the one that’s wrong and that our other polls are right. But you’re absolutely right that it could ultimately work the other way. ♦
