The Times’ Nate Cohn on Elizabeth Warren’s Odds Against Donald Trump
On Monday, the New York Times and Siena College released a poll of how Donald Trump is faring against three leading Democratic Presidential opponents—Joe Biden, Bernie Sanders, and Elizabeth Warren—in six critical swing states, all of which Trump won in 2016. The results contain bad news for Warren, despite her strong showing with Democratic-primary voters in Iowa; against President Trump, she performs worse than Biden or Sanders, with Trump leading or tied in five of six swing states. Biden leads or is tied with Trump in five of the six states, while the Times/Siena poll shows Trump and Sanders running essentially even. The Times’ Nate Cohn, who oversaw the poll, wrote of Warren, “not only does she underperform her rivals, but the poll also suggests that the race could be close enough for the difference to be decisive.” Cohn also noted that the poll suggests that Trump may have a greater advantage in the Electoral College in 2020 than he did in 2016, suggesting that the President could again win the election while losing the popular vote.
For further insight into the results, I recently spoke by phone with Cohn. (Cohn and I worked together at The New Republic and remain friends.) During our conversation, which has been edited for length and clarity, we discussed why some Biden voters dislike Warren, the odds of a Democratic upset in Texas, and whether other pollsters are making the same mistakes they did in 2016.
What is your biggest takeaway from all of the data you collected?
The President is better positioned in the battleground states than he is nationwide, and remains really competitive in the states likeliest to decide the election, despite all that’s happened to him.
What would you imagine the national numbers are if these are the numbers in the swing states you polled?
Over the summer, we did an analysis of what we thought the President’s approval rating was in every state. And in these states we thought the President’s approval rating was five points better than nationwide. So, if Biden is up two in these states, I would say maybe that he is up seven nationwide. That would make Warren up three nationally.
Michigan and Pennsylvania surprised me a little bit. You have Warren losing by six in Michigan, and Biden tied. Biden is up three in Pennsylvania, while Warren is tied. Why might they be as close or closer than in Wisconsin, which you and others have long talked about as the most likely key swing state?
The first thing I would note is that individual state polls are going to be fairly noisy. And, in general, I would suggest that people take the data collectively rather than focus on the exact order of the states. That said, I could conjure up an argument about how Pennsylvania is a state where Elizabeth Warren’s style of reformist, idealistic progressivism is not likely to play as well as it would in Wisconsin. Pennsylvania is a state that has a tradition of being friendly to establishment Democrats in the primary, like Hillary Clinton in 2008 and 2016, while Wisconsin voted for Bernie in the 2016 primary and Obama in 2008. So I could imagine something like that explaining the difference, if it is real.
And Michigan?
I think the same story could potentially hold there. The other thing I would note about Michigan that is interesting in our sample is that we think the President is doing really well among non-voters in Michigan. It is the only state where the President does better among registered voters than likely voters. And if you narrow it even further, to people who have voted before, Biden’s lead grows even more. Our sample of non-voters is not particularly big in Michigan, but it is conceivable to me that the voters who are on the sidelines of politics there may be an untapped source of strength for the President. Again, it is a single-state poll, and we are talking about a sub-sample of that poll, but that is the story in this particular survey.
Arizona surprised me, too. It’s the only state with Warren leading. Biden is up five among registered voters. Were you surprised?
I was not. Over the summer, we estimated that the President was weakest in Arizona, among all of these states. Remember, this is a poll of registered voters, and so there are a large number of Hispanic voters who are on the sideline there. I think we had Joe Biden with more than a fifty-point lead among Hispanic voters in Arizona. And so in this poll of registered voters you have a more diverse electorate than the one you have seen on Election Day, and that’s an advantage Democrats can claim that they wouldn’t get in a state like Michigan or Pennsylvania, where most of the non-voters are a bunch of white working-class voters who would be relatively favorable to the President.
Would one upshot of what you have said be that, if I told you on Election Day, 2020, that there was a massive turnout, that would imply good news for the President in a place like Michigan, and good news for the Democratic nominee in Arizona or Texas? Or is that too simplistic?
I don’t think that is too simplistic. I am not saying it is a sure thing. It would depend on who is mobilized to the polls. It is not just the opportunity but how you take advantage of it. But I think that, in general, the Democrats have this upside that they have not realized in states like Texas or Arizona, while I think there is a case that it is the reverse for Republicans in the north. Or, at least, the Democrats don’t have that kind of upside.
What do the Arizona numbers make you think about Texas? What was your guess about Trump’s approval over the summer?
When we estimated the President’s standing in Texas over the summer, we thought it was basically the same as it was in Pennsylvania or Michigan or Wisconsin among registered voters. And, again, there is a difference between the registered-voter population in Texas and the one that has ever shown up to vote before. If you look at our poll in Arizona, it is consistent with that. It has Joe Biden up by five in a state that Hillary Clinton lost by almost four. That is a nine-point swing. I would note that Hillary Clinton lost Texas by nine. I don’t think it is crazy that there could be a nine-point swing between the 2016 vote and the registered-voter population in Texas. It could even be more. Texas is a more diverse state. It’s very conceivable to me that we would have found Biden ahead in Texas if we had conducted a poll there.
So you think that, even if the registered-voter advantage wouldn’t translate into Democrats being favored there, it would be close enough that Democrats should compete there?
I would want to have actual data from Texas before making that conclusion and spending the millions of dollars that it will take to compete in Texas. But, I think, if you were to extrapolate from our data, you would think there would be a real opportunity for Democrats in Texas. And I would add that these are polls of people who are registered today, and millions of people are going to register over the next year. That would tend to exaggerate the trends that we see among new voters in this data. Texas may be a narrow Biden lead in registered voters now, but, after another year of young and nonwhite and newly eligible voters, perhaps it would be more than that.
Who are the Biden-but-not-Warren voters? How do you think about them as a group?
I think they are about six per cent of the electorate, and I think it is useful to divide them into three groups. One group thinks Warren is too far to the left ideologically. They are relatively well-educated and live in the suburbs, they consider themselves moderate, they voted for Trump or Gary Johnson, they do not like single-payer health care. That vote makes sense for me, and I think they are logical. I think it is consistent with a lot of evidence about the role of ideology in political elections.
The next group is not so obviously disposed against a left-leaning candidate but really doesn’t like Warren that much. One question we asked that may be telling about the reason they don’t like Warren is whether they thought that the women who run for President aren’t very likable. About forty per cent of the voters who support Biden but not Warren said they agreed with that statement. I think this group also holds some conservative views on cultural issues. They believe whites face about as much discrimination as nonwhite people do, or they think political correctness has gotten out of control in this country. This is a disproportionately working-class group, and it is disproportionately male. There is less evidence that gender plays a huge role in American elections. Most of that data is from congressional races. I think it is conceivable that it is a bigger factor here, or at least in this case.
Then there is a final group who just don’t know very much about her. I would think she could have success in mobilizing this group to her side over the next year. And, to be clear, she could persuade the other voters, as well, but I think she faces real obstacles now that she will have to surmount.
One of the striking findings was, as you wrote in your story, that “Nonwhite Biden supporters are likelier than white Biden voters to say they would choose Mr. Trump over Ms. Warren.” How do you understand that?
If we just step back for a second, we found across the board that Democrats were not doing as well with nonwhite voters against the President as Hillary Clinton was in similar polling in 2016. I think that, whether it is the economy, or whether it is because the nonwhite population is not as woke as one might assume, and not as offended by the President’s conduct as you might think, that the President has made some inroads—not many, but some—among voters who we wouldn’t typically think of as receptive to his cause. And this should not be totally unfamiliar for Democrats. Back in 2004, the polls said the same thing about Bush against Kerry. Granted, there was the security explanation, but this idea that an incumbent President with the advantage of incumbency and name recognition running on a culturally populist message might make a few more inroads among nonwhite voters than we might think wouldn’t be a huge surprise. The idea that there might be a group of voters who are a little Trump-curious, and Warren might not be the right candidate for them, doesn’t seem like a huge surprise.
Do you have enough data to see whether there is a giant gender gap among nonwhite voters?
We did find evidence of a large gender gap among nonwhite voters. I think Elizabeth Warren was only receiving about forty-nine per cent of nonwhite men in our surveys.
Who are the Warren-but-not-Biden voters? There are not nearly as many of them.
There aren’t as many as them, and so the sample sizes are getting really small. I didn’t see a clear message from them. They didn’t like Biden, but it wasn’t like they were a bunch of very liberal voters, either. They weren’t overwhelmingly women. It wasn’t obvious what the source of the issue was, other than I am inclined to assume that for whatever reason they don’t like Joe Biden much. There is always this sort of churn in the electorate.
What does your poll do that other polls don’t do that leads you to believe your data is worth paying attention to?
I think we do a lot of things, but I would emphasize two. One is that we are able to adjust our sample on party registration for these states. So, when you look at a poll in Arizona, that is a poll where registered Republicans outnumber Democrats by four points, just like in reality. Our poll in Pennsylvania is the opposite. That sample is forty-seven-per-cent registered Democrats and thirty-nine-per-cent registered Republicans. So you can have confidence that, on the single characteristic that is the most likely to predict whether you like the President or not, we have taken steps to make sure it is appropriately balanced. There are very few public polls that do that.
The second thing I would note is that our poll is constructed to account for the educational composition of the electorate. In 2016, many states pollsters didn’t do that, and as a result they underestimated President Trump’s support, because well-educated voters are likelier to respond to telephone surveys. That problem continues today, and a number of the state polls you see that show the Democrats posting wide leads over the President have not been adjusted to account for the reality that we live in a country where only about thirty per cent of the population has a college degree.
So you are still concerned pollsters are making the same mistake.
I am. In 2018, there were a lot of state polls that overestimated the Democrats in a lot of white-working-class states. On Election Night, it was a big surprise that Claire McCaskillwas losing when there were polls showing her up. Same thing for Joe Donnelly, in Indiana. Democrats lost governors’ races in Iowa and Ohio where they led in pre-election polls. Obviously, the polls over all weren’t all that inaccurate, but, in a lot of the same states, they were biased in the same way. And that’s when Donald Trump wasn’t the person on the ballot. He is the one who really generates the largest education split. So the idea it could be back in full force in 2020 isn’t crazy to me. I am not saying it is assured, but it strikes me as a possibility.
Some people have suggested that general-election polling this far out from an election isn’t helpful. You write in your story that, “On average over the last three cycles, head-to-head polls a year ahead of the election have been as close to the final result as those taken the day before.” Why might this be? And do you think this would hold with lesser-known candidates?
I don’t think it would hold with lesser-known candidates, or candidates who have not been attacked and who have not faced criticism from their opponents or a full-fledged campaign. There are many cases where I wouldn’t want to take a poll and results a year out as an indication of anything. But, when it comes to Trump, Warren, Sanders, and Biden, I think it’s a little different. They are well-known candidates who have been in the fray. The President has certainly been under attack. I would think a large number of voters basically know what they need to know about those folks. A lot could change, but I do think it means that it isn’t unreasonable to take a measure of where things are.
What other swing states interest you from these numbers?
I would considerGeorgia. I feel like we know that Georgia is a state where voters are deeply divided about the President. I would be surprised if we learned something like Biden up five or Trump up five. I think it could be an important part of the swing-state picture, but I am not sure I have that much uncertainty, because the state is deeply polarized along racial lines. I am curious about some of the peripheral northern battlegrounds, like Maine, New Hampshire, and Minnesota, not because I think those states are necessarily going to be top-tier battlegrounds but because it might help confirm the general pattern in the Midwest in our polling, or call it into question. And, by the way, if we are right in Pennsylvania or Michigan, then I think those states would be fairly competitive.
You also did a survey of Iowa, in which you found Warren further behind than anyone else, despite all the time she has spent in the state, and despite the fact that she led your primary poll among Democrats. What did you make of that?
The Iowa poll was done after everything else, so we had already seen Warren underperforming by about four points across the battlegrounds. And so it was: there is that again. I would say I don’t think it is particularly good news for Warren. If there was a state where she might have hoped to have made more inroads as the result of active campaigning, that would have been the state where it would happen, and it didn’t. It doesn’t rule out the possibility it will happen in the future, but it at least crosses off one potential argument on her behalf.
Your poll included a note that the Michigan data was hard to gather. What, exactly, was the problem?
Michigan has been tough for us for a while. We just have very low response rates, and a lot of numbers were people saying the person we were asking for doesn’t live there. The people apparently are pretty mean to our interviewers, too. The survey responses don’t come in as balanced as we want them to, which requires more weighting. In this case, we reduced the number of interviews compared to everywhere else. I guess my view is that I don’t think this poll is terrible or something. I think it still obeys the rules of the margin of error. But, by the measures that we use to evaluate our own polling, it’s objectively the worst of the samples. So I would say I have more uncertainty about that result than the others.