Harris is gaining in post-debate polls

11 quick thoughts on the state of the race

Hello readers,

I’m going to employ a little trick that I use from time to time when I’m really crunched for time: writing up some observations as a series of bullet points. But the headline is that the bold prediction I made after last week’s debate looks to be coming true: Kamala Harris is moving up in the polls enough that the model is converging back toward 50/50 in the Electoral College.

1. Harris is pretty clearly getting a bounce in national polls. She’s now up by 2.9 points in our polling average, as compared with 2.2 points on the day of the debate (and 2.0 points on the day after the debate, when there wasn’t yet any post-debate polling included in the averages).

2. Arguably — arguably! — this is even a little conservative. Among the most recent polls that conducted some interviewing on Sunday or Monday, Harris has a larger lead: 4.6 points. So it’s possible that Harris is benefiting not just from the debate but also from the favorable news coverage that it’s brought her. However, these are a favorable set of polls for Harris (they have Democratic-leaning house effects), and the model is generally pretty smart about how to weigh recency versus other factors, so I wouldn’t get carried away with this.

3. It’s also always worth keeping in mind that the election isn’t being held today. The headline of the weekend was another assassination attempt against Donald Trump. This one didn’t get very far, fortunately. I don’t want to speculate about its impact on the polls, if any, but the news cycle turns over quickly. And the fundamentals of the race still probably favor Trump.

4. We’re really lacking for high-quality state polls, and to the extent there have been some, they’ve been more mixed for Harris. She got a high-quality poll showing her ahead by 3 points in Pennsylvania, for instance, but another one showing her behind 2 points there. Although the model somewhat tries to anticipate movement in state polls based on its trendline adjustment, it can be slightly conservative about this. State polls are really what matter the most to the model in the end.

5. Because there’s not a lot of recent state data, the model is still relying heavily on pre-debate state polls, in some cases including polls that were conducted shortly after the DNC. In our forecast, these polls were subject to our much-disputed convention bounce adjustment, so this is still having some effect on our Nov. 5 projection until more recent data replaces them.

6. It’s also worth remembering that the cohort of polls Harris was getting just before the debate were generally pretty mediocre for her. Trump was gaining on her in most swing states even without applying the convention bounce adjustment, and high-quality national polls like NYT/Siena showed a tied national race, which implied that Harris was an underdog in the Electoral College. This data is still working its way out of the system, too.

7. There’s now almost a 25 percent chance that Harris wins the popular vote while losing the Electoral College (and only a 0.2 percent chance of the other way around). This gap has continued to grow. And it can make poll-reading really counterintuitive. You’ll see lots of headlines saying that Harris is leading — but our elections aren’t determined by the popular vote. In 2016, Hillary Clinton won the popular vote by 2.1 points and lost the Electoral College. In 2020, Joe Biden won by nearly 4.5 points but barely emerged victorious in tipping-point states like Wisconsin. So if Harris is ahead by 3-ish points — in between Clinton and Biden — you can see why the model thinks of the race as a tossup.

8. It’s very easy to cherry-pick data without really meaning to — if you’re a Democrat, for instance, by ignoring a high-quality poll from AtlasIntel that showed Trump ahead nationally. It’s important to include even apparent outlier polls in the average; that’s what averages are for.

9. In fact, after Labor Day, many political partisans totally lose their minds, especially on Twitter and other social media platforms. So, it’s even more important to trust the process.

10. One quick non-polling comment: I’d be wary of any conclusion that Trump and JD Vance passing along conspiratorial claims about Haitian immigrants is some sort of savvy political strategy, such as by putting immigration back in the news. Trump and Vance have made a lot of mistakes in the campaign, and Trump often doesn't react well when he’s losing ground in the polls.

11. Silver Bulletin is now including polls like this one from Monmouth University in our polling averages and forecasts. For the past couple of election cycles, Monmouth has been doing something rather strange. For instance, they’ll ask whether people are considering voting for Harris, and then ask a second question about whether people are considering voting for Trump. Our guess is that they’re doing this to duck accountability and avoid compromising their strong pollster rating, since by some extremely literal-minded standard, they aren’t technically asking which candidate voters would prefer. But given that very few Americans are considering voting for both Harris and Trump, that the topline numbers resemble horse race polls (in their recent survey, 48 percent of voters are considering Harris and 43 percent are considering Trump — the +5 margin is similar to other recent national polls) and that we don’t want to incentivize Monmouth’s manipulative strategy, we’re now treating these as though they’re horse race polls, and they’ll affect Monmouth’s pollster rating going forward.