“Anyone is welcome to propose a better name for synthetic past vote, especially because people seem to think it's an AI thing, which I definitely didn't anticipate. People are calling synthetic respondents AI-based, and people are confused about this. All it is is that we're correcting for a couple of things that we observe are clearly wrong in recall vote.”
GD POLITICS · Policy · November 2026
Context
Nate Cohn explains the Times/Siena weighting method called "synthetic past vote", which readers keep confusing with AI-generated survey respondents. It is only an adjustment to how people's reported past votes are used.
Transcript
…in the electorate to differ and potentially significantly from the result of the last election. And I think that's also really important. If you were waiting the Hispanic vote in Texas by recall vote to be whatever your estimate was for 2024, I think you would have been way off. And we weren't.
…when you're done waiting your respondents. Now, it sounds like as that has become increasingly a crutch for pollsters, you still, you're not, I mean, it sounded like actually in 2024, you were interested in relying a little bit on it, but it sounds like maybe you're more skeptical about that now.
So stepping back, the reason why we haven't followed the industry down this path is because when we try and use it on our election, our previous data, it doesn't help. If we, and I think that when other pollsters wait on recall vote, it does a lot more for them than it's done for us. I don't know if that's because we have higher quality samples on the front end or if it's because actually I kind of think that's what it has to be. I think other pollsters are just getting much more Democratic skewed samples on the front end. This is like the only thing that's getting them back in the ballpark and they're pulling the lever. That's just not what we see. You know, in 2020, our polls do not get better with recall vote. They get worse. It's in the APOR report. There's a whole section where they talk about an unnamed pollster. We're the pollster that is not helped by recall vote with the data that we provided to them for their reanalysis. Doesn't help us in 2022, doesn't help us in 2024. Nonetheless, though, after those three elections, we can see some patterns where, as you put it, recall vote maybe could have helped us a little bit. It could…
And so that's a change that you made between 2024 and 2026.
Yes. And I should say, by the way, I think this is really important. I don't know if there are other public pollsters who made a change after 2024 that gave them results that they really liked for 2024. I haven't reported what our results are for 2024, but I'll just say like we would have been…
Yeah, I think that's something that's so important to keep in mind. People are always like, oh, well, the polls got it wrong last time. It's like, well, a lot of pollsters, especially the most rigorous pollsters, are iterating and trying new things. Are there lessons that you've taken from the past…
More from GD POLITICS
