How do news outlets know when a race can be called on election night? How confident are they that they’re right when they do? How much can be predicted ahead of time? Luckily, Zachary Donnini, Head of Data Science at VoteHub, is joining us to answer all those questions — as well as whether there was anything unusual about California’s primary election results, and how the electoral landscape is shaping up for Democrats in 2026.
Together with Senior Editor Tim Dickinson, Donnini explains how he and his team prepare ahead of election night, what kind of data goes into their models, and what kind of political environment Democrats will have to traverse come November.
Make sure to keep up with Donnini on his Substack, Data and Divergence.
Zachary Donnini is a Data Science master’s student at Harvard University and Head of Data Science at VoteHub. He previously worked in Data Science and Race Call at Decision Desk HQ and graduated from Yale University with majors in Statistics & Data Science and Economics.
Join the fight now and get 25% off an annual subscription
Contrarians in Court bring hundreds of cases and legal matters against Trump’s regime — by joining our community today, YOU help STOP the chaos!
The following transcript has been edited for formatting purposes.
Tim Dickinson
Hey, this is Tim Dickinson for The Contrarian. We’re joined today by Zachary Donnini, head of data science for VoteHub. Zachary, how you doing today?
Zachary Donnini
I’m pretty good, thank you very much for having me on, I appreciate it.
Tim Dickinson
The California, primary results have been in the news this week, along with a lot of hubbub about the way California counts, and whether there was some regularities. Based on the science, did you see anything at all irregular in what was going on in California?
Zachary Donnini
Yeah, so, before election night, we are preparing for weeks. for election night, and even the days after. So, you know, when we’re preparing, the number one question in our head is, how are the final results going to differ from what we see on election night? So, we get a lot of data. We’ve recalled all 59 counties in California in the week before election night. We got files sent to us by the Secretary of State of California. And when we’re combing through that data. we had, you know, very, you know, strong beliefs that the later ballots were going to be more Democratic and younger, and then we saw that happen. So for us, it wasn’t too much of a surprise.
Tim Dickinson
So this sort of initial red wave dissipating into a blue, cool, results. It was totally expected by you and anticipated by your data.
Zachary Donnini
Exactly. I think, you know, there’s obviously a lot of uncertainty, right? It’s not like we knew what was going to happen exactly in the week after the election. I think I went on a podcast on Wednesday morning, so right after election night, and I said I thought the LA mayor race was going to be pretty close, but I described it as leaning towards Spencer Pratt. Right? So that turned out to be wrong. The shift was a bit bigger than we thought, and Nithya Raman ended up, you know, winning relatively easily in the end. But it’s not like we thought that outcome was close to possible, or even very unlikely, right? It was, you know, just, like, a small surprise.
Tim Dickinson
Any other thoughts about the California election? Either, like, variance from the data, or what to expect going forward?
Zachary Donnini
So, I think it’s difficult, for us, right, when, you know, to kind of do our job and be able to communicate, you know, with the public when so many votes are outstanding after.
So, like, I love New Jersey as an example. New Jersey counts about 92% of its ballots, usually on election night. It’s very easy for us to say, hey, this race is super close, it’s within a percent, let’s wait a little while, right? And people just really intuitively understand that. It’s a little harder when you count about half your votes on election night, and we have to say, oh, you know, like, someone might be up by double digits Wednesday morning that they could lose, you know. It’s much more difficult for people to understand.
Tim Dickinson
I wanted to talk about the midterms, and you guys, your forecast right now is very favorable for the Democrats in the House, and sort of a toss-up. In the Senate. Tell us about your model and where you see the lay of the land as we speak today.
Zachary Donnini
Yeah, so I think the number one most important thing in a, you know, forecast model is just one number, which is we are trying to predict what we call the national political environment in November. And by national political environment, we just kind of mean, like, if generic Democrat and generic Republican are on the ballot, and people go to the polls, you know, what are they feeling, right? Who do they pick? So the key, you know, input into that prediction right now is the generic ballot polling average. So right now, our generic ballot polling average, in which pollsters literally ask. Generic Democrat, generic Republican for Congress, Who are you voting for? Right now, it’s D plus 5.4, which is, you know, quite good for Democrats. It would be their best year since, you know, 2018, nationally. But, if anything, we expect that to shift more towards Democrats throughout the year, because in past midterm cycles, the out party tends to gain more throughout the year, and do better as it goes on. As I like to say, you know, waves can sometimes take a long time to break.
Tim Dickinson
And I don’t think anybody is surprised to hear that Democrats are sort of generally favored in the House. What’s the Senate landscape? It seems sort of, right now, kind of on a nice edge, and I’m wondering what you’re seeing and where the key states are that are going to help make that difference.
Zachary Donnini
So the Senate landscape is incredibly chaotic, right? There are four states, Georgia, North Carolina, Maine, and Michigan, that Democrats just really need to win if they’re gonna flip the chamber. They need to win those four states. Those are all states that either Trump carried by narrow margins, or in the case of Maine, Harris won in 2024. But it’s not clear that that’s going to be easy. Susan Collins is a very strong performer in Maine, and there are some questions around Platner. In Michigan, you know, the Democratic primary is very chaotic. Then beyond that, there are these states, Alaska, Iowa, Ohio, Texas, that Trump won by double digits in 2024. And Democrats need to win a couple of those as well, to be able to flip the chamber, which is gonna be really hard.
Tim Dickinson
I saw that you guys have some prediction market modeling in your data set. I wonder if you could talk to that, and whether, you know, because we always, particularly at this stage in an election, we have numbers that we can go by, polling that we can go by. And then there’s sort of the vibes question, right? Like, Talarico seems to have the vibes, and Platner may be not, right, in Maine. And so help us understand how those prediction markets now factor into your modeling.
Zachary Donnini
So, like, what I like to, you know, tell people is this is like a quantitative forecast, right? So let’s assume something, you know, that may have happened in a recent election, you know, happens, where someone has a terrible debate performance, such as Joe Biden in 2020 before, right? Just a really bad debate performance, a scandal. There’s no input into our forecast that’s some, you know, binary option. Like, it’s like, like. cheated on wife in Croatia while on a congressional trip, right? Or something like that. Or convicted of insider training, right? We don’t have, you know, a binary variable for those types of things to put into our forecast, because it just wouldn’t make sense. We can’t train on it. But, that is something that goes into the conventional wisdom, and we are very careful with how we integrate prediction markets. We don’t just, like, plop the odds you see on, you know, a call sheet, like, into the forecast. We look at the orders and the activity, and the order book and stuff. But, historically, prediction markets have been, you know, pretty good at predicting outcomes of elections, and we believe, you know, they summarize the conventional wisdom in a, you know, meaningful way, which is predictive.
Tim Dickinson:
So it just adds a different kind of input that’s a little bit qualitative, but people are actually putting some money and data behind it.
Zachary Donnini
Exactly, exactly.
Tim Dickinson
Of those Senate races, which ones in your model… is there variability there? Are you seeing ones that, you know, if you had to guess, it’s going to come down to X, Y, and Z?
Zachary Donnini
Yeah, I think, one of the, like, more interesting, you know, Senate races that, you know, everyone is talking about is Texas, right? Where, there’s a lot of questions around both nominees. So, for instance, you know, the models, you know, the model just sees James Tallarico as a strong fundraiser with limited electoral experience and, you know, an average track record, right? And then it sees Ken Paxton as, you know, someone who has experience, but has been an electoral underperformer in the past, who’s not a strong fundraiser, right? So, you know, there’s… people just have, like, different gut feelings about which candidates are strong, and that comes through in polling. But I think a big question is, you know, what that race looks like in polling in September. You know, we try to do our best to predict you know, candidate strength through things like experience, electoral track record, fundraising. But it’s really hard to do, right? These models, you know, aren’t always on point, and where polling will be in, you know, October will tell us a lot.
Tim Dickinson
Right. Can you let us under the hood a little bit about… I mean these models are very sophisticated, and so when you were talking about how you guys, you know, talk to the counties in California and pick up, I guess, latest Fed registration data? What are you looking for? So how do you corral all of these inputs, and how much of it is just the computers going out and pulling stuff, and how much is it actual, like, human labor trying to gather extra intelligence?
Zachary Donnini
So, The core of our prediction models are something we call the fundamentals, which is just kind of like facts about a district. So it’s like, how did this district vote in recent presidential elections, congressional elections? Who’s the incumbent? You know, what is the demographic makeup of this district? Or also just, you know, some stuff to tell the model what we’re dealing with. Like, is it a midterm year or a presidential year, right? And we collect all of that data from, like. About a decade of past races. And then we train, like, statistical machine learning models on those past races. And we use those past races to try to predict the future after we collect the, you know, the same data for future races. So yeah, as the interns who work with me, and myself as well know, there’s a lot of data collection and, you know, quality, you know, assurance and checking that goes into that.
Tim Dickinson
You used to work at Decision Desk, and I used to work at Rolling Stone, and we would always, sort of use you kind of as, like, an early alert system. You guys always seem to be calling races maybe 20 minutes before other people did, and, you know, we always had to go off AP or NBC or whatever. I’m just curious, but—and I don’t recall a case where you were wrong, and so I wondered… I was always curious—when you guys were able to predict the outcomes of races a little quicker than your competition in the old… wearing your old hat. What was that about? What was the juice there?
Zachary Donnini
Yeah, so I’ll also say we have a race called us at VoteHub. We still, you know, we’ve never gotten a race wrong, which is, you know, very exciting. We’ve been calling races since last, you know, December. And I think it’s, there’s a couple things. So, first of all, you know, at VoteHub, we prioritize accuracy above all else. We do not want to miss any race calls, ever, because we believe it’s harmful publicly. And, you know, sometimes that means, you know, waiting a long time if we believe, you know, things are… things are unclear or something.
But in the sense of, you know, maybe getting an edge and be able to call things, you know, accurately, with certainty before other people, I think it comes down to a lot of, you know. research before. And to be clear, you know, like, NBC, you know, CNN, the AP, you know, all these race call desks are really good at their job. NBC and the AP, you know, regularly call races with less than half of the votes reporting, and they haven’t made a mistake in a federal or statewide race for 8 years. So, you know, I’m not here to diss anyone else at all. Everyone in the industry has been doing a fantastic job. In the last, you know, 6 to 8 years. But, A lot of, you know, an edge might come from research, right? So knowing, you know, getting data from counties about, you know, how many people are going to turn out, or something like that, or what the remaining ballots look like. So at VoteHub, I believe we were first to call for Nithya Rahman over Spencer Pratt. On Sunday night. And we were able to do that even though Raman had this tiny lead, because we knew what types of ballots were remaining. We knew they were super young and super democratic, and we had seen them breaking for Raman in large numbers, so we were able to, you know, know that will continue because of the ballots remaining.
Tim Dickinson
Let’s dig a little bit into the… and under the hood and the house. I mean, there’s so many races that we could look at. Is it a certain kind of race, or are there, Which are the… which, like, what kind of granularity does your data give you that you’re saying you can sort of, like, tease out some trends that are going to be important in the house?
Zachary Donnini
Yeah. So, something we’re always watching in the House is something, like, we call, like, overperformance, or candidate quality, or candidate strength, or stuff like that. So, in 2024, and I don’t quite remember the exact numbers, but there were around 20 seats where Democrats won, a House seat Trump won, and there were only, I think, 3 where Republicans won a House seat that Harris won. So, basically, Democratic House candidates in these competitive districts were punching above their weight. They were able to win seats that Trump won. And this should be concerning for Republicans, because as we said, you know, the environment looks like it’s going to be much bluer in 2026, and if Democrats are winning a lot of seats that they shouldn’t even be winning based on that environment for the second straight cycle in a row, that’s how the Democratic majority gets up to, like, 235, or something like that.
Tim Dickinson
What is your high water mark? I mean, if there is indeed a sort of late crashing… Blue wave, what kind of slit could we be looking at?
Zachary Donnini
So, the Republicans have gained a lot in the U.S. House, map-wise in the last two years, right? So, they’ve drawn a lot of these new gerrymanders that are very effective. So, at some point, if we’re talking about a really blue year, so think something goes really, really wrong, either, you know, maybe with Iran, or, like, economically or something like that, and we get a true blue wave, you know, as bluer than 2018, a lot of those gerrymanders could break, right? So if we’re talking about, you know, a 10% chance or a 5% chance of things happening, Democrats could get really high, be, like, 250, 260, these seats. because these maps that are drawn to, you know, distribute votes efficiently for Republicans do mean there’s a lot of seats that Trump won by about 10 to 15 that, you know, could all flip blue at once, and then the dam breaks. But again, you know, right, high water, we’re talking about, you know, 5% chance, 10% chance. It’s very unlikely. You know, it’s the same chance the U.S. makes the semifinals in the World Cup or something, right?
Tim Dickinson
Right, well, we can, we can, we can hope on many fronts, can’t we? Zachary, are there any other, thoughts, or just sort of, sort of behind-the-scenes stuff at VoteHub that you want watchers to be aware of?
Zachary Donnini
Yeah, I think something I always say is, like, I go out, I do a lot of, like, you know, interviews, I go on TV, I go on podcasts, etc. But, you know, VoteHub has, you know, so many hardworking people who make, first of all, make all of our models run. They, you know, scrape tons of super, you know, incredible, super detailed data for us to play with. They’re why stuff looks so pretty on the front end. So I guess I’d say, you know, VoHub is, you know. We’re really trying to scale up, and we have, you know, so many, so many cooks in the kitchen right now, trying to make things, you know, run smoothly, be as pretty as possible, find as much cool data as possible for people.
Tim Dickinson
And I’ll just pay you a compliment. I’m a new user to the site, and just logged on, and it’s like, oh, this is all very intuitive and really easy to grab the information that you’re looking at. It’s like, oh, here are the chances in the Senate, and this is what the map looks like, and here’s the house. And so, just as a resource for people who are you know, treating politics a little bit like sports or the World Cup, it’s like a place to go and really just, you know, get what you’re looking for quickly.
Zachary Donnini
Yeah, exactly. Yeah, thank you very much, I appreciate that.
Tim Dickinson
Alright, well, I hope we have you back on as the midterm progresses here. It’s been nice talking to you.
Zachary Donnini
Yeah, thank you very much, thank you so much for having me on.
Tim Dickinson
Take it easy.
