Split Ticket

Political analyst Harrison Lavelle ’25 makes sense of predictions in an election that was “very, very close.”

Words by Kara Pothier

Harrison Lavelle '25 / Photo credit: Michelle Gustafson

With a newly inaugurated president, it’s easy for most Americans to make the 2024 campaign season — with its relentless horse-race polling and often biased chatter — a distant memory. But for Harrison Lavelle ’25, the poll numbers, and politics in general, are not something he views casually.

Lavelle, a political science and international studies major, is a founder of Split Ticket, an election analysis website that he started in 2021. With a team of five, all under the age of 30 and several still in college, Split Ticket has some of the youngest analysts in the field, and their 2024 forecasts were among the most accurate. The political world took note as Split Ticket prediction models appeared in media outlets such as Politico, The New York Times, and the BBC. After the election, they contracted with The Washington Post to write two articles per month and to continue their forecasts for the 2025 governor races in Virginia and New Jersey, and then the 2026 midterms.

We talked to Lavelle about what it is like to be in the political hot seat.

TCNJ Magazine: Why did you start Split Ticket?

Harrison Lavelle: I followed politics closely starting in 2016. I met some friends on Twitter because I was part of a thriving election community there. We noticed that the polling was off that year; there was an underreporting of the vote for Donald Trump. We began to map and model elections ourselves, and after the 2020 election we started our own website for election analysis.

TM: How is Split Ticket different from other sites?

HL: Our main goal from the beginning was to give people the data with nonpartisan analysis. We felt too many personal views about individual races were coming into the polling coverage. We wanted to give better insight free from pundit narratives, and we hope our quantitative, statistical models will better predict election outcomes.

TM: Explain what you mean by models.

HL: You can think of a model as a program that runs a number of simulations, and the aggregate of those simulations is more or less the prediction we get. Then we create graphic interactives on our website to help people visualize the numbers. In 2024, we had a model for the Senate, House, and presidential races to show how likely it would be for a candidate to win. And for the House and the Senate we also gave the percent chance a party had of taking over each chamber.

TM: Do you have a particular expertise when it comes to the analysis?

HL: I bring a lot of knowledge regarding political geography to the scene. Especially when it comes to the House model, it requires knowledge of individual districts and dynamics, and so it is nice to have personal knowledge to check the results from models against.

TM: The polls showed the 2024 presidential election to be pretty much a dead heat, but there were differences in predictions from poll to poll. Why is that?

HL: The bulk of the polling suggested the race would come down to the wire, and we were no different. Each model has its own methodology, and there are some differences between them, and those can produce slightly different outcomes. We pull each poll that 538 [an established public opinion polling website] collects, but then we drill down deeper. Our aggregates control for poll age, pollster quality, whether the poll is of likely or registered voters, sample size, and whether a poll is from a partisan source or not. So for example, we give more weight to the most recent polls and to nonpartisan polls. And sometimes we do our own polls, too.

TM: Are we in unusual times that everything is so close?

HL: I would say that this was really the closest election from a polling and prediction standpoint since the Barack Obama and Mitt Romney campaign in 2012. In 2016, Hillary Clinton was actually strongly favored to win pretty comfortably according to all outlets, and that ended up being wrong. And in 2020, Joe Biden was expected to win by a lot more than he actually won by. I think pollsters have tried to account for the failings, if I can call them that, so we have seen changes to sampling methods.

TM: Was there any outcome that surprised you in 2024?

HL: One thing that was interesting was the popular vote win. Donald Trump was the first Republican to win the popular vote in 20 years. But for the most part, nothing was too surprising. Even with our prediction that Kamala Harris had a slightly better chance to win, there was still a decent probability in our model that Trump would sweep all of the swing states, which he did. Overall, our forecast ended up being one of the most accurate of the cycle. There were 525 federal elections in the United States this year. Of those, our forecast called all but 12 correctly, for a 97% accuracy rate.

TM: Do you think the public has that same impression or do you think they came away feeling like the pollsters got it wrong again?

HL: A lot of people read polls as an indicator of certainty, which is not what they’re supposed to do. The polls are just meant to show you the range of possible outcomes. But I do think the narrative on polling needs to change, because if polls show a toss-up, it’s a toss-up, right? People don’t like it because they want a definitive answer. But it’s kind of disingenuous to give you a definitive answer because we really don’t know.

TM: Do you think polling is still relevant even in our polarized political climate?

HL: Oh, yeah, because polling is still the best thing that we do have to predict outcomes.