In the days after Democrats won control of Virginia’s state legislature, Charles “Chaz” Nuttycombe was focused on the results in house of delegates districts 41 and 82, both of which you’ve probably never heard of.
Neither of the competitive races would determine which party controlled the Virginia legislature, but it was one of a handful where votes were still being counted and the results too close to call. In the lead-up to election day, Nuttycombe, a 24-year-old senior at Virginia Tech, had predicted that the Republican candidates would win both. But his final forecast in Virginia gave Democrats a 61% chance of winning control of the house of delegates and a 71% chance of holding control in the state senate.
When he spoke with the Guardian the day after the election, he had already correctly predicted 100% of the results in every other Virginia state legislative race – 98 other house of delegates seats and 40 in the state senate. Eventually, both races were called for Nuttycombe, giving him a perfect forecast.
It was an astonishing feat that underscored the niche Nuttycombe has carved out predicting races at the state legislative level.
Nuttycombe runs the forecasting site cnalysis.com, and these little-known legislative races are his expertise. While the science of forecasting presidential, gubernatorial, congressional and senatorial races has exploded in recent years, Nuttycombe is one of the only forecasters focused on the 7,383 state legislative districts across the country.
His focus underscores the rising awareness of the importance of state legislatures in US politics. Long overlooked by parties and reporters, there has been a much greater understanding of the consequential power state legislatures have to set policies on issues like abortion, gun rights, education and voting. Just a handful of races in a single chamber can determine which party has control.
“Your state legislature is going to affect your day-to-day life a lot more than Congress is,” Nuttycombe said. “State legislative elections are a million times more important than congressional elections, but I’m obviously biased on that front.”
The effort can be much more difficult than forecasting a congressional race. Many of the candidates who run for the seats have no national profile. Polling, if it exists at all, is sparse. The site’s GIS team also breaks down data to figure out how state legislative districts voted in prior elections. Tracking down data from states can be a nightmare, since every state formats their information differently and some charge for it (the site also relies on precinct-level election data collected by the non-profit Voting and Election Science Team at the University of Florida).
“It’s a monster endeavor to cover legislative races in multiple states, so most analysts don’t even attempt it,” said Dave Wasserman, a well-respected forecaster and election analyst at the Cook Political Report. “Big credit to Chaz Nuttycombe for having his finger on the pulse of every race in Virginia on Tuesday. He’s a rising star in our field.”
Nuttycombe’s interest in state legislatures started in 2017, when he was starting his senior year in high school. He saw both professionals and amateurs posting their predictions on Twitter. He began offering his own, just for fun, and began doing some volunteer work with Decision Desk HQ, an online election forecasting website.
He immediately caught the attention of J Miles Coleman, who was working for the site and is now a forecaster for Sabato’s Crystal Ball at the University of Virginia. Going into the election that year, Republicans held a 66-34 advantage in the house of delegates, and the conventional wisdom was that Democrats could pick up 10 or so seats on a good night. Nuttycombe was much more bullish on their prospects and thought they had a chance to get a majority, and he was right. Republicans came away from the election with a 51-49 majority in the legislature, only winning the 51st seat after a tied race was determined by picking the winner from a hat.
“He must have been like 17 or 18. I tell you, he was into every race, he knew all the candidates. And just had this kind of hustle to him that was hard to find,” Coleman said. “Basically Chaz will spend his weekend going through campaign finance reports for legislative races. I don’t know anyone else who does that to that extent.”
Nuttycombe decided to turn his predictions into a full-blown website in 2019. He reached out to other people who were analyzing nitty-gritty election data to forecast results in an online community called #electiontwitter.
“There are some of us sickos who stay up all night talking about poll numbers or precincts. I think that’s probably been good for him too,” Coleman said.
In 2021, at a chance hangout watching fireworks on the Fourth of July, he met Aidan Howard, then a rising junior studying geographic systems at Virginia Tech. Afterwards, Nuttycombe asked him if he would be interested in joining the site and working on political maps though Howard had no political experience. He sent Nuttycombe samples of his work – fire prevention and flood maps – and joined the site.
Some of the people on the cnalysis team – there are nine in total – have never met in person. They coordinate over Twitter, Slack and Discord. Nuttycombe relies on donations, Substack subscriptions, a small amount of ad revenue and some work for clients to pay a few a modest hourly wage (he declined to share the hourly rate, but said it was above the minimum wage anywhere in the country).
Another member of the cnalysis team – someone who goes by the X handle @cinyc9 – helped Nuttycombe understand how to break down electoral data to the most granular level possible, and then reallocate it to current precinct boundaries.
“Work-wise it’s cool that I assisted in it. But to me its a bigger win on a personal level because Charles is a good friend and knowing that I got to help him do something that’s been his dream for years … it means a lot to me knowing that I helped my friend achieve something he really wanted to achieve,” said Howard, who is now the site’s GIS technician.
One of the people Nuttycombe got in touch with was Jack Kersting, then a college student at the University of Alabama, who had been making his own maps focused on congressional, presidential and Senate maps.
Kersting, 22, is now the site’s chief oddsmaker, and builds the model that forecasts the chances of legislative control in each chamber. This year he built a live model for the forecast that ingested results from Virginia’s department of elections and provided real-time updates on election odds. He spent 50 to 60 hours on it over the last month.
“This was the first thing I’ve ever done like this. It was very satisfying in the end,” said Kersting, who is now getting a masters in finance.
Nuttycombe bases his predictions on a combination of previous election results in a district, campaign finance reports and internal campaign and party data he gets “through the grapevine”. He uses that information to assign each race a rating – toss-up, tilt, lean, likely, very likely and solid. The team at cnalysis feeds that information into a model that does 35,000 simulated election outcomes to predict a chamber’s outcome.
There have been learning moments since he began forecasting. In 2018, he overestimated Democrats in rural areas and underestimated them in the suburbs, he wrote in a blogpost on Tuesday titled “The 2023 Virginia election was easy to predict.” In 2020, he said he paid too much attention to campaign finance data and polling.
In 2022, Nuttycombe and cnalysis made forecasts in 83 of the 88 chambers. He made predictions in 3,380 races and was wrong in just 190 of them, a nearly 95% accuracy rate, according to his tally. His error rate, he said, was in part because he didn’t give his team enough time to analyze election data for new state legislative districts. He keeps a spreadsheet tracking the biggest missed prediction with an explanation of why the forecast was off.
Nuttycombe said he usually works on the site in the evenings, after classes, the gym and dinner, and balances it with a full course load (he’s taking 18 credits both this semester and next as he finishes his degree in political science).
The Democratic Legislative Campaign Committee (DLCC), which focuses on state legislative races, is aware of Nuttycombe and was following his work this year. The group relies on its own in-house data for forecasts, but they were watching Nuttycombe’s as well and could see it was consistent with internal projections.
“It’s hard to not take him seriously when what we were tracking internally was very similar to what he was tracking with his analysis,” said Abhi Rahman, a DLCC spokesman. “He’s definitely a very talented forecaster.”
Nuttycombe hasn’t been shy about his success, but acknowledges that he’s learned a lot since he began forecasting.
“There will be races in even-numbered years where I’m dead wrong. Maybe upwards of 10 races where I’m dead wrong. It’s just a resource thing. I maybe missed a scandal or some sort of development. Or a candidate does really, really well,” Nuttycombe said.
He’s also learned how to factor things into his forecast that can be difficult to quantify. In Virginia, for example, a Democrat in a competitive race this year had a scandal involving allegations she and her husband livestreamed sex acts. After the Washington Post broke the story on 11 September, Nuttycombe moved the state from a “toss up” to “tilt R”. In October, he moved the seat even more safely in the Republican column. The Republican candidate wound up winning by two points.
Nuttycombe plans to work on the site full-time after he graduates in the spring and is already planning out ways to grow his effort. While most of the country will be focused on the high-stakes presidential race next year, there will be thousands of state legislative races to analyze and predict.
“I’ll do this until they put dirt over me,” he said.