Predicting Virginia: A Brief Retrospective
I was a rising senior in high school when I cast my first-ever set of state legislative predictions. On August 25th, 2017, I went to the Virginia Public Access Project’s page to find a map of the House of Delegates, took a screenshot, traced the district lines with my mouse in Adobe Photoshop, and colored in the districts with my district-by-district ratings. Once I finished the map and made the ratings, I posted them to Twitter.
I began writing that year for DecisionDeskHQ, with J. Miles Coleman (who now works at Larry Sabato’s Crystal Ball at the University of Virginia’s Center for Politics) and forming my prediction maps for the House of Delegates using ArcGIS. Six articles for DecisionDeskHQ on each competitive district later, I finalized my predictions for the House of Delegates on November 6th.
I’ll admit to being somewhat infatuated with the competition, but I had only seen two people even attempt to evaluate the House of Delegates districts in Virginia that year: Mr. I’ve Seen Enough himself, Dave Wasserman at the Cook Political Report, and HeartCooksBrain on DailyKos, AKA “Opinion Haver” AsInMarx on Twitter. At age 18, I overperformed my expectations and “bested” Dave, though the article I’ve attached was written in September and I don’t believe he updated his predictions on the House of Delegates before Election Day. I fared about as well as AsInMarx, or a neck ahead, depending on your system of values since I eliminated Toss-Ups to decide an ultimate winner.
I loved reading both of these guys’ work then as much as I do now, and I was surprised with how well I did once the results were finalized: I correctly predicted 96/100 seats in the House of Delegates with Democrats picking up 11 seats as others, including then-Governor McAuliffe, predicting that Democrats would do slightly worse than that. This score, by the way, includes David Yancey (R) being declared the winner after his name was drawn out of a ceramic bowl once he tied with Shelly Simonds (D) in HD-94, which is where I was born 24 years ago.
In 2019, after a successful 2018 cycle, I jumped back into focusing on Virginia and ended up doing better than I did in 2017, successfully predicting 99/100 House of Delegates seats as well as 38/40 State Senate seats. In 2020 I started CNalysis alongside the founding staff: cinyc9, Jackson Martin, Allie Geier, and Blaine Monroe. Down the line, Chris Leonchik, Jack Kersting, Aidan Howard, and Elliot Li would join us.
The year after in 2021, I correctly predicted that Glenn Youngkin would be elected Governor with a Republican House of Delegates, though I believed at the time there would be more ticket-splitting for the Lieutenant Governor and Attorney General races for Democrats. Still, I knew that if Youngkin hit my ceiling for him, and if there was less ticket-splitting than expected, Republicans could pull off a statewide sweep. In the House of Delegates, I’d circle back to getting 96/100 seats right.
The Virginia 2023 Election
I’ve never worked harder on a Virginia off-year election cycle than this one. Our GIS team, Aidan Howard and cinyc9 were instrumental in our success by calculating how each of the new districts voted from 2019-2022 in each of those elections. Once I had the 2021-2022 data at the beginning of the year, I created my initial ratings and Jack Kersting created the forecast pages, running the ratings through 250,000 simulations back in March.
After a big primary season, I spent most of my summer traveling through Scotland after saving up enough money. When I got back, the team and I hit the ground running. Outside of Virginia, our intern Elliot and our campaign finance analyst Chris helped with work that needed to be done in New Jersey and Mississippi, which also had state legislative elections. Jackson helped keep the lights on by building our Bang for Your Buck Model for the Substack, generating subscriptions from people who wanted to know where they could spend their money for the most direct impact.
I asked cinyc9 and Aidan in September to figure out how each district voted in 2019, as I believed this would help strengthen my confidence in our forecast. Using that data I made appropriate adjustments to what I believed to be the expected outcome in the final stretch of the campaign. The initial ratings from March used an average of 4 different environmental electoral scenarios using the 2021 and 2022 data and the trends in each district from 2021-2022. When I got the 2019 data I fine-tuned what is known as the “base rating” in my methodology by using the electoral data and 19-21 trends and 19-22 trends. This didn’t change the forecast much once I initially got the 2019 data, but was critical to helping pick the closest races.
In addition to this data I had factored in the campaign finance reports into our forecast and made rating changes when needed. In years past, I had weighed this too heavily in my ratings so I made sure I was not only cautious but also paying attention to the spending habits of each campaign, noting to whom and what they were allocating their funds.
Another oversight I had made in years past in Virginia was only reaching out to Democrats for their available data. From 2017-2019, I was publicly rooting for the Democratic Party, but when I started CNalysis in 2020 I pledged to be nonpartisan in my work. In 2021 I began to build relationships with the Republicans here in Virginia who were familiar with my work, which helped point me in the right direction in predicting a Republican House of Delegates and Governor Glenn Youngkin.
This year I had more access to internal data on both sides of the aisle than ever before. I’m fortunate to have spoken with many people on both sides of the aisle who worked tirelessly to pull their party over the top in what has proven to be a highly competitive general election.
I gathered internal polls for almost every single state legislative district across the Commonwealth, which helped put my predictions in perspective. The ratings didn’t rely on these polls, especially since there were results in them that I knew made them outliers based on previous election results, but they certainly helped track movement across the Commonwealth. This metric was very helpful in determining the preexisting electoral environment. Once I had figured out the electorate would be somewhere from D+1 to D+4, the 2019 helped put the down ballot tendencies in perspective for what a Democratic-leaning environment could look like in Virginia on the state legislative level.
In the end, I had come to the conclusion written in my final prediction piece that based on polls, other private data, and special election data throughout the year, most of the competitive races would be at least 3 points to the left of the 2021 gubernatorial margin. That proved to be correct, with two regions exempted: Prince William County and Southside Virginia. I anticipated this, and thus predicted Republicans would win HD-22 in Prince William as well as HD-82 and SD-17 in Southside.
General conversations with Republicans and Democrats also put things in perspective for me, in addition to the data. I spoke with Republicans who thought they’d get 53 seats in the House of Delegates, and Democrats who believed they’d get 23 seats in the State Senate. Both of these were fair estimates as to where things would land, but each side lives in their own bubble as they only see their own data, which can be slightly biased in their favor. Getting to reach out to both sides and having access to their data makes it far less likely to become ensnared in one side’s bubble.
One of the most important data points that would be shared with me was early voting data. We can see how many people cast early votes in each city/county in Virginia and by what method, but we cannot see what type of voter was participating in early voting. Unless, of course, you have access to something like VAN or i360. After I spoke with operatives from both parties, I came to the conclusion a few weeks into October that the new early votes for the Republicans consisted mostly of the same voting bloc that was already going to come out on Election Day, with the exception of Southside. That area was the only one I saw that had these low to mid-propensity Republican voters make up a sizable share of the early vote electorate. So while others were predicting that Stirrup would win HD-21 thanks to the L2 political data which showed incredible numbers for the Republicans, I already knew that these were the voters concerned with the data center and thus came out to vote early.
Of course, I cannot name those who contributed to my successful analysis across the Commonwealth (but y’all know who you are, and I thank you). Speaking with a forecaster and letting them know what your internal data shows is always seen as risky, and can land you out of a job sometimes. This fear is understandable and several people on both sides would refuse to answer my questions or calls. All love to them. The biggest worry, of course, is that someone in my line of work would leak it to the other side.
I have a saying: I don’t care who wins, I care about being right. Leaking one side’s data to the other is detrimental to my work as it can cut me off from sources willing to help me out. Everything I’ve gathered is in my notes and spreadsheets accessible to me, myself, and I.
In addition to all this, I do try and learn my lessons in every race I get wrong to make sure I don’t repeat the same mistakes. In the 2018 midterms, I overestimated Democrats in rural areas and underestimated them in the suburbs. In 2019, I was too bullish on Democrats in Southside Virginia. In 2020, I weighed campaign finance data and polling too heavily. In 2021, overestimated how much ticket-splitting there’d be and didn’t weigh prior election data enough. In 2022, I didn’t give our GIS team enough time to figure out how all of the new state legislative districts had voted in years prior, and didn’t pay attention to a polling aggregate that was skewed to favor Republicans. My grandfather, Charlie Nuttycombe, had his own saying: “practice with persistence to perform your potential.” After six years, I’ve finally performed my potential.
A lot of people have asked me how I’m feeling now that I’ve finally achieved this. To be honest, I thought I’d be jumping up and down with joy. That’s not to say I’m not happy, I am. But compared to the three times prior I’ve predicted Virginia elections, this one felt easy to predict. That’s thanks to my experience and my relationships with both Democrats and Republicans in Virginia, as well as the hard work I and my team have put into this.
There were some close calls, of course, so part of my success was luck. I came pretty close to getting a 139/140 or worse. Notably, on Election Day I said that based on turnout that day Lily Franklin (D) was on track to defy my final “Lean R” rating for my home district. After checking on the 4 PM turnout on campus, when every voting booth was full and students were filling their ballots on foldable tables, a new unsettling thought entered my head: “I think she’s probably gonna win now. I’m going to possibly get every district right this year but my own.” It was a close call, but that didn’t turn out to be the case, though I had admittedly underestimated Lily Franklin.
Statistically, there were plenty of universes where instead of nailing this election I had my worst set of Virginia election predictions yet. I would sometimes get asked by Republicans and Democrats alike: “What if you’re wrong this year?”
I guess we’ll never know.