The Prognosticators: Harris Class Attempts to Predict 2022 Midterm Election Results

With just hours to go before the midterm elections, pundits, pollsters, and betting markets are feverishly jostling to correctly predict who will control the U.S. Senate. So are the students in the Harris School of Public Policy’s Science of Elections and Campaigns course.

In this autumn-quarter elective course taught by Assistant Professor Alexander Fouirnaies, students are perfecting predictive models that they hope will pick winners — and boost their grade — in the 35 Senate races at stake on Nov. 8, a number that includes Oklahoma’s special election.

Students in the class — typically in their second year of graduate school — are being taught to take “the Harris approach to studying elections,” Fouirnaies said. That means, he said, “looking at the scientific evidence we have on different campaign activities and thinking systematically about how to predict elections.” 

His main goal, Fouirnaies said, is teaching the students to not just take statements at face value. “When people tell you, ‘Oh yeah, this type of campaign activity works,’ I want students to be able to critically look at the evidence to see whether it works or not.” 

Divided into two parts, the class asks students to draw on skills learned in their rigorous first-year Core classes, said Fouirnaies, a political scientist who has been teaching the class each fall since 2016. He took the baton from the class creator, Professor Anthony Fowler, who previously taught it.

In the first part, discussion centers on candidate stances and chances as well as polls and predictive models, Fouirnaies said, with students learning how to predict election outcomes in a systematic way. In the second part, the focus shifts to campaign technologies and exploration of the causal effect of different types of campaigns. “So, for example,” Fouirnaies said, “what's the effect of knocking on people's doors or sending out mailers or TV ads?”

One of the class exercises each year is to predict an upcoming election. This year, that prediction is about the U.S. Senate races.

“Usually, students do a pretty good job and build some pretty powerful predictive models,” he said. Sometimes, he said with a laugh, “they get close to beating FiveThirtyEight analyst Nate Silver and other professional pollsters.”

Students brush up their models until the last minute. But Madeleine Roberts, MSCAPP ’24, in late October shared an early peek at her work.

“In my current model, the Democrats are projected to pick up two, maybe three, seats in the Senate,” Roberts said, noting that this projection would keep Democrats in control. “As the midterms near and October Surprises arise, it will be exciting to incorporate new data into my model and see how the projections change.”

For her model, Roberts said she “used state-based polling data that I weighted based on the recency of the data, the pollster ratings, and whether either of the candidates were incumbents. I also incorporated past election results – to include voter patterns – and fundraising totals for the individual campaigns.”

Many of the students who take this class, Fouirnaies said, want to go on to work on elections or run for office. Roberts, a self-described “voracious consumer of electoral politics” said she is hoping to pursue a “data-based role within a political campaign, election forecasting institution, or research foundation.”

“This class has been extremely helpful in cementing my understanding of elections in the United States and the strategies that campaigns exercise to understand and persuade possible voters,” she said. The knowledge and modeling skills she has learned, she added, will be essential in her career.

The class, Fouirnaies noted, draws a diverse group of students, American and international. 

“The international angle brings something really interesting because political systems are so different across the world,” he said. Most of the available evidence is focused on U.S. elections, he said. But he wants students to think about the extent to which they would be able to take what they’ve learned and implement something similar in New Zealand or India.

From New Zealand, but with ties to Japan and India, class member Riki Fujii-Rajani, MPP ’23, said that while “things are so different here in the U.S., there are takeaways I can apply to other countries – especially around how to critically evaluate information and extract what is useful.”

Fujii-Rajani said she took the class “because I’ve always been interested in politics, especially the behind-the-scenes work and analysis to arrive at a strategy to appeal to voters or key stakeholders. My career plan is in progress: I know I want to end up in a policymaking space. But I came to Harris to broaden my horizons after a few years of central banking experience. This class is helping me do just that.”

Her predictions were on Democratic vote share, she said, and suggest Democrats, who currently hold 14 of the contested Senate seats, will win only 13. “Although,” she added, “this is very close!”

For her model, Fujii-Rajani relied on vote share and party affiliation of the incumbent from past Senate elections, presidential approval ratings, and consumer sentiment.

“These annual election predictions are a fun exercise,” Fouirnaies said. “It's in real time. And students have to consider many factors, ‘OK, what kind of data can I get? What do I think matters the most? How can I systematically train different models and then evaluate which model is the best in the historical perspective, and then try to apply that to predict the upcoming election?’”

“It’s also at the borderline between science and art,” he explained. “You need a systematic approach, but there's also part of it where you need creativity to think about different ways of doing things. Students can experiment and sometimes they find out that an idea was horrible. And sometimes they find out that an idea was terrific.”

Fujii-Rajani said she was surprised by the many ways in which extracting information from data can go wrong. 

“It’s easy to read media articles or poll results and believe it to be the ‘truth’ since the data says so. But collecting and analyzing data is a process that’s full of trade-offs,” she said. “It’s definitely useful to understand how the metaphorical sausage is made!”

For Roberts the biggest takeaway from the class “is that election polling is not broken.” 

“Media outlets like to contend that polling is in crisis,” she said.  But she noted that she’s learned that after the 2020 presidential election, “polling institutions turned inward, reflected on their errors, and to this day, continue to readjust their polling techniques.”

Fouirnaies – whose own research concentrates on the political economy of elections, digging into fundamental questions about accountability, representation, and institutions – said the belief that polls are useless today is “definitely one of the ideas that students typically have when the class starts. But that’s not based on any science. On average, polls are actually doing quite a good job.” 

But whatever his students bring to the class, “one of the things that I really want them to take away is if you care about policy, you should also care about campaigns.”

“It's where policymaking starts, this whole chain leads to a policy in the end,” he said. “And if you care about policies, you have to also care about campaigns. And if you care about making progress on a particular issue, you should care about how to organize an effective campaign – that’s how policymaking happens.”

To read about the class's predictions in the previous midterm, read "Harris and the Power of Election Prediction".