skip to main content

Election Analytics Website Presents Perfect Merger of Computer and Social Science

10/16/2020 9:35:07 AM Aaron Seidlitz, Illinois CS

Since 2008, Illinois CS professor Sheldon H. Jacobson has worked with students to create, develop and publish the Election Analytics @ Illinois website. Over those 12 years, there have been many great reasons for Jacobson to continue the project.

Sheldon H. Jacobson
Sheldon H. Jacobson

Chief among those reasons is the understanding that undergraduate computer science students gain. Primarily, they realize their computing skills can impact projects that have national implications.

When Jacobson explains how this website tracks and informs the public about the presidential and senatorial elections in the United States, he looks for a gleam in a student’s eye. To the professor, this shows interest in combining an interest in CS with social impact.

"My involvement in many interesting applications over the years proves to my students that Illinois Computer Science reaches a very large footprint of influence,” Jacobson said. “The real experience with our election analytics website is for our students to apply the things they’re learning in the classroom.

“It helps that they do it for an eager audience, because everybody wants to know what’s going to happen before it happens.”

Three of the four students currently powering the election analytics website are CS majors.

Each found themselves interested in coding at a young age. As they have grown within the CS program, however, each found that projects like this one spark a deeper passion for the industry.

Sophomore Marzuk Rashid noticed Jacobson’s announcement about the project last year. He immediately discussed the opportunity with fellow classmates and expressed an interest to work with Jacobson.

Rashid said he now optimizes the backend of the site to make the model run faster. He also designed an interactive histogram visual and created an expected electoral vote visual. He’s excited to apply his growing skillset in computing with a genuine interest in politics that dates to his junior high years.

“Our project is similar to what Nate Silver does with his site, FiveThirtyEight, and I thought I would enjoy working in the same way,” Rashid said. “It’s impossible to predict the future, because sometimes we encounter an unprecedented black swan event. But we can make observations from past patterns to inform ourselves about what might happen.”

This year's Election Analytics @ Illinois website has four team members, including three undergraduate students from Illinois CS.
This year's Election Analytics @ Illinois website has four team members, including three undergraduate students from Illinois CS.

As Rashid noted, predicting the outcome of such a high-profile event is difficult. That’s where trust in their professor’s methodology comes into play.

Jacobson based his model off Bayesian statistics, which is an interpretation of probability where probability expresses a degree of belief in an event. He first utilized the model for the 2008 election, and he explained its effectiveness in the 2009 research paper.

Jacobson’s work aims to collect and address the overwhelming amount of data swirling about during election season. The site then turns that data into a much less daunting read for casual followers of the electoral races.

Despite the challenges to accurate forecasting – such as slim margins in state-by-state results that affect the electoral college outcome – Jacobson said his method has stayed the same since 2008.

But he and the students have occasionally tweaked it. Recently, the group did decide to weight certain polls differently than others. Yet, he remains confident in the method and the result.

“We take the probabilities from each state and the District of Columbia, to form a mosaic of all the possibilities for who’s going to win the White House or who’s going to control the Senate,” Jacobson said. “I’m very big on storytelling. When I say I use data-driven storytelling, it means that I try to ensure that our data helps explain a narrative that people finding interesting and informative.

“Accordingly, our students working on projects like this all have a sense of social altruism. They are doing something for the betterment of others and society.”

Still, as certain as they are in the method and their interaction with it, outcomes do not represent what will happen as much as what could happen.

Very few results proved that better than the 2016 election, for which Jacobson said their site produced 24 potential outcomes. Only one of those represented a Donald Trump victory and Hillary Clinton loss.

Despite the odds, that is exactly what did occur – an outlier so unheard of that Jacobson can only compare it to one other presidential election.

“When we did a post-analysis of the election, we found an uncanny comparison with the 1948 election between Truman and Dewey,” Jacobson said. “The parallels between the two were striking. There were three states in 1948 that determined the election, and there were three states that did the same in 2016.

“This was a once-in-a-lifetime event – since it took 60 or 70 years between instances.”

During this election, Illinois CS sophomore Christian Sparks enjoyed learning more about potential outcomes. His primary takeaway stems from Democratic candidate Joseph Biden and “how shockingly consistent his polling has been over the past few months.”

Still, he pairs Biden’s current lead in the race with an understanding of what occurred in 2016.

“A lot of the presidential models in 2016 were far too aggressive and didn’t show their information in a way that made sense to users,” Spark said. “In 2020, many forecasters, including us, have changed the way they show the results of their models to make it more understandable to users.

“The models we make are extremely useful, but we need to make sure that we explain to users the risks and assumptions that we make in our models.”