Published by the Students of Johns Hopkins since 1896
September 22, 2021

FiveThirtyEight’s Micah Cohen and Galen Druke explain election forecasting

By ARIELLA SHUA | October 18, 2020



Micah Cohen and Hopkins alum Galen Druke explained the mechanics of FiveThirtyEight's election forecasting model.

The Political Science Steering Committee (PSSC) at Hopkins hosted Galen Druke, a Hopkins alum, and Micah Cohen of the political analysis website FiveThirtyEight on Oct. 13. Junior Leland Held, vice president of faculty outreach for PSSC, moderated the event.

FiveThirtyEight was founded in 2008 by statistician Nate Silver. Named after the number of electors in the U.S. Electoral College, the organization is known for its political polling models and data-driven analysis. The website features articles and podcasts analyzing political poll averages, including forecasts for presidential and congressional elections.

Cohen discussed the forecast model, one of the website’s most popular features, as a tool that helps people more accurately understand the results of upcoming elections.

“An election model is just a set of rules set down in advance that govern how we take in and interpret data,” he said.

Cohen, a former reporter at the New York Times, is the managing editor of FiveThirtyEight. Druke is the producer and host of its podcast.

Druke also is a Hopkins alum who graduated from the Krieger School of Arts and Sciences in 2012. He was an intern at WYPR, Baltimore’s local public radio station, where he said he got his first experience in journalism that eventually led him to work for FiveThirtyEight.

Junior Riya Jain attended the session to better understand the role polls play in politics.

“I spend a lot of time looking at the debates, looking at the candidates,” she said. “I look at [political elections] from a very policy-heavy perspective, and also how they appear in the media. I don’t really look at the more logistical part of it — the numbers, the statistics, the polling.”

Cohen and Druke explained that people have a difficult time understanding the implications of polls.

Many depend on polling data to provide a mental shortcut in understanding the political environment. The likelihood of the outcomes resulting from the polls, however, are less well understood.

FiveThirtyEight rates pollsters according to factors such as partisan bias and polling method. It then averages the polls together to determine its polling average for national popular vote and individual state races.

Druke spoke about the website’s signature election models, which are derived from the polling data.

“In our forecast we show bands that fall within an 80% likelihood, such as with the popular vote forecast,” Druke said. “We also show the different outcomes that could result from the data of the polls. It helps people understand uncertainty in ways that they can’t with the other ways we cover elections.”

Senior Nina Tophoff occasionally checks FiveThirtyEight and receives regular polling updates from the New York Times. She wanted a deeper understanding of how political polls model public sentiment.

“I went because polls were so off last time,” she said. “How do I know what I’m seeing is what’s happening? How do I know these polls won’t be off?”

In 2016, FiveThirtyEight’s model showed Democratic candidate Hillary Clinton to have a 71% chance of winning the election. However, several states predicted to swing in Clinton’s favor went to Republican candidate and eventual winner Donald Trump.

Cohen explained that the website had written articles prior to the 2016 election warning voters that Trump was only a polling error behind. He noted that he himself did not predict Trump to win despite the fact that he knew intellectually that it could happen.

Druke added that some pollsters in 2016 did not properly weigh their polls by education in swing states like Michigan and Wisconsin, creating a skew in the Democrats’ favor. They have since made adjustments to their methods.

“Pollsters often have a hard time reaching various groups of people,” he said. “One is non-college-educated people. Two, people of color. Three, younger people.”

Both Cohen and Druke stated there was nothing to add to the current model that would improve it.

“The better thing to do is, look at my model and consider: Does it make sense in the real world?” Cohen said. “The model looks at public opinion, looks at the economy, looks at whether the president is an incumbent or not. That’s basically it.”

Though Jain appreciates FiveThirtyEight’s polling analysis, she stated that polls do not sway her political opinion.

“I don't choose who I vote based on the polls,” she said. “It’s one of the biggest problems, where people look at the polling and see who the front runner is and decide their vote isn’t worth it. The value of polling is important for campaign officials and understanding how to run a campaign. I don’t know how useful it is for the average voter.”

Cohen and Druke noted that polling has been relatively stable in 2020.

Polling average from RealClearPolitics, a competitor website to FiveThirtyEight that aggregates polling data, shows that support for Trump never went beyond the 40% to 44% range in the past six months. Vice President Joe Biden’s polling fluctuated between 47% and 52%, maintaining a fairly stable lead over the president.

FiveThirtyEight’s presidential election forecast gives Democratic candidate Biden an 87% chance to win the election as of this writing.

Druke explained that the pandemic has damaged Trump’s political outlooks, but most external factors have done little to change the polls. Cohen added that the model does not account for any attempts to manipulate the vote or reshape the rules of the election.

Tophoff praised the decision to bring Cohen and Druke to speak.

I’m glad they’re doing these kinds of things,” she said. “They’re helping us understand the news. And it’s cool that one [speaker] was a Hopkins alum.”

Chris H. Park contributed reporting to this article.

Comments powered by Disqus

Please note All comments are eligible for publication in The News-Letter.

News-Letter Special Editions