2020 Election Forecast

2021 Shortlist

Country/area: United States

Organisation: FiveThirtyEight

Organisation size: Big

Publication date: 12 Aug 2020

Credit: Many, many people. Please see complete list at the bottom of the project.

Project description:

The FiveThirtyEight 2020 election forecast combines sophisicated statistical modeling of election results with innovative data visualization techniques. For this year’s version we prioritized accessibility for a wide range of audiences, usability of our data visualizations, and strong integration with our other editorial content. The result is a project that makes a complicated and commonly misunderstood subject easy to understand without skimping on detail.

Impact reached:

This project informed a vast audience about the range of possible electoral outcomes in a critical election year. It sparked interest in the poltiical process amongst our readership, informed our reporting and was widely cited by other digital and broadcast media. By any quantitative measure it was the most successful project FiveThirtyEight has ever created.

But perhaps it’s greatest success is in what it did not do. Our focus on accessibility this year led to numerous design tweaks that demystified and simplified the forecasting process, heading off confusion and misinpretation. In particular, we used user experience research and reader feedback to ensure the uncertain nature of the forecast was well-understood by a wide variety of readers, including those with limited statistical literacy. We innovated new data visualization forms (the highly-regarded “ballswarm”) and added a carefully designed mascot (“Fivey Fox”) to guide readers through statistical concepts and on to our other elections coverage.

This project also provided the raw materials for other projects, including the interactive version of the forecast we published in October.

Techniques/technologies used:

The forecast uses FiveThirtyEight’s standard technical stack which involves a Node based build system on the backend and a lot of vanilla javascript on the frontend. We used D3 for charting. The forecast model itself is written in STATA. Some automation was written in Ruby.

What was the hardest part of this project?

The hardest part of this project was finding the right ways to communicate the level of detail our existing readers expect, while not overwhelming or confusing new readers or those who’s first exposure to FiveThirtyEight is the forecast. This a delicate balancing act which entailed extensive user research, endless rounds of design mocking, and hours of debate amongst the team. We further invested a great deal of time into careful choice of words, integration of other explanatory reporting, and accessible design components, such as screenreader summaries for data visualizations.

What can others learn from this project?

Every fresh project is an opportunity to improve on what’s come before. As a team we are proud of the intentions we brought to this project. We were aware of the public critiques that had been leveled against previous version of the forecast. Rather than react defensively we engaged in a comprehensive effort to build a forecast that not only engaged all readers, but truly informed them. The use of complicated statistical techniques need not create a black box. A first time reader to the forecast can walk away just as well-informed as someone who understands monte carlo simulations and the nitty gritty of details of the electoral college.