The analyses here are mostly reproductions of figures from two articles: Allen et al (2020) and Muise et al (2022). The underlying datasets are derived from large panels (50-100k) of individuals who agreed to have their desktop internet browsing and TV viewing behavior passively tracked by the Nielsen company.

Nielsen is a U.S.-based data analytics company that uses representative panels to estimate media consumption trends on TV, web, and mobile platforms. Every month, these panels are adjusted, where new panelists are added, and current panelists can choose to no longer participate. These panelists are usually involved in the panel for at least a year but no more than two years.

We note that partisan content labels are not provided by Nielsen. Rather, we categorize content at the level of TV programs and web domains into news, and further into right-leaning news, left-leaning news, and hyper-partisan news using categorizations and methods popular in the academic literature. In particular, we place online news domains on a left-right ideological spectrum with a popular method that assigns an ideology score for domains based on the partisanship of the users who share content from the domain on Twitter/X. As a disclaimer, aggregation methods are our own, as Nielsen does not provide ideological labels to TV programs nor defines what programs are considered right-wing or left-wing.

For more details on how we define news content, see methods section of the 2020 paper, and for more details on how we define echo chambers, see the methods section of the 2022 paper.


Allen, Jennifer, Baird Howland, Markus Mobius, David Rothschild, and Duncan J. Watts. “Evaluating the fake news problem at the scale of the information ecosystem.” Science advances 6, no. 14 (2020): eaay3539.

Muise, Daniel, Hosseinmardi, Homa, Howland, Markus Mobius, David Rothschild, and Duncan J. Watts. “Quantifying partisan news diets in Web and TV audiences.” Science advances 8, no. 28 (2022): eabn0083.