Article-Level Slant and Polarization of News Consumption on Social Media

Abstract

There is widespread concern that the online news ecosystem produces highly polarized content and that extreme content gets further amplified through social media distribution channels. Methodological limitations in estimating content-based slant at the article level have made evaluating these claims difficult. We use data on the near universe (∼ 1 million) of hard news articles published online by the top 100 U.S. news outlets in 2019, in combination with recent advances in natural language processing, to obtain a measure of content-based slant at the article level. We report five main findings. First, the majority (65%) of the variance in slant across articles arises within outlets, rather than across outlets. Second, most news produced is centrist, but the tails of the slant distribution are thick. Third, using article-level engagement data from the universe of U.S. Facebook users, we find that extreme content is much more likely to be shared widely on Facebook than moderate content. Fourth, consistent with concerns over echo chambers, left-and right-leaning users get exposed to and consume highly congenial news articles on Facebook, often from the same outlet. Fifth, polarization in news consumption, defined as the scaled difference between the average slant consumed by liberals and by conservatives, is higher than previously thought, because most existing measures do not take into account the fact that partisans consume pro-attitudinal news even within outlets.