Do Online Intermediaries Increase the Frequency and Diversity of News Exposure? Cross-Country Evidence from Web Tracking and Surveys

Recent studies have shown that online intermediaries such as social networking sites, search engines and email portals increase the frequency and diversity of news exposure. This evidence relies on a mix of survey-based studies and passively recorded web tracking data, thereby providing robust and important evidence against the notion of widespread "filter bubbles" caused by algorithmic recommender systems. At the same time, all existing research suffers from at least one of the following limitations: (1) The frequency of news exposure is usually measured at the news media brand (offline) or web domain level (online) instead of the individual article level. This ignores that significant shares of contents, especially if encountered on social networking sites, are non-political. (2) While previous research shows that the number of news sources used correlates with the diversity of a user's news diet, a more direct measure of diversity requires the analysis of the actual contents that users have seen. (3) Most studies have not distinguished non-regular - and therefore possibly incidental - news exposure via online intermediaries from regular, typically more intentional or routinized forms of news consumption. (4) Evidence based on individual-level web tracking data comes exclusively from single-country case studies, whereas patterns of news exposure are clearly contingent on the political and media systems in which it is embedded. The present study seeks to address these limitations by analyzing 150 million website visits by 7,729 study participants to investigate the effects of online intermediaries on news consumption in France, Germany, Italy, Spain, UK, and the US. In combination with survey data, the web tracking data allow us to measure news consumption on a very fine-grained level and to link online behavior to demographic and other individual-level covariates. A large-scale web crawling of the actual contents that participants saw is combined with automated text analysis methods to create three outcome measures: (1) Exposure to different news types, based on outlet classifications (e.g., legacy press, public broadcasting, hyperpartisan news); (2) exposure to political news, based on a machine learning model trained on articles from the politics sections of the included news sites; and (3) exposure to diverse topics, using a measure derived from topic models and text similarity scores. For relating the political and diversity measures per visited URL to the use of online intermediaries, we apply a statistical model that distinguishes between stable between-person differences and within-person effects, the random-effects within-between (REWB) model. The REWB model allows us not only to compare within- and between-person effects, but also to investigate how effects of intermediaries on the amount and diversity of online news exposure vary across countries. First results for Germany show that intermediaries increase the frequency and diversity of news use. However, we expect to find considerable heterogeneity across countries, where the use of Facebook for political news and the political parallelism of media systems vary considerably. We discuss the implications of our findings for debates about filter bubbles, online selective exposure and political knowledge acquisition in high-choice media environments.