The distribution and consumption of news on social media is a central concern for the future of journalism as social media platforms have become a prominent component of the news ecosystem. And, after enduring criticism about their possible role in undermining local journalism, social media platforms have become more active in their efforts to help sustain and promote local journalism.
One effort in this regard has been Facebook’s Today In feature, which was launched in 2018 and identifies and highlights local news stories for Facebook users. Originally launched in six communities and later expanded to 400 (Editor’s note: Facebook announced Thursday that it’s expanding Today In to more than 6,000 cities), Today In uses machine learning to identify local news stories across Facebook and displays them in a dedicated tab for users of the feature.
The nuts and bolts of Facebook’s local news efforts have largely remained within a black box. In an effort to open that black box, Facebook made a month’s worth of data collected for the Today In feature available to us so that we could analyze the local news available on Facebook. We combined these data with user interaction data from Facebook’s CrowdTangle platform and U.S. Census data in order to get a sense of how Facebook users engage with the different types of local news stories that are available to them, and how the availability of local news differs across different types of communities.
It is important to note that our data are from February of 2019, when the Today In feature was available in about 400 communities, and when the criteria Facebook employed for categorizing a story as local were a bit more geographically stringent than those now being employed to expand the availability of the feature. As Facebook noted at this earlier stage, “about one in three users in the U.S. live in places where we cannot find enough local news on Facebook to launch Today In.”
More information about the data and our methodological approach can be found in our full report. Here, we briefly summarize some of the key findings.
We were particularly interested in the extent to which the local news available on Facebook served a “critical information need,” a concept that we have used in previous research to get a sense of the extent to which local news stories are facilitating an informed citizenry. We found that the local news aggregated by Facebook often covers critical information needs, although our interaction data show there may be demand for more content of this type.
Using an automated content analysis approach we were able to categorize 149,283 of 313,787 stories in the data set. We found that 59 percent of the stories we were able to categorize served a critical information need. Emergencies (28 percent) was the most common critical information need served by local news stories in the categorized group. Aside from critical information needs, 31 percent of stories categorized covered sports and 9 percent were obituaries. These findings on their own show the potential of Facebook’s local aggregation to discover and feature content that serves critical information needs, but also highlight the challenges given the amount of stories devoted to sports and memorials. The mix of local news stories available on Today In is limited, to some extent, by the types of stories that local news sources choose to post, and then is also a function of how the selection algorithm chooses from amongst these available stories.
What we found particularly interesting was how user engagement (Facebook likes, shares, and reactions) on local news stories differed across different story types. We did not have data on how many clicks individual posts received. Users engaged most frequently with stories on emergencies, transportation and health.
Layering the story data with interaction data shows a clear appetite on the part of consumers for stories serving critical information needs; despite their prevalence in the story count, stories on sports and death are clicked on with less frequency. Surprisingly, we found that not only were local political stories the least common type of critical information need story in our data set, but political stories also generated the lowest levels of engagement of any story type. If we interpret engagement as an indicator of demand, then this finding is a bit discouraging.
Matthew Weber is an associate professor at the University of Minnesota’s Hubbard School of Journalism and Mass Communication and the Cowles Fellow of Media Management. He is the co-principal investigator of Duke’s News Measures Research Project.
Peter Andringa is a Robertson Scholar at Duke University and the University of North Carolina at Chapel Hill. He is a student researcher in Duke’s DeWitt Wallace Center for Media & Democracy and a fellow in the UNC School of Media and Journalism’s Emerging Technologies Lab.
Philip M. Napoli is the James R. Shepley Professor of Public Policy in the Sanford School of Public Policy at Duke University, where he is also a faculty affiliate with the DeWitt Wallace Center for Media & Democracy. He is the co-principal investigator of the News Measures Research Project.