Assistant Professor @UTAustin @UTiSchool. @Stanford and @OhioState alum.
- Tweets674
- Following1.4K
- Followers1K
- Likes4K
ALT First page of the article in Political Communication, 2026, volume 43, issue 6, pages 851–887. Title: Multi-Platform Referrers of Misinformation: A Comparative Analysis of Misinformation Visits Referred by Facebook, Twitter, Instagram, Reddit, YouTube, Snapchat, and TikTok. Authors: Ross Dahlke, Ryan C. Moore, Danya Adib-Azpeitia, Johan Ugander, and Jeffrey T. Hancock. The abstract says existing research counts only misinformation visits that directly follow social media. The authors call this approach the direct referrer paradigm. They introduce the embedded referrer paradigm, which uses counterfactual simulations to count social media's role across people's whole browsing journeys. The data are 3 months of web browsing from 1,240 Americans (21 million website visits). Effects under the embedded referrer paradigm are 1.5 to 15+ times larger than effects under the direct referrer paradigm.
ALT Figure 1, a diagram with two panels. Panel A, "Captured by Direct Referrer Paradigm," shows one arrow from a Facebook box to a Misinformation Website box. Panel B, "Captured by Embedded Referrer Paradigm But Not Direct Referrer Paradigm," shows a chain of four boxes: Facebook, then Curiosity, then Intermediate Website, then Misinformation Website. In Panel A, the direct referrer paradigm gives Facebook all the credit for the misinformation visit. In Panel B, the direct referrer paradigm gives Facebook no credit, even though the path started on Facebook. The embedded referrer paradigm captures Facebook's role by simulating the person's path as if they could not visit Facebook.
ALT Figure 2, a flow diagram of the study's analysis steps. A: Web browsing data from desktop and laptop computers, smartphones, and tablets, collected August 24 to December 7, 2020. The data are shown as a table of user IDs, timestamps, and URLs, with a survey alongside. B: The browsing data give each person a transition matrix, the probability of moving from one website to another. C: When an observed path is about to reach a social media platform, such as Facebook, the path is diverted to a different platform using experimentally derived diversion ratios. D: The diverted path is simulated to completion with the person's transition matrix. E: Multi-level models compare visits to misinformation websites in the simulated paths with visits in the observed paths.
ALT Table 1, "Direct versus embedded referrer paradigm results." The table gives the estimated percentage reduction in visits to misinformation websites if each of seven platforms were removed. Estimates appear under both paradigms, for all misinformation and separately for conservative, liberal, and non-ideological misinformation. The data cover 1,240 participants and 395,936 sessions. For all misinformation, removing Facebook cuts visits by 9.25% under the direct paradigm and by 14.07% under the embedded paradigm. Twitter: 2.17% versus 12.27%. YouTube: 2.08% versus 4.92%. For all misinformation, Instagram, Reddit, Snapchat, and TikTok have estimates near zero, and their confidence intervals include zero. The largest gaps are for liberal misinformation: Facebook 11.43% versus 58.56%, YouTube 1.93% versus 29.30%, and Twitter 3.59% versus 24.22%.
ALT Title page of journal article "The Consumption of Pink Slime Journalism: Who, What, When, Where, and Why?" by Ross Dahlke, Ryan C. Moore, Priyanjana Bengani, and Jeffrey T. Hancock, published in Digital Journalism. Abstract describes a study combining web browsing behavior of 1,238 American adults (21M website visits) during the 2020 election with a database of 1,627 pink slime outlets and 132,237 articles. Finds 3.7% of American adults (~9.5M people) visited pink slime sites; consumption differs from general misinformation and local news; topics produced and promoted differ from topics consumed; Google and Facebook were prominent referrers; news desert residents were not more likely to consume pink slime. Keywords: pink slime, digital trace data, audience analysis, local news, misinformation, journalism, 2020 election, imagined audiences.
ALT Figure 1: nine-panel grid showing proportion of participants exposed to Pink Slime (left column), Misinformation (middle), and Local News (right) websites, broken out by Age (top row), Media Diet Slant on a 1–10 liberal-to-conservative scale (middle row), and Race (bottom row). Each panel plots point estimates with confidence intervals. Pink slime exposure is low across all subgroups (mostly under 10%) with no strong demographic pattern. Misinformation exposure rises sharply with age, peaking around 50% for the 65+ group, is higher among more conservative media diets, and is lower among Black respondents (~17%). Local news exposure is fairly even across age and race, slightly higher for more liberal media diets.
ALT Figure 4: four U.S. maps in a 2x2 grid showing where individuals live (dots) and where the local news or pink slime sites they visited claim to be based (arrow tips). Top row shows local news visits; bottom row shows pink slime visits. Left column shows non-Trump supporters (blue); right column shows Trump supporters (red). Local news maps for both groups are densely covered with arrows across the country, indicating widespread visits. The two pink slime maps are nearly empty by comparison, with only a handful of arrows, and those arrows frequently span long distances across state lines—suggesting people typically visit pink slime sites that are not local to where they live. Alaska and Hawaii are not included.
ALT Table 1: Topic Model Results for 18 topics, with columns for whether the topic is programmatically generated, percent of articles produced, percent consumed by readers, percent of Facebook ads, and percent of tweets. Notable mismatches between production and consumption: Gas Prices is 21.5% of articles produced but 0% consumed; Iron and Metals is 18.4% produced, 0% consumed; Professional Licensing is 12.5% produced, 0% consumed. Conversely, Politics is just 1.4% of articles but 55.9% of consumption. Illinois Public Salary Data is 4.1% produced and 12.7% consumed. COVID and Election Conspiracies is 2.0% produced, 10.8% consumed, and 61.2% of Twitter promotion. Local News is 50.5% of Facebook ads and 33.7% of tweets but only 4.5% of articles and 5.9% consumed.
ALT Three texts from the same person. The first reads "Do you have advice on what I should put I the thread promoting commit on twitter?" The second reads "I could use some help." After those texts is the read receipt. A third, unread, message follows "Guys????"
