Thanks, Ben!

Screenshot of a paper titled “Contextualizing Misinformation: A User-Centric Approach to Linguistic and Topical Patterns in News Consumption,” authored by Ross Dahlke and colleagues. The abstract says the study uses web-browsing data from 1,240 U.S. adults during the 2020 election to compare misinformation and hard news. It finds that misinformation people consumed was generally easier to read, more negative in tone, and more morally framed, with substantial variation across topics and across groups such as older adults and Republicans.

Image description Two side-by-side horizontal bar charts compare topic distributions for hard news and misinformation. Hard news is led by general news at 35.9%, followed by U.S. electoral politics at 27.0%, social issues at 17.1%, COVID-19 at 14.2%, and health at 5.7%. Misinformation is much more concentrated in U.S. electoral politics at 53.0%, followed by social issues at 23.7%, COVID-19 at 11.6%, general news at 6.8%, and health at 4.9%.

Two stacked line charts show how topic shares changed over time from late August to early December 2020, with a vertical marker at Election Day 2020. In misinformation, U.S. electoral politics rises sharply in October and November and becomes the dominant topic around the election. In hard news, general news remains largest throughout, while U.S. electoral politics also spikes around Election Day before declining afterward.

Image description Two stacked line charts show how topic shares changed over time from late August to early December 2020, with a vertical marker at Election Day 2020. In misinformation, U.S. electoral politics rises sharply in October and November and becomes the dominant topic around the election. In hard news, general news remains largest throughout, while U.S. electoral politics also spikes around Election Day before declining afterward.

Most web browsing studies analyzing news and misinformation operate at the domain level. Work by me,
@fangjingtu.bsky.social et al., scrapes the content from web visits to go beyond the source to the content level, finding significant topical and linguistic variation doi.org/10.1145/3757571

Style and substance on The Alex Jones Show predict InfoWars sales: a multi-modal analysis of a media empire Alex Jones, a prominent conspiracy theorist, has garnered substantial influence and wealth through his InfoWars media empire, which includes The Alex Jones Show and InfoWars.com. This study leverag...

Cannot believe I got to work with such an incredible group of scholars on a topic that I care so deeply about! Check out our new article on the political economy of Alex Jones, wherein we find that his presentation and topic choices predicted his merch sales.

www.tandfonline.com/doi/full/10....

Style and substance on The Alex Jones Show predict InfoWars sales: a multi-modal analysis of a media empire Alex Jones, a prominent conspiracy theorist, has garnered substantial influence and wealth through his InfoWars media empire, which includes The Alex Jones Show and InfoWars.com. This study leverag...

doi.org/10.1080/1369...

Screenshot of the title page of a journal article in Information, Communication & Society by Ross Dahlke and coauthors. The article is titled “Style and substance on The Alex Jones Show predict InfoWars sales: a multi-modal analysis of a media empire.” The abstract explains that the study combines daily InfoWars sales data from 2016 to 2018 with linguistic, auditory, and topical features from Alex Jones’s radio show and online articles, finding that some styles and topics predict next-day sales.

Image description Line chart showing daily InfoWars sales in dollars from January 2016 through December 2018. Sales are highly volatile, with frequent spikes, but generally rise from relatively low levels in early 2016 to a higher and more sustained range through 2017 and 2018, often around $100,000 to $300,000 per day, with occasional peaks approaching $1 million.

Image description Multi-panel figure showing daily trends in selected themes and styles in Alex Jones radio shows and InfoWars news articles from 2016 to 2018. The left column tracks radio show content including Power, Bio, Achieve, Focus Future, and Money; the right column tracks article content including Power, Achieve, Money, Anger, and Focus Future. Gray daily values are overlaid with smoothed trend lines, showing that some themes shift gradually over time while others remain fairly stable.

Image description Multi-panel figure showing daily trends in major topics in Alex Jones radio shows and InfoWars news articles from 2016 to 2018. Radio show panels include Nationalism, Politicians, Show Slogans, Promotions, and Fake News; article panels include Trump, Scientific Discoveries and Controversies, Media and Politics, Attacking Democrats, and Global Conflicts. Smoothed trend lines show modest but noticeable changes over time, including persistent attention to Trump and politics in articles and nationalism and political messaging in radio content.

New from me, @yunkangyang.bsky.social @jolukito.bsky.social @jasong.bsky.social @m-dot-brown.bsky.social @beccalew.bsky.social: analyzing sales data released from InfoWar's court case, we find that certain styles (linguistic and auditory) used by Alex Jones on his radio show predict next-day sales

Style and substance on The Alex Jones Show predict InfoWars sales: a multi-modal analysis of a media empire Alex Jones, a prominent conspiracy theorist, has garnered substantial influence and wealth through his InfoWars media empire, which includes The Alex Jones Show and InfoWars.com. This study leverag...

Oh man, I've been waiting so long for this to be out in the world!

This is an incredible study. Must read.

www.tandfonline.com/doi/full/10....

If you're around Madison, come to Ryan's talk this Friday!

The first ever graduate seminar I ever took was basically a whole semester on Habermas, and it’s shaped my thinking about communication, media, and the public sphere since. RIP to a legend

Ross Dahlke presents in a lecture room, gesturing toward a slide titled “Ideological Information Environments” with bullet points comparing the open web and personal messaging.

Ross Dahlke points at a projected slide with two bar charts comparing “Web Browsing” vs. “Personal Messaging,” showing congruent vs. cross-cutting content for Democrats and Republicans.

Ross Dahlke speaks at the front of a conference room while a slide shows scatterplots of “Percentage of Messages by Type,” with colored dots representing different messaging platforms.

Ross Dahlke leads a small-group meeting in the Moeller Lab “Media Research” room, seated at a conference table with laptops while a wall screen displays a simple diagram about “Untrustworthy Websites.”

Thanks to University of Iowa School of Journalism and Mass Communication for hosting me for their Communication and Media Colloquium series and to Sang Jung Kim, @bingbingzhang.bsky.social and Jamil Marques for inviting me to your classes!

pc: @kaipingchen.bsky.social

Via this pipeline we have collected $37.8B in trading volume

Electoral prediction markets: a quantitative description

Two scatter plots showing the relationship between trades and comments. R = 0.28

Thanks to UW IT for inviting me to present my Polymarket data collection pipeline at Google Cloud Platform Research Day. Pipeline enabled by cloud computing: $40B in volume, 500M trades, and 1.25M comments with continuous collection. Preprints and public datasets coming soon!

Interesting postdoc position on AI and journalism with @tomasdodds.bsky.social and the @publictechmedialab.bsky.social at the UW-Madison School of Journalism and Mass Communication. Tomas is a good egg, so do apply 🥚

wisconsin.wd1.myworkdayjobs.com/UW_Madison/j...

Happy holidays from Lydia, Ross, and Boomer

Happy holidays from the Dahlke-McComas family! @lydiamccomas.bsky.social

New York post the red apple

Madison’s red mayor

Who did it better? Mamdani or Soglin?

Wisconsin official typifies a new era in election administration Lydia McComas, 28, decided in college that she wanted to work in elections, and focused her education around that goal. Veteran officials are looking to people like her as turnover accelerates.

So proud of my wife for starting as the new City Clerk of Madison, WI! www.votebeat.org/wisconsin/20...

Ross Dahlke. Speaker. Join me at the Trust & Safety Research Conference. September 25-26. Stanford University Alumni Center.

Image description Journal of Online Trust & Safety. Volume 3, Issue 1. September 2025. ISSN: 2770-3142.

Image description Vol. 3 No. 1 (2025) Untrustworthy Website Exposure and Election Beliefs: Selective Exposure and Ideological Asymmetry Peer-reviewed Articles https://doi.org/10.54501/jots.v3i1.250 Published 2025-09-12 Authors Ross Dahlke University of Wisconsin-Madison https://orcid.org/0000-0002-5179-2525 Jeffrey Hancock https://orcid.org/0000-0001-5367-2677 Keywords Digital trace data double machine learning data science false beliefs causal inference Categories Conference Proceedings How to Cite Dahlke, R., & Hancock, J. (2025). Untrustworthy Website Exposure and Election Beliefs: Selective Exposure and Ideological Asymmetry. Journal of Online Trust and Safety, 3(1). https://doi.org/10.54501/jots.v3i1.250

Image description Figure 1. Timeline of data collection.

I am excited to present as part of the #TSRConf 2025 Conference Proceedings of the Journal of Online Trust and Safety at @stanfordcyber.bsky.social. Happy to have this paper published doi.org/10.54501/jot...

Screenshot of a journal article titled “The power of alignment: how personalized information shapes voter decisions” by Nikandros Ioannidis in the Journal of Information Technology & Politics. The abstract summarizes a field experiment using a Voting Advice Application (VAA) during Cyprus’s 2021 elections, finding that personalized political information boosted participation by up to 10 percentage points, but did not meaningfully shift vote intention toward more ideologically congruent parties.

Image description Flowchart of the experimental design. Participants (N ≈ 17,000) were randomly assigned to one of five groups: control, Party Rankings, Map Eco-Social, Map Eco-CyProb, or Map CyProb-Social. All groups were asked about demographics, policy preferences, and past vote. Treatments received personalized VAA output and were later surveyed on vote intentions.

Image description Top panel shows number of VAA users per day from May 21–30, 2021, peaking on May 22. Bottom panel shows cumulative number of users, rising steadily across the same period. Below, a horizontal bar chart shows sample VAA output: seven parties with positive affinity scores (yellow bars) and two parties with negative affinity (red bars). Scores range from −37 to +40.

Image description Coefficient plot from probit models predicting election participation across four treatment arms: Party Rankings, Map Eco/Social, Map Eco/CyProb, and Map Cyprob/Social. All treatments show positive effects on turnout likelihood compared to control, with Party Rankings showing the largest and most precise effect.

In an experiment with ~4% of the electorate of Cyprus, personalized affinity information increased electoral participation and encouraged party consideration but did not shift voting intentions, finds Ioannidis doi.org/10.1080/1933...

Screenshot of a journal article titled “The urban–rural divide in policy priorities across time and space” by Yildirim and Solvig in Political Science Research and Methods. The abstract summarizes an analysis of 850 U.S. surveys (1939–2020) showing persistent but modest urban–rural differences in top policy concerns, with partisan identity more predictive than geography. Keywords include partisanship, geography, and public opinion.

Image description Six line graphs showing urban–rural gaps in prioritization of budget deficit, agriculture, moral values, immigration, economy, and tax issues from 1960 to 2020. Rural residents more often prioritize budget, agriculture, values, and immigration. Urban and rural trends on economy and tax converge more closely. Rural lines are generally above urban, indicating stronger issue salience.

Image description Six-panel plot showing how partisanship and residence jointly shape issue priorities over time (1960–2020). Rural Republicans rank budget, values, and immigration highest. Urban and rural Democrats differ little from each other and show lower prioritization of conservative-coded issues. Economy and tax are prioritized similarly across groups, with convergence over time. ⸻

Image description Six-panel plot showing partisanship × geography effects on civil rights, crime, education, foreign policy, health, and drugs (1960–2020). Urban/rural gaps are small relative to partisan gaps. Republicans (urban and rural) consistently prioritize crime and drugs more than Democrats. Democrats emphasize civil rights, education, and health more. Foreign policy converges by 2020.

While an urban-rural divide persists in policy priorities, partisan affiliation is a stronger predictor of priorities than geographic location, finds Yildirim & Solvig in @psrm.bsky.social doi.org/10.1017/psrm...

Screenshot of study abstract titled “Does threat increase conservatism?” summarizing three large U.S. experiments (Ns = 1000, 889, 843) testing threat effects on ideology. Despite successful threat manipulations, no effects are found on conservatism or personality × threat interactions. Concludes field should move beyond threat-based explanations.

Bar plots from three studies testing effects of threat on ideology. Each plot shows coefficient estimates for two threat conditions vs. control across ideological outcomes. Study 1 shows null effects on global ideology, healthcare, and economic policy. Study 2 adds measures like Right-Wing Authoritarianism (RWA) and Social Dominance Orientation (SDO), also showing no threat effects. Study 3 adds race threat and race policy; again, no consistent significant effects.

Two panels of coefficient plots from Studies 2 and 3 showing threat effects on specific policy attitudes (e.g., abortion, immigration, guns). Each dot represents a treatment effect estimate vs. control. Across dozens of items, no consistent ideological shifts emerge in response to unemployment, healthcare, or race threats.

Interaction plots from all three studies testing if personality (openness, conscientiousness) moderates threat effects on ideology. Across economic, global, and healthcare ideology—as well as RWA and SDO—no consistent threat × personality interactions emerge. One weak effect in Study 2 flagged for negative bias, but generally null.

Experimental manipulation of threat exposure has a null effect on ideological conservatism, finds @abbycassario.bsky.social
et al. osf.io/preprints/ps...

Candidates Be Posting: Multi-Platform Strategies and Partisan Preferences in the 2022 U.S. Midterm Elections - Josephine Lukito, Maggie Macdonald, Bin Chen, Megan A. Brown, Stephen Prochaska, Yunkang ... In this multi-platform, comparative study, we analyze social media messages from political candidates (N = 1,517) running for Congress during the 2022 U.S. Midt...

🚨New publication in Social Media + Society🚨

Candidates Be Posting: Multi-Platform Strategies and Partisan Preferences in the 2022 U.S. Midterm Elections

And it's open access!

journals.sagepub.com/doi/full/10....

Yeahhhh, I literally texted my wife during it that I was on a manel

World Salon Surveillance Capitalism: Who owns your data? www.world-salon.com

Excited to be on this panel discussing surveillance capitalism today!

CCCR logo, 100 Days Under Trump: Public Reactions to Attacks on American Governance & Institutions

🚨 CCCR has a new survey report out today! 🚨

"100 Days Under Trump: Public Reactions to Attacks on American Governance & Institutions"

The report draws on our Apr/May YouGov panel survey of US adults, following our Oct 2024 survey w/ recontacts + a sample refresh. 1/
cccr.wisc.edu/wp-content/u...

Excited to see the research brief option!

So many great Computational Methods sessions coming up at #ICA25!!

Check them out, and we look forward to seeing you there!
👇👇👇

AI is Perceived as Less Trustworthy and Less Effective when Using Emotional Arguments to Moderate Misinformation

Image description Figure 1. Perceived quality by moderator identity and argument type Three bar plots show perceived credibility, informativeness, and transparency as a function of moderator identity (human vs. AI) and argument type (rational vs. emotional). Rational arguments are rated significantly higher than emotional ones across all dimensions (p < .001), and human moderators are rated higher than AI within each argument type. Error bars reflect 95% confidence intervals.

Image description Two bar plots display mean perceived trustworthiness. The left panel shows human moderators are rated more trustworthy than AI (p = .004). The right panel shows rational arguments are rated as more trustworthy than emotional ones (p = .005). Error bars represent 95% confidence intervals.

Image description Figure 3. Efficacy across misinformation contexts by moderator and argument type Five panels show perceived efficacy of moderation across different misinformation scenarios (GMO, general, aligned beliefs), comparing human and AI moderators using rational vs. emotional arguments. Human moderators using rational arguments are rated as significantly more effective in several contexts (p < .05), with AI-emotional combinations rated lowest overall.

People negatively evaluate AI moderators that use emotional arguments rather than rational arguments, finds Silver, Williams-Ceci, & @informor.bsky.social doi.org/10.1145/3706...

Community Fact-Checks Trigger Moral Outrage in Replies to Misleading Posts on Social Media

Image description Figure 2. Summary statistics for misleading posts and replies before and after fact-check display Panel (a) shows a time series line chart of the rolling average number of misleading source posts with community notes from January to April 2023, trending upward over time. Panel (b) displays two horizontal bars comparing the proportion of positive vs. negative sentiment in source posts, with negative sentiment dominating. Panel (c) contains horizontal bars comparing six emotions (anger, disgust, fear, joy, sadness, surprise) in source posts, with anger and disgust most prevalent. Panels (d–f) show CCDFs: (d) total reply count per post, (e) post age at the time of community note display, and (f) the proportion of replies that occurred before vs. after the display. Panels (g–i) present line charts tracking hourly averages of reply sentiment/emotion (negative, anger, surprise) from 16 hours before to 16 hours after note display, showing modest increases after the note appears.

Image description Figure 4. Predicted effects of note display on sentiments and emotions in replies Panels (a–h) are separate line charts with scatter overlays, each representing one sentiment or emotion. The x-axis spans from -16 to +16 hours relative to the display of a community note. Each chart includes a blue line for pre-display averages, a yellow line for post-display averages, and a shaded 95% confidence interval around the predicted effect. Predicted increases are most visible in anger (c), disgust (d), and negative sentiment (b), with flat or minimal changes for joy (f), sadness (g), and positive sentiment (a).

Image description Figure 6. Regression estimates for reply sentiment and emotion after community note display Eight coefficient plots (panels a–h) showing the estimated effects of key predictors—including whether a note was displayed, post age, and source sentiment—on each type of sentiment or emotion in replies. The estimates are split by whether the source post was political or not, with red and blue error bars representing separate groups. Vertical lines represent the 95% confidence intervals around each coefficient. Displaying a community note is positively associated with anger, disgust, and negative sentiment, especially for political posts. ⸻

When Community Notes inform users about falsehoods on X posts, the replies to the post have more negativity, anger, distrust, and moral outrage, finds Chuai ‪et al. dl.acm.org/doi/10.1145/...

Problematic social media use is associated with believing in and engaging with fake news

Image description Figure: Interaction between problematic social media use and engagement with false vs. real content. Five panels (A–E) show line graphs with problematic social media use on the x-axis and outcome variables on the y-axis: credibility (A), intention to click (B), like (C), comment (D), and share (E). Each graph shows separate lines for false (blue dashed) and real (red solid) content. In panels A, B, and E, false content shows stronger increases in credibility and engagement as problematic use rises, with significant or marginal interaction effects. In panels C and D, both content types rise similarly, with only main effects significant. Results suggest that problematic social media use is associated with higher belief in and engagement with false news, more so than real news in some cases.

Higher problematic social media use is correlated with engaging false information online, finds Meshi & Molina t.co/NokaLxCGVn

Thinking Hard, Thinking Smart: How News Users’ Cognitive Traits Guide Their Responses to Fact-Checks

Image description A flowchart diagram illustrating a randomized controlled experiment with three arms: (1) a treatment group split into two subconditions—Forewarning and No-Forewarning—and (2) a Control group. • Participants in the treatment subgroups receive either a forewarning or a filler before viewing four news articles, each accompanied by a fact-check. • Control group participants receive filler content, followed by the same four news articles without fact-checks. • All participants complete a post-test questionnaire at the end.

Image description A bar chart with the y-axis labeled “Perceived Truthfulness” (range 0 to 6) and four x-axis groups representing combinations of two traits: Need for Cognition (NFC: high or low) and Cognitive Reflection (CR: high or low). • Each group contains two bars—dark gray for “Fact-check True” and light gray for “Fact-check False.” • In all groups, True-rated items are perceived as more truthful than False-rated ones, but the difference is largest for participants high in both NFC and CR. • Differences in perceived truthfulness between fact-check conditions diminish among participants low in NFC or CR.

Image description Four path models labeled (a) through (d), each showing a mediation analysis by trait group: (a) High NFC × High CR (b) High NFC × Low CR (c) Low NFC × High CR (d) Low NFC × Low CR • Each diagram includes three variables—Fact-check (0=False, 1=True), Perceived Truthfulness, and Sharing Intention—connected by arrows with regression coefficients. • In all four panels, Fact-check strongly predicts Perceived Truthfulness (significant in all groups). • Perceived Truthfulness significantly predicts Sharing Intention in all models. • Direct effects of Fact-check on Sharing Intention are only significant in panel (b), where a negative coefficient suggests that High NFC but Low CR participants reduce sharing when content is marked false.

Those high in need for cognition and cognitive reflection ability are more receptive to fact checks, finds Lee & Chung doi.org/10.1080/2167...

Candidate B-Roll as Super PAC Subsidy

Image description Table 1: Visual Resource Provision by Chamber as a Proportion of Total Races, 2018-2022 Cycles.

Image description Table 4, Breakdown of Advertising Employing Candidate/Party-Provided Visual Resources 2018-2020.

Image description Table 2, Division of Red Boxing/Visual Resource Permission by Candidate in Cycle 2018-2022.

In a political era of Super PACs, congressional candidates seek to maintain control over their visual image through visual "b-roll", effectively subsidizing outside organizations, finds ‪@gfoysutherland.bsky.social‬ doi.org/10.1177/1532...