iiinova — where your digital self keeps evolving

Megan Brown @m-dot-brown.bsky.social · 10/06/2026

9/ The broader implication: the concern that recommendation algorithms are radicalization engines appears overstated, at least as of fall 2020. The subtler effect is a platform nudging all users toward ideological moderation while deepening content homogeneity over time.

222

Megan Brown @m-dot-brown.bsky.social · 10/06/2026

8/ Beyond YouTube, our audit methodology is designed to be extensible. The same framework can be applied to TikTok's For You Page, Facebook's News Feed, and X's Home Timeline — any recommendation system of interest to scholars studying the consequences of platform design.

212

Megan Brown @m-dot-brown.bsky.social · 10/06/2026

7/ Critically, we find no evidence of radicalization pathways — the theorized process by which echo chambers and rabbit holes together steer users toward more extreme content. The popular narrative that YouTube systematically radicalizes its users is not supported by our data.

100

Megan Brown @m-dot-brown.bsky.social · 10/06/2026

6/ Finding 3: We find strong evidence of content rabbit holes. The algorithm's primary function appears to be maximizing continued engagement by recommending content similar to what the user is currently watching, producing an increasingly narrow content loop.

100

Megan Brown @m-dot-brown.bsky.social · 10/06/2026

5/ Finding 2: We do find a platform-wide bias toward moderately conservative content — across users of all ideological starting points. Importantly, however, this nudge is toward the moderate right, not toward the extreme content that dominates public concern.

100

Megan Brown @m-dot-brown.bsky.social · 10/06/2026

4/ Finding 1: We find only limited evidence that YouTube's recommendation algorithm pushes users into ideological echo chambers. Where echo chamber effects do appear, they are better explained by user self-selection than by platform recommendations.

200

Megan Brown @m-dot-brown.bsky.social · 10/06/2026

3/ Our solution: an audit study fielded in fall 2020 with real U.S.-based YouTube users, where we manipulated how users select recommended videos. This lets us hold user behavior constant and isolate the independent effect of the algorithm on downstream recommendations.

100

Megan Brown @m-dot-brown.bsky.social · 10/06/2026

2/ Prior research has struggled to answer this cleanly. Watch history data can't disentangle user self-selection from platform effects. Anonymous scraping methods strip out the prior behavioral data the algorithm actually relies on, undermining construct validity.

100

Megan Brown @m-dot-brown.bsky.social · 10/06/2026

2/ We define three concepts: echo chambers (ideologically homogeneous recommendations), rabbit holes (increasingly niche content), and radicalization pathways (the two combined, pushing users to extremes).

100

Megan Brown @m-dot-brown.bsky.social · 10/06/2026

1/ Our new paper in Political Communication asks: does YouTube's algorithm push users into echo chambers, rabbit holes, or radicalization pathways? 🧵

120

Reposted by Megan Brown

Ulrike Klinger @ulrikeklinger.bsky.social · 04/06/2026

Wanna learn how the new #DSA EU data access to platform works for you, even outside Europe? Join our #ICA26 Bluesky workshop on Sat June 6 at 9:00 am in CTICC1, Lev 1, 1.52 with @claesdevreese.bsky.social @dscheykopp.bsky.social @jolukito.bsky.social @kwelle.bsky.social Daniel Angus, Megan Brown

0166

Reposted by Megan Brown

Jakob Ohme @dscheykopp.bsky.social · 04/06/2026

🚨 Did you know that researchers can now apply for access to platform data under the EU’s Digital Services Act? Join our #ICA26 Bluesky Workshop on Saturday (9:00–10:15) to learn how DSA data access works in practice, what infrastructures are available, and how to get started. #DSA40

0103

Reposted by Megan Brown

Ross Dahlke @rossdahlke.bsky.social · 30/03/2026

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

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.

23612

Reposted by Megan Brown

Michael Zimmer @michaelzimmer.bsky.social · 20/11/2025

Really exicted to finally see our paper in print: “Web scraping for research: Legal, ethical, institutional, and scientific considerations” A great interdisciplinary effort with @m-dot-brown.bsky.social, @orangechair.org, Gabe Maldoff, @solmg.bsky.social, @zevesanderson.com doi.org/10.1177/2053...

doi.org

Sage Journals: Discover world-class research

Subscription and open access journals from Sage, the world's leading independent academic publisher.

184

Megan Brown @m-dot-brown.bsky.social · 20/11/2025

Grateful for the collaboration with excellent researchers for this paper @orangechair.org, Gabe Maldoff, @solmg.bsky.social, @zevesanderson.com, & @michaelzimmer.bsky.social

011

Megan Brown @m-dot-brown.bsky.social · 20/11/2025

Thrilled to have a new article published in @bigdatasoc.bsky.social! 🥳🎉📊 With scraping becoming a more common data collection strategy for internet researchers, we cover the legal, ethical, institutional, and scientific ramifications researchers should consider. doi.org/10.1177/2053...

doi.org

Sage Journals: Discover world-class research

Subscription and open access journals from Sage, the world's leading independent academic publisher.

182

Megan Brown @m-dot-brown.bsky.social · 11/07/2025

6/ This has implications for both research and industry: ⚠️ Fairness evaluations should be context-specific 🤔 Model choice alone will not solve bias 🔍 Human disagreements are part of the complexity—not a flaw

120

Megan Brown @m-dot-brown.bsky.social · 11/07/2025

5/ We also find that the difficulty of the labeling task is most predictive of LLM agreement with human annotators.

110

Megan Brown @m-dot-brown.bsky.social · 11/07/2025

4/ Key finding: While LLMs disagree with human annotators on the basis of demographics, it tends to be in the same directions on the same demographic categories within the same dataset. In other words, the direction of bias is not LLM-specific, but dataset-specific.

120

Megan Brown @m-dot-brown.bsky.social · 11/07/2025

3/ This study evaluates LLM annotations across 4 datasets and tasks, analyzing whether these models disproportionately reflect majority group opinions.

110

Megan Brown @m-dot-brown.bsky.social · 11/07/2025

2/ Prior research praises LLMs for their high accuracy, precision, recall, and F1 scores in labeling tasks—but also raises concerns about bias, especially around sensitive or polarizing content (e.g., toxicity).

110

Megan Brown @m-dot-brown.bsky.social · 11/07/2025

Can large language models (LLMs) fairly annotate data on contentious topics? Our new paper dives into this question—looking at whether LLM-generated labels reflect diverse viewpoints or skew toward majority perspectives. The results are surprisingly nuanced. 🧵

1184

Reposted by Megan Brown

Hannah Waight @hwaight.bsky.social · 18/06/2025

I am thrilled to share a new article in Sociological Methods & Research, “Quantifying Narrative Similarity Across Languages”. My co-first author Sol Messing and our collaborators developed a new approach to measuring “narrative similarity” between texts: journals.sagepub.com/doi/10.1177/...

journals.sagepub.com

Quantifying Narrative Similarity Across Languages - Hannah Waight, Solomon Messing, Anton Shirikov, Margaret E. Roberts, Jonathan Nagler, Jason Greenfield, Megan A. Brown, Kevin Aslett, Joshua A. Tuck...

How can one understand the spread of ideas across text data? This is a key measurement problem in sociological inquiry, from the study of how interest groups sh...

35827

Reposted by Megan Brown

Kevin Yang @yang3kc.bsky.social · 19/05/2025

[New WP] With the closure of major social media APIs and the new data access mandates under DSA, we enter what we call the "post-post-API" era. But have researchers obtained the data they need? Our recent survey (180) + interview (19) study suggests a stark reality. 🔗 arxiv.org/abs/2505.09877 1/3

1215

Reposted by Megan Brown

Jo(sephine) Lukito @jolukito.bsky.social · 19/05/2025

So thrilled to have worked on this important piece with @yang3kc.bsky.social @m-dot-brown.bsky.social and Kayo Mimizuka. Data access for independent researchers is at such a critical juncture

032

Reposted by Megan Brown

Maggie Macdonald, PhD @mgmacdonald.bsky.social · 17/03/2025

Special thanks to Mango Brown and Taylor Swift's 'Mr. Perfectly Fine' for their help in getting this paper over the finish line

011

Reposted by Megan Brown

Maggie Macdonald, PhD @mgmacdonald.bsky.social · 17/03/2025

So excited to finally see this out! It was the first paper I started during my postdoc at @csmapnyu.org

194

Reposted by Megan Brown

NYU's Center for Social Media, AI, and Politics @csmapnyu.org · 17/03/2025

It's well known that politicians take more extreme positions during primaries. In @electoralstudies.bsky.social, we find this shift is much more likely when incumbents in safe seats face a well-funded primary challenger. 🧵👇 authors.elsevier.com/a/1kn5KxRaZn...

176

Reposted by Megan Brown

Patrick Wu @patrickwu.bsky.social · 05/03/2025

@adambonica.bsky.social showed ideology predicts which agencies experience DOGE layoffs. But what other factors could be driving this? Using a generative LLM-derived measure, I find agencies perceived as knowledge institutions are more likely to experience layoffs, even controlling for ideology. 🧵

Scatterplot showing various U.S. government agencies plotted with the total staff (on a log scale) on y-axis versus likelihood of being perceived as a knowledge institution on the x-axis. Red dots indicate agencies that have experienced DOGE layoffs, while gray dots indicate agencies without layoffs. Agencies like NIH, NSF, CDC, and NOAA appear on the right side (more likely to be perceived as knowledge institutions), while agencies like ICE, DEA, and Secret Service appear on the left side (less likely to be perceived as knowledge institutions).

612859

Reposted by Megan Brown

Jo(sephine) Lukito @jolukito.bsky.social · 13/01/2025

🚨New Publication in New Media & Society🚨 Co-First-Authored w/ @m-dot-brown.bsky.social @meredithpruden.bsky.social & @markriedl.bsky.social Making academia suck less: Supporting early career researchers studying harmful content online through a feminist ethics of care. jlukito.com/s/brown-et-a...

Cover for Making academia suck less: Supporting early career researchers studying harmful content online through a
feminist ethics of care

613635

Reposted by Megan Brown

Rebekah Tromble @rebekahtromble.bsky.social · 07/11/2024

They don’t need an excuse. They’ll claim we gave them an excuse no matter what we do. “Be careful what you say” is precisely how authoritarians achieve compliance w/o lifting a finger. And yet, even w/ compliance, they will still attack. Instead, may I suggest taking a look at researchersupport.org

researchersupport.org

Researcher Support Consortium

326976

Reposted by Megan Brown

Rebekah Tromble @rebekahtromble.bsky.social · 14/03/2024

I’ll keep working with academics, civil society researchers, and journalists—including via the Coalition for Independent Technology Research—to continue these important accountability efforts. 12/ (independenttechresearch.org/introducing-...)

independenttechresearch.org

Introducing the DSA Data Access Audit - Coalition for Independent Technology Research

In a year of more than 70 global elections, the need for independent researchers to have access to social media data has never been greater. That’s why the Coalition for Independent Technology Researc...

044

Reposted by Megan Brown

NYU's Center for Social Media, AI, and Politics @csmapnyu.org · 15/02/2024

🚨Our new paper in Political Analysis presents a novel, cross-platform method for estimating the ideology of YouTube videos. What we found: it is possible to do this at scale with an efficient, automated method! 🧵1/

11612