YouTube's algorithm is *not* radicalizing people according to a study by leading scholars that examined 29 million YouTube viewing sessions and recently appeared in a prestigious peer reviewed journal: pnas.org/content/118/32…
There is more research/work to be done (especially with experimental designs and on other platforms), but this is the most comprehensive and careful analyses I've yet seen by @homahmrd@aaronclauset@duncanjwatts@markusmobius@DavMicRot and Amir Ghasemian.
And to be clear this does not indicate there are not major issues with extreme content on YouTube (and other platforms)... only that the algorithm does not seem to be the most likely culprit for producing extremism.
As many have argued (e.g. @kmmunger), we need more research on the supply side of extremism... and, I'd add, more careful theories of how our attempts to gain status interact with the technical affordances of our platforms and the information ecosystem more broadly.
And, more broadly, perhaps we all need a lot more humility vis-a-vis this complex, multifaceted issue. We are still very early in our understanding of how extremism evolves on social media, and there may not be many simple answers.
And to be clear I don't think this means that platforms could not do more to prevent/discourage extremism either; we all need to get more creative about how we think about the bigger picture and ask how the fundamental design features of social media engender different behavior.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
1/ Do you feel hopeless about political polarization on social media? Introducing a new suite of apps, bots, and other tools that you can use to make this place less polarizing from our Duke Polarization Lab: polarizationlab.com/our-tools
One of the biggest problems with social media is that it amplifies extremists and mutes moderates, leaving us all feeling more polarized than we really are. Our tools can help you avoid extremists and identify moderates with whom you might engage in more productive conversations.
The Duke Polarization Lab’s Bipartisanship Leaderboard identifies politicians, celebrities, activists, journalists, media outlets, and advocacy groups whose posts get likes from people in both parties: polarizationlab.com/bipartisanship…
Many have expressed skepticism about calls for healing and #depolarization over the past few days. But what does the latest research indicate about the prospects for reconciliation? Let’s look at the #SocSciResearch...1/9
An experiment that asked Democrats and Republicans to discuss politics in person for just 15 minutes improved their attitudes towards each other by *70 percent* compared to a control group. See @m_levendusky ‘s forthcoming book “Our Common Bonds” 2/9
Brief cross-party conversations seem to be effective even if they occur on an online chat platform created to pair partisans to discuss controversial topics (~9 point increase on a 0-100 point “feeling thermometer”. See @erinrossiter’s job market paper (then hire her ;) 3/9
1/n Are you a *complete beginner* in computational social science who wants to learn how to code? I'm happy to announce our new "coding bootcamp" video tutorials for the Summer Institutes in Computational Social Science: compsocialscience.github.io/summer-institu…
2/n I cover everything from setting up Rstudio to data cleaning (and "wrangling"), visualization, programming, modeling, communicating (w/Markdown, Rpres, and Shiny) as well as collaboration w/Github
3/n Though there are MANY great intro tutorials out there, this one is designed with computational social scientists in mind-- it offers examples w/ Twitter data, mobility data for COVID-19, and data about the opioid crisis.
1/n How do computational social scientists land non-academic jobs? I asked this question to a panel of senior leaders in for-profit and non-profit companies on a wonderful webinar yesterday, and I’d like to share what I learned:
2/n The cadence of non-academic work is very different. Academics like to take their time developing the perfect research design, but in other settings, people need answers, fast. Also, many academics are used to working alone, whereas most non-academic work is team-based.
3/n You need a good elevator pitch—one that makes it immediately clear how you can add value to a business or organization, BUT
1/n Did Russian trolls actually influence the attitudes and behaviors of U.S. social media users? Our Polarization Lab’s new article suggests the answer might be “no” pnas.org/content/early/…
2. Many people think Russian trolls exerted strong influence upon U.S. social media users because of the sheer scale and apparent sophistication of their techniques. There is also anecdotal evidence that IRA accounts succeeded in inspiring American activists to attend rallies.
3. Though many studies have analyzed the content and strategy of these campaigns, to our knowledge, no studies have examined whether they actually shaped the attitudes or behaviors of large groups of U.S. social media users.
1/5 Interested in learning how to collect and analyze social media data using topic models, text networks, or word2vec? I'm pleased to announce I am releasing an open source version of my "Text as Data" class from Duke's Data Science program: cbail.github.io/textasdata/Tex…
2/5 The course website (above) includes tutorials on a range of subjects with annotated R code. The class assumes basic knowledge of R and describes the techniques we use in the @polarization lab to run studies like this: pnas.org/content/115/37… and this: pnas.org/content/113/42…
3/5 All datasets used in the tutorials are hosted on my Github site, which also includes all source files for the tutorials themselves: github.com/cbail