Philine Widmer

As you noted, we find that the effects are driven by Republicans and Independents. We have about 5K participants in total, but the asymmetric design means the subsamples for each direction (switching on vs. off) are smaller, which limits our power to test for detailed treatment heterogeneity.

Vielleicht könnte unsere Studie (heute erschienen) zu den politischen Effekten von X für euch von Interesse sein:
www.nature.com/articles/s41...
bsky.app/profile/phin...

Read the full paper in Nature:

www.nature.com/articles/s41...

With @pinchofdata.bsky.social, Roland Hodler, and Ekaterina Zhuravskaya

@pse.bsky.social @bocconi.bsky.social @unisg.ch

We ran this experiment independently: no platform cooperation, no corporate funding. Instead, it was funded by the Swiss National Science Foundation SNSF.

Important nuances: the algorithm did NOT change party identification or affective polarization over the seven-week study period. It shifted policy views and attitudes on current events, not deep partisan identity.

This is a behavioral lock-in: once you follow new accounts promoted by the algorithm, they stay in your feed regardless of which feed setting you use.

Why this asymmetry? We also found that X's algorithm promotes political activist accounts (especially conservative ones) and demotes traditional news media. In response to algorithm exposure, users follow these accounts and continue to follow them even after switching to the chronological feed.

But here's the twist: switching the algorithm OFF did not reverse the effects. Users who went from algorithmic to chronological feeds showed no significant change in political attitudes. The effects of algorithm exposure are sticky.

The key finding is that switching from a chronological to an algorithmic feed shifted users toward more conservative positions: policy priorities, views on Trump’s criminal investigations, and attitudes toward the war in Ukraine all shifted to the right.

We recruited ~5,000 active X users in the US and randomly assigned them to use either the algorithmic feed ("For You") or the chronological feed ("Following") for seven weeks in summer 2023. We measured their political attitudes before and after with surveys.

Can feed algorithms shape what people think about politics? Our paper "The Political Effects of X's Feed Algorithm" is out today in Nature and answers "Yes."

www.nature.com/articles/s41...

@essobecker.bsky.social @elliottash.bsky.social

mpwz_cepr_schedule_15_16_september_2025 10th Monash–Paris–Warwick–Zurich–CEPR Text-As-Data Workshop ON ZOOM (https://ethz.zoom.us/j/61047338571) 15–16 September 2025 (online-only) All times in ET (US East) · CEST (Europe) · AEST (Australia)...

Join us NOW ONLINE to hear from Tarek Hassan and Anton Korinek -- discussing timely topics on AI! These are the concluding lectures of the 10th MPWZ-CEPR workshop (Zoom link in the program): docs.google.com/document/d/1...

Thanks to all presenters for the insights of the past 2 days :)

Registration is open for the 10th Text-As-Data Workshop. Join us online on 15-16 September for this special anniversary edition!
Organisers: @elliottash.bsky.social @essobecker.bsky.social @phinifa.bsky.social
Programme and Register: cepr.org/events/10th-...
#EconSky #EconConf

The 10th anniversary edition of our MPWZ-CEPR Text-as-Data workshop will feature the following special sessions 👇

🎉 Join us for the 10th anniversary of the MPWZ-CEPR Text-As-Data Workshop!
📅 September 15-16 | Online (Zoom link in program)
We're celebrating a decade of text-as-data research with a lot of research presentations and some special events: t.co/9q05IcW33b

Looking for panelists: editors at Econ journals

Are you an editor at an Econ journal? Would you like to share your thoughts & experience on AI in reviewing/publishing (e.g., AI-gen reviews)? The next MPWZ-CEPR conference (15-16 Sept, ONLINE) will host a panel on this. Please DM. #econtwitter

A keynote by Anton Korinek on AI in research, and additionally a panel on AI in the review process!

This edition will entail the usual presentations of papers in any field using text or other unstructured data PLUS ....

@elliottash.bsky.social @essobecker.bsky.social

Paper Submission: 10th Monash-Warwick-Zurich-CEPR Text-as-Data Workshop Submit papers by August 25th, 2025 AoE. The Workshop will be held on Zoom on September 15th-16th, 2025. We welcome submissions from all institutions and all fields in social science and data science u...

Last day to submit your work to the next edition of our text-as-data workshop series: docs.google.com/forms/d/1xay...
This is the ANNIVERSARY edition: the co-organizing Monash, Warwick, ETH Zurich, Paris School of Economics, and CEPR are thrilled to meet you at the 10th (!) edition!

Registration is open for the 9th Text-As-Data Workshop!
When: 28-29 April 2025
Where: Online
Organisers: @elliottash.bsky.social @essobecker.bsky.social @phinifa.bsky.social
More information: cepr.org/events/9th-m...
#EconSky #Workshop #Conference

For friends of text & images (or other unstructured data), submit your work until tomorrow to our workshop co-hosted with CEPR. It is online with slots for all timezones. The format has existed since 2021. Many drafts from earlier editions are now in great journals. Join the tradition :)

Are you using text, images, audio, or videos in your research? Submit the 9th edition of our online workshop!

The extended submission deadline is *15 March 2025*