Back to the directoryFind in the timeline
Bakshy, Messing and Adamic on news exposure on Facebook
Bakshy, Messing and Adamic (2015), described as Facebook data science researchers, used Facebook data on 10.1 million US users to compare friends, News Feed ranking and clicks as limits on cross-cutting news. Critics objected to the sample and to treating choice and ranking as rivals.
- Where it sits
- Academic psychologyResearch published for other researchers and students.
- Evidence level
- ContestedThe evidence conflicts or is disputed, for example after failed replications.
- Organizations
- Facebook (Meta)An organization is listed where the sources concern it. The text says exactly how.
- Sources inspected
- 5 of 54 with the relevant text, company document or archive item checked, 1 as an abstract or reference article, 0 not verified.
What happened
In May 2015 Science published a paper by Eytan Bakshy, Solomon Messing and Lada Adamic on how much news that crosses ideological lines reaches Facebook users. The study used de-identified data from 10.1 million US users who report an ideological affiliation, for six months from July 2014 to January 2015. Press coverage describes the authors as Facebook data science researchers, and their own account of the work sits on Facebook's research blog.
The psychological idea
The design compares stages of exposure. Ideology comes from self-reported affiliation on a five-point scale, and an article's score is the average affiliation of the users who shared it, so the blog says it measures the alignment of the audience, not bias. The authors compared the share of cross-cutting hard news among what friends shared, what News Feed showed and what users clicked.
In practice
As reported on the authors' company blog, on average 23 per cent of users' friends claim the opposing ideology. Of hard news friends shared, 29.5 per cent crossed ideological lines, 28.5 per cent of what appeared in News Feed did, and 24.9 per cent of what users clicked did. The authors conclude that network composition matters most, individual choice also matters, and ranking has a smaller effect.
Why it mattered
The study entered the filter bubble debate, and critics responded within days of publication. It is an example of a platform's own researchers publishing evidence about the platform's algorithm in a major journal.
Limits and qualifications
The authors worked for the company whose algorithm they assessed, and the figures quoted here come from their company blog, so this is not an independent audit. Fortune reported critics saying the sample was under 4 per cent of Facebook's user base; the blog says 9 per cent of US users self-report an affiliation. Sandvig and Tufekci argued that choice and ranking cannot be compared as rivals. Jurgenson argued they are additive. Facebook said individual choices matter more than algorithms.
Evidence behind the claims
The study analyzed de-identified data from 10.1 million US Facebook users and concluded that individuals' own choices limited exposure to cross-cutting content more than algorithmic ranking did.
On average 23 per cent of friends claim the opposing ideology; 29.5 per cent of hard news shared by friends, 28.5 per cent in News Feed and 24.9 per cent of clicked news was cross-cutting.
Data covered July 2014 to January 2015; ideology was self-reported on a five-point scale, article scores averaged sharers' affiliations and measure audience alignment, and the blog says 9 per cent of US users self-report an affiliation; the authors conclude network composition matters most and ranking has a smaller impact.
Journalist's Resource describes Bakshy, Messing and Adamic as Facebook data science researchers (Adamic also affiliated with the University of Michigan), and Cyborgology says all three are Facebook researchers; the blog is hosted on Facebook's research site.
Fortune reports critics saying the study covered under 4 per cent of Facebook's user base, that Sandvig called it the 'not our fault' study, that Tufekci called the comparison apples to oranges, and that Pariser said only Facebook-employed researchers could access the data.
Facebook's response, as reported by Fortune, was that the study establishes that individual choices matter more than algorithms.
Jurgenson, writing on Cyborgology, argued that the sample covered only users who list political views and that user choice and algorithmic filtering are additive and shape each other, not separable alternatives.
People involved
- Eytan Bakshy. Lead author, described as a Facebook data science researcher
- Solomon Messing. Co-author, described as a Facebook data science researcher
- Lada Adamic. Co-author, described as a Facebook data science researcher and affiliated with the University of Michigan
- Christian Sandvig. Critic, University of Michigan
- Zeynep Tufekci. Critic, sociologist
- Nathan Jurgenson. Critic, writer on social media
In the Psy101 course
- OpenStax Psychology 2e, section 12.3: Attitudes and PersuasionChapter 12, Social Psychology. Context. Nearest related section; it does not discuss this topic by name.Nearest related section: it covers persuasion and attitude change, not selective exposure to news or feed ranking.
- OpenStax Psychology 2e, section 2.2: Approaches to ResearchChapter 2, Psychological Research. Method. The section discusses this topic.The section covers correlational research and what it cannot show about cause and effect, relevant to an observational exposure comparison.
This is course relevance only. It is not evidence for the historical or corporate claims on this page.See the full course map.
Sources and inspection
5 of 5 sources were opened and read at the level shown. A source marked as not verified supports no claim on this page by itself.
- [1]Exposure to diverse information on FacebookBakshy, E; Messing, S; Adamic, L. Meta (Facebook) Research blog, 2015. Company document. Company document.Company document readRead the authors' company blog post as returned by the fetch tool: percentages, sample, ideology measure, conclusions. The page does not state author affiliations or the 10.1 million figure. Company self-report, not an independent audit.
- [2]Exposure to ideologically diverse news and opinion on FacebookBakshy, E; Messing, S; Adamic, L. Science (record on SSRC MediaWell), 2015. Journal article. Primary source.Abstract or publisher record onlyRead the abstract as summarized on MediaWell. The Science page returned 403, so the paper itself was not read.
- [3]Facebook filter bubble doubtsFortune. Fortune, 2015. News report. Critique or replication.Relevant text checkedRead the article as returned by the fetch tool. It reports the critiques and Facebook's response; it does not say whether the authors work for Facebook.
- [4]Cyborgology post on the Facebook news feed studyJurgenson, N. Cyborgology (The Society Pages), post 19947, 2015. News report. Critique or replication.Relevant text checkedRead the post as returned by the fetch tool (a JSON endpoint of the blog). It is signed by Nathan Jurgenson, who states he works for Snapchat, and calls the authors all Facebook researchers. Date not confirmed.
- [5]Journalist's Resource expert commentary on the Facebook studyJournalist's Resource. Journalist's Resource, Harvard Kennedy School, 2015. News report. Secondary account.Relevant text checkedRead the article as returned by the fetch tool: author descriptions, the small fraction of users with a self-reported affiliation, and expert comments.
Access note: The Science paper returned 403 and was not read; the figures come from the authors' company blog and a MediaWell abstract summary. Sandvig's, Tufekci's and Pariser's original posts were not opened.
Independent source review: Claude review agent, independent of the research agent, 2026-10-21. Verdict: pass-with-edits.