Facebook weighted the angry reaction five times a like and then found angry posts carried more misinformation
Hook
From 2017 Facebook’s feed counted an angry reaction as five likes. Its own data scientists confirmed in 2019 that angry-reaction posts were more likely to carry misinformation and toxicity.
Claim
Nobody at the platform has to want outrage. A reward weight that scores reactions can produce it. (Counter: the weight was not only on “angry”, see below.)
Story
Source: Washington Post, 26 Oct 2021, “Five points for anger, one for a ‘like’”, by Jeremy B. Merrill and Will Oremus, based on internal documents given to the SEC and, in redacted form, to Congress by the lawyers of whistleblower Frances Haugen, a former Facebook project manager on civic misinformation (TechCrunch). I read the syndicated copy (Boston.com).
What the Post reported:
- 2017: when emoji reactions launched, the ranking algorithm treated each as five times a like (all of angry, love, haha, wow, sad, not only angry). Facebook had told Mashable in 2017 it weighted reactions “a little more than likes”.
- 2018: weight cut to four times a like.
- 2019: data scientists confirmed posts with angry reactions were disproportionately likely to include misinformation, toxicity and low-quality news. In April 2019 a mechanism to demote content with disproportionately angry reactions was added. A late-2019 proposal to set several reactions to 1x or 0x was dropped at the last minute.
- 2020: weight cut to 1.5 times a like. The Post says that angry reactions drove 429 million clicks a week against 63 billion for likes.
- “Last September” (the Post’s wording; read as September 2020 given the article date): angry set to zero, love and sad raised to 2x. After the change users saw less misinformation, “disturbing” content and graphic violence, and user activity was unaffected (per documents; Nieman Lab: “there wasn’t a trade-off after all”).
- Frances Haugen to British Parliament: “Anger and hate is the easiest way to grow on Facebook.” She testified to the US Senate commerce subcommittee on 5 Oct 2021 that engagement-based ranking amplifies content that draws strong reactions (TechCrunch).
What Facebook said:
- Spokesperson Dani Lever: “We continue to work to understand what content creates negative experiences, so we can reduce its distribution”, including content with “a disproportionate amount of angry reactions”.
- Lars Backstrom, vice president of engineering: “Like any optimization, there’s going to be some ways that it gets exploited or taken advantage of.”
- Mark Zuckerberg, 5 Oct 2021, on the whistleblower claims: “At the heart of these accusations is this idea that we prioritize profit over safety and well-being. That’s just not true.” (CBS). One outlet reports he called the idea that Facebook pushes angering content for profit “deeply illogical” (seen in search, not opened).
- Facebook renamed its parent company Meta on 28 Oct 2021, two days after the Post story (Facebook, Instagram and WhatsApp kept their names). Today the company is Meta; the feed in the story is Facebook’s.
Sources
- Washington Post via Boston.com, 26 Oct 2021: https://www.boston.com/news/technology/2021/10/26/five-points-for-anger-one-for-a-like-how-facebooks-formula-fostered-rage-and-misinformation/
- Nieman Lab, 26 Oct 2021: https://niemanlab.org/2021/10/more-internal-documents-show-how-facebooks-algorithm-prioritized-anger-and-posts-that-triggered-it
- The Hill, 26 Oct 2021: https://digital-release.thehill.com/?p=578548
- TechCrunch, Haugen Senate testimony, 5 Oct 2021: https://www.techcrunch.com/2021/10/05/facebook-whistleblower-frances-haugen-testifies-before-the-senate/amp
- CBS News, Zuckerberg response: https://www.cbsnews.com/amp/sanfrancisco/news/facebooks-mark-zuckerberg-fires-back-coverage-misrepresents-our-work-and-our-motives/
- Reason (rebuttal), 29 Oct 2021, Liz Wolfe: https://reason.com/2021/10/29/if-you-think-facebook-is-full-of-dubious-outrage-bait-wait-til-you-see-the-companys-critics
- Wikipedia was used only to find leads and is no longer cited.
Check before posting
- The Washington Post original (paywall) was not opened; the syndicated copy was. The Techdirt rebuttal (all five reactions weighted 5x, not only angry) and Meta/Clegg’s own statements were not opened (403 / not found).
- The Senate testimony transcript was not opened; only press summaries.
- Resolved 2026-11-04: names (Merrill, Oremus; Haugen; Lever; Backstrom), 5x in 2017, 4x in 2018, 1.5x in 2020, zero in September (the Post’s “last September”, year not stated in text), 2019 finding. The note’s “from 2017 angry = 5 likes” is true but applied to all reactions.
- Still open: whether the zero weight applied to all users or to a test group; the Post text does not say.
Counterargument
Two sides. (1) “It’s the incentives” can excuse choices: the company could have set the weight to zero earlier. (2) Pro-Meta rebuttal: the five-times weight applied to all reactions including love, which is far more common than angry; the algorithm rewards strong emotion in general; news media have similar incentives; the company tested and reduced the weight over time, and its staff warned about it (Reason, Techdirt). Meta also denies it pushes angering content for profit.
Post angles
Facebook (Croatian)
- Facebook je 2017. svaku reakciju, pa i “ljutito”, računao pet puta vrijednije od lajka. Interni podaci iz 2019. pokazali su da takve objave češće nose dezinformacije. Težina za “ljutito” spuštena je tek kasnije (na 1,5 pa na nulu). Nitko nije morao željeti bijes: dovoljno je da formula nagradi reakciju.
- Kako vi biramo što ćete čitati?
LinkedIn (English)
- Facebook’s feed once weighted every emoji reaction, “angry” included, at five likes. Internal data in 2019 showed angry-reaction posts carried more misinformation; the angry weight was cut over 2018-2020 and later set to zero. No villain required: a scoring rule is an incentive.
- Incentives explain, they do not excuse. The company could have set the weight to zero sooner.