Influence
Part 9  The Marketplace of Attention
Chapter 262 of 360

The Facebook Files: Knowing the Harm and Shipping Anyway

Frances Haugen was a product manager on Facebook's civic integrity team. Before leaving in May 2021 she copied tens of thousands of pages of internal research and gave them to the Wall Street Journal, to Congress, and to the Securities and Exchange Commission.

The documents are the most substantial internal record of a platform's own understanding of its effects that exists.

Two findings dominate.

The first concerns Instagram and adolescent girls. Internal research slides stated that thirty-two percent of teenage girls said that when they felt bad about their bodies, Instagram made them feel worse, and that among teenagers who reported suicidal ideation, a meaningful proportion traced it to Instagram. The company disputed the characterization and released some of the underlying material, arguing the studies were small and that many teenagers reported benefits. Both things can be true and the internal slides said what they said.

The second concerns ranking. In 2018 Facebook changed its algorithm to weight what it called meaningful social interaction, giving comments and reshares more weight than passive consumption. Internal analysis subsequently found that the change had increased the reach of divisive, angry and sensational content — because those are what produce comments and reshares. Political parties in several countries told Facebook directly that they had shifted their communications toward negative and polarizing material because that was what the system distributed.

The mechanism is not obscure and the company's own researchers stated it. Engagement-weighted ranking selects for whatever produces engagement, and what produces engagement in humans is high-arousal negative affect. Rozin and Royzman's work on negativity bias establishes the asymmetry: negative information is attended to more, weighted more heavily, and processed more thoroughly than equivalent positive information, because for most of evolutionary history missing a threat cost more than missing an opportunity.

For adolescents, upward social comparison — Festinger's mechanism from chapter 28's neighborhood — operates on a feed composed of other people's curated best moments, which is a comparison set with no ordinary life in it.

The scandal is epistemic rather than technical. The harm was measured internally, by the company's own researchers, using the company's own data, and the ranking was retained.

Haugen testified before the Senate on 5 October 2021.

The case

Frances Haugen’s disclosure of internal Meta research to the Wall Street Journal and US Congress in September-October 2021, including studies on Instagram’s effects on teenage girls and on engagement ranking that amplified outrage; she testified before the Senate on 5 October 2021.

The mechanism

Internal research described a mechanism the company had built deliberately: engagement-weighted ranking selects for high-arousal negative affect, because anger and social comparison produce more interaction — consistent with the negativity bias documented by Paul Rozin and Edward Royzman. For adolescents, upward social comparison in Festinger’s sense compounds body-image distress, producing measurable self-harm risk. The scandal’s core is epistemic: the harm was measured internally and the ranking retained.

What this chapter covers

  1. A product manager copies the research
  2. Engagement ranking selects for outrage
  3. Documents leaked and Senate testimony, October 2021
  4. Teen users report comparison-driven distress
  5. Whistleblower publishes the internal studies

Cross-ref: Attack — Reward Unpredictably