The mismeasure of misinformation
Social media giants keep trotting out jaw-dropping stats about fake accounts and rule-violating posts they're removing. But the number that matters most is how much misinformation remains up.
Driving the news: CEOs of Facebook, Google and Twitter will recite more numbers Thursday, when their CEOs testify at a marquee hearing before a Congressional committee investigating online misinformation.
The big picture: These companies report takedown numbers to show us they're working hard to police their digital precincts. Critics counter that the size of the numbers tells us that the problem remains out of control.
Either way, this data is impossible to assess without some measure of the total universe of misinformation from which the companies' enforcement actions are subtracting.
- Even after the platforms have removed mountains of posts and accounts, we know there's still some volume of misinformation online — whether it's conspiracy theories, anti-vaccination propaganda, or false election claims — because reporters keep unearthing examples. But we don't know exactly how much.
- Researchers and third parties — like Avaaz, a nonprofit that released a report this week on election-related misinformation that Facebook disputes — try to fill that gap, but they don't have the same access to realtime data as the company.
Why it matters: Without knowing the total extent of social media misinformation and how much is left after the platforms' enforcement, all the takedown data in the world won't tell you whether the platforms are winning or losing the fight.
Here's why this data is so elusive:
First, we want to put a number on posts the companies missed in their enforcement efforts — but the companies can't count what they failed to catch.
- In its regular transparency reports, Facebook does report what it calls a "prevalence rate" for different kinds of content violations — a sort of "parts per million" measure, like we use for air pollution, that estimates how often readers are likely to encounter such content.
- "Prevalence" is a fascinating metric that gives us a sense of the relative volume of different kinds of problem posts, and over time it can offer some sense of which kinds Facebook is making headway against.
- But the categories themselves — including "Adult Nudity and Sexual Activity," "Bullying and Harassment," "Dangerous Organizations: Organized Hate," and 9 others — are limited.
This is where a second problem emerges: "Misinformation" itself isn't one of Facebook's categories, everyone defines it differently, and the issues where the public, politicians and the press focus their outrage keep evolving.
- False claims about election results and false information about COVID vaccines are the two likeliest kinds of misinformation to be on the committee's agenda Thursday.
- Yet neither of those types fits into any of the categories of content violation that Facebook has tracked in its past transparency reports.
- By the time the company does start tracking them, we may well face another crisis involving a new kind of misinformation problem.
Between the lines: Even when a misinformation category is well-defined and enforcement is effective, social media platforms face a tough challenge with so-called "borderline content" — posts that don't outright violate a service's rules but still raise questions about misinformation-rife realms like election fraud or vaccines.
- Removing such posts raises free-speech questions, but leaving them untouched can put the public in danger. Those concerns motivated Facebook's recent efforts to study and counter vaccine hesitancy, as reported in the Washington Post.
Of note: Researchers — including that Facebook study — keep finding that social media misinformation with high impact gets spread by a relatively small number of users and groups. Solving the problem will likely involve finding ways to act against them.