Internal emails: the data Facebook provided to researchers studying misinformation since 2019 included about half of its US users, not all as it then claimed
More than three years ago, Mark Zuckerberg of Facebook trumpeted a plan to share data with researchers about how people interacted …
New York TimesDavey Alba
Context & Ripple Effects
This lands in the middle of Facebook's long disclosure-accuracy problem: from the Cambridge Analytica episode, where its own accounting of exposed users grew from roughly 50M to 87M, to internal docs showing it gave ~150 companies broader data access than disclosed while saying it hadn't violated its FTC decree. Zuckerberg publicly framed the 2019 researcher-data program as full transparency into how people interacted on the platform.
The internal emails reported here undercut that framing — the corpus covering misinformation research since 2019 reached about half of US users, not all. It also matters because Facebook's own researchers found US users are four times more likely to see misinformation via a share of a share, meaning what falls outside the shared half of the dataset is precisely where the platform's own work says virality concentrates.
First-order effects
Researchers who built findings on the shared dataset were working with about half of US users while believing they had all of them, so conclusions drawn under the program carry an unexamined coverage gap.
Facebook's claim of having provided all users' data is directly contradicted, handing its critics and oversight partners concrete evidence that its self-reported transparency has been overstated.
Second-order effects
Regulators and the FTC — already the reference point after the ~150-companies disclosure — have grounds to treat Facebook's representations about data programs as requiring independent verification rather than company attestation.
Academic partners and rival platforms face pressure to audit and disclose the actual population coverage behind their own research-access programs, since 'full data' claims are now demonstrably contestable.
Third-order effects
If the pattern holds — companies defining their own transparency metrics, internal records later contradicting them — the likely structural response is mandated third-party auditor access to platform data, shifting research legitimacy from corporate disclosure to externally verified datasets.
The trend: Platform transparency is being adjudicated less by company statements and more by leaked internal records, forcing a shift toward independently audited data access.
New (w/ @RMac18 assist): More than three years ago, Mark Zuckerberg unveiled a plan to share Facebook data with outside researchers, so that academics could study misinformation on the site. But the data set had a major flaw. https://www.nytimes.com/...
It's also interesting how this flaw was discovered in the first place. @fabiogiglietto took Facebook's shelved Q1 report on Widely Viewed Content, then compared it to the researchers data set. That's when he found a huge discrepancy. https://www.nytimes.com/... https://twitter.co…
It's a bit inside-baseball academia, but a lot of work has been trashed because of this mistake. One researcher expressed concern that their doctoral students' degrees could be at risk. https://www.nytimes.com/... https://twitter.com/...
The data included the interactions of about 50% of Facebook's U.S. users, not 100% of them as the company had described the data set would include. Researchers have been working with this data set for TWO YEARS, getting academic research papers ready. https://www.nytimes.com/...
As one of the research teams working on this data... this isn't the first error in the data, but this one is egregious. How do we build research on top of these opaque data — having lost trust in their curation? https://twitter.com/...
“Facebook sent flawed data to misinformation researchers”. “The data included the interactions of only about half of Facebook's U.S. users ... not all of them as the company had said” What a f*ck-up this is, and really good @fabiogiglietto spotted it. https://www.nytimes.com/...
“One researcher said doctoral degrees were at risk because of the mistake, while another expressed concern that Facebook was either negligent or, worse, actively undermining the research.” https://twitter.com/...
And either way, Facebook culture will draw the lesson that trying to share accountability data only for them criticized, so they should even *try* to be accountable in the future. They'll see any work with credible researchers as a PR risk.
Another Friday Facebook news dump: Researchers who relied on FB-provided data to study user behaviors were told this week the underlying data they were using was flawed. This will have significant impact on research about the platform. https://www.nytimes.com/...
When called in front of congress 3 years ago, Zuck tried to play-act accountability by saying they'd share key data with researchers. But FB made a fundamental error in gathering that data. https://www.nytimes.com/... Since then all the links to the original announcement are dead…
This thread is WILD. And it is case in point against closed datasets. “It's a great demonstration that even a little transparency can provide amazing results.” @fabiogiglietto, the researcher who found the bug. https://twitter.com/...
When Facebook makes a data error — like this, or the one about how many people were watching video streams — it materially impacts the livelihoods of so many people, because Facebook is such a powerful distributor that industries and jobs rely on it. This is awful. https://twitte…
This is bonkers. Is Facebook even capable of handling the monster it created? Example after example suggests no. I am so sorry for all the researchers whose work is harmed here ↘️ https://twitter.com/...
Wait. Wut. We knew there was risk in Facebook providing the data but they only included the posts of what were likely the most partisan users (those whose ideologies could be easily understood)????? That's absolutely nuts and damages countless academic projects and scrutiny. http…
So Facebook's plan to release alternative data to push back on a Twitter account they didn't like accidentally revealed an error that invalidated years' worth of academic research. Can't make it up! https://twitter.com/...
Facebook: Researchers should use data Facebook provides. Research using independently collected data is inaccurate and misleading about the platform. Also Facebook: https://twitter.com/...
A very good example of why the people who work in this world should be highly skeptical of FB's various attempts to leverage their credibility to create a certified field of “disinformation” https://www.nytimes.com/...
Either Facebook is too incompetent internally to share the data necessary to enforce a bare minimum of accountability, or Facebook is so corrupt that they undermined the data-sharing process, knowing it would be years before the truth came out. Which is it?
There's a reason data professionals live and die by the expression “garbage in, garbage out”. Unfortunately Facebook continues to exemplify the “garbage in” part. This was a heartbreaking call to witness today, just shocking for early PhDs, and heck, late career colleagues too ht…
There's a lot of trust that FB needs to win back from researchers, journalists and the wider public. But hopefully they end up understanding that MORE transparency leads to outcomes like this where they can make the data they share better and more accurate https://www.nytimes.com…
Facebook withdrew half of their data without telling researches who study misinformation. FB took out *all* posts of users without (yet) clear political leanings:That's where you'd expect the biggest impact of misinformation! But hey, they've now got glasses we'll put on our face…
This is such a Facebook story. In this case it might not even be them being shady (which is the default) but them doing the kind of unprofessional work one sees all over their products. https://twitter.com/...
One reason independent research of the platforms is crucial is to allow researchers to verify that the data they're getting is reliable. Turns out the concern is not hypothetical — for two years, researchers have been using data from FB that is flawed. https://www.nytimes.com/...