View profile

Facebook vs. Independent Research

Dispatches from our founder
This Week
Hello, friends,
This week Facebook took the aggressive step of disabling the Facebook accounts of independent researchers at New York University who were studying ads on its platform, claiming that the project violated user privacy. This move sparked an immediate outcry as everyone from senators to members of the academic community called out the company’s hypocrisy and lack of transparency. 
And, of course, for us at The Markup, it’s personal. Not only is our Citizen Browser project similarly providing public accountability for Facebook’s algorithms, but the NYU Ad Observer project was an outgrowth of work that I did with my team at ProPublica. 
It all started back in 2017, after the presidential election, when everyone was trying to figure out whether President Trump’s Facebook ads might have been the dark horse that decided the election. At that time, there was no way to see political ads—or any other ads—on Facebook unless they were shown to you in your own personalized News Feed.
It seemed to me that political advertising should be viewable by the public. So while drinking beer and eating bratwurst at a data journalism conference in Hamburg, Germany, with my then-colleague Jeff Larson, I had the insight that we should build a way for Facebook users to voluntarily share their ads with us. 
We sold some of our German colleagues on the idea, and when we came back to New York we got to work building a browser extension to offer to the public. We launched the tool in September 2017. It collected the ads from users’ News Feeds and used machine learning to classify them as political or nonpolitical. The political ads were available in a public database.
The tool allowed our German reporter friends to discover that Facebook had allowed political ads whose origins were untraceable to appear on its platform, and for us to reveal that in the U.S., Facebook had allowed political ads that were actually scams and contained malware, among many other insights.
But in 2019, after I had left ProPublica, Facebook blocked that publication’s ability to collect ad targeting information from the tool. NYU researchers eventually agreed to take over the project and expand upon it. They launched their new tool—which collects both Facebook and YouTube ads—in March 2020 while still maintaining the original tool installed by users of the ProPublica extension.
By 2020, things had changed. Facebook provided more transparency into political ads, offering a publicly available library of them. But it didn’t and still doesn’t offer visibility into the crucial ad targeting information that allows advertisers to choose who sees their ads. The NYU researchers have worked to combine Facebook’s public data with the targeting data they received from the users of their crowdsourcing tools.
Strangely, in this recent shutdown, Facebook left the crowdsourcing tools untouched but shut down the NYU researchers’ personal Facebook accounts, which essentially prevented them from using Facebook’s publicly available research tools. At The Markup, we released a public statement decrying Facebook’s actions and stating our continued commitment to independent accountability journalism.
To understand more of what happened with Facebook and NYU, I spoke with Laura Edelson, a Ph.D. candidate in computer science at New York University Tandon School of Engineering and the lead researcher behind NYU Cybersecurity for Democracy, which operates Ad Observer and Ad Observatory, a site for the public to explore trends in Facebook advertising.
The following interview has been edited for brevity and clarity.
Laura Edelson
Laura Edelson
Angwin: So, you’ve had a busy week. It’s confusing to me that Facebook didn’t take action against the Ad Observer browser extensions that you offer to users who share data with you but instead acted against you personally.   
Edelson: It doesn’t make sense to me that Facebook’s goal in taking these actions is about protecting user privacy. If Facebook honestly thought that our browser extension compromised user privacy in any way, they would have taken action to stop our browser extension. They would have sued us. They would have taken technological measures. They would have tried to get our extension kicked out of the Chrome and Firefox store. 
If this was really about user privacy to Facebook, they would have to take some of those steps, but they haven’t. Instead they’ve attacked our ability to do other kinds of research about Facebook. So if Facebook is saying this is about user privacy, then its actions just don’t line up.
Angwin: Instead they disabled your accounts that you use to access their ad library and CrowdTangle. What is that data used for?
Edelson: As I’m sure you know, there are millions of Facebook political ads in a year. And while we have a really strong install base of about 16,000 users, they see a very small fraction of the overall Facebook ad landscape. 
So we actually consider the Facebook ad library data to be our primary source of data for the Ad Observatory, which is our public website. We supplement that website with political data that we collect from Ad Observer. But most of the data comes from Facebook’s ad library API [Application Programming Interface].
Angwin: And you accessed that library from your personal Facebook accounts because Facebook doesn’t have a way to set up professional research accounts, correct? So did you lose your personal Facebook accounts?
Edelson: That’s right. I have no place to post cute dog pictures.
Angwin: So what is lost now that you have lost access to these Facebook services?
Edelson: The Ad Observer browser extension will continue on unimpeded. It’s our tools for the public, our tools for civil society, our tools for other researchers that are all shut down.
The Ad Observatory is shut down. We also have a private tool that we provided to civil society groups so that they could monitor for misinformation and hate. That doesn’t work anymore  because it relied on ad library data.
This cuts off data to all of the researchers who take data from us because the [Facebook] ad library is really hard to work with. It is a pretty serious technical lift to collect all of it. We had permission from Facebook to share ad library data with other people who had permission to use it.
Angwin: Stepping back a bit, what were you looking for when you started this area of research?
Edelson: I think most people who come into this space, they’re interested in politics. They’re interested in what the advertisers are doing. I’m interested in that too, but I’m a cybersecurity researcher first. 
I’m actually primarily interested in what Facebook does and how Facebook behaves. So one of the most important questions that I have when I’m looking at a system like the ad library is just: How good is Facebook at identifying political ads?
That’s a really core question because understanding how good they are at doing that job tells us how useful the ad library is. If they’re missing a lot of ads, that’s a pretty serious asterisk against any research that’s done with that ad library. If they are not doing a good job, it also means that we probably need other transparency mechanisms, because right now the ad library is how Facebook says that researchers and journalists should understand political ads. 
The conclusion that we’ve been able to come to because of data from Ad Observer and other sources is that Facebook’s [ad library] is about 90 percent accurate. In other words, they catch about 90 percent of ads that aren’t voluntarily disclosed by advertisers, which is pretty good. The problem is that means there are hundreds of thousands of ads that are political and aren’t disclosed as such and never make it into the ad library.
Angwin: Is there also an error the other way, of classifying nonpolitical ads as political?
Edelson: Anecdotally, we have all seen many examples of false positives. We actually have forthcoming research about this. We’ve submitted it to peer reviewed journals so I can’t really talk about the conclusions yet.
Angwin: So, you got this project up and running in the spring of 2020. When did your dispute with Facebook begin?
Edelson: We had a conversation with them over the summer of 2020, when they told us they didn’t like this project. You know, they don’t like a lot of things we do. I didn’t expect that the things would escalate to anything like a cease and desist. And we certainly didn’t think that they would reach the point where we’re at now where they’ve terminated our Facebook accounts.
Just to be clear, we were negotiating with them about changes we could make or things that Facebook could do to make the Observatory not necessary. And, well, obviously those have not been successful.
Those talks were going on up until last week. 
Angwin: Facebook’s allegation was about privacy. But as I understand it, the extension only collects ads. So what is the alleged privacy violation of ads?
Edelson: To my knowledge, and this has bounced around, what Facebook is saying now is that advertiser data, like the advertiser page name and the advertiser page itself, is private user data. That is the case they’re making. 
Angwin: But those are public pages.
Edelson:  That’s correct. You cannot advertise on Facebook unless you’re a public page. 
This is where we just have a factual disagreement with Facebook. We do not see public advertiser pages as private.
Angwin: One thing I have wondered is why Facebook came after your project rather than other Facebook monitoring tools like WhoTargetsMe or our Citizen Browser project.
Edelson: To my knowledge, the primary difference between the Ad Observatory and other projects is that we publish data. We publish data about political ads and how political ads are targeted. 
In all fairness, I do think that Facebook has a rational economic interest in controlling information about their ad marketplace. They control a pretty powerful chunk of the overall ad ecosystem. And one of the ways that you control the market is you control information about that market. 
We represent a pretty significant threat to that dominance of market information because, to my knowledge, there is no other publicly available source of information about political ad targeting
As always, thanks for reading.
Julia Angwin 
The Markup
From The Markup
Anti-Semitic Merchandise Persists on Pinterest, Despite Restrictions
Uber Requires Nondisclosure Agreement Before Helping Carjacked Driver
Is the Cookie Web Tracker Dying?
P.S. For more from Julia Angwin and Hello World, look here. To receive the latest from our Citizen Browser project, sign up here, and so you can keep up on all the news from The Markup, sign up here, and we’ll email you every time we publish about the ways powerful actors are using technology to change society, usually two to three times a week.
This email doesn't track you when you open it or click on any links. To learn more read our Privacy Policy.
In order to unsubscribe, click here.
If you were forwarded this newsletter and you like it, you can subscribe here.
The Markup - The Markup P.O. Box 1103 N.Y., N.Y. 10159