Mary Ann Badavi
Over the course of the past year of researching disinformation and social media, I’ve run into one problem again and again: accessing data is extremely challenging. In “Build It! Data Is Never Raw, It’s Collected,” Sarah Williams describes the method of participatory data collection, saying that “these projects help build communities that together gather the evidence to both fight for and understand their rights.” When it comes to the problem of social media data collection, there have been many impressive efforts by academics, journalists, and the general public to try to understand the black boxes through which technology companies operate.
Arguably the biggest of these black boxes is operated by Facebook. There are many outstanding questions about the company’s algorithmic operations, from how it serves ads to how it labels disinformation. One news outlet, The Markup, has created a participatory data mapping system in order to try to answer some of these questions. Citizen Browser, introduced in January, allows a panel of Facebook users to voluntarily share data on the posts and ads they see in their feed. The panel is composed of 2,000 paid participants across 48 U.S. states, spanning the political and demographic spectrum.
In her piece, Williams brings up the fact that “the act of collecting the data becomes the primary learning tool.” Indeed, Citizen Browser has already unearthed some interesting insights, simply by trying to verify claims that Facebook itself has made. On January 19, The Markup’s writers Leon Yin and Alfred Ng wrote a piece called “Facebook Said It Would Stop Pushing Users to Join Partisan Political Groups. It Didn’t.” They used Citizen Browser to show that despite Facebook claiming they had excluded political groups from its recommendation algorithm, the data collected by its participants showed that wasn’t the case. Not only did Facebook continue to push political groups on its users, the groups it did recommend were heavily influenced by the user’s political affiliation.
While these two visualizations may not be maps in the traditional sense, they do map data across the political spectrum, using participatory data collection methods to do so. When you hover over one of the blocks in the first graph, you can see some of the ads in question. I find the visualization itself a little confusing: while it’s clear that Trump voters are receiving more political group recommendations that Biden voters, I’m curious about what other actions these voters are taking on Facebook that might lead to their seeing more political recommendations. Do they like political pages? Are they part of other political groups? Or is it simply based on demographic data that Facebook collects? I’m particularly curious about these questions as they relate to non-voters: are non-voters simply not engaging with any political content at all?
The second visualization shows the political groups that were most often recommended to Citizen Browser participants on both sides of the aisle. Here, the data reveals an even more concerning revelation: right-wing groups like “Tucker Carlson Fox News” are recommended 10% more often than left-wing ones like “Dr. Fauci Fan Club.” I believe that percentage should be emphasized more in order to highlight these points. And while having the group images looks nice from a visualization standpoint, they’re not actually that relevant to the points the data is trying to reveal. In addition to making the percentage numbers larger, it might be helpful to visualize by the number of members—in the screenshot above, you can see that both conversative-leaning groups have more members than the liberal-leaning groups, but the visualization is not organized to show that.
Overall, I do think the maps are effective in their main goal: Facebook says it’s not recommending political groups anymore, yet this data clearly shows that it is. If participatory data collection’s main aim is to subvert institutions and advocate for reform, this was successful.
When thinking about how to apply these practices to a map of my own, I wanted to take a similar approach of using data to shine a light on obfuscated social media algorithms. Using Facebook’s Top 10, a Twitter account from a New York Times journalist who tracks where the Facebook posts with the most engagement are coming from, I tracked the political leanings of over 50 popular Facebook pages from July 2020 to March 2021.
Facebook has repeatedly claimed to be a politically neutral company, even as they face accusations from Republicans that they’re stifling conservative voices. However, data from Facebook’s Top 10 clearly shows the opposite: the vast majority of political posts with the highest engagement levels are from right-wing sources.
In these visualizations, viewable in full on Glitch, I used similar tactics to Citizen Browser in mapping across the political spectrum and using traditional political colors to differentiate the data. But in both visualizations, I’ve tried to place an emphasis on the scale of the disparity between right-wing and left-wing content. In the bubble plot to the left, the size of the circles correlates to the number of posts from a specific user that made Facebook’s Top 10; anti-woke, partisan right voices like Dan Bongino and Ben Shapiro vastly outpace the most engaging partisan left account, CNN. And in the election-style chart to the right, I simply counted how many posts came from right-wing vs left-wing sources: clearly, the right dominates here too.
While my map isn’t participatory, I do have similar aims as participatory data collection efforts: learn through parsing the data, challenge institutions, and raise further questions by presenting an incomplete picture and showing just how much is unknown. To complete that picture, we need more data transparency from the likes of Facebook and other tech companies. Hopefully, projects like Citizen Browser will push them to get there.