The next speaker in this session at the Social Media & Society conference in Glasgow is my excellent colleague Laura Vodden, presenting our reflections on the experience of doing research using the Meta Content Library clean-room environment (and we have just published a new article in Political Communication Report on the clean-room model as well).
The MCL is Meta’s core access model for data from Meta platforms; it replaced the previous platform CrowdTangle in 2024, similarly offering a Web interface and API but transitioning to a clean-room environment within which all serious data work is meant to be conducted. Access is available predominantly to qualified academic or research institutions, with some allowances for public-interest organisations, but not to journalists or independent researchers.
Clean-room access is provided via two models: the Virtual Data Enclave (VDE) is hosted by SOMAR at the University of Michigan (which now incurs a usage fee), and rather clunky: it requires VPN access, two-factor authentication, a remote desktop connection to a Windows Virtual Machine, a remote desktop connection from there to a Linux VM, and the launch of a Jupyter Notebook to finally connect to the database itself. That Notebook environment had only very limited functionality, was slow and laggy to use, and supported only very limited export of results. Content deleted from Meta’s platforms had to be removed from the VDE every 180 days.
Alternatively, the newer Secure Research Environment is hosted by Meta itself, and accessed more simply within a Web browser using the Amazon Secure WorkSpaces plugin; this vastly improved the login and usability experience. It also provides access to a much broader range of packages for use in the Jupyter Notebook. However, it institutes a mandatory and complete data deletion process that occurs on the first day of each month, complicating more long-term analysis processes and hitting data access rate limits.
This can be understood as a kind of hostile architecture limiting meaningful research uses of these platforms. It discourages cross-platform analyses as the clean-room is an isolated space; privileges quantitative analysis as no qualitative coding and close reading tools are available; and reduces the potential depth and length of analysis in research processes. Standard research tool development practices are also made difficult by an inability to connect to GitHub and similar code development tools.
There is a significant need to continue constructive conversations with Meta and its representatives on this tool (and in fairness, the MCL team itself is responsive to many of these concerns, but limited in its ability to address them) – and we are also actively building a community of MCL users in order to better gather and represent our concerns towards the company on a collective basis.
