Collaborative Research: Learning, Behavior, and Design in Diffusion Processes

University Of Pennsylvania, Philadelphia PA

Investigators

Abstract

This award funds research that uses economic theory and laboratory experiments to study information and learning on social media platforms. Social media are an increasingly important source of news for many people. The content that users see on such platforms depends on what other users choose to post and share. It also depends on the algorithms that the platform developers use to generate news feeds. The researchers will investigate how the decisions of the users and developers of social media platforms affect what people learn. What types of content are users likely to see? How do people process this content to form beliefs about the relevant issues? When are these beliefs likely to be accurate? The research relates to recent debates about whether certain user and developer choices contribute to the spread of misleading or incorrect information on social media. For example, does a social media news feed that focuses on showing the most popular content (as opposed to random content) help or hurt the accuracy of people's beliefs in the long run? In the first part of the project, the researchers will develop a theoretical model describing social learning in settings where people learn by posting, sharing, and re-sharing copies of “signals” (e.g., news stories) about the state of the world. The model will focus on how people's beliefs and behavior co-evolve when the information diffusion process driven by people's actions also influences their learning. This research will yield results about how the platform's signal-sampling algorithm, which determines which signals get shown to users, affects the accuracy of social learning and the extent of agreement in people's beliefs. The second part of the project will test these predictions in a laboratory experiment. The investigators will conduct social-learning games where subjects see predecessors' signals and choose to endorse a subset of those signals. In choosing which past signals to show to future users, different treatments will vary the weight put by the algorithm on endorsements by previous users. The third part of the project will deal with the “tipping point” in complex diffusion models. This part of the research will identify simple conditions on individual behavior that determine whether there exists a “tipping point” where the diffusion discontinuously switches from (a) reaching a very small fraction of the population to (b) reaching a large fraction of the population. The authors will investigate whether such transitions always happen smoothly. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.