This research program seeks to discover which interventions most reduce social polarization in deliberative groups and populations. Prior research on deliberation has identified many features – of individual participants, the circumstances under which they come together, and the topics they consider – that are potentially relevant to the success of a given exercise. However, the current state of research does not provide clear guidelines for how these features influence the process of deliberation. As a result, we have trouble using our existing knowledge to predict which interventions are most likely to succeed in a given situation.

Our inability to contextualize knowledge about deliberation is an important example of what has been called a “generalizability crisis” in social science. Experiments in deliberation typically vary only one or two parameters of theoretical interest, while keeping many others fixed. Because different studies vary from one another along many potentially relevant dimensions, and because we have no way to express how large or small these differences are or how important we expect them to be to the outcomes we care about, it is essentially impossible to make “apples to apples” comparisons across different contexts. Consequently, we lack a rigorous way to specify over which range of conditions we expect any of our existing studies to generalize.

Duncan Watts’ Computational Social Science Lab at the University of Pennsylvania is working to develop a platform for making and testing predictions about polarization, related to the generalizability problem described above. Designing and building the core infrastructure for high-throughput experiments that will power the open science platform is the pilot phase of this two-stage project. As a longer term goal, the team aims to organize a community of researchers and practitioners around this platform to explore the space of possible experimental contexts, and ultimately to build replicable, cumulative, and useful knowledge.