High-Throughput Experiments In Small-Group Deliberation

The effectiveness of small group deliberation, like many problems in social science, depends on context. There is no one-size-fits-all solution for improving deliberation outcomes, so it is important to understand how interventions can be tailored to suit specific deliberative situations. However, traditional experimental methods, which are focused on finding support for particular theories, tend to limit examination to a few interventions in a small number of contexts, creating the possibility for overly generalized conclusions.

Rather than attempt to determine universal laws of deliberation, we approach the problem as a map-making exercise. This approach involves three methodological advancements: 1) automated data collection, 2) space-aware statistical models that make local generalizations from each sample, and 3) adaptive algorithms that determine the most important samples to collect next, given existing data. We use this approach to evaluate which interventions best support cross-partisan dialog for differing group compositions and levels of topic partisanship.