Signal Understanding Human Perception of Representation in Citizens' Assemblies: An Empirical Study

Summary

Citizens' assemblies are meant to mirror the wider population, but organizers who fill seats by quota must choose which attributes count, and satisfying every quota can still leave out a dimension that citizens care about. Using randomized conjoint experiments on general-purpose and climate-focused assemblies, the study finds that demographics shape perceived representation, while political alignment and topic-specific traits such as concern about climate weigh more heavily. In the climate setting both kinds of attribute stay influential, with political alignment showing the larger estimated effect. Panels stratified on demographics, even with political alignment added, reproduce the climate-concern distribution of the observed pool no better than uniform random draws, so a topic-specific attribute may need its own explicit quota. Two learned models, a learned metric and a utility model conditioned on each respondent, predict the choices of held-out respondents with substantial accuracy but do not fully capture their judgments. The authors advise designers to weigh political and topic-specific dimensions alongside demographics, not to rely on correlated proxies to cover omitted attributes, and to use predictive models to diagnose how profiles drive representation choices.