Belief polarization can be caused by disagreements over source independence: Computational modelling, experimental evidence, and applicability to real-world politics

A large literature debates whether belief polarization, in both experiments and real-world political opinion data, is the result of biased forms of reasoning like biased assimilation and motivated reasoning, or if it can be caused by rational reasoning. We present evidence for the plausibility of a novel Bayesian mechanism of experimental and real-world political belief polarization involving perceptions of source independence. We show, using a novel Bayesian network, that when presented with conflicting testimony from two source groups, Bayesians should update towards the position of the group they deem to be more independent, meaning those who disagree about which group that is should polarize. We find in a pre-registered experiment (N = 351) that human participants polarize under these conditions. We then find in a UK study (N = 507) and a pre-registered US replication (N = 300) that, using a novel scale instrument, real-world partisans (Labour, Conservative, Republican, and Democrat) perceive their party's supporters to be more independent than the opposing party's supporters, with large average effect sizes (d = 0.87 UK, d = 0.82 US), suggesting the conditions are in place for such polarization to occur in the real world. Accordingly, we find that those who view their party's supporters as more independent than their opponents to the greatest extent have the most polarized beliefs, even after controlling for partisanship and affective polarization. Overall, our results highlight perceptions of testimonial independence as a plausible mediator of experimental and real-world belief polarization.