Proceedings of the Annual Meeting of the Cognitive Science Society

When multiple sources agree, their testimony provides stronger evidence if they are independent. But how do people infer whether sources are coordinated? We developed a Bayesian model predicting that dependency inferences should depend on three factors: verbal similarity between reports, the knowability of the topic, and the diversity of possible expressions. In a pre-registered experiment (N = 156), participants viewed social media posts varying in these dimensions and judged the likelihood of coordination. Results revealed a striking dissociation: participants were highly sensitive to verbal similarity (__ = .283), inferring substantially more dependency when posts were near-identical versus substantively similar but differently worded. However, participants were insensitive to knowability and expressive similarity, contrary to normative predictions. These findings suggest people rely on a simple "copy detection" heuristic based on surface similarity rather than engaging in full Bayesian integration of domain-relevant probabilistic information.