Proceedings of the Annual Meeting of the Cognitive Science Society

A central challenge in belief updating is how individuals evaluate evidence when multiple competing hypotheses are plausible. Although prior work shows that belief updating can align with Bayesian principles, this evidence typically involves intuitively directional cases. It remains unclear whether similar rational trajectories are observed in counter-intuitive settings, where accumulating supportive evidence can increase the plausibility of an alternative explanation. In the present study, participants evaluated a scenario in which five sequentially presented confessions were diagnostic either of collective guilt or of coercive interrogation. Early confessions led to increases in belief in both guilt and force, indicating non-zero sum updating. However, contrary to a priori model predictions, participants did not exhibit the non-monotonic pattern whereby belief in force overtakes belief in guilt. Participant-specific Bayesian Network models showed closer alignment with observed trajectories, yet revealed systematic under-weighting of force and over-weighting of guilt. Overall, belief updating reflected sequential evidence integration rather than full Bayesian re-computation.