Direct perception, heuristic reasoning, or Newtonian mechanics? A Bayesian approach to naturalness and causality judgments in collision events

This study examines how individuals evaluate the naturalness and causality of object collisions. Participants observed horizontal collisions varying in mass ratio, velocity, and restitution coefficient. Three groups judged whether: (1) the collision appeared natural or unnatural; (2) the post-collision motion of the initially stationary object was caused by the collision; or (3) the post-collision motion of the initially moving object was caused by the collision. Three computational models were tested: a Newtonian model assuming veridical internalization of mechanics, a Heuristic model based on approximate knowledge, and a Michottean model grounded in perceptual features. A Bayesian approach estimated individual decision thresholds and adjusted dichotomous responses. Predictions were compared against raw and threshold-adjusted judgments. The Newtonian and Heuristic models outperformed the Michottean model. The Heuristic model showed a slight advantage for unadjusted responses, but not after threshold adjustment. Participants attributed causal influence to stationary objects when prompted, indicating limits to causal asymmetry—that is, the tendency to interpret collisions as involving unidirectional transmission of impetus from a moving to a stationary object. These results demonstrate the value of quantitatively contrasting models that assume veridical physical knowledge with those positing only approximate representations.