A probabilistic constraint satisfaction model of information distortion in diagnostic reasoning.

Information distortion is a cognitive bias in sequential diagnostic reasoning. It means that assumptions about the diagnostic validity of later evidence are distorted in favor of the leading hypothesis. Therefore the bias contributes to a primacy effect. Current parallel constraint satisfaction models account for order effects and coherence shifts, but do not explain information distortion. As an alternative a new, probabilistic constraint satisfaction model is proposed, which considers uncertainty about diagnostic validity by defining probability distributions over coherence relations. Simulations based on the new model show that by shifting distributions in favor of the leading hypothesis an increase in coherence can be achieved. Thus the model is able to explain information distortion by assuming a need for coherence. It also accounts for a number of other recent findings on clinical diagnostic reasoning. Alternative models and necessary future research are discussed.