Decades of research have repeatedly shown that people perform poorly at understanding and estimating conditional probabilities that are inherent in Bayesian reasoning problems. Prior work indicates that base rate neglec(BRN)may be the key to misjudgment of Bayesian reasoning problems. For example, researchers have explored visualization and causal Bayesian framework to eliminate BRN. Although the reasoning performance has been improved to some extent, it is still obscure as to how BRN biases Bayesian reasoning. To solve this puzzle, Hattori and Nishida (2009) proposed the equiprobability hypothesis. It is believed that BRN in Bayesian reasoning is related to whether reasoners can recognize the imbalanced probability structure (IPS) — the difference between marginal probabilities of two target events. According to the equiprobability hypothesis, people perform poorly in the Bayesian reasoning, not because the probabilistic inference is intrinsically difficult, but because two processes — the one to recognize the task structure and the other to infer the structure on the basis of the equiprobability assumption — compete with each other. Accordingly, if the default equiprobability hypothesis could be dissolved somehow, IPS of the task would be recognized, and hence, individual responses should roughly conform to the Bayesian theorem.

Given the findings of prior studies, we aim to identify factors that influence performance on Bayesian reasoning. Implied by equiprobability hypothesis, if Bayesian reasoning tasks could highlight IPS, individual reasoning performance should be facilitated. Consequently, we hypothesize that some evidence information and self-reference can help human recognize the IPS. To test our hypothesis, we employed two kinds of reasoning tasks: X syndrome task (negative context) and rehabilitation task (positive context). Each task has 4 variants differing in evidence information and the self-reference level. Evidence information could manipulate the noticeability of IPS. For example, “cough” (in X syndrome task) and “folk prescription” (in rehabilitation task) are respectively better than “positive result” and “specific drug” in illustrating IPS.

Based on the equiprobability hypothesis, this research conducted two experiments. In Experiment 1, we employed a 2 (evidence information: positive test or cough) × 2 (self-reference: related to oneself/not) between-subjects design. In Experiment 2, we also employed a 2(evidence information: specific drug or folk prescription) × 2 (self-reference: related to oneself/not) between-subjects design. The dependent variable was Bayesian reasoning performance, with 3 criteria: reasoning correct rate, reasoning accuracy, and response type.

The results showed that: (1) In both positive and negative medical diagnosis context, participants recognizing IPS performed generally better than those not recognizing IPS. (2) In the negative medical diagnosis context, the reasoning performance in the problem “related to oneself” was better than that in problems “not related to oneself”, while in the positive medical diagnosis task, higher self-reference level significantly reduced BRN. (3) When the collected evidence information prompted people to recognize the IPS, the reasoning performance discrepancy of self-other was related to event value. The research results supported the equiprobability hypothesis, which showed that if evidence information can help human recognize IPS, Bayesian reasoning performance will be improved, while “self-reference” may affect reason performance through two mediators — “processing depth” and “subjective desire”.