中国科学院行为科学重点实验室知识库

风险决策主流理论以期望法则的计算为核心假设,认为决策者需要进行"加权求和"的期望值计算过程.为检验这一基本假设,本研究采用相同的视觉材料设计了两类任务:将"%"看成"概率"的概率决策任务(风险决策)和将"%"看成"比例"的比例决策任务(计算决策).两任务均要求被试选择自己偏好的选项.考虑到被试在比例决策任务中须执行"加权求和"的期望值计算,方能使其收益最大化,我们推测,若被试在概率决策任务中也是根据"加权求和"规则做出决策,我们应当会发现影响"期望值计算"的诸因素(如计算难度、算术能力)同时影响两类任务.结果发现:计算难度和算术能力只影响比例决策任务绩效(其反应时更长,基于期望值选择的比例更高),但不影响概率决策任务绩效.相反,与"期望值计算"无关的"概率能力"和"感觉寻求人格"只影响概率决策任务绩效,但不影响比例决策任务绩效.该结果表明,概率能力和人格对风险决策的行为绩效起调节作用,人们在风险决策中可能并未按照期望法则的精确计算方式做出决策.期望法则家族理论将"加权求和"结果作为人类风险偏好的单一指数,或许是一种人为创造的错误指数. 

Mainstream theories of risky decision making assume that decision maker employs the expectation rule - computing expected value or utilities by summing all the weighted risky outcomes. To test this general hypothesis, the present research developed two tasks by using visually identical experimental materials: Probabilistic Decision (PROB) task and Proportional Decision (PROP) task, in which the same sign “%” denotes probability or proportion respectively. Participants were asked to choose the option they preferred in both tasks. Given that choosing the option with greatest value in PROP tasks requires an expectation computation, we speculated that, if participants do follow the same rule to make a risky decision in PROB tasks, the factors that affect expected value or utility computation, such as computational difficulty and arithmetic ability, should affect their performance in both tasks. We found that the resulting responses to the two tasks did not consistently fit the prediction of the expectation rule: participants’ response times were shorter in PROB task than in PROP tasks, and proportions of their expected value (EV)-based choices were lower in PROB task than in PROP tasks. We also found that participants’ task performances in the two tasks were moderated by disparate variables, indicating that the underlying mechanisms were different in these two tasks. In particular, the arithmetic ability moderates performance of PROP tasks, but not of PROB tasks, whereas, the numeracy ability and sensation-seeking trait moderates performance of PROB tasks, but not of PROP tasks. These results suggested that individuals might not make risk decisions by performing an expectation computation as predicted by the expectation rule, and that using an EV-based index to prescribe human risky preferences appears to be an artificial or false index of risk preference.