基于期望值最大化的理论何时失效:风险决策中为自己-为所有人决策差异的眼动研究

When expectation-maximization-based theories work or do not work:An eye-tracking study of the discrepancy between everyone and every one

主流的风险决策理论专家发展了一系列基于期望值最大化(expectation-maximization)的理论,以期捕获所有人的风险决策行为。然而大量证据表明,这些基于期望值最大化的理论并不能如同描述性理论那样理想地描述单一个体的决策行为。本研究采用眼动追踪技术,系统考察了个体在为所有人决策与为自己决策时的风险决策行为及信息加工过程的差异。本研究发现,基于期望值最大化的理论可捕获为所有人决策或为自己多次决策时的情况,却不能很好捕获个体为自己进行单次决策时的情况。本研究结果有助于理解基于期望值最大化的理论与启发式/非基于期望值最大化的理论的边界,为风险决策理论的划分和发展提供实证参考。

Mainstream theorists in risky decision-making have developed various expectation-maximization-based theories with the ambitious goal of capturing everyone’s choices.However,ample evidence has revealed that these theories could not capture every individual’s(“every one’s”)actual risky choice as descriptive theories.Substantial research has demonstrated that people do not follow the logical process suggested by expectation-maximization-based theories when making risky choices but rather rely on simplifying heuristics.From our perspective,the possible reason why mainstream decision theorists did not abandon the framework of expectation is that these theorists never doubted the validity of the expectation rule as a descriptive rule in describing decision-making under risk.We believe that expectation-maximization-based theories may capture risky choices when individuals make decisions for everyone.However,whether these theories could capture risky choices when individuals make decisions for themselves cannot be taken for granted.We thus used an eye-tracking technique to explore whether a theory for everyone would work well for every one.A total of 52 college students participated in the experiment.Three risky choice tasks were conducted in the present study:a D-everyone task,a D-multiple task,and a D-single task.In the D-everyone task,participants were asked to choose the more optimal option out of two options under the assumption that their selection would be the final decision for everyone who was facing the same choice—that is,everyone would be subject to the same choice but could receive different outcomes.In the D-multiple task,participants were asked to choose between the two options under the assumption that their selection would be applied a total of 100 times.In the D-single task,participants were asked to choose between the two options under the assumption that their selection would be applied only once to themselves.The participants’eye movements were recorded while they performed the tasks.Behavioral results revealed that,compared with the D-single task,participants selected more choices correctly predicted by EV and EU theories,and took a longer time to make a decision in the D-everyone and D-multiple tasks.Furthermore,eye movement measurements revealed the following.(1)The scanpath patterns of the D-everyone task and D-multiple task were similar but different from those of the D-single task.(2)The depth of information acquisition and the level of complexity of information processing in the D-everyone task and D-multiple task was higher than that in the D-single task.(3)The direction of information search in the D-everyone task and D-multiple task was more alternative-based than that in the D-single task.(4)The eye-tracking measures mediated the relationship between the task and the EV-consistent choice.In summary,behavioral and eye movement results supported our hypotheses that participants were likely to follow an expectation strategy in the D-everyone and D-multiple tasks,whereas they were likely to follow a heuristic/non-expectation strategy in the D-single task.We found that expectation-maximization-based theories could capture the choice of an individual when making decisions for everyone and for self in a multiple-play condition but could not capture the choice of an individual when making decisions for self in a single-play condition.The evidence for the discrepancy between everyone and every one,which was first reported in our study,implied that the possible reason why expectation-maximization-based theories do not work is that a default compatibility between the full set(everyone)and the subset(every one)does not exist.Our findings contribute to an improved understanding of the boundaries of expectation-maximization-based theories and those of heuristic/non-expectation models.Our findings may also shed light on the general issue of the classification of risky decision-making theories.