从风险决策中的多次博弈到单次博弈：量变还是质变?

诺贝尔经济学奖获得者Samuelson于1963年发现人们在单次和多次博弈条件下决策行为不一致。文章综述了两种博弈条件下人们决策行为的差异并质疑了这种差异的传统理论解释机制。描述或预测决策行为的风险决策理论其实只采用了一种评价法则—— 期望法则, 始终没有跳出“最大化”的窠臼。基于实验证据, 我们推测, 多次博弈时人们遵守了期望法则, 而单次博弈时人们所遵循的是非补偿性法则。从多次博弈到单次博弈, 不单单是一种博弈次数上的变化(量变), 而是代表了从期望法则(补偿性法则)到非补偿性法则两种策略之间的转变(质变)。最后, 文章介绍了单次、多次博弈问题在医疗、应急管理以及投资领域的体现, 并呼吁更多的研究者关注单次、多次博弈问题。     

风险决策和跨期决策的过程比较：以确定效应和即刻效应为例

风险决策和跨期决策与人类生存发展密切相关, 且两类决策在理论发展、行为效应及神经基础等方面具有相似性。为检验二者是否具有共同过程机制, 本研究以风险决策中的确定效应和跨期决策中的即刻效应为例, 采用眼动追踪技术比较了它们的局部、整体过程及模型拟合。辅以贝叶斯因子分析实验数据表明：二者的主要过程特征均相似, 且更符合非折扣模型假设; 二者在加工复杂程度等少数特征上有所不同; 确定和即刻信息在加工方向等特征上存在特异性。这表明二者可能具有共同的核心决策规则：两类决策更可能遵循非折扣模型预期的简捷、启发式规则, 而不是折扣模型所假设的补偿性、基于选项规则。本研究为建立两类决策的共同解释框架做出了有益尝试, 并为决策比较研究方法提供新的方向。     

决策过程的追踪:基于眼动的证据

利用过程追踪技术实现对"决策过程"的追踪,或能真正解答"人类是如何做出决策"这一科学问题。在各种过程追踪技术中,眼动技术由于具有无干扰性、适用范围广泛、采集信息多样等独一无二的优势,已被广泛应用于各种决策研究中,为验证和比较不同决策理论的不同假设和预测做出了贡献。本文介绍了眼动技术如何被应用于验证补偿性决策理论与非补偿性决策理论对风险决策过程的不同假设,眼动技术如何用于检验多次决策与单次决策的决策过程差异,以及眼动技术如何用于检验"剔除–关注模型"与"齐当别决策模型"对多属性决策中共享信息发挥作用机制的不同预测。     

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

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。     

不确定性风险选择的抱负水平—相对效用整合理论

本文认为不确定性风险决策备择方案的选择标准既不是规范效用理论所主张的效用最大化,也不只是研究者们基于西蒙有限理性决策理论而提出的抱负水平,而是存在于决策不同阶段上的、由启发式策略控制着的这两种标准的整合,我们称之为抱负水平－相对效用整合标准。为此,我们在研究抱负水平结构的基础上提出了负险选择的抱负水平－相对效用整合理论及其支持性研究证据,并就有关问题进行了讨论与思考。指出抱负水平－相对效用整合是风     

论证据决策理论的困境与出路

证据决策理论(evidential decision theory)要求决策者的行动要最大化地满足主体的需求(desirability),其方法就是将决策者的决策基于条件证据期望结果(conditional evidence expected out-comes)。也就是说,只要先前发生的事件与决策者的期望之间具有高度相关性,决策者就应当根据这     

风险决策中的领域特异性

风险领域特异性指某特定领域内的风险决策和偏好可能会受到领域内特有的风险因素影响, 因而会导致领域特异性的行为。基于期望法则的风险决策理论认为, 风险倾向具有跨情境的稳定性。但近20年行为决策领域的证据表明, 风险倾向具有领域特异性, 其机制探索、测量工具和影响因素方面均取得了积极进展。从机制探索历程来看, 以往主流机制着重从风险−回报框架解构领域特异性的风险行为, 近年研究在丰富其机制检验证据的同时, 开始从进化、人格、动机等新的视角进行解读; 在测量工具层面, 领域特异性风险量表仍占主导地位, 并在多种文化、内容领域和群体间得到了进一步检验和扩展, 其他基于不同内容领域划分的风险领域特异性量表也具有较好的信效度; 在影响因素层面, 大量研究从遗传、环境、个体差异等角度探索了导致风险领域特异性的前因变量。未来研究方向应注重整合各风险内容领域, 在更加细化的风险领域检验现有理论, 并进一步探索个体层面的领域特异性规律。     

风险决策心理因素的理论综述

风险决策理论对于人们是否作出冒险行为的选择提出了两种比较典型的解释:一种解释把风险决策归因于人们共有的基本过程,即“较冷”的心理和认知过程。这些理论认为风险选择是由人类基本的心理和感觉机制引起的。而另一种解释则把风险决策归因于“较热”的情感和动机过程。这些理论认为,情境和人格因素会增强风险决策的动机并导致风险决策个体差异的存在。本文对这两种解释的不同理论作了较为系统的论述。     

模糊决策的认知神经机制

模糊决策是特殊、复杂的风险决策, 还是一种独立的决策类型, 是当前认知神经科学争论的热点问题。一系列的fMRI研究分别得出了不同结论, 但大多数证据支持模糊决策和风险决策之间存在分离, 模糊决策有其独特的决策机制; 模糊决策的认知神经机制也是一个亟待解决的问题。因此本研究拟应用事件相关电位技术、生物反馈技术和基因技术, 采用IGT和GDT任务范式, 通过网络成瘾人群和正常人群的对比研究, 去探索模糊决策和风险决策之间的分离; 应用事件相关电位技术和磁共振技术, 采用IGT范式和选瓶任务范式, 研究模糊决策的认知神经机制; 并从临床角度进一步验证上述结果。该项目的开展, 有助于拓广模糊决策的研究领域和视野, 对理解人类在模糊情境下的决策机制以及模糊决策和风险决策之间的关系, 具有重要的理论意义; 对临床上成瘾人群、脑损伤患者的认知诊断和治疗以及现实中企业和个人的决策, 具有重要的参考价值。     

测验理论的新发展:多维项目反应理论

多维项目反应理论是基于因子分析和单维项目反应理论两大背景下发展起来的一种新型测验理论。根据被试在完成一项任务时多种能力之间是如何相互作用的,多维项目反应模型可以分为补偿性模型和非补偿性模型两类。本文在系统介绍了当前普遍使用的补偿性模型的基础上,指出后续研究者应关注多维项目反应理论中多级评分和高维空间的多维模型、补偿性和非补偿性模型的融合、参数估计程序的开发和多维测验等值四个方面的研究。     

I felt low and my purse feels light: depleting mood regulation attempts affect risk decision making

We propose that negative affect can induce people to engage in risky decisions. We test two alternative hypotheses as to how this effect may emerge. The mood repair hypothesis states that risky choices in risk decision making serve as a means to repair one's negative affect. The depletion hypothesis, in contrast, states that risky choices in risk decision making are the mere consequence of a state of depletion resulting from engagement in active mood regulation attempts. The results of a first laboratory study establish a link between risky choices in risk decision making and negative affect. Subsequent experiments provide evidence that depletion due to active mood regulation attempts, rather than mood repair, is the underlying process for this link. Copyright © 2008 John Wiley & Sons, Ltd.     

迫选规则体验法: 检验规范性和描述性风险决策理论的新尝试

为寻求检验规范性和描述性风险决策理论的通用标准, 本研究以期望价值理论和齐当别抉择模型为例, 探讨了“迫选规则体验法”的适用性。被试为120名大学生, 实验任务为要求被试分别完成自主决策(采用未知规则: 真规则)和规则迫选决策(遵循给定规则: 假规则)任务, 并对决策后的情感和认可程度进行评定。研究发现: (1) 被试在自主决策条件下比在规则迫选条件下体验到的正性情感程度更强, 负性情感的程度更弱; (2) 被试在自主决策与规则迫选决策两种条件下做出的相同决策越多, 该被试对迫选规则更加认可并体验到的正性情感程度越强, 负性情感的程度越弱; (3) 与期望价值理论相比, 齐当别抉择模型可能符合更多决策者的实际决策规则。这些结果表明, 作为检验规范性和描述性风险决策理论的新尝试, 迫选规则体验法可能更有助于回答“决策者实际采用的决策规则是什么”的问题。     

A Scanpath Analysis of the Risky Decision‐Making Process

In the field of eye tracking, scanpath analysis can reflect the sequential and temporal properties of the cognitive process. However, the advantages of scanpath analysis have not yet been utilized in the study of risky decision making. We explored the methodological applicability of scanpath analysis to test models of risky decision making by analyzing published data from the eye‐tracking studies of Su et al. (2013); Wang and Li (2012), and Sun, Rao, Zhou, and Li (2014). These studies used a proportion task, an outcome‐matched presentation condition, and a multiple‐play condition as the baseline for comparison with information search and processing in the risky decision‐making condition. We found that (i) the similarity scores of the intra‐conditions were significantly higher than those of the inter‐condition; (ii) the scanpaths of the two conditions were separable; and (iii) based on an inspection of typical trials, the patterns of the scanpaths differed between the two conditions. These findings suggest that scanpath analysis is reliable and valid for examining the process of risky decision making. In line with the findings of the three original studies, our results indicate that risky decision making is unlikely to be based on a weighting and summing process, as hypothesized by the family of expectation models. The findings highlight a new methodological direction for research on decision making. Copyright © 2016 John Wiley & Sons, Ltd.     

Consequences of Regret Aversion: Effects of Expected Feedback on Risky Decision Making

Previous research has considered the question of how anticipated regret affects risky decision making. Several studies have shown that anticipated regret forces participants towards the safe option, showing risk-aversion. We argue that these results are due to the previous confounding of the riskiness of the options with the feedback received. Our design unconfounds these factors, and we predict that participants will always tend to makeregret-minimizingchoices (rather than risk-minimizing choices). We present three experiments using a “choices between equally valued alternatives” paradigm. In these experiments we manipulate whether the risky or safe gamble is the regret-minimizing choice by manipulating which gamble(s) will be resolved. As predicted, participants tend to choose the regret-minimizing gamble in both gains and losses and in both relatively high risk and relatively low risk pairs of gambles. We consider the implications of these results for the role of regret in choice behavior.     

Linguistic–Numeric Presentation Mode Effects on Risky Option Preferences

In this study, we examine preferences between lotteries with chances presented either numerically or linguistically. Presentation mode is predicted to affect preferences due to the perception of linguistic chance as skewed distributions of risk. Based upon weighting functions incorporating risk/uncertainty aversion from ambiguity theory and cumulative prospect theory, we predict that presentation mode effects on risky choices will be detectable in very small risks and in large risks. In two experiments, subjects chose between both gain and loss lotteries with constant payoffs and equivalent numeric and linguistic chances. Presentation mode affected choices when chances were above 50%, where lotteries with numeric chances were more frequently preferred in gains while lotteries with linguistic chances were more often preferred in losses. The effect of presentation mode for low-chance lotteries (5% and less) also affected choices such that numeric choices were generally preferred more frequently in losses and linguistically expressed choices were generally preferred more often in gains. Overall, these results suggest that theories of the effects of second order uncertainty on risky choice may be used to model decisions involving linguistic risk. They also suggest that the study of the perception of linguistic risk assessments can provide insight into the cognitive processing behind the weighting functions proposed to depict decision under risk and uncertainty. Finally, the results have practical implications since information providers can affect decision makers’ choices by controlling presentation mode in such a way as to alter the relative attractiveness of uncertain events.     

How Affect Shapes Risky Choice: Distorted Probability Weighting Versus Probability Neglect

People's choices between prospects with relatively affect‐rich outcomes (e.g., medical side effects) can diverge markedly from their choices between prospects with relatively affect‐poor outcomes (e.g., monetary losses). We investigate the cognitive mechanisms underlying this “affect gap” in risky choice. One possibility is that affect‐rich prospects give rise to more distortion in probability weighting. Another is that they lead to the neglect of probabilities. To pit these two possibilities against each other, we fitted cumulative prospect theory (CPT) to the choices of individual participants, separately for choices between options with affect‐rich outcomes (adverse medical side effects) and options with affect‐poor outcomes (monetary losses); additionally, we tested a simple model of probability neglect, the minimax rule. The results indicated a qualitative difference in cognitive mechanisms between the affect‐rich and affect‐poor problems. Specifically, in affect‐poor problems, the large majority of participants were best described by CPT; in affect‐rich problems, the proportion of participants best described by the minimax rule was substantially higher. The affect gap persisted even when affect‐rich outcomes were supplemented by numerical information, thus providing no support for the thesis that choices in affect‐rich and affect‐poor problems diverge because the information provided in the former is nonnumerical. Our findings suggest that the traditional expectation‐based framework for modeling risky decision making may not readily generalize to affect‐rich choices. Copyright © 2015 John Wiley & Sons, Ltd.     

Modeling the Dynamics of Risky Choice

Individuals make decisions under uncertainty every day. Decisions are based on incomplete information concerning the potential outcome or the predicted likelihood with which events occur. In addition, individuals' choices often deviate from the rational or mathematically objective solution. Accordingly, the dynamics of human decision making are difficult to capture using conventional, linear mathematical models. Here, we present data from a 2-choice task with variable risk between sure loss and risky loss to illustrate how a simple nonlinear dynamical system can be employed to capture the dynamics of human decision making under uncertainty (i.e., multistability, bifurcations). We test the feasibility of this model quantitatively and demonstrate how the model can account for up to 86% of the observed choice behavior. The implications of using dynamical models for explaining the nonlinear complexities of human decision making are discussed as well as the degree to which the theory of nonlinear dynamical systems might offer an alternative framework for understanding human decision making processes.     

Risk preferences in young children: early evidence of individual differences in reaction to potential gains and losses

New methods were developed for studying risky decision making in children as young as age five. Each child was given a block of ‘gain’ trials, for example, a choice between a sure gain of one prize and a 50:50 chance of gaining either two prizes or no prize, and a block of ‘loss’ trials, for example, a choice between a sure loss of one prize and a 50:50 chance of losing either two prizes or no prize. We were thus able to compare risky choice for gains and losses at the level of the individual child. In each of two experiments a variety of individual difference variables were measured, including in Experiment 2, the child's parent's scores on the same task. Across experiments, the preponderance of choices was of the risky option. However, most children and adults made more risky choices in the domain of losses than in the domain of gains. Predictors of individual differences in children included shyness, impulsivity, and the risk taking of the child's parent. We suggest that methods are now in place to encourage further studies of decision processes in young children. Copyright © 2003 John Wiley & Sons, Ltd.