正当化需求对决策策略的影响
文献类型：学位论文
| 作者 | 杨雨佳
|
| 答辩日期 | 2022-06 |
| 文献子类 | 硕士 |
| 授予单位 | 中国科学院大学 |
| 授予地点 | 中国科学院心理研究所 |
| 其他责任者 | 栗胜华 |
| 关键词 | 决策正当化 决策策略分类 模型拟合 眼动 机器学习算法 |
| 学位名称 | 理学硕士 |
| 学位专业 | 应用心理（风险判断与决策） |
| 其他题名 | The Effect of Justification on Decision-making Strategies |
| 中文摘要 | When the outcome of a decision affects others, decision makers must explain their decision-making process and the results to others for support. This demand for explanation increases the need for justification. As most decisions in real life impact others, decision-making justification is not only a widespread phenomenon, but also one of many monitoring strategies used by organizations. Bettman et al. (1998) firstly proposed “maximizing justification” as one of the main goals in decision making. Unfortunately, not much has been known regarding decision-making justification to date. To fill in this gap, I conducted two studies to investigate the effect of justification on decision-making processes and strategies in this thesis. Additionally, most strategy identification approaches in previous studies focus only on decision outcomes. The lack of technical tools for integrating process and outcome data limits the precision and resolution of strategy identification. Furthermore, although some existing approaches do integrate multiple data types, they cannot classify strategies trial-by-trial. These limitations lead to the poor performance of extant classification approaches, and challenge the accuracy and validity of research findings. In this thesis, I attempted to combine eye-tracking process data with machine-learning algorithms to present an innovative method for distinguishing decision-making strategies in a more precise and accurate manner. Study 1 focuses on the impact of different decision-making scenarios (risky vs. non-risky choices) and levels of justification demand (low: private decision; medium: the presence of relevant others; high: the need to explain to relevant others) on decision-making strategies. We classified strategies by the Multiple-Measured Maximum Likelihood (MMML) method based on behavioral data. The results show that in both scenarios, the majority of subjects were classified as using a weighting- and-adding compensatory strategy. For non-risky choices, different levels of justification had no effect on strategies used. For risky choices, demand of justification increased the proportion of individuals using compensatory strategies, but the effect was not statistically significant. There are several limitations of the strategy classification method used in Study 1 that may explain why the experimental results were insignificant. Therefore, in Study 2, a new strategy classification method was used to explore the effect of justification levels in risky decisions. The new method is called Choice-Learning- Training-Classification (CLTC), and it integrates eye-tracking data and machine- learning algorithms in strategy classification. We compared CLTC with MMML in classification outcomes. The results show that the classification accuracy of CLTC was much higher than that of MMML for prescribed-strategy decision data, and the classification results of CLTC and MMML differed greatly for free-choice data. Moreover, because CLTC can classify subjects’ strategies trial-by-trial, its results show that increasing the demand for justification raised the rate of trials in which subjects used compensatory strategies, although the overall strategy remained largely unchanged. This is one of the possible reasons for why the effect of justification in Study 1 could not be observed based on changes in the overall strategy. This thesis tries to bring the topic of justification in decision making back to research attention, and makes several contributions to decision-making research. First, it places the study of decision making in social contexts, improving its ecological validity; second, it shows preliminarily that the need for justification has an impact on the use of decision-making strategies and encourages decision-makers to integrate information in a more compensatory manner; third, it provides a theoretical basis for the proper use of justification supervision; and lastly, the strategy classification method applied in this thesis is innovative, has high accuracy and granularity, and can be highly useful for future decision research. The method can be used for investigating important but under-studied topics, such as adaptive decision-making. |
| 英文摘要 | 决策者的决策过程和方式往往会被决策结果相关他人所评价，此时决策者需要向他人解释以获得支持，这种对外解释的需求是一种典型的决策正当化需求。我们在生活中的很多决策多少都会对他人产生一定影响，因此对决策的正当化不仅是一个常见的现象，甚至被很多组织当作对决策者的监察手段之一。Bettman等人早在1998年就提出将“最大化决策正当性”作为决策者的四大决策目标之一，第一次系统性的将正当化作为影响决策过程的重要因素带入研究视野。但遗憾的是，决策中的正当化这一课题到目前还没有得到足够的关注和理解。本研究通过两项子研究探索了正当化需求对决策过程，特别是决策策略的影响。在研究决策策略时，现有策略识别方法大多仅依靠决策结果，缺乏将过程数据与结果数据整合起来、更为有效识别决策策略的技术方法。即使少数现有方法能整合不同数据类型，这些方法也难以做到对决策者在单次决策中使用的策略进行分类。这些问题导致了现有决策策略分类手段相对粗糙，进而影响研究结果的精度和效度。因此，本研究还试图结合眼动追踪的过程数据和机器学习算法，提出一种创新性的、能更细致准确区分决策策略的方法。 研究1初步探究了决策情景(风险决策或非风险决策)和正当化需求的不同水平(低:个体决策;中:有相关他人存在;高:需向相关他人解释)对决策策略的影响。我们采用多重测量最大似然法(Multiple-Measured MaximumLikelihood, MMML)对行为实验数据进行决策策略分类，分类结果表明大多数被试在两种决策情景下都被分类为使用加权平均的补偿性决策策略。正当化需求的不同水平对非风险决策的决策策略影响不大，但会使风险决策中使用补偿性策略的被试增多。不过这一增加在统计上并不显著。 研究1中的策略分类方法只能给出系列决策中的总体策略，但现实决策中决策者的策略却会随时改变，这可能导致因变量测量不准进而使研究结果有误。因此研究2采用新的策略分类方法对研究1中表现出一定趋势的实验条件进行再探索。具体说来，我们运用选择一学习一建模一分类(Choice-Learning-Training-Classification, CLTC)这一结合眼动数据和机器学习算法的新范式对每一试次的决策策略精细化分类，探讨正当化需求(中vs.高)对风险决策中的决策策略的影响。同时与MMML的对比测试了CLTC的分类效果。结果表明，在规定策略决策的策略分类上，CLTC的分类准确性远高于MMML;在对自由决策的策略分类上，CLTC与MMML的结果有明显的差异。CLTC对被试每一试次下的决策策略分类的结果表明，正当化需求的增加难以改变被试在所有试次中采用的总体策略，但会显著提高被试使用补偿性决策策略的试次的比率。这也是研究1中无法从总体策略改变上看出正当化效果的可能原因之一。 作为将决策中的正当化拉回研究者视野的探索性研究，本研究将决策放入社会情景，在提升研究生态效度的同时，初步证明了正当化需求对决策策略使用存在影响，会在一定程度上促使决策者对信息进行更充分的运用，为正当化监察的恰当运用提供理论基础。同时，研究中提出的策略分类范式也为今后的决策策略研究提供了准确率与分辨率更高的分类方法，可成为研究诸如“适应性决策”等重要课题的有力工具。 |
| 语种 | 中文 |
| 源URL | [http://ir.psych.ac.cn/handle/311026/43206]  |
| 专题 | 心理研究所_社会与工程心理学研究室 |
| 推荐引用方式 GB/T 7714 | 杨雨佳. 正当化需求对决策策略的影响[D]. 中国科学院心理研究所. 中国科学院大学. 2022. |
入库方式： OAI收割
来源：心理研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。