非交互个体的估计值分布研究及其在群体智慧中的应用  

Research on the distribution of non-interactive individual’s estimates with applications in the wisdom of crowds

作　　者：杨雷[1] 曹希雯 YANG Lei;CAO Xiwen(School of Business Administration,South China University of Technology,Guangzhou 510641,China)

机构地区：[1]华南理工大学工商管理学院,广州510641

出　　处：《系统工程理论与实践》2021年第10期2733-2747,共15页Systems Engineering-Theory & Practice

基　　金：国家自然科学基金(71771092)。

摘　　要：进行高风险决策和预测未知事件均涉及到大量估计任务.准确地估计对个体是困难的,群体智慧提供一种在估计任务中辅助决策的方法.群体环境中非交互个体的估计值是随机变量,而现有文献对这一随机变量分布类型的设定存在矛盾.研究群体智慧需要确定个体估计值的分布规律.本文选取五个检验分布,对个体估值数据进行K-S检验,并提出相对差异度准则确定最优拟合分布.结果表明对数正态分布为最佳拟合分布的次数远高于其它分布类型,且其相对差异度显著小于其他检验分布.不论群体规模大小,不论估计值的数量级大小和估计任务难度,对数正态分布均是个体估计值分布的最佳拟合.基于对个体估计值分布的确定,研究设计两个仿真实验:实验一比较聚合统计量的准确性,结果表明几何平均是比算术平均更合理准确的聚合统计量;实验二探究群体多样性和群体智慧效度的关系,结果表明在个体估计不准确的情况下,提高群体多样性是提高群体智慧的有效途径.用对数正态分布可以模拟出各种规模大小人群的估计值,从而为学者深入研究群体智慧提供一种高效率、低成本的研究工具.Making high-risk decisions and predicting unknown events involve a large number of estimation tasks.Making accurate estimates is difficult for individuals.Fortunately,the wisdom of crowds provides a method to assist decision-making in estimation tasks.The estimate of non-interactive individual under group conditions is a random variable,while the existing literature has a contradiction in the assumption of its distribution type.The determination of distribution type plays a crucial role in the study of the wisdom of crowds.In this paper,five test distributions are selected to perform Kolmogorov-Smirnov test on the experimental data,and a criterion of relative difference is proposed to determine the optimal fitting distribution.Results show that logarithmic normal distribution is selected as the best fitting distribution far more frequently than other distribution types,and the relative difference of logarithmic normal distribution is significantly lower than that of other test distributions.The optimal fitting distribution of individual estimates is logarithmic normal distribution,regardless of group size,the magnitudes of values and task difficulty.Based on the determination of distribution type,we design two simulation experiments.One to explore the accuracy of aggregation statistics,and the other to explore the relationship between group diversity and the wisdom of crowds validity.Results show that the geometric mean is a more reasonable and accurate aggregate statistic than the arithmetic mean,and improving group diversity is an effective way to enhance the wisdom of crowds when individual estimates are not accurate.The logarithmic normal distribution can be used to simulate the estimates of various group sizes,thus providing an efficient and low-cost research tool for scholars to study the wisdom of crowds in depth.

关 键 词：群体智慧 估计判断 分布研究 群体决策 

分 类 号：C93[经济管理—管理学]