个体概率估计值极端化校准聚合方法的群体智慧效度研究

Study on wisdom of crowds validity of calibration polymerization method for individual probability estimation based on extremization

群体智慧研究表明,聚合个体判断而形成的群体判断将优于单个个体的判断准确度.为了进一步提高群体智慧效度,改善聚合质量,消除个体在判断和启发过程中存在的系统性偏见,本文构建了极端化转移的多个聚合模型,设立准确度和优胜度两个指标来定量化评价个体概率估计值聚合的群体智慧效度.通过在职工商管理硕士研究生的概率判断实验,获取个体概率估计值数据.基于个体概率估计数据,比较研究个体概率聚合的多种顺序策略和区域策略,完成了极端化函数选择和参数优化.研究发现:1)只对区间[0, 0.45]和[0.55, 1]内的概率估计值进行极端化,局部极端化策略比全局极端化策略的绩效更优;2)先平均后极端化(ATC)策略优于先极端化后平均(CTA)策略;3)在先平均后极端化(ATC)策略中,反S型极端化函数能够对Karmarkar极端化函数做出改善;4)极端化校准聚合比传统的算术平均方法的绩效更优.论文发现了聚合个体概率估计值一种更优的策略方法,在相同个体估值条件下,可显著提高群体智慧效度.

The study of wisdom of crowds shows that the group judgment formed by aggregating individual judgments is more accurate than the individual judgments. In order to improve the aggregation quality and eliminate the systematic bias in individual judgment and heuristic process,this paper concretely constructs a variety of aggregation models based on extreme ideas. Two indicators, accuracy and superiority, were set up to quantitatively evaluate the wisdom of crowds validity of aggregation. The data of individual probability estimation were obtained through the probability judgment experiment on Master of Business Administration. Based on the data of individual probability estimation, Several sequential strategies and regional strategies of individual probability aggregation are compared, and the extreme function selection and parameter optimization are completed. The results show that: 1) If only the probability estimates within intervals [0, 0.45] and [0.55, 1] are polarized, the local extreme strategy performs better than the global extreme strategy. 2) Average-then-calibrate strategy is better than calibrate-then-average strategy;3) In the average-then-calibrate strategy, the anti-S-type polarizing function proposed in this paper can outperform the classical polarizing function. 4) The performance of extreme calibration aggregation is better than that of traditional arithmetic average method. This paper proposes a better strategy of aggregating individual probability estimates, which can significantly improve the wisdom of the crowds validity under the same individual valuation conditions.