Aggregation of Probabilistic Opinions

该建议侧重于单个决策者（DMs）用于组合来自多个非独立来源的概率信息的聚合过程。典型的例子是，投资者要综合各种财务顾问对某些股票升值机会的预测，而病人必须综合各种专家对某一特定医疗程序成功机会的信息。本文描述了一个模型，该模型假设(A) DM通过对各种预测进行平均来组合信息，以及(b) DM对总体的置信度与（可能加权的）平均预测的方差呈负相关。这个模型被用来得出一系列关于影响和驱动决策者信心的因素的预测。他们将在一系列实验中进行测试，以澄清(a) DM所使用的汇总规则的性质，(b)影响DM对最终汇总的信心的因素，(c)汇总与它们所激发的信心之间的依赖的性质和程度，以及(d)决定DM对某些顾问的偏好的因素。除了有助于我们理解收集意见所涉及的基本心理过程之外，研究结果还对许多领域中出现的几个具有重大实际意义的问题产生了直接影响，这些领域包括但不限于金融、军事和医疗决策。结果将提供（至少部分）解决以下问题的指南：(a) DM应该咨询什么样的顾问，以及多少顾问才能达到期望的信心水平？(b)什么因素使某些顾问在某些类型的决策问题中更或更不具有吸引力？(c)顾问之间惊人的高度相关性的来源是什么？各种因素的相对重要性是什么？人们如何利用这些信息以最佳方式选择顾问？

This proposal focuses on the aggregation process used by individual Decision-Makers (DMs) to combine probabilistic information from multiple, non-independent sources. Typical examples are investors who combine forecasts from various financial advisors regarding the chances of certain stocks to appreciate in value, and patients who must aggregate information from various experts about the chances of success of a given medical procedure.A model is described that assumes that (a) the DM combines information by averaging the various forecasts, and (b) the DM's confidence in the aggregate is inversely related to the variance of the (possibly weighted) mean forecast. This model is used to derive a series of predictions about the factors that affect and drive the DM's confidence. They will be tested in a series of experiments that would clarify (a) the nature of the aggregation rules used by DMs, (b) the factors that affect the DMs' confidence in the final aggregate, (c) the nature and level of dependence between the aggregates and the confidence they inspire, and (d) the factors that determine the DM's preference for certain advisors.In addition to contributing to our understanding of the basic psychological process involved in aggregating opinions, the results have immediate implications for several questions of considerable practical interest that arise in many fields including, but not restricted to, financial, military and medical decision-making. The results will provide (at least partial) guidelines for approaching questions such as:(a) What kind of, and how many, advisors should a DM consult to achieve a desired level of confidence?(b) What factors make some advisors more or less attractive in certain types of decision problems?(c) What are the sources of the surprisingly high correlation among advisors? What is the relative importance of the various factors and how can one use this information to choose advisors in an optimal fashion?