Doctoral Dissertation Research in DRMS: The coupled impact of conflict and imprecision of multiple forecasts

技术描述人们经常依赖多个专家的预测来做出决策。这包括日常决策，如利用多种天气预报，以及改变生活的决定，如寻求多个医生对严重诊断的意见。以前的研究区分了相互矛盾和不精确的预测。当多个顾问提供不同但精确的预测时，就会出现冲突（例如，一个专家预测6英寸的雪，另一个专家预测只有1英寸）。当顾问们同意他们的不精确性时，观察到不精确性（例如，两位专家预测1到6英寸的雪）。以前的模型分别处理冲突和不精确性，但它们很少被很好地区分，而且往往是相关的（例如，一个专家预测1到5英寸的雪，另一个预测2到6英寸）。该提案探讨了这两个因素（冲突和不精确）的各种组合如何改变人们的看法和选择的理论假设的基础上，冲突和不精确的功能的基本属性的预测集。最终目标是确定最佳模式的聚合和提出多种预测调用准确的看法的信息。该研究计划包括一系列在线实验，涉及全国代表性样本，比较冲突和不精确性的各种组合，这些组合因两个关键因素（相似性和对称性）的类型和程度而异。相似性是指预测之间的关系，有三类：不重叠的不相交集合，部分重叠的相交集合，以及一个集合完全嵌入另一个集合的嵌套集合。对称性指的是集合围绕中心的平衡（或者在数学上，所有预测的平均值与其中位数的相对偏差）。不对称的方向可以变化，因此正（负）偏斜的集合具有较少的高（低）值。参与者将在定义明确的领域中查看几组区间预测，并将估计最可能的值、可能值的范围，并根据关键属性（例如，模糊性、可信性、信息性等）。我们还将操纵主题域（使用金融，健康和政治背景）来测试结果的普遍性。更广泛的意义和重要性分析将量化的影响，各种因素操纵的决策者？决策 我们还将使用各种降维和分类技术（多维缩放与聚类分析配对），根据“心理距离”映射各种投影集，以帮助理解驱动人们反应的认知过程。这项研究是一个重要的一步，以改善沟通的风险和不确定性的基础上经验观察心理学原则。这些步骤对于弥合专家和外行之间的差距至关重要，因为非专家往往会受到信息呈现方式的影响。这些结果与军事情报、气候预报等许多领域有关，他们必须做出谨慎的决定，投入他们有限的时间和金钱，以减少专家之间和专家内部的不确定性。

Technical DescriptionPeople often rely on projections from multiple experts to make decisions. This includes daily decisions like utilizing multiple weather forecasts as well as life-changing decisions like seeking multiple doctors' opinions about a serious diagnosis. Previous research has differentiated between conflicting and imprecise forecasts. Conflict is observed when multiple advisers offer different, but precise, forecasts (e.g. one expert projects 6 inches of snow and another projects only 1 inch). Imprecision is observed when the advisers agree in their imprecision (e.g. both experts forecast 1 to 6 inches of snow). Previous models treat conflict and imprecision separately, but they are rarely well-differentiated and often correlated (e.g. one expert predicts 1 to 5 inches of snow and another predicts 2 to 6 inches). This proposal examines how various combinations of these two factors (conflict and imprecision) alter people's perceptions and choices based on the theoretical hypothesis that conflict and imprecision are functions of the underlying attributes of the forecast sets. The ultimate goal is to determine optimal modes of aggregating and presenting multiple forecasts to invoke accurate perceptions of the information. The research plan includes a series of online experiments involving nationally-representative samples comparing various combinations of conflict and imprecision varied by the type and degree of two key set factors, similarity and symmetry. Similarity refers to the relationship between the forecasts and has three categories: disjoint sets that do not overlap, intersecting sets that partially overlap, and nested sets where one set is fully embedded in the other. Symmetry refers to the balance of the sets around the center (or mathematically, the relative deviation of the mean of all forecasts from their median). The direction of asymmetry can vary, so positively (negatively) skewed sets have fewer high (low) values. Participants will view several sets of interval forecasts in well-defined domains and will estimate the most likely value, range of possible values, and rate the sets on key attributes (e.g., ambiguity, credibility, informativeness, etc.). We will also manipulate the topic domains (using finance, health, and politics contexts) to test the generalizability of the results. Broader Significance and ImportanceThe analysis will quantify the effects of the various factors manipulated on the decision makers? decisions. We will also use various dimensionality reduction and classification techniques (multidimensional scaling paired with cluster analysis) to map the various projection sets based on their "psychological distances" to help understand the cognitive processes that drive people's responses. This study is an important step toward improving the communication of risk and uncertainty based on empirically observed psychological principles. Such steps are vital to bridging the gap between experts and laypeople because non-experts are often disproportionally influenced by how information is presented. These results are relevant to many domains such as military intelligence, climate forecasting, etc., which must make careful decisions to invest their scarce time and money to reduce uncertainties between and within experts.