Models and Model Checking for Spatially-Varying Environmental Hazards and Decision Problems

有关公共卫生和公共支出的决定往往必须基于高度不确定的数据；例如，考虑到衡量个人经济决策和公共项目和政策影响的质量参差不齐。在环境政策领域，为知情决策所需的诸如污染物测量等数据通常是稀疏的，在空间上分散的，并且存在很大的测量误差。监管机构和其他政策制定者对环境和其他公共决策问题中典型的巨大不确定性的一种反应是使用“保守”（通常是夸大的）对暴露、风险等的估计。然而，由于认识到政策应以对可能的成本和收益的评估为基础，近年来越来越多地使用收益-成本分析。需要向前迈出的关键一步，特别是对于空间变化的环境危害，是校准风险估计：这意味着建议在不同领域采取不同的行动方针和不同的数据收集策略。为了有效地做到这一点，对相关暴露和风险进行空间建模是有用的。该项目的目标是在不确定性（由于信息不完整）和真正的潜在可变性的情况下，为空间变化的危险开发更可靠的模型和模型检查方法。近年来，统计学领域在利用贝叶斯方法对复杂数据结构建模方面取得了很大进展。需要取得更多进展的领域和研究者计划开展的工作包括模型拟合、计算、模型检查和使用图表或地图显示推论。研究人员计划特别关注模型检查和图形方法的使用，以建立对模型拟合结果的信心，以便个人和决策者将拥有值得信赖的工具，使他们在做出决策时更好地考虑不确定性和可变性。作为一个重要的例子，研究人员建议在对来自许多来源的家庭氡数据进行综合分析的基础上，在家庭氡风险补救的背景下发展他们的模型。

Decisions concerning public health and public expenditures must often be based on highly uncertain data; for example, consider the uneven quality of measurements of individual economic decisions and effects of public programs and policies. In the field of environmental policy, data such as pollutant measurements that are required for informed decisions are usually sparse, spatially dispersed, and subject to substantial measurement error. One response by regulators and other policy makers to the large uncertainties typical of environmental and other public decision problems has been the use of `conservative` (often inflated) estimates of exposure, risk, etc. However, the recognition that policy should be based on assessment of both likely costs and benefits has led to increased use of benefit-cost analysis in recent years. A key step forward that needs to be made, especially for spatially-varying environmental hazards, is to calibrate risk estimates: this means recommending different courses of actions and also different data-gathering strategies in different areas. In order to do this effectively, it is useful to spatially model the relevant exposures and risks. The goal of this project is to develop more reliable methods of models and model-checking for spatially-varying hazards, in settings with uncertainty (due to incomplete information) and also true underlying variability. In recent years, much progress has been made in the field of statistics in modeling complex data structures using Bayesian methods. Areas in which more progress needs to be made and on which the investigators plan to work include model fitting, computation, model checking, and display of inferences using graphs or maps. The investigators plan to particularly focus on the use of model-checking and graphical methods to build confidence in the results of the modeling fitting, so that individuals and policy-makers will have trustworthy tools to allow them to take better account of uncertainty and variability when making decisions. As an important example, the investigators propose to develop their model in the context of remediation of risks from home radon, based on a combined analysis of home radon data from many sources.