Assessing the Accuracy of Self-reported Pollution Data

信息提供越来越多地被用作一种监管工具。环境保护局的有毒物质释放清单（TRI）计划要求处理特定化学品阈值的制造设施每年报告这些有毒物质的释放和转移。本研究旨在探讨TRI数据的准确性以及导致自我报告错误或逃避的因素。本研究使用两种不同的方法来评估TRI数据的准确性。对于TRI报告所涵盖的12种化学品，EPA使用全国范围内的监测网络对这些化学品的空气浓度进行采样。地理信息系统（GIS）软件允许人们确定哪些污染设施在环保局监测范围内。因此，研究人员可以比较监测数据中测量的污染趋势如何与污染设施自我报告的空气排放趋势相匹配。为了调查监测数据和TRI数据之间的潜在差异，分析还将探讨周围社区的性质，州环境执法以及公司和设施级别的特征如何影响报告的TRI空气排放的表观准确性。将报告的TRI数字与排放数字的预期分布进行比较，提供了评估污染数据准确性的第二种方法。关键的见解是，如果设施估计排放量时有向下的偏差，数字的总体分布将不会遵循与监测数据中实际数字相同的模式。这项研究将证明，选择物理监测和数据模式的统计偏差分析的程度，可以帮助确定自我报告的数据的准确性。

Information provision is increasingly used as a regulatory tool. The Environmental Protection Agency's Toxics Release Inventory (TRI) program requires manufacturing facilities that handle threshold amounts of specific chemicals to report yearly their releases and transfers of these toxic substances. This project investigates how accurate TRI data are and what factors give rise to errors or evasion in self-reporting.The proposed research uses two different methods to assess the accuracy of TRI data. For 12 of the chemicals covered by TRI reporting, the EPA samples air concentrations of these chemicals using a network of monitors across the country. Geographic information systems (GIS) software allow one to determine which polluting facilities are within range of the EPA's monitors. The researchers can thus compare how measured trends in pollution from monitoring data match self-reported trends on air releases by the polluting facilities. To investigate potential divergences between the monitoring data and TRI figures, the analysis will also explore how the nature of the surrounding community, state environmental enforcement, and company and facility-level characteristics affect the apparent accuracy of the reported TRI air emissions. The comparison of reported TRI figures with expected distributions of emissions digits offers a second way to assess the accuracy of pollution data. The key insight is that if facilities are estimating emissions with a downward bias, the overall distribution of digits will not follow the same pattern as the actual digits in the monitor data. This research will demonstrate the degree that selected physical monitoring and analysis of data patterns for statistical bias can help determine the accuracy of self-reported data.