Addressing Disclosure Risk of Contextualized Microdata in Survey Design

该项目旨在增加有关一个人的邻里和个人特征，行为和健康结果的详细研究数据的可用性，这些信息对于研究关键的国家问题至关重要，例如健康差异。然而，必须在方便访问这些数据和保护研究参与者的匿名性之间取得微妙的平衡。为了应对对情境化微观数据日益增长的需求，大型国家调查通常会收集有关其主题的个人和地理属性的详细信息。然而，当数据被准备用于公共使用的文件时，这些重要的细节中的许多都被压制或粗糙化，以保护受访者的匿名性。这些限制减少了重要科学研究的机会，并给必须执行限制性数据使用协议的生产商和经销商带来了昂贵的负担。 很少有人知道如何保护被告的身份（即，披露风险）受到发布微数据文件的影响，微数据文件包含每个主题周围的县、区域、区块组和1/2英里地理区域的上下文属性。考虑到在研究开始时确定的因素，目前尚不清楚不同程度的敏感信息或不同的抽样设计和分析目的如何影响情境化微观数据的披露风险。至于在收集数据之后准备分发研究档案时通常处理的因素，目前尚不清楚选择不同的变量发布或应用各种统计技术限制披露在多大程度上影响了披露风险和数据的科学价值。有了这些决定因素的先验知识，数据生产者将能够预测有多少和哪些受访者有披露的风险，并调整其数据收集方法以保护他们。这种调整将保持和加强数据的广泛传播效用。此外，还可以衡量影响数据收集效率的因素，从而可以估计与修改抽样设计以实现披露目标有关的调查费用。 因此，本项目力求将披露风险纳入评估调查设计所用的概念和经验框架。在此过程中，我们首先开发和验证模型，预测不同抽样设计下的调查数据的组成。接下来，我们开发用于评估披露风险，分析效用和披露调查成本的措施和方法，这些措施和方法最适合评估抽样和数据库设计。最后，我们进行模拟，以收集风险，效用和成本的估计与广泛的抽样和数据库设计特征的研究。

DESCRIPTION (provided by applicant): This project seeks to increase the availability of detailed research data about a person's neighborhood and individual characteristics, behaviors, and health outcomes, information which is crucial for research on critical national issues, such as health disparities. However, a delicate balance must be struck between providing easy access to these data and protecting the anonymity of study participants. Responding to the rising demand for contextualized microdata, large national surveys typically collect meticulous information about their subjects' personal and geographic attributes. When data are prepared for public-use files, however, much of this important detail is either suppressed or coarsened to protect the anonymity of respondents. These limitations reduce opportunities for important scientific research and impose costly burdens on producers and distributors who must implement restrictive data use agreements. Little is known about how the ability to protect a respondent's identity (i.e., disclosure risk) is affected by releasing microdata files that contain the contextual attributes of counties, tracts, blockgroups, and 1/2-mile geographic areas surrounding each subject. Considering factors that are determined at the outset of a study, it is not known how disclosure risk of contextualized microdata is affected by varying levels of sensitive information, or different sampling designs and analytical purposes. Turning to factors that are usually addressed after data collection when research files are prepared for dissemination, it is not known to what extent that disclosure risk and the scientific value of data is affected by the selection of different variables for release or application of various statistical techniques to limit disclosure. With a priori knowledge of these determinants, data producers will be able to anticipate how many and which respondents are at risk of disclosure, and adapt their data collection methods to protect them. Such adjustments will preserve and enhance the utility of the data for broad dissemination. Also, factors that affect data collection efficiencies can then be measured, allowing for the estimation of survey costs associated with modifying sampling designs to meet disclosure goals. Hence this project seeks to incorporate disclosure risk into the conceptual and empirical frameworks used in the evaluation of survey designs. In so doing, we first develop and validate models that predict the composition of survey data under different sampling designs. Next we develop measures and methods used in the assessments of disclosure risk, analytical utility, and disclosure survey costs that are best suited for evaluating sampling and database designs. Lastly we conduct simulations to gather estimates of risk, utility, and cost for studies with a wide range of sampling and database design characteristics.