NCRN-MN: Triangle Census Research Network

许多联邦统计机构的主要任务是向公众传播数据，以便进行二次分析。然而，由于意外的机密性泄露、无反应和错误数据的风险，以及收集许多详细属性的调查成本的增加，传播面临越来越大的挑战。三角人口普查研究网络(TCRN)将开发广泛适用的方法，以改变和改进联邦统计系统的数据传播做法。特别是，TCRN将根据专门为纵向和多层次高维数据建立的灵活的非参数贝叶斯模型，开发发布多个推算的合成数据集的理论和方法，从而推动以高质量和可接受的保密泄露风险传播公共使用数据的方法和工具。TCRN将开发在编辑后的数据中纳入调查权重的方法，这些方法可以改善统计估计，而不会导致保密披露。该项目还将开发计算机系统框架，向二级分析员提供对编辑数据的推断质量的反馈，并将开发基于线性规划和贝叶斯建模的融合创建综合联想表的理论和方法。TCRN将改进处理丢失和错误数据的方法和做法，方法是开发框架，将统计学和运筹学的范例结合起来，同时对丢失数据进行补充，并对错误数据进行编辑。该项目还将开发非参数贝叶斯方法，用于纵向和多层次的高维缺失数据的多重补偿。最后，为了加强各机构综合来自多个来源的信息的能力，技合研究网络将开发一些方法，供各机构和二级分析员使用，以便在不完美的记录链接环境中适当地解释推论中的不确定性，并通过多重推定的数据集在公共使用的数据产品中传递这种不确定性。TCRN还将开发统计方法，将来自不依赖记录链接的多个数据来源的信息合并在一起。TCRN方法的发展将改变统计机构在统计披露限制、缺失数据和综合信息方面处理数据传播的方式。这些进展将为联邦机构提供发布效用更高的数据产品的选择，从而导致科学进步和改善政策制定。TCRN将把这些方法应用于人口普查局的主要数据产品，从而改进对这些数据集的数百次二次分析。TCRN的跨学科团队将使用这些数据产品来回答对政策制定具有重要影响的老龄化、经济和社会福利方面的问题。作为研究的组成部分，TCRN将参与并向联邦机构的博士后研究员、研究生和统计学家提供教育机会，从而培养和培训数据传播研究和实践方面的未来领导者。这项活动得到了NSF-人口普查研究网络资助机会的支持。

A primary mission of many federal statistical agencies is to disseminate data to the public for secondary analysis. However, dissemination is increasingly challenging due to risks of unintended confidentiality breaches, nonresponse and faulty data, and the costs of mounting surveys that collect many detailed attributes. The Triangle Census Research Network (TCRN) will develop broadly applicable methodologies that will transform and improve data dissemination practice in the federal statistical system. In particular, the TCRN will advance methodologies and tools for disseminating public use data with high quality and acceptable risks of confidentiality breaches by developing theory and methodology for releasing multiply imputed, synthetic datasets based on flexible, nonparametric Bayesian models built specifically for high-dimensional data with longitudinal and multi-level aspects. TCRN will develop approaches for including survey weights in redacted data that can improve statistical estimation without leading to confidentiality disclosures. The project also will develop the framework for computer systems that provide secondary analysts with feedback on the quality of inferences from redacted data, and it will develop theory and methodology for creating synthetic contingency tables based on fusions of linear programming and Bayesian modeling. The TCRN will improve methodology and practice for handling missing and faulty data by developing frameworks for simultaneous imputation of missing data and editing of faulty data by integrating paradigms from statistics and operations research. The project also will develop nonparametric Bayesian methodology for multiple imputation of missing data in high dimensions with longitudinal and multi-level aspects. Finally, to enhance agencies' abilities to integrate information from multiple sources, the TCRN will develop methods that agencies and secondary analysts can use to properly account for uncertainty in inferences in imperfect record linkage settings, as well as to pass on that uncertainty in public use data products via multiply imputed datasets. TCRN also will develop statistical approaches to combining information from multiple data sources that do not depend on record linkage.The methodological developments of the TCRN will transform the way statistical agencies handle data dissemination with regard to statistical disclosure limitation, missing data, and integrating information. These developments will offer federal agencies options for releasing data products with increased utility, leading to advances in science and improved policy making. The TCRN will apply the methodologies to major Census Bureau data products, thereby improving the hundreds of secondary analyses of these datasets. The interdisciplinary team of the TCRN will use these data products to answer questions in aging, economics, and social welfare that have important implications for policy making. As an integral part of the research, the TCRN will involve and offer educational opportunities to postdoctoral fellows, graduate students, and statisticians at federal agencies, thus developing and training future leaders in data dissemination research and practice. This activity is supported by the NSF-Census Research Network funding opportunity.