CAREER: The Intersection of Spatial Statistics and Differential Privacy

该职业奖将开发生成高质量、空间参考的公共使用数据的方法，同时解决数据保密性问题。获取高质量的公共使用数据对许多研究学科至关重要。然而，对人口规模较小的精细地理区域(例如人口普查区域)的分析往往产生统计上不可靠的推断。较小的区域也可能包含很少的研究参与者，从而增加了关于参与者的敏感信息泄露的风险，如个人的疾病或就业状况。该项目将在正式隐私文献和空间统计文献之间建立一个统一的框架，对隐私考虑和由此产生的数据的效用给予同等重视。这项研究的结果将对学术研究人员和联邦统计机构的工作人员都有价值。研究人员将与疾病控制和预防中心和国家卫生统计中心的研究人员合作。调查员将为联邦统计机构的工作人员举办关于空间统计和数据隐私的讲习班和短期课程。该项目还将创造贝叶斯推理和统计计算方面的本科生研究机会，并提供与空间统计和数据隐私相关的教育机会。该项目将开发贝叶斯统计方法，以生成空间参考合成数据，达到或超过美国联邦统计机构目前实施的隐私保护。空间统计文献中的小面积估计方法提供了一个框架，以利用数据中的复杂依赖关系来提高估计的精度。数据隐私文献中出现的方法可以用来掩盖或以其他方式对这些领域隐藏信息，以保护向数据主体作出的隐私保证，以换取他们的参与。综上所述，这两种方法在提供准确可靠的当地估计和需要模糊小区域估计与驻留在其中的数据主体之间的详细联系之间存在分析上的紧张。该项目将解决以下问题。首先，该项目将设计一个统计框架，以产生由空间参考的合成总量计数数据组成的大规模、不同于私人的公共使用数据储存库。这项工作的一个关键方面将是在计算效率和数据效用之间取得平衡。其次，该项目将为从一大类空间模型合成数据建立标准，以满足正式的隐私保护。这项工作的结果将是在实用方面提供实质性收益的方法，并有助于将数据分析和合成数据生成的任务结合起来，以避免重复。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

This CAREER award will develop methods for generating high-quality, spatially referenced public-use data while addressing data confidentiality concerns. Access to high-quality public-use data is critical for many research disciplines. However, analyses of fine-scale geographic regions with small population sizes (e.g., census tracts) often yield statistically unreliable inference. Small areas also may contain few study participants, thus increasing the risk of disclosure of sensitive information about a participant, such as an individual's disease or employment status. This project will create a unifying framework between the formal privacy literature and the spatial statistics literature that gives equal weight to privacy considerations and the utility of the resulting data. The results of this research will be of value both to academic researchers and staff at the Federal statistical agencies. The investigator will collaborate with researchers at the Centers of Disease Control and Prevention and the National Center for Health Statistics. Workshops and short courses will be developed by the investigator on spatial statistics and data privacy for staff at the Federal statistical agencies. The project also will create undergraduate research opportunities in Bayesian inference and statistical computing and provide educational opportunities related to spatial statistics and data privacy.This project will develop Bayesian statistical methods for generating spatially referenced synthetic data that achieve or exceed the privacy protections currently implemented by U.S. Federal statistical agencies. Small area estimation methods from the spatial statistics literature provide a framework to leverage complex dependencies in the data to improve the precision of an estimate. Emerging methods from the data privacy literature may be used to mask or otherwise conceal information from these areas to protect the privacy guarantees made to the data subjects in exchange for their participation. Taken together, these two approaches present an analytic tension between providing accurate and reliable local estimates and the need to obscure detailed linkage between small area estimates and the data subjects residing therein. This project will tackle the following issues. First, the project will devise a statistical framework for producing massive, differentially private public-use data repositories comprised of spatially referenced synthetic aggregate count data. A key aspect of this work will be to strike a balance between computational efficiency and data utility. Second, the project will establish criteria for synthetic data from a broad class of spatial models to satisfy formal privacy protections. The result of this work will be methods that provide substantial gains in utility and help combine the tasks of data analysis and the generation of synthetic data to avoid redundancies.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.