CORE--STATISTICAL DATA ENCLAVE

核心——统计数据飞地

• 批准号: 6352569
• 负责人: P.J. ERIC STALLARD
• 金额: $ 11万
• 依托单位: DUKE UNIVERSITY
• 依托单位国家: 美国
• 财政年份: 2000
• 资助国家: 美国
• 起止时间: 2000-09-15 至 2001-06-30
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/6352569
• 关键词: aging; behavioral /social science research tag; biomedical facility; data collection methodology /evaluation; health care model; human data; human old age (65+); human population study; human very old age (85+); longitudinal human study; mathematical model; model design /development; statistical service /center; statistics /biometry

In Core E we will perform two major tasks. One task is the development of new analytic models for dealing with longitudinal data from both national and local-area studies of elderly and oldest-old populations. The second task is the investigation of statistical models for masking crucial variables in a data set to prevent individuals from being identified from data sets-yet generate data sets that retain considerable and sufficient detail for multiple research purposes. In fact, on a fundamental mathematical level these two analytic tasks are related-though possibly as inverse problems. In analyzing longitudinal data for human populations the assessments of the age trajectories of health and functional changes are generally "incomplete," i.e., measurements occur at fixed times often several years apart. Thus there is considerable "missing data" on the processes being sampled in most longitudinal observation plans. The data masking problem is the reverse problem, i.e., what data has to be "masked" or made "missing" to prevent individuals from being identified. Thus, the measurement can either be removed or made "fuzzier" by reporting less detailed measurement intervals or by adding random noise to the variable. We will start by creating models with the basics of the Missing Information principle, the super-population theory of sampling (appropriate for modeling processes, the E-M algorithm and models of partly observed stochastic processes, and apply those models and concepts to different types of data (i.e., survey reports, medical cost data, genetic or DNA data) to see how to best proceed in a..) Analysis, and b.) Masking, with each. It is anticipated, that apart from the standard approaches to assessing confidentiality (e.g., as practiced by NCHS or RTI; see Core C), that multivariate procedures will be explored for data masking that are based on stochastic process models, and that can use mathematical and statistical principles to approximate multivariate distributions to the degree desired.

在核心E中，我们将执行两个主要任务。一项任务是开发新的分析模型，以处理来自国家和最古老的人群的国家和地方地区研究的纵向数据。第二个任务是研究统计模型，以掩盖数据集中的关键变量，以防止个人从数据集中识别出迄今为止生成的数据集，这些数据集保留了大量和足够的详细信息，以实现多个研究目的。实际上，在基本的数学层面上，这两个分析任务是相关的 - 尽管可能是反问题。在分析人口纵向数据时，对健康和功能变化的年龄轨迹的评估通常是“不完整的”，即，在固定时间经常相隔几年的固定时间进行测量。因此，在大多数纵向观察计划中采样的过程中存在相当大的“丢失数据”。数据掩盖问题是反向问题，即必须“掩盖”或“缺少”哪些数据以防止个人被识别。因此，可以通过报告较少详细的测量间隔或通过向变量添加随机噪声来删除或制造“模糊”。我们将首先创建具有缺失信息原理的基础的模型，即取样的超人选理论（适用于建模过程，E-M算法和部分观察到的随机过程的模型，并将这些模型和概念应用于不同类型的数据（即，调查报告，调查报告，医疗成本数据，遗传或DNA数据），以查看如何最佳地进行分析。可以预见的是，除了评估机密性的标准方法（例如，根据NCHS或RTI的实践；请参见Core C），还将探索多元程序以基于随机过程模型的数据掩盖，并且可以使用数学和统计原理来近似多元分配，以将多变量分配到近似于所需的程度。