Impact of the implementation of differential privacy to decennial census data in understanding of health disparities in the United States

摘要 到2020年，基于差异隐私算法的新的差异避免方法的实施。 人口普查数据和其他数据产品可能会影响这些数据的有用性、估计和得出的比率的准确性 以及对健康差距等社会现象的批判性知识。美国人口普查局发布了一系列 允许在传统隐私和提议的隐私下产生的计数之间进行比较的演示产品 算法。我们通过研究县级年龄别死亡率的变化来测试差异隐私的后果 总人口和种族/民族比率(ASMR)。我们问分母如何变化，由于 实施不同的隐私，产生不同的ASMR。我们还建议通过以下方式审查趋势 比较2000-2010年两组ASMR死亡率比较得出的结论 分母的分母。此外，我们建议量化因实施差别私隐而产生的变化。 从回归模型得出的经验关联性。我们通过检查死亡率来实现这些目标 使用不同隐私技术制作的2010年十年计数的不同年份，并将它们与制作的年份进行比较 使用传统的方法。这些发现将突出实施差异隐私的后果 研究人为的种群组成变化所产生的比率变化。总体而言，我们将演示 使用从拟议的披露避免方法衍生的数据产品，同时强调关键 科学理解可能受到负面影响的情况。

Summary The implementation of new differential-avoidance methods, based on differential privacy algorithms, to the 2020 U.S. Census data and additional data products may affect the usefulness of these data, the accuracy of estimates and rates derived from them, and critical knowledge about social phenomena such as health disparities. The U.S. Census has released a series of demonstration products that allow comparisons between the counts produced under traditional and the proposed privacy algorithm. We test the ramifications of differential privacy by studying changes in the county-level age-specific mortality rates (ASMRs) for the overall population and racial/ethnic groups. We ask how changes in the denominators, due to the implementation of differential privacy, result in different ASMRs. We further propose the examination of trends by comparing the conclusions derived from 2000-2010 comparisons of ASMRs using mortality rates produced using both sets of denominators. Further, we propose quantifying the changes produced by the implementation of differential privacy in the empirical associations derived from regression models. We address these aims by examining mortality rates produced using different vintages of the 2010 decennial counts produced using differential privacy and comparing them with those produced using the traditional methods. These findings will highlight the consequences of implementing differential privacy for research examining changes in rates produced by artificial population composition changes. Overall, we will demonstrate the challenges of using the data products derived from the proposed disclosure avoidance method while highlighting critical instances where scientific understandings may be negatively impacted.