Methods to Protect Privacy in State Longitudinal Data Systems Research Files

联邦资助的州学生纵向数据系统（SLDS）的扩展提供了丰富的数据来源，在STEM教育研究和评估中具有巨大的潜力。然而，《家庭教育权利和隐私法》的要求阻碍了对这些数据的访问。在这项研究中，研究人员将研究统计披露控制的两种一般方法如何使各州在遵守FERPA标准的同时共享数据。研究人员将使用风险分层的插补和数据掩蔽，包括来自五个同意授权使用其数据的州的数据的变量交叉表。他们将创建受保护的数据集，研究这些数据集以确定它们是否可以防止泄露，并进行一些分析以确定它们是否产生与使用原始数据进行的相应分析基本相同的答案。保护隐私是对不断增长的数据的关注的一个重要组成部分，这些数据正在收集关于所有社会成员的数据。然而，收集无法用于合法研究和评估的数据会降低收集的价值。各国需要更好的机制来确保其收集的数据的隐私，同时还需要保证其向研究人员公布的数据结果将为研究和评价问题提供有效的答案。该项目将建立和测试数据屏蔽模型，这是有效使用大规模状态纵向数据系统的必要基础设施。

The expansion of federally funded state student longitudinal data systems (SLDS) provides a rich source of data that has great potential in STEM education research and evaluation. However, access to that data has been hampered by the requirements of the Family Education Rights and Privacy Act (FERPA). The researchers in this study will examine the ways in which two general approaches to statistical disclosure control will enable states to share data while still complying with the standards of FERPA. The researchers will use imputation and data masking of risk strata with the inclusion of cross tabulations of variables with data from five states that have agreed to authorize the use of their data. They will create protected datasets, study them to determine whether they protect against disclosure and carry out a number of analyses to determine whether they yield essentially the same answers as the corresponding analyses using the original data.Protecting against the loss of privacy is an essential component of the concerns about the ever-growing data that are being collected about all members of society. However, collecting data that are not available for legitimate research and evaluation decreases the value of that collection. States need better mechanisms to ensure privacy of data they collect. At the same time they also need assurances that the findings from the data they release to researchers will provide valid answers to the research and evaluation questions posed. This project will build and test the data masking models that are the necessary infrastructure for the effective use of large scale state longitudinal data systems.