Formal Privacy for Modern Nonparametric Statistics
Modern nonparametric (NP) statistics is an increasingly important and expanding tool set in data analytics as more large, complex data are gathered and analyzed. However, corresponding privacy concerns that arise require novel methods to balance privacy guarantees with statistical utility. NP methods present a unique challenge for privacy because the resulting summaries can contain significant amounts of individual-level information.
Modern NP statistics consists of tools to analyze data without assuming the data are distributed according to some pre-specified parametric family, such as assuming the data is distributed normally. Often, the goal of NP statistics is to estimate a function (e.g., probability density or regression function) with only limited assumptions, such as the number of derivatives.
Some content is only viewable by ASA Members. Please login or become an ASA member to gain access.
Hello. I am the Disclosure Review Board Chair at the Department of Education. I work in the Student Privacy Policy Office (SPPO) under the direction of the Senior Agency Official for Privacy (SAOP). I have two related questions. First, how may I access Chance articles dealing with privacy and managing reidentification risk? Second, I Chair the Managing Reidentification Risk interagency working group, which is sponsored by the Federal Privacy Council (FPC). I came across an AmStat section or working group focused on privacy statistical disclosure methods. For the life of me I cannot locate that web page again. Can you direct me to that group? And is it possible to be put in touch with someone in the group? I recall the page I stumbled across contained no emails.