ASA Connect

Hi members! The ASA has been asked for input on the National Privacy Research Strategy… specifically, which research areas outlined in the document would be the most promising for collaboration between statisticians and security/privacy researchers. Here are the topics:

• Foster multidisciplinary approach to privacy research and solutions;
• Understand and measure privacy desires and impacts;
• Develop system design methods that incorporate privacy desires, requirements, and controls;
• Increase transparency of data collection, sharing, use, and retention;
• Assure that information flows and use are consistent with privacy rules;
• Develop approaches for remediation and recovery; and
• Reduce privacy risks of analytical algorithms.

Some proposed topics include record linkage, data dissemination strategies, data access protocols, input and integration, more research on evaluating sensitive variables in administrative and survey data, and collection and use of essential versus auxiliary information. What are your thoughts? Does anyone know of any statisticians active in these areas of research? Are there any topics not listed that should be considered? Let us know by replying here or emailing me at nussbaum@amstat.org. Thanks!

------------------------------
Amy Nussbaum
Science Policy Fellow
American Statistical Association
------------------------------

One thing that's not statistical but should be considered is avoidance of things that waste time.  Here are some HIPA examples:

Doctors won't email me or let me email them, because email is "not secure", but they are perfectly happy to fax things even though most of us use fax machines that are shared with co-workers.

I need to sign an authorization permitting my doctor to give me copies of my medical records.

Many government offices will let you fax things to them but won't let you email things.  Social Security discards records so fixing mistakes often means start from scratch.

Social Security Supplemental Security Income, Social Security Disability, and Division of Developmental Disabilities don't share information with each other so one has to start from scratch with each one.

Always think about the Law of Unexpected Consequences.

------------------------------
Emil M Friedman, PhD
emilfriedman@gmail.com
http://www.statisticalconsulting.org

Amy,

Micro-aggregation methods assure privacy and confidentiality of patients (experimental units) and allow sharing of otherwise sensitive information. But some micro-aggregation strategies are much more useful to both researchers and the general public than others. The current federal guideline is that each aggregated statistic to be revealed (published) must be derived from at least 11 experimental units.

The non-parametric / robust strategy for statistical inference known as Local Control is based upon a micro-aggregation strategy fully consistent with propensity scoring theory ...but is more "fine." Stan Young and I have published several papers describing the LC approach and illustrating its application to the largest publicly available data sets we could find or create ourselves.

Much more research on (and scientific awareness of) this promising new combination of fundamental statistical concepts is badly needed. LC combines estimation of entire treatment effect-size distributions with prediction of individual estimates using confounding characteristics, looking for "fixed" (rather than random) effects.  Sound applications of LC Strategy have great potential.

------------------------------
Robert Obenchain
Principal Consultant
localcontrolstatistics.org

Hi all! Thanks so much for the suggestions.... I've also gotten several emails and I appreciate everything so much! 

------------------------------
Amy Nussbaum
Science Policy Fellow
American Statistical Association