Blog Viewer

Q&A with Michael Hawes, senior advisor for Data Access and Privacy at the U.S. Census Bureau

  
In an effort to help the community better understand the U.S. Census Bureau's decision making process in applying differential privacy to the 2020 Decennial Census, the ASA P&C committee asked Michael Hawes to answer the following questions. We thank Michael for his time and his responses.

Q: What methods and metrics are the Census Bureau using to measure and evaluate potential changes in the level of privacy and accuracy? How is the Bureau comparing the relative value of these changes?

A: The framework of differential privacy allows the Census Bureau to precisely quantify the maximum amount of information that an attacker seeking to re-identify respondents in our published data could possibly learn about them. That threshold is established in what is known as the “privacy-loss budget,” which then determines how much noise must be injected into the published statistics.  Much like a monetary budget, the lower your privacy-loss budget, the less privacy protection you are willing to give up (but the more noise that will be added to the data).  A higher privacy-loss budget will lead to more accurate data, but with a corresponding reduction in the degree of privacy protection afforded to the respondents. 

The Census Bureau’s Data Stewardship Executive Policy Committee (DSEP) has not yet set the privacy-loss budget for the 2020 Census. That decision will be made based on an extensive analysis and evaluation of the accuracy of our statistics at a wide range of privacy-loss budgets, and weighing the resulting fitness-for-use of these data for an array of priority use cases against the risk to our respondents at those corresponding levels of accuracy. 

To facilitate these evaluations and decision-making, the Census Bureau has been actively engaging with our data users and stakeholders to learn and document how they use census data, and what their fitness-for-use requirements for those uses are. We are working closely with our National Advisory Committee and the Census Scientific Advisory Committee, and with expert groups from the National Academies of Sciences, Engineering, and Medicine’s Committee on National Statistics. We have also engaged extensively with numerous federal agencies, state and local governments, and American Indian and Alaska Native tribal leaders, along with professional associations, the academic community, and other stakeholders. Based on the feedback we have received from this engagement, we have developed an extensive set of metrics that capture these diverse uses of Census data, and have been using those metrics to evaluate our implementation of differential privacy for the 2020 Census. As new use cases are brought to our attention, we are improving and expanding these evaluation metrics accordingly.  


Q: What thresholds or standards is the Census Bureau employing to determine whether the level of privacy or accuracy is acceptable?  

A: DSEP has not yet made the final policy decisions governing the setting and allocation of the privacy-loss budget for the 2020 Census. As such, the thresholds or standards for sufficient privacy protection and for sufficient fitness-for-use for the priority uses of census data are still under discussion. The determination of these thresholds for appropriate levels of accuracy and privacy protection will be based on the extensive stakeholder engagement discussed above and the recommendations we receive from our advisory committees, along with consultation with our Office of General Counsel and members of the privacy and civil rights communities. In making their decisions, DSEP will also consider the results of extensive qualitative and quantitative research that the Census Bureau has conducted on public attitudes towards privacy and data use. 

0 comments
9 views

Permalink

Tag