Formal Privacy for Multi-stage Probability Samples

Principal Investigator: Chris Clifton

Census products based on sampling, such as the U.S. Census Burueau's  American Community Survey, use a complex weighting process to improve accuracy.  We have been studying how to apply formal privacy methods to weighted samples.  A straightforward application of differential privacy results in unacceptably high noise:  as the potential exists for a highly weighted individual to have a huge impact on results, considerable noise is needed to protect against this possibility.

This project is developing techniques to address these issues:  providing formal privacy guarantees while releasing data with quantifiable data value.


Other PIs: Keith Merrill (Brandeis University)

Students: Shawn Merrill Eric Hanson (Brandeis University)

Keywords: Differential Privacy