U.S. flag

An official website of the United States government

Skip Header


Preliminary Report on Differentially Private Post-Stratification

Written by:
Working Paper Number ced-wp-2023-004

Abstract

Differential Privacy protects individual privacy by adding random noise conditioned on the impact a single individual can have on an analysis outcome. With weighted data, the impact of an individual would appear to be based on their weight, but it can be greater than this. We show that post-stratification, where weights are assigned to ensure that results on a sample conform to known population totals, can if done naively result in potentially unbounded impact of an individual; it is not only an individual’s own weight that affects the result, but that individual’s impact on other weights in the post-stratification process. We present a probabilistic post-stratification method that enables us to bound the impact of an individual on an overall query such that it is feasible to satisfy differential privacy while maintaining the reduction in sampling error from post-stratification.

Page Last Revised - June 9, 2023
Is this page helpful?
Thumbs Up Image Yes Thumbs Down Image No
NO THANKS
255 characters maximum 255 characters maximum reached
Thank you for your feedback.
Comments or suggestions?

Top

Back to Header