Princeton Election Consortium

Innovations in democracy since 2004

Outcome: Biden 306 EV (D+1.2% from toss-up), Senate 50 D (D+1.0%)
Nov 3 polls: Biden 342 EV (D+5.3%), Senate 50-55 D (D+3.9%), House control D+4.6%
Moneyball states: President AZ NE-2 NV, Senate MT ME AK, Legislatures KS TX NC

What is differential privacy and could it affect redistricting?

June 2nd, 2021, 5:05pm by Zachariah Sippy

In March, Alabama sued the U.S. Department of Commerce. At the heart of their suit is the claim that the Census Bureau’s disclosure avoidance system— designed to protect the privacy of census respondents, inhibits the state from accurately redistricting.

Alabama, along with more than 15 other states, has argued that the Bureau’s privacy scheme is “arbitrary and capricious,” and may even violate the Voting Rights Act and/or Fifth Amendment.

For decades, the Census Bureau has been able to protect individuals’ data by using a simple swapping technique. But in recent years computer scientists and statisticians developed a better system: “differential privacy,” a method of adding small random numbers to block-level counts.

Alabama and other states fear that this new system is insufficiently accurate. Indeed, they’ve pointed to research by Christopher Kenny and other students of Prof. Kosuke Imai at Harvard University, that seems to argue as such.

But Kenny and Imai’s research has not yet been through peer review. Therefore our group at Princeton University, the Electoral Innovation Lab, did a rigorous review of the claims made in the Harvard paper.

Our research found four major problems in their research. And overall, we were not convinced that the Census Bureau’s policy has the potential to impede redistricting. Indeed, we found that any resulting differences from the privacy scheme when aggregated on scale were smaller than the rounding error.

For a deeper dive into our research, click here for our full analysis of the Disclosure Avoidance System.

Tags: Redistricting

3 Comments so far ↓


    Have Kenny and Imai had a chance to your review? If so, what do they say?

    • Sam Wang

      I believe they have moderated their statements somewhat, especially those with practical consequences. They responded on their website.

  • James McDonald

    It seems one could run a program to double-check any proposed map using the original unmodified data, but answer only with an overall assessment.

    That should suffice to dispel any arguments that the map violated any criteria, while revealing effectively zero privacy information.

    It wouldn’t even need to be a zero-knowledge proof. You just need to keep the data confidential.

Leave a Comment