By Wendy Underhill
While we wait for redistricting data to arrive in August, there’s a little-known but hot-hot issue in the redistricting world: differential privacy, the system chosen by the U.S. Census Bureau to protect the privacy of census responses.
Differential privacy is one option to ensure disclosure avoidance; any disclosure avoidance option involves reporting data that is somewhat different than what was collected. The technical term is “adding noise.”
One more thing makes this issue particularly interesting: It’s not at all clear if the use of differential privacy has an advantage for either political party, or how it could impact the number of districts affected by the Voting Rights Act.
As attorney Jason Torchinsky put it in an argument earlier this year before a federal court in Alabama: “The way [Census] data is perturbed or moved around the state will have an impact on whether Alabama can draw or will need to draw no majority-minority districts for Congress, one majority-minority district for Congress, or potentially even two majority-minority districts for Congress.”
I’m neither an attorney nor a statistician. In this context, I am merely an observer and occasional reporter. With that note, here’s what I can report.
Alabama’s attorney general has sued the bureau, saying that the bureau’s proposal for how to report data—that is, with the application of differential privacy—will mean the data isn’t fit for use in redistricting.
Much of the argument centers on whether the population totals at the census block level—the smallest unit of census geography—will be accurate. Those who are concerned say that without accurate census block data, there’s no way to build electoral districts with any precision in terms of equal population. This applies to congressional districts, legislative districts and all kinds of local districts.
Rather than centering the argument on “how many census blocks are noisy,” one could ask, “is there an adequate noise-canceling effect once many census blocks are aggregated into electoral districts?” We know that at the state level, population and racial and ethnic characteristic data will be “true” for the state. Anything below that, we will have noise, under the current plan. But will the changes that noise represents have a material impact on redistricting, especially in smaller population districts or majority-minority districts?
To get to that answer, on April 28 the bureau released a data set researchers can review. Many have done so, and they’ve come up with very different answers:
The Southern Demographic Association has collected several more state-specific studies, and Esri has provided a blog post that helps explain how data users can creating a web map to compare the differential privacy values released by the bureau on April 28 with the published 2010 data values.
The bureau has heard from these researchers and your feedback is most welcome, too. It’s due by May 28.
Wendy Underhill is the director of NCSL's Elections and Redistricting Program.