By Wendy Underhill
Everyone is talking about whether a citizenship question will be on the 2020 census. We will know soon enough, when the U.S. Supreme Court decides Department of Commerce v New York. That decision will come any day now.
But there’s another question relating to the census: How will the census continue to ensure that data for individuals is kept private?
First, some background. All individuals’ answers to the census are to be kept confidential for 72 years, by law. To fulfill this requirement, the census has long added a bit of “noise” at the smallest reporting levels so that, say, an American Indian woman in her 60s cannot be identified in a block of just a dozen households where only one responded with that profile.
The system for protecting that individual has involved moving a bit of data (part of her answer) to another block and swapping in some other household’s data. By this system, the data at the block level is just a bit fuzzy—but the totals are accurate.
Turns out, it is theoretically possible to use the many data tables provided by the census in conjunction with myriad other data sources to cross-tabulate and find the census answers provided by some people. To protect data, the U.S. Census Bureau is working on a better approach called “differential privacy.” The new concept includes injecting noise, in a controlled manner, at a variety of levels of geography, not just at the smallest unit.
Is this good? I am not the least bit qualified to opine on this question. It is a question of privacy vs. data accuracy, and anyone who is qualified will have to explain how they decide where to fall between absolute privacy and absolute accuracy.
I can say that at a pre-conference to NCSL’s Get Ready to Redistrict seminar last week in Providence, R.I. (the first of five), bureau staff faced questions from two dozen redistricting mavens. One high-level question was whether total population for each state would be accurate. We were assured that yes, total population at the state level will be precise.
The next big question didn’t have quite such a clear response: Will census data below the state level that is intentionally different from the actual count be adequate from a legal standpoint, and particularly in relation to compliance with the Voting Rights Act? The consensus is that judges will have to make that determination.
See the presentation by Michael Hawes, of the Census Bureau, to understand the intricacies of this question more fully.
Wendy Underhill is the director of NCSL's Elections and Redistricting Program.