This week, the U.S. Census Bureau released a bulletin that sounds a lot like it is a response to a report released by MALDEF and other civil rights groups earlier in the week. That report warns that the Bureau’s application of differential privacy (DP) statistical techniques that alter census redistricting data to protect privacy, is likely to frustrate implementation of the Voting Rights Act during the redistricting process, ultimately disenfranchising minority voters.
In particular, the MALDEF report claims that recent samples of 2010 census data treated with various degrees of differential privacy (DP) techniques will obfuscate the true population count of minority populations in many areas. Ultimately, they suggest that this will likely result in an increase in “false positive” majority-minority districts.
On Tuesday, the Census Bureau released a bulletin titled “Meeting Redistricting Data Requirements: Accuracy Targets,” which gave some detail about how the Bureau has approached the difficult task of balancing data accuracy with data privacy. To be clear, the Bureau has released over the past 18 months or so, several demonstration files for data users to assess. The files allowed users to compare 2010 census data treated with DP to untreated 2010 census data. (2010 census data was not treated with DP when it was officially released in 2011).
Since then, there have been various reports by states, demographers and other data professionals that the demonstration products produced inaccurate and in some cases “unusable” data for redistricting. Each succeeding demonstration product released by the Bureau was aimed at weakening the effect of DP on the data in the hope to make it usable for redistricting and other purposes. However, the most recent demonstration file continues to receive criticism from key users in the community. These concerns raise the question of whether the Bureau has had enough time to consider the ramifications of DP. Here are some key takeaways from the Bureau’s bulletin:
- The Bureau has recently been turning its attention to “optimizing and tuning the parameters of the algorithm to further improve accuracy . . . [and] is still evaluating the empirical results of these experimental runs.
- As part of its quality checks, it has created an accuracy target to “ensure that the largest racial or ethnic group in any geography entity with a total population of at least 500 people is accurate to within 5 percentage points of their enumerated value at least 95% of the time.”
- There will be less than 4 weeks between the time that the Bureau’s Data Stewardship Executive Policy Committee makes a final determination of the precise parameters for DP, and the final production run of 2020 census redistricting data.
For more information on the Bureau’s Disclosure Avoidance strategy and demonstration data and Metrics, click here.
