Electoral Innovation Lab Comment on Kenny et al. Census DAS Evaluation

The plaintiffs in Alabama v. Department of Commerce (case no. 3:21-cv-00211-RAH-ECM-KCN) have filed a supplemental statement to their expert report in which they attach a working paper by Christopher Kenny and other students, working in collaboration with Kosuke Imai of Harvard University. Prof. Imai is a recognized expert in automated methods for drawing district maps. Kenny et al. report results of applying ensemble simulation methods to the Census Bureau’s DAS 12.2-demonstration data set, in which noise was added to 2010 Census data. For comparison they do calculations using the Census 2010 data release, in which privacy protection was accomplished using swapping, an older method of disclosure avoidance. Kenny et al. report differences between their simulations under the two conditions, and conclude that these differences arise from bias. They assert that these biases are large enough to make it difficult to comply with redistricting requirements. They conclude that such issues can be avoided by reverting to the swapping method or by suppressing some block-level Census tables.

That working paper has not been through peer review. We therefore performed our own examination of the manuscript. Our group at Princeton University, the Electoral Innovation Lab, is expert in analysis of election and redistricting data. One of our projects, the Princeton Gerrymandering Project, does ensemble analysis in its own work, and we are published in this area. We are therefore qualified to comment on the work of Kenny et al.

See ALARM's response to concerns here: https://alarm-redist.github.io/posts/2021-06-02-das-evaluation-faq/