Zack Stanton for Politico:

For the past five years, Duchin has led Tufts’ Metric Geometry and Gerrymandering Group, a lab that has quietly upended conventional wisdom about how gerrymandering works by approaching the issue less as a political problem than a mathematical one. As the country sprints into a new redistricting cycle, understanding redistricting in those terms has taken on new importance—especially in light of a controversial change to the Census Bureau data that will be used to draw the new district maps.

This year, for the first time, the Census Bureau has added random noise to its data that makes it slightly inaccurate at the smallest, most zoomed-in level, but accurate at an aggregate, wide-angle view. The approach, known as “differential privacy,” aims to protect the anonymity of census respondents amid a glut of third-party online data that could otherwise make it possible to personally identify census respondents. The move has prompted a wave of criticism that redistricting based on those “noisy” numbers will be inaccurate.

Duchin, who has studied the Census’ use of differential privacy for the past year, has come to a different conclusion: that, in terms of drawing districts and enforcing Voting Rights Act provisions, the effect of the noise is negligible. But, in something of a surprise, Duchin also found that this noise might actually make it more difficult to do extreme gerrymandering in the new districts—which could actually complicate partisans’ designs for the 2022 congressional maps.

“If you build your district starting with the tiniest particles—in other words, if you do the practices that are associated with gerrymandering and make microdetailed plans—[differential privacy] is going to mess up your numbers more than if you start with larger units and only use the little units to tune at the end,” said Duchin.