Harvard Law School’s Election Law Clinic filed this amicus brief today, on behalf of Jowei Chen and me, in Alexander v. South Carolina State Conference of the NAACP. The brief explores the uses of redistricting algorithms in racial gerrymandering litigation. Here are some excerpts from the introduction:

[T]he brief explains why computational redistricting can be probative in the racial-gerrymandering context. The basic logic is that racial-gerrymandering claims focus on the intent of mapmakers, and computational redistricting can be a helpful way to produce evidence of mapmakers’ intent. Consider a district attacked as a racial gerrymander and defended on the basis that one or more of nonracial criteria A, B, and C predominantly account for the district’s creation. A computer algorithm can be instructed to incorporate criteria A, B, and C—but to ignore racial data—and to churn out large numbers of districts in the vicinity of the disputed district. If these computer-generated districts significantly differ demographically from the disputed district, that’s supportive evidence for the inference that race predominantly drove that district’s formation. Had race not been the primary factor, that district would likely have had a different demographic makeup, one in the range of the computer-generated districts.

Critically, the emphasis on intent in racial-gerrymandering claims distinguishes this context from other areas where this Court has been skeptical of computational redistricting. In Rucho, ensembles of computer-generated maps were offered as the benchmark for determining partisan effect—a “baseline from which to measure how extreme a partisan gerrymander is.” 139 S. Ct. at 2505. Likewise, in Allen, Alabama argued that “millions of possible districting maps for a given State” should constitute the “race-neutral benchmark” relative to which the effect of racial vote dilution should be assessed. 143 S. Ct. at 1506. The Court properly rejected Alabama’s claim on several grounds, one of which was that racial vote dilution “turns on the presence of discriminatory effects, not discriminatory intent.” Id. at 1507. Unlike racial vote dilution, though, racial gerrymandering turns on the presence of racial intent, not racial effects. Computational redistricting can therefore be probative here for precisely the reason it was inapt in Allen—its ability to shed light on mapmakers’ motives.

It’s true, as the Court pointed out in Allen, that it’s generally infeasible for computer algorithms to enumerate every lawful map for a jurisdiction. See id. at 1514 (“What would the next million maps show?”). But mathematical proofs show that modern algorithms can produce—and mounting empirical evidence demonstrates that they often do produce— representative map ensembles with the same statistical properties as the entire map universe. See, e.g., Benjamin Fifield et al., The Essential Role of Empirical Validation in Legislative Redistricting Simulation, 7 Stat. & Pub. Pol’y 52 (2020) [Fifield et al., Essential Role]. The Court was also correct in Allen that the inclusion of different criteria in algorithms can “yield different benchmark results.” 143 S. Ct. at 1513. But in racial-gerrymandering cases, experts rely on the criteria specified by jurisdictions, not whichever parameters they happen to prefer. Experts also should and do conduct robustness checks to investigate if their conclusions hold when they vary the instructions for their algorithms.