The Rise of the Hal-mander: Is Gerrymandering by Algorithm the Next Frontier of Partisan Gerrymandering?

• Author: Harry A. Levin
• Position: Georgetown University Law Center, J.D. 2022; Haverford College, B.S. 2014
• Pages: 891-922

The Rise of the Hal-mander: Is Gerrymandering by

Algorithm the Next Frontier of Partisan

Gerrymandering?

HARRY A. LEVIN*

TABLE OF CONTENTS

INTRODUCTION. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 892

I. BACKGROUND . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 895

A. BACKGROUND ON REDISTRICTING . . . . . . . . . . . . . . . . . . . . . . . . . . . . 895

B. REDISTRICTING REQUIREMENTS AND CRITERIA . . . . . . . . . . . . . . . . . . 898

1. Population Equality . . . . . . . . . . . . . . . . . . . . . . . . . . . . 898

2. Race-Based Redistricting . . . . . . . . . . . . . . . . . . . . . . . . 899

3. Compactness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 900

4. Preservation of Communities of Interest . . . . . . . . . . . . . 901

5. Preservation of Political Boundaries . . . . . . . . . . . . . . . . 901

6. Competitiveness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 902

7. Proportionality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 902

8. Measuring Redistricting Criteria . . . . . . . . . . . . . . . . . . . 902

C. REDISTRICTING TECHNOLOGY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 903

1. Computational Limits of Redistricting Technology . . . . . 904

2. Types of Plan Generators . . . . . . . . . . . . . . . . . . . . . . . . 905

3. Types of Plan Evaluators. . . . . . . . . . . . . . . . . . . . . . . . . 907

II. EXPECTATIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 907

III. LEGAL LIMITS OF REDISTRICTING BY ALGORITHM . . . . . . . . . . . . . . . . . . . 909

A. REDISTRICTING BY ALGORITHM AND NONDELEGATION . . . . . . . . . . . 910

B. REDISTRICTING BY ALGORITHM AND THE ELECTIONS CLAUSE . . . . . . 912

* Georgetown University Law Center, J.D. 2022; Haverford College, B.S. 2014. © 2023, Harry A.

Levin. I thank Professor Matt Blaze and Professor Sorelle Friedler. I would also like to thank the editors

of The Georgetown Law Journal for their thoughtful feedback throughout the review and editing

process.

891

C. PARTISAN GERRYMANDERING BY ALGORITHM: HAL-MANDERING. . . . 913

1. Minimal Federal Limits to Hal-mandering. . . . . . . . . . . . 914

2. Some States Limit Partisan Gerrymandering but Not

Hal-mandering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 914

IV. TECHNOLOGY MAKES MORE MESS: RISKS OF AUTOMATION . . . . . . . . . . . 917

A. NEGATIVE IMPACTS OF AUTOMATION . . . . . . . . . . . . . . . . . . . . . . . . . 917

1. Automation Is Still Political . . . . . . . . . . . . . . . . . . . . . . 917

2. Automation Removes the Democratic Ideal. . . . . . . . . . . 919

3. Automation Faces Computational Limits . . . . . . . . . . . . 920

B. AUTOMATION LOOSENS OTHER REDISTRICTING LIMITS . . . . . . . . . . . . 920

CONCLUSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 922

INTRODUCTION

We cannot automate partisan gerrymandering out of existence, but we might

perpetuate it through automation. Every ten years, around the United States, state

legislatures and independent commissions draw new districts for congressional

and state legislative elections, and many of these mapmakers rely on complex

software and computer algorithms.

1

Sam Levine, ‘From Dark Art to Dark Science’: The Evolution of Digital Gerrymandering,

GUARDIAN (Aug. 22, 2021, 4:00 AM), https://www.theguardian.com/us-news/2021/aug/22/gerry

mandering-us-electoral-districts-congress [https://perma.cc/37FL-VDVN].

A redistricting consultant can use an algo-

rithm to generate thousands of gerrymandered maps that comply with state and

federal redistricting requirements.

2

Reformers also craft redistricting proposals

through algorithms.

3

Like the consultant who generates thousands of gerryman-

dered maps, a reformer can generate thousands of nongerrymandered

4

maps that

comply with state and federal redistricting requirements.

5

Whether to gerryman-

der more efficiently under the cover of a facially neutral algorithm or to prevent

gerrymandering by handing control to a facially neutral algorithm, states might

consider an automated redistricting process.

One example can be found in North Carolina. North Carolina decided its state

legislative districts in 2019 with a nearly automated redistricting process.

6

See Common Cause v. Lewis, No. 18 CVS 014001, 2019 WL 13198027, at *3 (N.C. Super. Ct.

Oct. 28, 2019); Miles Parks, A Surprise Vote, Thrown Phone and Partisan ‘Mistrust’ Roil N.C. as Maps

After a

state court ruled that North Carolina’s state senate and house districts were

1.

2. Id.

3. See infra Section I.C.2.

4. It is important to note that these maps are not necessarily objective because algorithms require

choices, which have political consequences. See infra Section IV.A.1.

5. See infra Section I.C.2.

6.

892 THE GEORGETOWN LAW JOURNAL [Vol. 111:891

Are Redrawn, NPR (Sept. 16, 2019, 5:19 AM), https://www.npr.org/2019/09/16/760177030/a-surprise-

vote-thrown-phone-and-partisan-mistrust-roil-n-c-as-maps-are-redrawn [https://perma.cc/7WW9-AZNG].

unconstitutional because of partisan bias, the Republican majority had to redraw

the districts.

7

The first step of the process was entirely automated. University of

Michigan Professor Jowei Chen, who testified as an expert witness against the

State in the litigation, produced a large set of redistricting plans for North

Carolina’s state senate and house.

8

The mapmakers ranked all of the computer-

generated plans based on traditional redistricting criteria and selected the top five

plans.

9

The legislators used a state lottery machine to randomly select one of the

five redistricting plans to use as a base plan.

10

Parks, supra note 6; Miles Parks (@MilesParks), TWITTER (Sept. 11, 2019, 8:28 PM), https://

twitter.com/MilesParks/status/1171943607441530880 [https://perma.cc/8K82-NQJ5].

This new plan was enacted after a

few minor modifications, such as unpairing incumbents who were placed in the

same district.

11

On its face, this process in North Carolina is what many reformers have called

for. The legislature relied predominantly on a redistricting algorithm and facially

neutral instructions to select its redistricting plans. However, the enacted state

house districts had fewer Democratic-leaning seats than 95% of a sample of simu-

lated plans.

12

In reviewing the new plan, the court admitted that mapmakers may

not have selected the optimal plan, but it was reasonable to rely on an automated

process despite evidence that the resulting districts were motivated by partisan-

ship.

13

Thus, North Carolina’s nearly automated redistricting process was affirmed

by the court.

A second example can be found in Mexico. Mexico has relied on a nearly auto-

mated redistricting process for decades. Like the United States, Mexico’s lower

legislative body has single-member districts that are apportioned to its states.

14

The National Electoral Institute of Mexico, an independent administrative body,

is responsible for drawing these districts.

15

Alejandro Trelles, Micah Altman, Eric Magar & Michael McDonald, No Accountability Without

Transparency and Consistency: Redistricting-by-Formula in Mexico 2 & n.3 (Jan. 8, 2021) (unpublished

manuscript) (available at https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3762805 [https://perma.

cc/4WHG-FDVK]).

Since 1996, the Institute has relied on

a redistricting algorithm that weighs traditional redistricting criteria to generate

7. Parks, supra note 6; Common Cause v. Lewis, No. 18 CVS 014001, 2019 WL 4569584, at *135–

37 (N.C. Super. Ct. Sept. 3, 2019).

8. Parks, supra note 6.

9. Id.; Lewis, 2019 WL 13198027, at *3.

10.

11. Lewis, 2019 WL 13198027, at *7.

12. Plaintiffs’ Response to Legislative Defendants’ Reply on Remedial Plans at 1–2, Lewis, No. 18

CVS 014001 (N.C. Super. Ct. Oct. 7, 2019). More specifically, Dr. Chen produced two sets for his

analysis: Set 1, which followed traditional districting principles, and Set 2, which avoided pairing

incumbents. Lewis, 2019 WL 4569584, at *18, *22. The remedial plans had fewer Democratic-leaning

seats than 94.6% of Set 1 and 97.8% of Set 2. Plaintiffs’ Response to Legislative Defendants’ Reply on

Remedial Plans, supra.

13. See Lewis, 2019 WL 13198027, at *3.

14. Alejandro Trelles, Micah Altman, Eric Magar & Michael P. McDonald, Open Data,

Transparency and Redistricting in Mexico, 23 POLI

´TICA Y GOBIERNO 331, 334–35 (2016).

15.

2023] THE RISE OF THE HAL-MANDER 893