Amicus Curiae Brief — Gill v. Whitford, 138 S. Ct. 52 (2017) (No. 16-1161)

No. 16-1161

IN THE

Supreme Court of the United States

BEVERLY R. GILL, et all,

Appellants,

Vv.

WILLIAM WHITFORD, et al.,

Appellees.

On Appeal from the United States District Court

for the Western "istrict of Wisconsin

BRIEF OF ERIC MCGHEE AS AMICUS CURIAE

IN SUPPORT OF NEITHER PARTY

CHRISTOPHER S. ELMENDORF DANIEL F. KOLB

UNIVERSITY OF CALIFORNIA, Counsel of Record

DAVIS SCHOOL OF LAW LYNN E. BUSATH

400 Mrak Hall Drive JOHN HAMMOND HEATH

Davis, CA 95616 ELIZABETH A. TIPPETT

(530) 762-5756 DAVIS POLK & WARDWELL LLP

cselmendorf@ucdavis.edu 450 Lexington Avenue

New York, NY 10017

(212) 450-4000

Daniel.Kolb@DavisPolk.com

August 10, 2017

WALSON-EPES PRINTING CO., INC. — (202) 789-0086 — WASHINGTON, D. C. 20002

TABLE OF CONTENTS

TABLE OF AUTHORITIES ...................:0..cc0cee000

INTEREST OF AMICUS CURIAE.................2.0..-

INTRODUCTION AND SUMMARY OF

I. THE EFFICIENCY GAP IS A

SIMPLE, HISTORICALLY GROUNDED

METRIC OF PARTISAN ADVANTAGE,

RESPONSIVE TO CONCERNS THAT

JUSTICES OF THIS COURT HAVE

A. The Efficiency Gap Measures the

Votes Wasted By Packing and

Cracking Opposing-Party Voters

Among Legislative Districts................

B. The Efficiency Gap’s Implementation

of the Concept of “Partisan

Symmetry” Is Historically Grounded

and Does Not Depend on Implausible

ea

C. The Efficiency Gap Is Calculated

Using Familiar Concepts, Methods,

and Data Sources, and Is Not Easily

I hi incnsnediistendendibdistiinigeninindies

D. Because the EG Provides a Snapshot

Picture of Relative Wasted Votes,

Assessments of Partisan Fairness

Benefit From Further Analysis ..........

(1)

12

i

TABLE OF CONTENTS—Continued

Il. THE OBJECTIONS TO THE EFFI-

CIENCY GAP AS A MEASURE OF

PARTISAN ADVANTAGE ARE EASILY

| _ e S E 18

A. The Efficiency Gap Does Not Rest on

a Mistaken Conception of Wasted

Weiaiiiiis:ictihctietensimichencnitinieinienintianiaenpuitisicaalinss 18

B. The Putative Volatility of the Effi-

ciency Gap Is a Feature, Not a Flaw ... 22

C. The Efficiency Gap Does Not Account

for All Democratic Values—Nor

ERAS ESE I Ae eee 23

D. The Efficiency Gap Is Not Biased

Toward Democrats ...................ss-:s0000+ 25

E. The Efficiency Gap Does Not Require

Proportional Representation............... 27

Il. THE EFFICIENCY GAP IS CLOSELY

RELATED TO THE OTHER MEA-

SURES OF PARTISAN SYMMETRY,

A. Arithmetically, the Symmetry Metrics

Are Closely Related to One Another... 30

B. Empirically, Little Turns on the

Choice Among Symmetry Metrics in

Competitive States Such = as

I iiinenincienibinbedinlediitaiieaaapmieeine 32

IEEE intictinnenssopesccvscceneentnnsupniontiiesavagiecnse 34

ili

TABLE OF AUTHORITIES

CASES Page(s)

Baker v. Carr,

i sespcnmtiebanncenedn 24

Bartlett v. Strickland,

TT ee ne 22

City of Mobile, Ala. v. Bolden,

ST nincsicnenrnennnansopeduannenesnces 28

Davis v. Bandemer,

A 6, 19, 28

Frank v. Walker,

768 F.3d 744 (7th Cir. 2014)..........00.000.... 17

Johnson v. De Grandy,

ee 17

League of United Latin American Citizens

v. Perry (LULAC),

Re ee SR ccccveccccecsseccssssccsvcessnens passim

League of Women Voters of North Carolina

v. Rucho,

No. 1:16-CV-1026, 2017 WL 876307

ALE ER 12

Reynolds v. Sims,

BT ncrnicacnachcascodeminteniiveubees 28

Vieth v. Jubelirer,

Se ee Ne HE Arcnccacenenccncvssennscsscneeseoes passim

Whitcomb v. Chavis,

| en 28

Whitford v. Gill,

218 F.Supp.3d 837 (2016)............-...0..000: passim

iv

TABLE OF AUTHORITIES—Continued

COURT FILINGS Page(s)

Brief Amicus Curiae of the Republican

National Committee in Support of

Appellees, Davis v. Bandemer, 478 U.S.

109 (1986) (No. 84-1244), 1985 WL

Brief for Appellants, Whitford v. Gill, 218

25

pl passim

Brief of Amici Curiae Professors Gary King

et al., in Support of Neither Party,

LULAC, 548 U.S. 399 (2006) (No 05-204),

EE ARSE ie

Brief of Amici Curiae The Republican

National Committee and the National

Republican Congressional Committee in

Support of Appellants, Whitford v. Gill,

218 F.Supp.3d 837 (No. 16-1161), 2017

Weep EI ccsvssncececevccrcccsconcnscosnets 21-22, 25, 26

Brief of Amici Curiae Wisconsin State

Senate and Wisconsin State Assembly in

Support of Appellants, Whitford v. Gill,

218 F.Supp.3d 837 (2016) (No. 16-1161),

Re I recs ccccscescseceserensesseseuseess

Plaintiffs’ Trial Brief, Whitford v. Gill, 218

F Supp.3d 837 (2016) (No. 15-cv-421-bbc)...

OTHER AUTHORITIES

Andrew Gelman & Gary King, A Unified

Method of Evaluating Electoral Systems

and Redistricting Plans, 38 AM. J. POL.

ee I iaiiiicissacaentcekiniatiadephescsieensis

18

33

Vv

TABLE OF AUTHORITIES—Continued

Page(s)

Andrew Gelman & Gary King, Enhancing

Democracy Through Legislative Redistrict-

ing, 88 AM. POL. Sci. REV. 541 (1994) ....... 10, 26

Barry C. Burden, Candidate Positioning in

US Congressional Elections, 34 Brit. J.

is Sy i i ecseveseicocsencocinnensivecyesecs 20

Bernard Grofman & Gary King, The Future

of Partisan Symmetry as a Judicial

Test for Partisan Gerrymandering after

LULAC v. Perry, 6 ELECTION L.J. 2

DANIEL KREISS, PROTOTYPE POLITICS:

TECTINOLOGY-INTENSIVE CAMPAIGNING

AND THE DATA OF DEMOCRACY (2016) ....... 20

David S. Lee et al., Do Voters Affect or Elect

Policies? Evidence from the U.S. House,

119 Q.J. ECON. 807 (2004) ............000cccce0e 20

Devin Caughey et al., Partisan Gerry-

mandering and the Political Process:

Effects on Roll-Call Voting and State

Policies, 16 ELECTION L.J. (forthcoming

2017), available at http://cwarshaw.scri

pts.mit.edu/papers/CTW_efficiency_gap_

170515.pdf (May 15, 2017) .................00008- 20

EITAN HERSH, HACKING THE ELECTORATE:

How CAMPAIGNS PERCEIVE VOTERS

RRC AND SS ob EI A I bes Ean 20

vi

TABLE OF AUTHORITIES—Continued

Page(s)

Eric McGhee, Measuring Efficiency in

Redistricting, 16 ELECTION L.J. (forth-

coming 2017), available at https://papers.

ssrn.com/sol3/papers.cfm?abstract_id=30

UTE Gicebccduisdadninennténidmetsaneineneertsounneaniees passim

Eric McGhee, Measuring Partisan Bias in

Single-Member District Electoral Systems,

39 LEGIsS. STUD. Q. 55 (2014).................... passim

Eric McGhee, Memo to the Indiana

Legislature on Efficiency ‘ap Imputa-

tions, Aug. 5, 2016, availalie at https://

papers.ssrn.com/sol3/papers.cfm?abstrac

EIT ceccatnddentucdecinseneinnantcnnsenentiness 14-15

JOHN STUART MILL, CONSIDERATIONS ON

REPRESENTATIVE GOVERNMENT (Prome-

theus Books 1991) (1861)......................00. 27

Keith T. Poole, Changing Minds? Not in

Congress!, 131 PUB. CHOICE 435 (2007)... 20

Michael D. McDonald & Robin E. Best,

Unfair Partisan Gerrymanders in Politics

and Law: A Diagnostic Applied to Six

Cases, 14 ELECTION L.J. 312 (2015)......... 30

Nicholas O. Goedert, Gerrymandering or

Geography? How Democrats Won the

Popular Vote but Lost the Congress in

2012, RES. & POL., Apr.—June 2014.......... 26

vii

TABLE OF AUTHORITIES—Continued

Page(s)

Nicholas Stephanopoulos, The Causes and

Consequences of Gerrymandering, 59

Wo. & MArY L. REV. (forthcoming 2017),

available at https://papers.ssrn.com/sol3/

papers.cfm?ahstract_id=2990638............. 26

Nicholas Stephanopoulos & Eric McGhee,

Partisan Gerrymandering and _ the

Efficiency Gap, 82 U. Cul. L. REV. 831

(ieee inesecscvedsocceconcosessecescnnsesosenseopepssesosesied passim

Nolan McCarty et al., Does Gerryman-

dering Cause Polarization?, 53 Am. J.

ie Bs GH IP ceccccccscccccccccecocececssecsees 20

Richard G. Niemi & Simon Jackman, Bias

and Responsiveness in State Legislative

District-ing, 16 LeGcis. STUD. Q. 183

EE a SERIES SSIES NA MES SIEBER ae ee 26

RODERICK J. A. LITTLE & DONALD B. RUBIN,

STATISTICAL ANALYSIS WITH MISSING

ee eueneness 14

Samuel S.H. Wang, Three Tests for Practi-

cal Evaluation of Partisan Gerryman-

dering, 68 STAN. L. REV. 1263 (2016)....... 30

Simon Jackman, Assessing the Current

Wisconsin State Legislative Districting

Plan, Whitford v. Gill, No. 15-cv-421-bbe

(W.D. Wis. Nov. 21, 2016)................:scce000 11, 25

Simon Jackman, Rebuttal Report, League

of Women Voters of North Carolina v.

Rucho, No. 1:16-CV-1026 (M.D.N.C.

ane ihtatecndepsotnnctecssuccicapbesecbepsodasesepens aaa passim

TABLE OF AUTHORITIES— Continued

Page(s)

State Legislative Election Returns (1967—

2010), Inter-University Consortium for

Political and Social Research, available at

http://www.icpsr.umich.edu/icpsrweb/ICP

SR/studies/34297, Study No. 34297.......... 11

Stephen Ansolabehere et al., Candidate

Positioning in U.S. House Elections, 45

AM. J. POL. SCI. 136 (2001) .........ccccceseceeees 20

BRIEF OF ERIC MCGHEE AMICUS CURIAE

IN SUPPORT OF NEITHER PARTY

INTEREST OF THE AMICUS CURIAE'

Amicus curiae Eric McGhee is a political scientist

and research fellow at the Public Policy Institute of

California (“PPIC”). PPIC is a nonprofit, nonpartisan

organization dedicated to informing and improving

public policy in California.? McGhee developed the

“Efficiency Gap” measure of partisan advantage on

which the plaintiffs and the court below relied. He

holds a Ph.D. in political science from U.C. Berkeley

and has published numerous papers in top peer-

reviewed journals.

He takes an interest in this case because the plain-

tiffs and the lower court have relied on his work, and

because he wants to ensure that this Court’s decision

is not based on a misunderstanding of the Efficiency

Gap and the other metrics of partisan advantage that

may be used in analyzing a map of legislative districts.

' Pursuant to Rule 37.6, amicus affirms that no counsel for a

party authored this brief in whole or in part and that no person

other than amicus and its counsel made a monetary contribution

to its preparation or submission. The parties’ letters consenting

to the filing of amicus curiae briefs have been filed with the

Clerk’s office.

* McGhee’s affiliation with PPIC is given here as biographical!

information; his findings in this area have not been reviewed or

endorsed by PPIC, and PPIC takes no position on the use of this

research for any purpose. The views expressed in McGhee’s brief

in this matter are his personal views, and not those of PPIC.

2

INTRODUCTION AND

SUMMARY OF ARGUMENT

Amicus files this brief on behalf of neither party to

explain the Efficiency Gap (“EG”) measure of partisan

advantage and respond to questions about it. This brief

does not argue for or against affirmance of the lower

court’s decision, nor does it propose a specific doctrinal

standard or test for liability. Amicus’s aims are purely

explanatory.

Part I below provides an introduction to the EG,

including an explanation of how the EG relates to mea-

sures of partisan symmetry—the relative opportunity

for each party to convert votes into seate—previously

considered by this Court. The EG may be used in all

states, including those that do not have competitive

elections. It is an easy-to-calculate, hard-to-manipulate

measure of partisan advantage that is grounded both

in recognized concepts of political fairness and in our

nation’s actual history of elections since the adoption

of the one person, one vote principle.

Part II responds to objections that have been raised

as to the use of the EG. The objections raised are

easily met.

Part III explains the conceptual and empirical rela-

tionships between the EG and other proposed mea-

sures of partisan symmetry. The main takeaways are

that the EG has distinct advantages over the other

measures and in politically competitive states such as

Wisconsin all of the symmetry metrics lead to similar

conclusions.

Amicus submits that the EG is a very useful meas-

ure of the partisan advantage resulting from a full

map of districts for a legislative body. The lower

court’s reliance on the EG as evidence of partisan

3

advantage was reasonable, and certainly not grounds

for reversal.

ARGUMENT

I. THE EFFICIENCY GAP IS A SIMPLE,

HISTORICALLY GROUNDED METRIC OF

PARTISAN ADVANTAGE, RESPONSIVE

TO CONCERNS THAT JUSTICES OF THIS

COURT HAVE RAISED.

The Efficiency Gap is a reliable measure of the effect

of a partisan gerrymander for three principal reasons.

First, the EG is predicated on the widespread under-

standing that the whole point of a partisan gerryman-

der is to distribute opposing-party supporters ineffi-

ciently so their votes are not converted into seats while

distributing one’s own supporters efficiently so their

votes are converted into seats.

Second, the EG is a practical, historically grounded

implementation of partisan symmetry—the relative

opportunity for each party to convert votes into state

legislative or congressional seats. An EG of zero math-

ematically indicates symmetry, and, further, that

the relationship between the parties’ vote shares and

seat shares under the map at issue aligns with the

historical norm in state legislative and congressional

elections.

Several Justices of this Court have expressed inter-

est in using symmetry to guide the evaluation of par-

tisan gerrymandering claims, but also concern that

quantification of a map’s asymmetry not depend on

unrealistic electoral scenarios. See League of United

Latin American Citizens v. Perry (LULAC), 548 U.S.

399, 420 (2006) (plurality opinion of Kennedy, J.)

(“The existence or degree of asymmetry may in large

part depend on conjecture about where possible vote-

4

switchers will reside... .({Wle are wary of adopting a

constitutional standard that invalidates a map based

on unfair results that would occur in a hypothetical

state of affairs.”); id. at 465 (Stevens, J., concurring in

part and dissenting in part); id. at 483-84 (Souter, J.,

joined by Ginsburg, J., concurring in part and dissent-

ing in part); id. at 491-92 (Breyer, J., concurring in

part and dissenting in part). Addressing that concern,

the EG reflects actual election results, and its robust-

ness can be tested with electoral fluctuations typical

for the state in question.

Third, calculation of the EG is straightforward.

There is little room for expert witnesses to distort it to

portray the challenged map in a deceptively good or

bad light.

The picture of partisan advantage provided by the

EG is a snapshot of a particular election—of votes

actually cast and seats actually won or lost. Before

invalidating a map, a court may consider evidence of

whether and if so how the EG is likely to vary in future

elections—or has varied in other elections—held

pursuant to the map. Such evidence can be generated

using statistical methods that are standard in the

political science literature and widely employed by

mapmakers who design partisan gerrymanders.

A. The Efficiency Gap Measures the Votes

Wasted By Packing and Cracking

Opposing-Party Voters Among Legis-

lative Districts.

Presented first in a peer-reviewed academic paper,

see Eric McGhee, Measuring Partisan Bias in Single-

Member District Electoral Systems, 39 LEGIS. STUD. Q.

55 (2014) (hereinafter McGhee, Measuring Partisan

Bias), the EG metric of partisan advantage is grounded

5

in a very simple and familiar idea: A redistricter

seeking partisan advantage distributes voters among

districts to increase the number of seats his party will

obtain for a given share of the popular vote. Because

winning more seats is the point of a partisan gerry-

mander, a measure that purports to quantify the

effect of alleged partisan gerrymanders should clas-

sify a map as more advantageous to the redistricting

party (relative to some other map) if the map would

give the redistricting party more seats for the same

share of the vote. Amicus calls this the “Efficiency

Principle.” See McGhee, Measuring Partisan Bias,

supra, at 61; Eric McGhee, Measuring Efficiency in

Redistricting, 16 ELECTION L.J. (forthcoming 2017),

available at https://papers.ssrn.com/sol3/papers.cfm?

abstract_id=3007401 [hereinafter McGhee, Measuring

Efficiency in Redistricting).

In developing the EG, McGhee built on observations

of this Court and many others concerning the two prin-

cipal ways by which redistricters may increase the

number of seats held by their party without an associ-

ated increase in their party's popular vote. One is

to “crack” groups of opposing-party voters, spreading

them among several legislative districts in which they

will be outnumbered by voters of the party drawing

the map. The other is to “pack” voters of the opposing

party into districts that are already likely to be won

by opposing-party candidates, increasing those candi-

dates’ vote shares while effectively preventing the

opposing party from winning more seats.

Whether packed or cracked, such votes are wasted

because they do not contribute to victories. McGhee,

Measuring Partisan Bias, supra, at 56-57; McGhee,

Measuring Efficiency in Redistricting, supra, at 2-4.

See also Vieth v. Jubelirer, 541 U.S. 267, 286 n.7 (2004)

6

(using the terms cracking and packing to characterize

partisan gerrymanders); Davis v. Bandemer, 478 U.S.

109, 116—17 n.6 (1986) (stacking and cracking). A suc-

cessful gerrymander will saddle the opposing party

with more wasted votes than the mapmakers’ party,

thereby garnering more legislative seats for the party

in control.

The EG quantifies the relative difference in the

parties’ wasted votes. It sums across all the districts

in the map (1) the number of votes cast for each party’s

losing candidates, and (2) the number of votes cast for

each party’s winning candidates in excess of the 50%

that assures victory.* The first sum captures gerry-

mandering by cracking, the second, gerrymandering

by packing.

Next, a party’s wasted-vote total is subtracted from

the other party’s wasted-vote total, and the difference

is divided by the total number of votes in the election.

The result—the EG—.is the difference in wasted votes

as a percentage of the total vote. Restated in math

notation:

EG =

>¥(Party X wasted votes) - }}(Party Y wasted votes)

>) (Party X votes + Party Y votes)

To illustrate, if there are 100,000 voters in a district

and 55,000 vote for the Democratic candidate while

* Arguably, votes for the winner are not wasted except insofar

as the winner’s vote share exceeds 50% + 1, rather than 50%. This

minor modification of the formula is inconsequential for real-

world maps of legislative districts, with tens of thousands of

voters per district. Either version of the formula satisfies the

Efficiency Principle. See generally McGhee, Measuring Efficiency

in Redistricting, supra.

7

45,000 back the Republican, the outcome of the race

contributes 45,000 wasted votes to the Republican

sum, 5,000 wasted votes (55,000-50,000) to the Demo-

cratic sum, and 100,000 votes to the denominator.

The EG can be calculated from the perspective of

either party, i.e., with “Democrats” or “Republicans” as

Party X in the above expression. When lawyers and

commentators talk about whether the EG is “large” or

“small,” they are referring to the absolute value of the

EG. Because such discussions about size are always

about the absolute value of the EG, it does not matter

which party’s wasted vote total is used as the first

term or the second term in the numerator of the EG.

As a sum across districts, the EG quantifies the total

or aggregate partisan advantage conferred by a map

of legislative districts, rather than how much a par-

ticular district benefits either party. To crack and to

pack is to shift voters among districts; accordingly, any

measure of the effect of a partisan gerrymander

must account for the distribution of voters among

districts, not simply the characteristics of a single

district viewed in isolation. In layman’s terms, the EG

is simply the net impact of packing and cracking

across all the districts in a map.

B. The Efficiency Gap’s Implementation of

the Concept of “Partisan Symmetry” Is

Historically Grounded and Does Not

Depend on Implausible Hypotheticals.

There is an important relationship between the EG

and the concept of partisan symmetry. Long used by

political scientists to define fairness in two-party sys-

tems, the concept of partisan symmetry was addressed

by the Justices of this Court in LULAC. The EG

employs the idea of partisan symmetry in a tractable,

8

historically grounded manner that responde specifi-

cally to Justice Kennedy’s critique in LULAC of the

then-available symmetry metric.

A map of legislative districts is considered symmet-

ric, as political scientists use the term, if it is expected

to reward each party with the same share of seats for

a given share of the two-party popular vote. Sym-

metry so defined is indifferent to what political scien-

tists call responsiveness, that is, to the rate at which

seats switch hands as the parties’ vote shares shift.

Any level of responsiveness is compatible with sym-

metry, as symmetry merely means (1) that each party

enjoys the same seat share for a given vote margin

above 50%, and (2) that each party obtains 50% of the

seats if both parties get 50% of the vote.

To identify the expected or usual! relationship between

votes and seats under a particular map of legislative

districts, analysts will often shift (hypothetically) a

party’s vote share in every seat up or down by the

same amount and record the party’s new seat share

given this shift. If this exercise is repeated for all poss-

ible vote shares, the hypothetical vote shares and seat

shares can be plotted against each other in a “votes-to-

seats” curve. (See Part III, infra, for illustrations.)

The votes-to-seats curve is a useful tool for under-

standing how a redistricting plan is likely to drive

results under circumstances that have not occurred.

However, while this votes-to-seats curve is useful for

thinking about symmetry in the abstract, no political

party will ever experience the full range of vote shares

from 0 to 100 percent, in a given state and over the

lifetime of a given map. The votes-to-seats curve is

empirically grounded only for vote shares within the

9

recent range of variation in actual vote shares in the

state.

Recognizing as much, analysts do not usually quan-

tify the curve’s asymmetry over the full span from a

vote-share of 0% to a vote share of 100%. Instead, they

focus on the most normatively significant point—

a vote share of 50%—and estimate the seat shares

that each party would likely receive when they earn

the same share of the statewide vote. The difference

between either party’s seat share and 50% at this

point is called “partisan bias” or “Gelman-King bias.”

See, e.g., Andrew Gelman & Gary King, A Unified

Method of Evaluating Electoral Systems and Redis-

tricting Plans, 38 AM. J. POL. Sci. 614, 536 (1994).

Gelman-King bias was briefed in LULAC. See Brief of

Amici Curiae Professors Gary King et al., in Support

of Neither Party 14—15, LULAC, 548 U.S. 399 (No 05-

204), 2006 WL 53994 (suggesting legal standard based

on estimated seat-share difference at 50% vote share).

Yet the Gelman-King measure may still depend

on hypothetical electoral scenarios with little basis in

real-world voting patterns. For example, in a state

where Republicans regularly win about 65% of the

two-party vote in state legislative elections, Gelman-

King bias requires a prediction about the number of

seats that Democrats would carry if the parties’ vote

shares were significantly different from the actual

balance of party voter percentages—e.g., if each party

won 50% of the vote. Such dramatic, short-term parti-

san reversals are fanciful. They exist only in imagined

realms of unheard-of calamity or radically overhauled

party platforms and leadership. It’s anyone’s guess

which voters would switch sides in these scenarios.

Yet without predictions about the geographic distribu-

tion of vote-switchers, there’s no way to predict how

10

many seats would be won by each party in the coun-

terfactual election. See LULAC, 548 U.S. at 419-20

(Kennedy, J. articulating such concerns).

Acknowledging the force of these objections, propo-

nents of the Gelman-King measure have confirmed

that it should only be used in competitive states, i.e.,

where each major party usually receives close to 50%

of the two-party vote. See Bernard Grofman & Gary

King, The Future of Partisan Symmetry as a Judicial

Test for Partisan Gerrymandering after LULAC v.

Perry, 6 ELECTION L.J. 2, 31 (2007); Andrew Gelman

& Gary King, Enhancing Democracy Through Legisla-

tive Redistricting, 88 AM. POL. SCI. REV. 541, 545 (1994).

But this limitation puts courts in the awkward posi-

tion of making case-by-case determinations of whether

a state is sufficiently competitive for the metric of par-

tisan advantage to be used.

The EG offers a ready solution to this problem.

It is a version of symmetry that uses actual election

outcomes—the votes actually cast and the seats actu-

ally won in any state. It does not depend on fanciful

hypotheticals at all.‘ Simple algebra establishes that

an EG of zero will be observed if and only if the parties’

vote shares and seat shares in an election correspond

to a point on the line which passes through the 50%

vote, 50% seats point and has a slope of 2. See McGhee,

Measuring Partisan Bias, supra, at 79-80. This line

is symmetrical with 2:1 responsiveness, meaning that

for every 1% increase in a party’s vote share above

* Analysts do consider hypotheticals when investigating the

durability of the observed EG, and in districts where seats are

uncontested. But these inquiries are grounded in the observed

range of variation in the state, in contrast to the counterfactuals

sometimes required to calculate Gelman-King bias. See infra Part

LD.

11

50%, that party secures an additional 2% of the legisla-

tive seats. Thus, an EG of zero means that the election

results accord with partisan symmetry—without requir-

ing estimation of a hypothetical votes-to-seats curve.

Two-to-one responsiveness is not a strange anomaly.

The existence of a “winner’s bonus” is a well-known

feature of single-member district, plurality-winner elec-

toral systems. See, e.g., Grofman & King, supra, at 9;

Vieth, 541 U.S. at 357—58 (Breyer, J., dissenting). The

2:1 relationship reflected in an EG of zero is a particu-

larly helpful guidepost because it corresponds to the

actual historical relationship between votes and seats

in American elections in the one person, one vote era.

To assess this relationship, plaintiffs’ expert Simon

Jackman, then a tenured professor at Stanford

University,° analyzed the canonical dataset® of post-

malapportionment redistricting maps, compiled by

experts and covering nearly all states from 1972 to the

present. See Simon Jackman, Assessing the Current

Wisconsin State Legislative Districting Plan, Whitford

v. Gill, No. 15-cv-421-bbe (W.D. Wis. Nov. 21, 2016)

{hereinafter Jackman Wisconsin Report] at 33 fig.

11. He calculated vote and seat shares for every elec-

tion held under each map, and the results cluster

neatly around the votes-to-seats curve implied by an

EG of zero. There are some outliers, but for the most

part, the historical practice of districting in the United

States has produced roughly symmetric maps with

° Jackman is now CEO of the United States Studies Centre at

the University of Sydney, in his home country of Australia.

® State Legislative Election Returns (1967-2010), Inter-

University Consortium for Political and Social Research, avail-

able at http//www.icpsr.umich.edu/icpsrweb/ICPSR/studies/34297,

Study No. 34297.

12

2:1 responsiveness. This finding is uncontested. See

Whitford v. Gill, 218 F.Supp.3d 837, 904 (2016).

In Vieth, 541 U.S. 267, Justice Kennedy posited that

litigants might in a future partisan gerrymandering

case uncover “historical guidance” that would shed light

on appropriate standards. Id. at 308-09 (Kennedy,

J., concurring in judgment). He also noted that “new

technologies may produce new methods of analysis

that make more evident the precise nature of the bur-

dens gerrymanders impose on the representational

rights of voters and parties.” Id. at 312-13. The record

in this case bears out Justice Kennedy's hope. The

plaintiffs relied on a new technology (the EG metric),

and they developed a record that is based on the actual

history of United States elections. This history shows

that the EG is grounded in a well-rooted (though previ-

ously unarticulated) tradition concerning the relation-

ship between a major party’s support in the electorate

and its share of seats in the legislature.

C. The Efficiency Gap Is Calculated Using

Familiar Concepts, Methods, and Data

Sources, and Is Not Easily Manipulated.

The reliability of the EG as a measure of partisan

advantage is further indicated by the lack of disagree-

ment among expert witnesses as to its calculation. In

the present case, defendants’ experts did not challenge

the EG calculations at all. In the North Carolina case,

League of Women Voters of North Carolina v. Rucho,

No. 1:16-CV-1026, 2017 WL 876307 (M.D.N.C. 2017),

defendants’ expert was able to show EG results quite

different than the plaintiffs’—but only by erroneously

defining the EG as the deviation from proportional

representation. See Simon Jackman, Rebuttal Report,

League of Women Voters of North Carolina v. Rucho,

13

No. 1:16-CV-1026 (M.D.N.C. 2017) at 9-10 [herein-

after Jackman, Rucho, Rebuttal Report] (showing con-

vergence of defense-side and plaintiff-side calculations

upon correction of this definitional error).

It is true that calculation of the EG is not entirely

mechanical, but expert witnesses have very limited

discretion. Discretion may enter at two points. First,

for ease of computation and because of data limita-

tions, experts sometimes use a simplified approxima-

tion of the full EG formula; the simplified version is

only exactly correct if the number of votes is constant

across districts and no one votes for third-party or

independent candidates.’ But because the number

of voters does not vary greatly across districts, and

because minor party and independent candidates rarely

garner many votes, the simplified EG is extremely

highly correlated with the “full” version. See McGhee,

Measuring Efficiency in Redistricting, supra, at 40—

43.° As a defense expert stated in this case, the simpli-

fied method provides “an appropriate and useful sum-

mary.” Whitford, 218 F.Supp.3d at 907.

' The simplified formula is: EG = (Party X Seat Margin) — 2 *

(Party X Vote Margin). The seat margin is the party's percentage

of legislative seats, minus 50%; the vote margin is tho party's

percentage of the total statewide vote in legislative elections,

minus 50%. See McGhee, Measuring Partisan Bias, supra.

®* Amicus has developed a slightly revised version of the for.

mula that is more appropriate when turnout deviations are

significant. In this version the baseline 2-to-1 seats-to-votes ratio

is maintained and the substantive implications of the measure

are very similar. As noted, for American elections generally and

the Wisconsin plan in particular, all versions of the EG are highly

correlated with each other. See McGhee, Measuring Efficiency in

Redistricting, supra.

14

Second, it is best practice to calculate the EG using

“imputations” for the two-party vote in uncontested

seats, and different experts may use slightly different

imputation methods. See generally Nicholas Stepha-

nopoulos & Eric McGhee, Partisan Gerrymandering

and the Efficiency Gap, 82 U. CHI. L. REV. 831, 850—

53 (2015); McGhee, Measuring Partisan Bias, supra.°

The imputed two-party vote for a district represents

the expected vote if the seat had been contested. In

theory, one could compute the EG using actual vote

totals in districts where the incumbent went unchal-

lenged. But that would be tantamount to saying that

supporters of the out-of-power party who have been

“cracked” into districts held by strong, unchallenged

incumbents of the gerrymandering party were not in

fact rendered impotent—even though a protest, write-

in vote would have been pointless, and even though

their vote would have been tabulated as wasted if a

challenger had run and lost. Imputation allows the EG

to account for votes that probably would have been

wasted had a challenger run.

The imputed vote for a legislative district is based

on the actually observed relationship in a state

between the district-level vote shares of partisan can-

didates in statewide elections and the vote shares of

legislative candidates running to represent the dis-

trict. The associated methods are standard in the sta-

tistical literature,’° and generally accepted methods

° Imputation is also used when computing other measures of

partisan bias. See, e.g., Andrew Gelman & Cary King, A Unified

Method of Evaluating Electoral Systems and Redistricting Plans,

38 Am. J. POL. Sci. 514, 549-50 (1994).

10 See generally RODERICK J. A. LiTrLE & DONALD B. RUBIN,

STATISTICAL ANALYSIS WITH MISSING DATA (2d ed. 2014); Eric

McGhee, Memo to the Indiana Legislature on Efficiency Gap

15

yield very similar EG estimates. See, e.g., Jackman,

Rucho, Rebuttal Report, supra, at 9-10 (showing

convergence of EG calculations by different experts);

id. at 16-17 (reporting 98% correlation between EGs

calculated using different imputation approaches).

Notably, none of the defense-side experts in this or

the North Carolina case where the EG was offered

questioned the imputations.

D. Because the EG Provides a Snapshot

Picture of Relative Wasted Votes, Assess-

ments of Partisan Fairness Benefit

From Further Analysis.

The EG is calculated using actual outcomes in a

given election. The number of votes deemed wasted

for each party in a district depends on which candidate

actually won that seat, and, save for uncontested seats,

on the actual number of votes cast for each party’s

candidate in that district.'' The EG thus provides an

election-specific “snapshot” picture of relative wasted

votes. This snapshot helps to establish that the “feared

inequality” is not “hypothetical,” see LULAC, 548 U.S.

at 420 (plurality opinion of Kennedy, J.), but before

invalidating a map, courts should determine from

the evidence whether a large observed EG is likely

to persist. See Whitford, 218 F.Supp.3d at 898-910;

Stephanopoulos & McGhee, supra, at 887-90. If a

large EG was simply the result of one party having a

Imputations, Aug. 5, 2016, available at https://papers.ssrn.com/

8013/papers.cfm?abstract_id=3007178.

1! Even when the “simplified” method is used, the EG relies on

actual election results; the simplified method just uses statewide

totals (seat margin, and legislative vote margin) rather than

summing wasted votes on a seat-by-seat basis. See supra Part I.C.

16

good year and sweeping all the competitive districts,

there would be no reason for a court to step in.

To gauge persistence of partisan advantage, expert

witnesses perform sensitivity tests.'* (Gerrymanderers

do too."*) A sensitivity test is a statistical projection

of potential future election results, grounded in the

historically observed range of variation in elections in

the state. Professor Jackman’s sensitivity tests below

indicated, and the district court found, that the pro-

Republican EG in the Wisconsin map was very likely

to persist throughout the decennial period. Whitford,

218 F.Supp.3d at 905-06. This finding is hardly a

surprise, given that sensitivity tests conducted by

the expert whom the Wisconsin redistricters hired to

evaluate potential maps reached the same conclusion.

See id. at 857-58.

Judicial inquiries into partisan fairness might also

consider the expected EG: the EG as the map-maker

sees it, with an eye to the future but without knowing

which election years will be banner years for each

party, which incumbents will die or retire, and which

seats will be contested by strong challengers. The

expected EG is simply the predicted EG under a range

of electoral scenarios, weighted by the likelihood of

those scenarios.'* Calculating the expected EG is just

The district court used the term “swing analysis” to describe

sensitivity tests. See Whitford v. Gill, 218 F.Supp.3d 837, 899—

905 (2016).

‘8 An expert hired by the firm that designed the map at issue

in this case carefully investigated how proposed plans would

likely “perform” for the Republican Party under a range of vote-

share scenarios. See Whitford, 218 F.Supp.3d at 857-58.

Because the expected EG accounts for the range of electoral

scenarios, it won't be thrown off by the idiosyncrasies of any given

election, such as which candi. ates opted to run, what strategies

17

like calculating the realized EG, except that the two-

party vote is imputed for every seat, not just the uncon-

tested ones."° This kind of analysis is common for

those who draw gerrymandered maps."®

The expected EG measures the opportunity to secure

representation under a given map (i.c., the field on

which the political parties do battle), whereas the real-

ized EG measures outcomes (i.e., the results of the

battle—votes cast and seats won or lost). Though

amicus takes no position on whether the expected

EG should be used in a given case, he observes that

in related contexts, courts have often favored such

opportunity-based standards. See, e.g., Johnson v. De

Grandy, 512 U.S. 997, 1013-14 & n.10 (1994) (distin-

guishing opportunity and outcome standards for vote-

dilution claims under the Voting Rights Act, and

expressly rejecting the latter); Frank v. Walker, 768

F.3d 744, 748—49, 752—54 (7th Cir. 2014) (Easterbook,

J.) (defending opportunity-not-outcome standards for

“vote denial” cases under the Equal Protection Clause

and the Voting Rights Act).

they pursued, and the ups and downs of presidential and guber-

natorial approval ratings.

° The expected EG was calculated for a pending case in North

Carolina, which was brought before any elections had been held

under the map at issue. The forecasts used in pre-election calcu-

lations proved to be extremely accurate. See Simon Jackman,

Rucho, Rebuttal Report, League of Women Voters of North

Carolina v. Rucho, No. 1:16-CV-1026 (M.D.N.C. 2017) at 16 [here-

inafter Jackman, Rucho, Rebuttal Report) (“(T)he efficiency gap

predicted ... for an electoral environment like that of 2016 was

accurate to within a percentage point.”).

‘6 See supra note 9.

18

Il. THE OBJECTIONS TO THE EFFICIENCY

GAP AS A MEASURE OF PARTISAN

ADVANTAGE ARE EASILY MET.

It has been said that the EG rests on a mistaken

conception of wasted votes; that it is excessively or

misleadingly volatile; that it fails to account for demo-

cratic values other than partisan symmetry; that it is

“biased” toward the Democratic Party; and that it

requires proportional representation. These objec-

tions are easily met.

A. The Efficiency Gap Does Not Rest on a

Mistaken Conception of Wasted Votes.

Some critics maintain that no summary measure

of wasted votes should be used to gauge the effect

of a partisan gerrymander because all votes have

some potential to affect the representative’s behavior.

Because of this, the critics say, there simply is no iden-

tifiable class of “wasted” votes. See Brief for Appel-

lants at 49-50, Whitford v. Gill, 218 F.Supp.3d 837

(2016) [hereinafter Appellants’ Brief]; Brief of Amici

Curiae Wisconsin State Senate and Wisconsin State

Assembly in Support of Appellants 18-19, Whitford v.

Gill, 218 F.Supp.3d 837 (2016) (No. 16-1161), 2017 WL

1506064. Other critics accept the idea of tabulating

wasted votes, but argue that the EG equation uses an

incorrect definition or standard to summarize relative

wasted votes. See, e.g., Whitford, 218 F.Supp.3d at

957—59 (Griesbach, J., dissenting) (arguing that votes

for the winner should be deemed “wasted” insofar as

the winner’s vote «hare exceeds that of the runner-up,

rather than 50%); cj. McGhee, Measuring Efficiency in

Redistricting, supra (investigating this and other pro-

posed revisions to the EG formula). Neither objection

has merit.

19

The all-votes-may-have-influence argument overlooks

the fact that winning more seats without winning over

more voters is the point of a partisan gerrymander.

The EG is thus grounded in the Efficiency Principle:

Any measure of partisan effect must indicate a greater

advantage for a party when the seat share for that

party increases without any corresponding increase

in its vote share. See Part 1.A, supra. Indeed, it is

precisely because votes cast by legitimate voters are

valued in a democracy that normative concerns are

raised if one party’s votes are devalued because they

are less effectively converted into seats than another

party's votes due to partisan gerrymandering. a

Justice Kennedy noted in Vieth, “The inquiry .

whether political classifications were used to Ag a

group’s representational rights.” Vieth 541 U.S. at 315.

In the absence of some “compelling interest,” id., that

concern is not ameliorated because the votes, although

devalued in terms of winning seats (the partisan

gerrymanderer’s concern), might still have some valuc

in influencing an office holder of the opposing party.

Moreover, the “all-votes-may-have-influence” argu:-

ment elevates a theoretical possibility above every-

thing that is actually known about elections and rep-

resentation.’’ A legislator’s margin of victory has very

little effect on how she or he votes un bills; by contrast,

the effect of replacing a Democratic legislator with a

' The theoretical possibility is not legally determinative.

Though a plurality of this Court once treated the presumption

that voters who support the losing candidate have influence over

the winner as a reason not to strike down partisan gerrymanders,

that plurality also recognized that the presumption is subject

to “actual proof.” See Davis v. Bandemer, 478 U.S. 109, 131-

32 (1986).

20

Republican or vice versa is huge, holding constant dis-

trict characteristics. See, e.g., Stephen Ansolabehere

et al., Candidate Positioning in U.S. House Elections,

45 AM. J. PoL. Sci. 136 (2001); Barry C. Burden,

Candidate Positioning in US Congressional Elections,

34 Brit. J. POL. Sci. 211 (2004); Devin Caughey et al.,

Partisan Gerrymandering and the Political Process:

Effects on Roll-Call Voting and State Policies, 16

ELECTION L.J. (forthcoming 2017), available at http://

cwarshaw.scripts.mit.edu/papers/CTW_efficiency_gap

_170515.pdf (May 15, 2017); David S. Lee et al.,

Do Voters Affect or Elect Policies? Evidence from the

U.S. House, 119 Q.J. ECON. 807 (2004); Nolan McCarty

et al., Does Gerrymandering Cause Polarization?,

53 Am. J. POL. Sci. 666, 671 fig. 3 (2009); Keith T.

Poole, Changing Minds? Not in Congress!, 131 PUB.

CHOICE 435 (2007).

Lack of representation for citizens who vote for the

losing candidate extends even to rudimentary matters

of constituent service. In just the past few years, data

vendors have revolutionized campaigns by merging

official state voter files with consumer databases, creat-

ing estimates of the partisanship and turnout pro-

pensities of every registered voter. See generally

DANIEL KREISS, PROTOTYPE POLITICS: TECHNOLOGY-

INTENSIVE CAMPAIGNING AND THE DATA OF DEMOCRACY

(2016); EITAN HERSH, HACKING THE ELECTORATE: How

CAMPAIGNS PERCEIVE VOTERS (2015). Sophisticated

legislators are now using such estimates to screen

constituent inquiries, disregarding citizens whom the

data vendors classify as unlikely to vote for the incum-

bent. See HERSH, supra, at 200-05.

Some critics have usserted a related point: that

voters make complex decisions but the EG assumes

they only care about parties. See, e.g., Brief of Amici

21

Curiae The Republican National Committee and the

National Republican Congressional Committee in

Support of Appellants 41-50, Whitford v. Gill, 218

F.Supp.3d 837 (No. 16-1161), 2017 WL 1548280

(hereinafter RNC Amici Brief]. This is incorrect. The

EG does not ignore independent or swing voters.

Rather it requires that neither party be forced to

win over more of them to avoid a “wasted vote” deficit.

Moreover, the sensitivity testing described above can

explore the consequences of significant nonpartisan

voting behavior, should evidence of such behavior

emerge in the future.

As for the criticism that the EG formula uses an

incorrect definition to summarize relative wasted votes,

amicus has for a forthcoming article inve tigated pro-

posed reformulations, including Judge Griesbach’s

argument that votes cast for the winner should be

counted as wasted to the extent that they exceed the

second-place candidate’s vote share (rather than 50%).

See McGhee, Measuring Efficiency in Redistricting,

supra. Amicus’s analysis shows that the reformula-

tions would imply a level of responsiveness (in the trans-

lation of votes into seats) that deviates from the ! 2

votes-to-seats curve historically typical of American

elections, and would be far less consistent with the

Efficiency Principle. It follows that a legal standard

resting on Judge Griesbach’s preferred definition of

wasted votes could disrupt American electoral tradi-

tions, while doing a poor job of identifying maps that

distribute opposing-party voters for maximum parti-

san advantage.

Judge Griesbach’s argument also elides the fact that

in our essentially two-party system a candidate cannot

be sure of winning without 50% of the vote because

votes not cast for the candidate of one party are almost

22

always cast for the candidate of the other party.

Unlike Judge Griesbach’s baseball example based on

runs not vote percentages, a Republican who wins

with 52% of the vote to the Democratic opponent’s 48%

would likely have lost had the Republican’s vote share

fallen to 48%. To similar effect, if Republicans com-

prising 4% of the district had been moved into another

district and replaced with non-Republican voters, the

Republican candidate almost surely would have lost.

The 50% threshold embodied in the EG formula indeed

has “special significance[] in the democratic process.”

Bartlett v. Strickland, 556 U.S. 1, 19 (2009) (plurality

opinion of Kennedy, J.).

B. The Putative Volatility of the Efficiency

Gap Is a Feature, Not a Flaw.

Like a number of commentators, Judge Griesbach

has challenged the EG because it may vary from elec-

tion to election, particularly under maps in which

there are many competitive districts. See Whitford,

218 F.Supp.3d at 959-62 (Griesbach, J., dissenting);

see also Appellants’ Brief, supra, at 51-52.

This objection reflects a failure to appreciate two

very fundamental points. First, any measure of parti-

san advantage that takes account of which party actu-

ally wins seats has some potential to vary from election

to election, particularly if the map in question has

a substantial number of competitive districts.“ Such

When the number of legislative districts is very small, the

standard percentage form of the EG should be converted to raw

seats to avoid exaggerating either the size or volatility of the bias

in the plan. For example, in a two-seat plan a party can increase

its seat share (and so its EG) by 50% by winning just one more

seat. It would be better to present this change as the one-seat

shift that it is. Small-seat plans like this are easily identified

and managed in any potential litigation. Cf. Stephanopoulos &

23

“volatility” simply reflects actual partisan advantage

in given elections.

Second, no proponent of judicial action as to parti-

san gerrymanders maintains that a legislative map

should be deemed constitutionally suspect merely on

the basis of the EG observed in a single election. It is

imp tant to confirm through sensitivity testing that

the EG is likely to be durable."” See Stephanopoulos &

McGhee, supra, at 887-90; Whitford, 218 F Supp.3d at

898-910.

In sum, the so-called volatility of the EG simply

reflects realities on the ground that vary from election

to election. Far from being a weakness, it is actually

a safeguard against excessive judicial involvement in

partisan gerrymandering cases, relative to measures

that might lead courts to think that an observed parti-

san asymmetry is more stable than it actually is.

C. The Efficiency Gap Does Not Account

for All Democratic Values—Nor Should

It.

The EG has been faulted for its failure to account for

certain democratic values. Judge Griesbach observed,

for example, that the EG “counts” all districts equally

(a wasted vote is a wasted vote, regardless of the

district in which it was cast), whereas as a functional

matter the median district is more important than the

others. See Whitford, 218 F Supp.3d at 952-53. And

he pointed out that large FGs could materialize for

McGhee at 887-89 (proposing a two-seat threshold for cases

about congressional redistricting).

® The volatility objection would also vanish if courts grounded

liability rulings on the expected EG, rather than the realized EG.

See supra Part I.D.

24

entirely legitimate reasons, as may occur if districts

were drawn for congruence with political subdivision

boundaries in a state where supporters of one political

party happened to be more efficiently distributed with

respect to those boundaries than supporters of the

other party. Id. at 962-63.

Such objections are easily met because the EG only

measures partisan symmetry, not everything that

might be deemed good or bad about a map of legisla-

tive districts. The argument that the EG “fails to

account for everything” just restates the principal

argument against judicial intervention in the malap-

portionment cases, namely, that because the equal-

population norm is not the exclusive consideration in

designing maps for fair and effective representation,

the norm should not be judicially enforced az all.

See Baker v. Carr, 369 U.S. 186, 301-24 (1962)

(Frankfurter, J., dissenting). The argument is no

stronger today than it was in 1962.

To address potentially competing values, courts may

construct a framework under which other values can

justify legislative maps that deviate substantially

from the equality norm. Amicus takes no position on

what the substantial-asymmetry trigger for height-

ened scrutiny should be, or on what state interests are

weighty enough to justify large asymmetries.

Similarly, if courts conclude that there is no reason

to intervene unless the plaintiff political party would

have some chance of winning the median district under

a fair map, courts could require plaintiffs to prove both

substantial partisan bias under the challenged map

and potential-majority status under an unbiased map.

25

D. The Efficiency Gap Is Not Biased

Toward Democrats.

It has been suggested that the EG’s appealing veneer

conceals a biased mechanism for indicting Republican

gerrymanders while insulating Democratic gerryman-

ders from scrutiny. See Appellants’ Brief, supra, at

50—61; RNC Amici Brief.

Notably, those who assert that the EG is “biased”

toward Democrats haven’t actually shown or even

advanced a theory as to how EG calculation could lead

to exaggerated estimates of the partisan asymmetry

of Republican gerrymanders and deflated estimates

of the partisan asymmetry of Democratic gerryman-

ders.” Nor is it the case, historically, that the EG has

shown a consistent Republican advantage in enacted

maps. In the 1970s and 1980s (when the RNC favored

judicial policing of partisan gerrymanders, contrary to

its position today*') enacted maps generally had a pro-

Democratic EG, whereas recent maps have tended to

favor Republicans (hardly surprising given Republi-

can control of state legislatures). See Stephanopoulos

& McGhee, supra, 882-84 fig. 8; Jackman Wisconsin

Report, supra, at 44. The EG also suggests that

” They rely instead on hypothetical examples of imaginary

gerrymanders under utterly implausible political conditions.

Compare Brief of Amici Curiae The Republican National Commit-

tee and the National Republican Congressional Committee in

Support of Appellants 16-22, Whitford v. Gill, 218 F.Supp.3d 837

(No. 16-1161), 2017 WL 1548280 (presenting hypothetical exam-

ple from the defendants’ expert in pending North Carolina case),

with Jackman, Rucho, Rebuttal Report, supra, at 14—16 (explain-

ing the ridiculous features of this example).

*! See Brief Amicus Curiae of the Republican National Commit-

tee in Support of Appellees *3-8, Davis v. Bandemer, 478 US.

109 (1986) (No. 84-1244), 1985 WL 670030.

26

Democratic and Republican gerrymanders are equally

effective: Redistricting under unified Democratic

control shifts the EG in the pro-Democratic direction

by about the same amount as redistricting under

unified Republican control shifts the EG in the pro-

Republican direction. See Nicholas Stephanopoulos,

The Causes and Consequences of Gerrymandering, 59

Wm. & MARY L. REV. figs. 1 & 2 (forthcoming 2017),

available at https://papers.ssrn.com/sol3/papers.cfm?

abstract_id=2990638; see also Nicholas O. Goedert,

Gerrymandering or Geography? How Democrats Won

the Popular Vote but Lost the Congress in 2012, RES. &

POL., Apr._June 2014.”

To the extent those arguing “Democratic bias” have

a discernible theory, it is not about the EG per se, but

about the very concept of partisan symmetry. The

argument is that the fairness (for equal protection

purposes) of a map should not be gauged by symmetry—

i.e., an equal chance for left-of-median and right-of-

median voters to secure a legislative majority—but

instead by “traditional” districting principles such as

compactness and contiguity with political subdivision

boundaries. See, e.g, Appellants’ Brief, supra, at

50-51 (arguing against a “zero-gap baseline”); RNC

Amici Brief, supra, at 8-14. This issue goes beyond

the scope of this brief, though there are good reasons

for this Court to choose the symmetry norm,” while

“ Similar results on the effects of unified control have been

found using other measures of partisan symmetry. See Andrew

Gelman & Gary King, Enhancing Democracy Through Legislative

Redistricting, 88 AM. POL. Sci. REV. 541, 553 (1994); Richard

G. Niemi & Simon Jackman, Bias and Responsiveness in State

Legislative Districting, 16 LEGIS. STUD. Q. 183, 195 (1991).

*™ Justice Kennedy in Vieth warned against legal standards

that valorize compactness and contiguity with subdivision bound-

aries because such criteria could “benefit one political party over

27

embedding that norm within a doctrinal framework

that allows states to justify deviations as may be

reasonably necessary to achieve other legitimate

objectives.

E. The Efficiency Gap Does Not Require

Proportional Representation.

Appellants’ objection that the EG requires propor-

tional representation confuses two potential meanings

of the term “proportional representation.” See Appel-

lants’ Brief, supra, at 49-650. The classical or Millian

usage denotes an electoral system in which small

groups of citizens who share an interest can band

together and secure a fraction of legislative seats

approximately equal to the group’s share of the

voting-eligible population. See JOHN STUART MILL,

CONSIDERATIONS ON REPRESENTATIVE GOVERNMENT

146 (Prometheus Books 1991) (1861) (“In a really

equal democracy, every or any section would be

represented [with seats in the legislature], not

disproportionately, but proportionately. ... Man

for man, [minorities] woulc be as fully represented

as the majority.”). Though the Millian usage

remains standard, one might also use “proportional

representation” to describe any electoral system in

which there exists some usual relationship between a

another.” Vieth v. Jubelirer, 541 U.S. 267, 308-09 (2004) (Kennedy,

J., concurring in judgment). The normal-map alternative to the

“zero-gap baseline” is vulnerable to Justice Kennedy’s objection

because it could result in judicial invalidation of symmetric maps

when applied to a state where one party’s voters are more effi-

ciently distributed than the other’s. For example, a zero-EG plan

in a state where the putatively normal maps favored Republicans

by 10% would, as a matter of law, have a pro-Democratic “bias”

of 10% (the difference between the EG of the enacted map, zero,

and the typical EG of putatively normal maps, 10%).

28

political party’s vote shares and seat shares. Appel-

lants and Judge Griesbach invoke the latter concep-

tion when they criticize the EG for implying a 2:1

“winner’s bonus,” i.e., seat margins about twice as

large as vote margins. See Whitford, 218 F.Supp.3d at

947-51 (Griesbach, J., dissenting); Appellants’ Brief,

supra, at 49—50.

By contrast, this Court, in the many cases in which

it has “set its face against the claim .. . that the

Constitution somehow guarantees proportional repre-

sentation,” City of Mobile, Ala. v. Bolden, 446 U.S.

55, 79 (1980) (plurality opinion), has consistently used

the term in the standard Millian sense. See, e.g.,

Whitcomb v. Chavis, 403 U.S. 124, 156 (1971) (revers-

ing district court’s constitutional vote dilution holding,

on the ground that the lower court’s decision

“[expressed] the proposition that any group with dis-

tinctive interests must be represented in legislative

halls if it is numerous enough to command at least

one seat”) (emphasis added); Bolden, 446 U.S. at 75—

80 (discussing Whitcomb—and the prospect of various

and sundry vote-dilution claims from, e.g., “union ori-

ented workers, the university community, [and] reli-

gious or ethnic groups”—in rejecting the notion that

Reynolds v. Sims’ (377 U.S. 533 (1964)) recognition of

a right to “equally effective voice” implies “propor-

tional representation as an imperative”); Bandemer,

478 U.S. at 145-47 (O’Connor, J., concurring in the

judgment) (positing if partisan gerrymandering claims

were justiciable, this would lead to a guarantee of

“rough proportional representation for all political

groups,” and an unmanageable proliferation of claims

from “political, religious, ethnic, racial, occupational,

and socioeconomic groups”) (emphasis added); Vieth,

541 U.S. at 288 (plurality opinion) (“[The Constitution]

nowhere says that farmers or urban dwellers, Christian

29

fundamentalists or Jews, Republicans or Democrats,

must be accorded political strength proportionate to

their numbers.”).

Nothing about the EG or other proposed symmetry

metrics would favor proportional representation in the

Millian sense. To the contrary, as Part 1.B of this brief

explained, an EG of zero implies not proportional rep-

resentation but the very winner’s bonus that has been

historically typical of United States elections.

It is possible that if this Court upholds the lower

court’s decision, minor parties might start bringing

partisan-gerrymandering claims. But the EG metric

wouldn’t do them any good. Ina forthcoming publi-

cation, see McGhee, Measuring Efficiency in Redis-

tricting, supra, amicus shows that the EG can be

extended to a world in which minor parties win a non-

trivial share of the vote, but this extension of the EG

does not indicate “bias” against minor parties—even

from a complete lack of legislative seats—unless the

party wins a share of the vote that far exceeds what

minor parties have historically achieved in U.S.

elections.**

Ill. THE EFFICIENCY GAP IS CLOSELY

RELATED TO THE OTHER MEASURES

OF PARTISAN SYMMETRY, BOTH

ARITHMETICALLY AND EMPIRICALLY.

Appellants fault respondents for encouraging courts

to consider a supposed “hodgepodge” of metrics of

partisan bias or symmetry. Appellants’ Brief, supra,

at 45-48. This is said to leave district courts with

* Formally, there is no EG bias against the party (from a lack

of seats) until the party wins a fraction of the statewide vote in

excess of 1/2k, where k is the number of parties.

30

unbounded discretion to “pick a winner” in each case.

Id. at 46-47. But all the metrics are tied together by

a common concept—the neutral principle of partisan

symmetry—and the metrics are arithmetically related

to one another and highly correlated in politically

competitive states like Wisconsin. What social science

has produced is not a “hodgepodge” but a set of closely

related analytical tools for measuring a common

concept.

A. Arithmetically, the Symmetry Metrics

Are Closely Related to One Another.

Scholars have variously measured partisan sym-

metry with Gelman-King bias, discussed in Part I.B

above; the EG; the “mean-median difference,” which

is the gap between a party’s mean district-level vote

share and its median district-level vote share;* and

the “average winner's margin difference,” which is the

difference between the average margin of victory of the

winning candidates of each party.”

The most salient difference among these metrics is

that the EG can be used to evaluate maps in non-

competitive states, while the Gelman-King and mean-

median metrics would depend on fanciful counter-

factuals in such states.”” For that reason the propo-

nents of Gelman-King recommend using it only in

competitive states. See supra Part I.B.

*% Michael D. McDonald & Robin E. Best, Unfair Partisan

Gerrymanders in Politics and Law: A Diagnostic Applied to Six

Cases, 14 ELECTION L.J. 312, 316 (2015).

* Samuel S.H. Wang, Three Tests for Practical Evaluation of

Partisan Gerrymandering, 68 STAN. L. REV. 1263, 1303—04 (2016)

(calling this the “lopsided-margins test”).

27 This difference may prove critical for partisan gerrymander-

ing claims about congressional districts, because partisan bias (or

31

But in competitive states like Wisconsin, the differ-

ences are less important than the similarities. Figure

1, below, shows a hypothetical votes-to-seats curve,

together with the quantities corresponding to the

mean-median difference, and Gelman-King bias. As

can be seen, the mean-median difference indicates the

statewide two-party vote margin (from 50%) a party

needs to control 50% of the seats, whereas Gelman-

King bias indicates the seat margin that a party

receives if it wins 50% of the vote. Indeed, the mean-

median difference is equivalent to Gelman-King bias

divided by the slope (responsiveness) of the votes-to-

seats curve at the tied-election point. See Jackman,

Rucho, Rebuttal Report, supra, at 8.

If one calculates the EG at these points in the seats-

votes curve, the measures are almost perfect substi-

tutes for each other. Gelman-King bias and the EG

are mathematically identical when both parties have

50% of the vote, and when the median is 50% (giving

both parties half the seats) the EG is precisely twice

the mean-median difference. See McGhee, Measuring

Efficiency in Redistricting, supra. Furthermore, it has

been shown that the “average winner’s margin differ-

ence” is mathematically identical to the EG under

all conditions, not just at certain counterfactual points

on the votes-to-seats curve. The bottom line is that,

although the EG has distinct advantages as explained

in Part I.B., all of these measures are closely related

its absence) in Congress as a whole depends on redistricting

decisions made in uncompetitive as well as competitive states.

An additional seat for Congress is equally significant whether

it comes from an uncompetitive state like California or Texas

or a competitive one like North Carolina or Pennsylvania. See

Stephanopoulos & McGhee, supra, at 861.

32

to one another, and indeed the EG can be used as a

substitute for any of them if desired.

Figure 1. Dlustration of Votes-to-Seats Curve,

and Two Measures of Partisan Symmetry

ary

4

B. Empirically, Little Turns on the Choice

Among Symmetry Metrics in Competi-

tive States Such as Wisconsin.

While the plaintiffs relied on the EG in this case, it’s

very unlikely that the trial court’s finding that the

Wisconsin map conferred a large, durable advantage

on the Republican Party would have been any differ-

ent had the plaintiffs instead used Gelman-King bias

or the mean-median difference.

Simon Jackman has calculated Gelman-King bias

and the EG for state legislative maps and congres-

sional maps from 1972 to the present. He found that

in competitive states (such as Wisconsin and North

Carolina), the two measures are extremely highly cor-

related, but that they often diverge in uncompetitive

33

states. See Plaintiffs’ Trial Brief at 29-31, Whitford

v. Gill, 218 F.Supp.3d 837 (2016) (No. 15-cv-421-bbc)

(discussing and il) ustrating state legislative findings);

Jackman, Rucho, Rebuttal Report, supra, at 2-8,

(reporting congressional findings). The divergence in

uncompetitive states is not surprising; it reflects the

speculative nature of projecting seat-shares at the 50%

vote-share point in those states where one party con-

sistently dominates the other, as explained in Part I.B.

Jackman also found the mean-median difference

and Gelman-King bias to be highly correlated with one

another, in both competitive and uncompetitive states.

See Jackman, Rucho, Rebuttal Report, at 8. This cor-

relation is also to be expected given the mathematica]

connection between the two measures noted above.

Amicus has calculated for this brief the EG, Gelman-

King bias, and mean-median difference of the current

Wisconsin assembly map for 2016. Despite widespread

claims of a political transformation in Wisconsin in the

2016 presidential election, Republicans received just

52% of the two-party vote in the state’s Assembly

elections.” Republicans nonetheless turned this nar-

row margin into 65% of the seats. The EG for the

election favored Republicans, and by a substantial

11 percentage points. Gelman-King bias (13%) and the

mean-median difference (7%) also indicate that the

map continues to favor Republicans.”®

%8 This is the vote share calculated with imputations for uncon-

tested seats (i.e., estimating the parties’ vote shares in these dis-

tricts had the incumbent been challenged by a candidate of the

other party).

* This is consistent with record evidence showing that, in

Wisconsin, the EG and Gelman-King bias have been very closely

correlated in competitive election years. See Brief of Plaintiffs

34

The bottom line is that, in Wisconsin and other

competitive states a map that has been shown to be a

“symmetry outlier” by one measure is also likely to be

an outlier using the other metrics.

CONCLUSION

Amicus hopes that the explanation of the Efficiency

Gap provided in this brief will help this Court make

an informed judgment about whether to employ the

EG, alone or in conjunction with the larger family of

partisan symmetry metrics, in determining the consti-

tutionality of partisan gerrymanders.

Respectfully submitted,

CHRISTOPHER S. ELMENDORF DANIEL F. KOLB

UNIVERSITY OF CALIFORNIA, Counsel of Record

DAVIS SCHOOL OF LAW LYNN E. BUSATH

400 Mrak Hall Drive JOHN HAMMOND HEATH

Davis, CA 95616 ELIZABETH A. TIPPETT

(530) 752-5756 DAVIS POLK & WARDWELL LLP

cselmendorf@ucdavis.edu 450 Lexington Avenue

New York, NY 10017

(212) 450-4000

Daniel.Kolb@DavisPolk.com

Counsel for Amicus Curiae Eric McGee,

Public Policy Institute of California

August 10, 2017

at 26-31, Whitford v. Gill, 218 F.Supp.3d 837 (2016) (No. 15-cv-

421-bbc).

” That said, if one uses the EG, one can consider a wider range

of states and historical periods when determining the level of bias

sufficient to mark a map as an outlier. The reason is that the EG

can be computed for any map in any state, whereas Gelman-King

bias and the mean-median difference are applicable only to states

in which the parties are pretty evenly matched.