Considering the Prospects for Identifying a Gerrymandering … · 2016-06-03 · the winning candidate takes two forms. Andrew Hacker, who refers to the winner’s wasted votes as

Considering the Prospects for Identifying

a Gerrymandering Standard

Robin Besta

Shawn J. Donahueb

Jonathan Krasnoc

Daniel B. Maglebyd

Michael D. McDonalde

Abstract

Courts and the States themselves have found it difficult to rid redistricting processes of partisan manipulation The knotty problem, legally and procedurally, is that no proposed standard has found acceptance as a convincing means for identifying whether a districting plan is a partisan gerrymander with knowable harmful effects. We review five standards for curbing gerrymandering in order to diagnose their shortcomings and to see whether it might be possible to get purchase on what is needed to overcome them. We take as our perspective how easily manageable and effective each would be to apply at the time a redistricting authority decides where to draw the lines or, post hoc, when a court is asked to decide whether a gerrymander has been enacted. Upon review we conclude that a combination of the two symmetry standards offer the best prospects but only if recognize that gerrymandering’s harm takes two forms: diluting votes and curtailing speech.

_______________________________ Department of Political Science, Binghamton University, Binghamton, NY 13902-6000;

Phone; (607) 777-2252; FAX (607) 777-2675 a Assistant Professor of Political Science, [email protected] b PhD Student in Political Science, [email protected] c Associate Professor of Political Science, [email protected] d Assistant Professor of Political Science, [email protected] e Professor of Political Science, [email protected]

1

Partisan gerrymandering drains the energy and vitality from elections as

instruments of democracy. By packing one set of partisans in a small number of

districts, opposition partisans can maintain a legislative majority even without winning a

vote majority. By spreading one set of partisans evenly across very many districts, the

minority partisan voices in a legislative chamber can be all but silenced. Nevertheless,

despite the democratic infirmities associated with gerrymandering, policing let alone

eliminating the practice has proven difficult. The problem, legally and procedurally, is

that no proposed standard has found acceptance as a convincing means for identifying

whether a districting plan is a gerrymander with knowable harmful effects.

Justice Scalia, announcing the Court’s judgment in Veith, notes that “[t]he term

‘political gerrymander’ has been defined as ‘[t]he practice of dividing a geographical area

into electoral districts, often of highly irregular shape, to give a political party an unfair

advantage by diluting the opposition’s voting strength’” (Vieth v. Jubelirer, 2004, 271 n.

1, quoting Black’s Law Dictionary 1999, 696). Finding intention and observing weirdly

shaped districts are seldom difficult (as in Davis v. Bandemer 1986; Veith v. Jubilier

2004, LULAC v. Perry 2006), but finding a standard that identifies a party’s unfair

advantage because the opposition party’s votes have been diluted has proved elusive.

We review five proposed standards for identifying gerrymandering’s harmful

effects. All five have been presented to courts in recent years, but to date none has been

found actionable. Our purpose is to spot the shortcomings of each and to see whether it

might be possible to gain purchase on what is needed to surmount them. The five

include counting wasted votes for one set of partisans versus the other (Stephanopolous

and McGhee 2014), applying a vote-denominated symmetry test (McDonald and Best

2015), comparing the number of districts won in an enacted plan against expected wins

from a null set of neutrally drawn plans (Chen and Rodden 2013a), adopting a seat-

denominated symmetry test (Grofman and King 2007), and setting up a three-prong test

for proportionality, lack of responsiveness, and vote-denominated symmetry (Wang

2015). Stephanopolous and McGhee along with McDonald and Best concern themselves

principally with diluted votes; Chen and Rodden as well as Grofman and King focus

2

principally on unfairness to parties; and Wang tackles both unfairness to parties and

diluted vote weights.

At each turn in our reviews we take as our perspective how easily manageable and

effective each would be to apply at the time a redistricting authority decides where to

draw the lines or, post hoc, when a court is asked to decide whether a gerrymander has

been enacted. Our evaluations show that both counting wasted votes and comparing

actual to expected district wins are manageable but not effective; neither focuses on

structural features of the districting plan in direct association with gerrymandering.

Vote-denominated symmetry is both manageable and effective but applies only to

packing forms of gerrymandering. Seat-denominated symmetry applies to packing and

packing-plus-cracking gerrymanders, though not packing by itself, but is not entirely

manageable because its reading of gerrymanders is situational and requires reliance on

nonfactual hypotheticals. The three-prong approach fails on its own terms because the

prongs do not fit together as a coherent whole and, worse, can operate at cross-purposes.

All in all, the reviews lead to these surprising, tentative conclusions. Only the vote-

denominated symmetry standard is capable of identifying gerrymandering’s harm as

unfairness to a party because its supporters’ votes have been diluted. Expanding the

concept of harm to cover unfairness to a party as such requires extending the offense to

silencing or otherwise controlling the voice of the people in violation of the First

Amendment.

Wasted Votes

Counting and comparing wasted votes is the basis for the efficiency gap standard

proposed by and Stephanopoulos and McGhee (2015; see McGhee 2014 for the

underlying social science thinking). A 2015-16 challenge to Wisconsin’s Assembly

districts relies on it to charge partisan gerrymandering (Whitford v. Nichol 2015, cv-

421). The approach proceeds from the insight that both winners and losers “waste” votes

by inefficient allocation in an election. That is, any votes above the winning number for

the winner plus all votes for the loser are wasted in that they contribute nothing of

3

determinative importance to deciding who wins. So, in an election decided by a 60-40

margin, the winner wastes 10 percentage points above 50% (setting aside ties for the

sake of simplicity), while the loser wastes all 40 percentage points. Comparing the

magnitude of the waste on both sides, 10 versus 40, shows an efficiency gap (of 30

points) favoring the winner. McGhee and Stephanopoulos argue that in a non-

gerrymandered system both sides waste the same number of votes, so ideally the

efficiency gap should equal zero.

Their claim is appealingly simple, straightforward, and intuitive. We know that

gerrymanders function by inequitably allocating votes in some fashion. The question is

whether comparing inefficiencies in this way fares well as a manageable and effective

metric for detecting partisan gerrymanders.

On the question of manageability, its simple computation would seem to clear the

bar, but even the counting process is an open question.1 It runs into a second

manageability problem when deciding how many wasted votes signal a gerrymander.

Because no democratic or legal principle answers the question of how many wasted votes

are needed to say a plan is a gerrymander, the approach calls for comparisons to the

historical record in the same jurisdiction and contemporaneous results in other

jurisdictions. Such relative baselines beg the question of whether what occurred

previously in the same jurisdiction or is occurring contemporaneously in other

jurisdictions are results contaminated by gerrymandering.

Effectiveness, too, runs into two problems. First, and simply, under single-

member district rules votes are wasted for reasons other than gerrymandering. One

needs to look no further than a simple example of a congressional district in a one-

district state such as Delaware to see this. Unless the vote splits 75-25, one wastes more

votes than the other, this despite the fact that a gerrymander is impossible in a one-

district state. Maybe the efficiency gap is useful only in multi-district situations, but that

1 While the arithmetic is simple, and in that sense clears the bar, the definition of votes wasted by the winning candidate takes two forms. Andrew Hacker, who refers to the winner’s wasted votes as excess votes, defines them as one more than the votes received by the losing candidate (Hacker 1964, 55-7). McGhee (2014) and Stephanopolous and McGhee (2015) define a winner’s excess/surplus/wasted votes as votes beyond 50%+1.

4

can’t be true either. In a three districts state, a symmetrical distribution of 48-52-56 has

a gap of +8.3 in favor of the majority party and is, by the Stephanoplous and McGhee

criterion a gerrymander. Of course, if the vote shifted uniformly to 46-50-54, there is no

gerrymander, even though it is the same districting plan. Then, if votes shift another two

points to 44-48-52, the gerrymander would be said to run in the direction opposite of

what was inferred from the original 48-52-56 distribution. In fewer words, reading a

gerrymander from the efficiency gap can and often will vary depending on the

percentage level of the votes a party receives. Again we see, as in the one-district

example, that the efficiency gap is not solely recording a structural feature of the

districting plan as it relates to gerrymandering.

The second effectiveness problem is that a necessary condition for equalizing

wasted votes is a majoritarian seat-vote ratio of two to one (assuming equal turnout in all

districts and accepting the McGhee counting of wasted votes rather than Hacker’s; see

fn. 1). Every one percentage of the majority party’s vote above 50 wins that party an

additional two percent of the districts—55 percent of the vote wins 60 percent of the

seats. Why a two-to-one ratio is desirable is an open question, except to say that it

equalizes wasted votes. Equally damning, false positives and false negatives are likely to

abound. Perfectly symmetrical vote distributions can fail the two-to-one standard.2 As

well, a two-to-one ratio can be achieved precisely because a plan has a strong asymmetry

bias.3 So, despite the claim that the efficiency gap standard comports with and arises

from the idea of partisan symmetry, it doesn’t really.4

The wasted vote approach has clear intuitive appeal, and its computation is

easy—just count and compare wasted votes using straightforward arithmetic. Its

2 The 48-52-56 vote distribution discussed above offers a simple example of a false positive; 52 percent of the vote wins 66.7 percent of the seats—an 8.33 ratio instead of the 2.00 ratio. Thus, the efficiency gap identifies this plan as biased in favor of the majority party when it clearly is not.

3 A 40-40-60-65-70 vote distribution offers a simple example of a false negative. The distribution is asymmetrical (median 60 and mean 55), but the efficiency gap shows an equal number of wasted votes. Votes are five points above 50, and seats are ten points above 50. Thus the majoritarian ratio is two-to-one despite the asymmetry of the distribution.

4 See Stephanopoulos and McGhee (2015, 2-3 and passim) and Whitford v. Nichol (2015, 15-19) for claims about the relationship between symmetry and the efficiency gap.

5

downsides? Its definition of wasted votes is shaky. It under-achieves on the question of

manageability and overreaches on questions of effectiveness. While the arithmetic is

easy, evaluating the results requires using a relative comparison to the historical record

of elections in the same jurisdiction or to election results in other jurisdictions. A

historical comparison is liable to perpetuate gerrymanders in earlier years; comparison

to other jurisdictions leaves one wondering whether the baseline involves a mix of fair

and unfair outcomes elsewhere. Overreaching on effectiveness comes from two sources.

On is its variable reading of gerrymandering depending on the levels of the vote split

system wide. The other is the implied premise that single-member district elections are

completely fair if and only if they operate with a swing ratio of 2.0. Few single-member

district systems actually operate that way generally, and probably none do so

persistently. In effect, therefore, the wasted vote approach is an implicit indictment of

the single-member district rules. That is a nonstarter in American politics. It substitutes

a two-to-one majoritarian ratio for the view that fairness requires proportional

representation—i.e., a one to-one majoritarian ratio—which has not found acceptance in

American politics and law (Bandemer 1986, 129-30).

Vote-denominated Symmetry

The vote-denominated symmetry approach or, as McDonald and Best (2015)

refer to it, the equal vote weight standard, has been proposed by amici (Hebert and

Lang 2015) at the remedy stage of the Virginia litigation that earlier found the State’s

congressional districts to be an unconstitutional racial gerrymander (Page v. Virginia

State Board of Elections 2014). The approach relies on two easily observed facts: (1)

compare the median and mean district vote percentages, and (2) see whether majority

rule is violated. The point is to identify packing forms of gerrymandering on the thought

that packing is the prevalent form of gerrymandering alleged in recent decades. When

one group of partisans is relatively more packed than the other, a districting plan has the

potential to violate the widely embraced principle of equal vote weights and, from the

unequal weights, to entrench one party in majority status.

6

In all, the standard for a factual identification of a gerrymander rests on three

ideas.

1. Leading indicator: Asymmetrical packing exists when the median district vote

percentage for one party is persistently lower than its mean district vote

percentage.

2. Objectionable harm: A vote weight inequality is clearly identifiable when one

set of partisan voters casts a majority of the votes but carries less than a majority of

the districts, because violating majority rule occurs only when all votes do not

count equally.

3. Cause: District line placements are the known cause of the unequal vote weights.

Votes counted system-wide contribute equally to the count. Votes counted after

division into districts changes only the manner of counting. To the extent the two

forms of counting do not produce the same result the difference must be caused by

the line placements.

Its relative lack of aggressiveness in attacking gerrymandering comes from

majority rule’s central role in identifying gerrymandering as the cause of unequal vote

weights. This could prove disquieting in either of two situations.

In a jurisdiction with a predominant party, violations of majority rule are not

observed because one party always wins a vote majority. In this situation a median-

mean difference that violates majority rule is out of the question. So, for example, by

this standard the Texas congressional district, 2004 and 2010, could not be considered a

gerrymander because in those years Democrats never cast a vote majority for any Texas

office (McDonald and Best 2015, 327-8). A more aggressive symmetry requirement

could be used to challenge this sort of situation on two fronts. It could question whether

the majority party should be allowed to win an extra-large seat majority, somehow

defined. It could also be used to challenge the lack of responsiveness to vote shifts

because very many districts have been made noncompetitive—e.g., in Texas 28 of 32

districts didn’t undergo switched partisan control after the State Legislature drew the

lines in 2003.

7

Also tied to a focus on violating majority rule is this second disquieting situation.

Equal median and mean district vote percentages indicate only average symmetry, not

full-scale symmetry. Reaching for a full- or at least a fuller-scale approach would be

more aggressive. For example, a five district plan applied to two-party competition that

has (expected) Republican district vote percentages of 44, 46, 51, 52, and 62 is

symmetrical via the equal vote weight standard but asymmetrical under a full-scale

symmetry requirement. The median and mean are both 51. Thus, average symmetry is

upheld inasmuch deviations above and below the mean of 51 both average 6. Majority

rule is also preserved; the vote majority holds a 3 to 2 seat majority. Full-scale symmetry

goes wanting however, because something like uniform vote swings would result in

Republicans winning only three seats with 52 percent of the vote—an upward shift of 1

point resulting in a 45, 47, 52, 53, 63 distribution—but Democrats win 4 seats when they

have 52 percent of the vote—after a downward shift of three points resulting in a 41, 43,

48, 49, 59 distribution. While majority rule is maintained under both vote swings, the

idea of equality is strained because different rewards (seats) are acquired from the same

resources (votes).

The equal vote standard has pros and cons. It could be applied with relative ease.

The findings of fact are simple: compare the median and mean district percentages and

check for violations of majority rule. However, it is not as thoroughly effective as some

might demand. It holds predominant majorities safe from gerrymandering allegations,

and it under-reaches in situations when vote shifts could produce different seat

outcomes despite each party receiving the same vote percentages.

Comparison to a Null Set

This approach switches emphasis from diluted votes to unfairness to parties. If an

enacted plan is an outlier in a null set’s distribution, one can infer that it was probably

intended to hold a partisan advantage. As will soon become apparent, however, a crucial

question involves figuring out what characteristic needs to be evaluated in terms of its

distribution—districts won, perhaps the vote distribution’s number of competitive

8

districts, or perhaps its standard deviations, maybe its skewness, or maybe even its

kurtosis?5

Comparisons to a null set figured prominently in reports and testimony in the

challenge to Florida’s post-2010 congressional districts (i.e., Romo v. Detzner 2014).

The basic idea of using computers for this purpose has been around at least since

William Vickrey made the point more than a half-century ago (Vickrey 1961), and a few

pioneers succeeded in advancing the idea in modest ways in the 1960s and 70s (Nagel

xxxx; Engstrom and Wildgen 1977). With advances in processing speed, the approach

was ready for a full-scale application (e.g., Cirincione, Darling, and O’Rouke, 2000;

Altman and McDonald 2011; Chen and Rodden 2013a), or so it seemed in the run up to

the Florida proceedings. Both Darling along with Chen and Rodden produced null sets

(see Darling 2013; Chen and Rodden 2013b; 2014), and Rodden testify at length. In the

end, however, neither the reports nor Rodden’s testimony received any mention in the

trial court or subsequent court decisions.

For what it says about manageability, the court’s silence is disquieting. The

silence may have been benign. In the face of the smoking gun evidence of violating

Florida’s intent standard, the court might well have reasoned that nothing as

sophisticated as a null set was needed.6 Perhaps, however, the court was dissuaded from

crediting the method with probative value because one report identified a few contiguity

problems (Hodge 2013) and another report, plus testimony, questioned whether the

Chen-Rodden null set was randomly generated since no one can know the characteristics

of the population of all possible plans (McCarty 2013; 2014). Or, perhaps and more

simply, the black box nature of the method left the court unsure what reliable

conclusions could be drawn.

5 See Chen and Rodden (2013a) for an analysis that points to the number of wins; for the importance of the number of competitive districts see Engstrom and Wildgen (1977); on the standard deviation see, among others, Gudgin and Taylor (1979); on skewness see, among others, McDonald and Engstrom (1989); on kurtosis see Brady (1988).

6 The facts revealed such damning evidence as Republican legislators and their operatives enlisting mapmaking confederates to submit “citizen constructed plans” under fake names and writing scripts for “concerned citizens” to present the operatives’ ideas at public meetings (Romo v. Detzner 2014, 20-31).

9

Because the null set approach has yet to be tried and tested in a full form

application, questions about its effectiveness are open. Still, this much can be said. Not

much thought has gone into how the null set could be used to detect gerrymandering

beyond forming a baseline to say whether an enacted plan is an outlier in the null set

distribution and, on that basis, probably indicates a gerrymander. Engstrom and

Wildgen (1977, 469-70) evaluate a plan in regard to how many competitive districts it

contains. Cirincione et al. (2000), Darling (2013), along with Chen and Rodden (2013a,

2014) evaluate a plan in regard to the number of districts in which each racial group or

political party holds a majority. We have to suppose that focusing solely on the central

tendency is not enough. Why? Missing versus hitting the mark of the central tendency

leaves one wondering what, if any, harm has been created. Of the large number of plans

a computer can draw, the expected number of competitive districts or of majority-held

districts is liable to include some proportion that square with the expectation—i.e., the

central tendency—but involve differential packing, differential cracking, or both.

As example of the problems encountered when focusing on a central tendency

consider Chen and Rodden’s attempt to indicate a gerrymander by counting Bush or

McCain district wins across Florida, in both their academic and trial related work (Chen

and Rodden 2013a, 2013b, 2014). As noticed and noted by Darling (2013) and by

McCarty (McCarty 2013; 2014), a match or mismatch between expected and observed

number of districts carried is not a per se robust and structural feature of a districting

plan. The match or mismatch varies depending on the vote percentage won (this is a

problem similar to that encountered when using the efficiency gap). A packing

gerrymander that all but guarantees that a party win, say, 55 percent of the districts

whether it wins, say, 40, 50, or 60 percent of the vote—which is the type of result a

packing gerrymander can and often does produce—will sometimes match the expected

number of districts carried and other times will not, depending on a party’s system-wide

vote percentage. In different words, the contours of a districting plan interact with a

party’s system-wide level of support to produce more, equal, or fewer than expected

10

wins. As a consequence, the interaction produces variable readings of whether one and

the same plan is or is not a gerrymander.7

Using computer generated districts to form a null set holds promise. It removes

all but inadvertent partisan effects in its construction of a null set and thus supplies a

strong basis for probabilistic inferences about intentions. One problem it has to

overcome is making the computer processing more intuitive and transparent. Another

pressing matter is choosing a benchmark other than the expected number of competitive

districts or the number of district wins. The question is this: by what guidelines, by what

standard, can we know a partisan gerrymander has been chosen versus avoided? The

approach supplies a useful tool, but we need to figure out how to make it transparent and

how to use it effectively.

Seat-denominated Symmetry

A proposal for a seat-denominated symmetry approach goes back decades

(Gelman and King 1994). It has found favor among political scientists (e.g., Engstrom

2013; McGann at al. 2015; 2016). To some extent it has also found favor among

members of the Supreme Court when amici (King et al. 2005) proposed it for

consideration in LULAC v. Perry (2006; for a detailed discussion of the Justices’

positions taken on the approach see Grofman and King (2007, 1-6).

The approach, which Grofman and King refer to simply as a partisan symmetry

standard, relies on an equal opportunity notion of fairness. Within practical and

probabilistically knowable limits, each party is expected to win the same seat percentage

for the same vote percentage. If 55 percent of the vote results in Democrats winning 7 of

10 seats, 70 percent, then Republicans should win 70 percent of the seats when they win

55 percent of the vote. The partisan symmetry standard shares the same concern for

7 Darling analyzed his 5,000 map null set for nine pre-2012 statewide Florida elections in addition to the McCain-Obama presidential contest. For the McCain-Obama contest he found, as did Chen and Rodden, the expected number of McCain wins under the 2012 lines was 14, whereas the enacted districting plan had McCain wining 17—a result observed in less than one percent of the null set plans. However, his analysis of the nine other elections showed the actual versus expected wins either matched (three elections), differed by one in favor of Republicans (three elections), or differed by one or two in favor of Democrats (three elections).

11

asymmetry that violates majority rule as the vote-denominated approach, but it adds a

requisite symmetrical operation of the swing ratio. In the competitive range of two-party

vote splits, perhaps inside the 40 to 60 range, if Democrats win one more seat with an

additional three percent of the vote, then Republicans should be expected to add one seat

when their vote increases by three percentage points. Its attention to the swing ratio

bears a similarity to the wasted vote approach; however, it deviates by being agnostic

about the magnitude of the ratio, provided that the effect of the swing is symmetric.

The approach runs into three difficulties, two principled and one practical. One

principled difficulty is that the harm it identifies is not directly connected to voters.

Another is that when the harm exists it could be situational rather than fixed. The

practical problem is producing assurance that the situational harm is recognizable.

The simplest way to see both principled complications is by returning to the

examples used to point to a shortcoming of the vote-denominated approach. There we

had a five Democratic two-party percentage district vote distribution of 44, 46, 51, 52,

and 62. The median and mean are equal and therefore vote-denominated asymmetry as

a leading indicator of gerrymandering is missing. However, as discussed above, a three

point uniform shift in favor of the Republicans, moving the median and mean to 54,

leaves them with three district wins, while a three point swing in favor of Democrats

leads to four district wins. The different outcomes signal differential treatment

disadvantaging Republicans, but it is not clear that a vote weight difference is necessarily

implied. In both situations, majority rule is maintained. Thus we do not know based on

the majority rule marker that votes are unequally weighted. What we do know is that the

swing is not symmetrical. With Republicans at 52 through 57 percent of the vote they

win three districts, but with Democratic votes at 52 through 57 percent they win four

districts. Are those facts enough to say Republican votes carry less weight than

Democratic votes? Yes, perhaps, in that range of circumstances. If, however, the vote

swings a uniform 7 points in favor of Republicans, giving them 58 percent, they win 5

districts; but at 58 percent of the vote Democrats win only 4 districts. In that situation,

Republicans are advantaged and, thus, seemingly have a vote weight advantage. Again,

12

we do not know that votes count differently, because, at all levels of the vote, majority

rule is preserved. What we do know is that in some situations Democrats are

advantaged, and in other situations Republicans are advantaged. The differential

treatment is situational.

Because we cannot say with assurance that votes are weighted unequally in the

scenarios presented, given there is no violation of majority rule at any vote level, the

most we can say is that the votes count differently. How differently? That depends on

nature and magnitude of swing in each district. Therein resides the practical problem. If

the swing is uniform and a 3 point uniform swing toward Democrats is more likely than a

5 point uniform swing toward the Republicans, then Republicans do not have the same

opportunity as Democrats to win four seats. If the swing is non-uniform—i.e., mixed in

the sense that some districts swing more than others—we need to know more, much

more. Getting an assured handle on what else we need to know was the apparent

stopping point for Justice Kennedy when he remarked favorably on a seat-denominated

symmetry approach but said courts are “wary of adopting a constitutional standard that

invalidates a map based on unfair results that would occur in a hypothetical states of

affairs” (LULAC v Perry 2006, 420).

The partisan symmetry standard is more comprehensive than the equal vote

standard. It takes account of both packing and cracking effects. To do so, however, it

can overreach in principle by locating the harm as unequal treatment of parties whether

or not we can know that voters suffer harm. It can also overreach in practice by

requiring a supporting analysis that makes some decision makers wary of enforcing it.

Three-prong approach

Because an effective gerrymander can be achieved through so-called packing,

cracking, and their combination, it would seem a good idea to use multiple criteria to

identify them. That is the apparent thought standing behind Samuel Wang’s proposed

three-prong test (Wang 2015a). In an amicus brief to the Supreme Court in Harris v.

Arizona Redistricting Commission (2016), Wang proposed adopting his ideas as a basis

13

for ruling that the Commission’s actions do not amount to a gerrymander (Wang

2015b).8

Wang’s three prongs are grounded in concerns for (a) a less than justifiable

degree of seat-vote proportionality, (b) under-responsiveness of seat shifts from vote

shifts, and (c) asymmetry in the vote distribution.

1. Proportionality: Seat to vote responsiveness is within a range between

proportionality and what could be expected from the seat-vote relationship in

other states (plus allowance for random variation).

2. Under-responsive seat shifts to vote shifts: This is indicated by

unequal average lopsidedness in the vote distribution, unequal average values

of each party’s winning margin above 50 (plus allowance for random

variation).

3. Symmetry: A party’s median district percentage equals its mean district

percentage (plus allowance for random variation).

Giving consideration to three prongs has the appearance of a more

comprehensive set of concerns than the preceding four approaches. That much can be

granted, but having three prongs to deal with creates manageability problems. For

instance, Wang advises in one statement that the three prongs can be used “separately or

together” (Wang 2015a, 22), but he later advises that the degree of proportionality

(prong 1) might often be ignored except to serve as a substitute when the others are

impossible to apply (Wang 2015a, 46). Natural questions are these: is satisfying one of

the prongs enough to say no gerrymander exists; is violating one of the prongs enough to

say a gerrymander does exist?

8 His amicus proposal appears to have misapprehended the issue in Arizona. Appellants claimed that the Commission gerrymandered by under-populating districts favoring Democrats. If true, the bias introduced is a form of malapportionment/turnout bias, which requires a comparison that Wang did not take into account. That is, Wang proposed a median-mean comparison in order to indicate advantaged gained from under-populated districts, but to make the proper comparison for turnout bias the mean district vote percentage needs to be compared to the statewide vote percentage, not to the median district percentage (see McDonald and Best 2015, 317-8).

14

Managing the three prongs separately or together is also a fundamental concern

for questions of effectiveness. The different prongs can provide indications running in

opposite directions. For example, a five district distribution of 40, 40, 60, 60, 60,

satisfies both proportionality (prong 1) and equal average lopsidedness (prong 2) but

fails the symmetry standard of prong 3 (median 60 and mean = 52). A swing ratio could

reside within the bounds of acceptable proportionality but fail on both lopsidedness and

symmetry. And, a districting plan could fail the lopsidedness test due to incumbency

effects, if one party has more incumbents running, even in the absence of

gerrymandering. Who decides how many prongs to apply? Requiring failure on all three

prongs simultaneously leaves an opportunity for the original mapmakers to satisfy any

one prong of their choosing while enacting a gerrymander that would be indicated by

either or both of the other two prongs. In all, and in other words, the three prongs do

not work together as a coherent framework.

Evaluating gerrymanders through three different tests has an intuitive feel.

However, it raises difficult questions of both manageability and effectiveness because, as

it stands, no compelling coordinating principle supplies clarity about whether a

gerrymander exists.

What Can Be Done?

Our review indicates that the key to identifying and combatting partisan

gerrymanders requires putting symmetry at the forefront. This is in part a result of a

process of elimination and in part an implication of symmetry squaring with the

definition of gerrymandering’s harm. The efficiency gap and the three-prong approach

are nonstarters. The efficiency gap has to be eliminated as a viable approach because its

core concept of wasted votes may relate to gerrymandering in some circumstances but

clearly is not an indicator of any structural feature of a district plan having to do directly

with gerrymandering. The Wang three-prong approach is ambitious but lacks a coherent

structure upon which one could sustain a conclusion that a district plan is or is not a

gerrymander. The null set approach is a starter. However, its black box nature has to be

15

made transparent. Also, its focus has to be on something other than the number of

districts wins outside typical expectations because expected wins are a variable

consequence matching and mismatching actual wins depending on a party’s level of

support.

It is not just that the symmetry approaches are preferred by default. A focus on

symmetry stays true to the harm element in gerrymandering’s definition. It identifies

favorableness to one party by diluting votes for the other party. One can’t help but

notice, however, the two symmetry tandards do so as a duality. Vote-denominated

symmetry emphasizes vote dilution; seat-denominated symmetry emphasizes party

favoritism. It is preferred to join the two, allowing their union to speak to both

definitional elements. Unfortunately, their union by simple means poses a problem. If

the notion of harm is to be read as a conjoined concept, wherein un-favorableness to a

party is necessarily the consequence of vote dilution, seat denominated symmetry has

nothing to add to vote-denominated symmetry. Seat-denominated symmetry identifies

vote dilution as the source of un-favorableness if and only if vote-denominated

symmetry has already reached the same conclusion.

We consider this unfortunate because it disregards the widely held sense that

gerrymandering tilts the partisan playing field even if it does not do so solely by diluting

votes. In what sense does the tilted field have a connection to voters? The answer in an

extreme circumstance is that a gerrymander can silence minority voices by creating

super-majoritarian results,9 such as when a congressional delegation is composed of

members of one political party across an entire decade. In more limited circumstances,

manipulation of the swing ratio, which is an important concern for seat-denominated

symmetry, can distort the voice of the people to the liking of the government charged

with drawing the districts. As we see it, therefore, it is not that harm to a political party

9 This could be relevant to situations when a congressional delegation is composed of members of one political party across an entire decade. The outcomes in Massachusetts or Nebraska between 2002 and 2012 could be instances of super-majoritarian cracking; no Republican won election to the U.S. from Massachusetts and no Democrat won election from Nebraska during that time. Of course, detailed checks are required before being able to say whether gerrymandering or something else stands behind these outcomes.

16

as such is outside the harm produced by gerrymandering. Rather, the harms are

twofold: vote-denominated symmetry is an indicator of violations of equal protection;

seat denominated symmetry is an indicator of violations of silencing or distorting

speech, speech that is a keystone element of the democratic process.

Conclusion

The task we set for ourselves here has been to engage in a diagnostic for the

purpose of gaining insight as to why it has proved so difficult to establish a standard for

detecting gerrymanders. On that score, if only on that score, we have made some

headway. The efficiency gap, a null set focused on observed versus expected district

wins, and Wang’s three-prong approach can be set to the side as unworkable. Symmetry

stands up as a plausible and useful standard, but the duality embedded in the concept

requires a redefinition of the duality of the harms that gerrymander can produce.

The missing element in the equal vote weight standard (vote-denominated

symmetry) is taking account of the manipulated swing ratios. The missing element in

the partisan symmetry standard (seat-denominated symmetry) is an assurance that votes

have been diluted. To move beyond diagnosis to prescription is a long step that we are

not yet prepared to take. Our proposal for combining the duality of the symmetry

standards is to insist on two requirements to hold safe a plan from the charge that it is

likely to produce one or another of gerrymandering’s two harms: plausible evidence that

(1) majority rule is unlikely and unnecessarily persistently violated and (2) a plan does

not have an anomalous swing ratio.

17

References

Altman, Micah & Michael P. McDonald. 2011. “BARD: Better Automated Redistricting.”

Journal of Statistical Software 42: 2-36.

Brady, David W. 1988. Critical Elections and Congressional Policy Making. Stanford,

CA: Stanford University Press.

Chen, Jowei and Jonathan Rodden. 2013a. “Unintentional Gerrymandering: Political

Geography and Electoral Bias in Legislatures.” Quarterly Journal of Political

Science 8: 239-69.

Chen, Jowei and Jonathan Rodden. 2013b. “Report on Computer Simulation of Florida

Congressional Districting Plans” (February 15). Available at Justin Levitt’s All

about Redistricting website.

Chen, Jowei and Jonathan Rodden. 2014. “Response to Expert report of Stephen W.

Hodge” (February 18). Available at Justin Levitt’s All about Redistricting

website.

Cirincione, Carmen, Thomas A. Darling, and Timothy O’Rourke. 2000. “Assessing

South Carolina’s 1990 Congressional Districting.” Political Geography 19: 189-

211.

Darling, Thomas A. 2013. “Report of Thomas A. Darling in Romo v. Detzner” (April 8).

Available at Justin Levitt’s All about Redistricting website.

Davis v. Bandemer. 1986. 478 U.S. 109.

Engstrom, Erik J. 2013. Partisan Gerrymandering and the Construction of American

Democracy. Ann Arbor, MI: University of Michigan Press.

Engstrom, Richard L. and John K. Wildgen. 1977. “Pruning Thorns from the Thicket:

An Empirical Test of the Existence of Racial Gerrymandering.” Legislative

Studies Quarterly 2:465-79.

Gelman, Andrew and Gary King. 1994a. “A Unified Method of Evaluating Electoral

Systems and Redistricting Plans.” American Journal of Political Science 38: 514-

54.

Grofman, Bernard N. 1983. Measures of Bias and Proportionality in Seats-Votes

Relationships. Political Methodology, 9:295-327.

18

Grofman, Bernard N. and Gary King. 2008. “The Future of Partisan Symmetry as a

Judicial Test for Partisan Gerrymandering after LULAC v. Perry.” Election Law

Journal 6: 2-35.

Gudgin, Graham and Peter J. Taylor. 1979. Seat, Votes, and the Spatial Organisation of

Elections. London, UK: Pion.

Hacker, Andrew. 1964. Congressional Districting: The Issue of Equal Representation

(revised ed.). Washington, DC: Brookings.

Harris v. Arizona Independent Redistricting Commission. 2016. U.S. Supreme Court

(slip opinion).

Hebert, J. Gerald and Danielle Lang. 2015. “Memorandum of Amici Curiae Common

Cause and Virginia New Majority Regarding Proposed Remedial Plans.” U.S.

District Court for the Eastern District of Virginia (3:13-cv-678).

Hodge, Stephen W. 2013. “Assessments of the Redistricting Simulations of Professors

Chen and Rodden” (November). Available at Justin Levitt’s All about

Redistricting website.

Hill v. Detzner. 2015. No. 3:15-cv-00380 (N.D. Fla.).

King, Gary, Bernard Grofman, Andrew Gelman, and Jonathan Katz. 2005. “Amicus

Brief in Jackson v. Perry Submitted on Behalf of Neither Party, in the Supreme

Court (No. 05-276).

LULAC v. Perry. 2006. 548 U.S. 399.

McCarty, Nolan. 2013. “Response to Chen and Rodden’s ‘Report on Computer

Simulation of Florida Congressional Districting Plans’” (No data stated).

Available at Justin Levitt’s All about Redistricting website.

McCarty, Nolan. 2014. “Response to Jowei Chen and Jonathan Rodden’s February,

2014 Report” (April 28). Available at Justin Levitt’s All about Redistricting

website.

McDonald, Michael D. and Richard L. Engstrom. 1989. "Detecting Gerrymanders." In

Bernard Grofman (ed.) Toward Fair and Effective Representation: Political

Gerrymandering and the Courts. New York, NY: Agathon, 178-202.

McDonald, Michael D. and Robin E. Best. 2015. “Unfair Partisan Gerrymandering in

Politics and Law: A Diagnostic Applied to Six Cases.” Election Law Journal 14:

312-30.

19

McGann, Anthony J., Charles Antony Smith, Michael Latner, and Alex J. Keena. 2015. “A

Discernable and Manageable Standard for Partisan Gerrymandering.” Election

Law Journal 14: 295-311.

McGann, Anthony J., Charles Antony Smith, Michael Latner, and Alex J. Keena. 2016.

Gerrymandering in America: The House of representative, the Supreme Court

and the Future of Popular Sovereignty. New York: Cambridge University Press.

McGhee, Erin. 2014. “Measuring Partisan Bias in Single-Member District Electoral

Systems.” Legislative Studies Quarterly 55: 68–69.

Nagel, Stuart. 1965. Simplified Bipartisan Computer Redistricting.” Stanfor Law

Review 17: 863-69.

Page v. Virginia State Board of Elections. 2014. U.S. District Court for the Eastern

District of Virginia (3:13-cv-678).

Romo v. Detzner 2012. Final Judgement, 2nd Judicial Circuit Court (CA-412).

Stephanopoulos, Nicholas and Eric McGhee. 2015. “Partisan Gerrymandering and the

Efficiency Gap.” University of Chicago Law Review 82: 831-900.

Tufte, Edward R. 1973. “The Relationship between Seats and Votes in Two-Party

Systems.” American Political Science Review 67: 540-54.

Vickery, William. 1961. “On the Prevention of Gerrymandering.” Political Science

Quarterly. 76; 105-10.

Wang, Samuel S.-H. 2015a. “A Three-prong Standard for Practical Evaluation of

Partisan Gerrymandering” (December version). Available at SSRN, id=2671607.

Wang, Samuel S.-H. 2015a. “Brief of Amicus Curiae Samuel S. Wang, Ph.D. in Support

of Appellees.” U.S. Supreme Court, Harris v. Arizona Independent Redistricting

Commission (14-232).

Whitford v. Nichol. Case: 3:15-cv-00421. (Wisc. E.D. 2015).

Veith v. Jubelirer. 2004. 541 U. S. 267.

20