Competition and Redistricting
in California:
Lessons for Reform
An IGS Study Funded by
The James Irvine Foundation
Bruce E. Cain
Karin Mac Donald
Iris Hui
with the assistance of
Nicole Boyle
Anita Lee
Alex Woods
INSTITUTE OF
GOVERNMENTAL
STUDIES
UNIVERSITY OF CALIFORNIA AT BERKELEY
Competition and Redistricting in California:
Lessons for Reform
A Study Funded By the James Irvine Foundation
by
Bruce E. Cain
Karin Mac Donald
Iris Hui
With the assistance of
Nicole Boyle
Anita Lee
Alex Woods
Institute of Governmental Studies
University of California, Berkeley
February 2006
1
Introduction
The defeat of California's Proposition 77 marks a new phase in the redistricting reform debate.
The fact that this specific measure failed, however, does not imply that the prospects for change
are dead. Proposition 77 was perceived as flawed in many specific ways that can be remedied in
future proposals. The purpose of this report is to look at the function of redistricting criteria— in
particular, political competition— and to derive some lessons that might instruct any future
attempts to amend the line- drawing process in California.
There is nothing straightforward or simple about redistricting. Indeed, the process has become
more difficult over time as court decisions and new statutes have incrementally added criteria to
the initial “ one person, one vote” requirement. Redistricting now requires line- drawers to
incorporate what we will refer to as primary and secondary levels of criteria. Primarily,
redistricting must equalize the populations in contiguous districts, and comply with the Voting
Rights Act. These are federally mandated rules that cannot be over- ridden by secondary criteria.
The next level of criteria includes rules about compactness, communities of interest and city and
county boundaries, nesting and the like that are established by state law or state constitutions.
Federal and state court decisions provide explicit guidance in the interpretation of some of these
criteria; for example, the equal population criterion for congressional districts has been
interpreted to mean that districts can only differ by a few people, far less than 1 percent. Other
criteria are more vague and largely left to the line- drawers’ discretion.
Over the past few years, the discussion over redistricting principles has focused on a new
criterion: competitiveness. The purpose of this project was to discover how many potentially
competitive seats could be constructed hypothetically, and then how implementing other criteria
affected that number. Given the nature of redistricting law and public expectations, it is not
sufficient to simply know how many potentially competitive seats can be drawn. It is also
important to recognize the cost of creating competitive seats in terms of other goals such as
fairness to racial and ethnic minorities, observing communities of interest, keeping districts
compact, and the like.
The following pages report the results of a study funded by The James Irvine Foundation and
conducted by the Institute of Governmental Studies at UC Berkeley. In the sections that follow,
we:
1. describe the methods of the study;
2. discuss the results;
3. make some recommendations based on what we found.
Our basic conclusion is that the ability to achieve a high level of potentially competitive seats is
greatly limited by other redistricting criteria, the uneven political demography of the state and
the advantages of money, name recognition, and staff resources that incumbents enjoy in the
state legislature and Congress. We recommend against any specific attempt to define
competitiveness or to specify a given number of competitive seats in any proposed new
redistricting law. Instead, we could recommend that if any language about competitiveness is
considered for inclusion in a new law, that it be very general. Because there are so many
2
different perspectives in this state about fairness and what matters in redistricting, any proposed
line- drawing process should have guarantees for the public submissions of proposals, open
meetings and a diverse membership.
Method and Research Design
For this study, we used a team of graduate and undergraduate students and one Geographic
Information Systems ( GIS) Specialist as our technical line- drawing team. Two of the
undergraduate students had no previous training in GIS, and only one team member had
redistricting experience. This team drew statewide plans for California’s Congressional and
Assembly districts using specific sets of criteria. Some plans began with the existing majority
minority districts, lines, i. e. the status quo; other plans were drawn ‘ free- hand.’ Our team
adhered to strict population equality, drawing contiguous districts that were at minimum as
compact as the status quo, and in most cases more compact. We then varied three other criteria:
maximizing the number of majority minority districts, minimizing the number of county and/ or
city splits and maximizing the number of competitive seats. Altogether we drew over 30
statewide Congressional plans and 21 Assembly plans. 1 In the process, we considered multiple
definitions of competitiveness and majority minority districts. We did not use incumbent
addresses or take geopolitical bases into consideration. We report 2000 registration figures to
make it easier to compare our plans to the current districts as those are the same data the state
used in its last redistricting.
It is important to note that an infinite number of plans can be drawn under even the simplest
criteria. Our study does not attempt to provide one answer to a question that has many. We also
did not fine- tune our plans to the degree necessary to submit them to the Legislature. For
instance, in some of the plans we did not clean up all the small Census place splits. Census
places often are non- contiguous and cleaning up a plan can add many hours to a line drawing
exercise. We kept the population deviation under 1% but made no attempt to drive it down to one
person. Rather, our plans were intended as heuristic devices, illustrating some key points about
the trade- offs inherent in redistricting and the likely political effects of new districts. It should
be noted that we did not use the Community of Interest criterion in our exercise because it is
difficult to impossible to implement without public testimony. We included the drawing of
‘ square box’- type plans as one of our experiments, to simulate the kind of automated, stripped
down redistricting process ( compact, equally populated and devoid of potential human/ political
interference) that some people have argued for over the years.
Our basic findings for Congressional lines are as follows:
1. Plans that balance all the criteria ( population equality, contiguity,
compactness, minimizing county splits, preserving the VRA seats and enhancing
1 In addition to these, we also developed more than twenty other plans to examine other hypotheses, such as how the
criteria specified in Prop 77 would affect the redistricting process. The Assembly plans developed were also utilized
to assess different ways of nesting two Assembly Districts in one Senate district. Results on nesting will appear in a
supplemental report. For this report, however, we concentrated on the fifty- two plans developed to examine trade-offs
among constraints.
3
competitiveness) would create between 12 – 14 Congressional seats ( 13 on
average) in the range between a 3 percent Republican registration advantage and a
10 point Democratic registration advantage2.
2. Districts in that range will be contested more heavily but small registration
margins do not necessarily predict turnover since other factors matter significantly
such as incumbency, money advantages, national tides and candidate quality. In
the redistricting plan drawn by the Court in 1991, only 14 of the 260 California
Congressional races ( i. e. 5%) between 1992 and 2000 resulted in party turnover.
3. Plans that maximized competitiveness and ignored city/ county lines and
the integrity of the VRA districts create on average as many as 18 to 25 districts
in the potentially competitive range, but they would be subject to serious legal
challenges and much controversy in the affected local communities.
4. Political geography and the VRA give the Democrats a big edge in safe seats over
the Republicans. No plan, no matter who draws it, can change that. Barring a
heavily biased Republican plan, the Democrats are unlikely to drop below 26
seats in Congress and the Republicans could fall to 14.
As for the State Assembly, we found that:
5. Out of eighty Assembly districts, plans that aim to maximize the number of
potentially competitive seats could produce between 21 to 30 seats in the 3 point
Republican and 10 point Democratic registration range.
6. Among plans that balance all other redistricting criteria, between 12 to 17 seats
( 15 on average) would fall in that range.
7. Similar to the Congressional races, due to incumbency advantage and other
factors, a slim party registration difference does not necessarily translate into a
narrow vote margin. Among the 400 Assembly races that took place between
1992 and 2000, only 22 ( 6%) resulted in party turnover. Ten of these races ( 45%)
occurred in districts with a party registration difference in the 3 point Republican
and 10 point Democratic range. Contrary to conventional expectation, none of
these party turnover races happened in districts with a party registration difference
within 3 percentage points. In fact, several Republican candidates were able to
win in districts with high concentration of Democratic voters.
We considered most of the commonly used redistricting criteria and conducted a series of
experiments by observing or relaxing some of the constraints. With over fifty plans developed,
we came to the following conclusion about trade- offs among those criteria.
2 We use this range because an evaluation of partisan races during the 1990s shows that inside the 0- 3% Republican
to 0- 10% Democratic advantage range, seats have the highest likelihood to actually turn over. In fact, only two
Congressional seats that switched party control did not fall into that range, and they were products of extraordinary
circumstances.
4
8. The conventional belief that majority minority districts tend to be non- competitive
and dominated by the Democratic Party still holds. Yet with changing
demographic composition and partisan alignment, it is now feasible to draw one
or two majority minority districts that might be potentially competitive.
9. Plans that placed a heavy emphasis on compactness and minimizing city/ county
splits made it hard to achieve the political goals of more competitive seats and
preserving majority minority representation.
A Review of Redistricting Criteria
Any single goal in a redistricting will be highly constrained by the other criteria that must be
followed to achieve a legal redistricting plan. Meeting one criterion will ultimately lead to a
trade- off with another criterion. Below, we briefly explain the most commonly used redistricting
criteria.
1. Equal Population
Making district populations equal is the rationale for changing district lines at the beginning of
each decade. Districts are supposed to be balanced in population to ensure that the one person -
one vote principle will not be violated. For example, if one district had 100 residents and one had
1000, the residents of the first district would essentially have ' more' representation than those of
the second. Thus a vote in the first district would be valued differently than in the second.
The equal population criterion has become more and more narrowly defined over the past
decades. Under current case law, congressional districts are held to ' strict scrutiny' meaning that
they can not deviate from the ideal population3 by more than a few people. For legislative
districts, this criterion is not as narrowly interpreted. Most experts will advise keeping deviations
below 10 percent, preferably much below that, to avoid claims of malapportionment. Big cities
or counties are often split in order to adhere to the equal population requirement. San Francisco
County had a total population over 776,000 according to the 2000 Census. Yet the ideal
population for a congressional district is only 639,088. Figure 1 shows that the County must be
split into two Congressional Districts, Districts 8 and 12.
3 The ideal population is computed by dividing the total population of the State by the number of districts.
5
Figure 1. San Francisco County and Congressional Districts 8 and 12.
2. The Voting Rights Act
California is ‘ covered’ under Section 5 of the Voting Rights Act ( VRA). This means that a
redistricting plan must be precleared by the Department of Justice ( DOJ) before it can go into
effect. The DOJ will evaluate plans for retrogression, i. e. they make sure that minority
populations in 4 counties4, and in districts that are part of those counties, are not weakened in
their potential political power under the new district lines. Line- drawers are severely limited in
their creativity in those counties. Because these counties and their districts are necessarily part of
the congressional plan, the entire State plan is affected and has to be precleared. A redistricting
also must not run afoul of Section 2 of the VRA. The simplest explanation of the Section 2 non-dilution
standard is that in racially polarized areas in which minority groups constitute a majority
in a district, groups should not be split up but rather kept whole.
4 The counties are Kings, Merced, Monterey and Yuba
6
3. Contiguity
Contiguity is the most basic of all redistricting criteria, but even it has had its challenges. Some
districts are contiguous because they are connected by a bridge. As illustrated in figure 2, the
current Congressional District 7 spreads across Solano and Contra Costa county and is connected
via the Carquinez and Benicia Bridges. The general rule of thumb is that districts have to be
connected in some way, and the more connected they are, the less controversial this criterion is.
Figure 2. Congressional District 7
4. Compactness
Compactness, on the other hand, has been interpreted in many different ways. There are
currently at least 7 different compactness measures that are commonly used and that are part of
the redistricting software we used for this study. 5 When our line- drawers were instructed to
draw compact districts, they would, in absence of any compactness measure, attempt to draw
box- like districts that did not have too many edges or ‘ fingers.’ Figure 3 shows some illustrations
of box- like districts we drew in Los Angeles County, which are more compact than, for
example, the District in the above diagram.
5 Maptitude for Redistricting 4.7, by Caliper Corporation.
7
Figure 3. Example of Compact, Box- like Districts
5. Respect for City and County Boundaries
This criterion seeks to minimize the number of times that district boundaries split local
jurisdictions. A notable point is that many cities are actually not contiguous. Figure 4 and 5
highlight the city boundaries of Bakersfield and Fresno ( shaded in green). There are often
outlying areas that redistricters need to pick up to keep a city whole. City boundaries are often
not very compact. California’s counties, while more compact, are in many cases too large to be
contained in one district. Some cities are equally subject to mandatory splitting to achieve
equally populated districts. For the purpose of this study, as is commonly done in redistricting,
we use Census places to mean cities. Census place designations consist of cities and
unincorporated areas. There are 1081 Census places in California.
8
Figure 4. Example 1 of non- contiguous, non- compact city boundaries: Bakersfield
Figure 5. Example 2 of non- contiguous, non- compact city boundaries: Fresno
9
6. Communities of Interest
This criterion is the most vaguely defined, and the one that is often most important in its
application when a decision must be made about where to split a city or a county. At its highest
level of application, a community of interest could be a city or a county because of the common
interest of a respective jurisdiction. It could also be a region, such as the central valley or the
coastal communities. On the smallest level, a community of interest might be a neighborhood, a
redevelopment district or an area that encompasses a group of activists advocating for a common
goal. A Community of Interest is most often identified during the process of public hearings in
which testimony is provided and areas are defined. Groups like the Asian Pacific American
Legal Center and the Mexican American Legal Defense and Educational Fund held their own
workshops in communities throughout the state during the last redistricting process to collect
information under this criterion. Elected officials are also often helpful in providing information
about existing communities of interest in their districts. In the absence of current public
hearings, testimony and hence available data, this study did not have the benefit of being able to
utilize this criterion. Rather than introducing our own biases by including only some
Communities of Interest with which we were personally familiar, we decided to exclude this
criterion altogether.
Potential and Actual Competitiveness
Competition is the new buzz word when redistricting criteria are discussed. Whether districts
could indeed be drawn to be potentially competitive is a complex question that has as much to do
with the electoral geography at hand as with the definition of what it means to have a
competitive seat. One issue is clear, however: one would be much more successful in drawing
potentially competitive seats using political data in the process than one would be if political data
could not be used.
The discussion of competition has been grossly oversimplified and the answer of whether a
district is indeed competitive is highly nuanced. There are many different measures of
competition that have been used. Some evaluate a district based on the party registration of
voters, others look at election outcomes. How one assesses a district depends on how one looks
at it; for example, is a district competitive within a 3, 5 or 7 percent spread of registration?
Given that Democrats tend to have lower level of turnout, should Democratic registration be
weighted differently than Republican? And how does the increasing number of voters that
decline to state their party affiliation factor into the equation?
In this report, we focus on one measure of competition, the 0- 3% Republican to 0- 10%
Democratic advantage registration range, or simply referred to as the ‘ 3- 10’ range. Using party
registration data, we calculate the percentage of registered Democrats and Republicans by
dividing the number of registered Democrats ( or Republicans) by the total number of registered
voters in the district. Then we calculate the difference in party registration. For example, district
1 has 30% registered Republicans and 35% registered Democrats, the difference in party
registration is 5 percentage points ( 35%- 30%). In other words, the Democratic Party has a 5
percentage point party registration advantage in this district. If district 2 has 40% registered
Republicans and 38% registered Democrats, then the Republican Party enjoys a 2 percentage
10
point party registration lead. Therefore both districts fall into the 3- 10 percent range. Looking at
all the districts within a plan, we then count how many districts have a 0 to 10 percentage point
Democratic advantage and how many districts have a 0 to 3 percentage point Republican
advantage. We arrived at this ‘ 3- 10’ measure via analysis of the Congressional and State
Assembly races in the 1990s in California, which show that races within that range of
registration are most likely ( while still highly unlikely) to experience seat turnover6. We also
evaluated other measures of competitiveness which we found to be less predictive. ( See
Appendix I.)
While party registration is the most common measure by which the balance of partisans is
assessed, districts that look potentially competitive based on their registration figures do not
necessarily predict competitive races. Many factors determine the outcome of elections
including incumbency, which can add as much as a 5 to 7 point advantage, the amount of money
spent, the quality of the candidates, and the like. As a consequence, even seats with narrow
registration margins do not frequently change party hands. On the other hand, seats with
registration differences outside of what we have defined as the range of potentially competitive
seats occasionally experience party turnover.
During the nineties, a decade in which Congressional races were fought in districts drawn by the
court masters, there were 5 cycles of 52 races between 1992 and 2000 for a total of 260
Congressional contests. Of those, only 14 ( 5%) resulted in a change in party control from either
Democrat to Republican or vice versa. Of the 37 races with registration differences of three
points or less only 6 ( 16%) resulted in party change. In fact, 4 of the seats ( CD1: Hamburg-
Riggs- Thompson: CD15: Mineta- Campbell- Honda; CD36 Harman- Kuykendall- Harman; and
CD49 Schenk- Bilbray- Davis) accounted for 8 of the party changes. The other 6 seats only
changed once.
The other side of the coin is that seats that do not seem to be competitive on paper can
sometimes experience a party turnover. A good example of this is CD1 which never had a
Democratic registration advantage of less 13.5%. Yet, Dan Hamburg, a Democrat, lost to Frank
Riggs, Republican, in 1994 and the seat was held by the Republicans until 1998 when Democrat
Mike Thompson was elected. Here the factor was the division between the Democrats and the
Green party. A less dramatic example was Lynn Schenk’s victory in 1992 in a seat that was just
outside the 3% Republican range ( 42.8 Republican to 39.12 Democratic) in the so- called " year of
the woman."
Because Assembly districts are smaller than Congressional districts, candidates’ personalities
and political experience sometimes over- ride advantages in political affiliation. Party registration
difference becomes relatively less important in predicting the actual competitiveness of races.
Out of the 400 races contested between 1992 and 2000 ( 80 districts by 5 election cycles), only 22
( 6%) resulted in party switches. One would expect these party turnovers to have taken place in
districts with razor thin party registration difference. The reality was contrary to such
expectation. None of the turnover races occurred in districts in which the party registration
difference was less than 3 percentage points. Ten turnovers ( 45%) took place in districts with a
6 A seat turnover happens when the political party affiliation of the winner switches from one party to another in two
consecutive elections. The turnover can be from Democratic control to Republican or vice versa.
11
0- 3% Republican to 0- 10% Democratic advantage registration range. 7 Some Republican
candidates were able to win in districts with heavy concentrations of Democrats. For example,
Bruce McPherson, a moderate Republican, was first elected to the State Assembly District 27 in
left- leaning Santa Cruz in 1993. Another example was Brooks Firestone ( AD35). He won the
seat in 1994 where the Democratic Party had a 9 percentage point lead in registration and
received over 65% of the votes in his 2nd term.
Adding Competition — Congressional Level
The 2001 redistricting resulted in a bipartisan plan. This means that the parties compromised
and agreed to a fixed share of the seats. To ensure that the seat shares did not change, potentially
marginal districts were made safer. This was accomplished by concentrating Democratic voters
in districts held by Democrats, and Republican voters in districts held by Republicans, making
all previously marginal seats safer. Figure 6 shows the distribution of seats by party registration
margins. It clearly shows that the 2001 redistricting contained no seats in the range between the
0 to 3% Republican advantage and 0 to 10% Democratic advantage.
Figure 6: Distribution of Party Registration for current Congressional districts
7 AD24: Cunneen- R ( 1998) – Cohn- D ( 2000); AD25: Snyder- D ( 1992) – House- R ( 1994); AD35 O’Connell- D
( 1992) – Firestone- R ( 1994) -- Jackson- D ( 1998); AD43: Rogan- R ( 1994) – Wildman- D ( 1996); AD44: Hoge- R
( 1994) -- Scott- D ( 1996); AD54: Karnette- D ( 1992) – Kuykendall- R ( 1994) -- Lowenthal- D ( 1998); AD61: Aguiar-
R ( 1996) – Soto- D ( 1998); AD80 Bornstein- D ( 1992) -- Battin- R ( 1994).
12
What is the range of possibility with respect to seats in this potentially competitive range?
Assuming for illustrative purposes that it would be legal to conduct a minimal redistricting,
drawing only equally populated, compact districts, and ignoring all other federal and state
considerations, we developed five such plans. We averaged the results from these five plans to
examine how many seats would fall into one of the twelve ranges of party registration ( used in
figures 6 through 13). The average Democratic versus Republican registration differences are
displayed in figure 7. On average, these “ random box” plans put 13 seats in the potentially
competitive range. Another way of looking at this is that this random map- making created 40
safe Democratic and Republican seats: a stark reminder that California's political geography
accounts for a large portion of the non- competition in the state.
Figure 7: Random Box Plans - Distribution of Average Party Registration
13
At the opposite end of the spectrum, we created a set of plans that only maximized competition
but were subject to equal population and reasonable compactness. A map of this sort would also
not be legal, but it does give an idea of the potential upper bound on attempts to create more
competitiveness. This is displayed in figure 8. It shows that on average there were 20 seats in the
potentially competitive range. Still, even a plan that placed competitiveness above everything
else yielded 22 safe Democratic seats and 11 safe Republican ones.
Figure 8: Competitiveness Maximization Plans - Distribution of Average Party Registration
14
The final illustration of the number of seats that can be created in the potentially competitive
range is what we termed the ‘ fully balanced’ plan. It is important to keep in mind that even ' fully
balanced' in this study does not equate: having considered all applicable redistricting criteria.
For this study, we did not include the Community of Interest criterion, which can have an effect
on the outcome of any plan. For this exercise, we drew five plans that took into account equal
population, kept the existing number of majority minority districts, were reasonably compact,
minimized county splits and maximized the number of seats in the potentially competitive range.
Even with all of these constraints, we were able to create on average 13 districts in the 3 point
Republican to 10 point Democratic range.
Figure 9: Fully Balanced Plans - Distribution of Average Party Difference
15
Adding Competition — State Assembly Level
None of the Assembly districts adopted as part of the bipartisan plan of 2001 have party
registration differences ( based on 2000 registration figures) within 3 percentage points.
However, five out of eighty seats fall in the range between 3.1 and 10 point Democratic
registration advantage.
Figure 10: Existing State Assembly Districts - Distribution of Party Registration
This part of our study began with the drawing of compact, equally populated districts, without
the use of party registration data or consideration of other redistricting criteria. These districts
were drawn without any political consideration. We produced four of these ‘ random box’ plans.
The resulting districts fared better than the 2001 bipartisan plan in terms of the number of
potentially competitive seats. Figure 11 shows that, on average, 17 seats would fall into the 3
point Republican and 10 point Democratic registration range. The increase in the number of
potentially competitive seats was made possible by a reduction of safe seats from both parties.
Under the 2001 bipartisan plan, the Republican Party held 16 seats with at least a 10 percentage
point registration advantage, and Democrats had 46. In these random box plans, the number of
safe Republican and Democratic districts would be reduced to 13 and 40 respectively.
16
Figure 11: Random Box Plans - Distribution of Average Party Registration
The next set of plans was produced with the single goal of maximizing the total number of
potentially competitive seats. We used Census block level party registration data to locate
partisan clusters. The results gave us a good estimate on the upper bound one could achieve
without considering legal ramifications. In contrast to the bipartisan plan, 10 seats could be
added to the missing range within a 3 point registration difference. Another 16 seats could fall
between the 3.1 and 10 Democratic registration lead. In other words, 26 seats could be in the
potential toss- up range which might result in party turnover.
17
Figure 12: Competitiveness Maximization Plans - Distribution of Average Party Registration
The last set of plans produced were the ‘ fully balanced’ plans. Mappers first observed the
Federal redistricting criteria, i. e equal population, contiguity and the Voting Rights Act. Then
they attempted to draw compact and potentially competitive districts while minimizing county
and city splits. Referring to figure 13, these plans on average produced 15 seats in the 3 point
Republican and 10 point Democratic registration range.
18
Figure 13: Fully Balanced Plans - Distribution of Average Party Difference
Minority Representation and Potential Competitiveness
Majority- minority districts fulfill the descriptive representational needs of minority groups and
often act as springboards for ethnic minorities to launch a political career. As most minorities
tend to identify with the Democratic Party, majority minority seats are often viewed as districts
without any real electoral competition. However with the changing residential patterns and
partisan alignment of ethnic minorities in California, we noticed that there are areas where one
might be able to draw a majority- minority district with a close party registration difference.
Figure 14 shows an example in the Anaheim- Santa Ana area of Orange County. District 12
( shaded in green) has 59.5% Latinos and 1.4% African- Americans, 43% of the voters are
registered Democrats and 40% are registered Republicans. Figure 15 displays another possibility
in the Central Valley. District 32 is made up of the South- West part of Fresno county ( primarily
outside the city of Fresno), and corners of Kings and Tulare county. It has 54% Latinos with a
close match of registered Democrat and Republicans ( both at 43% of the registered voters). The
small party registration gap might intensify electoral competition, especially in an open- seat race.
19
Figure 14. Potentially Competitive Majority- Minority District in Orange County
20
Figure 15. Potentially Competitive Majority- Minority District in the Central Valley
Trade- offs and Constraints
Like other public policy decision making processes, the redistricting process begins with a set of
criteria and priorities that limit the possible outcomes. Although redistricting is becoming more
complex and legal constraints since 1962 are more severe, map drawers can still navigate within
the constraints and carve plans to suit their political or social agenda. In practice, redistricting
negotiations often occur behind closed doors. We know that map makers consider a gamut of
factors before coming up with a plan. But what precisely is the effect of each constraint on their
decision making? How does the combination of different criteria affect the final outcome?
What would happen if more/ fewer constraints were in place? To better understand the impact of
each redistricting criterion, we conducted a number of experiments in line- drawing where we
relaxed some constraints and applied others, and then examined how that would affect the map
one could produce.
The primary goal of these experiments is to understand the trade- offs among criteria and
priorities. As discussed in the above section, the equal population requirement often results in
splitting densely populated counties and cities. Because minority groups are sometimes
geographically dispersed, map drawers may need to cut across counties or cities to group ethnic
21
communities into one district. In these cases, such districts would be less compact. Besides, the
creation of majority- minority districts usually implies reduced ability to draw potentially
competitive seats, perhaps with the exceptions of the Santa Ana area and the outskirts of Fresno
County. In the following sections, we report how different variations of constraints affect
minority representation, political subdivisions, compactness and potential competitiveness.
All plans followed the equal population and contiguity requirements. Mappers were instructed to
draw districts as compact as possible. Additionally, they focused on three major redistricting
constraints which were ‘ switched on or off’ in the experiment. These three constraints were 1)
fulfilling the Voting Rights Act requirements and drawing majority- minority districts; 2)
preserving political subdivisions by minimizing county and city splits; 3) drawing potentially
competitive districts. Our experiment began with the ‘ random box’ plans. These random box
plans consisted of drawing contiguous equal population districts using only Census demographic
data ( i. e. without any political data). These plans were the closest scenario to using a computer
program to automatically draw districts. We ignored any VRA considerations, as well as city and
county boundaries. Next, we added political data and attempted to draw plans that maximized a
single constraint. The purpose was to estimate the ‘ upper bound’--- how far could we go if we
concentrated our efforts on that single dimension? What is the maximum number of majority-minority
or potentially competitive districts we could obtain within political/ geographical limits?
What is the least number of counties and Census places we need to split in order to derive equal
population districts? Then our experiment increased the complexity of map making by
considering two constraints at a time. Lastly, to mimic the actual redistricting process, we
developed the ‘ fully balanced’ plans which took all the criteria into consideration8.
There are infinite ways to divide the state population into equally populated districts. The
experience of our map drawers is illustrative of this point. We can not stress enough the extent of
variability among those drawing lines. No two line- drawers produce the same maps. Each
mapper has preconceived notions regarding what a compact district is, or which areas should be
put under the same district. There is also a learning curve --- the more one maps, the easier it
becomes to locate ethnic or partisan areas, or to find geographic units with a particular share of
the population. In order to obtain different perspectives, we assigned at least two mappers to
develop plans under each of the combination of constraints. No fewer than two plans ( five for the
‘ fully balanced’ plans) were drawn to gauge the range of possibilities. 9 The summary statistics
are displayed in table 1a ( for Congressional districts) and 2a ( for State Assembly districts), the
full results are reported in table 1b and 2b. 10 Please see Appendix II for all referenced tables.
8 Please note that ' fully balanced' here does not mean that ALL legally required redistricting criteria were taken into
consideration. ' Fully balanced' plans were only developed with the criteria used for this study, Communities of
Interest, for example, were not taken into consideration.
9 We began with Congressional districts. After developing over fifty Congressional plans, our mappers had become
very proficient in identifying minority and partisan areas. Instead of having each mapper develop more than one
Assembly plan under each combination of constraint, we assigned each plan to at least two mappers. Each mapper
produced one plan for comparison.
10 Table 1a and 2a display ONLY the averages from a number of plans. Readers should refer to the detail table, table
1b and 2b, to get a full range of possible values.
22
The first row in table 1a ( highlighted in gray) lists the major statistics for the 2001 bipartisan
Congressional plan which serves as a baseline for comparison. Table 1a comprises five major
sections. The first section lists the constraint( s) considered. An ‘ X’ in the column implies that the
constraint was applied while an empty space indicates the constraint was relaxed. The second
section shows the average number of Latino and Black seats these plans could produce. Under
‘ political subdivision’ are figures on the number of counties and Census places split. 11 For
compactness, there are many measures available. Experts in the field do not agree on one single
indicator. The redistricting software used for this study, Maptitude, includes seven measures of
compactness. 12 As these measures tend to agree with each other, for stylistic simplicity, we only
included the mean score and standard deviation of two measures, Roeck13 and Schwartzberg, 14
in the table. 15 Using the party registration data in 2000, the last section (‘ Potential
Competitiveness’) reports the number of potentially competitive seats where the party
registration difference between the two major parties fall within 7, 5, 3 percentage points. In
addition, we also count the number of seats in the 3 point Republican, 10 point Democratic
‘ possible toss- up’ range. 16
The random plans produced the most compact districts on average. The mean score for the
Roeck measure was 0.49, highest among all the plans developed. It was significantly higher than
the score for the 2001 bipartisan plan ( 0.33) which was generally perceived as an incumbent
protection plan. However, the high compactness score we achieved with these plans did come
with a cost. Without a conscious effort to boost minority representation, the number of Latinos
seats fell to 7, below the existing standard of 10. As shown in figures 4 and 5, county and city
boundaries never have smooth edges. Predictably, the pure pursuit of rectangular- or circular-shaped
districts resulted in more counties and Census place splits. Over 200 Census places were
split in this exercise, and 41 out of 58 counties were divided on average to give way to compact
districts.
The next three sets of plans gave us the ‘ what- if’ scenarios: what if we just pursue one goal
single- mindedly? In terms of minority representation, we would expect an 8 seat increase in the
number of districts with at least 50% Latinos ( totaling 18) over the existing Congressional plan.
As for potentially competitive seats, up to 20 seats could be added in the 3 point Republican, 10
point Democratic registration difference range. Not all of the seats in this potential toss- up range
11 Here we only report the number of counties or Census places that are split/ not split. More statistics ( such as the
number of time a district is split) are available upon request.
12 The seven measures are Roeck, Schwartzberg, Perimeter, Polsby- Popper, Population polygon, Population Circle,
Ehrenburg.
13 It is an area- based measure that compares each district to a circle, which is considered to be the most compact
shape possible. For each district, the Roeck computes the ratio of the area of the district to the area of the minimum
enclosing circle for the district. The range goes from 0 ( least compact) to 1 ( most compact).
14 Schwartzberg is a perimeter- based measure that compares a simplified version of each district ( excluding
complicated coastlines) to a circle. For each district, the test computes the perimeter ratio of the simplified version
of the district to the perimeter of a circle with the same area as the original district. The district is simplified by only
keeping those shape points where three or more areas in the base layer come together. Water features and a
neighboring state also count as base layer areas. This simplification procedure can result in a polygon that is
substantially smaller than the original district, which can yield a ratio less than 1 ( e. g. an island has a 0 ratio). A
score closer to 1 is more compact than a score further away from 1, i. e. a score of 0.8 is more compact than 1.5.
15 Statistics for other measures are available upon request.
16 Refer to the Appendix I for other measures of potential competitiveness.
23
would result in party turnover, but one might expect more heated electoral competition especially
in open races. The current 2001 plan did a superior job in preserving county and Census place
boundaries, given other redistricting constraints. Out of 58 counties and 1,081 Census places,
only 22 counties and 65 Census places were divided. Our plans that attempted to minimize splits
and ignored all other redistricting considerations beat the current Congressional plan by merely 2
counties and 17 Census places.
By contrasting the three sets of single constraint plans, we noticed a few intriguing results.
Among these sets, plans that maximize minority representation were the least compact, with a
mean Roeck score of 0.38. The reason behind this is that ethnic minorities in California have
become more residentially dispersed overtime. Mappers often had to reach out far to locate
pockets of ethnic communities. The implication is that if one’s goal is to enhance minority
representation, one would need to lower the compactness standard. Confirming this conventional
belief, minority representation is often increased at the expense of electoral competitiveness.
Moving from plans that maximize the number of majority- minority seat to plans that maximize
competitiveness would result in a hefty reduction of Latino seats ( from 18 to only 6 seats). Plans
that either boost minority representation or electoral competitiveness were made possible by
bisecting existing political boundaries. In a rather extreme case, boundary integrity of 40
counties and 204 Census places were sacrificed to create 20 potentially competitive seats in the
‘ toss- up’ range.
We next enhanced the complexities by considering two constraints simultaneously along with the
equal population, contiguity and compactness requirements. The additional constraint hindered
one’s ability to advance a single dimension. For example, in our single constraint plans, we
demonstrated that it was feasible to have 18 districts with at least 50% Latinos in the population.
When we attempted to preserve existing city and county boundaries, the number of Latino seats
dropped to 8.17 And when we included competitiveness, the number was reduced by 4 to 14. A
similar pattern applied to competitiveness. Our single constraint plans that maximize potential
competition had on average 20 seats in the likely toss- up range. Once we considered additional
constraints, this number slid to 11.18
We developed five fully balanced plans to mimic an actual redistricting process. Although our
plans were not professionally or legally polished enough to submit as real proposals, our
mappers did consider all given criteria in the drawing process. As a whole, these plans on
average could produce 12 Latino seats, two of these comprised at least 65% Latinos in the
population. According to the 2000 Census, Asians and African Americans made up 6.7% and
10.9% of the California population respectively. However, because Asians’ residential patterns
are more scattered geographically, it is much harder to create majority Asian than majority Black
districts. Our fully balanced plans had two districts where the percentage of African Americans
exceeded 30% of the population, but only one for Asians. Compared to the 2001 bipartisan plan,
17 Our plans did not fare as well as the 2001 Congressional plan in terms of county and Census places splits. We
suspect this may due to a learning effect. It took our mappers some time to juggle multiple criteria. Perhaps if these
plans were redrawn again, we might get more Latino seats with fewer county and city splits.
18 An intriguing surprise is that the plans that maximize minority representation while minimizing city and county
splits somehow ended up with more potentially competitive seats in the 3 point Republican, 10 point Democratic
registration difference range. This may also be due to learning effect. If more majority- minority districts were
created, the number of potentially competitive seats would decrease.
24
our fully balanced plans had more compact districts with far more seats in the potentially
competitive range. In other words, without protecting incumbents’ geopolitical interests, there
was more room for improvement in both compactness and potential competitiveness. Our plans
had slightly more county and city splits than the current 2001 bipartisan plan. Twenty three
counties and 92 Census places were split, as compared to 22 counties and 65 Census places for
the current Congressional plan. With more time and patience to fine- tune our maps, we believe it
is feasible to reduce these splits without compromising competitiveness or minority
representation.
Table 2a presents the summary statistics for our State Assembly plans. Again, this table averages
results from several plans within each variation of constraints. ( Refer to table 2b for the full
range of values.) Findings for the Assembly plans parallel those for the Congressional plans.
Given a single constraint, out of 80 Assembly districts, our plans produce 22 Latino seats or 26
seats in the 3 point Republican, 10 point Democratic registration difference range. Because
Assembly districts have a smaller ideal population, inevitably more counties and cities must be
divided to attain equal population. Our single constraint plans which attempted to minimize
dividing cities and counties fared slightly better than the current bipartisan plan. We managed to
split 16 fewer cities, at the expense of reducing minority representation. Dual constraints
prevented mappers from blatantly overemphasizing one particular dimension. Progressively,
when multiple redistricting criteria were considered at once, our fully balanced plans produced
22 Latino seats, 4 African- American seats and 3 Asian predominant districts. In regards to
competitiveness, we could expect to see 15 seats in the 3 point Republican, 10 point Democratic
registration range.
Discussion
The trade- offs of elevating one criterion over another, and the interplay and effects that multiple
criteria have on each other and consequently on the outcome of a redistricting plan has been
illustrated in the experiments above. In sum, here are some conclusions:
A strict application of the equal population criterion is the single biggest constraint on keeping
cities and counties from being split. The narrower this criterion is applied, the more severe its
effect will be on all other redistricting criteria, including those that we did not evaluate in this
study specifically, such as preserving communities of interest within district boundaries.
California is ' covered' under Section 5 of the Voting Rights Act. This means that districts that
are completely or partially part of 4 counties must be drawn such that protected minority
populations are not made worse off in terms of their opportunity to elect a candidate of their own
choice after the redistricting than before. Any redistricting must take these seats into
consideration in order to not violate federal law. No redistricting plan can go into effect until the
Department of Justice has verified that no ' retrogression' has taken place. In addition to those
seats, there are additional majority minority districts that are currently in effect and that
redistricters should either preserve or add to.
25
Voting Rights Act and majority minority districts are much more likely to be non- competitive
than districts that do not preserve high concentrations of minority populations within the same
district. They are also less likely to be very compact. In order to be in compliance with federal
law, any redistricting in California must allow for less compact and less competitive districts in
these areas. Most importantly, there is a clear inverse relationship between the number of seats
that could potentially be competitive and the number of majority minority seats.
Preserving city and county lines also places a real constraint on competitiveness. California's
political geography is such that Democrats predominate in many urban areas and Republicans in
suburban and rural areas. When city and county boundaries are kept intact, the consequence is a
baseline of non- competitiveness in most areas.
Critics of the current, legal, California plan have described its districts as ' ugly' or
gerrymandered. In its most basic definition, a gerrymander is a district plan that is designed for
either racial or political purposes. There is a tendency to decry every district that does not look
like a box as a gerrymander. But our study shows that just because a district is non- compact does
not mean it is a gerrymander. Redistricting criteria, especially the preservation of city and
county boundaries, can place severe constraints on compactness because the boundaries of those
jurisdictions are non- compact. Few cities in California are box- shaped. Furthermore, many
cities have outlying, non contiguous areas that have to be picked up to keep the respective cities
whole. Thus the canvass on which districts are constructed is already biased toward non-compactness
before one line has been drawn. Counties are, generally speaking, more compact
than cities, but because their populations are often larger than the districts', splitting them cannot
be avoided in many circumstances and they cannot be used as building blocks as readily as cities.
Minority populations in California do not always reside in compact neighborhoods. More often
than not, drawing legally required majority minority seats necessitates the drawing of non
compact districts.
Data Problems and Mid- Decade Redistricting
Redistricting usually begins as soon as the Bureau of the Census releases its first dataset after the
Census collection. The Census is conducted every ten years and data are released for California
roughly one year after the data collection. The Census Bureau releases the PL94- 171 dataset,
which is also known as the ' redistricting data' on the Census block level. A Census block in
urban areas roughly corresponds to a city block; it is larger in rural areas. Census blocks are
essentially the building blocks for electoral districts. Census blocks are the smallest geographic
unit on which data are reported. They can be aggregated to Census tracts, and to most larger
geographies that are essential in the redistricting process, such as cities and counties. However,
due to the strict population equality requirements, especially when drawing congressional seats,
the Census block as a reporting unit of data is most important.
Most states operate under fairly tight deadlines to complete redistricting. Thus, the process
begins as soon as the Census data are released. The deadlines ensure that districts are drawn as
close to the original date of data collection as possible, when the data are still ' fresh.' It is a
commonly accepted, unavoidable fact that Census data becomes more outdated as the decade
26
progresses, being much more accurate closer to the collection date. Thus, the population of the
electoral districts also shifts and what starts out as an apportioned plan, with districts equal in
population, becomes in most cases a plan that has to be adjusted after the next Census is
collected, when a new 100 percent head- count of the population is released.
The total population for California in the year 2000 as reported by the Census based on its 100
percent count of the population was 33,871,680. High mobility, developments and immigration
patterns among other variables all serve to outdate these data relatively quickly. Throughout the
decade, the Census Bureau and various other government agencies and departments release
estimates on the growth of the population for various jurisdictions like States or Counties. There
are no data sources available between Census data collections that systematically, statistically,
and reliably report population figures on a small unit of analysis, such as the Census block, or
even on the Census tract. Errors in population estimates vary, but they are larger on small
geographic units and tend to ' wash out' on larger units like on the State or County level.
By 2005, California's Department of Finance ( DOF) reported total population estimates of
36,810,35819 an increase of 2,938,678. The DOF in the same report provides estimates on the
city and county levels on how that growth is distributed. For example, the county of Riverside
grew by 3.8% between 2004 and 2005 while the county of Alameda only grew by 0.7%. The
smallest level estimates provided in a systematic way are by city.
If a redistricting were to begin in 2006, for instance, the first question would be: what is the ideal
population for each district? Currently available Census data are outdated but estimates show
that California’s population has increased by about 3 million people. Depending on the data
source the estimates of the population vary20, but we do know for certain that California has
grown substantially since 2000. What we do not know is the exact distribution of that growth.
Without the availability of systematically collected, recent data on a small geographic unit, it
would be impossible to draw equal population districts. Consider that we would not even know
whether the ideal population of districts be should assessed by dividing the current total
population estimate by the number of seats? Or should the 2000 Census population be used to do
this? If congressional districts are held to strict scrutiny under the equal population requirement,
how are line- drawers supposed to meet this criterion in the absence of current data? It seems that
malapportionment would be guaranteed. Furthermore, estimates of racial and ethnic populations
are known to have large error margins21, ( Black population estimates, for example, have a 10.3%
error in one report) which would make the drawing of Voting Rights seats ambiguous at best.
All of this discussion leaves out the obvious point that mid- decade redistricting opens the door to
political mischief. If the majority party has the ability to re- do redistricting whenever it was
19 State of California, Department of Finance, E- 1 City / County Population Estimates, with Annual Percent Change,
January 1, 2004 and 2005. Sacramento, California, May 2005.
20 For 2004, the Census estimates California's population at 35,893,799 ( Source: U. S. Census Bureau, 2004
Population Estimates, Census 2000) while California's DOF estimates it at 36,144,000. ( Source: California State
Department of Finance, Demographic Research Unit, E- 1: City/ County Population Estimates with Annual Percent
Change January 1, 2003 and 2004.)
21 Current Population Survey Basic Report, March 2004 Data: California; Appendix A: Standard Errors ( SE) and
Confidence Intervals ( CI) for Selected Measures; California State Department of Finance, Demographic Research
Unit.
27
politically advantageous to do so, the legitimacy of California’s political system would suffer.
Based on the data problems and the potential for abuse, we strongly urge that any future
redistricting proposal prohibit the option of mid- decade redistricting unless ordered by a court.
Who draws the lines - Mapper Effects and Time Lines
Our study did not set out to evaluate or address any potential effect it might have to move this
process from a large, diverse, group of elected representatives to a small group of ( most likely)
homogeneous, appointed special masters. We also did not attempt to evaluate our maps
qualitatively depending on how much time went into constructing them. During the five month
process of conducting our experiments, we did, however, have an opportunity to observe four
mappers from different backgrounds in the construction of a variety of maps.
First, it is easy to underestimate the time it takes to draw statewide plans even with the new
technology. We were furthermore surprised at the variation in learning curves and how
personality traits factored into how well districts were drawn. How many city and county splits
a map had turned out in some cases to be a direct factor of how much time a mapper spent on
that criterion and how much patience the mapper had. It also mattered whether the mapper was
given a baseline of how many city/ county splits the 2001 map had, and whether they were
instructed to do better or not. ' Cleaning up' a map to minimize city and county splits and come
up with the best possible scenario under the given criteria could add between 10 and 15 hours to
a plan. The same was true when an error was found in a map that mandated a change. The
process of incorporating any change in an existing plan added many hours to the final project.
At times, mappers would start over from scratch, explaining that this would be faster then to
incorporate a change.
When we began this study, we did not set out to measure biases that mappers might insert in the
process. Our mappers spent more time mapping areas that they were familiar with than areas
that they did not know. There was also a tendency to try and repeat the splitting of a district
along the same line, if that line was ' in the proper location' according to that mapper. For
example, one mapper has very strong feelings about where the city of Fresno should be split and
' mysteriously' her maps all split the city in the same place.
Even when our mappers did not know areas, they tended to develop biases in how districts
should be built. When we tried to minimize that bias by varying the starting point of a map, i. e
starting at the upper left corner and the next map with the same criterion from the lower right,
they would tend to make roughly the same decisions in, for example, uniting minority
populations for a VRA plan.
In sum we found that the quality of a plan very much depends on how much time is spent on
constructing it. Biases are also introduced depending on who the mappers were and which areas
they were familiar with. If an already existing plan receives public input and is then changed to
reflect that input, it will have a ripple effect through the entire plan and add a considerable
amount of time to the redrawing process.
28
Conclusion
In the current Congressional plan, there are no districts in the range between 3% Republican and
10% Democratic. Taking into account various constraints and trade- offs, it might be possible to
get about 13 districts back into the missing range. It would be best if the decision to modify
redistricting criteria and processes were made without the immediate prospect of an impending
election since both parties have something to gain or lose from a new round of redistricting. Had
the 2001 redistricting contained 13 seats in the potentially competitive range, it is possible that
the Democrats would have lost a few seats in 2002 and 2004. A redistricting in 2006 might have
cost the Republicans a few seats given prevailing national and historical trends ( i. e. mid- term
elections for second term Presidents). Knowledge of who wins and loses can easily undermine
the search for the best process. It is better to consider procedural changes behind the “ veil of
ignorance,” to borrow from philosopher John Rawls’ phrase. Otherwise it is all about making
changes for short term gain.
But it is also important to understand that redistricting is limited in its capacity to create a
heavily competitive state. Even plans that ignore constitutional and good government criteria for
the sake of maximizing competitiveness still leave well over half the state in safe seats. The
sources of electoral safety to a greater degree lie in our choices to live with like- minded people
and in socially homogenous areas. Moreover, even when districts are potentially competitive,
they do not become actually competitive unless there are good candidates with well- financed
campaigns. And even then, the number of seats that will turnover will likely be as low as when
the court masters drew the lines in 1991 ( i. e. 14 out of 260 races).
Some will say that even if turnover will never be high, competitiveness is a good in itself
because it will improve the behavior of elected officials. In a future report, we will examine the
widespread speculation that coming from a marginal seat creates moderation and a greater
willingness to negotiate among representatives.
Based on our study, we recommend the following:
1. If language about competitiveness is included in a redistricting law, the language should
only be general. Forecasting competitive seats is a tricky business and as our study
indicates, the effort to maximize the number of competitive seats can wreak havoc on
other criteria such as communities of interest and fairness to racial and ethnic minorities.
Moreover, the definition of competitiveness will likely change over time depending upon
the behavior of independents and the relative loyalty rates of partisans.
2. Given the diversity of criteria and the different perceptions that people will have about
what is important in redistricting, we believe that the redistricting body should be
diversely composed. Language in the proposal should urge that consideration be given to
geographic, gender and racial and ethnic balance to the degree possible.
3. Communities, groups and individuals should have the right to observe the line- drawing
process and to submit plans of their own. This ensures that all perspectives are put on the
29
table. As we noted in our study, line- drawers are inevitably influenced by their own
biases and habits. Confronting the ideas of those outside the process is the best way to
ensure that a broad number of options are considered.
4. The fairness of a various proposals cannot be considered without having political data
and openly assessing the implications of where the lines are placed. A provision that
denies the redistricting body the use of political data gives advantages to political
consultants and insiders whose business is to know precinct returns and groups that have
the resources to collect political data on their own. It is fairer to have this information in
the public realm for all to share.
30
Appendix I: Measuring Potential Competitiveness
We considered three measures to assess the potential competitiveness of a district; party
registration, previous statewide offices and presidential election outcomes. Party registration is
the most convenient measure. Individual- level party registration files are usually available
through local County Registrars. The Statewide Database, the non- partisan redistricting database
for the State of California, at the Institute of Governmental Studies ( UC Berkeley), provides
voter registration data on the Census block level for the entire State of California. Users can load
these data into their redistricting software and can immediately begin drawing. They can pre-define
‘ potential competitiveness;’ for example, a district is deemed ‘ potentially competitive’ if
the difference in party registration between the two major parties is within 7 ( or 5 or 3)
percentage points. 22 Users can aggregate registration data up to the district level and calculate
the percentage of registered Democrats and Republicans out of total number of registered voters.
They can take the difference between the percentage of registered Democrats and Republicans
and count how many districts fall into their pre- defined range. For our report, we chose the 3
point Republican, 10 point Democratic registration difference as our potentially competitive
range based on our analysis of the actual races in the 1990s. For the report on Prop. 77 published
by the Rose Institute in 2005, the authors extended their range to a 5 point Republican, 10 point
Democratic registration advantage. 23 Tables 3a and 3b compare our indicator with the Rose
Institute measure. Our fully balanced Congressional plans would on average have 13 seats in the
3 point Republican, 10 point Democratic registration range, 15 if we extended the range to 5
point Republican. As for our fully balanced Assembly plans, stretching the range would capture
17 seats instead of 15.
One caveat about using party registration data: party registration can change noticeably between
years. We drew our plans using 2000 party registration data ( as these were used in the 2001
round of redistricting). Except for a few outliers, almost all the Congressional plans experienced
reductions in the number of potentially competitive seats. For example, for our fully balanced
Congressional plans, the 2000 party registration data suggested we could expect 13 seats in the
3- 10 range. Yet the number dropped to 12 if we re- analyzed these districts using 2004 party
registration data. 24 This is partly due to a general realignment in partisanship in California. The
coastal areas remain liberal- leaning while the inlands have turned increasingly conservative. In
areas with new, rapidly growing settlements, such as the Central Valley and the Inland Empire,
partisan realignment seems to tip towards the Republican Party. In other words, even if mappers
22 Note that these cut- off points are arbitrary in nature. These party registration ranges capture seats that may have a
good chance of party turnover. Actual party turnovers can and do occur outside these ranges.
23 Johnson, Douglas, Elise Lampe, Justin Levitt, Andrew Lee. 2005. ‘ Restoring the Competitive Edge: California’s
Need for the Redistricting Reform and the Likely Impact on Proposition 77.’ The Rose Institute of State and Local
Government, Claremont McKenna College.
24 Results for Assembly plans were rather mixed. Of 21 Assembly plans, 12 experienced increases in the number of
potentially competitive seats in the 3 point Republican, 10 point Democratic registration range when 2004 instead of
2000 registration data were used. Ten plans experienced a reduction. The mixed results may be explained by the fact
that Assembly districts are smaller, and hence regional partisan swings can result in bigger fluctuation in party
registration. More qualitative analyses may explain why some areas showed bigger partisan swings than others.
31
intentionally created some potentially competitive districts there, any narrow registration
difference might be washed away as time goes by.
In addition to party registration, we constructed potential competitiveness measures based on
previous vote outcomes. We created ‘ normal vote’ measures by combining results for the six
statewide races25 in 1998 and 2002. As these statewide races tend to be less high- profile and
voters usually vote along party lines, the purpose of this measure is to estimate the underlying
partisanship of districts. By combining the 6 statewide races in two election cycles, we averaged
out the quality of the candidates, the differences in money raised, and other campaign related
factors. The pooled series also smoothed out fluctuations across time. Using this normal vote
measure, our balanced criteria plans produced on average 11 Congressional seats or 14 Assembly
seats in the margin between 3% Republican and 10% Democratic registration advantage.
Comparing registration differences with a normal vote score that combined statewide races
shows that registration constitutes a good part of office- holding destiny. Party registration and
normal vote are highly correlated. Our fully balanced Congressional plans had 14, 11 and 7 seats
within 7, 5 and 3 point registration margins. Using the normal vote measure, we got 14, 9, 7 seats
respectively in the 7, 5, and 3 point range.
The third set of measures of potential competitiveness was constructed by using actual
presidential election results in 2000. The 2000 presidential race between George W. Bush and
Vice President Al Gore was one of the closest races in recent history. Despite the fact that party
registration tends to overstate the actual vote margin, the Presidential vote in 2000 is generally
close to the party registration distribution. In our balanced Congressional plan, there were 11
districts with Bush v Gore margins of 7 or less, 9 with 5 or less and 6 with 3 or less as compared
to 14, 11 and 7 districts in terms of registration margins of 7, 5 and 3 points. Bush ran a little
behind but voting seems to have followed party registration fairly well. We observed a similar
pattern in our fully balanced Assembly plans. Based on 2000 party registration data, 18, 13 and
7 seats were in the 7, 5 and 3 registration margins, contrasted to 12, 8 and 6 seats in the 7, 5 and
3 presidential vote margins. In sum, by comparing the registration margins with the normal vote
measure ( i. e. the average margin of the statewide races below the Governor) and the Presidential
vote in 2000, one can conclude that party registration is generally a good on- average predictor of
vote margin.
25 The six races are Lieutenant Governor, Secretary of State, Attorney General, Controller, Treasurer and Insurance
Commissioner.
32
Page 1
Table 1a. Summary Statistics for Congressional Plans
A. Minority Representation B. Political Subdivision C. Compactness D. Potential Competitiveness
Plan Type
Draw
Competitive
Seat
Respect
County/
City
Boundary
Draw
Majority-
Minority
Seat
No. of Seat
( 50%+
Hispanics)
No of Seat
( 65%+
Hispanics)
No. of
Seat
( 30%+
Blacks)
No. of
Seat
( 30%+
Asians)
No. of
County Not
Split
No. of
County
Split
No. of
Census
Places Not
Split
No. of
Census
Place Split
Roeck
Mean
Roeck
Standard
Deviation
Schwartz-berg
Mean
Schwartz-berg
Standard
Deviation
No. of Seat
( Party
Registration
Diff within 7
pct pt)
No. of Seat
( Party
Registration
Diff within 5
pct pt)
No. of Seat
( Party
Registration
Diff within 3
pct pt)
No. of Seat
( Party
Registration
Diff 3 pt
Rep, 10 pt
Dem)
2001 Congressional 10 4 2 0 36 22 1016 65 0.33 0.10 2.31 0.61 3 0 0 0
Random Box 7 2 1 1 17 41 869 212 0.49 0.12 1.57 0.23 14 11 7 13
Single Constraint
Max MM X 18 1 2 1 30 28 875 206 0.38 0.12 1.91 0.37 12 9 4 9
Min Split X 7 2 1 2 38 20 1033 48 0.41 0.12 1.71 0.30 14 10 5 11
Max Competition X 6 2 1 1 18 40 877 204 0.45 0.12 1.67 0.27 24 17 11 20
Double Constraints
Min Split + Max MM X X 8 2 1 1 34 24 912 169 0.41 0.12 1.65 0.26 15 12 7 14
Max Comp + Max MM X X 14 2 2 1 26 32 895 186 0.37 0.12 1.93 0.51 10 6 5 11
Max Comp + Min Split X X 8 3 1 1 35 23 1036 46 0.39 0.12 1.79 0.30 12 7 5 11
All Constraints
Fully Balanced X X X 12 2 2 1 35 23 989 92 0.39 0.11 1.77 0.30 14 11 7 13
Table 2a. Summary Statistics for State Assembly Plans
A. Minority Representation B. Political Subdivision C. Compactness D. Potential Competitiveness
Plan Type
Draw
Competitive
Seat
Respect
County/
City
Boundary
Draw
Majority-
Minority
Seat
No. of Seat
( 50%+
Hispanics)
No of Seat
( 65%+
Hispanics)
No. of
Seat
( 30%+
Blacks)
No. of
Seat
( 30%+
Asians)
No. of
County Not
Split
No. of
County
Split
No. of
Census
Places Not
Split
No. of
Census
Place Split
Roeck
Mean
Roeck
Standard
Deviation
Schwartz-berg
Mean
Schwartz-berg
Standard
Deviation
No. of Seat
( Party
Registration
Diff within 7
pct pt)
No. of Seat
( Party
Registration
Diff within 5
pct pt)
No. of Seat
( Party
Registration
Diff within 3
pct pt)
No. of Seat
( Party
Registration
Diff 3 pt
Rep, 10 pt
Dem)
2001 Assembly 17 6 3 4 31 27 987 94 0.36 0.10 2.04 0.45 9 4 0 5
Random Box 11 4 2 4 26 32 869 212 0.47 0.11 1.51 0.20 19 12 7 17
Single Constraint
Max MM X 22 3 4 3 26 33 855 226 0.41 0.12 1.72 0.31 17 10 7 13
Min Split X 11 5 2 3 31 27 1004 78 0.43 0.11 1.69 0.28 22 16 10 17
Max Competition X 12 5 2 3 26 33 854 228 0.47 0.11 1.58 0.21 31 18 10 26
Double Constraints
Min Split + Max MM X X 22 4 4 4 28 31 959 122 0.40 0.11 1.80 0.31 15 10 7 17
Max Comp + Max MM X X 25 5 4 4 30 28 865 217 0.39 0.11 1.79 0.32 23 14 5 16
Max Comp + Min Split X X 11 5 2 3 28 30 1001 81 0.42 0.12 1.84 0.35 32 19 12 25
All Constraints
Fully Balanced X X X 22 5 4 3 32 26 993 88 0.38 0.11 1.89 0.36 18 13 7 15
Notation Roeck Schwartzberg
Max MM-- Maximize number of majority- minority seat 1= most compact Closer to 1= more compact
Min Split-- Minimize County/ City Split 0= least compact
Max Comp-- Maximize number of potentially competitive seat
Appendix 2
Table 1b. Full Results for All Congressional Plans by Variation of Constraints
A. Minority Representation B. Political Subdivision C. Compactness
Page 2
Plan
Type
Draw
Competiti
ve Seat
Respect
County/
City
Boundary
Draw
Majority-
Minority
Seat
No. of Seat
( 50%+
Hispanics)
No of Seat
( 65%+
Hispanics)
No. of Seat
( 30%+
Blacks)
No. of Seat
( 30%+
Asians)
No. of
County
Not Split
No. of
County Split
No. of
Census
Places Not
Split
No. of
Census
Place Split
Roeck
Mean
Roeck
Standard
Deviation
Schwartz-berg
Mean
Schwartz-berg
Standard
Deviation
2001 Congressional 10 4 2 0 36 22 1016 65 0.33 0.10 2.31 0.61
Random Box Plans
1 5 3 1 1 17 41 889 192 0.51 0.13 1.49 0.28
2 8 1 1 1 20 38 874 207 0.47 0.12 1.52 0.19
3 7 1 1 1 23 35 892 189 0.48 0.13 1.52 0.20
4 8 2 1 2 14 44 852 229 0.49 0.11 1.65 0.23
5 9 3 1 2 13 45 838 243 0.48 0.12 1.66 0.25
Mean 7 2 1 1 17 41 869 212 0.49 0.12 1.57 0.23
Single Constraint-- Max Minority Representation
6 X 18 1 3 1 32 26 870 211 0.39 0.11 1.79 0.28
7 X 19 1 3 1 32 26 864 217 0.40 0.11 1.80 0.30
8 X 18 2 1 2 27 31 892 189 0.36 0.13 2.13 0.54
Mean 18 1 2 1 30 28 875 206 0.38 0.12 1.91 0.37
Single Constraint-- Min County/ City Split
9 X 7 2 1 2 37 21 1036 45 0.42 0.11 1.71 0.31
10 X 6 2 0 2 39 19 1030 51 0.39 0.12 1.71 0.28
Mean 7 2 1 2 38 20 1033 48 0.41 0.12 1.71 0.30
Single Constraint-- Max Competition
16 X 6 1 1 1 25 33 874 207 0.42 0.13 1.67 0.28
17 X 6 2 2 2 21 37 877 204 0.40 0.13 1.75 0.29
18 X 6 2 1 1 15 43 895 186 0.48 0.10 1.63 0.22
19 X 4 2 1 1 15 43 868 213 0.47 0.13 1.64 0.27
20 X 6 2 0 1 13 45 871 210 0.47 0.13 1.66 0.27
Mean 6 2 1 1 18 40 877 204 0.45 0.12 1.67 0.27
Dual Constraints-- Min County/ City Split + Max Minority Representation
11 X X 10 1 2 2 31 27 906 175 0.40 0.11 1.63 0.23
12 X X 10 2 2 2 32 26 934 147 0.40 0.13 1.75 0.29
13 X X 6 2 0 1 35 23 916 165 0.42 0.11 1.61 0.24
14 X X 6 2 1 1 33 25 888 193 0.42 0.12 1.63 0.25
15 X X 6 1 1 0 37 21 918 163 0.43 0.12 1.63 0.27
Mean 8 2 1 1 34 24 912 169 0.41 0.12 1.65 0.26
Dual Constraints-- Max Competition + Max Minority Representation
21 X X 16 1 3 1 23 35 852 229 0.38 0.13 1.97 0.40
22 X X 16 1 3 1 23 35 900 181 0.38 0.12 1.90 0.44
23 X X 11 4 1 1 29 29 926 155 0.36 0.11 1.96 0.57
24 X X 12 3 2 2 28 30 903 178 0.37 0.13 1.89 0.62
Mean 14 2 2 1 26 32 895 186 0.37 0.12 1.93 0.51
Dual Constraints-- Max Competition + Min County/ City Split
25 X X 8 3 0 1 34 24 1046 35 0.39 0.11 1.77 0.28
26 X X 7 2 1 0 36 22 1025 56 0.39 0.12 1.80 0.32
Mean 8 3 1 1 35 23 1036 46 0.39 0.12 1.79 0.30
Fully Balanced
27 X X X 14 1 3 0 35 23 893 188 0.40 0.10 1.73 0.23
Appendix 2
28 X X X 10 2 1 2 32 26 996 85 0.40 0.12 1.75 0.31
29 X X X 14 1 2 0 36 22 1020 61 0.39 0.11 1.86 0.29
30 X X X 10 2 2 1 36 22 1020 61 0.38 0.11 1.77 0.32
31 X X X 10 2 2 2 36 22 1016 65 0.39 0.11 1.76 0.34
Mean 12 2 2 1 35 23 989 92 0.39 0.11 1.77 0.30
Table 2b. Full Results for All State Assembly Plans by Variation of Constraints
A. Minority Representation B. Political Subdivision C. Compactness
Page 3
Plan
Type
Draw
Competiti
ve Seat
Respect
County/
City
Boundary
Draw
Majority-
Minority
Seat
No. of Seat
( 50%+
Hispanics)
No of Seat
( 65%+
Hispanics)
No. of
Seat
( 30%+
Blacks)
No. of
Seat
( 30%+
Asians)
No. of
County
Not
Split
No. of
County
Split
No. of
Census
Places
Not Split
No. of
Census
Place
Split
Roeck
Mean
Roeck
Standard
Deviation
Schwartz-berg
Mean
Schwartz-berg
Standard
Deviation
2001 Assembly 17 6 3 4 31 27 987 94 0.36 0.10 2.04 0.45
Random Box Plans
1 13 3 3 4 33 25 887 194 0.46 0.12 1.51 0.20
2 11 4 3 3 34 24 890 191 0.47 0.12 1.51 0.21
3 9 4 2 3 16 42 842 239 0.48 0.10 1.51 0.19
4 10 4 1 5 20 38 858 223 0.47 0.11 1.51 0.21
Mean 11 4 2 4 26 32 869 212 0.47 0.11 1.51 0.20
Single Constraint-- Max Minority Representation
5 X 23 3 4 3 27 31 837 244 0.41 0.12 1.71 0.27
6 X 21 3 3 3 24 34 873 208 0.41 0.12 1.73 0.35
Mean 22 3 4 3 26 33 855 226 0.41 0.12 1.72 0.31
Single Constraint-- Min County/ City Split
7 X 12 5 2 3 35 23 1019 62 0.43 0.11 1.63 0.26
8 X 10 5 2 3 27 31 988 93 0.42 0.11 1.74 0.30
Mean 11 5 2 3 31 27 1004 78 0.43 0.11 1.69 0.28
Single Constraint-- Max Competition
11 X 11 4 2 2 31 27 858 223 0.47 0.11 1.57 0.20
12 X 12 5 2 3 20 38 849 232 0.46 0.11 1.58 0.21
Mean 12 5 2 3 26 33 854 228 0.47 0.11 1.58 0.21
Dual Constraints-- Min County/ City Split + Max Minority Representation
9 X X 23 4 4 4 29 29 930 151 0.42 0.11 1.70 0.27
10 X X 21 4 3 3 26 32 988 93 0.38 0.11 1.90 0.35
Mean 22 4 4 4 28 31 959 122 0.40 0.11 1.80 0.31
Dual Constraints-- Max Competition + Max Minority Representation
13 X X 25 3 4 4 31 27 862 219 0.40 0.11 1.71 0.26
14 X X 24 6 4 3 29 29 867 214 0.38 0.11 1.87 0.38
Mean 25 5 4 4 30 28 865 217 0.39 0.11 1.79 0.32
Dual Constraints-- Max Competition + Min County/ City Split
15 X X 11 4 1 4 32 26 992 89 0.41 0.12 1.77 0.33
16 X X 10 5 2 2 24 34 1009 72 0.42 0.12 1.90 0.36
Mean 11 5 2 3 28 30 1001 81 0.42 0.12 1.84 0.35
Fully Balanced
17 X X X 27 4 4 2 32 26 998 83 0.39 0.12 1.82 0.29
18 X X X 20 5 4 4 30 28 991 90 0.38 0.11 1.91 0.42
19 X X X 20 6 4 3 31 27 998 83 0.38 0.11 1.95 0.36
20 X X X 20 6 4 4 33 25 987 94 0.37 0.12 1.89 0.34
21 X X X 22 6 4 3 33 25 989 92 0.38 0.11 1.90 0.43
Mean 22 5 4 3 32 26 993 88 0.38 0.11 1.89 0.37
Appendix 2
E. Rose
Plan
Type
Draw
Competiti
ve Seat
Respect
County/
City
Boundary
Draw
Majority-
Minority
Seat
No. of Seat
( Party
Registration
Diff within 7
pct pt)
No. of Seat
( Party
Registration
Diff within 5
pct pt)
No. of Seat
( Party
Registration
Diff within 3
pct pt)
No. of Seat
( Party
Registration
Diff 3 pt Rep,
10 pt Dem)
No. of Seat
( Party
Registration
Diff within 7
pct pt)
No. of Seat
( Party
Registration
Diff within 5
pct pt)
No. of Seat
( Party
Registration
Diff within 3
pct pt)
No. of Seat
( Party
Registration
Diff 3 pt Rep,
10 pt Dem)
No. of Seat
( Party
Registration
Diff within 7
pct pt)
No. of Seat
( Party
Registration
Diff within 5
pct pt)
No. of Seat
( Party
Registration
Diff within 3
pct pt)
No. of Seat
( Party
Registration
Diff 3 pt Rep,
10 pt Dem)
No. of Seat
( Vote
Margin
within 7 pct
pt)
No. of Seat
( Vote
Margin
within 5 pct
pt)
No. of
Seat
( Vote
Margin
within 3
pct pt)
Rose
Report
Measure
( 5pt Rep &
10 pt Dem)
3 0 0 0 0 0 0 2 0 0 0 1 1 1 0 0
1 12 8 8 14 16 11 8 16 17 14 8 12 14 12 8 14
2 12 9 5 13 11 7 5 11 11 8 7 12 7 5 4 15
3 15 11 5 12 16 8 7 11 12 12 5 9 11 8 4 15
4 15 11 8 14 13 8 6 12 11 10 6 9 8 8 2 15
5 18 14 7 13 12 8 7 12 14 10 6 9 13 13 7 17
Mean 14 11 7 13 14 8 7 12 13 11 6 10 11 9 5 15
Single Constraint-- Max Minority Representation
6 X 11 9 5 11 11 7 6 10 7 6 4 10 9 6 2 13
7 X 14 9 5 11 11 8 5 11 8 6 3 10 7 6 4 12
8 X 11 8 3 6 9 6 4 7 8 3 1 5 6 5 3 9
Mean 12 9 4 9 10 7 5 9 8 5 3 8 7 6 3 11
Single Constraint-- Min County/ City Split
9 X 14 9 2 9 11 5 2 9 10 8 4 9 5 5 4 13
10 X 14 11 8 13 13 8 3 8 12 8 6 11 10 9 6 15
Mean 14 10 5 11 12 7 3 9 11 8 5 10 8 7 5 14
Single Constraint-- Max Competition
16 X 25 18 13 20 20 15 9 13 17 13 9 14 17 13 12 22
17 X 29 18 14 25 23 17 10 17 20 14 8 13 16 13 8 26
18 X 21 14 8 18 20 12 8 14 17 14 8 12 13 10 8 20
19 X 23 16 10 18 21 16 9 12 16 12 8 13 16 8 4 22
20 X 24 17 10 19 19 15 11 14 17 13 10 14 15 9 1 22
Mean 24 17 11 20 21 15 9 14 17 13 9 13 15 11 7 22
Dual Constraints-- Min County/ City Split + Max Minority Representation
11 X X 16 12 5 14 15 10 6 12 13 11 7 11 12 8 6 15
12 X X 15 10 7 12 12 9 5 9 12 9 6 9 9 8 6 13
13 X X 12 9 7 14 13 10 9 13 12 10 5 11 8 3 2 14
14 X X 15 13 8 16 14 11 8 14 15 11 5 12 8 7 6 19
15 X X 15 14 9 16 15 11 8 14 15 12 7 14 10 6 5 18
Mean 15 12 7 14 14 10 7 12 13 11 6 11 9 6 5 16
Dual Constraints-- Max Competition + Max Minority Representation
21 X X 10 6 5 10 9 5 3 7 9 3 2 11 7 6 3 10
22 X X 8 5 5 11 10 6 4 10 10 8 5 12 12 10 8 11
23 X X 12 7 5 12 10 8 6 13 8 7 5 8 8 7 4 12
24 X X 11 6 5 11 10 8 6 11 10 9 4 11 10 6 3 11
Mean 10 6 5 11 10 7 5 10 9 7 4 11 9 7 5 11
Dual Constraints-- Max Competition + Min County/ City Split
25 X X 9 5 3 7 8 5 3 6 8 6 2 7 7 5 4 7
26 X X 15 8 6 14 13 8 6 12 12 7 5 14 7 6 4 15
Mean 12 7 5 11 11 7 5 9 10 7 4 11 7 6 4 11
Fully Balanced
27 X X X 16 11 7 14 14 13 10 13 13 9 8 11 11 9 6 17
28 X X X 14 10 6 12 15 10 5 12 14 9 5 10 12 11 10 14
29 X X X 16 12 7 13 14 14 10 14 13 9 8 11 11 9 5 16
30 X X X 10 9 7 12 11 8 5 11 13 7 6 11 7 6 3 13
31 X X X 16 11 7 12 13 12 7 11 15 11 7 11 12 8 7 14
Mean 14 11 7 13 13 11 7 12 14 9 7 11 11 9 6 15
Appendix 2
Table 3a. Results for Various Measures of Potential Competitiveness for All Congressional Plans
Random Box Plans
2001 Congressional Plan
A. 2000 Party Registration B. 2004 Party Registration C. Normal Vote D. 2000 Presidential
Page 4
E. Rose
Plan Type
Draw
Competitiv
e Seat
Respect
County/
City
Boundary
Draw
Majority-
Minority
Seat
No. of Seat
( Party
Registration
Diff within 7
pct pt)
No. of Seat
( Party
Registration
Diff within 5
pct pt)
No. of Seat
( Party
Registration
Diff within 3
pct pt)
No. of Seat
( Party
Registration
Diff 3 pt Rep,
10 pt Dem)
No. of Seat
( Party
Registration
Diff within 7
pct pt)
No. of Seat
( Party
Registration
Diff within 5
pct pt)
No. of Seat
( Party
Registration
Diff within 3
pct pt)
No. of Seat
( Party
Registration
Diff 3 pt Rep,
10 pt Dem)
No. of Seat
( Party
Registration
Diff within 7
pct pt)
No. of Seat
( Party
Registration
Diff within 5
pct pt)
No. of Seat
( Party
Registration
Diff within 3
pct pt)
No. of Seat
( Party
Registration
Diff 3 pt Rep,
10 pt Dem)
No. of
Seat ( Vote
Margin
within 7
pct pt)
No. of
Seat ( Vote
Margin
within 5
pct pt)
No. of
Seat ( Vote
Margin
within 3
pct pt)
Rose Report
Measure
( 5pt Rep &
10 pt Dem)
9 4 0 5 7 2 1 8 6 4 0 6 5 1 1 8
1 18 11 5 15 16 12 5 17 17 12 8 18 14 11 7 20
2 23 15 9 25 22 20 12 24 24 18 13 19 20 14 6 26
3 22 13 9 15 17 13 9 16 20 14 10 19 17 12 9 18
4 14 8 4 13 14 11 9 16 16 13 7 14 17 10 8 15
Mean 19 12 7 17 17 14 9 18 19 14 10 18 17 12 8 20
Single Constraint-- Max Minority Representation
5 X 20 10 9 15 15 10 5 16 18 8 4 15 12 8 6 16
6 X 14 10 5 11 13 10 7 13 12 7 4 11 11 9 6 14
Mean 17 10 7 13 14 10 6 15 15 8 4 13 12 9 6 15
Single Constraint-- Min County/ City Split
7 X 20 16 9 17 17 15 10 18 17 13 8 18 14 10 6 23
8 X 24 16 11 16 20 16 10 16 17 14 8 15 18 16 7 20
Mean 22 16 10 17 19 16 10 17 17 14 8 17 16 13 7 22
Single Constraint-- Max Competition
11 X 32 19 12 30 26 23 11 20 22 18 12 19 17 14 10 32
12 X 29 17 8 21 22 15 10 18 19 14 9 16 14 9 4 26
Mean 31 18 10 26 24 19 11 19 21 16 11 18 16 12 7 29
Dual Constraints-- Min County/ City Split + Max Minority Representation
9 X X 12 8 5 16 15 9 5 16 18 10 4 14 11 8 8 18
10 X X 18 12 9 17 15 12 10 16 14 9 6 13 10 4 3 19
Mean 15 10 7 17 15 11 8 16 16 10 5 14 11 6 6 19
Dual Constraints-- Max Competition + Max Minority Representation
13 X X 25 15 5 19 20 15 8 21 19 11 7 19 16 13 7 22
14 X X 21 12 4 13 18 12 4 16 12 10 5 14 9 5 1 15
Mean 23 14 5 16 19 14 6 19 16 11 6 17 13 9 4 19
Dual Constraints-- Max Competition + Min County/ City Split
15 X X 33 19 12 25 27 17 14 21 27 18 10 17 21 16 11 29
16 X X 31 19 11 24 24 16 9 20 24 21 12 17 21 15 10 26
Mean 32 19 12 25 26 17 12 21 26 20 11 17 21 16 11 28
Fully Balanced
17 X X X 16 12 8 12 17 14 9 15 18 12 6 12 15 11 6 13
18 X X X 18 14 7 15 15 13 7 16 12 10 6 16 11 7 7 17
19 X X X 18 12 8 17 18 10 7 18 16 11 7 14 11 7 2 18
20 X X X 20 13 7 16 17 12 8 18 17 13 7 16 12 9 6 19
21 X X X 18 13 6 14 16 10 6 14 14 11 6 13 10 8 7 16
Mean 18 13 7 15 17 12 7 16 15 11 6 14 12 8 6 17
Page 5 Appendix 2
Random Box Plans
D. 2000 Presidential
2001 Assembly Plan
Table 3b. Results for Various Measures of Potential Competitiveness for All State Assembly Plans
A. 2000 Party Registration B. 2004 Party Registration C. Normal Vote