HomeMy WebLinkAboutDiamond Bar Racially Polarized Voting Analysis.pdf1 | Page
City of Diamond Bar – Racially Polarized Voting Analysis
The California Voting Rights Act (CVRA) was enacted in 2002 and focuses exclusively on the use of at-
large election systems. As defined in the law, at-large systems include any election method except
single-member districts in which only the area voters select their representative.
The law does not create any oversight agency or empower any state or regional agencies to implement
the law; instead, it is left to the courts. Unlike federal Voting Rights Act cases, CVRA suits can be filed in
local courts and litigation costs are fully recoverable from the successful plaintiff. In order to be
successful, the plaintiffs must only prove that racially polarized voting exists and that any protected
subgroups could influence elections under a different system.
Racially polarized voting is where a protected minority group has a preference for one candidate or
issue, while the majority has a preference for another. In order to establish racially polarized voting,
California law requires courts to look to methodologies used in applicable federal cases to enforce the
federal Voting Rights Act.
In Thornburg v. Gingles, 478 U.S. 30 (1986), the U.S. Supreme Court noted that reliable inferences about
voting behavior could be derived from a number of techniques, including homogenous precinct analysis
(HPA) and regression analysis.
One circumstance that may be considered under CVRA is the extent to which candidates who are
members of a protected class and who are preferred by voters of the protected class have been elected
to the governing body of a political subdivision. This is considered probative evidence, but it alone does
not preclude or create a successful claim under CVRA.
In the precedent-setting Gingles case, the Supreme Court upheld the trial court’s decision that racially
polarized voting existed in North Carolina. The evidence included a statistical analysis showing that the
African American support for black candidates was overwhelming in almost every election. In all but five
of 16 primary elections, African American voters showed strong support for African American
candidates.
In contrast, the trial court found that a substantial majority of white voters would not vote for an African
American candidate. In the general elections, white voters almost always ranked black candidates either
last or next to last in the multicandidate field, except in heavily Democratic areas where white voters
consistently ranked black candidates last among the Democrats, if not last or next to last among all
candidates.
The court went on to state:
“…Multimember districts may impair the ability of blacks to elect representatives of their choice
where blacks vote sufficiently as a bloc as to be able to elect their preferred candidates in a black
majority, single-member district and where a white majority votes sufficiently as a bloc usually
to defeat the candidates chosen by blacks. It is the difference between the choices made by
blacks and whites — not the reasons for that difference — that results in blacks having less
2 | Page
opportunity than whites to elect their preferred representatives. Consequently, we conclude that
under the ‘results test’ of § 2, only the correlation between race of voter and selection of certain
candidates, not the causes of the correlation, matters.”
While modern-day California is very different than North Carolina several decades ago, the precedents
set in these cases still have a direct impact on how at-large elections are viewed and analyzed.
Measuring Degree of Polarized Voting
Redistricting Partners utilizes multiple methods for determining racially polarized voting for the
purposes of the CVRA:
Homogeneous Precinct Analysis
The first level of data analysis is of voting patterns in homogenous census blocks – small areas
that are composed of a single racial group. The voting patterns of minorities in these blocks are
analyzed and compared to similar areas with very few minority voters.
In the absence of exit polls and direct access to individual ballots, this common measure of
racially polarized voting provides a high-confidence way to see voting patterns. Since census
blocks are usually not exclusively one race, blocks with greater than 80% or more individuals of a
single race are considered homogeneous. In order to achieve statistical validity, the analysis
should include a large number of homogenous precincts. In some parts of the state, aggregation
of many census blocks will provide a final analysis of several thousand individual vote results in a
cluster that is 90% or more of one single race.
Regression / Trend Line Analysis
A trend line analysis is done using all the precinct-level election results from a candidate race or
ballot measure. The results for each precinct are placed in a formula with a variable to be
studied, such as ethnicity of that precinct based on surname analysis from the voter file or
Census Bureau figures on the ethnicity of the Citizen Voting Age Population. The data points are
each individually plotted with a regression to overlay a trend line. This trend line will show how
the vote for or against a candidate or ballot measure increases or decreases as the variable
changes.
The resulting formula in the format of Y=mX+B, with m=slope, provides a quick way to compare
the trend between different groups. A large positive slope shows a relationship between votes
and that ethnic group, suggesting bloc voting. A negative slope would suggest that the group is
bloc voting against the candidate or issue.
This formula is supplemented by an “R-squared” value, which is the correlation coefficient. This
is the value which frames the mathematical equation as being predictive of an impact or not. A
high R-square of “1” would mean that the two subjects being studied are perfectly correlated,
and a “0” would be perfectly uncorrelated. However, in this kind of analysis, there are rarely
absolutes —there are ranges. In this work we see that a range of 0.8 to 0.4 is very strong
correlation, a range of 0.3 to 0.2 is moderate correlation, 0.2 to 0.1 as slight correlation, and
anything under 0.1 is not enough correlation to provide any kind of real support to a claim.
3 | Page
Ecological Inference
The next step in the process after performing a regression is a method of making deeper
conclusions regarding the voting patterns of protected classes using Ecological Inference. This
takes the aggregated data from each precinct, uses the data from the regression, and
mathematically calculates a likely range of support for candidates based on any highly-
correlated results.
Due to the fact there are not areas with very few Latino or Asian residents to compare heavily
Asian/Latino areas with, this analysis relies almost entirely on the regression method and ecological
inference. This provides the greatest statistical framework for analysis, and it responds to the tools
which would most commonly be used in a CVRA complaint.
Regression Analysis Examples
Regression analysis is not something many people encounter in their regular course of work, so here are
some examples of regressions utilizing data from an ethnically diverse city in Northern California. This is
intended to show how the methodology can show or disprove correlations.
Positive Correlation
In this first example, we have two things that should be related – the size of the family and the number
of people in a household. Common sense would dictate that households with large families would have
more people under one roof, and households with roommates, or non-family members, would generally
have fewer people in them.
The regression takes data from the census and places family size and the number of people in the
household on opposite axes. This creates a scatterplot with a dot for the value of each thing being
measured – with one for each geographic area (in this case, census blocks).
4 | Page
The line of best fit is Y= 0.59x + 1.87 with an R-squared value of 0.54. There are two important pieces of
information that can be culled from these calculations. The 0.59 slope of the line shows that the two
variables, family size and household size, are positively linked and that their growths align. The R-
squared number measures how densely packed the individual dots are to the line of best fit. The denser
these dots are packed, the higher the R-squared value and the overall correlation between these two
variables. The closer the R-squared is to 1.0, the higher the correlation between the two variables. An R-
squared of 0.54 shows a high correlation between these variables, and the conclusion from this
scatterplot would be: As family size grows, so does household size.
Lacking Correlation
For something that should have no correlation, we can look at data on how many women are in each
household and how many Asians. There is no reason for these two datasets to have any relationship –
Asian residents are no more likely to be female or male, and women are no more likely to be Asian than
men are.
Utilizing the same scatterplot and line of best fit, along with the creation of a regression formula, we can
see that there, in fact, is no correlation between these two measures. The scatterplot has no upward or
downward direction, and the line of best fit is completely flat.
The formula created from this data for the line of best fit is Y= 0.016x + 0.50 with a miniscule R-squared
of .0006. The slope of the line of best fit is very close to zero, illustrating that the relationship between
the two variables being analyzed is also close to zero. The R-squared value of 0.0006, again nearly zero,
shows that there is very little correlation between these two variables, as expected. This substantiates
what would be known intuitively – there is no relationship between the number of women and the
number of Asian residents in each household.
5 | Page
Negative Correlation
The final example is of negative correlation: two data sets that can be shown to have opposite
relationships. When one value goes up, the other value goes down.
For this, we have data on the number of seniors in each household, and the number of children.
As can be seen in this scatterplot, there is a downward direction of the dots, but the data is not narrowly
clustered around the line of best fit. It is a bit fatter than the first example with lots of datapoints far
from the line of best fit.
The formula created from this data is Y= -1.36x + 1.06 with a small R-squared of 0.11. The slope of the
line of best fit, -1.36, substantiates what would be known intuitively – there is a negative relationship
between the number of seniors and the number of children in each household. However, the strength of
this relationship is not as great as the first example; the dots are not narrowly packed along the line of
best fit. Instead, the dots are dispersed throughout the scatterplot. Mathematically, this low correlation
is shown in the small R-squared of 0.11.
In summary, the more the data is directional (going up or down, rather than being all in a random
scatterplot like a shotgun blast), the more there is a seeming relationship between the two datapoints.
That relationship is described in a formula, and the most important part of that formula is the R-squared
value. In racially polarized voting analyses, scatterplots with a positive or negative direction and medium
to high R-squared values mean that race was a polarizing factor in the political decisions being made.
6 | Page
Tools of the Analysis
In conducting this analysis, we will utilize four main datasets built specifically to the City of Diamond Bar:
1) 2015-2019 American Community Survey (ACS) dataset, which provides estimated total
population and Citizen Voting Age Population (CVAP). This dataset is based on sample data
and statistical estimates and can be useful in understanding protected minority classes and
their representation in different portions of the district boundaries.
2) Registration data to understand the voter makeup of the city and recent vote history, including
the age, ethnicity (utilizing surname, birthplace and language ballot access), partisanship, and
other factors which help provide understanding of the populations that have voted in recent
local and statewide elections.
3) Election history and results for statewide, congressional and local elections, along with the
geographic layers for precinct boundaries, which have changed for each election during the
time covered in this analysis.
4) 2020 Census Redistricting Data [P.L. 94-171] Summary File, adjusted for prison population
reallocation by the California Statewide Database. This is the dataset utilized for the creation
of districts in a CVRA conversion or redistricting.
City of Diamond Bar Population and Ethnicity
Based on the US Census, adjusted for prison population by the California Statewide Database, the City of
Diamond Bar has a total population of 55,181.
This can be further broken down to 38,179 Citizen Voting Age Population (CVAP), or 34,267 registered
voters. The federal Voting Rights Act places a priority on the ethnic composition of the CVAP as it is a
proxy for eligible voters and allows for an analysis of potential voting strength of protected classes
without having to adjust for common lower registration and participation rates for these minority
populations.
The majority of residents in the City of Diamond Bar are Asian, totaling 59.1% of the population and
account for 52.8% of the Citizen Voting Age Population. The city is also home to a sizeable Latino
population, accounting for 19.7% of the total population and 19.3% of the CVAP. In total, there are
14,607 registered Asian voters accounting for 42.6% of registered voters, and there are 6,575 registered
Latino voters, accounting for 19.2% of all voters. Right off the bat, what this tells us is even though the
Asian population represents the majority of the voting age population, they can still be denied their
candidate of choice if non-Asian voters pick a different candidate.
Asians have only recently become the majority of voting age citizens in the City of Diamond Bar. In 2010,
the Asian CVAP was 45.9% and the Latino CVAP was 20.1%.
The other protected ethnic groups in Diamond Bar – Native Americans and Black – comprise less than
4% of the total population and are not a part of this analysis due to their smaller size.
7 | Page
Prior Election History
In analyzing elections, we can look at exogenous elections (elections that go beyond the city’s borders,
such as statewide or county office, but reviewing just those precincts within the city) or endogenous
elections (prior elections for the agency being analyzed).
Statewide and Congressional Elections
In statewide and congressional candidate contests, Diamond Bar has been consistently supportive of
Democrats, but are still relatively moderate. The following are vote breakdowns for a few elections that
are commonly analyzed, the 2018 Gubernatorial election, with two white candidates, and the contests
for Attorney General with a Latino Democrat and White Republican and congress, specifically CA-39,
with a Latino Democrat and an Asian Republican.
Governor 2018
Candidate Votes
Gavin Newsom (D) 9,837 54.7%
John Cox (R) 8,147 45.3%
Attorney General 2018
Candidate Votes
Xavier Becerra (D) 9,996 56.9%
Steven Bailey (R) 7,577 43.1%
U.S. House of Representatives 2018
Candidate Votes
Gil Cisneros (D) 10,624 60.2%
Young Kim (R) 7,024 39.8%
8 | Page
Statistical Analysis
Utilizing these same elections, we can produce a regression analysis with a plot of the winner’s votes by
precinct, the racial composition of the precinct, a line of best fit and resulting regression formula.
2018 Latino Voters
2018 Governor Newsom 2018 AG Becerra
2018 Gil Cisneros 2018 Young Kim
9 | Page
2018 Asian Voters
2018 Governor Newsom 2018 AG Becerra
2018 Gil Cisneros 2018 Young Kim
10 | Page
2018 White Voters
2018 Governor Newsom 2018 AG Becerra
2018 Gil Cisneros 2018 Young Kim
All three of these races show evidence of racial polarization, most notably seen in Latino voters, where
we see moderately high R-squared values and steep lines of best fit, telling us that the more a precinct is
comprised of Latino voters, the more support Newsom, Becerra and Cisneros enjoy. We see the same
voting patters with white voters, where the more a precinct is comprised of white voters, the more
support the Democratic candidates received. However, the R-squared values are substantially lower,
suggesting there is little correlation between a white voter’s ethnicity and their support for a candidate.
11 | Page
Asian voters, on the other hand, have the opposite voting pattern. The lines of best fit for the
Democratic candidates are all negative, suggesting that as precincts contain more Asian voters, the less
support those candidates receive. Here, we see potential for bloc voting against the Asian voters’
candidates of choice.
Using Ecological Inference, we can take the aggregate result and break it down into predicted outcomes
just among Latino, Asian and white voters. This method is intended as a way of identifying the
directional support based on ethnicity. If there is no racially polarized voting, the results would be the
same for each racial group, but when there is polarized voting, it can identify when the candidate
choices of the majority are overwhelming the candidate choices of the protected minority group.
Governor 2018
Asian Support Latino Support White Support
Gavin Newsom (D) 48.6% 76.8% 50.1%
John Cox (R) 51.4% 23.2% 49.9%
Attorney General 2018
Asian Support Latino Support White Support
Xavier Becerra (D) 49.9% 85.0% 49.9%
Steven Bailey (R) 50.1% 15.0% 50.1%
U.S House of Representatives 2018
Asian Support Latino Support White Support
Gil Cisneros (D) 57.0% 75.5% 57.7%
Young Kim (R) 43.0% 24.5% 42.3%
In all three of these races, we Asian and White voters support each candidate at almost identical rates,
suggesting there is little polarization occurring. However, Latino voters have substantially different
support for candidates, which tells us there is polarized voting occurring for at least one substantially
large protected class in Diamond Bar.
Local Elections
There have been recent local elections in Diamond Bar which have been more competitive. Given the
City Council is composed of five members, these contests elect two or three candidates each election
cycle for four-year terms. Because they all run together, no candidate has received more than 32% over
the past three election cycles.
November 2015
City Council (2 elected)
Candidate Votes
Robert Nishimura 2,149 29.2%
Carol Herrera* 2,216 30.1%
12 | Page
Robert Velker 730 9.9%
Ruth Low* 2,266 30.8%
November 2018
City Council (3 elected)
Candidate Votes
Andrew Chou* 7,483 19.0%
Nancy A Lyons* 9,880 25.0%
Jimmy Lin 6,366 16.1%
Robert Bob Velker 4,081 10.3%
Steve Tye* 8,231 20.9%
Joe Weng 3,430 8.7%
November 2020
City Council (2 elected)
Candidate Votes
Ruth Low* 12,748 31.9%
Stan Liu* 8,135 20.4%
Tommy Orona 2,321 5.8%
Jennifer Mahlke 6,475 16.2%
Aaron McElerea 1,689 4.2%
San Castorena Jr. 3,652 9.1%
Bill Rawlings 4,926 12.3%
Statistical Analysis
Utilizing many of these same elections, we can also produce a regression analysis for these local
contests.
Local results provide us another opportunity to look for elections that could be polarized. Often,
statewide elections are driven by partisanship, making it harder to separate out the power of partisan
labels versus racial polarization. In local contests, without identified partisanship, the weight that voters
are giving to perceived ethnicity of candidates can be more instructive.
13 | Page
2018 City Council Race
Asian Voter Support
Jimmy Lin Nancy Lyons Steve Tye
Latino Voter Support
Jimmy Lin Nancy Lyons Steve Tye
White Voter Support
Jimmy Lin Nancy Lyons Steve Tye
14 | Page
Here, again, we see bloc voting among white and Latino voters, supporting Nancy Lyons and Steve Tye,
who were the two candidates selected to serve on the City Council, and Asian voters supported Jimmy
Lin, who placed fourth. The R-squared values are particularly high for all three ethnicities in regards to
Lin and Lyons’ races, suggesting that there is a strong relationship between a voter’s race and their
support or opposition to a candidate.
2020 City Council Race
Asian Voter Support
Ruth Low Jennifer Hahlke Stan Liu
Latino Voter Support
Ruth Low Jennifer Hahlke Stan Liu
15 | Page
White Voter Support
Ruth Low Jennifer Hahlke Stan Liu
In the 2020 city council races, we, again, see significant levels of racial polarization and similar voting
patterns as 2018. White and Latino voters continued to bloc voting against the Asian candidates of
choice, and the R-squared values are high across all races and ethnicities, suggesting there is strong
correlation between a voter’s race and their support of a candidate. This time, however, the both
candidates supported by Asian voters, Ruth Low and Stan Liu, were elected.
After examining city council races, we can conclude with a high degree of confidence that there is a
racial polarization occurring at the local level.
Going further into analyzing the citywide elections, we used Ecological Inference to predict the
outcomes just among Asian, Latino and white voters to identify directional support based on
race/ethnicity and to see which candidate was the candidate of choice among their demographic, if
there was one. If there is no racially polarized voting, the results would be the same for each racial
group.
2018 City Council Race
Asian Support Latino Support White Support
Andrew Chou* 41.2% 19.7% 50.2%
Nancy A Lyons* 38.1% 61.9% 66.5%
Jimmy Lin 55% 20.6% 15%
Robert Bob Velker 9.6% 12.8% 45.4%
Steve Tye* 38.8% 25.7% 58.0%
Joe Weng 16.7% 0% 36.7%
In this race, we see Asian voters denied their candidate of choice. Jimmy Lin was the only candidate with
over 50% Asian support, yet, he came in fourth place overall. All three candidates who won, Andrew
Chou, Nancy Lyons and Tye, all had over 50% support from white voters in this predicted model.
16 | Page
2020 City Council Race
Asian Support Latino Support White Support
Ruth Low* 54.2% 0% 54.5%
Stan Liu* 45.7% 3.8% 20.8%
Tommy Orona 3.9% 24.1% 3.0%
Jennifer Mahlke 10.9% 31.5% 34.1%
Aaron McElerea 4.8% 6.2% 8.5%
San Castorena Jr. 0% 15.3% 28.3%
Bill Rawlings 11.3% 18.6% 25.8%
In the 2020 race, the candidates who were elected were supported by Asian and white voters, but saw
single digit or no support from Latino voters. And while Asian voters were not denied their candidate of
choice, Ruth Low, there is clearly evidence of racially polarized voting continuing to play a role in picking
winners and losers.
Majority Minority District
For the purposes of this analysis, we were able to successfully create two maps with four out of five
districts being composed of over 50% Asian CVAP. This flags the important issue of Section 2 compliance
under the federal Voting Rights Act, and demonstrates that a federally protected class under the VRA
could potentially have their voting power diluted if they were districted in a way that would separate
them.
Map A
Map B
In this case, we see similar results across all five districts, regardless of being a majority minority district
or not.
DISTRICT Pop Asian Latino Biden Trump Newsom Cox
1 10,991 56% 20% 59% 40% 53% 47%
2 11,079 61% 15% 59% 41% 51% 49%
3 11,043 61% 20% 59% 41% 55% 45%
4 10,848 51% 18% 61% 39% 58% 42%
5 11,220 36% 23% 61% 39% 57% 43%
DISTRICT Pop Asian Latino Biden Trump Newsom Cox
1 11,378 58% 21% 59% 41% 55% 45%
2 10,952 58% 14% 59% 41% 53% 47%
3 11,005 53% 20% 61% 39% 57% 43%
4 10,894 59% 19% 59% 40% 52% 48%
5 10,952 36% 22% 61% 39% 57% 43%
17 | Page
Conclusion
While Diamond Bar has an Asian CVAP of over 50% they still have polarized voting which can show
differences in voting patterns when Latino and white voters are bloc voting against their candidates.
Similarly, voting preferences of Latinos can be seen as polarized against white and Asian residents.
This may not happen in every election cycle, but the evidence that it can and does suggests that the
ability of Asians and Latinos to elect their favored candidates may be impaired.
We have observed high and middling levels of racial polarization occurring at the local level and state
levels, and we have identified two maps with four majority minority districts. Going to districted
elections may have made the difference for at least one candidate of choice for Asian voters between
getting elected to city council and not.
The tools commonly used by any potential plaintiff, such as regression analysis and ecological inference,
show statewide and local municipal elections differed in support among Latino, Asian and non-Latino or
Asian residents in a substantial manner, and that could be used to demonstrate how the at-large system
in the City of Diamond Bar could potentially deny Asian and Latino voters the opportunity to elect their
candidates of choice.
It appears Asian and Latino voters could have a greater ability to influence the outcome of an election
with districted elections, which is the standard under the California Voting Rights Act.
We are happy to review this analysis with legal counsel for the city or present these findings.