University of Southern California Dissertations and Theses

Location, location, location: a spatial econometric analysis of place-context effects in Los Angeles mayoral elections

(USC Thesis Other)

Location, location, location: a spatial econometric analysis of place-context effects in Los Angeles mayoral elections

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content LOCATION, LOCATION, LOCATION:
A SPATIAL ECONOMETRIC ANALYSIS OF PLACE-CONTEXT
EFFECTS
IN LOS ANGELES MAYORAL ELECTIONS

by

Jason Alan McDaniel

Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(POLITICAL SCIENCE)

December 2007

Copyright 2007 Jason Alan McDaniel

ii
DEDICATION

To my partner,
Casey,
whose constant support
and patient understanding
made this possible

iii
ACKNOWLEDGEMENTS

This project is the culmination of years of inspiration, support, and
encouragement by many people. I gratefully acknowledge Janelle Wong, the chair of my
dissertation committee, who energetically took on this odd little orphan of a project, and
whose effort and attention, often from across the country, were crucial to its completion.
I am indebted to Jeb Barnes whose teaching and analytical approach to political science
have inspired me in countless ways. I am grateful to Phil Ethington for his enthusiastic
commitment to expanding my intellectual horizons and developing my theoretical
understanding of place and space.
I am especially thankful to Judith Grant, who inspired me as a young
undergraduate with her brilliant teaching of political theory, and who has mentored me
throughout my graduate career.
I owe a debt of gratitude to Ernesto Calvo, who patiently guided a young graduate
student through some difficult methodological territory. Similarly, I thank Gary King
and Andrew Gelman for their patient responsiveness to my methodological inquiries. I
owe many thanks to both Luc Anselin and Wendy K. Tam Cho for their pioneering work
in spatial econometrics and their tireless efforts to make spatial analysis accessible to the
larger scholarly community.
I want to thank the faculty of the Department of Political Science at the University
of Southern California for fostering a supportive and challenging intellectual
environment. I am grateful to Raphael Sonenshein, who is the model of the politically
engaged public scholar of Los Angeles politics, for allowing me to waylay him with my
questions in the hallways during his time at USC. I also acknowledge Eric Garcetti, who
as a politician that combines intellect with ambition is an inspiration to many.
iv
TABLE OF CONTENTS
Dedication........................................................................................................................... ii
Acknowledgements............................................................................................................ iii
List of Tables .......................................................................................................................v
List of Figures.................................................................................................................... vi
Abstract............................................................................................................................. vii
Chapter 1 The Politics That Places Make: A Review of the Literature............................1
Political Geography: Politics in its Place........................................................ 4
Voting Behavior: Loss of Context ................................................................ 15
Contextual Determinants of Political Behavior ............................................ 22
Conclusion: Outline of the Dissertation........................................................ 27
Chapter 2 Place-Context and Political Behavior: Ontological and Methodological
Debates...........................................................................................................29
Place-Context and Ontological Individualism: The Agnew-King Debate.... 31
Methodology: The Ecological Fallacy, Survey Data and Aggregate Data
Analysis......................................................................................................... 36
Dissertation: Methods and Data .................................................................... 43
Spatial Econometric Methods ....................................................................... 55
Chapter 3 Place, Space, and Urban Voting Behavior: A Spatial Analysis of the
2001 Los Angeles Mayoral Election..............................................................57
Los Angeles 2001 Mayoral Election............................................................. 59
Data and Methods.......................................................................................... 63
Spatial Data Analysis .................................................................................... 64
Conclusion: Place-Context and Neighborhood Effects................................. 94
Chapter 4 Place-Context and Racial Group Voting Behavior: A Spatial Analysis of
the 2005 Los Angeles Mayoral Election........................................................97
Racial Voting in Los Angeles ..................................................................... 100
Data and Methods........................................................................................ 102
Discussion of Results .................................................................................. 113
Conclusion: The Interaction of Race and Place .......................................... 132
Chapter 5 Conclusion: Implications and Future Research............................................135
Place-Context as Causal Mechanism .......................................................... 136
Place-Context and Historical Institutionalism............................................. 141
Political Participation and Local Democracy.............................................. 142
Bibliography ....................................................................................................................146
v
LIST OF TABLES
Table 2.1 Descriptive Statistics GWR-EI Estimates of Racial Group Vote......................46
Table 2.2 Comparison of Aggregate GWR-EI Estimates of Racial Group Vote ..............52
Table 3.1 Latino Voters OLS and Spatial Lag Regression Models..................................84
Table 3.2 White Voters OLS and Spatial Error Regression Models .................................87
Table 3.3 Asian Voters OLS and Spatial Lag Regression Models....................................90
Table 3.4 Black Voters OLS and Spatial Lag Regression Models...................................93
Table 4.1 Estimated Racial Group Vote, Los Angeles 2005 ...........................................103
Table 4.2 Moran’s I Spatial Autocorrelation Statistic .....................................................107
Table 4.3 Latino Voters Global OLS and Spatial Lag Regression Models.....................116
Table 4.4 Latino Villaraigosa Voters Local Spatial Interaction Models ........................117
Table 4.5 Latino Hahn Voters Local Spatial Interaction Models ...................................118
Table 4.6 White Voters Global OLS and Spatial Lag Regression Models.....................120
Table 4.7 White Villaraigosa Voters Local Spatial Interaction Models..........................121
Table 4.8 White Hahn Voters Local Spatial Interaction Models....................................122
Table 4.9 Black Voters Global OLS and Spatial Lag Regression Models .....................125
Table 4.10 Black Villaraigosa Voters Local Spatial Interaction Models .......................126
Table 4.11 Black Hahn Voters Local Spatial Interaction Models ..................................127
Table 4.12 Asian Voters Global OLS and Spatial Lag Regression Models ...................129
Table 4.13 Asian Villaraigosa Voters Local Spatial Interaction Models .......................130
Table 4.14 Asian Hahn Voters Local Spatial Interaction Models ..................................131
vi
LIST OF FIGURES
Figure 2.1 Distributions of GWR-EI Estimated Dependent Variables, 2001 Election .....47
Figure 2.2 Distributions of GWR-EI Estimated Dependent Variables, 2001 Election .....48
Figure 2.3 Distributions of GWR-EI Estimated Dependent Variables, 2005 Election .....49
Figure 2.4 Distributions of GWR-EI Estimated Dependent Variables, 2005 Election .....50
Figure 2.5 Percentile Maps GWR-EI Estimated Racial Group Vote, James Hahn,
Los Angeles 2001 ..............................................................................................................53
Figure 2.6 Percentile Maps GWR-EI Estimated Racial Group Vote, Antonio
Villaraigosa, Los Angeles 2005.........................................................................................54
Figure 3.1 Spatial Empirical Bayes Smoothed Box Maps, Los Angeles 2001..................68
Figure 3.2 Moran’s I Scatterplot Graphs, Racial Group Vote Hahn .................................69
Figure 3.3 Moran’s I Scatterplot Graphs, Racial Group Vote Villaraigosa.......................70
Figure 3.4 LISA Cluster Maps, Racial Group Vote Hahn, 2001.......................................74
Figure 3.5 LISA Cluster Maps, Racial Group Vote Villaraigosa, 2001............................75
Figure 3.6 Geographic Regions of Los Angeles................................................................76
Figure 4.1 GWR-EI Estimated Racial Group Vote For Antonio Villaraigosa,
Los Angeles 2005 ............................................................................................................106
Figure 4.2 Geographic Regions of Los Angeles..............................................................111

vii
ABSTRACT
Scholars of voting behavior are increasingly turning to the contextual
environment for explanations for why people vote they way they do. Many scholars,
particularly those that are concerned with racial/ethnic and urban voting, are unsatisfied
with analysis that treats individual voters as isolated from their social networks of friends,
family, and neighbors. I argue that the concept of context and contextual effects should
be connected to particular places, such as urban neighborhoods. As a theoretical
foundation for this argument, I develop the concept of place-context. Place-context
channels the flow of political information and stimuli that influence voting behavior.
I investigate the debates concerning the best methodological techniques for the
study of place-context effects. Given the widely acknowledged problems associated with
survey research and racially polarized voting, I use spatially located aggregate data; the
dependent variables of interest percentage vote choice in each census tract by each of
four racial groups are estimated via a geographically weighted approach to ecological
inference that allows spatial dynamics to be take into account. The estimated data are
examined with spatial econometric methods that allow for the visualization and modeling
of behavioral diffusion through space.
The Los Angeles mayoral elections in 2001 and 2005, both of which involved
James Hahn and Antonio Villaraigosa, are analyzed for place-context effects on the
voting behavior of four racial groups. The findings indicate the presence of many
instances of neighborhood effects whereby voters were affected by the diffusion of
behavior from their neighbors. Additionally, the dynamics of racial group cooperation
viii
and conflict, as expressed through voting, is examined. The findings suggest that place
and racial context interact at the neighborhood level, such that, instead of general
findings of either conflict or cooperation, instead the results indicate the existence of
complex patterns of both cooperation and conflict diffusing through space from
neighborhood to neighborhood.
1
CHAPTER 1 THE POLITICS THAT PLACES MAKE: A REVIEW
OF THE LITERATURE
There are two different neighborhoods, each in different parts of a large city. The
neighborhoods differ in their racial, ethnic, economic, and educational makeup, but not
radically so. One neighborhood contains more white residents than the other one, and
fewer Latino residents. Both neighborhoods contain a small but significant percentage of
African-American and Asian residents. Both neighborhoods contain a mixture of
homeowners and renters, and the residents tend to be above the city average in terms of
income and education but neither is among the socioeconomic elite of the city.
One key difference between the two neighborhoods is that they tend to vote for
different candidates when it comes to local, citywide, statewide, and national elections.
Moreover, the residents of one neighborhood tend to vote at higher rates than the other.
What is the explanation for these differences in political behavior? To answer that
question we would delve deeper than social background characteristics. We would look
into individual attitudes and preferences such as political party attachment, ideological
self-identification, orientation towards the issues and events of the day, and differing
assessment of particular candidates. If we took all of the individual attitudinal
differences and controlled for socioeconomic characteristics, we would be able to explain
most of the variation in the way these neighborhoods tend to vote.
However, consider the following hypothetical situation. Entirely at random, half
the residents of each neighborhood are switched from one place to the other. As
2
aggregate units, will the voting trends of each neighborhood continue as before? Will the
residents who switched places continue to vote as they did before or will they vote
differently? According to the dominant models of political behavior the answer is no: the
voting patterns of the neighborhoods will not continue as before because the vote choices
of the individuals will not change depending on their place of residence. Those voters
who switched neighborhoods will continue to vote as the models would predict because
neighborhood differences have no effect on the vote decision. Although some studies do
find that clustering within a neighborhood can accentuate the effects of some variables
such as social status, the general consensus of studies of neighborhood effects on political
behavior is that neighborhood contextual environment is not a significant determinant of
the vote (Eulau and Rothenberg 1986; Foladare 1968; Huckfeldt 1979; Huckfeldt 1986;
King 1996).
Why is it reasonable to assume that a change in environment will have no effect
on the vote decision? Does a person’s situation within a specific place-based social
context have no effect on political behavior? Over three decades of social science
research into the “power of place” among geographers, sociologists, and historians—
research almost completely ignored by political scientists—makes it very difficult to
accept the proposition that the location of the voter is irrelevant to her voting behavior
(Agnew and Duncan 1989; Agnew and Smith 2002; Baybeck and Huckfeldt 2002; Dear
1990; Ethington 2001; Foladare 1968; Harvey 1990; Huckfeldt 1979; Kohfeld and
Sprague 2002; Lefebvre 1991; Soja 1989; Sui and Hugill 2002). So we need to ask: is
there anything about a neighborhood or locality, independent of and apart from the
3
individual characteristics of the residents who live there, that can affect political
behavior? What kind of effects, if any, can be measured? And, what is the causal
mechanism that explains the influence of place?
This dissertation attempts to answer those questions. The argument begins from
the position that within the political science discipline the understanding of place and its
effects on voting behavior has been limited. Utilizing a more sophisticated understanding
of place borrowed from the field of political geography, I argue that place can affect
political behavior in both the short and long-term. Despite the fact that voting is an
individual act that occurs in a private booth, voters do not vote in isolation from each
other or the place in which the vote decision is made. Place can be altered by the people
who exist there, but places can also change the behavior of the people who move through
them. Neighborhoods are durable institutions with different histories, cultures, and
norms of behavior. Place can also have a mediating effect on the way voters interact with
each other, perceive their social environment, and receive and process political
information. Additionally, some political geographers suggest that place can play
important roles in the formation of political identity (Agnew 2002; Forest 1995; Forest
2002; Wacquant 1994). This claim will be taken up in more detail later, but if
substantiated, it does illustrate that the way we think of place and its effects on voting
need to be expanded upon.
In order to support my argument I rely on several different bodies of literature. I
begin by asking the question: what is place, and how can it be used as an analytical
concept to improve our understanding of politics? To answer the question I look to
4
theories of political geography, in particular the work of Henri Lefebvre and John
Agnew. I then move to an exploration of the basics of voting behavior theory and how
place has been incorporated. That discussion leads us to the literature of contextual
effects, the only field within political science that is explicitly concerned with place and
political behavior. I apply the insights gained from political geography to a critique of
contextual effects research. I find that although contextual effects analysis is specifically
concerned with place (e.g. the neighborhood effect), the theories and methodologies
employed reflect a vague and under theorized conception of the connection between
context and place. I argue that more rigorously connecting context and place will offer
substantive insights into the ways that place affects voting behavior. Also, my study will
shed light on the lingering question of the causal mechanism of social context.
POLITICAL GEOGRAPHY: POLITICS IN ITS PLACE
Investigations into the interaction of place and political behavior are not new.
Political scientists have long been interested in the regional variety of politics in America,
with the work of V.O. Key being perhaps the best example of political research that is
conscious of the peculiarities of place and community (Key 1949). More recently, some
political scientists have turned to an examination of the contextual environment of
political behavior (Books and Prysby 1991; Carsey 1995; Eulau and Rothenberg 1986;
Huckfeldt 1979; Huckfeldt 1980; Huckfeldt, Plutzer, and Sprague 1993; Leighley 2001;
Oliver 2001; Oliver and Mendelberg 2000; Oliver and Wong 2003; Voss 2004). This
research often, but not always, contains analysis of political behavior as it is connected to
5
places such as cities, neighborhoods, and churches. Political geographers have come to
understand that “politics is everywhere” and examine political behavior at various
geographic scales, from the local to the global (Agnew 1996a; Agnew 2002; Flint 1996b;
O'Loughlin 2002; Ward and O'Loughlin 2002). Although they have not merged together,
within the past decade these two realms of contextual political research have come to
intersect and engage each other on theoretical, substantive, and methodological grounds
with beneficial results.
1

Given the extensive research into the connection between place and politics, why
are these two bodies of literature insufficient to answer the questions that I pose in this
dissertation? That is, why is further research necessary? To put it simply, despite
increasing interdisciplinary cross-pollination between the two branches of contextual
effects research, there is disagreement between many political scientists and political
geographers over issues of methodology and ontology. Political geographers often make
claims for an ontological divide: the study of individual political behavior cannot be
separated from its geographical location and social context.
Because the distribution of phenomena on the earth’s surface has been well
documented in thousands of studies and by simple observation, we know that
clustering of like objects, people, and places is the norm … geographers argue
that the dynamics of human interaction in communities of kindred individuals,
driven by needs of security and familiarity and/or by fears of the dissimilar, give
rise to a ‘contextual’ element that is more than simply the sum of the effects of the
community composition (O’Loughlin 2002, 219).

1
For examples see special issues of Political Geography, Volume 14, number 6/7 (1995),
Volume 15, number 2 (1996), and Volume 21(2002) and Political Analysis, Volume 10,
number 3 (2002).
6
O’Loughlin’s position regarding the impact of place on human behavior illustrates a
defensiveness among political geographers regarding assertions about the importance of
place and context. However, I would argue that rather than seeing an ontological chasm
separating the two disciplines, there is room for productive engagement along substantive
and methodological dimensions. Place is an integral part of political processes and
behavior, and its impact can be observed and measured. Therefore, questions about the
impact of place should be added to our investigations of political behavior. At the same
time political geographers’ conception of political behavior tends to be vague and limited
(Agnew 1996a; King 1996). For the most part, political geographers have not
substantively engaged with the established foundations of political behavior research.
From The American Voter to the contributions of rational choice theory and political
psychology, political geographers tend not to cite or build upon the scholarship of
political scientists. They have generally been content to map particular manifestations of
political behavior rather than attempt causal analysis that connects place and politics. I
argue that it is possible to bridge the theoretical and methodological disconnect by taking
seriously the conceptual framework of political geography and applying it to an engaged
critique of the political science literature of political behavior in order to make a creative
and innovate case for the substantive claims of the geographers: that place and
environmental context matter to study of individual behavior. Before reviewing the
foundations of political behavior research, I turn to an examination of the literature of
political geography and ask the question: what, exactly, is place?
7
Place, Space and Context
Throughout America one can hear the oft-repeated claim that the country is
changing for the worse, and it is often due to the irretrievable loss of “something special.”
The concern emanates from across the political spectrum, and usually manifests itself
around the changing nature of American communities. Many decry the homogenization
of America due to the proliferation of corporate franchises. Longtime urban dwellers
often lament the encroachment of processes of gentrification that alter the precious
character of their city. Planners and local policy experts point to the problems caused by
the seemingly implacable force of suburban growth and urban sprawl. Local politicians
and public intellectuals criticize the ways in which globalization and immigration are
fundamentally altering the cultural and economic landscape of America. There is a
feeling that the world is smaller, and that countries and cultures are more interconnected
than they ever have been. Location on the globe seems to matter less and physical
distance is unimportant. At the same time, many struggle to maintain a connection to
community, to those places that mean “something special” to them.
This debate helps to illuminate the meaning of place as an analytical concept
developed by scholars of human and political geography. In order to understand the
meaning of place it first must be distinguished from the concept of space. “The relative
distinctiveness of places can decline in the face of pressures for sameness, as when
supermarket chains, motel chains, fast-food restaurants reproduce the same images from
coast-to-coast” (Agnew and Smith 2002, 6). What is really being lamented is the loss of
8
place; the uniqueness of American communities is being paved over by a combination of
local and global economic processes. Space is often appropriated, shaped, commanded
over, and controlled. Its rough edges and exceptional diversity are smoothed away
(Harvey 1989). Instead of a collection of unique places, America is becoming one giant
homogenized space.
While geographical places are supposedly becoming less important, space –
spatial theory and spatial analysis – has become more important. According to Henri
Lefebvre, French social theorist of urban space, the concept of space has been largely
absent from Western thought. That is, physical, practical, social space has been absent.
Instead, Western thought – especially since Descartes – has developed in mental space,
philosophical space, and linguistic space. From mental space we get a profusion of so-
called spaces, all of which are detached from physical, social space. “We are forever
hearing about the space of this and/or the space of that: about literary space, ideological
space, the space of the dream, psychoanalytic topologies, and so on and so forth.
Conspicuous by its absence ... is not only the idea of ‘man’ but also that of [physical]
space” (Lefebvre 1991, 3). By separating mental and physical space, a duality is created
whereby mental/philosophical space is presented as the site of activity, agency,
knowledge, and power.
Conversely, physical space is presented as a neutral container of human existence.
Space “is equivalent, practically speaking, to a set of institutional and ideological
superstructures that are not presented for what they are ... alternatively it assumes an
outward appearance of neutrality, of insignificance, of semiological destitution, and of
9
emptiness (or absence)” (349). According to Lefebvre, it is erroneous “[t]o picture space
as a ‘frame’ or container into which nothing can be put unless it is smaller than the
recipient, and to imagine that this container has no other purpose than to preserve what
has been put in it” (94). It is an error to view space as a neutral container because that
obscures the social character of space and the way it is organized by political power.
“[S]pace is neither a mere ‘frame’, after the fashion of the frame of a painting, nor a form
or container of a virtually neutral kind, designed simply to receive whatever is poured
into it” (94). Space is portrayed as a natural and neutral realm, the realm of the abstract,
the general, and the homogenous. It lacks the imprint of particular human experience; it
is the plane of undifferentiated individuals without apparent agency subject to universal
causal influences.
Rather than the abstract generality of space, place is particular and specific. Place
is defined by the presence of human identity and social experience, In making the
distinction between space and place, political geographer John Agnew emphasizes human
experience:
Place represents the encounter of people with other people and things in space. It
refers to how everyday life is inscribed in space and takes on meaning for
specified groups of people and organizations. Space can be considered ‘top-
down,’ defined by powerful actors imposing their control and stories on others.
Place can be considered as ‘bottom-up,’ representing the outlooks and actions of
more typical folk. Places tend to be localized when associated with the familiar,
with being ‘at home.’ But they can also be larger areas, depending upon patterns
of activities, network connections, and the projection of feelings of attachment,
comfort, and belonging … [Place] is also the spaces of everyday life that continue
to be formed and reformed in a technologically and economically shrinking
world. (Agnew and Smith 2002, 5)
10
Place denotes specific locations that contain a wide array and unique configurations of
individuals, institutions, and social networks, all of which have a dialectical relationship
with the places in which they are grounded. The individuals and institutions transform
places through their actions. And, in turn, their behavior is influenced by place in
numerous ways over both the short and the long-term. Places are irreversibly
transformed by the removal of individuals and institutions. It follows that those people,
social networks, and institutions will be transformed by changing places.
Place and space are distinct and yet exist only in relation to each other. They are
“necessarily interrelated geographical scales which are always defined in reference to one
another” (Agnew 1996, 130). What connects the two analytical scales of place and
space, what Agnew calls “local” and “global,” is human activity that expresses a local
particularity of social relations, cultural norms, and behaviors, while also being subject to
more general, or global, causal influence. Agnew combines these two perspectives of
place and space together into a concept that he calls, variously, hierarchical-geographical
context, strong contextualism, or “context-as-place.” Agnew intends for it to be a richer
and broader conception of context that can bridge the desires of political scientists to find
universal and generalized causal influences, and the objective of geographers to call
attention to local particularity. “In a contextual view human action is seen as threading
out from the here and now of face-to-face social interaction into more extensive fields of
mediated interaction managed by institutions and organizations. In this way social
relations can be thought of as stretching over time and space yet linked to the concrete
production of individual attitudes and behavior” (131). A context-as-place perspective
11
preserves the theoretical connection to individual activity while allowing for multiple
causal forces to be considered at multiple geographic and temporal scales. “From this
point of view, context refers to the hierarchical (and non-hierarchical) ‘funneling’ of
stimuli across geographical scales or levels to produce effects on politics and political
behavior. These effects can be thought of as coming together in places where micro
(localized) and macro (wide-ranging) processes of social structuration are jointly
mediated” (132).
Agnew’s context-as-place is designed to integrate the geographies of place at
various scales – the local, the nation-state, and the global – into the study of human
behavior, particularly, in his case, the study of regional differences in Italian
parliamentary voting. Geographers are especially attuned to the ways that nation-state
territoriality and spatial boundaries shape structures of power. As such, Agnew is critical
of scholars who fail to consider and integrate the influence of forces at higher geographic
scales beyond the local. According to Agnew, these scholars are “local contextualists”
that are propagating a weak version of contextual thought, rather than Agnew’s preferred
“strong contextualism.” By rejecting “local contextualism,” Agnew makes the claim that
one should not study contextual effects unless they are integrated into a strong and
holistic hierarchical-geographic ordering of contextual influences at many different
geographic scales. Agnew not only seeks to deeply connect context to place, but also to
theorize a strong context-as-place that emphasizes the geographic interconnectedness of
behavior from the local to the global. Agnew sees context-as-place as a regime of norms
and practices that is non-severable from each particular and unique place.
12
This position is understandable from the perspective of a geographer that places a
theoretical primacy on geographic influences. But, it may be construed as a deterrent for
students of local political behavior that wish to incorporate rigorous concepts of place
and space into their analysis by quantifying the observable aspects of place. I argue that
those of us who are interested in neighborhood effects on voting can still benefit from
Agnew’s aim of connecting place and context. Those who are interested in contextual
effects on political behavior, especially local context, should strongly ground context in
place, rather than defining context as a vague concept that is too broad to have real
theoretical meaning. Researchers should take care to incorporate the observable
measures that are connected to the contextual environment of place. Therefore, for the
specific purposes of my research into the effects of place on voting behavior, I will use
the terms place-context or place-based context to differentiate my particular
understanding of place and context from Agnew’s theoretically more formal “context-as-
place.” Additionally, the term place-context serves to draw a distinction between my
work and that of other contextual effects scholars who are not specifically dealing with
issues of place and space.
What are the implications and benefits of a place-context approach to the study of
local voting behavior? First, contextual effects analysis suffers from a lack of common
understanding of what, exactly, context is. There are those who examine racial context
(Gay 2004; Kohfeld and Sprague 1995; Liu 2001), and those who examine social
networks as context (Eulau and Rothenberg 1986; Huckfeldt, Johnson, and Sprague
2005; Zuckerman 2005); at the same time other scholars look at partisan context (Ceaser
13
and DiSalvo 2006), while others examine economic context (Books and Prysby 1999).
This leads to confusion about definitions of context, about context is operationalized for
quantitative analysis, and how to interpret findings across various types of context. The
key insight of the political geographers is that context is rooted in particular places.
Place-context, rooted in localities, neighborhoods, or regions is a way to capture a variety
of contextual forces.
Second, a place-context approach requires accounting for the spatial
embeddedness of individuals, local organizations and institutions, and other political
actors. This differs markedly from the standard approach to political science analysis that
sorts individuals into social categories that are detached from place, such as can be found
in most research that use either survey data or aggregate census data. Recalling
Lefebvre’s spatial theory, Agnew explains: “the categorical approach [to political
research] suffers from a sort of ‘agnosia‘ or disorder of perception in which
representations of space only set boundaries for non-spatial processes. Space is thereafter
eliminated from theoretical consideration” (Agnew 1996a, 131). A contextual approach
should treat political processes as operating in and through space rather than external to
it. Further, while the contextual unit of analysis should still emanate from the individual,
it necessarily recognizes that the individual is situated within a geographic environment
of social and political connections. Political and social actors are mediated by the spaces
that they move through; the places they inhabit, and that they act upon. The interactions
of peoples, groups, politics and economic transactions, power and weakness are all
embedded in social spaces. Institutions of politics and society – institutions related to
14
local governmental authority, or to economic activity, arts, society and culture – are also
embedded in space and place.
Third, the everyday experience of place will shape the political behavior of
individuals at a variety of geographic and temporal scales, including the relatively short-
term. Political information, and attitudes towards issues and candidates are “funneled” to
the individual voter through processes that are not external to place-based context, but
rather integrally connected to it. Issues are experienced and evaluated by the voter
through the lens of place. Candidates make a variety of appeals to voters, not the least of
which is an understanding and experience of particular places, regions, and contextual
cultures and behaviors. Political information is often disseminated geographically, and
campaigns are highly cognizant of the geography of various communities to which they
target their appeals. Within the literature, these forces, which are commonly understood
to affect political behavior, are rarely situated in context.
How does a place-context perspective differ from that usually employed in
political science? In order to answer that question, I turn to an examination of the
development of political behavior research in political science and the associated
evolution of contextual effects analysis of political behavior. Does the literature situate
individual behavior in place-based context? Is there recognition that individual and
social group identity formation is influenced by place, and that it varies geographically?
Although the literature posits a wealth of forces and influences upon individual behavior,
is there recognition that those influences are funneled through place-based context, and
can vary from place to place, even at a relatively short time scale?
15
VOTING BEHAVIOR: LOSS OF CONTEXT
There is an increasing interest among political scientists in contextual
explanations of political behavior. Many of those scholars cite an interest in moving
away from a focus on socially isolated individuals – a paradigm which they believe is
dominant in political science – and moving towards a view of individuals that are situated
within networks of social and political connections that complexly influence voting
decisions. “Many [political scientists] now do emphasize the need for contextualized and
path dependent explanations. That, in turn, represents something of a retreat away from
generality and toward particularity, away from universality and towards situatedness, in
the explanatory accounts we offer for political phenomena” (Goodin and Klingemann
1996, 22). However, a tension exists between the contextual analysts and those more
mainstream scholars whose substantive research agendas do not involve a concern for
political context. The strain can be traced to the evolutionary leap made by the Michigan
researchers, a leap away from the sociological model developed by the Columbia
scholars towards a socio-psychological model based on survey research data. The
tension is caused by methodological and substantive differences that tend to relegate
contextual research to the sidelines of the discipline. However, I argue that there is a lost
tradition of thinking about political behavior connected as it is situated in context.
Specifically, there is a tradition of linking political behavior to particular places and
locales that has been disregarded. That tradition can be recovered and expanded upon by
incorporating a place-context perspective. In the sections below, I review the literature of
voting behavior and explore the lost tradition of place-based research. From that basis I
16
offer a critique of contemporary contextual effects research that builds on a connection to
place-context.
The modern discipline of voting behavior research begins with the Columbia
model as developed in the 1954 book Voting (Berelson, Lazarsfeld, and McPhee 1954).
The Columbia model of voting behavior distinguished two different models of voting.
The consumer preference model of voting – which likened voters to consumers choosing
between appealing products – was rejected in favor of the sociological model, which
focused on the connection between social group attachment and factors such as
education, income, social class, and religion (Berelson, Lazarsfeld, and McPhee 1954;
Niemi and Wiesberg 2001). In The American Voter, researchers from Michigan broke
from the social group perspective and established the standard for voting behavior
research through the sophisticated and elegant use of survey research tailored to
understanding the individual vote decision (Campbell et al. 1960). The Michigan model
concluded that voter attitudes and orientations towards political parties, issues, and
candidates determined vote decisions. The Michigan model is built around a “funnel of
causality” that extends from each individual vote decision back through time to
incorporate both short-term and long-term causal mechanisms. Among other things,
short-term effects include the influence of campaign advertising and exposure to the
opinions of friends and family. Long-term effects are seen as formative influences on
partisan identification, the primary vote determinant, and include socio-economic status,
sociological characteristics such as race, ethnicity, and religion, and the important
influence of parental factors (Miller and Shanks 1996). The “funnel of causality”
17
includes “region” at the mouth of the funnel as a long-term effect. The region in which a
voter lives affects educational and occupational possibilities, which affect social status.
Even more important, in the funnel of causality, region affects partisan identification.
This is especially true for regions that have historical ties to one party or another,
particularly the Democratic Party in the South. Regional variation is presented as a factor
in the variation of partisan identification around the country.
The appearance of The American Voter with its powerful application of individual
survey data was somewhat contemporaneous with an emerging critique of the use of
aggregate ecological data – such as U.S. Census data, which is more amenable to the
study of social groups, race and ethnicity, etc. – to make causal inferences about
individual political behavior (King 1997). In the wake of this critique the discipline
shifted increasingly towards a methodological and theoretical focus on individual voters,
often shorn of any context or social connection. What often goes unsaid is that there are
theoretical assumptions about causation underlying the focus on individual behavior and
the use of survey data. Several scholars, usually associated with contextual research,
have begun to criticize the theoretical attachment to the uncontextualized individual voter
(Cho 2003; Huckfeldt, Johnson, and Sprague 2005; Oliver 2001; Voss 2004; Zuckerman
2005). The behavioral revolution – the use of survey data to explore determinants of
individual behavior –started political science down the path, and Downsian rational
choice theory accelerated the pace, whereby causation in political behavior research was
a product of atomistic individuals. “Attitudes about candidates, policies, and issues
proximate to the vote obtain theoretical primacy. Calculations about asocial self-interest
18
predominate. The explanatory principles of the social logic of politics recede into the
analytical distance” (Zuckerman 2005, 16).
Despite the emerging critique of mainstream political behavior research,
practitioners of contextual research have often been forced to be defensive with respect to
their theory, methods, and area of interest, and much effort has been spent justifying the
idea that “context matters” (Agnew 1996a; Agnew 1996b; Flint 1996a; King 1996). One
strategy, often used to mount the defense, is to attempt to re-connect political behavior
research to the Columbia school sociological model of politics. In a recent book, The
Social Logic of Politics, the editor points out that the study of political behavior in
political science started out from the position that was consistent with the principles of
“the social logic of politics” (Zuckerman 2005). The importance of social group
attachment and “the complexity of social ties” as influences on voting was a primary
theme of Berelson, Lazarsfeld, and McPhee’s sociological study, Voting (Berelson,
Lazarsfeld, and McPhee 1954). However, The American Voter relegated “the social logic
of politics” – the social context and social interaction among family, friends, and
neighbors – into the background. The Michigan scholars even went so far as to define
social group attachment in terms of individual perception. They “conducted surveys that
examine individuals but ignore the members of their social circles, and they transformed
social groups into objects of individual identification” (Zuckerman 2005, 11).
I am sympathetic to the critique developed by the contextual scholars that political
behavior research is focused on causation that derives from socially isolated individuals,
and I align myself with the substantive agenda of their research. However, I believe that
19
the authors of The Social Logic of Politics are misguided in dismissing the Michigan
model of the “funnel of causality” and returning to the sociological model of political
behavior. The dichotomy is not between social context versus individuals, but rather
about harnessing our understanding of individual voter behavior while incorporating a
rigorous concept of context. In fact, I would point out that the Michigan model of voting
behavior did not entirely ignore context. There is a rich vein of contextual theory located
within The American Voter that has been almost entirely ignored by the discipline; a
substantive vein that can be developed using a place-context perspective. The key to
uncovering the importance of context in The American Voter is to focus on place.
Campbell, Converse, Miller, and Stokes consider several ways that place variation can
influence the individual voter. For instance, formal institutional rules and informal
cultural norms and rules of behavior vary from place to place, and will have a varying
effect on the vote decision:
Other more elusive forces requiring conception in purely political terms shape
behavior as well. We shall complete our discussion of the political context of the
voting act by considering … the rules and norms for behavior that vary by
political sub-community across the national electorate. In some instances these
norms have become institutionalized in the legal regulations that circumscribe
political participation from area to area; in others they remain informal matters of
community sentiment, reflecting local differences in political tradition. (Campbell
et al 1960, 119)

Generally, Campbell et al. consider place variation and social context to have a long-term
effect upon voting behavior. From the quote above we get a picture of place-variable
behavior as a product of long-term historical and cultural forces. The long-term effect is
similar to the way the “funnel of causality” considers demographic background.
20
Nonetheless, their examination of these long-term contextual factors leads them to the
conclusion “that electoral behavior can be understood only through a comprehension of
the context in which it occurs” (289).
Campbell et al., comment further on the importance of place by considering
population movement in the United States. “The winning of the West, the growth of the
city, and the rise of suburbia have contemporary as well as historical interest for us in our
present analysis” (442). In an extended exegesis about urban population shifts that
echoes the my hypothetical at the start of the chapter, Campbell et al. suggest a line of
research for future inquiry having to do with urban versus suburban residential and social
mobility. The authors speculate about specific place differences having a constructive
effect on political behavior and vote choice through long-term cultural norms and familial
connections:
It may be that Republicans, despite their own metropolitan origins, are more often
linked through enduring family ties to ancestral beginnings in small cities, towns,
and on farms. Although at least one generation removed, they may reflect the
consequences of an earlier era of urbanization and central city growth. With a
heritage of values and traditions congenial to non-metropolitan life they may seize
the opportunity created by occupational success and escape from the metropolis.
The urban Democrat, on the other hand, may be the child of a thoroughly urban
culture. Whether the metropolis of his family line once was Warsaw or Rome or
Dublin, or Boston, Baltimore, or New York, he may accept more often the way of
life of the metropolitan center. For him occupational success may provide the
means to exploit and enjoy the treasures of the city, not freedom to leave the
familiar for the unknown. (467)

This line of thought does suggest that the Michigan paradigm reduces the experience of
place, the variable experience of different cities, into a product of individual attitudes,
rather than as part of a social experience of place context. Yet there is very much a sense
21
that political behavior is embedded in place, which will have a direct effect on voting
behavior. “The American political system … embraces a variety of political sub-
communities. Each of these communities is a pervasive medium within which behavior
must occur. And each leaves some characteristic impress on that behavior” (266). This
line of reasoning embodies the concept of context-as-place outlined by Agnew; place and
space should be incorporated as more than just a neutral medium of political behavior.
But the authors do not follow through with a full evocation of a theory of context
and place. They admit that they do not have the data to fully explore the implications of
their theory in regards to place context. The authors speculate about Republican
newcomers to the city and the effect on urban politics.
Even assuming that these Republican newcomers may be unable to achieve
complete insulation from the politics of their new neighbors, social integration
(and the subsequent acceptance of community norms) appears to be a very slow
process. It is a matter of critical significance for metropolitan politics, however,
whether these Democratic influences may in time make themselves felt. (471)
They doubt the existence of short-term effects of place variation on voting behavior. Yet
despite this doubt, they do present the possibility of place variation, and suggest the
possibility of future debate. Campbell et al believe that place variation, especially urban
versus suburban, will be a slow process having long-term effects on voting behavior; yet
the possibility exists for short-term behavior to be influenced by place variation. This
question about the possibility of contextual influences in the short and long term is one
that was largely abandoned by mainstream political behavior research. It was left to
those interested in contextual effects to take up the question.
22
CONTEXTUAL DETERMINANTS OF POLITICAL BEHAVIOR
In the pages of The American Voter, interest in the impact of the contextual
environment on voters is connected to places such as cities and urban ethnic
neighborhoods. Although Campbell, Converse, Miller, and Stokes did not themselves
pursue contextual research, their work did inspire the work of early contextual effects
analysts such as Robert Huckfeldt and John Sprague. However, as I will show below,
contextual effects research within the political science discipline has lost its connection to
places. Place-context is not considered part of the causal process, but rather treated as a
surrogate for other variables, such as the racial and ethnic composition, which are
considered to be independent of place. Because of this, contextual effects research
suffers from a lack of conceptual consensus about the definition of context. Instead a
variety of vague definitions of context abound within the discipline: ranging from the
most common definition of social interaction with friends and neighbors (Achen and
Shively 1995; Huckfeldt 1986; Huckfeldt, Plutzer, and Sprague 1993), to the
psychological perception of the partisan environment (Burbank 1995), the objective
racial and ethnic composition of census tracts (Oliver and Mendelberg 2000; Oliver and
Wong 2003), the characteristics of racial and ethnic conflict or cooperation in urban areas
(Kaufmann 2003), the performance of the state and local economy (Books and Prysby
1999), and the size of particular suburban cities (Oliver 2001; Oliver 2003). I argue that
it is necessary to consider context connected to place in order to effectively capture the
variety of contextual forces that determine local voting behavior.
23
In general, contextual effects are seen as external influences on political behavior.
Agnew criticizes this approach to contextual analysis because it fails to consider the
particularity of place; the unique configurations of social and political relations and
contextual influences that vary from place to place. “Contextual effects are usually seen
as external effects on individual behavior arising from social interaction within an
environment … but the hierarchical-geographic structuring of that environment is not
examined” (Agnew 1996a, 133). Theoretically and methodologically, contextual effects
analysis by political scientists tends to treat geographic place as a neutral container rather
than as a rich medium that constitutes and structures political behavior through a variety
of contextual processes at specific localities. In this way, geographic place is treated as a
surrogate or placeholder for other, more important variables such as race/ethnicity or
socioeconomic status. Place is treated as an “epiphenomenal or residual influence –
either a superficial representation of a deeper set of influences (geography as merely a
surrogate) or a trivial influence, to be taken into account when all others have been
considered” (Johnston and Pattie 2005, 187).
Still, I argue that contextual effects analysis is almost always nominally or
implicitly concerned with the effects of place. One of the founders of modern contextual
research, Robert Huckfeldt, started from a belief in the importance of urban
neighborhoods, particularly the racial and ethnic composition and social dynamics of
assimilation or conflict within those neighborhoods (Huckfeldt 1979; Huckfeldt 1980;
Huckfeldt 1986). Yet, Huckfeldt never grounded his contextual research in a theory of
geographic place, and his research moved beyond a concern for geographic places and
24
urban neighborhoods. His theory of context directs attention almost entirely toward the
influence of social and personal networks, friendship groups, and family connections
(Baybeck and Huckfeldt 2002; Huckfeldt, Johnson, and Sprague 2005; Zuckerman 2005).
The causal mechanism that Huckfeldt uses to explain the influence of context is social
interaction and personal contact within various network contexts, including home, work,
and church. Social interaction networks structure the flow of political information and
have a determining affect on resulting political behavior. Huckfeldt acknowledges that
geographic space can have a bounding effect on the particular configurations of social
networks, but he emphasizes that social space – the variety of personal and social
network connections – is not coincident with geographic space (Baybeck and Huckfeldt
2002; Huckfeldt, Johnson, and Sprague 2005). Huckfeldt’s emphasis on social
interaction and the flow of political information within networks has produced valuable
findings about social context. Yet, placing theoretical emphasis on social interaction as
the dominant cause of contextual effects leads Huckfeldt away from the ontological
connection between context and place. The result is the neglect of non-social interaction
contextual forces that are all geographically situated to a greater or lesser degree. For
instance, there are a multiplicity of contextual forces that are tied to particular places such
as the way evaluations of candidates, issues, and events are mediated by place, the way
political information is channeled through local media, and the way political, civic, and
governmental institutions educate and target local voters.
Books and Prysby (1991) make a similar criticism of Huckfeldt’s focus on the
context of social interaction. Through their review of contextual effects research, they
25
develop a theory of – and suggest a methodological model for researching – contextual
effects. Their theory also focuses on the way that social context interaction is an
important component that structures the flow of political information. Yet, by locating
context geographically they are able to move beyond social network interaction to
examine multiple ways in which voters receive political information that influence their
vote choice. For instance, “the information received by a voter during an election
campaign – whether through grassroots party activity, paid media advertisements, news
coverage, or other sources – may depend in large part on where the voter lives” (Books
and Prysby 1991, 34). Books and Prysby also include the contextual influence of local
institutions, which is again something neglected by a social interaction-based theory of
context. Local institutions – governmental, social, or political – will vary from place to
place, and so their contextual influence will vary from place to place. A final advantage
to the approach outlined by Books and Prysby is attention to structural factors, such as
the density of political organizations, or the activity level of a particular political party
organization. “Structural properties like these cannot be measured simply by adding
individual characteristics. They must be detected by observing the context itself” (Books
and Prysby 1991, 65).
Huckfeldt criticizes Books and Prysby for going too far towards a geographical
model of context rather than the more pliable concept of “social space” that Huckfeldt
uses by concentrating on geographical units of analysis (Huckfeldt 1992). However, I
believe that Books and Prysby do not go far enough towards situating context in
geographic space and place. The theory of context developed by Books and Prysby is
26
most consistent with the concept of context-as-place outlined by Agnew, but it should be
expanded upon in order to fully capture the place-based component of local voting
behavior. Books and Prysby locate context geographically, which allows them to account
for the many place-based contextual influences. Also, a geographically based theory of
context is sensitive to the forces of time. Places change over time, and Books and Prysby
pay close attention to the temporal dynamics of changing contexts. Yet, the theory still
treats contextual effects as external influences on political behavior rather than as
political processes that are inherently spatial and integrally connected to particular places.
For instance, Books and Prysby account for the contextual variation in racial group
composition of neighborhoods. However, they do not consider how the everyday
experience of place and space may cause interesting variations in racial group behavior.
Nor, do Books and Prysby consider how the behavior of racial groups in one
neighborhood may influence their neighbors in varying ways. That is, behavioral effects
may move through space and place. A theory of context that treats place as more than a
neutral medium for a variety of external contextual effects will fail to consider the ways
in which place influences the everyday lived experience of voters and, thereby, influence
the development of their political attitudes. Place-based context incorporates a process of
feedback between individuals and groups and the places they inhabit. Treating
contextual effects as external to place fails to adequately theorize the iterative feedback
processes of place context.

27
CONCLUSION: OUTLINE OF THE DISSERTATION
At the beginning of this chapter I proposed a hypothetical thought-experiment that
brings to light a number of questions regarding the influence of place and space on urban
political behavior. In this chapter, I have established that there is reason to believe that
place matters as a determinant of voting behavior. I have traced the way in which place
and politics have been analyzed in political science, and established that there is a
foundation for place-context research located within the mainstream of the discipline. In
order to investigate the impact of place, I develop the concept of place-context, which, I
argue, will provide a productive basis for future contextual effects research. Place-
context structures the various and often vaguely defined types of context that are studied
by political scientists, and allows the researcher to situate voters and other political
participants in concrete places.
In chapter two, I inquire into the best method of studying place-context and voting
behavior. In order to answer that question, I delve deeper into Agnew’s contention that a
place-context perspective entails an ontological understanding that is different from that
of most voting behavior research. The ontological debate leads to an examination of
various methodological strategies for studying place-context. I argue that a method of
ecological inference combined with spatial econometric data analysis is the best approach
to studying place-context.
Chapters three and four are applications of the theory and method described above
to cases of voting behavior in Los Angeles. Chapter three is an analysis of the 2001
mayoral election in which I look for the independent effects of place as a vote
28
determinant. In chapter four, I extend the analysis to the 2005 mayoral election, and
examine the interaction of place and race as determinants of racial group voting behavior.
I conclude the dissertation in chapter five with a discussion of the implications of
my findings regarding place-context for the larger fields of urban politics and political
participation. I also discuss the areas where further research is required, including
analysis of the causal mechanism of contextual effects, more detailed analysis of the
ways in which local institutions shape place-context, and the role of self-selection as an
alternative explanation for contextual effects.

29
CHAPTER 2 PLACE-CONTEXT AND POLITICAL BEHAVIOR:
ONTOLOGICAL AND METHODOLOGICAL
DEBATES
*

In Chapter One, I argue that there is good reason to believe that place and space
have been inadequately considered within the discipline of political science, and that
place-context is an important factor in the voting behavior of city dwellers. However,
even when political scientists are interested in contextual effects, the concept of context
tends to be vaguely and variously defined. Context, whether the researcher is studying
racial attitudes, demographic composition, socioeconomic environment, or the partisan
milieu of a community, begins from a connection to particular places. Place-context
shapes the way voters receive and evaluate political information, including issue and
campaign effects, and the way voters form attitudes about particular candidates.
There are certain scholars of political behavior in particular that are turning to the
study of context and contextual effects. Those scholars tend to be unsatisfied with the
individually focused causation that is the hallmark of political behavior research since the
advent of the Michigan school. They are motivated by a sense that individual behavior,
in addition to being a product of individual decisions and preferences, is also subject to
causal forces that are rooted in the contextual environment. It should be noted that
variables rooted in the contextual environment might be difficult to capture with surveys

*
Portions of this chapter appear previously in Ethington, Philip J., and Jason A.
McDaniel. 2007. “Political Places and Institutional Spaces: The Intersection of Political
Science and Political Geography.” Annual Review of Political Science 10:127-142.

30
of individual attitudes. Because of this, contextual effects researchers often turn to
aggregate data as a more accurate representation of context. Yet, contextual effects
analysis tends to be undertaken from an intellectually defensive standpoint that is related
to the inadequate conceptualization of context within political science, as well as
disciplinary critiques of methodological practice of using aggregate data. As I argue in
Chapter One, I believe that contextual approaches to political behavior should be rooted
in place-context theoretical orientation, and analysis of the various “types” of context
(racial, economic, partisan) should incorporate the way these various contextual
influences flow through and arise out of connections to particular places. In this chapter,
I explore the questions of methodology that surround contextual analysis, especially
related to the proper use of aggregate ecological data such as that collected by the U.S.
Census Bureau. However, as I review in more detail below, this practice has been
criticized as a form of the ecological fallacy – incorrectly inferring individual behavior
from aggregate data. Many political geographers respond to this criticism based on their
belief in an holistic ontology of place and space that eludes the practice of quantitative
analysis. In the next section, I review the ontological debate and the associated
methodological debate. In the process I argue that debates about ontology are misguided
in that they close off areas of scholarly inquiry; moreover, advances in methodological
practice allow us to further consider substantive questions of place-context and political
behavior without retreat to an exclusionary ontology. Once I make that argument, I will
detail my methodological approach to the study of place-context effects on urban voting
behavior.

31
PLACE-CONTEXT AND ONTOLOGICAL INDIVIDUALISM: THE AGNEW-
KING DEBATE
The debate over whether or not context matters, and if so, how best to study it, is
represented by three distinct positions. The first position, developed mainly by political
geographers, holds that context matters, and that its effects cannot be easily reduced to
quantitative measures, especially those associated with typical public opinion surveys
(Agnew 1996b). The second position, represented here by Gary King, asserts that
context does not matter based on previous research findings, and that, moreover, it should
be the focus of serious scholarship (King 1996). However, I argue that a third position
exists, one that argues for the importance of context, but does not necessarily accept the
ontological position of political geographers. Instead, creative methodological
techniques, such as social network analysis (Huckfeldt, Johnson, and Sprague 2005) and
spatial econometrics (Cho 2003) can be used to push through the ontological veil.
Political geographer John Agnew suggests that his concept of context-as-place
derives from an ontological perspective that is different from that normally taken within
the political science discipline. Agnew’s position is neatly captured in a debate with
noted political scientist Gary King that takes place in the pages of the journal Political
Geography (Agnew 1996a; Agnew 1996b; Flint 1996b; King 1996). Political
geographers, according to Agnew, focus “on trying to discover the ways in which
geography ... mediates between people, on the one hand, and political organization, on
the other” (Agnew 2002, 12). This research agenda leads to a concern for the context –
geographical, social, and political – in which politics happens and political behavior takes

32
place. “There is a persisting tendency to insist that politics cannot be adequately
understood without understanding the geographical contexts in which it takes place, from
global geopolitics at one end of the scale to local politics at the other” (Agnew 2002, 12).
Political issues and government policy, Agnew argues, have geographically variable
impacts, and political messages are often effectively targeted and shaped according to
variations in political and human geography. “In summary, the hierarchical-geographical
context or place channels the flow of interests, influence and identity out of which
political activities emanate. This approach assumes, therefore, that political behavior is
inevitably structured by a changing configuration of social-geographical influences as
global-local connections shift over time. The configurations of causal influences all
relate back to the historical geography of the world economy at any particular time.
However, because of differences in prior experiences no place can be reduced to them”
(Agnew 1996a, 133).
Gary King, in his response to Agnew, challenges the usefulness of contextual
research. “Political geographers should not be so concerned with demonstrating that
context matters" (King 1996, 160). King argues that context rarely does matter, and that
scholars should not be concerned to show that it does matter. "For social science
phenomena we understand, political context rarely makes a huge difference … who you
talk to, the types of people you live near, the nature of your community, your political
geography, all have some effect on the vote and on political opinions, but all empirical
evidence seems to indicate that the effect is relatively small. The geographic variation is
usually quite large to begin with, but after we control for what we have learned about

33
voters, there isn’t much left for contextual effects. So, in a narrow sense, geography
matters, but contextual effects do not” (King 1996, 160).
Agnew replies that King is reducing the concept of context and the study of
contextual effects to a simplified form of the “neighborhood effect” that has been studied
periodically in political science literature (Foladare 1968). While much more
sympathetic to the work of contextual effects scholars than King, Agnew also criticizes
research that neglects the integration of context and place. Agnew calls this form of
contextual analysis “local contextualism” and criticizes it for being intellectually
defensive. Practitioners of local contextualism, in Agnew’s view, treat the local context
as separate from the global or (macro) context. Moreover, contextual effects are
aberrational in that those affected by context are seen as deviations from the behavioral
norm. Contextual effects, then, are seen as external effects acting upon the neutral plane
of space and place. Context is seen as another cause, another explanation of individual
behavior. Context-as-place, on the other hand, understands context as integral to the
formation of political identity and political behavior. Context-as-place cannot be
separated so easily from political behavior.
At bottom, the Agnew-King debate reflects a fundamental disagreement on
underlying premises. Agnew finds King to be “dismissive of the geosociological element
in [his] argument. King’s alternative is ontological (and methodological) individualism”
(Agnew 1996b, 165). The “point is precisely that we can never satisfactorily explain
what drives individual choices and action unless we situate the individuals in the social-
geographical contexts of their lives … These include much more than the neighborhood

34
effects King wants to reduce context to. Above all, they involve paying close attention to
the geographical levels or scales that frame explanation. In other words, the causes of the
political beliefs and actions of individuals are organized geosociologically” (Agnew
1996b, 165). Agnew is making a strong claim for the importance of context-as-place by
pointing out that King is making an unacknowledged ontological argument that is at odds
with the way political geographers understand the causal forces of context-as-place.
Agnew is arguing for the idea that the very elements of place are non-severable the
behavior of individuals and organizations embedded in unique spaces. Unique
combinations of place related elements that cannot be reduced and isolated in regression
analysis.
While I agree that it is important to point out the ontological underpinnings of
King’s refutation of context, I do not think that taking place and context seriously
requires making ontological claims for the absolute uniqueness and particularity of places
such that place-context effects cannot be meaningfully measured and incorporated into
creative quantitative studies of contextual effects. The debate between Agnew and King
should not be thought of as an unbridgeable gap of ontological individualism versus a
holistic ontology of place and space. Rather, the real divide is one of causation; political
geographers make claims for unique and complex causal effects of context that are more
than the sum of the separate, measurable variable factors. While I agree that it is
important to point out the ontological underpinnings of King’s refutation of context, I do
not think that taking place and context seriously requires making ontological claims for
the absolute uniqueness and particularity of places such that place-context effects cannot

35
be meaningfully measured and incorporated into creative quantitative studies of
contextual effects.
At the same time, King does make overly strong claims for the unimportance of
context and contextual variables. There is room for those who are unsatisfied with the
kind of uncontextualized, individual-focused analysis that predominates to make
substantive arguments about the importance of place to political behavior by using
observable place-context variables. We can do a better job of incorporating the
observable dimensions of place that capture the complexities of place and space in order
to enrich our correlation and regression models of individual political behavior. There
may still be an ontological realm where the complexity and particularity of place and
space cannot be properly understood no matter how creative the methodological
approach. But, our task is to push our models and methodological approaches as far as
we can in order to gain substantive insight into the complex ways that place-context
influences political behavior.
For some researchers that will mean devising surveys that attempt to capture
contextual elements (See for instance Baybeck and Huckfeldt 2002, Huckfeldt, Johnson,
and Sprague 2005, and Johnston and Pattie 2005). But, other researchers will choose to
delve into the myriad of aggregate data and ecological analysis in order to investigate
contextual effects (See for instance Gimpel and Cho 2004, Liu 2001, and Oliver 2000).
However, for those who choose to pursue ecological analysis, there are methodological
hurdles that must be overcome. In the next section, I turn to a review of the
methodological critique of ecological analysis.

36
METHODOLOGY: THE ECOLOGICAL FALLACY, SURVEY DATA AND
AGGREGATE DATA ANALYSIS
In addition to the debate over ontology and causation, contextual effects scholars
often face questions and methodological critiques related to the choice to use aggregate
data in their political behavior research. In recent years scholars are finding publishing
outlets for their work on context and political behavior (Liu 2001; Lublin and Voss 2002;
Voss 2004). However, despite the progress made by contextual effects researchers, there
is still a debate regarding the ecological fallacy and the use of aggregate data to make
inferences about individual behavior rather than more accepted methodological approach
that sees finely crafted opinion survey data as the best way to research individual vote
choices. In this section, I outline the debate over the use of aggregate data. I show that
the critique of ecological data analysis has shaped the development of contextual effects
research away from place-based analysis towards a less geographically bound emphasis
on “life space” and social networks.
The appearance of The American Voter heralded a behavioral revolution within
the study of political behavior. The Michigan scholars developed a model based on
surveying individual social and psychological factors that emphasized partisan
attachment as the primary determinant of the vote choice (Campbell et al. 1960). The
success of the Michigan model resulted in a loss of interest in contextual and regional
variation in favor of broad national surveys. Contextual effects scholars offered an
alternative that harkened back to the work of the Columbia school of voting behavior and
the work of V.O. Key. Much of this work began as an investigation of the interaction

37
between places (neighborhoods usually) and political behavior, and evoked Irving
Foladare’s study of neighborhood effects on voting behavior (Foladare 1968). Robert
Huckfeldt, in particular, focused on urban neighborhoods as the starting place of social
context (Huckfeldt 1979; Huckfeldt 1980). In these studies Huckfeldt combines
aggregate and individual survey data to situate individual political participation within
neighborhood social context. Huckfeldt’s findings indicated that neighborhood social
context enhanced political participation, depending upon social status.
However, the use of aggregate data to infer individual behavior has been widely
criticized (Duncan and Davis 1953; Goodman 1953). This criticism, of course, was one
of the primary reasons for the decline of ecological analysis, and the rise of individual
survey analysis. Therefore, contextual effects scholarship, as practiced by Huckfeldt and
others, proceeded cautiously in the use of aggregate data. The use of aggregate data to
capture individual relationships and infer contextual social interaction presented a
problem, in that the causal mechanism of context had to be assumed rather than be
conclusively proven (Eulau and Rothenberg 1986). It was assumed that behind all of the
aggregate data was a myriad of social interactions at the neighborhood level that could be
observed and used to infer contextual effects on political behavior. Eulau and
Rothenberg, sympathetic to contextual analysis, responded to the methodological
criticism by making a theoretical distinction between social context as individual-based
social interaction networks and the physical environment, which had the connotation of
place. They constructed survey data that captured interpersonal social interaction. They
found that neighborhoods as places had no impact on political behavior, but rather

38
neighborhoods as defined by social networks – that is, as unconnected to actual places –
influenced political behavior and vote choice. Context was defined, therefore, as “life
space,” and the casual driving force of “life space” was social interaction within personal
networks. In response to this, Huckfeldt, who has become the leading political science
practitioner of contextual effects analysis, moved away from a geographical focus on
neighborhood context, towards a social network model (Baybeck and Huckfeldt 2002;
Huckfeldt 1992; Huckfeldt, Johnson, and Sprague 2005). Huckfeldt has developed a
variety of sophisticated methods to study context as social interaction within personal
networks. His findings provide strong evidence for the importance of social interaction,
and the flow of information through social networks, as an influential force for political
behavior. However, by following Eulau and Rothenberg, and focusing almost
exclusively on social interaction as the causal mechanism of context, Huckfeldt has
moved away from a concrete sense of space and place.
Given that I am interested in investigating contextual effects that are connected to
specific place-contexts rather than assuming that social interaction is the only mechanism
driving contextual effects, what is the best method for studying the impact of place-
context on voting behavior? Books and Prysby (1991) review the state of methodological
practice and offer a strategy for correctly modeling contextual effects. In their review
they point out that the state of the art in contextual effects modeling, best exemplified by
Huckfeldt, only captures social interaction effects. They note the need to expand
contextual effects modeling to capture effects not based on social interaction. Their
conclusion is that a structural equation model based on Erbring and Young (1979) is the

39
best approach because it is dynamic with respect to change over time, and treats context
as an endogenous feedback process. However, their model is not dynamic with respect to
geography. There are several examples of creative multi-level and spatial econometric
models that capture place-based dynamics. For instance, Johnston and Pattie (2005)
elegantly use data from the British Election Survey in addition to aggregate data to create
a multi-level model of voting behavior that finds significant place-based contextual
effects. Cho (2003) uses spatial econometric methods in her research into spatial
contagion effects among Asian-American political contribution networks.
Although survey data have been successfully utilized by leading contextual
effects scholars, I believe that individual-level vote choice data estimated via ecological
inference, and analyzed for place-context variation using spatial econometrics is the best
approach. This approach begins with Gary King’s (1997) A Solution to the Ecological
Inference Problem. The ecological inference problem, commonly known as the
ecological fallacy, occurs “when one is interested in the behavior of individuals, but has
data only at an aggregated level (e.g., precincts, hospital wards, counties). This data
limitation creates a situation where the behavior of individuals must be surmised from
data on aggregated sets of individuals rather than on individuals themselves” (Cho and
Manski 2006, 1). In his book, King develops a statistically complex, Bayesian
influenced, model of ecological inference that is a significant advance over previous
methods. King’s EI model has provoked criticism, but with the vigorous defense of its
author, is increasingly accepted as the best method for inferring individual-level behavior
from aggregate data.

40
Stephen Voss (2004) strongly defends King’s EI model, and argues that there are
several reasons why it is the best method for studying contextual effects. First, Voss
points out that estimates generated with King’s EI compare quite favorably to existing
real world data.
3
Second, Voss points out that King’s EI model is particularly useful in
studying those questions for which survey data have been shown to produce problematic
results. For instance, surveys have been shown to produce unreliable results in cases of
racial polarization (Barreto et al. 2006; King, Rosen, and Tanner 2004). Finally, Voss
points out that King’s EI model is particularly suited to separating contextual effects from
the demographic composition of a voting precinct, because it can easily incorporate
statistical controls for race/ethnic composition.
Although acknowledged as a breakthrough for ecological analysis, King’s EI
model has been the subject of criticism among methodologists. One of the primary
methodological criticisms of King’s EI comes from practitioners of spatial econometrics,
who point out that one of the primary assumptions of King’s EI model is that there be no
spatial autocorrelation in the data. Cho points out that this is very unrealistic in practice
because spatial autocorrelation is a common feature of real world aggregate data (Anselin
and Cho 2002; Cho 1998). In response, King claims that the consequences of violating
the assumption “are not very serious” (King 1997, 164). However, Anselin and Cho
(2002) and Calvo and Escolar (2003) provide compelling evidence to show that King’s
EI produces biased estimates in the presence of spatial autocorrelation.

3
Methods of ecological inference are often tested against voting data from Louisiana,
which, unlike most states, requires the collection of race/ethnicity information when
voting.

41
To counter the problem of biased performance in the presence of spatial
autocorrelation, Calvo and Escolar develop a geographically weighted autoregressive
approach (GWR-AR) to ecological inference that relies upon the geographically weighted
regression (GWR) method of Fotheringham, Brundson, and Charlton (2002). Just as in
King’s EI, GWR produces a local parameter estimate for each observation of a dataset.
However, GWR uses distance weights to account for the spatial variation among
variables, which produces estimates that weight more heavily those observations that are
closer to each other. Calvo and Escolar’s GWR-AR procedure is used to produce a
parameter estimate that captures the spatial structure of the data. This parameter can be
included in King’s EI (GWR-EI) in the form of a spatial covariate that properly accounts
for the spatial autocorrelation present in the data, and thus produces estimates that are
unbiased to spatial autocorrelation. The use of the spatial covariate fits into what King
calls the “extended model” of ecological inference, which King suggests can often be
used to produce better estimates than the simple EI model.
Another persistent criticism of King’s EI is that the individual-level estimates
should not be used as dependent variables in linear regression models, because it can,
under certain circumstances produce biased estimates (Herron and Shotts 2003).
Additionally, Herron and Shotts criticize the practice as logically inconsistent because
King’s EI assumes no aggregation bias, yet “a linear regression which uses King-based
point estimates as dependent variables … can imply the existence of aggregation bias”
(2004, 173). King vigorously defends his procedure by pointing out that the critique was
anticipated in his 1997 book presenting the model. Using the estimates in a linear

42
regression model is not the best solution, according to King, who prefers graphical
investigation of the data using scatterplot graphs and what he calls “tomography plots”
(King 1997; King and Adolph 2003).
Nonetheless, King points out, and Herron and Shotts agree, that there are two
relatively simple options to get around the problem of regression models using King’s EI-
based estimated dependent variables producing biased results (King et al. 2003). The
first, and best, is to use the “extended model” of King’s EI by including a covariate

in all
stages of the ecological inference process. The extended model can produce tighter
estimates because of the inclusion of additional information via a covariate. The second
option, which King believes is not always necessary, is to use Weighted Least Squares
regression instead of Ordinary Least Squares.
Given the benefits of ecological inference, and the vigorous effort to engage with
and respond to its critics, King’s EI model is increasingly being accepted as a valid
component of the methodological toolkit. This is evidenced by the fact that many
political scientists have produced peer-reviewed publications using King’s EI model.
Burden and Kimball (1998) were one of the first published studies to use first-stage
King’s EI estimates in their examination of split-ticket voting. Gay (2001) also uses first-
stage King’s EI estimates in her study of Black Congressional representation and political
participation. There are also several studies that use EI estimates in second-stage
analysis, developing regression models with EI estimated dependent variables. Tolbert
and Hero (2001) study the impact of racial/ethnic context on white voting for racially
charged ballot propositions in California. Liu (2003) models estimated dependent

43
variables for his study of white crossover voting in deracialized contexts. Lublin and
Voss (2002) use estimated dependent variables in second-stage regression analysis of
Francophone support for Quebec sovereignty. Voss and Miller (2001) use King’s EI
estimated dependent variables to study white backlash voting in Kentucky. Ecological
inference will always involve a certain level of uncertainty. And, while some estimates
will be more uncertain than others, the procedure provides a measure of uncertainty that
can be incorporated into second-stage analysis. Moreover, it is often the case that
estimating quantities of interest with ecological inference procedures is the only, and
perhaps even the best choice available to study certain topics, especially those involving
spatially located racial group voting behavior.
Calvo and Escolar’s geographically weighted ecological inference (GWR-EI)
procedure satisfies both critiques of King’ EI model mentioned above. It is essentially an
“extended model” version of King’s EI that controls for spatial autocorrelation. As such
it will produce estimates that satisfy the criteria laid out by King, Adolph, Herron, and
Shotts for second-stage regression models involving EI-estimated variables without
biased results. The level of uncertainty of the estimates is reflected in the standard errors,
which can be used as one step in the process of determining if the estimates are
acceptable.
DISSERTATION: METHODS AND DATA
Geographically Weighted Ecological Inference (GWR-EI) forms the foundation
for my methodological approach to the study of place-context effects. My methodology

44
can be summarized in four steps. First, I develop a spatially located dataset (in GIS
shapefile format) that combines US Census measures of the City of Los Angeles with raw
vote counts for the 2001 and 2005 Los Angeles mayoral run-off elections. Vote counts
are aggregated from the voting precinct to the larger census tract level. Second, I
estimate the percentage of each of four racial groups (Asian, Black, Latino, White) that
voted for each of the two candidates in the 2001 and 2005 Los Angeles mayoral run-off
elections, using Calvo and Escolar’s GWR-EI model. These estimated variables are the
dependent variables of interest that I will visualize and explore further using spatial
econometric techniques to assess the spatial structure of the data. Finally, I will model
the estimated racial group vote variables using spatial econometric regression techniques
that will be used to assess place-context spatial effects on racial group voting behavior.
The GWR-EI process itself involves two stages. The first stage estimates racial
group voter turnout percentage. The second stage is an extension of the basic model that
produces estimates of racial group vote choice percentage. Calvo and Escolar use the
following equation to describe the first stage GWR-EI:
T
i
= B
1
X
i
+ B
2
(1 - X
i
) + B
3
R
i
+ B
4
Z
i +
µ
ι

(2.1)
where T
i
represents the racial group voter turnout for each location, X
i
represents the
racial group percentage of voting age population, µ
i
is the error term, R
i
is the raw turnout
percentage, ballots cast divided by registered voters, and Z
i
is the spatial covariate that
controls for spatial autocorrelation. The spatial covariate Z
i
is obtained by running a
Geographically Weighted Regression of predicted turnout on the residuals of a
Goodman’s naïve regression, as described in Calvo and Escolar (2003). The second

45
stage (EI2) estimates of racial group vote percentage brings additional information into
the equation in the form of V
i
, the percentage of votes cast for each candidate, based on
total votes cast.
Table 2.1 displays the descriptive statistics for the estimated racial group vote
choice measures, including the mean, standard deviation, and standard errors of the
estimates. Figures 2.1 – 2.4 contain frequency distribution histograms and plotted normal
curve for the GWR-EI estimated dependent variables of racial group vote percentage for
each candidate in the 2001 and 2005 Los Angeles mayoral elections. Generally, the data
fit within normal unimodal distribution, though there are some outliers for White and
Asian voters in both 2001 and 2005. Also, not all variables are distributed symmetrically
around the mean: particularly the Black 2001 estimates for both candidates. The estimate
of Black Villaraigosa 2001 voters is particularly unsymmetrical. Also, notice the
distribution of Black voters in 2005, clustered very tightly around the mean.

46

Table 2.1
Descriptive Statistics GWR-EI Estimates of Racial Group Vote
Los Angeles 2001 Los Angeles 2005
Racial Group Vote Mean St. Dev. Mean St. Dev.
Asian Vote Hahn .66
(.030)
.06 .54
(.024)
.09
Asian Vote Villaraigosa .34
(.026)
.05 .46
(.002)
.09
Black Vote Hahn .89
(.007)
.05 .41
(.008)
.03
Black Vote Villaraigosa .23
(.002)
.11 .57
(.007)
.04
Latino Vote Hahn .11
(.005)
.06 .19
(.003)
.09
Latino Vote Villaraigosa .89
(.005)
.05 .81
(.002)
.09
White Vote Hahn .61
(.004)
.09 .47
(.000)
.09
White Vote Villaraigosa .39
(.003)
.09 .52
(.000)
.09
Note: GWR-EI standard errors in parentheses.

Figure 2.1 Distributions of GWR-EI Estimated Dependent Variables, 2001 Election
47
Asian Percent V ote Hahn 2001 Asian Percent V ote Villaraigosa 2001
Black Percent V ote Hahn 2001 Black Percent V ote Villaraigosa 2001
Figure 2.2 Distributions of GWR-EI Estimated Dependent Variables, 2001 Election
48
Latino Percent V ote Hahn 2001 Latino Percent V ote Villaraigosa 2001
White Percent V ote Hahn 2001 White Percent V ote Villaraigosa 2001
Figure 2.3 Distributions of GWR-EI Estimated Dependent Variables, 2005 Election
49
Asian Percent V ote Hahn 2005 Asian Percent V ote Villaraigosa 2005
Black Percent V ote Hahn 2005 Black Percent V ote Villaraigosa 2005
Figure 2.4 Distributions of GWR-EI Estimated Dependent Variables, 2005 Election
50
Latino Hahn 2005
Latino Villaraigosa 2005
White Hahn 2005
White Villaraigosa 2005

51
There are several ways to judge the quality of the estimates produced by the
GWR-EI procedure. King’s EI software provides a variety of diagnostic graphs to test
the accuracy of the estimates, including a graph that he calls a tomography plot
4
. I run a
series of diagnostic tomography plots with likelihood contours at the 90% and 95%
confidence level for each of the 16 estimated racial voter turnout and 16 estimated racial
group vote variables (eight turnout and eight vote percentage variables estimated for each
of the two elections being analyzed). Based on the diagnostic tomography plots, I am
satisfied that the GWR-EI procedure produced quality estimates.
Additionally, King suggests that EI estimates are best tested against a researcher’s
substantive knowledge of the local political processes being studied. Although real-
world data does not exist to test the GWR-EI estimates, they can be compared against
existing exit poll surveys of the 2001 and 2005 Los Angeles mayoral election. Table 2.2
displays the results of the two-stage estimation of racial group vote choice at the
aggregate, citywide level, compared with estimates produced by exit poll surveys. The
results of the GWR-EI procedure stand up quite well when compared to the exit poll
surveys produced by the Los Angeles Times and Loyola Marymount University,
especially given the disagreement between those two polls in 2005 with respect to
racially polarized voting (Barreto et al. 2006).

4
See King 1997 p. 114 for King’s discussion of tomography plots as diagnostic tools for
ecological inference.

52

Table 2.2
Comparison of Aggregate GWR-EI Estimates of Racial Group Vote
Los Angeles 2001 Los Angeles 2005
Racial Group Vote GWR-EI LAT GWR-EI LAT LMU
Asian Vote Hahn .66
(.030)
.65 .54
(.024)
.56 .59
Asian Vote
Villaraigosa
.34
(.026)
.35 .46
(.002)
.46 .41
Black Vote Hahn .89
(.007)
.80 .41
(.008)
.52 .42
Black Vote
Villaraigosa
.23
(.002)
.20 .57
(.007)
.48 .58
Latino Vote Hahn .11
(.005)
.18 .19
(.003)
.16 .14
Latino Vote
Villaraigosa
.89
(.005)
.82 .81
(.002)
.84 .86
White Vote Hahn .61
(.004)
.59 .47
(.000)
.50 .43
White Vote
Villaraigosa
.39
(.003)
.41 .52
(.000)
.50 .57
Note: GWR-EI standard errors in parentheses.

The comparisons presented above are tests of the aggregate quantities of interest
rather than the actual point estimates or individual-level behavior. Figure 2.5 and Figure
2.6 are percentile maps of the GWR-EI estimated racial group vote percentages for 2001
and 2005. These maps are quite consistent with what a researcher who is knowledgeable
of local politics would expect. These maps support the conclusion that the GWR-EI
process created reliable estimates of racial group vote choice in both the 2001 and 2005
Los Angeles mayoral elections.
53
Figure 2.5 Percentile Maps GWR-EI Estimated Racial Group V ote for James Hahn, Los Angeles, 2001
54
Figure 2.6 Percentile Maps GWR-EI Estimated Racial Group V ote For Antonio Villaraigosa, Los Angeles 2005

55
SPATIAL ECONOMETRIC METHODS
The GWR-EI procedure described above serves two purposes. First, because it is
a version of King’s “extended model,” it produces estimates that are unbiased when used
as dependent variables in further second-stage analysis such as regression modeling. The
second purpose is that the inclusion of the spatial covariate controls for the presence of
spatial autocorrelation in the original voting data. This allows the data to be analyzed
with spatial econometric data analysis and regression models as developed by Luc
Anselin (1988).
Spatial econometric methods are designed to explore and understand spatial
processes and spatial structure, and within the proper theoretical context, to establish an
understanding of spatial dependence in the observed behavior. Spatial econometric
techniques allow behavioral observations to be located in space, and detect behavioral
diffusion through space via spatially lagged variables. The spatial dimensions of the
underlying behavior, in this case racial group vote choice within a census tract, can be
captured via the use of spatially lagged variables. A spatial lag variable is an average of
the neighboring locations that is computed via a spatial weights matrix that assigns value
based on contiguity or distance. A spatial lag can be used to compare an observation
with its neighbors, and more specifically to detect the presence of spatial effects, such as
the diffusion of behavior through space. In other words, spatial lag variables can detect
neighborhood effects, behavioral differences caused by location in a particular
neighborhood. Neighborhood effects are very difficult to detect using dummy variables

56
and/or survey data, and, as mentioned above, most research into neighborhood effects has
failed to find such effects. Further, spatial lag regression can be used to statistically
account for the presence of neighborhood effects, and provide a sense of which predictor
variables best explain the behavioral diffusion in the form of neighborhood effects.
It is important to contrast the use of spatially lagged variables to model the spatial
structure of political behavior with the use of dummy variables. Dummy variables are
often used in ordinary regression models as the traditional way to control for geographic
variation in the data, providing a rough sense of location. Spatial lag variables are not
dummy variables, and, in addition to maintaining geographic location, the analysis of
spatially lagged variables can provide finely grained information about behavioral
variation and diffusion through space.
Location in place and space, and the influence of the surrounding environment
should have measurable effects on the voting behavior of the Los Angeles electorate.
The detection and analysis of behavioral diffusion between neighborhoods is the primary
goal of my dissertation. Neighborhood effects are a key indicator of the importance of
place-context. I will use the spatial econometric methods described here to visualize and
model voting behavior in order to assess place-context effects.

57
CHAPTER 3 PLACE, SPACE, AND URBAN VOTING BEHAVIOR:
A SPATIAL ANALYSIS OF THE 2001 LOS ANGELES
MAYORAL ELECTION
In Chapter One, I argue that place matters to voting behavior. I also argue that, in
the study of political behavior, context has been severed from place. As a corrective
mechanism, I suggest that Agnew’s concept of context-as-place is the necessary starting
point for contextual effects analysis of political processes and mechanisms. In Chapter
Two, I develop the argument that the most appropriate way to capture place-context
effects is by using a method that combines a geographically weighted approach to King’s
Ecological Inference model with spatial econometric techniques such as indicators of
spatial autocorrelation and spatial regression models. I also argue that this type of
analysis will tend towards identifying political processes and behaviors that vary across
space and place. Instead of attempting to divine general laws of place-context, the
analysis will point to the complexity of behavior in space. In the present chapter I
combine place-context theory with spatial econometric practice in order to explore the
ways that place affects urban voting behavior.
Why urban voting behavior? Place suggests an attention to geographic scale
below the national and state-level. Large urban elections are ideal for the study of place-
context and voting. In 1999, at the end of an interesting article into the distinctions of
urban and non-urban presidential electorates, Sauerzopf and Swanstrom (1999) issue a
call for further research into understanding contextual effects on urban voting behavior.

58
The authors based their call in some part on the work of Eric Oliver, whose analysis of
suburban democracy and political participation has consistently found that context – city-
level qualities and compositional characteristics of suburban areas – has an effect on the
levels and quality of suburban political behavior (Oliver 1999; Oliver 2000; Oliver 2001).
Although Oliver does not engage directly with a theory of place, nor does his analysis
look for spatial effects, his thoughtful work carefully analyzes the particular quality of
places in suburban America. Although he is able to make generalizable claims about
suburban political behavior, Oliver’s quantitative analysis takes pains to treat each suburb
as a unique context. Thus, Oliver’s work is consistent with a concept of context-as-place.
Among those scholars who study cities and urban political processes, there has
been an increase of interest in contextual effects, yet very little of that work is grounded
in Agnew’s concept of place-context. As I argue in Chapter One, I trace the lack of
interest in places and political science to the early contextual effects analysis of Eulau
and Rothenberg and Robert Huckfeldt (Eulau and Rothenberg 1986; Huckfeldt 1986).
Huckfeldt identifies various processes of social interaction among family, friends, and
neighborhoods as the causal mechanism of contextual effects. Huckfeldt’s research has
evolved into sophisticated analyses of individuals and their social networks, and thus has
moved away from contextual analysis that is situated in place and space (Huckfeldt,
Johnson, and Sprague 2005).
Similarly, very little scholarship employs spatial econometric methods to examine
contextual effects in space and place. Wendy K. Tam Cho is the scholar that most
effectively carries the torch of spatial analysis of political behavior. Cho argues that

59
Huckfeldt’s claim that social interaction is the sole determinant of context effects is
insufficient (Cho and Rudolph 2007). Her findings indicate that that social interaction
alone cannot explain contextual effects on political behavior. Although Cho explains that
the research design use of spatial econometrics does not allow her to make strong
statements about the causal mechanism of contextual effects, she does speculate, based
on her empirical findings that “low intensity environmental cues” – elements of the
contextual environment, the prevalence of campaign signs in a neighborhood for
example, that do not involve conscious social interaction among individual voters – are
part of the contextual effects story.
Karen Kaufmann has also responded to the call and developed a specific
contextual theory of urban voting behavior (Kaufmann 2004). From a contextual effects
point of view, Kaufmann believes that racial context is the most important aspect of
urban voting behavior. Her theory rests on the idea that individual perceptions of racial
group conflict or cooperation will activate the race as a salient issue, and determine the
impact that race has on individual elections. However, although her work is concentrated
on particular cities, Kaufmann is uninterested in place-context below the city level. As
such, place is a variable missing from her theory of urban voting behavior.
LOS ANGELES 2001 MAYORAL ELECTION
The 2001 Los Angeles mayoral election was very important for a variety of
reasons. It marked a decisive turn in the city’s politics towards the left, after eight years
of a Republican mayor. Additionally, the election revolved around three issues that

60
concerned the future of Los Angeles: secession, racial and ethnic transformation brought
about largely by immigration from Latino America and Asia, and an economy that was
becoming increasingly bifurcated between rich and poor (Meyerson 1999).
In addition to the issues of racial diversity and economic ideology, the election
also concerned secession, an inherently spatial question about whether or not Los
Angeles would be the first major city to break itself into smaller spatial regions and
political units (Purcell 2001). Although the issue of secession was not on the ballot
during this mayoral election, the future of a united Los Angeles definitely formed the
backdrop for the elections. The stakes were high (Meyerson 1999; Meyerson 2001).
Antonio Villaraigosa made history as the first Latino candidate to make the run-
off in modern Los Angeles history. He ran a vigorous campaign based on building a
multi-racial coalition, taking up the mantle of the biracial Bradley coalition, while also
emphasizing a progressive ideology based on the support of organized labor (Cleeland
2001; Gorov 2001; Rainey and Shuster 2001). Villaraigosa worked hard to court African
American voters, deploying many of his organized labor resources to walk precincts
specifically in African-American neighborhoods (Rainey and Shuster 2001).
Villaraigosa's team waged an aggressive campaign to break up the bloc vote
among African Americans for Hahn in South Los Angeles. But despite the use of
precinct teams to contact thousands of voters directly, Villaraigosa mustered just
one in five black votes (Rainey and Krikorian 2001).

Villaraigosa’s effort failed to garner enough African-American votes in combination with
his Latino base, and liberal, white, Westside voters to defeat James Hahn, the white
liberal Democrat who had a strong electoral base in the Africa-American community.

61
Moreover, Hahn succeeded in appealing to moderates and conservatives voters,
primarily those in the Western San Fernando Valley, who had previously voted for
Riordan. Many credit Hahn’s success in courting a Valley constituency on two issues:
public safety and secession, both of which had racial overtones. Hahn’s campaign
crafted a campaign advertisement that evoked race-baiting campaigns of Los Angeles’
past by super-imposing Villaraigosa’s image with a crack pipe. This ad was
devastatingly successful in evoking fear based on public safety and racial change.
Kaufmann applies her theory of urban voting to the 2001 Los Angeles mayoral
election. She finds that the racial context of perceptions of group competition between
Latinos and African-Americans was the most salient factor in the election outcome.
Because of the racial context Latino candidate Antonio Villaraigosa was unable to attract
significant support from the African-American community, despite several endorsements
from prominent African-American political leaders. Instead, African-American voters
gave large majorities to the white candidate, James Hahn. Contrasting with Kaufmann’s
argument, Abrajano, Nagler, and Alvarez (2005) argue that race and perceptions of racial
group competition were not the most salient aspect of the 2001 election. The authors
point out that this particular election presented the opportunity for a “natural experiment”
regarding the effects of race versus the impact of issues and ideology. Abrajano, Nagler,
and Alvarez point out that another Latino on the ballot in 2001, Rocky Delgadillo, was
victorious in his campaign for the City Attorney position. Further, Delgadillo, unlike
Villaraigosa, gained the support of a significant majority of the African-American
electorate. The authors argue that Villaraigosa’s ideological position, as the more liberal

62
of the mayoral candidates, and his position on issues of importance to African-American,
were the real factors behind his failure to assemble a cross-racial coalition of Latino and
African-American voters. As with Kaufmann’s analysis, the focus is, of course, on
individual behavior, and the scale of analysis is the entire city of Los Angeles. Abrajano,
Nagler, and Alvarez do incorporate three geographic regions into their analysis as control
variables, but regional variation is not their focus, nor do they find the regional control
variables to be significant in three out of four of their regression models.
The dynamics of the 2001 election – the racial/ethnic cleavages and the spatial
issue of secession – make it an excellent case for a place-context analysis. There has
been no examination of urban voting behavior that combines a place-context perspective
with spatial econometric analysis. I argue that situating voters in place, adjusting the
scale of analysis to include the neighborhood level will produce significant additions to
our understanding of urban voting behavior.
In addition to being a debate about the salience of race/ethnicity to voting
behavior, the debate between Kaufmann and Abrajano, Nagler, and Alvarez can be
characterized as a debate over whether or not urban voting behavior is different from the
voting behavior in national elections. Kaufmann makes the argument that urban voting is
different because of the contextual environment of racial attitudes. Race matters more at
the urban level than it does at the national level. Abrajano, Nagler, and Alvarez argue
that urban voters behave in accordance with the standard issue-space model, making vote
choices based on their assessment of candidates positions within the issue and ideological
space.

63
My argument is that place context is the missing theoretical concept that can
provide a bridge between the two positions. Place context situates racial group dynamics,
activates issue positions, and is one of the factors that channel the flow of political
information. There has been no examination of urban voting behavior that combines a
place-context perspective with spatial econometric analysis. I argue that situating voters
in place and adjusting the scale of analysis to include the neighborhood level will produce
significant additions to our understanding of urban voting behavior.
The questions that motivate this chapter are related to whether or not place and
space played a role in the 2001 Los Angeles mayoral election. Does place-context
matter? If so, which places? Aside from the contextual effects of neighborhood racial
composition, partisan composition, and class composition, is it possible to isolate
independent neighborhood spatial effects? Is there evidence of neighborhood effects, as
well as regional effects? Is there evidence of place effects related to racial context? If so,
that will suggest further study into the interaction of race and place.
DATA AND METHODS
The variables of interest for this study are the percent vote by each racial group
for each of the two candidates in the 2001 Los Angeles general mayoral election. The
City of Los Angeles does not collect information regarding the race of registered voters,
and the secret ballot prevents the collection of race/ethnicity information. As previously
mentioned in Chapter Two, it is possible to estimate percent vote by race using Gary
King’s solution to the ecological inference problem (1997). This process recovers, as

64
best as possible, individual-level behavior using the available aggregate data. However,
one of the assumptions of King’s model is that there be no spatial autocorrelation in the
data, and King’s EI model has been shown to produce biased results in the presence of
spatial autocorrelation (Anselin and Cho 2002; Cho 1998). Spatial autocorrelation occurs
when there is nonrandom clustering of similar values at geographic locations. Finding,
exploring, and modeling spatial autocorrelation are some of the primary tasks of spatial
econometric data analysis, which, when properly undertaken can lead to analytical
findings of spatial neighborhood effects, the diffusion of behaviors through space from
neighborhood to neighborhood. In order to preserve and correctly assess the effects of
spatial autocorrelation, I estimate the eight variables of interest using Calvo and Escolar’s
geographically weighted approach to King’s EI model (GWR-EI), which was designed
explicitly to estimate local vote data (2003). Using GWR-EI I was able to estimate
turnout rates and percent vote by racial group for each unit of aggregation.
The neighborhood is one of the geographic scales at which voting behavior will
be analyzed. Of course, neighborhoods are nebulous things, and there are no official
neighborhood boundaries. As such, census tracts are used to closely approximate
neighborhoods as the basis for the aggregation of individual votes into ecological units of
analysis.
SPATIAL DATA ANALYSIS
The substantive theoretical focus of this chapter is the impact of place and space
on urban voting behavior. As such, the methodological focus will be on visualization of

65
the voting data, and spatial data analysis, including spatial regression to model spatial
effects. One of the most common techniques used to begin exploring spatial structure in
ecological data is to visualize the data through various percentile maps, which display
data using a color ramp scheme. An enhanced variation of a percentile map is the box
map – similar to a box plot in that observations are grouped according to standard
deviations from the mean – which is used to find outlier observations. Figure 3.1
displays four box maps of the 2001 vote percent data.
5
The data displayed in these maps
have been smoothed via a Spatial Empirical Bayes procedure that involves “shrinking”
observational values towards the mean of a geographically moving window of neighbors
around each census tract observation. The Spatial EB box maps in Figure 1 highlight
regional trends in the voting data, with orange colors indicating high levels of support,
blue indicating lower levels of support, and red indicating high outlier observations. The
most noticeable result can be seen in the map of white Hahn voters, which contains
several strong outlier observations in the northern areas of the San Fernando Valley.
Additionally, if we look closely, we can see that there is an outlier in the Harbor area.
These results show that Hahn ran very strongly among white voters in the Valley and the
Harbor, but less well among white voters in the central portions of the City. In contrast,
the Asian Hahn map appears almost a mirror image of the white Hahn map, but the
regional division is between East and West, rather than North and South. The Latino
Villaraigosa box map indicates that Villaraigosa did best in the central portions of the

5
All of the maps, spatial statistics, and spatial regression models in this chapter were
created using Luc Anselin’s GeoDa spatial analysis software package (Anselin 2003).

66
city, but also in the East Valley and Harbor regions. Figure 3.1 also displays the results
for Black Villaraigosa voters. It should be noted that citywide Black voters strongly
supported Hahn. But the map in Figure 3.1 suggests that there is interesting regional
variation among Black voter support for both candidates. Overall, the maps in Figure 3.1
suggest broad regional divisions within the electorate. For each of the racial groups, the
regional trends demonstrated by the maps indicate the need for further statistical inquiry
into the causes of the spatial dynamics in the data.
To further explore the spatial structure of voting in the 2001 election, we turn to
an examination of statistical indicators of spatial autocorrelation. There are two types of
spatial structure, spatial dependence, and spatial heterogeneity (Anselin 1988). Spatial
dependence is associated with neighborhood effects, whereas spatial heterogeneity is
associated with regional effects. An empirical finding of neighborhood effects is
consistent with behavioral contagion, or the diffusion of behavior between neighboring
locations. That is, the likelihood of a particular voting behavior in one neighborhood is
increased by the voting behavior of its neighboring locations. Regional effects, on the
other hand, are spatial effects that cannot be directly linked to behavioral diffusion from
neighbors.
The Spatial EB box maps in Figure 3.1 strongly suggest regional effects, but are
not helpful in assessing evidence of local neighborhood effects. A good way to ascertain
evidence of neighborhood effects is to further statistically explore the nature of any
spatial autocorrelation, that is, correlation between spatial location and observed value in
the data. The most common method of assessing spatial autocorrelation is the Moran’s I

67
statistic. Figure 3.2 and Figure 3.3 display scatterplot graphs of Morans’ I statistic for
each of the eight racial group percent vote variables. Moran’s I is designed to indicate
both positive spatial autocorrelation (geographic clustering of similar values) and
negative spatial autocorrelation (geographic clustering of dissimilar observations).
Positive spatial autocorrelation is displayed in the upper right and lower left quadrants of
the graphs, and negative spatial autocorrelation is displayed in the upper left and lower
right quadrants. The slope of the regression line is the value of the Moran’s I statistic.
The graphs in Figure 3.2 and Figure 3.3 demonstrate significant and high levels of
positive spatial autocorrelation for all variables of interest.
6
These results indicate that
there is a large amount of highly significant non-random global spatial autocorrelation, or
clustering, in the data. This means that there are strong levels of locational similarity in
the data; people who live near each other tended to vote in similar ways. This is hardly
surprising given the strong regional patterning seen in Figure 3.1. The one exception is
Black Hahn voters, whose data exhibits almost no significant spatial autocorrelation.

6
Significance levels for the Moran’s I statistic are calculated using a random permutation
procedure, which produces a pseudo-significance level for each statistic of p < .001 with
999 permutations.
Figure 3.1. Spatial Empirical Bayes Smoothed Box Maps, Los Angeles 2001
68
White Hahn Voters
Black Villaraigosa Voters
Latino Villaraigosa Voters
Asian Hahn Voters
Lower Outlier
< 25%
25% - 50%
50% - 75%
> 75%
Upper Outlier
Spatial Empirical Bayes
Smoothed Box Maps

69
Figure 3.2 Moran’s I Scatterplot Graphs, Racial Group Vote Hahn

70
Figure 3.3 Moran’s I Scatterplot Graphs, Racial Group Vote Villaraigosa

71
This result means those Black voters that supported Hahn did so regardless of
their location. This makes sense given the high levels of support within the Black
community for Hahn. It also means, given the significant spatial autocorrelation among
those few Black voters that did support Villaraigosa, that spatial location mattered a great
deal to Black Villaraigosa voters. Similarly, although Latino voters did strongly support
the Latino candidate, Villaraigosa, spatial location is correlated to that support. This
indicates the presence of spatial clustering among Latino voters, both those who
supported Villaraigosa, and the relatively small proportion that supported Hahn. The
results of the Moran’s I analysis is strongly suggestive of neighborhood effects on racial
group voting.
Since we are interested in local neighborhood effects, further analysis of local
indicators of spatial autocorrelation (LISA) can provide helpful insights. LISA cluster
maps display significant clusters of positive spatial autocorrelation as well as significant
spatial outliers (negative spatial autocorrelation). The map colors correspond to the
quadrants of the Moran’s I scatterplot graphs: red indicates high-high clusters and dark
blue indicates low-low clusters, each of which correspond to the positive spatial
autocorrelation quadrants of the Moran’s I graph (upper-right and lower-right); light blue
indicates low-high spatial outliers (low values surrounded by high values), and pink
indicates high-low spatial outliers (high values surrounded by low).

72
Figure 3.4 and Figure 3.5 display LISA cluster maps for each of the eight
variables of interest.
7
Figure 3.4 displays the Hahn vote, and Figure 3.5 displays the
Villaraigosa vote. The cluster maps reveal a large number of local clusters. The LISA
Cluster maps are strongly suggestive of neighborhood spatial effects in the vote choice
data. Given the high slope values for Moran’s I for almost all of the variables, the strong
patterns seen in the LISA cluster maps are not surprising. These results indicate that the
regional patterns illustrated in Figure 3.1 may, in fact, be related to neighborhood effects.
That is, the results may indicate the diffusion of voting behavior from one neighborhood
to another in certain locations in the city.
There are several patterns that are especially noteworthy. Of particular interest
are the blue low-low clusters for Latino Villaraigosa voters in South L.A., which indicate
the possibility of neighborhood effects significantly lowering Latino support for
Villaraigosa in that region. Further, this region of Los Angeles contains a dense African-
American population, and is, as Figure 3.4 indicates, an area of high-high spatial clusters
of African-American Hahn voters. The two clusters suggest a very interesting
neighborhood effect involving Latino and African-American voters in South L.A. that
significantly altered Latino support for Villaraigosa.
The large clusters of high-high spatial autocorrelation for White Hahn voters in
the San Fernando Valley demonstrate how strong Hahn’s support was among white
residents of this region. It is also interesting to note the similarity between the patterns

7
The LISA cluster maps are calculated with second order Queen contiguity spatial
weights, and are displayed at the p < .01 significance level.

73
for the Asian Villaraigosa vote and the Black Villaraigosa vote. On a city-wide level,
both groups strongly supported Hahn, but in terms of local spatial patterns, there is a
strong East-West divide, with large numbers of high-high clusters in the West Valley and
West L.A. regions for both variables, and several low-low clusters in the Downtown and
South L.A. regions.
However, the spatial patterns seen in the LISA cluster maps may be spurious
spatial effects rather than evidence of neighborhood effects and behavioral diffusion.
Non-random spatial clustering is a necessary but not sufficient condition of making an
analytical finding of spatial dependence and neighborhood effects. Although the patterns
are statistically non-random, they may still be artifacts of the distribution of racial group
population, or other variables such as population distribution along class and partisanship
dimensions. Spatial regression models can be used to determine whether there are true
neighborhood effects present in the data, while controlling for other variables such as
race/ethnicity, socioeconomic class, and partisanship.
There are several patterns that are especially noteworthy. Of particular interest
are the blue low-low clusters for Latino Villaraigosa voters in South L.A., which indicate
the possibility of neighborhood effects significantly lowering Latino support for
Villaraigosa in that region. Further, this region of Los Angeles contains a dense African-
American population, and is, as Figure 3.4 indicates, an area of high-high spatial clusters
of African-American Hahn voters. The two clusters suggest a very interest neighborhood
effect involving Latino and African-American voters in South L.A. that significantly
altered Latino support for Villaraigosa.
74
Figure 3.4. LISA Cluster Maps, Racial Group V ote Hahn, 2001
White Hahn
Black Hahn
Latino Hahn
Asian Hahn
75
Figure 3.5. LISA Cluster Maps, Racial Group V ote Villaraigosa, 2001
White Villaraigosa
Black Villaraigosa
Latino Villaraigosa
Asian Villaraigosa

76
Figure 3.6 Geographic Regions of Los Angeles

77
The large clusters of high-high spatial autocorrelation for White Hahn voters in
the San Fernando Valley demonstrate how strong Hahn’s support was among white
residents of this region. It is also interesting to note the similarity between the patterns
for the Asian Villaraigosa vote and the Black Villaraigosa vote. On a city-wide level,
both groups strongly supported Hahn, but in terms of local spatial patterns, there is a
strong East-West divide, with large numbers of high-high clusters in the West Valley and
West L.A. regions for both variables, and several low-low clusters in the Downtown and
South L.A. regions.
However, the spatial patterns seen in the LISA cluster maps may be spurious
spatial effects rather than evidence of neighborhood effects and behavioral diffusion.
Non-random spatial clustering is a necessary but not sufficient condition of making an
analytical finding of spatial dependence and neighborhood effects. Although the patterns
are statistically non-random, they may still be artifacts of the distribution of racial group
population, or other variables such as population distribution along class and partisanship
dimensions. Spatial regression models can be used to determine whether there are true
neighborhood effects present in the data, while controlling for other variables such as
race/ethnicity, socioeconomic class, and partisanship.
SPATIAL REGRESSION MODELS
Broadly speaking there are two types of spatial regression models that are of use
for the current researcher. The first, spatial lag regression, utilizes a maximum likelihood

78
estimator, and incorporates a spatially lagged form of the dependent variable as an
explanatory variable. The spatial lag model is expressed as:
y = ρWy + Xβ + ε (3.1)
Where ρ (Rho) is the spatial autoregressive coefficient, Wy is the spatially lagged form
of the dependent variable created via an N x N spatial weights matrix, Xβ is the
explanatory variable, and ε is the error term. A significant spatial lag regression model
will have a significant spatial lag term ρ, which indicates that the dependent variable is
partially explained by the geographically weighted average of its neighbors, even when
controlling for covariates. Following Luc Anselin and Wendy K. Tam Cho, I interpret
this result as a neighborhood effect that is independent of racial/ethnic population
distributions, socioeconomic status, and partisanship (Anselin 1988; Anselin and Cho
2002; Cho 2003). A neighborhood effect means that voting behavior is diffusing through
space from one place to its neighbors. The spatial lag model does not tell us what is the
mechanism causing the neighborhood effect; it simply provides evidence that the
behavioral diffusion is taking place. It may be that the neighborhood effect is caused by
the social interaction of friends and neighbors (Huckfeldt 1986; Huckfeldt, Plutzer, and
Sprague 1993), or it may be caused by a reaction to low-level environmental place-
context cues (Cho and Rudolph 2007). The spatial lag model does tell us about the
independent effects of place and space upon voting behavior.
The second type of spatial regression is the maximum likelihood spatial error
model, which can be expressed as:
y = Xβ + ε, where ε = λWε + ξ (3.2)

79
In the spatial error model, the spatial autoregressive coefficient λ (Lamda) and spatial
weights matrix W are placed into the error term. The spatial error model is used to model
spatial autocorrelation that is not captured by the measured independent variables. A
proper spatial error specification would indicate that there are neighborhood effects in the
data that are caused by unmeasured independent variables. For instance, the impact of
targeted campaign voter mobilization and persuasion, as well as the effects of a salient
political issue such as Valley secession, may cause neighborhood effects that are difficult
to measure, and therefore indirectly captured by a spatial error model.
How does one decide which spatial regression model is more appropriate? There
is a diagnostic procedure that guides the researcher through the choice of spatial
regression model towards the most appropriate model specification. The GeoDa software
provides a set of Lagrange Multiplier (LM) diagnostic tests for the type of spatial
dependence present in the data that are used to guide model choice
8
. However, in
practice, data often exhibit both spatial lag as well as spatial error dependence. Goodness
of fit statistics, particularly log likelihood can be used to guide model choice.
9

Tables 3.1 – 3.4 display the spatial regression models, as well as the companion
OLS models for each of the eight dependent variables. For the spatial lag models,
comparing the change in coefficient and significance levels between OLS and spatial lag
model facilitates interpretation of the spatial effects. The spatial lag model will often

8
See Anselin (Anselin 1988) for the derivation of Lagrange Multiplier diagnostic tests
and for their use in guiding model choice.
9
Increase in the absolute value of the log likelihood value indicates improved fit.
Although R
2
is provided for the spatial regression models, it is important to note that this
is a pseudo-R
2
, and is not appropriate to compare with OLS fit statistics (Anselin 2005).

80
exhibit an increased significance level for an explanatory variable, or even a change in
sign directionality. In these two cases, accounting for the spatial autocorrelation in the
data by incorporating the spatially lagged dependent variable as an explanatory clarifies
the stronger or changed relationship between the dependent variable and the explanatory
variable. The other possibility is that the explanatory variable is reduced in significance
level by an order of magnitude between the OLS and spatial lag specification. This result
means that the explanatory power of the independent variable is, in fact, due to the
influence of neighboring observations.
In order to model the independent effects of space and place, I include several
covariate explanatory variables that affect vote choice. To control for racial group
population distribution, I have included measures of racial group population density.
Population density is a measure of population per square mile, and is obtained by
dividing the Census population count into the area of the census tract. I use the density
measures for several reasons. First, density measures are less collinear than percent
race/ethnic population.
10
Second, I would argue that density more accurately reflect the
way individuals experience population as they move through space, which experience
may be reflected in their vote choice.
I also use the Census measure of percent native born in order to capture some
effects of immigration. I control for socioeconomic status by including the Census

10
The regression models do exhibit some multi-collinearity issues, but they are related to
the socioeconomic status variables rather than the racial population density measures.
Maximum Likelihood spatial regression is robust to violations of the multi-collinearity
assumption.

81
measures of median household income and percent of residents who have obtained a
bachelors degree. I include the Census residential stability measure, percent that have
been in the same residence for five years or more, to capture the effects community
cohesion and stability.
Partisan attachment and ideological self-identification are two of the most reliable
predictors of voting behavior (Campbell et al. 1960; Miller and Shanks 1996). Of course,
attitudinal measures of partisan attachment and ideological self-identification are very
difficult to measure within an aggregate ecological framework. Nonetheless, I include a
measure of Republican Party registration, aggregated from the vote precinct to the
Census tract level to partially, and crudely, capture some effects of partisanship.
Although Los Angeles elections are non-partisan, the majority of candidates in recent
years have been affiliated with the Democratic Party. Additionally, Los Angeles County
is one of the most reliable bastions of Democratic Party support in statewide and
presidential elections. Because of this, I find that percent Republican registration is a
more useful variable than Democratic Party registration.
Levels of voter turnout and mobilization are also an important predictor of vote
choice. Additionally, voter turnout reflects to some extent the GOTV efforts of the
various candidate campaigns. Campaigns are increasingly sophisticated at voter
mobilization that targets specific neighborhoods and areas of the city. I use the GWR-EI
approach previously discussed to estimate turnout rate by racial group at the census tract
level. I include this variable in each regression model in order to control for the effects of
voter turnout.

82
Finally, I include regional spatial dummy variables in order to capture and model
the regional effects that appear in the Spatial EB box maps in Figure 3.1. Figure 3.6 is a
map that depicts the geographical layout of the seven regional dummy variables. These
regional boundaries are not taken from any official designation, but are rather determined
from an analysis of geographic boundary markers such as the Hollywood Hills, police
department division jurisdiction lines, and major freeways and roads. They are designed
to capture regional divisions without adhering to any arbitrary political / bureaucratic
boundaries that sort residents regardless of any exiting organic community ties.
Results: Latino Voters
The regression model results for Latino voters demonstrate strong spatial
neighborhood effects. The spatial lag term ρ is highly significant, and the spatial lag
provides improved fit, as illustrated by the increased log likelihood and R
2
statistics. By
comparing the OLS and spatial lag specifications for both Latino Hahn and Latino
Villaraigosa voters we can see that the addition of the spatial lag term caused decreases in
significance level and coefficient for the Latino turnout variable, percent Republican
registration, and all of the significant spatial dummy variables. This means that part of
the explanatory power of these variables is in fact explained by the voting behavior of the
neighboring locations. This is interpreted as the diffusion of behavior between
neighborhoods independent of the compositional context of racial population density,
socioeconomic status, party registration, and residential stability.

83
The results also indicate that most of the spatial autocorrelation is caused by
spatial dependence neighborhood effects rather than the regional effects of spatial
heterogeneity. The regional spatial dummy variables are significant in the OLS models,
but all are significant altered in the spatial lag model. Even South L.A., the only regional
dummy to remain significant in the spatial lag specification, is strongly affected by the
spatial lag term. This indicates that the spatial cluster in South L.A. is related to a
neighborhood effect.

84
Table 3.1
Latino Voters OLS and Spatial Lag Regression Models
% Latino Hahn % Latino Villaraigosa
OLS Spatial Lag OLS Spatial Lag

Spatial Lag (ρ) --- .704***
(041)
--- .679***
(.043)
Constant .187***
(.014)
.075***
(.014)
.838***
(.014)
.258***
(.039)
Latino Turnout .362***
(.062)
-.110*
(.055)
.232***
(.062)
.024
(.055)
Income
(thousands)
-.0002*
(.0001)
-.0002*
(.0001)
.0002
(.0001)
.0002
(.0001)
% Bachelors Degree -.003
(.020)
.003
(.018)
.007
(.020)
-.006
(.018)
% Republican
Registration
.157***
(.026)
.112***
(.024)
-.161***
(.026)
-.125***
(.024)
% Native -.036*
(.018)
-.029*
(.016)
.034*
(.018)
.034*
(.016)
Residential Stability -.074***
(.020)
-.066***
(.018)
.065***
(.019)
.058***
(.017)
Latino Pop Density
(thousands sq./mi.)
.00007
(.0002)
-.00002
(.0002)
-.0001
(.0002)
.00007
(.0002)
Black Pop Density
(thousands sq./mi.)
.002*
(.0009)
.002**
(.0008)
-.003***
(.0009)
-.003***
(.0008)
White Pop Density
(thousands sq./mi.)
-.0004
(.0005)
.0001
(.0004)
.0009*
(.0005)
.0001
(.0004)
Asian Pop Density
(thousands sq./mi.)
.0003
(.0005)
-.0006
(.0004)
.0003
(.0005)
.001**
(.0004)
Spatial Dummies
West Valley --- --- --- ---
East Valley .008*
(.005)
-.002
(.004)
.012*
(.007)
.002
(.004)
Downtown -.035***
(.007)
-.004
(.006)
.037***
(.007)
.006
(.006)
South LA .079***
(.008)
.029***
(.007)
-.077***
(.008)
-.031***
(.008)
Harbor .019**
(.008)
-.002
(.007)
.019**
(.008)
.003
(.007)
Mid City .008
(.006)
.004
(.006)
-.004
(.006)
-.005
(.006)
West LA -.007
(.007)
.002
(005)
.010*
(.007)
-.001
(.006)
R
2
.474 .596 .472 .574
Log Likelihood 1507.33 1592.11 1515 1587.76
Note: N = 837 census tracts. Dependent variable is percent racial group vote for each candidate.
Standard errors in parentheses. Bolded results indicate a change in significance level between
OLS and spatial lag regression models. *** p < .001, ** p < .01, * p < .10

85
The contextual environment of Black population density positively influenced
Latino Hahn voters. The likelihood of Latino voters supporting Hahn increased with
Black population density. The significance of Black population density increases in the
spatial lag model, which means that taking into account spatial dependence further
clarified the relationship between Latino voters and Black population density. This is an
independent racial context effect that demonstrates a level of cooperation between Latino
and Black voters, probably concentrated in the South L.A. region. This result stands in
contrast to Kaufmann’s findings of competition and conflict between Latino and Black
voters as the driving force of the 2001 election, and it demonstrates the importance of a
place-context perspective. Yes, citywide election results point toward a level of conflict
between the interests of the Black and Latino communities when it comes to voting for
mayor. However, if we drill down to the neighborhood level, situating voters in place-
context, we find strong evidence of cooperation; Latinos would be more likely to support
the white candidate favored by a majority of African-Americans in South L.A.
What explains the current results involving behavioral diffusion and racial context
effects? Are the observed effects caused by social interaction between Latino and Black
voters (friends, neighborhoods, churchgoers, etc.)? My analysis here does not directly
assess individual social interaction, nor does it allow for definitive statements regarding
the causal mechanism of the contextual effects. However, the results do provide
evidence of both independent place effects as well as evidence of voters responding to
their race context environment.

86
Results: White Voters
Table 3.2 displays the spatial error regression model results for White voters. The spatial
error term λ is highly significant, and the log likelihood shows increased fit from the OLS
model. This result means that there are spatial effects in the data of white voters, but that
they are best explained by unmeasured independent variables. What variables are
causing the spatial effects? It is impossible to say for certain given the current study
design, however, my speculation points toward two factors: 1) campaign effects, and 2)
effects related to the issue of Valley secession. I particularly emphasize the impact of
secession, which was an important issue during the 2001 election, yet has been relatively
unaddressed by Kaufmann and Abrajano et al. Secession is an issue that is intimately
connected to the concept of place, with Valley secessionists attempting to carve out a
place-identity and a political culture that is distinct from that of the rest of the city of Los
Angeles (Finnegan 2001; Gorov 2001). Although Abrajano et al attribute issues and
ideology as the driving force of the election, they gloss over the issue of secession. I
would argue that this omission occurs because secession does not fit neatly onto the
liberal-conservative ideological continuum (Hogen-Esch 2001). There are longstanding
economic, political, and cultural aspects to the various secession movements in the
Valley.

87
Table 3.2
White Voters OLS and Spatial Error Regression Models
% White Hahn % White Villaraigosa
OLS Spatial Error OLS Spatial Error

Spatial Error
Lamda (λ)
--- .667***
(.050)
--- .764***
(.039)
Constant .510***
(.021)
.513***
(.024)
.541***
(.021)
.486***
(.023)
White Turnout .433***
(.035)
.487***
(.041)
-.288***
(.035)
-.025***
(.040)
Income
(thousands)
-.0001
(.0002)
-.0001
(.0002)
.0001
(.0002)
.00003
(.0002)
% College Edu. -.113***
(.030)
-.137***
(.030)
.118***
(.030)
.094**
(.029)
% Republican
Registration
.477***
(.039)
.510***
(.047)
-.443***
(.038)
-.370***
(.046)
% Native -.122***
(.028)
-.133***
(.028)
.053*
(.027)
.072**
(.027)
Res. Stability -.046
(.032)
-.055*
(.029)
-.047
(.031)
-.057*
(.028)
Latino Pop Density
(thousands sq./mi.)
.0006
(.0003)
.0005*
(.0003)
-.001***
(.0003)
-.001***
(.0003)
Black Pop Density
(thousands sq./mi.)
.002
(.001)
.0002
(.001)
-.001
(.001)
.001
(.001)
White Pop Density
(thousands sq./mi.)
-.001
(.0008)
-.0006
(.0007)
.001*
(.0007)
.001*
(.0007)
Asian Pop Density
(thousands sq./mi.)
-.001
(.0007)
-.0008
(.0008)
.0007
(.0007)
.0006
(.0007)
Spatial Dummies
West Valley .069***
(.012)
.065***
(.019)
-.077***
(.012)
-.063**
(.021)
East Valley .018*
(.010)
.017
(.017)
-.034***
(.010)
-.021
(.019)
Downtown .028**
(.010)
.011
(.017)
-.042***
(.010)
-.002
(.018)
South LA --- --- --- ---
Harbor .031*
(.013)
.022
(.027)
-.041**
(.013)
-.024
(.032)
Mid City -.004
(.009)
-.002
(.014)
-.0003
(.009)
.018
(.015)
West LA -.019*
(.012)
-.017
(.019)
.013
(.012)
.031*
(.021)
R
2
.542 .622 .552 .664
Log Likelihood 1172.33 1236.92 1186.88 1285.35
Note: N = 837 census tracts. Dependent variable is percent racial group vote for each candidate.
Standard errors in parentheses. Bolded results indicate a change in significance level between
OLS and spatial error regression models. *** p < .001, ** p < .01, * p < .10

88
There are also elements of racial resentment in the politics of secession. The
variable that makes these issues difficult to locate in standard issue-space analysis, or
within standard theories of race and ideology is the impact of place. Many of these
political, economic, cultural, and racial ingredients are tied up in the propagation of a
place-identity associated with the Valley, and the desire to control urban space and local
economic development (Hogen-Esch 2001; Hogen-Esch 2002; Purcell 2001; Sonenshein
and Hogen-Esch 2006). Secession, particularly the case of the San Fernando Valley, “is
overtly spatial in that it involves restructuring existing state territory … [it] is also a
struggle over space … [and] how urban space should be used” (Purcell 2001, 630). For
these reasons, I believe that attitudes towards secession and the mobilization efforts of
Valley secession organizations is the unmeasured variable causing the strong spatial
effects seen among White Hahn voters in the West Valley.
11

As for the impact of racial context, white voters did not appear to be strongly
influenced by the racial population density environment. There is an effect related to
Latino population density, which can be interpreted as a form of racial group conflict;
increases in Latino population density increased the likelihood that white voters
supported Hahn. Additionally, increases in white population density increase the
likelihood of support for Villaraigosa by white voters. This points towards white voters
in the denser areas of West L.A. and the Mid City regions giving higher levels of support
to Villaraigosa. This confirms the LISA cluster analysis from Figure 3.5 that show

11
Although Hahn strongly courted Valley secessionists during the 2001 election, once
elected he was a crucial part of the effort to defeat secession.

89
several clusters of high support for Villaraigosa stretching from the Westside through the
central areas of the city.
Results: Asian Voters
The LISA cluster maps for Asian voters suggest that there was significant
variation across space for Asian voters, despite the fact that as an electoral bloc, Asians
strongly supported Hahn. Yet, the spatial lag models in Table 3.3 do not quite confirm
the LISA analysis. The spatial lag term is significant for Asian Hahn voters, but only at
the p < .05 significance level. The result indicates that there was a certain amount of
behavioral diffusion affecting Asian Hahn voters, especially in the Downtown, East
Valley, and Harbor regions of the city. These regions correspond to the high-high spatial
clusters seen on the LISA map in Figure 3.4. The Asian Hahn spatial lag model shows
that the addition of the spatial lag term had very little effect on the regional spatial
dummy variables West Valley, Mid City, and West L.A. Instead of neighborhood
effects, the spatial clusters seen in these regions can be classified as regional effects that
are related to spatial heterogeneity in the data; they are not effects of spatially dependent
behavioral diffusion.

90
Table 3.3
Asian Voters OLS and Spatial Lag Regression Models
% Asian Hahn % Asian Villaraigosa
Model Type OLS Spatial Lag OLS Spatial Lag

Spatial Lag (ρ) --- .157*
(.074)
--- .326***
(.054)
Constant .789***
(.018)
.677***
(.055)
.238***
(.010)
.143***
(.019)
Asian Turnout .137***
(.041)
.126**
(.040)
.016
(.023)
.037*
(.023)
Income
(thousands)
-.0003*
(.0001)
-.0003*
(.0001)
-.0001
(.00009)
-.0001
(.00009)
% Bachelors Degree .005
(.026)
.008
(.026)
.065***
(.015)
.056***
(.014)
% Republican
Registration
-.073*
(.033)
-.072*
(.032)
.179***
(.019)
.160***
(.018)
% Native -.167***
(.023)
-.158***
(.023)
.043***
(.013)
.027*
(.012)
Residential Stability -.064**
(.024)
-.059*
(.024)
.074***
(.014)
.062***
(.013)
Latino Pop Density
(thousands sq./mi.)
.00004
(.0003)
-.00004
(.0003)
-.0005***
(.0001)
-.0004*
(.0001)
Black Pop Density
(thousands sq./mi.)
-.002*
(.001)
-.002
(.001)
.009***
(.0007)
.007***
(.0007)
White Pop Density
(thousands sq./mi.)
.0008
(.0006)
.0009
(.0006)
-.00003
(.0004)
.000003
(.0003)
Asian Pop Density
(thousands sq./mi.)
-.0001
(.0006)
-.0002
(.0006)
-.0004
(.0004)
-.0003
(.0003)
Spatial Dummies
West Valley .017*
(.010)
.018*
(.010)
-.008
(.006)
-.010*
(.006)
East Valley .029***
(.009)
.027**
(.009)
-.019***
(.005)
-.014**
(.005)
Downtown .031***
(.009)
.027**
(.009)
-.027***
(.005)
-.016**
(.005)
South LA --- --- --- ---
Harbor .049***
(.012)
.044***
(.012)
-.020**
(.006)
-.014*
(.006)
Mid City .019*
(.008)
.016*
(.008)
-.015***
(.004)
-.011*
(.004)
West LA .019*
(.011)
.018*
(.010)
-.033***
(.005)
-.028***
(.006)
R
2
.374 .378 .613 .631
Log Likelihood 1300.42 1302.47 1773.22 1789.68
Note: N = 837 census tracts. Dependent variable is percent racial group vote for each candidate.
Standard errors in parentheses. Bolded results indicate a change in significance level between
OLS and spatial lag regression models. *** p < .001, ** p < .01, * p < .10

91
In contrast, the spatial lag term for Asian Villaraigosa is highly significant, and
alters the impact of most of the significant explanatory variables, including all of the
regional dummy variables. Thus, there is a large amount of behavioral diffusion that
impacts the voting behavior of Asian Villaraigosa voters, independent of the
environmental context of racial population density, partisanship, and socioeconomic
status.
Interpretation of racial context is difficult with respect to the voting behavior of
Asians in Los Angeles. As previously mentioned, the LISA analysis shows striking
similarities between Asian Villaraigosa voters and Black Villaraigosa voters.
Additionally, as Black population density increases, Asian Villaraigosa vote increased,
and the Asian Hahn vote decreased. It is difficult to make strong assertions regarding
racial group cooperation; nonetheless, the results indicate that Villaraigosa gained
consistent and similar levels of electoral support from both Asians and Blacks in
particular places in the city. Moreover, the voting behavior of Asians and Blacks in those
places was strongly influenced by the behavior of their neighbors.
Again, as with the evidence of neighborhood effects among Latino voters, I
cannot make claims about the exact causal mechanism driving the behavioral diffusion of
Asian Villaraigosa voters.

92
Results: Black Voters
The model results for Black voters are not as illuminating as the previous analysis, as
illustrated in Table 3.4. First, neither the spatial lag nor the spatial error models fit the
Black Villaraigosa data. Thus, despite the high levels of spatial autocorrelation seen in
the Moran’s I graph in Figure 3.2, we cannot draw any conclusions regarding
neighborhood effects or regional effects. We also cannot corroborate the findings
regarding racial context effects between Black and Asian and Latino voters for
Villaraigosa.
A spatial error model was fit to the Black Hahn voter data. Given the low levels
of spatial autocorrelation seen in the Moran’s I graph, it is not surprising to see the
relatively low R
2
fit. Still, we can conclude that there is evidence of neighborhood
effects amongst Black Hahn voters. The neighborhood effects are related to unmeasured
explanatory variables not present in the regression models, rather than behavioral
diffusion from neighboring places.
What unmeasured variable is causing the spatial dependence in the Black Hahn
data? As in the White voter data, I argue that the issue of Valley secession is part of the
explanation. But, the effect is the opposite of that seen with White voters. According to
the Los Angeles Times polling of opinions towards secession, Blacks generally did not
express favorable opinions of Valley secession efforts, regardless of geography (Pinkus
and Darling 2002). Hahn’s attempts to court Valley secessionists may have backfired

93
Table 3.4
Black Voters OLS and Spatial Lag Regression Models
% Black Hahn % Black Villaraigosa
OLS Spatial Error OLS Spatial Error

Spatial Error (λ) --- .172*
(.089)
--- .031
(.097)
Constant .946***
(.016)
.945***
(.017)
-.003
(.002)
-.003
(.004)
Black Turnout .347***
(.033)
.368***
(.034)
.974***
(.004)
.973***
(.004)
Income
(thousands)
-.0004**
(.0002)
-.0004**
(.0002)
.00002
(.00002)
.00002
(.0002)
% Bachelors Degree -.022
(.025)
-.024
(.025)
.003
(.003)
.003
(.003)
% Republican
Registration
-.015
(.031)
-.013
(.033)
.0005
(.004)
0005
(.004)
% Native -.109***
(.023)
-.117***
(.023)
.005*
(.003)
.005*
(.003)
Residential Stability -.092***
(.025)
-.091***
(.025)
.011***
(.003)
.011***
(.003)
Latino Pop Density
(thousands sq./mi.)
-.0003
(.0002)
-.0003
(.0002)
-.00006*
(.00003)
-.00006*
(.00003)
Black Pop Density
(thousands sq./mi.)
.0002
(.001)
.0007
(.001)
.00007
(.0001)
.00007
(.0001)
White Pop Density
(thousands sq./mi.)
-.0002
(.0006)
-.0002
(.0006)
-.00006
(.00007)
-.00006
(.00007)
Asian Pop Density
(thousands sq./mi.)
-.0005
(.0006)
-.0005
(.0006)
-.00005
(.00008)
-.00005
(.00008)
Spatial Dummies
West Valley -.007
(008)
-.005
(.009)
-.0006
(.001)
-.0007
(.001)
East Valley .003
(.007)
.004
(.008)
-.001
(.0008)
-.001
(.0009)
Downtown --- --- --- ---
South LA .021*
(.008)
.019*
(.009)
-.002*
(.001)
-.002*
(.001)
Harbor .010
(.009)
.011
(.011)
-.002
(.001)
-.002
(.001)
Mid City .007
(.007)
.008
(.007)
-.0005
(.0009)
-.0005
(.0009)
West LA .014*
(.008)
.016*
(.009)
-.001
(.001)
-.001
(.001)
R
2
.139 .144 .996 .996
Log Likelihood 1354.97 1356.22 3061.90 3061.94
Note: N = 837 census tracts. Dependent variable is percent racial group vote for each candidate.
Standard errors in parentheses. Bolded results indicate a change in significance level between
OLS and spatial error regression models. *** p < .001, ** p < .01, * p < .10

94
among Black Valley residents, contributing to the low-low spatial clusters seen in Figure
3.3 LISA cluster map of Black Hahn voters.
Another unmeasured factor that may explain the neighborhood effects of Black
Hahn voters is the so-called “Kenny Hahn Effect.” According to analysts of Los Angeles
politics, much of the support for James Hahn by Black voters was an effect of the legacy
of his father, Kenneth Hahn, a legendary politician who strongly advocated civil rights in
Los Angeles (Meyerson 2001).
CONCLUSION: PLACE-CONTEXT AND NEIGHBORHOOD EFFECTS
Place context theory and spatial econometric analysis is not a substitute for
attitudinal survey analysis of urban voting behavior. However, the empirical results
produced in this chapter offer clear evidence that place and space are important missing
variables that should be accounted for when studying urban voting. Urban voters are
influenced by their place-context, through neighborhood behavioral diffusion, and
through their contextual environment of racial density, partisanship, and socioeconomic
status.
Additionally, there are several examples of racial context and place-context
interacting to produce effects that vary from place to place. The most prominent example
of this was found among Latino and Black Hahn voters in South L.A. that point towards
the interaction of place-context and racial context effects.
In addition to contributing to our understanding of the importance of place-
context to urban voting behavior, my findings of neighborhood effects are also relevant

95
for interpretations of the 2001 election. Most analyses focus on two issues to explain the
results. First, Villaraigosa failed to achieve a broad multi-racial coalition that appealed to
African-American voters. Additionally, scholars such as Kaufmann have used this result
to point to the importance of racial group conflict in determining election results.
However, my results show that a place-context perspective which analyzes voting at the
neighborhood level complicate these analyses. At the neighborhood level, I found
significant evidence of cross-racial behavioral contagion between Latino and African-
American voters, which suggests that racial group conflict is not necessarily the
inevitable fate of the Los Angeles electorate. Villaraigosa’s efforts to attract African-
American voters may not have been reflected in an electoral victory, but my findings
indicate that those African-American voters that did vote for him were spatially clustered
in certain Los Angeles neighborhoods. Moreover, although analyses suggest that Hahn
ran a racially divisive campaign, when one looks at the neighborhood level it is possible
to see evidence of African-Americans positively influencing Latinos to vote for Hahn.
The second point of most analysis is that Hahn successfully portrayed
Villaraigosa as unacceptably liberal, especially on issues of public safety that were of
great concern moderate and conservative voters, many of whom live in the Valley. The
evidence of my research in this chapter suggests that a spatial issue, in the form of Valley
secession, played an important role, even when controlling for partisanship.
The results of this chapter suggest that a place-context perspective complicates
contextual analyses of racial group conflict and cooperation. In the next chapter, I turn to

96
a more specific examination of the interaction between racial context and place-context
effects.

97
CHAPTER 4 PLACE-CONTEXT AND RACIAL GROUP VOTING
BEHAVIOR: A SPATIAL ANALYSIS OF THE 2005
LOS ANGELES MAYORAL ELECTION
The study of race and ethnicity in American politics has often been intertwined
with specific places, regions, and geographic cultures. Much of the research into the
politics of race/ethnicity in America is descended from the work of V.O. Key who had a
clear understanding of the importance of race to the politics of the South (Key 1949).
The intertwining of place and ethnicity was a subject that was taken up by Robert Dahl
(Dahl 1961). The assimilation of white ethnic groups into the politics of New Haven was
a crucial aspect of Dahl’s pluralist theory of politics. In general, ethnic assimilationists
believed that social class would overcome ethnic cleavages. Dahl’s theory of ethnic
assimilation was later challenged by scholars who found that ethnic bloc voting continued
to be a powerful influence on politics and voting (Parenti 1967; Wolfinger 1965).
As the demographics of American cities shifted –African-American migration to
cities, and white flight to the suburbs – the focus of study often shifted to the examination
of race and the phenomenon of racial cleavage in the electorate. Some scholars, whose
unit of analysis was the city, examined the dynamics of racial coalition politics as
practiced by urban political elites (Browning, Marshall, and Tabb 1996; Sonenshein
1993). For those who were more interested in the individual unit of analysis – and less
interested in the politics of place – racial or power threat became the leading model of
racial voting behavior and political attitudes. According to this model, white racial

98
attitudes and voting behavior are motivated by perceptions of threat connected to the
presence and political empowerment of African Americans (Giles and Evans 1985; Giles
and Evans 1986; Giles and Hertz 1994; Pettigrew 1971).
By turning their attention to places such as cities and neighborhoods, contextual
effects scholars such as Robert Huckfeldt, take a different approach to racial/ethnic
voting. Proponents of contextual effects analysis believe that environmental and social
context can have a determining effect on racial attitudes and voting behavior. For
instance, Robert Huckfeldt found that “the persistence of ethnic politics is often rooted in
a supportive social environment” (Huckfeldt 1986, 83). In particular, Huckfeldt found
that the neighborhood and family social contexts were the most important to ethnic
voting behavior. Social interaction within tight-knit ethnic neighborhoods was, they
argued, an important aspect of political cohesiveness and ethnic bloc voting behavior.
Many scholars, influenced by Huckfeldt, pursued contextual effects analysis that
challenges the racial threat model of political attitudes and voting behavior. These
scholars find that social contact and interaction between members of different
racial/ethnic groups can have a positive effect on attitudes, producing more tolerance for
diversity, and a higher likelihood of voting for out-group political candidates (Carsey
1995; Oliver and Mendelberg 2000; Oliver and Wong 2003).
More recently, there have been several attempts to add depth and complexity to
the study of racial contextual effects. Claudine Gay has found that “neighborhood
quality” is an important aspect of environmental context (Gay 2004). Rodney Hero and
Caroline Tolbert have produced a body of research that focuses on differentiating and

99
classifying racial context effects. They have found that racial behavior will differ
depending on whether the racial context is moderately diverse, bifurcated, or
homogenous (Hero and Tolbert 1996; Tolbert and Hero 2001). Karen Kaufmann
develops a contextual theory of urban voting behavior that depends on how racial group
members perceive their various environments of group conflict or group cooperation
(Kaufmann 2003; Kaufmann 2004). Baodong Liu attempts to bridge the divide between
the idea of racial threat / group conflict and positive social contact / group cooperation
(Liu 2001). Liu points out that the two different models of racial voting behavior present
a false dichotomy, one that disappears depending upon the context being studied.
These studies all show that racial attitudes and voting behavior vary depending on
how and what type of context is studied. What unites these studies is the concept of
place-context. Political science research into contextual effects tends not to consider
place a variable that can have a causal affect on voting behavior. What is missing from
the debate over racial voting behavior is a serious consideration of place-context theory,
and methodological approaches that take space and place into account. Although many
contextual effects scholars take places such as cities and neighborhoods as their canvas,
much of the work is uninformed by a theory of place-context, which requires that places
be treated as more than neutral containers for social group identity formation and
interaction (Agnew 1996a). As such, context is often defined as the racial/ethnic
composition of a neighborhood or city. This approach fails to consider the ways in which
place and racial identity interact and intertwine, with each having an effect on the other
(Agnew and Smith 2002; Saito 1998).

100
I argue that contextual analyses of racial voting behavior should incorporate a
place-context perspective. Once we introduce space and place into the equation, we
would expect the racial context effects to vary across space and place. Instead of general,
or global, patterns of racial voting behavior - polarized racial bloc voting, or positive
contact effects – we would expect to see a complex mixture of behaviors that vary from
place to place, neighborhood to neighborhood, even within the same city. If my
argument is correct, it means that studying race and ethnic voting behavior, even from a
contextual point of view, and failing to consider the interaction of place and race will
produce incorrect findings involving general patterns at one level, while variable effects
will be evident once we drill down from a global (city-wide) level to a local
(neighborhood) level.
RACIAL VOTING IN LOS ANGELES
To test my argument, I look at racial voting in recent Los Angeles mayoral
elections. In 2001, Villaraigosa ran quite clearly as the progressive Latino future of Los
Angeles. In 2005, however, Villaraigosa appeared to pursue a strategy that was
cognizant of the ability of Hahn to racialize the 2001 campaign (Wright Austin and
Middleton IV 2004). Villaraigosa attempted to project a much more pragmatic and
results-oriented campaign focused on issues rather than explicitly on building an
ideologically progressive multiracial coalition.
Three recent studies of racial voting in Los Angeles find strong evidence for
racially polarized voting in 2001 and 2005 elections. Karen Kaufmann found that the

101
primary motivating factor for James Hahn’s victory in 2001 was the perception of group
conflict established in the minds of African-American voters (2004, 176). Despite the
attempts of the Villaraigosa campaign to recruit African-American political elites in an
effort to build a cross-racial coalition, most individual black voters felt that there interests
ran counter to the interests of a Latino led Villaraigosa governing regime. Similarly,
Wright Austin and Middleton IV (2004), find that Villaraigosa was unable to overcome
polarized racial voting behavior in 2001. Sonenshein and Drayse (2006) examine the
spatial dynamics of racial/ethnic coalition voting for the 2001 and 2005 elections, and
also find that city-wide voting patterns indicate competitive or conflictual relations
between African-Americans and Latinos. I argue that using a place-context perspective,
one that takes seriously the interaction of place and race, will call these findings into
question. Each of these studies finds global, or citywide, patterns of racial bloc voting,
which leads the authors to generally conclude that voting in Los Angeles is driven by
racial group conflict. Yet, none of the studies place racial voting behavior in place-
context by examining the interaction of place and race at levels below the citywide. I
argue that placing Los Angeles racial voting behavior in spatial context will produce
evidence of both racially polarized or group conflict voting, as well as evidence of
positive contact or group cooperation voting.
12
This finding would be consistent with
Liu’s (2001) argument that general findings of conflict versus cooperation present a false

12
I have elected to use the terms group cooperation and group conflict when discussing
the voting behavior of and between racial groups. However, it should be noted that it is
difficult to assess attitudes of racial threat, conflict, or cooperation when conducting
contextual analysis using aggregate ecological data rather than attitudinal surveys,
especially in a complex multiracial / multiethnic environment.

102
distinction that, more accurately, reflects which context is being studied. In this case, I
would argue that what matters is which place-context is being studied.
DATA AND METHODS
In addition to place-context theory, a method of inquiry using spatial
econometrics will contribute to the debate by allowing researchers to: 1) treat
neighborhoods and other geographic units of analysis as places, distinct and yet
connected spatially with their surroundings, and 2) examine how racial context effects, as
well as other types of contextual effect, propagate through space and place in the form of
behavioral contagion.
The variables of interest for this study are the geographically located percentage
of each racial group (Latino, African-American, White, and Asian) that voted in each
census tract for the two candidates in the 2005 Los Angeles mayoral run-off election.
13

Because of the secret ballot, information about the race/ethnicity characteristics of voters
is not available. Moreover, exit poll surveys are not appropriate for this study because
they do not provide enough geographically located observations, and because several
studies have shown surveys to be somewhat unreliable when it comes to issues of racial
attitudes and voting behavior (Voss 2004). Therefore, I utilize Calvo and Escolar’s
(2003) geographically weighted approach to King’s (1997) ecological inference (GWR-
EI) solution in order to estimate the dependent variables of interest, the percentage of

13
Unfortunately, the Census does not provide population information about Jewish
ancestry. As Sonenshein’s (2000) work about the importance of Jewish voters to Los
Angeles politics indicates, it would be very useful to include this information.

103
each racial group that voted for the two candidates, Hahn and Villaraigosa. The
estimation is a two-stage process involving first, estimating the racial group turnout
percentage, followed by the percentage of each racial group that voted for each of two
candidates. Table 4.1 presents the estimated aggregate or global results for the eight
variables of interest compared with 2005 exit polls from Loyola Marymount University
and the Los Angeles Times.

Table 4.1
Estimated Racial Group Vote, Los Angeles 2005
Racial Group Candidate
Vote
GWR-EI LAT LMU
Asian Vote Hahn .54
(.024)
.56 .59
Asian Vote Villaraigosa .46
(.002)
.44 .41
Black Vote Hahn .41
(.008)
.52 .42
Black Vote Villaraigosa .57
(.007)
.48 .58
Latino Vote Hahn .19
(.003)
.16 .14
Latino Vote Villaraigosa .81
(.002)
.84 .86
White Vote Hahn .47
(.000)
.50 .43
White Vote Villaraigosa .52
(.000)
.50 .57
Note: Table compares GWR-EI estimates with exit polls from Los Angeles
Times and Loyola Marymount University. GWR-EI standard errors in
parentheses.

The results from the GWR-EI compare favorably to the two exit polls, especially
considering the significant differences between the two polls. The GWR-EI estimates are

104
more similar to the LMU exit poll, which, unlike the Los Angeles Times poll employs a
racially stratified homogenous precinct approach to more accurately estimate racial group
voting (Barreto 2006). The most significant differences are the estimates for Black
voters; the Los Angeles Times poll indicates that a 52% majority of Blacks voted for
Hahn. But the LMU poll and the GWR-EI estimates show Villaraigosa gaining a strong
majority of Black voters at 57%.
In terms of interpreting the results for evidence of racial group competition or
cooperation, the global estimates provide relatively little information. It is also important
to note that the global estimates produce a picture of racial group voting behavior at the
citywide level. From that perspective, the racial group vote percentages can fit within an
interpretation that emphasizes competition and conflict among groups. For instance,
according to the LMU exit poll, Hahn gained a strong 57% majority of support from
white voters, and 60% of Asian voters. The Los Angeles Times numbers are slightly
different, but still suggest an image of a city divided between white and Asian voters on
one side and Black and Latino voters on the other. If we are to examine racial group
competition between Black and Latino groups, the exit polls give us mixed information.
According to the Los Angeles Times numbers, Blacks and Latinos in Los Angeles 2005
did not see common interests in voting for the minority candidate, Villaraigosa.
However, the GWR-EI and LMU estimates present a different picture of Blacks voting
for Villaraigosa. The citywide estimates do not present a clear picture of racial group
voting. The advantage of the GWR-EI method is that it produces local estimates, which
can be used to further analyze racial group voting.

105
Figure 4.1 contains four maps that display the census tract vote percentage, by
racial group, for Villaraigosa. There are clearly geographic areas of strength for each
candidate. In particular, Hahn performed best in the Harbor and West Valley areas of the
city, whereas Villaraigosa’s base was in the Mid City, West L.A., and East Valley
regions. The maps in Figure 4.1 suggest a certain level of spatial structure.
In order to assess the statistically significant level of spatial autocorrelation, I
explore the data through the use of Moran’s I statistics. Moran’s I is a test of global
spatial autocorrelation in the data (Anselin 1988). The statistic is the slope of a
regression line indicating linear association between observation values and the weighted
average of its neighbors. Positive spatial autocorrelation indicates spatial clustering of
high and low values, and, given the right temporal or theoretical dynamics, can imply
spatial contagion or cooperation. Negative spatial autocorrelation indicates a
checkerboard pattern, or spatial clustering of unlike values, and can imply competition or
conflict. Table 4.2 contains the Moran’s I values for each of the eight variables of
interest, all of which illustrate positive global spatial autocorrelation.
14
The Moran’s I
values for White and Latino voters exhibit strong positive correlation between location
and vote. This implies that White and Latino voters that reside together tended to vote
similarly. However, as in the 2001 data, for Black voters, there is a weak, though
positive, correlation between location and vote choice. Similarly, the Asian vote data

14
Moran’s I statistics and the spatial lag regression models require a spatial weights
matrix. All of the statistics and spatial lag regression models were calculated using 2
nd

order, Queen contiguity weights. All spatial statistics and regression models were created
using Luc Anselin’s GeoDA spatial analysis software package (Anselin 2003).
106
Figure 4.1 GWR-EI Estimated Racial Group V ote For Antonio Villaraigosa, Los Angeles 2005

107
exhibits positive spatial autocorrelation, though location is not as strongly correlated
with vote as it is for White and Latino voters. Location does not seem to matter as much
to Black and Asian voters as it does for White and Latino voters.

Table 4.2 Moran’s I Spatial Autocorrelation Statistic
Variable Moran’s I
Asian Vote Hahn .2215
Asian Vote Villaraigosa .2596
Black Vote Hahn .1286
Black Vote Villaraigosa .0601
Latino Vote Hahn .6289
Latino Vote Villaraigosa .6303
White Vote Hahn .5115
White Vote Villaraigosa .4525
Note: All statistics are significant at p < .001, and computed
using 2
nd
order Queen contiguity weights.

The presence of positive spatial autocorrelation in the data is theoretically
consistent with behavioral contagion, the diffusion through space of similar values of the
observed variable. That is, the observed voting behavior is at least partly dependent on
neighboring values. In terms of racial voting behavior, contagion would indicate spatial
context effects – voting behavior in a neighborhood would be partially dependent on the
behavior of neighboring areas.
The Moran’s I results suggest interesting spatial effects and behavioral contagion
of racial voting behavior in the data. In order to more fully understand the spatial effects
on racial voting, multivariate spatial lag regression is used to model the data (Anselin,
Florax, and Rey 2003; Anselin 1988). Spatial lag regression, using a maximum
likelihood estimator, incorporates a spatially lagged form of the dependent variable (a

108
spatially weighted average of its neighboring values) as an explanatory variable in the
regression equation.
A spatial lag regression model can be expressed using this equation:
y = ρWy + Xβ + ε (4.1)
Where ρ is the spatial autoregressive coefficient, Wy is the spatially lagged form of the
dependent variable created via an N x N spatial weights matrix, Xβ is the explanatory
variable, and ε is the error term. A significant spatial lag model with significant spatial
lag term indicates that the dependent variable can be partially explained by the
neighboring values. This result can be interpreted as a behavioral contagion spatial effect
whereby the observed behavior is partially caused by the behavior of the neighbors. In
the present study, such a result can be interpreted as the diffusion through space of
similar racial group voting behavior.
The analysis and interpretation of spatial effects does not end with a significant
spatial lag term. It is possible to assess the explanatory forces associated with the spatial
effect by comparing the Ordinary Least Squares (OLS) and spatial lag model
specifications. When comparing the OLS and spatial lag models, there are three possible
changes. First, the significance level of an explanatory variable may be significantly
reduced in the spatial lag model. This indicates that the spatial lag term absorbed the
partial effect of the variable, such that a portion of the explanatory power of that variable
is due to neighboring quantities. This can be interpreted as a spatial effect related to the
particular explanatory variable. Second, a variable may display increased significance
level in the spatial lag model. This indicates that modeling the spatial autocorrelation

109
clarified the presence of a significant relationship between the explanatory and dependent
variables. For instance, the effects of income may not be significant in the OLS
specification, but becomes significant in the spatial lag model. Accounting for the spatial
autocorrelation present in the data revealed the significant relationship. Similarly, the
spatial lag model may clarify the direction of the variable effect, which may reverse from
one model to the other. A variable that indicates a negative correlation coefficient in the
OLS model may become positive in the spatial lag model.
The regression model tables are organized by racial group. Tables 4.3, 4.6, 4.9,
and 4.12 display the global OLS and spatial lag model results for each of the eight racial
group dependent variables. These models are “global” or citywide models in that I am
interested in the effects associated with racial population density and other explanatory
variables at the citywide level. Each model contains ten explanatory variables plus a set
of spatial dummy variables that correspond to the seven regions of the City of Los
Angeles. Figure 4.2 illustrates the geographic layout of the seven spatial regions that I
will use throughout the analysis. These spatial regions were designed not to correspond
with arbitrary political or bureaucratic sorting of residents, but rather to approximate
regional geography and community boundaries. The spatial dummy variables are
included to measure the independent significance of geographic regions on racial voting
behavior. However, it should be noted that the effect of dummy variables is distinctly
different from spatial lag variables. The spatial dummy variables are designed to allow
me to isolate neighborhood effect interaction with racial context at geographic scales
below the citywide.

110
Racial population density measures are included in the models in order to assess
the relationships of cooperation or conflict between racial groups. Population density is
obtained by dividing population by the area of each census tract. The resulting measure
reflects the population per square mile. The decision to use population density per square
mile is somewhat of a departure from other studies of racial context and voting behavior.
Among geographers, the use of density measures is quite common. However, political
science quantitative studies typically use the racial group proportion of the total or voting
age population. Interestingly, many studies of racial context and voting behavior use the
term “racial population density” to discuss the effects of racial context, while at the same
time using the variable proportion of the population as the quantitative measure.
15
I
would argue this reflects an understanding that population density measures more
appropriately capture the spatial dimension of racial context, and provide a better
quantitative measure of the way in which race is experienced through place.
Several contextual studies have found that the quality of a neighborhood –
sometimes this means socio-economic composition and other times it may indicate the
physical character of a neighborhood – is an important impact on voting behavior (Gay
2004; Sonenshein and Drayse 2006). In order to test this dimension, I include a measure
of what I call residential stability, the census measure of percent of the population who
have lived in the same house for five years or more. Admittedly, this is a crude measure
of neighborhood quality, and much better measures are needed. But, I would argue the
residential stability variable does capture some portion of neighborhood character.

15
See, for instance, Kohfeld and Sprague (1995) and Liu (2003)

111
Figure 4.2 Geographic Regions of Los Angeles

112

In addition to racial group population density and residential stability measures, I
included controls for racial group turnout by including the first-stage GWR-EI estimation
result. I control for the effects of socio-economic status (SES) are by including census
tract measures of median household income and percent that have obtained a college
degree. Also included is a measure of native-born status in order to capture effects
related to immigration. Finally, Republican percentage of registered voters is included to
capture partisanship effects. Perhaps the most important weakness of ecological analysis
is the inability to include any sophisticated attitudinal measures of ideological self-
identification or partisan attachment. According the Los Angeles Times exit poll of the
2005 mayoral election, the large majority of voters (70%) registered Democratic.
Additionally, Los Angeles elections are non-partisan. Therefore, despite the drawbacks,
as an ecological measure, percent Republican registration does, I believe, crudely
approximate some of the partisan cleavages in Los Angeles elections.
A key part of my argument is that the interaction of racial context and place will
produce variations in behavior from place to place. Racial context may affect white
Hahn voters differently in the East Valley than in West L.A. To test this argument, in
addition to the global models, I develop local spatial interaction models that are designed
to assess the ways in which the explanatory variables interact with the local spatial
dummy variables. Tables 4.4, 4.5, 4.7, 4.8, 4.10, 4.11, 4.13 and 4.14 display the results
for the local spatial interaction models. Each local spatial interaction model contains the
same ten standard explanatory variables mentioned above, one spatial dummy variable,

113
and nine spatial interaction cross-product terms based on that spatial dummy. These
tables display only the results for the spatial interaction terms, omitting for the purposes
of clarity the complete model results. For example, Table 4.4 reports the two local spatial
interaction models for Latino Villaraigosa voters, one for Downtown spatial interaction
and another for South L.A.
DISCUSSION OF RESULTS
Given the levels of spatial autocorrelation reported earlier, it is no surprise that the
all of the spatial lag regression models contain highly significant spatial lag terms. The
global model tables in particular indicate the presence of numerous behavioral contagion
spatial effects that vary from place to place (the bolded results).
Residential stability is significant for white, black, and Asian voters. For white
and black voters, stability negatively affects the vote for Villaraigosa. In terms of spatial
analysis, it is important to point out that the spatial lag model reverses the directionality
of stability, from positive to negative. For Asian voters, residential stability increases the
vote for Villaraigosa. This is especially interesting given the contrast with education:
support for Villaraigosa increased among the highly educated of all racial groups. It
suggests that stable neighborhoods, with high percentages of long-term home ownership
preferred the status quo. Considering the high levels of support for Villaraigosa by
Latino voters, it is unsurprising that variations in residential stability do not affect the
vote. Republican party registration is the most consistently significant variable, reliably

114
indicating that those neighborhoods with higher levels of Republican registration were
more likely to vote for Hahn.
The spatial dummy variables demonstrably indicate the spatial divisions of the
electorate in Los Angeles. In contrast to the 2001 election, Villaraigosa in 2005, made
impressive geographic inroads into both portions of the Valley, expanding his geographic
base from the Downtown and Mid City areas. Only the Harbor region is a positive for
Hahn, and it this is one of the most noteworthy results of my analysis. Table 4.3
illustrates that the Harbor variable is highly significant and positive for Latino Hahn
voters, and negative for Latino Villaraigosa voters. This result confirms that place-
context has a significant impact on Latino voting behavior, even the capacity to alter the
level of support for a Latino candidate.
The following sections will explore the interaction of place and race in more
detail by focusing on the racial context effects on voting behavior.
Racial Context: Latino Voters
Decades of research into voting behavior, especially into the ways that racial
identity and attitudes can impact voting, leads us to expect that Latino voters will produce
large majorities for a Latino candidate. Thus it comes as no surprise that Latino voters
overwhelmingly supported Villaraigosa in the 2005 mayoral election, in all areas of the
city. Given Villaraigosa’s ability to mobilize Latinos in favor of his candidacy, the
question is whether or not race and place-context can have an effect on even the voting

115
behavior of Latinos in this election? My analysis shows that, even in this case, the
interaction of place and race has an impact on voting behavior.
There is evidence, though it is difficult to interpret, that Latino Villaraigosa voters
were partially motivated by the racial context of their spatial environments. As Table 4.3
illustrates there is evidence of positive contextual interaction between Latino and black
voters. As black population density increases, Latino support for Villaraigosa decreases.
Additionally, as shown in Table 4.5, those Latinos that did vote for Hahn in the Harbor
region were positively influenced by white population density. As white population
density increases, Latino voters in the Harbor region were more likely to vote for Hahn.
This is a powerful example of the interaction of race and place context. However, there
is evidence of conflict between Latino and white voters in general. In Table 4.3 we see
that as white population density increases, Latinos are more likely to support
Villaraigosa, and less likely to support Hahn.

116
Table 4.3
Latino Voters Global OLS and Spatial Lag Regression Models
% Latino Villaraigosa % Latino Hahn
OLS Spatial Lag OLS Spatial Lag

Spatial Lag (ρ) --- .331***
(.035)
--- .519***
(.049)
Constant .850***
(.019)
.571***
(.034)
.141***
(.018)
.068***
(.018)
Latino Turnout .100**
(.036)
.093**
(.033)
-.095**
(.034)
-.086**
(.032)
Income
(thousands)
.0003
(.0001)
.0003
(.0001)
-.0003*
(.0001)
-.0003*
(.0001)
% Bachelors Degree .087**
(.029)
.070**
(.027)
-.075**
(.028)
-.045*
(.026)
% Republican
Registration
-.623***
(.037)
-.545***
(.035)
.597***
(.035)
.439*
(.036)
% Native .004
(.026)
.017
(.024)
-.009
(.025)
-.021
(.023)
Residential Stability .027
(.028)
.011
(.026)
-.022
(.027)
-.004
(.025)
Latino Pop Density
(thousands/mi.
2
)
-.0004
(.0003)
-.0002
(.0003)
.0005
(.0003)
.0003
(.003)
Black Pop Density
(thousands/mi.
2
)
-.005***
(.001)
-.005***
(.001)
.005***
(.001)
.004***
(.001)
White Pop Density
(thousands/sq.mi)
.002**
(.0007)
.001*
(.0007)
-.002**
(.0007)
-.001*
(.0007)
Asian Pop Density
(thousands/mi.
2
)
-.001*
(.0007)
-.0005
(.0007)
.001*
(.0007)
-.0003
(.0006)
Spatial Dummies
West Valley .029***
(.009)
.037***
(.007)
-.028***
(.009)
-.032***
(.008)
East Valley .028***
(.008)
.033***
(.007)
-.025***
(.007)
-.023***
(.007)
Downtown .035***
(.008)
.021**
(.007)
-.032***
(.008)
-.013*
(.007)
South LA -.054***
(.009)
-.037***
(.009)
.051***
(.009)
.026**
(.008)
Harbor -.201***
(.012)
-.121***
(.013)
.196***
(.011)
.077***
(.014)
Mid City --- --- --- ---
West LA .032***
(.009)
.026***
(.008)
-.031***
(.008)
-.018*
(.008)
R
2
.592 .638 .589 .651
Log Likelihood 1211.95 1257.48 1244.05 1300.14
Note: N = 837 census tracts. Dependent variable is percent racial group vote for each candidate.
Standard errors in parentheses. Bolded results indicate a change in significance level between
OLS and spatial lag regression models. *** p < .001, ** p < .01, * p < .10

117
Table 4.4
Latino Villaraigosa Voters Local Spatial Interaction Models
Downtown South LA
Model Type OLS Spatial Lag OLS Spatial Lag

Spatial Lag (ρ) --- .569***
(.035)
--- .561***
(.034)
Constant .899***
(.024)
.401***
(.035)
.892***
(.024)
.409***
(.034)
Spatial Dummy -.033
(.043)
-.016
(.035)
.027
(.034)
.016
(.027)
Space * Income .003**
(.001)
.003**
(.0009)
.001
(.003)
.001
(.001)
Space * % Bach Deg -.326**
(.129)
-.133
(.104)
.245
(.203)
.134
(.161)
Space * % Repub Regis -.532**
(.201)
-.490**
(.161)
-.985***
(.274)
-.677**
(218)
Space * % Native .205*
(.100)
-.0008
(.080)
-.072
(.137)
-.063
(.109)
Space * Res Stability -.069
(.101)
.013
(.081)
-.143
(.123)
-.065
(.098)
Space * Latino Pop Density
(thousands/mi.
2
)
.001
(.001)
.0008
(.009)
.001
(.001)
-.0003
(.001)
Space * White Pop Density
(thousands/mi.
2
)
.007
(.01)
.001
(.01)
.07
(.09)
.05
(.07)
Space* Black Pop Density
(thousands/mi.
2
)
-.01
(.009)
-.005
(.007)
.008*
(.005)
.006
(.004)
Space * Asian Pop Density
(thousands/mi.
2
)
.01**
(.004)
.007*
(.007)
.02
(.03)
.03
(.03)
R
2
.353 .577 .347 .579
Log Likelihood 1018.52 1181.89 1014.83 1184.36
Note: Table displays spatial interaction cross-product terms. Full model includes additional
predictor variables (turnout, income, % bach deg, % repub regis, stability, and four racial group pop
density measures). N = 837 census tracts. Dependent variable is percent Latino Villaraigosa voters.
Standard errors in parentheses. Bolded results indicate a change in significance level between OLS
and spatial lag regression models. *** p < .001, ** p < .01, * p < .10

118

Table 4.5
Latino Hahn Voters Local Spatial Interaction Models
Harbor Mid City
Model Type OLS Spatial Lag OLS Spatial Lag

Spatial Lag (ρ) --- .489***
(.047)
--- .737***
(.035)
Constant .165***
(.018)
.092***
(.017)
.144***
(.025)
.036*
(.019)
Spatial Dummy -.312*
(.122)
-.347**
(.111)
.066**
(.026)
.067***
(.019)
Space * Income -.005**
(.001)
-.003*
(.002)
.0004
(.0006)
.0004
(.0005)
Space * % Bach Deg .333
(.226)
.018
(.209)
.330***
(.086)
.142*
(.062)
Space * % Repub Regis -.029
(.232)
.105
(.214)
-.331**
(.121)
-.134*
(.088)
Space * % Native .804***
(.181)
.702***
(.165)
-.338***
(.075)
-.214***
(.054)
Space * Res Stability .280*
(.135)
.163
(.124)
.131*
(.061)
.039
(.045)
Space * Latino Pop Density
(thousands/mi.
2
)
.0008
(.002)
.001
(.002)
.002**
(.0007)
.0005
(.0005)
Space * White Pop Density
(thousands/mi.
2
)
.02**
(.007)
.01*
(.006)
-.004*
(.002)
-.001
(.001)
Space* Black Pop Density
(thousands/mi.
2
)
-.03*
(.02)
-.04**
(.01)
.004
(.004)
.003
(.003)
Space * Asian Pop Density
(thousands/mi.
2
)
-.003
(.02)
.006
(.01)
.003
(.002)
.0006
(.002)
R
2
.611 .669 .338 .643
Log Likelihood 1266.74 1323.75 1044.56 1276.95
Note: Table displays spatial interaction cross-product terms. Full model includes additional
predictor variables (turnout, income, % bach deg, % repub regis, stability, and four racial group pop
density measures). N = 837 census tracts. Dependent variable is percent Latino Hahn voters.
Standard errors in parentheses. Bolded results indicate a change in significance level between OLS
and spatial lag regression models. *** p < .001, ** p < .01, * p < .10

119
Racial Context: White Voters
The behavior of white voters is of particular interest in this study for several
reasons. First, much of the debate concerning racial context voting is centered on the
behavior of white voters in the presence of racial diversity. Second, Los Angeles has a
particularly loaded history of both racial conflict and racial cooperation. Liberal white
voters were an integral part of the successful Bradley coalition (Sonenshein 1993). But,
more recently, white voters seemed reluctant to vote for minority candidates such as
Mike Woo in 1993 and Antonio Villaraigosa in 2001, despite the hopes of many analysts
for the creation of successful multiracial coalitions in Los Angeles following Bradley.
The results of the current study are mixed when it comes to racial context and white
voters, but they do support the argument that place context influences the way that racial
context will affect voting behavior. Table 4.6 illustrates the global results for white
voters. Although Latino population density does not appear to be a significant factor,
Black and Asian population density are both negative factors for white Villaraigosa
voters, and positive factors for white Hahn voters. As minority population density
increased, white voters were less likely to cross racial lines to support the minority
candidate. Moreover, as we can see when comparing the OLS to spatial lag models, the
significance level of both Black and Asian population density were affected by the
presence of the spatial lag term. This implies that behavioral contagion through space is
negatively associated with minority population density for white voters.

120
Table 4.6
White Voters Global OLS and Spatial Lag Regression Models
% White Villaraigosa % White Hahn
OLS Spatial Lag OLS Spatial Lag

Spatial Lag (ρ) --- .677***
(.054)
--- .671***
(.052)
Constant .570***
(.027)
.233***
(.036)
.425***
(.022)
.094**
(.034)
White Turnout -.330***
(.066)
-.309***
(.058)
.333***
(.057)
.336***
(.052)
Income
(thousands)
-.0001
(.0002)
-.0001
(.0002)
.00003
(.0002)
.00006
(.0003)
% Bachelors Degree .115**
(.038)
.052
(.036)
-.085**
(.033)
-.025
(.031)
% Republican Regis -.405***
(.049)
-.277***
(.048)
.409***
(.042)
.269***
(.041)
% Native .027
(.034)
.057*
(.031)
-.018
(.029)
-.059*
(.028)
Residential Stability -.051
(.036)
-.055*
(.033)
.056*
(.031)
.053*
(.028)
Latino Pop Density
(thousands/mi.
2
)
-.00001
(.0004)
-.0001
(.0004)
.00008
(.0003)
-.00009
(.0003)
Black Pop Density
(thousands/mi.
2
)
-.01***
(.002)
-.008***
(.002)
.01***
(.001)
.009***
(.001)
White Pop Density
(thousands/mi.
2
)
.005***
(.0009)
.004***
(.0009)
-.005***
(.0009)
-.004***
(.0007)
Asian Pop Density
(thousands/mi.
2
)
-.004***
(.0009)
-.003**
(.0009)
.005***
(.0008)
.004***
(.0007)
Spatial Dummies
West Valley .063***
(.015)
.037**
(.015)
-.088***
(.011)
-.049***
(.013)
East Valley .052***
(.013)
.019
(.013)
-.081***
(.011)
-.033***
(.011)
Downtown .008
(.013)
.003
(.012)
-.031**
(.011)
-.016*
(.011)
South LA --- --- --- ---
Harbor -.126***
(.017)
-.035*
(.017)
.102***
(.015)
.024*
(.014)
Mid City .017
(.011)
.007
(.011)
-.039***
(.010)
-.026**
(.009)
West LA .065***
(.015)
.019
(.015)
-.086***
(.013)
-.038**
(.013)
R
2
.436 .511 .514 .586
Log Likelihood 983.71 1034.52 1106.12 1164.75
Note: N = 837 census tracts. Dependent variable is percent racial group vote for each candidate.
Standard errors in parentheses. Bolded results indicate a change in significance level between OLS
and spatial lag regression models. *** p < .001, ** p < .01, * p < .10

121

Table 4.7
White Villaraigosa Voters Local Spatial Interaction Models
Mid City East Valley
Model Type OLS Spatial Lag OLS Spatial Lag

Spatial Lag (ρ) --- .793***
(.039)
--- .798***
(.038)
Constant .609***
(.029)
.175***
(.032)
.599***
(.072)
.187***
(.031)
Spatial Dummy -.138***
(.031)
-.118***
(.025)
-.018
(.075)
-.064
(.063)
Space * Income
(thousands)
-.00006
(.0007)
-.0003
(.0006)
-.0004
(.0007)
-.0005
(.0006)
Space * % Bach Deg -.251**
(.100)
-.212**
(.083)
.0006
(.092)
-.015
(.078)
Space * % Repub Regis .284*
(.142)
.183
(.118)
-.102
(.118)
-.057
(.099)
Space * % Native .357***
(.087)
.328***
(.072)
.085
(.111)
.063
(.093)
Space * Res Stability -.067
(.071)
-.022
(.059)
.057
(.102)
.091
(.086)
Space * Latino Pop Density
(thousands/mi.
2
)
-.0007
(.0008)
-.0006
(.0007)
.002
(.002)
.003*
(.001)
Space * White Pop Density
(thousands/mi.
2
)
-.0005
(.002)
.001
(.002)
.005
(.004)
.005*
(.003)
Space* Black Pop Density
(thousands/mi.
2
)
-.010*
(.005)
-.010**
(.004)
-.009
(.020)
-.010
(.003)
Space * Asian Pop Density
(thousands/mi.
2
)
.009***
(.002)
.007***
(.002)
-.020*
(.010)
-.010
(.010)
R
2
.335 .530 .323 .510
Log Likelihood 914.55 1046.83 906.92 1029.08
Note: Table displays spatial interaction cross-product terms. Full model includes additional
predictor variables (turnout, income, % bach deg, % repub regis, stability, and four racial group pop
density measures). N = 837 census tracts. Dependent variable is percent white Villaraigosa voters.
Standard errors in parentheses. Bolded results indicate a change in significance level between OLS
and spatial lag regression models. *** p < .001, ** p < .01, * p < .10

122

Table 4.8
White Hahn Voters Local Spatial Interaction Models
Harbor West L.A.
Model Type OLS Spatial Lag OLS Spatial Lag

Spatial Lag (ρ) --- .636***
(.049)
--- .801***
(.036)
Constant .407***
(.020)
.111***
(.030)
.327***
(.024)
-.012
(.025)
Spatial Dummy -.246*
(.144)
-.260*
(.131)
.281**
(.095)
.261***
(.076)
Space * Income
(thousands)
-.003
(.002)
-.0003
(.002)
-.0002
(.0004)
.00004
(.0003)
Space * % Bach Deg .595*
(.267)
.214
(.247)
.134*
(.085)
.025
(.068)
Space * % Repub Regis -.151
(.274)
-.008
(.251)
.506***
(.142)
.402***
(.114)
Space * % Native .572**
(.213)
.389*
(.195)
-.512***
(.107)
-.353***
(.086)
Space * Res Stability .168
(.159)
.093
(.146)
-.229**
(.085)
-.226***
(.068)
Space * Latino Pop Density
(thousands/mi.
2
)
.002
(.002)
.001
(.007)
-.004
(.003)
-.002
(.003)
Space * White Pop Density
(thousands/mi.
2
)
.020**
(.008)
.009
(.007)
.003
(.003)
.001
(.002)
Space* Black Pop Density
(thousands/mi.
2
)
-.050**
(.002)
-.050*
(.002)
.008
(.006)
.002
(.005)
Space * Asian Pop Density
(thousands/mi.
2
)
-.002
(.020)
.010
(.020)
-.009*
(.004)
-.007*
(.003)
R
2
.540 .607 .386 .596
Log Likelihood 1128.79 1187.42 1007.57 1169.26
Note: Table displays spatial interaction cross-product terms. Full model includes additional
predictor variables (turnout, income, % bach deg, % repub regis, stability, and four racial group pop
density measures). N = 837 census tracts. Dependent variable is percent white Hahn voters.
Standard errors in parentheses. Bolded results indicate a change in significance level between OLS
and spatial lag regression models. *** p < .001, ** p < .01, * p < .10

123
Additionally, as white population density increases white voters were more likely
to support Villaraigosa. This result points to the difficulty of placing complex multiracial
context analysis within the dichotomous conflict / cooperation models. White voters in
Los Angeles are more likely to support a minority candidate when they live in more
homogenous white areas, a result that does not seem consistent with the idea of positive
racial contact, but also not entirely consistent with theories of group threat.
The local spatial interaction models complicate the analysis of white voters, as
shown in Tables 4.7 and 4.8. Both tables show that the impact of place/race interaction
changes the way white voters respond to racial context. Table 9 reveals that white
Villaraigosa voters were positively influenced by Latino population density. The
relationship becomes more apparent in the spatial lag model. Moreover, Table 4.8
illustrates that white voters were less likely to support Hahn in West L.A. as Asian
population density increases. Similarly, white voters in the Harbor region were less
likely to support Hahn as black population density increases. The spatial lag models
show that both of these results exhibited behavioral contagion through space. Again, it is
difficult to place these results within the racial cooperation / conflict dynamic, but I
interpret the results to mean that white voters were more willing to cross racial lines to
support a minority candidate in the presence of more dense minority populations in West
L.A. and the Harbor area.

124
Racial Context: Black Voters
Unlike the analysis of white voting behavior, the behavior of black voters
provides strong evidence of positive contact and racial group cooperation, at both the
global and local levels. As Table 4.9 illustrates, black Villaraigosa voters were positively
affected by Latino population density, and the spatial lag model makes this relationship
even more apparent. Looking at the local spatial interaction models in Table 4.10 we see
that the relationship between black and Latino population changes depending on location.
In the Mid City area, a racially diverse region where Villaraigosa performed strongly, the
positive influence of Latino population density on Black support for Villaraigosa
continues, and exhibits spatial contagion. However, in the East Valley, a predominantly
Latino area, Latino population density is a negative influence on black Villaraigosa
voters. This result is more consistent with a racial group conflict interpretation.
However, Table 4.11, the local spatial interaction models for Black Hahn voters,
presents a more ambiguous picture of racial context voting. In the Downtown region, we
see that Black voters were more likely to support Hahn as Latino population density
increases. Is this evidence of racial group conflict? It certainly appears that in the
Downtown region, unlike Black voters in other areas of Los Angeles, Black voters were
influenced by Latino population density away from supporting the Latino candidate.
That result seems consistent with some form of group conflict.

125
Table 4.9
Black Voters Global OLS and Spatial Lag Regression Models
% Black Villaraigosa % Black Hahn
OLS Spatial Lag OLS Spatial Lag

Spatial Lag (ρ) --- .367***
(.037)
--- .339***
(.038)
Constant .439***
(.017)
.257***
(.023)
.356***
(.013)
.228***
(.019)
Black Turnout .376***
(.049)
.279***
(.047)
.367***
(.039)
.296***
(.038)
Income
(thousands)
.0002
(.0001)
.0002*
(.0001)
.00008
(.00009)
.00008
(.00009)
% Bachelors Degree .040*
(.018)
.048**
(.017)
.022
(.014)
.028*
(.014)
% Republican Regis .114***
(.022)
.080***
(.022)
-.082***
(.018)
.080***
(.017)
% Native -.042**
(.015)
-.042**
(.015)
-.062***
(.012)
-.057***
(.012)
Residential Stability .078***
(.016)
-.035*
(.022)
.034**
(.013)
.031*
(.012)
Latino Pop Density
(thousands/mi.
2
)
.0006**
(.0002)
.00006***
(.0002)
.00004
(.0002)
.0001
(.0001)
Black Pop Density
(thousands/mi.
2
)
.003***
(.0008)
.002**
(.0008)
.002***
(.0006)
.002***
(.0006)
White Pop Density
(thousands/mi.
2
)
-.0006
(.0004)
-.0005
(.0004)
.0003
(.0004)
.0002
(.0003)
Asian Pop Density
(thousands/mi.
2
)
-.0004
(.0004)
-.0005
(.0004)
-.0004
(.0003)
-.0003
(.0003)
Spatial Dummies
West Valley -.010
(.006)
-.006
(.005)
-.006
(.005)
.001
(.004)
East Valley -.009*
(.005)
-.003
(.004)
-.003
(.004)
.001
(.004)
Downtown --- --- --- ---
South LA -.008
(.006)
-.004
(.006)
-.013**
(.005)
-.011*
(.004)
Harbor .025***
(.007)
.017**
(.007)
-.029***
(.006)
-.014*
(.006)
Mid City -.014**
(.005)
-.012**
(.005)
-.020***
(.004)
-.017***
(.004)
West LA -.005
(.006)
-.004
(.006)
-.007
(.005)
-.003
(.005)
R
2
.154 .249 .247 .314
Log Likelihood 1611.18 1655.42 1802.93 1837.60
Note: N = 837 census tracts. Dependent variable is percent racial group vote for each candidate.
Standard errors in parentheses. Bolded results indicate a change in significance level between OLS
and spatial lag regression models. *** p < .001, ** p < .01, * p < .10

126

Table 4.10
Black Villaraigosa Voters Local Spatial Interaction Models
East Valley Mid City
Model Type OLS Spatial Lag OLS Spatial Lag

Spatial Lag (ρ) --- .366***
(.039)
--- .391***
(.034)
Constant .443***
(.017)
.254***
(.025)
.458***
(.016)
.259***
(.022)
Spatial Dummy .069*
(.032)
.086**
(.030)
.001
(.013)
.002
(.011)
Space * Income -.0001
(.0003)
-.00002
(.0003)
-.0007
(.0003)
-.0001
(.0003)
Space * % Bach Deg -.054
(.040)
-.022
(.037)
.307***
(.041)
.296***
(.038)
Space * % Repub Regis -.016
(.051)
-.034
(.047)
-.273***
(.058)
-.259***
(.053)
Space * % Native -.015
(.048)
-.001
(.045)
-.295***
(.036)
-.292***
(.033)
Space * Res Stability -.058
(.044)
-.109**
(.041)
.207***
(.029)
.209***
(.027)
Space * Latino Pop Density -.004***
(.007)
-.002***
(.0007)
.0009**
(.0003)
.0005*
(.003)
Space * White Pop Density -.0008
(.0002)
-.002
(.0001)
-.001
(.002)
-.0008
(.0008)
Space* Black Pop Density .02*
(.007)
.003
(.007)
.007***
(.002)
.008***
(.002)
Space * Asian Pop Density .009*
(.006)
.004
(.005)
-.001
(.001)
.0008
(.0009)
R
2
.161 .246 .243 .349
Log Likelihood 1614.75 1653.87 1657.59 1715.12
Note: Table displays spatial interaction cross-product terms. Full model includes additional predictor
variables (turnout, income, % bach deg, % repub regis, stability, and four racial group pop density
measures). N = 837 census tracts. Dependent variable is percent black Villaraigosa voters. Standard
errors in parentheses. Bolded results indicate a change in significance level between OLS and spatial
lag regression models. *** p < .001, ** p < .01, * p < .10

127
Table 4.11
Black Hahn Voters Local Spatial Interaction Models
South LA Downtown
Model Type OLS Spatial Lag OLS Spatial Lag

Spatial Lag (ρ) --- .352***
(.039)
--- .339***
(.039)
Constant .355***
(.014)
.224***
(.019)
.344***
(.014)
.220***
(.018)
Spatial Dummy .001
(.013)
.0004
(.013)
-.002
(.017)
.002
(.016)
Space * Income .0004
(.0007)
.0003
(.0007)
.005
(.004)
.0004
(.0004)
Space * % Bach Deg -.167*
(.081)
-.129*
(.076)
.033
(.051)
.037
(.048)
Space * % Repub Regis -.031
(.110)
-.029
(.103)
.020
(.079)
.031
(.075)
Space * % Native .091*
(.055)
.065
(.051)
.121**
(.039)
.097**
(.037)
Space * Res Stability -.096*
(.049)
-.075
(.046)
-.129***
(.040)
-.121**
(.038)
Space * Latino Pop Density .00008
(.0006)
.0001
(.0006)
.0009*
(.0004)
.0008*
(.0004)
Space * White Pop Density -.02
(.03)
-.02
(.03)
-.02**
(.005)
-.01**
(.005)
Space* Black Pop Density .0005
(.0002)
.001
(.002)
-.003
(.004)
-.003
(.003)
Space * Asian Pop Density .003
(.01)
.009
(.01)
.005**
(.002)
.004**
(.002)
R
2
.212 .291 .233 .349
Log Likelihood 1784.24 1823.60 1795.19 1715.12
Note: Table displays spatial interaction cross-product terms. Full model includes additional predictor
variables (turnout, income, % bach deg, % repub regis, stability, and four racial group pop density
measures). N = 837 census tracts. Dependent variable is percent black Hahn voters. Standard errors in
parentheses. Bolded results indicate a change in significance level between OLS and spatial lag
regression models. *** p < .001, ** p < .01, * p < .10

128
Racial Context: Asian Voters
The analysis of Asian voters in the 2005 L.A. mayoral election demonstrates
again the importance of modeling racial context as it is situated in place, and with an
understanding that place-context works at different levels of analysis. A standard
analysis of Asian voting in the 2005 mayoral election would have looked at exit poll
results and noted that Asians were the only group that voted a clear majority for Hahn. It
is difficult from this perspective to conclude that there is evidence of racial conflict
between Asians and whites in Los Angeles. Based on further attitudinal responses, a
researcher might conclude that Asian voters were reluctant to support at Latino candidate
based on racial attitudes, or feelings of group conflict. Moreover, a contextual analyst
that looks only at the citywide level of analysis will not find evidence of conflict. As
Table 4.12 illustrates, only the black and Asian population density measures are
significant, neither result indicating a clear racial group conflict. White and Latino
population density measures are not significant to Asian voters at the global level.
However, the local spatial interaction models portray a different story. Table 4.13
indicates that the Asian Villaraigosa vote declines as Latino population density increases
in the Downtown area. Conversely, in the Mid City region we see the opposite result;
Asian support for Villaraigosa increases as Latino population density increases. We see a
continuation of the pattern in Table 4.14. In Downtown, Asian support for Hahn
increases with white population density. But, in the Harbor region, Asian support for
Hahn decreases as white density increases. This result suggests a level of conflict
between Asians and whites in the Harbor region.

129
Table 4.12
Asian Voters Global OLS and Spatial Lag Regression Models
% Asian Villaraigosa % Asian Hahn
Model Type OLS Spatial Lag OLS Spatial Lag

Spatial Lag (ρ) --- .388***
(.052)
--- .401***
(.052)
Constant .175***
(.028)
.046
(.032)
.772***
(.031)
.522***
(.043)
Asian Turnout .368***
(.035)
.343***
(.034)
-.222***
(.039)
-.208***
(.037)
Income
(thousands)
-.0001
(.0003)
-.0002
(.0002)
-.0002
(.0003)
-.0001
(.0003)
% Bachelors Degree .014
(.040)
.023
(.038)
.091*
(.044)
.087*
(.042)
% Republican Regis .320***
(.050)
.255***
(.049)
-.248***
(.055)
-.217***
(.053)
% Native .161***
(.035)
.135***
(.033)
-.256***
(.038)
-.226***
(.036)
Residential Stability .089*
(.037)
.085*
(.035)
-.008
(.041)
-.013
(.036)
Latino Pop Density
(thousands/mi.
2
)
-.00006
(.0004)
-.00002
(.0004)
-.0006
(.0004)
-.0003
(.0004)
Black Pop Density
(thousands/mi.
2
)
.005**
(.002)
.003*
(.002)
.0002
(.001)
.001
(.002)
White Pop Density
(thousands/mi.
2
)
.001
(.0009)
.001
(.0009)
.0002
(.001)
-.0002
(.001)
Asian Pop Density
(thousands/mi.
2
)
.004***
(.0009)
.003***
(.0009)
-.004***
(.001)
-.004***
(.001)
Spatial Dummies
West Valley -.001
(.016)
-.014
(.015)
.017
(.017)
.026
(.017)
East Valley -.014
(.014)
-.016
(.013)
.031*
(.015)
.029*
(.014)
Downtown .008
(.014)
-.018
(.014)
-.002
(.016)
.014
(.015)
South LA --- --- --- ---
Harbor .067***
(.018)
.023
(.017)
-.041*
(.019)
-.003
(.019)
Mid City -.004
(.012)
-.010
(.015)
.012
(.014)
.008
(.013)
West LA -.030*
(.016)
-.038
(.015)
.025
(.017)
.027*
(.017)
R
2
.324 .379 .224 .291
Log Likelihood 942.445 972.25 866.87 898.37
Note: N = 837 census tracts. Dependent variable is percent racial group vote for each candidate.
Standard errors in parentheses. Bolded results indicate a change in significance level between OLS
and spatial lag regression models. *** p < .001, ** p < .01, * p < .10

130

Table 4.13
Asian Villaraigosa Voters Local Spatial Interaction Models
Downtown Mid City
Model Type OLS Spatial Lag OLS Spatial Lag

Spatial Lag (ρ) --- .395***
(.049)
--- .388***
(.049)
Constant .148***
(.026)
.018
(.029)
.186***
(.028)
.465
(.032)
Spatial Dummy .019
(.047)
-.002
(.044)
.0006
(.029)
-.006
(.028)
Space * Income -.0001
(.000001)
-.0003
(.001)
.001
(.0007)
.001*
(.0007)
Space * % Bach Deg -.206
(.139)
-.194
(.131)
-.143
(.096)
-.155*
(.090)
Space * % Repub Regis -.419*
(.217)
-.281
(.205)
-.293*
(.135)
-.198
(.128)
Space * % Native .201*
(.108)
.197*
(.101)
.170*
(.083)
.188*
(.078)
Space * Res Stability .017
(.109)
-.021
(.103)
-.077
(.068)
-.103*
(.064)
Space * Latino Pop Density -.004**
(.001)
-.003*
(.001)
.003***
(.0008)
.002**
(.0007)
Space * White Pop Density -.02*
(.01)
-.02*
(.01)
-.0008
(.002)
-.0006
(.002)
Space* Black Pop Density -.02*
(.009)
-.02*
(.009)
-.008*
(.005)
-.008*
(.004)
Space * Asian Pop Density .03***
(.005)
.03***
(.04)
-.007**
(.002)
-.006*
(.002)
R
2
.348 .411 .337 .399
Log Likelihood 957.89 993.94 950.93 986.03
Note: Table displays spatial interaction cross-product terms. Full model includes additional predictor
variables (turnout, income, % bach deg, % repub regis, stability, and four racial group pop density
measures). N = 837 census tracts. Dependent variable is percent Asian Villaraigosa voters. Standard
errors in parentheses. Bolded results indicate a change in significance level between OLS and spatial
lag regression models. *** p < .001, ** p < .01, * p < .10

131
Table 4.14
Asian Hahn Voters Local Spatial Interaction Models
Downtown Harbor
Model Type OLS Spatial Lag OLS Spatial Lag

Spatial Lag (ρ) --- .384***
(.050)
--- .332***
(.052)
Constant .811***
(.028)
.563***
(.041)
.757***
(.027)
.556***
(.041)
Spatial Dummy -.005
(.051)
.008
(.047)
.264
(.191)
.178
(.183)
Space * Income .00007
(.001)
.0002
(.001)
.004
(.003)
.001
(.003)
Space * % Bach Deg .162
(.151)
.153
(.142)
-.978**
(.355)
-.762*
(.342)
Space * % Repub Regis .337
(.236)
.235
(.223)
.705*
(.365)
.673*
(.349)
Space * % Native -.116
(.117)
-.122
(.110)
-.224
(.283)
-.069
(.271)
Space * Res Stability -.116
(.119)
-.067
(.112)
-.529**
(.213)
-.412*
(.204)
Space * Latino Pop Density .004***
(.001)
.003**
(.001)
-.003
(.003)
-.002
(.003)
Space * White Pop Density .02*
(.01)
.02
(.01)
-.03**
(.01)
-.02*
(.01)
Space* Black Pop Density .02*
(.01)
.02
(.01)
.003
(.03)
.03
(.03)
Space * Asian Pop Density -.03***
(.005)
-.02***
(.005)
.04
(.02)
.02
(.02)
R
2
.261 .326 .266 .310
Log Likelihood 887.27 919.69 890.05 911.63
Note: Table displays spatial interaction cross-product terms. Full model includes additional predictor
variables (turnout, income, % bach deg, % repub regis, res stability, and four racial group pop density
measures). N = 837 census tracts. Dependent variable is percent Asian Hahn voters. Standard errors in
parentheses. Bolded results indicate a change in significance level between OLS and spatial lag
regression models. *** p < .001, ** p < .01, * p < .10

132
CONCLUSION: THE INTERACTION OF RACE AND PLACE
Place-context alters the determinative impact of variables such as social class,
residential stability, and even variations in the racial context environment. The
explanatory impact of these forces is partially due to spatial contagion from
neighborhood to neighborhood, which is consistent with the concept of place-context
presented in this dissertation. This finding confirms the findings related to the
examination of the 2001 Los Angeles mayoral election in Chapter 3. Voting in Los
Angeles is spatially structured with numerous instances of behavioral contagion situated
in particular locations.
The primary focus of this chapter is to look deeper into the spatial dynamics of
racial voting behavior and the interaction effects of racial context and place. Based on
the theory of place-context and the findings of spatial contagion from the previous
chapter, my argument is that two forces of place and race will interact to produce
complex patterns of racial voting that defy general patterns. The finding that Harbor area
Latino voters were less likely to support Villaraigosa is illustrative of the importance of
place-context. Although it may not have changed the overall outcome, or the generally
overwhelming support given by Latinos to Villaraigosa, it does suggest that we will not
fully understand racial voting unless we take place-context into consideration.
These findings are consistent with the work of scholars, such as Liu (2001), that
argue that analysis of urban voting behavior and political attitudes should move beyond
searching for general patterns of racial cooperation versus conflict. Once political

133
behavior is situated in place-context, we see complex behavioral dynamics associated
with both conflict and cooperation that diffuse through space and place in ways that defy
general patterns.
Previous examinations of racial context and voting in recent Los Angeles
elections focused on patterns of group conflict, especially between Black and Latino
groups. Overall, these results stand in contrast to studies that only find racial group
conflict between Black and Latino voters in Los Angeles. On the global, citywide level,
the findings may be consistent with Kaufmann’s contention that racial context consists
partially of individual perceptions of group conflict. It is certainly possible that
perceptions of conflict between black and Latino groups were not prominent in 2005.
My argument, however, is that one type of relationship may be apparent at the citywide
level, different relationship dynamics will exist at the neighborhood level, and they will
vary from place to place. My analysis suggests that, in several Los Angeles
neighborhoods, place-context and racial voting are interacting in such a way as to
produce group cooperation between Blacks and Latinos. At the same, areas such as the
East Valley indicate the presence of racial group conflict voting. Thus, although the
findings of cooperation rather than conflict between Black and Latino groups does not by
itself contradict Kaufmann’s theory of the urban voter, the finding of conflict in the East
Valley makes it clear that place-context is an additional dimension that must be taken into
account in order to properly understand racial context and voting behavior in a
multiracial environment.

134
Contextual effects of race occur in place and through space. Racial context
effects vary from place to place, neighborhood to neighborhood. Although it is tempting
for scholars of racial context to find general patterns of positive social interaction and
group cooperation, or attitudes and behaviors associated with feelings of racial threat and
group conflict, my analysis indicates that findings not situated within space and place
may be misleading. In order to properly assess contextual environments of conflict or
cooperation, it is necessary to consider the interaction of place and race, and that place-
context may alter the way that racial conflict and cooperation is expressed through
voting.

135
CHAPTER 5 CONCLUSION: IMPLICATIONS AND FUTURE
RESEARCH
This dissertation attempts to contribute to the study of political behavior, and
especially to answer questions about how voters are influenced by the contextual
environment that surrounds them. I begin with a review of the literature that found
political scientists tend to ignore place and space as factors that influence political
behavior. This holds even for those political scientists who are interested in contextual
explanations of politics. I argue that the concept of context-as-place, borrowed from
political geography, is a productive theoretical perspective that provides a way to bring
together the various ways in which political scientists deploy contextual analysis. Place-
context structures and channels the many causal forces, the multiple strands of identity,
culture, and political information that are situated within the built environment of
American cities and influence the way voters behave.
Place-context analysis is a conceptual bridge between scholars of individual
political behavior and those scholars who are exploring social context as a critique of
previous behavioral political science. In fact, I point out that there is a strand of place-
context thinking contained within the founding text of American political behavior
researcher, The American Voter. This suggests that a place-context theoretical
foundation for contextual effects researcher is not entirely out of the mainstream of
political science, but rather offers a corrective to the tendency to reduce political behavior
to the attitudinal preferences of isolated individuals. Consistent with the oft-stated desire

136
of contextual scholars to examine political behavior in ways that reflect the complexities
and particularities of political phenomena are situated within a web of social and
community relations, I develop a method of analyzing place-context effects that is based
on recent advances in ecological inference and spatial econometrics.
Once we conceive of political and social context as place-context, the question
then becomes whether or not place-context is an influential determinant of political
behavior, or whether contextual effects are merely artifacts of the demographic
composition of neighborhoods and communities. My findings show that place-context
has an independent effect on voting behavior, and that voting varies from place to place
through a behavioral contagion process even while controlling for many of the common
determinants of the vote choice like partisanship, class, and race/ethnicity. I also find
intriguing interaction effects between place and racial context, including behavior that
varies from cooperation to conflict and from place to place within the same election.
PLACE-CONTEXT AS CAUSAL MECHANISM
The research in this dissertation has important implications for several bodies of
political science literature. First, my findings regarding the myriad ways that place-
context influences voting behavior have implications for the debate over the causal
mechanism of contextual effects. Another way to think of this question is, what, exactly,
is going on here? What is causing the observed behavioral effects?
There are three basic approaches to the question of what is, exactly, the
operational mechanism that causes context to matter to political behavior. The most

137
common understanding is that social interaction is the engine driving social context.
Voters are influenced by the information they receive through personal, even face-to-
face, interaction with their friends, family, and neighbors. The empirical findings
provided by Huckfeldt and his colleagues over the past few decades are strong evidence
for the conclusion that social interaction is a powerful causal mechanism of contextual
effects. Yet, social interaction is not the entire story of context, and a singular focus on
interaction leads away from a focus on place. Burbank offers a political psychological
explanation for the causal mechanism of contextual effects. His findings emphasize the
determining influence of individual perception of the partisan environment for political
behavior (Burbank 1995; Burbank 1997). Burbank finds no evidence for the social
interaction or personal contact model of context. Cho and Rudolph (2007) add to the
debate through examination of the spatial structure of political participation. They find
that social interaction and other individual level variables such as SES cannot fully
explain the spatial patterns in their data. Social interaction is an important factor, but not
the only cause of contextual effects.
The implication of this result is that, with respect to political participation,
geographical proximity matters in ways that are not sufficiently accounted for by
existing theories of contextual effects and the mechanisms are likely characteristic
of diffusion processes. Psychologically, our results also imply that the genesis
and spread of ideas is not wholly dependent upon explicit communication with
other people. Simply observing those around us, but not communicating with
them, may be how we pick up many of our ideas, a point that is often
unacknowledged in the literature because adult life seems to revolve so much
around social communication. (Cho and Rudolph 2007, 24)

They suggest as an alternative causal mechanism that people are influenced by what they
call “low-intensity environmental cues.” The idea is that as people move through space

138
and place they are influenced by elements of their environment (they suggest the
profusion of lawn signs in particular neighborhoods as a possibility) that have nothing to
do with social interaction between individuals. These “low-intensity environmental
cues” cause the diffusion of political information throughout particular place-based
contexts. The causal mechanism suggested by Cho and Rudolph is entirely consistent
with a theoretical commitment to place-context. Moreover, it is an indication that a focus
on context as social interaction within personal networks is important, but not sufficient
for a complete understanding of social context.
My results support Cho and Rudolph’s conclusion that there is a spatial
component to the causal mechanism of contextual effects. Place and space are influential
determinants of political behavior independent of social interaction networks of friends
and families. Those researchers who have focused on social interaction have done
important work to advance the understanding of context and social networks. However,
by locating context in place, my research points the way towards future research into
understanding the cause of contextual effects beyond social interaction. Social
interaction is an important determinant of voting behavior, and social interaction is
bounded by space and place. But, there is more to place-context than social interaction.
Cho and Rudolph hesitate to speculate further about the theoretical implications
of “low-intensity environmental cues” as causal mechanism of context. However, I argue
that their results provide a link between rational choice theories and contextual theories
of voting behavior once we conceive of the environmental cues as heuristic information
shortcuts. Heuristic information shortcuts are recognized as important instruments of

139
rational vote choice under a theory of low-information rationality (Cutler 2002; Popkin
1991). Absent the costs of total information awareness, voters utilize informational
shortcuts associated with partisanship and socio-demographic background to evaluate
political campaigns, candidates, issues, and events. My contention is that, at least on a
local level, voters used place-based heuristics – information shortcuts gleaned from their
neighborhoods and other places important to them – to evaluate their attitudes towards
issues, their response to candidates and campaigns, and to shape their sense of political
identity.
For example, a voter might use a candidate’s background in a particular
neighborhood to evaluate that candidate’s worthiness. The voter might be impressed by a
candidate’s work with a local neighborhood organization. She might be more receptive
to a campaign speech about education if it takes place at a neighborhood elementary
school. At the same time, the voter might evaluate the incumbent’s job performance
based on her experience of graffiti in her neighborhood, or the general decline of local
shops. Her political identity might also be shaped by her experience of and connection to
a particular neighborhood, and this might dispose her to view one candidate more
favorably than another. She might use the political endorsement granted by local
organizations and local news media as an information shortcut for candidate evaluation.
Political campaigns will tailor campaign literature and cable television advertisements to
specific neighborhoods.
Contextual effects produced by such place-based heuristics, or “low-intensity
environmental cues,” are not caused by social interaction with friends and neighbors.

140
Additionally, they are not caused by individual perception of partisan context. They are
practically unique effects associated with particular places. However, the research
presented here cannot directly address the questions about place-based heuristics. They
only provide indirect evidence of their effects. More research is required to more fully
understand the ways that voters are influenced by place-heuristics. For instance, analysts
of campaigns and elections rarely find strong effects produced by knowledge of issues or
by campaign tactics. Yet, if we situate voters attitudes towards issues and the effects of
campaign tactics in place-context we may see impacts on behavior that are not seen larger
scales such as statewide or national level elections.
SELF-SELECTION EFFECTS
Self-selection bias is another possible explanation for many of the findings
regarding contextual effects. Critics may charge that the observed effects related to
place-context are in reality produced by individual self-selection, or the tendency of
people to sort themselves into likeminded communities. Instead of observing real
behavior influenced by the place-context environment, we are seeing effects related to
pre-existing attitudes regarding residential choice. Concerns over self-selection bias are
particularly acute among those studying racial attitudes amid the context of residential
segregation (Oliver and Wong 2003; Pettigrew 1998). Oliver and Wong (2003) as well
as Oliver (2001) institute individual-level controls for selection effects by including
measures of socioeconomic status and home ownership. Although the researched
presented in this dissertation does control for neighborhood-level measures of SES and

141
partisanship, these are not adequate measures to control for individual self-selection
choices. Another way to control for self-selection effects is to incorporate the dimension
of time. Voss (2004) suggests that empirical research that is dynamic with respect to
time as well as context variation will do a better job of separating out self-selection
effects. Further research is required to more fully understand the impact of self-select
effects on political behavior, especially by researchers who are interested in place-
context.
PLACE-CONTEXT AND HISTORICAL INSTITUTIONALISM
My research into place-context effects also has implications for scholars of
historical institutionalism and American political development. Institutionalists have
been very successful in delineating the ways in which institutions establish the norms for
political behavior, assign roles, and operate within the boundaries of institutional rules
and procedures. The work of institutionalists is often replete with spatial metaphors –
institutions are “sites of political conflict, and the “path dependence” of political
processes – but they have not thought enough about the implications of these spatial
metaphors. Institutions have “spatial footprints.” They can have an enormous amount of
influence on the places over which they have jurisdiction. Many institutions of
government authority exhibit a literate control over space and place.
More research from contextual effects scholars is needed to incorporate
institutions into the place-context environment. Local knowledge is required to assess
the ways in which institutions set the conditions for place-context political behavior.

142
Institutionalists have done a good job of bringing historical thinking back into political
analysis, but they also need to incorporate spatial thinking into their institutional
analyses. Institutionalists need to think more about the ways in which institutions shape
place-context, and also about how place-context behavior feeds back into the institutional
rules and procedures.
POLITICAL PARTICIPATION AND LOCAL DEMOCRACY
The research presented in this dissertation has implications for political scientists
that study civic engagement, social capital, and participation in urban political coalitions.
The ongoing theme of this dissertation is that places matter to politics, that the often
unique qualities of particular places, the culture, institutions, relations, and people that
characterize particular places, will shape the nature of politics, especially at the local
level. However, the thesis of much modern political research is that places are losing
their particularity; the bonds of community social interaction are eroding. Unique places
have been turned into homogenized spaces, controlled and shaped by various powerful
forces such as government authority or corporate capital. Robert Putnam’s research on
social capital can be read from this point of view as the vanishing of places and
communities from the American landscape. Putnam’s work on social capital suggests
that social capital is a function of individual and community connectivity (Putnam 2000).
Putnam identifies two types of community social capital, community bonding which
tends to be exclusionary in nature, and community bridging which is more inclusive,
encouraging people to reach across boundaries and ameliorate differences. Social capital,

143
especially the type that bridges and strengthens community bonds has declined. Putnam
states “Americans today feel vaguely and uncomfortably disconnected” (402). The
copious evidence produced in his opus shows that this feeling is the product of real
changes and has real consequences. “Americans are right that the bonds of our
communities have withered, and we are right to fear that this transformation has very real
costs [emphasis original]” (402).
Scholars of racial context and political coalitions see a withering in the ability of
urban leaders to build cross-racial political coalitions that can lead to a lessening of racial
tension in American cities. Both Kaufmann (2004) and Sonenshein and Drayse (2006)
conclude in their studies of racial context in Los Angeles that the political process is not
resolving racial conflict but rather enhancing it. Their analysis sees Los Angeles politics
lacking in social capital, lacking the types of community networks that bridge racial
divides. By situating racial context within particular place-contexts, my research
provides evidence that at the neighborhood level, it is possible to see instances of both
inclusive, bridging interaction as well as exclusionary and conflictual interaction. Place-
context fostered a level of interaction among racial groups that, from the perspective of
racial polarization, had a positive impact on racial group voting behavior, and the ability
of communities to form bridges across racial lines.
Clarence Stone conducts research into urban civic capacity, a concept that is
similar to Putnam’s social capital, but which is focused on the ability of urban political
leaders to build governing coalitions and achieve policy goals (Sonenshein and Drayse
2006; Stone 2001). He notes that civic capacity varies from place to place in American

144
communities. My findings indicate that there are particular neighborhoods and
communities in Los Angeles that are more likely to be fertile ground for those interested
in sowing the seeds of grassroots engagement in local communities. The place-context
environment of some neighborhoods is amenable for those political leaders who may
want to build civic capacity in Los Angeles in order to accomplish important policy
objectives.
Does the homogenization of American urban space and erosion of community
social capital affect the quality of local democracy? In his work on suburban context
Oliver engages with the concept of social capital and civic engagement to examine the
nature of democracy in American suburbs (Oliver 2001). Oliver’s analysis comes to two
somewhat conflicting conclusions about democracy in suburbia. First, because of their
reduced size, suburbs do foster some types of political participation. Rather than the
wasteland of civically unengaged residents, suburbs are sometimes venues for vibrant
political participation. However, suburbanization tends to institutionalize racial and class
segregation behind municipal boundaries. From a regional perspective, the increasing
suburbanization and political fragmentation of American metropolitan areas prevents
resolution of political conflict within democratic processes, and erodes the quality of
local democracy.
Control over space and place can have enormous consequences for how well local
political processes operate. Oliver shows that one of the most common forms of spatial
control – suburbanization and political fragmentation – is having a corrosive effect on
local democracy. My inquiry into the politics that places make shows that places have a

145
role in producing a diversity of political outcomes. Place-context environments can
foster positive and negative effects on the quality of political participation in the
democratic process. Vibrant places – neighborhoods and communities that incorporate
diversity and whose environments are conducive to the resolution of conflict and tension
– are important indicators of the quality of local democracy in urban America. This
understanding should guide further research into the ways that place-context can shape
political behavior.

146
BIBLIOGRAPHY
Abrajano, Marisa A., Jonathan Nagler, and R. Michael Alvarez. 2005. A Natural
Experiment of Race-Based and Issue Voting: The 2001 City of Los Angeles
Elections. Political Research Quarterly 58 (2):203-218.

Achen, Christopher H., and W. Phillips Shively. 1995. Cross-Level Inference. Chicago:
University of Chicago Press.

Agnew, John. 1996a. Mapping politics: how context counts in electoral geography.
Political Geography 15 (2):129-146.

Agnew, John. 1996b. Maps and models in political studies: a reply to comments.
Political Geography 15 (2):165-167.

Agnew, John. 2002. Making Political Geography. New York: Oxford University Press.

Agnew, John, and James Duncan, eds. 1989. The Power of Place: Bringing Together
Geographical and Sociological Imaginations. London: Unwin Hyman.

Agnew, John, and Jonathan M. Smith, eds. 2002. American Space / American Place:
Geographies of the Contemporary United States. New York: Routledge.

Anselin, L., R.J.G.M. Florax, and S.J. Rey, eds. 2003. Advances in Spatial Econometrics.
Methodology, Tools and Applications. Berlin: Springer Verlag.

Anselin, Luc. 1988. Spatial Econometrics: Methods and Models. Dordrecht: Kluwer
Academic Publishers.

Anselin, Luc. 2003. GeoDA 0.9 User's Guide: Spatial Analysis Laboratory, University of
Illinois, Urbana-Champaign.

147
Anselin, Luc. 2005. Exploring Spatial Data with GeoDa: A Workbook. Urbana, IL:
University of Illinois, Spatial Analysis Laboratory.

Anselin, Luc, and Wendy K. Tam Cho. 2002. Spatial Effects and Ecological Inference.
Political Analysis 10 (3):276-297.

Barreto, Matt A., Fernando Guerra, Mara Marks, Stephen A. Nuño, and Nathan D.
Woods. 2006. Controversies in Exit Polling: Implementing a Racially Stratified
Homogenous Precinct Approach. PS: Political Science and Politics 39 (July):477-
483.

Baybeck, Brady, and Robert Huckfeldt. 2002. Urban contexts, spatially dispersed
networks, and the diffusion of political information. Political Geography 21:195-
220.

Berelson, Bernard R., Paul F. Lazarsfeld, and William N. McPhee. 1954. Voting: A Study
of Opinion Formation in a Presidential Campaign. Chicago, Ill.: University of
Chicago Press.

Books, John W., and Charles L. Prysby. 1991. Political Behavior and the Local Context.
New York: Praeger.

Books, John W., and Charles L. Prysby. 1999. Contextual Effects on Retrospective
Economic Evaluations: The Impact of the State and Local Economy. Political
Behavior 21 (1):1-16.

Browning, Rufus P., Dale Rogers Marshall, and David H. Tabb. 1996. Racial Politics in
American Cities. 2nd ed. Princeton, New Jersey: Princeton University Press.

Burbank, Matthew J. 1995. The Psychological Basis of Contextual Effects. Political
Geography 14 (6/7):621-635.

Burbank, Matthew J. 1997. Explaining Contextual Effects on Vote Choice. Political
Behavior 19 (2):113-132.

148
Burden, Barry C., and David C. Kimball. 1998. A New Approach to the Study of Ticket
Splitting. In Controversies in Voting Behavior, 4th ed., edited by R. G. Niemi and
H. F. Weisberg. Washington, D.C.: CQ Press.

Calvo, Ernesto, and Marcelo Escolar. 2003. The Local Voter: A Geographically
Weighted Approach to Ecological Inference. American Journal of Political
Science 47 (1):189-204.

Campbell, Angus, P. Converse, W. Miller, and D. Stokes. 1960. The American Voter.
Chicago: University of Chicago Press.

Carsey, Thomas M. 1995. The Contextual Effects of Race on White Voter Behavior: The
1989 New York City Mayoral Election. The Journal of Politics 57 (1):221-228.

Ceaser, James W., and Daniel DiSalvo. 2006. Midterm Elections, Partisan Context, and
Political Leadership: The 2006 Elections and Party Alignment. The Form 4 (3).

Cho, Wendy K. Tam. 1998. Iff the Assumption Fits.: A Comment on the King Ecological
Inference Solution. Political Analysis 7:143-163.

Cho, Wendy K. Tam. 2003. Contagion Effects and Ethnic Contribution Networks.
American Journal of Political Science 47 (2):368-387.

Cho, Wendy K. Tam, and Charles F. Manski. 2006. Cross-Level / Ecological Inference.
In Oxford Handbook of Political Methodology.

Cho, Wendy K. Tam, and Thomas J. Rudolph. 2007. Emanating Political Participation:
Untangling the Spatial Structure Behind Participation. British Journal of Political
Science 37 (1).

Cleeland, Nancy. 2001. Unions Criticize Leader's Push for Villaraigosa. Los Angeles
Times, June 7, 2001, A27.

149
Cutler, Fred. 2002. The Simplest Shortcut of All: Sociodemographic Characteristics and
Electoral Choice. The Journal of Politics 64 (2):466-490.

Dahl, Robert. 1961. Who Governs? Democracy and Power in an American City. New
Haven, Conn.: Yale University Press.

Dear, Michael. 1990. Postmodern Geographies: The Reassertion of Space in Critical
Social Theory (Book Review). Annals of the Association of American
Geographers 80 (4):649-654.

Duncan, Otis Dudley, and Beverly Davis. 1953. An Alternative to Ecological Correlation.
American Sociological Review 18:665-666.

Erbring, Lutz, and Alice A. Young. 1979. Individuals and Social Structure: Contextual
Effects as Endogenous Feedback. Sociological Methods and Research 7:396-430.

Ethington, Philip J. 2001. Ghost Neighborhoods: Space, Time and Alienation in Los
Angeles. In Looking for Los Angeles: Architecture, Film, Photography, and the
Urban Landscape, edited by C. Salas and M. Roth. Los Angeles: Getty Research
Institute.

Ethington, Philip J., and Jason A. McDaniel. 2007. Political Places and Institutional
Spaces: The Intersection of Political Science and Political Geography. Annual
Review of Political Science 10:127-142.

Eulau, Heinz, and Lawrence Rothenberg. 1986. Life Space and Social Networks as
Political Contexts. Political Behavior 8 (2):130-157.

Finnegan, Michael. 2001. Mayoral Candidates Vow Support for Vote Next Year on
Secession. Los Angeles Times, B3.

Flint, Colin. 1996a. Whither the individual, whither the context? Political Geography 15
(2):147-151.

150
Flint, Colin. 1996b. Wither the Individual, Whither the Context? Political Geography 15
(2):147-151.

Foladare, Irving S. 1968. The Effect of Neighborhood on Voting Behavior. Political
Science Quarterly 83 (4):516-529.

Forest, Benjamin. 1995. West Hollywood as Symbol: The Significance of Place in the
Construction of Gay Identity. Environment and Planning D: Society and Space
13:133-157.

Forest, Benjamin. 2002. A New Geography of Identity? Race, Ethnicity, and American
Citizenship. In American Space / American Place: Geographies of the
Contemporary United States, edited by J. Agnew and J. M. Smith. New York:
Routledge.

Fotheringham, A. Stewart, Chris Brunsdon, and Martin E. Charlton. 2002.
Geographically Weighted Regression: The Analysis of Spatially Varying
Relationships. London: John Wiley & Sons, Ltd.

Gay, Claudine. 2001. The Effect of Black Congressional Representation on Political
Participation. American Political Science Review 95 (3):589-602.

Gay, Claudine. 2004. Putting Race in Context: Identifying the Environmental
Determinants of Black Racial Attitudes. American Political Science Review 98
(4):547-562.

Giles, Michael W., and Arthur Evans. 1985. External Threat, Perceived Threat, and
Group Identity. Social Science Quarterly 66:50-66.

Giles, Michael W., and Arthur Evans. 1986. The Power Approach to Intergroup Hostility.
Journal of Conflict Resolution 30 (469-486).

Giles, Michael W., and Kaenan Hertz. 1994. Racial Threat and Partisan Identification.
American Political Science Review 88 (2):317-326.

151
Goodin, Robert E., and Hans-Dieter Klingemann. 1996. Political Science: The Discipline.
In A New Handbook of Political Science, edited by R. E. Goodin and H.-D.
Klingemann. New York: Oxford University Press.

Goodman, Leo. 1953. Ecological Regressions and Behavior of Individuals. American
Sociological Review 18:663-664.

Gorov, Linda. 2001. LA Vote Is Seen Moving City to Left, Democrats Dominate; Face
Issues on Wages, Housing and Secession. Boston Globe, June 7, 2001, A3.

Harvey, David. 1989. The Urban Experience. Oxford: Basil Blackwell.

Harvey, David. 1990. The Condition of Postmodernity: An Enquiry into the Origins of
Social Change. Cambridge: Blackwell Publishers.

Hero, Rodney E., and Caroline J. Tolbert. 1996. A Racial/Ethnic Diversity Interpretation
of Politics and Policy in the States of the U.S. American Journal of Political
Science 40 (3):851-871.

Herron, Michael C., and Kenneth W. Shotts. 2003. Using Ecological Inference Point
Estimates in Second-Stage Linear Regressions. Political Analysis 11 (1):44-64.

Herron, Michael C., and Kenneth W. Shotts. 2004. Logical Inconsistency in EI-Based
Second Stage Regressions. American Journal of Political Science 48 (1):172-183.

Hogen-Esch, Tom. 2001. Urban Secession and the Politics of Growth: The Case of Los
Angeles. Urban Affairs Review 36 (6):783-809.

Hogen-Esch, Tom. 2002. Recapturing Suburbia: Urban Secession and the Politics of
Growth in Los Angeles, Boston, and Seattle, Depart of Political Science,
University of Southern California, Los Angeles.

Huckfeldt, Robert. 1979. Political Participation and the Neighborhood Social Context.
American Journal of Political Science 23 (3):579-592.

152

Huckfeldt, Robert. 1980. Variable Responses to Neighborhood Social Contexts:
Assimilation, Conflicts and Tipping Points. Political Behavior 2 (3):231-257.

Huckfeldt, Robert. 1986. Politics in Context: Assimilation and Conflict in Urban
Neighborhoods. New York: Agathon Press, Inc.

Huckfeldt, Robert. 1992. Review of Political Behavior and the Local Context by John W.
Books and Charles L. Prysby. The Journal of Politics 54 (4):1190-1192.

Huckfeldt, Robert, Paul E. Johnson, and John Sprague. 2005. Individuals, Dyads, and
Networks: Autoregressive Patterns of Political Influence. In The Social Logic of
Politics: Personal Networks as Contexts for Political Behavior, edited by A. S.
Zuckerman. Philadelphia: Temple University Press.

Huckfeldt, Robert, Eric Plutzer, and John Sprague. 1993. Alternative Contexts of
Political Behavior: Churches, Neighborhoods and Individuals. The Journal of
Politics 55 (2):365-381.

Johnston, Ron J., and Charles J. Pattie. 2005. Putting Voters in their Places: Local
Context and Voting in England and Wales, 1997. In The Social Logic of Politics:
Personal Networks as Contexts for Political Behavior, edited by A. S.
Zuckerman. Philadelphia: Temple University Press.

Kaufmann, Karen M. 2003. Black and Latino Voters in Denver: Responses to Each
Other's Political Leadership. Political Science Quarterly 118 (1):107-125.

Kaufmann, Karen M. 2004. The Urban Voter: Group Conflict and Mayoral Voting
Behavior in American Cities. Ann Arbor: University of Michigan Press.

Key, V.O. 1949. Southern Politics in State and Nation. New York: Alfred P. Knopf.

King, Gary. 1996. Why Context Should Not Count. Political Geography 15 (2):159-164.

153
King, Gary. 1997. A Solution to the Ecological Inference Problem: Reconstructing
Individual Behavior from Aggregate Data. Princeton: Princeton University Press.

King, Gary, and Christopher Adolph. 2003. Analyzing Second-Stage Ecological
Regressions: Comment on Herron and Shotts. Political Analysis 11 (1):65-76.

King, Gary, Christopher Adolph, Michael C. Herron, and Kenneth W. Shotts. 2003. A
Consensus on Second-Stage Analyses in Ecological Inference Models. Political
Analysis 11 (1):86-94.

King, Gary, Ori Rosen, and Martin A. Tanner, eds. 2004. Ecological Inference: New
Methodological Strategies. Cambridge: Cambridge University Press.

Kohfeld, Carol W., and John Sprague. 1995. Racial context and voting behavior in one-
party urban political systems. Political Geography 14 (6/7):543-569.

Kohfeld, Carol W., and John Sprague. 2002. Race, space, and turnout. Political
Geography 21:175-193.

Lefebvre, Henri. 1991. The Production of Space. Translated by D. Nicholson-Smith.
Oxford: Blackwell Publishers, Inc.

Leighley, Jan E. 2001. Strength in Numbers: The Political Mobilization of Racial and
Ethnic Minorities. Princeton: Princeton University Press.

Liu, Baodong. 2001. Racial Contexts and White Interests: Beyond Black Threat and
Racial Tolerance. Political Behavior 23 (2):157-180.

Liu, Baodong. 2003. Deracialization and Urban Racial Contexts. Urban Affairs Review
38 (4):572-591.

Lublin, David, and D. Stephen Voss. 2002. Context and Francophone Support for the
Sovereignty of Quebec: An Ecological Analysis. Canadian Journal of Political
Science 35 (1):75-101.

154

Meyerson, Harold. 1999. Cattle Call: The Race to Succeed Richard Riordan - and to
Shape the Future of Los Angeles - Has Already Begun. LA Weekly, August 20,
1999.

Meyerson, Harold. 2001. A City Hesitates at Political Change. New York Times, June 8.

Miller, Warren E., and J.M. Shanks. 1996. The New American Voter. Cambridge, Mass.:
Harvard University Press.

Niemi, Richard G., and Herbert F. Wiesberg, eds. 2001. Controversies in Voting
Behavior. 4th ed. Washington, DC: Congressional Quarterly Press.

O'Loughlin, John. 2002. The Exploratory Geography of Weimar Germany: Exploratory
Spatial Data Analysis of Protestant Support for the Nazi Party. Political Analysis
10 (3):217-243.

Oliver, J. E. 1999. The Effects of Metropolitan Economic Segregation on Local Civic
Participation. American Journal of Political Science 43 (1):186-212.

Oliver, J. E. 2000. City Size and Civic Involvement in Metropolitan America. The
American Political Science Review 94 (2):361-373.

Oliver, J. E. 2001. Democracy in Suburbia. Princeton: Princeton University Press.

Oliver, J. E., and Tali Mendelberg. 2000. Reconsidering the Environmental Determinants
of White Racial Attitudes. American Journal of Political Science 44 (3):574-589.

Oliver, J. Eric, and Janelle Wong. 2003. Intergroup Prejudice in Multiethnic Settings.
American Journal of Political Science 47 (4):567-582.

155
Oliver, J.E. 2003. Mental Life and The Metropolis in Suburban America: The
Psychological Correlates of Metropolitan Place Characteristics. Urban Affairs
Review 39 (2):229-253.

Parenti, Michael. 1967. Ethnic Politics and the Persistence of Ethnic Identification.
American Political Science Review 61:717-726.

Pettigrew, Thomas F. 1971. When a Black Candidate Runs for Mayor: Race and Voting
Behavior. In People and Politics in Urban Society, edited by H. Hahn. Beverly
Hills, CA: Sage.

Pettigrew, Thomas F. 1998. Intergroup Contact Theory. Annual Review of Psychology
49:65-85.

Pinkus, Susan, and Jill Darling. 2002. Poll Analysis: L.A. Breakup Proposal. Los Angeles
Times, July, 02, 2002.

Popkin, Samuel L. 1991. The Reasoning Voter: Communication and Persuasion in
Presidential Campaigns. 2nd ed. Chicago: University of Chicago Press.

Purcell, M. 2001. Metropolitan political reorganization as a politics of urban growth: the
case of San Fernando Valley secession. Political Geography 20:613-633.

Putnam, Robert D. 2000. Bowling Alone: The Collapse and Revival of American
Community. New York: Simon & Schuster.

Rainey, James, and Greg Krikorian. 2001. Hahn Won on His Appeal to Moderates,
Conservatives. Los Angeles Times, June 7, 2001, A1.

Rainey, James, and Beth Shuster. 2001. Mayoral Rivals Vie for Black Votes; Election:
Villaraigosa and Hahn announce raft of endorsements. Los Angeles Times, May 8,
2001, B3.

156
Saito, Leland T. 1998. Race and Politics: Asian Americans, Latinos, and Whites in a Los
Angeles Suburb. Chicago: University of Illinois Press.

Sauerzopf, Richard, and Todd Swanstrom. 1999. The Urban Electorate in Presidential
Elections, 1920-1996. Urban Affairs Review 35 (1):72-91.

Soja, Edward W. 1989. Postmodern Geographies: The Reassertion of Space in Critical
Social Theory. London: Verso.

Sonenshein, Raphael J. 1993. Politics in Black and White: Race and Power in Los
Angeles. Princeton, New Jersey: Princeton University Press.

Sonenshein, Raphael J. 2000. The Distinctiveness of Jewish Voting: A Thing of the Past?
Urban Affairs Review 35 (3):358-389.

Sonenshein, Raphael J., and Mark H. Drayse. 2006. Urban electoral coalitions in an age
of immigration: Time and place in the 2001 and 2005 Los Angeles mayoral
primaries. Political Geography 25:570-595.

Sonenshein, Raphael J., and Tom Hogen-Esch. 2006. Bringing the State (Government)
Back In: Home Rule and the Politics of Secession in Los Angeles and New York
City. Urban Affairs Review 41 (4):467-491.

Stone, Clarence. 2001. Civic Capacity and Urban Education. Urban Affairs Review 36
(5):595-619.

Sui, Danile Z., and Peter J. Hugill. 2002. A GIS-based spatial analysis on neighborhood
effects and voter turn-out: a case study in College Station, Texas. Political
Geography 21:159-173.

Tolbert, Caroline J., and Rodney E. Hero. 2001. Dealing with Diversity: Racial/Ethnic
Context and Social Policy Change. Political Research Quarterly 54 (3):571-604.

157
Voss, D. Stephen. 2004. Using Ecological Inference for Contextual Research. In
Ecological Inference: New Methodological Strategies, edited by G. King, O.
Rosen and M. A. Tanner. New York: Cambridge University Press.

Voss, Stephen D., and Penny Miller. 2001. Following a False Trail: The Hunt for White
Backlash in Kentucky's 1996 Desegregation Vote. State Politics and Policy
Quarterly 1 (March):63-82.

Wacquant, L. J. D. 1994. The New Urban Color Line: The State and the Fate of the
Ghetto in PostFordist American. In Social Theory and the Politics of Identity,
edited by C. Calhoun. Oxford: Blackwell.

Ward, Michael D., and John O'Loughlin. 2002. Spatial Processes and Political
Methodology: Introduction to the Special Issue. Political Analysis 10 (3):211-216.

Wolfinger, Raymond. 1965. The Development and Persistence of Ethnic Voting.
American Political Science Review 59:896-908.

Wright Austin, Sharon D., and Richard T. Middleton IV. 2004. The Limitations of the
Deracialization Concept in the 2001 Los Angeles Mayoral Election. Political
Research Quarterly 57 (2):283-293.

Zuckerman, Alan S., ed. 2005. The Social Logic of Politics: Personal Networks as
Contexts for Political Behavior. Philadelphia: Temple University Press.

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

The new dynamics of bureaucratic autonomy: courts, litigation, and agencies in the modern American state

PDF

Reducing gang involvement through employment: a pilot intervention

PDF

Can election reforms improve democracy?

PDF

Greening law: a socio-legal analysis of environmental human rights in India

PDF

(In)visible Asian American women: an analysis of how stereotypes in political cartoons reveal symbolic annihilation

PDF

Politics is something we do together: identity and institutions in U.S. elections

PDF

Are you as old as you feel? A quantitative and qualitative analysis of leader age and foreign policy decision-making

PDF

Concrete utopia: the development of roads and freeways in Los Angeles, 1910-1950

PDF

A spatial analysis of human trafficking in Greater Los Angeles

PDF

The geographic connotations of reincarceration: a spatial analysis of recidivism in Washington State

PDF

Creating cities and citizens: municipal boundaries, place entrepreneurs, and the production of race in Los Angeles county, 1926-1978

PDF

Toward counteralgorithms: the contestation of interpretability in machine learning

PDF

Understanding bilingual Latino parents’ experiences of their children’s autism services in Los Angeles: a critical ethnography

PDF

Participation in higher education diversity, equity, and inclusion work: a relational intersectionality of organizations analysis

Asset Metadata

Creator McDaniel, Jason Alan (author)

Core Title Location, location, location: a spatial econometric analysis of place-context effects in Los Angeles mayoral elections

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Political Science

Publication Date 11/14/2007

Defense Date 08/29/2007

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag contextual effects,Los Angeles,OAI-PMH Harvest,place,voting behavior

Place Name California (states), Los Angeles (city or populated place), USA (countries)

Language English

Advisor Wong, Janelle S. (committee chair), Barnes, John E. (committee member), Ethington, Philip J. (committee member)

Creator Email jasonamcdaniel@mac.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-m923

Unique identifier UC1217653

Identifier etd-McDaniel-20071114 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-591134 (legacy record id),usctheses-m923 (legacy record id)

Legacy Identifier etd-McDaniel-20071114.pdf

Dmrecord 591134

Document Type Dissertation

Rights McDaniel, Jason Alan

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Repository Name Libraries, University of Southern California

Repository Location Los Angeles, California

Repository Email cisadmin@lib.usc.edu