Documentos de Académico
Documentos de Profesional
Documentos de Cultura
1 Eitan
Hersh is Assistant Professor, Department of Political Science, Yale University. Address: Institute for
Social and Policy Studies, Yale University, 77 Prospect Street, P.O. Box 208209, New Haven, CT 06520-8209;
Email: eitan.hersh@yale.edu; Phone: 203-436-9061. Clayton Nall is Assistant Professor, Department of Political
Science, Stanford University. Address: 616 Serra Street Rm 100, Stanford CA 94305; Email: nall@stanford.edu;
Phone: (650)725-4076.
2 We thank David Broockman, Anthony Fowler, Andrew Gelman, Justin Grimmer, Lauren Davenport, Jacob
Hacker, David Laitin, and participants in the Stanford Political Science Methods Workshop for comments. Thanks
to Yales Institution for Social and Policy Studies and the Center for the Study of American Politics for financial
support and to Jonathan Rodden for sharing a combined version of the Harvard Election Data Archive precinct
data.
Abstract
Why does the relationship between income and partisanship vary across U.S. regions? Some answers
have focused on economic context (in richer environments, economics is less salient), while others have
focused on racial context (in diverse environments, rich voters oppose support for poor minorities). Using
73 million geocoded registration records and 185,000 geocoded precinct returns, we examine incomebased voting across a diverse set of local areas. We show the political geography of income-based voting
is inextricably tied to racial context, not economic context. Within homogeneously non-Black areas,
an areas wealth has no bearing on the individual income-party relationship. The correlation between
income and partisanship is strong in heavily black areas of the Old South and weaker nearly everywhere
else, including urbanized areas of the South. Differences between racially diverse local areas of the
South and comparable areas elsewhere account for state-level differences in income-based voting.
Introduction
Why are people in some places more likely to vote with their alleged income-based class interests than
people in other places? What is it about certain areas of the United States that make the incomepartisanship relationship stronger or weaker? While the political consequences of citizens alleged false
consciousness has been a topic of intrigue at least since Marx, recent work in behavioral social science
has sought to explain deviations from class-based voting (typically operationalized in terms of voter
income) across U.S. states and across countries. Within American politics, these explanations fall primarily under two schools of thought. The first, exemplified by the research of Gelman et al. (2008), has
focused attention on how the wealth or modernization of a state corresponds to the income-partisanship
relationship. The key finding is that within wealthier state jurisdictions, the relationship between income
and partisanship is weaker than it is within poorer state jurisdictions. The second school has focused
attention on race. For example, Alesina and Glaeser (2004) suggest that much of the regional difference
in support for social welfare policies and income redistribution (and support for the political party advocating redistribution) may be related to the level of racial heterogeneity in the region. The key finding
is that jurisdictions that are more racially diverse exhibit stronger partisan differences between rich and
poor.
These two schools of thought offer useful paradigms for understanding cross-regional patterns in
income-based voting. The choice of paradigm yields different hypotheses about the micro-level factors
that affect the voting patterns of rich and poor citizens. Under the paradigm focused on economic context,
we might look to how post-materialistic preferences, such as religion, affect voters in places that are
predominantly affluent versus areas that are predominantly poor. Under the racial context paradigm, we
might look to explanations of racial threat, sorting, and historical migration as factors that explain why
rich and poor vote the same in some places and differently in others.
In this paper, we sort out these paradigms by showing that racial context trumps economic context
in structuring the geographic patterns of income and partisanship in the United States. Whatever microlevel processes are at the root of jurisdictional differences in the income-partisan relationship, those
1
processes are closely tied to racial context, and they are only tied to economic context to the extent that
economics and race are collinear. By narrowing the set of viable explanations for regional variation
in income-based voting to those bearing on racial context, we move the literature a step forward in
understanding geographic variation in the political fissures between the wealthy and the poor.
Building on research on racial heterogeneity and redistribution, our efforts here expand on the racialcontextual paradigm, focusing on geographic variation in racial and political orders, and capitalizing
on new data sources that permit study of voter behavior at much lower geographic levels than could
previously be achieved using either survey data or aggregate-level election returns. Just as opposition to
redistribution has been predicted by white reaction to the local prevalence of racial minorities, we expect
that the strength of the relationship between income and support for the party that favors economic
redistribution will vary with local racial composition. We test this expectation using full-population
datasets of all 73 million voters who register with one of the two parties in party registration states, as
well as a geographic file containing data on the 185,000 precincts from all 49 states in which election
returns are organized by precinct and published. Once we account for the racial composition of small
geographic areas within states, the relationship between the wealth of a jurisdiction and income-based
voting vanishes.
We show that variation in the income-partisan relationship is tied, foremost, to the racial composition
of local areas. In the homogeneously white areas where most Americans live, the relationship between
income and partisanship does not vary with the wealth of the community nor the wealth of the state. Only
in local areas with substantial minority populations do regional differences emerge. In the Northeast and
Midwest, more affluent voters living in districts with larger proportions of African-Americans are only
slightly more Republican than low-income voters in these areas. However, in rural areas with high concentrations of minorities, particularly in the Black Belt areas of the South and some agricultural areas
of the West such as Californias Central Valley, the relationship between income and partisanship is
much stronger than would be expected otherwise. This is not only because poor blackswho are overwhelmingly Democraticare a larger share of the population and contribute to a stronger income-party
relationship, but because affluent non-blacks living in these areas tend to register and vote for Repub2
licans at much higher rates than those living among black voters elsewhere. We show that alternative
explanations for this pattern, including religiosity and other indicators of social conservatism and liberalism, provide only minimal additional predictive value. Regional differences between the preferences
of whites living in close proximity to racial minorities largely accounts for geographic variation in the
income-party relationship.
Toblers Law, the Modifiable Areal Unit Problem, and the Geography of the Left
Our impetus for exploiting variation in income-based voting at low geographic levels, and our hypotheses
resultant from that decision, are rooted in three principles that inform us about the appropriate estimation
of geographic aggregate statistics, contextual effects, and their influence on redistribution preferences.
The first principle, known as Toblers Law, states, Everything is related to everything else, but
nearer things are more closely related than distant things (Tobler, 1970). Toblers Law helps us to
understanding where the income-party relationship ought to be strongest: if racial or economic context
corresponds with variations in income-base voting, we expect that variation to occur at the local level.
Toblers Law is consistent with much of the behavior contextual effects literature, which has emphasized
the importance of voter behavior within areas that are typically much smaller than states (Gimpel and
Schuknecht, 2004). While much of that literature relies on arbitrary aggregate-level data (notably, metro
areas and counties), it constitutes nearly a century of scholarship that has consistently demonstrated that
voting behavior is geographically contingent and local, even if it has not established that differences in
voter behavior are due to exogenous contextual effects.1
As early as Tingstens study of Swedish voting behavior (Tingsten, 1937) and Keys scholarship on
county-level variation in Democratic primary voting across the South (Key, 1949), research founded
on local-level election results and surveys has established that models that account for neighborhood,
district, or county-level social context more accurately predict political behavior than analyses using
1
For recent scholarship on the specific geographies that voters use as their contexts, see Wong et al. (2012).
individual-level voter predictors alone (e.g., Lazarsfeld 1948; Katz and Lazarsfeld 1955; Huckfeldt and
Sprague 1995; Sampson, Raudenbush and Earls 1997; Oliver 1999; Gay 2006; Putnam 2007). Recent
research based on geographically coded voter files suggests that Keys (1949) racial threat hypothesis
appears in low-level neighborhood contexts as well (Enos, 2010). British political geographers of the
1960s and 1970s found that class-based voting varied according to the geographic distribution of voters:
working class voters living within more working-class communities tended to exhibit stronger support
for the Labour Party than working class voters living elsewhere (Butler and Stokes, 1969, 146). Research
on racial contextual effects commonly identifies effects at very local levels (Gay, 2002, 2006). Similarly,
major changes that may predict differences in income effects, such as income segregation (Reardon and
Bischoff, 2011; Saez, 2008) and partisan gaps between core and periphery, have been observed within
metropolitan areas, which are still substantially smaller geographic regions (in most cases) than the states
widely used when relying on survey data (Chen and Rodden, 2011). All together, this work suggests that
when studying the relationship between geographic context and the income-party relationship, we should
investigate contextual correlates at levels well below state lines.
The second principle is a seemingly mundane regularity in the study of geographic statistical methodology known as the Modifiable Areal Unit Problem, or MAUP (Fotheringham and Wong, 1991). The
MAUP tells us that parameters estimated on aggregate data are vulnerable to how the data are grouped
for aggregation. Of special relevance to our claims here, the MAUP tells us that geographic correlations
based on state-level data may obscure substantial within-state variation in these correlations. The MAUP
means that as we dig beneath the state-level, we cannot take for granted that prior results identified at the
state- or country-level will map on to local areas.
Models that assume the independent and identical distribution of voter preferences within states (even
conditional on demographics) may obscure key aspects of income-party relationship, particularly if this
relationship is geographically contingent (Cho and Rudolph, 2008).2 The caricature of the stereotypical
rich, secular New York Democrat and her rich, religious Alabama Republican counterpart may only hold
2
We use the term income effects and income-based voting as a shorthand for the bivariate relationship between
personal income and partisan outcomes, not to suggest that income has a causal effect on party and vote choice, or even that
income is the primary consideration in voters minds.
within some subregions of each of those states. For example, unlike the urban areas of the South, the
Black Belta set of counties with a history of slave-based cotton plantations and, later, sharecropping
under Jim Crow lawshas a high concentration of black poverty and some of the highest rates of overall
Democratic voting. Similarly, the persistence of Democratic voting among key white ethnic groups in
the Northeast (Gimpel and Cho, 2004) has helped maintain the partys dominance in those areas. The
same can be said of areas dominated by ethnic and religious groups in which the Republican Party holds
sway, such as the Mormon-dominated areas of Utah, parts of Idaho, and northern Nevada (Gimpel and
Chinni, 2011). In short, there are persistent political caricatures of voters in different regions of the
U.S., but those caricatures have within-state, not just across-state, variants.3 Our interest in local context,
then, stems not just from the idea that local forces may be more potent in influencing political behavior
than coarser geographic phenomena, but that comparisons across large boundaries like states obscures
important within-state differences.
The first two principles, Toblers Law and the MAUP, justify our investigation of contextual effects
at local levels. When we examine the income-party relationship within local geographic areas, then,
what do we expect to find? Prior research, both in studies of countries and of U.S. states, falls within
two primary schools of thought about the contextual correlates of income-based voting. One school is
centered around the effects of economic context. For example, Gelman et al. (2007), building on work by
Huber and Stanig (2007) and Inglehart (1990), suggest that economic issues might well be more salient
in poorer states (p. 74). The theory offered is that in areas that are more modernized, or that have higher
average living standards, or that have higher incomes, economics becomes a less salient part of a voters
political calculus. Post-materialist concerns are not just a function of ones own economic condition
but, under this view, they arise depending on the local economic conditions one sees on a day-to-day
basis. Beyond this line of reasoning, there are many other theories focused on the role of economic
context in driving support for redistributive policies and/or support for parties that advocate or oppose
redistribution. For example, research on income inequality (Melzer and Richard, 1981; Romer, 1975),
3
Wong et al. (2012) address the implications of choice of context for individual-level effects, assessing how individuals define the geographic scope of their neighborhoods. Their findings are an extension of MAUP reasoning to individual
psychology, while our results are more focused on the basic statistical problem.
natural resource abundance (Auty and Gelb, 2000), and class bias in the electorate (Hill and Leighley,
1992) test variation in political support by measures of economic context. Whether economic context
is measured by GDP, average income, inequality, or another measure, these theories share in common a
focus on the economic conditions of a jurisdiction affecting individual-level attitudes or voting behavior.
The main alternative school of thought, especially pertinent to American politics, is that preferences
for and attitudes towards economic redistribution are influenced by response to out-group threat or outgroup dislike (e.g. Alesina and Glaeser 2004; Luttmer 2001; Gilens 1999). Hypotheses under this school
of thought include that people are predisposed to feel sympathetic to those who look like them but not
to those who look different, and that politicians capitalize on racial differences by pitting groups against
one another for electoral gain. Whatever the micro-level foundations, Alesina and Glaeser (2004) have
shown that racial heterogeneity in a jurisdiction explains much more of the variation in economic preferences than do economic factors like inequality. As our third motivating principle, we apply Alesina and
Glaesers result about the preeminent role of racial context, previously found to correlate with redistributive preferences, to geographic variation in the income-party relationship. We argue that racial context,
not economic context, ought to explain why in some places rich and poor vote alike and in other places
they do not. Local areas are typically racially homogenous. In the absence of racial diversity in these
local areas, we do not expect to see wide gaps between rich and poor. If nearly all voters in a local area
are white, for example, we expect to see small differences in average support for the Republican Party
across income levels.
We expect that most variation in income-based voting occurs in racially diverse localities. In such
areas, the poor end of the income spectrum is often overwhelmingly black, and black voters of all income levels everywhere are overwhelmingly Democratic (Dawson, 1995). In diverse local areas where
wealthier voters living near blacks (and, increasingly, other racial minorities) are politically liberal, the
relationship between income and partisanship will be weak. But in areas where wealthy voters living
near blacks are politically conservative, the relationship between income and partisanship will be quite
strong. The former condition is likely to be especially common urban areas with cosmopolitan voters (as
Gelman et al argues), whereas the latter condition rarely exists in central cities but may be more likely
6
in suburban and rural areas of the U.S. South. To the extent that the relationship between income and
partisanship varies across local jurisdictions in the United States, then, we expect that variation to be
contained in racially diverse areas.
Our contribution, then, is to test three empirical hypotheses about the role of local context in the
strength of income-based voting: First, after accounting for racial context, we do not expect incomebase voting to vary with area-level income. Second, in homogenous white areas, we expect rich and poor
voters to have minimal differences in their political affiliations and vote choices. Third, in heterogenous
places, we expect to see regional variation in the income effect. In some parts of the country, poor
minorities live in close proximity to rich, white, conservatives. In other parts, they live near rich, white,
liberals. We expect strong income-base voting patterns only in areas that are both racially diverse and
where the rich people are historically racially conservative, namely the South. Only when both these
conditions are met (diverse districts with racially conservative Whites) do we expect sizeable partisan
gaps between rich and poor.
We test these hypotheses using data sources that have been unavailable to previous researchers. Past
work on the political geography of income-based voting has relied primarily on survey measures of
self-reported income, party identification, and vote choice, or on geographically coarse aggregate-level
election data. Because nationally representative surveys such as the American National Election Study
and even the National Annenberg Election Study have small sampleseach NES release is sample of
only about 2,000 respondentsthe studies commonly sidestep questions of sub-state geographic variation
(Gelman and Little, 1997; Park, Gelman and Bafumi, 2004; Lax and Phillips, 2009; Vigdor, 2006).
In some cases, scholars supplement these data with analysis of aggregate correlations in counties and
other similarly sized units below the state level (Gelman et al., 2008).4 This paper expands on these
4
Among recent contributions to this literature, recent work has bridged across multiple surveys Tausanovitch and Warshaw
(2011) to provide similar estimates within congressional districts. Other work has provided estimates of the voting behavior
of racial and economic subgroups within states (Ghitza and Gelman, 2012), but like many existing models, these have been
limited to estimation of statewide effects within non-geographic subgroups.
research findings using 73 million party-registration records and presidential election results from more
than 185,000 precincts. These data, which even a decade ago were not available due to limitations in
technology and data quality, now permit us to explore the relationship between geographic context and
income-based voting with extremely high resolution.
The first of our data sources is a voter file containing 73 million voter records from states that require
party registration. We apply a combination of nonparametric comparisons and multi-level modeling
to voter registration data from Catalist, a leading Democratic data vendor that compiles, cleans, and
keeps current registration records from every state.5 Our analysis using these data is restricted to the
29 states that invite registrants to register with a party (or as independent or unaffiliated voters). The
second data set, compiled through the Harvard Election Data Archive is a record of the 2008 presidential
election returns from 185,000 precincts, which allows us to estimate the relationship between precinctlevel income and the 2008 Republican presidential vote in the 49 states that have published and made
available their data (Ansolabehere and Rodden, 2011).
The Catalist voter database permits us to make individual-level inferences about partisan identification by linking party registration records to block-group level demographic characteristics. The version
of the Catalist database made available to academic researchers links individual registrants to the Census block-group of their residential addresses. The median household income of each registrants block
group in the 2000 Census is used as a proxy for their income class. These data are represented in the
Catalist data as categorical variables that place each voter into a block-group income category at $20,000
intervals, ranging from $20,000 or less to $200,000 income (the top-coded value reported in the Census
data). Each voter i also associated with the racial composition of his or her block group.6
To study the link between local racial and economic context and voter partisanship, we analyze the
Catalist data at the state house district level. We have several reasons to study these relationships within
state house districts. Foremost, these are the bodies that elect representatives to the lower chamber of
5
These data are explained in more detail in Ansolabehere and Hersh (2012).
Individual-level race data is only available in some Southern states. While we would prefer to have individual-level
demographic information, given the level of racial segregation in the United States, using block-group level race data is
unlikely to induce sufficient ecological bias to reverse our findings. We discuss the merits of using block-group income as a
measure of economic class in the Online Appendix.
6
the state legislatures, making them the lowest-level geographic areas that elect politicians responsible
for major policymaking. These legislators set state welfare policy and are responsible for state-to-local
transfers that result in both income redistribution and geographic redistribution. So, the relationship
between district income and partisanship within these districts has important implications for each state.
State house districts are also useful to study because they are approximately comparable across states
and there are nearly 5,000 of them across the country. Because state house districts are numerous and
typically equal in population, they are especially useful in the study of metropolitan areas, which are
generally contained within just a few populous counties but are almost always divided into many more
state legislative districts.7
To measure district income, we take the unweighted mean of the median household block group
income of every registered voter in each district. We similarly calculate the black population percentage
in the district by averaging the black population percentage in registrants block groups in each district.8
Thus, we study the relationship between a registered voters individual-level partisanship, the income
of their block group, and the income and race of their district. For registrants in either of the major
parties, we have measures of party, income, racial and economic context, and geography for over 73
million voters. Note that at the end of the manuscript and in a detailed online appendix, we offer a
detailed justification for the use of party registration as a measure of partisanship, block group income
as a measure of income, and state house districts as a measure of local area.
Though recent scholarship has justified use of party registration states by arguing that they are representative of the nation at large (see Abrams and Fiorina 2012; McGhee and Krimm 2009), we supplement
our analysis by using additional data from the Harvard Election Data Archive (hereafter, HEDA) (Ansolabehere and Rodden, 2011), a GIS database containing 2008 presidential election results from 49
states.9 We merge the precinct-level data with block-group level income and race data from the 2000
7
Unlike states and counties, the boundaries of state house districts are drawn strategically, subject to legal requirements
about population size, racial composition, contiguity, compactness, and other criteria imposed by courts. That these districts
are drawn purposefully is not a weakness in our research design, as many so-called gerrymanders act to combine communities of interest (Forest, 2004).
8
This will provide a registered-voter-weighted average of the the local population figures, which will differ slightly from
the black proportion of the district population.
9
The omitted state, Oregon, conducts its elections by mail and does not release precinct-level voting returns.
Census using a spatial join in ArcGIS.1011 After these Census data were aggregated by precinct, each
precinct was linked the Census shapefile of 2006 state house districts and the ESRI county shapefile by
way of a spatial join (of the Census, 2006). This was achieved by first defining the precincts as points,
then assigning them to the county or district in which they fell. The resulting precinct-level data set
contains 185,002 precincts distributed across districts and counties in all 49 states. Summary statistics
for these data appear in Appendix Table A-3. Table 1 summarizes the two data sources.
Table 1: Summary of Data Sources
Individual
Units
Aggregate
Units
Coverage
Catalist Data
73,170,970 D. and R. Registrants
with block group (BG) income measure
State House Districts
with aggregated BG inc. and race
29 Party Registration States
HEDA Data
185,002 precincts
with BG income measure
Counties and State House Districts
with aggregated BG inc. and race
49 States with Published Data
To test the relationship between racial context and income-based voting, we present a series of results
that vary in the strength of their assumptions and progressively bore below state boundaries. We adopt
two approaches in our analysis of the Catalist party registration data. First, we present nonparametric
summaries of Republican registration within different income groups, taking advantage of the abundant
Catalist data. Then, adopting stronger modeling assumptions, we estimate a series of hierarchical linear
models in which both the group level intercept and income coefficients are allowed to vary (Gelman
and Hill, 2007). We estimate these district-specific effects of income on Republican registration among
two-party registrants within state house districts in the 29 party-registration states, and for precincts
within districts (and in the appendix, within counties) for the 49 precinct-data states. We present the
model results using cartograms, maps in which the area of each geographic unit is proportional to its
population.
Finally, we present a set of regressions in which district-level random-effect estimates of the districtlevel deviation in the rich-poor partisan gap are defined as the outcome variable, and various social and
10
Race and population data were obtained from the ESRI block group layer (ESRI, 2008). Aggregate and median household
income for the block group were obtained from the National Historical GIS (Fitch and Ruggles, 2003). Block groups were
converted to points, and data from all points within each precinct were averaged to generate the precinct-level data.
11
Block groups from states that do not appear in the HEDA data were excluded from the spatial matching procedure.
10
cultural variables are used as the explanatory. The purpose of this analysis is to test the marginal explanatory value of competing social explanations commonly offered for income-based voting: religiosity, rural
culture, education, and family structure.
To build intuition about the strength of the income-party relationship within districts, our empirical
investigation begins with scatter plots based on aggregate-level regressions of district-level partisanship
against district-level income. Using the full dataset of registered Democrats and Republicans in 29 states,
we calculate, in Figure 1, the percent Republican of each state house district and the average income level
of each district. We organize the party registration states into four approximately even groups, based on
the states income. We calculate state income by aggregating the block group median income value for
each voter in the jurisdiction. Similar to state income, the district income is calculated by averaging
each registered voters median block group income within a district. We plot the proportion of two-party
Republican registration against the logged mean district income, and then fit a lowess curve to each
states set of state house districts.
We display Figure 1 because it draws immediate attention to four facets of local political geography
that might be obscured by state-level analysis. First, the figure indicates the large variance in districtlevel income. State house districts in the U.S. differ in average income levels by as much as $130,000.
By comparison, the richest and poorest states differ in average income by a much narrower $35,000.
Second, in all but two of the 29 states plotted (Wyoming and Oregon), on a state-wise basis richer districts
are more Republican districts. Unlike states, rich localities are more Republican, and this holds in rich
liberal states like Connecticut as much as it holds in poor conservative states like Louisiana. Third, notice
that while there is state-by-state variation in the nature of the slope lines that are plotted, that variation
appears to be completely unrelated to state-level income. Each quadrant of Figure 1 contains states
with very steep curves, gradual curves, and exponential-like curves. Fourth, and perhaps most important
for our argument, while the connection between state-level income and the bivariate relationship of
district income and district partisanship appears to be tenuous in Figure 1, observers familiar with the
11
Middle-Poor States
WY WYWY
WY
WY
WY
WY
WY
WY
WY
WY
WY
WY
WY
WY
WY WY
WYWY
WY
WY
WY
WY
KY WY WY WY
ME
WY
WY LA
WY
WY
WY
KY
OK
KY
WY
WY
WY
KY WY
OK WY OK OK
WYSD
OK
WV
NM
SD
WY
WYSD
NM
SD
OK
NM SD
WY
SD
NM
WY
WY
OK
NM
KY KY KY SD
OK
OK
KY
OK
OK
LA
OK OKOK LA
WY
OK
OK
SD
LA
SD
NM
WY
NM
ME
WY
WY
SD
OK
NM
OK
WV
ME
OKWY
OK
OK
ME
NM
WY
LA
SD WY
OK
OK
LA
NM
WY
KY NM
ME
ME
ME
WY
LA
ME
KY
SD
WV
SD
OK
ME
OKLA
ME
SD
ME
ME
OK
OK
OK
LA
WY
WV
KY
KY
ME
KY KYKY
SD
WV
OK
ME
ME
LA
LA
LA
OK
LA
ME
ME
LA
SD
WV
LA
SD
OK
ME
ME
NMNM
NM
LA
SD
ME
ME
KY
SD
OK
OK
ME
KY
LA
ME
ME
LA
ME
SD
ME
ME
OK
OK
ME
ME
ME
ME
SD
ME
ME
MEKY
ME
WV
SD
ME
WV
LA
OK
OK
SD
ME
LA
SD
LA
ME
ME
WV
ME
OK
NM
ME
NM
KY
SD
OK
ME
ME
ME
NM
ME
LA
LA
SD
ME
ME
ME
ME
WY
SD
LA
ME
OK
OK
ME
NM
ME
WV
WY
ME
MEME KY
WV
ME
ME
SD
ME
LA
KY
KY
ME
ME
ME
WY
KY
KYNM
LA
LA
LA
OK
OK
ME
ME
WV
WV
ME
NM
OK
NM
ME
ME
ME
SD
ME
ME
WY
LA
OK
ME
SD
ME
KY
LA
ME
LA
OK
ME
NM
KY
LA
OK
WV
ME
OK
ME
NM
SD
ME
WV
ME
KY
ME
NM
LA
OK
ME
WV
ME
ME
OK
ME
ME
ME
WY
WY
NM
ME
WY
LA
LA
KY
NM
KY NM
ME
ME
SDNM
ME
WV
LA
WV
OK
WV
ME
ME
WV
WV
KY
LA
NM
ME
WV
LA
OK
ME
SD
OK
NM
KY
ME
OK
OK
ME
OK
ME
OK
ME
NM
KY WV
ME
LA
OK
ME
NM
NM
LA
OK
KYME KY
ME
ME
KY
ME
OK
NM
KY
OK
WV
KY
KY
ME
NM
WV
ME
KY
KY
ME
WV
KY
KY
OK
ME
WV
ME
LA
OK
KY
KY
OK
NM
ME
ME
KY
OK
OK
NMLALA
NM
WV
ME
SD
SD
ME
KY
LA
WV
NM
KY
ME
LALA
OK
LA
KY
LA
OK
ME
OK
KYKYLA
ME
OK
WV
WV
KY
KY
NM
KY
LA
LA
OK
ME
NM
OK
OK
WV
KY
KY
ME
ME
NM
OK
ME
ME
WV
LA
NM
ME
LA
NM
KY
LA
WV
KY
LA
OK
OK
WV
LA
WY
ME
ME
KY
NM
WV
WV
KY
LA
OK
ME
KY
LA
OK
NM
NM
KY
KY
LA
ME
KY
KY
ME
KY
WV
WV
OK NM
KY
ME
OK
WV
OK
LA
ME
OK
WV
KY
LA
KY
KY
LA
ME
WV
ME
OK
KY
LA
NM
KY
LA
LA
OK
NM
WV
OK
KY
WV
LA
OK
SD
ME
LA
OK
WV
NM
OK
NM
KY
KY
OK
KY
LA
LA
NM
LA
NM
WV
KY
WV
KY
KY
ME
KY
ME
WV
KY
KY
ME
SD
OK
KY
LA
OK
NM
LA
OK
KY
ME
ME
KY
KY
LA
ME ME
KY
LASD
OK
NM
NM
KY
WV
LA
NM
NM
NM
NM OK
LA
ME
WV
LA
NM
KY
LA
KY
KY
KY
LA
NM
WV
LA
NM
OK
OKLA NM
LAWV
LA
OK
OK
WV
KY
LA
ME
LALA
LA
WV
NM
LA
NM
ME
LALA
OK
LA
OK KY
ME
NM
KY
NM
LA
LA
NMNM
LALA
WVLA
LALA
WV
KY WV OK
WV
LA
LA LA
LA
LA
KY
FL
IA
IA
NENE
NC
IA
NE
IA PA
PA
PA
FL
NENENE
NE
FL FL
NE
NE
NE
NC
PA
IA
NE
PAPAPAPA
PA
NE
PA
NC
NE
NC
NE PA
IANE
FL
IA
NC
OR
PA
FL IAFL
NC
NC
PANC
OR
PANE
PA
NC
PA
PA
PA
PA
IAPA
NE
NE
NENC
NE NEPA
PA
PA
IA
PA
FL
OR
PA
NC
NE
OR
NC
FL
NE
PA
NE
PA
IA
IA
NE
NE
NC
OR
PA
PAIA
NC
PA
IA
NC
FL
NE
PA
PA
IA
IA
PA
PA
PA
PA
IA
NC
FL
PAPA
NC
FL
PA
IA
NC
OR
PA
NC
NE
OR
PA
IA
IA
IA
FL
FL
FL
IA
NC
FL
NE
OR
PA
FL
PA
PA
IA
OR
PA
PA
IA
OR
PA
PA
IA
NC
NC
FL
NE
NE
PAIA
PA
IA
NE
PA
PA
PA
PA
NC
NC
FL
FL
FL
OR
NC
FL
FL
FL
OR
NC
FL
FL
OR
PA
PA
PA
FL
FL
FL
FL
PA
PA
PA
OR
IA
IA
NC
NC
NC
NC
FL
FL
OR
OR
PA
PA
PA
IA
NC
NC
FL
FL
FL
OR
PA
IA
NC
FL
FL
FL
NE
PA
PA
PA
IA
NC
FL
OR
PA
NC
NC
FL
NE
OR
IA
IA
NC
FL
NE
PA
PA
IA
IA
NC
FL
FL
IA
NC
NC
FL
FL
FL
PA
IA
IA
PA
PA
NC
FL
OR
NC
OR
IA
FL
FL
FL
PA
PA
IA
IA
FL
IA
NC
NC
FL
FL
PA
PA
PA
IA
PA
PA
PA
IA
IA
IA
FL
FL
OR
NE
PA
IA
NC
FL
FL
FL
FL
FL
IA
IA
NC
FL
PA
IA
IA
IA
FL
FL
OR
PA
PA
PA
IA
OR
IA
NC
NCPA
FL
NE
PA
IA
FL
FL
PA
IA
NC
FL
FL
PA
PAPA
IA
FL
OR
PA
PA
IA
NC
FL
OR
OR
FLPA
NE
PA
PA
IA
FL
FL
NE
IA
IA
NC
FL
NC
NC
OR
PA
NC
NC
NC
OR
OR
PA
PA RI FL
PA
FL
OR
PA
IA
IA
IA
FL
PA
FL
PA
FL
PA
PA
NC
RI
IA
IA
NC
FL
OR
PA
PA
IA
IA
NC
FL
NE
IA
IA
OR
PA
OR
PA
IA
OR IA
NC
PAPA
PA
IANCPA
PA
NE
OROR
PA FL
FL
OR
PA
PA
PA
OR
IA
PA
IA
NC
NC
OR
OR
PA
NC
OR
PA
IAOR
RI
OR
NC
NC
RIOR
IAOR
OR
PARI
PA
PA
PA
OR
NC
RI OR
NC
PA
PA
NC
NC
PA
PA
NC
PA
PA
PA
RIRI
IA
PA
NC
PA
IA
NE
IA
NE
PA
RIRI
FL
NE
NC
NC
OR
IA
IA
PA
NC
NC
IA
NC
PA
PA
RI
NC
NC
FL
OR
IA
IA
NC
FL
NENC
PA
IA
FL
OR
RI
IA
NC
NC
NE
OR
RI
PA
PA
NC
NC
FLRI
PA
RI
FL
PA
PA
RI
NC
PA
FL
IA
NC
RI
IA
NC
FL
FL
FL
PA
PA
IA
PA
RI
RI
PA
PA
IA
FL
FL
PA
FL
RI
PA
IA
FL
NC
NC
FL
PA
NC
IA
RI
PA
PA
IA
PA
FL
PA
PA
IA
NC
RI PA
NCPA
PA
RI
OR
NC
NC
PA
RI
RI
PA
RI
RINCRI
PA
PA
PA
NC
NC
NC
FL
IA
NC
PA
ORRI
NC
PA
PA
PA
RI
FL
PA
RIOR
NC
FLRI
FL
RI
RI
PA
NC
PA
FL
PA
RI
NC
NC
RI
IA
RI
PA
RI
IA
RI
NCNC
NC
FL
OR
FL
PA
PA
RIRI
FL
RI
NC
NC
NC
FL
PARI
NC
PAOR
PA
RI
RI
RIOR
NC
NC
RI
RI
NC
PA
FL
FL
PA
PANC
RI
FLPA RI
NC
FLNC
FL
PA NE
RI
RI
FL
RI
RI RI
RI
RIRI
FL
PAPARI
OR
RI
PA
RI
RI
OR
PA
RI
RI
RI
PA
PA
RI
PARI
PA
RI
RI PA
PA
PA
PA RI
PA
RIRIR IPA
PA
Poorest States
KY
Ln($12)
Ln($20)
Ln($33)
Ln($55)
Ln($90)
Log of Mean District Income in $10K
Ln($150)
Ln($12)
Ln($20)
Ln($12)
1
Pr. Republican of Two-Party Reg.
.2
.4
.6
.8
UT
UTUT
UT UT UT
UT
UT
UT
UT
UT UT
UT
UT
UT
UT
UT
UT
UT UT UT
UT UTUT
UT
UT UT
DE
UTUT
UT
UT
UT
UT
UT
UT
KS
UT
UT
UT
KS
UT UT
UT
KS
UTUTUT
KS KS
KS
KS
KS
KS
KS KS
KS
UT
UT
KUT
S UTUTUT
KS
KSKS
UT
UT KSKS
KS
KS
KS
UT
KS
UT
KS
KS
KS
KS
KS
KS
UT
KSKSKS
UT
KS UT
KS
KS
UT
KS
KS
UT
KS KS KS
KS
KS UT
KS KS
KS
NV
UT
KS
UT
KS
KS
NH
NH
KS
KS
KS
KS
KS
KS
NH
NH
KS
NY
AZ
UT
KS
AZ
AZ
KS
KS
NY
UT
NH
KS
KS
KS NV
AZ
KS KS
KS
NYAZ
NHNHKSNV
KS
KS
KS
KS
KSUT
NYNY
KS
KS
NY
AZKS
KS
NYKS
NH
AZ
KS
NY
NH
NH
KS
NV
NY
NH
KSUT
NH
NH
NH
AZ
KS
KS
KS
NV
NY
KS
NY
KS
KS
NY
NH
NY
NH
NH
AZNH
NH
NH
NH
NHNY
KSNY
NV
AZ
AZ
NH
AZ
KS
NV
NH
NHKS
NV
NV
NY
NH
NH
NY
NY
NH
NY
NH
NH
NY
NV
NY
NH
KS
KS
UT
NH
NH
AZNY
NY
NH
NH
NH
AZ
NY
NH
NH
NHNY
NH
NV
NY
NH
NY
KS
NY
NH
UT
NY
NH
KS
NH
DE
KS
NH
NY
KSNY
NH
KS
NY
AZ
KS
NH UT
KS
NY
NY
NH
NV
NH
NY
UTNH
NH
KS
NH
NH
AZ
KS
NY
NH
DE
NY
NYNY
NY
KSNHDEKS
KS
NHNH
NV
NY
NY
KS
DE
NH
NY
NY
NH
KS
NH
NY
UT
NH
NH
NH
KSUT
AZ
DE
DE
NVNH
DE
AZ
DE
NY
UT
NH
DE
NH
NH
DE NY
NY
NH
DE
NY
NY
KS
DE
DE
AZ
NH
AZ
DE
KS
KS
NVNH
KS
AZ
NY
NH
NH
NH
KS
NVNH
NY
NVDE
DE
NH
NH
NH
DE
DE
NY
NH
NY
NY
NY
NH
NH
NH
NY
DE
NH
NY NY NY
NH
NY
NV
NH
NH
NH
UT
NY
KS
NV
KS
NY
NV
NV
NH
NY
NH AZ
NY
NH
NV
NV
NH
DE
NYNV
NYNY
DE
NV
NYDE
NH
DE
NV
NY
NY NY
DEUT
DE
NY
NH
NV
AZ
DE
NY
NV
NY
DE
KS
NH
NY NY
KS AZ
NY
NY
NH
NH
NH
DE
NV
NV
NH
NV
NY
NV
KS NV DE
DE
NH
NY
NH
DE
AZ
AZ
AZ
KSKS
NY
NY
AZ
NV
NV
NY
DE
NY
DENY
KSNY
KS
NY
DE NY
NY
NY
NYAZ NY
NY
KS
NYNY NY
NY NH
KS
NYNY
KS
NV
DE
NY NV
NV
NV
NY
NY NY
NY
NY NY
NY
NYNY
NY
NY
NY NY
NY
NY
NY
NY
NY DE
KS
DE DE
DE NY
NY
KS
NY
NYNY
NY NYNY NY
NY
NY
NY
NY
NY NY
NYNYNY
NY
NY
NY
NYNY
NY
NY
NY
Ln($20)
Ln($33)
Ln($55)
Ln($90)
Log of Mean District Income in $10K
Ln($150)
Richest States
AZ
Middle-Rich States
Ln($33)
Ln($55)
Ln($90)
Log of Mean District Income in $10K
Ln($150)
Ln($12)
AK
CA
AK AK
AK
CO
AKAK
AK AK
AK
AK
AK
CO
CO
CO
AK
AK CO
AKCO
CO
AK
AK
MD CO
CO
CT
NJ
CT
AK
AK AK
CO
CO CO
CO
MD CA CA
CO
CO
CO
CO
CO
COAK
MD
AKMD
AKCA
CAAKCA
CO
MDCO
AK
CO
CA
CTCT
NJ
MA
CO
CT
MD
NJCO
CA
CA
CA
CA
NJ
AK
NJ
CA
AK
CA MD
CT
AK
CA
AK
CA CT
AK
CT
MD
NJ
CO
AK
CA
CACA CT
CT
CT
CO
MD
NJ
CA
MDCO
CT
NJ
CA NJ
MD
MDCT
CT
CO
CA
CA
CA
MD
CT
NJ
CA
CO
MD
CT
AK
CA
CT
CTMD
CT
CA
MD
CT CT
MD
MD
CA
CT
CT
MD
CT CT
CO
CT
CA
CA
CT
CT
AK
CO
CT
MD
MD
CT
CO
MD
MA
AK
CT NJ
CT NJ CT
MD
CA
CT
CT
AK
CO
CO
AK
CO
MD
MD
CT
COCO
NJ
NJ
CA
CT
CT
MD
CO COCO
MA
CT
CT
CT
CT
CO
AK
CO
CT
CT
MD
MA
CO
MA
MD
MA
CA
CT
CTMA
CT
MD
MDMD AKNJ
CT
CT
CT
CT
CTMA
CT CT
NJ
CT
NJ
CT
MA
CA
CT
CO
MA
CO
MA
NJ
CA
CA
CT
CA
CT
CT
CT
CO
MA
CT
CT
MA
CT
NJ
CO
MA
MA
CA
CT
MA
CA
CT
CT
MA
CT
CT
MA
CA
MA
MD
CA
CO
CO
MD
MA
AK
MA
MA
MA
CA
CTNJ CA
CT
CT
CT
MA
AK
CO
MD
MA
CT
CO
MA
MA
CA
CA
CT
MD
CA
CT
CT
CT
MA
CT
CA
CO
MA
MD
CO
MA
CA
CT
MA
NJ
AKMA
CA
CTCO
MA
MD
MD
CT
MA
NJ
CA
CA
CT
CT
CT
CT
MMA
D
NJCT
CT
CT
CO
MA
CA
CT
MA
MA
MAMA
MA
CA
CA
CO
MD
CA
CA
MA
CT
MA
CA
MA
CA
CT
MA
MA
CT
CA
CT
NJ
CT
MA
CT
MA
MA
MA
CA
CA
CT
MD
CA
CA
MA
MA
MA
MA
CTCA
MA
MA
MA
MA
CT
MA
CT
CT
CO
MD
MD
AK MD
CA
CA
CT
MA
MA
MD
NJ
CA
CT
CO
NJ
MD
CO MA
MD
CT
MA
MA
CT
MA
CT
MA
MA
NJ
CT
CO
CA
CA
MA
MA
CTMA
MD
MA
CT
MA
MA
NJ
CT
CT
MA
MA
MACT
MA
MA
MD
NJ
MA
CT
MA
NJ
NJ
CT
MA
MA
MD
MA
MA
MA
MA
CA CA
CT
CTCO
MA
MA
MA
MD
MD
MA
CT
MA
MA
MA
CT
CO
CT
MA
CO
NJMA
NJ
MA
MA
NJ
MA
MA
CA
MA
CA
MA
CACO
CT
MA
MA
CA
CT
MA
MA
MD
CT
MA
MA
CT
CTMA
CT
MA
MA
CT
MACO
CT
MA
MA
NJ
CA
MA
CT
MA
MA
MA
CA
MA
MA
NJ
MA
MA
MA
MD
MA
MA
MA
NJNJCAMD
MA CTCTMA
NJ
MA
MA
MA
CT
MA
MD CAMAMA MD
CA CA
CT
NJ
CT
MA
MA
MA
NJMD MD
MA
MD
CT
MA
MD
MA
MD
CT MA
MD
MA
MA
MA MAMD
MACTMA
CT
CTMAMD
CT CT
MD
CT
CT
MA
MA
CT
CT CTCT
CT
CT
Ln($20)
Ln($33)
Ln($55)
Ln($90)
Log of Mean District Income in $10K
Ln($150)
Note: Figures are derived from the full population of registered Democrats and Republicans in party registration states. N=
73,170,970. Average median household income is the unweighted average of the median household income of block groups
in the district; partisan registration is obtained from individual-level voter file records.
American political landscape will notice a connection between the geography of race and the districtlevel relationship. The states that have very Republican wealthy districts and very Democratic poor
districts are states that have pockets of high-poverty minority areas. These include rich states like New
Jersey and Connecticut as well as poor states like Louisiana and Oklahoma. States that have a relatively
flat relationship include both rich and poor states, including Wyoming, Kentucky, Iowa, Oregon, Utah
and New Hampshire. All of these states are less than 2% African-American, with the exception of
Kentucky, which in the 2010 Census had the lowest African-American population share (7.71%) of
any state south of the Mason-Dixon Line. Other states have not a linear relationship but a curvilinear
relationship. In South Dakota, Nebraska, Delaware, and California, pockets of poor racial-minority
12
districts appear to explain the relatively flat relationship between income and party in all but the poorest
parts of the state.
4.1
Having viewed the data in the aggregate, we now examine the rate of Republican registration among
voters falling within each block-group income category within state house districts and within states. We
stratify the state house districts based on their average income levels and on their racial compositions.
Our first empirical strategy is to exploit our massive data sets to adopt as few modeling assumptions as
possible and estimate the difference in partisan registration across income groups in different geographic
settings. The large number of observations in the voter file goes far in overwhelming the curse of
dimensionality problem by brute force, allowing us to plot conditional means, without any models or
semiparametric smoothing techniques such as locally weighted regression (Cleveland, 1979).
Figure 2 presents a nonparametric tests in the form of a set of graphs (lattice plots) that demonstrate the role of racial and economic context in the relationship between block-group median household
income (expressed in $20,000 income intervals starting at income levels of $20,000 and below and
capped at $100,000+) and the Republican proportion of two-party registrants within each income category.12 These proportions are calculated and plotted for registered voters in block groups that fall within
three state-house-district income categories: those with average block-group median income of less than
$40,000, $40,000 to $60,000, and more than $60,000. Each of these categories is represented by a different line style on the graph. Two sets of such lines are presented: red lines represent voters in the
wealthier 14 party-registration states in our sample, while purple lines represent voters in states in the 15
poorer party-registration states. Each of these estimates is further subdivided into graphs based on the
districts black population percentage: 0 to 5 percent, 5 to 10 percent, 10 to 25 percent, and 25 percent or
more black. Block-group median household income is on the horizontal axis. To cite one example from
the figure, the solid red line in the upper left plot shows the relationship between block-group income
12
These bins are not selected arbitrarily; rather they are how the data are grouped by Catalist. Because there are so few
block groups with median household income above $100,000, income groups above this level were collapsed into the top
category.Such plots summarize bivariate relationships within researcher-defined strata (Becker, Cleveland and Shyu, 1996).
13
and partisanship in districts that are poor, less than 5% black, and located in the rich states.13
If there was a tendency of voters in poor states to vote their class interests more than voters in rich
states, we would expect to see the purple lines (representing poor states) to have steeper slopes than the
set of red lines (representing rich states). Likewise, if there was a tendency for voters in poor districts to
vote with their class interests, we would see different slopes in the richest districts (shown with dotted
lines) and the poorest districts (shown with solid lines). As the graph demonstrates, these patterns do
not generally hold. The key exception is in districts with larger proportions of African-Americans.
In districts with more African Americans, we finally see a divergence between the richest half of the
states and the poorest half. While we are not attributing these findings to a causal effect of contextual
racial composition, the findings presented here show that economic explanations are, at a minimum,
descriptively explained by behavior among rich voters who live among black voters in rich states, versus
those who live among black voters in poor states.
State-by-state differences in the income-party relationship appear to rest almost entirely on differences in behavior in areas with sizeable black populations. Once state house districts with sizeable
African American populations are excluded from the analysis, the link between income and partisanship
does not vary meaningfully with respect to either district-level or state-level contextual income. This fact
is important, because districts that are less than 5% black contain over half the U.S. population. These
are typically white voters who live in rural and suburban areas and small cities and have minimal social
exposure to African-American citizens. The substantial overlap among block groups in both rich and
poor districts and rich and poor states indicates that homogeneously white districts are not the source of
major differences between so-called rich and poor states. A second, related finding is that the rich-poor
gap is only meaningfully correlated with state income when we limit the analysis to state house districts
that have sizable black populations. Unsurprisingly, places with a larger proportion of black voters have
poor voters who are much more Democratic than poor voters in other settings. This is unsurprising
because black voters everywhere register with the Democratic Party at rates of about 90 percent, and
poor voters in districts with large black minorities (or black majorities) are disproportionately black. In
13
In the Online Appendix (Table A-1), we conduct a version of this analysis using regression techniques.
14
.1
.1
Proportion Republican
.2
.3
.4
.5
Proportion Republican
.2
.3
.4
.5
$0-20K
$100K+
$0-20K
$100K+
.1
Proportion Republican
.2
.3
.4
.5
Proportion Republican
.2
.3
.4
.5
.6
.6
.1
$0-20K
$0-20K
Med.$100K+
Household Income
$0-20K
$100K+
.1
Proportion Republican
.2
.3
.4
.5
.6
.6
.6
Rich States
Poor States
$100K+
Note: Figures are derived from the full population of registered Democrats and Republicans in party registration states. N=
73,170,970.
rich states, the poorest voters in the most black districts affiliate with the Republicans at rates below ten
percent, a number that is only slightly higher in black areas in poor states.
Here appear the sub-state origins of the red-state, blue-state paradox (Gelman et al., 2008) and,
potentially, the local origins of opposition to redistributive policy in poor states: a huge gap exists
between rich voters who live in more heterogeneous racial contexts in the poor states and rich voters who
live in more heterogeneous contexts in richer states. Voters with block group incomes over $100,000 in
heavily black districts in rich states affiliate with the Republicans, on average, at just over 20 percent,
while voters in the same income group in poorer states affiliate with Republicans at twice that rate.
Several plausible explanations exist for these differences other than the legacies of local race relations
15
and white response to perceived racial threat, but man of these can be dismissed. One might think that
these differences can be attributed to larger middle class black populations in black districts in less
poor states who may be only slightly more likely to be Republican than poor black voters. This is
unlikely, both because the black middle and upper class are increasingly geographically segregated from
poor blacks (Reardon and Bischoff, 2011), and their partisan voting, which remains almost entirely
Democratic (Dawson, 1995), probably explains very little of the partisan difference between rich and
poor areas within such districts. Instead, it is much more likely that differences at the upper end of the
income spectrum can be explained by voter behavior among whites consistent with Keys racial threat
hypothesis. For our purposes, it is irrelevant whether this is a result of individual-level contextual effects,
or is due to the persistence of partisanship arising from social interactions in these areas over multiple
decades.14 In the Appendix (Figures A-6 and A-7) we replicate Figure 2 but restrict the data to block
groups that are less than 10% black and more than 90% non-Hispanic white. While the income-party
intercept is somewhat higher (due to the removal of most black voters in such districts) we find that the
divergence in voting behavior in high-minority districts varies even when we limit the analysis to voters
in block groups that are 90-100% white, allowing us to much more safely assume that the voters being
studied are white voters. These results uphold the strength of the income-party relationship.
15
Another plausible reaction to the results presented in this section is that perhaps they are attributable
differences in a handful of states. We address this concern in several ways. First, in the Appendix
(Figure A-5), we present a graphical summary of the party registration gap between rich and poor voters
in districts of different racial compositions by state. The results confirm the divergence in behavior
among voters in rich and poor states, but the differences are concentrated in districts that are more than
10% black. Second, we present the analysis in Figure 2 separately for each state, displaying estimates for
nearly 2,000 conditional means. These graphs provide a transparent summary of the partisan registration
14
For an excellent discussion of the false choice of contextual versus compositional effects, see Sampson 2012, Ch. 1.
Another feature of Figure 2 establishes the primacy of race: in poor states, no state house districts with average blockgroup median incomes above $60,000 per year are more than 25% African-American. While this could be seen as a weakness
of the nonparametric analysis, in fact it reinforces our findings that race and income are inseparable in the American context,
and income effects should not be discussed without accounting for the geographic distribution of black poverty and its role in
partisan identification.
15
16
of voters in each race-income stratum in each group of state house districts for all party-registration
states. Third, in the Appendix (Figure A-8), we replicate Figure 2 using precinct data from 49 states, and
find the same pattern. Finally, in the next section we adopt slightly stronger modeling assumptions by
using the Catalist data and presidential election results data in multi-level models.
We now continue to concentrate on local areas to demonstrate geographic clustering in the income-party
relationship. For each state house district, the Catalist data provides counts of voters in each income
category, as well as state house district-level racial and income variables. We use these frequencies in
a hierarchical linear model in which the Republican registration proportions in each income bin in each
district are weighted according to the total number of two-party registrants in each district income bin.
We then estimate a model in which mean Republican registration is the dependent variable and intercept
and the effect of the categorical income variable are allowed to vary within district. The six income
categories are included as a set of dummy variables in the six income categories, using as the omitted
base category block-group income of $0 to $20,000. Both the district level intercept, and district-level
coefficients on the categorical variables are allowed to vary. This model can be written as follows:
where yij represents the Republican share of the two-party vote in for unit i situated in district j, represents an overall intercept term, and 2 . . . 6 represent fixed effects for each income category controlling
for the estimated district-level random effects. The random intercept, j , is the intercept coefficient
estimated by partially pooling the mean of observations i located in district j to the global mean intercept. Our primary substantive interest, though, is in the coefficients on the categorical income variables.
These similarly vary by district and are represented by the term k j for each income category k = 2 . . . 6,
where the base category is the bottom income category. These random effect estimates in each community lend themselves to easy interpretation. The coefficient on the top income category (for voters in
17
block groups with income of $100,000 or above) can be interpreted as the district-specific difference
in the gap between the richest and poorest voters. Higher positive (negative) values indicate a higher
(lower) income-party gradient between the poorest and richest voters than would be expected otherwise.
Finally, the error term, ij is independently and identically distributed in each district j. The random
effects and errors are each assumed to be normally distributed with constant mean and variance in each
income category k (Gelman and Hill, 2007, 258):
kj N (k , 2k )
kj N (k , 2k )
ij N (j , 2j )
This model is represented using the R function lmer applied to panel data containing two-party registration data and voter numbers in each district-income bin. (For additional details on the implementation
of these models in the lme4 package (Bates, Maechler and Bolker, 2011), see the Online Appendix.)
We apply a similar varying-intercept, varying-slope model to estimate the income-party relationship
using HEDA precinct data clustered within state house districts, defining the McCain share of the twoparty vote as the outcome variable and defining the income variable as the average of the block-group
level median household income values in each precinct. This model can then be estimated using the
same methods, allowing the relationship between precinct-level income and the two-party vote to vary
within state house districts. The hierarchical model for the income-party relationship for precinct i within
district j is then
yij = + xi + j xi[j] + j xi[j] + ij
where is a general intercept term, is a general coefficient on xi , the precinct-level median household
income variable expressed in tens of thousands of dollars (after accounting for district-level random
effects), and j and j are, respectively, the random intercept and income coefficients for precincts in
each district j. This model is estimated using similar assumptions using lmer (see Appendix).
18
We are most interested in the random effect for the income slope in each state house district, which
is analogous to an interaction term between income variable and the district or county dummy variable.
Rather than presenting such estimates using the usual dot plots or confidence interval plots, we integrate
this with geographic knowledge by presenting our results in maps, but with one more added change
to traditional approaches. While basic choropleth maps have been used to present model estimates,
the weakness of maps that preserve areas or distance is that the area of the units presented is rarely
proportional in area to the amount of data used. As a result, most maps of political quantities in the
United States devote a majority of their area to unpopulated areas such as wheat fields of Kansas and
the barren Nevada and Utah high desert, while high-density population centers such as New York City
and Long Island are nearly imperceptible to the naked eye. We avoid this problem by presenting our
results using cartograms, maps in which geographic units have been distorted so that their areas are
approximately proportional to their population, using Gastner and Newmans diffusion-based cartogram
method (Gastner and Newman, 2004) as implemented in ArcGIS (Gross, 2009).16 The size of the district
presented on each of these maps will be approximately proportional to the 2007 population of the county
or counties that it overlaps. For each model, the j random income effects estimated for each grouping
geography j are presented as a heat map constituting twelve equal intervals. Areas in which the marginal
district-level difference is below expectations appear at the green end of the scale, while areas with a
higher than expected income-party relationship appear at the redder end of the scale.
5.1
A cartogram of state-house-district level random income effects in Figure 3 visualizes interregional differences in partisan voting and highlights the importance of sub-state context. Here, we present 6 ,
the random effect coefficient on the indicator for registered voters in block groups with incomes above
$100,000 per year, relative to a base category of households earning $20,000 per year or less. Thus,
this map captures the magnitude of the rich-poor gap relative to a national baseline. These random
effects on the Republican proportion of two-party registration between voters in the richest versus poor16
19
est block groups range from -0.34 to 0.45. Large swaths of the three party-registration states of the
old ConfederacyNorth Carolina, Florida, and Louisianahave a larger income effect than would otherwise be expected, most notably in the rural Black Belt areas of those states. Similarly, the ethnically
diverse and economically polarized areas of Californias Central Valley region, an area dominated by
poor Latino agricultural workers, have a strong relationship between income and party. The relationship between income and Republican partisanship is lower than would otherwise be predicted in most
metropolitan areas, especially in districts of the Northeast megalopolis running from DC to Boston that
have been the subject of blue-state voter stereotyping. However, overall there is little evidence that
major differences in voting behavior are explained well by state-level contexts.
What are the sources of these major differences? Based on the nonparametric analyses, we again
examine the explanatory value of the racial composition of places. We replicate the racial stratification
in the earlier analysis by subsetting the random effects estimates to display only districts with the highest
and lowest African-American populations. In Figure 4 we show districts that are at least 10% black and
less than 5% black. This demonstrates the link between the racial composition of places and incomebased voting. The variation in voting behavior in homogeneously white areas varies substantially within
states, but does not differ dramatically across states (top panel). The regional divide in the income-based
party affiliation shows up, however, in the map of black-dominated state house districts (bottom panel).
Except in a few urbanized areas of the South where urban blacks live among cosmopolitan whites, such
districts are in rural Black Belt areas where partisan affiliation is strongly related to income. Note, in
particular, the large-magnitude effect in the districts along the Mississippi River in Louisiana and in
northern and eastern counties of North Carolina. Both of these areas have a legacy of slavery and black
sharecropping, particularly relative to the Appalachian areas of those states.
These findings are not merely a consequence of an accounting identity related to the nearly universal
Democratic affiliation of black voters who fall near the bottom end of the economic spectrum. If we
isolate the results to include data from block groups that are not more than 10% African-American
(emulating the results that appear in Figure A-7), areas identified as having the strongest income-party
divide continue to stand out (see Figure A-9 in the appendix). Voters from block groups with few African20
Figure 3: Income Effects are Stronger than Expected in Districts with Rural Minority Poverty
Legend
-0.343 - -0.277
-0.276 - -0.211
-0.210 - -0.145
-0.144 - -0.0786
-0.0785 - -0.0125
-0.0124 - 0.0535
0.0536 - 0.120
0.121 - 0.186
0.187 - 0.252
0.253 - 0.318
0.319 - 0.384
0.385 - 0.450
Cartogram of estimates of the marginal effect of income on individual Republican party registration, using data on
Republican and Democratic registration in block-group income categories in party-registration states. Varying shades of
green indicate a less positive (or more negative) marginal district-specific effect between income and party, while varying
shades of red indicate a more positive (or less negative) marginal relationship.
21
Figure 4: Income Effects Are Stronger than Expected in Rural Areas with Minority Poverty: Districts
5% or > 10% Black
Within-District Random Effect of Income on Republican Registration
Shift from <$20,000 Block-Group Income to $100,000+ Block-Group Income
State House Districts Up to 5% Black
Legend
-0.343 - -0.277
-0.276 - -0.211
-0.210 - -0.145
-0.144 - -0.0786
-0.0785 - -0.0125
-0.0124 - 0.0535
0.0536 - 0.120
0.121 - 0.186
0.187 - 0.252
0.253 - 0.318
0.319 - 0.384
0.385 - 0.450
Legend
-0.343 - -0.277
-0.276 - -0.211
-0.210 - -0.145
-0.144 - -0.0786
-0.0785 - -0.0125
-0.0124 - 0.0535
0.0536 - 0.120
0.121 - 0.186
0.187 - 0.252
0.253 - 0.318
0.319 - 0.384
0.385 - 0.450
This cartogram estimates the marginal effect of income on Republican party registration, using data on major-party
registrants in block-group income categories within state house districts in party-registration states, screened for racial
composition. Top map: districts up to 5% black. Bottom map: districts greater than 10% black. Areas excluded from the
selected racial stratum appear in gray.
22
Americans in the cities and suburbs of the Northeast Corridor continue to stand out for their low levels of
income-based voting, while the magnitude of income-based voting in non-black block groups in districts
that are more than 10% black continues to display the same sorts of regional differences: the incomeparty relationship is stronger in such places in the Deep South for which we have data than elsewhere.
(Key exceptions include a few areas outside the South with high concentrations of black voters who are
heavily segregated from non-urban white voters. See, for example, the areas around Gary, Indiana and
Benton Harbor, Michigan. Regional differences in income-based voting appears to be driven not just by
the behavior of black voters, but also by differences of white voters living in proximity to black voters.
The findings of strong income-based party polarization in Californias Central Valley also appear more
pronounced in these analyses in the Appendix.
To bolster these findings, we present the results of district-level analyses using precinct-level income
and presidential results. These results rely on aggregate-level precinct data, but highlight even more
clearly the unusually strong relationship between income and party across rural areas with a history of
racial and economic inequality. Figure 5 presents a cartogram of the slope random effect, j , which
captures the district-level deviation of the income coefficient from the expected value. The randomeffects coefficient on income (measured in tens of thousands of dollars) ranges from -0.174 to 0.171. The
Black Belt along the Mississippi River and throughout the Deep South appears much more prominently
in this graph, as does Californias Central Valley and predominantly Hispanic Rio Grande area. However,
except in these areas, there is little systematic clustering of these random effects.
While the importance of race seems self-evident in the results that we just presented, we next assess
the relative importance of race as an explanatory variable. Considering that other factors, like culture
and education, have been advanced as explanations of deviations from class-based voting, we aim to
assess their relative influence. To convey the relative influence of non-racial and non-economic expla-
23
Legend
-0.115 - -0.093
-0.092 - -0.070
-0.069 - -0.048
-0.047 - -0.025
-0.024 - -0.003
-0.002 - 0.020
0.021 - 0.042
0.043 - 0.065
0.066 - 0.088
0.089 - 0.110
0.111 - 0.133
0.134 - 0.155
No Data
Cartogram of estimates of the marginal effect of income on the 2008 McCain Presidential Vote, based on data from the
Harvard Election Data Archive. Varying shades of green indicate a lower income effect than expected, while varying shades
of red indicate a higher effect than expected.
24
nations for deviations from income-based voting, we present results in several different ways. We first
test the marginal effect of a series of variables selected to represent cultural explanations for deviations
from income-based voting. Overall, these models suggest that cultural explanations for deviations from
income-based voting provide minimal additional predictive value beyond what is provided by the geography of race.
In Figure 6, we present the coefficients from several regressions. The dependent variable in these
regressions is the value that is plotted for each district in Figure 3 - a measure of where the incomepartisanship relationship is strong and where it is weak. We seek to determine whether cultural variables
explain district-by-district variance in the income-partisan relationship. In all models here, we include
as explanatory variables the districts income, the districts percentage black and the interaction of those
terms. Then, one at a time, we add to this model the districts percentage identifying as evangelical
Protestant, percent married with children, percent with a bachelors degree, and percent rural. Figure
6 plots the coefficient estimates on each of these added variables. The local prevalence of evangelical
Protestantism has minimal bearing on the magnitude of the gap; after accounting for the districts race
and income, a one-standard-deviation shift in the percent evangelical is associated with only a two-point
increase in the rich-poor gap. A one-standard-deviation shift in the percentage married with children
corresponds to just over a one-point increase in the district-level random effect. A one-standard deviation
difference in the percentage of residents with at least a bachelors degree corresponds to a negative twopoint difference in the random effect. While each of these alternative explanations, save percent rural,
yields differences in the expected direction and at a 5% significance level, the associated effect sizes, as
represented by standardized beta coefficients, are quite small.
For another look at these alternative explanations, we return to a version of Figure 5 in which we
measure which districts have a stronger and weaker relationship than expected, but this time we control
for the district percentage black, percentage evangelical Christian, percentage college-educated, percentage married with children, and percentage rural. Figure 7 plots the results. Here again we are looking at
where the partisan difference between voters in the richest block groups and the poorest block groups is
smaller or greater. Even when we control for additional district-level variables, it is still districts largely
25
Standardized Coefficient
0.01
0.02
0.03
% Evangelical
% With BA
% Rural
Explanatory Variable
Standardized Coefficient
Standardized coefficients for marginal effect of three alternative cultural explanations of the rich-poor Republican
Explanatory
Power of
Social Variables
(Net
Income,
Rural)
registration gap. Estimates
are standardized
coefficients,
after inclusion
of District
full set of Race,
interactions
of district
median
household income and percentage black (top panel), and median household income, percentage black and percentage rural
(bottom panel).
0.03
0.02
in the Southern
black belt in which the relationship between income and partisanship is unusually strong.
0.01
0.00
0.01
Discussion
0.02
7.1
0.03
% With BA
Our analysis relies on party registration statistics from voter registration files and on Census block-group
Explanatory Variable
income measures, as well as on geographic measures of state house districts and precincts in addition to
county and states. Scholars may have questions about how choices of measures and boundaries affect
our analysis. Here and in the Online Appendix we address these concerns.
First, by merging individual-level registration data to survey items measuring party identification,
vote choice, and income, and by merging the registration data to aggregate-level measures of election
outcomes, we address the use of party registration and Census income measures. In the appendix, we
compare measures of party identification, self-reported vote choice, party registration, and precinct Presidential vote choice. We also compare self-reported income to median Census block group income. By
26
Figure 7: Income-Based Voting is Strongest in the Black Belt, Even After Accounting for District-Level
Social Variables
Within-District Effect of Precinct Income (Tens of Thousands of Dollars) on McCain Vote
All Districts, Controlling for % Black, % Evangelical, % With BA Degree, % Married With Children, % Rural
Legend
-0.115 - -0.093
-0.092 - -0.071
-0.070 - -0.049
-0.048 - -0.027
-0.026 - -0.005
-0.004 - 0.016
0.017 - 0.038
0.039 - 0.060
0.061 - 0.082
0.083 - 0.104
0.105 - 0.126
0.127 - 0.148
No Data
Cartogram of estimates of the marginal effect of income on the 2008 McCain Presidential Vote based on data from the
Harvard Election Data Archive, after controlling for district percentage black, percentage evangelical Christian, percent
college-educated, percent married with children, and percent rural. Varying shades of green indicate a lower income effect
than expected, while varying shades of red indicate a higher effect than expected..
27
using two data sources, registration records and precinct returns, we overcome the limitations of trying
to measure partisan affiliation using either a single presidential election result or party-registration data.
While they offer different measures of partisan support, both party registration and election results effectively capture the strength of party affiliation. Party registration expresses a decision to stand publicly
with one party, a behavioral measure of party identification that is not vulnerable to survey biases that
occur, among other reasons, due to over-sampling of politically engaged respondents.
Second, we use each voters census block group median income as a proxy for his or her income
class. This represents each voters income class in terms of the median household income among,
approximately, the voters nearest one thousand neighbors (block groups have about a thousand people).
When we refer to rich and poor voters in this article, we are inferring that voters in the poorer
block groups tend to be in a lower-income social class and those in the richer block groups tend to be
in a higher-income social class. This will, of course, introduce some error into the analysis, but blockgroup level income is the lowest-level income estimate available for the full population. The Census
block-group measure is helpful for another reason: while sample-selection and reporting biases exist in
political surveys like the ANES, and especially problematic in questions about income, the block-group
median household income measure, as an indicator of social class, may be less vulnerable to biases than
survey self-reports.
Third, one may fairly worry that state house districts are themselves subject to the Modifiable Areal
Unit Problem and limit the validity of our results. To be sure, it is possible that levels of partisan registration and vote choice in a district will vary with the drawing of district lines, but these gerrymanders must
still cope with local geography. Regardless of the extent of partisan and racial manipulation through
the drawing of noncompact districts, those drawing districts must still operate over a narrow geographic
region that may limit their latitude to draw plans with respect to income, race, and partisan composition.
For gerrymandering to affect our results, it would need to overwhelm such geographic constraints. While
the drawing of legislative lines may dramatically change the racial composition of districts, regional geographic limitations limit the effectiveness of such gerrymanders. Our cartographic evidence shows that
income effects are not limited to isolated, gerrymandered districts, but typically cluster across multi28
ple adjacent districts. Moreover, with the exception of much of the Mountain West, the income effects
that we identify in the Black Belt are observed in both county and district data (see Online Appendix),
suggesting that they are not an artifact of redistricting. Finally, even if the effects here result from redistricting, districts represent political areas of interest; policy is made by legislators elected within these
particular sets of boundaries. In the unlikely event that our results would change due to the selection of
alternative geographic boundaries, the difference across state house districts inside and outside the Deep
South (and a few other agrarian areas) would still demonstrate how voter behavior differs across regions
within policy-relevant boundaries.
7.2
It is worth considering that it might not be rich Southern whites whose behavior is exceptional in our
analysis, but rather rich, non-southern whites. In racially diverse, mostly urban areas of the North, rich
voters are, on average, almost as Democratic as poor voters. We find this to be true even when looking
at rich voters who live in block groups that are over 90% non-Hispanic white. This behavior, implied
succinctly in the question Whats the Matter with Connecticut?, is as puzzling as the unusual strength
of the income-party relationship in places like the Black Belt and California Central Valley (Gelman
et al., 2007).
In future research, we look forward to not only engaging with individual hypotheses that explain the
political divisions in the diverse areas of the South, but also accounting for the political agreement in the
diverse areas of the North. Given our findings, future research that examines the micro-foundations of
income-based voting ought to give particular attention to race-oriented models, like threat, race-based
sorting, as well as racial liberalism (e.g. Tesler and Sears (2010)), rather than to economic and nonracial cultural models. In contemplating such models, priority should be placed on the persistence of
voting behavior and partisanship (Green, Palmquist and Schickler, 2002). While short-term contextual
effects may be at work, we need not attribute the patterns arising in the South or the non-South to the
causal effect of racial context. Voters living in ethnically diverse settings may vote the way they do
because of reactions to context, but while scholarship has presented contextual effects and individual-
29
level selection effects as competing explanations for citizens geographic distribution, selection into and
out of neighborhoods is itself a response to contextual influences (Sampson, 2012).
Conclusion
We find evidence in both individual and aggregate data that local racial context is a dominant factor
explaining the strength of income-based voting. The relationship between income and partisanship does
not vary with state income, and does not even vary much with lower-level geographic contextual income,
except in places with more racial diversity. Surely unsurprising to scholars of racial and ethnic politics,
the strength of income-based voting is bound up with race. It is only within the racially diverse political
subunits of red states and blue states that we see the largest differences in incomes effect on party.
The evidence helps sort out competing paradigms for understanding the political fissures between rich
and poor. Theoretical models of economic voting predict that rich and poor voters should have opposing
political preferences oriented around redistribution (Melzer and Richard, 1981). But empirically, it is
well known that there is substantial cross-state and cross-jurisdiction variation in where rich and poor
voters are politically polarized and where they are not. That the economic context of a locality or a
state, as measured by average income, does not explain variation in the income-party relationship (once
racial diversity is accounted for) calls into question the post-materialist paradigm. Rather, the evidence
presented here is more consistent with the racial context paradigm of (Alesina and Glaeser, 2004): in
the absence of perceived racial threat, issues of redistribution are less politically salient and we would
expect minimal partisan differences between rich and poor.
At the same time, much of the variation of the strength of the income-party relationship can be attributed to the partisan composition of different ethnic groups. That places with a history of endemic
racialized poverty have a stronger income-party relationship is unsurprising given the known and increasing Democratic orientation of racial minorities. Such groups essentially pull down the intercept,
increasing the income-party slope even as other voters preferences remain the same. But it is especially
interesting that these race-based influences on the income-party relationship do not appear everywhere.
Thus, our findings highlight that variation in the income-party relationship is about more than individual
30
models of political behavior; the findings also point to the importance of persistent geographic patterns
in racial settlement and the long-term durability of local partisanship.
The nonparametric and model-based findings presented here reinforce findings based on survey data,
while delving more deeply into the geographic origins of cross-state differences in income-based voting.
Substituting geographically precise micro-level data for survey-based measures that are typically bereft
of fine-grained geographic information, our study offers a new interpretation of findings that income is
a stronger predictor of voting in some places than in others. Among the important implications of our
findings is that large parts of the South are still exceptional in the sense that the Republican Party
has substituted for the old Democratic Party among whites living in racially diverse rural areas, wealthy
voters in such areas are among the strongest supporters of the Republican Party, and race remains an
important factor defining the regions politics. While these results do not establish contextual effects,
they demonstrate that differences in the white voter behavior in areas of racial and ethnic diversity are
at the heart of the income-party relationship. Whether these differences result from contextual effects or
sorting of different types of white voters into different neighborhoods, local political geography provides
one answer to state variation in income-based voter behavior.
Finally, our results help us better understand American political representation. They may explain
previous findings linking racial diversity and the American single-member-district system to the limited
adoption of redistributive social policies (Alesina and Glaeser, 2004; Boix, 2003). If the U.S. had a proportional representation system, the relationship between economic voting and racial/economic context
would be a subject of considerably less substantive importance. However, in a system in which local
areas elect their district representatives, the clustering of voters into homogenous and heterogeneous
areas bears directly on how different classes of voters are represented by elected leaders. Our results
suggest that support for the party of economic redistribution (the Democrats) may be a result of regional
variation in the types of white voters living amidst racial diversity. Varying support for redistribution
may be a result of voters reaction not to statewide racial context, but to effects of local racial context
that translate into the election of different types of representatives. Local political geography is essential
to understanding such preferences.
31
References
Abrams, Samuel and Morris Fiorina. 2012. The Big Sort That Wasnt: A Skeptical Reexamination.
PS: Political Science and Politics 45(2):203210.
Alesina, Alberto and Edward Glaeser. 2004. Fighting Poverty in the United States and Europe: A World
of Difference. Oxford: Oxford University Press.
Ansolabehere, Stephen and Eitan Hersh. 2012. Validation: What Big Data Reveal About Survey Misreporting and the Real Electorate. Political Analysis 20(4):437459.
Ansolabehere, Stephen and Jonathan Rodden. 2011. Harvard Election Data Archive. online.
URL: http://projects.iq.harvard.edu/eda
Auty, Richard M. and Alan H. Gelb. 2000. Political Economy of Resource Abundant States. Paper
prepared for the Annual Bank Conference on Development Economics, Paris.
Bates, Douglas, Martin Maechler and Ben Bolker. 2011. lme4 Package. Online.
Becker, Richard A., William S. Cleveland and Ming-Jen Shyu. 1996. A Tour of Trellis Graphics. Technical report Bell Laboratories Murray Hill, NJ: .
Boix, Carles. 2003. Democracy and Redistribution. New York: Cambridge University Press.
Butler, Daniel M. and Eleanor Neff Powell. 2012. Understanding the Party Brand. Working paper.
Butler, David and Donald Stokes. 1969. Political Change in Britain: Forces Shaping Electoral Choice.
New York: St. Martins Press.
Chen, Jowei and Jonathan Rodden. 2011. Urban Form and Electoral Bias: Evidence from the U.S.
States. American Political Science Association Annual Meeting.
Cho, Wendy K Tam and Thomas J Rudolph. 2008. Emanating Political Participation: Untangling the
Spatial Structure Behind Participation. British Journal of Political Science 38(02):273289.
Cleveland, William S. 1979. Robust Locally Weighted Regression and Scatterplots. Journal of the
American Statistical Association 74(368):829836.
Dawson, Michael C. 1995. Behind the Mule. Princeton: Princeton University Press.
Enos, Ryan D. 2010. What Tearing Down Public Housing Projects Teaches Us About the Effect of
Racial Threat on Political Participation. Unpublished.
ESRI. 2008. U.S. Census Block Groups. In StreetMap North America. ESRI.
Fitch, Catherine A and Steven Ruggles. 2003. Building the National Historical Geographic Information
System. Historical Methods 36(1):4151.
Forest, Benjamin. 2004. Information Sovereignty and GIS: The Evolution of Communities of Interest
in Political Redistricting. Political Geography 23(4):425451.
32
Fotheringham, A.S. and D.W.S. Wong. 1991. The Modifiable Areal Unit Problem in Multivariate
Statistical Analysis. Environment and Planning A 23(7):10251044.
Gastner, Michael T and Mark E J Newman. 2004. A Diffusion-Based Method for Producing DensityEqualizing Maps. Proceedings of the National Academy of Sciences 101(20):7499.
Gay, Claudine. 2002. The Effect of Black Congressional Representation on Political Participation.
American Political Science Review 95(3):589602.
Gay, Claudine. 2006. Seeing Difference: The Effect of Economic Disparity on Black Attitudes Towards
Latinos. American Journal of Political Science 50(4):997.
Gelman, Andrew, Boris Shor, Joseph Bafumi and David Park. 2007. Rich State, Poor State, Red State,
Blue State: Whats the Matter With Connecticut? Quarterly Journal of Political Science 2(345367).
Gelman, Andrew, Boris Shor, Joseph Bafumi, David Park and Jerome Cortina. 2008. Red State, Blue
State, Rich State, Poor State: Why Americans Vote the Way They Do. Princeton, NJ: Princeton University Press.
Gelman, Andrew and Jennifer Hill. 2007. Data Analysis Using Regression and Multilevel/Hierarchical
Models. New York: Cambridge University Press.
Gelman, Andrew and Thomas Little. 1997. Poststratification Into Many Categories Using Hierarchical
Logistic Regression. Survey Methodology 23(2):127135.
Ghitza, Yair and Andrew Gelman. 2012. Deep Interactions With MRP..
Gibson, Keegan and Daniel Gleason. 2012. Obamas WV Problem Stretches Into PA. Politics PA .
Accessed August 12, 2012.
URL: http://www.politicspa.com/obamas-wv-problem-stretches-into-pa/36833/
Gilens, Martin. 1999. Why Americans Hate Welfare: Race, Media, and the Politics of Anti-Poverty
Policy. Chicago: University of Chicago Press.
Gimpel, James and Dante Chinni. 2011. Our Patchwork Nation: The Surprising Truth About the Real
America. New York: Gotham.
Gimpel, James G and Jason E Schuknecht. 2004. Patchwork Nation: Sectionalism and Political Change
in American Politics. Ann Arbor: Michigan.
Gimpel, James G. and Wendy K. Tam Cho. 2004. The Persistence of White Ethnicity in New England
Politics. Political Geography 23:9871008.
Green, Donald P., Bradley Palmquist and Eric Schickler. 2002. Partisan Hearts and Minds: Political
Parties and the Social Identities of Voters. New Haven: Yale University Press.
Gross, Tom. 2009. Cartogram Geoprocessing Tool version 2. Online.
Hill, Kim Quaile and Jan E. Leighley. 1992. The Policy Consequences of Class Bias in State Electorates. American Journal of Political Science 36(2):351365.
33
Huber, John D. and Piero Stanig. 2007. Why Do the Poor Support Right-Wing Parties? A CrossNational Analysis. Presented at theRSF Inequality Conference, UCLA.
Huckfeldt, Robert R and John D Sprague. 1995. Citizens, Politics and Social Communication. Cambridge: Cambridge University Press.
Inglehart, Ronald. 1990. Culture Shift in Advanced Industrial Society. Princenton: Princeton University
Press.
Katz, Elihu and Paul F. Lazarsfeld. 1955. Personal Influence: The Part Played by People in the Flow of
Mass Communications. Glencoe, Ill.: Free Press.
Key, V O. 1949. Southern Politics in State and Nation. New York: Vintage.
Lax, Jeffrey and Justin Phillips. 2009. How Should We Estimate Public Opinion in the States? American Journal of Political Science 53(1):107121.
Lazarsfeld, Paul F. 1948. The Peoples Choice: How the Voter Makes Up His Mind in a Presidential
Election. 2nd ed. New York: Columbia University Press.
Luttmer, Erzo. 2001. Group Loyalty and Taste for Redistribution. Journal of Political Economy
109(3):500528.
McCarty, Nolan, Keith Poole and Howard Rosenthal. 2006. Polarized America: The Dance of Ideology
and Unequal Riches. Cambridge, MA: MIT Press.
McGhee, Eric and Daniel Krimm. 2009. Party Registration and the Geography of Party Polarization.
Polity 41(3):345367.
Melzer, Allan H. and Scott F. Richard. 1981. A Rational Theory of the Size of Government. Journal
of Political Economy 89(5):914927.
of the Census, United States. Bureau. 2006. State Legislative Districts. Technical report Washington,
DC: . Downloaded 15 December 2012.
URL: http://www.census.gov/geo/www/cob/sld metadata.html
Oliver, J. Eric. 1999. The Effects of Metropolitan Economic Segregation on Local Civic Participation.
American Journal of Political Science 43(1):186212.
Park, David, Andrew Gelman and Joseph Bafumi. 2004. Bayesian Multilevel Estimation with Poststratification: State-Level Estimates from National Polls. Political Analysis 12(4):375385.
Putnam, Robert D. 2007. E Pluribus Unum: Diversity and Community in the Twenty-first Century.
Scandinavian Political Studies 30(2):137174.
Reardon, Sean and Kendra Bischoff. 2011. Income Inequality and Income Segregation. American
Journal of Sociology 116(4):10921153.
Romer, Thomas. 1975. Individual Welfare, Majority Voting, and the Properties of a Linear Income
Tax. Journal of Public Economics 4(2):163185.
34
Saez, Emmanuel. 2008. Striking It Richer: The Evolution of Top Incomes in the United States. Pathways pp. 67.
Sampson, R J, S W Raudenbush and F Earls. 1997. Neighborhoods and Violent Crime: A Multilevel
Study of Collective Efficacy. Science 277(5328):918.
Sampson, Robert. 2012. The Great American City. Chicago: University of Chicago Press.
Tausanovitch, Chris and Jonathan Warshaw. 2011. Testing Theories of Representation in Congress,
State Legislatures and Cities Using a Super Survey..
Tesler, Michael and David O. Sears. 2010. Obamas Race: The 2008 Election and the Dream of a
Post-Racial America. Chicago: University of Chicago Press.
Tingsten, Herbert. 1937. Political Behavior: Studies in Election Statistics. London: P.S. King.
Tobler, Waldo R. 1970. A Computer Movie Simulating Urban Growth in the Detroit Region. Economic
Geography pp. 234240.
Vigdor, Jacob. 2006. Fifty Million Voters Cant Be Wrong. NBER Working Papers (12371).
Wong, Cara, Jake Bowers, Tarah Williams and Katherine Drake Simmons. 2012. Bringing the Person
Back In: Boundaries, Misperceptions, and the Measurement of Racial Context. Journal of Politics
74(4):11531170.
35
Online Appendix
This Appendix contains a series of figures, tables, and methodological discussions referenced in the text.
A.1
The dependent variable throughout our study has been an estimate of voter partisanship. In the main
analysis, we measure partisanship based on party registration data. Since party registration is neither the
only basis nor the most common basis for studying partisanship, the validity of the measure merits some
attention. There are essentially three common ways to study the partisanship of voters: one based on vote
choice, a second based on self-reported identification with political parties (i.e. party ID), and a third
based on registration status. Each of these three measures - vote choice, party ID, and party registration
- has advantages and disadvantages for researchers. In this section, we demonstrate that while these
measures differ slightly, in recent years they have been highly correlated and each measure an aspect of
underlying partisan identity.
Vote choice represents the clearest translation of voter preferences into political outcomes. However,
political scientists have studied party ID instead of vote choice in large part because party ID captures
more stable partisan preferences than voters decisions in any one election. As McCarty, Poole and
Rosenthal (2006) argue in support of using party ID, the measure is less influenced by election-specific
factors (75). But party ID has drawbacks too. Since it can only be estimated through surveys, the
measure may be corrupted by sample selection and misreporting biases, and national surveys will not be
useful for studies of small sub-populations. Party registration is advantageous in that it is available at the
individual-level in public records; however, it is only available in three-fifths of the states that collect the
data. Fortunately, the states that have party registration are quite representative of the entire nation, as
Abrams and Fiorina (2012) and McGhee and Krimm (2009) discuss.
The theoretical motivation for studying vote choice and party ID have been well established in the
literature. The basis for studying party registration is not as well established. Thus we offer three
justifications for our use of party registration data as a measure of partisanship. Our first justification
is that it is independently an interesting measure worthy of our attention. Our second justification is
36
that party registration is very closely related to vote choice and party ID both at the individual-level
and aggregate levels. This is especially apparent in the case of party identification which is nearly
perfectly collinear with party registration. Our third justification is seen in a robustness check by which
we replicate with individual-level survey responses one of the graphs we made using party registration
and census income data.
Party registration is, in itself, an important phenomenon in American politics. In many states, party
registration is a pre-requisite for voting in primaries. Therefore the population of registered party affiliates represents the universe of the core voters who determine the party nominees. Furthermore, in every
state with party registration, these data provide the basis for political mobilization. Campaigns utilize
the public records to decide which voters and which geographic areas merit their attention. Thus, the
connection between party registration and other measures of partisanship aside, it is useful to understand
the variance in party registration across voters and across places.
Our second justification is that in addition to the independent virtues of party registration, the measure is also very highly correlated with vote choice and party ID. To show this, we first rely on the 2008
Cooperative Congressional Election Study (CCES) that has previously been merged with public records
of party affiliation (Ansolabehere and Hersh, 2012). Among respondents who are registered Republican
according to public records, 96% of them identify as Republican, 89% reported voting for John McCain
for President, and 91% said they voted for the Republican candidate for state representative. For respondents who are registered Democratic, 95% identify as Democratic, 87% reported voting for Barack
Obama, and 95% reported voting for the Democratic candidate for state representative. Notice there is
slightly more defection for Presidential voting compared to down-ballot voting, which is not surprising given that voters have much less information about non-Presidential races and therefore rely more
heavily on partisanship for their vote choice in these contests.
Figure A-1 shows the state-by-state relationships between party registration stemming from public
records and survey-based measures of partisanship. The first sub-plot simply shows the individual-level
tetrachoric correlations between party registration and each of the survey measures. The second subplot shows the near perfect collinearity between party registration and party ID. Only one state, Utah, is
37
Individual-Level Correlations
GOP Two-Party PID
.4
.6
.2
Pres. Vote
WY
UT
DC
.2
.4
GOP Two-Party Reg.
.6
.2
.4
GOP Two-Party Reg.
.6
.8
.8
McCain Pres. Vote Share
.2
.4
.6
LA
OK
AK
KY NC
NEAZ
NV
SD
KS
WV NM
NH
FL
PANJ CO
OR
CTME
CA
MD
MA NY
RI
DE
UT
DC
SH Dist. Vote
WY
Party ID
MA
OKAK
LANE
NH
KS
AZ
ME
FL
NC
SD
NV
CO
NJ
NM
CT
PA
WVKY
OR
CA
MDDE
NY
RI
.8
LAOK
NH
KS
NE
AZ
NC
NV
FL
CO
PANJSD
KY
ME
AK
CT
OR
NM
CA
WV
NY
MD
RI
DE
WY
UT
MA
.2
.4
GOP Two-Party Reg.
.6
.8
Note: There are 9,621 weighted respondents with non-missing values on two-party PID and two-party registration. There
are 9,743 respondents with non-missing values on two-party registration and two-party Presidential vote choice. And there
are 8,421 weighted respondents with non-missing values on two-party registration and two-party state representative vote
choice. Data source: 2008 CCES matched with Catalist voter registration data.
38
slightly off-diagonal in this graph. To the extent that scholars value party ID as a measure of partisanship,
the individual-level and state-level relationships on display indicate that party registration is a valid proxy
for party ID.
In addition to studying party ID, we investigate the relationship between party registration and selfreported vote choice in the 2008 Presidential race and in the respondents state house district election.
At the state level, the relationship between party registration and vote choice is very tight - especially
with respect to voting in the down-ballot races such as for state house district representative. We display
statistics about voting for state representatives because most of our analysis in this manuscript investigates how partisanship varies according to state house district.
Figure A-2: Party Registration Compared to Precinct Returns
Party Registration vs. Presidential Vote Returns
by State
KS
SD NE
AZ
by State
UT
.8
WY
AK
.8
UT
CO
NHFLIA
OK
ME
NV
PA NC
NM
CA
DE NJ
KY
WV
CT
LA
NYMD
MA
RI
SD
AZ
CO
NH OKFL
IA
MENV
NC
PA
NM
CA
DE NJKY
WV
LA
NY MD CT
RI
.2
.4
.6
GOP Two-Party Precinct Returns
.8
MA
.2
.4
.6
.8
GOP Two-Party Precinct Returns
.2
.4
.6
GOP Normal Vote Based on Precinct Returns
.8
UTUT
UTWY
UT
WY
UT
UT
UTUT
WY
WY
UTWY
UT
UT
WY
UTUT UT
KS
WY
WY
UT
UT
UT
UT
WY
UT
UT
UT
UT
UT
WY
IA
UT
UT
UT
AK
UT
WY
District-Level Corr: 0.75 NH UTUT
UT
WY
ME
UT
UT
AK
WY
KS
UT
UT
UT
KY
WY
ME
KS
WY
WYUT
WYWY
UT
UT
KS
UT
AK
CO
AK
UT
KS
AK
WY
WY
KS
UT
UT
UT
AK
AK
KS
WY
KS
SD
AK
KS
KS
AK
KS
KS
WY
NH UTUT
KS
KS
KS
UT
UT
PA
AK
AK
OK
KS
UT
KS
WY
UT
WV
KS
UT
AK
KS
WY
WY
SD
CO
AK
KY
AK
PA
WY
AK
KS
KS
KS
KS
OK
KS
UT
AK
OK
KY
SD
NC
AK
KY
WY
UT
NE
OK
NE
IA
KS
WY
KS
WY
SD
CO
CO
FL
IA
KS
IA
KS
KS
UT
UT
UT
ME
ME
ME
NE
NE
CT
OK
KS
NH
KS
KY
WY
ME
OK
PA
PA
UT
CO
FL
KS
KS
WY
KS
UT
MD
ME
NE
CT
CO
NH
NM
KS
KS
LA
MEUT
NM
CO
FL
IA
PA
KS
PA
KY
UT
NE
NH
KS
KS
PA
WY
NE
OK
WY
SD
NE
NH
NM
NY
KS
KS
KS
KS
CO
CO
CO
PA
WY
KS
KS
NE
NC
NE
CO
OK
KS
KS
NC
NH
NJ
NH
OK
KS
PA
SD
NE
CO
AZ
NY
OK
KS
NE
NE
IA
KY
KS
NC
NE
OK
PA
OK
PA
KS
ME
NE
NC
CO
NM
NM
NV
UT
ME
NC
AK
PA
PA
NE
NE
CT
FL
FL
KS
UT
NE
NE
FL
PA
IA
KS
AZ
NY
NY
PA
KS
PA
PA
KS
ME
CO
AZ
AK
NY
NY
PA
OK
OK
IA
WY
KS
NH
AZ
AK
NV
WY
KS
KS
NC
N
H
CO
AK
AK
IA
WY
KS
LA
KY
CA
NH
AZ
NY
PA
KY
UT
NE
CO
AZ
NY
OK
PA
IA
KY
SD
UT
NH
OK
FL
FL
PA
SD
CO
CT
KS
SD
WY
CA
AZ
NM
NM
NY
PA
OK
KS
UT
NE
CA
NJ
NY
PA
LA
LA
NC
DE
NY
NY
OK
PA
PA
PA
WY
KS
ME
NC
NE
NY
OK
PA
KS
SD
MD
NH
CO
AK
PA
OK
FL
WY
SD
NC
NC
MD
NE
CO
CT
PA
FL
PA
KS
ME
ME
NC
NH
CO
KS
PA
SD
SD
NC
NC
CA
NH
AZ
AZ
NM
CT
AK
NV
NY
OK
IA
PA
KS
LA
MD
NC
NH
NH
NM
NV
NY
PA
WY
CA
CT
PA
PA
KY
KS
NC
NE
NC
CA
CA
CA
AZ
NY
PA
OK
FL
PA
KS
KS
SD
LA
ME
ME
MD
NHNH
DE
AZ
PA
FL
IA
WV
KS
PA
SD
LA
ME
ME
MD
MD
NE
NE
CA
CA
NH
CO
AZ
AZ
CT
AZ
NY
NY
PA
FL
FL
IA
WY
WY
ME
NC
CA
CT
NH
CT
NV
PA
OK
NY
FL
PA
WY
KY
LA
ME
NC
CA
CA
CO
NJ
CT
AK
PA
PA
OK
IA
KS
PA
PA
ME
NC
NC
NH
CT
DE
NV
NY
PA
PA
WY
ME
ME
NC
NE
CO
CT
NY
FL
FL
FLFL
IA
KS
KS
SD
LA
MD
MD
NC
NE
NH
NJ
NH
CT
AK
NY
NY
FL
FL
IA
PA
KY
ME
NC
NJ
NH
AZ
NM
NM
NY
PA
OK
IA
WV
KS
SD
UT
NC
AZ
NV
PA
OK
OK
FL
IA
LA
ME
ME
ME
MD
NE
DE
CA
CA
CT
NY
PA
PA
PA
IA
IA
IA
KS
KY
ME
MD
NH
NC
NC
NV
PA
NY
F
L
FL
WV
WY
WV
PA
KY
KS
ME
ME
MD
NH
NJ
NM
FL
KS
PA
WV
ME
NJ
NJ
NH
OK
FL
PA
KY
UT
NC
NC
NE
CA
CT
CT
NY
PA
NY
OK
OK
FL
FL
FL
ME
ME
ME
NC
NC
CA
NH
NC
CA
NJ
CO
AZ
NV
PA
PA
FL
FL
IA
PA
PA
KY
MD
CA
CO
CT
AZ
NY
NY
OK
FL
FL
FL
PA
IA
PA
WY
SD
SD
ME
CA
PA
PA
PA
ME
ME
ME
CA
NH
NH
DE
AZ
AK
NY
OK
OK LA LA
FL
IA
IA
PA
IA
IA
WY
SD
ME
ME
MD
NC
PA
OK
PA
FL
FL
IA
PA
SD
S
D
ME
NE
NC
CA
C
A
CT
OK
IA
WY
LA
LA
ME
ME
MD
MD
NH
NM
NM
CT
NY
NY
FL
IA
FL
IA
KY
KY
UT
MD
MD
NC
NE
NH
NH
OK
FL
IA
WY
KY
DE
DE
NH
CT
NM
NY
PA
NY
IA
IA
KS
ME
ME
NJ
NH
NY
FL
FL
IA
WV
KS
UT
NC
CO
CT
CO
CT
NV
OK
PA
FL
FL
FL
FL
IA
PA
LA
MA
MD
NC
NC
NE
CA
CO
CO
CT
DE
DE
CT
AK
NY
PA
FL
KS
ME
ME
ME
ME
NH
NH
NH
NH
CO
CO
NH
CA
NM
NV
OK
FL
FL
WV
WV
SD
NC
NC
NE
DE
CT
DE
FL
IA
KS
MD
NC
DE
NV
FL
KS
KS
KY
MA
CO
CO
CT
N
V
FL
IA
KY
LA
KS
ME
MD
NC
CA
NC
CO
CT
NH
NM
AK
OK
FL
FL
WV
KS
KY
KS
WV
MD
NH
CT
CT
NM
CT
NY
FL
IA
KY
ME
ME
NC
NH
NH
CO
NH
CT
NV
PA
PA
FL
IA
IA
KS
KS
RI
WY
UT
ME
ME
ME
MD
MD
NH
NH
NY
NY
FL
F
L
KS
SD
LA LA
ME
ME
ME
CA
CT
CA
NH
DE
NH
AK
PA
OK
PA
PA
OK
FL
IA
IA
WV
LA
LA
KY
ME
NH
NH
NE
DE
NV
CT
PA
IA
IA
IA
ME
MD
NC
NE
DE
CO
NH
AZ
OK
OK
FL
FL
IA
PA
WY
WV
WV
SD
LA
NC
NC
NH
NH
DE
NJ
NM
IA
KY
LA
ME
NC
AZ
NM
NY
PA
FL
KS
WY
SD
SD
LA
ME
NJ
NH
NV
NV
NY
PA
PA
FL
FL
IA
PA
KS
KS
MA
NC
NC
NH
CT
OK
OK
IA
PA
KS
MD
NE
NE
NH
CO
NM
OK
PA
KY
LA
KY
ME
ME
NH
NH
NJ
NJ
DE
CT
IA
WV
KY
KY
KY
NC
CA
CA
CT
NM
NV
PA
PA
NY
OK
IA
KS
LA
LA
ME
ME
ME
NH
CA
NH
DE
AZ
NM
NM
FL
SD
WV
ME
MA
MA
MD
NC
NC
NC
NH
NM
AK
NY
NV
NY
NY
OK
PA
IA
WV
RI
KY
ME
ME
MA
MD
NH
NH KS
PA
PA
RI
LA
LA
LA
W
V
ME
ME
MA
NE
DE
NE
CT
PA
FL
FL
WV
ME
NH
CO
CT
CT
NV
OK
IA
PA
WV
MA
MD
NH
CO
NH
NM
NM
NM
CT
OK
OK
PA
NC
CT
NY
PA
OK
PA
FL
WV
PA
KY
ME
ME
NE
CT
CO
NM
NM
NV
PA
DE
IA
IA
WY
ME
MA
ME
CA
CA
NH
IA
PA
WV
RI
SD
ME
NJ
NH
CT
NM
CT
CT
PA
OK
NY
OK
PA
MA
NY
OK
FL
PA
WV
MA
ME
ME
NC
CT
OK
IA
KY
RI
LA
LA
LA
LA
CO
NV
NY
PA
IA
WV
WV
KY
MA
ME
CA
CA
DE
DE
CO
CT
NM
OK
KY
LA LA
KY
MA
NC
CA
NH
NH
NH
NH
CT
NV
CT
NV
NY
PA
NY
KS
WV
MA
CO
NV
KSDE
KY
MA
ME
NCNC
CT
CT
NY
IA
MA
MD
NC
NE
CA
OK
IA
PA
IA
WV
KY
MA
ME
NC
DE
NM
IA
RI
LA
ME
MA
ME
MA
CT
NH
CT
AK
IA
RI
KS
LA
ME
NC
DE
CT
NM
NM
NY
NY
PA
OK
OK
PA
MA
NC
CO
CT
CT
CT
NM
CT
NY
PA
PA
NY
WY
KY
MA
ME
NH
CA
NH
CT
NV
NY
PA
PA
IA
RI
MA
MA
MD
CA
NV
NY
N
Y
PA
PA
IA
RI
SD
KY
ME
MD
ME
MA
MD
NC
NC
CT
NM
FL
PA
PA
KY
RI
LA
KY
ME
ME
NC
DE
PA
PA
LALA
ME
NC
NH
DE
CA
NY
NY
OK
OK
IA
PA
RI
LA
ME
CA
CA
CT
CO
NM
NV
PA
OK
FL
RI
SD
KY
MA
ME
MA
MA
NC
AZ
IA
WV
KS
ME
ME
CA
CO
NY
KY
PA
KY
KY
KY
KY
MA
MD
NC
CA
DE
CO
NJ
AZ
CT
CT
NV
NY
PA
PA
LA
ME
MA
MA
NC
CA
CT
CO
CT
OK
NY
OK
FL
PA
KY
UT
WV
ME
MA
NM
NM
CT
NM
FL
IA
PA
KY
LA
LA
MA
NC
ME
CO
CO
AZ
OK
WV
MA
MD
NC
DE
NJ
NY
OK
FL
IA
PA
PA
RI
KY
WV
CA
DE
NH
CT
NM
NM
NV
NY
FL
KS
PA
KY
ME
ME
MA
NC
NC
NH
CT
NH
CT
NM
NV
FL
IA
KY
MA
MD
NJ
NV
NM
OK
WV
KY
LA
LA
MA
NC
NE
AZ
CT
AK
OK
OK
RI
KY
RI
ME
ME
NC
NJ
CT
CO
NY
OK
FL
IAIA
IA
WV
KS
SD
KY
MA
ME
MA
CA
CT
PA
FL
RI
KY
KY
MA
MA
MD
MD
NE
CA
CT
CT
NV
PA
PA
IA
LA
KY
ME
ME
CA
CA
AZ
NY
OK
WV
KY
MA
NC
CO
CT
CT
NM
FL
PA
PA
RI
KY
MD
NC
NC
CT
AZ
PA
PA
FL
FL
PA
WV
WV
RI
LAOK
KY
CT
PA
PA
OK
IA
KY
RI
RI
LA
LA
MA
MA
MA
CO
NJ
PA
OK
PA
IA
KY
MA
MA
MA
NH
CT
NV
FL
ME
MD
NC
NH
NM
AK
NY
KY
MD
MA
NC
CT
CA
CT
RI
LA
ME
MD
MA
NC
CA
OK
OK
IA
RI
RI
NJ
PA
KS
RI
RI
RI
LA
KY
ME
MA
ME
PA
FL
PA
DE
CT
NJ
NJ
NJ
NV
NY
WV
RI
KY
LA
LA
MD
NC
CA
DE
CO
NH
NM
CT
CT
CT
OK
IA
MA
ME
ME
NC
WV
KY
MA
NC
NC
NJ
NE
CT
NY
LA
KS
KY
MA
NC
NC
NH
NC
CA
AZ
OK
PA
PA
KY
KS
LA
WV
MA
MA
NC
AZ
CT
NY
OK
PA
KY
LA
LA
LA
ME
CT
NV
NY
PA
IA
WV
MA
ME
MD
NC
CA
NV
OK
OK
FL
WV
LA
KY
KY
MA
ME
CA
CO
CA
CO
CT
CT
CT
CT
NY
PA
KY
RI
LA LA
ME
MA
NH
CT
NJ
NH
NH
NH
NY
OK
PA
RI
LA
KY
MA
NJ
NY
PA
KS
MA
NM
NM
CT
NM
NY
PA
OK
IA
WV
KS
RI
LA
LA
KY
MA
NC
NC
NM
PA
PA
RI
KY
LA
MA
MA
CA
CT
NY
LA
LA
ME
MA
ME
MD
MA
MA
NC
CT
NY
NY
KS
MA
MA
MA
NY CO
PA
PA
PA
RI
ME
MA
MA
NC
DE
CA
NM
FL
RI
RI
MA
MA
MA
ME
NY
PA
PA
RI
LA
KY
MD
MA
NC
NM
IA
WV
KY
WV
RI
RI
KY
KY
WV
MA
CA
CT
NM
IA
RI
MA
NC
NC
CT
OK LA
FL
KY
RI
KY
MA
NH
NJ
NJ
CT
NM
NM
IA
PA
RI
RI
LA
LA
WV
MA
MA
CT
DE
NY
WV
RI
LA
LA
MD
NH
NY
OK
IA
KS
KY
CT
CT
NY
FL
FL
RI
RI
KY
MA
MA
NC
NC
NJ
NY
FL
FL
KY
MA
NC
CT
NY
ME
MA
CA
NJ
NJ
NM
PA
WV
KY
RI
MA
NC
CO
CT
NM
LA
LA
MA
MD
ME
MD
CA
NM
NY
KY
ME
MA
NC
DE
DE
IA
KY
ME
CO
PA
IA
RI
MA
MA
MA
CT
NV
PA
OK
MA
ME
NC
NC
NC
CT
NY
PA
FL
PA
MA
MA
CA
NM
NM
NV
RI
MA
NC
CA
NM
OK
OK
FL
WV
OK
WV
LA
LA
MA
NJ
OK
PA
KY WV
LA
LA
MA
CT
PA
LA
MA
MA
NC
CA
CO
CT
NY
RI
RI
LA
MA
MA
NH
DE
NJ
NJ
NY
FL
KY
MD
MA
MA
NY
NY
LA
CA
NY
NY
NY
PA
MA
MA
MA
CT
NY
LA
ME
NJ
CT
KS
LA
LA
MA
MA
CT
NM
FL
RI
RI
MA
CT
MA
MA
NC
CA
CT
NY
NY
RI
LA
MA
MD
MA
MA
MA
NM
NY
FL
LA
NY
PA
KY
LA
DE
PA
MARI
CA
PA
MA
NE
PA
RI
MD
NY
FL
KS
RI
MA
NY
WVWV
MA
DE
NY
ME
MA
OK
RI
MA
MD
ME
MD
NJ
NY
FL
FL
PA
RI
RI
RIMA
RI
LA
LA
NY
MA
NY
RI
MA
NM
PA
MA
MA
MD
NY
CT
NY
OK
MA
CT
CT
PA
RI
MD
MD
M
D
CT
LA
NY
MA
NY
PA
PA
RI
RIMA
LA
CT
NY
NY
NY
WV
RI
MA
MA
MA
CT
PA
PA
RI
RI
RI
LA
NY
PA
PA
RI
LA
MA
PA
PA
CT
NJ
CT
CT
NY
NY
NY
LA
MA
MD
LA
MA
NY
RI
LA
CT
NY
MA
CT
NY
NY
NY
MA
CT
PA
PA
MD
CT
RI
RI
LA
CT
RI
MD
NMRI
UT
UT
UT
WY
WY
UT
UT
UT
UT
WY
WY
UT
UT
WY
UT
UT
UTUT UT
WY
KS
WY
UT
UT
WY
UT
UT
WY
UT
UT
UT
UT
WY
UT
WY
UT
IA
UT
UT
NH
UT
AK
UT
WY
UT
WY
ME
UT
UT
AK
WY
KS
UT
UT
WY
UT
KY
ME WY
UT
KS
WY
UT WY
KS
UT
AK
AK
CO
UT
KS
UT
UT
AK
WY
WY
KS
UT
UT
UT
AK
AK
KS
KS
WY
KS
SD
AK
KS
AK
KS
WY
NHUTUT
UT
KS
KS
KS
UT
PA
AK
AK
OK
KS
UT
KS
WY
UT
UT
WV
KS
AK
KS
WY
WY
SD
AK
CO
AK
PA
WY
KS
AK
KS
KS
OK
UT
KS
AK
OK
NC
AK
WY
UT
NE
OK
NE
IA
KS
WY
WY
KS
WY
SD
CO
CO
CO
FL
UT
IA
KS
IA
KS
KS
KS
UT
UT
ME
ME
ME
NE
OK
PA
KS
NE
NH
KS
WY
ME
OK
PA
KS
UT
PA
CO
FL
KS
WY
KS
UT
MD
MESD
CT
NH
NM
CO
CO
KS
KS
MEUT
LA
NM
CO
FL
IA
PA
KS
KS
PA
KY
UT
NE
NH
KS
KS
WY
PA
NE
OK
WY
SD
NE
NM
NY
KS
KS
KS
CO
CO
PA
WY
KS
NC
NE
NC
OK
CO
KS
KS
NC
NH
NH
NJ
OK
KS
PA
SD
NENE
AZ
NY
CO
KS
NE
IA
KS
KY
NC
NE
OK
OK
PA
PA
ME
NE
NC
NM
NM
NV
CO
UT
ME
NC
AK
KS
PA
PA
NE
FL
FL
KS
UT
NE
NE
NM
NY
FL
PA
IA
KS
KS
NY
NY
PA
KS
PA
ME
AZ
NY
NY
OK
AK
OK
CO
PA
IA
WY
KS
NH
NV
AK
WY
KS
KS
NC
NH
AK
AK
CO
IA
LA
NH
AZ
PA
UT
NE
AZ
NY
OK
PA
I
A
UT
SD
NE
NJ
NH
OK
FL
FL
PA
SD
LA
CO
KS
WY
SD
NM
NM
AZ
OK
OK
PA
PA
KS
UT
NE
NJ
CA
NY
PA
SD
LA
LA
NC
NY
DE
NY
PA
PA
WY
KS
ME
NC
NC
NE
OK
KS
SD
MD
NH
AK
OK
CO
PA
FL
WY
SD
MD
NC
NC
NE
CO
PA
FL
PA
KS
ME
ME
NC
NH
NY
CO
SD
SD
NC
NH
NM
NV
NY
AZ
AZ
AK
IA
PA
KS
MD
LA
NC
NH
NH
NM
NV
NY
PA
WY
PA
PA
KS
NC
NC
NE
CA
AZ
NY
OK
FL
PA
KS
KS
SD
LA
ME
MD
ME
NH
DE
FL
PA
IA
KS
WV
PA
SD
LA
MD
MD
ME
ME
NE
NE
NH
AZ
AZ
NY
CO
PA
FL
FL
WY
WY
ME
NH
NV
OK
PA
PA
WY
LA
ME
NC
NJ
CA
AK
NY
OK
CO
PA
IA
KS
PA
PA
ME
NC
NC
NH
NV
NY
CO
NY
DE
PA
WY
ME
ME
NC
NE
NH
NY
CO
FL
FL
FLFL
FL
IA
KS
KS
SD
MD
MD
NE
NH
NH
NJ
AK
NY
PA
FL
IA
PA
ME
ME
NE
NC
NH
NJ
NM
NMLA
AZ
NY
OK
PA
IA
WV
KS
SD
UT
NV
AZ
OK
IA
IA
KS
LA
ME
ME
MD
ME
NE
NM
NY
CA
CA
NY
DE
PA
PA
IA
IA
IA
KS
MD
ME
NC
NH
NV
NY
FL
WV
WY
PA
KS
ME
ME
MD
NH
NJ
NM
FL
KS
PA
ME
NC
NJ
NJ
NH
OK
FL
PA
IA
IA
PA
KY
UT
NC
NE
OK
NY
OK
PA
FL
F
L
FL
ME
ME
NC
NC
NC
NV
CA
CA
CA
AZ
CO
PA
PA
FL
IA
PA
PA
MD
CA
AZ
NY
NY
NY
OK
CO
FL
IA
PA
PA
WY
SD
SD
ME
PA
FL
PA
ME
ME
ME
ME
NH
NH
CA
OK
AK
NY
O
K
DE
FL
FL
IA
IA
IA
IA
PA
WY
SD
MD
ME
ME
NC
NY
OK
PA
FL
PA
FL
IA
PA
SD
ME
NC
NE
CA
OK
IA
WY
LA
MD
MD
LA
ME
ME
NH
NM
NM
NY
FL
IA
UT
MD
M
D
NC
NE
NH
NH
OK
FL
IA
WY
LACA
NH
NM
NY
NY
DE
DE
IA
IA
KS
LA
ME
ME
NH
NJ
NY
FL
FL
IA
IA
WV
KS
UTSD
NC
NV
CO
CO
CT
FL
FL
FL
PA
FL
IA
PA
MD
LA
NC
NC
MA
NE
CA
AK
NY
CO
CO
PA
FL
DE
DE
ME
ME
ME
ME
NH
NH
NH
NH
NM
OK
CO
CO
FL
FL
WV
SD
NC
NE
DE
DE
FL
IA
KS
MD
NC
NV
D
E
FL
IA
KS
NC
MA
N
V
CO
CO
FL
IA
KS
SD
LAFL
KY
ME
MD
NC
NC
NH
NM
CA
AK
CO
CO
OK
FL
FL
WV
KS
KS
MD
NH
NM
NY
IA
ME
ME
NC
NH
NH
NH
NV
CO
PA
FL
PA
IA
IA
KS
KS
WY
UT
ME
MD
ME
MD
ME
NH
NH
NY
NY
FL
FL
LA
ME
NH
NH
CA
AK
OK
DE
FL
PA
PA
IA
IA
IA
LA
LA
ME
NE
NV
DE
P
A
IA
IA
IA
ME
MD
ME
NC
NE
AZ
OK
OK
OK
CO
DE
FL
FL
IA
PA
WY
SD
LA
NC
NC
NH
NH
NJ
NM
DE
IA
ME
NC
NM
AZ
NY
PA
FL
KS
WY
PA
KS
SD
LA
ME
LA
NH
NJ
NV
NV
NY
PA
FL
FL
IA
IA
PA
KS
KS
LA
NC
MA
NC
NH
OK
OK
IA
PA
WY
KS
MD
NE
NE
NH
NM
OK
CO
PA
ME
ME
NH
NH
NJNJ
DE
IA
ME
NC
NM
NV
CA
NY
OK
PA
PA
IA
KS
KS
LA
ME
ME
ME
NH
NH
NM
NM
AZ
NY
CA
DE
FL
SD
ME
MD
M
D
MA
MA
NC
NH
NM
NY
NV
NY
NY
OK
AK
PA
IA
WV
ME
MD
ME
ME
NC
MA
NH
PA
MA
LA
ME
LA
ME
LA
NE
NE
DE
PA
FL
FL
ME
NH
NV
OK
CO
IA
WV
PA
MD
MA
NH
NH
NM
NM
NM
OK
OK
CO
WV
PA
NC
NC
NH
NY
OK
PA
PA
FL
WV
PA
KY
ME
ME
NE
NM
NM
NV
CO
DE
CT
PA
IA
IA
WY
ME
ME
MA
NH
NY
IA
PA
SD
ME
NH
NJ
NM
OK
NY
CT
DE
PA
PA
MA
NY
OK
CT
FL
PA
ME
MA
NC
OK
IA
LA
LA
LA
LA
ME
LA
NV
NY
CO
INH
A
PA
ME
LA
MA
NM
OK
CO
DE
DE
IA
LA
NC
MA
NH
NH
NH
NH
NV
NV
NY
NY
PA
IA
WV
KS
MA
NC
NV
NY
CO
KS
KY
ME
MA
NC
NY
IA
MD
MA
NC
NE
OK
IA
PA
ME
MA
NC
NM
NV
NY
DE
LA
MA
ME
ME
MA
NH
NH
AK
IA
KS
RI
LA
ME
MA
NC
NC
NM
NM
CA
NY
OK
OK
DE
PA
NM
NY
NY
CO
OK
PA
PA
WY
MA
ME
NH
NH
NH
NH
NV
CA
NY
PA
PA
IA
MA
MD
MA
NV
NY
DE
IA
PA
SD
MD
ME
ME
MD
ME
NC
MA
NM
FL
KY
KY
LA
ME
MA
MA
NC
DE
PA
LA
MD
ME
NC
OK
O
K
NY
DE
NY
IA
PA
LA
ME
NM
NV
OK
CO
PA
FL
SD
ME
MA
NC
MA
MA
AZ
IA
WV
KS
LA
ME
ME
NY
CO
PA
KY
MD
MA
NC
MA
NH
NJ
NV
AZ
CA
NY
CO
DE
PA
KS
PA
LA
ME
MA
MA
NC
OK
OK
CO
FL
WV
PA
MA
UT CA
ME
NM
NM
NM
FL
IA
PA
LA
LA
NC
ME
MA
AZ
CO
OK
WV
MD
MA
NC
NJ
NV
NY
OK
DE
FL
IA
WV
PA
PA
MA
NH
NM
NM
NV
NY
OK
DE
FL
KS
PA
MA
ME
ME
NC
NC
NH
NH
NM
NV
FL
IA
MD
MA
NJ
NV
NM
LA
LA
MA
NC
NE
AZ
OK
OK
AK
ME
ME
NC
NJ
NY
OK
CO
FL
IAIA
IA
WV
KS
SD
ME
MA
MA
OK
PA
FL
KY
MD
MD
MA
MA
NE
NV
PA
FL
IA
IA
LA
ME
ME
OK
AZ
NY
WV
LA
MA
MA
NC
NM
CO
FL
PA
PA
MD
NC
MA
NC
AZ
PA
PA
FL
FL
WV
WV
PA
LA
OK
PA
PA
LA
LA
MA
MA
NJ
OK
CO
PA
FL
PA
IA
MA
NC
MA
NH
NV
CT
FL
MD
ME
MA
NC
NH
NM
AK
NY
MD
MA
NC
OK
LA
MA
MD
ME
NC
MA
OK
OK
IA
NJ
PA
KS
KYNV
LA
ME
ME
MA
FL
PA
NH
NJ
NJ
NJ
NV
NY
DE
LA
MD
MD
LA
NC
NH
NM
CA
CO
DE
ME
MA
ME
NC
WV
MA
NC
NC
NE
NJ
NY
KS
LA
MA
NC
NC
NH
AZ
OK
PA
PA
KS
LA
NC
MA
MA
AZ
OK
NY
DE
PA
LA
LA
LA
LA
ME
NY
PA
IA
MD
ME
NC
MA
NV
OK
OK
NY
PA
FL
KY
LA
ME
MA
MA
CO
NY
ME
NH
NH
NH
NJ
NH
OK
NY
LA
MA
NJ
PA
KS
MA
MA
NM
NM
NM
NY
PA
IA
WV
KS
LA
KY
LA
NC
MA
NC
NM
PA
PA
LA
MA
MA
MA
NY
LA
ME
MD
ME
LA
MA
NC
MA
MA
NY
NY
KS
MA
MA
NY
PA
PA
PA
ME
NC
MA
NM
DE
FL
ME
MA
MA
MA
COCO
NY
PA
PA
LAMA
MD
NC
MA
NM
IA
WV
KY
MA
MA
NM
IA
MA
NC
NC
OK
FL
MA
NH
NJ
NJ
NM
NM
IA
WV
PA
LA
MA
LA
MA
MA
NM
NY
DE
WV
LA
LA
LA
MD
NH
OK
NY
NY
KS
NY
FL
RI
KY
NC
NC
MA
MA
NJ
FL
FL
MA
NC
ME
MA
NJ
NJ
NM
CA
PA
WV
MA
NC
NM
LA
MD
ME
MD
LA
MA
NM
NY
WV
KY
ME
NC
MA
DECO
DE
IA
ME
MA
CA
CO
CO
PA
IA
MA
MA
MA
NV
OK
PA
ME
MA
NC
NC
NC
PA
FL
PA
MA
NM
NM
NV
NC
MA
NM
NV
OK
OK
FL
WV
OK
LA
LA
MA
NJ
OK
NY
PA
KY
LA
LA
MA
PA
LA
MA
MA
NC
MA
LA
MA
MA
MA
NH
NJ
NJ
DE
NY
FL
MD
MA
MA
NY
LA
NY
NY
NY
PA
MA
MA
MA
NY
LA
ME
NH
NJ
KS
LA
LA
MA
MA
NM
FL
MA
MA
MA
NC
NY
N
Y
MD
MA
LA
MA
MA
MA
NM
NY
FL
LAME
MA
NY
PA
KY
LA
DE
PA
MA
NC
PA
MA
NE
MD
MA
NY
FL
KS
RI
MA
NY
MA
NY
DE
MA
OK
MD
MD
ME
MA
NJ
NY
FL
FL
PA
LA
LA
NY
MA
NY
MA
NM
PA
MD
MA
MA
NY
OK WV
NY
MD
MA
PA
MD
MD
MD
CT
NY
LA
NY
MA
NY
PA
PA
LA
NY
MA
MA
MA
NY
PA
LA
NY
LA
MA
NY
PA
PA
MA
NJ
NY
PA
LA
MD
MA
LA
MA MA
NY
NY
NY
LA
NY
MA
NY
MA
PA
PA
MD
LA
CT
MD
NM
1
GOP Two-Party Reg.
.2
.4
.6
.8
WY
AK
KS
NE
DC
.2
.4
.6
.8
GOP Normal Vote Based on Precinct Returns
Note: Precinct returns in Oregon were not available for use in this graph. Additionally, DC precinct returns are not available
for the subplots estimating normal vote. In the Catalist 1% sample, 691,345 registered Democrats and Republicans were
matched by Census block group to the vote returns from their precincts. Data source: Catalist 1% analytics sample and
Ansolabehere/Rodden precinct returns.
39
Aside from the survey analysis, we also compare party registration to actual vote returns. We do
this by combining two sources of data. First, for a 1% sample of voters in Catalists database, we have a
record of each registrants party affiliation and the census block group of his or her residence. Second, we
utilize precinct-level vote returns collected by Stephen Ansolabehere and Jonathan Rodden. By merging
these files based on block group, for 1% of all registered voters we have both their registration status and
the vote distribution in the precinct in which they live.
In Figure A-2, we aggregate the registration and vote choice data to the state level and to the state
house district level. We show the relationship by aggregating precinct-level Presidential vote returns from
the 2008 election and also by aggregating a precinct-level normal vote calculated in Ansolabehere and
Rodden (2011). The normal vote measure averages returns from down-ballot contests and from prior
years, but the inputs to the measure vary based on data availability across states.
As with down-ballot voting measured on the survey, the normal vote measure ought to be a more
stable signal of partisanship than voting returns from a single Presidential election. Indeed, the normal
vote measure is more highly correlated with party registration than Presidential returns are in the state
house district graphs. In the statewide graph, the correlation measure is lower for the normal vote only
because we were not able to use data from Washington DC in the normal vote subplots. While both
measures stemming from vote returns are clearly very similar to the party registration measure, here
we see that there are several outlier states. Kentucky, West Virginia, Louisiana and Oklahoma vote
somewhat more Republican than their registration records suggest. Nevertheless, notice that these states
are not outliers in the survey comparisons. To the extent that party ID, party registration, and vote choice
pick up on slightly different aspects of partisanship, the registration and ID measures are closer to one
another and are both slightly different from vote choice in some states.
As another look at the robustness of party registration data, we rely on data shared by Daniel Butler
that indicates the party affiliation of each state legislator elected in the 2010 midterm election (see Butler
and Powell (2012)). We merge Butlers data with the proportion of registered Republicans in each
district, as derived from the Catalist estimates. In Figure A-3, the Republican vote share is binned
into 10-percentage point groupings. For instance, the first grouping includes all districts that are 0-
40
Pr(Representative is Republican)
.25
.5
.75
531
263
96
51
20
452
429
112
<10% R
220
332
40-50% R
Percentage Republican of D and R Registrants in District
>90% R
Note: Data on legislator partisanship provided curtesy of Daniel Butler. Districts in Nebraska are not included because their
representatives are non-partisan. The numbers in the figure represent the number of districts that fall into each
party-registration bin.
10% Republican. Each point is represented by the number of districts falling into each bin. The figure
demonstrates the strong relationship between party registration and electoral outcomes in state house
districts.
The close relationships demonstrated here among different measures of partisanship not only help to
justify our reliance on the most under-utilized of these measures, but it also helps advance scholarship on
partisanship in general. Partisanship as measured by party ID, registration, or vote choice, and measured
in surveys or in public records, exhibits a remarkable degree of conformity. Each of these measures
has virtues and disadvantages, and that is why we replicate our analysis using alternative measures.
Nevertheless, future projects that will be unable to replicate their analyses with multiple measures of
partisanship may gain confidence in their results from the comparisons illustrated here.
A.2
We compare our income-based survey results to data from the 2008 Cooperative Congressional Election
Survey, which has previously been linked to Catalist voting records and therefore allows us to compare
41
block group income with self-reported income (Ansolabehere and Hersh, 2012). Figure A-4 shows that
comparison. Apart from the close relationship between the income measures displayed in this graph,
we also learn from the CCES that the block group measure picks up on other aspects of income class.
For example, 75% of voters in block groups in which the median income is greater than $100,000 report
owning their home. In block groups in which the median income is less than $20,000, only 35% report
owning their home. Voters in poor block groups are also much more likely to report that jobs in their
local area are scarce.
Figure A-4: Relationship between Self-Reported Family Income and Census Block Group Median
Household Income
$150 +
$120-$150
$100-$120
$80-$100
$70-$80
$60-$70
$50-$60
$40-$50
$30-$40
$25-$30
$20-$25
$15-$20
$10-$15
$0-$20
$20-$40
$40-$60
$60-$80
$80-$100
$100 +
Census Block Group Median Household Income (in $20K groupings)
Mean
95% CI
IQR
Means, 95% confidence intervals and interquartile ranges are shown. Data generated from 2008 CCES matched to Catalist
Census data for 24,107 CCES respondents who were identified in registration records. The CCES asks respondents to report
income within the ranges identified on the y-axis.
A.3
To reinforce the intuition of the nonparametric averages, we present least squares regressions comparable to the nonparametric regression, using the full population of registered Democrats and registered
Republicans. We first regress the binary dependent variable, Republican registration, on a registered
voters block group income and district income, employing state fixed-effects. The block group income
variable takes on values 1, 3, 5, 7, 9, and 11, corresponding to the center of each of six major income
42
categories: $0-$19,999, $20,000-$39,999, and so on, up to a top category of $100,000 and up. In a
second regression, we add as a covariate the black proportion of district population. In a third regression,
we add the interaction between the district income and district race measures.
The results of this analysis (Table A-1) convey the importance of district racial composition to the
income-party relationship and show that the effect of district income evaporates once district race is
accounted for. The first column shows that, after accounting for state fixed effects, Census block-group
income, and state house district median income, Republican registration increases by 2.5 percentage
points for every additional $10,000 in median block-group income, and by an additional 1.7 percentage
points for every additional $10,000 in district income.
While the Model 2 and Model 3 coefficient on block-group level income remain identical to Model
1, the state-house-district income effect is reversed in Model 2 and reduces to 0 in model 3. This change
has been picked up in the district-level race term. In model 2, Republican registration is now expected to
drop by 6.3 percentage points for every additional 10 point increase in the black share of the population,
a huge effect size. It is especially noteworthy that the addition of the contextual race variable does little
to affect the coefficient on the block-group income variable, but erases the contextual income effect.
A.4
While the district-level analysis presented in Figure 2 makes clear that race explains differences in the
relationship between contextual income and partisan identification, it does so by disregarding the labels
43
of the rich and poor states in which different types of districts are located. We now disaggregate the result
by presenting summaries of the four sets of estimates that we presented in Figure 2, but representing them
now by displaying the low-income and high-income estimates in each state, in each racial-composition
interval. These results appear in Figure A-5. In each group, rather than presenting the average Republican
registration in each $20,000 income interval, we present for each state, the Republican registration in
block groups with less than $20,000 income and those with over $100,000 average median block-group
income. The Republican registration in low-income block groups in each type of state house district is
represented by an L, and in high-income block groups it is represented by an H. The income category
of each state is represented by red and purple. The number of districts in each state that meet the stated
racial composition criteria appear in parentheses with each interval estimate. Notice that all 29 party
registration states are represented in the upper left subplot, but some are missing from the other sub-plots
since they lack districts that meet the racial criteria.
Presenting effect sizes separately by state confirms the intuition conveyed when the data from rich
and poor districts are completely pooled. Voters in the racially homogeneous areas of rich and poor states
are not substantially different in their voter registration behavior. In some states, the rich-poor partisan
gap is bigger than others, but the gap is as big in New Jersey as Louisiana and as big in Connecticut
as Kentucky. The states only sort into the familiar red-state/blue-state rich-state/poor-state categories
with respect to income-based voting when we confine the analysis to districts containing large numbers
of African Americans. In such districts, the gaps between the richest and poorest voters are smallest
in states considered traditional blue states: Massachusetts, Rhode Island, Oregon, Pennsylvania, and
Maryland. As we demonstrate with maps, the blacks who live in these blue states overwhelmingly live in
urbanized districts where non-black voters are also heavily Democratic. On the other end of the scale are
a set of traditionally Republican states where blacks live in rural poverty and wealthy non-black voters
living in proximity are much more Republican.
A.5
44
Figure A-5: State-Level Variation: State-Level Differences Rooted in Racially Heterogeneous Districts
Districts Less Than 5% Black
NE
.9
H KS
L
H
(75)
AK
OK
SD
AZ
NC
CO
H
LA
KY H
L FL IA H PA NHNV H
H NM
H
NJ
DE
H
CT
L (43)NY (95) H H H CA H H L WV
ME H
H
L
H
H
OR
L
(15) H
(30) H
L L (16) H
H
L H
L
L
H
L L L (37)
RI L
L
(38)(90)
MA(53)
L
L (28)
(151)
(71)
(136)L(103)
(25)
(58)
H
L (58)
L H L (65)
(52)
L
L
L (4)
(47)
(69)
(60)
MD L
L
(59)
H
L
L H
OR
(10)
H
H L
L (1)
RI
CO
L
L (10)
L L
(23)
(1)
AK
FL
H
CA
(41)
(6)
MA L
(16)
(3)
L
(8)
AZ
IA
WV
L H
NE
L H
L
NY (7)
(7)
PA
CT
H L
(13)
(1) (29)
(21)
(11)
L
(28)
L L L L
L
(19)
(16)
(9)
WV
H
L
(3)
H L
L (21)
H
L
(1)
(20)
RI
H
L
(6)
CO
CA
H
MA
(7)
L
(7)
H
L
(4)
KS
H H
L H
AZ
PA
L L H H
(23) (39)
(3)
(1)
L
(21)
H
L
(41)
(8)
L
(12)
IA
NY
H
L
(9)
L L
(19)
H
NJ
CT
KY
(3)
(4)
(11)
.1
MD
NE
FL NC
H
NV
(23)
DE
OK
LA
Proportion Republican
.3 .4 .5 .6 .7
.2
AK
.8
.9
.9
NV
H
H
H H
(28)
KY
(16)
OK
H H
KS
.8
Proportion Republican
.3 .4
.5 .6 .7
.2
.1
0
.1
(129)
MD
NJ
(7)
NM
H H
DE
.2
(3)
(16)
NC
.1
.2
(95)
LA
.8
H
L
Proportion Republican
.3
.4 .5 .6 .7
.9
.8
WY
Proportion Republican
.5 .6 .7
.3 .4
UT
NE OK
H H
KS KY DE
NV
LA
NC FL
MD CA CT
CO
H H
NY
H H H H
NJ
L
L
H L
H H H
L
L (47)
L
L
(5)
(42)
H H
L
L (2) L
(6)
(18)
(7)
H RI L L L L L
(4)
L
(2)
(2)
L
L L
(5)
(19) (20)
(16)
(5)
H (1)
(28)
(7) (1)
MA
OR
PA
Note: L represents voters in low-income ($0-20,000) block groups, while H includes voters in block groups with
median income of $100,000 and above. Purple lines are poor states, red lines are rich states. Number of districts included in
calculation appears in parentheses. Figures are derived from the full population of registered Democrats and Republicans in
party registration states. N= 73,170,970. Percentage black and median household income are derived from Year 2000
Census block group statistics; partisanship is derived from individual-level voter file records.
45
Figure A-6: The Link Between Income Effects and Local Racial Context Persists When Only Non-Black
Block Groups Included
.1
.1
Proportion Republican
.2
.3
.4
.5
Proportion Republican
.2
.3
.4
.5
$0-20K
$100K+
$0-20K
$100K+
.1
Proportion Republican
.2
.3
.4
.5
Proportion Republican
.2
.3
.4
.5
.6
.6
.1
$0-20K
$0-20K
Med.$100K+
Household Income
$0-20K
$100K+
.1
Proportion Republican
.2
.3
.4
.5
.6
.6
.6
Rich States
Poor States
Note: Only registrants residing in block groups less than 10% black are included.
46
$100K+
Figure A-7: The Link Between Income Effects and Local Racial Context Persists When Only White
Non-Hispanic Block Groups Included
in Varying Racial and Economic Contexts, White Non-Hispanic Block Groups Only
State House Districts 0-5% Black
.6
Proportion Republican
.2
.3
.4
.5
Proportion Republican
.2
.3
.4
.5
.1
.1
$0-20K
$100K+
$0-20K
$100K+
.1
Proportion Republican
.2
.3
.4
.5
Proportion Republican
.2
.3
.4
.5
.6
.6
.1
$0-20K
Med.
Household Income
$100K+
$0-20K
$100K+
.1
Proportion Republican
.2
.3
.4
.5
.6
.6
Rich States
Poor States
Note: Only registrants residing in block groups more than 90% non-Hispanic white are included.
47
$100K+
A.6
Figure A-8: The Link Between Income Effects and Local Racial Context Persists When Precinct Data
are Used in Place of Registration Data
.6
.1
.1
$0-20K
Precinct Income
$100K+
$0-20K
Precinct Income
$100K+
.1
.5
.5
.6
.6
.1
$0-20K
$0-20K
Med.
Household Income$0-20K
$100K+
$100K+
.1
Proportion Republican
.2
.3
.4
.5
.6
.5
Precinct Income
Precinct Income
Rich States
Poor States
$100K+
Note: This graph replicates the analysis if Figure 2 with precinct data from 49 states. For each precinct, we generate an
income level based on the block group(s) located in that precinct. We show the relationship between precinct income and
McCain vote share in the 2008 election. As in Figure 2, this relationship is shown in three levels of state house district
income, four levels of state house district race, and two levels of state income. The same divergence between poor states and
rich states emerges in the heavily black districts here as in the registration data.
48
A.7
Summary statistics for the state house district data used in multilevel modeling appear in Table A-2.
Summary statistics for the HEDA data appear in Table A-3.
Variable
Block Groups Per District
Two-Party Registrants
Proportion R Regis.
Proportion Black, 2000
Mean SD
Min. Max.
404
447
17
3,198
28,460 31,746 1,149 230,600
0.432 0.189 0.030 0.941
0.080 0.146 0
0.928
Table A-2: Summary statistics for state house districts used in multilevel modeling (n=2,570).
Variable
Total Votes
McCain Votes
Obama Votes
McCain Proportion of 2-Party Vote (%)
Median HH Inc., 1999 (x$10,000) (Mean of BG Med HH Inc)
Pct. Black, 2000 (Total Black Pop./Total Pop.)
Mean
766
351
415
45
4.59
11.2
SD
615
364
357
21
2.24
20
Min.
1
0
0
0
0
0
Max.
15,020
8,419
11,290
100
20
99.5
Table A-3: Summary statistics of Harvard Election Data Archive precinct data merged with 2000 Census
block-group data (n=165,631).
49
A.8
Legend
No Data
REincgroupinc100up
-0.343752 - -0.090903
-0.090902 - -0.059212
-0.059211 - -0.038957
-0.038956 - -0.019833
-0.019832 - -0.002100
-0.002099 - 0.015755
0.015756 - 0.034408
0.034409 - 0.058717
0.058718 - 0.093806
0.093807 - 0.335929
Legend
No Data
REincgroupinc100up
-0.343752 - -0.090903
-0.090902 - -0.059212
-0.059211 - -0.038957
-0.038956 - -0.019833
-0.019832 - -0.002100
-0.002099 - 0.015755
0.015756 - 0.034408
0.034409 - 0.058717
0.058718 - 0.093806
0.093807 - 0.335929
Replication of Figure 4 restricted to Census Block Groups that are 5% black and more than 10% black.
50
A.9
The multilevel model using the Catalist data translates into the following model in lmer:
lmer.out <- glmer(pctr2pincgroup+(1+incgroup|stdist),
weights=tot2p, data=shpanel)
where pct2rp is the Republican proportion of two-party registration, incgroup is a factor variable
for income group, and stdist is a grouping variable for each state house district. Each of the districtincome bin observations in the panel data is weighted using the variable tot2p, the total number of
voters in each income category reporting affiliation with one of the two parties.
The multilevel model using the HEDA precinct data is represented via lmer as follows:
lmer.out<-lmer(pctr08medhhinc10k+(1+medhhinc10k|county), data=precinct)
where pctr08 is the McCain proportion of the two-party vote in each precinct, medhhinc10k is
the unweighted average of the median household income in the block groups overlapping each district,
measured in tens of thousands of dollars, and county groups precincts into counties.
Figure A-10: State House District Aggregates Follow Individual-Level Income-Party Patterns, State
Aggregates Do Not
States
.1
.1
Proportion Republican
.2
.3
.4
.5
Proportion Republican
.2
.3
.4
.5
.6
.6
$0-20K
$100K+
$0-20K
$100K+
Poorest States
Richest States
Poorest Dists.
Richest Dists.
36 Pt. Diff.
29 Pt. Diff.
18 Pt. Diff.
24 Pt. Diff.
Note: Figures are derived from the full population of registered Democrats and Republicans in party registration states. N=
73,170,970. Median household income are derived from Year 2000 Census block-group statistics; partisanship is derived
from individual-level voter file records.
party registration states. The left panel of Figure A-10 presents Republican registration as a function
of block-group income, stratifying according to state-level income. Echoing Gelman et al. (2008), the
poorest voters (those in block groups with median income under $20,000) register as Republicans at a
rate of only twenty to thirty percent, with poorer voters in the poorer states actually slightly more likely
to register Republican (8 percentage points more likely) than poor voters in richer states. However,
as block-group level income increases, the gap between voters in rich and poor states grows. At the
highest income levels, the gap in Republican registration and voting between rich states and poor states
reaches 14 percentage points. Only 49% of members of the top income category (block-group income
of $100,000 or more) in rich states are registered as Republicans, while over 63% of the richest voters in
poor states are registered as Republicans.
When we focus on state house districts as income contexts rather than states, the familiar paradoxical relationship between income context and income-based voting behavior flips. The right panel of
Figure A-10 presents this result. The poorest voters in poor state house districts register as Republicans
at a rate of only 17 percent while poor voters in richer state house districts register at an average rate
52
of almost 28 percent. In rich state house districts, 51% of voters in the $100,000-plus income bracket
are registered Republicans, but only 35% of the richest voters in the poorest state house districts are
registered Republicans, a number that is almost as low as the percentage of Republicans among poor
voters in the richest districts.17 Thus, there is something about the lower-level contexts in all states that
sets them apart from states, and we would not expect predictions about the income-party relationship
obtained using states to transfer to the lower geographic levels more typically used in scholarship on
geographic context.
Figure A-11: Replication of Figure A-10 using Self-Reported Partisanship and Income Fields
States
10
10
Pct. Republican
20
30
40
Pct. Republican
20
30
40
50
50
60
60
$5K
Self-Reported Income
Poorest States
$170.5K
$5K
Richest States
Self-Reported Income
Poorest Dists.
$170.5K
Richest Dists.
Note: N= 10,095 self-reported weighted Democrats and Republicans in party registration states. Lines are created with
lowess smoothing. Data comes from the 2008 CCES matched to census block via Catalist (Ansolabehere and Hersh 2012).
We can also replicate this result with survey data rather than party registration data. We replicate
Figure A-10 using party identification and self-reported income rather than party registration and block
group income. In making A-11, we divide the sample into state-level and district-level income quartiles. For these quartiles, we use block group income. For the richest and poorest states and the richest
and poorest districts, we then measure the relationship between Republican self-identification and selfreported income. As in Figure A-10, the gap between rich and poor is larger in poor states than rich
states but is bigger in rich districts than poor districts.
17
The thresholds for binning states or state house districts into rich and poor states are based on top and bottom quartiles
of the average of the block-group median income in party registration states.
53
A.10
State-by-State Analysis of Party Registration, Block-Group Income, District Income, and District Percent Black
We provide the equivalent of Figure 2 for each individual state. By doing so, we are displaying conditional means analogous to coefficients from a regression model in which party affiliation is regressed on
block group income (6 categories), district income (3 categories), district race (4 categories), and state of
residence (29 categories), with all independent variables interacted. Though some of these interactions
have missing values, the graphs subdivide the 73 million-person sample into 2,088 subpopulations. In
each graph that follows, the number of districts represented in each line is indicated in parentheses in
red. The number of observations (Democratic and Republican registrants) in each cell is indicated above
each estimate.
A.11
The relationship between local racial composition and income-based voting becomes more obvious in
the map of precinct-level income effects that are allowed to vary by county (Figure A-12). These results
throw into sharper relief the heterogeneity that exists within different racial contexts. In key urbanized
areas of the South, notably in areas such as Atlanta and Miami, the relationship between income and
partisanship is relatively flat, while in areas along the Mississippi River the relationship is steep. This is
of vital importance to claims about income-based voting within states. Were we to assemble a new state
out of the metropolitan areas of the New South such as Atlanta and Nashville, the income-based voting
in these states would look much like states of the Northeast. If we instead constructed a state consisting
of the Black Belt counties of North Carolina, Georgia, Alabama, Tennessee, and Florida, income-based
voting would be much stronger in such states than in the states given to us by history.
The map of 2008 precinct-level results presents a few surprising results that differ from party registration figures. Some of these differences may be an artifact of the 2008 presidential election, but
others suggest explanations related to the different measures. One is that the income-based differences
in partisan registration that appear in registration data in Californias Central Valley do not appear on
this map. This may itself be an artifact of the modifiable areal unit problem (especially since counties
54
in the Western region are typically larger than state house districts), but it may also relate to a partisan
shift towards the Democrats in such counties due to the disproportionate impact of the 2008 foreclosure crisis. The other interesting finding is that across the rural and small-city areas of Pennsylvania,
income-based voting is especially pronounced. The origins of this behavior are unclear. Many of the
Appalachian counties where the marginal effect of income on partisanship was larger than expected are,
for example, counties in which Obama lagged Hillary Clinton in the primaries (Gibson and Gleason,
2012). But most of these deviations pale in comparison to the major polarization that appears on the
map: voters in precincts in counties in the Northeast Corridor from Northern Virginia to New Hampshire
engage in much less income-based voting than we would expect from precinct-level income alone, while
the rural counties of the South and across isolated patches of the countrys midsection engage in much
more income-based voting. Our analysis suggests that the reason for this difference has to do with voting
patterns in areas with blacks, not with areas that are homogeneously white.
A.12
The relationship between local racial composition and income-based voting becomes more obvious in
the map of precinct-level income effects that are allowed to vary by county (Figure A-12). These results
throw into sharper relief the heterogeneity that exists within different racial contexts. In key urbanized
areas of the South, notably in areas such as Atlanta and Miami, the relationship between income and
partisanship is relatively flat, while in areas along the Mississippi River the relationship is steep. This is
of vital importance to claims about income-based voting within states. Were we to assemble a new state
out of the metropolitan areas of the New South such as Atlanta and Nashville, the income-based voting
in these states would look much like states of the Northeast. If we instead constructed a state consisting
of the Black Belt counties of North Carolina, Georgia, Alabama, Tennessee, and Florida, income-based
voting would be much stronger in such states than in the states given to us by history.
The map of 2008 precinct-level results presents a few surprising results that differ from party registration figures. Some of these differences may be an artifact of the 2008 presidential election, but
others suggest explanations related to the different measures. One is that the income-based differences
in partisan registration that appear in registration data in Californias Central Valley do not appear on
55
Figure A-12: Cartogram of Random Effects for Income, Precincts Within Counties
Within-County Random Effect of Precinct Income (Tens of Thousands of Dollars) on McCain Vote
All Counties
Legend
-0.100 - -0.0799
-0.0798 - -0.0598
-0.0597 - -0.0397
-0.0396 - -0.0196
-0.0195 - 0.000558
0.000559 - 0.0207
0.0208 - 0.0408
0.0409 - 0.0609
0.0610 - 0.0811
0.0812 - 0.101
0.102 - 0.121
0.122 - 0.141
No Data
Cartogram of county-specific marginal effect of each $10,000 of average block group median household income in each
precinct on the McCain vote share in the 2008 presidential election. Varying shades of green indicate districts in which the
correlation between block group income and partisan registration are below the national median among state house districts,
while varying shades of red indicate places where this relationship is above the national median.
56
this map. This may itself be an artifact of the modifiable areal unit problem (especially since counties
in the Western region are typically larger than state house districts), but it may also relate to a partisan
shift towards the Democrats in such counties due to the disproportionate impact of the 2008 foreclosure crisis. The other interesting finding is that across the rural and small-city areas of Pennsylvania,
income-based voting is especially pronounced. The origins of this behavior are unclear. Many of the
Appalachian counties where the marginal effect of income on partisanship was larger than expected are,
for example, counties in which Obama lagged Hillary Clinton in the primaries (Gibson and Gleason,
2012). But most of these deviations pale in comparison to the major polarization that appears on the
map: voters in precincts in counties in the Northeast Corridor from Northern Virginia to New Hampshire
engage in much less income-based voting than we would expect from precinct-level income alone, while
the rural counties of the South and across isolated patches of the countrys midsection engage in much
more income-based voting. Our analysis suggests that the reason for this difference has to do with voting
patterns in areas with blacks, not with areas that are homogeneously white.
57
Alaska
State House Districts 5-10% Black
.9
.9
20248
15913
1950
10860
96
4736
204
9(8)
(19)
(3)
2933
7459
305
4047
174
39
1820
2560
14058
3590
1724
745
1694
375
585
364
201
(1)
(5)
17
75
(1)
.2
.1
.1
$0-20K
$100K+
.8
.9
3996
1786
219
30
64
4(2)
(1)
4054
324
59
1272
1976
11
.1
.2
Proportion Republican
.3
.4
.5
.6
.7
Proportion Republican
.3
.4
.5
.6
.7
13
168
71
11043
3132
.2
Proportion Republican
.3
.4
.5
.6
.7
.8
24
$0-20K
$100K+
$0-20K
$100K+
Arizona
State House Districts 0-5% Black
.8
.8
.9
.9
150717
34691
(5)
11703
(12)
3635
896
(11)
478580
30816
96
706
21499
14960
59
(1)
.2
1137
.1
.1
$0-20K
$100K+
Proportion Republican
.3
.4
.5
.6
.7
.8
.9
324
30
(1)
1272
11
1976
.1
.2
Proportion Republican
.3
.4
.5
.6
.7
12828
1101
257994
45931
416863
114678
59410
60044
.2
Proportion Republican
.3
.4
.5
.6
.7
60632
149590
234902
$0-20K
$100K+
$0-20K
$100K+
California
State House Districts 0-5% Black
.8
1210857
244540
15114
59237
301043
725343
1214154
72494
(26)
691090
(23)
932858
577
299
23365
(3)
27816
(2)
45026
478219
43790
8898
42100
$0-20K
$100K+
98074
176542
51685
698
78417
88049
(11)
180
(3)
198653
51701
$0-20K
$100K+
.9
.9
23564
6619
328711
36715
(5)
122482
225605
36781
55177
(2)
Proportion Republican
.3
.4
.5
.6
.7
.8
.8
Proportion Republican
.3
.4
.5
.6
.7
640378
5443
398328
.1
.1
Proportion Republican
.3
.4
.5
.6
.7
1722047
243977
774944
.2
14987
96997
.2
Proportion Republican
.5
.6
.7
.3
.4
.8
.9
.9
54884
$0-20K
105942
48244
.1
.1
6032
.2
.2
72466
$100K+
39885
22291
$0-20K
158319
195704
138321
34482
2325
27052
190
28210
(2)
(3)
309
$100K+
Colorado
State House Districts 0-5% Black
.8
4295
40779
212418
188752
160800
81105
319935
256981
10105
137457
42088
2718
100648
(19)
14890
645(23)
(11)
10086
13848
64929
(4)
332
969
3920
27187
6105
2615
14121
7786
6284
7497
480
4830
329
(1)
(1)
21522
.1
.1
$0-20K
$100K+
$0-20K
$100K+
90
20
8500
74
235
21298
7620
289
18776
1264
(1)
1205
(3)
31440
Proportion Republican
.3
.4
.5
.6
.7
.8
.8
.9
.9
.2
2363
1299
1607
441
8756
18955
19678
9551
4786
4046
(1)
.1
.1
400
(1)
4483
.2
Proportion Republican
.5
.6
.7
.3
.4
476
22891
4091
.2
Proportion Republican
.3
.4
.5
.6
.7
158278
.2
Proportion Republican
.3
.4
.5
.6
.7
.8
.9
.9
$0-20K
$100K+
$0-20K
$100K+
Connecticut
State House Districts 0-5% Black
.8
12525
323
27873
4104
2642
201482
58040
5(30)
(1)
415
12
30
21352
15311
1580
49
.2
698
969
105964
131443
(64)
530
Proportion Republican
.3
.4
.5
.6
.7
116019
114209
6101
220
1083
$0-20K
$100K+
1398
14122
781
3872
8(7)
766
(2)
(10)
4364
15051
$0-20K
$100K+
.9
.9
1547
10639
580
(1)
156
767
697
(11)
46
(9)
$0-20K
$100K+
208
4737
.2
37203
12983
64
.1
500
28023
38382
2758
13145
958
Proportion Republican
.3 .4 .5 .6 .7
.8
.8
Proportion Republican
.3 .4 .5 .6 .7
2742
.2
.1
4877
35521
10713
.1
.1
.2
Proportion Republican
.3
.4
.5
.6
.7
.8
.9
.9
2021
17899
$0-20K
18805
34746
24827
8545
1599
(7)
1604
281
298
(8)
$100K+
Delaware
State House Districts 0-5% Black
.8
(3)
6976
19
5506
1454
2183
20658
447
$0-20K
15278
221
(5)
125
(5)
3167
91
$100K+
$0-20K
$100K+
764
32030
44071
29490
369
4137
62426
9877
2512
4694
21973
19247
1270
42
(3)
(7)
1623
(11)
1218
33
54
(2)
(5)
20096
.2
542
104
Proportion Republican
.3
.4
.5
.6
.7
.8
.8
.9
.9
.2
Proportion Republican
.3
.4
.5
.6
.7
38996
10914
2489
24306
.1
.1
.2
Proportion Republican
.3
.4
.5
.6
.7
11597
6544
.2
Proportion Republican
.3
.4
.5
.6
.7
.8
.9
.9
.1
.1
62
$0-20K
$100K+
277
2541
$0-20K
25829
2591
2908
500
59
14091
$100K+
Florida
State House Districts 0-5% Black
.8
691552
394938
40650
475
3275
69394
321445
14365
84183
55571
74910
753
(28)
29823
(7)
(3)
18563
76779
481669
12342
746015
231529
1014182
57439
376418
40818
40908
92717
7942
44117
1905
(29)
(2)
2679
(10)
25857
28683
767
.1
.1
.2
Proportion Republican
.3
.4
.5
.6
.7
899036
158476
2678
71353
.2
Proportion Republican
.5
.6
.7
.3
.4
.8
.9
.9
$0-20K
$100K+
$0-20K
$100K+
180196
402012
25778
180703
4630
45092
1167
(11)
10955
(12)
557215
281854
Proportion Republican
.3
.4
.5
.6
.7
Proportion Republican
.3
.4
.5
.6
.7
.8
.8
.9
.9
2035
47388
(18)
272101
32193
24137
692314
.2
.2
9265
.1
.1
142996
$0-20K
$100K+
$0-20K
$100K+
Iowa
State House Districts 0-5% Black
.8
170610
35952
288380
17965
16748
5807
1138
(6)
(38)
1140
472
(46)
12449
1953
.2
9437
2576
1008
17909
21164
2780
23302
9196
36
513
(3)
(4)
71
1264
.1
.1
$0-20K
$100K+
Proportion Republican
.3
.4
.5
.6
.7
.8
.9
49
247
7457
4943
33
(3)
19902
.1
.2
Proportion Republican
.3
.4
.5
.6
.7
339
367522
152992
7862
33498
72455
11203
.2
Proportion Republican
.7
.3
.4
.5
.6
.8
.9
.9
$0-20K
$100K+
$0-20K
$100K+
Kansas
State House Districts 0-5% Black
1175
422
7286
87606
82181
173770
55731
102666
58345
5281
555
(21)
670
347
(33)
(41)
14128
2005
.8
(2)
12663
47
498
11048
99
(5)
1875
179
14099
40852
109
3528
(9)
.1
.1
$0-20K
$100K+
$0-20K
8814
20292
809
7385
17157
271
68
37
(4)
(5)
1431
66
10569
23815
.1
.2
.1
$0-20K
(5)
.2
3099
$100K+
33
Proportion Republican
.3
.4
.5
.6
.7
.9
169
5162
.8
Proportion Republican
.3
.4
.5
.6
.7
1489
10868
5794
24771
71
657
.2
Proportion Republican
.3
.4
.5
.6
.7
286329
38175
5153
50420
.2
Proportion Republican
.3
.4
.5
.6
.7
.8
.9
.9
$100K+
3669
$0-20K
$100K+
Kentucky
State House Districts 0-5% Black
.8
66814
202888
340
16606
807634
140134
61290
245088
146738
19691
13735
60878
9075
1702
2712
(5)
3193
266
(16)
(44)
168
82791
13754
3823
$100K+
64590
175451
$0-20K
12344
37980
10384
1016
(12)
74992
$100K+
.9
.9
2166
143
55251
1140
35040
52037
10987
4474
56
(4)
441
(4)
26031
.1
.1
.2
6155
Proportion Republican
.3
.4
.5
.6
.7
.8
.8
Proportion Republican
.3
.4
.5
.6
.7
19251
14094
3543
(2)
225(7)
.1
.1
$0-20K
.2
3679
23259
6158
.2
.2
2328
Proportion Republican
.3
.4
.5
.6
.7
Proportion Republican
.3
.4
.5
.6
.7
.8
.9
.9
$0-20K
$100K+
698
2069
(1)
1447
9082
685
16141
635
350
(5)
65834
3723
19358
$0-20K
13612
$100K+
Louisiana
State House Districts 0-5% Black
.9
.9
.8
.8
17
84
1885
54
(3)
93
26788
5795
11051
19
21257
75667
14512
5730
81
48808
7517
165
17107
379
1690
.1
.1
$0-20K
$100K+
$0-20K
$100K+
11245
4043
125270
104039
4673
99145
343462
3807
.2
31767
827
10805
Proportion Republican
.3
.4
.5
.6
.7
.8
Proportion Republican
.7
.3
.4
.5
.6
8279
64063
7479
771(1)
9851
(24)
(16)
.8
.9
.9
1747
23410
120682
4296
55711
546573
10058
29799
1895
150
2205
(42)
50
(5)
.1
179534
.1
.2
21745
27076
6616
32
190
(2)
1749
(8)
(3)
.2
Proportion Republican
.3
.4
.5
.6
.7
4361
27565
(1)
.2
Proportion Republican
.5
.6
.7
.3
.4
$0-20K
$100K+
$0-20K
$100K+
Massachusetts
State House Districts 0-5% Black
16712
13717
4801
481
3212
96812
(63)
(56)
235
Proportion Republican
.3
.4
.5
.6
.7
159466
383311
32829
129340
21164
.2
1705
31672
168459
54025
247961
159803
(10)
.1
.1
.2
Proportion Republican
.3
.4
.5
.6
.7
.8
.8
.9
.9
$0-20K
$100K+
287
4803
6896
30813
30504
4154
$0-20K
19108
59637
9867
8745
2534
17343
4600
171
3154
2577
1365
62(2)
(9)
(5)
$100K+
$0-20K
5681
145
20440
93
10363
4706
22
3811
2606
1634
1417
5
(5)
(1)
.2
6440
5604
1068
30718
4950
.1
.2
28
.1
Proportion Republican
.3
.4
.5
.6
.7
Proportion Republican
.3 .4
.5 .6 .7
.8
.8
.9
.9
$100K+
558
10217
$0-20K
11710
45916
15686
17469
1412
2013
492
232
796
128
(2)
(5)
$100K+
Maryland
State House Districts 0-5% Black
.8
2809
570
12827
5424
16831
65110
68415
237
36728
117219
85
49001
1817
(4)
51
(3)
761
42364
(8)
15
(1)
14252
6579
1027
665
1830
9996
33630
17333
93021
58033
270
81658
56857
79381
46
2805
(3)
78250
(6)
.1
.1
.2
Proportion Republican
.3
.4
.5
.6
.7
42769
165
.2
Proportion Republican
.5
.6
.7
.3
.4
.8
.9
.9
$0-20K
$100K+
$0-20K
$100K+
1452
104365
238091
95800
856
92347
87991
(10)
(10)
.2
.1
$0-20K
11428
62088
.1
22982
189612
117480
17004
Proportion Republican
.3
.4
.5
.6
.7
Proportion Republican
.3
.4
.5
.6
.7
1705
.2
.8
.8
.9
.9
$100K+
675
4351
51805
$0-20K
24429
179664
92695
78698
171276
144668
79449
5829
89685
26211
1737
(7)
6837
(7)
12988
(6)
$100K+
.8
Proportion Republican
.3
.4
.5
.6
.7
Proportion Republican
.3
.4
.5
.6
.7
.9
.9
1019
256637
50801
154995
4507
65086
11565
16393
2379
5182
234
1039
210(56)
(5)
1416
451
(90)
104(23)
521(21)
5974
386
(59)
161
641
1269
.1
.1
$0-20K
$100K+
$0-20K
$100K+
7736
43241
1810
230
121204
26985
111276
11573
23075
10121
74468
18299
989(5)
7736
571(15)
1688
(49)
424579
37067
800
48
5762
11
4
(1)
1636
.1
.1
.2
Proportion Republican
.3
.4
.5
.6
.7
.8
.8
.9
.9
.2
Proportion Republican
.3
.4
.5
.6
.7
1559
9525
23976
.2
.2
1152
10376
26
6889
45108
51689
25153
176515
36452
100317
41936
$0-20K
$100K+
$0-20K
$100K+
Nebraska
State House Districts 0-5% Black
.8
80265
119492
54078
29479
8213
746
2241
261
21380
(8)
4422
(14)
718
(21)
5372
7561
1629
7771
18441
3270
38
226
210
73
35
6833
(2)
(1)
135
.1
.1
.2
Proportion Republican
.3
.4
.5
.6
.7
277199
100614
18477
73664
32208
4150
.2
Proportion Republican
.3
.4
.5
.6
.7
.8
.9
.9
$0-20K
$100K+
$0-20K
$100K+
.8
Proportion Republican
.3
.4
.5
.6
.7
1554
98
4305
.2
9683
(1)
.1
122
$0-20K
$100K+
2110
34
6541
(1)
7674
.2
253
681
240
.1
.8
Proportion Republican
.3
.4
.5
.6
.7
.9
.9
4296
$0-20K
10126
22
1186
100
11
(1)
$100K+
New Jersey
State House Districts 0-5% Black
.9
.8
178627
(11)
3370
132535
64456
21772
15395
142782
1075
193356
(5)
108654
Proportion Republican
.3
.4
.5
.6
.7
.9
273687
3998
36664
36463
54818
4412
32880
(5)
273
(2)
.1
$100K+
$0-20K
$100K+
5059
219667
62516
(5)
88807
14753
$0-20K
11511
15398
23677
2255
15605
12597
71124
.2
27957
108835
(7)
58957
.1
4663
122943
92824
23663
Proportion Republican
.3 .4 .5 .6 .7
.8
.9
.9
.8
Proportion Republican
.3 .4
.5 .6
.7
.2
120983
31561
.2
6393
$0-20K
.1
60980
121264
1037
.1
.2
Proportion Republican
.5
.6
.7
.3
.4
.8
$100K+
11351
2781
13472
$0-20K
11018
26420
77302
12496
504
9251
(1)
8627
(3)
74
(1)
$100K+
Nevada
State House Districts 0-5% Black
.8
1695
32738
14095
6766
162724
117982
48802
169993
1895
9110
48492
168
78233
(8)
697
59
(15)
(2)
1878
3775
584
7148
2939
33151
$0-20K
$100K+
2353
1919
33352
11574
18593
$0-20K
$100K+
.9
.9
1215
61
(3)
16693
31273
6224
108
2623
14
564
12
1046
(1)
300
2141
9829
3331
416
(1)
2382
8906
8930
.1
.1
30
13640
.2
238
Proportion Republican
.3
.4
.5
.6
.7
.8
.8
Proportion Republican
.3
.4
.5
.6
.7
(3)
167
66427
.2
40(8)
.1
.1
.2
Proportion Republican
.3
.4
.5
.6
.7
15680
.2
Proportion Republican
.7
.3
.4
.5
.6
.8
.9
.9
$0-20K
$100K+
$0-20K
$100K+
New York
State House Districts 0-5% Black
.8
6876
54526
408616
300773
773987
282173
258345
600131
28215
55141
278199
12123
3249
204354
(27)
(23)
1550
34155
17423
(21)
74623
125869
245300
6806
46820
28984
$0-20K
$100K+
104407
109322
18727
(10)
5037
3155
4696
(11)
(7)
$100K+
37205
92237
$0-20K
256169
79574
134570
233231
31665
24514
2838
(3)
8497
870(11)
(9)
2846
11663
.2
14131
143740
32740
17006
377983
.1
5558
56636
22335
Proportion Republican
.3 .4 .5 .6 .7
.8
.9
.9
.8
37704
.2
Proportion Republican
.3
.4
.5
.6
.7
163071
193417
41209
$0-20K
.1
12571
212627
131301
.1
.1
.2
Proportion Republican
.3
.4
.5
.6
.7
682923
.2
Proportion Republican
.3
.4
.5
.6
.7
.8
.9
.9
$100K+
$0-20K
76327
612944
187518
163191
83721
2743
3438
(7)
(21)
19030
$100K+
North Carolina
State House Districts 0-5% Black
.8
558
190
32516
14957
259(2)
1015
(11)
(3)
5643
168
3325
2727
160090
139576
15715
$0-20K
$100K+
49959
7770
297
16773
1487
(5)
(11)
121
(7)
$0-20K
$100K+
17675
146035
79100
385061
303
120520
79963
10533
814
36765
33908
(4)
329
5150
(14)
(21)
6918
308508
272780
Proportion Republican
.3
.4
.5
.6
.7
.8
.9
.9
.8
Proportion Republican
.3
.4
.5
.6
.7
52900
67350
1740
496
19030
233751
$0-20K
$100K+
19720
2673
252311
.2
.1
.1
12412
68561
4271
923
(14)
(28)
655163
200184
16084
.2
79726
217109
96045
.1
.1
Proportion Republican
.3
.4
.5
.6
.7
2572
28748
633
.2
145
54604
304109
3674
56117
78266
14816
.2
Proportion Republican
.3
.4
.5
.6
.7
.8
.9
.9
80734
14563
$0-20K
$100K+
Oklahoma
State House Districts 0-5% Black
.9
.9
22401
182
5806
168521
81450
25132
65331
10370
(16)
449
105733
(39)
4410
41614
490956
.8
(3)
3616
11921
Proportion Republican
.3
.4
.5
.6
.7
Proportion Republican
.5
.6
.7
.3
.4
.8
3649
11609
16661
2287
32224
65063
58371
66231
3430
618
4161
(9)
197
(19)
2316
190797
.1
.1
.2
.2
19688
$0-20K
$100K+
$0-20K
$100K+
(3)
6406
26910
1832
18840
592
24486
392
196
(8)
70572
Proportion Republican
.3
.4
.5
.6
.7
800
.8
.9
.9
.8
2143
Proportion Republican
.3
.4
.5
.6
.7
1203
54
(4)
1931
7850
.1
.1
.2
.2
14578
$0-20K
$100K+
37008
11328
$0-20K
$100K+
Oregon
State House Districts 0-5% Black
2140
391837
265726
180110
471410
57832
58099
147943
17785
29145
27235
1215
14719
14420
500(5)
(22)
5741
16245
.1
.1
$0-20K
$100K+
.1
.2
Proportion Republican
.3
.4 .5 .6 .7
.8
.9
2001
$0-20K
837
.2
(31)
Proportion Republican
.3
.4
.5
.6
.7
Proportion Republican
.3
.4
.5
.6
.7
11592
.2
.8
.8
.9
.9
19651
10763
790
1994
83
(1)
$100K+
15633
11880
955
$0-20K
100
40
(1)
$100K+
Pennsylvania
State House Districts 0-5% Black
.8
1052
44839
1282365
516653
1128388
279567
392343
270923
326485
41595
31767
191063
2044
118181
5864
(25)
(56)
1047
(55)
81705
15030
Proportion Republican
.3
.4
.5
.6
.7
Proportion Republican
.3
.4
.5
.6
.7
.8
.9
.9
15363
123075
$0-20K
11065
95006
10184
608
20599
(5)
373
6098
(11)
(13)
218370
8699
29387
$100K+
$0-20K
$100K+
83368
57185
39171
72590
6695
45048
13990
16414
4576
4347
(3)
2415
1513
(5)
8479
223954
(11)
$0-20K
90886
169960
145
.1
450
3928
68016
Proportion Republican
.3 .4
.5 .6 .7
21628
.2
Proportion Republican
.3
.4
.5
.6
.7
.8
.8
.9
.9
.1
.2
229064
35620
.1
.1
.2
.2
622
54857
57996
81708
$100K+
$0-20K
382116
11536
8327
3780
22150
1132
1997
671
(1)
2966
(18)
$100K+
Rhode Island
State House Districts 0-5% Black
4096
3062
48414
18679
105799
556
(9)
13
1177
(8)
(42)
72
10024
.1
.1
14623
2080
5448
5798
101
Proportion Republican
.3
.4
.5
.6
.7
232
17796
32707
.2
3300
.2
Proportion Republican
.3
.4
.5
.6
.7
.8
.8
.9
.9
$0-20K
$100K+
5123
$0-20K
1615
(9)
319
$100K+
Proportion Republican
.3 .4 .5 .6 .7
.8
.8
.9
.9
2925
6182
746
$0-20K
13805
725
2255
267
898
1269
$100K+
25
.2
19
374(5)
(1)
.1
Proportion Republican
.3 .4 .5 .6 .7
.2
51
.1
30147
6721
1636
$0-20K
3058
444
0
(1)
$100K+
2859
420
314
157(14)
(23)
1949
6789
86902
77097
5332
12938
8295
30891
618
2362
10577
(41)
240(16)
(18)
.2
$100K+
6
(2)
.9
3772
576
184
213512
45747
42141
.1
.1
$0-20K
16
72
20893
75051
968
21963
46540
981
4843
8994
3166
91
503
78
31
(35)
46
(23)
2705
.1
.2
.3
Proportion Republican
.4 .5 .6 .7 .8
103081
75504
.8
39863
95729
67038
219425
4094
Proportion Republican
.3
.4
.5
.6
.7
37987
.2
Proportion Republican
.3
.4
.5
.6
.7
.8
.9
.9
$0-20K
$100K+
$0-20K
$100K+
West Virginia
State House Districts 0-5% Black
.8
186
668
79669
546
41899
71
(5)
289
(42)
547145
.2
1530
8831
32293
2027
4909
125
3684
49
107
(5)
(3)
45805
94138
21984
.1
.1
$0-20K
$100K+
Proportion Republican
.3
.4
.5
.6
.7
.8
.9
13
(3)
1825
104
48
14701
4402
.1
.2
Proportion Republican
.3
.4
.5
.6
.7
21316
35893
5793
8851
.2
Proportion Republican
.5
.6
.7
.3
.4
.8
.9
.9
$0-20K
$100K+
$0-20K
$100K+