Article
Voter Ideology,
Economic Factors,
and State and Local
Tax Progressivity
Public Finance Review
41(2) 177-202
ª The Author(s) 2013
Reprints and permission:
sagepub.com/journalsPermissions.nav
DOI: 10.1177/1091142112467528
pfr.sagepub.com
John M. Foster1
Abstract
This study examines the relationship between voter ideology and the
distribution of tax burdens across income groups using state and local data,
aggregated at the state level, for 1995, 2002, and 2007. I find that average voter
liberalism is positively related to subnational tax progressivity. However, the
effects are economically insignificant. A state’s ethnic demographic context
appears to be more important. The ethnic congruence between the poor and
the nonpoor is positively related to progressivity and the effects are
economically significant. The tension between ethnic groups, measured with
an index of residential segregation, is inversely related to progressivity. The
effects are larger in magnitude than those of average voter liberalism. It is
possible that the ethnic demographic context reflects aspects of voter preferences that are not captured by measures of voter ideology.
1
James W. Martin School of Public Policy and Administration, University of Kentucky,
Lexington, KY, USA
Corresponding Author:
John M. Foster, James W. Martin School of Public Policy and Administration, University of
Kentucky, 405 Patterson Office Tower, Lexington, KY 40506, USA.
Email: [email protected]
Downloaded from pfr.sagepub.com at PENNSYLVANIA STATE UNIV on February 19, 2016
178
Public Finance Review 41(2)
Keywords
ideology, tax progressivity, fiscal federalism
State and local taxes are important in magnitude. In 2007, state and local tax
revenue constituted approximately 11 percent of total personal income. Many
scholars, policy makers, and commentators argue that combined federal,
state, and local tax burdens should be allocated across households according
to their ability to pay or that tax systems should be ‘‘progressive.’’ Conventional public finance theory posits that redistribution should be carried out
primarily by the federal government for two reasons: first, there may be
national spillovers in the benefits for altruistic voters; second, redistribution
carried out at the state or local level could prompt the in-migration of poor
households and out-migration of the rich.1 While the federal system is
generally recognized as progressive, a few states have complemented the
progressive federal income tax with progressive tax systems of their own.
On average, across all states, however, low-income households pay a larger
share of their income in state and local taxes than do wealthy households. In
other words, state and local tax systems tend to be regressive, although there
is significant variation in the degree of regressivity. One way to measure
progressivity is to take the ratio between the average tax rate faced by a
state’s top income quintile and that faced by its bottom quintile. According
to this measure, the most progressive state and local tax system (New York)
was more than three times as progressive as the least (Washington) in 2007.
Differences in the tax policies adopted by governments reflect
differences in priorities and economic circumstances. Policy makers have
to compete for office. State and local tax policy, therefore, is likely to be
shaped by voter preferences, among other factors. These tax policy
preferences of voters may be shaped by their narrow financial interests and
their conceptions of economic justice. The latter are likely to be grounded in
established ideologies. Past studies on state and local tax progressivity have
used measures of voter ideology as an explanatory variable, but the
measures utilized either did not account for change in the electorate’s
ideological orientation over time or failed to account for differences in the
intensity of an electorate’s liberalism, along with other limitations.
The models estimated in this study used the measure of the average
voter’s liberalism developed by Berry et al. (1998). The analysis utilizes
state and local incidence data on the forty-eight contiguous states for
1995, 2002, and 2007. I find that average voter liberalism is positively
Downloaded from pfr.sagepub.com at PENNSYLVANIA STATE UNIV on February 19, 2016
Foster
179
related to state and local tax progressivity. However, the impact is small in
magnitude. A state’s ethnic demographic context has statistically and
economically significant effects on the progressivity of the tax structures
utilized by state and local governments. Holding other relevant factors
constant, state and local tax structures appear to be more progressive in
states in which the poor and nonpoor tend to be of the same ethnicity. The
tension between ethnic groups, measured with an index of ethnic residential
segregation, is inversely related to progressivity. The effects of these
variables are considerably larger than that of the average voter’s liberalism
even with the latter variable held constant. This suggests that the ethnic
congruence between the poor and the nonpoor and the amiability of
relations between ethnic groups reflect components of voter preferences
that may not be captured by measures of general voter ideology. This is the
first study to examine the association between ethnic residential segregation
and state and local tax progressivity.
State and Local Tax Structure and Progressivity
The progressivity of a government’s tax system is determined by its mix of
tax instruments and its rates assigned to each base. The burden of some
taxes, such as the personal income tax, can be adjusted based on the
characteristics of the taxpayer and, thereby, can be made progressive. On
the other hand, it is difficult to adjust the burdens imposed by general sales
and excise taxes so these taxes tend to be regressive.2 Property taxes could
perhaps be made progressive, but they tend to be regressive in practice. Of
course, lifetime incidence is different from incidence in terms of annual
income. Metcalf (1994) finds that the regressivity of consumption taxes and
the progressivity of income taxes are lower from the standpoint of lifetime
incidence than when measured annually. The estimates of incidence utilized
in this article are based on the latter.3
In 2007, state and local governments derived about 23 percent of their
tax revenue from the personal income tax, 34 percent from sales and excise
taxes, and 30 percent from property taxes. Local governments, alone,
obtained around 72 percent of their tax revenue from property taxes.
Because of the high degree of reliance on sales and excise taxes by most
state governments and the high degree of reliance on property taxes by most
local governments, state and local tax systems tend to be regressive on
average. In 2007, the average tax rate paid by the top-income quintile was
around 74 percent of that borne by the bottom quintile, on average. There is,
Downloaded from pfr.sagepub.com at PENNSYLVANIA STATE UNIV on February 19, 2016
180
Public Finance Review 41(2)
however, substantial variation across states. The standard deviation (SD) of
this percentage is equal to about 30 percent of the mean.
The progressivity of state and local tax structures in terms of annual
income is determined by reliance on income taxes relative to consumption
and property taxes as well as the structure of the income tax that is utilized.
The states with the most progressive tax systems are those which rely heavily
on the personal income tax. Those with the most progressive income tax
systems have a relatively large number of tax brackets, impose relatively high
marginal tax rates on the top income bracket, and utilize low-income tax
credits. The exemption of food from the sales tax tends to reduce the regressivity of the sales tax (Institute on Taxation and Economic Policy 2009).
Conceptual Framework and Previous Empirical
Research
Some voters may view the adoption of more progressive tax policies as a
means to promote economic justice while others may view it as punishment
for hard work and ingenuity. When elected representatives design a tax
structure, they have to take conflicting interests into account. How do the
representatives weight these different interests? In the United States,
politicians have to compete for office. Thus, the determination of tax policy
should be examined in the context of electoral competition.
Hettich and Winer (1999) developed a theoretical model in which the
process of electoral competition forces political agents to take the
preferences of the electorate into account when designing the tax structure.
In their model, citizens vote for the platform that gives them the highest
expected utility. Candidates running for office choose the fiscal platform
that maximizes expected voter support. The weight that a particular
citizen’s welfare receives can depend on the responsiveness of that citizen’s
voting behavior to fiscal policy, whether he or she is a member of an
organized interest group, and on the size of that group.
The politically optimal tax structure may depend on voters’ financial
interests and on their conceptions of economic justice. Analyses of survey
data on support for redistribution by government conducted by Fong
(2001) and Alesina and La Ferrara (2005) suggest that respondents’
preferences for redistribution are influenced not just by their economic and
demographic characteristics but also by the extent to which they believe that
the economic system is ‘‘fair.’’ Respondents who believed that society
provides equal opportunities to all and that economic well-being is mainly
determined by one’s own effort and ability tended to be less supportive of
Downloaded from pfr.sagepub.com at PENNSYLVANIA STATE UNIV on February 19, 2016
Foster
181
redistribution, ceteris paribus. Emphasis on the role of personal effort and
ability in the determination of one’s position in the income distribution tends
to be associated with conservative views while belief in the dominance of
luck and social connections tends to be associated with more liberal
perspectives. Fong finds that the marginal effects of respondents’ beliefs
concerning economic justice on support for redistribution are large in magnitude, even with respondents’ economic and demographic characteristics held
constant. The marginal effects of those beliefs are twice as large as those of
income. Voters’ conceptions of economic justice are likely to be grounded in
established ideologies.
Political scientists and economists previously have examined the
relationship between state and local tax progressivity and measures of the
average voter’s ideology. At least four measures of the average voter’s
ideology have been used. Erikson, Wright, and McIver (1993) pool survey
data across numerous years to obtain state-representative samples and
measure average citizen liberalism in each state with the difference between
the percentages of respondents who identified themselves as ‘‘liberal’’ and
those who identified as ‘‘conservative.’’ They find that this measure is
statistically significant and positively related to state and local tax
progressivity. There are three significant limitations to the validity of the
Erikson, Wright, and McIver measure of average voter liberalism. Because
the authors obtained representative state samples by pooling many years of
surveys, the measure does not change over time. A second limitation of the
measure is that it does not capture differences in the intensity of a
respondent’s liberalism, as the response was only ‘‘liberal,’’ ‘‘moderate,’’
or ‘‘conservative.’’ Another potential problem is that ideological
self-identification may not map consistently into policy preferences.
According to Stimson (1991, 61), ‘‘the symbols, personalities, ideas, and
images underlying ‘liberal’ and ‘conservative’ include much content that is
other than prescriptions for public policy.’’ Thus, the Erikson, Wright, and
McIver measure may be a weak measure of operational ideology or a weak
measure of the kinds of policies that citizens prefer.
Chernick (1992) measures average voter liberalism with the average
American Federation of Labor-Congress of Industrial Organizations
(AFL-CIO) rating of state assemblies. An important limitation of the
measure used by Chernick is that it only gauges the ideological orientation
of the citizens who voted for the winning candidates. He finds that this
measure is positively related to state and local tax progressivity. Morgan
(1995) uses Holbrook-Provow and Poe’s (1987) measure of general voter
conservatism. The measure is the average Conservative Coalition score for
Downloaded from pfr.sagepub.com at PENNSYLVANIA STATE UNIV on February 19, 2016
182
Public Finance Review 41(2)
the state’s congressional delegation. Morgan finds that the ideology
measure is inversely related to state and local tax progressivity. Like the
measure used by Chernick (1992), the Holbrook-Provow and Poe (1987)
measure does not take the preferences of those who voted for the opposition
into account.
Fletcher and Murray (2008) measure average voter liberalism with the
share of votes received by Al Gore in 2000. This measure does not capture
differences in the intensity of liberal ideology among different electorates.
The authors use the United Way’s State of Caring Index as a proxy for
citizens’ tastes for redistribution. The constructive validity of this measure is limited because government redistributive policies and private
charity may be substitutes. They find that both variables are statistically
insignificant.
The analysis that follows uses a measure of average voter liberalism that
was developed by Berry et al. (1998). The measure is intended to gauge differences in the intensity of voter liberalism across states, capture changes over
time, and summarize the ideological orientation of the entire electorate.
Empirical Approach
The analysis utilizes data on the combined state and local tax burdens for
each income quintile in each state provided by Citizens for Tax Justice and
the Institute on Taxation and Economic Policy (1996) and the Institute on
Taxation and Economic Policy (2003, 2009). All three studies used the
same methodology. The income quintiles are specific to each state. The
authors utilized data on a stratified random sample that includes all
household types and is statistically valid at the state level. The studies
present the average tax rates (i.e., the percentages of income paid to
taxes) for each state-specific income quintile for 1995, 2002, and
2007.4 The average tax rates consist of burdens imposed by personal
and corporate income, general sales, property, cigarette, gasoline, and
beer taxes. The studies allow portions of certain tax burdens to be
exported to nonresidents. None of the studies takes tax importing into
account. Corporate income, property, and retail sales taxes levied on
firms could be passed forward to in-state or out-of-state consumers or
passed backward to in-state workers and/or in-state or out-of-state capital owners depending on the tax instrument and on the characteristics
of the industries affected by the tax. The methodology is described in
detail in Appendix B.
The empirical model can be summarized as follows:
Downloaded from pfr.sagepub.com at PENNSYLVANIA STATE UNIV on February 19, 2016
Foster
183
Pit ¼ b0 þ b1 CLit þ b2 GLit þ b3 Xit þ b4 Eit þ b5 WPjt þ Fi þ Yt þ eit ;
ð1Þ
where i and t are subscripts for the state and year, respectively. The subscript j
denotes neighboring states. Pit refers to one of three measures of state and local
tax progressivity: the top quintile’s average tax rate as a percentage of that
faced by the bottom quintile (Top-to-Bottom %), the average tax rate faced
by the top quintile as a percentage of that faced by the middle quintile
(Top-to-Middle %), and the average tax rate faced by the middle quintile as
a percentage of that faced by the bottom quintile (Middle-to-Bottom %). CLit
and GLit denote the average voter’s and the government’s liberalism, respectively. The vector Xit consists of relevant state demographics. The vector Eit
contains measures of potential state tax exporting capacities. WPjt is the
weighted average progressivity measure of neighboring states or the spatial lag
of the dependent variable. Fi and Yt denote state- and year-fixed effects, respectively. The random error is denoted by eit.
I use a measure of mean voter ideology that was developed by Berry et al.
(1998; hereafter referred to as the BRFH measure). This measure has not been
used in previous studies of state and local tax progressivity. The authors use the
average of ratings of incumbent Congress members in each district by the
Americans for Democratic Action (ADA) and the AFL-CIO’s Committee
on Political Education (COPE) to quantify ideology. They assume that the
ideological orientation of state legislators is the same as that of their congressional counterparts. Additionally, they assume that a challenger’s ideology
score is equal to the average score of all incumbents in the state from the same
party. The authors assume that the ideological orientations of the candidates
are approximations of those of the voters who support them. Each electoral district’s ideology score is the weighted average party ideology where the weights
are each party’s vote share. From the district scores, the authors calculate a
state-level unweighted average. A higher score denotes greater liberalism.
Since it is based on congressional votes, the BRFH measure is
potentially a more valid measure of the average voter’s operational
ideology than are alternative measures that have been previously used in
this literature. The strength of the measure, however, is limited by the extent
to which political parties deviate from the platforms desired by voters. If
voters merely choose the ‘‘lesser evil,’’ then the BRFH measure is a limited
measure of voter ideology. Another potential limitation of the measure is
that it assigns equal weight to each voter. It is likely to be the case that
political influence varies. Schlozman, Verba, and Brady (1999) find that the
wealthy are significantly more likely than the less well off to donate dollars
Downloaded from pfr.sagepub.com at PENNSYLVANIA STATE UNIV on February 19, 2016
184
Public Finance Review 41(2)
and time to campaigns and participate in protests. Thus, the weight that a
voter’s utility receives may be directly related to his or her income.
Developing a state-level, income-weighted, average voter ideology measure
presupposes the availability of a survey data set, with a state-representative
sample, that reports respondents’ views of the economic system, policy
preferences, and demographic characteristics. The surveys utilized by Fong
(2001) and Alesina and La Ferrara (2005) contain the relevant information,
but they are representative only at the national level. The sample used by
Erikson, Wright, and McIver (1993) is representative at the state level. The
limitations of their ideology measure are described in the third section.
If the ideologies of the political parties are not identical, then majority
rule will yield a government with an ideological orientation that may differ
from that of the average voter. A particular party’s share of political power
increases if it wins the governorship. For these reasons, the model includes
the BRFH measure of government liberalism in addition to the measure of
citizen liberalism. To calculate their measure of government liberalism,
Berry and coauthors first assume that the minority party’s maximum share
of power is 40 percent while the majority’s minimum share is 60 percent. If
the majority’s share of power exceeds 60 percent, then it is assumed to have
complete control. The governor and the legislature are assumed to have
equal power. The authors assume that the ideological orientation of state
officials (including the governor) is the same as that of the state’s
Congressional delegation, as measured by the ADA and COPE. The authors
estimate ‘‘government liberalism’’ by taking the weighted average of the
ideological scores of both houses of the legislature, where the weights are
each party’s effective power, and adding the governor’s ideology score.
A higher value indicates a greater degree of liberalism. The BRFH
government liberalism measure does not necessarily reflect the ideological
orientation of the individual state legislators. Thus, the validity of the measure depends on the degree of ideological homogeneity within each party.5
Because government liberalism is a function of average voter liberalism, the
models include only the component of government liberalism that is not
explained by average voter liberalism. More specifically, I regressed
government liberalism over average voter liberalism, took the residual, and
include the residual in the models instead of the raw government liberalism
measure.6 I expect both average voter liberalism and the government
liberalism residual to be positively related to state and local tax
progressivity.
The models include measures of state demographic characteristics that
may influence voter demand for tax progressivity. The politically optimal
Downloaded from pfr.sagepub.com at PENNSYLVANIA STATE UNIV on February 19, 2016
Foster
185
level of redistribution may be lower in states that are ethnically heterogeneous. Research conducted by social psychologists, political scientists, and
economists indicates that people tend to be more generous toward others
who are more similar to them ethnically, racially, linguistically, and so
on.7 These results may stem from ethnic or racial bias in the perceived
worthiness of the recipients of assistance. The nonpoor may tend to attribute
the misfortunes of poor people who are ethnically, racially, or linguistically
different from them to moral weakness while attributing poverty among
members of their own groups to bad luck. Fong and Luttmer (2011) find that
nonblack potential donors tend to view recipients as being less worthy of aid
when they believe that the recipients are predominantly black. Ethnic
heterogeneity may also influence the demand for redistribution on the
revenue side. One reason is that the nonpoor may be more concerned about
reducing the average tax rates faced by the poor when the two groups tend to
be of the same ethnicity. Additionally, the nonpoor may be willing to pay
higher tax rates to support redistributive spending if the beneficiaries tend
to be members of their ethnic groups. Hero (1998) examines the relationship
between state ethnic heterogeneity and state and local tax progressivity and
does not find a significant relationship. However, the author does not
control for state tax exporting capacities or spatial dependence. The ethnic
similarity between the poor and the nonpoor is the dimension of ethnic
heterogeneity that is likely to be the most relevant for the political economy
of tax progressivity. Thus, I attempt to measure the ethnic congruence
between the poor and the nonpoor directly by calculating the probability that
two randomly drawn individuals, one from the poor population and the other
from the nonpoor population, are of the same ethnicity (Ethnic Congruence).
If we denote the proportion of the nonpoor who are in ethnic group k as nonpoork and the proportion of the poor who are from group k as poork, then
X
nonpoork poork ;
Ethnic congruence ¼
8k
where k ¼ Hispanics, non-Hispanic whites, non-Hispanic African Americans, other non-Hispanic.
The ‘‘nonpoor’’ group consists of individuals with household incomes
above the state median while the ‘‘poor’’ group consists of individuals with
household incomes in the bottom fifth of the state’s income distribution.
The ethnic congruence measure does not account for variation in the
amiability of relations between ethnic groups. If greater diversity leads to
greater contact between individuals of different ethnic groups, then it may
Downloaded from pfr.sagepub.com at PENNSYLVANIA STATE UNIV on February 19, 2016
186
Public Finance Review 41(2)
actually increase support for redistributive fiscal policies. Roch and Rushton (2008) examine the effects of racial diversity and residential segregation
on white voter support for an Alabama state referendum that would have
substantially increased the progressivity of the state’s tax structure. They
find that white support for the measure at the county level is significantly
and negatively affected by the degree of segregation but insignificantly
affected by the African American share of county population. Their results
suggest that it is the degree of tension between racial groups—and not racial
diversity itself—that influences support for redistributive fiscal policies. I
use the state-level analog of the county-level measure used in their study.
Let xq be the proportion of state’s white population living in census tract
q and let yq be the proportion of state’s nonwhites living in census tract
q. Then
Segregation ¼
1 X :
x
y
q
q
8q
2
If each census tract has the same proportion of white and nonwhite residents
as the state as a whole, then this segregation index equals zero. An index
value of 100 indicates complete segregation. The measure can be
interpreted as the proportion of the state’s nonwhite population who would
need to move to a different tract in order to reduce segregation to zero.8 As
was mentioned earlier, greater diversity may increase support for redistributive fiscal policy by increasing contact between different ethnic groups.
However, individuals who are already comfortable with other ethnic groups
may choose to live among them. Regardless of whether residential segregation reinforces or reflects animosity between ethnic groups, it can be viewed
as a proxy for the degree of ethnic group tension. The African American
share of state population is included in the models because Fong (2001) and
Alesina and La Ferrara (2005) find that African Americans tend to be more
supportive of redistribution. If the BRFH measure of average voter ideology
is a precise measure of the average voter’s taste for redistribution, then the
ethnic demographic variables should be redundant. They are included in
case the BRFH voter ideology measure does not fully capture the
electorate’s redistributive preferences.
The models also include the percentage of a state’s population age
sixty-five or older. This group is likely to have a high consumption-toincome ratio. Consequently, the elderly may prefer greater reliance on
income taxes relative to consumption taxes. Thus, I expect the elderly
population share to be associated with a progressive distribution of tax
burdens. The model includes the percentage of workers who are labor union
Downloaded from pfr.sagepub.com at PENNSYLVANIA STATE UNIV on February 19, 2016
Foster
187
members since labor unions tend to promote redistributive fiscal policies. I
also control for income inequality, measured with the ratio between the
mean and median household incomes, since a relatively high degree of
inequality may create political pressure for redistribution (Hettich and
Winer 1999; Chernick 2005). Numerous empirical studies of the political
economy of subnational tax and spending policies control for per capita
income and the poverty rate. Thus, those variables are controlled in the
models estimated here.
The equations include measures of a state’s capacity to export taxes to
nonresidents via the federal deductibility of state and local taxes and
interstate commerce. The percentage of tax returns with itemized deductions
is used as a proxy for the deductibility offset for residents. Since highincome households are more likely to itemize, federal deductibility lowers the
political costs associated with taxing these households. Feldstein and Metcalf
(1987) make a strong case for the possibility that the itemization rate is subject
to reverse causation.9 To address this issue, I use a generalized method of
moments (GMM) model, which is described in detail in the next subsection.
The models include proxies for a state’s capacity to export taxes through
interstate commerce. The Sales Activity Index is a proxy for the importance
of tourism in a state’s economy. It is the ratio between a state’s per-capita
retail sales and the national average. Tourism provides opportunities for consumption tax exporting, which can influence state and local tax progressivity
in two ways. Consumption tax exporting opportunities may encourage
reliance on the sales and excise taxes, which reduce progressivity. However,
the fact that nonresidents bear some of the burden allows governments to
raise a given amount of revenue with lower consumption tax rates. Tourism
may also encourage greater reliance on license fees, which may align the
benefits obtained by tourists more closely with the costs associated with providing the public services that they utilize. For these reasons, the expected sign
on the Sales Activity Index is ambiguous. The presence of mineral resource
industries can also provide opportunities for states to shift taxes to nonresidents, either through the sale of these goods outside of the state or through
backward shifting to investors.10 Firms involved in these industries tend to pay
sales, corporate income, and property taxes. Portions of all of these taxes could
potentially be exported, so it is unclear how the importance of mineral
resources in a state’s economy could influence state and local tax structure.
The politically optimal tax structure for a particular state may be
influenced by the tax structures adopted by its neighbors. WPjt denotes the
weighted average progressivity measure of state i’s neighbors or the spatial
lag of the dependent variable. I experiment with two weighting matrices that
Downloaded from pfr.sagepub.com at PENNSYLVANIA STATE UNIV on February 19, 2016
188
Public Finance Review 41(2)
are frequently used in the literature: the simple contiguity matrix and the
population-weighted contiguity matrix. The simple contiguity matrix assigns
a weight of 1/n to each neighboring state, where n equals the number of neighbors, and a weight of zero to nonneighboring states. It calculates the
unweighted average tax rate of neighboring states. With the populationweighted contiguity matrix, the tax variable of each neighbor is weighted by
its share of the total population of neighboring states since a state may be more
(less) responsive to the tax policies imposed by their larger (smaller) neighbors. As with the simple contiguity matrix, nonneighbors receive a weight
of zero. The population weights are based on state population averaged across
the sample period. As with all models of strategic interaction between governments, the sign on the spatial lag is theoretically ambiguous (Brueckner and
Saavedra 2001). Since the tax structure adopted by state i may influence those
adopted by neighboring states, WPjt will be treated as endogenous.
The models control for year-fixed effects. State-fixed effects are likely to
be relevant. However, the ratio between the average within SD of each tax
variable and its mean tends to be very small (see table 1), meaning that state
tax structures changed little during this period. Thus, most of the variation
resulted from variation between states. To allow the between-state variation
to play a greater role in the analysis, I use indicator variables for the nine
Census divisions (with New England as the reference category) instead of
state-fixed effects. The results must be interpreted with caution since the
regressors may be correlated with the state-fixed effects and the Census
division indicators may not completely control for states’ time-invariant,
unobserved characteristics.11 Table 1 presents summary statistics. The data
sources are presented in appendix A.
Prior to estimating the above model, there are econometric issues that
need to be addressed. There are two potentially endogenous variables: the
itemization rate and the neighbors’ weighted average tax burden measure
or the spatial lag. To address the potential endogeneity, I use a GMM
estimator with clustering by state. I use three instruments for the itemization
rate: its tenth-order lag, the percentage of household heads who are married,
and the average number of children per household. The tenth-order lag of
the itemization rate should be recent enough to be correlated with the
current value but a deep enough lag to be uncorrelated with the error term.
Edmiston and Spong (2012) find that married households and households
with children are more likely to itemize. Thus, I expect the percentage of
householders who are married and the average number of children per
household to be positively related to the itemization rate. I do not expect
these variables to be directly related to state and local tax progressivity.
Downloaded from pfr.sagepub.com at PENNSYLVANIA STATE UNIV on February 19, 2016
Foster
189
Table 1. Summary Statistics
Variable
M
Top-to-Bottom %
Top-to-Middle %
Middle-to-Bottom %
Q1 Rate
Q2 Rate
Q3 Rate
Q4 Rate
Q5 Rate
Average Voter Liberalism
Government Liberalism
Residual
Ethnic Congruence
Segregation Index
African American
Age 65þ
Union Density
Per Capita Income (thousands
of year 2005 dollars)
Sales Activity Index
Mining
Marriage Rate
Number of Children Per
Household
Overall SD Within SD
Min
Max
76.7
85
89.2
10.796
9.807
9.458
9.071
8.059
50.12
0
20.2
13.9
13.8
2.128
1.603
1.532
1.603
1.869
15.8
23.673
5.987
5.719
6.920
0.945
0.623
0.515
0.451
0.611
7.669
17.222
31.2
130.1
47.5
112.7
53.9
125
4.7
17.6
4.7
14.3
4.8
14.3
4.3
14.6
3.013 13.833
8.45
95.972
45.566 49.911
0.593
53.503
10.527
12.799
11.933
31.883
0.171
10.724
9.559
1.614
5.463
5.947
0.044
2.801
0.378
0.322
1.417
3.875
0.312
29.052
0.35
8.593
3
20.936
0.967
75.118
37.028
18.524
27.9
52.728
1.014
2.531
53.203
0.704
0.131
5.269
3.608
0.092
0.066
1.189
2.275
0.051
0.78
0.018
50.8
0.56
1.51
34.65
69.1
1.24
Following Rork (2003), I use the spatial lags of the exogenous regressors
as instruments for the spatial lag of the own-state’s tax measure. These are
valid instruments provided that they are jointly correlated with the spatial
lag of the own-state tax variable, uncorrelated with own-state unobservables, and not directly related to the own-state tax variable. I also included
the tenth-order lag of the average income tax share of own-source revenue
for neighboring states.
The instruments satisfy two criteria: from Wald tests, I find that the
instruments are jointly correlated with the endogenous regressors12; in overidentification tests, I fail to reject the joint exogeneity of the surfeit of
instruments at the 95 percent confidence level. These tests provide evidence
of the validity of the instruments.
To determine whether the use of GMM was justified, I carried out
the Durbin-Wu-Hausman endogeneity test. The null hypothesis of
Downloaded from pfr.sagepub.com at PENNSYLVANIA STATE UNIV on February 19, 2016
190
Public Finance Review 41(2)
Table 2. Pooled GMM Estimates for the Progressivity Equations.a
(1)
Top-toBottom %
(2)
Top-toMiddle %
(3)
Middle-toBottom %
With simple contiguity weights
Average Voter
Liberalism
Government Lib.
Residual
Ethnic
Congruence
Segregation
Index
African American
Age 65 þ
Union Density
Per Capita Inc.
($1,000)
Income Inequality
Poverty
Itemizers
Sales Activity
Index
Mining
Neighbors’
Progressivity
Constant
R2
0.259**
0.067
(.011)
(.341)
0.038
0.013
(.271)
(.606)
60.49***
14.54**
(<.0001)
(.045)
0.575*** 0.360***
(<.0001)
(.001)
0.069
0.181
(.718)
(.146)
0.832
0.818
(.565)
(.365)
0.029
0.301
(.929)
(.149)
0.459
0.377
(.323)
(.166)
5.409
0.946
(.731)
(.908)
2.290***
0.711***
(<.0001)
(.005)
2.714***
1.175***
(<.0001)
(<.0001)
29.967*** 21.009***
(.007)
(.002)
0.247
0.168
(.336)
(.369)
0.740*** 0.777***
(<.0001)
(<.0001)
37.05
142.2***
(.371)
(<.0001)
0.60
0.59
0.247***
(<.0001)
0.003
(.913)
54.12***
(<.0001)
0.143
(.175)
0.270**
(.039)
0.604
(.475)
0.355
(.138)
0.318
(.426)
7.188
(.576)
2.041***
(<.0001)
1.957***
(<.0001)
8.159
(.327)
0.447***
(.005)
0.675***
(<.0001)
33.11
(.284)
0.59
(4)
Top-toBottom %
(5)
Top-toMiddle %
(6)
Middle-toBottom %
With population-contiguity weights
0.205*
0.041
(.071)
(.574)
0.014
0.007
(.742)
(.804)
57.16***
12.42*
(<.0001)
(.089)
0.577*** 0.436***
(<.0001)
(<.0001)
0.127
0.259**
(.544)
(.040)
0.435
1.050
(.729)
(.152)
0.140
0.388*
(.694)
(.052)
0.152
0.359
(.783)
(.198)
1.428
5.939
(.910)
(.463)
2.998***
1.402***
(<.0001)
(<.0001)
2.639***
1.305***
(<.0001)
(<.0001)
25.156** 16.143**
(.020)
(.026)
0.0901
0.199
(.756)
(.342)
0.577*** 0.454***
(<.0001)
(.001)
5.793
108.3***
(.875)
(<.0001)
0.56
0.58
0.259***
(<.0001)
0.001
(.981)
57.08***
(<.0001)
0.199
(.113)
0.200
(.148)
0.175
(.842)
0.552**
(.030)
0.163
(.722)
1.003
(.929)
2.242***
(<.0001)
1.969***
(<.0001)
14.660*
(.093)
0.383**
(.027)
0.519**
(.003)
18.23
(.514)
0.58
Note: N ¼ 144. p ¼ values in parentheses.
*p < .10. **p < .05. ***p < .01.
a
Year indicators and indicators for the nine Census divisions (with New England serving as the
reference category) were included in the models. The coefficients from those indicators were
omitted from the table to conserve space.
exogeneity was rejected at the 95 percent confidence level for two of
the three progressivity equations. Thus, the discussion of the results
focuses on the GMM estimates.
Results
Table 2 presents the GMM regression results for the factors that influence
the progressivity of state and local taxes. Regardless of which spatial
Downloaded from pfr.sagepub.com at PENNSYLVANIA STATE UNIV on February 19, 2016
Foster
191
weighting matrix is used, average voter liberalism is positively and
significantly related to the middle quintile’s average tax rate as a percentage
of that of the bottom quintile (Middle-to-Bottom %). In other words, there is
robust evidence of a positive relationship between average voter liberalism
and progressivity toward the low end of the income distribution. It is also
positively related to the average tax rate of the top quintile as a percentage
of that of the bottom quintile (Top-to-Bottom %). However, the effect is
statistically significant only at the 90 percent confidence level when the
population-weighted contiguity weights are used. When average voter
liberalism is statistically significant at traditionally accepted confidence
levels, the effect of an increase in that variable of one SD falls between 4 and
5.5 percent of the mean value of the progressivity measure. Thus, it appears
that the effects of average voter liberalism on state and local tax progressivity
are economically insignificant.13,14 The estimated effects may have been
larger if differences in political influence among voters could have been taken
into account. The BRFH measure of average voter liberalism is a voter
population-weighted average. Political influence may be positively correlated
with wealth. If this tends to be the case, then an income-weighted average
ideology measure may more accurately gauge the influence of voter ideology
on state and local tax progressivity. Government liberalism does not appear to
be a significant determinant of progressivity.15
State ethnic demographics appear to significantly influence the distribution
of tax burdens across income groups. The ethnic congruence between the poor
and the nonpoor is significantly and positively related to all three progressivity
measures.16,17 Table 3 presents the effects of the ethnic demographic variables
on each income quintile’s average tax rate. The estimates were obtained with
the GMM estimator, using the same independent variables and excluded
instruments that were used in the estimation of the progressivity equations.
We see that Ethnic Congruence is inversely related to the average tax rate
imposed on the bottom quintile and positively related to the average tax rate
faced by the top two quintiles. It is also positively related to the middle quintile’s average tax rate. However, its statistical significance is sensitive to the
spatial weighting matrix that is used. These results suggest that regressive taxation is more politically costly when there is greater ethnic congruence between
the poor and the nonpoor. One possible explanation for the positive association
between Ethnic Congruence and the average tax rates faced by the wealthy is
that those taxpayers may get less disutility from paying taxes when members of
their own ethnic group(s) constitute a large share of the beneficiaries.
The effects of Ethnic Congruence on the progressivity measures are large in
magnitude. A one-SD increase in Ethnic Congruence is predicted to increase
Downloaded from pfr.sagepub.com at PENNSYLVANIA STATE UNIV on February 19, 2016
192
Public Finance Review 41(2)
Table 3. Pooled GMM Estimates of the Effects of Voter Ideology and Ethnic Demographics on Average Tax Rates.a
Q1 Rate
Q2 Rate
Q3 Rate
Q4 Rate
Q5 Rate
Estimates from equations using the simple contiguity spatial weights
Average Voter
0.005
0.009
0.013*
0.005
0.012*
Liberalism
(0.652)
(0.193)
(0.087)
(0.491)
(0.078)
Ethnic Congruence 4.418*** 0.541
2.197*
3.301**
2.535**
(0.004)
(0.655)
(0.078)
(0.010)
(0.023)
Segregation Index
0.084*** 0.076*** 0.064*** 0.059*** 0.039**
(<.0001) (<.0001) (<.0001) (<.0001)
(0.001)
African American
0.051*
0.011
0.014
0.013
0.036*
(0.056)
(0.657)
(0.514)
(0.509)
(0.051)
R2
0.42
0.44
0.56
0.69
0.76
Estimates from equations using the population-weighted contiguity spatial weights
Average Voter
0.007
0.006
0.013
0.011
0.021***
Liberalism
(0.512)
(0.465)
(0.108)
(0.163)
(0.008)
Ethnic Congruence 4.101*** 1.439
3.115*** 4.126*** 2.983***
(0.009)
(0.249)
(0.004)
(<.0001)
(0.002)
Segregation Index
0.082*** 0.071*** 0.058*** 0.048*** 0.024*
(<.0001) (<.0001) (<.0001) (<.0001)
(0.059)
African American
0.053** 0.007
0.009
0.009
0.033*
(0.043)
(0.767)
(0.667)
(0.654)
(0.078)
R2
0.39
0.43
0.57
0.70
0.75
Note: N ¼ 144. p ¼ values in parentheses.
*p < .10. **p < .05. ***p < .01.
a
The models include the regressors listed in table 2, along with the Census division and year
indicators. The spatial lag of the relevant average tax rate was used instead of the spatial lag of a
progressivity measure. The unabridged results are available upon request.
the Top-to-Bottom % by about 13 percent from the mean regardless of which
spatial weighting matrix is used. The effect of a one-SD increase in Ethnic Congruence on the Middle-to-Bottom % is equal to around 10 percent of the mean
of that progressivity measure with both specifications of the spatial lag. The
effect is a little more than twice that of average voter liberalism. These
results suggest that the ethnic similarity between the poor and the nonpoor
is an important determinant of the average tax rates imposed on the poor
relative to those faced by middle and upper-income households. Ethnic
Congruence is positively related to the Top-to-Middle %, but the effects are
small in magnitude.
Downloaded from pfr.sagepub.com at PENNSYLVANIA STATE UNIV on February 19, 2016
Foster
193
These results are at odds with those presented by Hero (1998), who does
not find a significant relationship between ethnic heterogeneity and state
and local tax progressivity. It is important to note that Hero does not control
for tax exporting capacities and spatial dependence. I obtain results that are similar to his when I exclude the tax exporting variables and the spatial lag (these
results are available upon request). I find that both the Sales Activity Index and
the fitted values of the itemization rate from the first-stage regression are significantly correlated with Ethnic Congruence.18 The correlations between the spatial lags of the progressivity measures and Ethnic Congruence are insignificant.
However, the fitted values of the spatial lags tend to be highly correlated with
one or more of the other demographic variables, which tend to be significantly
correlated with Ethnic Congruence.19 These relationships may explain the differences between the results presented here and those obtained by Hero. Excluding the tax exporting variables and ignoring spatial dependence may produce
biased results.
The degree of tension between ethnic groups, measured with an index of
residential segregation also appears to be significantly associated with state
and local tax progressivity. A one-SD increase in residential segregation is
associated with a reduction in the Top-to-Bottom % of roughly 8 percent of
the mean of that progressivity measure. It is also inversely related to the
other two progressivity measures, but the effects are smaller in magnitude.
The residential segregation index is positively related to the average tax
rates faced by all five income quintiles. However, the effects are larger for
lower-income groups. These results suggest that greater tension between
ethnic groups is associated with higher state and local expenditures as a
share of state personal income but that these expenditures are financed with
relatively regressive tax structures. Social and economic divisions between
ethnic groups may be directly related to spending on law enforcement and
corrections. However, an examination of the effects of ethnic group tension
on the composition of subnational government expenditures is beyond the
scope of this study.
Like Chernick (2005), I find that the progressivity of neighboring states
is inversely related to own-state progressivity. The spatial lag is significantly
and negatively related to all three of the progressivity equations. This means
that if neighbors’ tax structures become less (more) progressive on average,
then a particular state’s tax structure will become more (less) progressive.
These results may reflect the development of tax havens within regions.
A state’s capacity to export taxes through federal deductibility has a
statistically and economically significant impact on state and local tax
progressivity. The size of the deductibility offset for a state is increasing in the
Downloaded from pfr.sagepub.com at PENNSYLVANIA STATE UNIV on February 19, 2016
194
Public Finance Review 41(2)
percentage of tax returns with itemized deductions. An increase in the itemization rate of one SD is predicted to prompt an increase in the Top-to-Bottom %
equal to approximately 26 percent of the mean. These results suggest that the
indirect subsidy offered by deductibility has been a strong motivator for the
adoption of progressive tax structures. These findings are consistent with those
obtained by Chernick (1992, 2005).
Tax exporting capacities arising from interstate commerce appear to exercise
influences on the distribution of subnational tax burdens that are slight in economic terms but statistically significant. The Sales Activity Index is inversely
related to the Top-to-Bottom % and the Top-to-Middle % while the mining share
of state private sector GDP is positively related to the Middle-to-Bottom %.
Discussion and Concluding Comments
The survey-based research conducted by Fong (2001) and Alesina and La Ferrara (2003) suggests that citizens’ preferences for redistribution are significantly
affected by conceptions of fairness even when variables reflecting the extent to
which they are likely to benefit financially from redistributive policies are held
constant. Conceptions of fairness are likely to be grounded in established ideologies. This study examines the relationship between average voter liberalism
and the determination of tax progressivity by state and local governments. I find
that the impact of average voter liberalism is statistically significant and positive.
However, the relationship does not appear to be economically significant. These
results may reflect measurement error. Thus, continued experimentation with
the measurement of voter ideology may yield valuable results.
On the other hand, the ethnic congruence between the poor and nonpoor and
the degree of tension between ethnic groups appear to have substantial impacts
on state and local tax progressivity. When the poor and nonpoor tend to be of
the same ethnicity, state and local governments tend to utilize tax structures
that are more progressive, ceteris paribus. The amiability of ethnic group relations, measured with an index of ethnic residential segregation, is inversely
related to progressivity. Thus, the downward pressure on support for progressive taxation generated by ethnic dissimilarity between the poor and nonpoor
may be offset by greater congeniality between ethnic groups. These results
indicate that voter preferences concerning tax progressivity are not completely
determined by narrow self-interest. Voters appear to consider the impacts of
state and local fiscal policies on others. However, the weight that the nonpoor
attach to the well-being of the poor appears to be substantially shaped by ethnic
group loyalty and by the degree of tension between ethnic groups. The fact that
these variables are significant even with average voter liberalism held constant
Downloaded from pfr.sagepub.com at PENNSYLVANIA STATE UNIV on February 19, 2016
Foster
195
suggests that the interaction between state ethnic demographics and voter
altruism may not be captured by measures of general voter ideology.
The results presented here suggest that changes in ethnic demographics
could lead to changes in state and local tax progressivity. Population
projections developed by the US Bureau of the Census suggest that significant
changes in the nation’s ethnic composition will occur in the proceeding
decades. From 2010 to 2050, the Hispanic share of the population is projected
to rise from 15 percent to 24 percent. The effects that these changes will have
on the ethnic similarity between the poor and the nonpoor will depend on the
upward mobility of Hispanics in the proceeding decades. If the growth of the
Hispanic population increases the Hispanic share of the poor population without leading to an increase in the Hispanic share of the nonpoor population that
is at least equal in magnitude, then it may place downward pressure on popular
support for progressive taxation. The effects may be less pronounced if ethnic
differences become less divisive.
Appendix A
Data Sources
Variable
Source
Average Tax Rates
Citizens for Tax Justice/Institute on Taxation and
Economic Policy (2003); Institute on Taxation and
Economic Policy (2009)
Average Voter Liberalism, Richard Fording’s website: http://www.bama.ua.edu/
Government Liberalism
*rcfording
Segregation Index
William H. Frey and the University of Michigan Social
Science Data Analysis Network
Itemization Rate
US Internal Revenue Service, Statistics of Income,
various years
Sales Activity Index
US Department of Commerce, Bureau of the Census,
Statistical Abstract of the United States, various years
Mining, Per-Capita Income US Bureau of Economic Analysis, various years
Calculated by author using data from US Department of
Income Inequality,
Commerce, Bureau of the Census; United States
Number of Children
Department of Labor, Bureau of Labor Statistics,
Per Household, Ethnic
Annual Social and Economic Supplement Survey, various
Congruence, Marriage
years
Rate
Age 65þ, African
US Department of Commerce, Bureau of the Census,
American
http://www.census.gov/popest/archives/
Union Density
Hirsch, Macpherson, and Vroman (2001)
Poverty Rate
University of Kentucky Center for Poverty Research
Downloaded from pfr.sagepub.com at PENNSYLVANIA STATE UNIV on February 19, 2016
196
Public Finance Review 41(2)
Appendix B
Methodology Used to Develop the Average Tax Rates
The tax burden estimates used here were provided by the Citizens for
Tax Justice and the Institute on Taxation and Economic Policy
(1996) and the Institute on Taxation and Economic Policy (2003,
2009). The authors used samples of IRS tax returns, data from the
US Census Bureau, and other sources to estimate the personal income,
general sales, cigarette, gasoline, beer, corporate income, and property
tax burdens for samples of households that are representative at the
state level. The authors use state and local tax provisions to estimate
the income tax burdens for each household. Estimates of the homeowner property tax burden borne by itemizing homeowners were
derived from IRS tax return data. The burdens for nonitemizers were
imputed using data from the Census and other sources.
The authors estimated the sales tax burdens for their samples by
imputing expenditures for numerous items and calculating the tax
burdens based on those expenditures. More specifically, the authors
utilized data from the Consumer Expenditure Survey (CES) to
estimate a model of household expenditure on each item. The authors
then used the estimated models to impute the consumption expenditures for the households in their state-level samples. The sales and
excise tax burdens were estimated by applying state and local sales
and excise tax provisions.
Business property and corporate income tax burdens were
estimated by first allocating the national total business property and
corporate income burden to the states in relation to each state’s share
of national capital income. The authors proceeded to allocate a particular state’s business property and corporate income tax burden
across that state’s income groups according to each group’s share
of state capital income. In states that imposed business property and
corporate income taxes that were significantly above the national
median, the authors assigned the excess to either in-state wages or
in and-out-of-state consumption depending on the activity.
The authors also accounted for sales and excise taxes levied on
firms. These levies were divided into those paid by industries
producing for domestic markets and those producing for national
Downloaded from pfr.sagepub.com at PENNSYLVANIA STATE UNIV on February 19, 2016
Foster
197
markets. Taxes on domestic market items were assigned to residents
of each state in proportion to their share of the state’s consumption
expenditures on each item. Taxes on national market items were
assigned to national consumption with an adjustment to reflect a
portion assumed to be retained in state. In states that imposed
relatively high sales and excise taxes on national market activities,
firms may not be able to pass the entire burden forward to consumers
because of competitive factors. Because capital and labor are
immobile in the short run, portions of these taxes may be shifted back
to wages and capital. For states with national market business tax
burden that were significantly above the national median, the authors
divided the excess evenly between in-state wages and capital.
Declaration of Conflicting Interests
The author declared no potential conflicts of interest with respect to the research,
authorship, and/or publication of this article.
Funding
The author received no financial support for the research, authorship, and/or
publication of this article.
Notes
1. See Tresch (2002, 832–928) for a review of the theoretical literature on fiscal
federalism.
2. The progressivity of the tax system can be enhanced by imposing relatively high
tax rates on commodities consumed disproportionately by the rich.
3. The development of cross-sectional data on lifetime tax incidence is likely to
involve significant data analysis challenges.
4. Tax burden estimates for earlier time periods are available. The Citizens for Tax
Justice (CTJ) produced estimates for 1985 and 1991 in CTJ (1991). The study
used a different sampling strategy and different shifting assumptions than those
used by Citizens for Tax Justice and Institute on Taxation and Economic Policy
(1996) and the Institute on Taxation and Economic Policy (ITEP; 2003, 2009).
Estimates for 1977 were provided by Phares (1980). While the CTJ/ITEP
studies were based on microsimulation approaches, Phares used aggregate state
data to allocate total tax burdens to income groups. Chernick (2005) pooled the
Phares estimates with the CTJ estimates to form a three-year panel. Data on all
of the regressors in the models presented here are available only for 1990
onward. Thus, I was only able to add the 1991 CTJ cross section to the panel.
Downloaded from pfr.sagepub.com at PENNSYLVANIA STATE UNIV on February 19, 2016
198
5.
6.
7.
8.
9.
10.
11.
12.
Public Finance Review 41(2)
To determine whether the 1991 cross section could be pooled with the 1995,
2002, and 2007 cross sections, I carried out the Chow test for parameter
constancy. The null hypothesis of parameter constancy was rejected at the 99
percent confidence level for all of the equations.
Erikson, Wright, and McIver (1993) developed a measure of mean state ‘‘elite
ideology’’ based on survey responses from political party elites within each
state. However, this measure dates back to the mid-1980s.
Using the raw government liberalism measure does not substantially affect the
results. Using the residual instead simplifies the interpretation of the results
since one is not required to take into account the indirect effect of average voter
liberalism on progressivity through its effect on government liberalism.
See Fong and Luttmer (2011) for a review of the literature.
I obtained the segregation data from William H. Frey and the University of
Michigan Social Science Data Analysis Network. The authors calculated the
segregation index from 1990 and 2000 decennial Census data and from
American Community Survey data pooled from 2005 to 2009. I used the
estimates derived from ACS data for 2007 and produced estimates for 1995 and
2002 with linear interpolation. The authors calculated the indices for white–
African American, white–Hispanic, and white–Asian segregation. I use a
weighted average of these indices for which the weights are each nonwhite
group’s share of the sum of the state’s Hispanic, non-Hispanic African
American, and non-Hispanic Asian populations.
If income tax rates are relatively high in a particular state, then more taxpayers
are likely to itemize. This would cause an upward bias.
If a state dominates a natural resource market, then it may be able to shift a
portion of its business taxes forward to nonresident consumers. Otherwise, any
exporting of these taxes that occurs will be backward to nonresident investors
(Gade and Adkins 1990).
I estimated the models with the GMM fixed effects estimator and found that the
fitted values and the estimates of the fixed effects are significantly correlated in
every case. This suggests correlation between the state fixed effects and at least
one of the regressors. The Census division indicators may only account for a
portion of the variation that is explained by the state-fixed effects.
Consequently, the estimates presented here must be interpreted with caution.
Evidence of identification of the endogenous variables does not rule out the
possibility that the instruments are only weakly correlated with the endogenous
regressors. Weak identification can potentially produce a serious degree of
small-sample bias. I obtained the Kleibergen and Paap weak identification test
statistics and compared them to the critical values provided by Stock and Yogo
(2005). These critical values are defined so as to ensure that the bias of GMM is
Downloaded from pfr.sagepub.com at PENNSYLVANIA STATE UNIV on February 19, 2016
Foster
13.
14.
15.
16.
17.
199
no more than 10 percent of the bias of OLS at the 95 percent confidence level. In
all but two of the equations, the Kleibergen-Paap statistic exceeded the
Stock-Yogo critical value. The test statistic failed to reject the null hypothesis
of weak identification for the equations for the Top-to-Bottom and Top-toMiddle percentages, with the spatial lag calculated with the populationweighted contiguity weights. For both equations, the Kleibergen-Paap test
statistic exceeds the Stock-Yogo test statistic that ensures that the bias of GMM
is no more than 20 percent of the OLS bias. The Kleibergen-Paap test statistic is
robust to within-cluster correlation in the errors. However, the Stock-Yogo critical values require an assumption of i.i.d. errors. Critical values for the
Kleibergen-Paap statistic have not been developed. This is problematic in this
case since within-state error correlation could be present. When one is dealing
with potential within-cluster error correlation in a GMM context, Baum, Schaffer,
and Stillman (2007) recommend reporting the Kleibergen-Paap statistic and using
the Stock-Yogo critical values with caution.
I also estimated the equations with the Erikson, Wright, and McIver (1993)
measure of average liberalism. The coefficient on this measure is significant
only in the equation for the Middle-to-Bottom %. The magnitude of the effect
is not substantially different from that of the BRFH voter liberalism measure.
These results are available from the author upon request.
Replacing Census division-fixed effects with state-fixed effects has a significant impact on the results. When state-fixed effects are included and the population
contiguity weights are used, average voter liberalism has a small negative effect on
the Top-to-Middle %. It is otherwise statistically insignificant. These results may
be attributable to the low degree of within-variation in average voter liberalism.
The within-SD of average voter liberalism is equal to 11 percent of the mean.
Both the government liberalism residual and the raw government liberalism
measure are insignificant even when average voter liberalism is excluded.
Excluding government liberalism or the residual measure does not substantially
affect the coefficient on average voter liberalism.
Excluding Ethnic Congruence and residential segregation from the model does
not significantly affect the coefficient on average voter liberalism and vice
versa.
When state-fixed effects are used, the coefficient on Ethnic Congruence is
positively related to the Top-to-Bottom %. The coefficient is about half the size
of the coefficient that is obtained when the Census division fixed effects are used.
It is only statistically significant when the population-weighted contiguity
weights are used. It is insignificant in the equations for the other two progressivity
measures. These results may be attributable to the low within-state variation in
Ethnic Congruence. The within-SD is equal to about 7 percent of the mean.
Downloaded from pfr.sagepub.com at PENNSYLVANIA STATE UNIV on February 19, 2016
200
Public Finance Review 41(2)
18. Ethnic Congruence is positively correlated with the Sales Activity Index but
negatively correlated with the itemization rate. The absolute values of the
correlation coefficients are close to 0.2.
19. For example, the correlation between the spatial lag of the Top-to-Bottom %
and the poverty rate is .42. The correlation between the poverty rate and Ethnic
Congruence is .35.
References
Alesina, Alberto, and Eliana La Ferrara. 2005. ‘‘Preferences for Redistribution in the
Land of Opportunities.’’ Journal of Public Economics 89:897–931.
Anselin, Luc, and Harry H. Kelejian. 1997. ‘‘Testing for Spatial Error Autocorrelation in the Presence of Endogenous Regressors.’’ International Regional Science
Review 20:153–82.
Baum, Christopher F., Mark E. Schaffer, and Steven Stillman. 2007. ‘‘Enhanced
Routines for Instrumental Variables/GMM Estimation and Testing.’’ Stata
Journal 7:465–506.
Berry, William D., Evan J. Ringquist, Richard C. Fording, and Russell L. Hanson.
1998. ‘‘Measuring Citizen and Government Ideology in the American States,
1960-1993.’’ American Journal of Political Science 42:327–48.
Brueckner, Jan K., and Luz Saavedra. 2001. ‘‘Do Local Governments Engage in
Strategic Tax Competition?’’ National Tax Journal 54:231–53.
Chernick, Howard. 1992. ‘‘A Model of the Distributional Incidence of State and
Local Taxes. Public Finance Review 20:572–85.
Chernick, Howard. 2005. ‘‘On the Determinants of State and Local Tax Progressivity
in the U.S. National Tax Journal 58:93–112.
Chernick, Howard. 2010. ‘‘Redistribution at the State and Local Level:
Consequences for Economic Growth. Public Finance Review 38:409–49.
Citizens for Tax Justice. 1991. A Far Cry from Fair: CTJ’s Guide to State Tax
Reform. Washington, DC: Citizens for Tax Justice.
Citizens for Tax Justice and the Institute on Taxation and Economic Policy. 1996.
Who Pays? A Distributional Analysis of the Tax Systems in All 50 states.
Washington, DC: Citizens for Tax Justice and the Institute on Taxation and
Economic Policy.
Edmiston, Kelly D., and Kenneth Spong. 2012. ‘‘Tax Incentives for Homeownership and the Provision of Local Public Services.’’ Public Finance Review 40:
116–44.
Erikson, Robert S., Gerald C. Wright, and John P. McIver. 1993. Statehouse
Democracy: Public Opinion and Policy in the American States. New York:
Cambridge University Press.
Downloaded from pfr.sagepub.com at PENNSYLVANIA STATE UNIV on February 19, 2016
Foster
201
Feldstein, Martin S., and Gilbert E. Metcalf. 1987. ‘‘The Effect of Federal Tax
Deductibility on State and Local Taxes and Spending.’’ Journal of Political
Economy 95:710–36.
Fletcher, Jason M., and Matthew N. Murray. 2008. ‘‘What Factors Influence the
Structure of the State Income Tax?’’ Public Finance Review 36:475–96.
Fong, Christina. 2001. ‘‘Social Preferences, Self-interest, and the Demand for
Redistribution.’’ Journal of Public Economics 82:225–46.
Fong, Christina, and Erzo F. P. Luttmer. 2011. ‘‘Do Fairness and Race Matter in
Generosity? Evidence from a Nationally Representative Charity Experiment.
Journal of Public Economics 95:372–94.
Gade, Mary N., and Lee C. Adkins. 1990. ‘‘Tax Exporting and State Revenue
Structures.’’ National Tax Journal 43:39–62.
Hero, Rodney. 1998. Faces of Inequality: Social Diversity in American Politics.
New York: Oxford University Press.
Hettich, Walter, and Stanley Winer. 1999. Democratic Choice and Taxation: A
Theoretical and Empirical Analysis. New York: Cambridge University Press.
Hirsch, Barry T., David A. Macpherson, and Wayne G. Vroman. 2001. ‘‘Estimates
of Union Density by State.’’ Monthly Labor Review 124:51–55.
Holbrook-Provow, Thomas M., and Stephen C. Poe. 1987. ‘‘Measuring State
Political Ideology.’’ American Politics Quarterly 15:399–416.
Institute on Taxation and Economic Policy. 2003. Who Pays? A Distributional
Analysis of the Tax Systems in All 50 States. 2nd ed. Washington, DC: Institute
on Taxation and Economic Policy.
Institute on Taxation and Economic Policy. 2009. Who Pays? A Distributional
Analysis of the Tax Systems in All 50 States. 3rd ed. Washington, DC: Institute
on Taxation and Economic Policy.
Luttmer, Erzo F. P. 2001. ‘‘Group Loyalty and the Taste for Redistribution.’’
Journal of Political Economy 109:500–28.
Metcalf, Gilbert. 1994. ‘‘The Lifetime Incidence of State and Local Taxes: Measuring
Changes During the 1980’s.’’ In Tax Progressivity and Income Inequality, edited
by Joel Slemrod, 59–88. New York: Cambridge University Press.
Morgan, David R. 1995. ‘‘Tax Equity in the American States: A Multivariate
Analysis.’’ Social Science Quarterly 75:510–23.
Phares, Donald. 1980. Who Pays State and Local Taxes? Oelgeschlager, UK: Gunn &
Hain.
Roch, Christine H., and Michael Rushton. 2008. ‘‘Racial Context and Voting
Over Taxes: Evidence from a Referendum in Alabama.’’ Public Finance
Review 36:614–34.
Rork, Jonathan C. 2003. ‘‘Coveting thy Neighbor’s Taxation.’’ National Tax
Journal 56:775–87.
Downloaded from pfr.sagepub.com at PENNSYLVANIA STATE UNIV on February 19, 2016
202
Public Finance Review 41(2)
Schlozman, Kay Lehman, Sidney Verba, and Henry E Brady. 1999. ‘‘Civic
Participation and the Equality Problem.’’ In Civic Engagement in American
Democracy, edited by Theda Skocpol and Morris P. Fiorina, 427–59.
Washington, DC: Brookings Institution Press.
Stimson, James A. 1991. Public Opinion in America: Moods, Cycles, and Swings.
Boulder, CO: Westview Press.
Stock, James H., and Motohiro Yogo. 2005. ‘‘Testing for Weak Instruments in IV
Regression.’’ In Identification and Inference for Econometric Models: A
Festschrift in Honor of Thomas Rothenberg, edited by Donald W. K. Andrews
and James H. Stock, 80–108. Cambridge University Press.
Tresch, Richard W. 2002. Public Finance: A Normative Theory. 2nd ed. London:
Academic Press.
Author Biography
John M. Foster is a doctoral candidate in the Martin School of Public Policy and
Administration at the University of Kentucky. His research interests include state
and local public finance and the economics of education.
Downloaded from pfr.sagepub.com at PENNSYLVANIA STATE UNIV on February 19, 2016