1MASSACHUSETS LIBRARIES MAY19 Do the Rich Speak Louder?:

Do the Rich Speak Louder?:
Examining whether U.S. Senators differentially respond to their
constituents by income across issues
by
Elisha W. Heaps
A.B. Government
Harvard University, 2010
SUBMITTED TO THE DEPARTMENT OF POLITICAL SCIENCE IN PARTIAL
FULFILLMENT OF THE REQUIRMENTS FOR THE DEGREE OF
MASTER OF SCIENCE IN POLITICAL SCIENCE
AT THE
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
1MASSACHUSETS
February 2014
INSTrUWE
OF TECHNOLOGY
MAY19 2014
LIBRARIES
@ 2014 Massachusetts Institute of Technology. All rights reserved.
Signature of Author
Department of Political' cience
October 31, 2013
Certified by
Christopher Warshaw
Assistant Professor of Political Science
Accepted by
Roger D. Peterson
Arthur and Ruth Sloan Professor of Political Science
Chairman, Graduate Program Committee
2
Do the Rich Speak Louder?:
Examining whether U.S. Senators differentially respond to their
constituents by income across issues
by
Elisha W. Heaps
Submitted to the Department of Political Science
on October 31, 2013, in partial fulfillment of the
requirements for the degree of
Master of Science in Political Science
Abstract
This thesis examines the relationship between public opinion and the way senators vote
on specific issues, and how this "responsiveness" might vary across income groups. The
independent variable of interest, state-level income group preference, is estimated using multilevel regression and poststratification (MRP) analysis. This is an improvement over earlier
methods, particularly when modeling income group level opinion where there are insufficient
sample sizes in national surveys. Income group opinions are found to be distinct across issue
areas and the top ten percent of the income bracket are found to hold different opinions when
compared to a more inclusively defined high-income group. Ideal point estimation is used
to generate the dependent variable of senator responsiveness based on roll call votes. The
first-stage MRP estimates of state-level income group opinion are then regressed on the corresponding senators' ideal points by issue area. While this paper expected the second stage
analysis to support an Instructed-delegate model of responsiveness, where senators vote in
accordance with constituents' interests, no evidence of such a relationship is found, even at
the aggregate opinion level. The evidence suggests that senators are looking elsewhere when
making their policy decisions.
Thesis Supervisor: Christopher Warshaw
Title: Assistant Professor of Political Science
3
4
Acknowledgments
This thesis would not have been possible without the support and encouragement of many
individuals. I cannot thank them enough for all of their help throughout this entire process.
First and foremost, I would like to thank my thesis supervisors, Professor Christopher Warshaw and Professor Adam Berinsky. Professor Warshaw generously spent many hours helping
me to apply MRP analysis, think through validation tests, develop modeling for the second
stage analysis, and construct a strong argument for my thesis. I was very fortunate to have
the opportunity to build on his own great accomplishments in the field. Professor Berinsky critically evaluated my work, which helped me develop a deeper understanding of the
topic and create well-formed methodological justifications for my modeling choices. It was
in Professor Berinsky's class that the idea for this project was born and, with his teaching,
I learned how to critically engage with the current scholarship and gained a more nuanced
understanding of the nature of public opinion and the challenges of using survey data. They
have both been true mentors and I owe them the greatest thanks.
I would also like to thank Professor Teppei Yamamoto who helped me apply the statistical methods used in this paper and provided me with a solid methodological background
in the classroom.
Chad Hazlett and Michael Sances, graduate students in the Political
Science Department, were both very generous with their time and I greatly appreciate all
of their help and advice. The comments and feedback from my cohort were also very helpful.
Finally, I would like to thank my family and friends for support, especially my husband,
Kevin. He provided me with advice, emotional support, and helped me locate the sources of
several pesky coding errors. He also helped to edit this Masters thesis, just as he had done
with my undergraduate honors thesis. I owe him my deepest and most heartfelt thanks.
5
Introduction
In this paper I examine the relationship between public opinion and the way senators vote
on specific issues, and how this "responsiveness" might vary across income groups. More
specifically, I will test the hypothesis that there is greater congruence between the way senators vote and the opinions of their higher-income constituents in certain policy areas (such
as economic policy), but not in others (such as social welfare). I first ask the question: do
constituents have opinions that vary by income level? I then test to see whether senators
are more responsive to a particular economic group when it comes to making policy for the
nation.
"Let them eat cake"
If the affluent have an oversized impact on the behavior of their elected officials scholars
should question what it means to have "true" democratic representation.
In the words of
V. 0. Key, "Unless mass views have some place in the shaping of policy, all the talk about
democracy is nonsense" (1961).
While we might not expect nor want a perfect correlation
between public opinion and government policy, enduring inequalities based on income status may be problematic for society. More privileged groups might try to shut out weaker
actors (perhaps by creating barriers to participation) in order to maintain their advantage,
undermining democracy (Schattschneider 1975). In any political conflict, the powerful seek
to privatize the issue to prevent the mobilization of the people, while the weak will try to
publicize the conflict so as to tip the balance of power. Government is an instrument of this
socialization of conflict.
Economic inequality has been increasing for the past thirty years in the United States.
This increase is compounded by the fact that responsiveness to the wealthy may have also
grown over this period (Gilens 2012, 252). Some argue this extreme concentration of income
gains at the top may in part be brought on by government policy perpetuating this "winner-
6
take-all" pattern (Hacker and Pierson 2010). This means influence is shifting toward a class
that looks increasingly dissimilar from the rest of America. If the wealthy have distinctive
preferences that are meaningfully different from those of other Americans, this may result
in a system of government that is not protecting the interests of the economic underclass.
The American representative republic is designed to protect and attend to the interests of
all citizens and as such, elite favoritism could pervert the government's purpose. While high
income individuals are generally more highly educated, they may lack information and experience to balance their views on public policy, or they may not care about advancing interests
other than their own. Therefore, first examining how the income preferences diverge and
then testing how they compare to their senators' votes will yield valuable insights about the
representative system.
Cacophony or Coherent?
One challenge to this line of inquiry is that the public may not have a clear opinion on
policy. If no clear opinion can be interpreted by the government, then the government cannot respond to it. Downs argues that it is rational to be ill-informed about public affairs,
conditional on trusting others to make decisions on one's behalf (1957). Lupia finds a solution to this problem of low information with the finding that people are able to form opinions
based on cues that would essentially match informed decision-making (1994). However, Lupia's work may be critiqued as only being the case for a small percentage of policies where
such cues are readily accessible to the public.
While Converse argues that individual opinion data is inconsistent and incoherent (1964),
Page and Shapiro show that in aggregate, public opinion is "stable and sensible" (1992, 1).
Fluctuations in opinion that might be observed at the individual level are absent when assessed en masse. Achen also responded to Converse's study and found that individuals' low
response stability was primarily due to measurement error (1975). While the empirical find-
7
ings of Converse's study have been scrutinized, he also suggests that most people structure
their thinking in terms of social groups. An income group may be one example of such a
reference group or an aggregation of related reference groups. Kinder expands on Converse's
reference and suggests that citizens first recognize the policy that benefits their group and
then form their opinions accordingly (2003). Dawson found that black Americans use group
identity to form policy preferences, which helps to explain why black party identification
is cohesive and relatively uniform (1994).
This suggests that it is possible to see distinct
preferences formed by income group either on the basis of membership within that particular
income group or because many readily identifiable subgroups can also be sorted into income
categories, such as a community church.
Only in rare cases are people be able to see the direct benefits of a policy to themselves, such
as the Vietnam War draft and Social Security benefits (Erikson and Stoker 2011; Campbell
2005). The received wisdom is that self-interest is not relevant in the formation of political
opinion (Sears et al. 1980). Therefore, while I would expect income group opinions to be
distinct and identifiable, they most likely formed from associations with people in a similar
income bracket.
Low-income individuals are more likely to belong to the same churches or bowling leagues
just as high-income individuals are more likely to frequent the same stores and belong to
the same country clubs. This sorting by income happens naturally, and as consequence,
different preferences concerning national policy are likely to emerge.
There are different
opinion-leaders in different social circles and it would be expected that these associations
simultaneously homogenize in-group preferences and perpetuate differences between groups.
Berelson, Lazarsfeld, and McPhee argue that political opinions are formed through the smallgroup contexts of friends, family, and coworkers and these groups tend to be homogeneous
in makeup (1954).
Expanding on this theory of social group opinion formation, Huckfeldt
and Sprague theorized that members of a political majority were less likely to recognize a
8
member of a political minority in their community (1987). If individuals fail to interact with
people from a different point of view or, if they are failing to recognize that they are doing so,
then it reasonable to expect that income group level differences to be real and self-sustaining.
When the People Talk, Does. Congress Listen?
In his book, Polyarchy, Robert Dahl wrote that "a key characteristic of a democracy is
the continuing responsiveness of the government to the preferences of its citizens, considered as political equals" (1).
There are three main categories of scholarship (as outlined
by Miller and Stokes (1963)) that have examined the way in which Congress represents its
constituents: the Instructed-delegate model, the Burkean model, and the Responsible party
model.
According to the Instructed-delegate model, representatives act in accordance with the preferences of their constituents. As such, representation is a form of "delegated" authority. The
way in which this particular model of responsiveness is measured is by observing the correspondence between congressional behavior and constituent opinion. One approach looks
to see how policy conforms with the corresponding changes in popular opinion. Page and
Shapiro find that changes in public preferences of a large enough magnitude correspond with
policy outcomes (1983).
Additionally, several scholars have found evidence that when the
public opinion has turned in favor of change, policy will match in the following years, particularly for issues the public finds important (Monroe 1979; Erikson, MacKuen, Stimson 2002).
Other scholars examine the link between constituent opinion and the behavior of elected
representatives or candidates across districts and between senators of the same state, finding positive relationships between constituents' opinions and their representatives' voting
behavior (Ansolabehere, Snyder, and Stewart 2001; Stimson, MacKuen, and Erikson 1995).
In line with this approach, Miller and Stokes examined the correlations between measures
9
of candidate and voter opinion on the issues of civil rights, social welfare, and foreign policy
and found that the opinions of those elected into office correlated most strongly with district
opinion, particularly in the area of civil rights (1963). They sampled 116 House districts in
the 1958 elections, however the samples varied, with sometimes less than 5 individuals in a
district. This lack of data results in inefficient estimates of voters' preferences at substate
levels. Achen argued that Miller and Stokes' correlation coefficients were not an adequate
measure of representation since large correlations can be observed when opinions are far
apart and small correlations can be observed when they are close together (1978). Contrary
to Miller and Stokes, Achen found that those who were not elected were better representatives of voters than winners and Achen did not observe significant variation across issue
areas. However, using a different methodology, Achen also found that there was a strong
positive association between district and member opinion.
Finding representative measures of opinion has been a persistent problem in this literature. Brace and his collaborators also examined state level opinion using representative
samples from a survey, however those estimates were also highly variable when examined
at the level of specific issues (2002). Another study by Clinton combined surveys to increase the sample size, resulting in over 100,000 combined responses, however there were
still fewer responses for specific issue areas (2006). Taking the mean or median of a small
sample of responses does not help with the issue of high variance. Clinton's study found
that House representatives were not completely responsive to district mean voter and did
not find strong evidence supporting party voting behavior in the House either, casting some
doubt on the Instructed-delegate model and party-based models of Congress. Recent studies
using multilevel regression and poststratification (MRP) to model constituent opinion at the
state or district level have found a solution to the problem of sample size (Lax and Phillips
2009; Park, Gelman and Bafumi 2004; Warshaw and Rodden 2012). Bayesian statistics and
multilevel modeling better leverage information about the respondents' demographics and
geography to provide estimates of opinion. While many studies found evidence to support
10
the Instructed-delegate model of representation, measures for constituents' opinion might be
improved with new methods like MRP, particularly when examining subpopulations.
In the Burkean model, representatives act in accordance with the interests of their constituents, instead of constituents' preferences. Voters thus entrust their representatives with
the responsibility of crafting policies that would best serve their interests. This model would
fit with Mayhew's description of representatives' behavior as advertising and credit-claiming
(1974). He argued that there was very little position-taking, though this was not as much
the case with senators, as representatives would focus on distributing particularistic benefits
to their home districts. Since reelection was the proximate goal, representatives would seek
to please their constituents to retain or capture their votes for the next cycle. Now whether
constituents would be most pleased with representatives who acted on behalf of their interests or preferences may be open for debate, regardless of the specific genre of scholarship,
Mayhew's argument unquestionably places the legislator in a position where he must be in
some way responsive to his constituents.
Fenno is arguably another scholar of the Burkean tradition (1977). While what representatives do in the capital is important, what really matters (from the electoral perspective) is
how they spend their time in the district, since this is how they cultivate trust. A representative's homestyle, or the tool by which constituent trust is cultivated, is his allocation
of resources, presentation of self, and his explanation for activity in Washington. It is not
the policy that is relevant, since constituents care more about the representative's homestyle.
However, policy does have some relevancy, even in the Burkean model, insofar as a representative's behavior in Washington cannot be used to subvert constituent trust. Challengers can
set the agenda in political campaigns to highlight inconsistencies in the incumbent's voting
record. Fenno argues that members are probably only constrained by one or two issues, but
on the majority of votes they can do as they wish as long as they are able answer for it
11
later. For example, Powell's amendment on HR7535, which called to block appropriations
of taxpayer money to construct segregated schools, created a strategic voting scenario. The
Democrats supported federal aid to schools and the Republicans did not. However, Southern
Democrats who were otherwise in favor of the aid, did not like the amendment for racial
integration. Some Republicans voted for the amendment since then they could be sure that
the bill would not pass. Likewise, some Northern Democrats votes against the amendment,
knowing that their Southern counterparts would vote against the bill if the amendment were
added. Ultimately, this was a killer amendment, which Republicans knew would have the bill
rejected. However, the Republicans who voted for the Powell amendment and the Northern
Democrats who voted against it would still have to face their constituents, and likely with
the vote pulled out of the context of the strategic voting environment. Trust was crucial
in order for this voting scenario to occur and some members just could not take that risk.
Representation and support are inseparable as members don't want to alienate their constituency. According to Fenno, the challenger controls which decisions a representative will
need to explain to his constituency. However, with information more accessible than ever
before, challengers may no longer be the powerful agenda-setters they once were. Though in
a more recent study, Ansolabehere, Snyder, and Stewart also argued that the more vulnerable the incumbent, the more attentive to the mainstream of the electorate he will be (2001).
In the Responsible party model, constituents express their policy preferences by casting
their votes for a particular party's platform where members serve as agents of the national
party. Miller and Stokes critique this model as not having a well-thought out concept of
representation (1963). The voters either affirm or reject the party platform and are viewed
as part of a national constituency instead of a local one. Scholars such as Schattschneider
(1975) and former President Wilson did not find a system of district-based forces in the form
of preferences or interests normatively good for government since it prevented members from
coordinating on national level policy. This is primarily due to the fact that the local politics
median voter is not the same as the national median voter.
12
In empirical support of the
Responsible party model, Ansolabehere, Snyder, and Stewart show how House candidates
do not converge to the district median, though more liberal (conservative) candidates run in
more liberal (conservative) districts, so there is responsiveness in this way (2001). However,
another study found that voters appear to punish candidates who vote too often at either
ideological extreme, using ADA scores (Canes-Wrone, Brady, and Cogan 2002).1
In all three of these models of representation the vote is the primary mechanism by which
members are tied to their constituents, be they preferences or interests. Stimson argued how
citizens may act in a reactionary or thermostatic way when voting for elected officials in an
attempt to keep them in line with their interests (2004). Public opinion is always moving
but the pace of it may vary, with opinion changing rapidly in times of crisis or in judging
an incumbent's performance in the months prior to an election. This agrees with Fiorina's
argument that retrospective evaluations of performance drive vote choice (1981). However,
politicians may rationally anticipate opinions prior to their elections (Stimson, MacKuen,
and Erikson 1995).
Politicians reconcile their own ideal point for policy with that of the
median voter in their home district, understanding that they cannot drift too far away from
the median in their behavior. "Hardly indifferent, these politicians are keen to pick up the
faintest signals in their political environment. Like antelope in an open field, they cock their
ears and focus their full attention on the slightest sign of danger" (Stimson, MacKuen, and
Erikson 1995, 559).
Representatives must be mindful of the opinions of the people who
allow them to keep their jobs. With the electoral goal being the necessary antecedent to all
other goals, representatives rationally respond to the preferences or interests of their constituencies. Of course elected officials may feel a degree of social pressure to respond to their
constituents in addition to the electoral incentive (Glynn et al. 1999, 303).
Voters also tend to elect people like themselves, so representation may be symbolic or descrip'This study used the presidential vote as a catch-all proxy for district public opinion. Presidential vote
share, while commonly used, is not an ideal measure since it not possible to disaggregate public opinion into
individual issues to examine the relationship between district level preferences and the voting behavior of
their representatives.
13
tive (Glynn et al. 1999). This may also be reinforced by the representative's presentation
of self. As an example of a person-to-person homestyle Fenno includes the following quote,
"if a man takes a bite of your chewing tobacco - or better still if he gives you a bite of his
chewing tobacco - he'll not only vote for you, he'll fight for you" (1977, 900). By sharing
a bite of chewing tobacco, the representative is presenting himself as an equal. This is an
activity one might do with a neighbor or friend, which probably made this constituent more
likely to trust that this candidate would represent his interests. This type of homestyle is
more common in homogeneous districts where it is easiest to relate to a common culture
or demographic. However, symbolic representation may be more than just a way to garner
trust, as it may also shape the way policy is made. A representative who descriptively looks
like his constituents may be more likely to act in their interests.
How Might Differential Responsiveness Persist in a Democracy?
The demographic makeup of Congress is not descriptively representative of the American
public. Some studies assert that the demographic makeup of the legislature matters both
in terms of representing the needs and preferences of constituents.2
In a randomized field
experiment in India by Chattopadhyay and Duflo, areas that had more women Village Council members in head positions were found to have different types of public goods provided
(2004). They found that female leaders invest more in infrastructure that is directly relevant
to the needs of the women in the village. Additionally, a ruling class that does not look like
the people it represents may send messages to voters about efficacy, such as who belongs in
the political process (see working paper by Posner and Kramon on how having a coethnic
political leader positively affects primary school attendance in Kenya).
This may suggest that individual senators vote according to their personal preferences. A
2
Much of the research on descriptive representation is in the field of comparative politics. A small sample
of the vast literature includes: Posner 2005 examining ethnic politics in Zambia; Matland 1993 studying
female representation in Norway; and Chattopadhyay and Duflo 2004 examining female representation in
India.
14
U.S. study found that legislators with more daughters had more feminist voting records
(Washington 2006). Similarly, those who come from more privileged economic backgrounds
are more likely to shape policy in favor of the wealthy. Carnes found evidence that previous professional histories are related to a member's voting record (2011). He found that
representatives with different occupational backgrounds, coming from similar districts, vote
differently. A congress with members from poorer backgrounds would therefore be expected
to adopt politics more in line with the interests of the lower and middle-income groups. 3
However, the upper class bias of Congress does not vary over time, while policy responsiveness does (see Gilens 2011). This suggests that the responsiveness gap cannot be entirely
explained by personal preferences.
Some scholars have suggested that an association between high-income voters' opinion and
senators' votes might be due to the fact that the public adopts the views of its policy-makers.
Since high-income constituents are more likely to read the newspaper or watch the news,
they receive these elite cues, aligning their opinions with elite behavior. Page and Shapiro
believe that collective deliberation relies on such cue-taking from the press and public figures
(1992). Zaller agrees, arguing that individuals rely on elite leadership cues for topics beyond
their understanding (1992, 14). Stimson argues that a small minority of interested members
of the public may be the opinion leaders for the rest of the population (2004).
However, Druckman and Nelson provide evidence against elite cue-taking with their study
that found that conversation can mitigate such influence (2003). When members of a group
received a news story with different framing and then engaged in conversation, the framing
effects went away. This may be taken as evidence against this alternative causal story. As
long as alternative frames are accessible, elite influence is muted and the public is freer to
develop different opinions.
Political resource theory asserts that higher-income individuals are more politically engaged
3 Carnes found no association between members' voting records and their outside income or wealth (2011).
15
since they have the greater resources of time, money, and education. The rich have a greater
ability to make campaign donations, which can buy them influence (Brady, Verba, and
Scholozman 1995). This is particularly true of those at the the very top end of the U.S.
income bracket. A survey of the top one percent in Chicago found that two-thirds of respondents reported contributing money to politics compared to just fourteen percent of the
general population (Page, Bartels, and Seawright 2013).
Given this differential political
engagement of those at the very top, examining their opinions at the national level would
be especially important in the study of this question. When this participation is stratified
along socio-economic lines, "participatory distortion" exists whereby those who participate
(wealthy constituents) send biased signals to policy makers and these signals in turn result
in biased policy outcomes (Verba, Schlozman, Brady 1995, 468), presenting a problem for
democracy.
Hall and Wayman (1990) found that members of Congress are more responsive to organized
business interests than the unorganized masses, even when voters have strong preferences
and the issue is considered important. Organized political action groups use targeted contributions to the members and chairs of particular congressional committees who are in
the best positions to promote their interests (Romer and Snyder 1994). These groups will
specifically add or drop members who change their committee assignments to areas that are
no longer relevant to the interest group. Larger organized interest groups, like a regional
steel producer, may be able to offer politicians larger financial contributions, but if they
can also guarantee more strength of direct voter support (from their employees), then their
contributions are scaled down (Bombardini and Trebbi, Working). This may be marshaled
as further evidence that the wealthy and monied interests have the capability to disproportionately influence national policy. While collective action by large groups is difficult to
achieve even when there are shared interests, small groups are better able to organize and
press the government for policies that support their interests, sometimes at the cost to the
majority (Olson 1965). Small groups are able to organize because they are brought together
16
by selective incentives or particularlized benefits. In a large group, because all share in the
spoils there is an increased incentive to free ride and not participate, which could leave the
interests of the middle class and poor may be largely ignored.
Mayhew argues that since there are high information costs for constituents to faithfully
keep track of their representative's voting record, there may be occasions such as those described by Hall and Wayman (1990) where members act against constituent interests when
it may further reelection prospects in another way.
One such way would be responding
to monied interests who promise to support future campaigns.
Mayhew's representatives
are vote-maximizing agents, not those who seek programmatic impacts. Fenno discusses a
member's constituency as nests of concentric circles, where the innermost circle, his personal
contacts, are his strongest supporters (1977). This is then followed by the primary, reelection and geographic outer circles. The representative's first responsibility is satisfying the
demands of his inner-most circles since without them he cannot be reelected. These inner
circles are disproportionately occupied by the wealthy and other monied interests.
The arguments concerning descriptive representation, elite cue-taking, and differential allocation of political resources and ability to organize can all at least partially explain why
certain segments of the population, such as the wealthy, may be able to exert more influence
over policy-makers.
More research must be performed in order to be able to conclusively
adjudicate between these competing explanations. The task of this paper, however, will be
to examine the nature of the relationship between public opinion and policy making, not to
test the causal mechanism.
Differential Responsiveness By Income
As already discussed, there are many scholars who have examined questions exploring the
relationship between public preferences and government policy-making. However, there are
17
only a few authors who have directly examined the question of differential responsiveness of
legislators to their constituents by income group. In his book, Unequal Democracy, Bartels
examines this question using Poole/Rosenthal NOMINATE scores and several key roll-call
votes of senators (2008).
He finds that senators were consistently more responsive to the
opinions of high-income constituents, however these conclusions might have been driven by
the fact that he selected votes that would generate results conforming with his argument.
Using the same data, I successfully replicated the part of Bartels' study that was examining
specific roll call votes, but in more recent roll call votes (from 1999-2004) I found his key
statistically significant finding regarding responsiveness to the high income on the issue of
minimum wage may have been outlier.' Though much like Bartels found, for the issues of
minimum wage and abortion, very few of the estimated effects were statistically distinguishable from zero. In Figures 61 to 66 in the Appendix I show, using the same estimation
technique, how the probit estimates on the votes from my study compare to Bartels' probit estimates on minimum wage and abortion.'. My survey data, though large, were not
representative at the state level and instead of a general measure for ideology I used the
issue-specific ideological measure.6
Erikson and Bhatti replicate the part of Bartels' study using NOMINATE scores, but they
were not able to reproduce his findings using two more recent datasets (forthcoming). They
believe this was likely due to insufficient variation in opinion across income groups for the
more recent time period of their study. While it may be possible that senators are only
responding to their high-income constituents' preferences, if these preferences look like the
preferences of the lower-income constituents then there is no way to measure this differential
effect. It is also possible that Erikson and Bhatti's measure of policy preference does not
adequately capture the variation across income groups. A more direct ideological measure,
4
None of the point estimates for high income opinion on the thirty-four recent votes fall within the
confidence interval for the minimum wage vote Bartels selected.
5 Bartels' estimate for minimum wage was based on votes for the Minimum Wage Restoration Act of 1989
(HR2) and his estimate for abortion on the Harken motion to table the Armstrong amendment (HR5257)
6
Though this was rescaled to conform with Bartels' ideology measure on a -1 to 1 scale of liberal to
conservative.
18
relating to specific policy areas, would have perhaps produced different results.
Bruner Ross, and Washington (2011) matched legislative and constituent votes on ballot
initiatives in California and also did not find that the rich are better represented than the
poor. In their study as well, the views of the rich and poor were highly correlated, and as
such, both groups were found to be well-represented.
Tausanovitch found that the expressed preferences of the wealthy are better represented
in both the Senate and House than the poor using a continuous measure of political preferences based on responses to policy questions (working). He believes this is the reason Erikson
and Bhatti are not able to find results in their study using just a 5-point ideology measure.
The continuous measure of preferences provides more information about individuals in policy
space. The dependent variable consists of all roll calls taken in both chambers from the 108111 Congresses. However Tausanovitch has not explored the effect of income group opinion
across issues. He also rather arbitrarily picked his income categories, classifying those with
household incomes less than $25,000 as poor and respondents with household incomes over
$100,000 as rich.
Gilens has attempted to answer a similar question by looking at how responsive the government as a whole is to the populace by income, instead of just examining the responsiveness of
senators, like Bartels (2012; 2008). Gilens looks at how this differential responsiveness varies
across the policy areas of foreign policy & national security, social welfare, economic policy,
and religious issues, running twelve separate logit regressions. His dependent variable is the
policy outcome, coded with a 1 if the proposed policy change took place within four years of
the survey and 0 if it did not. He also uses opinion measures relating to specific issues, rather
than using a general ideological measure, like Bartels. Gilens found that the responsiveness
gap between the wealthy and the poor was negligible on issues of social welfare and largest
for foreign policy.
19
While roll-call votes neglect to account for the power of agenda-setting in determining which
issues are considered and which are ignored, I argue that it is first necessary to examine the
responsiveness of one body of government at a time before looking at interactions between
bodies necessary to enact policies. Perhaps the failure of representation does not stem from
any one branch, but is distributed amongst all of them. Alternatively, it may be that one
body of government is responsible for the bias that Gilens finds in policy outcomes. It may
also be by the collaborative efforts between the bodies of government where representation
fails. It is important to analyze each separately so that this differential responsiveness may
be clearly attributed to its source.
Informed by the choices of Bartels, Tausanovitch, and Gilens, I examine the differential
responsiveness of senators to low, middle, and high-income opinions across the issue areas of
economic policy, social policy & civil rights, foreign affairs, and healthcare. I also adopt the
perspective of the Instructed-delegate model of representation and expect to find evidence
of a positive relationship between constituents' opinions and the voting behavior of their
representatives.
Overview of Project
I begin by using a large and more diverse set of data sources and surveys to allow me to
examine public opinion at the issue-specific level. Since these surveys are not representative
samples of each state, as the ANES survey data used by Bartels, I perform multilevel regression and poststratification (MRP) analysis. This approach allows me to more accurately
represent public opinion at the state level and for specific income groups by combining survey and census data (see Warshaw and Rodden 2012). Previous work has not been able to
answer this specific question due to insufficient sample sizes in national surveys. I am able
to connect the vote of a senator to his or her constituents' opinions across a broad range of
issues. I then examine these opinions at the fine-grained level of income within each state.
20
Once I have these estimates of income group opinion by state and issue area, I use them
as the independent variables of interest in the second stage.7
For the second stage I first
estimate the ideal point (on the liberal to conservative scale) for each senator and for each
issue area. I then take these estimates and use them as the dependent variable in a standard
OLS regression with the MRP estimates of opinion and a dummy variable for whether or
not the senator is in the Republican Party.8
A discussion of the results and substantive
implications follows.
Data
I use IPUMS for the necessary census data (5% sample of population from the year 2000)
like income, race and percent urban for each state. I then combine 3 large-N surveys to
measure pubic opinion across a variety of issues: the 2000 and 2004 National Annenberg
Election Survey and the 2006 Cooperative Congressional Election Survey (CCES). I chose
twelve issues where I could find similar question wording across at least two different surveys. This yielded an average of 56,707 responses per issue area. The issues are minimum
wage, social security, estate tax, abortion, gun control, gay marriage, school vouchers, environment, trade, war in Iraq, immigration, and health insurance. Responses are coded as
1 if the position is supported, and 0 otherwise.9
I performed factor analysis to make sure
that these questions across different surveys were statistically similar, such that variations
across each of these surveys mainly reflect variations in fewer unobserved variables, namely
public opinion. Across most issue areas, the proportion of the variance explained by the first
7
Unfortunately, I am not able to propagate the uncertainty of these Stage 1 estimates into the second
stage. However, much of the current MRP literature does not do so either.
8I also use another second stage estimation technique, a generalized estimating equation model for correlated data. In this model the dependent variable consists of all roll call votes in a given issue area, which are
regressed on the same independent variables used in the ideal point regression. I then compare my results
from each of these estimation strategies and discuss their substantive implications. The results from this
method and interpretation are available in the Appendix
9
Following Lax and Phillips (2009) I coded "don't know" as 0. Arguably, one might assert that it indicates
a deference to the status quo. Missing responses were dropped.
21
component was greater than 0.7, with the exception of gay marriage and Social Security.1 0
As the respondents were not the same across surveys I compared respondents by income
group, sex and race. The results suggest not only that public opinion was stable across the
time period of these three surveys, but also the questions across surveys are comparable
enough to be combined. I also performed factor analysis by each respondent for questions
across different issue areas for each of the three surveys. I found that the issue areas were
sufficiently distinct to warrant their individual treatment. Comparing each issue area against
each other, I found the proportion of the variance explained by the first component was less
than 0.7."
The surveys identify the home state of each respondent so they can be used in MRP analysis."
Each respondent also has associated demographic information such as gender, race,
education, and income. I examine three sessions of the Senate (106-108), which corresponds
to the time frame, 1999-2004. In the 2000 election the Senate went from a Republic majority
to being evenly split between Republicans and Democrats. The Senate then went back to
being a Republican majority for the 108th Congress.
For the issue area and ideological direction of the senators' roll-call votes, I use data from
the Policy Agendas Project. This organization uses NOMINATE to score the ideological direction of a given vote, which has been a generally accepted metric in the American political
science literature."
Using this dataset and the roll call datasets from the Senate website, I
coded the Senate responsiveness dependent variable. I coded the votes such that a 1 corresponded to the conservative position on an issue and a 0 corresponded to the liberal position.
I recoded the survey responses to questions asked in the liberal direction on minimum wage,
gun control, environment, and health insurance, so that all of the responses were coded in
10
Data available upon request.
"Data available upon request.
12
Only the CCES data contains respondents from Alaska and Hawaii.
3
1 Poole and Rosenthal apply a spatial voting model to congressional roll-call data to estimate ideology in
Congress. Legislators ideal points are then estimated within choice spaces on various dimensions.
22
the conservative direction to match the votes. Votes which could not be classified as liberal
or conservative based on their NOMINATE score, which mostly included unanimous votes,
were not used in the estimation.
The reported top household income category for all three surveys was "more than $150,000".
This category, alone, was made the high income category. The low income category includes
all respondents with household incomes below $25,000 and the middle income category captures those earning $25,000 - $150,000.
While household income is not an ideal measure,
as I am unable to adjust for the number of people in the household, these thresholds are
sufficiently extreme that a poor household is unlikely to be misclassified as a rich household
and a rich household for a poor one. To my knowledge, this is the first study measuring
income group public opinion where the high income threshold is set as high as $150,000, representing the top 10 percent of the US household income distribution. This is likely because
previous studies were not able to correct for the scarcity of data for individuals at the top.
MRP is uniquely suited for estimation where limited data is available. I also run the same
set of analysis with the more equal income breaks (equal in terms of survey categories) of
keeping low income as less than $25,000, but setting middle income as $25,000 to $50,000,
and high income as above $50,000 (above the median for the time period). These are the
same income breaks that Bartels uses in his study. Although both Bartels and Bhatti find
their results are unchanged when using different income thresholds, they have not tried this
drastic a change. A recent survey of top earners in Chicago found they are more socially
liberal and economically conservative than others (Page, Bartels, and Seawright 2013). This
suggests that a national study examining the preferences of the top tenth would be an important contribution to the field. In order to test how responsive senators are more generally
to their constituents, I also run the analysis without any income breaks, with state opinion
estimated by issue area.
23
Method: MRP and Poststratification
MRP models each response as a function of its respective demographic and geographic characteristics. It assumes that effects within a group of variables are related to each other
by their hierarchical structure. MRP applies Tobler's Law such that individual observations that are close together in space or on certain dimensions are assumed to have more in
common. The multilevel model pools group-level variables toward their mean, with greater
pooling when the variance is small and more smoothing when there are fewer respondents in
the group. MRP produces more accurate and robust estimates than other forms of estimation like disaggregation, particularly in estimating opinion in small states and for the very
wealthy, which have fewer respondents (Warshaw and Rodden 2012).
In the model below I estimate each respondent's preferences as a function of his or her demographic characteristics and state (the senator's constituency). Individual i indexes a, r-g, s
for age (younger than 27, 28-37, 38-57, 58-77, 78+) race & gender (white male, black male,
hispanic male, other race male, white female, black female, hispanic female, other race female), and state, respectively.
-i,,:,,b
represents the fixed effects for the interaction between
each income group and the percent living in an urban area. The model allows individual
attributes to predict state and income level ideology. It also manages to correct for some of
the shortcomings of the survey sample. Surveys tend to oversample older respondents and
this might particularly be true for those in the top income group. This model is able to
compensate by re-weighting opinion estimates to better reflect the actual population such
that the under-sampled populations, like young high income respondents (captured by age
and income covariates), receive a greater weight. Additionally, the model incorporates both
within and between state geographic variation.
24
Hierarchical model for individual respondent:
Pr(yi
=
ageU ~
Cla"
+ yincurb + arace-,ender
1) =og-O
(,
agA)a
ar-gender ~a(0,
astate)
age
= 1, ...
)5
,
,cegender)
rg
8
1,
Following previous work with MRP, I assume that the effect of demographic factors does not
vary geographically (see Warshaw and Rodden 2012). In the state level of the model, each
state is allowed to vary with its own intercept. The state effects are modeled as a function of
its region of the country, the state's average income, and the percent of the state's residents
living in urban areas.
~
V7low-inc
(state
- income,
+
ymid -
income, + -yhi-inc - income.,
2
)
The next step is poststratification where the estimates for each respondent demographic
geographic type (7rt) is weighted by the percentages of each type in the actual state populations (Nt).
~mrp
states
tEst
=
Twt
(EtEsNt
There are 51 states (including D.C.) with 120 potential demographic types in each, which
yields 6,120 possible combinations of demographic and state values but not all of these combinations were observed in the data.
For each state and income group combination, the average opinion over each type is calculated. These estimates become the independent variables of interest for the second stage
of the estimation.
25
Figure 1
Minimum Wage
UT
NH
WY
AZ
MT
AK
OH
GA
U
0
A
0
*A
0
*A
0
A
**A
**A
U
*A
U
*
A
*
0 A
U
q
A
0
*A
U
0 A
N
*
A
U
* A
N
*
A
*
*A
E
4
0*A
gA
0
U
0
A
*A
0
0*A
*
0
A
**A
N
* A
U
*A
0
* A
*
* A
N
0 A
u
*A
0
0 A
0
*A
*A
0
U
*
A
*
9
A
*A
0
S S
U
a
m
*A
U
*
A
N
0 A
ID
HI
WI
RI
Co
OK
TX
Sc
ME
KS
NE
FL
MN
MO
OR
LA
SD
TN
(n ND
IN
WA
KY
MI
IA
PA
NV
CA
NC
VT
WV
AL
NM
CT
MS
DE
AR
NJ
MD
VA
IL
MA
NY
DC
(56,260,16)
(19,154,14)
(17,63,2)
(121,667,58)
(41,106,6)
(9,89,8)
(243,1147,48)
(144,902,79)
(46,174,7)
(14,62,3)
(160,660,32)
(18,98,0)
(85,483,38)
(80,301,18)
(334,2033,183)
(75,322,19)
(44,164,4)
(89,331,8)
(44,129,5)
(309,1685,125)
(109,528,53)
(138,598,30)
(118,508,30)
(70,288,19)
(30,92,3)
(120,496,27)
(20,68,1)
(137,595,19)
(150,823,57)
(102,339,10)
(234,1134,67)
(78,341,14)
(229,1161,67)
(42,283,22)
(396,2601,367)
(160,719,46)
(17,53,4)
(78,196,6)
(92,344,20)
(47,227,17)
(39,266,42)
(49,135,7)
(16,82,6)
(104,239,9)
(64,651,80)
(52,541,74)
(82,691,102)
(165,1125,118)
(63,429,56)
(242,1347,146)
(1,3,1)
oA
S
S
*
A
0
N
*
a
a
m
S
w
0.0
0
A
0
A
q A
*
A
0 A
* A
*
A
0.2
0.4
0.6
MRP Estimated Opinion
26
0.8
1.0
Figure 2
Minimum Wage
(Equal Breaks)
U
U
UT
NH
WY
ID
AZ
ME
MT
OH
AK
OK
SD
MO
U
A
A
A
A
0
0
U
us
A
A
A
A
A
A
A
A
*
U
0
0
U
U
me
OR
*
TN
KS
FL
ND
WI
MN
HI
RI
NE
U
U
*
*A
*
*
U
U
0
*
*
U
N
0
S
*
*
*
*
*
SC
*
IN
CO
U
*S
(D
*
o
*
C
C
SVT
5
& WV
TX
LA
IA
GA
KY
AL
MS
MI
CA
NM
NV
PA
VA
WA
NC
CT
AR
DE
NJ
IL
MA
NY
MD
DC
U
U
*
*
*
*
*
U
U
a
U
U
A
A
A
A
A
A
C
C
*
* A
*
*
U
0
a
*
*
*
U
0.0
A
A
*
*
U
U
U
0
S
0
U
U
N
CA
*
*
*
A
A
A
A
A
A
A
A
A
A
A
A
A
A
A
A
CA
*
U
(56,186,90)
(19,84,84)
(17,46,19)
(46,132,49)
(121,384,341)
(44,110,58)
(41,79,33)
(243,757,438)
(9,49,48)
(80,219,100)
(30,66,29)
(138,404,224)
(118,325,213)
(120,344,179)
(89,238,101)
(309,1049,761)
(20,48,21)
(160,434,258)
(109,315,266)
(14,32,33)
(18,59,39)
(44,97,37)
(75,207,134)
(137,400,214)
(85,257,264)
(17,38,19)
(78,142,60)
(334,1170,1046)
(70,191,116)
(78,247,108)
(144,517,464)
(102,217,132)
(92,238,126)
(49,89,53)
(234,707,494)
(396,1370,1598)
(47,121,123)
(42,167,138)
(229,734,494)
(82,373,420)
(150,469,411)
(160,451,314)
(39,138,170)
(104,184,64)
(16,52,36)
(64,317,414)
(165,641,602)
(63,219,266)
(242,782,711)
(52,256,359)
(1,2,2)
*A
0
*
*
*
A
A
A
A
A
A
A
A
A
A
0.2
0.6
0.4
MRP Estimated Opinion
27
0.8
1.0
Figure 1 and Figure 2 show the estimated opinions for increasing the minimum wage
of income groups by state.14 Figure 1 defines income as the top ten percent, and Figure 2
has more equal income breaks as discussed in the previous section. The square represents
the poor, the circle represents the middle-income, and the triangle represents the wealthy
respondents. The numbers on the right are the sample sizes for the poor, middle income,
and rich respectively in each state. A one on the X axis corresponds with the conservative
opinion that the minimum wage should not be increased, while a zero denotes an approval for
the increase. The figures of the MRP estimates for the other eleven issue areas are provided
in the Appendix as Figures 2-24.
Interestingly, the top ten percent are more liberal than when high income is defined as
above $50,000. However, they are still less likely to support the increase in minimum wage
than the middle or low income respondents. Low income respondents are most likely to
support the increase than any other group.
The top 10 percent are also more liberal on the issue of abortion than the other high income
group. The high income group is overall more likely to be against placing restrictions on
abortion while the low income group holds the most conservative position. This may be because lower income individuals are typically more religious or traditional than upper income
groups.
The high income are less likely than the poor or middle class to agree that the government should do more to protect the environment. The poor are most liberal on this issue.
This is likely because the high income fear government encroachment on their business interests.
Self-interestedly, the high income favor eliminating the estate tax and this is much more
14
Included in the Appendix is this same figure but for MRP estimates of pooled constituent opinion (no
income group breaks).
28
apparent for the top ten percent. The low income are most likely to oppose the elimination
of the estate tax.
There does not appear to be a clear income group level opinion for the issue of gun control,
regardless of income breaks.
This may be more determined by where in the country the
respondent lives.
Both the high and low income respondents are more liberal on the issue of gay marriage
than the middle income. The reason for this is less clear.
High income respondents are less likely to support a federal health insurance plan than
the middle and low income groups. Low income respondents are most in support of the program. The top ten percent appear more conservative on this issue than the more inclusive
high income group.
The high income are much less likely to want to restrict immigration than the low and
middle income groups. The top ten percent are much more liberal on this issue than the
more inclusive high income group. This accords with the Hainmueller and Hiscox's finding
that the highly educated are more likely to support immigration than those with less education (2010), as income and education are highly correlated.
The low income respondents are much less likely to believe the situation in Iraq was worth
going to war over. This may be because they more acutely felt the costs, since low income
youths are much more likely to enlist. Interestingly, the top ten percent are more liberal on
this issue than the more inclusive high income group.
The low income respondents are least likely to support privatizing Social Security, while
the high income are most in favor of its privatization. Again, we see the top ten percent is
29
more liberal on this issue than the more inclusive high income group.
A clear distinction across the income groups does not arise on the question of whether
school vouchers should be an option. However, the top ten percent are slightly more likely
to prefer the privatization than the other two groups. This same pattern does not appear
for the income groups with equal breaks.
The top ten percent are much more likely to be pro-free trade, while the low income respondents are most against free trade. The more inclusive high income group is also more
in favor of free trade than the other groups but the disparity is less substantial.
There are apparent differences in opinion across income groups for many of these issue
areas. Interestingly, the top ten percent are more liberal on issues like minimum wage, abortion, gay marriage, Social Security, and the Iraq War than the more inclusive high income
group. They are also more conservative than the inclusive group on issues like the estate
tax, school vouchers, and free trade." I found that these differences between income group
opinions were statistically significant. The difference in means between the top ten percent
and the more inclusively defined high income group opinions were statistically distinct as
well. Tables presenting results from the t-tests are provided in the Appendix.
As a validation test of the MRP estimates, I checked to see how correlated they were with
Tausanovich and Warshaw's state level ideology measure from their 2013 paper, "Measuring
Constituent Policy Preferences in Congress, State Legislatures, and Cities". Their estimates
were produced using Item Response Theory and were found to outperform previous measures
of citizens' policy preferences.
In their paper, Tausanovitch and Warshaw create a measure for ideology at a state level,
1 5Though "free trade" would technically be "liberal" in the classical economic sense of the term, here by
"conservative" I am referring to an alignment with the Republican Party.
30
aggregated across all issue areas and income groups. In contrast, in this paper I develop
my estimates by issue area and a number of different aggregations of income level. Figure 3
shows one example of how my state-level opinion MRP estimates by issue area are positively
correlated with Tausanovitch and Warshaw's general estimate of state opinion when looking
at all income levels in aggregate. The validation plots for the other eleven issue areas are
contained in the Appendix as Figures 14 to 24 with correlations included in the legend of
each plot.
Figure 3
Minimum Wage
(No Income Breaks)
U
U
-r
Co
* All Incomes (Cor. 0.558)
0.12
0.14
0.16
0.18
0.20
0.22
0.24
0.26
State-level Opinion, MRP Estimates
For ten of the twelve issue areas, the relationship between the two measures are positively
correlated.16 The remaining two issue areas, school voucher and free trade, despite having a
negative correlation, show little variation in the opinions across states for these issues (10%
of the scale, as opposed to ranges more typically around 30 to 40%); as such correlations are
not a meaningful measure of performance. This overall pattern of positive correlation is to be
expected despite the fact that Tausanovitch and Warshaw's measure was not issue-specific,
16In particular, middle-income group opinion was consistently found to be the most highly correlated
income group opinion with their measure largely because this group constitutes majority of the population.
31
as both measures are obtained using survey responses from the same time period and all
three surveys used in this paper were also used in Tausanovitch and Warshaw's study (in
addition to six others).
Figure 4
Minimum Wage
Minimum Wage
(Equal Breaks)
Ln
0
o
*
Ci
.Q
A
AA
AA AL
.2
AL
AL~
A
o
- /
A
*A
A
A Lw
LnAe
Cr
o1
0
LA
A
1P
S
*1
A
Low Income (Cor 0.200)
" Midde income (Cor. 0.708)
A High icome (Cor. 0.121)
A
o
II
0.05
0.10
0.15
I
1
0.20
0.25
0.30
*Low Inome (C r .0.189)
* Middle kcome (Cor. 0.45)
A High Income (Cor. 0.559)
0.35
0.10
0.05
0.15
Minimum Wage
(No Income Breaks)
0
0
(a
0)
U)
P
0
0
*
U
U
0
0
0
w
All
_
0.12
0.20
0.25
State-level Opinion, MRP Estimates
State-level Opinion, MRP Estimates
0. 14
0.16
State-level
0.20
0.22
Opinion, MIRP
Estimates
0.18
32
Incomes (Cor. 0. 715)
0.24
0.26
0.30
0.35
I further examined the correlation between my MRP estimates and state weighted averages of the raw survey data in order to better validate in an issue specific manner, especially
for school voucher and free trade in which the low variation across states made it difficult
to compare to Tausanvitch and Warshaw's more general measure. Figure 4 presents this
relationship for the issue area of minimum wage. The raw data validation plots for the other
eleven issue areas are available in the Appendix as Figures 26 to 36. The simpler disaggregated measures were often used in the literature to date examining legislative responsiveness
to public opinion (Brace et al. 2002; Clinton 2006; Miller and Stokes 1963), however given
the paucity of the number of respondents, estimates of opinion for subpopulations have a
large amount of uncertainty (Achen 1978; Lax and Phillips 2009; Warshaw and Rodden
2012). However, the MRP estimates should still be positively correlated with these simpler
measures when the sample size is large, and there is evidence of this linear relationship across
the issue areas.
As expected, MRP estimates of constituent opinion in aggregate by state is highly correlated
with the simpler estimates across issues, indicating a positive linear relationship. Here the
sample sizes are large since the opinion of particular income groups is not being estimated.
This increases the precision of the state weighted averages, making them more similar to the
MRP estimates of state opinion. The ten issue areas found to be positively correlated with
Tausanovitch and Warshaw's measure are also correlated with the raw weighted averages,
with some issues having correlations well above 0.9. As with my MRP estimates, the raw
weighted averages for the remaining two issue areas also demonstrate low levels of variation
in opinions across states. This confirms that the low variation is not unique to MRP estimation. Despite having little variation, the MRP estimated opinion regarding free trade was
well correlated with the raw weighted averages, and the MRP estimated opinion regarding
school vouchers was one of the most highly correlated issue areas with the raw weighted
averages. This is further evidence that the first-stage MRP estimation was successful.
33
Since averages are highly variable in small sample sizes across issue areas and income, it
is best to separately consider the performance of the middle income MRP estimates as this
is the largest subpopulation. 1 7 Compared to the other income groups, middle income opinion estimates are most highly correlated with the simpler weighted averages of opinion, both
when middle income is described as those earning $25-$50,000 as well as when those earning
$50-$150,000 are also included. These state-level middle-income opinion MRP estimates are
as well correlated with the middle-income raw weighted averages as observed at the aggregate state-level. This, again, validates the first-stage estimation.
However, when comparing the MRP estimates for high and low income opinion to the raw
weighted averages the correlations are not as strong. This is largely due to the increased
variability that results from these sample sizes being smaller than the middle income group.
For low-income opinion, the correlation between my MRP estimates and the simple weighted
averages are positive across all issue areas, although in some cases this correlation is very
weak.18 For high-income opinion, the correlations between the MRP estimates and the simple weighted averages were better for some issue areas when high income was defined as only
the top 10%, while for others correlations improved when high income was defined more
broadly, even though it included a more economically diverse set of respondents (increasing
the variation); in all cases though, the correlations were not as strong as they were for middle
income. This is in fact, specifically why I have chosen to use MRP. Previous scholarship (Lax
and Phillips 2009; Warshaw and Rodden 2012) have shown that MRP results in estimates
that are more reliable with smaller errors. This is particularly true when dealing with sample
sizes that are small, like high and low income respondents to a set of national surveys being
divided by state.
In general, weighted averages of income group opinion by state result in highly variable
17
Middle-income respondents also tend have have lower variance in their responses compared to the lowincome group more generally. This is often credited to the higher levels of education associated with the
middle class.
18
The definition of the low income group does not change in the two different aggregations by income level.
34
estimates due to insufficient sample size. As would be expected, middle income opinion,
the income group with the largest share of the population per state, was found to have the
highest correlations with my MRP estimates across the issue areas. The high positive correlations of aggregate state level opinion with the corresponding MRP estimates provides the
strongest support for the validity of the first-stage estimation.
Method: Ideal Point Estimation and Regression
Now that I have constructed the independent variables of interest for state-income group
opinion, I next construct the dependent variable. Using the CVP estimation procedure created by Fowler and Hall (working), I estimate the ideal point for each senator for each issue
area. I coded the senators' vote as liberal or conservative based on their NOMINATE score
and a conservative vote was set equal to one and zero otherwise. This eases the interpretation with a positive coefficient corresponding to the conservative position on the issue by
the respondents. I then regressed legislator fixed effects and bill fixed effects (to control for
the content of each bill, which is important for comparing legislators who did not vote on
the same subset of bills) on the recoded vote variable for each issue area. I used the median
legislator as the omitted category for the legislator fixed effects. This means the coefficient
on a legislator's fixed effect is equal to the probability (relative to the median legislator)
that the legislator votes in the conservative direction on any given bill. The ideal point
can change depending on the senator's voting record in any given issue area. For instance,
Republican Senator from RI, Lincoln Chafee's ideal point for minimum wage places him just
barely in the liberal direction, while his ideal point for abortion places him in the conservative direction.
I chose this method for ideal point estimation because it performs better with a smaller
set of votes than other estimation techniques. Since I am looking within specific issue areas,
pooling the votes for the time period 1999-2004 still only leaves me with a small number
35
of votes. With only 35 votes and over 100 legislators the results are expected to achieve a
0.05 mean absolute deviation (5 percentage points between the "true" ideal point and the
estimated value). DW-NOMINATE would require 100 bills to provide accurate estimates
(Fowler and Hall, working). It was also found to be highly correlated with the first dimension
of DW-NOMINATE.
After estimating the ideal points for each senator for each issue area, I then ran a standard OLS regression of the ideal points on the first-stage MRP estimates for low, middle,
and high income opinion as well as a dummy variable for whether or not the senator was
Republican. I then block-boot-strapped the standard errors, by resampling the senators with
replacement and reconstructing the dataset such that the senators that were included twice
in a sample have all their bills in there twice and the ones the did not get sampled were
left out. I then reran both regressions generating a distribution of new estimates for the
confidence intervals. My results are shown in Table 1 below for all twelve issue areas.
Table 1: CVP Model Regression Table
Intercept
Low Income Opinion
Middle Income Opinion
High Income Opinion
Minimum Wage
0.32*
Abortion Rights
0.56*
Social Security
0.16
Estate Tax
-0.47*
(0.17, 0.51)
(0.37, 0.80)
(-0.55, 0.92)
(-0.94, -0.10)
1.37*
1.67
-0.65
0.34
(0.21, 2.74)
(-0.91, 3.44)
(-1.37, 0.14)
(-1.02, 1.93)
-1.16
-3.41
0.88
0.30
(-3.64, 0.54)
(-5.99, 0.03)
(-1.99, 3.68)
(-2.11, 2.21)
-0.42
1.17
-0.58
0.10
(-1.22, 0.78)
(-0.35, 2.45)
(-2.05, 0.94)
(-0.99, 1.58)
Republican senator
-0.25*
-0.31*
-0.04*
0.12*
Adj R 2
(-0.28, -0.22)
0.61
(-0.36, -0.26)
0.65
(-0.07, -0.01)
0.06
(0.10, 0.14)
0.40
123
118
N
123
Bootstrapped confidence intervals in parentheses
* indicates significance at p < 0.05
122
36
Table 1: CVP Model Regression Table
Intercept
Low Income Opinion
Middle Income Opinion
High Income Opinion
Republican senator
Adj
R2
Gun Control
0.12
Gay M arriage
0.8 7*
(-0.12, 0.30)
(0.72, 1.01)
0.36
1. 39
(-1.09, 1.87)
(-0.65, 2.80)
-2. 75*
-0.54
(-2.94, 2.25)
0.05
(-4.44, -0.21)
-0. 07
(-1.41, 1.13)
(-0.80, 0.55)
-0.17*
-0. 27*
(-0.21, -0.14)
0.36
(-0.31, -0.24)
0. 68
-0.42
(-1.42, 0.70)
Environment
-0.08
(-0.36, 0.16)
-0.67
(-1.46, 0.09)
-0.09
(-1.59, 1.30)
0.99
(-0.44, 2.68)
-0.11*
(-0.13, -0.09)
0.41
118
(-0.16, -0.10)
0.24
118
School Voucher
1 3
118
N
Bootstrapped confidence intervals in parentheses
* indicates significance at p < 0.05
-0.10
(-0.35, 0.20)
-0.15
(-1.24, 1.09)
0.92
(-2.03, 3.56)
-0.13*
Table 1: CVP Model Regression Table
Ira q War
Free Trade
).10
0.13
(-8.18.1 o-5, 0.19)
(-0.11, 0.40)
).02
-4.76*
Low Income Opinion
(-0.3 3, 0.32)
(-7.65, -1.57)
- 0.41
6.79*
Middle Income Opinion
(-1.0 9, 0.21)
(1.63, 11.62)
).28
-1.58*
High Income Opinion
(-2.83, -0.10)
(-0.3 1, 0.95)
-( .05*
-0.27*
Republican senator
(-0.0 7, -0.02)
(-0.31, -0.22)
0.42
Adj R 2
).07
123
123
N
in
parentheses
intervals
confidence
Bootstrapped
* indicates significance at p < 0.05
Intercept
Immigration
0.87
(-0.96, 2.19)
-0.72
(-5.33, 4.85)
-2.15
(-7.46, 2.31)
1.94
(-0.25, 3.88)
0.34*
(0.27, 0.42)
0.31
116
Federal Health Insurance
0.07*
(0.03, 0.11)
-0.09
(-0.41, 0.23)
0.50
(-0.39, 1.44)
-0.59*
(-1.20, -0.02)
0.04*
(0.03, 0.05)
0.21
118
The only positive and statistically significant results were found for senators' responsiveness to low income opinion on minimum wage and middle income opinion on free trade. A
ten percent increase in low income opinion against raising the minimum wage corresponds to
a 13.7 percentage point increase in the likelihood of their senators voting in the conservative
direction. This is contrary to Bartels' finding that senators only responded to the preferences
of the wealthy.
Negative and statistically significant effects were found for the middle income opinion for
37
banning gay marriage, low and high income opinion on free trade, and the wealthy opinion
on federal health insurance. It is not likely that the senators are actively choosing to do
the opposite of a particular income group's opinion. Most likely the senators are voting in a
position that is most dissimilar to this income group opinion for other reasons.
Figure 5
Minimum Wage
N1
0
U.
0
A
0A
A
*
A
A
A
0
A
A
A.
.rm1
gA.
.02
U
a
goo
EU
me
a0-
0
*
MU
C/)
M
U
a
0
opU
Eg
as
0.U .
0
~
*~
6-
AA
A
*
A
U
0
AO
0
a*
A
A
AA
A AAA AA
" A
A
A&
A
A
A
A
A
O
A
0
6
qt
*
A*
0
A
A
A
A
A
0
ELow nom
*Middnle ncom
A
ihIncome
1P
A
A
Ate
*
A
A
A
N
U.
A
4
I
I
I
I
I
1
7
0.05
0.10
0.15
020
0.25
0.30
0.35
-
State-level Opinion MRP Estimates
Figure 5 shows that the MRP estimates are not particularly well correlated with the
senator's ideal points. Figures for the other eleven issue areas are included in the Appendix
(as Figures 38-48) and the same is true for many of them. This would suggest income group
opinion is not particularly predictive of senators' voting behavior.
However, another explanation for the lack of correlation may be that the votes used to
generate the CVP ideal points are not particularly representative of senators' voting behaviors (since this includes strategic and party-line votes). Tausanovitch and Warshaw's
measure for ideology was already found to be well correlated with roll call votes, so as check
I examined their measure's correlation with the roll call votes used in this study and the
38
CVP ideal points generated from them. Figure 6 plots this lack of correlation with their
measure as well. Figures for the other eleven issue areas in the Appendix (50-60) also show
this lack of correlation. This either means that these particular roll calls are not representative of senators' voting behavior on the issues or that by issue, senators are not particularly
responsive to constituent opinion.19
Figure 6
Minimum Wage
U
a
C1
6
0
d
a ON
I
a
IwU
a
am
a
U
aaU
s
0
o
ana
Nw
0
I
I
-0.4
-0.2
I
0.0
0.2
J
0.4
T &W State-level Ideology Measure
Table 2 shows the estimates for all twelve issue areas for equal income breaks.
19 Future study could eliminate all votes except those that are votes on bills, instead of amendments to
bills or motions. I did not exclude them for this current study since scholars on this topic, such as Bartels,
did not do so (see Footnote 4).
39
Table 2: CVP Model Regression Table (Equal Breaks)
Intercept
Low Income Opinion
Middle Income Opinion
High Income Opinion
Minimum Wage
0.38*
Abortion Rights
0.66*
Social Security
0.44
Estate Tax
-0.23
(0.21, 0.57)
(0.46, 0.93)
(-0.08, 0.99)
(-0.93, 0.21)
2.15*
0.64
-0.59
-0.40
(0.71, 3.58)
(-2.92, 3.16)
(-1.96, 0.98)
(-1.69, 1.35)
-2.25*
-2.15
-0.46
1.82
(-4.42, -0.34)
(-7.28, 4.66)
(-5.26, 4.09)
(-2.04, 4.37)
-0.32
0.62
0.20
-1.06
(-1.08, 0.48)
(-3.21, 3.76)
(-2.55, 2.97)
(-3.13, 2.25)
Republican senator
-0.25*
-0.30*
-0.04*
0.12*
Adj R 2
(-0.28, -0.22)
0.60
(-0.36, -0.26)
0.64
(-0.06, -4.79-10-3)
0.06
(0.10, 0.14)
0.41
123
118
N
123
Bootstrapped confidence intervals in parentheses
* indicates significance at p < 0.05
122
Table 2: CVP Model Regression Table (Equal Breaks)
Intercept
Low Income Opinion
Middle Income Opinion
High Income Opinion
Gun Control
0.15
Gay Marriage
0.81*
School Voucher
-0.13
Environment
0.23*
(-0.15, 0.37)
(0.64, 0.98)
(-0.35, 0.09)
(0.05, 0.39)
-0.15
(-2.16, 2.52)
1.50
1.02
(-1.07, 2.49)
-1.00
-0.79
(-2.33, 0.79)
2.25
-1.30*
(-2.41, -0.39)
2.18*
(-5.07, 6.86)
(-2.64, 0.89)
(-0.95, 5.43)
(0.65, 3.82)
-1.57
-1.38*
-1.05
-1.38*
(-5.73, 3.40)
(-2.33, -0.34)
(-2.35, 0.28)
(-2.65, -0.11)
Republican senator
-0.17*
-0.27*
-0.11*
-0.12*
Adj R 2
(-0.21, -0.13)
0.36
(-0.31, -0.24)
0.68
(-0.13, -0.09)
0.42
(-0.16, -0.09)
0.25
118
118
N
118
123
Bootstrapped confidence intervals in parentheses
* indicates significance at p < 0.05
40
Table 2: CVP Model Regression Table (Equal Breaks)
Intercept
Low Income Opinion
Middle Income Opinion
Free Trade
0.15
Iraq War
0.13*
Immigration
0.19
Federal Health Insurance
0.11*
(-0.24, 0.52)
(0.04, 0.22)
(-1.03, 1.02)
(0.05, 0.16)
-3.25
-0.35
1.96
-0.21
(-6.97, 0.19)
(-1.01, 0.26)
(-3.60, 10.23)
(-0.82, 0.35)
3.84
0.66
-4.02
0.48
(-2.14, 10.55)
(-0.81, 2.19)
(-13.36, 3.04)
(-0.70, 1.73)
Republican senator
-0.48
(-2.31, 1.00)
-0.26*
-0.57
(-1.63, 0.45)
-0.05*
1.90
(-0.60, 4.35)
0.34*
-0.60
(-1.42, 0.16)
0.04*
Adj R 2
(-0.31, -0.21)
0.40
(-0.07, -0.02)
0.07
(0.27, 0.42)
0.31
(0.03, 0.05)
0.20
116
118
High Income Opinion
N
123
123
Bootstrapped confidence intervals in parentheses
indicates significance at p < 0.05
The only positive and statistically significant results were found for senators' responsiveness to low income opinion on minimum wage and middle income opinion on the environment.
Here middle income is more narrowly defined than it was in the results for Table 1, which
affects the model's estimates. A ten percent increase in low income opinion against raising
the minimum wage corresponds to to a 21.5 percentage point increase in the likelihood of
their senators voting in the conservative direction. It is also of note that the sign on the
coefficient for high income opinion on the environment became negative and statistically
significant, despite being positive for the income breaks in Table 1. This shows how these
results are sensitive to how income group is defined.
Negative and statistically significant effects were found for the middle income opinion on
minimum wage, the high income opinion on banning gay marriage, and the low and high
income opinion on the environment. Substantively, I do not think this is particularly meaningful.
Figure 7 shows the lack of correlation of the MRP estimates for income opinion with the
senators' ideal points.
41
Figure 7
Minimum Wage
(Equal Breaks)
0*
.
A
~i~tm,,t
AAA
AA
*AA,
*U
AA
*A
*
0
so
r i*
mu
*
tA
*
h
*
a
A
AD.
0
.
m
**
0.
:.
A.
AL~
0
A0
A
*
U
k
A
* A &
A
Ak
AA
A
LA
0
*.
ELowncome
*Middle Income
A High Income
I
I
I
0.05
0.10
A
00A
U
00
A
A
A
A
00
I
I
I
1I
0.15
0.20
0.25
0.30
0.35
State-level Opinion MRP Estimates
Table 3 shows the aggregate constituent opinion estimates for all twelve issue areas (with
no income breaks).
Table 3: CVP Model Regression Table (No Income Breaks)
Minimum Wage Abortion Rights
0.26*
0.71*
(0.12, 0.44)
(0.60, 0.84)
Constituent Opinion
-0.67
-1.05*
(-1.55, 0.03)
(-1.35, -0.73)
Republican senator
-0.25*
-0.30*
(-0.28, -0.22)
(-0.35, -0.26)
Adj R 2
0.59
0.64
N
123
122
Bootstrapped confidence intervals in parentheses
* indicates significance at p < 0.05
Intercept
42
Social Security
0.26
(-0.03, 0.58)
-0.44
(-1.01, 0.11)
-0.04*
(-0.07, -0.01)
0.06
123
Estate Tax
-0.45*
(-0.75, -0.22)
0.72*
(0.29, 1.26)
0.12*
(0.10, 0.15)
0.41
118
Table 3: CVP Model Regression Table (No Income Breaks)
Intercept
Constituent Opinion
Gun Control
0.16*
Gay Marriage
0.82*
School Voucher
-0.03
Environment
0.09*
(0.08, 0.25)
(0.68, 0.95)
(-0.17, 0.10)
(0.01, 0.17)
-0.27*
-1.48*
0.17
-0.01
(-0.51, -0.04)
(-1.85, -1.12)
(-0.10, 0.47)
(-0.30, 0.24)
Republican senator
-0.17*
-0.27*
-0.11*
-0.13*
Adj R 2
(-0.21, -0.14)
0.37
(-0.31, -0.24)
0.67
(-0.14, -0.10)
0.42
(-0.16, -0.09)
0.23
118
118
N
118
123
Bootstrapped confidence intervals in parentheses
* indicates significance at p < 0.05
Table 3: CVP Model Regression Table (No Income Breaks)
Republican senator
Free Trade
0.30*
(0.11, 0.52)
-0.32
(-0.95, 0.27)
-0.26*
Iraq War
0.11*
(0.01, 0.21)
-0.16
(-0.39, 0.05)
-0.05*
Immigration
-0.17
(-1.12, 0.50)
0.16
(-0.72, 1.33)
0.35*
Federal Health Insurance
0.09*
(0.04, 0.13)
-0.32*
(-0.48, -0.19)
0.04*
Adj R 2
(-0.30, -0.21)
0.39
(-0.07, -0.02)
0.08
(0.27, 0.42)
0.32
(0.02, 0.05)
0.19
116
118
Intercept
Constituent Opinion
N
123
123
Bootstrapped confidence intervals in parentheses
* indicates significance at p < 0.05
Figure 8 shows that the senators' voting behavior as represented by their ideal points
is not particularly well correlated with constituent opinion more generally on the issue of
minimum wage. While a relationship is not detected for many of the other issue areas,
senators' voting behavior does appear positively correlated with overall constituent opinion
for the issue of estate tax. Constituent opinion in aggregate is negatively correlated with
senators' voting records on abortion rights, gun control, gay marriage, and federal health
insurance. For these issues, senators may have other reasons or incentives for casting their
vote in opposition to mass opinion. Since a positive and significant relationship is not
observed between senators' votes and aggregate public opinion, it makes the lack of findings
by income group more expected.
43
Figure 8
Minimum Wage
(No Income Breaks)
U)U
0.14
0.18
0.16
0.20
0.22
0.24
0.26
State-level Opinion MRP Estimates
Conclusion
Schattschneider cautioned that the "the flaw in the pluralist heaven is that the heavenly
chorus sings with a strong upper-class accent" (1975, 34-35), however this warning was
unsupported by the data. I did not find evidence that senators' votes are positively and
significantly correlated with the opinions of the wealthy or any income group, looking within
specific issue areas. While Brunner, Ross, and Washington (2011) also did not see an income
group preference, this was because their measures of income group opinion were highly correlated, not because they did not see a correlation between legislative voting and constituent
opinion more generally. In contrast, I found that income group opinions were distinct. Tausanovitch (working), Bartels (2008), and Gilens (2012) also found income group preferences
to be distinct, but unlike their studies, I failed to find a positive association between public
opinion and senators' voting behavior.
As previously discussed, this may be due to the fact that the sample of votes used was
44
too inclusive, resulting in a dependent variable not particularly reflective of senators' behavior. The second-stage results were very sensitive to income group breaks and this inability
to disconfirm the null hypothesis that senators are not responding to a particular income
group was robust across two different second stage model estimations.2 0 Since Tausanovitch
and Warshaw's measure of aggregate state-level ideology was also found to be uncorrelated
with the senators' voting behavior, despite the high positive correlation found in their own
study (2013), this suggests that the sample of votes is somehow different. The first-stage
MRP estimates were found to be positively correlated both with Tausanovitch and Warshaw's measure for ideology and the state-level weighted averages of the raw survey data, as
expected. Therefore, the null second-stage results are likely not due to measurement error
in the estimates of state-level income group preferences.
The lack of results may be particularly surprising when one considers how survey organizations ask about the most salient and important issues and underrepresent those that are
more obscure or technical. Since the relationship between public preferences and senators'
votes is expected to be stronger for salient issues, this would upwardly bias estimates.
The absence of a positive relationship between senators' voting behavior and public opinion
more generally does not accord with much of the existing literature. Several studies have
found a positive relationship between constituents' opinions and their representatives' voting behavior (Ansolabehere, Snyder, and Stewart 2001; Miller and Stokes 1963; Page and
Shapiro 1983; Stimson, MacKuen, and Erikson 1995; and many others). This null finding
also runs counter to the intuition of the Instructed-delegate model of representation where
the representative is supposed to act based on the preferences of his constituents. This was
the perspective I adopted for this paper and it was not supported by the evidence. Under the
Responsible party model I would also expect evidence of a positive relationship since it also
shares the idea of popular control, but at the national level instead of local one. However,
'0I also estimated the second stage using GEE models. For a detailed discussion of the method and results,
please see the Appendix.
45
in the second stage analysis, the party of the senator was only found to be positive and
significant for the issue area of estate tax. If senators were dutifully voting in support of
their party's platform, this variable would explain most of the variation in the model, but
that is not the case.
Under the Burkean model, the representative is granted a bit more leeway and is expected
to act in accordance with constituents' interests, but not necessarily their preferences. However, it would be expected that when voting in accordance with interests, there would still
be some observed positive relationship with preferences. In theory people should have some
preference for the policies that serve their interests, although self-interest does not appear
to drive preference formation (Sears et al. 1980).
Mayhew and Fenno both stressed how
elected officials are primarily concerned with their reelection since that is how all other goals
of office were achieved, so it is unexpected to find that senators vote in ways unrelated to
popular opinion. While Fenno acknowledged that members' behavior may be largely unconstrained so long as it could be defended come time for reelection and the member voted with
constituents on key issues (1977, 912), it is doubtful that this argument could be used to
account for a complete lack of a relationship. However, perhaps for the one or two issues that
really matter to constituents, there is better evidence of a positive relationship. The salience
of an issue may also vary state to state and so the relationship would otherwise be masked in
this analysis. Future analysis could better account for this possibility of state-variable issue
salience.
Should this be the case, then the analysis could be used to support a primarily
Burkean model of representation with elements of the Instructed-delegate model for issues
of high salience to particular states and income groups. While finding a true measure for
constituents' interests is perhaps unrealistic, the lack of a relationship with constituents'
preferences suggests that senators are looking elsewhere when making their policy decisions.
While this analysis was not suggestive of the Instructed-delegate model responsiveness, it did
reveal that not only are income group opinions different across issue areas, but the top ten
46
percent were found to behave differently from a more inclusively defined high-income group.
They were found to be more liberal on issues like minimum wage, abortion, gay marriage,
Social Security, and the Iraq War than the more inclusive high income group. They are also
more conservative than the inclusive group on issues like the estate tax, school vouchers, and
free trade. I also found that on issues like gun control, income group opinion aggregation is
less meaningful.
The distinctiveness of income-group level opinion provides some evidence in support of opinion formation theories at the level of social groups. Kinder suggested that citizens recognize
when a policy benefits their social group in making their decision to support it (2003). Likewise, Dawson believes that group identity is used to form policy preferences which accounts
for the remarkable unity of the Black vote (1994).
Since individuals do not tend to form
political opinions on the basis of self-interest (Sears et al. 1980), the fact that income groups
have distinct opinions is likely due to class-based socialization. Political opinions are formed
through the contexts of friends, family, and personal acquaintances (Berelson, Lazerfeld, and
McPhee 1954) and minority opinions can be difficult to recognize (Huckfeldt and Sprague
1987).
Since individuals tend to self-segregate by income group, the fact that they hold
different opinions from each other is not the necessary outcome, but it is a highly probable
one. These projected differences in income group opinions are supported by the first-stage
MRP analysis.
While this paper has conclusively shown that the wealthy have different political opinions
from low and middle-income individuals, the U.S. Senate's model of representation and pattern of responsiveness is left more of an open question. However, this analysis has exposed
important considerations for modeling responsiveness and will hopefully help to guide the
direction of future research on the topic.
47
Future Research
While the methods presented in this paper constitute an improvement over earlier methods, particularly when modeling income group level opinion where there are insufficient
sample sizes in national surveys, additional analysis would increase confidence in these findings. Conducting MRP analysis and the second stage analysis as a Bayesian model would
better capture uncertainty in the estimates (particularly with regard to propagating Stage
I uncertainty into Stage II). Estimating senators' ideal points using Bayesian item response
theory will also help capture unobserved variables of interest. For instance, the salience of
a particular issue may vary by each state and by each income group, but this cannot be
directly estimated. It may be that high income individuals care most about economic issues
and low income individuals care most about issues relating to social welfare. It is also possible that two income groups consider the same issue salient but have different preferences and
the senators must choose which group to appease. Item response theory will allow this key
moderator to be included in the estimation. A statistical method allowing for issue salience
to be flexibly modeled in this way would also provide more support for a theoretical hybrid
model of Burekan and Instructed-delegate representation.
Future researchers may also create a more issue-specific way to classify votes as liberal
or conservative by using substantive knowledge to choose different senators as anchors for
a given issue. In this analysis I relied on NOMINATE scores to code whether a vote was
liberal or conservative. However, this measure may be too general and result in inaccurate
codings of ideology for votes within certain issue areas. While I would expect this measure
to perform well for economic issues, it likely does not perform as well for social issues.
More data over a longer time frame will increase the ability to assess the effects that elections
have on senators' responsiveness. This would test whether my results are unique to a specific
time period or partisan make-up of the Senate.
48
Regardless of the model used, second stage results are predictive quantities and only meant
to be suggestive of a possible or lack of causal relationship between income group opinions
and the way senators vote. While this analysis did not find there to be a positive relationship
indicating responsiveness to constituent opinions, it is possible that with different roll-call
votes such a relationship may be detected. If a relationship indicating responsiveness is observed then it is possible for it to vary by income group, since income group opinions were
conclusively found to be distinct across a majority of issues.
There are several theories about how income group preferences may be differentially translated into the observed voting behavior of the political elite. Descriptive representation,
elite cue-taking, and the non-uniform allocation of political resources all offer different stories about how it might be possible to craft policy that is more line with the preferences
of a certain segment of the population in a democracy theoretically open to all. Scholars
who found that representatives who look like their constituents are also more likely to craft
policy preferred by those constituents (see Chattopadhyay and Duflo 2004; Carnes 2011; Posner 2005 and many others) could easily extend this explanation to an observed upper-class
bias in policy-making, as many members of Congress came from wealthy and well-connected
families (Hess 1966). Though, this theory would fail to explain the observed variation in
responsiveness over time. Top-down political behavior scholars might argue that the wealthy
are more likely to read or watch the news than other socio-economic classes and thus are
more likely to receive elite framing of issues, consequently adopting the same viewpoints
(Page and Shapiro). However, this would only persist in the absence of alternative frames
(Druckman and Nelson 2003). Perhaps the most compelling story is that the wealthy have
more political resources of time, money, and education which allow them to be better able
to engage in the political process (Brady, Verba, and Schlozman 1995). Additionally monied
interests tend to be better organized, and by leveraging the representatives' electoral incentive, can gain disproportionate influence over policy (Hall and Wayman 1990; Mayhew
49
1974; Fenno 1977). These competing explanations for potential gaps in responsiveness to the
opinions of the low versus high-income groups deserve careful examination in future research.
50
Appendix
51
MRP Estimates of Public Opinion by State and Income Group
52
Figure Al
Minimum Wage
(56,260,16)
(19,154,14)
(17,63,2)
(121,667,58)
(41,106,6)
(9,89,8)
(243,1147,48)
(144,902,79)
(46,174,7)
(14,62,3)
(160,660,32)
(18,98,0)
(85,483,38)
(80,301,18)
(334,2033,183)
(75,322,19)
(44,164,4)
(89,331,8)
(44,129,5)
(309,1685,125)
(109,528,53)
(138,598,30)
(118,508,30)
(70,288,19)
(30,92,3)
(120,496,27)
(20,68,1)
(137,595,19)
(150,823,57)
(102,339,10)
(234,1134,67)
(78,341,14)
(229,1161,67)
(42,283,22)
(396,2601,367)
(160,719,46)
(17,53,4)
(78,196,6)
(92,344,20)
(47,227,17)
(39,266,42)
(49,135,7)
(16,82,6)
(104,239,9)
(64,651,80)
(52,541,74)
(82,691,102)
(165,1125,118)
(63,429,56)
(242,1347,146)
(1,3,1)
U
0
A
*A
*A
*
0
A
U
U
*A
0 A
U
*A
U
A
U
0
A
0
A
U
OA
U
0
*
A
*
0
A
*
*A
A
U
0
*A
*
S
OL
0*A
*A
0
A
U
0
**A
**A
*0
A
*
A
a
* .A
**A
*
*A
*
e A
*0
A
**A
e A
o*A
U
*A
A
U
*
0
A
U
* A
U
a&
SL
* A
N
e A
0
0 A
S
E
A
S
*A
ok
U
A
0
*
*
A
U
a
* A
A
U
0
U
*A
S
a A
A
0
U
UT
NH
WY
AZ
MT
AK
OH
GA
*
ID
HI
WI
RI
CO
OK
TX
SC
ME
KS
NE
FL
MN
MO
OR
LA
SID
STN
ND
IN
WA
KY
MI
IA
PA
NV
CA
NC
VT
WV
AL
NM
CT
MS
DE
AR
NJ
MD
VA
IL
MA
NY
DC
0.0
0.2
0.6
0.4
MRP Estimated Opinion
53
0.8
1.0
Minimum Wage
(Equal Breaks)
(56,186,90)
(19,84,84)
(17,46,19)
(46,132,49)
(121,384,341)
(44,110,58)
(41,79,33)
(243,757,438)
(9,49,48)
(80,219,100)
(30,66,29)
(138,404,224)
(118,325,213)
(120,344,179)
(89,238,101)
(309,1049,761)
(20,48,21)
(160,434,258)
(109,315,266)
(14,32,33)
(18,59,39)
(44,97,37)
(75,207,134)
(137,400,214)
(85,257,264)
(17,38,19)
(78,142,60)
(334,1170,1046)
(70,191,116)
(78,247,108)
(144,517,464)
(102,217,132)
(92,238,126)
(49,89,53)
(234,707,494)
(396,1370,1598)
(47,121,123)
(42,167,138)
(229,734,494)
OA
A
U
*
A
0
A
U
A
U
*
fleA
A
A
*
U
A
0
a
*
A
N
A
e
*
A
A
*
*
*A
a
A
*
N
A
*
*
*
A
*
A
0
*
A
U
A
U
*
A
0
U
A
e
U
A
0
U
A
0
U
A
*
U
UOA
U
A
A
*
U
A
0
U
A
U
S
A
*
U
A
U
0
E 0 A
* A
U
A
0
U
0
A
U
A
U
0
A
0
U
A
0
U
U
*A
A
*
U
A
0
U
e A
U
A
e
U
A
0
U
A
e
U
A
0
U
A
U
*
*A
U
A
0
U
A
U
&
UT
NH
WYa
ID
AZ
ME
MT
OH
AK
OK
SD
MO
OR
TN
KS
FL
ND
WI
MN
HI
RI
NE
SC
IN
CO
SVT
fn Wv
TX
LA
IA
GA
KY
AL
MS
MI
CA
NM
NV
PA
VA
WA
NC
CT
AR
DE
NJ
IL
MA
NY
MD
DC
0.0
0.2
(82,373,420)
(150,469,411)
(160,451,314)
(39,138,170)
(104,184,64)
(16,52,36)
(64,317,414)
(165,641,602)
(63,219,266)
(242,782,711)
(52,256,359)
(1,2,2)
0.4
0.6
MRP Estimated Opinion
54
0.8
1.0
Minimum Wage
(No Income Breaks)
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
UT
NH
AZ
AK
WY
ID
MT
GA
HI
CO
OH
RI
TX
WI
ME
OK
FL
KS
OR
MN
SD
MO
NE
SC
WA
ND
rn TN
IN
NV
CA
KY
e
MI
IA
LA
VT
PA
CT
NC
NM
WV
NJ
AL
MD
DE
VA
IL
AR
MS
MA
NY
DC
(332)
(187)
(846)
(106)
(82)
(227)
(153)
(1125)
(79)
(606)
(1438)
(116)
(2550)
(852)
(212)
(399)
(2119)
(428)
(656)
(690)
(125)
(766)
(178)
(416)
(1030)
(89)
(643)
(751)
(347)
(3364)
(451)
(1435)
(433)
(377)
(74)
(1457)
(347)
(925)
(291)
(280)
(795)
(456)
(667)
(104)
(875)
(1408)
(352)
(191)
(548)
(1735)
(5)
U
U
0.0
0.2
0.4
0.6
MRP Estimated Opinion
55
0.8
1.0
Figure A2
Abortion Rights
mAO
IN
MS
AK
UT
AR
WV
LA
KY
AL
OK
A
S"
a
A
A
In
O
A
A
A
A
A
SD
A
A
TN
TX
HI
ND
Sc
KS
MO
(442,1445,53)
(390,829,37)
(10,95,8)
(261,969,48)
(442,991,53)
(410,785,19)
(562,1415,94)
(654,1604,80)
(642,1664,95)
(523,1357,57)
(158,346,10)
(793,2277,131)
(2278,7316,607)
(14,65,3)
(128,280,4)
(542,1553,90)
(436,1308,56)
(796,2491,123)
(231,664,21)
(253,703,28)
(840,3195,283)
(1519,4936,233)
(718,2708,119)
(90,268,9)
(1049,3237,202)
(1116,3975,249)
(628,2398,171)
(813,2597,122)
(191,432,18)
(276,751,43)
(1707,5406,302)
(1044,4179,390)
(546,2142,148)
(1772,5939,498)
(689,3043,320)
(466,1890,123)
(79,306,26)
(199,866,48)
(111,394,22)
(534,1767,95)
(756,2844,163)
(352,2129,236)
(2701,10314,1275)
(501,2823,392)
(242,718,29)
(105,603,48)
(221,1165,158)
(1778,6050,676)
(479,2209,229)
(95,302,19)
(51,179,31)
a
A
U
* U
*
a
*U
m
O
A
A
A
A
ID
NE
GA
OH
WI
WY
NC
MI
5
O MN
IA
MT
NM
PA
IL
AZ
FL
VA
CO
DE
NV
RI
OR
WA
MD
CA
NJ
ME
NH
CT
NY
MA
VT
DC
A
A
A
a
on
m
An
An
w * A
* U
m
emn
S
A
U
A
O
Am
A em
A o n
A emn
A
0 n
on
A
A
wK
A
em
A
**
A oEn
em
A
A
A en
A
o I
Am
Am
A
*
a
A
*
n
A
o
a
A e m
e n
A
A
*
0
o
0
A
A
o
n
nI
A
0.0
0.2
0.6
0.4
MRP Estimated Opinion
56
0.8
1.0
Abortion Rights
(Equal Breaks)
IN
MS
AR
UT
AK
WV
LA
KY
AL
OK
(442,1074,424)
(390,620,246)
(442,796,248)
(261,721,296)
(10,52,51)
(410,621,183)
(562,1033,476)
(654,1197,487)
(642,1249,510)
(523,1079,335)
(158,273,83)
(793,1729,679)
(14,35,33)
(2278,4923,3000)
(128,210,74)
(542,1125,518)
(840,2101,1377)
(436,995,369)
(796,1840,774)
(231,522,163)
(253,524,207)
(1519,3595,1574)
(718,1939,888)
(90,207,70)
(1049,2323,1116)
(1116,2699,1525)
(628,1630,939)
(191,351,99)
(276,519,275)
(813,1969,750)
(1707,3858,1850)
(1044,2786,1783)
(546,1473,817)
(1772,4132,2305)
(689,1948,1415)
(466,1277,736)
(79,209,123)
(199,577,337)
(111,277,139)
(2701,6335,5254)
(534,1308,554)
(352,1255,1110)
(756,1894,1113)
(501,1606,1609)
(242,537,210)
(105,364,287)
(221,687,636)
(1778,4001,2725)
(479,1341,1097)
(95,228,93)
(51,105,105)
NA
A ON
A U
AM
A a
A
Aa
em
SD
An
A m
As
As
TN
HI
TX
ND
Sc
GA
KS
MO
ID
NE
OH
WI
WY
NC
MI
A MN
A e
A *m
Am
As
mA
As
AN
An
A *
An
U
Ae m
A 9m
AOM
A o w
A a
A 9 a
An
A * m
MT
NM
IA
PA
IL
AZ
FL
VA
CO
DE
NV
RI
CA
OR
MD
WA
NJ
ME
NH
CT
NY
MA
AM
A *m
Am
A 0 N
A e
w
Ae N
AM
A *
A *
A*
A* w
A *
A*
A
VT
0.0
0
n
n
*
A
DC
m
a
m
0.2
0.6
0.4
MRP Estimated Opinion
57
0.8
1.0
Abortion Rights
(No Income Breaks)
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
IN
MS
AR
WV
UT
KY
LA
AK
AL
OK
SD
TN
HI
TX
ND
SC
KS
MO
GA
ID
OH
NE
WI
NC
WY
MT
MI
NM
IA
MN
PA
AZ
IL
FL
VA
DE
CO
NV
RI
ME
OR
CA
WA
MD
NJ
NH
CT
NY
MA
VT
DC
(1940)
(1256)
(1486)
(1214)
(1278)
(2338)
(2071)
(113)
(2401)
(1937)
(514)
(3201)
(82)
(10201)
(412)
(2185)
(1800)
(3410)
(4318)
(916)
(6688)
(984)
(3545)
(4488)
(367)
(641)
(5340)
(1070)
(3532)
(3197)
(7415)
(2836)
(5613)
(8209)
(4052)
(411)
(2479)
(1113)
(527)
(989)
(2396)
(14290)
(3763)
(2717)
(3716)
(756)
(1544)
(8504)
(2917)
(416)
(261)
U
U
U
U
U
0.0
0.2
0.4
0.6
MRP Estimated Opinion
58
0.8
1.0
Figure A3
Environment
10
U
WY
SID
UT
ND
OR
WV
(44,111,3)
(18,45,0)
(37,58,1)
(56,151,8)
(26,51,1)
(94,270,11)
(79,126,3)
(31,59,1)
(104,226,4)
(87,164,11)
(64,131,7)
(143,427,21)
(162,446,13)
(85,172,7)
(114,312,10)
(143,456,14)
(86,316,16)
(45,132,8)
(111,420,22)
(216,623,39)
(134,265,7)
(112,352,11)
(77,236,15)
(302,788,32)
(62,108,3)
(486,1178,89)
(174,415,19)
(128,312,11)
(103,280,11)
(11,47,2)
(318,872,32)
(584,1783,173)
(17,104,4)
(21,45,1)
(213,716,47)
6
S
46
6
U
*A
U gA
0
A
U
U
*
*A
U
U
*A
U
*A
U
*A
0
U
*A
U
U
oh
U
S
A
U
*A
*
A
U
B A
E A
0Ok
U
A
U 0 A
*0 A
*0 A
A
0
9
A
*
*
* * A
U
*A
e
A
** A
*
A
*9
A
A
mA
A
U S
A
U *
A
A
U
*
U *
A
U
S
A
A
*
A
U
*
U
A
U 0
A
*
A
a
*
A
me
A
A
me
MT
OK
AR
NE
WA
MO
MS
AZ
WI
CO
ME
MN
MI
KY
IN
KS
OH
NM
TX
r TN
AL
IA
DE
PA
CA
NH
VT
IL
VA
LA
GA
CT
NV
MI D
NC
RI l
SC
FL
MA
NY
NJ
*
DC
HI
AK
0.0
*
0.2
(159,555,46)
(136,242,14)
(161,526,40)
(34,207,15)
(36,120,3)
(81,358,30)
(217,583,27)
(34,57,6)
(114,283,10)
(335,897,68)
(102,402,29)
(376,1082,89)
(108,508,67)
(20,38,7)
(0,0,0)
(0,0,0)
A
0.6
0.4
MRP Estimated Opinion
59
0.8
1.0
Environment
(Equal Breaks)
w
h
S
L
0 A
ID
WY
SD
UT
ND
OR
WVV
MT
OK
NE
AR
MO
WA
AZ
WI
MS
ME
IN
MN
KY
MI
N
U
U
U
U
U
U
U
0
0
(44,89,25)
(18,37,8)
(37,49,10)
(56,118,41)
(26,38,14)
(94,223,58)
(79,112,17)
(31,50,10)
(104,185,45)
(64,99,39)
(87,136,39)
(162,372,87)
(143,321,127)
(114,229,93)
(143,346,124)
(85,131,48)
(45,103,37)
(112,274,89)
(111,314,128)
(134,211,61)
(216,441,221)
(86,240,92)
(77,192,59)
(302,635,185)
(62,85,26)
(174,341,93)
(128,242,81)
(318,662,242)
(17,66,42)
(486,865,402)
(103,218,73)
(11,33,16)
(584,1218,738)
(21,36,10)
(159,399,202)
(213,508,255)
(136,183,73)
(34,140,82)
(34,43,20)
(36,87,36)
(161,378,188)
(114,212,81)
(217,449,161)
(81,222,166)
(335,705,260)
(376,771,400)
(102,267,164)
(20,27,18)
(108,315,260)
(0,0,0)
(0,0,0)
As
A
A
sA
*A
*A
0A
BA
SL
0 A
BA
EA
U
*
A
0 A
w e
* A
U
m * A
0BA
0 A
0*A
*
0
A
0
A
N
U B A
U
BA
A
N
a A
0
9
A
w* A
*BA
oA
9
a A
m eA
a oA
A
U
*
u 9 A
mBA
U
*
A
U
*
A
Co
KS
OH
NM
STN
CO AL
PA
NH
TX
IA
DE
CA
VT
VA
IL
LA
CT
RI
NV
GA
SC
NC
MD
FL
NY
MA
DC
NJ
Hi
AK
MBA
U
U
M
A
*
A
BA
mBA
ne
Bo
B
0.0
0.2
A
A
A
0.6
0.4
MRP Estimated Opinion
60
0.8
1.0
Environment
(No Income Breaks)
ID
WY
SD
ND
UT
OR
MT
WV
OK
NE
WA
AR
MO
ME
U
(158)
(63)
(96)
(78)
(215)
(375)
(91)
(208)
(334)
(202)
(591)
(262)
(621)
(185)
(613)
(553)
(436)
(475)
(418)
(406)
(264)
(878)
(328)
(1122)
(394)
(608)
(125)
(67)
(451)
(1222)
(60)
(173)
(1753)
(2540)
(976)
(256)
(760)
(469)
(97)
(159)
(727)
(827)
(392)
(407)
(1300)
(533)
(1547)
(683)
(65)
(0)
(0)
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
WI
MN
AZ
IN
CO
KY
MS
MI
KS
OH
IA
* TN
NH
VT
AL
PA
DE
NM
TX
CA
IL
CT
VA
MD
RI
NV
GA
NC
LA
SC
FL
MA
NY
NJ
DC
HI
AK
U
U
U
U
0.0
0.2
0.6
0.4
MRP Estimated Opinion
61
0.8
1.0
Figure A4
Estate Tax
OK
KY
IN
LA
AR
wV
A
A
A
A
*
0
0
0
(38,111,4)
(59,114,3)
(50,144,4)
(60,140,6)
(47,79,7)
(41,76,3)
(14,30,2)
(9,17,0)
(27,63,4)
(54,126,7)
(76,161,6)
(10,41,1)
(230,578,52)
(74,268,24)
(73,314,22)
(75,183,13)
(53,170,4)
(27,76,3)
(9,24,0)
(107,289,16)
(141,475,21)
(37,71,4)
(22,68,4)
(6,31,0)
(10,34,2)
(68,239,7)
(49,251,32)
(20,49,0)
(88,190,9)
(109,305,18)
(42,144,11)
(250,833,100)
(137,397,14)
(110,312,32)
(50,128,6)
(26,57,1)
(21,100,13)
(48,154,4)
(22,38,0)
(61,209,11)
(38,100,6)
(5,28,3)
(52,212,15)
(168,509,41)
(34,73,5)
(63,185,15)
(39,179,21)
(6,14,4)
(179,534,65)
(0,0,0)
(0,0,0)
A
A
*0
A
N q
A
U * A
*
*
A
1
*
A
a *
A
*
0
A
U
0
A
U
S
A
A
*
*
A
*
*
A
A
U
0
A
U
*
A
m * A
*
*
A
*
*
A
*
*
A
*
*
A
A
SNJ
*
*
A
*
*
A
*
*
A
*
*
A
0
9
A
*
0
A
*
*
A
*
A
*
*
*
A
0
0
A
E
E
MT
ND
ME
SC
AL
NH
TX
GA
VA
MO
IA
NE
VT
$
*
*
0
a
0
NC
PA
MS
NV
WY
RI
WI
ID
TN
MI
AZ
CA
OH
IL
OR
NM
CT
CO
*
SD
WA
KS
DE
MN
FL
*
a
*
*
A
.
9
*
*
0
*
*
*
A
A
A
A
A
A
A
A
A
A
*
UT
MA
MD
DC
NY
HI
AK
*
*
*
U
U
0.0
0.2
*
A
0.6
0.4
MRP Estimated Opinion
62
0.8
1.0
Estate Tax
(Equal Breaks)
(38,90,25)
(59,87,30)
(50,108,40)
(60,109,37)
(47,65,21)
(41,65,14)
(14,26,6)
(9,13,4)
(27,50,17)
(54,87,46)
(76,119,48)
(10,28,14)
(74,198,94)
(9,20,4)
(230,418,212)
(53,132,42)
(75,144,52)
(27,60,19)
(73,223,113)
(107,219,86)
(141,368,128)
(37,56,19)
(6,27,4)
(22,50,22)
(20,40,9)
(88,149,50)
(68,185,61)
(10,25,11)
(109,208,115)
(49,167,116)
(42,108,47)
(50,109,25)
(250,561,372)
(137,308,103)
(26,46,12)
(110,228,116)
(22,33,5)
(21,72,41)
(48,108,50)
(38,72,34)
(61,143,77)
(5,23,8)
(52,151,76)
(168,379,171)
(34,55,23)
(63,124,76)
(39,121,79)
(6,9,9)
(179,352,247)
(0,0,0)
(0,0,0)
* A
0 A
* A
*
* A
0
* A
U * A
a * A
a 0A
*A
0
* A
*0 A
* A
0
*0 A
A
*
0
A
0 A
9
* A
* * A
**
A
*& A
0 A
9
*0 A
* * A
**
A
* A
* * A
o A
**
A
*
0 A
*
*
A
* 0
A
**
A
*
*
A
0 A
e
A
m
*
A
**
me A
*
*
A
*
A
m 0 A
*0
A
U
0
A
* 0 A
*
0
A
0
A
U
A
U 0
A
U
*
0
A
U
A
* 0
OK
KY
IN
LA
AR
V
N
N
MT
ND
ME
Sc
AL
NH
GA
me
VT
TX
IA
MO
NE
VA
NC
PA
MS
WY
NV
'D
TN
5 WI
RI
MI
NJ
AZ
OR
CA
OH
NM
IL
SD
CT
CO
KS
WA
DE
MN
FL
UT
MA
MD
DC
NY
HI
AK
0.0
0.2
0.4
0.6
MRP Estimated Opinion
63
0.8
1.0
Estate Tax
(No Income Breaks)
KY
IN
OK
AR
MT
ME
ND
NH
WV
LA
SC
AL
VT
GA
WY
IA
NE
MO
VA
NC
TX
RI
ID
CA
NJ
c PA
C )AZ
MI
NV
TN
MS
OH
WI
CT
OR
SD
IL
CO
KS
MN
NM
DE
WA
MA
UT
FL
MD
DC
NY
H1
AK
(176)
(198)
(153)
(133)
(46)
(94)
(26)
(52)
(120)
(206)
(187)
(243)
(33)
(366)
(37)
(227)
(106)
(271)
(409)
(412)
(860)
(46)
(69)
(1183)
(332)
(637)
(197)
(432)
(94)
(287)
(112)
(548)
(314)
(134)
(184)
(60)
(454)
(206)
(144)
(279)
(84)
(36)
(281)
(263)
(112)
(718)
(239)
(24)
(778)
(0)
(0)
U
w
U
U
S
U
N
a
S
E
0
U
S
U
E
S
E
E
E
0
0
N
0
0
0.0
0.2
0.6
0.4
MRP Estimated Opinion
64
0.8
1.0
Figure A5
Gun Control
U
0
U
0
U
*
A
U
0
A
a 0 A
U
0
A
*
*
A
U 0
A
MT
WY
SD
ID
IA
AR
OK
NV
ND
WV
KY
Co
KS
NM
NE
TN
WI
AZ
OR
MO
UT
MS
OH
TX
GA
IVMN
Ci MI
SC
LA
PA
VT
AL
NH
FL
VA
WA
ME
DE
CA
IL
NC
MD
RI
CT
NY
NJ
MA
IN
DC
HI
AK
0.0
(78,177,5)
A
A
(39,109,3)
(68,132,5)
(98,238,6)
(363,1130,47)
(191,397,28)
(255,550,22)
(78,304,14)
(69,113,2)
(184,309,9)
(301,689,39)
(204,750,46)
(181,514,29)
(136,261,12)
(109,301,12)
(354,935,55)
(272,1062,45)
(230,748,39)
(220,685,34)
(343,1014,43)
(107,390,20)
(165,365,16)
(650,1976,95)
(1052,2779,221)
(365,1225,101)
(287,985,51)
(475,1535,87)
(258,663,28)
(279,587,45)
(783,2185,116)
(46,133,5)
(294,746,44)
(45,227,17)
(786,2181,200)
(322,1256,129)
(325,1108,58)
(98,302,16)
(30,125,12)
(1229,4031,488)
(450,1689,139)
(460,1373,89)
(166,845,91)
(44,138,13)
(103,468,57)
(796,2507,271)
(243,1163,161)
(228,954,97)
(170,475,21)
(25,89,20)
(0,0,0)
(0,0,0)
*oA
NO
* 0A
0 e A
Se A
*
N
A
a SL
U
ueA
a eA
0 ek
le A
&
N 4
weoA
0 oA
mA
A
on
me
Am
A
ofn
An
AM
Am
AE
Am
A ofn
A EN
A * w
A En
A
A
0
Acm
*
*
U
0.2
0.6
0.4
MRP Estimated Opinion
65
0.8
1.0
Gun Control
(Equal Breaks)
MT
.
0
WY
(78,147,35)
(39,93,19)
(68,111,26)
(98,186,58)
(363,870,307)
(191,316,109)
(255,439,133)
(78,220,98)
(69,85,30)
(184,256,62)
(301,527,201)
(204,545,251)
(181,413,130)
(136,198,75)
(109,225,88)
(272,789,318)
(354,739,251)
(220,543,176)
(107,307,103)
(343,780,277)
(230,563,224)
(165,272,109)
(650,1477,594)
(1052,2016,984)
(365,833,493)
(287,699,337)
(475,1079,543)
(258,496,195)
(279,433,199)
(783,1586,715)
(45,140,104)
(46,106,32)
(294,586,204)
(786,1587,794)
(325,775,391)
(322,834,551)
(98,234,84)
(1229,2620,1899)
(30,84,53)
(460,1043,419)
(450,1167,661)
(166,522,414)
(44,102,49)
(103,273,252)
(796,1712,1066)
(243,716,608)
(228,604,447)
(170,380,116)
(25,54,55)
(0,0,0)
(0,0,0)
U0
SD
U
U
U
ID
IA
AR
OK
NV
ND
WV
KY
CO
KS
NM
NE
WI
TN
OR
UT
MO
AZ
MS
OH
TX
GA
§MN
4
&
e
U
U
U
&
&
0 4
E SL
0 L
11&
0
1&
w OL
0
N
N S
E4
0 @A
0 S
&
EA
ROL
cn MI
N
Sc
LA
PA
NH
vT
AL
FL
WA
VA
ME
CA
DE
NC
IL
MD
RI
CT
NY
NJ
MA
IN
DC
HI
AK
a
*eA
I
Ab
NA
k
6 N
Aem
A
1
Ea
A
S
A*
4N
&
A
QE E
0.0
S
0.2
0.6
0.4
MRP Estimated Opinion
66
0.8
1.0
Gun Control
(No Income Breaks)
U
MT
WY
SD
ID
IA
AR
ND
OK
NV
WV
KY
CO
NM
KS
NE
(260)
(151)
(205)
(342)
(1540)
(616)
(184)
(827)
(396)
(502)
(1029)
(1000)
(409)
(724)
(422)
(1379)
(1344)
(517)
(1017)
(939)
(1400)
(4052)
(546)
(2721)
(1691)
(1323)
(2097)
(289)
(949)
(184)
(3084)
(911)
(3167)
(1084)
(416)
(1707)
(1491)
(5748)
(167)
(1922)
(2278)
(1102)
(195)
(628)
(3574)
(1279)
(1567)
(666)
(134)
(0)
(0)
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
WI
TN
UT
AZ
OR
MO
TX
MS
OH
GA
MN
C MI
NH
SC
VT
PA
LA
FL
AL
ME
VA
WA
CA
DE
NC
IL
MD
U
U
U
U
U
RI
CT
NY
MA
NJ
IN
DC
HI
AK
U
U
0.0
0.2
0.4
0.6
MRP Estimated Opinion
67
0.8
1.0
Figure A6
Gay Marriage
Am e
* *
S
0
An 0
SD
TN
AL
OK
IN
AR
MS
(16,59,0)
(94,379,25)
(59,218,16)
(62,207,12)
(45,271,13)
(53,153,5)
(35,110,3)
A
4
U
NA 0
A
WV~
0
m(51,107,3)
* *
KY
ID
Sc
(69,261,12)
(28,128,6)
(55,227,17)
(25,70,2)
(11,41,0)
(57,198,13)
(32,101,5)
(37,161,11)
(285,1457,132)
(88,570,53)
(111,505,36)
(108,411,20)
(195,871,55)
(8,39,2)
(14,29,3)
(4,29,3)
(93,482,30)
(10,113,9)
(55,205,9)
(78,511,78)
(121,406,18)
(168,745,44)
(132,754,85)
(5,51,4)
(27,154,14)
(94,426,51)
(152,677,42)
(70,386,35)
(21,107,5)
(75,349,29)
(4,21,0)
(200,1091,86)
(12,50,1)
(42,346,50)
(88,361,25)
(51,417,69)
(343,1980,259)
U A*
A
MT
0
*
a
Am e
A 0
ND
LA
NE
UT
TX
GA
NC
MO
PA
WY
VT
AK
WI
NH
(7 KS
VA
IA
OH
IL
DE
NV
MN
MI
AZ
ME
CO
HI
FL
RI
MD
OR
NJ
CA
WA
NM
CT
MA
NY
DC
a
M
A
SA
&
0*
*
*
A N
*
A Ne
A *e
An g
A a
m A
An
u
Am
a As
flo
A
&
0
0
q
*
S
OA
AM 0
m
M
SA 0
A
No
*
A
m
Ae
M
e
M e
mAe
A u g
0*
mAe
(122,609,44)
&0
Ame
Nh0
A
(41,141,12)
(28,154,38)
(46,308,32)
(192,986,113)
(4,12,4)
m
00
0.0
0.2
0.6
0.4
MRP Estimated Opinion
68
0.8
1.0
Gay Marriage
(Equal Breaks)
aA
SD
AL
TN
OK
IN
AR
WV
MS
KY
ID
ND
MT
SC
NE
LA
NC
GA
Mo
UT
VT
PA
TX
WY
NH
AK
N
a
m
(16,41,18)
(59,146,88)
(94,259,145)
(62,152,67)
(45,172,112)
(53,118,40)
(51,79,31)
(35,68,45)
(69,185,88)
(28,95,39)
(11,30,11)
(25,53,19)
(55,154,90)
(32,74,32)
(57,137,74)
(111,315,226)
(88,314,309)
(108,287,144)
(37,112,60)
(14,21,11)
(195,569,357)
(285,868,721)
(8,27,14)
(10,56,66)
(4,16,16)
(78,263,326)
(93,311,201)
(55,152,62)
(121,304,120)
(21,75,37)
(94,252,225)
(152,413,306)
(5,28,27)
(27,90,78)
(168,485,304)
(132,442,397)
(70,250,171)
(75,202,176)
(200,706,471)
(4,10,11)
(12,33,18)
(88,227,159)
(51,214,272)
(42,158,238)
(122,354,299)
(41,77,76)
(28,80,112)
(343,1096,1143)
(46,163,177)
(192,581,518)
(4,7,9)
0A
OA
A*
U &
U 4
0
U
0 A
* h
ho
mAO
U As
NA 0
0
0A
SAO
A*
A0
U 4A
A a
a
SA
U
EL
Am
0
SVA
M
0W1
A
a A*
A*
Am *
KS
IA
ME
MN
MI
DE
NV
OH
IL
AZ
Co
FL
HI
RI
OR
NJ
MD
WA
NM
CT
CA
MA
NY
DC
0
e
e
m As
mA
0 A
A
m1A
0
Ae
m Ae
&0
m Ae
mAe
mA 9
mAe
1&
w Ae
SAe
M *
A
0
01A
A
&
m Ae
0.0
0.2
0.4
0.6
MRP Estimated Opinion
69
0.8
1.0
Gay Marriage
(No Income Breaks)
SD
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
TN
AL
IN
OK
AR
MS
WV
KY
ND
ID
MT
SC
NE
LA
UT
TX
NC
MO
PA
GA
WY
AK
NH
WI
IA
KS
VA
OH
DE
NV
ME
IL
MI
MN
AZ
CO
HI
FL
RI
MD
OR
CA
NJ
WA
NM
CT
MA
NY
DC
(75)
(498)
(293)
(329)
(281)
(211)
(148)
(161)
(342)
(52)
(162)
(97)
(299)
(138)
(268)
(209)
(1874)
(46)
(652)
(539)
(1121)
(711)
(49)
(36)
(132)
(605)
(545)
(269)
(667)
(957)
(60)
(195)
(133)
(971)
(871)
(571)
(491)
(453)
(25)
(1377)
(63)
(438)
(474)
(2582)
(537)
(775)
(194)
(220)
(386)
(1291)
(20)
U
U
0.0
0.2
0.4
0.6
MRP Estimated Opinion
70
0.8
1.0
Figure A7
Federal Health Insurance
NE
ND
*
*
SI D
*
ID
U
A
A
A
A
A
U
A
U
0
A
U
*
A
U
A
*
*
U
o
A
A
U
0
0
A
U
A
0
U
U
*
A
e A
U
e
0
A
A
U
*
U
* A
*
A
U
* A
0
0
A
U
**
A
0
A
U
0
A
U
0
A
U
A
U
0
U
*
A
*0
A
U
0 A
A
U
0
* A
A
U
0
U
* A
0
A
U
oA
U
U *A
A
U
0
U
A
0
A
U
A
U
A
U
0
0
U
*
*
0
0
U
IN
TN
MO
IA
MT
TX
MN
GA
AL
CO
WY
MS
LA
OH
KY
NV
Wi
NM
SCA
(5
A
A
0
*
*
OK
KS
UT
(143,361,12)
(71,111,2)
(92153,2)
(121,281,6)
(293,611,13)
(221,578,31)
(141,434,18)
(290,789,32)
(423,1073,52)
(402,1113,41)
(337,943,41)
(89,176,2)
(1251,3101,237)
(319,1072,59)
(445,1351,96)
(345,782,43)
(225,832,44)
(47,123,1)
(229,421,13)
(337,684,40)
(756,2155,94)
(350,714,28)
(91,349,11)
(340,1202,44)
(145,284,9)
(1547,4597,478)
(569,1510,84)
(360,1194,57)
(261,735,27)
(410,1436,121)
(276,824,43)
(38,140,10)
(214,468,24)
(295,754,28)
(128,339,16)
(560,1699,90)
(904,2439,189)
(540,1856,132)
(905,2348,117)
(225,339,9)
(68,177,12)
(49,243,15)
(286,1333,173)
(202,964,96)
(109,546,54)
(55,147,4)
(982,2811,280)
(289,1056,91)
(36,95,17)
(0,0,0)
(0,0,0)
A
A
0
U
NC
WA
OR
VA
AZ
DE
AR
SC
ME
MI
FL
IL
PA
WV
RI
NH
NJ
MD
CT
VT
NY
MA
DC
HI
AK
U
U
U
0.0
*
0.2
OA
*A
A
0.6
0.4
MRP Estimated Opinion
71
0.8
1.0
Federal Health Insurance
(Equal Breaks)
ND
NE
SD
ID
OK
KS
IN
UT
IA
U
*
*
U
U
MT
U
U
MO
TN
MN
TX
WY
AL
CO
GA
MS
wi
OH
KY
NV
LA
NC
OR
CO CA
NM
ME
WA
AR
AZ
VA
SC
DE
MI
FL
PA
wv
IL
NH
RI
Vr
NJ
MD
CT
NY
MA
DC
HI
AK
0
U
*
N
U
U
U
U
U
0 A
0 A
U
S A
A
0 A
*
A
sA
*
A
*
A
A
*
A
*
A
0 A
0
A
0 A
*
A
0 A
*
A
0
A
**A
*
A
0
A
*
e
A
*
A
**
A
**
A
sA
9
0
A
(71,85,28)
(143,272,101)
(92,126,29)
(121,223,64)
(293,494,130)
(221,456,153)
(290,626,195)
(141,339,113)
(337,753,231)
(89,148,30)
(402,876,278)
(423,873,252)
(319,799,332)
(1251,2294,1044)
(47,102,22)
(345,609,216)
(225,617,259)
(445,969,478)
(229,323,111)
(340,917,329)
(756,1672,577)
(350,562,180)
(91,250,110)
(337,521,203)
(569,1157,437)
(261,589,173)
(1547,3059,2016)
(145,222,71)
(128,271,84)
(360,856,395)
(214,384,108)
(276,629,238)
(410,1015,542)
(295,567,215)
(38,100,50)
(560,1222,567)
(904,1829,799)
(905,1768,697)
(225,291,57)
(540,1319,669)
(49,163,95)
(68,131,58)
(55,119,32)
(286,820,686)
(202,624,436)
(109,355,245)
(982,1979,1112)
(289,690,457)
(36,60,52)
(0,0,0)
(0,0,0)
usA
U
*
*
a
0.0
*
A
0
A
*
A
0*
A
*
0
A
U
*
A
U
S
A
*
*
A
U
*
A
A
U
S
A
USA
U
0
A
NA
*
A
*
U
*
A
0
*
A
a * A
A
mU
*
A
0.2
0.4
0.6
MRP Estimated Opinion
72
0.8
1.0
Federal Health Insurance
(No Income Breaks)
U
U
U
U
ND
NE
SD
ID
KS
OK
IN
[A
MN
UT
MT
MO
TN
TX
CO
WY
GA
WI
AL
NV
OH
ME
CA
WA
NC
VA
65 OR
KY
AZ
MS
DE
LA
AR
NM
SC
MI
IL
NH
FL
PA
WV
NJ
(184)
(516)
(247)
(408)
(830)
(917)
(1111)
(1321)
(1450)
(593)
(267)
(1556)
(1548)
(4589)
(1101)
(171)
(1892)
(1586)
(1170)
(451)
(3005)
(483)
(6622)
(1611)
(2163)
(1967)
(1023)
(1092)
(1143)
(663)
(188)
(1061)
(706)
(438)
(1077)
(2349)
(2528)
(307)
(3532)
(3370)
(573)
(1792)
(257)
(206)
(1262)
(709)
(4073)
(1436)
(148)
(0)
(0)
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
RI
VT
MD
CT
NY
MA
DC
HI
AK
U
U
U
0.0
0.2
0.4
0.6
MRP Estimated Opinion
73
0.8
1.0
V
0
m
0
0
II--
C4O
o
7-
ob
-<>>
M
1
-m>
Ibi CO
K
State
ZN-OMXWO
-
- -
-
- --
-
-
wiI><M
~
-8)
--
z
N
)
-
N)
- -
CO
~
(
-o O;K
-4
-4.
7
.r-ZCflZ
-N
NP
0
U
I
00.
Eo
Ew
U
4
a
zza
2
i a
58-
a
-
ia
4n
g-:;.8 . akkv! g ---u
w
2
f
44
e
4
te
' ---
*
-
8
-
EE
-
"--1
8
4 4-44 444 4 4
Rs'
4 4
:9i -. * 2.g
g----
D
eieis52
0
26
444444444444444444444444444444
59 #5
-O-(
CL
0
w
D.2E
-O 0
q
0
Immigration
(No Income Breaks)
IN
MS
TN
SC
AL
LA
NE
OK
NC
SD
WY
ND
AR
WV
MO
KY
ME
GA
OH
VA
NJ
KS
TX
NH
MD
MI
(I
(475)
(387)
(880)
(629)
(709)
(583)
(296)
(522)
(1254)
(136)
(88)
(104)
(408)
(332)
(953)
(676)
(291)
(1120)
(1751)
(1129)
(1055)
(497)
(2681)
(184)
(714)
(1374)
(656)
(672)
(2001)
(848)
(254)
(2026)
(107)
(152)
(399)
(258)
(590)
(140)
(932)
(1474)
(947)
(350)
(261)
(840)
(964)
(3837)
(2381)
(113)
(93)
(0)
(0)
a
U
U
U
U
U
U
U
U
U
U
U
CO
AZ
PA
MN
ID
FL
DE
MT
CT
NM
OR
RI
wl
IL
U
U
IA
UT
NV
MA
WA
CA
NY
VT
DC
HI
AK
U
0.0
0.2
0.4
0.6
MRP Estimated Opinion
76
0.8
1.0
Figure A9
Iraq War
u
OK
h
A
A 0
U
NE
ID
AL
KY
WV
So
I
IA
WY
KS
GA
LA
TX
SC
MO
IN
MS
ND
AR
TN
WI
OH
AK
NV
CO MN
NC
PA
ME
VA
MI
AZ
CO
FL
DE
OR
NJ
NH
WA
IL
RI
HI
MA
NM
CA
VT
MD
(153,425,23)
(74,328,20)
(61,157,8)
(68,219,10)
(68,230,12)
(163,516,45)
(188,592,34)
(91,243,7)
(32,106,3)
(343,1143,51)
(20,81,3)
(131,418,24)
(228,1115,113)
(136,430,44)
(665,2601,248)
(154,503,35)
(249,859,45)
(46,269,13)
(86,250,15)
(34,92,0)
(116,308,21)
(216,765,53)
(190,922,62)
(428,1668,91)
(4,28,3)
(57,285,26)
(206,827,79)
(282,1068,78)
(501,1873,131)
(63,233,11)
(177,994,149)
(338,1357,90)
(161,704,58)
(162,668,47)
(504,2065,197)
(17,108,13)
(161,646,42)
(156,921,146)
(29,215,20)
(256,1085,78)
(310,1457,167)
(18,124,7)
(4,21,0)
(142,740,88)
(89,254,19)
(741,3506,518)
(23,91,5)
(102,657,106)
(70,341,66)
(489,2070,255)
(10,44,18)
*
A
U
UT
MT
A*
N
A*
U
0
*
A 0
0
A
Ae
Ae
Ae
A
U
N
E
U
Ae
*A
h
U
h
Ae
A.
A *
A *
U
M
U
m
0
0
AD
A *
A *
i
*
U
0
$
U
N
A *
4
E
A *
A 0
*
4
U
*
A*
0 A
U
4
oA
U
a
4
U
A.
m
*0 A
*
Ae
mA
m
e
N
Ae
4
A*0
a&
N
0
a
4
A eu
4
CT
S
NY
DC
N
A
N
0.0
&
0.2
4
A
0.6
0.4
MRP Estimated Opinion
77
0.8
1.0
Iraq War
(Equal Breaks)
OK
UT
MT
NE
ID
AL
KY
IA
0
w
U
0
U
N
4
S
N
WV
A
A
SD
0
WY
KS
TX
LA
MO
GA
IN
ND
SC
AR
MS
TN
(153,323,125)
(74,244,104)
(61,122,43)
(68,155,74)
(68,173,69)
(163,378,183)
(188,418,208)
(343,851,343)
(91,187,63)
(32,78,31)
(20,65,19)
(131,317,125)
(665,1656,1193)
(136,308,166)
(249,606,298)
(228,663,565)
(46,171,111)
(34,67,25)
(154,358,180)
(116,233,96)
(86,172,93)
(216,540,278)
(190,610,374)
(428,1135,624)
(4,15,16)
(57,186,125)
(206,503,403)
(501,1252,752)
(63,160,84)
(282,713,433)
(177,557,586)
(161,469,293)
(162,430,285)
(338,878,569)
(161,445,243)
(504,1344,918)
(17,65,56)
(29,118,117)
(256,654,509)
(18,84,47)
(156,503,564)
(310,902,722)
(4,10,11)
(23,68,28)
(89,157,116)
(741,2011,2013)
(142,419,409)
(102,334,429)
(70,172,235)
(489,1265,1060)
(10,24,38)
oA
*A
*A
sA
o
*A
o
A
*A
OI
0A
o
A
o
A
N
U
a
OL
0
U
A
e
U
6
o
A
* A
o
*
o
A
* A
U
U
U
WI
OH
AK
NV
(n MN
PA
ME
NC
VA
AZ
Co
MI
OR
FL
DE
NH
WA
RI
NJ
IL
HI
VT
NM
CA
MA
MD
CT
NY
DC
0 A
o
oA
o A
6
&
U
U
A
U
U A
* A
o
4
U
E
*A
&
0 A
o
A
* A
* A
o
U
U
A
N
*A
4
o A
o A
U
w
U
A
S
U
AM
U
U
U
U
U
N
U
0.0
*
0.2
4
A
*A
oA
OA
oA
A
0.6
0.4
MRP Estimated Opinion
78
0.8
1.0
Iraq War
(No Income Breaks)
(422)
(601)
(226)
(310)
(297)
(141)
(104)
(814)
(1537)
(724)
(341)
(573)
(3514)
(1456)
(1153)
(328)
(126)
(1034)
(692)
(35)
(610)
(445)
(307)
(1174)
(368)
(2187)
(351)
(1428)
(1112)
(923)
(2505)
(877)
(1320)
(849)
(1785)
(2766)
(138)
(264)
(1223)
(119)
(1419)
(149)
(1934)
(25)
(362)
(970)
(4765)
(865)
(477)
(2814)
(72)
w
0
U
U
UT
OK
MT
ID
NE
SD
WY
KY
IA
AL
WV
KS
TX
GA
MO
IN
ND
TN
SC
AK
LA
AR
ME
WI
NV
OH
CI)MS
NC
MN
AZ
PA
CO
VA
OR
MI
FL
DE
NH
NJ
VT
WA
RI
IL
H1
NM
MA
CA
MD
CT
NY
DC
U
1
0
0
a
U
a
w
E
a
U
U
U
U
0
0
0
U
U
0.0
0.2
0.4
MRP Estimated
0.6
Opinion
79
0.8
1.0
Figure A10
Social Security
*
A
0
A
0
A
*
A
0
A
* A
*
A
0
A
0
A
A
* A
0 A
*A
* A
*A
* A
UT
LA
TX
U
U
SC
NC
OH
OK
GA
VA
TN
KS
CO
NE
AL
MS
WA
1D
MI
IA
KY
FL
MO
ME
NV
RI
(D
Z PA
Cz Wi
IN
AZ
SD
IL
CA
DE
MD
NM
WY
HI
NJ
NH
OR
MN
DC
AK
U
U
N
E
U
U
a
*
N
(103,367,17)
(257,556,35)
(925,2511,195)
(228,564,18)
(426,1211,79)
(610,1762,70)
(227,503,20)
(343,1113,92)
(279,1163,106)
(312,820,42)
(163,461,23)
(182,682,44)
(112,268,7)
(279,656,35)
(153,333,11)
(293,960,53)
(86,223,5)
(430,1375,69)
(275,912,35)
(263,571,23)
(699,2018,165)
(291,868,35)
(91,254,18)
(72,296,13)
(44,135,11)
(663,1972,87)
(257,974,34)
(189,519,20)
(209,692,32)
(62,116,5)
(406,1529,124)
(1070,3622,418)
(35,115,11)
(156,746,81)
(126,235,10)
(33,100,1)
(4,9,2)
(203,1034,137)
(40,218,15)
(196,613,32)
(258,878,40)
(21,77,14)
(2,15,0)
(43,125,9)
(200,830,81)
(59,96,1)
(196,368,20)
(179,300,11)
(718,2222,227)
(93,399,51)
(74,153,5)
eA
* A
U
*
U
*
U
* A
0 A
A
*A
0A
* A
*A
4L
U
e
M
*A
0
0
4
U
*A
*A
*A
0 A
N
U
*
e
U
a
a
4
* A
*A
O
a
U
U
4
Ab
a
*
VT
MA
ND
AR
wv
NY
CT
MT
0
0
a
N
2
U
A
4
*
&
G
O
A*
As
AD
4
A
0 A*
0.0
0.2
0.6
0.4
MRP Estimated Opinion
80
0.8
1.0
Social Security
(Equal Breaks)
UT
LA
SC
TX
U
0
OH
0
OK
KS
NC
GA
VA
CO
TN
NE
AL
ID
MS
IA
MO
KY
WA
FL
MI
NV
ME
PA
IN
Cn AZ
w1
RI
U
U
*
U
U
U
m
U
U
m
U
a
U
1
*
U
*
*
*
sD
0
*
CA
NM
WY
HI
E
*
U
IL
0
DE
MN
MD
NH
AK
*
N
E
0
N
OR
0
M
0
VT
ND
NJ
AR
MA
DC
N
N
z
a
NY
CT
0
*
0.0
0.2
(103,286,98)
(257,423,168)
(228,419,163)
(925,1781,925)
(610,1342,490)
(227,399,124)
(163,362,122)
(426,922,368)
(343,767,438)
(279,791,478)
(182,485,241)
(312,660,202)
(112,209,66)
(279,523,168)
(86,179,49)
(153,243,101)
(275,703,244)
(291,661,242)
(263,442,152)
(293,680,333)
(699,1480,703)
(430,954,490)
(72,208,101)
(91,200,72)
(663,1445,614)
(189,409,130)
(209,512,212)
(257,728,280)
(44,95,51)
(62,96,25)
(1070,2344,1696)
(126,178,67)
(33,82,19)
(4,7,4)
(406,1060,593)
(35,78,48)
(258,630,288)
(156,463,364)
(40,130,103)
(2,9,6)
(196,496,149)
(43,96,38)
(59,72,25)
(203,647,524)
(196,296,92)
(200,525,386)
(21,45,46)
(179,251,60)
(74,124,34)
(718,1505,944)
(93,243,207)
e
*A
o
a&
o A
oA
N
wv
MT
A
A
A
0
A
0
A
*
A
0
A
*
A
0
A
*
A
0 A
*
A
* A
*
A
* A
A
0 A
* A
o A
e A
* A
* A
* A
* A
o
* A
9
0 A
* A
* A
e A
*A
* A
oA
*A
o A
o A
o A
o A
0 A
eA
4
0
0
*
U
U
o A
o A
A
O
oA
o A
0.4
0.6
MRP Estimated Opinion
81
0.8
1.0
Social Security
(No income Breaks)
U
UT
TX
VA
CO
GA
LA
OH
SC
NC
KS
OK
WA
MO
AZ
NE
ID
(487)
(3631)
(1548)
(908)
(1548)
(848)
(2442)
(810)
(1716)
(647)
(750)
(1306)
(1194)
(933)
(387)
(314)
(497)
(2882)
(1222)
(1174)
(183)
(381)
(1874)
(363)
(857)
(5110)
(970)
(17)
(134)
(2722)
(1265)
(273)
(177)
(728)
(190)
(983)
(1374)
(2059)
(15)
(371)
(161)
(841)
(156)
(1176)
(1111)
(112)
(232)
(3167)
(543)
(584)
(490)
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
U
MS
FL
IA
TN
SD
NV
MI
ME
KY
i CA
C75 AL
AK
WY
PA
WI
NH
VT
IN
RI
MD
NJ
IL
HI
NM
DE
OR
ND
MN
MA
DC
MT
NY
CT
AR
WV
0.0
0.2
0.4
0.6
MRP Estimated Opinion
82
0.8
1.0
Figure All
School Voucher
TX
NJ
LA
IL
NY
NM
AZ
PA
GA
CA
MD
DC
DE
FL
RI
NV
CO
4m
0 EL
= A
w
A
A m
ow
4 U
= A
a
A
es A
em
a A
1
w A
w
A
No
A
0
A
n o
W A
No
A
a A
a A
a A
so
m A
a A
A
WA
GA
0 A
Eo A
CT
q
w
UT
$
(1843,4928,388)
(417,2049,298)
(475,1054,71)
(823,2839,252)
(1450,4417,496)
(217,481,23)
(406,1383,85)
(1399,3958,224)
(661,2144,188)
(2217,7256,836)
(286,1481,150)
(47,152,27)
(61,211,19)
(1381,3948,344)
(89,270,19)
(147,543,22)
(358,1321,79)
(174,857,107)
(529,1912,79)
(191,663,30)
(326,656,26)
(1199,3542,166)
(297,819,32)
(682,2074,102)
(589,2215,198)
(311,717,43)
(523,1234,71)
(447,1162,65)
(319,922,46)
(199,546,22)
(567,1900,97)
(628,1784,86)
(636,1688,95)
(394,1177,57)
(521,1196,68)
(840,2398,143)
(178,465,14)
(401,1667,161)
(496,1760,114)
(84,413,31)
(847,2653,162)
(71,194,7)
(426,980,37)
(323,549,12)
(186,526,23)
(141,304,10)
(101,197,3)
(122,246,6)
(75,231,13)
(0,0,0)
(0,0,0)
OEA
o &
a
A
a
A
A
U
MS
OH
IN
IA
VA
AR
AL
SC
KS
NE
WA
MO
TN
OR
KY
NC
m
ID
mA
MA
MN
NH
e N
*A
so A
MI
MA
WY
OK
WV
ME
a A
mA
A
MT
ND
SD
VT
HI
AK
0.0
0.2
0.6
0.4
MRP Estimated Opinion
83
0.8
1.0
School Voucher
(Equal Breaks)
TX
NJ
NY
NM
LA
IL
CA
AZ
DC
PA
MD
GA
DE
FL
RI
NV
CT
CO
OH
UT
IA
(1843,3531,1785)
(417,1223,1124)
(1450,3040,1873)
(217,369,135)
(475,785,340)
(823,1987,1104)
(2217,4707,3385)
(406.1034,434)
(47,93,86)
(1399,2926,1256)
(286,939,692)
(661,1485,847)
(61,147,83)
(1381,2880,1412)
(89,199,90)
(147,383,182)
(174,530,434)
(358,957,443)
(1199,2675,1033)
(191,505,188)
(682,1588,588)
(529,1425,566)
(311,585,175)
(589,1493,920)
(326,506,176)
(297,648,203)
(319,719,249)
(447,870,357)
(636,1316,467)
(523,942,363)
(567,1341,656)
(199,403,165)
(394,917,317)
(628,1363,507)
(401,1058,770)
(521,927,337)
(178,372,107)
(496,1253,621)
(840,1789,752)
(84,259,185)
(847,1866,949)
(426,799,218)
(71,152,49)
(323,449,112)
(186,406,143)
(141,254,60)
(101,151,49)
(122,201,51)
(75,178,66)
(0,0,0)
(0,0,0)
Ao w
A
Ae u
Ae
m
MA
A
WA
me A
E
WI
AM
AR
VA
MS
IN
rn KS
me
"AA
mA
me A
MA
m
SC
TN
AL
WA
NE
OR
MO
MA
KY
ID
MN
NC
NH
MI
OK
WY
WV
ME
MT
ND
A
mA
mA
me A
hem
a
Eb
M
A
A
SD
Vr
HI
AK
0.0
0.2
0.4
0.6
MRP Estimated Opinion
84
0.8
1.0
School Voucher
(No Income Breaks)
TX
NJ
NY
LA
IL
NM
AZ
PA
GA
CA
MD
DC
DE
FL
RI
NV
CO
CT
OH
IA
UT
WI
VA
AR
IN
VMS
(7159)
(2764)
(66)
(1600)
(3914)
(721)
(1874)
(5581)
(2993)
(10309)
(1917)
(226)
(291)
(5673)
(378)
(712)
(1758)
(1138)
(4907)
(2858)
(884)
(2520)
(3002)
(1071)
(1148)
(1008)
(1674)
(1287)
(1828)
(2419)
(2564)
(767)
0
U
U
11
0
0
E
a
U
U
U
U
U
W SC
KS
AL
TN
WA
NE
MO
OR
MA
KY
ID
MN
NC
NH
MI
WV
W4Y
OK
ME
MT
ND
VT
SD
HI
AK
U(2498)
(1628)
U(2229)
(1785)
(657)
(2370)
(3381)
(528)
(3662)
(884)
U(272)
(1443)
(735)
U
U(455)
(301)
U(319)
(374)
(0)
(0)
0.0
0.2
0.4
0.6
MRP Estimated Opinion
85
0.8
1.0
Figure A12
Free Trade
U.e
U.e
U.e
*
0
U.e
U.e
*
0
U.o
*
0
U.e
* 0
U.e
*
0
U.e
FL
NJ
TX
CO
UT
NM
MD
AZ
DC
NV
RI
Hl
VA
MN
DE
CA
NY
MA
CT
IL
PA
GA
*
A
A
A
A
A
A
A
A
A
A
A
A
A
A
0
A
A
U.
U.e
U.e
U.e
*
0
*
(292,1626,118)
(61,611,70)
(301,2029,174)
(83,477,39)
(58,257,14)
(43,228,17)
(47,545,65)
(119,648,58)
(1,3,2)
(41,277,22)
(15,93,0)
(14,65,3)
(71,637,105)
(99,500,51)
(14,77,6)
(381,2550,361)
(229,1285,124)
(56,407,48)
(33,275,33)
(168,1136,113)
(224,1081,62)
(139,854,73)
(91,324,7)
(65,265,14)
(35,121,5)
(68,316,21)
(164,655,33)
(76,286,14)
(156,812,55)
(10,95,8)
(44,170,6)
(117,509,28)
(17,148,13)
(18,62,2)
(96,330,12)
(28,88,2)
(80,290,14)
(238,1103,48)
(103,480,28)
(141,589,26)
(229,1124,67)
(19,71,1)
(88,336,19)
(38,154,1)
(36,97,4)
(151,682,39)
(40,129,6)
(100,221,9)
(72,188,4)
(145,622,21)
(15,51,3)
A
A
A
A
A
A
0
KS
A
A
A
LA
NE
$ SC
rn wI
*
*
IA
WA
AK
ID
OR
NH
WY
KY
0
0
* 0
U.e
*
A
A
A
A
A
A
A
A
A
A
0
SD
A
OK
OH
TN
*
*
MO
MI
ND
AL
ME
MT
NC
MS
AR
Wv
A
0e
0
A
A
A
0
U.e
U.e
A
A
A
A
A
A
A
A
A
A
* 0
U.e
00e
U.e
IN
A
VT
0.0
0.2
0.6
0.4
MRP Estimated Opinion
86
0.8
1.0
Free Trade
(Equal Breaks)
TX
FL
NM
UT
Co
AZ
CA
NV
NJ
HI
NY
RI
MD
DC
MN
DE
VA
KS
CT
GA
PA
MA
NE
IL
IA
E LA
65 AK
WA
OR
ID
WY
SD
WI
Sc
NH
KY
OK
ND
MO
TN
AL
ME
MT
MS
MI
OH
NC
AR
VT
WV
IN
A
a
m
B
A
A
A
A
0
0
N
m
A
A
NE
A
s
f
a
a
A
A
A
A
A
A
A
A
me
me
U.
*o
E
m
*
so
U.
U.
me
sE
s
me
U.
*
*
No
w
a
NU
E 0
No
me
we
a
119
BC
US
w.
US
nS
0E
US
se
me
me
U
US
0.0
0.2
(301,1160,1043)
(292,1000,744)
(43,122,123)
(58,179,92)
(83,261,255)
(119,362,344)
(381,1334,1577)
(41,161,138)
(61,295,386)
(14,35,33)
(229,749,660)
(15,57,36)
(47,258,352)
(1,1,4)
(99,288,263)
(14,47,36)
(71,339,403)
(91,235,96)
(33,141,167)
(139,481,446)
(224,679,464)
(56,207,248)
(35,93,33)
(168,660,589)
(76,206,94)
(65,175,104)
(10,52,51)
(156,471,396)
(117,323,214)
(44,129,47)
(18,44,20)
(28,62,28)
(164,428,260)
(68,198,139)
(17,80,81)
(96,214,128)
(80,212,92)
(19,50,22)
(141,398,217)
(103,331,177)
(88,228,127)
(38,103,52)
(36,70,31)
(40,81,54)
(229,700,491)
(238,717,434)
(151,424,297)
(100,173,57)
(15,36,18)
(72,136,56)
(145,422,221)
A
A
A
A
A
A
A
A
A
A
A
A
A
A
A
A
A
A
A
A
A
A
A
A
A
A
A
A
A
A
A
A
A
A
A
0.6
04
MRP Estimated Opinion
87
0.8
1.0
Free Trade
(No Income Breaks)
NJ
FL
TX
CO
MD
DC
UT
AZ
NM
NV
RI
VA
HI
CA
MA
MN
DE
NY
CT
IL
GA
PA
KS
WA
LA
NE
AK
SC
IA
WI
OR
NH
WY
ID
SD
KY
OK
MO
OH
MI
TN
ND
AL
MT
NC
ME
MS
AR
IN
WV
VT
(742)
(2036)
(2504)
(599)
(657)
(6)
(329)
(825)
(288)
(340)
(108)
(813)
(82)
(3292)
(511)
(650)
(97)
(1638)
(341)
(1417)
(1066)
(1367)
(422)
(1023)
(344)
(161)
(113)
(405)
(376)
(852)
(654)
(178)
(82)
(220)
(118)
(438)
(384)
(756)
(1389)
(1420)
(611)
(91)
(443)
(137)
(872)
(193)
(175)
(330)
(788)
(264)
(69)
M
U
U
U
0.0
0.2
0.4
0.6
MRP Estimated Opinion
88
0.8
1.0
Validation:
Scatterplots of MRP Estimates V.S. Tausanovitch & Warshaw
Measures for Ideology
89
Figure A13
Minimum Wage
(Equal Breaks)
Minimum Wage
0
0
00
A
A
*
00
*0
U
A
I
A
*/
U
0
0'
A
0
0)
00
a
I
0.05~~
0.0
01
U
A Low Income (Cor. 0.576)
* Middle Income (Cor 0.643)
A High Income (Cor. 0.263)
q
0.10
A
A
0.20
0
AA
AA
AA
A
0
In
.0Midd0
A
A
0
A
-
0.25
0.30
0.05
0.35
State-level Opinion, MRP Estimates
0.10
0.15
0.2A
A.3
U)
0~
_
ED|
an
0
. All Incomes (Cor.
0.12
0.20
0.14
0.16
0.18
0.20
0.22
State-level Opinion, MRP Estimates
90
0.24
0.558)
0.26
m
(
r..9)
3
Low Inome (Car. 0.602)
:Middle Income (Car. 0.695)
A High Inome (Car 0.514)
0.25
State-level Opinion, MRP Estimates
Minimum Wage
(No Income Breaks)
0
0
A
A
___ -------
0.15
AA
A A
A
.6
I I
01
AA
Co
0.05
*
0
A A
*A
*A&
O
A
A
*A,
*
*
' A
**
i
.
A
kA
e"A''A
.11 **
0
A
1
1
0.30
0.35
Figure A14
Abortion Rights
(Equal Breaks)
Abortion Rights
0
An
0
129
. OIL A *0
A
64'
'L
1%64:
0A
A A
A ON
A*
M
a
A A* 0
A AO
NO
R.O. no
A a
A
:0
A
0
03
CD
ci,
A
q
Lawicome (Cor0.774)
(Cor
High
aicome(Cor
SMidde
A
0.1
0.2
State-level
0.3
04
Opinon, MRP
income
A
0
0.830)
0808)
0.5
0
0.1
0.2
Estimates
Low Inome (Cor. 0.787)
MkMW ftme (Cor. 0.824)
A High InoDme (Cor. 0.8W)
0.3
Abortion Rights
(No income Breaks)
Ln
0
L
8
0un
-to
i-
0
.
0.2
0.4
State-level Opinion, MRP Estimates
0.4
0.3
State-level Opinion, MRP Estimates
91
Alt
incomes (Cor. 0.824)
0.5
0.5
Figure A15
Environment
Environment
(Equal Breaks)
0
0
I
0
0i
a
a o A nomA (Cr
AH
m C r
c Ahi
.
6))
0)
.6
H
.8
I-
-]
U
A
S
Low
A
P
Income
(Cor,
0.588)
Income (Cr 0.640)
Income (Car 0.680)
Middle
A
0.2
0.3
0.2
0.5
0.4
0.3
Environment
(No Income Breaks)
(0
0
eU
0
S-
01
in
Ci,
6
0
m All Incomes (Cor. 0.610)
0.20
0.4
State-level Opinion, MRP Estimates
State-level Opinion, MRP Estimates
e
Hgh
0.25
0.30
0.35
0.40
State-level Opinion, MRP Estimates
92
0.45
0.50
0.5
Figure A16
Estate Tax
(Equal Breaks)
Estate Tax
0
0
a
2
A
0
**
*
%o
.
op~ *
AA*
I0
*
0o
0%
AA
0
A
A
*O
U
A
*A
00
A
I-
A
0
AA
A ,
0
AAA
A
A,
A
*
0soA
0*
0~
M
A
A
A
U S
At
A
U
nee(a.002
o
*
*
0
0.5A
0.5
0.0 04
Stteee OiinMREsm A
I-
A
a
P
I
0.35
* Low Income (Cor. 0 605)
* Middle Income (Cor. 0.629)
A High Income (Cor 0.251)
I
0.40
I
0.45
I
0.50
I
0.55
I
0.60
0.65
1 '
0.70
0.35
State-level Opinion, MRP Estimates
Estate Tax
(No Income Breaks)
In
a
a
a
0
b
a
.5
05
.6
F-
0?
All Incomes (Cor. 0.608)
I
T
-
.
0.44
0.46
0.48
0.50
0.52
0.54
State-level Opinion, MRP Estimates
93
0.56
.6
06
Figure A17
Gun Control
(Equal Breaks)
Gun Control
0
0
I
0,
FC?
0,
I.-
*
*
0
A
0.0
I
I
I
0.1
0.2
0.3
0.4
*
0.853)
0.705)
0A78)
Law Oncome (Car
Middle
(Car
High
(Car
Income
Income
I
I
i
0.5
0.6
0.75
A
U
Low
0
0.1
0.2
State-level Opinion, MRP Estimates
0.3
U,
0
0
an
TD
Ga
o -
Ui
0-
I_
0
.
S
0.1
0.2
I
All
Incomes (Cor. 0.700)
I
0.3
0.4
State-level Opinion, MRP Estimates
94
0.5
(Car.
MiddleIncome
A
High
0.4
State-leval Opinion, MRP Estimates
Gun Control
(No income Breaks)
Income
*
0.847)
(Cr. 0.705)
Income (Cor. 0.713)
0.5
0.6
Figure A18
Gay Marriage
(Equal Breaks)
Gay Marriage
6P
U
0
I
9
0
A
A .A
.
e
I
A
HAhicm
(o
A
.5)
3:
.8
Low
A
-
0-
0
0.15
0.20
0.25
0.30
0.35
0.40
0.20
0.50
0.45
0
0
0.25
0.30
I
O.35
I
0.40
0
0.45
State-level Opinion, MRP Estimates
State-level Opinion, MRP Estimates
Gay Marriage
(No Income Breaks)
U)
0
A
2
8)
.
0
.5
UA
Ui
llIcms(a.086
0)
I
I
0.25
0.30
1
0.35
1
0.40
State-level Opinion, MRP Estimates
95
Income (Car 0758)
SMiddleIncome (Cor. 0.822)
A High Income (Car. 0.810)
-
0.45
0.50
1
0.50
Figure A19
Federal Health Insurance
(Equal Breaks)
Federal Health Insurance
0
0
0
i
A
-
A
00
A
AHn4
A
A
OA
A
0
.
0V
*
U
U
go
(A
)
A
A A
A
A
A
AA
A
AL
n
A
A
A
A
A
9-
C?
'0
3O
A
M
ue.
*
MMdl
* *Cr
0
L
4)
AA
nom
nom Cr.065
4
06
CfO
a
.
A
0
q
0.1
04
0.3
*.
A
0.2
05
0
Low Incom (Car. 0.679)
*Middle Incomfe (Car 0.837)
A Highr kncme (Car. 0.733)
0.3
0.4
0.1
0.5
0.2
A
.
U)
0
6,
0
0j
.%
CI,
I-
0
.
0.20
0.30
0.25
0.35
State-level Opinion, MRP Estimates
96
All
Incomes (Cor. 0.789)
0.40
.
*Law Incme (Car. 0.655)
Middle kicome (Car 0.834)
A High Income (Car. 0.829)
0.3
0.4
State-level Opinion, MAP Estimates
Federal Health Insurance
(No Income Breaks)
0.15
A
0
State-level Opinion, MRP Estimates
(9
6,
00.
*.
.
0.5
Figure A20
Immigration
(Equal Breaks)
Immigration
&A
ADAN
A
P
A
A A
A AA*A
A
A
A
AE
:3
A
AA
A*A
.
A
A
A&
.7
A
A
t
A
A
01
a
0 _A
A4L 0 NA
01
0
A
**
A
AA
I
A
A
A
an*
A
aA on
A
e
A
AA
Cr.
A
A
A
0m
m
U AN-
'AA
%L
one
A
A
A
*a
on
A
on
ON
A
C?
**
c-
9
A
A
Income (Cor 0,577)
Middle hicome (Cot 0,604)
A High, Ine
(Cot 0.639)
ULow
7
0
I
0.6
0.7
0.8
0.9
0.60
-
0.65
State-level Opinion, MRP Estimates
N
I
I
I
0.70
0.75
0.80
U)
0
w
0
_0
9 U)
0 -
0
UAll
0.70
0.85
State-level Opinion, MRP Estimates
Immigration
(No Income Breaks)
e
Low Income (Cor. 0.521)
Middle Income (Cot 0.571)
A High Income (Cor o'6W)
0.75
0.80
0.85
State-level Opinion, MRP Estimates
97
Incomes (Car. 0.632)
0.90
0.90
Figure A21
Iraq War
(Equal Breaks)
Iraq War
0
0
B
*
I
A
*
'87A
0
0
AAA
*A
AA
*
A
.5
.5
.5
AA
Co
0)
.5
H
.5
I-
A
q
Low Income (Cor 0 746)
Middle Income (Cr 0.939)
A High Income (Cor 0.895)
I
I
I
0.1
0.2
0.3
I
0.4
.
.
0.5
0.6
A
A
0.3
0.2
0.1
0.5
State-level Opinion, MRP Estimates
State-level Opinion, MRP Estimates
Iraq War
(No Income Breaks)
0)
0
0
aE
a Nis
@1
0 1
0
U
U
a
Aom
0)
|-
0
SAll
0.2
0.4
Low Income (Cor. 0.753)
Middle Income (Cor. 0.939)
High Income (Cor 0.924)
0.4
0.3
State-level Opinion, MRP Estimates
98
Incomes (Cor. 0.922)
0.5
0.6
0.6
Figure A22
Social Security
(Equal Breaks)
Social Security
0
0
..
,
*
A
.
A.
*A 0
*
A.
4
8*,e
A
,
A
A
A
0
.
.
A
A
A
I
A
0~
A
As
Il
*
U0
U
~
.*A
S
0
.A1A
*
*
*
.4
4
A A
A
e A
A
0
.
-
,eA
A
A
e
~eA AA
.
A
ci-
A
*.
01
.A.
pn
A
.
E
M
A
s
ci'
06
I-
E
A
0
SI
0.40
0.45
0.50
0.55
0.60
*0
Low Income (Cor. 0451)
* Middle Income (Cor 0 448)
A High Income (Cor. 0.256)
I
I
0.65
A
ci
0.40
0.70
0.45
0.50
0.55
Social Security
(No Income Breaks)
0,
0
a
9
0
-
a
.
..
2
*0
*13
-r
a
Cl,
0,
0-
06
I-
0
.
0.50
060
State-level Opinion, MRP Estimates
State-level Opinion, MRP Estimates
a
aLow income (Cor. 0.424)
Middle Income (Car. 0.534)
AHigh Income (Car. 0379)
0.52
0.54
0.56
Alt
0.58
State-level Opinion, MRP Estimates
99
Incomes (Cor. 0.403)
0.60
0.62
0.65
0.70
Figure A23
School Voucher
(Equal Breaks)
School Voucher
Ui
0
0
A4
A
A
0A
A
a 0
a *A
AA
a
A
(0Cor.A
*A
**A
A
A
U..
A
A
so
A
GO
A
A
0
*~~
0
AA
A
A
.4
-.-
**WA
A
0.
A
A
Ae
A
U
As
s9
113
03
00
1-
*A
Low Inoome (Cor -0.438)
Middle Income (Cor. -0 283)
A High Income (Car. -0.090)
0.45
0.55
0.50
0.60
0.40
0.45
School Voucher
(No Income Breaks)
03
0
%
UO
ID
ID
U
0
U
UIN
ID
-r
ID
Al
CI,
I-
0
0.40
0.50
State-level Opinion, MRP Estimates
State-level Opinion, MAP Estimates
a
Low
income (Cor. -0.459)
a Middle Income (Cor. -0328)
A High Income (Cor. -0.058)
1I
II
0.40
Aa
0
0.45
0.50
State-level Opinion, MRP Estimates
100
Uncms(or
039
0.55
0.60
Figure A24
Free Trade
(Equal Breaks)
Free Trade
0
0
as
am
A
A Le
-
AA A
1
A
AA A
A
A
AAA
AA
0
a%
eee
0 A
A
0.
f.4 '
A,
Q
as
A
A
AAA
A
A
Em.ee
A
A
A,
A
A
A
A
A
A
0A
A
A
A
8
a
.L06
A
A
A
CUS
A
A
*mSon
d.2
I
A
_A
A
A
mEeA
ee0
ag
A
AA
. ~
40
.6
1-
.6
I-
m
0
Low Mocwme)Cor -0.252)
Middle Inomre (Car. -0 343)
Income (Car. -0.416)
-
a Low Income (Car -0.213)
a
0
Middle income (Cor. -0.266)
SHigh Income (Cor. -0.408)
AHigh
0.2
0.4
0.3
0.5
i
0.20
0.6
0.25
0.30
State-level Opinion, MRP Estimates
Free Trade
(No Income Breaks)
40
0
a
U)
U
a
0?
-e
62
0)
.5
I.-
0
.
0.25
0.35
0.40
State-level Opinion, MRP Estimates
All
0.30
State-level Opinion, MRP Estimates
101
Incomes (Cor. -0.394)
0.35
0.45
Validation: Scatterplots of MRP Estimates V.S. Weighted Averages of Statelevel Opinion
102
Figure A25
Minimum Wage
(Equal Breaks)
Minimum Wage
LO
AA
A
A
0
o
0
M
*
0
.2
0
00
A
1A
0*
A
a
0.15
0.20
o
A
no
e(Cr
.19
A
A A
0.10
A
A
AkA
0.05
A
A
.)
8
AA
A A
A
(
C.
*Low Icome (Car 0.200)
*Mde Inoor-e (Cor 0 708)
A High icme (Cor 0.121)
0.25
0.30
*Middle hcm (Car 0,465)
A High Inome (Car. 0.559)
0.35
.
I
I
0.05
0.10
0.15
0.20
0.25
State-level Opinion, MRP Estimates
State-level Opinion, MRP Estimates
Minimum Wage
(No Income Breaks)
0
C.,
0
KU
U)
*
U
U
U
0)
U
9
*
U
*
0
0-
I
0
U
0
U
*
U
*
0
Co
U
I
II
I
1
.1
0.12
0.14
0.16
At) Incomes (Cor. 0.715)
I
I
I
I
I
0.18
0.20
0,22
0.24
0.26
State-level Opinion, MRP Estimates
103
0.30
0.35
Figure A26
Abortion Rights
(Equal Breaks)
Abortion Rights
0
S.
%
AA
0n
.
07
0
U
U
CD,
A
UA
A
A
4A&
0
A
* Low kncome (Cor 0.752)
SMiddle hicome (Cor. 0.941)
A High icome (Cor. 0.498)
0.1
0.2
0.3
0.4
S Low
.
A
0.5
I
I
I
I
0.1
0.2
0.3
0.4
0.5
State-level Opinion, MRP Estimates
Abortion Rights
(No Income Breaks)
01
1
0
0
a
0.2
0.3
All Incomes (Cor. 0.994)
0.4
State-level Opinion, MRP Estimates
104
Inome (Car 0.477)
I
State-level Opinion, MRP Estimates
-i
Income (Car. 0.739)
icome (Car. 0.940)
Middle
High
0.5
Figure A27
Environment
(Equal Breaks)
Environment
0R
C
a
0
#
0o
*S
A
A
A
.2
A
C
Co'
*
I
0.2
*Low hroe(Car
A
I
0.3
Midde
9
0.594)
0.928)
ncome)(Cor
High bncoms (Car. -0.214)1
0.5
0.4
0.2
State-level Opinion, MRP Estimates
0.3
State-level
_
1
aU
01
.0
.U
l
noe
(o.091
Crj
I
I
I
I
I
0.20
0.25
0
0.30
03
O 35
0.40
State-level Opinion, MRP Estimates
105
0.45
0.50
I
Income (Cor. 0587)
Income (Cor.0.958)
A High Income (Cor. 0. 175)
0.4
Opinion, MRP Estimates
Environment
(No Income Breaks)
CP
0
A
Low
SMiddle
0.5
Figure A28
Estate Tax
(Equal Breaks)
Estate Tax
C?
0
*A
.
.It
C?,
*
00
A
01
.2
6
A
*
0
A
.
so
A~
A
6
A
01
64
m
an
6
0)
A
A
A
SLow IncomeCo. 0.541)
SMiddle income (Cor 0 562)
A High Income (Cor 0.112)
' I
0.35
I
I
I
0.40
0.45
0.50
0.55
0.60
I
_17 '
0.65
0.70
* Low Income (Cr. 0.535)
* Middle Income (Cor. 0.737)
A High Income (Cor 0,006)
0.35
0.40
0.45
State-level Opinion, MRP Estimates
Estate Tax
(No Income Breaks)
*
U
*
a
U
.
CL
U
53-)
V
01
61
6
Ua
53)
6i
6
All Incomes (Cor. 0.725)
.
0.44
0.50
0.55
State-level Opinion, MRP Estimates
0.46
0.48
0.50
0.52
0.54
State-level Opinion, MRP Estimates
106
0.56
0.60
0.65
Figure A29
Gun Control
(Equal Breaks)
Gun Control
0
Ci
gO
**
0
A
LO-
A
6
A
17
A
A
2
0
0
.2
6
6
..
Ae
6
As
A
.A Al
CD
A
A
A*
* Low icame (Cor. 0.733)
* Middle Income (Cor 0 962)
A High income (Cor. 0.473)
I
0.0
I
0.1
I
I
I
I
I
I
0.2
0.3
0.4
0.5
0.6
0.7
A Low Income (Cor. 0.732)
* Middle income (Cor. 0.972)
A High Income (Cor. 0.388)
0.1
0.2
State-level Opinion, MRP Estimates
0.3
State-tevel Opinion, MRP Estimates
Gun Control
(No Income Breaks)
Co
0
A
0
U
0.1
0.4
0.2
0.3
0.4
State-level Opinion, MRP Estimates
107
All
Incomes
(Car.
0.5
0.993)
0.5
0.6
Figure A30
Gay Marriage
(Equal Breaks)
Gay Marriage
%e
A
.6*
*
0m
a
*U
*A
A A
A
A
Hg
AN
A
7~
.05
-
01
A
A
Au
En
-
n
nII
(
AA
.
(
A A
AAA
A
A
A
AA
f
MA
A Low Income (Cor. 0.682)
: Midde Income (Cor 0.723)
A High Income (Cor. 0.183)
0.15
0.20
0.25
0.30
0.35
0.40
0.45
0
0.50
0.20
0.25
0.30
State-level Opinion, MRP Estimates
0.35
0.40
0.45
State-level Opinion, MRP Estimates
Gay Marriage
(No Income Breaks)
IPD0
d
01
0
a
6j
.
0.25
aLow Inome (Cor. 08682)
Middle Income (Cor. 0.834)
A High Income (Car. 0.102)
0
0.30
0.35
0.40
State-level Opinion, MRP Estimates
108
All
Incomes (Car. 0.870)
0.45
0.50
0.50
Figure A31
Federal Health Insurance
(Equal Breaks)
Federal Health Insurance
0
.
A
0
gI
C5
A
*
0
C2
d
%
*0
9
**
A
0
A
A
A
0
00
41
A
j
-'
A
00
AA
gInoe(Ar.011
A A
EALOW
Income (Cor 0.714)
SMiddle
income (Cor 0 946)
A High Income (Cor 0.116)
i
0.1
0.2
0.3
0.4
0.2
0.1
0.5
State-level Opinion, MRP Estimates
Federal Health Insurance
(No Income Breaks)
0
.0
0
01
dl
0
0
0
t2
0
All Incomes (Cor. 0.968)
0.15
0.3
0.4
State-level Opinion, MRP Estimates
0.20
0.25
0.30
0.35
State-level Opinion, MRP Estimates
109
0.40
0.5
Figure A32
Immigration
(Equal Breaks)
Immigration
0
CP
*
0
0
,.
.
00
Co
**
e
0
0.
*
A*
A
0
A
N
A
A
A
A
AA IA:
A
C.
AA
o
AA
A'
.AA
AA
a
-A
17
0
* Low Income (Cor 0 552)
* Middle Income (Cor 0.497)
A High Income (Cor. -0.218)
0.6
0.7
A
a Low Income (Cor. 0.490)
* Middle Income (Cor. 0.649)
A High Income (Cor. -0.324)
I
I
I
I
0.65
0.70
0.75
0.80
I
0.60
0.9
0.8
Immigration
(No Income Breaks)
-
U)
C0
0
01
-
0
CD
0)
Co
0
0n
0
Co
0
0
n All Incomes (Cor. 0.941)
I
0.70
1
1
0.85
0.90
State-level Opinion, MRP Estimates
State-level Opinion, MRP Estimates
Co
0)
0
U
LA
A~
A
7
.
0.75
0.80
0.85
State-level Opinion, MRP Estimates
110
0.90
Figure A33
Iraq War
(Equal Breaks)
Iraq War
0i
0
A
A
A
A
A
.0
6\
*.0
.2
AA*
A
AA
A
A
A
SA
.5
S
0s
C,
A
-'
[o Low noame (Car
0.39)
6
U
A
I
SI
0.2
0.3
0.4
0.5
0.6
0.1
0.3
0.2
Iraq War
(No Income Breaks)
10
0
U
U
Irv
a,
0I
IU
0,
U
N
6
0.2
0.4
Low Income (Cor. 0.398)
Middle Income (Cor. 0.938)
High Income (Cor 0. 140)
I
I
0.5
State-level Opinion, MRP Estimates
State-level Opinion, MRP Estimates
0.4
0.3
State-level Opinion, MRP Estimates
S11
0.5
0.6
A
A
M Midde kncome (Car. 0 928)
A High income (Cor -0.047)
0.1
A
AII
A.O
0.6
Figure A34
Social Security
(Equal Breaks)
Social Security
~.
.?
*
..
-
31
a
0*
e
p.
A
6
A
A
A
A
AAA
C
AA
It
A
'7N-
02
A
A
SI
0.45
0~
Low Income (Cor. 0.307)
Middle Income (Cor 0.555)
High Income (Cor. 0.099)
TI - - - 0.50
0.55
0.60
0.65
I
0.70
A Low Income (Cor. 0.296)
. Middle Income (Cor. 0.619)
A High income (Cor. 0.238)
I
I
I
0.40
0.45
State-level Opinion, MRP Estimates
0.50
0.55
0
01
a,
All Incomes (Cor. 0.412)
I
0.50
0.60
State-level Opinion, MRP Estimates
Social Security
(No Income Breaks)
CL
A
A
A
A
0.40
&A
A AA
0
A
AA,
I
I
I
I
0.52
0.54
0.56
0.58
State-level Opinion, MRP Estimates
112
0.60
0.62
0.65
0.70
Figure A35
School Voucher
(Equal Breaks)
School Voucher
0
.0*
.*4.*
0
0-
000.
0
0
p
.0
*
0
0A
0
A
A
0"
0
*
0
*1
U
0.0
0
0
A
A
A
A
A
2
.2
0.0 04
0
@ 0
0
*
A
A
A
AA
.5
.
75
a
0
A
AM
A
-9
07
g
A
0
AA
AAA
A
A
A~
Law Income (Can. 0,352)
Middle Income (Car 0 653)
High Income (Cor 0.561)_
__T . ....
0.40
0.45
0.50
0.55
0.60
0.40
0.45
State-level Opinion, MRP Estimates
e
-
61
01
C
61
Mm
.
0
0.40
H
0.50
0.55
State-level Opinion, MRP Estimates
School Voucher
(No Income Breaks)
0
U
0
Low income (Car. 0.351)
Middle Income (Co. 0.339)
A High Income (Car. 0.609)
0
0.45
0.50
State-level Opinion, MRP Estimates
113
le Icms Cr 099
0.60
Figure A36
Free Trade
(Equal Breaks)
Free Trade
0
.a
C?
LIS
6
me%0
A
0
*
**
A.
AMA
A
E*
"We
6
A Low
Income (Cor 0.225)
Income (Cor 0 575)
U0
UA
0.4
0.3
0.5
0.6
I
I
0.20
0.25
Low Income (Car. 0.244)
Middle Income (Car. 0.387)
A High Income (Car. 0.706)
0
03
0.30
0.35
Free Trade
(No Income Breaks)
0
A
0)
Co
.0
0O
'0
C
0
.
0.25
0
0.40
State-level Opinion, MRP Estimates
State-level Opinion, MRP Estimates
0.30
State-level Opinion, MRP Estimates
114
All
Incomes (Car. 0.528)
0.35
AAA
U
U
A High Income (Cor 0.680)
0.2
A
AA
A
0
* Middle
M
AA
03)
A
AA
A
A
A
0
0.45
Tables Showing T-Tests of MRP Estimates
115
Table Al: T-Tests of MRP Estimates
Income
Opinion
Low vs. Middle
Low vs. Super-High
Middle vs. Super-High
Super-High vs. Equal Breaks High
Equal Income Breaks
Minimum Wage
p-value
3.26-10-35*
3.71. 1044*
Low vs. Middle
Low vs. High
Middle vs. High
Abortion Rights
p-value
0.10
Social Security
p-value
Estate Tax
p-value
1.24-1017*
1.79.1012*
5.
10-7*
8.02.10-4*
0.40
0.03*
4.29 10-44*
3.39- 10-30*
6.55- 10-4*
0.78
6.78.10-24*
0.17
4.33.10-43*
8.36-10-14*
2.74.10-44*
5.54.10-24*
2.84.10-3*
1.63.10-38*
5.49-10-9*
5.90.10-31*
1.42-10-13*
0.12
2.88- 10-43*
2.62 10-29*
4.13. 1017*
indicates significance at p < 0.05
Table A2: T-Tests of MRP Estimates
Income
Opinion
Low vs. Middle
Low vs. Super-High
Middle vs. Super-High
Super-High vs. Equal Breaks High
Gun Control
p-value
0.18
0.17
0.71
0.85
Gay Marriage
p-value
0.22
0.12
0.77
1.32.105*
1.81.10-4*
0.76
1.20.10-4*
0.01*
School Voucher
p-value
0.51
2.81.10-3*
6.31.105*
1.01.10-3*
Environment
p-value
8.71.106*
3.15-1017*
6.33-10-6*
0.02*
Equal Income Breaks
Low vs. Middle
Low vs. High
Middle vs. High
indicates significance at p < 0.05
0.53
0.99
0.44
0.02*
0.08
1.21.10-4*
1.43.10-11*
2.95.103*
Table A3: T-Tests of MRP Estimates
Income
Opinion
Low vs. Middle
Free Trade
p-value
6.49.1010*
Iraq War
p-value
1.40.10-14*
Immigration
p-value
0.02*
Fed. Health Insurance
p-value
2.68.1016*
Low vs. Super-High
Middle vs. Super-High
1.38-1035*
1.89-10-29*
4.48-10-14*
5.78-10-18*
2.50-1014*
1.78.10-25*
2.29-10-7*
Super-High vs. Equal Breaks High
Equal Income Breaks
5.58-10-19*
6.99-10-5*
0.18
Low vs. Middle
Low vs. High
2.62.10-3*
5.77-10-26*
0.46
3.52-1012*
9.04-10-11*
6.17.10-13*
*
Middle vs. High
indicates significance at p < 0.05
6.13-10-22*
0.37
0.06
6.17.10-13*
1.17-10-16*
0.11
116
7.71.10-25*
2.59-10-7*
Scatterplots of MRP Estimates (IV) V.S. Senators' CVP Ideal Points (DV)
117
Figure A37
Minimum Wage
(Equal Breaks)
Minimum Wage
A: *
*
0
*A
*~A
*u
&
on
%
*
a..
0 :;:gi.£AA*
*A
%
A
UA
*
A
A
A A
*I
A m A AA
A
**
e A * :J'f*~
A A A J. A
00 640.0
0
A
A
A
AD*
00.
A
AA
AA
N~
AA
F
A
A*
M m
AA
A
A
so
a
NO
C\!
IA.
A
UAI
*
2
o
%
*
EU
mm 3
0
S
A
A*
U
0~
A
A
A A
*~
0
-:
A
A 0A
CL
A
A
m
VON
.
A
*
A
~0 ~*
* Og
3mb *
m
0
AA
AA
*A AAA
Ae AA
AA
AA
At
AUA
Au
A
A
0
A
A
A
A
cci
leO
a
In
7r-
A
A
m~owkicam
A
A
A~~
A
0
Low Income
*Middle Income
AHigh Icome
I
I
I
0.05
0.10
0.15
I
1
0.25
0.30
I
0.20
0.35
0.05
0.10
0.15
State-level Opinion MRP Estimates
0.20
0.25
State-level Opinion MRP Estimates
Minimum Wage
(No Income Breaks)
('m
on
0
*
U
U
0~
NOm
IN
0
A
HAhkcm
II
II
cli
I
I
I
I
I
I
i
0.14
0.16
0.18
0.20
0.22
0.24
0.26
State-level Opinion MRP Estimates
118
0.30
0.35
Figure A38
Abortion Rights
(Equal Breaks)
Abortion Rights
A
AsMO
A
*
A
0
w
m
A
do
As9
M
0
*A
A
n
C%
0
AbiwfMwW&0
A0% AMW06A
A
U
A
&A
A
a
A
a
AA*A
AWO
A
IUA 0
A
AA
A
C-
a-
AAUit
A
w
A
AS
AAC
A
9
A
A
.
Low Income
.
MuiddeIncom.e
A
High
0.1
* Low Income
* Middle Income
A High Income
Income
I
I
I
0.2
0.3
0.4
State-level
Opinion
0.5
a
*
.
0.3
a
.
U
.U
0
*
U
U
Un
a
0.2
E
UA~A
A
em
0.4
State-level Opinion MRP Estimates
MRP Estimates
Cs
M
a
a
a
.
ag~
a
ae
U
a
ona
0.4
0.3
State-level Opinion MRP Estimates
119
U
aU
0
I
0.2
CNI
0
SM
A
Abortion Rights
(No Income Breaks)
0
-*
S
A
CD
N
*S
A
U~i
0.5
0.5
Figure A39
Environment
(Equal Breaks)
Environment
*
0
A
fW
0~
*
U
*
0
aA0
U
0~
*
pSA
A
m
tlA
~~%AA
A
A
jA &A a-
C,;
.
A
A
0
A
A
N ;10
A
0
*A
iii
Low kconie
*Middle mInone
A High Income
1
AnIe
M
M MA r
a
0.3
A
%
A
A
R
siA
te
0.4
State-level Opinion MRP Estimates
Environment
(No Income Breaks)
1P0
0
04
0
0
0
0-
0-
0.20
f
A
0tt-ee Opno
0.2
0.5
State-level Opinion MRP Estimates
*ii
o
1
0.4
0.3
as
a Low Incorm
* Middle Income
A High Incom
0
mA
M
me
n, Hig
AAA
A
US
I
0.2
f*
A
aUA
0
A
2
A
A
a
*
i*dA
*AS,
C-)
2
A
e 0
A
N0
04
A
0.25
0.30
0.35
State-level Opinion
120
0.40
MRP Estimates
0.45
0.50
0.5
Figure A40
Estate Tax
(Equal Breaks)
Estate Tax
0
*A
0
M
A
A
0
U
U
0
4A
H.g
M
**/A
*
A
A
a-
0~
0
U
A
so
A
P
0
A
LO
..
~6
A*
A A
&A
.
B
A
*w*%.n'omAA
L
0
2
Al'
%e
ci)
0*0*
U.
Low Inoe
*Middle ncoe
AHigh Income
I *.
:
A
*A
U
*
A
A
A
I
I
I
I
0.45
0.50
0.55
1
0.60
~
%j
0.40A
A
C\3
0.40
-
0
07
0.65
0.70
I
0.45
0.40
Estate Tax
(No Income Breaks)
N
0
7~
LI
0~
0
U
U)
S.40.0
*
~
*A
AA
A
U
6-
ME
on
-
Wo
E
I
Ci
I
i
I
I
I
1
0.44
0.46
0.48
0.50
0.52
0.54
State-level Opinion MRP Estimates
121
At
A
A
AA
A
A
I
I
0.50
0.55
State-level Opinion MRP Estimates
State-level Opinion MRP Estimates
a,
E
A
I.t060
A
06
R
ULow Income
*Midce Income
A High Income
i ~
p
~A 0
1
0.56
0.60
0.65
Figure A41
Gun Control
(Equal Breaks)
Gun Control
B
A
0
0
ae *AA
A* A
A
A
A
0 a
WamL
*&
A
*A
A
A au
of
A0
P0A
0~
4-
r
AA
eO
B
Al Ad
AA
a(L
A
B
A
C0
CdI
BAA
%
A
BA
*
I
0.1
*nA
*A1
Income
A High Income
A High Irwcome
0.0
Middle
1
La
I
I
I
0.2
0.3
0.4
0.5
0.1
0.7
0.6
0.3
0.2
State-evel
State-level Opinion MRP Estimates
Gun Control
(No Income Breaks)
CD.
Un
Cd
*
0
0I
U
a
U
U
U
U
*0 0
0D
(D
'a.
U.
*
II
0.2
0.3
1
1
0.4
0.5
State-level Opinion MRP Estimates
122
0.4
Opinion MRP
Estimates
0.5
0.6
Figure A42
Gay Marriage
(Equal Breaks)
Gay Marriage
AM
A
A
0
A
go
.
AA
e
.
;_MA
MRA
AA
C
A
MpAo Am on o o
a
A.
AiI~
A
WinRAO
: 0
A
42
0
A
a.
ma
N
U
A
*
A
1
A M
A *I M*
0%
0.30
gA
0.35
Mea
0
0
SAA :0o
S
w A
A S
M
0.40
0g4
A*
AA
U
*U
- f AA*
AAin'
a
A
0cu A
A
AAA
S
MS
0.45
0.25
0.50
0.30
0.35
0.40
0.45
State-level Opinion MRP Estimates
Gay Marriage
(No Income Breaks)
-
-
.i
Rf . 0 .
.
U
*
n
urn
U
n
0
0
.n
In
me
ag in=
in
0
*0
Un
mea
0
4)
*in
a
.n
0
0.30
.
Af
om A
* Low Income
* Middle Income
A High Income
State-leve Opinion MRP Estimates
Q_
A
MAU
0
US
0
M.
A
0.25
N6w
0
0
Low Income
Middl Income
A High=Icm
0.20
a
S
Ae'left
* .0
16
Oa&&*&
4U
d&
An :A Aft,
IV,
AU
A~
*
a
EmO
SA
A
"I
U
A
M
0.15
0
0
*0
A
0.35
State-level
0.40
Opinion MRP Estimates
123
0.45
0.50
0.50
Figure A43
Federal Health Insurance
(Equal Breaks)
Federal Health Insurance
A
*
*
*
6s
a
A
U *
a-
*
0
*~
U
*
=4
0~
ONE.
an
A
04
91
.
aS
A
A&
5
% PA
.UU~
*
U A.
0
4A
,
-- - -T-
0.2
A
A
a
A
A
0
AA
-S
A
U,
AA
ApA
A A
A
A
o *
OU :
11111 A"*
A
6
am
m
m IN
A
a:
S
U
*Al
*
A
a
:0o
in0omA06
A
A
A
A
AAAA
A
A
4fA
"*S
E
AA
a
-
0
a-
A A
AS A
S
A
0
A
A
A
a
* Low Income
* Middle Income
A High Income
A
A
5
m
A.
A
:IaA
A
A 4At
#%~AAA A 4J&
~
EU
4u~EEEE
T
a
A
*
*
&
A
S
S
6
A
A
A
A
A
A
*Mididle Income
A High Income
I
I
0.3
0.4
0.2
0.5
0.3
State-level Opinion MRP Estimates
State-level Opinion MVRP Estimates
Federal Health Insurance
(No Income Breaks)
0
a
9
-
*a
0
0
Sa
a
* *L
U
mu
ml:.=
0
u
u
0 a
a
a
0.20
0.4
0.30
0.25
State-level
0.35
Opinion MRP Estimates
124
0.40
0.5
Figure A44
Immigration
(Equal Breaks)
Immigration
A
A
A
A
A
A
A
AtA
A.
A
A All A
A
0
*A
o ao o
n
ms
A
m
em
21
A
AA
A
Un
rn
A
AS a
A
U
0
A
s~
ea .lo.
em A
A
soe
U
E
AA A
Urnsa
a
A"
A
A
AA
ma.
use
AA
MeA
A
AA
OmAE
ema
.
6 a
AD a
on
A
0
CL
0
Ln
0
A
A
A
A
AA
hA
e
A
A
a
A
Anne
ee.o
I
0.60
I
0.65
I
0.70
I
0.75
AA
AA
e0
U
mbo
An
I
0.80
0.65
0.90
State-level Opinion MRP Estimates
I
I
1
0.70
0.75
0.80
aa
ina
a
a
-n
*
-
-
*
.m
-
a
*0
0
U
U
U
mm
u
in
0.70
a
0.75
0.80
0.85
State-level Opinion MRP Estimates
125
.a
a
0
0.85
State-level Opinion MRP Estimates
Immigration
(No Income Breaks)
-0
ee a
A
A
*Low &icme
* Middle Income
A High Income
1
0.85
e Em
a A
A
i
amon
* Middle Income
A High Income
I
0.55
A A m m.
0.90
0.90
Figure A45
Iraq War
(Equal Breaks)
Iraq War
C?
AM
U
A
N
CA
0
*
04
u*w.
AO
A AM
a*
*
Aj
AAV.
A
aT
U AM 0~
.. ....
o .. .e
INN
Cbu
A"
Ib
A
Hig
a
A
A
0
*
C,
A
E
111
C
t
U
A
A
"A
a-
*e~A
9
0
A
A.
o
0A
A
A0
moD
a
A
A
A
A..A
A
.
6
AcoLmo
01
C4
9-
0.2
0.3
0.4
0.5
* Low kIcome
* Middle Income
A High Icomie
I
0.3
0.6
0.4
State-level Opinion MRP Estimates
State-level Opinion MRP Estimates
Iraq War
(No Income Breaks)
0
C4)
0
0
No
0n
6-4
I wa 0
0.35
0.5
0.40
a
a
0.45
110wa
0.50
State-level Opinion MRP Estimates
126
0.55
0.6
Figure A46
Social Security
(Equal Breaks)
Social Security
0
A
a
Ci
0
0l
Cii
@0
soA*A
am II
A
*A
OP~
A
0
04
0
"
0 mL *a
.
.
a.
*
*La
.
6A
A
AJ&8u*..9 *
SA
0
(0
A
A
.Ae
6A
*:&A A
10
A
A*
~
*~
A
&A
.
"
0
Ao
I
I
0.45
0.50
.0
SI
I
0.60
AA
.ho0 000
Ak
ii
0.55
AA
A
A
-a*
* Low Income
* Middle Income
A High Income
I
A
AAA
do
*
* Low Income
.
Middle Income
A High Income
0.40
*A
.
A&
EM%
*A
A
e60***
A
&A
0.65
0.70
0.40
0.50
0.45
*A
I
I
0.55
0.60
A
State-leve Opinion MRP Estimates
State-level Opinion MRP Estimates
Social Security
(No Income Breaks)
0
0
a
U)
IN
au
IN 0
6-
0-
I
I
1
I
I
I
1
0.50
0.52
0.54
0.56
0.58
0.60
0.62
State-level Opinion MRP Estimates
127
A
1
1
0.65
0.70
Figure A47
School Voucher
(Equal Breaks)
School Voucher
C4
0
A
* 0 0
AW
me
A
0
*
0
ON
A
Am
gAAOp
A&NJ *
A
A
A
A Ajt
OW
**
O
A
A
*
c
a
-P'.
0
M
A
AA
A
0
A
41110 *
*J0A
a
,
A
Ai
A
cmA
ON
A
ANA
AAU
A
A
0
T
Ae
41*
A
A
A
I
&~ A
0.45
0.40
0.60
01
0
U
U
*
U
U
U
0
U
*
Eu
U
U
*
U
*
*
*
-
EU.
U
0
0
U
*fl
U
*E
U
U
U
3U
*
U
U
U
*
*
Eu
U
U
*
U
U
*
U
U
U
*U
*
*
*
U
U
U
Eu...
~
*
*
0
U
E
a
U
U
U
0.50
0.45
State-level Opinion MRP Estimates
128
U
U
Eg
0.40
A
*
A,
0 AM*
M
A
wc
A
A
0.50
State-level Opinion MRP Estimates
School Voucher
(No Income Breaks)
(f)
A.
4
I
0.55
State-level Opinion MRP Estimates
U)
0
A
A
mc
01i
Low Income
Middle Income
AHigh Income
0.50
Ac
me
*Low In.oNO
*Middle Income
A High Income
0.45
A
0
8 f '%AN
Ac
A
N1
0.40
A
A
C%
*W
*U
A
A Oh*
4AA
'At~~
@A01
f
DA As
L No
o
*
so
N&
A
'~A~
A A
cA
a0
U
A
0
C-
0
AgA a
A
0.55
0.60
Figure A48
Free Trade
(Equal Breaks)
Free Trade
Co
*
A
0
*
A
Sa
min
0
A
A
A
*:
*
a
1%
:0
*~
N S
*C
e
e~e
00I
0E
A
A
e
AA
A
U~
M
N
A
GA AA
AA
AhA
A AA
*~mme68
0
ye
*1
*
log
*ee
AA
AA A
AA
A
an
cc
CO us
w *
*
2
w
em
oj
a t
0
N
-0
-
me
;6
0
;;
1 6
AnoA
0.3
0.4
A A
A A
A
A
A
AA A*
A
A
AA
A
A
A
A
A
A
A
A A
At0
oUt
LOPW%
~
a Low lncA
High
OR
0.2
A A
*
A A
A
soA
AAA
Lo
A
40 i*
of*e
a.
A
~A
*5
A
A
Me
*-
5*
A
A
A
AA A
A
6
A
A
N
III
A
A
A
A
AA
A
A
A
me
A
A
A
A
0 e.
S
A AA
&A
A
A
A A
j148.
A£
9L
AA
AA,
AAZ
A
*Middle Icome
AHigh Icome
0.5
0.20
0.6
0.25
0.30
Free Trade
(No Income Breaks)
C0
<o
0
me
0
noU
ii
-C
'.
I
a
a
On
U
a
0
II
U
0
C4
0.25
0.35
0.40
State-level Opinion MVRP Estimates
State-level Opinion MRP Estimates
0.30
State-level Opinion MRP Estimates
129
0.35
a
0.45
Scatterplots of Tausanovitch & Warshaw Ideology Measure V.S. Senators'
CVP Ideal Points (DV)
130
Figure A49
Minimum Wage
c'J
0
o
N
0
05
0
N
U
0-
ON9
.U=
*E
C4J
C
0
co
V~j
0;
I
-0.4
-0.2
0.0
0.2
T & W State-level Ideology Measure
131
0.4
Figure A50
Abortion Rights
Co
0~
*
e
-
.
0)
C
0
a-i
0i
V
. Ue
a0
0)
U
0
CU
C
Ua
CU
a
U
e
U
N
0
-0.4
-0.2
0.0
0.2
T & W State-level Ideology Measure
132
0.4
Figure A51
Environment
co
Um
a,
C
0
0~
U)
V
cm
U
a0
*ii,
o
0
0
M
Cu
C
0)
Co
U
_
M M
-
M
M
U-
M
cmJ
M
U
N
EU
M
M
-0.4
0.0
-0.2
0.2
T & W State-level Ideology Measure
133
0.4
Figure A52
Estate Tax
c;-
U,
a.
a-0
0
U
= =
0
a-
*
0
E.
U
U
=
U
-w
0
Um
0
C,,
0
(4;
-0.4
-0.2
0.0
0.2
T & W State-level Ideology Measure
134
0.4
Figure A53
Gun Control
IVj
6
U
Um
0E
aC
>
N
cmJ
0
=
a~~
ms
0m
0
=
0
E m
m
a
=
U
0U
*m
0
-0.4
-0.2
0.0
0.2
T & W State-level Ideology Measure
135
0.4
Figure A54
Gay Marriage
,IJ.
6
U m
c\
m
m
a.
m m
*0
-0
m Um
m
m
*
m
C
Um
m
m
m
m
*U
-0.4
-0.2
m
ma m m
mI&=m
m m
N~
6
U
m
0.0
m
m m
m
EU
0.2
T & W State-level Ideology Measure
136
0.4
Figure A55
Federal Health Insurance
0
LO
eU
,bN
ME~ a
0
C
es
'"
0
0
.
mo
m
a
m==m,U0
"
0m
m
LO
a
I
07
-0.4
-0.2
0.0
0.2
T & W State-level Ideology Measure
137
0.4
Figure A56
Immigration
UEU
C',
C
6.
E
-O
0
0)
V
0~
0
~CI~
0
Cu
C
0
0
00 o
a'
M M MU
woO
C',
O-
-0.4
-0.2
0.0
0.2
T & W State-level Ideology Measure
138
0.4
Figure A57
Iraq War
UU
U
6v
0
.
-
0
U
a_
0
-
0
-
m
a
g
U
e=
m M
M=ma ME
m' M
M
%M
S
0i
0
M m
0
U
EU
0~j
M
U
*
U
U
M
.
-0.4
-0.2
0.0
0.2
T & W State-level Ideology Measure
139
0.4
Figure A58
Social Security
U
0
,-
-
EU
U
un
U
a
U
e
I
U
a
17
*
a
a
e
enU
U
U
U
U
0m
-0.4
-0.2
*
*
*
U
0.0
u
0.2
T & W State-level Ideology Measure
140
0.4
Figure A59
School Voucher
cm
U
0
-
Q_
6 -
>)
C
V
0
-7
U
*u
U
U
*
W
e
U
Cm
*
-0.4
-0.2
0.0
.
0.2
T & W State-level Ideology Measure
141
0.4
Figure A60
Free Trade
co
U
U
U
U
C
c
~0
-,
cmJ
0;
0
Euu
U
C
Cf)
0
0
C~J
0
-0.4
-0.2
0.0
0.2
T & W State-level Ideology Measure
142
0.4
Bartels Replication with Recent Senate Votes
143
Figure A61
Minimum Wage, Low Income Opinion
mwOOl
mw002
mw003
mw004
mw005
mw006
mw007
mw008
mw009
mw010
mwOl11
---
--
-------
mw012
mw013
mw014
mw015
mw016
mw017
mw018
mw019
mw020
mw021
mw022
mw023
mw024
mw025
mw026
mw027
mw028
mw029
mw03O
mw031
mw032
mw033
mw034
I Bartels' estimate
-100
-50
0
coefficient
144
50
100
Figure A62
Minimum Wage, Middle Income Opinion
mwOOl
mw002
mw003
mw004
mw005
mw006
mw007
mw008
mw009
mw010
mwOl01
-
p
mw012
e-
I
mw013
mw014
mw015
mw016
mwOl 7
mw018
mwOl 9
mw02O
mw021
mw022
mw023
mw024
mw025
mw026
mw027
mw028
mw029
mw03O
mwO3l
mw032
mw033
mw034
I
-
Bartels' estimate
--e-+I
I
i
-80
-60
I
I
-40
-20
coefficient
145
0
20
40
Figure A63
Minimum Wage, High Income Opinion
mwOOl
mw002
mw003
mw004
mw005
mw006
mw007
mw008
mw009
mw010
mwol11
I
--
,--+-'I
p
I
mw012
mw013
mw014
mw015
pe
-
mw016
mw017
mw018
mw019
mw02O
mw021
-'
--- +--
mw022
mw023
mw024
mw025
mw026
mw027
*
mw028
mw029
mw03O
mw031
mw032
mw033
I
*
mw034
'I
0
-50
coefficient
146
Bartels' esmate
50
Figure A64
Abortion Rights, Low Income Opinion
abO01
ab002
ab003
ab004
ab005
ab006
ab007
abOO8
abOO9
abOl 0
abOl1
ab012
ab013
abOl4
ab015
ab016
abOl 7
ab018
ab019
-.
p
S
e
I Bartels' estimate
0
-50
coefficient
147
50
Figure A65
Abortion Rights, Middle Income Opinion
------
abO01
ab002
ab003
ab004
ab005
ab006
ab007
ab008
abOO9
abOl 0
abOl1
ab012
ab013
abOl 4
abOl 5
ab016
abOl7
ab018
abOl 9
1
p
p
'p
I
-----
w
0-I-F
p
I Bartels' estimate
-60
-40
0
-20
coefficient
148
20
40
60
Figure A66
Abortion Rights, High Income Opinion
-2+
abO01
ab002
ab003
ab004
ab005
ab006
ab007
ab008
ab009
abOl 0
abOl1
abOl 2
ab013
ab014
ab015
ab016
abOl 7
ab018
ab019
p
p
I
p
p
I.
ip
p
p
Bartels' estimate
I Bartels' estimate
I
p
-100
-50
9
0
coefficient
149
50
100
Method: GEE Models
150
I implement another second stage estimation technique to measure the responsiveness
of the senators to their constituents by income group and across issue areas. The Generalizing Estimating Equation Model (GEE) allows for correlation across senators' votes on a
particular issue. The primary quantity of interest is the effects of the independent variables
(income group opinion MRP estimates) on senators' votes, fixed within clusters by state,
which makes the GEE model a suitable choice (see Zorn 2001 for further discussion).
GEE modeling does make the additional assumption that all of the votes are drawn from
the same conditional mean distribution and are exchangeable for the purposes of estimation.
However, it is able to avoid many of the rigid assumptions that accompany OLS estimation.
The parameter estimates are the population-averaged effects and, as GEE is a semiparametric regression technique, it relies only on the correct model specification and mild regularity
conditions in order to produce consistent estimates. By contrast, the estimates in generalized
linear models are more sensitive to the variance structure. The estimates are obtained using
the Newton-Raphson algorithm, whereby the score (s(6(t)) determines whether to move right
(if positive) or left (if negative) and the Hessian (H(6(t)) or second derivative assesses the
curvature. The Newton-Raphson algorithm is defined as 0(t+1) = 0 (t) - H(O(t))-1(0() which
converges to the maximum value, estimating 3. The Hessian of the solution to the parameter
estimation can then be used to calculate robust standard error estimates. Additionally, like
CVP estimation, it does not require a large number of votes for reliable estimates.2 1 The
substantive results are shown in Figures A67, A68, and A69 below for extreme, equal and
no income breaks respectively.
"For the purposes of comparing the two methods, the same votes were used.
151
Figure A67
Republican Senator's change in likelihood of voting for issue when
income group moves from 25th to 75th opinion percentile (in favor)
*Low Income
*Middle Income
0 High broomse
&
I
I
Minimumn
Wage
I
Abortion
R tona
social
Security
U-
*
r
Estate
Tax
Gun
Control
Gay
Marage
School
Voucher
1I
Environment
-1El1' "I
Free
Ttade
Iraq
war
Federal
Heattir
krmgraton
Insurance
Since GEEs are just one form of a marginal model, the most standard approaches for
interpreting these models are also useful for understanding the substantive meaning of GEE
estimates. Figures A67 and A68 show the changes in predicted values for each income group
and issue area. A separate GEE model was run for each issue area.
With high income classified as the top ten percent, the only positive and statistically significant results were for the low income opinion on banning gay marriage and middle income
opinion on promoting free trade. The middle income opinion on restricting abortion, the
middle income opinion on banning gay marriage, and the low and high income opinion on
promoting free trade were all negative and statistically significant, though likely not sub152
stantively meaningful.
Though not statistically significant, the coefficients for low and middle income opinion for
not being in favor of raising the minimum wage were positive and negative respectively.
Substantively, this would mean a Republican senator whose low income constituents moved
from the 25th to the 75th percentile (toward not favoring an increase in the minimum wage),
a shift equivalent to moving from PA to KY, holding the other constituents at the median
has a predicted 1.8 percentage point increase in his likelihood of voting against raising the
minimum wage. Similarly, a Republican senator whose middle income constituents moved
from the 25th to the 75th percentile, a shift equivalent to moving from WV to OK, holding
the other constituents at the median has a predicted 0.56 percentage point decrease in his
likelihood of voting against raising the minimum wage.
Below in Table A4 is an assessment of the goodness of fit of the GEE model for each issue
area. Receiver operating characteristics are used to visualize the performance of a binary
classifier by plotting the true positive rate against the false positive rate. If the area under
the ROC curve is equal to one half then it is performing no better than random chance. It
is clear that the GEE models are performing poorly.
Table A4: GEE Models, Area under the ROC curve
AUC
AUC
AUC
Minimum Wage
Abortion
Environment
Estate Tax
0.647
0.713
0.572
0.587
Gun Control
Gay Marriage
Health Insurance
Immigration
0.587
0.701
0.525
0.661
Iraq War
Social Security
School Voucher
Free Trade
0.526
0.531
0.561
0.664
153
Figure A68
Republican Senator's change in likelihood of voting for issue when
Income group moves from 25th to 75th opinion percentile (in favor)
(Equal Breaks)
*Lowe Inome
*Middle Income
I~~I
Minimumn
Wage
Abortion
ihts
Social
Seourlty
Estate
Tax
.IncoIi
g
Gun
Control
Gay
Marriage
School
Voucher
Environment
Hi
Free
ide
Iraq
War
rimnigratlon
e
Fedal
Health
isurance
With more equal income breaks, there are no positive and statistically significant results
are for any issue area. For gay marriage, high income opinion was found to be negative and
significant, but likely not substantively meaningful. This demonstrates how the results are
sensitive to income breaks.
For these income breaks as well, the area under the ROC curve shows that the model fit is
poor for many of the issue areas.
154
Table A5: GEE Models (Equal Breaks), Area under the ROC curve
AUC
Minimum Wage
0.645
Gun Control
AUC
0.588
Iraq War
AUC
0.528
Abortion
0.713
Gay Marriage
0.701
Social Security
0.532
Environment
0.572
Estate Tax
Health Insurance
Immigration
0.588
0.525
0.658
School Voucher
Free Trade
0.562
0.655
For the GEE models where constituent opinion was pooled across income groups (see
Figure A69), a statistically significant relationship between senators' voting records and the
opinions of their constituents was found only for the issue areas of estate tax, gay marriage,
and abortion.
However, only for estate tax was this relationship a positive one, where
senators were voting in the direction of public opinion. Therefore, the lack of a relationship
with income group opinion is understandable.
155
Figure A69
Republican Senator's change in likelihood of voting for issue when
constituents move from 25th to 75th opinion percentile (in favor)
(No Income Breaks)
.
H
I
I
I
I
H
0
Miru
Wage
Abortion
Rights
social
Seturity
Estate
Tax
Gun
Ctotrol
Gay
Marriage
School
Voruchrer
Erronmert
Free
Trade
Iraq
War
Federal
Heath
ImrmWation
Inlsurance
Table A6: GEE Models (No Breaks), Area under the ROC curve
AUC
AUC
AUC
Minimum Wage
0.645
Gun Control
0.587
Iraq War
0.526
Abortion
0.712
Gay Marriage
0.700
Social Security
0.529
Environment
0.565
Health Insurance
0.525
School Voucher
0.563
Estate Tax
0.586
Immigration
0.645
Free Trade
0.650
The area under the ROC curve shows that the model fit is poor for many of the issue
areas when constituent opinion is estimated in aggregate.
156
Notably, despite the different modeling assumptions made by the CVP Ideal Point estimation and the GEE models, both performed similarly. Without any income breaks, when
constituent opinion was a single independent variable, the CVP and GEE model estimates
were all in the same direction with the exception of the issue area of the environment, however here the estimate was close to zero. The CVP estimates were all in the same direction as
the GEE estimates for the equal income breaks with the exception of low income opinion on
gun control, but this estimate was also close to zero. When the income breaks were defined
in terms of the extremely wealthy, the estimates were in the same direction across all three
income groups for seven out of twelve issue areas. For both models of estimation it was
clear that some of the results were very sensitive to how income group breaks were defined.
It is unlikely that negative effects are substantively meaningful without a compelling story
for why senators would choose to purposefully act against their constituents' opinions. This
suggests that income group opinion, and constituent opinion more generally is not especially
predictive of senators' votes.
157
References
Achen, C. H. 1975. Mass Political Attitudes and the Survey Response. American Political
Science Review 69, 1218-1231.
1978. Measuring representation. American Journal of PoliticalScience 22, 475-510.
Ansolabehere, S., J. M. Snyder, and C. Stewart, III. 2001. Candidate Positioning in US
House Elections. American Journal of Political Science 45 (1), 136-59.
Bartels, L. 2008. Unequal Democracy: The PoliticalEconomy of the New Gilded Age. New
York: Russell Sage Foundation.
Brace, P., Sims-Butler, K., Arceneaux, K., and M. Johnson. 2002. Public Opinion in
the American States: New Perspectives Using National Survey Data. American Journal
of Political Science 46 (1), 173-89.
Bombardini and Trebbi. Working Paper. Votes or Money? Theory and Evidence from
the US Congress.
Brady, H. E., Verba, S., and K. L. Schlozman. 1995. Beyond SES: A Resource Model
of Political Participation. American Political Science Review 89, 271-294.
Brunner, Ross, and Washington. 2011. Does Less Income Mean Less Representation?
Campbell, A. L. 2005. How Policies Make Citizens: Senior Citizen Activism and the
American Welfare State. Chapters 1-5.
Canes-Wrone, B., Brady, D. W., and J. F. Cogan. 2002. Out of step, out of office: Electoral
158
accountability and House members' voting. American Political Science Review 96, 127140.
Carnes, N. 2011. Class and Representation: Legislators' Social Backgrounds and Economic
Policy Choices. Ph.D. dissertation, Princeton University.
Chattopadhyay, R. and E. Duflo. 2004. Women as Policy Makers: Evidence From a
Randomized Policy Experiment in India. Econometrica 72(5), 1409-1443.
Clinton, J. 2006. Representation in Congress: Constituents and Roll Calls in the 106th
House. Journal of Politics 68 (2), 397-409.
Converse, P. E. 1964. "The Nature of Belief Systems in the Mass Publics," in David Apter
(ed.) Ideology and Discontent. New York: Free Press.
Dahl, R. 1971. Polyarchy: Participationand Opposition. New Haven: Yale University Press.
Dawson, Michael. 1994. Behind the Mule: Race and Class in African American Politics.
Princeton: Princeton University Press. Chapters 3, 5-8.
Downs, A. 1957. An Economic Theory of Democracy. New York: Harper and Row.
Chapters 11-13.
Druckman, J. N. and K. R. Nelson. 2003. Framing and Deliberation: How Citizens'
Conversations Limit Elite Influence. American Journal of Political Science.
Erikson, R. S. and Y. Bhatti. Forthcoming. How Poorly are the Poor Represented in
the US Senate?
159
Erikson, R. S., MacKuen, M. B., and J. A. Stimson. 2002. The Macro Polity. New York:
Cambridge University Press.
Erikson, R. S., and L. Stoker. 2011. Caught in the draft: The effects of Vietnam draft
lottery status on political attitudes. American PoliticalScience Review, 105 (2), 221-237.
Fenno, R. F. 1977. U.S. House members and their constituencies: An exploration. American
Political Science Review 71, 883-917.
Fiorina, Morris. 1981. Retrospective Voting in American National Elections. New Haven:
Yale University Press. Chapters 1 and 5.
Fowler, Anthony, and Andrew Hall. Working Paper. Conservative Vote Probabilities: An
Easier Method for the Analysis of Roll Call Data.
Gilens, M. 2005. Inequality and Democratic Responsiveness. Public Opinion Quarterly
69, 5.
-
2012. Affluence and Influence: Economic Inequality and Political Power in America.
New York: Russell Sage Foundation.
Glynn, C. J., Herbst, S., O'Keefe, G. J., Shapiro, R. Y., and L. Jacobs. 1999. "Public
Opinion and Policy Making," in Carroll J. Glynn, Susan Herbst, and Garrett J. O'Keefe
(ed.) Public Opinion. Boulder, CO: Westview Press.
Hacker J. S. and P. Pierson. 2010. Winner-Take-All Politics: Public Policy, Political
Organization, and the Precipitous Rise of Top Incomes in the United States. Politics
160
and Society 38 (2), 152-204.
Hainmueller, J. and M. J. Hiscox. 2010. Attitudes Toward Highly Skilled and Low-Skill
Immigration: Evidence From a Survey Experiment. American Political Science Review
104(1).
Hess, S. 1966. America's Political Dynasties. Garden City, NY: Doubleday & Company,
Inc.
Huckfeldt, Robert, and John Sprague. 1987. Networks in Context. American Political
Science Review 81(4), 1197-1216.
Key, V. 0. 1961. Public Opinion and American Democracy. Knopf.
Kinder, D. R. 2003 "Belief Systems after Converse." in Michael MacKuen and George
Rabinowitz, ed, Electoral Democracy. Ann Arbor: University of Michigan Press.
Lax, J. and J. Phillips. 2009. How Should We Estimate Public Opinion in the States?
American Journal of Political Science 53(1), 197-121.
Lupia, A. 1994. Shortcuts Versus Encyclopedias: Information and Voting Behavior in
California Insurance Reform Elections. American Political Science Review 88, 63-76.
Matland, R. E. 1993. Institutional Variables Affecting Female Representation in National
Legislatures: The Case of Norway. The Journal of Politics 55 (3), 737-755.
Miller, W. E., Kinder, D. R., and S. J. Rosenstone, and the National Election Studies.
NATIONAL ELECTION STUDIES, 1988-1992 Merged Senate File [dataset]. Ann Arbor,
161
MI: University of Michigan, Center for Political Studies [producer and distributor], 1999.
Miller, W. E., and D. E. Stokes. 1963. Constituency Influence in Congress. American
Political Science Review 57(1), 45-56.
Monroe, A. D. 1979. Consistency between Public Preferences and National Policy
Decisions. American Politics Quarterly 7, 3-19.
Page, B. I., Bartels, L. M., and J. Seawright. 2013. Democracy and the Policy
Preferences of Wealthy Americans. Perspectives in Politics 11(1).
Page, B. I. and R.Y. Shapiro. 1983. Effects of Public Opinion on Policy. American
PoliticalScience Review 77, 175-90.
-
1992. The Rational Public: Fifty Years of Trends in American Policy Preferences.
Chicago: University of Chicago Press. Chapters 1, 2, 8 and 9.
Park, D., Gelman, A., and J. Bafumi. 2004. Bayesian Multilevel Estimation with
Poststratification: State-Level Estaimates from National Polls. PoliticalAnalysis. 12(4),
375-385.
Posner, D. 2005. "Ethnicity and Ethnic Politics in Zambia," in Institutions and Ethnic
Politics in Africa. New York: Cambridge University Press, pp. 91-129.
Posner, D. and E. Kramon. Working Paper. Ethnic Favoritism in Primary Education in
Kenya.
Olson, M. 1965. The Logic of Collective Action. Cambridge: Harvard University Press.
162
Romer, T. and J. M. Snyder, Jr. 1994. An Empirical Investigation of the Dynamics of
PAC Contributions. American Journal of PoliticalScience 38(3), 745-769.
Sears, D. 0., R. R. Lau, T. Tyler, and A. M. Allen Jr. 1980. Self-Interest versus
Symbolic Politics in Policy Attitudes and Presidential Voting. American PoliticalScience
Review 74, 670-684.
Schattschneider. 1975. The Semi-Sovereign People: A Realist's View of Democracy in
America. Hinsdale, Illinois: The Dryden Press.
Stimson, J. A. 2004. Tides of consent: How Public Opinion Shapes American Politics.
Cambridge: Cambridge University Press, pp. 1-82.
Stimson, J. A., M. B. MacKuen, and R. S. Erikson. 1995. Dynamic Representation.
American Political Science Review 89(3), 543-65.
Tausanovitch. Working Paper. Income and Representation in The United States Congress.
Tausanovitch, Christopher and Christopher Warshaw, 2013. Measuring Constituent
Policy Preferences in Congress, State Legislatures, and Cities. The Journal of Politics 75
(2), 330-342.
United States Senate. "Roll Call Tables." http://www.senate.gov/pagelayout/legislative
/a-three-sections-with-teasers/votes.htm.
Accessed October 25, 2012.
Verba, Sidney, Kay L. Schlozman, and Henry E. Brady. 1995. Voice and Equality: Civic
Voluntarism in American Politics. Cambridge: Harvard University Press. Chapters 16
163
-17.
Washington, E. 2006. Female Socialization: How Daughters Affect Their Legislator Fathers'
Voting on Women's Issues. NBER Working Paper 11924.
Warshaw, Christopher and Jonathan Rodden. 2012. How Should We Measure DistrictLevel Pubic Opinion on Individual Issues? The Journal of Politics 74(1), 203-219.
Zaller, J. 1992. The Nature and Origins of Mass Opinion. Cambridge: Cambridge
University Press. Chapters 1-5.
Zorn, C. J. W. 2001. Generalized Estimating Equation Models for Correlated Data: A
Review with Applications. American Journal of PoliticalScience 45(2), 470-490.
164