Congruence, Responsiveness, and Representation in
American State Legislatures*
Boris Shor†
May 14, 2014
Abstract
Two problems hinder the ability of scholars to assess the quality of representation of state-level
public opinion by elected representatives. First, the main tool measuring ideology in public
opinion has historically been self-reported, but this is now well known to be severely plagued
by measurement error. Second, and far more binding, we lack a common scale on which to
place both constituents and representatives. While the literature has addressed a number of
methods estimating a common space for politicians’ ideal points across political institutions,
little work exists that incorporates citizens into this space. e uni ed methodology in this paper solves both problems in order to assess representation of constituents by their individual
state legislators, the parties in the state legislatures, and the state legislature as a whole. Bridging is accomplished using policy preference questions from a state and congressional candidate
survey administered since the early 1990s. I ask those questions in my own 2008 survey of
over 4,200 citizen respondents, representative at the state and national levels. us, citizens
and state legislators can be located on the same ideological scale. I employ multilevel regression with poststrati cation to model state-level public ideology and obtain aggregate opinion
estimates for all 50 states and 1942 upper chamber state legislative districts. State legislators
and chamber and party medians are responsive to public opinion, but they are very oen incongruent to it. Democrats and Republicans diverge from district and state opinion, but in an
asymmetric fashion, with Republicans considerably more distant.
EARLY DRAFT: PLEASE DO NOT CITE WITHOUT PERMISSION
*
Paper prepared for the 2014 Annual Meeting of the Midwest Political Science Association, Chicago, IL. is paper
emerges from earlier work I did with Nolan McCarty. e data in this paper relies on the hard work of the following: Jon
Rogowski, Chad Levinson, Mateusz Tomkowiak, Michelle Anderson, Stuart Jordan, Johanna Chan, Sarah McLaughlin
Emmans, and Lindsey Wilhelm. anks to Andrew Gelman, David Park, Gerald Wright, Will Howell, Stephen Jessee,
Michael Bailey, Jeffrey Lax, Justin Phillips and seminar participants in Chicago, Princeton, UNC, and Stanford for
comments on earlier ideas. anks to Project Vote Smart for making their NPAT data and questions available to me. I
welcome comments and questions. Any errors are my own.
†
Harris School of Public Policy Studies, University of Chicago; [email protected]
1
1 Introduction
Normative democratic theory establishes the benchmark of representativeness for democratic
politics. We can disaggregate this into two standards: responsiveness and congruence. e rst denotes the idea that legislators, either by selection or adaptation, respond to their constituents’ policy preferences. us, liberal publics should be represented by liberal representatives (and similarly
for conservatives). Yet responsiveness is a relatively weak standard; conservative representatives
of liberal public opinion could be perfectly responsive, as long as they responded on the margins
to changes in opinion. Congruence is a far higher standard. In cardinal terms, the preferences of
constituents and the representative should match in some common metric.
Assessing congruence in the representational relationship is beset by two problems. First, how
should we ascertain the unobservable set of policy preferences that individuals possess–and are
these views consistent or constrained? Second, how can we measure those preferences on a common
scale with that of their representatives at multiple levels of government? Both questions has long
bedeviled attempts to assess congruence at the state level (Lax and Phillips 2009a; Matsusaka 2010),
so analysts have typically had to fall back on responsiveness as a benchmark (Erikson, Wright and
McIver 1993).
At the individual level, the main measurement tool on ideology writ large has long been some
variant of a self-reported ideology measure. at is, people are asked in a survey to rate themselves
from strongly liberal to strongly conservative on some scale. Of course, this measure is notoriously
noisy and crude. e terms “liberal” or “conservative” do not refer to some xed external scale, but
rather are dependent on context, both in time and place, ltered through contemporaneous events
in particular places. Conservative means one thing in Texas–and something entirely different in
Massachusetts. It may mean one thing in the 1980s under Ronald Reagan, and something else
entirely in the 2000s under George W. Bush. Today, the pejorative connotations of “liberal” affect
even people who otherwise hold liberal policy preferences. In sum, these labels mean different
things to different people, which is problematic if we hope to get at some latent, comparable trait.
us, either our survey instruments are too weak to identify underlying ideology, or perhaps there
is no coherent, consistent ideology in most people to measure in the rst place (Converse 1964).
Achen (1975) and Ansolabehere, Rodden and Snyder (2008) conceptualize the problem of determining individual ideology from the perspective of measurement error. is perspective suggests alternatives to abstract single-item questions about ideology such as multi-item issue preference scores which evince far greater individual stability than self-reported ideology. ey also
show that these scales approach the stability of party identi cation in applications like vote choice.
While undoubtedly a major advance in measurement, the choice, weightings, and groupings of the
2
components of these indices seem arbitrary. Beyond received wisdom, for example, what suggests
breaking up ideology into moral and economic groupings, but not foreign policy?
For politicians, the estimation of spatial models of roll call voting and their interpretation as
measures of unobserved ideology has been one of the most important developments in the study
of Congress and other legislative and judicial institutions. e seminal contributions of Poole and
Rosenthal (1991, 1997) launched an extensive literature that is marked by sustained methodological
innovation, new estimators (Martin and Quinn 2002; Clinton, Jackman and Rivers 2004; Jackman
2004; Poole 2000) e scope of application has expanded greatly from the original work on the
U.S. Congress. Spatial mappings and ideal points have now been estimated for the Supreme Court
(e.g. Martin and Quinn 2002; Bailey and Chang 2001; Bailey, Kamoie and Maltzman 2005), U.S.
presidents (McCarty and Poole 1995), a large number of non-U.S. legislatures (Londregan 2000b;
Morgenstern 2004), the European Parliament (Hix, Noury and Roland 2007), and the U.N. General
Assembly (Voeten 2000).
e most painful limitation of any set of ideal point estimates for politicians, or of conventional
ideological measures of individuals, is that they cannot be compared to each other directly on the
same scale. While responsiveness in representation could be assessed with measures like presidential vote or self-identi cation (Clinton 2006), the same is not true of congruence.
is paper proposes a uni ed solution to this common metric problem. e scale derives from a
hitherto-underemployed voter information questionnaire developed by a private organization and
administered to politicians across all state legislatures and Congress for over a decade, beginning
in the mid 1990s. Politicians are jointly scaled with constituents through the use of a new survey
sampling 4,200 Americans. Such a strategy allows, for the rst time, measuring ideological distributions of state legislative chamber and party medians, individual state legislators, and citizens all
on the same scale. ese scores can then be applied to address theoretical debates in social science
and substantive public policy problems.
At the macro level, I assess the degree to which state legislative chamber and party medians
are congruent with state-level public ideology. I nd signi cant heterogeneity. While state-level
public opinion varies across states, it is in the main rather moderate. In contrast, in most states,
state legislatures are considerably more extreme in their ideological leanings.
See Poole (2005) for an introduction and survey of ideal point estimation.
3
2 Methodology
2.1
Common Space Ideal Points
To grossly simplify, statistical identi cation of ideal points comes from data on how oen legislators vote with other legislators on a common set of roll calls. We identify a legislator as a conservative because he is observed voting with other conservatives more frequently than he is observed
voting with moderates, which he does more oen than he votes with liberals. But when we want to
make comparisons across political institutions, we are at a loss to make such comparisons. Being
a conservative in the Massachusetts Senate is quite different from being a conservative in the US
House.
e need for comparable preference estimates across political institutions is hardly new. Efforts
to place multiple institutions in a common space rely, in varying ways, on bridge actors. ere have
been efforts to produce common ideological scales for the US House and Senate (Poole 1998; Groseclose, Levitt and Snyder 1999), for presidents and Congress (McCarty and Poole 1995), for presidents, senators, and Supreme Court justices (Bailey and Chang 2001; Bailey, Kamoie and Maltzman
2005), for Supreme Court and Court of Appeals justices (Epstein et al. 2007), and state legislators
and Congress (Shor, Berry and McCarty 2010; Shor and McCarty 2011). Similar issues arise in the
estimation of dynamic models (Poole and Rosenthal 1997; Martin and Quinn 2002). Identi cation
of the models relies on the existence of bridge actors who make choices that can be construed as votes
in multiple settings. For example, Bailey and Chang (2001) compares Congress and the Supreme
Court by leveraging the fact that legislators oen opine on the cases that the justices have voted on.
Common scales are identi ed by the analyst’s assumptions about the consistency of behavior when
a bridge actor moves from one setting to another.
Yet the limitations of using bridge legislators is obvious; they can not help us to construct a
common ideological space between politicians and constituents. Instead, I adopt the framework
of an earlier paper (Shor and McCarty 2011), where I use bridge questions from the NPAT survey
of Project Vote Smart rather than individuals. I exploit this survey of state and federal legislative
candidates–described in more detail below–that has been continually administered in largely identical form every election cycle since 1996. I conduct an online survey that asked questions from the
legislative survey to 4,200 respondents. Treating citizens as if they were legislators, I use standard
ideal point estimation techniques to derive spatial location parameters for congressional and state
legislative incumbents, as well as my survey respondents.
Similarly, Gerber and Lewis (2004) use the initiative in California, limiting its application to that state.
Ideal point estimates are from Bayesian item-response theory models (Jackman 2000; Martin and Quinn 2002;
Clinton, Jackman and Rivers 2004; Jackman 2004), but Poole-Rosenthal NOMINATE scores work just as well.
4
For incumbents in state legislatures, I transform their roll call based ideal point measures into
NPAT common space measures. Aer estimating roll call ideal points for legislators in each state, I
project them into the space of NPAT ideal points using OLS. e tted values of these regressions
generate predicted NPAT scores for the legislative non-respondents. e crucial assumption is that
legislators who are NPAT respondents act consistently whenever they are making choices about
policy, whether it be during a roll call or answering an election survey. If they do, at least on average,
the resulting estimates will be comparable across states, chambers and time. It is not necessary for
legislator NPAT respondents themselves to be representative of the universe of legislators (Shor and
McCarty 2011).
2.2
Modeling Aggregate Opinion
To evaluate how well individual state legislators, state legislative parties, and state legislatures as
a whole represent their constituents, we need measures of state- and district-level public opinion.
is is difficult to accomplish, as sample sizes on typical national surveys dwindle rapidly for all but
the largest states. A common response to this problem is disaggregation at the state (Erikson, Wright
and McIver 1993) and now district levels (Warshaw and Rodden 2012; Tausanovitch and Warshaw
2013). Here, multiple national polls are pooled so as to generate sufficiently large sample sizes by
state or district.⁴ e principal difficulty of such a method is the sheer amount of data required,
which oen necessitate pooling polls over a decade or more. is approach radically limits the
number of applications for aggregating opinion to important political constituencies (for example,
Bafumi and Herron (2010) only display common space results for a subset of states for this reason);
even the large survey data presented here with 4,200 respondents only has 24 states with 50 or more.
e data requirements are drastically more onerous when attempting to estimate opinion at the
congressional or state legislative district level.⁵
Instead, I turn to the newly developed technique of modeling aggregate opinion with multilevel (or hierarchical) regression combined with poststrati cation (MRP) (Park, Gelman and Bafumi 2004). Here, opinion is modeled with both demographic information and non-demographic
geographic effects. e multilevel structure of the model improves the estimation of these effects
through partial pooling, which is a more reasonable alternative to complete pooling (ignoring geography entirely) or no pooling (ignoring out-constituency variation, eg xed effects) (Gelman and
Hill 2006). Mechanically, aer the multilevel model is estimation, predictions are made for each
demographic-geographic respondent cell. en, these predictions are poststrati ed by detailed
⁴Disaggregation, in turn, was a response to weaknesses of a prior method of simulating state-level public opinion
using only the demographic differences between states.
⁵Tausanovitch and Warshaw (2013) pool surveys containing over 350,000 respondents for a decade to estimate district level opinion.
5
Census population data about the sizes of these cells in the states. Lax and Phillips (2009b) use
simulations and validation via real election returns to conclude that MRP offers superior recovery
of state-level opinion to disaggregation, at fractions of the required sample sizes. Warshaw and Rodden (2012) conduct similar exercises for state Senate districts and nd that MRP efficiently estimates
opinion despite the extreme sparseness of the data.
Typical uses of MRP in applied research, such as Lax and Phillips (2009a) and Kastellec, Lax
and Phillips (2010), involve estimation of state-level public opinion from dichotomous preference
questions. For example, state level support for particular Supreme Court nominations is modeled
from a binary response for support or opposition by respondents. I adopt the MRP approach, but
adapt it for individual-level ideal points (themselves estimated from preference questions). is use
of continuous measures as an individual response is novel, but poses no special difficulties for MRP.
e approach has been used elsewhere, including Kousser, Phillips and Shor (2014).
To be speci c, I model each individual i’s ideal point as a function of their demographics (indexed j, k, m and n for gender, race, age and education), and either their state (s) or district (d) of
residence, using a varying intercept approach.
state|district
gender
age
race
edu
N P ATi = β 0 + αj[i]
+ αk[i]
+ αm[i]
+ αn[i]
+ αs|d[i]
(1)
e demographic and state intercepts are themselved modeled as drawn from a normal distribution
with mean zero and estimated variance, as follows:
2
αjgender ∼ N (0, σgender
), for j = 0, 1
(2)
2
αkrace ∼ N (0, σrace
), for k = 0, ..., 4
age
2
αm
∼ N (0, σage
), for m = 0, ..., 3
2
αnedu ∼ N (0, σedu
), for n = 0, ..., 4
District effects are modeled as a function of their region, their state, and the 2008 presidential election returns by district:
region
state
2
αddistrict ∼ N (αs[d]
+ αr[d]
+ β prez · prezd , σdistrict
), for d = 1, ..., 1942
(3)
State effects are modeled as a function of their region, and the 2008 presidential election returns by
state:
region
2
), for s = 1, ..., 50
αsstate ∼ N (αr[s]
+ β prez · prezs , σstate
(4)
With region a varying intercept term similar to the demographic terms:
2
αrregion ∼ N (0, σregion
), for j = 1, ..., 4
6
(5)
Poststrati cation proceeds aer predictions are made for each demographic-geographic type, which
number 10,000 (50 states with 200 demographic categories) for states and 388,400 for state senate
districts (1,942 districts with 200 demographic categories). Populations of each of the types are
obtained from the Census’ 5% Public Use Microdata Sample. Predictions are weighted by the population, and summed over each state or district, to derive the mean public ideology by state or
district.
3 Data
3.1
Legislators: NPAT Survey and Roll Call Data
To scale members of Congress, state legislators, and individuals on a common ideological space,
I begin with 17 years (1996-2013) of the National Political Awareness Test (NPAT) survey administered by Project Vote Smart, a nonpartisan organization that disseminates information on campaigning politicians to the public at large. Since the survey is answered before November of an election year, I drop all observations for all subsequent losers to get at the population of all incumbents.
e questions asked by NPAT cover a wide range of policy matters, including on national security,
social issues, scal policy, environmentalism, criminal justice, and many more. e questions on
the NPAT do change somewhat over time. But while some topics like stem cell funding come and
go, the survey asks many other questions consistently and repeatedly. Most useful, a large proportion of the questions asked of state and congressional candidates are identical across the country.
is large set of common questions provides signi cant leverage for making cross-state, cross-level,
and longitudinal comparisons.
Despite the richness of the data, use of the NPAT surveys has been limited. Ansolabehere, Snyder
and Stewart (2001) use the 1996 and 1998 surveys to distinguish between the in uence of party and
preferences on roll call voting. is is because of the less than universal number of incumbents who
answered the survey. e alternative used here is the exploitation of roll call voting data, which has
been the staple of the empirical spatial voting literature. Congressional roll call data is from Keith
Poole’s Voteview web site. For state legislatures, I use a new comparative data set of state legislative
roll call votes that covers over 21,000 legislators from all 50 states from over a decade from Shor and
McCarty (2011).
3.2
Individuals: CCAP Survey
I administered a large subset of the NPAT survey to individuals via several waves in the 2008 Cooperative Campaign Analysis Project (CCAP), a joint venture of 27 research teams. e multiwave
panel was administered online by YouGov/Polimetrix with roughly 20,000 people (representative
at the state level) completing all six waves. Details on the construction of the sample and validation
7
exercises can be found in Jackman and Vavreck (2010). Pooled together, a common set of questions
are asked of all the respondents, while a subset of representative respondents answered individual
team questions.
In two long-form CCAP survey waves conducted in March and September 2008, I asked 76
dichotomous questions taken verbatim from the 2008 Congressional and state legislative NPAT
questionnaires. I matched these question with NPAT questionaires from 2006 and earlier. In the
short-form CCAP survey waves conducted in September 2008, I asked a subset of 15 questions
asked in the long-form survey. e consequences of the smaller number of bridge actors is one of
efficiency, not bias.⁶
4 Descriptives
4.1
Fit and Dimensionality
For each state roll call record, I estimated one and two dimension spatial models. Following
Poole and Rosenthal (1991, 1997), I assess the models based on the overall classi cation success
as well as the aggregate proportionate reduction in error (APRE).⁷ Despite considerable cross-state
variation, average classi cation and APRE scores for state legislatures are quite comparable with
that of Congress. Moreover, the improvement in t afforded by a second dimension is minimal.
us, policy preference appear adequately described in one dimension in the states, just as it is in
Congress.
Estimation statistics from the CCAP survey sample, in comparison with legislative institutions,
show that the ts are poorer, though not much more so. us, the rst dimension classi es some
73% of NPAT answers correctly, with a 28% APRE.⁸ is compares with the poorest tting states,
with Nebraska at 79% correct classi cation and Arkansas at 26% APRE. e second point is that the
improvement in classi cation for using two dimensions is more substantial relative to that of most
states and Congress, but only a touch larger than the Illinois legislature, which gives the largest improvement in adding a second dimensions. In sum, we can safely say that one ideological dimension
dominates the CCAP survey responses to NPAT questions.
is nding is remarkable on a number of levels. It undermines Converse (1964) and his classic argument that the mass public is ideologically unconstrained–that belief in general ideological
⁶e standard deviation of the estimate ideal points is higher for the short-form waves, but the mean is not different, as expected. Note that similar studies (Bafumi and Herron 2010; Jessee 2009, 2010) use only 10-15 questions for
bridging.
⁷e APRE measures the improvement in classi cation relative to a null model where all votes are cast for the
winning side. is is a more realistic benchmark than classi cation success, where even the naive model can do well
on.
⁸Note that these statistics are from a CCAP-only estimation; they are not contaminated by including legislators.
8
principles do not constrain more speci c policy preferences. e necessary inference from such an
argument is that policy preferences for the vast majority should not be correlated in a consistent way.
Individuals hold ideological opinions as a grab-bag, plucked almost randomly. ey (as opposed
to elites) can not be classi ed on the le-right ideological continuum because of their incoherent
beliefs. If true, this should manifest itself in very poor t statistics from an ideal point estimation,
and/or highly multidimensional preferences, as beliefs in one area do not correlate with those in
another.
It has been thought that the ubiquity of one dimension in legislative politics had to do with
institutional or party factors. For example, there may be very high returns to party discipline in
establishing party brand names, which may necessitates simple and consistent voting behavior to
more easily be identi ed by the public at large. But the unidimensionality of public opinion points
to causes that transcend features of legislative politics, such as the politico-cultural or philosophical
fusing of intellectual traditions into the public discourse, or the psychological need for some sort of
consistency in preferences. It is not clear why environmentalism necessarily hangs together with a
desire for more union prerogatives, but it does – for people as well as politicians.
4.2
Validation
Before proceeding, it is instructive to compare the estimated common space ideal points for individuals with other measured political variables. e correlation of the ideal points with the standard seven-item party identi cation question is 0.66. e correlation with the ve-item ideological
self-placement question is 0.72.⁹ e scores are both highly substantively and statisically signi cant
predictors of responses to issue preference questions, including those on abortion, prison, and guns,
even controlling for party. ey are also a very powerful predictor of presidential vote choice, also
controlling for party identi cation.
At the aggregate level, I begin with the comparison to a measure of district ideology developed
by Tausanovitch and Warshaw (2013). ey measure the ideology of all states and state legislative
districts prior to the 2012 redistricting, pooling responses from a variety of surveys over a decade.
In their analysis, the number of respondents per state averaged around 7,000 and the state senate
districts around 180, orders of magnitude more than this application. Yet Figure 1 shows that the
correlations between their measure and ours is quite impressive; 0.92 at the state level and 0.63 at
the district level. ⁰ is is a good sign that the survey and methodology have produced an accurate
estimate of constituent preferences at both of these levels.
⁹e correlation is 0.77 for the simple three-question ideological index suggested by Ansolabehere, Rodden and
Snyder (2008).
⁰By comparison, the raw data without MRP correlate at 0.54 and 0.32, respectively.
9
State Ideology and Chris^2 2008
ID
OK
MS
LA
0.2
MT
TX
GA
MO
AZ
Chris^2 State Mean
SC
NV
IA
CO PA
MI
NH
0.0
−0.2
MD
HI
RI
WIVA
AL
NE
UT
WY
NDKY
IN
WVKS
AK
OH
NC
FL
MN
OR NM
ME
IL
DE
WA
CA
TN AR
SD
NJ
CT
NY
MA
−0.4
VT
−0.2
0.0
0.2
Estimate State MRP Mean
(a) State
MRP Means vs Raw Chrises District Opinion
Raw Chris^2 District Opinion
1
0
−1
−1.0
−0.5
0.0
0.5
Estimate 2008 District Score
(b) District
Figure 1: State and District level validation. District MRP estimates are highly correlated with state and district
level ideology from Tausanovitch and Warshaw (2013).
10
5 District Level Representation
e payoff of the joint scaling exercise is that we can now compare legislator and district opinion
directly. Figure 2 plots the two against each other. Legislator opinion, on the whole, is responsive
to district opinion. Separating districts by the party of the legislator shows that Democrats are
more responsive than Republicans. Yet this result mirrors earlier ndings using the proxy of district
presidential vote for district opinion (Shor and McCarty 2011).
e new nding shown in Figure 2 is the common space distance of legislators from their districts. e grey line in the plot is the 45 degree line of perfect congruence. Democrats are typically
(but not always) more liberal than their districts, and Republicans are almost always more conservative. Yet Republicans are typically more distant from their districts, as highlighted in the density
curve of Figure 2.
What factors account for the variation in the distance between legislators and their districts? My
outcome variable is the absolute value of distance (since Republicans’ distance is typically positive
and Democrats’ negative). I regress a number of predictors on the absolute distance. e rst of
these is political party. ere have been a number of arguments (Hacker and Pierson 2005; McCarty, Poole and Rosenthal 2013; Mann and Ornstein 2013) that Republicans are becoming polarized much faster than Democrats. If district opinion moves slower than elite preferences, we should
expect Republicans at a time of extreme polarization to be driing farther from public opinion than
Democrats.
Groseclose (2001) argues that incumbents have strategic reasons to locate closer to the center,
since this would amplify their valence advantages over their challengers. Alternatively, incumbents
could trade policy distance for their valence advantages, allowing them to locate farther away than
their challengers. I include an interaction between party and incumbency to account for heterogeneity in the incumbency effect by party.
For re-election minded politicians, close races should attune their incentives closer to that of
their constituents (Mayhew 1974). I measure electoral tightness by the vote margin of the state
legislator over the runner up in the last election prior to 2008. A higher margin should induce
moderation by legislators towards the distract mean.
Rodden et al. (2014) argue that heterogeneous districts themselves lead to polarized, and hence
extreme legislators. e simple intuition is based on the idea that candidates must choose platforms
in the presence of uncertainty over the median voter (Calvert 1985; Wittman 1983). e greater the
uncertainty, the more candidates move towards their party’s more extreme ideological preferences.
e intuition is that when district opinion is unimodal, the median voter on election day will be
largely predictable, constraining candidates. In contrast, when voters are more evenly distributed
11
throughout the ideological spectrum or even polarized into a bimodal distribution, there is more
uncertainty about the identity of the median voter on election day, and hence weaker electoral constraints on candidates’ ideological positioning. In other words, when districts are moderate–but
only as a consequence of internal divisions–they tend to elect more extreme legislators. is is because normal Downsian pressures to converge are balanced against the returns to turning out your
own base. For example, this is seen in large legislative districts where red precincts in the outlying
areas surround deep blue areas like college towns. ey are moderate only because they are deeply
internally divided, with a balance between highly liberal and conservative voters, not because the
voters are moderate and unimodal. So the expectation is that more heterogeneity should induce
greater distance from distract means.
e results in Table 1 largely con rm these hypotheses. Republicans are considerably more distant than Democrats, though the effect is moderated by incumbency. Republican challengers who
win their state Senate races before 2008 are quite extreme. In fact, the incumbency effect operates
only in the Republican party; as Model 2 shows, there is little differences, ceteris paribus, between
Democratic incumbents and challengers. is seems quite consistent with a polarizing party environment where successive generations of Republicans are ever more polarized, while Democrats
are more stable. Legislators who win by tighter margins are indeed closer to their districts. And
more heterogenous districts lead to increasingly distant legislators.
12
Table 1: Distance Models
Dependent variable:
Absolute Distance
(1)
(2)
∗∗∗
0.141∗∗∗
(0.034)
Republican
0.059
(0.018)
Incumbent
−0.055∗∗∗
(0.021)
Vote Margin
0.064∗∗
(0.027)
0.058∗∗
(0.027)
District Opinion Polarization
0.681∗∗∗
(0.216)
0.684∗∗∗
(0.218)
−0.113∗∗∗
(0.039)
Republican * Incumbent
−0.988∗∗
(0.490)
Constant
Observations
Log likelihood
Akaike Inf. Crit.
Bayesian Inf. Crit.
−0.004
(0.027)
1, 088
−211.541
437.083
472.027
∗
Note:
13
−1.027∗∗
(0.494)
1, 088
−209.740
435.479
475.416
p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01
MRP Means vs Legislator Ideology
State Senator Ideal Point
1
0
−1
−0.50
−0.25
0.00
0.25
0.50
Estimate 2008 District Score
(a) Scatter
Congruence by Party
1.2
Density
0.9
0.6
0.3
0.0
−2.0
−1.5
−1.0
−0.5
0.0
0.5
1.0
Distance to Raw District Mean
(b) Density
Figure 2: Congruence at the district level.
14
1.5
2.0
5.1
State Level Representation
Figure 3 plot state opinion against aggregate state legislative ideology, measured as the average
of the chamber medians. Opinion is responsive but frequently incongruent.
State Opinion and Legislative Ideology 2008
OK
MO
GA
ND
SC
0.5
Averaged Chamber Medians
MS
OH
TXAZ
VA
FL
LA
PA
MI
NV
DE
0.0
WI
IL
OR
ME
UT ID
SD
AKKS
IN
AR
TN
MT
NC
WV
KY
NE
WY
AL
RI WA
VTNY
HIMD
−0.5
MA
CT
−1.0
IA
CO
NM
MN
NJ
NH
CA
−0.2
0.0
0.2
Estimate State MRP Mean
Figure 3: Congruence at the state level.
5.1.1 Chamber Medians
Figure 4 plot state opinion against legislative ideology, measured as chamber medians (Figure 5 recon gures the plot to more clearly communicate the extent of state-level incongruence
of opinion). Opinion is responsive but frequently incongruent. Republican-held chambers are
more conservative than state opinion, while Democratic-held chambers are mostly, but not always,
more liberal than their states (Mississippi is the most prominent exception). Here, too, the nding
from the district-level analysis holds up. Republican-led chambers appear to be more distant than
Democratic-led chambers.
15
Chambers and the Public, 2008
House
Senate
OK
SC
WI
Chamber Median
DE
NY
HI
MA MD
MT
NC
PA
IL
ME
RI WA
VT
TX
AZ
LA
DE
NM
WA IL OR CO
MD
VT
HI
MN
MA
NJ
NJ
KS
WY
MT
RI
NY
NM
ID
UT
NE
AL
ME
NV
NH IA
MI
IN
AK
SD
OK
TN
AR
WV
WV
IN
CO
MN
CT
MI
NV
SC
VA
PA
FL
AK
TN KY
AL
OR
0.0
KY
ND
OH
MS
ND UT
AZ
ID
MS
OH SD
VA
TX
WY
LA
KS
FL
AR
0.5
−0.5
MO
GA
GA
MO
NC
IA
WI
CT
CA
NH
−1.0
CA
−0.2
0.0
0.2
−0.2
0.0
0.2
State MRP Mean
(a) Scatter
Congruence by Party
H
S
2.0
Density
1.5
1.0
0.5
0.0
−2.0 −1.5 −1.0 −0.5
0.0
0.5
1.0
1.5
2.0
−2.0 −1.5 −1.0 −0.5
0.0
0.5
1.0
1.5
2.0
Distance to State Mean
(b) Density
Figure 4: Congruence at the chamber median level. Party assigned to who controls the chamber.
16
State
2008
MA
CA
VT
HI
NY
RI
MD
DE
WA
CT
IL
OR
ME
NM
MN
CO
NJ
PA
MI
NV
NH
SC
WI
IA
VA
FL
MO
GA
MS
TX
NC
OH
MT
LA
AZ
TN
SD
IN
WV
AK
KS
AR
ND
KY
AL
NE
UT
ID
OK
WY
party
D
Public
R
chamber
House
Public
Senate
−1.0
−0.5
0.0
0.5
Score
Figure 5: Representation of state opinion by chamber medians in common NPAT ideological space. Circles
indicate a MRP point estimate of mean ideology in the state. Chambers are colored according to party control as
of 2008.
17
5.2
Party Medians
Incomplete
Figure 7 continues the exercise, this time displaying state opinion plotted against party medians
within the two chambers. Parties are clearly responsive to public opinion, but with Democrats more
so than Republicans. And again, Republican medians are more conservative than are Democratic
medians liberal.
Figure 6 plots state opinion means and chamber party medians directly on the le, and on the
right reformats the plot to highlight the ideological distance between the two. Since some party
medians are more moderate than state opinion, the use of an absolute (or folded) distance measure
would be misleading for these cases. Instead of using the absolute value of distance, I have created an
alternative measure of distance, extremism, which multiplies the Democratic medians by negative
1. us, positive values for both Democrats and Republicans indicate increasing extremism relative
to the state, while (infrequent) negative values imply moderation relative to the state.
Several patterns become apparent. First, there is an extreme heterogeneity in the range of distance. States like California and Colorado have highly distant party medians, while at the other
extreme, Alaska and Rhode Island have very moderate party medians. Nevertheless, party medians
are still, on the whole, more extreme than state opinion. ere are outliers like Alaska, where 3 out
of 4 medians are more moderate than the state, and New Jersey, whose Republicans are to the le
of state public opinion. ird, party medians are correlated in extremism relative to state opinion.
Party medians tend to be distance simultaneously for both parties.
Now we investigate the variation in the congruence of party medians relative to state opinion
means. e outcome variable is the absolute distance between the two.
Republicans party medians are expected to be further from state opinion than Democrats.
Chambers that are more competitive should inspire moderation by parties who converge in
Downsian fashion. Competition is calculated as per Holbrook and Van Dunk (1993), but at the
chamber level.
States vary dramatically in population and the number of legislative districts. e ratio of the
two I call district magnitude. e idea is that a larger magnitude makes the monitoring problem
more difficult, and allows for ideological dri from state opinion.
Table 2 shows the results of the model.
18
2008
2008
MA
CA
VT
HI
NY
RI
MD
DE
WA
CT
IL
OR
ME
NM
MN
CO
NJ
PA
MI
NV
NH
SC
WI
IA
VA
FL
MO
GA
MS
TX
NC
OH
MT
LA
AZ
TN
SD
IN
WV
AK
KS
AR
ND
KY
AL
NE
UT
ID
OK
WY
CA
CO
AZ
NM
MI
WA
WI
OH
MO
NV
MD
MN
MT
ID
FL
NC
UT
ME
OR
TX
GA
VA
TN
IA
NH
PA
SC
IN
AR
WY
MS
ND
AL
OK
KS
CT
KY
VT
DE
MA
NJ
IL
NY
NE
SD
HI
LA
WV
RI
AK
party
D
Public
R
chamber
House
Public
Senate
−1.5
−1.0
−0.5
0.0
0.5
1.0
party
D
R
chamber
H
S
0.0
Score
0.5
1.0
Extremism
(a) Score
(b) Extremism
Figure 6: Lemost plot shows state public opinion mean and party medians within chambers on the common
NPAT scale. Rightmost plot is ideological distance between state opinion means and chamber party medians.
Positive values indicate party medians are more extreme than the state; negative values are more moderate than
the state (eg, AK). Le plot is sorted by score, right plot is sorted with by extremism.
19
Party Medians and the Public, 2008
House Democrats
House Republicans
CO
CA
1.0
WA
0.5
AK
VT
ME
WI
NMMN MI
NVSC
MD
OR
PANH
DE
MS LA
0.0
Chamber Median
−0.5
−1.0
−1.5
PA SC
RI DE
NV
IL
NH IA
VA
HI
VT
ME
MA NY WA CT
CO
OR
MN
FL
WI
MD
MI
NM NJ
GA
NC
AR
OK
KY
SD
WV NDAL
TN
KS
IN
VA
IA FL
MOOH
AZ
GA
MSMT
TX
NC
LA
IL
CT
MA HI RI
NY
TN AR
IN
ND
AL
SD KS KY
WV
OK
UT ID
WY
AK
NJ
UT ID WY
MT
TX
MO
OH
AZ
CA
Senate Democrats
Senate Republicans
CO
CA
1.0
MD WA
AK
0.5
HI
VT
OK
0.0
DE
RI
−0.5
−1.0
MS LA
MA
HI
VT
MD
NY
CA
SC
IL OR
IA
NMMN PANV
ME
NJ
WA
CO
FL
MI WI
CT
VA
GATX
MO MT
NC
OH
AZ
NE
AR
SD
WV ND
TN
IN KS KY
AL
MA
MS
MO
GA
AZ
NC
MT
OH
TXLA
NM
NV
MI WI
OR
SC
ME MN PANH IA
VA
DE
FL
AR
OK
KY
NDAL
NE UT ID
IN
SD
AK
WY
WVKS
TN
CT
IL
RI
NY
NJ
WY
UT
ID
NH
−1.5
−0.2
0.0
0.2
−0.2
0.0
0.2
State MRP Mean
(a) Scatter
Congruence by Party
H
S
1.5
Density
1.0
0.5
0.0
−2.0 −1.5 −1.0 −0.5
0.0
0.5
1.0
1.5
2.0
−2.0 −1.5 −1.0 −0.5
0.0
0.5
1.0
Distance to State Mean
(b) Density
Figure 7: Congruence at the party median level, within chambers.
20
1.5
2.0
Table 2: Party Median Distance Models
Dependent variable:
Absolute Distance from State Mean
0.097∗∗∗
(0.033)
Republican Median
−0.005∗∗
(0.002)
Chamber Competitiveness
Initiative
0.069
(0.052)
−1.396∗∗∗
(0.481)
Interest Group Density
Term Limits
0.047
(0.057)
Logged District Magnitude
0.069∗∗∗
(0.023)
Legislative Professionalism
0.079
(0.233)
Constant
0.421∗∗∗
(0.142)
Observations
Log likelihood
Akaike Inf. Crit.
Bayesian Inf. Crit.
198
−20.282
60.564
93.447
∗
Note:
21
p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01
6 Discussion and Conclusion
My goal in this paper is the creation of an ideological common space that includes individuals
and state legislators (individually and aggregated at the chamber and party levels) in common space.
e presence of those scores–varying across space and time–open up the possibility for exploring
any number of interesting questions in representation, lawmaking, and public policy. I illustrated a
few examples in this paper.
State legislators and state legislative chamber and party medians are responsive to variation in
district and state-level public opinion. But they are quite oen strikingly incongruent to it. Public
opinion is considerably more moderate than elected officials’ ideology. e variation in this incongruence is partly explained by political party. Republicans are considerably more incongruent than
Democrats, at the individual and aggregate levels. Other factors include structural features of state
institutional environments, like district magnitude.
is pattern of responsiveness combined with incongruence has been noted before, but only at
the level of individual policy issues (Lax and Phillips 2009b,a). e ndings here suggest a broader
conclusion. e policy views of elected officials are systematically distant from their constituencies.
Getting good measures of public opinion on a common scale with their representatives is going to be useful in illuminating the role of institutions as moderators of incongruence. Public and
elite preferences are insufficient, on their own. is is because those preferences are ltered through
the institutional environment that characterizes each state. e amount of state institutional heterogeneity is staggering. ese include supermajoritarian decision rules, legislative professionalization,
term limits, campaign nance regulations, primary institutions, and much more. For example, the
heterogeneity in the quality of representation could be in part caused by institutional differences
between states. (Gerber 1996; Lax and Phillips 2009a; Matsusaka 2010). An integrated perspective
is required to illuminate state policy outcomes, requiring all three parts–public opinion, legislative
ideology, and institutions.
22
References
Achen, Christopher. 1975. “Mass Political Attitudes and the Survey Response.” American Political
Science Review 69:1218–1223.
Ansolabehere, Stephen, James Snyder and Charles Stewart. 2001. “e Effects of Party and Preferences on Congressional Roll-Call Voting.” Legislative Studies Quarterly 26(4):533–572.
Ansolabehere, Stephen, Jonathan Rodden and James Snyder. 2008. “e Strength of Issues: Using Multiple Measures to Gauge Preference Stability, Ideological Constraint, and Issue Voting.”
American Political Science Review 102(2):215–232.
Bafumi, Joseph and Michael C. Herron. 2010. “Leapfrog Representation and Extremism: A
Study of American Voters and eir Members in Congress.” American Political Science Review
104(3):519–542.
Bailey, Michael A., Brian Kamoie and Forrest Maltzman. 2005. “Signals from the Tenth Justice: e
Political Role of the Solicitor General in Supreme Court Decision Making.” American Journal of
Political Science 49(1):72–85.
Bailey, Michael and Kelly H. Chang. 2001. “Comparing Presidents, Senators, and Justices:
Interinstitutional Preference Estimation.” e Journal of Law, Economics, and Organization
17(2):477–506.
Calvert, Randall L. 1985. “Robustness of the multidimensional voting model: Candidate motivations, uncertainty, and convergence.” American Journal of Political Science pp. 69–95.
Clinton, Joshua D. 2006. “Representation in Congress: Constituents and Roll Calls in the 106th
House.” e Journal of Politics 68(2):397–409.
Clinton, Joshua, Simon Jackman and Douglas Rivers. 2004. “e Statistical Analysis of Roll Call
Data.” American Political Science Review 98:355–370.
Converse, Phillip E. 1964. e Nature of Belief Systems in Mass Publics. In Ideology and Discontent,
ed. David E. Apter. New York: Free Press.
Epstein, Lee, Andrew Martin, Jeffrey Segal and Chad Westerland. 2007. “e Judicial Common
Space.” Journal of Law, Economics, and Organization 23:303–325.
Erikson, Robert S., Gerald C. Wright and John P. McIver. 1993. Statehouse Democracy: Public
Opinion and Policy in the American States. New York: Cambridge.
Gelman, Andrew and Jennifer Hill. 2006. Data Analysis Using Regression and Multilevel/Hierarchical
Models. Cambridge University Press.
Gerber, Elisabeth R. 1996. “Legislative Response to the reat of Popular Initiatives.” American
Journal of Political Science 40(1):99–128.
23
Gerber, Elisabeth R. and Jeffrey B. Lewis. 2004. “Beyond the Median: Voter Preferences, District
Heterogeneity, and Political Representation.” Journal of Political Economy 112(112):1364–1383.
Groseclose, Tim, Steven D. Levitt and Jr. Snyder, James M. 1999. “Comparing Interest Group Scores
across Time and Chambers: Adjusted ADA Scores for the U.S. Congress.” e American Political
Science Review 93(1):33–50.
Groseclose, Timothy. 2001. “A Model of Candidate Location When One Candidate Has a Valence
Advantage.” American Journal of Political Science 45:862–886.
Hacker, Jacob S and Paul Pierson. 2005. Off center: e Republican revolution and the erosion of
American democracy. Yale University Press.
Hix, Simon, Abdul Noury and Gerard Roland. 2007. Democratic Politics in the European Parliament.
Cambridge University Press.
Holbrook, omas M and Emily Van Dunk. 1993. “Electoral Competition in the American States.”
American Political Science Review 87(04):955–962.
Jackman, Simon. 2000. “Estimation and Inference are Missing Data Problems: Unifying Social
Science Statistics via Bayesian Simulation.” Political Analysis 8:307–332.
Jackman, Simon. 2004. “Bayesian Analysis for Political Research.” Annual Review of Political Science
7:483–505.
Jackman, Simon and Lynn Vavreck. 2010. “Primary Politics: Race, Gender, and Partisanship in the
2008 Democratic Primary.” Journal of Elections, Parties, and Public Opinion 20(2):153–186.
Jessee, Stephen A. 2009. “Spatial Voting in the 2004 Presidential Election.” American Political Science
Review 103:59–81.
Jessee, Stephen A. 2010. “Partisan Bias, Political Information and Spatial Voting in the 2008 Presidential Election.” Journal of Politics 72:327–340.
Kastellec, Jonathan P., Jeffrey R. Lax and Justin H. Phillips. 2010. “Public Opinion and Senate Conrmation of Supreme Court Nominees.” Journal of Politics 72(3):767–784.
Kousser, ad, Justin H. Phillips and Boris Shor. 2014. “Reform and Representation: Assessing
California’s Top-Two Primary and Redistricting Commission.”.
Lax, Jeffrey and Justin Phillips. 2009a. “Gay Rights in the States: Public Opinion and Policy Responsiveness.” American Political Science Review 103(03):367–386.
Lax, Jeffrey R. and Justin H. Phillips. 2009b. “How Should We Estimate Public Opinion in e
States?” American Journal of Political Science 53(1):107–121.
Londregan, John B. 2000b. Legislative Institutions and Ideology in Chile. New York: Cambridge
University Press.
24
Mann, omas E and Norman J Ornstein. 2013. It’s even worse than it looks: How the American
constitutional system collided with the new politics of extremism. Basic Books.
Martin, Andrew D. and Kevin M. Quinn. 2002. “Dynamic Ideal Point Estimation via Markov Chain
Monte Carlo for the US Supreme Court.” Political Analysis 10(2):134–153.
Matsusaka, John G. 2010. “Popular Control of Public Policy: A Quantitative Approach.” Quarterly
Journal of Political Science 5:133–167.
Mayhew, David R. 1974. Congress: e Electoral Connection. New Haven: Yale University Press.
McCarty, Nolan, Keith T. Poole and Howard Rosenthal. 2013. Political Bubbles: Financial Crises and
the Failure of American Democracy. Princeton: Princeton Univ Press.
McCarty, Nolan M. and Keith T. Poole. 1995. “Veto Power and Legislation: An Empirical Analysis of Executive and Legislative Bargaining from 1961 to 1986.” Journal of Law, Economics, &
Organization 11(2):282–312.
Morgenstern, Scott. 2004. Patterns of Legislative Politics: Roll Call Voting in the Latin America and
the United States. Cambridge: Cambridge University Press.
Park, David K, Andrew Gelman and Joseph Bafumi. 2004. “Bayesian Multilevel Estimation with
Poststrati cation: State-level Estimates from National Polls.” Political Analysis 12:375–385.
Poole, Keith. 2005. Spatial Models of Parliamentary Voting. Cambridge: Cambridge University
Press.
Poole, Keith and Howard Rosenthal. 1991. “Patterns of Congressional Voting.” American Journal of
Political Science 35(1):228–278.
Poole, Keith and Howard Rosenthal. 1997. Congress: A Political-Economic History of Roll Call Voting.
New York: Oxford University Press.
Poole, Keith T. 1998. “Recovering a Basic Space from a Set of Issue Scales.” American Journal of
Political Science 42:954–993.
Poole, Keith T. 2000.
8:211–237.
“Non-parametric Unfolding of Binary Choice Data.” Political Analysis
Rodden, Jonathan, Boris Shor, Christopher Warshaw, Christopher Tausanovitch and Nolan McCarty. 2014. “Geography and Polarization.”.
Shor, Boris, Christopher Berry and Nolan McCarty. 2010. “A Bridge to Somewhere: Mapping State
and Congressional Ideology on a Cross-Institutional Common Space.” Legislative Studies Quarterly 35(3):417–448.
Shor, Boris and Nolan McCarty. 2011. “e Ideological Mapping of American Legislatures.” American Political Science Review 105(3):530–551.
25
Tausanovitch, Chris and Christopher Warshaw. 2013. “Measuring Constituent Policy Preferences
in Congress, State Legislatures, and Cities [Forthcoming].” Journal of Politics .
Voeten, Erik. 2000. “Clashes in the Assembly.” International Organization 54:185–215.
Warshaw, Christopher and Jonathan Rodden. 2012. “How Should We Measure District-Level Public
Opinion on Individual Issues?” Journal of Politics 74(1):203–219.
Wittman, Donald. 1983. “Candidate motivation: A synthesis of alternative theories.” e American
political science review pp. 142–157.
26