ACCOUNTABILITY IN A F EDERAL S YSTEM
S TEVEN M ICHAEL ROGERS
A D ISSERTATION
P RESENTED TO THE FACULTY
OF
IN
P RINCETON U NIVERSITY
C ANDIDACY FOR THE D EGREE
OF
D OCTOR OF P HILOSOPHY
R ECOMMENDED FOR ACCEPTANCE
BY THE
D EPARTMENT OF
P OLITICS
A DVISER : N OLAN M C C ARTY
S EPTEMBER 2013
c Copyright by Steven Michael Rogers, 2013.
All rights reserved.
Abstract
Theories of political accountability suggest that officeholders should be electorally punished
when they perform poorly or fail to represent their constituents. This dissertation evaluates
the extent to which these theories apply to state legislatures by addressing the question:
do elections hold state legislators accountable for their own performance? In analyses
of state legislative challenger decision-making, party performance, and representatives’
roll-call activity, I find little evidence of electoral accountability in state legislatures. Over
a third of state representatives do not face major party challengers in the general election,
and when state legislators face competition, voters do not appear to reward or punish
state legislators for state-level policy outcomes, their legislative records, or their general
performance. Instead of serving as a referendum on state legislators’ own actions, state
legislative elections are dominated by national politics. State legislators affiliated with
the president’s party - especially during unpopular presidencies - are the most likely to be
challenged, and compared to individuals’ assessments of the state legislature, changes in
presidential approval have at least three times the impact on voters’ decision-making in state
legislative elections. Thus, while state legislatures wield considerable policy-making power,
elections appear relatively ineffective in holding state legislators accountable for their own
performance and lawmaking.
iii
Acknowledgements
Advisers, professors, and friends taught me many things about myself and politics
while drafting this dissertation, but at this point, I wish I learned how to be a better writer.
This weakness of mine may worry those who are about to read the dissertation I wrote, but it
concerns me here because no sentences I write can adequately express my gratitude to Larry
Bartels and Nolan McCarty. Each provided me guidance and remarkable patience, but most
importantly, Nolan and Larry pushed me to consider politics from different perspectives
while commonly doing what they thought was best for me. For this, I am grateful. I am
also indebted to Brandice Canes-Wrone, Sarah Binder, and especially Josh Clinton for
their constructive comments and continued encouragement from both near and afar (and
sometimes near again). I can only hope to provide my future students the dedication my
advisers gave me.
While writing this dissertation, I was fortunate to be welcomed by two departments
who provided me many happy hours. Faculty residing in the halls of Robertson, basement
of Corwin, and Ingram Commons helped me hone this dissertation’s argument through
conversations, seminars, and generously commenting on drafts. My Princeton and Vanderbilt
experiences would not have been possible without the help of Michele Epstein, Jayne
Cornwell, Natasha Duncan, and Keith Whittington - who I commend for his excellent
service as director of graduate studies. I am additionally grateful to Doug Arnold for
agreeing to serve on the defense examination committee.
I extend special thanks to my fellow graduate students for their continued comradery
and support. I was lucky to be joined in the bullpen by individuals like Matt Tomkowiak,
Alex Bolton, and Matt Tokeshi, and my friends provided invaluable escapes both at the
poker table in Princeton and trivia nights in Nashville. I was very fortunate to have Jeff
Tessin’s and Nick Carnes’ guidance and sympathetic ears from the beginning to the end of
my graduate career, and I could not imagine my time at Princeton without Scott Abramson,
iv
Michael Donnelly, and especially Deborah Beim. I am thankful for our trips to Hopewell,
their pull-out couch, and most of all friendship.
This dissertation would not have been possible without the data collection efforts of
(as an empiricist I am embarrassed to say) too many to count. I thank Michelle Anderson,
Adam Bonica, Michael Davies, Carl Klarner, Peter Koppstein, Boris Shor, the Center for
Congressional and Presidential Studies, the Center for the Study of Democratic Politics, the
Center for the Study of Democratic Institutions, the National Committee for an Effective
Congress, the National Institute for Money in State Politics, the Sunlight Foundation, and
numerous individuals from Secretaries’ of States offices or local boards of elections for their
too often underappreciated efforts.
Most importantly, I am grateful for the continued love and support of Karen, Bill, and
Alex Rinehart, along with Peter and Susan Rogers - to whom this dissertation is dedicated.
v
Dedicated to Susan Etta Rogers
vi
Contents
1
2
3
4
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
iii
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
iv
Introduction
1
1.1
Challengers, Parties, & Individual Representation . . . . . . . . . . . . . .
3
1.2
Accountability in a Federal System . . . . . . . . . . . . . . . . . . . . . .
8
Competition in Legislative Elections
10
2.1
Challenger Entry & Strategy . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2
Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3
Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.4
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
How Parties Perform in Office & Elections
32
3.1
Requirements for Collective Accountability . . . . . . . . . . . . . . . . . 34
3.2
State-Level Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.3
Individual-Level Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.4
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
Accountability for State Legislative Roll-Calls & Ideological Representation
57
4.1
Holding “Out of Step” Legislators Accountable . . . . . . . . . . . . . . . 59
4.2
Accountability for Unpopular Roll-Call Votes . . . . . . . . . . . . . . . . 64
vii
5
4.3
Accountability for Ideological Representation . . . . . . . . . . . . . . . . 69
4.4
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
Conclusion
84
A Appendix for Chapter 2
92
B Appendix for Chapter 3
99
B.1 Supplementary Estimates for Seat Change Analysis . . . . . . . . . . . . . 99
B.2 Supplementary Estimates for Survey Analysis . . . . . . . . . . . . . . . . 106
C Appendix for Chapter 4
113
C.1 Full Estimates of Individual Roll-Call Analysis . . . . . . . . . . . . . . . 114
C.2 Full Estimates of Ideological Representation Analysis . . . . . . . . . . . . 118
C.3 Alternative Measurement of Extremity . . . . . . . . . . . . . . . . . . . . 122
C.4 Electoral Accountability for Party Loyalty . . . . . . . . . . . . . . . . . . 125
Bibliography
129
viii
1
Introduction
“
All politics is local.
”
Tip O’Neill, Massachusetts House of Representatives, 1936 - 1952
Every two years voters go to the polls and elect individuals to represent them in American
legislatures. Once elected, little constrains officeholders’ behavior, but if these representatives govern irresponsibly, they can be replaced. By providing voters opportunities to
hold those in power accountable, American elections establish a fundamental connection
between citizens and elites that can motivate elected officials to act in the interests of those
they represent (Federalist 52).
For elections to be effective accountability mechanisms, G. Bingham Powell suggests at
least two benchmarks be met. Voters must “know who is responsible for making policy” and
“have a fair opportunity to cast a meaningful vote for or against the policymakers” (Powell
2000: 51). Despite Americans disinterest in politics (e.g. Delli-Carpini and Keeter 1996),
enough voters at times fulfill Powell’s first requirement to meaningfully carry out the second.
National elections appear to hold both political parties (e.g. Tufte 1975) and members of
1
Congress (e.g. Canes-Wrone et. al 2002) accountable for how they govern, suggesting that
there can be meaningful relationships between voters’ and legislators’ behavior.
Electoral connections help explain Congressional lawmaking (e.g. Fenno 1968; Mayhew
1974; Cox & McCubbins 1993), but most American lawmaking does not occur in Washington, DC. For every law that Congress passes, state legislatures pass 75. States regulate
the economy, provide health care, and even recently redefined the institution of marriage,
continuing a long-history of consequential policy-making. States’ voter identification laws
made headlines in the 2000s, and some see these policies as successors to earlier poll taxes.
Only when the Supreme Court overruled legislatures did these taxes end in the 1960s, and a
decade later, the high court again stepped in to strike down a Texas statute in Roe v. Wade.
The War on Drugs gained momentum when 49 states enacted mandatory minimums before
Congress took similar action in the 1980s (Murdoch 2001), and after the Brady Bill in the
1990s, 20 legislatures passed laws to make carrying a concealed handgun legal in a total of
38 states (Grossman and Lee 2008).
When drafting civil rights, public safety, and even immigration policies in state capitals,
little directly constrains state legislators’ actions. Many assume that competition for votes
pressures those in government to produce representative policies (e.g. Federalist 52; Downs
1957, Erickson, Wright, and McIver 1994). Others additionally assert that “a retrospective
voting electorate will enforce electoral accountability,” leading to desirable governance
(Fiorina 1981: 11; Key 1966). Median voter and retrospective voting theories provide
valuable insights regarding American government, but before leveraging these theories in
explanations of state politics, it is necessary to have evidence supporting their underlying
assumptions.
To determine whether such evidence exists, this dissertation addresses the question: do
elections hold state legislators accountable for their own performance?
The theories addressed by this dissertation are straightforward but fundamental. Theories
of elections predict that legislators will be electorally successful when they perform well
2
or act in constituents’ interests (Downs 1957; Key 1966), and I test whether this basic
prediction applies to state legislative elections. I focus on elections serving as a solution
to a moral hazard problem (e.g. Barro 1973; Ferejohn 1986) but acknowledge there are
many other and sometimes richer predictions (Fearon 1999; Ashworth 2012). For example,
factors such as voters’ partisanship and institutions also obviously influence elections and
are important parts to the accountability story (Campbell, Converse, and Stokes 1960). I aim
to account for these and other variables where possible and reasonable. The primary aim of
this dissertation, however, is to provide the strongest empirical investigation regarding the
relationship between state legislators’ behavior and election outcomes.
With this objective, I focus on the extent to which state legislators control their own electoral fate. Potential electoral rewards are intended to motivate legislators’ behavior, but even
with the considerable portion of lawmaking conducted by states, there are disproportionately
few analyses that address legislators’ electoral incentives to represent their constituents. The
most thorough studies of the electoral consequences an individual representative’s behavior
only examine two election years in less than a third of states (Hogan 2004; 2008). More
attention is given to the relationship between the economy and which party is successful in
state legislative elections (Berry, Berkman, and Schneiderman 2000; Campbell 1986; Chubb
1988), but outside of a single study that controls for divided government (Lowry, Alt, and
Ferree 1998), I know of no published work that accounts for which party controls the state
house. The scarce consideration of state legislators’ and their parties’ electoral incentives
leaves it largely unclear if median voter or retrospective voting theories meaningfully apply
to most state legislative elections.
1.1
Challengers, Parties, & Individual Representation
This dissertation adds clarity by addressing both voters’ and elites’ behavior in three
empirical chapters. My analyses evaluate the electoral consequences of state legislative
3
challenger decision-making, party performance, and representatives’ roll-call activity. My
findings, however, fail to deliver encouraging evidence for the prospects of state-level
accountability. Elections do not appear to hold state legislators accountable for state-level
policy outcomes, their legislative records, or their general performance, providing little
evidence that the assumptions underlying median voter or retrospective voting theories apply
to state legislatures. Instead of serving as a referendum on state legislators’ own actions,
state legislative elections are dominated by national politics.
My evidence for this argument begins where electoral competition typically starts: a
candidate’s decision to challenge the incumbent - which happens relatively infrequently in
state legislative elections. Rarely do more than 60% of sitting state house members face a
major party opponent, leaving millions of voters without the opportunity to hold their state
representative accountable. Political scientists show that challengers emerge more often
in professionalized legislatures (Squire 2000; Dunk and Weber 1997; Weber, Tucker, and
Brace 1991) or states with competitive districts (Hogan 2004; 2008). These explanations of
competition in state legislative elections provide insights regarding where state legislators
face challenges but surprisingly rarely account for variation across time.
Cross-time variation is important for assessing electoral accountability, especially when
one considers that few institutions change between elections but the rates of challengers
often do. For example in 2008, Republicans challenged less than 50% of sitting Democratic
state representatives, but over 66% of Democratic incumbents were challenged in 2010.
Going into these respective elections, both casual observers of politics and political elites
likely recognized the state of the economy would be electorally helpful for Democratic
politicians in 2008 but detrimental in 2010. If potential challengers consider dynamic
political conditions such as the economy in their decision-making, which state legislators
receive opposition and who voters can hold accountable will vary from one election to the
next. Gary Jacobson (1989) shows temporal political conditions relate to challenger entry
and levels of accountability in federal elections, but I do not know of research that provides
4
systematic evidence that state legislative challengers take advantage of favorable electoral
circumstances.
I remedy this oversight in the second chapter of this dissertation by providing evidence
of how economic and political contexts in addition to institutions influence state legislative
challenger entry over the past two decades. I find state legislators more often face major party
challengers during economic downturns, making incumbents who oversaw a weak economy
increasingly vulnerable in elections. There is little difference between the economy’s impact
on the likelihood that members of the state house majority or minority parties face an
opponent, but members of the governor’s party more frequently face challengers during less
prosperous times. By running against the governor’s copartisans in the legislature, these
candidates provide voters a greater number of opportunities to hold the governor’s party
collectively accountable for their management of state affairs.
To threaten incumbents’ reelection, it is necessary for elites to run against those in the
state house, but state-level collective accountability will only be brought about if there is a
meaningful relationship between state-level conditions and election outcomes for members
of the parties in power. In low-information elections such as those for the state legislature,
using retrospective evaluations and party labels potentially simplifies the accountability
process (e.g. Schattschneider 1942; Key 1961; Fiorina 1981). If voters sanction the party in
control of the legislature for producing ineffective policies, how legislative parties perform
in office will have implications for how they perform in elections.
A national look at how parties perform in legislative elections, however, makes it
questionable that state-level factors are the main determinant of state legislative elections.
Figure 1.1 plots the nationwide seat change for the Democratic party in state (black solid
line) and federal legislative elections (grey dashed line) over the past hundred years. Seats
clearly change party hands, but the similarity between federal and state elections is striking.
When Democrats took control of the U.S. House in 2006, this party also won hundreds
of state legislative seats. The story is similar for Republicans both in the 1994 and 2010
5
Figure 1.1: Democratic Seat Change in State House and U.S. House Elections
Nationwide proportion of seats won or lost by the Democratic party in state house or U.S. House
elections over the last hundred years.
elections. While the correlation (.96) between the solid and dashed lines is not definitive, it
strongly suggests there is a common dimension underlying both state and federal elections.
In the third chapter of this dissertation, I evaluate the degree to which state-level retrospective voting is responsible for the data presented in Figure 1.1. I investigate whether
the governing state parties and their members are electorally punished when they perform
poorly in office. My analyses specifically focus on how changes in state legislative seats
and votes for these parties respectively relate to objective measures of performance - such
as the state economy - and subjective measures - such as an individual’s assessment of the
legislature. Neither study produces compelling evidence that elections hold members of
the state house majority or governor’s party collectively accountable. The state economy,
state policy outcomes, or voters’ approval of the legislature appear to have little - if any consequences for members of the governor’s or state house majority party in state legislative
6
elections. My findings suggest that how state parties perform may matter at the margins but
overall have relatively small electoral implications.
Retrospective party voting can simplify the accountability process, but voters fundamentally elect individual representatives not parties. Spatial theories of electoral competition
expect legislators to receive fewer votes if they fail to represent their district on a broad
ideological dimension, all else equal. If voters respond to how well they have been represented by their individual legislator, it would provide evidence for median voter theories’
assumption that competition for votes pressures lawmakers to adopt policies consistent with
their constituents’ preferences.
To better determine whether this assumption applies to state legislatures, I examine
the electoral implications of dyadic representation in the fourth chapter of this dissertation. Specifically, I conduct two district-level analyses that evaluate the extent to which
voters sanction legislators who cast unpopular roll-call votes or provide poor ideological
representation of their districts. The first study uses referendum election returns to create
district-level measures of public opinion on the exact bills voted on by their legislators,
and the second employs a new dataset that links district-level measures of election returns,
constituent ideology, and legislative behavior from 1998 - 2008 for 38 states. Together, these
data permit the most empirically thorough cross-state electoral analysis of the relationship
between public opinion and individual state legislators’ behavior to date.
Consistent with the findings of the previous chapters, both district-level analyses in the
fourth chapter provide little support for the proposition that electoral connections exist in
most state legislatures. Only 2 of 10 states offer statistically significant evidence that districts
punish their legislators for unpopular roll calls, and only 9 of 38 states provide evidence
that legislators pay an electoral price for ideologically extreme representation. Statistical
analyses predict a state house member providing the worst instead of the best district-level
ideological representation in the country results in less than a 3% change in vote share - a
difference too small to change the outcome of over 97% of 2008 state legislative elections
7
where the incumbent sought reelection. Weak district-level electoral connections combined
with safe state legislative seats make it difficult for districts to throw “out of step” legislators
out of office.
1.2
Accountability in a Federal System
The findings of this dissertation provide little evidence that state legislators face meaningful electoral consequences for their own actions, and the results presented in the following
chapters additionally counter the claim that “the electorate often manages to make an effective separation of its choices of presidential and state candidates” (Key 1964: 307).
My analyses demonstrate that state legislators’ electoral fates are more closely tied to the
performance of The White House than the state house. State legislators affiliated with
the president’s party - especially during unpopular presidencies - are the most likely to be
challenged, and compared to individuals’ assessments of the state legislature, changes in
presidential approval have at least three times the impact on voters’ decision-making in
state legislative elections. My findings suggest state legislators’ own performance appears
to have relatively little electoral relevance, and national politics are a driving force in state
legislative elections.
It should be made clear that these assertions are not condemnations of the voter but
instead of the accountability process in American legislatures. Voters alone are not responsible for this lack of electoral accountability and even seem to try to reward or punish those
who they believe are in power. Additionally in instances where electoral connections appear
to exist in state legislatures, partisan districts make it difficult to throw bad policymakers out
of office. The analyses in this dissertation demonstrate that elites’ decisions (Chapter 2),
voters’ efforts (Chapter 3), and the electoral system (Chapter 4) all play critical roles in the
accountability story and should not be overlooked.
8
These findings support my argument that there is little evidence that elections hold state
legislators accountable for what they do, and both voters’ and elites’ decision-making within
a system of electoral competition are responsible for the weak incentives for representation
in state legislatures. The “little evidence” I find does not necessarily rule out the possibility
that elections hold legislators accountable. My statistical analyses could be underpowered
or testing the wrong theories. Each of the following chapters, however, employs the most
powerful tests to date with arguably the best available data. And even when following
research designs that produce evidence of electoral accountability in Congress, I find little
support for the proposition that how state legislators perform in office matters for how they
perform in elections.
It is ultimately up to the reader to determine whether state legislative elections meet his
or her standards for electoral accountability. I again offer Powell’s requirements to serve as
benchmarks. For elections to be effective accountability mechanisms, voters must “know
who is responsible for making policy” and “have a fair opportunity to cast a meaningful
vote for or against the policymakers.” I use these basic requirements to guide the subsequent
chapters’ analyses. If the evidence I provide satisfies the reader’s own standards, it should
be reassuring that the predictions of the founding fathers and electoral theorists successfully
apply to state legislatures. But if they do not, it casts doubt on the proposition that the
fundamental electoral connection between citizens and those who govern them meaningfully
exists.
9
2
Competition in Legislative Elections
The threat of being thrown out of office is intended to pressure lawmakers to govern
responsibly, but voters can only replace irresponsible representatives if there is an alternative
to the incumbents. American voters are nearly always provided an alternative choice to
reelecting their president, governor, or member of Congress and therefore have the ability
to remove these officials from power. However in 2012, Republicans reclaimed control of
the South Carolina state house before voters cast any ballots, as not enough Democratic
candidates ran to secure a Democratic majority. Similarly in Rhode Island, 53% of state
legislative seats only had a Democratic candidate, thereby deciding which party controlled
the state legislature without any elections taking place. In each of these states, voters had no
opportunity to throw the incumbent government out of the state house.
With so few Democrats deciding to run for the state legislature in South Carolina
and fewer Republicans in Rhode Island, holding state legislators accountable for their
policymaking becomes difficult, if not impossible, for voters. To induce representative
policymaking, median voter theories require at least a meaningful threat of a challenger
(e.g. Downs 1957). Without competition, Delaware or South Carolina legislators have
little electoral incentive to take their constituents’ interests into account as they consider
legalizing gay marriage or agreeing to federal health insurance exchanges. Challengers play
10
a critical role in the accountability process not only by providing alternatives to ineffective
governments but also by bringing unrepresentative policymaking to the attention of voters
(Arnold 1992). An explanation of electoral accountability in state legislatures, therefore,
requires identifying the conditions under which incumbent state legislators face competition
from other political elites.
Political scientists, however, know relatively very little regarding how state legislative candidates take advantage of changing electoral circumstances. Prior work on state
legislative competition focuses almost exclusively on cross-state differences and ignores
the influence of changing conditions within a state. I know of no analysis that considers
whether the strength of the economy relates to state legislative candidates’ entry decisions
despite findings from congressional elections regarding strategic challengers (Jacobson
1989). Neither federal nor state-level challengers enjoy losing, so it would seem plausible
that state legislative candidates are more likely to enter races when conditions are most conducive to winning. However if candidates take advantage of favorable political conditions,
voters’ opportunities to hold their representatives accountable will systematically vary as
the contexts surrounding elections change.
To assess the extent to which these opportunities differ, I study the relationship between
electoral contexts and challenger entry in state legislative elections. I analyze elections
from the past two decades to examine whether changing political conditions, such as
economic growth and presidential popularity, influence challenger decision-making at
the state legislative level. I show that both institutional and political contexts shape the
opportunities voters have to hold their state legislator accountable. Only 60% of incumbents
nationwide typically receive major party opposition, but challengers most often emerge in
professionalized legislatures, competitive districts, and during bad economies. Challengers
appear to take advantage of weak economies to run against vulnerable and unrepresentative
legislators, and the relationship between economic growth and challenger entry is strongest
for incumbents affiliated with the governor. Members of the president’s party, meanwhile,
11
are overall the most likely to be challenged, especially during unpopular presidencies. Both
state and national politics, therefore, appear to influence state legislative candidates’ decisionmaking and the opportunities voters have to hold their state representatives accountable.
2.1
Challenger Entry & Strategy
Figure 2.1: Comparison of Contestation in U.S. and State House Elections
Solid lines represent the proportion of U.S. House (grey) and state house (black) incumbents who
faced a major party challenger in elections from 1982 - 2010. The difference in rates in challenger
entry is consistently greater than 20%. Dotted lines illustrate the rates in which open seats have
candidates from both political parties.
Ideally, every incumbent would be challenged to give voters an opportunity to hold their
legislators accountable for poor representation (Key 1949). If one follows the accountability
benchmarks set by Powell, only with “a fair opportunity to cast a meaningful vote for or
against the policymakers” can voters create electoral incentives to motivate their representatives’ behavior (Powell 2000: 51). These opportunities are relatively few in state legislative
elections. Figure 2.1 presents the proportions of challenged incumbents in U.S. and state
house races over the last three decades. Rarely did more than 60% of state house incumbents
12
face major party opposition (black line) - a rate 20% lower than that of Congressional
elections (grey line).1 Over a fourth of state house incumbents had no opponent in either the
2010 general or primary election, and some incumbents never face challengers. For example,
Louisiana Republican Emile Bruneau Jr. was “reelected” without any competition for over
18 years. Voters in the 94th Louisiana state house district, therefore, had little chance to hold
their representative or the Republican legislative party accountable because no one decided
to challenge Bruneau.
Figure 2.2: Proportion Contested and Open Seats across States
The above illustrates the proportion of contested or uncontested incumbent and open seats in state
legislative elections from 2001 - 2010 by state. Proportions in each column sum to 1. Incumbents
often do not face major party challenges in South Carolina or Massachusetts, but most Minnesota
and Michigan state representatives face such challenges.
Bruneau is an extreme example, and the rates of challenger entry vary across the United
States. Figure 2.2 illustrates the levels of contestation in states that exclusively have singlemember districts. Every election year, over 90% of Minnesota state legislators face major
party opposition and relatively few open seats go uncontested by both political parties, but
less than 30% of Arkansas legislators regularly encounter challenges. To explain this cross1
State house competition may have not always been this meager (Ray and Havick 1981). Van Dunk and
Weber (1997: Figure 2) find over 75% of state house races were contested in 1972 but fell 10% by 1986.
During this period, the rates of incumbents seeking reelection did not decline (Jewell and Breaux 1988). Those
who sought reelection were highly successful in primaries (Grau 1981; Jewell and Breaux 1991) and over time
became less likely to be challenged in the general election (Jewell and Breaux 1988: 507) .
13
state variation, prior studies focus on institutions, such as legislative professionalism (Hogan
2004; Squire 2000; Dunk and Weber 1997; Weber, Tucker, and Brace 1991), campaign
finance laws (Mayer and Wood 1995; Werner and Mayer 2007; Hamm and Hogan 2008;
Malhotra 2008), or term limits (Forgette, Garner, and Winkle 2009; Cain, Hanley, and
Kousser 2006).2 Despite providing a better understanding of differences across states,
existing research does not explain variation within states from one election to the next.
Considering the sources of cross-time variation is important for assessing electoral
accountability in state legislatures. As referenced in the introduction of this dissertation, few
institutions changed between the 2008 and 2010 elections but who entered state legislative
elections did. In 2008, Republicans challenged less than 50% of sitting Democratic state
representatives, but over 66% of Democratic incumbents were challenged in 2010. Meanwhile, the comparable rate of Republican incumbents facing challengers fell from 59% to
49% across these two elections. These differences in challenger entry in 2008 and 2010
likely were not random, and the respectively higher rates of Democratic and Republican
challengers partially explain the parties’ successes in these wave elections.
Candidates from both the federal and state legislative levels selectively enter races
(Canon and Sousa 1992; Jacobson and Kernell 1983; Maestas et al. 2006). Jacobson argues
that congressional candidates strategically run for office to take advantage of a president’s
popularity. To provide evidence for his strategic entry theory, his study demonstrates
that presidential approval correlates with the percentage of quality challengers in U.S.
House elections (Jacobson 1989: Table 3). This suggests candidates anticipate presidential
popularity will influence their own electoral success. During an unpopular Republican
2
Fiorina (1994) suggests challengers in professionalized states tend to be more Democratic, but he only
examines Democratic seat shares and not challenger emergence itself. In a survey of state legislative caucuses,
leaders from legislatures with higher salaries indicated they were more involved in candidate recruitment
(Sanbonmatsu 2006). Research is inconclusive regarding the immediate impact of redistricting. Weber, Tucker,
and Brace (1991) discover little relationship between reapportionment and challenger entry, but Pritchard
(1992) finds higher levels of contestation in Florida state legislative elections following redistricting in the
1970s. Conflicting findings may be attributable to the lack of attention given to redistricting principles (Forgette,
Garner, and Winkle 2009). Prior work finds legislators representing majority-minority or partisan districts less
likely to be challenged (Squire 2000; Hogan 2003; 2004; 2008).
14
presidency, for example, Democratic congressional challengers will take advantage of the
anti-president sentiment and be more likely to run. If Democrats adopt this common strategy,
voters will have more opportunities to electorally sanction Republicans who perform poorly
and thereby strengthen the relationship between federal parties’ behavior in office and their
members’ probabilities of reelection.
The theory of strategic challengers most straightforwardly translates to the state-level
when thought of in the context of the governor and state legislature. Potential state legislative
challengers may anticipate the governor’s coattails influencing their own electoral success
(Hogan 2005). Members of the opposition party then should be more likely to run when the
governor is unpopular, and by taking advantage of anti-gubernatorial sentiment, challengers’
entry decisions will effectively connect the performance of the governor’s party to its members’ electoral security. Similar to the federal level, state legislative challengers’ strategies
thereby promote partisan collective accountability and create incentives for members of the
state parties to perform well. If incumbent party members fail to perform, they will receive
more challenges, reducing their chances of reelection.
A complication for the application of this theory of challenger entry to state legislative
elections is the fact that state legislatures are embedded within a federal system where
national politics can influence state-level electoral outcomes (Chubb 1988; Carsey and
Wright 1998). Similar to riding gubernatorial coattails within a state, a potential state
legislative candidate may anticipate taking advantage of an anti-presidential wave. Watergate
provides a possible example of candidates at the federal and state levels adopting a common
strategy. In 1974, President Nixon’s average approval rating was 25% and consistent with
Jacobson’s theory, Democrats challenged all but one of the 165 Republican U.S. House
members who sought reelection. Voters then had many opportunities to hold the federal
Republican party accountable. However, Democrats also challenged every Republican state
legislator in over 50 state legislative chambers, well over twice the comparable figure for
15
Democrats (Tidmarch, Lonergan, and Sciortino 1986). Democrats gained over 500 state
legislative seats in 1974, so their entry decisions helped secure legislative majorities.
If challengers’ entry decisions depend on both state and national conditions, the implications for accountability are not straightforward. In congressional elections, a challenger
that responds to national conditions arguably promotes collective partisan accountability,
as Congressmen have direct influence over the national government. State legislators, in
contrast, have little control over federal politics, so if challengers systematically respond to
national conditions, this does not directly strengthen electoral connections at the state-level.
A weak economy or an unpopular presidency may result in more state legislators facing
opponents, but prosperous times or popular presidents could have the opposite effect and
reduce competition. Incumbents then may foresee the ability to ride favorable national
political conditions to unopposed reelection and pursue state-policy goals with less fear of
being held accountable through electoral punishment. Challengers responding to national
political conditions, therefore, can increase elections’ constraint on some policymakers but
loosen it for others, regardless of political actors’ own actions.
Despite its important implications for accountability, it is relatively unknown if state
legislative challengers take advantage of favorable political conditions and the extent to
which these strategies have consequences for representation in state legislatures. Prior
work predominantly focuses on institutional variation across states without accounting for
differences within states.3 No existing research considers whether changes within a state’s
3
Hogan provides two analyses that investigate the role of legislators’ behavior. Studying fourteen states
in two election years, he finds mixed results between incumbents’ interest group ratings and whether they
are contested Hogan (2004; 2008). Weber, Tucker, and Brace (1991) discover income tax increases have
no relationship on challenger emergence in twelve of fourteen different states, but in a later study of all
states, Dunk and Weber (1997) find that raising income taxes increase incumbent contestation rates by two
percent. Looking at the legislative behavior-contestation relationship in the other direction, there is relatively
little association between whether an incumbent is challenged and their actions in the legislative chamber.
Uncontested incumbents miss only 1 - 2% more roll call votes (Konisky and Ueda 2011). There is no
relationship between contestation and bill enactment, but unopposed state senators, introduce 10% fewer bills
than senators who previously faced competition.
16
economy have any relationship with challenger entry, and political scientists give little
attention to the role of parties.4
To better understand whether state legislative candidates enter races during electorally
favorable political contexts, I examine elections over the past twenty years. If state legislative candidates are strategic similar to their congressional counterparts, changing political
contexts – either at the state or national levels – should influence their entry decisions,
resulting in more incumbents facing challengers during unfavorable conditions. If challengers anticipate state-level conditions will have a greater impact on the electoral fates
of those who control state government, the relationship between state economic growth
and challenger entry should be stronger in races where a reelection seeking legislator is
a member of the governor’s or state house majority party. Support for this hypothesis
suggests challengers’ decision-making gives voters more opportunities to hold state parties
collectively accountable for actions taken in the state house. If, however, state legislative
challengers take advantage of national political conditions, there should be a negative relationship between presidential popularity and the likelihood the president’s co-partisans
in the legislature receive opponents. Evidence for this hypothesis suggests that national
politics also determine state legislators’ electoral fates.
2.2
Data
To test these hypotheses, I examine elections from 1991 to 2010 in 37 states.5 The
dependent variable is whether a sitting state representative (who survived the primary) from
a single-member district received a major party opponent (Klarner et. al 2013; Shor &
McCarty 2011). I investigate how the probability of a challenger varies under different
4
Hogan (2004) includes a Democratic member dummy in his analysis of 1996 and 1998 elections. Again
studying these two years, Hogan (2008) includes a control for whether a legislator was in the majority party.
Of research conducted in 1980s and 1990s, only Tidmarch, Lonergan, and Sciortino only control for party,
presenting the percentages of uncontested seats by chamber, party, and year (1986: Table 3)
5
Excluded states are AZ, ID, MD, ND, NE, NJ, SD, WA, WV, VT, and sometimes NC due to multimember
districts. I exclude Louisiana due to its run-off system and Nebraska because it is nonpartisan.
17
institutional contexts along with conditions that change between elections, such as state
economic performance or presidential popularity. To measure economic growth within
a state, I rely on the annual change in logged state personal income as measured in the
second quarter.6 This approach follows studies of federal elections and is intended to capture
economic conditions for the approximate time period when many candidates decide to
challenge an incumbent (Moncrief 2001: Table 2.7). For national political contexts, I rely
on the president’s approval rating from the last Gallup poll before July to capture quarter
two attitudes towards the president. This variable is measured from 0 to 1.7
If state legislative candidates follow strategies similar to their federal counterparts, I
expect economic or political conditions’ impact on entry decisions to differ by which parties
controlled political institutions. To provide support for the aforementioned hypotheses,
statistical analyses should produce negative relationships between economic growth and the
likelihood of challengers opposing incumbents of the state house majority or governor’s
party. Otherwise stated, as the economy prospers, members of the parties in power in state
government are safer from competition. If challengers take advantage of national political
conditions, I expect a negative relationship between presidential approval and challenger
entry for members of the president’s party. To capture these party specific effects, I subset
the data by incumbents’ affiliations with the state house majority, governor’s, or president’s
party. Results are similar when pooling data and including interaction terms.
When investigating voters’ opportunities to throw unresponsive legislators out of office,
Hogan (2004) finds little relationship between state legislators’ policy responsiveness and
challenger entry. His analysis, however, does not consider economic contexts. To assess
whether incumbents who provide poor representation are more vulnerable to challengers
during certain political conditions, I recreate Hogan’s measure of “policy responsiveness” for
6
This economic measure is adjusted for inflation using CPI conversion factors. Findings using a national
economic measure produce results consistent with those below, but the magnitude and statistical significance
of party specific estimates are sensitive to the inclusion of the 2008 and 2010 elections. Results presented
below are not.
7
Results are similar when using annual multi-level regression with post-stratification (MRP) estimates of
state-level presidential approval.
18
each state legislator using roll-call ideal points instead of interest group scores. Specifically,
I regress a legislator’s ideal point (Clinton, Jackman, and Rivers 2004; Shor and McCarty
2011) on a measure of state legislative district ideology (Tausanovitch and Warshaw 2013)
and use the absolute value of the standardized residuals of this regression to capture responsiveness, as it reflects the degree to which a legislator’s roll-call behavior diverges from
constituent preferences.8 Legislative roll-call data is available through 2008 for most states
but estimates of district ideology predominantly rely on surveys from the late 2000s. Instead
of assuming district ideology remains constant over a decade, I use this responsiveness
measure to illustrate trends in the 2006 and 2008 elections.
Following prior work, I control for institutional variation across states and elections, such
as district size, presence of term limits, and whether an election took place in the south, the
midterm (e.g. 2006 or 2010), the off-year (e.g. 2007 or 2009), or after redistricting (Dubin
2007). Each estimation controls for a state legislature’s professionalism using Squire’s
index that accounts for differences across states in legislators’ pay, staff, and length of state
legislative session (Squire 2007). I, furthermore, control for political variables at the state
and district levels that potentially influence challengers’ decisions. For example, candidates
may not want to be part of a meaningless minority party or face an unfriendly district. To
account for such possibilities, I control for the pre-election seat share of the minority party,
district’s partisanship, and an incumbent’s past electoral success. I measure partisanship
using district-level presidential vote for the incumbent state legislator’s party.9 The number
of terms served by the state representative and his vote share in the most recent election
capture an incumbent’s previous electoral success.10
8
Bond, Covington, and Fleisher (1985) and Lascher, Hagen, and Rochlin (1996) also employ this type of
responsiveness measure, but this measure makes a potentially inappropriate assumption regarding the linear
relationship between legislator and district ideology (Matsusaka 2001). Standardizing the residual intends to
create comparability of responsiveness across states but legislators’ ideal points themselves are not estimated
across states.
9
My analysis is missing Gore-Bush vote for AR, CO, and MS and Kerry-Bush vote for FL and MS.
10
Expensive races also likely deter challengers, but campaign finance data is not available for the full set of
elections studied here. Table A.1 in the Appendix includes comparable estimates that control for the logged
average amount of contributions to winning candidates in a given year and state where data is available (Bonica
2013).
19
Table 2.1: Challenger Entry as a function of Institutional and Political Contexts
Variable
Change Annual Log Q2 State Personal Inc.
All Inc.
-2.821*
(0.416)
Q2 Presidential Approval
Minority Party Seat Share
Professionalism
Southern Dummy
Off Year Election
Logged District Size
Term Limits Enacted
First Election after Redistricting Dummy
Freshman Incumbent
Terms Served
District Pres. Vote of Incumbent’s Party
Prev. Incumbent Vote Share
Incumbent Previously Contested Dummy
Member of the Democratic Party
Midterm Election Dummy
Constant
Log-Likelihood
N
∗p
1.105*
(0.095)
0.496*
(0.106)
-0.580*
(0.025)
0.146*
(0.046)
0.100*
(0.018)
-0.053*
(0.019)
-0.016
(0.021)
-0.083*
(0.023)
-0.008*
(0.003)
-1.153*
(0.064)
-2.493*
(0.101)
-0.069
(0.037)
0.091*
(0.017)
-0.051*
(0.016)
1.568*
(0.173)
-16392.4
28436
Pres. Pty
-4.182*
(0.585)
-0.011*
(0.001)
1.250*
(0.134)
0.479*
(0.154)
-0.591*
(0.035)
0.120
(0.064)
0.136*
(0.026)
0.002
(0.027)
0.016
(0.038)
-0.087*
(0.032)
-0.008*
(0.004)
-1.433*
(0.092)
-2.327*
(0.144)
-0.011
(0.054)
0.215*
(0.028)
0.094*
(0.024)
1.654*
(0.254)
-8007.2
14321
˜Pres. Pty
-2.015*
(0.605)
0.000
(0.001)
1.066*
(0.137)
0.512*
(0.150)
-0.572*
(0.035)
0.218*
(0.067)
0.074*
(0.026)
-0.095*
(0.027)
0.175*
(0.035)
-0.073*
(0.032)
-0.008*
(0.004)
-0.903*
(0.090)
-2.592*
(0.143)
-0.088
(0.053)
-0.041
(0.029)
-0.198*
(0.024)
1.763*
(0.250)
-8223.6
14115
Gov. Pty
-5.387*
(0.610)
˜Gov. Pty
-0.824
(0.586)
Maj Pty
-2.816*
(0.526)
˜Maj. Pty
-2.902*
(0.685)
0.792*
(0.139)
0.495*
(0.149)
-0.511*
(0.033)
0.061
(0.066)
0.100*
(0.026)
0.014
(0.027)
-0.080*
(0.030)
-0.132*
(0.032)
-0.010*
(0.004)
-1.104*
(0.091)
-2.921*
(0.146)
-0.228*
(0.054)
-0.027
(0.024)
-0.012
(0.023)
2.202*
(0.252)
-7951.8
13820
1.511*
(0.133)
0.423*
(0.154)
-0.678*
(0.036)
0.270*
(0.064)
0.096*
(0.026)
-0.127*
(0.026)
0.043
(0.030)
-0.023
(0.032)
-0.006
(0.004)
-1.302*
(0.093)
-2.117*
(0.14)
0.055
(0.052)
0.205*
(0.025)
-0.093*
(0.023)
1.084*
(0.242)
-8367.5
14616
1.184*
(0.120)
0.565*
(0.139)
-0.622*
(0.032)
0.048
(0.060)
0.104*
(0.024)
-0.130*
(0.024)
-0.026
(0.027)
-0.079*
(0.029)
-0.006
(0.003)
-1.200*
(0.082)
-2.504*
(0.128)
-0.062
(0.047)
0.071*
(0.023)
0.008
(0.021)
1.578*
(0.224)
-10054.1
17501
1.526*
(0.185)
0.377*
(0.169)
-0.510*
(0.039)
0.290*
(0.072)
0.110*
(0.029)
0.067*
(0.031)
0.000
(0.035)
-0.085*
(0.036)
-0.014*
(0.005)
-1.428*
(0.115)
-2.505*
(0.164)
-0.104
(0.063)
0.044
(0.027)
-0.145*
(0.026)
1.466*
(0.284)
-6263.7
10935
≤ .05; Standard Errors in Parentheses
Probit estimates of the likelihood of a major party challenger contesting an incumbent in a state
house election from 1991 - 2010. First column pools all elections, and subsequent columns provide
estimates from data subset by the incumbent’s affiliation with the president’s, governor’s, or state
house majority party.
Given the dichotomous dependent variable of whether an incumbent receives a major
party challenger, I use probit regressions to estimate the relationship between challenger
entry and the above independent variables.11 To give substantive meaning to relationships,
I convert probit estimates to average predicted probabilities or differences in probabilities
across all observed values of the independent variables.
11
Results are similar when accounting for potential autoregressive errors in state-level least squares estimations where the dependent variable is the proportion of challenged incumbents.
20
2.3
Results
Voters have relatively few opportunities to hold their state legislator accountable. The
probability a state house incumbent receives a challenger is approximately 0.56. Figure
2.2 demonstrates that rates of challenger entry vary across the U.S., and statistical analyses
suggest institutional differences explain why legislators in some states are more likely to
face challengers. Using estimates from the first column of Table 2.1, Figure 2.3 shows the
estimated differences in average predicted probabilities for standard deviation increases in the
institutional or political variable listed on the Y-Axis.12 For example, the top circle indicates
legislators from more professionalized legislatures are more likely to face opposition. When
holding values of other independent variables at their true values, the predicted probability of
a major party challenger increases by over 2% if a legislature’s professionalism increases by
a standard deviation. Otherwise stated, state representatives from the most professionalized
legislature (California) are 9% more likely to receive a challenger than those from the least
professionalized legislature (New Hampshire), all else equal. Whether attracted by higher
salaries or a more desirable office, candidates seem to want to serve in a more professional
setting.
The results also suggest that challengers account for political conditions within the
legislature and district. Competitive legislative chambers, for example, attract more challengers, as summarized by the second circle from the top of Figure 2.3. A 9% increase
in minority party seat share increases the probability of a challenger by approximately
3.5%. This relationship may reflect legislative leaders’ recruitment of candidates to secure
or maintain majorities (Sanbonmatsu 2006) and implies that voters have a greater number of
opportunities to punish or reward the majority party when the legislative chamber margin is
slim.
12
Standard deviations for these variables are: Professionalism (.13); Minority Party Seat Share (.09); Prev.
Inc. Vote Share (.11); District Pres. Vote (.14); Campaign Contributions (1.05); and Q2 State Economy (.02).
21
Figure 2.3: Institutional and Contextual Influences on Challenger Entry
Using estimates from the first column of Table 2.1, the above illustrates the difference in the average
predicted probability of a major party challenger associated with a standard deviation change in
institutional, political, and economic variables, as listed on the Y-Axis. Horizontal black lines are
95% bootstrapped confidence intervals. The bottom two circles suggest that 2% state economic
growth decreases the probability of a major party challenger by 2%, and the influence of the
economy grows in competitive districts.
Challengers also appear to avoid electorally successful incumbents, unfriendly constituencies, and expensive races, as summarized by the third, fourth, and fifth circles of
Figure 2.3. A 11% increase in an incumbent’s previous vote share reduces the likelihood of
a challenger by over 10%. A standard deviation or 14% increase in presidential vote share
for the incumbent’s party within a district reduces the likelihood of a challenger by 5%.
This relationship implies that if districts are drawn with partisan bias there will be fewer
challengers. These types of districts are common in state legislatures (see Kernel density
plot in Figure 2.4). In 2004, Bush or Kerry received at least 60% of the vote in over 55% of
state house districts.
22
Institutional setups appear to influence challenger entry, and professionalizing legislatures or making districts more competitive likely gives voters more chances to hold
incumbent state legislators accountable. However, even extreme institutional arrangements
do not make state legislative elections as competitive as those for Congress. The predicted
probability that a challenger opposes a state legislative incumbent from the most professionalized legislature, where the minority party has 49% of seats, and represents a district where
President Bush received 50% of the vote is still less than that for an average member of the
U.S. House. Only professionalizing legislatures or making more partisan balanced districts,
therefore, may not be sufficient to increase state legislative challenger entry to levels found
in federal elections.
While institutions differ little from one election to the next, a state’s economy can
change dramatically. The statistical analyses in Table 2.1 suggest that challengers react to
these more dynamic conditions within a state. As illustrated by the sixth circle from the
top of Figure 2.3, income growth of about 2% in the second quarter of an election year
reduces the likelihood of a challenger to an incumbent by about 2%. This is consistent
with the hypothesis that candidates strategically enter races to take advantage of perceived
favorable contexts surrounding elections. It also implies that challengers create electoral
incentives for state representatives to stimulate the economy because economic prosperity
helps incumbents secure unopposed reelection.
State legislative candidates emerge during bad economies but their candidacy has little
meaning if they only contest electorally secure incumbents. The solid black line in Figure
2.4 plots the raw data proportions of challenged incumbents who are reelected by district
partisanship. Unsurprisingly, reelection-seeking state representatives are more likely to win
in favorably partisan districts. Incumbent legislators win 99% of the time in districts where
their party received at least two-thirds of the presidential vote but only 87% of the time in
districts where the incumbent’s party received half of the presidential vote (black line). In
these 50-50 districts, 72% of state house members were contested during bad economies
23
Figure 2.4: Incumbent Challenge and Reelection Rates by Partisanship of District
Lines reflect reelection and challenge rates for incumbents across different types of districts. The
kernel density plot (grey region) illustrates the distribution competitive of districts across the country.
Incumbents are overwhelming likely to be reelected (solid line), but they are most likely to receive
challenges in districts where the incumbent’s party receives less that 60% of the presidential vote
share, as indicated by the X-Axis. This rate of challenges is highest during bad state economies
(dotted line) - those with less than 3.0% income growth - as compared to good economies (dashed
line).
(dotted line) but only 65% were during good economies (dashed line). This 7% difference is
eight times as large as the comparable disparity for districts where the incumbent’s party
received two-thirds of the presidential vote and challengers have little chance of winning.
The largest difference in rates of challenger entry during bad and good economies are found
in more competitive districts. The raw data suggests challengers emerge during recessions
and do so more often in districts where they can win.
Statistical analyses provide further suggestive evidence for the patterns illustrated by the
raw data plots in Figure 2.4. The sixth circle of Figure 2.3 denotes the relationship between
income growth and challenger entry for all districts where incumbents sought reelection, and
the seventh circle illustrates the same relationship using data subset to districts where the
incumbent’s party received less than 60% of the presidential vote. The impact of 2% income
growth on challenger entry appears to be a third stronger in these more competitive seats
24
and greater than the impact of a standard deviation decrease in legislative professionalism.13
Challengers, therefore, give voters opportunities to hold their state legislators accountable
for their management of the economy in districts where incumbents face at least some threat
of defeat. If challengers only took advantage of economic conditions to contest electorally
secure incumbents, it would matter little for incumbents’ reelection prospects.
Figure 2.5: Incumbent Challenge Rates by Legislator Policy Responsiveness
Lines reflect challenge rates for incumbents across their different levels of policy responsiveness. The
kernel density plot (grey region) illustrates the distribution of the legislative responsiveness variable.
During bad economies (dotted line), challengers more often oppose unresponsive incumbents.
During good economies (dashed line), there is little relationship between legislators’ responsiveness
and challenge rates in state legislative elections.
Data, furthermore, suggest that the difference in challenge rates between strong and weak
economies concentrates amongst unrepresentative incumbents seeking reelection. Figure 2.5
plots challenge rates by legislators’ policy responsiveness, separating states with below and
above average economies. In states with weaker economies, represented by the increasing
dotted line, legislators who are less responsive more often face a major party opponent. The
relatively flat dashed line, however, implies this is not the case for unresponsive legislators in
13
See Table A.3 in Appendix for probit estimates. For all districts, a standard deviation increase in the
economy variable associates with a 1.9% decrease in the probability of a challenger. In competitive districts,
the comparable change in probability is 2.8%. The differences in differences of these probabilities, however, is
not statistically significant to the .05 level.
25
states with strong economies. Prosperous times, therefore, seem to protect unrepresentative
state legislators from major party competition.14
Figure 2.6: State Economy’s Influence on Challenger Entry
Using estimates from the respective columns of Table 2.1, the above illustrates the difference in
average predicted probability of a major party challenger associated with a standard deviation
increase in state economic growth. Top circle and dotted grey line represent the change in probability
for all incumbents, and other circles represent the probability increase for incumbents who belong to
different political parties (Y-axis). Horizontal black lines are 95% bootstrapped confidence intervals.
Economic growth of 2% decreases the likelihood of a challenger for all incumbents, but with the
same improvement in economic conditions, the governor’s legislative copartisans are 3.1% less
likely to receive a challenger than incumbents who are not members of the governor’s party.
The previous figures repeatedly provide evidence that strong economies shield some
incumbents from challengers, but members of the governor’s party electorally benefit most.
Using estimates from the respective columns of Table 2.1, Figure 2.6 illustrates the impact
of economic growth on the probability of a challenger for all incumbents (top circle) and
members of different parties (other circles as indicated by Y-axis).15 Income growth of 2%
reduces the average probability of a challenger to members of the governor’s party by 3.7%.
Comparable probabilities for legislators unaffiliated with the governor’s party, meanwhile,
14
This increased likelihood of unresponsive opponents receiving challengers during bad economies is
statistically distinguishable from zero (Table A.4).
15
For comparable party-specific probability estimates for institutional and other political variables, see
Figures A.1 and A.2 in the Appendix.
26
fall by less than 1%. This difference supports the hypothesis that challengers account for the
party in the governor’s mansion when hoping to take advantage of economic conditions, but
the analyses suggest there is not a similar relationship the party in control of the state house.
Figure 2.7: Predicted Probabilities of Incumbents Being Contested
Using estimates from the respective columns of Table 2.1, the above illustrates the average predicted
probabilities of a state house incumbent being opposed by a major party challenger. The top circle
and dotted grey line represent the probability for all incumbents, and other circles represent the
probabilities for incumbents who belong to particular political parties, as indicated by the Y-axis.
Horizontal black lines are 95% bootstrapped confidence intervals. Incumbents generally face
challengers 56% of the time, but the president’s state legislative copartisans are 4.5% more likely to
receive a challenger compared to state legislatures unaffiliated with the president’s party.
The economic relationship for the governor’s party parallels that found in congressional
elections. By taking on legislative members of the governor’s party during less prosperous
times, challengers strengthen collective accountability amongst members of state parties.
Statistical analyses, however, suggest candidates account for both state and federal actors. Figure 2.7 presents the average predicted probability for any incumbent receiving a
challenger (top circle) or incumbents of particular political parties, such as the president’s
co-partisans (second circle from top). A member of the president’s party is over 4% more
likely to face a challenger than those unaffiliated with the president’s party, all else equal.
This difference is greater than that for either the governor’s (1.6%) or state house (0.6%)
27
Figure 2.8: Predicted Probabilities of President’s Party & Not President’s Party
Members Being Challenged Adjusting Presidential Approval
Using estimates from the second column of Table 2.1, the above solid line represents the predicted
probability of an incumbent state legislator of the president’s party being challenged under different
levels of presidential approval. Grey regions represents 95% bootstrapped confidence intervals.
Only under popular presidencies is a member of the president’s party less likely to be challenged
than state legislators not affiliated with the President (dashed line).
parties. Members of the president’s party, therefore, are the least electorally secure from
opposition in state legislative elections.
Despite there being little formal connection between the president and state legislators,
the president’s legislative co-partisans are even more likely to be challenged when the
president is unpopular. Figure 2.8 illustrates this relationship by plotting the probability of a
challenge against a member of the president’s party under different levels of presidential
approval. When the president’s approval rating is 35%, the estimated probability of a
member of the president’s party being challenged is .63 (solid line). The comparable
probability for state legislators unaffiliated with the president’s party is only .53 (dashed
line).16
Only during extremely popular presidencies are members of the president’s party as
likely to be challenged as their partisan counterparts, and this relationship likely explains the
16
The disparity in the probability of challenges grows when considering elections since the 1980s using
a reduced model that does not control for district level presidential vote (see top center and right panels of
Figure A.2).
28
changing rates of challenger entry during the more recent Bush administration. Following
September 11th, President Bush enjoyed approval ratings exceeding 70% until the summer
of 2002, and in the November election, approximately 51% of both Democrats and Republicans faced major party competition. However, as Bush became unpopular, there were
consequences for his state legislative co-partisans. In each of the 2006 and 2008 elections,
almost 60% of Republican state legislators faced a Democrat challenger but most Democrat
incumbents went unopposed. National politics, therefore, appears to have a dramatic and
large effect over the opportunities voters have to hold their state representatives electorally
accountable before any elections take place.
2.4
Summary
The above findings imply potential state legislative candidates consider a range of factors
before challenging an incumbent. All else equal, challengers avoid unfriendly districts and
take advantage of weak economies. Together, Figures 2.4 and 2.5 illustrate that political
contexts not only influence trends in candidate entry but also interact in important ways with
challengers’ entry decisions regarding whether to oppose a particular state legislator. By
strategically contesting state representatives who oversaw weak state economies, represent
marginal districts, or are not responsive to constituents’ preferences, challengers’ strategies
promote both individual and collective-partisan accountability in state elections.
Challengers can help strengthen electoral connections, but it is important to understand
the consequences of these candidates’ decisions for accountability and representation. As
described by Jacobson, Congressional challengers who take advantage of favorable national
politics give voters more opportunities to hold the members of federal political parties
collectively accountable for their management of the national government. My analyses
imply state legislative challengers similarly respond to national political conditions, but this
29
does little to promote comparable party accountability at the state-level since state legislators
have little control over national politics.
My analysis of challenger entry sheds new light on how subnational candidates strategically consider both local and national political contexts before taking on incumbents.
The influence of national conditions in state legislative elections not only has implications
for state politics but also for how political scientists use states to study American politics.
State legislatures offer scholars opportunities to test theories of lawmaking beyond the
congressional setting, and electoral connections that are only possible with the threat of a
challenger underlie canonical explanations of legislative behavior (e.g. Mayhew 1974). If
elites in state legislative elections respond to national instead of state politics, how state
politicians achieve their electoral goals should differ from their congressional counterparts,
and consequently voters’ choices and decision-making in state elections will systematically
differ to those in federal contests.
This study of elite-level competition is the most thorough to date in terms of elections
considered but is limited by only examining when major party competition emerges. Legislative incumbents are strategic themselves (Cox and Katz 2002; Engstrom and Monroe
2006), and the economy or the president’s popularity may influence their decision to seek
reelection. Moreover, the general election is not the only place elites promote electoral
accountability. 96 legislators across the country lost their jobs through primary elections
in 2010, and recall elections more recently forced three members of the Wisconsin state
legislature from office. The threat of a primary challenge or a recall potentially strengthens
electoral connections in legislatures, and these institutions provide voters more opportunities
to hold their legislator accountable than those considered here. Future research, therefore,
should build on this and existing work to consider how macro political conditions influence
challenger decision-making beyond the general election context (Grau 1981; Jewell and
Breaux 1991; Hogan 2003).
30
However even across recall, primary, or general elections in 2010, over a thousand
incumbent legislators did not face any opponent. When state representatives like Emile
Bruneau Jr. of Louisiana do not encounter opposition for most of their career, voters
cannot hold their state representative electorally accountable, and if legislators recognize
how national forces influence their elections, they could become less concerned about
potential electoral ramifications for their own policymaking. This may already be the case
in Tennessee. After the 2012 candidate filing deadline, only 40% of Tennesseans approved
of the president’s performance, and at the time, Republican state house Representative Glen
Casada claimed “[t]hat is the biggest thing working for us: President Obama and the antipresident attitude” (Cass 2012). In terms of candidates emerging in Tennessee legislative
elections, Casada was likely right. Democrats chose not to challenge Republicans in 37 of
99 state house districts, and Republicans only had to win 13 of 45 contested elections to
retain their majority in the Tennessee state house. Over a third of Tennessee voters, therefore,
did not have an opportunity to hold Republicans accountable for eliminating the estate-tax,
curbing collective bargaining, or legislative redistricting.
This example from Tennessee is consistent with the main findings of this chapter. State
legislative challengers appear to take advantage of larger political contexts. In particular,
they run against members of the governor’s party when the state economy is weak or
the president’s party when the president is unpopular. The former strategy assumes a
relationship between state-level conditions and voting behavior in state legislative elections.
The next chapter will demonstrate this tactic is probably misguided, and candidates who
take advantage of national conditions likely have the best strategy.
31
3
How Parties Perform in Office &
Elections
The analysis of challenger entry presented in chapter 2 suggests that voters have relatively
few “fair opportunities” to hold their state legislators’ accountable, and these opportunities
vary across elections as challengers appear to respond to both state and national conditions
when deciding to take on an incumbent. During less prosperous times, opposition candidates
provide voters more opportunities to electorally punish those in state government who poorly
managed the state economy. Challengers responding to national conditions, however, does
little to hold state legislators accountable for their own behavior since state representatives
have little control over national politics.
A look at state legislative election outcomes from the late 2000s suggests the that tie
between national and state politics likely goes beyond challengers’ decisions. For example
in 2010, Republicans took control of the U.S. House, but fifteen state legislative chambers
also changed party hands. Democrats’ poor showings in state legislative races followed their
considerable successes in 2006 and 2008. Across these two earlier elections, Democrats
gained state house seats in all but seven states. It, however, seems unlikely that Democratic
32
state legislative parties performed so poorly in 2010 after doing so well such a short time
before. Regardless, hundreds of state legislative Democrats lost their jobs nationwide.
The apparent vulnerability of state legislators to national political tides further brings into
question how much state representatives’ electoral fates have to do with how they perform
in office. When state legislators consider curbing collective bargaining rights, theories
of retrospective voting suggest looming judgments at the ballot box create incentives for
legislators to act in voters’ interests (e.g. Key 1966). The premise of this expectation is that
voters will replace a policymaker who fails to perform, but in low-information elections
such as those for the state legislature, it is difficult for voters to evaluate their legislator’s
performance (Wolfson 1985; Layton 1998).
Party labels and retrospective voting offer simplicity to the accountability process
(Schattschneider 1942; Fiorina 1981). Instead of holding representatives accountable for
their individual behavior - which I address in the next chapter of this dissertation - voters
can assess the parties in power and hold their members collectively accountable. Political
scientists have given considerable attention to studying collective accountability in federal
elections (e.g. Tufte 1975; Jacobson 1989; Jones and McDermott 2004) but scarcely
consider whether the performance of state legislative parties affects electoral outcomes. More
fundamentally, prior work ignores basic requirements of retrospective voting, such as voters
knowing which party is in charge of the state legislature. If voters cannot credibly reward or
punish the incumbent government, it is unclear if canonical theories of retrospective voting
successfully apply to state legislatures.
To assess the extent to which elections provide accountability and create incentives for
state legislative parties and their members to act in the interest of voters, I examine state
legislative elections since the early 1970s. My analyses of election results and surveys
produce very little evidence of accountability. In this chapter, I find state legislative contests
are relatively insensitive to voters’ evaluations of the legislature or measures of state policy
performance, such as economic growth or crime reduction. Election outcomes instead are
33
largely determined by factors outside of state representatives’ control. National conditions
are more influential than state conditions, and presidential rather than state legislative
approval drives outcomes in state legislative elections. Misinformation further complicates
the accountability process as some voters reward the minority instead of the majority party
even when they approve of the legislature’s performance. The behavior of misinformed
voters and the prominent role that evaluations of federal actors play in state legislative
elections cast doubt that these contests effectively hold state legislative parties electorally
accountable for their own performance.
3.1
Requirements for Collective Accountability
Theories of political accountability suggest that those in government should be electorally
punished when they perform poorly in office. Otherwise, incumbents have little incentive
to act in voters’ interests, creating a moral hazard (Ferejohn 1986). In the context of state
legislatures, if a state legislator provides poor representation by supporting undesirable
policies or otherwise failing to perform, voters can hold this representative accountable by
voting against her in the next election. However, less than a fourth of voters know who their
state legislator is yet alone what their individual state representative is doing from day to day,
potentially limiting electoral accountability in state legislatures (Jewell 1982; Vanderbilt
Poll 2012).
To make the accountability process easier, voters can rely on party labels and reward
or punish legislators for their collective behavior. By connecting a party’s performance
in office to its members’ electoral success, this behavior establishes a system of partisan
collective accountability consistent with theories of retrospective voting (Key 1966; Fiorina
1981). If followed, it implies members of the state house majority party need to seek
voters’ approval to avoid being replaced. Retrospection simplifies the accountability process
(Downs 1957), but to hold state legislative parties accountable, voters still need to satisfy
34
Powell’s benchmarks for accountability. They must “know who is responsible for making
policy” and “cast a meaningful vote for or against the policymakers” (Powell 2000: 51).
Figure 3.1: Voter Knowledge of the Partisan Control of Political Institutions
The bars of the graph illustrate the proportion of Correct, Incorrect, and Not Sure responses regarding
which party controls various political institutions, listed on the X-Axis. Sample is registered voters
from the October 2010 ANES Evaluations of Government and Society Study. *‘Not sure’ was not
given as a response option for the Governor’s party identification question.
Powell’s first condition for accountability concerns “clarity of responsibility.” Without
knowing which party is responsible for policy outcomes, it is difficult for voters to correctly attribute retrospective evaluations. For example, divided government (Fiorina 1995;
Nicholson and Segura 1999) and federalism (Downs 1999; Cutler 2004) can obscure who is
responsible for policy and subsequently diminish levels of electoral accountability. In state
legislative elections, failure to have clarity of responsibility is prevalent. Figure 3.1 shows
that while a majority of voters can identify who controls higher levels of federal and state
government, most cannot correctly name which party controls the state house. This brings
into question whether enough voters have “clarity of responsibility” to produce effective
accountability through state legislative elections. It is not clear how an uninformed voter
would use elections to motivate the incumbent government to act in his interest. Misinformed
35
voters could even credit parties for outcomes not attributable to them, potentially punishing
governments that performed well. This misinformed retrospective behavior would weaken,
rather than, strengthen accountability in state legislatures.
Powell additionally requires there be a “meaningful” relationship between votes and
performance, but national tides in state legislative elections cast doubt that this electoral
connection exists. Recall, the striking similarity between the success of state and federal
legislative parties illustrated by Figure 1.1 in the introduction of this dissertation. In all
but four elections in the last hundred years, the party that gained seats in Congress also
made net gains state legislatures. For example despite state legislatures addressing issues
independent of one another, Democrats lost seats in all but three legislatures in 1994, and in
1974, Republicans lost seats in all but five. Health care reforms or Watergate may explain
the outcomes of federal elections in these years, but state legislators had little or nothing
to do with these federal events. State legislative parties, meanwhile, appear to be punished
along with their federal counterparts.
This pattern in election outcomes could reflect at least two types of voter behavior. Voters
may either use state legislative elections to punish political actors outside of the legislature
or heuristically rely on evaluations of party figureheads when making decisions in state
legislative contests. An example of the first type of behavior would be voters using state
legislative elections to repudiate an unpopular governor, similar to how federal midterm
elections are sometimes considered presidential referendums (Tufte 1975). By removing
members of the governor’s party from the state legislature, voters make it more difficult for
a governor to accomplish her legislative agenda and thereby hold her accountable for poor
performance. Voters may also use state legislative elections to express their displeasure with
the White House or punish the President (Piketty 2000; Kellerman 2008). Before modern
polling, presidents commonly used state elections to assess trends in the electorate (Geer
1996). If the president’s party loses state legislative elections, it not only reflects poorly on
this party but can also be detrimental to accomplishing the President’s goals. Following
36
the 2010 elections, President Obama was likely displeased with losing 680 state legislative
co-partisans before congressional redistricting or the implementation of “Obamacare.”
Instead of punishing the governor or president, the pattern in Figure 1.1 may alternatively
be explained by voters relying on assessments of party figureheads when formulating
opinions of more obscure state legislators. Evaluations of political executives are usually
more certain than their assessments of the state legislature. For example, twice as many
respondents to the 2008 Cooperative Congressional Election Study (CCES) were “Not sure”
whether they approved of their state legislature (20.8%) as compared to their governor
(9.8%). With undefined views of the state legislature, voters could heuristically turn to their
more accessible evaluation of the governor for a broader state party evaluation (Tversky
and Kahneman 1974; Kahneman and Frederick 2002). Similarly, someone who knows little
about state politics may rely on evaluations of national party figures. Only 1.6% of CCES
respondents were “Not sure” whether they approved of President Bush, potentially giving
nearly every voter an assessment of Republicans albeit from the federal level.
This dissertation does not disentangle whether voters intentionally use their state legislative votes to reward or punish the governor or president, or simply rely on evaluations
of party figureheads as heuristics to assess less salient elected officials. In either case, the
implication for accountability in state legislatures is the same: state legislators lack strong
incentives to act in their constituents’ interests. Voting against the governor’s party in
state legislative elections may help hold the governor accountable, but state representatives
themselves often have little control over gubernatorial performance. And they have even
less control over the performance of the president. Thus, to the extent that state legislative
elections are driven by presidential politics, they become “second-order” elections analogous
to European Parliament elections, in which votes are cast “on the basis of factors in the
main political arena of the nation” (Reif & Schmitt 1980: 9). Second-order elections are
unlikely to serve what is presumably elections’ first-order purpose - to hold state legislators
accountable for their own performance.
37
The connection between state legislators’ collective performance and electoral outcomes,
however, is unknown. In studies of parties in state legislative elections, the governor is the
center of attention (Bailey and Fullmer 2011; Chubb 1988; Hogan 2005; Lowry, Alt, and
Ferree 1998).1 Similarly, work on federalism finds that national conditions influence regional
elections (Glineau and Blanger 2005; Remmer and Glineau 2003; Rodden and Wibbels
2011) but never considers the party in control of the state legislature in the U.S. system
(Berry, Berkman, and Schneiderman 2000; Campbell 1986; Chubb 1998, see also Besley
and Case 2003). Existing research additionally does not account for voters’ evaluations or
knowledge regarding their state legislature.2
To determine the extent to which theories’ of political accountability and retrospective
voting predictions bear out state legislative elections, I test the hypothesis that the party
in control of the state house gains seats or votes following times of strong state-level
performance or high approval of the state legislature. Evidence for this hypothesis suggests
voters reward legislative parties that perform well and hold party members collectively
accountable. To more systematically investigate the patterns suggested by Figure 1.1, I
also examine the influence of non-legislative actors in state legislative elections. If voters
sanction or heuristically draw on evaluations of prominent party figures in state legislative
elections, I expect the governor’s or president’s co-partisans to lose seats or votes during
times of weak performance and low gubernatorial or presidential approval. This analysis is
the first to consider how the performances of both state legislative and federal actors affect
legislative election outcomes.
1
Lowry, Alt, and Feree (1998) focus on gubernatorial elections but do examine legislative seat change
under divided government. In this related literature, findings are mixed regarding whether governors electorally
benefit from strong state economies (Atkeson and Partin 1995; Carsey and Wright 1998; Ebeid and Rodden
2006; Javian 2011; Stein 1990). Despite political psychology research arguing voters distinguish between
presidential and gubernatorial responsibilities (Arceneaux 2006), Carsey and Wright (1998) and Javian (2011)
discover national forces such as presidential approval influence gubernatorial contests.
2
Some analyses include other assessment measures. Prior to Klarner’s (2010) election forecasts, state
legislative election work including a measure of presidential approval had aggregated dependent variables
such as nationwide control of legislative chambers (Simon, Ostrom, and Marra 1991; Fiorina 1994) or straight
ticket voting (Piereson 1975). Studies with gubernatorial approval employ limited samples and produce mixed
results (Mayo 2004; Bailey and Fullmer 2011; Folke and Snyder 2012).
38
I test these hypotheses using election results and surveys. I investigate how statewide
party seat change in legislatures statistically relate to measures of both economic and noneconomic performance. To complement this macro- or state- level examination of election
outcomes, I use surveys to study the relationship between an individual’s assessment of how
the legislature performed and his voting decision. This micro-level analysis additionally
explores the implications of voters incorrectly identifying which party controls their legislature and how this affects the incentives for the state house majority party to perform well.
Together, these macro and micro approaches provide a fuller understanding of collective
accountability in state legislative elections.
3.2
State-Level Analysis
If elections hold members of the state house majority party collectively accountable,
theories of retrospective voting suggest voters should reward the majority party with more
legislative seats after their policies produce good outcomes. To determine whether this is
the case, I study how various measures of government performance are associated with
party seat changes in state legislatures in the 1972 - 2010 elections. This macro-level
approach resembles that of previous state legislative work (Chubb 1988) but differs in two
key respects. First, I assess the electoral success of the state house majority party in addition
to the governor’s and president’s party to establish whether voters reward legislative parties
independent of their affiliation with the governor or president. My dependent variables
are the proportion change in state house seats for the state house majority, governor’s, or
president’s party with observations at the state-year level (Dubin 2007; National Conference
of State Legislatures).3
3
For example in Texas in 1998 under President Bill Clinton and Governor George W. Bush, the dependent
variable for “President’s Party” is the proportion change in Democratic seats in the Texas state house (-0.02),
but the dependent variable for “Governor’s Party” is the proportion change in Republican seats (0.02). I adjust
control variables, such as change in congressional vote, to be consistent with the dependent variable.
39
My analysis also differs from prior work in the measures of performance I examine.
Previous research only explores the relationship between fiscal or economic variables and
seat change, but I expand the focus to also consider how state governments perform managing
other policy areas. Education expenditures typically are the second largest appropriation in
state budgets, and state governments are responsible for a third of non-federal public safety
spending (Barnett 2011). Thus, I investigate whether voters reward state legislators for a
state’s improved educational performance or reduced crime in addition to examining the
traditional relationship between the economy and election outcomes.
To capture economic performance, the analysis presented below relies on the annual
change in logged real disposable income at the state and national levels, but results are similar
with unemployment or GDP measures (Table B.2).4 For non-economic policy performances,
I use the annual change in a state’s homicide rate or crime index (U.S. Department of
Justice) to measure state crime prevention, and for education, I rely on the changes in a
state’s average SAT score or the National Assessment of Educational Progress reading exam
(National Center for Education Statistics).5 If voters hold legislators accountable for a state’s
performance in economic or non-economic policy areas, I expect these measures to correlate
with seat change for the majority party.
To investigate the extent to which the performance of governors influence state legislative
elections, I conduct similar economic, crime, and education analyses for the governor’s party
in the legislature. To assess presidential influences, I examine seat change for the president’s
state legislative party both as a function of economic performance and presidential approval.
Findings do not change when the dependent variable is Democratic party seat change and independent
variables are economic measures interacted with the president’s, governor’s and state house majority party
(Table B.1). Results are also similar when examining state senate elections.
4
I convert disposable income data from the Bureau of Economic Analysis to 2010 real dollars using
Consumer Price Index factors and estimate the annual change. Each state in a specific year receives the same
measure of “Change in Logged National RDI,” and each observation receives a specific state-year “Change
in Logged State RDI” measure. Substantive results do not change when including lagged performance from
previous years.
5
Crime index, SAT, and NAEP data respectively cover the 1972 - 2008, 1988 - 2010, and 1998 - 2010
elections. I recenter SAT scores before 1995 to maintain comparability to contemporary scores. I thank Jason
Grissom and Alex Bolton for pointing me to the education measures.
40
For presidential approval, I rely on the last national Gallup poll before the November election.
This variable ranges from 0 to 1 and assumes uniformity in presidential approval across
states.
I estimate the relationship between party seat changes and performance measures using
Ordinary Least Squares regressions. Each estimation includes state fixed effects and two
control variables. To account for coattail effects or surges in partisan turnout, I control for
changes in a state’s congressional vote. Unlike presidential, gubernatorial, or senatorial vote,
this measure is consistently available in two year intervals as U.S. House elections closely
follow most state house election calendars. To control for a party’s electoral exposure, I
also include a state legislative party’s seat change in the previous election (Oppenheimer,
Stimson, and Waterman 1986).6
Theories of retrospective voting predict that electorates will punish or reward those
in government for how they perform, but the statistical analyses reported in Table 3.1
provide little evidence that this is the case in state legislative elections. The first three
columns of this table show the relationships between measures of economic performance
and changes in seats held by state house majority party. During prosperous state-level
economies, voters do not reward the party in control of the state house. The relationship
between changes in national real disposable income and state house majority party seats
likewise is insignificant and in the unexpected direction. Analyses in the third column of
Table 3.1, furthermore, suggest that members of majority party are not rewarded when their
state’s economy performs well relative to the national economy. These findings persist
under conditions where one might expect to see greater levels of collective accountability in
state legislatures, such as unified state governments, professional legislatures, or midterm
elections (Tables B.3, B.4, & B.5). Voters, therefore, do not seem to retrospectively reward
6
The congressional vote control is not available for state legislative elections that occur in the “off-year.”
New Jersey, Virginia, Mississippi, Louisiana, and sometimes Kentucky, therefore, are excluded from the main
analysis. Substantive results do not change when dropping this control and including these states. Results
additionally are not sensitive to excluding the exposure control variable, a particular state or year, state fixed
effects, or alternatively using year fixed effects and clustering standard errors by state.
41
Table 3.1: State House Majority Party Seat Change as a Function of Performance
Measures
Performance Measure:
Change in Logged State RDI
Econ.
-0.022
(0.111)
Change in Logged National RDI
Econ.
Econ.
Crime
Crime
Educ.
Educ.
-0.147
(0.146)
State Economy Relative to National Economy
0.077
(0.154)
Change in Homicide Rate
-0.002
(0.002)
Change in Logged State Crime Index
-0.027
(0.046)
Change in Logged Average State SAT Score
-0.042
(0.103)
Change in Logged Average State Reading Score
Previous Seat Change
-0.184*
(0.077)
0.310*
(0.035)
-0.070
(0.021)
0.222
867
Congressional Vote Change
Constant
R-Squared
N
∗p
-0.184*
(0.077)
0.311*
(0.035)
-0.068
(0.021)
0.221
867
-0.182*
(0.077)
0.310*
(0.035)
-0.071
(0.022)
0.220
867
-0.184*
(0.077)
0.311*
(0.035)
-0.071
(0.022)
0.220
867
-0.184*
(0.078)
0.278*
(0.036)
0.005
(0.003)
0.197
822
-0.187*
(0.126)
0.321*
(0.036)
-0.032
(0.020)
0.278
524
-0.386
(0.610)
0.122
(0.096)
0.370*
(0.069)
-0.094
(0.043)
0.487
198
≤ .05; Robust Standard Errors in Parentheses
OLS estimates of state house majority party seat change regressed on economic and non-economic
performance measures. Data for economy regressions cover 1972 - 2010 elections, and non-economic
regressions cover more recent elections as discussed in the main text. Observations are at the stateyear level and estimations include state fixed effects. No performance measure is statistically related
to seat change for the party in control of the state house.
state legislative parties for economic prosperity, providing these parties little electoral
incentive to produce polices that stimulate the state economy.
Voters also do not appear to hold legislative parties accountable for a state’s performance
in non-economic policy areas. The statistical analyses presented in the fourth and fifth
columns of Table 3.1 imply there is little relationship between changes in a state’s homicide
rate or crime index and seats won by the state house majority party. Similarly, there is no
statistical association between students’ improvement on standardized tests and legislative
seat change, as shown by the final two columns of Table 3.1. Therefore even if a state
legislature successfully implements effective education policies, the party in power receives
little electoral reward.
Findings are similar when examining seat change for the governor’s party in state legislative elections. Table 3.2 shows the relationships between objective state-level measures
42
Table 3.2: Governor’s Party Seat Change as a Function of Performance Measures
Performance Measure:
Change in Logged State RDI
Econ.
-0.051
(0.101)
Change in Logged National RDI
Econ.
Econ.
Crime
Crime
Educ.
Educ.
0.013
(0.144)
State Economy Relative to National Economy
-0.103
(0.150)
Change in Homicide Rate
0.007*
(0.002)
Change in Logged in State Crime Index
0.003
(0.045)
Change in Logged Average State SAT Score
-0.096
(0.080)
Change in Logged Average State Reading Score
Previous Seat Change
-0.244*
(0.059)
0.299*
(0.044)
0.009
(0.012)
0.195
861
Congressional Vote Change
Constant
R-Squared
N
∗p
-0.245*
(0.059)
0.298*
(0.044)
0.008
(0.013)
0.195
861
-0.244*
(0.060)
0.300*
(0.044)
0.008
(0.012)
0.196
861
-0.244*
(0.063)
0.300*
(0.040)
0.015
(0.016)
0.203
861
-0.243*
(0.061)
0.264*
(0.043)
0.009
(0.013)
0.177
816
-0.253
(0.132)
0.317*
(0.059)
-0.013
(0.011)
0.259
518
-0.450
(0.622)
0.060
(0.137)
0.392*
(0.117)
-0.004
(0.008)
0.367
197
≤ .05; Robust Standard Errors in Parentheses
OLS estimates of governor’s party seat change regressed on economic and non-economic performance
measures. Data for economy regressions cover 1972 - 2010 elections, and non-economic regressions
cover more recent elections as discussed in the main text. Observations are at the state-year level
and estimations include state fixed effects.
of government performance and changes in seats held by governor’s party in the state house.
The governor’s party does not systematically gain seats in the state house when the state
economy prospers, crime falls, or test scores rise. State legislative elections, therefore, do
not seem to be referendum on the state executive and appear ineffective at holding state
parties accountable for both economic and non-economic policymaking.
Whereas state parties’ performances seem to have little electoral relevance, national
conditions are a driving force in state legislative elections. The statistical analyses offered
by Table 3.3 provide compelling evidence that voters collectively reward members of the
president’s party in state legislative elections when the president is popular or the national
economy is strong. Results presented in the second column of this table suggest national
income growth of 2% approximately associates with a 1.8% change in seats, accounting for
over a third of the average party seat swing in state legislative contests. The comparison of
economic estimates from the second and third columns of Table 3.3 demonstrate that similar
43
Table 3.3: President’s Party Seat Change as a Function of Presidential Approval and
Economic Performance
Performance Measure:
Presidential Approval
Pres. Approval
0.144*
(0.025)
Change in Logged National RDI
National Economy
0.934*
(0.121)
Change in Logged State RDI
Previous Seat Change
Congressional Vote Change
Constant
R-Squared
N
∗p
State Economy
-0.327*
(0.077)
0.225*
(0.037)
-0.110*
(0.036)
0.248
867
-0.312*
(0.077)
0.220*
(0.034)
-0.054
(0.035)
0.267
867
0.406*
(0.092)
-0.310*
(0.079)
0.250*
(0.035)
-0.043
(0.034)
0.240
867
All
0.073*
(0.028)
0.750*
(0.192)
0.005
(0.126)
-0.322*
(0.077)
0.208*
(0.036)
-0.088*
(0.037)
0.273
867
≤ .05; Robust Standard Errors in Parentheses
OLS estimates of the proportion change in state house seats for the president’s party regressed on
economic performance and presidential approval measures. Data cover the 1972 - 2010 elections.
Observations are at the state-year level and estimations include state fixed effects.
to other federal systems national rather than state economic conditions are more influential
in regional legislative elections (Glineau and Blanger 2005; Remmer and Glineau 2003).
A similar relationship emerges when using a more direct measure of voters’ evaluations
of presidential performance. A 10% increase in presidential approval corresponds to a 1.4%
change in state house seats. The association between seat change in state legislatures and
presidential approval is robust to institutional contexts such as divided government and is
not isolated to a few states. When regressing president’s party seat change on presidential
approval in separate OLS estimations by state, positive relationships emerge in over forty
states (Figure B.1). Even when controlling for economic conditions, presidential approval
correlates with state legislative election outcomes, as shown by the final column of Table
3.3.
The statistical analyses presented in Table 3.3 suggest that events entirely beyond the
control of state representatives, such as presidential scandals, influence the outcomes of state
legislative elections (Kernell 1978). Meanwhile, the results presented in Table 3.1 imply
that effective state legislative policymaking produces little in the way of electoral rewards.
44
Together, these analyses suggest that state legislative elections are unlikely to produce the
sort of “electoral connection” and curb on moral hazards envisioned in theories of political
accountability.
3.3
Individual-Level Analysis
The state-level analysis’ findings suggest that state legislative elections do not hold the
state house majority party accountable for its performance but instead appear to be largely
national affairs. This interpretation presumes that economic performance shapes evaluations
of elected officials and voters know who to reward or punish. However, objective economic
performance measures do not always translate into subjective assessments of government
(De Boef and Kellstedt 2004) and not all voters know which party controls the legislature
(Figure 3.1).
To conduct an analysis that relaxes these assumptions, I examine surveys that asked
registered voters about their state legislatures. I first investigate whether individuals report
voting for the state house majority party when they believe the legislature is doing a good
job, as predicted by theories of political accountability. This largely replicates the state-level
study at the individual-level but without relying on objective measures of performance or
aggregated election outcomes. The second part of the individual-level analysis accounts for
levels of voter information and investigates whether voters punish or reward the party they
believe controls the legislature. This addresses the extent to which voters try to hold state
legislative parties accountable even if they lack information about who is in charge.
These investigations employ two sets of surveys. To study elections across the country,
I rely on the 2008 and 2010 CCES. YouGov Polimetrix conducted these online surveys in
two waves, interviewing the same respondents in October and November. In the first wave,
individuals were asked who controlled their state legislative chambers and whether they
approved of the state legislature. In the second wave, respondents stated how they voted in
45
their state legislative elections. To complement these recent nationwide surveys and examine
elections since the 1970s, I use New Jersey state polls conducted by the Eagleton Institute of
Politics. New Jersey state elections occur in the “off-year” separate from federal elections,
and voters’ decisions presumably should be less sensitive to federal influences.
In these analyses of surveys, the dependent variable of interest is state house vote choice.
For the CCES surveys, I code respondents’ Republican and Democrat reported vote choices
respectively 0 and 1.7 I estimate how vote choice varies as a function of voters’ approval
ratings of their state legislature, governor, and president while controlling for a respondent’s
party identification.8 On the CCES, voters indicated their approval rating of these political
actors on a five-point scale ranging from “Strongly disapprove” to “Strongly approve,” which
I code from -2 to 2. To maintain consistency with the dependent variable, positive values
denote approving of a Democrat state legislature. For example, if a respondent strongly
approved of a legislature with a Democrat state house majority party, “State Legislative
Approval” receives a value of 2, but if Republicans held the majority, this variable receives a
value of -2. I create similar variables for gubernatorial and presidential approval. Substantive
findings are similar when either using dummy variables for approval levels instead of a
cardinal measure or substituting voters’ assessments of the economy for their approval
ratings of political actors.
Similar to the state-level analysis, this research design presumes that individuals know
which party to retrospectively reward for good performance. Most voters, however, cannot
identify the party controlling their legislature and therefore lack clarity of responsibility.
Misinformed voters may still act retrospectively and try to hold legislative parties account7
2008 and 2010 surveys asked, “For whom did you vote for in the state legislative elections” in the
respondent’s lower chamber. In 2008, individuals could select a “Not Sure” response, but in 2010, this option
was unavailable. To simplify presentation, I focus on registered voters who gave a definitive Democrat or
Republican response. Findings are similar when including “Not Sure” responses in a multinomial probit
estimation (Table B.7)
8
The seven-point party ID measure provided by the CCES does not account for possible dual-party
identification (Jennings and Niemi 1966) Most voters do not have dual-party identifications. The considerable
variation in state and national parties’ ideology, however, may limit partisanship’s effectiveness in voting
decisions (Shor and McCarty 2011).
46
able, but they could punish or reward the wrong party. In the context of the state-level
analysis, an incumbent legislative party that effectively manages the economy or education
policy then may not receive a full electoral reward for their performance. Put differently,
voters may try to use elections to sanction but fail to do so due to misinformation, and the
null relationships between seat change and the performance measures in Table 3.1 could be
attributable to voters lacking clarity of responsibility.
To better understand the implications of low levels of voter information in state legislative
elections, I additionally examine whether voters punish or reward the legislative parties they
believe are in power. This second component of the individual-level analysis resembles the
first but takes advantage of a political knowledge question in the CCES survey. The survey
asked respondents which party they thought controlled the state house. With this variable, I
adjust a voter’s approval rating to account for which party they believed was in control of the
state house instead of which party actually controlled the state house. For example if a voter
strongly approved of their legislature’s performance and believed Democrats controlled the
state house, “State Legislative Approval (Belief)” receives a value of 2. If the same voter
instead thought Republicans controlled the state house, this variable then receives a value of
-2. Coefficients on these measures shed light on whether voters try to hold state legislative
parties accountable. I perform similar knowledge adjustments for gubernatorial approval.
Due to the lack of a presidential party knowledge question, I assume voters knew Bush in
2008 was a Republican and Obama in 2010 was a Democrat.
Before proceeding to statistical results, potential limitations of the individual-level
analysis deserve some attention. The CCES surveys allow me to examine the relationship
between voters’ legislative evaluations and their electoral decisions, but the nationwide
CCES samples are wealthier, better educated, and more politically interested than the general
population (Table B.6). They also more often identify the parties in power and therefore may
be better able to hold state legislators collectively accountable. CCES respondents, however,
are still less likely to correctly identify the state house majority party than the governor’s
47
or federal legislative parties. To account for some of these differences, I employ sample
weights provided by the CCES that account for age, education, gender, race, and turnout.9
To examine the relationship between a voter’s evaluations of political actors, party identification, and vote choice, I use a weighted probit analysis. Positive relationships between
state legislative approval and vote choice serve as evidence that voters who approve of their
state legislature reward candidates of the incumbent state house majority party and suggest
theories of retrospective voting apply to state legislatures. Positive coefficients on gubernatorial or presidential approval provide evidence that how the governor or president performs
influences legislative elections. To simplify interpretations, I convert probit estimates to
predicted probabilities in text and figures. For differences in predicted probabilities, I adjust
the variable of interest and hold other variables at their weighted sample means.
The CCES surveys provide some evidence that elections hold state legislators collectively
accountable. However, the strength of the connection between voters’ assessments of the
state legislature and their state house voting decisions is relatively weak, as shown by the
first and third columns of Table 3.4. When voters strongly approve of their state legislature
instead of strongly disapprove, the probability that they vote for a candidate of the state house
majority party increases by up to .12. This relationship implies that there are incentives for
legislative parties to act in their constituents’ interests to better their members’ reelection
chances and is the first known evidence that state legislative elections appear to provide
some accountability.
9
Measurement assumptions further limit the analysis. I assume approving of a Democratic state legislature
is the same as disapproving of a Republican state legislature. To investigate whether results are sensitive to
this assumption, I estimate the models with unadjusted approval variables on data subset by who respondents
thought controlled the governorship and state house. Conclusions do not change. Since the state legislative
approval question does not specifically ask about the state house or senate, I assume a respondent’s approval
rating is the same across both chambers. Conclusions do not change when data is subset to those who thought
the same party controlled both legislative chambers, which makes this assumption less concerning unless
respondents have different opinions regarding state senate and state house Democrats. There are a considerable
number of “Not sure” responses to the governor and state legislative approval and knowledge questions,
and I code these responses as a middle category to reflect uncertainty regarding whether the respondent
disapproves or approves of these political actors. “Not sure” respondents may have answered correctly if given
different closed item responses (Mondak 1999: 72). Main conclusions do not change when omitting “Not sure”
respondents (Table B.8).
48
Table 3.4: State House Vote Choice as a Function of Approval Ratings and Party ID
Election Year:
State Legislative Approval
2008
0.075*
(0.015)
State Legislative Approval (Belief)
Governor Approval
Party ID (7 pt)
Constant
Log-pseudolikelihood
N
∗p
2010
0.059*
(0.016)
0.110*
(0.018)
0.132*
(0.013)
Governor Approval (Belief)
Presidential Approval
2008
0.347*
(0.016)
0.507*
(0.010)
-0.277*
(0.027)
-4928.1
18182
2010
0.108*
(0.018)
0.105*
(0.014)
0.131*
(0.014)
0.342*
(0.016)
0.502*
(0.010)
-0.270*
(0.027)
-4922.9
18182
0.461*
(0.016)
0.472*
(0.014)
0.075*
(0.026)
-5200.3
27769
0.115*
(0.015)
0.453*
(0.016)
0.468*
(0.014)
0.077*
(0.024)
-5164.8
27769
≤ .05; Standard Errors in Parentheses
Probit estimates of state house vote choice as a function of voters’ assessments of political actors and
partisan identification. The first and third columns adjust state legislative approval by the party that
actually controlled the state house, and the second and fourth columns adjust approval by the party
a respondent believed controlled the state house. These data from the Cooperative Congressional
Election Studies are weighted to make them representative of the registered voters in the 2008 and
2010 elections.
State-level actors outside the legislature, such as the governor, also seem influential in
legislative elections. Strongly approving instead of strongly disapproving of the governor
changes the predicted probability of a state house vote by at least .18. Punishing an unpopular
governor’s legislative party can stall the governor’s legislative agenda, and the relationship
between vote choice and gubernatorial approval could reflect this tactic by voters.
Assessments of state-level actors play some role in state legislative elections, but findings
presented in Table 3.4 reaffirm that national politics matter much more. Shifts in presidential popularity from strongly disapproving to strongly approving can change predicted
probabilities of voting for the president’s co-partisans by at least .38. The relative impact
of presidential approval compared to state legislative approval is remarkable. Figure 3.2
summarizes predicted probabilities of voting for candidates of the state house majority or
president’s party using estimates from Table 3.4. In each panel, solid lines represent the probability of voting for the state house majority party under different levels of state legislative
approval, and dotted lines plot the probabilities of voting for a legislative candidate of the
49
Figure 3.2: Voter Behavior under different levels of State Legislative or Presidential
Approval
Comparisons of the relationships between an individual’s assessments of the state legislature or the
president and their state house voting decisions in the 2008 and 2010 elections. Solid lines represent
the predicted probability of voting for a candidate of the state house majority party under different
levels of state legislative approval, and dashed lines represent the probability of voting for a member
of the president’s party under different levels of presidential approval. Other variables are set to
their weighted sample means, and grey regions are 95% confidence intervals. The relative influence
of presidential approval is at least three times that of state legislative approval.
50
president’s party for given levels of presidential approval. With growing approval, predicted
probabilities of voting for these parties’ candidates increase, but changes in presidential
approval have at least three times the impact of comparable changes shifts in state legislative
approval. This relationship between presidential approval and state legislative vote choice
is robust. Levels of voter knowledge or divided state government have no attenuating
effect, and the relationship persists amongst wealthy, educated, or politically interested
voters (Table B.9). The correlation between state legislative vote choice and presidential
approval consistently emerges when estimating the model on data subset by state (Figure
B.2). Therefore in state legislative elections across the country, changes in presidential
approval appear to matter much more than shifts in state legislative approval even though
legislative parties control the legislature’s performance more than the president’s.
The findings from the 2008 and 2010 CCES provide persuasive evidence that national
politics influence state legislative elections. These analyses, however, only examine recent
state elections that coincide with federal contests. Some state elections, such as those in New
Jersey or Virginia, occur in the “off-year” (e.g. 2009 or 2011) separate from presidential or
congressional elections. When advocating off-year elections, New Jersey Governor Alfred
Driscoll asserted “the election for a Governor and for Assemblymen should not coincide with
a Presidential election. The importance of a gubernatorial election merits an election that
will not be overshadowed by a national contest for the Presidency” (New Jersey Constitution
Convention Proceedings 1947).10 While the focus of this study is whether assemblymen
are held accountable rather than governors, Driscoll’s overarching point regarding state
elections still applies. By being held separate from federal contests, off-year elections
should be less likely to be “overshadowed,” and New Jersey state legislative elections
provide an opportunity to test the robustness of the presidential approval findings under
electoral conditions presumably less sensitive to national politics.
10
This quote was found thanks to Bishop and Hatch (2012).
51
52
NJ-1973
0.216*
(0.060)
0.015
(0.068)
0.005
(0.078)
0.746*
(0.055)
-0.232*
(0.099)
0.141
(0.098)
-211.444
446
∗p
NJ-1979
0.252*
(0.074)
0.088
(0.067)
0.08
(0.069)
0.902*
(0.056)
-0.292*
(0.106)
-0.215*
(0.105)
-220.603
638
NJ-1983
0.210*
(0.064)
0.094
(0.066)
-0.033
(0.075)
0.711*
(0.059)
-0.048
(0.088)
0.109
(0.088)
-192.25
415
NJ-1985
0.323*
(0.082)
-0.012
(0.102)
0.151
(0.099)
0.860*
(0.084)
-0.052
(0.134)
0.043
(0.134)
-92.577
323
NJ-1987
0.179*
(0.059)
0.146*
(0.068)
0.022
(0.069)
0.665*
(0.053)
-0.098
(0.095)
-0.057
(0.095)
-195.878
509
≤ .05; Standard Errors in Parentheses
NJ-1975
0.218*
(0.050)
0.202*
(0.064)
0.013
(0.063)
0.707*
(0.050)
-0.498*
(0.082)
-0.262*
(0.080)
-353.944
654
NJ-1995
0.423*
(0.078)
0.228*
(0.085)
-0.017
(0.095)
0.758*
(0.069)
-0.267*
(0.103)
-0.182
(0.102)
-127.763
461
NJ-2007
0.228*
(0.063)
0.074
(0.055)
0.092
(0.058)
0.684*
(0.051)
-0.051
(0.094)
0.282*
(0.096)
-251.3
523
VA-2007
0.296*
(0.023)
0.180*
(0.043)
0.186*
(0.039)
0.613*
(0.043)
0.185*
(0.070)
0.573*
(0.072)
-452.272
1052
Ordered probit estimates of state house vote choice as a function of voters’ assessments of political actors and partisan identification. Column headings
indicate the state and year of the poll. The Eagleton Institute of Politics conducted the New Jersey polls, and The Washington Post conducted the
Virginia poll. In these off-year election states, presidential approval consistently correlates with state house vote choice.
Log-Likelihood
N
Intercept: Split Votes | D Votes
Intercept: R Votes | Split Votes
Party ID
State Legislative Approval
Governor Approval
Election Year:
Presidential Approval
Table 3.5: NJ and VA Off-Year State Legislative Voting as a Function of Approval Ratings and Party ID since the 1970s
I, therefore, examine New Jersey voters’ state legislative voting behavior using polls from
the Eagleton Institute of Politics. This investigation extends the individual-level analysis in
two respects that further test the robustness my findings regarding the impact of national
conditions. First, it analyzes elections that occur in the off-year. Second, it examines polls
from each of five presidential administrations since the 1970s instead of only more recent
elections. Similar to the CCES analysis, I estimate the relationship between vote choice and
a voter’s approval rating of the state legislature, governor, and president while controlling
for an individual’s party identification. To account for New Jersey’s multi-member districts
and options to vote for two Democrats, split the ticket, or vote for two Republicans, I
estimate this relationship with an ordered probit regression. Over the last forty years, the
Eagleton Institute had inconsistent question wordings for vote choice, approval, and party
identification questions. To maintain comparability to CCES estimates in Table 3.5, I code
response categories similar to the CCES analysis. The Appendix details coding and provides
results from an alternative model specification without these adjustments (Table B.10).
Results are similar.
Table 3.5 presents evidence that presidential influences in state legislative elections are
not solely a result of federal election coattails nor a recent phenomenon. In each Eagleton
poll, approving instead of disapproving of the president can change the probability of a state
legislative vote for the president’s party by at least .27. While gubernatorial politics matter
more in some elections than others, assessments of the New Jersey state legislature never
have a meaningful relationship with vote choice. The final column of Table 3.5 indicates
these off-year election findings are not confined to New Jersey, as national influences have
similar effects in Virginia legislative elections, which also occur in odd-numbered years.11
In spite of weaker results from off-year states, recent CCES surveys provide some
evidence that theories of retrospective voting apply to state legislatures. Misinformation,
11
Virginia results use a 2007 Washington Post Poll. Instead of a vote choice question, this survey asked
“Regardless of your local contest, which party would you like to see in control of the Virginia state legislature
after the November elections, the (Democrats) or the (Republicans)?” I code “Divided” responses as the
middle category.
53
however, may limit the interpretations of these findings and those from the state-level analysis
because voters do not always know who controls their state legislature. The statistical
analyses in the second and fourth columns of Table 3.4 relax informational assumptions
by adjusting approval ratings to account for the party a voter believed controlled the state
house or governorship. Compared to the first and third columns, nearly all estimates are
relatively stable, but the relationship between state legislative approval and vote choice
strengthens. Voters, therefore, appear to try to hold state legislators collectively accountable
but sometimes punish the party that actually is in power when they intend to reward them.
Figure 3.3 illustrates the magnitude of this effect using the 2010 elections. To plot
the predicted probabilities in this figure, I classify voters by whether they correctly or
incorrectly identified their state house majority party, and within each of these subsets,
I regress state house vote choice on the party identification and approval variables. The
estimation using the correct respondents produces a positive relationship between approval
of the legislature and voting for candidates of the state house majority party (dotted line),
but the corresponding relationship among incorrect respondents is negative (dashed line).12
In other words, when the state legislature performs well, voters who identify the state house
majority party electorally reward the party in power, but misinformed voters punish the
incumbent party. These counteracting votes contribute to the relatively weak aggregate
relationships between state legislative approval and vote choice displayed in Figure 3.2.
Misinformed voting behavior also helps explain the null results from the state-level
analysis. Some voters could recognize that the economy is weak or the state government’s
policies are ineffective but not know which party to blame. These voters may try to hold
legislative parties accountable by voting for or against the party they believe to be in charge,
but their mistaken attribution ultimately weakens the relationship between performance
and state house majority party seat change. This misinformed behavior then diminishes
state legislative elections’ effectiveness as an accountability mechanism and reduces the
12
There is no meaningful relationship between state legislative approval and vote choice amongst those who
were “Not sure” which party controlled their state house.
54
Figure 3.3: Voter Behavior of Informed and Misinformed Voters
Lines represent the predicted probabilities of an individual voting for the state house majority
party, and grey regions reflect 95% confidence intervals. Dotted and dashed lines use estimates
from separate regressions using samples of voters subset by whether they correctly or incorrectly
identified the state house majority party. By voting against members of the state house majority party,
even when they approve of the legislature’s performance, misinformed voters reduce the incentives
for incumbent state legislative parties to perform well.
incentives for legislative parties to perform well. While informed voters more often hold
the poorly performing state house majority party accountable by voting for members of the
opposition minority party, the dotted line in Figure 3.3 illustrates that the connection between
informed voters’ assessments of the state legislature and state house voting decisions is
weak, even amongst these informed voters.
3.4
Summary
Theories of retrospective voting and accountability suggest there should be an electoral
connection between how state legislators perform in office and elections, but it appears that
legislators fearing electoral retribution need to worry more about the president’s performance
55
than their own. Carsey and Wright argue that “national forces working through evaluations
of the president are a major influence on voting for governor” (1998: 1001). The above
findings imply that these forces are strong enough to reach state legislative chambers. State
representatives’ behavior and performance may matter at the margins, but evaluations of the
president more likely determine whether legislators are reelected.
The weak or null relationships between state legislative performance and electoral outcomes again cast doubt on whether state legislative elections effectively sanction legislative
parties. Retrospective party voting may not be the only way to hold state legislators accountable, but without evidence that how state parties perform in the legislatures matters in
elections, it brings into question some interpretations of election outcomes. For example
following the 2010 election, the Tennessee Senate majority leader claimed being returned
to power gave Republicans the “mandate here to lead and to govern” (Locker 2010). The
foundation of this mandate, however, is unclear. Most Tennesseans did not know Republicans already controlled the state’s Senate, and of those who did, most disapproved of the
state legislature’s performance (CCES 2010). The findings of this chapter suggest that the
Republican gains in the Tennessee state legislature should not necessarily be interpreted as a
mandate for a Republican state legislative party but perhaps instead as a referendum on a
Democratic President.
National politics influencing state legislative elections may ultimately lead to diminished
representation and unintended policy consequences. Without electoral accountability, there
is little that constrains lawmaking at the state level. Legislative parties can claim mandates
and adopt controversial collective bargaining policies or other laws with little electoral
consequence. More research is needed to fully understand the above findings’ implications
for representation. State legislators may still “run scared,” act in their constituents’ interests,
and produce representative policies (Erikson, Wright, and McIver 1994). The evidence,
however, strongly suggests that there is not much electoral reason for them to do so, as
elections do relatively little to hold legislators and their parties collectively accountable.
56
4
Accountability for State Legislative
Roll-Calls & Ideological Representation
The previous chapters provide little evidence that challengers’ or voters’ decisions are
sensitive to parties’ activities in the legislature. Despite being discouraging for the prospects
of accountability, these findings do not necessarily imply that electoral connections fail to
exist in state legislatures. Party labels make it easier to hold party members responsible
for their collective performance, but they may also oversimplify the accountability process.
Voters do not elect parties but individual representatives, and once in office, legislators’
primary responsibility is to represent their constituents’ - not their party’s - interests in the
state house.
Recognizing the dyadic nature of their representative duties, state legislators sometimes
break the party line. For example when the Maine state legislature legalized same-sex
marriage, Democratic Representative Lajoie voted with his district - where 73% of voters
opposed the legislation - instead of with his party. The bill passed, and Lajoie’s nay vote
put him in the minority in more ways than one. Unlike Lajoie, a majority of Maine state
house members voted against their district’s opinion. Electoral pressures intend to prevent
this type of unrepresentative behavior, but seemingly counter to the expectations of spatial
57
theories of elections (e.g. Downs 1957), most Maine legislators who voted against their
districts were reelected.
The electoral success of Maine legislators who cast unrepresentative votes highlights the
question of whether districts hold their legislators accountable for their individual behavior.
Political scientists, however, know little regarding district-level electoral connections in the
states. Studies of American elections overwhelming focus on the electoral implications of
congressional representation (e.g. Canes-Wrone et. al 2002; Carson et. al 2010, Jacobson
1993) despite state legislatures’ responsibility for a considerable portion of the lawmaking
in this country. As discussed in the introduction of this dissertation, state legislatures pass
75 laws for every law adopted by Congress, but the most extensive studies of the electoral
implications of individual state legislator’s behavior only examine two elections in less
than a third of states (Hogan 2004; 2008).1 Hogan’s work tells us the most we know about
district-level accountability in state legislatures, but limited data availability has prevented
political scientists from conducting more thorough examinations.
To better evaluate electoral connections in state legislatures, I assemble the largest
cross-state collection of measures of legislators’ and voters’ preferences and behavior at
the state legislative district-level. With these data I evaluate the electoral consequences
legislators face for casting unpopular roll-call votes and the extent to which ideologically
extreme legislators receive lower vote shares over the past decade. Together, these studies
arguably provide the most empirically thorough examination of state legislators’ electoral
incentives to represent their districts to date.
Neither analysis produces compelling evidence of accountability in state legislatures.
Across ten states, I find that voters reward or punish legislators for 4 of 23 examined rollcalls, and only 9 of 38 states produce statistically significant evidence that legislators pay
an electoral price for ideologically extreme representation. Weak electoral connections
1
Serra and Pinney (2004) examine the relationship between legislator casework and election returns, and in
unpublished work, Birkhead (2013) studies the relationship between legislator extremity and vote-share in the
2000 election.
58
combined with safe state legislative seats make it difficult to throw “out of step” legislators
out of office, thereby casting doubt on the proposition that state legislative elections create
incentives for representation.
4.1
Holding “Out of Step” Legislators Accountable
A central concern of democratic theory is ensuring policymakers represent their constituents, and theories of elections predict that competition for votes pressures lawmakers
to adopt policies consistent with their constituents’ preferences (Black 1958; Downs 1957;
Hotelling 1929). Spatial models of electoral competition often assume there is a connection between policymakers’ and voters’ behavior (see Grofman 2004 for a review). If the
predictions of these theories apply to district-level representation in American legislatures,
elections are a potential solution to the moral hazard problem posed by representative
government (Ferejohn 1986).
An intent of elections is to create incentives for those in government to represent their
constituents’ interests, but legislators face more than electoral pressures. Party obligations,
interest groups, and personal policy positions also weigh upon legislators’ decision-making
(Fenno 1978; Kingdon 1989; Hall and Wayman 1990). These cross-pressures are partly
responsible for the findings that legislators do not always provide ideological representation
(e.g. Miller and Stokes 1963; Shaprio et. al 1990; Bafumi and Herron 2010). States have
U.S. senators of different parties, and Figure 4.1 illustrates incongruence between voters’
and their representatives’ ideology in the U.S. House and state houses across the country.
Using ideal point estimates from 2004, the top left panel of this figure shows conservative
members of the U.S. House generally represent districts that supported Bush, but many
ideologically dissimilar members of Congress represent constituencies with similar political
opinions. Other panels of this figure suggest the pattern is similar in state legislatures.
For nearly every Republican state legislator in Maine, Michigan, and Kentucky, there is a
59
Democrat from a district with comparable favorability towards Bush.2 The relationships in
Figure 4.1 suggest that a considerable number of federal and state legislators are “out of
step” with their districts.
Figure 4.1: U.S. and State House Representation
U.S. and state house member ideal plotted points against 2004 presidential vote for Bush. Both
at the federal and state levels, legislators often provide substantially different types of ideological
representation despite coming from constituencies with comparable political preferences. Ideal
points across state legislatures made comparable using NPAT estimates of legislator ideology (Shor
and McCarty 2011).
When studying the U.S. House, political scientists repeatedly find that “out of step”
legislators face electoral ramifications for their unrepresentative behavior. Voters sanction
their member of Congress for unpopular roll-call votes concerning the budget, congressional
salaries, or health care (Jacobson 1993; Clark 1996; Ansolabehere and Jones 2010; Nyhan
et al. 2012), and there is evidence that ideologically extreme members receive lower vote
shares (Canes-Wrone, Brady, and Cogan 2002; Carson et al. 2010; Ladewig 2010). These
2
This could reflect presidential vote being a poor proxy for state-level ideology, but similar patterns occur
in Pennsylvania, whose state party ideology resembles the federal parties (Shor and McCarty 2011).
60
empirical findings support the proposition that district-level electoral connections exist in
the U.S. House and provide evidence that assumptions of median voter theorems apply to
Congress.
Recall Figure 4.1 suggests that both members of Congress and state legislators are “out
of step,” and given the aforementioned empirical findings from U.S. House elections, one
may expect state legislators to face electoral consequences similar to their congressional
counterparts. Key differences in the informational and institutional contexts surrounding
state legislative contests, however, may make it relatively more difficult for voters to hold
their state legislator accountable.
Again consider the benchmarks for accountability laid out by Powell at the beginning of
this dissertation. A voter must “know who is responsible for making policy,” but compared
to learning about their member of Congress, it is relatively costly for a voter to acquire
information about their state legislator and her actions. Down-ballot elections for the state
legislature are frequently overshadowed by federal contests and receive less than a fourth
of the amount of local news coverage of Congressional elections (Kaplan, Goldstein, and
Hale 2003; 2005). This lack of attention - partially attributable to an incongruence between
legislative districts and media markets (Gierzynski and Breaux 1996) - has been shown to
lead to less informed state legislative voters (Delli Carpini, Keeter, and Kennamer 1994) and
weaker electoral connections in Congress (Snyder and Stromberg 2008). With few major
party challengers (Chapter 2) and little media coverage, the burden of “sift[ing] through
incumbents records” then increasingly falls to the individual voter (Arnold 1992: 49), and if
“the acquisition of any nonfree political information data whatever is irrational” for most
voters (Downs 1957: 239), it may be reasonable that the electorate does not “know who is
responsible for making policy” in state legislatures (Jewell 1982b).
Even if some voters identify and cast ballots against unrepresentative state legislators,
it is questionable whether they are given a “fair opportunity” to throw an unrepresentative
legislator out of office. For example, partisan state legislative districts help insulate state
61
Figure 4.2: Swing Seats in U.S. and State Houses
The above shows the proportion of state legislative seats where George W. Bush’s vote share in the
2004 election was below 45%, above 55%, or between 45% and 55%. Across the country, 10% fewer
state legislative seats fall into this latter category compared to U.S. House districts.
legislators from electoral defeat. Figure 4.2 presents the proportion of state legislative
“swing” districts where President Bush received between 45 and 55 percent of the vote in
the 2004 election. Only eight states have a higher percentage of swing districts than the
U.S. House, and there are some states, such as Rhode Island, with no legislative districts
where Bush received more than 55% of the vote. These types of districts contribute to state
legislators enjoying healthy electoral margins and often no opposition. Figure 4.3 illustrates
that from 1998 - 2008, approximately 80% of American legislators received at least 60% of
the vote, and only 10% received less than 55% of the vote. A small number of incumbent
legislators, therefore, appear to be electorally vulnerable.
The lack of attention given to state legislative elections makes it hard for voters to
identify out of step state legislators, and partisan districts increase the difficulty of throwing
them out of office. Both of these factors are important because if voters do not respond to
how they have been represented by their state legislator enough to threaten an incumbent’s
job security, state representatives have less electoral incentive to represent their constituents.
To determine whether this critical relationship - for both electoral accountability and the
62
Figure 4.3: Incumbent Vote Shares in U.S. and State House Elections
The bars of the graph illustrate the proportion of US and state house incumbents that sought reelection
who received different levels of vote share. Data cover the 1998 - 2008 elections. Approximately
80% of legislators receive at least 60% of the vote each election.
assumptions that underlie median voter theories - emerges in state legislatures, I conduct
two analyses concerning the electoral implications of legislators’ roll call behavior. I first
take advantage of referendum election results to generate district-level measures of public
opinion on 23 bills enacted by state legislatures. With these measures of public opinion, I
examine the electoral consequences for legislators casting unpopular roll-call votes. Second,
I study the electoral implications of a state legislator’s full roll-call record. For this analysis,
I use ideal point estimates of state legislators’ ideology from 1998 - 2008 to evaluate the
extent to which voters’ provide electoral incentives for ideological representation in 38
states. These analyses test the hypotheses that state house members who cast unpopular
roll-call votes or are ideologically extreme relative to their districts will receive lower vote
shares.
63
4.2
Accountability for Unpopular Roll-Call Votes
State legislatures do not always adopt popular policies, and when this is the case, voters
in 21 states can force a “veto-referendum” to have a statewide election determine whether a
bill adopted by the legislature will become law. Political scientists have used referenda and
initiatives to study whether state legislators vote consistently with their districts’ preferences
(e.g. Gerber 1996; Snyder 1996; Lewis and Gerber 2004), but I do not know of any research
that uses referenda to determine the extent to which voters hold legislators accountable for
their unpopular roll-call decisions. Because referenda election returns reflect district opinion
on the exact bills considered by the state legislature, they provide an excellent opportunity
to study the relationship between legislator and voter behavior.
In the last 15 years, 14 states have held veto-referenda. Veto-referenda can repeal
bills considered by the legislature but only occur for adopted legislation, which introduces
potential case selection bias in this study. I aim to include the universe of referenda but
focus on the 10 states that make precinct-level referendum election returns readily available,
which restricts my current analysis to 23 bills.3 Table 4.1 provides brief descriptions of the
considered bills and reports their levels of support in the state house and electorate. Voters
vetoed 11 of the 23 bills, and in doing so overturned state legislatures’ decisions to legalize
gay marriage, expand health care, and create charter schools.
Veto-referenda range in prominence and attention received. For instance the Michigan
Chamber of Commerce alone spent more money supporting emergency manager reform than
was spent overall on an earlier referendum regarding Michiganders’ hunting rights. With
enough interest group activity and spending, voters become aware of seemingly obscure
issues. Over $100 million spent by California interest groups on referenda regarding slot
machines in Native American casinos contributed to increasing statewide voter awareness of
the issue by 43% in the course of a month (Field Poll Jan 2008). Interest groups spent at least
3
For example, I do not study multiple Oregon veto-referendum since this state only provides county-level
referendum election returns. Excluded states are MA, OR, SD, and UT.
64
Table 4.1: Referendum Descriptions
SB 267
State
House
Vote
27-11
Statewide
Referendum
Result
53.5-46.5
HB 2518
32-24
36.1-63.9
98,174
SB 1373
51-7
53.5-46.5
98,174
Require medium to large business to provide health care coverage
SB 2
46-32
49.2-50.8
Prop 94
Permits 5500 additional slot
machines at certain Native
American Casinos
SB 903
61-9
55.6-44.4
2008
Prop 95
Permits 5500 additional slot
machines at certain Native
American Casinos
SB 174
50-13
55.6-44.4
CA
2008
Prop 96
Permits 3000 additional slot
machines at certain Native
American Casinos
SB 175
61-9
55.5-44.5
CA
2008
Prop 97
SB 957
52-11
55.5-44.5
ID
2012
Prop 1
S1108
48-22
42.7-57.1
47,432
ID
2012
Prop 2
S1110
44-26
42-58
47,432
ID
2012
Prop 3
Permits 3000 additional slot
machines at certain Native
American Casinos
Limits agreements btwn. teachers and school boards and ends
issuing renewable contracts
Establishes teacher pay for performance based on test scores
Increase technology spending
in schools, with ability to offset
costs using teacher salaries
S1184
44-26
33.3-66.7
47,432
ME
2005
Question 1
Prevent discrimination in
employment, housing, education...based on their sexual
orientation
LS 1196
91-58
55-45
ME
2008
Question 1
Soda Tax to pay for Health
Care Program
LD 2247
82-62
35.4-64.6
ME
2009
Question 1
Gay Marriage
LD 1020
89-57
47.2-52.8
MI
2006
Prop 3
Authorizes Dove Hunting Season
PA 160
65-40
31-69
MI
2012
Prop 1
PA 4
62-48
47-53
MI
2002
Prop 1
PA 269
56-47
40.3-59.7
MT
2012
IR-124
SB 423
78-17
57.2-42.8
$38,071
24,337
ND
2012
Measure 4
Authorizes Governor to establish city manager upon state
finding financial emergency
Eliminates Straight Party
Ticket
Enact a Medical Marijuana Program
Discontinue University of
North Dakota Fighting Sioux
nickname and logo
SB 2370
63-31
67.3-32.7
$19,499
13,452
OH
2008
Issue 5
Limit interest rates on short
term loans to 28%
HB 545
68-27
63.7-36.3
WA
2009
Ref. 71
Grants domestic partners all
rights, responsibilities, and
obligations granted to married
couples
SB 5688
62-35
53.2-47.8
WA
2004
Ref. 55
Authorizes creation of Charter
Schools
ESSHB2295
51-46
41.7-58.3
WA
2007
Ref. 67
Allows consumers to collect
trip damages from their insurance company
SB 5726
59-38
56.7-43.3
State
Year
Referendum
Issue Description
Bill
AK
2000
Measure 6
AZ
1998
Prop 300
AZ
1998
Prop 301
Restrict Land and Shoot Hunting of Wolves
Federal Oversight of Medical
Marijuana
Eligible for Probation with 1st
or 2nd Marijuana Crime
CA
2004
Prop 72
CA
2008
CA
65
Spending
on
Referendum
$31,075,168
$172,698,243
$172,698,243
$172,698,243
$172,698,243
$1,554,715
$4,612,389
$10,495,539
$3,003,704
$9,165,638
Min.
Signatures
Required
19,242
354,817
411,345
411,345
411,345
411,345
49,458
55,087
55,087
159,000
161,304
151,328
$21,416,231
$4,849,167
$9,228,262
$19,216,157
241,365
120,115
96,881
109,864
$4 million campaigning for or against most of the referenda considered in this analysis.4
Given voters appear to be “educated by the initiative” (Smith and Tolbert 2004), employing a
sample of veto-referenda increases the likelihood of voter awareness of legislators’ positions
compared to employing most other legislation.
Using election returns from these referenda and legislators’ roll-calls, I estimate the
extent to which a district’s support for a state legislator’s position correlates with a contested
state house incumbent’s vote share (Klarner et. al 2012) using Ordinary Least Square
regressions. To measure district-level opinion of legislators’ positions, I aggregate precinctlevel referendum election returns to the state legislative district-level. For each district, I
adjust this variable - ranging from 0 to 100 - such that higher values indicate greater support
for their state representative’s roll-call vote. Among those who sought reelection and faced
a major party opponent within this study, state house members cast 747 (of 1220) roll-calls
that represented their district’s majority opinion. If voters electorally reward legislators for
taking positions their district favors, β1 in Equation 4.1 should be positive.
IncumbentV oteShare =β0 + β1 [Supportf orIncumbent0 sRollCall]
+ β2 [IncumbentP artyP residentialV ote]
+ β3 [IncumbentSpendingAdvantage]
(4.1)
+ β4 [IncumbentP reviousV oteShare]
+ β5 [IncumbentP reviouslyContested]
+ β6 [DemocraticDummy] + OLS regressions include controls common to studies of elections. To account for a
constituency’s partisanship, I control for presidential vote at the legislative district-level
averaged by redistricting cycle.5 Statistical analyses additionally account for incumbents’
4
These figures reflect the available National Institute for Money in State Politics tabulations of spending on
referendum since 2004.
5
I collected Gore-Bush vote by pre-2000 legislative district for 45 states and Obama-Romney vote for
MI, MT, and ND. The National Committee for an Effective Congress generously provide 2004 and 2008
66
fundraising advantages using the difference in logged campaign contributions between the
incumbent and challenger (Bonica 2013; NIMSP). Other controls include an incumbent’s
partisanship, previous vote share, and whether the state representative was contested in the
previous election. Table C.1 provides summary statistics of each of the variables used in
this and subsequent analyses in this chapter.
To provide legislators greater electoral incentives to represent their constituents, electorates need to punish those in government for making unpopular roll-call decisions, but
statistical analyses provide little evidence that this is the case for most roll-calls considered
here. The fourth column of Table 4.2 presents central results from separate OLS estimations
for each piece of legislation that faced a veto-referendum. For 19 of 23 bills, there is no
statistically significant relationship between voters’ and legislators’ behavior (p ≥ .1). Voters
do not seem to sanction individual legislators for taking unpopular positions on important
legislation regarding payday loans, charter schools, or drug laws. It should be noted that
the only statistically significant results are from elections with the most observations, and
null findings may be the result of underpowered tests. The magnitudes of coefficients,
furthermore, are larger for positions legislators took on issues recently prominent in state
legislatures, such as health care, gay rights, and collective bargaining.
Findings from Table 4.2 do reveal some evidence of district-level electoral connections
in the Maine and Michigan state legislatures. Returning to the example of gay marriage
in Maine from the beginning of this chapter, statistical analyses suggest that a state house
incumbent’s reelection vote share would increase by 1% with a 10% increase in district-level
support for their roll-call position. Providing further evidence of a relationship between
legislator and voter behavior, a 10% increase in support for a legislator’s roll-call concerning
GLBT discrimination in 2006 results in approximately a 1.5% increase in vote share for the
incumbent. These statistical results suggest that Representative Lajoie electorally benefitted
presidential vote by post-2000 legislative district for 48 states, and Tausanovitch and Warshaw (2013) supply
Obama-McCain vote for FL and MS. Gore-Bush vote is missing for AR, CO, and MS, and Kerry-Bush vote
is missing for FL and MS due to an inability to collect precinct election returns by state legislative district.
Polidata provide the presidential vote measure for Congressional elections.
67
Table 4.2: Relationships between District-Level Support of Legislator’s Roll-Calls
and Incumbent Vote Share
State
AK
AZ
AZ
CA
CA
CA
CA
CA
ID
ID
ID
ME
ME
ME
MI
MI
MI
MT
ND
OH
WA
WA
WA
Year of Ref.
2000
1998
1998
2004
2008
2008
2008
2008
2012
2012
2012
2005
2008
2009
2002
2006
2012
2012
2012
2008
2004
2008
2009
∗∗ p
Issue Area
Hunting
Medical Marijuana
Marijuana Sentencing
Health Care
Slot Machines
Slot Machines
Slot Machines
Slot Machines
Collective Bargaining
Teacher Pay
Education Spending
Gay Rights
Soda Tax
Gay Marriage
Straight Party Ticket
Dove Hunting
Emergency Managers
Medical Marijuana
Native American Mascot
Payday Loans
Charter Schools
Insurance Claims
Gay Rights
Coef. District Support of Roll Call
0.167 (0.311)
-0.118 (0.245)
0.076 (0.186)
0.188 (0.138)
0.029 (0.113)
0.085 (0.110)
0.038 (0.113)
0.072 (0.107)
0.333 (0.204)
0.159 (0.143)
-0.003 (0.078)
0.146* (0.081)
0.000 (0.063)
0.101** (0.046)
0.051 (0.208)
0.101** (0.046)
0.087** (0.043)
-0.023 (0.130)
-0.039 (0.022)
-0.112 (0.111)
0.028 (0.048)
0.023 (0.119)
0.103 (0.091)
N
16
26
25
51
40
36
40
36
28
28
28
103
86
104
51
104
84
51
24
47
63
62
62
R-Squared
0.79
0.64
0.67
0.89
0.94
0.93
0.94
0.93
0.83
0.82
0.81
0.44
0.53
0.64
0.86
0.64
0.95
0.69
0.93
0.71
0.91
0.79
0.83
≤ .05; ∗ p ≤ .10 Standard Errors in Parentheses
Relationship between district public opinion for incumbent’s roll-call vote and incumbent vote share.
“Coef. District Support of Roll-Call” is the OLS coefficient and standard error for β1 of Equation 4.1.
Positive coefficients indicate voters reward or punish legislators for their roll-call position, which
appears to be the case for 4 of 23 bills examined. See Tables C.2 to C.7 for full estimates.
from his vote on gay marriage but most of his colleagues lost votes. The strength of
electoral connections, however, varies by issue. Maine voters appear to hold their legislators
accountable for their positions on gay rights but not for enacting an unpopular soda tax to
pay for health care programs.
Michigan voters also at times seem to hold their legislators accountable. In 2012, a
veto-referendum overturned a party-line vote by the Republican legislature that enabled the
governor to appoint emergency managers in financially-struggling local governments. Some
legislators paid a price for toeing the party line. Using estimates from Table 4.2, a 10%
decrease in support for a Michigan state house member’s roll call resulted in approximately
a 0.9% loss in incumbent vote share. Analyses additionally suggest that Michiganaders
punished their legislators for their position taking on dove hunting but fail to provide
68
evidence that they held their representatives accountable for changes to electoral laws, such
as eliminating the straight party ticket.
Findings from Maine and Michigan provide some evidence that electoral connections
exist in these states, but the relationships between legislator and voter behavior likely are
not strong enough to affect most legislators’ job security. As shown by Figure 4.3, voters
reelect most state legislators with considerable electoral margins, reducing the meaningful
electoral impact of an unpopular roll-call. For example if every Maine state house member
voted with the majority of their district on gay marriage, statistical analyses predict that
only two additional incumbents would have held onto their seats, and likewise even if every
incumbent voted against their district, only two additional incumbents would have lost.
Similarly if every Michigan state house member either voted with or against their district on
emergency manager reform, statistical analyses predict that only one seat outcome would
change. The potential vote share losses attributable to a state representative’s roll-call on
these pieces of legislation, therefore, do not appear to threaten most incumbents’ reelection.
4.3
Accountability for Ideological Representation
The above findings suggest that few legislators would lose votes for unpopular roll-calls
and fewer would lose their jobs for providing poor representation. Considering a single
roll-call, however, may be too narrow a test of Downsian predictions. State legislators take
hundreds of positions each legislative session. For example the 124th Maine legislature both
legalized gay marriage and cut education spending, and Michigan representatives created
emergency managers and tried to establish stricter voter ID laws from 2011 to 2012. Each
position taken by state legislators on these issues reflects their ideological representation
of their constituents, and spatial theories of electoral competition fundamentally expect
legislators who fail to represent their district on a broad ideological dimension to receive
lower vote shares, all else equal.
69
To study whether voters hold their state representatives accountable for their overall
legislative record, I investigate the electoral implications for how state representatives
ideologically represent their district. For this analysis, I compile district-level measures
of legislative behavior, constituent ideology, and election outcomes from the 38 states that
exclusively had single-member districts in the 1998 - 2008 elections. The dependent variable
is a contested state house member’s two-party vote share, and the primary independent
variable of interest is a measure of a legislator’s ideological representation.
To test the claim that “ideological moderation should increase an incumbent’s vote
share” (Canes-Wrone et. al 2002: 129), prior work on Congress often measures legislators’
ideological extremity by taking “the absolute value of [DW-NOMINATE] scores so that
high values...indicate that a member has an “extreme” voting record” (Carson et. al 2010:
606).6 Using this absolute value approach, however, presumes an ideal point of zero to
be moderate across all districts and potentially overlooks ideologically similar legislators
representing dissimilar districts.
Figure 4.4: Measuring Ideological Extremity
Assuming Jefferson and Lincoln represent districts with similar preferences, measures of ideological
extremity do not necessarily correspond to legislators’ ideological distance from their district.
6
Carson et. al argue “voters place greater weight on partisanship than ideology when evaluating behavior
in Congress” (Carson et. al 2010: 601-2) and argue that a legislators’ party loyalty matters more for their
reelection prospects than ideology. Similar to the ideological extremity findings in the main text, a replication of
Carson’s analysis at the state legislative level provides little evidence of relationships between state legislators’
party loyalty scores and their vote shares in most states (See Table C.16.)
70
Figure 4.4 illustrates an example of how a candidate classified as “extreme” using
this absolute value measurement may be more representative than a “non-extreme” one.
In this example, Jefferson and Lincoln respectively have ideal points of -50 and 60 and
are competing for votes in a district whose ideal point is 25. When measuring extremity
by taking the absolute value of an ideal point, Lincoln is considered more extreme than
Jefferson with a score of 60 compared to 50. However, the ideological distance between
Lincoln’s ideal point and the district (35) is less than the comparable distance for Jefferson
(75). Lincoln’s ideology, therefore, is more congruent with the district’s preferences, but
in regards to how Lincoln’s ideological representation will influence his electoral fate,
empirical findings from some Congressional research suggest that the “extreme” Lincoln
will receive fewer votes than Jefferson. Spatial theories of elections, meanwhile, predict the
ideologically congruent Lincoln will receive more votes than Jefferson.
For my analysis, I aim to more closely follow the framework of spatial theories and
measure a legislator’s ideological representation or extremity relative to their district. My
available measures of legislators’ and voters’ ideology - ideal points and district-level
presidential vote - however are on different scales and not directly comparable. I, therefore,
impute district ideal points under a general assumption that Democrat and Republican
legislators on average represent their districts. Variation between a legislator’s and her
district’s ideal points serve as evidence of ideologically extreme representation.
LegislatorIdealP oint =φ0 + φ1 [RepublicanP resV ote]
(4.2)
+ φ2 [RepublicanP artyDummy] + EstimatedDistrictIdealP oint = |φ0 + φ1 [RepublicanP resV ote]| + λ|φ2 |
71
(4.3)
To create district ideal points and measure legislative extremity, I first estimate an ideal
point for each state legislator (Clinton, Jackman, and Rivers 2004; Shor and McCarty 2011).7
Second, I regress legislators’ ideal points on their district presidential vote and party dummy
using Equation 4.2 for each legislative session. The party dummy accounts for intradistrict
divergence or Democrat and Republican legislators providing different representation to
the same district, captured by φ2 (McCarty, Poole, and Rosenthal 2009). Third, I predict a
district ideal point using estimates from Equation 4.2 in Equation 4.3 where the final term
assumes that Republicans and Democrats equally misrepresent the same district by setting λ
equal to 0.5.8 With this district ideal point projected into the ideal point space of legislators,
I measure extremity by taking the absolute value of the difference between a legislator’s
ideal point and their district’s ideal point. For each session, I divide this measure by its
standard deviation to create comparability across time within legislatures. In the main text, I
focus discussion on findings using this distance measure but provide comparable estimates
using the absolute value and other extremity measures in the Appendix (Tables C.14 &
C.15). Outside of this difference in measuring extremity, I aim to follow the econometric
setup of Canes-Wrone et al by estimating Equation 4.4 for each state.9
7
I use a single ideal point in order to facilitate cross-state comparisons. Substantive results do not change
when using session specific ideal points or adjusting session-level ideal points following the approach used by
Groseclose, Levitt, and Snyder (1999).
8
When excluding this party dummy and doing a residual regression, I find evidence of district-level
electoral connections in fewer - but different - states.
9
My study of state legislative elections further differs from the Congressional analysis in two respects.
Given the relative infrequency that state legislative incumbents face major party challengers, the below findings
pool elections from 1998 - 2008 to increase statistical power. Second, I do not control for challenger quality
given the lack of availability of this measure at the state legislative-level. Main results from Congressional
analyses do not change when excluding this control. Canes-Wrone et. al do not control for previous incumbent
vote share, but the ideological extremity analysis in this chapter are not sensitive to including this control.
72
IncumbentV oteShare =γ0 + γ1 [ExtremityM easure]
+ γ2 [IncumbentP artyP residentialV ote]
+ γ3 [IncumbentSpendingAdvantage]
(4.4)
+ γ4 [F reshmanDummy] + γ5 [DemocraticDummy]
+ γ6 [P residentialApproval] + I contrast the strength of state legislative electoral connections to those in Congress by
conducting a comparable study of U.S. House elections for the same 1998 - 2008 time period
(McCarty, Poole, and Rosenthal 2008). Both the state legislative and congressional analyses
control for district level presidential vote and campaign contributions using the variables
described in the individual roll-call analyses.10 I additionally control for presidential approval
as measured by the last national Gallup poll before the November election. This variable
ranges from 0 to 100, and I adjust values to reflect incumbents’ affiliation with the president’s
party. Freshman is a dummy variable indicating if the incumbent just completed his first
term, and the Democratic party variable denotes whether the state house member seeking
reelection was a Democrat.
State-level ideal points are not directly comparable across states, thereby preventing
comparisons of γ1 across legislatures. To provide comparable meaning to estimates, I
calculate the differences in average predicted vote shares associated with a common shift in
legislator ideology. Specifically, I difference a legislator’s predicted vote share holding all
variables at their true values and a legislator’s predicted vote share after making her ideal
point a standard deviation more extreme or distant from their district. To make standard
deviation changes comparable across states, I utilize the NPAT Common Space (described
in greater detail below) and calculate a NPAT standard deviation within the ideal point space
of each legislature (Shor and McCarty 2011). This counterfactual ideological adjustment is
10
Due to a lack of contribution data, I omit elections from Oklahoma and Delaware in 1998.
73
approximately equivalent to the median member of the U.S. House in 2008 always voting
with the Democratic majority leader, Steny Hoyer, or roughly a 1 unit change on the Y-Axis
on the top left panel of Figure 4.1 and approximately a .7 unit change on other panels.
Table 4.3: Relationships between Incumbent’s Ideological Extremity and Vote Share
State
US
AL
AK
AR
CA
CT
CO
DE
FL
GA
HI
IA
ID
IL
IN
KS
KY
MA
ME
MI
MN
MO
MS
MT
NC
NM
NY
OH
OK
OR
PA
RI
SC
TN
TX
UT
VA
WI
WY
∗∗ p
Distance from District
-0.220* (0.125)
-0.230 (0.957)
-1.126 (1.260)
0.113 (1.061)
-0.131 (0.165)
0.244 (0.266)
-0.148 (0.311)
0.280 (0.375)
-0.322 (0.226)
-0.339 (0.332)
-0.866** (0.417)
-0.042 (0.152)
-1.121** (0.319)
-0.442 (0.291)
0.045 (0.260)
0.268 (0.343)
0.281 (0.554)
0.038 (0.312)
-0.171 (0.271)
-0.242 (0.163)
-0.110 (0.211)
-0.616** (0.20)
-0.809 (0.935)
0.301 (0.296)
0.397 (0.317)
0.585 (0.520)
-0.145 (0.268)
-0.097 (0.294)
-1.520** (0.473)
-0.509** (0.190)
-0.414 (0.264)
-1.653** (0.496)
-0.403 (0.615)
0.087 (0.633)
-0.714** (0.276)
-0.997** (0.432)
0.377 (0.528)
-0.966** (0.203)
0.040 (0.773)
R-Squared
0.64
0.57
0.49
0.54
0.84
0.60
0.81
0.82
0.68
0.50
0.26
0.51
0.70
0.79
0.59
0.46
0.29
0.46
0.26
0.83
0.56
0.56
0.60
0.28
0.56
0.62
0.81
0.72
0.52
0.78
0.63
0.69
0.46
0.49
0.56
0.63
0.44
0.59
0.47
N
1931
72
83
35
239
466
132
97
152
275
188
293
429
298
285
290
160
178
397
398
621
373
46
266
136
117
534
299
147
153
538
166
138
167
262
185
131
262
80
≤ .05; ∗ p ≤ .10 Standard Errors in Parentheses
Second column presents the estimates and standard errors for γ1 of Equation 4.4 using the sample
of contested state house incumbents who sought reelection from 1998 - 2008. Negative estimates
provide evidence for the hypothesis that ideological extreme legislators receive lower-vote shares, all
else equal. Full Estimates available in Tables C.8 to C.12.
The second column of Table 4.3 presents central results from estimations of Equation
4.4 for each legislature. These estimates are the OLS coefficient for the extremity variable
- as measured by legislators’ distance from their district - for Congress (US) and each
74
state house. Increases in incumbents’ ideological extremity produce statistically significant
decreases in vote shares in 9 of 38 states (p ≤ .1). Using these estimates, Figure 4.5 displays
the difference in average predicted incumbent vote share when a state’s legislator’s rollcall behavior undergoes a common ideological shift away from their district. A standard
deviation change in a legislator’s ideal point results in over a 4% predicted vote share loss
for incumbents in Oklahoma, Wisconsin, Missouri, and Utah, over twice the comparable
punishment endured by U.S. House members (1.9%). Despite the accountability findings
regarding individual roll-calls from Maine and Michigan, neither states’ elections produce
strong evidence for the hypothesis that ideologically extreme legislators receive lower vote
shares.
Figure 4.5: Predicted Changes in State House Incumbent Vote Share
Points represent average predicted vote-share loss associated with a common change in an all
incumbents’ ideal points for state house elections from 1998 - 2008. The point labeled US represents
the comparable estimate for the U.S. House. Black and grey bars are 95% and 90% bootstrapped
confidence intervals. Only 7 states produce statistically significant evidence of vote-share loss for a
standard deviation change in all legislators’ ideal points.
Predicted losses in vote share support the proposition that district-level electoral connections appear to exist in some states. The magnitudes of these relationships suggest
75
that if some legislators become more ideologically distant from their district they could
be voted out of office. When converting predicted vote share losses from Figure 4.5 into
seat losses, at least an additional 7% of Wisconsin and Oklahoma incumbents who sought
reelection in the past decade are predicted to lose their seats. Seat losses in Wisconsin
are partially attributable to this state having the fifth most swing seats - those where Bush
received between 45 - 55% of the vote in 2004 - in the country (Figure 4.2). Analyses
similarly predict seat losses in Oregon. In this state, there is a relatively weaker relationship
between ideological extremity and vote share, but over a third of districts are swing districts.
Only in these three states along with Idaho and Congress does a standard deviation change
in a legislator’s ideal point away from their constituency result in statistically significant
predicted seat losses (p ≤.1). Safe seats, therefore, appear to insulate ideologically extreme
legislators from losing their jobs, even in the Utah or Missouri state legislatures where
comparably strong electoral connections exist.11
Despite few legislators losing their jobs for how they represent their districts, Figure
4.5 illustrates that the strengths of electoral connections vary across states. Ideological
extremity has larger electoral consequences for legislators in Wisconsin than Wyoming,
and institutional features may be responsible for this variation. Legislators are more likely
to be reelected from professionalized legislatures such as the Wisconsin Assembly (King
1991; Berry, Berkman, and Schneiderman 2000) despite voters being more disapproving of
these types of legislatures (Squire 1993; Kelleher and Wolak 2007 but see also Richardson, Konisky, and Milyo 2011). Recent work additionally demonstrates that increases in
district population negatively associate with favorable opinions about state government
(Bowen 2013). These institutional features may relate to voters’ opinions and behavior
concerning their state legislature, but it is unclear if they contribute to holding ideologically
unrepresentative legislators accountable.
11
Kerry received 45 - 55% of vote share in only 16% of districts in these states.
76
To further investigate the electoral impact of ideological extremity across the country
and better determine why differences in Figure 4.5 exist, I pool observations from the 38
state legislatures. In these analyses, I examine whether extreme state representatives from
more professionalized legislatures or large districts are more likely to receive lower-vote
shares. I additionally revisit Hogan’s (2008) finding that increased incumbent spending
advantages decrease the likelihood a legislator is held accountable for unresponsive legislative behavior. To estimate these conditional effects, I interact my measure of ideological
extremity with variables that capture state legislative professionalism, district size, and
individual candidates’ fundraising.12 OLS coefficients on these interaction terms reflect how
states’ legislative institutions influence the strength of district-level electoral connections.
For comparable measures of ideology, I again calculate legislators’ ideological distance
from their districts using the method described above but replace legislators’ state-level
roll-call ideal points with NPAT ideal points (Shor and McCarty 2011). Shor and McCarty
estimate ideal points using survey responses of state legislators who responded to Project
Vote Smart’s National Political Awareness Test (NPAT), producing a common space of
state legislator ideology across states. For each state, legislators’ survey ideal points are
then regressed on their corresponding roll-call ideal point and a dummy variable for party.
Coefficients from this regression and all state legislators’ roll-call ideal points then predict
state legislators’ NPAT ideal point, which is assumed to be on a common scale across states.
It should be noted that NPAT measures of ideology are less polarized - and therefore likely
less extreme - and estimated with more error than the roll-call, state ideal points used in
the analyses presented in Table 4.3 (Ansolabehere, Snyder, and Stewart 2001), but they are
currently the best cross-state measure of legislator ideology (Berry et al. 2012).
To disentangle the influence of incumbent and challenger fundraising similar to Hogan’s
(2008) analysis, I take advantage of increased statistical power and replace the incumbent
12
Similar to earlier chapters, I use Squire’s professionalism index which accounts for differences across
states in legislators’ pay, staff, and length of legislative session. I measure constituency size using the logged
value of district population reported by the National Conference of State Legislatures.
77
spending advantage variable with separate logged measures of the amount raised by these
candidates. Likewise for presidential approval, I include the unadjusted presidential approval
rating from the last national Gallup poll before the election and this measure interacted with
a dummy variable denoting if a legislator was a member of the President’s party. Coefficients
on these variables reflect the relationship between presidential popularity and the respective
vote shares for state legislators not affiliated and affiliated with the president’s party. For
this cross-state study, OLS regressions include state and year fixed effects and other control
variables are similar to those described in the prior analyses.
When pooling states, statistical analyses from Table 4.4 suggest that unrepresentative
legislators receive lower vote shares, but the electoral consequences of legislators providing poor ideological representation are relatively small. Estimates from the first column
of this table suggest that going from the minimum to maximum observed values of the
ideological extremity variable results in only a 2.4% change in vote share. This predicted
loss - attributable to a voter having the least representative state legislator in the country
instead of the most - is equivalent to the estimated impact of a 9% change in district partisanship. Statistical analysis in the second column of Table 4.4 fail to provide evidence this
relationship is sensitive to extreme legislators representing larger districts or serving in more
professionalized legislatures.
My cross-state analysis resembles Hogan’s examination of the 1996 and 1998 state
legislative elections in 14 states. In that study, he shows that increased challenger spending
reduces incumbent vote shares and more so for unresponsive incumbents. Similar to Hogan,
I find that increased challenger fundraising decreases incumbent vote while incumbent
fundraising has little effect. For example, a challenger raising $20,000 instead of $10,000
for their campaign equates to approximately a 1% decrease in incumbent vote share. I,
however, do not find that increased challenger fundraising is more electorally advantageous
when running against an ideologically unrepresentative incumbent, as implied by the near
zero coefficient on “Challenger Contrib. X Extremity.” It should be noted that Hogan’s and
78
Table 4.4: Cross-State Regression using NPAT Measures of Ideological Extremity
Variable
Ideological Extremity
Extremity
-0.155*
(0.048)
Extremity X Professionalism
Extremity X District Size
Incumbent Contrib. X Extremity
Challenger Contrib. X Extremity
Inc. Party Presidential Vote
Logged Incumbent Contributions
Logged Challenger Contributions
Incumbent Previous Vote Share
Incumbent Previously Contested Dummy
Q4 Presidential Approval
Presidential Approval X Pres Party
Professionalism
Logged District Size
Democratic Party Dummy
Member of Pres Party
Freshman Incumbent
Midterm Election Dummy
Constant
R-Squared
N
∗p
0.288*
(0.006)
-0.09
(0.087)
-1.403*
(0.032)
0.245*
(0.008)
6.816*
(0.327)
-0.032*
(0.01)
0.139*
(0.011)
-1.597
(2.778)
5.382*
(1.6)
1.33*
(0.164)
-7.79*
(0.615)
0.675*
(0.152)
-0.542*
(0.261)
-12.609
(14.687)
0.67
8851
Extremity w/ Controls
-0.496
(0.754)
-0.668
(0.498)
0.053
(0.087)
-0.017
(0.037)
0.013
(0.013)
0.288*
(0.006)
0.003
(0.219)
-1.477*
(0.078)
0.245*
(0.008)
6.802*
(0.328)
-0.031*
(0.01)
0.139*
(0.011)
2.812
(4.248)
4.68*
(1.782)
1.338*
(0.164)
-7.809*
(0.616)
0.673*
(0.152)
-0.582*
(0.264)
-6.944
(16.083)
0.67
8851
≤ .05; Standard Errors in Parentheses
Relationship between state representative’s ideological extremity and vote share for contested
incumbents who sought reelection from 1998 - 2008. Ideal points made comparable across states
using NPAT surveys (Shor and McCarty 2011). Statistical analyses suggest more extreme legislators
receive lower-vote shares, but the strength of this relationship is relatively weak and not conditioned
by professionalism, district size, or campaign finance as described in the main text.
my measure of representation differ considerably, and I could not replicate Hogan’s finding
using the 1998 election with my measure.13
13
My measure of extremity more closely follows Hogan’s (2004) measure of policy responsiveness. In this
study from 2004, he does not consider whether there is the relationship between vote share and extremity
is conditional on challenger spending. In Hogan’s later 2008 work, his partisan policy position measure,
emphasizes whether legislators voted with their party to capture “extreme partisan voting records” (Hogan
2008: 863). Two legislators with the same interest group score who represent identical districts receive
79
Compared to the implications of national political contexts, representatives’ individual
legislative behavior matters relatively little for their election outcomes. A standard deviation
decrease in presidential approval has a greater electoral impact than a 3 standard deviations
change in the ideological distance measure for members of the president’s party. These
predictions regarding the impact of presidential approval comport relatively well with the
findings from the seat change analyses in the previous chapter. The president’s party expects
to gain 1.4% of state legislative seats with a 10% increase in presidential popularity (Chapter
3: Table 3.3). With a similar shift in presidential approval, findings in Table 4.4 suggest that
the president’s legislative co-partisans reap a 1.1% increase in vote share, all else equal.
When comparing the relationship of presidential approval for members unaffiliated and
affiliated with the president’s party, the sensitivity of the relationship between incumbent
vote share and presidential approval is over three times greater for members of the president’s
party. The predicted impact of a 10% decrease in presidential approval on non-president’s
party state legislators’ vote share, however, still exceeds that of a standard deviation decrease
for the ideological extremity measure. Recall from the second chapter of this dissertation
that challengers are more likely to take on the president’s state legislative copartisans when
the president is unpopular. This differential influence of presidential approval in Table
4.4 suggests these challengers have the right idea.14 Similar to the findings regarding state
legislative challengers and parties, national conditions’ influence on state legislative elections
imply legislators fearing electoral retribution need to worry more about the president’s
performance than their own.
oppositely signed “party policy positions” if these legislators belong to different political parties. A limitation
of this approach is that it automatically classifies conservative Democrats and liberal Republicans as extremely
representative regardless of their districts. For example, consider Representatives Jackson (D) and Bush (R)
who have ideal points of 75 and 60 but both represent a district with an ideal point of 50. Following Hogan’s
example in footnote 13, Bush’s “party policy position” would be +10 and Jackson’s would be -25. Bush is
considered more extreme, despite being more ideologically representative of his district.
14
Which incumbents are included in the sample may be biased by challenger decision-making. Drawing on
the findings from chapter 2 of this dissertation, I replicate the analysis conducted in Table 4.4 using a Heckman
selection model. For this analysis, I assume that chamber competition influences challenger entry but has no
direct relationship with incumbent vote share. Results from this analysis (Table C.13) are similar to findings
from the main text.
80
4.4
Summary
Findings from this study of district-level accountability suggest that most state legislators
have relatively weak electoral incentives to represent their districts. State representatives
rarely seem to receive lower vote shares for casting unpopular roll-call votes, and most
state legislative elections from the past decade produce little support for the proposition that
elections hold legislators accountable for providing poor ideological representation. Weak
electoral connections combined with partisan legislative districts make it difficult for voters
to throw unrepresentative incumbents out of office. Most state legislators, therefore, do not
appear to face meaningful electoral consequences for their own legislative behavior.
Being unable to remove unrepresentative legislators from power not only has implications
for representation in state legislatures but also for Congress itself. State-level electoral
competition can weed out undesirable politicians and create a farm team of tested “quality”
candidates for political parties (Jacobson and Kernell 1983). State legislative teams have
potential all-stars including Illinois state senator Barack Obama and busts, such as Nevada
assemblywoman Sharon Angle. State legislative elections, however, never cut Angle the twice selected worst Assembly member by the Las Vegas Review-Journal - off the
Republican farm team, better enabling her to spoil Republicans’ chances to oust Harry Reid
from the U.S. Senate in 2010 (Vogel 2010). Political scientists, therefore, may want to more
carefully consider whether previously holding elected office necessarily defines a quality
candidate.
The above findings provide little support for a key assumption of median voter theories,
but it is important to recognize the empirical and theoretical limitations of my analysis. My
empirical investigations of district-level accountability in state legislatures are the most
thorough to date in terms of elections considered, but the veto-referenda I study only shed
light on electoral connections in a subset of states for a limited set of issues. Additionally
while my measure of ideological extremity more closely follows the framework of spatial
theories compared those that rely on the assumption that an ideal point equal to zero
81
is “perfectly moderate” (Ladewig 2010: 503), it still does not accurately put voters and
legislators on a common ideological space (Achen 1978).
Of greater theoretical consequence, my analyses in this chapter give little consideration
to the challenger. Ferejohn assumes that “the challengers play no active role” (Ferejohn
1986: 8), but Downs and other theorists presume that voters respond to both “government
activities” and “the strategies of opposition parties” (Downs: 1957: 138).15 Recall from
the second chapter of this dissertation that voters’ choices in state legislative elections may
systematically vary due to larger political conditions or legislative leaders (Sanbonmatsu
2006), which potentially has implications for voters’ choices in the spatial model.
The 2010 New Hampshire state house elections underscores this point. The Republican
state party chair claimed future Speaker Bill O’Brien recruited challengers “not only to fill
seats, but to find people who were ideological - you know the right kind of Republican.
He wanted the strongest Anti-Obama types” (Glass 2012).16 At the time of the candidate
filing deadline, President Obama’s approval rating in New Hampshire was the lowest of
his presidency (Granite Poll 2010), and when considering the findings of the previous
chapter of this dissertation, O’Brien’s strategy was likely sound. In regards to this chapter’s
findings, O’Brien’s tactics illustrate that New Hampshire voters’ ideological choices in state
legislative elections were not random. Voters reelecting the incumbent may be a better
option than choosing an even less ideologically representative alternative, and if this is
the case in most state legislative elections, my findings would be more consistent with the
predictions of spatial theories.
Despite the limitations of this analysis, my research designs are comparable to those
employed by congressional studies, are the most powerful to date, and still fail to provide
15
Ansolabehere, Snyder, and Stewart (2001) provide empirical evidence for this assumption in congressional
elections. Fearon (1999) additionally presents the argument that elections are the solution to an adverse
selection instead of a moral hazard problem. The analysis in this chapter focuses more on retrospection and
sanctioning than selection, but the two interpretations of elections are not “mutually exclusive.” Unless voters
are purely prospective (e.g. comparing only campaign platforms), “successful selecting of good types implies
sanctioning bad types” (Fearon 1999: 57).
16
I thank Eric Lawrence for pointing me to this quote.
82
compelling evidence of individual accountability in most state legislatures. This does not
imply that electoral connections fail to exist in state legislatures, but similar to the findings
from the previous chapters, they provide little evidence that state legislative elections are an
effective means of accountability for legislators’ own actions.
83
5
Conclusion
Competition for votes is intended to motivate officeholders’ performance, but this
motivation is not easy to achieve when a state representative from Louisiana never faces any
opponent for 18 years. Motivating those in power can be increasingly difficult when voters
reward the minority party when they think the legislature is doing a good job or repeatedly
reelect the twice selected worst Assembly member to the Nevada state legislature. These
anecdotes and the previous analyses do not seem encouraging for the proposition that state
legislative elections provide meaningful accountability.
My study of state legislative accountability appears to provide little hope that elections
create incentives for state legislators to represent their constituents, but my findings actually
are the strongest evidence - at least that I am aware of - that how state legislators perform
has any implications for their own elections. Incumbents more often face a major party challenger during bad economies (Chapter 2); voters overall tend to reward or punish the party
in control of the legislature (Chapter 3); and some legislators face electoral consequences
for casting unpopular roll-calls (Chapter 4). Thus, it is too strong to say that relationships
between legislators’ behavior and election outcomes fail to exist in state legislatures. The
strength of these electoral connections, however, is perhaps underwhelming.
84
As it is a voter’s decision regarding how to cast his or her ballot, it is up to the reader
to determine whether my overall findings imply elections for the state legislature fulfill his
or her standards for electoral accountability. At the beginning of this dissertation, I offered
Powell’s requirements as a benchmark, and the subsequent chapters provide empirical
evidence addressing whether state legislative elections satisfy these conditions. I find that
millions of voters do not have a “fair opportunity” to replace their representative. Even more
voters do not know “who is responsible for making policy” in state legislatures, and there is
little evidence of “meaningful” electoral implications for legislators’ and their parties’ own
actions in most states. These relationships produce relatively weak electoral connections at least compared to those in Congress - and suggest the assumptions underlying Key’s and
Downs’ theoretical arguments do not necessarily apply to state legislatures as they do the
federal level.
Electoral connections are strongest when there are consequences at the ballot box for
representative’s own actions, and I, therefore, focused the previous discussion on the extent
to which state legislators control their own electoral fate. Theories of electoral accountability
suggest a relationship between election outcomes and state legislators’ behavior will force
legislators to fear a future opponent, anticipate a retrospective electorate, or cater to the
median voter when representing their constituents. If legislators fail to do so, their reelection
should become less certain. A takeaway from my work is that there appears to be little for
legislators to fear in state legislative elections, at least within what they can control.
It becomes normatively worrisome if those in power recognize that their electoral fates
are only weakly tied to their own actions. An intent of elections is to constrain legislators’
policymaking, but if going into the 2010 elections state legislative Republicans can claim
that “the biggest thing working for us: President Obama and the anti-president attitude,” it
is unclear why they would worry much about repercussions for their own actions, especially
when their claim turned out to be right.
85
If elections are driven by forces outside officeholders’ control, electoral connections
for these individuals only weaken from the ideal. Neither state legislative challengers’
consideration of their opponents’ affiliation with the president nor ballots cast for the state
legislature on the basis of how the president performs promote accountability at the statelevel. As discussed in the third chapter, contests for the state legislature appear to be
“second-order” elections where votes are cast “on the basis of factors in the main political
arena of the nation” (Reif & Schmitt 1980: 9) instead of serving their first-order purpose - to
hold the state government accountable for their own performance. State politics, therefore,
is not all local.
“The voter is not a fool.”
– Morris Fiorina (1981)
The first-order purpose of this dissertation is addressing accountability in state legislatures. I demonstrate that most legislators appear to have weak electoral incentives to perform
well or represent their constituents, thereby providing little support for the fundamental
assumptions underlying many theories of retrospective voting or the median voter. As
forewarned in the introduction of this dissertation, my findings should not be confused for
a condemnation of the voter. Elections involve voters, but they also include elites and the
institutions they create. Any explanation of elections - either behavioral or institutional requires considering each, and the findings of this dissertation illustrate that the electorate is
only partly responsible for the weak electoral connections in state legislatures.
Elites and electoral environments are key components of the accountability story. Returning to Powell’s benchmarks one last time, voters cannot hold their state legislator electorally
accountable without challengers providing them a “fair opportunity” to replace the policymakers. When parties secure legislative majorities before any elections take place, it makes
it difficult, if not impossible, for the voter to hold their state government accountable. Even
when elections provide voters an “opportunity” to replace their representatives, it sometimes
is not necessarily a “fair” one. Partisan legislative constituencies translate into electorally
86
safe seats, making it increasingly difficult for voters to throw out of step legislators out of
office.
As elites should not be overlooked, neither should voters’ efforts. Voters can be reasonable and at times behave in ways consistent with Key’s and Downs’ expectations. A primary
finding of this dissertation is that elections do not seem to provide the type of retrospective accountability laid out by Downs or Key, but chapter 3 provides a clear example of
retrospective behavior by voters. Recall that when the state legislature is perceived to have
performed poorly, voters who identify the state house majority party electorally sanction
the party in power. Misinformed voters - those who mistakenly believe the other party is
in power - meanwhile are more likely to vote for this party that poorly managed the state
government. Both informed and misinformed voters therefore act retrospectively by trying
to hold who they think is in power accountable, implying that Downs and Key at least are
right regarding retrospection. Voters’ retrospective behavior, however, leads to counteracting
votes and ultimately weakens the relationship between government performance and election
outcomes that is fundamental for electoral accountability.
The lack of accountability in state legislatures can be partially explained by another
insight by Downs. For Downs, information is critical but expensive, and he stresses it can
be irrational for voters to acquire political information. Since learning about one’s state
legislature can take considerable effort, it may be perfectly reasonable for voters to be
unaware of who their state representative is or not know which party controls their state
legislature. It is cheaper to learn something about national politics, and while perhaps a
questionable standard, my findings suggest at least voters are responding to politics when
casting their ballot for the state legislature. Nothing requires voters to meet the potentially
idealistic standards of Key, benchmarks of Powell, or propositions of Downs. If they fail to
do so, it is still part of the story where elections fail to provide meaningful accountability in
state legislatures, but it does not imply that voters are fools.
87
“Unless mass views have some place in the shaping of policy, all the talk
about democracy is nonsense.”
– V.O. Key (1961)
This dissertation tests the prediction that legislators experience more electoral success
when they perform well or act in constituents’ interests (Downs 1957; Key 1961). This
prediction underlies countless studies of elections and representation, but it is shortsighted to
only consider Key’s and Downs’ explanations of politics. I focus on elections as a solution to
a moral hazard (Ferejohn 1986) instead of adverse selection problem (Fearon 1999), and the
previous chapters do not fully characterize voters’ and elites’ parts in the accountability story.
State legislators could be selected or held accountable by party constituencies (Fenno 1978;
Masket 2011), and the findings regarding gay rights veto-referenda in Maine suggest that
the electorate punishes legislators for the positions they take on particular issues (Chapter 4;
Bovitz and Carson 2006; Canes-Wrone et. al 2011). These forms of accountability do not
necessarily create electoral incentives for state legislators to cater to the median voter in all
policy domains, but evidence for either of these hypotheses supports the proposition that
some type of electoral connection exists in state legislatures.
Explanations of voter or elite behavior obviously did not end with the works of Downs
and Key, but I encourage future research to give their theoretical contributions a richer
treatment than the dimensions of elections studied here. For example, Downs gives greater
consideration to voter information than this dissertation does justice, and despite my findings
regarding challenger entry and representation, there is much to learn regarding the ideology
of state legislative elites. My studies of challengers’, parties’, and legislators’ behavior help
establish a foundation for this future work on state legislative elections, and I look forward
to others refining and even challenging my findings when providing better explanations of
the electoral implications of state legislators’ behavior.
This dissertation provides evidence - or the lack thereof - regarding levels of accountability in state legislatures, but it is important to recognize that electoral accountability is not
88
required for legislators to act in their constituents’ interests. Erikson, Wright, and McIver
(1994) find a strong correlation between state public policy and opinion, and this dissertation
cannot refute that this relationship exists. The previous chapters on accountability do not
disprove claims that policy responsiveness in state legislatures is attributable to voters selecting representative legislators. Nor does it rule out that state legislators’ anticipate electoral
monitoring (Erikson, Wright, and McIver 1994: 247). However, given I find that most state
representatives are not punished at the ballot box for their actions in the legislature, what
my analyses do say is that state legislators likely anticipate too much.
The fear of electoral sanction alone may be sufficient to produce representative legislative
behavior (Fenno 1978; Cohen 1984; Arnold 1992) or perhaps only a small group of highly
attentive citizens is needed to induce representative policy-making (Erikson, Mackuen, and
Stimson 2002). Statehouse Democracy asserts that this is the case and provides valuable
insights regarding the role of ideology, parties, and state policy using arguably the best data
available at the time of its publication. The authors find liberal states produce liberal policies
using measures aggregated over a dozen years. State ideology, who runs state governments,
and laws, however, change more frequently, and it is unclear the extent to which short
term changes in policy are responsive to state public opinion or are the side effect of larger
political forces (Brace et al. 2004; Dubin 2007).
A contribution of this dissertation is the data it brings to the study of state legislative
elections. These data and measures recently introduced by others (e.g. Battista, Peress, and
Richman 2013; Bonica 2013; Lax and Phillips 2009; Tausonovitch and Warshaw 2013; Shor
and McCarty 2011) can help provide clarity to the question of whether state public policy
and individual representatives respond to public opinion. For example, Figure 4.1 implies a
lack of dyadic representation within legislative chambers, and recent research on state-level
policy responsiveness and state legislative candidates’ perceptions of their districts’ opinion
each discover “noncongruence in the conservative direction” (Lax and Phillips 2009: 383;
Broockman and Skovron 2013). A promising avenue of future work is taking advantage of
89
these new data to revisit the findings from Statehouse Democracy to more fully understand
representation in the states.
“It is one of the happy incidents of the federal system that a single courageous
state may, if its citizens choose, serve as a laboratory; and try novel social and
economic experiments without risk to the rest of the country.”
– Justice Louis D. Brandeis (1932)
New data better enables political scientists to answer the repeated calls to use states
to test theories traditionally examined in the national setting (e.g. Jewell 1982; Brace
and Jewett 1995; Morehouse and Jewell 2004). States can serve as both laboratories for
policymakers and political scientists, but the findings of this dissertation and other work
should temper temptations to draw broader inferences from subnational research. If national
politics dominate state legislative elections and “national forces working through evaluations
of the president are a major influence on voting for governor,” (Carsey and Wright 1998:
1001) expectations regarding voters’ and elites’ behavior may vary between the state and
national levels. Recognizing differences in the “laboratories of democracy” is necessary
when theory testing at the state-level.
State legislatures, for example, offer opportunities to study theories of legislative organization (Aldrich and Battista 2002; Anzia and Jackman 2013; Cox, Kousser, and McCubbins
2010), but as with retrospective voting and median voter theories, it is important to establish whether contexts satisfy theoretical assumptions. My findings suggest that voters
respond more to presidential rather than state politics, implying how state actors achieve
their reelection goals should differ from their federal counterparts. If this is the case, one
question my findings raise is why state legislative parties should care about their brand in
the same way as Congressional cartels (Cox and McCubbins 2005) - especially when voters
don’t know which brand is in charge (Figure 3.1). Studying states can shed light on how
legislatures operate, but even within the subfield of American political science, “American
Exceptionalism” perhaps applies to studying the federal government (Lipset 1997).
90
This dissertation has implications for how scholars study politics, and but its central
message concerns accountability in state legislatures. State legislators have considerable
authority over American lives. They determine who has the opportunity to vote, go to
college, and even get married, and elections are the primary instrument by which citizens
can respond and exert control over those who govern them. This fundamental democratic
connection can emerge through almost any election, but the threat of electoral accountability
only has meaning if there is a relationship between voters’ and legislators’ behavior, which
does not appear to be the case in state legislatures.
91
Appendix A
Appendix for Chapter 2
92
93
∗p
Pres. Pty
-4.904*
(0.647)
1.367*
(0.151)
0.461*
(0.174)
-0.571*
(0.038)
0.169*
(0.081)
0.306*
(0.048)
0.009
(0.031)
-0.394*
(0.045)
-0.105*
(0.036)
-0.011*
(0.004)
-1.462*
(0.107)
-2.468*
(0.162)
-0.066
(0.060)
0.063*
(0.030)
0.118*
(0.027)
-0.158*
(0.031)
1.413*
(0.275)
-6385.0
11433
˜Pres. Pty
-3.593*
(0.677)
0.972*
(0.154)
0.350*
(0.170)
-0.521*
(0.038)
0.352*
(0.083)
0.384*
(0.046)
-0.125*
(0.030)
-0.033
(0.044)
-0.065
(0.036)
-0.009*
(0.004)
-0.992*
(0.102)
-2.543*
(0.161)
-0.068
(0.059)
-0.004
(0.030)
-0.125*
(0.027)
-0.263*
(0.029)
1.654*
(0.271)
-6621.6
11353
Gov. Pty
-7.411*
(0.686)
1.210*
(0.161)
0.167
(0.169)
-0.485*
(0.037)
0.134
(0.086)
0.326*
(0.047)
0.044
(0.031)
-0.380*
(0.042)
-0.117*
(0.036)
-0.011*
(0.004)
-1.253*
(0.105)
-2.845*
(0.166)
-0.201*
(0.061)
-0.054
(0.027)
0.086*
(0.027)
-0.190*
(0.030)
1.968*
(0.279)
-6319.9
11027
≤ .05; Standard Errors in Parentheses
All Inc.
-4.237*
(0.463)
1.161*
(0.107)
0.357*
(0.121)
-0.555*
(0.027)
0.260*
(0.058)
0.367*
(0.033)
-0.064*
(0.021)
-0.241*
(0.029)
-0.079*
(0.025)
-0.010*
(0.003)
-1.242*
(0.073)
-2.461*
(0.114)
-0.065
(0.042)
0.055*
(0.019)
0.012
(0.019)
-0.227*
(0.021)
1.455*
(0.191)
-13096.7
22786
˜Gov. Pty
-1.478*
(0.645)
1.343*
(0.151)
0.489*
(0.176)
-0.665*
(0.040)
0.412*
(0.079)
0.360*
(0.047)
-0.173*
(0.030)
-0.107*
(0.041)
-0.031
(0.036)
-0.009*
(0.004)
-1.418*
(0.106)
-2.142*
(0.158)
0.032
(0.059)
0.139*
(0.028)
-0.070*
(0.026)
-0.227*
(0.030)
1.168*
(0.269)
-6705.8
11759
Maj Pty
-4.132*
(0.588)
1.086*
(0.135)
0.333*
(0.157)
-0.628*
(0.035)
0.140
(0.075)
0.355*
(0.042)
-0.125*
(0.027)
-0.286*
(0.037)
-0.078*
(0.032)
-0.008*
(0.004)
-1.384*
(0.096)
-2.422*
(0.146)
-0.054
(0.053)
0.034
(0.025)
0.086*
(0.024)
-0.205*
(0.026)
1.482*
(0.246)
-8028.8
13972
˜Maj. Pty
-4.435*
(0.759)
1.840*
(0.208)
0.366
(0.195)
-0.449*
(0.042)
0.457*
(0.092)
0.409*
(0.054)
0.031
(0.035)
-0.174*
(0.047)
-0.079
(0.041)
-0.015*
(0.005)
-1.422*
(0.128)
-2.503*
(0.185)
-0.085
(0.070)
0.013
(0.030)
-0.107*
(0.030)
-0.261*
(0.034)
1.266*
(0.316)
-5012.6
8814
Probit estimates of the likelihood of a challenger contesting an incumbent in a state house election from 1991 - 2010. Unlike Table 2.1, estimations
control for the logged average amount of contributions to winning candidates in a given year and state where data is available.
Log-Likelihood
N
Constant
Logged Average Amt. to Win Race (Year Average)
Midterm Election Dummy
Member of the Democratic Party
Incumbent Previously Contested Dummy
Prev. Incumbent Vote Share
District Pres. Vote of Incumbent’s Party
Terms Served
Freshman Incumbent
First Election after Redistricting Dummy
Term Limits Enacted
Logged District Size
Off Year Election
Southern Dummy
Professionalism
Minority Party Seat Share
Variable
Change Annual Log Q2 State Personal Inc.
Table A.1: Challenger Entry as a function of Institutional and Political Contexts: Control for Campaign Costs
94
∗p
Pres. Pty
-0.005*
(0.001)
-2.787*
(0.478)
1.228*
(0.109)
0.647*
(0.115)
-0.527*
(0.029)
0.068
(0.054)
0.065*
(0.021)
-0.026
(0.025)
-0.074*
(0.030)
-0.147*
(0.027)
-0.009*
(0.004)
-2.999*
(0.106)
-0.208*
(0.042)
0.053*
(0.021)
0.071*
(0.021)
1.926*
(0.209)
-11171.6
19563
˜Pres. Pty
0.005*
(0.001)
-1.741*
(0.436)
1.502*
(0.098)
0.665*
(0.098)
-0.525*
(0.028)
0.211*
(0.052)
0.042*
(0.018)
-0.087*
(0.024)
0.069*
(0.026)
-0.101*
(0.026)
-0.010*
(0.003)
-2.786*
(0.094)
-0.087*
(0.036)
0.076*
(0.022)
-0.161*
(0.021)
1.217*
(0.191)
-12832.1
22471
Gov. Pty
0.000
(0.001)
-3.364*
(0.448)
1.302*
(0.098)
0.820*
(0.097)
-0.481*
(0.028)
0.085
(0.052)
0.030
(0.019)
0.005
(0.025)
-0.036
(0.027)
-0.151*
(0.027)
-0.012*
(0.003)
-2.999*
(0.098)
-0.186*
(0.037)
0.030
(0.020)
-0.022
(0.021)
1.961*
(0.200)
-12135.9
21341
≤ .05; Standard Errors in Parentheses
All Inc.
0.000
(0.001)
-2.145*
(0.319)
1.362*
(0.072)
0.640*
(0.074)
-0.528*
(0.020)
0.132*
(0.037)
0.055*
(0.014)
-0.055*
(0.017)
0.002
(0.019)
-0.122*
(0.019)
-0.009*
(0.002)
-2.862*
(0.070)
-0.140*
(0.027)
0.055*
(0.014)
-0.037*
(0.015)
1.536*
(0.140)
-24126.3
42034
˜Gov. Pty
0.000
(0.001)
-0.709
(0.458)
1.443*
(0.108)
0.451*
(0.114)
-0.581*
(0.030)
0.188*
(0.054)
0.073*
(0.020)
-0.118*
(0.024)
0.047
(0.027)
-0.092*
(0.027)
-0.007
(0.003)
-2.724*
(0.101)
-0.093*
(0.039)
0.082*
(0.02)
-0.057*
(0.021)
1.151*
(0.199)
-11963.5
20693
Maj Pty
-0.002*
(0.001)
-2.182*
(0.399)
1.573*
(0.089)
0.553*
(0.094)
-0.557*
(0.026)
0.065
(0.048)
0.081*
(0.018)
-0.101*
(0.022)
0.035
(0.024)
-0.105*
(0.024)
-0.006*
(0.003)
-2.899*
(0.088)
-0.132*
(0.033)
0.094*
(0.020)
0.037
(0.019)
1.326*
(0.179)
-15011.0
26484
˜Maj. Pty
0.004*
(0.001)
-1.926*
(0.534)
1.105*
(0.142)
0.700*
(0.124)
-0.495*
(0.033)
0.239*
(0.061)
0.017
(0.022)
0.029
(0.028)
-0.069*
(0.031)
-0.146*
(0.031)
-0.016*
(0.004)
-2.877*
(0.117)
-0.192*
(0.047)
-0.021
(0.024)
-0.151*
(0.024)
1.894*
(0.228)
-9044.4
15550
Probit estimates of the likelihood of a challenger contesting an incumbent in a state house election from 1981 - 2010. Unlike Table 2.1, estimations do
not control for district level presidential vote, permitting estimations on data since 1981.
Log-Likelihood
N
Constant
Midterm Election Dummy
Member of the Democratic Party
Incumbent Previously Contested Dummy
Prev. Incumbent Vote Share
Terms Served
Freshman Incumbent
First Election after Redistricting Dummy
Term Limits Enacted
Logged District Size
Off Year Election
Southern Dummy
Professionalism
Minority Party Seat Share
Change Annual Log Q2 State Personal Inc.
Variable
Q2 Presidential Approval
Table A.2: Challenger Entry as a function of Institutional and Political Contexts: 1981 - 2010
95
∗p
Pres. Pty
-4.182*
(0.585)
-0.011*
(0.001)
1.250*
(0.134)
0.479*
(0.154)
-0.591*
(0.035)
0.120
(0.064)
0.136*
(0.026)
0.002
(0.027)
0.016
(0.038)
-0.087*
(0.032)
-0.008*
(0.004)
-1.433*
(0.092)
-2.327*
(0.144)
-0.011
(0.054)
0.215*
(0.028)
0.094*
(0.024)
1.654*
(0.254)
-8007.2
14321
˜Pres. Pty
-2.015*
(0.605)
0.000
(0.001)
1.066*
(0.137)
0.512*
(0.150)
-0.572*
(0.0350)
0.218*
(0.067)
0.074*
(0.026)
-0.095*
(0.027)
0.175*
(0.035)
-0.073*
(0.032)
-0.008*
(0.004)
-0.903*
(0.090)
-2.592*
(0.143)
-0.088
(0.053)
-0.041
(0.029)
-0.198*
(0.024)
1.763*
(0.250)
-8223.6
14115
0.792*
(0.139)
0.495*
(0.149)
-0.511*
(0.033)
0.061
(0.066)
0.100*
(0.026)
0.014
(0.027)
-0.080*
(0.030)
-0.132*
(0.032)
-0.010*
(0.004)
-1.104*
(0.091)
-2.921*
(0.146)
-0.228*
(0.054)
-0.027
(0.024)
-0.012
(0.023)
2.202*
(0.252)
-7951.8
13820
Gov. Pty
-5.387*
(0.610)
≤ .05; Standard Errors in Parentheses
1.105*
(0.095)
0.496*
(0.106)
-0.580*
(0.025)
0.146*
(0.046)
0.100*
(0.018)
-0.053*
(0.019)
-0.016
(0.021)
-0.083*
(0.023)
-0.008*
(0.003)
-1.153*
(0.064)
-2.493*
(0.101)
-0.069
(0.037)
0.091*
(0.017)
-0.051*
(0.016)
1.568*
(0.173)
-16392.4
28436
All Inc.
-2.821*
(0.416)
1.511*
(0.133)
0.423*
(0.154)
-0.678*
(0.036)
0.270*
(0.064)
0.096*
(0.026)
-0.127*
(0.026)
0.043
(0.030)
-0.023
(0.032)
-0.006
(0.004)
-1.302*
(0.093)
-2.117*
(0.140)
0.055
(0.052)
0.205*
(0.025)
-0.093*
(0.023)
1.084*
(0.242)
-8367.5
14616
˜Gov. Pty
-0.824
(0.586)
1.184*
(0.120)
0.565*
(0.139)
-0.622*
(0.032)
0.048
(0.060)
0.104*
(0.024)
-0.130*
(0.024)
-0.026
(0.027)
-0.079*
(0.029)
-0.006
(0.003)
-1.200*
(0.082)
-2.504*
(0.128)
-0.062
(0.047)
0.071*
(0.023)
0.008
(0.021)
1.578*
(0.224)
-10054.1
17501
Maj Pty
-2.816*
(0.526)
1.526*
(0.185)
0.377*
(0.169)
-0.510*
(0.039)
0.290*
(0.072)
0.110*
(0.029)
0.067*
(0.031)
0.000
(0.035)
-0.085*
(0.036)
-0.014*
(0.005)
-1.428*
(0.115)
-2.505*
(0.164)
-0.104
(0.063)
0.044
(0.027)
-0.145*
(0.026)
1.466*
(0.284)
-6263.7
10935
˜Maj. Pty
-2.902*
(0.685)
Probit estimates of the likelihood of a challenger contesting an incumbent in a state house election from 1991 - 2010. Data subset to races where the
incumbent’s party received 60% or less of the presidential vote within the district.
Log-Likelihood
N
Constant
Midterm Election Dummy
Member of the Democratic Party
Incumbent Previously Contested Dummy
Prev. Incumbent Vote Share
District Pres. Vote of Incumbent’s Party
Terms Served
Freshman Incumbent
First Election after Redistricting Dummy
Term Limits Enacted
Logged District Size
Off Year Election
Southern Dummy
Professionalism
Minority Party Seat Share
Q2 Presidential Approval
Variable
Change Annual Log Q2 State Personal Inc.
Table A.3: Challenger Entry as a function of Institutional and Political Contexts: 1981 - 2010
Table A.4: Challenger Entry as a function of Policy Responsiveness, Institutional,
and Political Contexts
Variable
Policy Responsivness
Minority Party Seat Share
Professionalism
Southern Dummy
Logged District Size
Term Limits Enacted
Freshman Incumbent
Terms Served
District Pres. Vote of Incumbent’s Party
Prev. Incumbent Vote Share
Incumbent Previously Contested Dummy
Member of the Democratic Party
Midterm Election Dummy
Constant
Log-Likelihood
N
∗p
All States
-0.005
(0.018)
2.884*
(0.305)
3.767*
(0.323)
-0.365*
(0.066)
-0.351*
(0.048)
0.087*
(0.043)
-0.072
(0.053)
0.002
(0.006)
-0.614*
(0.164)
-2.783*
(0.225)
-0.174*
(0.081)
-0.216*
(0.038)
0.009
(0.037)
4.574*
(0.423)
-3204.8
5662
Above Avg. Economies
-0.054*
(0.024)
2.398*
(0.440)
3.442*
(0.480)
-0.466*
(0.089)
-0.322*
(0.07)
-0.034
(0.057)
-0.128
(0.068)
0.001
(0.009)
-0.665*
(0.213)
-3.567*
(0.302)
-0.460*
(0.110)
-0.238*
(0.050)
0.011
(0.058)
5.478*
(0.619)
-1814.1
3172
Below Avg. Economies
0.070*
(0.029)
2.738*
(0.505)
3.672*
(0.471)
-0.071
(0.109)
-0.355*
(0.074)
0.521*
(0.081)
0.066
(0.091)
0.002
(0.008)
-0.704*
(0.263)
-1.499*
(0.349)
0.198
(0.125)
-0.184*
(0.06)
0.157*
(0.075)
3.286*
(0.644)
-1340.5
2490
≤ .05; Standard Errors in Parentheses
Probit estimates of the likelihood of a challenger contesting an incumbent in 2006 and 2008 state
house elections. Estimations include measure of “policy responsiveness” constructed from a state
legislator’s ideal point and Tausanovich and Warshaw’s measure of state legislative district ideology.
First column includes all 2006 and 2008 state house elections. Second and third columns subset
data by legislators from states with above and below average economies. Unresponsive legislative
behavior only increases the likelihood of a major party challenger during bad economies.
96
97
The top left panel of the above Figure illustrates the baseline average predicted probability of a challenger emerging against any incumbent (top circle)
or party subsets (other circles), as indicated by the Y-Axis. These probabilities are the respective zero points (black vertical lines) in other panels
illustrating differences in probabilities. Differences in average predicted probabilities follow the shift indicated at the top of each panel for each
observation with other variables held at true values. Horizontal black lines are 95% confidence intervals. The range of probabilities along the X-axis is
always .2, and a dotted grey line indicates the corresponding probability for “All Incumbents.”
Figure A.1: Probabilities of Challenger Emergence Across Parties
98
The first row of panels presents the difference in predicted probability of a challenger for all incumbents (top circle) or party subsets (other circles) that
corresponds with increases in income growth or presidential approval. White circles indicate the estimated difference is indistinguishable from zero.
Probabilities in the second row use estimates from more competitive districts. Incumbents are less likely to be challenged during prosperous economies,
but when comparing the leftmost panels, this relationship is stronger in more competitive districts. Members of the president’s party furthermore are
the most likely to be challenged during weak economies or unpopular presidencies.
Figure A.2: Probabilities of Challenger Emergence under Different Economic and Political Conditions
Appendix B
Appendix for Chapter 3
B.1
Supplementary Estimates for Seat Change Analysis
99
Table B.1: Democratic State House Seat Change as a Function of Economic Variables
Dependent Variable: Democratic Seat Change
National Economy x President’s Party
National Economy X Governor’s Party
National Economy X State House Party
National Economy
0.812*
(0.128)
0.015
(0.129)
-0.089
(0.137)
State Economy X President’s Party
State Economy X Governor’s Party
State Economy X State House Party
Previous Seat Change
Congressional Vote Change
President’s Party Dummy
Governor’s Party Dummy
State House Party Dummy
National Economy
-0.253*
(0.063)
0.183*
(0.037)
-0.041*
(0.004)
-0.004
(0.003)
-0.020*
(0.004)
-0.318*
(0.132)
State Economy
Constant
-0.046*
(0.022)
0.355
861
R-Squared
N
∗p
State Economy
0.352*
(0.100)
-0.042
(0.106)
-0.127
(0.110)
-0.259*
(0.062)
0.206*
(0.038)
-0.032*
(0.003)
-0.003
(0.003)
-0.020*
(0.004)
-0.161
(0.089)
-0.046
(0.026)
0.333
861
National and State
0.831*
(0.184)
0.156
(0.198)
-0.049
(0.186)
-0.017
(0.128)
-0.136
(0.144)
-0.047
(0.158)
-0.253*
(0.063)
0.182*
(0.038)
-0.041*
(0.004)
-0.004
(0.003)
-0.020*
(0.004)
-0.258
(0.189)
-0.066
(0.134)
-0.045*
(0.023)
0.357
861
≤ .05; Robust Standard Errors in Parentheses
OLS estimates of State House Democratic Party Seat Change as a function of economy variables
interacted with the parties that controlled the presidency, governorship, and state house. Observations
at the state-year level and include state fixed effects. Results are consistent with those in Tables 3.1,
3.2, and 3.3.
100
101
-0.301*
(0.062)
0.255*
(0.038)
0.010
(0.017)
0.221
867
Pres. Pty
-0.003*
(0.001)
∗p
-0.183*
(0.066)
0.310*
(0.037)
-0.009
(0.022)
0.220
867
St. Hse Pty
-0.001
(0.001)
-0.251*
(0.071)
0.267*
(0.040)
0.006
(0.015)
0.207
780
-0.002
(0.001)
Pres. Pty
-0.204*
(0.073)
0.302*
(0.042)
0.008
(0.018)
0.188
774
0.000
(0.001)
Gov. Pty
-0.132
(0.068)
0.306*
(0.036)
-0.015
(0.014)
0.239
780
0.000
(0.001)
St. Hse Pty
≤ .05; Robust Standard Errors in Parentheses
-0.245*
(0.063)
0.299*
(0.041)
0.016
(0.018)
0.196
861
Gov. Pty
-0.001
(0.001)
-0.020
(0.054)
-0.306*
(0.056)
0.264*
(0.040)
-0.007
(0.012)
0.218
867
Pres. Pty
-0.079
(0.054)
-0.245*
(0.059)
0.300*
(0.044)
0.012
(0.013)
0.197
861
Gov. Pty
-0.011
(0.052)
-0.183*
(0.052)
0.310*
(0.048)
-0.012*
(0.005)
0.220
867
St. Hse Pty
OLS estimates of seat change regressed on changes in national unemployment, state unemployment, and state GDP. Party of dependent variable
indicated by the column heading. Observations are at the state-year level and estimations include state fixed effects.
R-Squared
N
Constant
Congressional Vote Change
Previous Seat Change
Change in Logged State GDP
State Unemployment
Dep. Var. (Seat Change):
National Unemployment
Table B.2: Economic Models using Unemployment and State GDP
102
Pres Pty
National
1.236*
(0.209)
-1.344*
(0.606)
-0.312*
(0.053)
0.220*
(0.040)
0.086*
(0.043)
-0.050*
(0.017)
0.270
867
∗p
St. Hse Pty
National
-0.276
(0.260)
0.552
(0.679)
-0.185*
(0.051)
0.310*
(0.048)
-0.068
(0.040)
0.008
(0.012)
0.222
867
Pres Pty
State
0.322*
(0.157)
0.498
(0.530)
-0.310*
(0.056)
0.248*
(0.040)
0.025
(0.046)
-0.023
(0.017)
0.241
867
Gov. Pty
State
-0.096
(0.146)
0.269
(0.520)
-0.243*
(0.060)
0.298*
(0.044)
0.027
(0.035)
0.001
(0.015)
0.196
861
St. Hse Pty
State
-0.039
(0.154)
0.091
(0.507)
-0.184*
(0.052)
0.310*
(0.048)
-0.056
(0.041)
0.003
(0.012)
0.22
867
≤ .05; Robust Standard Errors in Parentheses
Gov. Pty
National
0.182
(0.241)
-0.747
(0.669)
-0.244*
(0.060)
0.299*
(0.044)
0.048
(0.033)
-0.005
(0.015)
0.196
861
Pres Pty
Relative
0.002
(0.189)
0.139
(0.959)
-0.304*
(0.056)
0.263*
(0.040)
0.032
(0.045)
-0.017
(0.018)
0.218
867
Gov. Pty
Relative
-0.307
(0.211)
1.362
(1.063)
-0.240*
(0.060)
0.298*
(0.044)
0.023
(0.034)
0.002
(0.015)
0.198
861
St. Hse Pty
Relative
0.053
(0.206)
0.188
(1.042)
-0.182*
(0.052)
0.309*
(0.048)
-0.057
(0.039)
0.003
(0.012)
0.221
867
OLS estimates of seat change regressed on economic variables while controlling for professionalism. State house majority parties from professionalized
legislatures are not more likely to receive a reward for economic prosperity. The president’s state legislative party receives less of a reward in
professionalized legislatures, consistent with the findings of Berry, Berkman and Schneiderman (2000). However, even in the most professionalized
legislature, California, national income growth of 2.5% results in over a 1% increase in seats for the president’s party. Observations are at the
state-year level and estimations include state fixed effects. Party of dependent variable indicated by the column heading.
R-Squared
N
Constant
Professionalism
Congressional Vote Change
Previous Seat Change
Economy X Professionalism
Dependent Variable
Economy Variable
Economy
Table B.3: Party Seat Change as a Function of Economic Performance Controlling for Professionalism
103
Pres Pty
National
0.992*
(0.176)
-0.136
(0.249)
-0.312*
(0.058)
0.220*
(0.038)
0.002
(0.006)
-0.028*
(0.014)
0.267
867
∗p
St. Hse Pty
National
-0.128
(0.214)
-0.048
(0.281)
-0.183*
(0.052)
0.307*
(0.047)
0.008
(0.008)
-0.014
(0.008)
0.222
867
Pres Pty
State
0.531*
(0.120)
-0.234
(0.186)
-0.310*
(0.061)
0.248*
(0.038)
0.005
(0.006)
-0.017
(0.016)
0.242
867
Gov. Pty
State
-0.076
(0.160)
0.069
(0.244)
-0.219*
(0.064)
0.298*
(0.039)
0.025*
(0.007)
-0.005
(0.017)
0.219
861
St. Hse Pty
State
-0.096
(0.152)
0.146
(0.212)
-0.182*
(0.053)
0.307*
(0.047)
0.004
(0.006)
-0.015*
(0.007)
0.222
867
≤ .05; Robust Standard Errors in Parentheses
Gov. Pty
National
-0.151
(0.207)
0.357
(0.298)
-0.219*
(0.064)
0.298*
(0.039)
0.020*
(0.007)
-0.004
(0.018)
0.221
861
Pres Pty
Relative
0.043
(0.199)
-0.030
(0.281)
-0.305*
(0.061)
0.264*
(0.038)
-0.001
(0.006)
-0.008
(0.016)
0.218
867
Gov. Pty
Relative
0.009
(0.242)
-0.137
(0.338)
-0.217*
(0.064)
0.298*
(0.04)
0.027*
(0.006)
-0.006
(0.017)
0.220
861
St. Hse Pty
Relative
-0.061
(0.233)
0.241
(0.335)
-0.181*
(0.053)
0.307*
(0.047)
0.007
(0.004)
-0.016*
(0.007)
0.222
867
OLS estimates of seat change regressed on economic variables while controlling for divided state government. State parties from unified government
are not more likely to receive a reward for economic prosperity. Observations are at the state-year level and estimations include state fixed effects.
Column headings list the party of the dependent variable and the economic measure used (e.g change in logged national real disposable income).
R-Squared
N
Constant
Divided Government
Congressional Vote Change
Previous Seat Change
Economy X Divided Government
Dependent Variable
Economy Variable
Economy
Table B.4: Party Seat Change as a Function of Economic Performance Controlling for Divided State Government
104
Pres Pty
National
0.894*
(0.128)
0.075
(0.238)
-0.376*
(0.140)
0.214
(0.190)
0.116*
(0.038)
0.165*
(0.059)
-0.010
(0.009)
-0.026
(0.016)
0.290
867
∗p
Pres Pty
Relative
0.036
(0.190)
-0.071
(0.291)
-0.354*
(0.140)
0.200
(0.194)
0.167*
(0.040)
0.137*
(0.060)
-0.014
(0.008)
-0.005
(0.017)
0.242
867
Gov Pty
National
-0.203
(0.147)
0.489
(0.313)
-0.243*
(0.117)
0.017
(0.185)
0.158*
(0.042)
0.240*
(0.069)
-0.009
(0.007)
0.015
(0.018)
0.212
861
Gov Pty
State
-0.128
(0.136)
0.171
(0.276)
-0.242*
(0.118)
0.016
(0.186)
0.161*
(0.043)
0.236*
(0.071)
-0.003
(0.007)
0.013
(0.018)
0.211
861
Gov Pty
Relative
-0.072
(0.198)
-0.074
(0.352)
-0.244*
(0.118)
0.016
(0.187)
0.161*
(0.042)
0.237*
(0.070)
0.001
(0.005)
0.009
(0.019)
0.210
861
≤ .05; Robust Standard Errors in Parentheses
Pres Pty
State
0.388*
(0.136)
0.000
(0.213)
-0.366*
(0.141)
0.205
(0.192)
0.155*
(0.040)
0.141*
(0.059)
-0.012
(0.009)
-0.012
(0.017)
0.262
867
St. Hse Pty
National
-0.242
(0.133)
0.223
(0.309)
-0.188
(0.103)
0.018
(0.178)
0.178*
(0.037)
0.218*
(0.066)
-0.010
(0.009)
-0.008
(0.011)
0.235
867
St. Hse Pty
State
-0.124
(0.108)
0.270
(0.202)
-0.192
(0.103)
0.029
(0.176)
0.180*
(0.037)
0.218*
(0.066)
-0.010
(0.007)
-0.010
(0.011)
0.236
867
St. Hse Pty
Relative
-0.034
(0.161)
0.320
(0.261)
-0.191
(0.104)
0.027
(0.178)
0.176*
(0.038)
0.221*
(0.067)
-0.005
(0.006)
-0.013
(0.011)
0.236
867
OLS estimates of seat change regressed on economic variables while controlling for midterm elections. Column headings list the party of the dependent
variable and the economic measure used (e.g change in logged national real disposable income). There is little difference between the impact of
national conditions on the president’s party’s seat change in state legislatures during midterm elections.
R-Squared
N
Constant
Midterm Election Dummy
Cong. Vote Change X Midterm
Congressional Vote Change
Previous Seat Change X Midterm
Previous Seat Change
Economy X Midterm
Seat Change for:
Economy Variable
Economy
Table B.5: Party Seat Change as a Function of Economic Performance Controlling for Midterm Elections
105
$20k - $40k
0.224
0.223
0.21
0.239
0.212
0.241
0.221
< $20,000
0.093
0.103
0.079
0.176
0.108
0.173
0.19
Not Interested
0.072
0.033
0.052
0.047
0.038
0.099
Strong R
0.187
0.239
0.202
0.14
Much Worse
0.137
0.568
0.782
0.663
0.23
0.231
Strongly Dis.
0.494
0.515
0.4
0.616
0.457
0.219
Income
2006 CCES
2008 CCES
2008 Sept. Panel ANES
2008 Time Series ANES
2010 CCES
2010 EGSS ANES
2009 Census/BLS
Political Interest
2006 CCES
2008 CCES
2008 Sept. Panel ANES
2008 Time Series ANES
2010 CCES
2010 EGSS ANES
Party ID
2006 CCES
2008 CCES
2008 Sept. Panel ANES
2008 Time Series ANES
2010 CCES
2010 EGSS ANES
National Economy
2006 CCES
2008 CCES
2008 Sept. Panel ANES
2008 Time Series ANES
2010 CCES
2010 EGSS ANES
Presidential Approval
2006 CCES
2008 CCES
2008 Sept. Panel ANES
2008 Time Series ANES
2010 CCES
2010 EGSS ANES
2008 Gallup (Oct. 10)
2010 Gallup (Oct. 18)
Somewhat Dis.
0.095
0.154
0.189
0.154
0.084
0.18
Worse
0.279
0.357
0.159
0.226
0.293
0.198
Republican
0.126
0.098
0.166
0.108
0.089
0.11
Slightly Int.
0.289
0.086
0.143
0.129
0.088
0.157
High School
0.312
0.348
0.16
0.332
0.199
0.298
0.31
Some HS
0.033
0.035
0.035
0.148
0.019
0.128
0.137
Education
2006 CCES
2008 CCES
2008 Sept. Panel ANES
2008 Time Series ANES
2010 CCES
2010 EGSS ANES
2010 Census
Not Sure/Niether
0.012
0.015
0.194
0.02
0.239
Not Sure
0.224
0.061
0.053
0.086
0.269
0.414
Lean R.
0.111
0.101
0.157
0.094
0.143
0.195
Moderately Int.
0.233
0.365
0.298
0.199
0.323
$40k - $60k
0.217
0.213
0.177
0.208
0.204
0.188
0.17
Some College
0.304
0.238
0.366
0.199
0.283
0.183
0.193
Somewhat App.
0.217
0.255
0.163
0.133
0.228
0.256
Better
0.205
0.009
0.002
0.019
0.194
0.143
Ind./Not Sure
0.113
0.102
0.318
0.339
0.112
0.027
Very Int.
0.639
0.648
0.32
0.354
0.675
0.265
$60k - $100k
0.273
0.264
0.288
0.253
0.271
0.253
0.216
Assoc. Degree
0.107
0.072
0.112
0.09
0.074
0.086
Strongly App.
0.182
0.062
0.053
0.096
0.211
0.106
Much Better
0.155
0.005
0.003
0.006
0.014
0.014
Lean D.
0.137
0.1
0.139
0.185
0.094
0.207
Extremely Int.
0.12
0.171
0.156
> $100,000
0.193
0.198
0.246
0.123
0.204
0.145
0.202
College Degree
0.148
0.207
0.245
0.144
0.284
0.198
0.18
Total Dis.
0.589
0.669
0.589
0.77
0.541
0.399
0.71
0.48
Democrat
0.13
0.103
0.22
0.273
0.11
0.148
Grad Degree
0.097
0.1
0.195
0.066
0.125
0.12
0.092
Table B.6: Comparisons of the CCES and ANES in 2008 and 2010
Total App.
0.399
0.317
0.216
0.229
0.439
0.362
0.25
0.44
Strong D.
0.197
0.258
0.251
0.172
Total Respondents
36421
32559
2628
2322
55248
1229
B.2
Supplementary Estimates for Survey Analysis
Table B.7: Weighted Multinomial Probit. DV: Reported State House Vote Choice
Base Outcome | Outcome
State Legislative Approval
-1 | 0
0.080*
(0.021)
-1 | 1
0.124*
(0.020)
State Legislative Approval (Belief)
Governor Approval
0.111*
(0.018)
PID (7pt)
Intercept
Log psuedolikelihood
N
∗p
0.238*
(0.023)
0.343*
(0.020)
-0.963*
(.035)
-8404.63
19522
-1 | 1
0.113*
(0.022)
0.177*
(0.024)
0.111*
(0.018)
0.235*
(0.023)
0.339*
(0.015)
-0.959*
(0.0354)
-8394.71
19522
0.209*
(0.018)
0.463*
(0.022)
0.718*
(0.013)
-0.365*
(0.036)
0.204*
(0.017)
Governor Approval (Belief)
Presidential Approval
-1 | 0
0.471*
(0.022)
0.727*
(0.014)
-0.377*
(0.036)
≤ .05; Standard Errors in Parentheses
Each the 2008, and 2010 surveys asked “For whom did you vote for in the state legislative election”
in the respondent’s lower chamber. In 2008, individuals could select a“Not Sure” response, but in
2010, this option was unavailable. The above replicates Table 3.4 estimates for 2008 but includes
“Not Sure” responses in a weighted multinomial probit estimation.
106
Table B.8: Democratic State House Vote Choice as a Function of Approval Ratings and
Party ID - Remove Not Sure Responses
Election Year:
State Legislative Approval
2008
0.146*
(0.026)
State Leg. Approval x State House Majority Party Belief
Governor Approval
2008
0.165*
(0.027)
0.201*
(0.025)
Governor Approval x Governor’s Party Belief
Presidential Approval
0.540*
(0.032)
0.456*
(0.013)
0.086*
(0.034)
-2907.9
10837
Party ID (7 pt)
Constant
Log-pseudolikelihood
N
∗p
2010
0.124*
(0.027)
2010
0.155*
(0.026)
0.152*
(0.024)
0.188*
(0.025)
0.533*
(0.032)
0.453*
(0.013)
0.091*
(0.033)
-2911.7
10837
0.660*
(0.031)
0.440*
(0.017)
0.576*
(0.05)
-3807.2
19681
0.165*
(0.026)
0.648*
(0.032)
0.437*
(0.017)
0.592*
(0.047)
-3777.6
19681
≤ .05; Standard Errors in Parentheses
Weighted probit estimates of state house vote choice as a function of voters’ assessments of political
actors and partisan identification, similar to those in Table 3.4. These data from the Cooperative
Congressional Election Studies, but do not include respondents who answered “Not sure” to any
relevant knowledge, evaluation, or vote choice question.
107
108
∗p
2008
Income
0.245*
(0.050)
0.036*
(0.015)
0.135*
(0.039)
-0.001
(0.011)
0.035
(0.045)
0.014
(0.012)
0.526*
(0.029)
-0.008
(0.008)
-0.077*
(0.024)
-0.031
(0.081)
-4604
17164
2008
Educ.
0.297*
(0.044)
0.015
(0.012)
0.126*
(0.039)
0.002
(0.010)
0.050
(0.045)
0.007
(0.011)
0.552*
(0.027)
-0.014*
(0.007)
-0.016
(0.019)
-0.228*
(0.073)
-4924.1
18182
2010
Educ.
0.477*
(0.046)
-0.004
(0.011)
0.093*
(0.042)
0.003
(0.010)
0.062
(0.046)
-0.001
(0.010)
0.514*
(0.039)
-0.012
(0.009)
-0.005
(0.017)
0.094
(0.078)
-5196.5
27769
2008
Interest
0.223*
(0.063)
0.052*
(0.024)
0.104
(0.059)
0.012
(0.021)
-0.009
(0.066)
0.032
(0.024)
0.572*
(0.038)
-0.029*
(0.014)
-0.086*
(0.037)
-0.075
(0.098)
-4894.5
18125
2010
Interest
0.347*
(0.057)
0.047*
(0.0210)
0.103
(0.061)
0.001
(0.022)
0.053
(0.066)
0.001
(0.024)
0.702*
(0.048)
-0.093*
(0.018)
-0.074*
(0.033)
0.262*
(0.088)
-5114.3
27648
≤ .05; Standard Errors in Parentheses
2010
Income
0.374*
(0.043)
0.028*
(0.012)
0.123*
(0.043)
-0.008
(0.012)
0.056
(0.049)
0.001
(0.013)
0.547*
(0.038)
-0.026*
(0.011)
-0.034
(0.020)
0.197*
(0.071)
-4649.9
24265
2008
Prof.
0.267*
(0.032)
0.318*
(0.110)
0.178*
(0.025)
-0.158
(0.084)
0.002
(0.029)
0.285*
(0.096)
0.487*
(0.019)
0.066
(0.067)
-0.033
(0.183)
-0.264*
(0.051)
-4903.8
18182
2010
Prof.
0.429*
(0.030)
0.103
(0.093)
0.160*
(0.028)
-0.178*
(0.079)
0.058
(0.030)
0.020
(0.114)
0.408*
(0.025)
0.226*
(0.079)
0.262
(0.198)
0.001
(0.048)
-5182
27769
2008
Div. Gov.
0.343*
(0.020)
0.004
(0.036)
0.124*
(0.016)
0.038
(0.028)
0.092*
(0.018)
-0.072*
(0.032)
0.523*
(0.012)
-0.055*
(0.021)
-0.077
(0.058)
-0.255*
(0.032)
-4914.6
18182
2010
Div. Gov.
0.478*
(0.018)
-0.066
(0.036)
0.106*
(0.016)
0.001
(0.036)
0.055*
(0.018)
0.009
(0.040)
0.470*
(0.016)
0.012
(0.032)
-0.014
(0.055)
0.078*
(0.031)
-5193.3
27769
Estimates follow the model from the main text but include interactions and controls for a respondent’s income, education, political interest and
controls for a state’s legislative professionalism or divided state government. The first three controls are measured categorically for each voter 1-5.
Professionalism is Squire’s index. Divided state government is a dummy variable. Professionalism increases the impact of national conditions, and
this result holds in using data from the 2006 CCES (not shown) but the 2010 estimate is not significant. Including the party ID and professionalism
interaction reduces the magnitude of this coefficient, possibly attributable to the relationship between approval and ideology in professional state
legislatures amongst conservatives, or those already opposed to Obama (Richardson, Konisky, and Milyo 2011). Estimates from 2008, however, suggest
state legislative approval matters more in professional legislatures. Results from 2010 are in the same direction but insignificant.
Log-pseudolikelihood
N
Constant
Control Variable (Listed in Column Heading)
Party ID x Control
Party ID (7 pt)
State Leg Approval x Control
State Legislative Approval
Governor Approval x Control
Governor Approval
Presidential Approval x Control
Year
Control Variable
Presidential Approval
Table B.9: State House Vote Choice as a Function of Approval and Party ID, with controls
109
∗p
1973
0.216*
(0.060)
0.015
(0.068)
0.005
(0.078)
1.492*
(0.111)
-0.232*
(0.099)
0.141
(0.098)
-211.444
446
1979
0.252*
(0.074)
0.088
(0.067)
0.080
(0.069)
1.804*
(0.113)
-0.292*
(0.106)
-0.215*
(0.105)
-220.603
638
1983
0.210*
(0.064)
0.094
(0.066)
-0.033
(0.075)
1.422*
(0.118)
-0.048
(0.088)
0.109
(0.088)
-192.250
415
1985
0.323*
(0.082)
-0.012
(0.102)
0.151
(0.099)
1.719*
(0.168)
-0.052
(0.134)
0.043
(0.134)
-92.577
323
≤ .05; Standard Errors in Parentheses
1975
0.218*
(0.050)
0.202*
(0.064)
0.013
(0.063)
1.414*
(0.101)
-0.498*
(0.082)
-0.262*
(0.080)
-353.944
654
1987
0.180*
(0.059)
0.146*
(0.068)
0.022
(0.069)
1.330*
(0.106)
-0.098
(0.095)
-0.057
(0.095)
-195.878
509
1995
0.423*
(0.078)
0.229*
(0.085)
-0.017
(0.095)
1.515*
(0.137)
-0.267*
(0.103)
-0.182
(0.102)
-127.763
461
2007
0.342*
(0.095)
0.111
(0.083)
0.139
(0.087)
1.368*
(0.103)
-0.051
(0.094)
0.282*
(0.096)
-251.300
523
Probit estimates of New Jersey state assembly vote choice as a function of approval variables and party ID accounting for question wording differences.
In each Eagleton survey, respondents were asked a 3-pt Party ID question. CCES asked a 7-pt. For comparability in Table 3.4, I then code Democrats
2, Independents 0, and Republicans -2. Similarly, the CCES also asked for approval ratings on a Strongly Disapprove to Strongly Approve scale. I code
these from -2 to 2. With Eagleton in 2007, voters could only respond Disapprove or Approve. To match CCES, I code these as -1.5 and 1.5 in Table 3.
In other years, Eagleton commonly asked ask ‘how good a job’ institutions were doing where respondents could respond Poor, Only Fair, No Opinion,
Good, and Excellent. I code these from -2 to 2 to match CCES. The above table presents results without these adjustments for the CCES (e.g. Party ID
on a -1 to 1 scale, and approval coded 0 or 1).
Log-Likelihood
N
Intercept: Split Votes | D Votes
Intercept: R Votes | Split Votes
Party ID
State Legislative Approval
Governor Approval
Election Year:
Presidential Approval
Table B.10: NJ and VA Off-Year State Legislative Voting as a function of Approval Ratings and Party ID since the 1970s - Alternative
Specification
Table B.11: 2010 Democratic State House Vote Choice as a Function of Approval Ratings
and Party ID - Correct and Incorrect Respondents
Election Year:
State Legislative Approval
Governor Approval
Presidential Approval
Party ID (7 pt)
Constant
Log-pseudolikelihood
N
∗p
Correct Respondents
0.116*
(0.021)
0.097*
(0.017)
0.470*
(0.021)
0.442*
(0.018)
0.099*
(0.035)
-3470.2
18477
Incorrect Respondents
-0.144*
(0.054)
0.117*
(0.044)
0.384*
(0.048)
0.45*
(0.039)
-0.107
(0.077)
-479.9
2134
≤ .05; Standard Errors in Parentheses
Probit estimates of state house vote choice as a function of voters’ assessments of political actors and
partisan identification. Estimates are used to create Figure 3.3. These data from the 2010 Cooperative
Congressional Election Study are subset by whether a respondent correctly or incorrectly identified
the state house majority party.
110
Figure B.1: State-level relationships between Seat Change and Presidential Approval
Plots of OLS estimates and 95% confidence intervals for state level regressions of president’s party
seat change on presidential approval. Data cover the 1972 - 2011 elections. Ignoring uncertainty
of estimates, over 40 states produce a positive relationship. Points labeled US and SH represent
comparable estimates for the U.S. House and all state house seats.
111
Figure B.2: State-level relationships between Vote Choice and Presidential Approval
Plots differences in predicted probabilities for voting for a state house member of the president’s
party, drawing from probit estimates on pooled data from the 2008 and 2010 CCES. I run probit
regressions by state, and using these state level estimates, I shift the presidential approval rating
from ‘Strongly Disapprove’ to ‘Strongly Approve’ and hold other variables at their state-level means.
A positive relationship between presidential approval and state house vote choice emerges for each
state.
112
Appendix C
Appendix for Chapter 4
Table C.1: Summary Statistics of Variables
Variable
Support for Incumbent’s Roll Call (Referendum Measure)
Incumbent Vote Share
Incumbent Previous Vote Share
Contested Incumbent
Freshman Dummy
Democratic Member Dummy
Presidential Approval (Adjusted)
Ideological Extremity - State Level
Ideological Extremity - National (NPAT) Level
Incumbent Contribution Advantage
Logged Incumbent Contributions
Logged Challenger Contributions
Incumbent Party Presidential Vote
State Legislative Professionalism
Logged District Size
Min.
23.88
15.68
36.66
0.00
0.00
0.00
12.00
0.61
0.71
-6.18
4.48
0.00
11.92
0.05
8.25
Summary Statistics of Variables used in Chapter 4 Analysis
113
Max
90.90
100.00
100.00
1.00
1.00
1.00
88.00
14.54
16.35
13.61
16.07
15.02
98.14
0.63
12.09
Median
55.54
77.48
72.95
1.00
0.00
1.00
53.00
4.67
5.19
1.25
10.66
9.10
59.93
0.17
9.84
Mean
53.09
79.76
77.89
0.53
0.24
0.53
50.40
5.04
5.58
1.95
10.65
8.71
60.19
0.20
9.88
SD
12.17
19.09
19.17
0.50
0.43
0.50
13.45
2.07
2.23
2.28
1.30
2.28
13.68
0.12
0.74
C.1
Full Estimates of Individual Roll-Call Analysis
Table C.2: Full Referendum Estimates: Part 1
Variable
Support for Legislator’s Position
Inc. Party Presidential Vote
Incumbent Previous Vote Share
Incumbents Contributions Advantage
Democratic Party Dummy
Incumbent Previously Contested Dummy
Constant
R-Squared
N
∗p
AK-2000-Measure-6
0.167
(0.311)
0.733*
(0.311)
0.76
(0.45)
0.014
(0.008)
0.172
(0.121)
0.303
(0.189)
-0.713
(0.371)
0.791
16
CA-2004-Prop-72
0.188
(0.138)
0.607*
(0.104)
0.051
(0.119)
0.004
(0.003)
-0.025
(0.018)
0.04
(0.043)
0.101
(0.057)
0.891
51
CA-2008-Prop-94
0.029
(0.113)
0.134
(0.118)
0.638*
(0.127)
0.003
(0.003)
0.089*
(0.014)
0.209*
(0.053)
-0.132
(0.089)
0.937
40
CA-2008-Prop-95
0.085
(0.11)
0.153
(0.125)
0.633*
(0.136)
0.003
(0.003)
0.092*
(0.015)
0.212*
(0.054)
-0.177
(0.102)
0.934
36
≤ .05; Standard Errors in Parentheses
Contested Incumbent Reelection Vote Share as a function of District Support for Incumbent’s
Roll-Call Position and control variables. Column heading denotes referendum. See Table 4.2 for
corresponding bills.
114
Table C.3: Full Referendum Estimates: Part 2
Variable
Support for Legislator’s Position
Inc. Party Presidential Vote
Incumbent Previous Vote Share
Incumbents Contributions Advantage
Democratic Party Dummy
Incumbent Previously Contested Dummy
Constant
R-Squared
N
∗p
CA-2008-Prop-96
0.038
(0.113)
0.136
(0.117)
0.636*
(0.127)
0.003
(0.003)
0.089*
(0.014)
0.208*
(0.053)
-0.135
(0.088)
0.937
40
CA-2008-Prop-97
0.072
(0.107)
0.159
(0.128)
0.629*
(0.137)
0.003
(0.003)
0.091*
(0.015)
0.205*
(0.056)
-0.163
(0.095)
0.934
36
ID-2012-Prop-1
0.333
(0.204)
0.625*
(0.122)
0.019
(0.223)
0.005
(0.004)
-0.053
(0.062)
-0.005
(0.087)
0.058
(0.206)
0.833
28
ID-2012-Prop-2
0.159
(0.143)
0.674*
(0.12)
0.073
(0.227)
0.004
(0.004)
0.005
(0.042)
0
(0.089)
0.056
(0.215)
0.823
28
≤ .05; Standard Errors in Parentheses
Contested Incumbent Reelection Vote Share as a function of District Support for Incumbent’s
Roll-Call Position and control variables. Column heading denotes referendum. See Table 4.2 for
corresponding bills.
Table C.4: Full Referendum Estimates: Part 3
Variable
Support for Legislator’s Position
Inc. Party Presidential Vote
Incumbent Previous Vote Share
Incumbents Contributions Advantage
Democratic Party Dummy
Incumbent Previously Contested Dummy
Constant
R-Squared
N
∗p
ID-2012-Prop-3
-0.003
(0.078)
0.691*
(0.124)
0.073
(0.243)
0.005
(0.005)
0.038
(0.032)
0.001
(0.093)
0.119
(0.242)
0.812
28
ME-2005-Question-1
0.146
(0.081)
0.174
(0.123)
0.449*
(0.108)
0.026
(0.015)
0.011
(0.019)
0.148*
(0.066)
0.002
(0.12)
0.438
103
ME-2008-Question-1
0
(0.063)
0.297*
(0.09)
0.438*
(0.088)
0.005
(0.006)
0.019
(0.025)
0.156*
(0.053)
0.02
(0.096)
0.531
86
ME-2009-Question-1
0.101*
(0.046)
0.409*
(0.082)
0.411*
(0.082)
0.002
(0.005)
-0.126*
(0.019)
0.115*
(0.037)
0.029
(0.085)
0.636
104
≤ .05; Standard Errors in Parentheses
Contested Incumbent Reelection Vote Share as a function of District Support for Incumbent’s
Roll-Call Position and control variables. Column heading denotes referendum. See Table 4.2 for
corresponding bills.
115
Table C.5: Full Referendum Estimates: Part 4
Variable
Support for Legislator’s Position
Inc. Party Presidential Vote
Incumbent Previous Vote Share
Incumbents Contributions Advantage
Democratic Party Dummy
Incumbent Previously Contested Dummy
Constant
R-Squared
N
∗p
MI-2002-Prop-1
0.051
(0.208)
0.462*
(0.109)
0.26*
(0.109)
0.018*
(0.005)
-0.024
(0.038)
0.111*
(0.05)
0.061
(0.108)
0.856
51
MI-2004-Prop-3
0.086
(0.231)
0.469*
(0.125)
0.236*
(0.118)
0.005*
(0.002)
-0.042
(0.042)
0.11*
(0.055)
0.087
(0.117)
0.83
51
MI-2012-Prop-1
0.087*
(0.043)
0.644*
(0.07)
0.06
(0.054)
0.01*
(0.002)
0.036*
(0.012)
-0.03
(0.028)
0.154*
(0.037)
0.946
84
MT-2012-IR-124
-0.023
(0.13)
0.255*
(0.091)
0.679*
(0.111)
0.012*
(0.006)
0.048*
(0.021)
0.239*
(0.047)
-0.209
(0.121)
0.69
51
≤ .05; Standard Errors in Parentheses
Contested Incumbent Reelection Vote Share as a function of District Support for Incumbent’s
Roll-Call Position and control variables. Column heading denotes referendum. See Table 4.2 for
corresponding bills.
Table C.6: Full Referendum Estimates: Part 5
Variable
Support for Legislator’s Position
Inc. Party Presidential Vote
Incumbent Previous Vote Share
Incumbents Contributions Advantage
Democratic Party Dummy
Incumbent Previously Contested Dummy
Constant
R-Squared
N
∗p
OH-2008-Issue-5
-0.112
(0.111)
0.264
(0.136)
0.431*
(0.191)
0.024*
(0.007)
0.047
(0.031)
0.139*
(0.065)
0.068
(0.157)
0.71
47
WA-2004-Measure-55
0.028
(0.048)
0.516*
(0.055)
0.413*
(0.071)
0.011*
(0.003)
0.005
(0.01)
0.135*
(0.027)
-0.104
(0.06)
0.912
63
WA-2008-Measure-67
0.023
(0.119)
0.553*
(0.112)
0.178
(0.102)
0.013*
(0.004)
-0.032*
(0.014)
0.061
(0.039)
0.116
(0.079)
0.787
62
WA-2009-Ref-71
0.103
(0.091)
0.47*
(0.147)
0.335*
(0.119)
0.026*
(0.004)
-0.106*
(0.017)
0.109*
(0.043)
-0.025
(0.075)
0.83
62
≤ .05; Standard Errors in Parentheses
Contested Incumbent Reelection Vote Share as a function of District Support for Incumbent’s
Roll-Call Position and control variables. Column heading denotes referendum. See Table 4.2 for
corresponding bills.
116
Table C.7: Full Referendum Estimates: Part 6
Variable
Support for Legislator’s Position
Inc. Party Presidential Vote
Incumbent Previous Vote Share
Incumbents Contributions Advantage
Democratic Party Dummy
Four Candidates in Current Race
Three Candidates in Previous Race
Four Candidates in Previous Race
Constant
R-Squared
N
∗p
AZ-1998-Prop-300
-0.118
(0.245)
0.41
(0.248)
0.183
(0.122)
0.015
(0.008)
-0.02
(0.03)
-0.031
(0.017)
0.028
(0.017)
0.027
(0.021)
0.076
(0.129)
0.644
26
AZ-1998-Prop-301
0.076
(0.186)
0.258
(0.193)
0.126
(0.111)
0.011
(0.008)
-0.006
(0.023)
-0.041*
(0.018)
0.021
(0.018)
0.024
(0.021)
0.107
(0.131)
0.669
25
ND-2012-Measure-4
-0.039
(0.022)
0.258*
(0.038)
0.743*
(0.108)
-0.003
(0.004)
0.033*
(0.012)
-0.053*
(0.009)
0.084*
(0.016)
-0.076
(0.052)
0.93
24
≤ .05; Standard Errors in Parentheses
Contested Incumbent Reelection Vote Share as a function of District Support for Incumbent’s
Roll-Call Position and control variables. Column heading denotes referendum. See Table 4.2 for
corresponding bills.
117
C.2
Full Estimates of Ideological Representation Analysis
Table C.8: Full Ideological Extremity Estimates: Part 1
Variable
Ideological Extremity
AL
-0.23
(0.957)
-0.042
(0.055)
0.25*
(0.057)
2.589*
(0.37)
-0.646
(1.608)
6.155*
(2.724)
44.944*
(6.729)
0.575
72
Presidential Approval
Inc. Party Presidential Vote
Incumbents Contributions Advantage
Freshman Incumbent
Democratic Party Dummy
Constant
R-Squared
N
∗p
AK
-1.126
(1.26)
0.192*
(0.085)
0.5*
(0.095)
3.252*
(0.527)
-1.379
(1.46)
7.884*
(2.893)
19.607*
(8.327)
0.489
83
AR
0.113
(1.061)
-0.066
(0.097)
0.581*
(0.141)
3.032*
(0.996)
3.528
(2.607)
12.663*
(4.138)
21.514*
(10.621)
0.541
35
CA
-0.131
(0.165)
0.149*
(0.023)
0.681*
(0.031)
0.445*
(0.095)
-0.238
(0.471)
-3.584*
(0.678)
16.767*
(2.045)
0.84
239
CT
0.244
(0.266)
0.087*
(0.028)
0.5*
(0.036)
1.729*
(0.162)
-0.473
(0.847)
-3.165*
(0.962)
28.811*
(2.641)
0.598
466
CO
-0.148
(0.311)
0.14*
(0.026)
0.66*
(0.037)
1.313*
(0.226)
-0.337
(0.642)
-2.448*
(0.741)
15.259*
(3.08)
0.81
132
DE
0.28
(0.375)
0.257*
(0.046)
0.51*
(0.047)
1.222*
(0.344)
-0.513
(1.358)
-0.459
(1.614)
21.865*
(3.612)
0.825
97
FL
-0.322
(0.226)
0.11*
(0.034)
0.399*
(0.046)
2.772*
(0.282)
-0.37
(0.745)
-1.131
(1.083)
29.627*
(2.739)
0.679
152
≤ .05; Standard Errors in Parentheses
OLS regressions where Dependent Variable is Contested Incumbent Two-Party Vote-Share. Independent variable is estimated ideological distance between legislator and his district.
Table C.9: Full Ideological Extremity Estimates: Part 2
Variable
Ideological Extremity
GA
-0.339
(0.332)
0.186*
(0.039)
0.323*
(0.028)
2.36*
(0.26)
-3.238*
(1.028)
-0.052
(1.064)
34.583*
(2.915)
0.503
275
Presidential Approval
Inc. Party Presidential Vote
Incumbents Contributions Advantage
Freshman Incumbent
Democratic Party Dummy
Constant
R-Squared
N
∗p
HI
-0.866*
(0.417)
0.147*
(0.067)
0.553*
(0.134)
0.635*
(0.243)
-1.137
(1.706)
-5.622
(3.456)
34.077*
(7.942)
0.263
188
IA
-0.042
(0.152)
0.04
(0.028)
0.464*
(0.049)
2.08*
(0.182)
-0.826
(0.665)
-0.121
(0.787)
30.33*
(3.151)
0.512
293
ID
-1.121*
(0.319)
-0.146*
(0.027)
0.698*
(0.036)
1.194*
(0.265)
-3.508*
(0.747)
13.918*
(1.363)
27.203*
(3.133)
0.698
429
IL
-0.442
(0.291)
0.097*
(0.027)
0.469*
(0.026)
1.836*
(0.142)
-0.552
(0.803)
0.493
(0.849)
30.624*
(2.037)
0.79
298
IN
0.045
(0.26)
0.184*
(0.031)
0.414*
(0.035)
1.429*
(0.157)
-2.247*
(0.931)
0.409
(0.906)
25.917*
(3.387)
0.591
285
KS
0.268
(0.343)
-0.037
(0.034)
0.299*
(0.039)
2.726*
(0.21)
-0.833
(0.816)
7.176*
(1.195)
38.261*
(3.419)
0.455
290
KY
0.281
(0.554)
0.185*
(0.06)
0.195*
(0.065)
2.088*
(0.333)
-1.584
(1.336)
1.459
(1.92)
36.494*
(4.763)
0.294
160
≤ .05; Standard Errors in Parentheses
OLS regressions where Dependent Variable is Contested Incumbent Two-Party Vote-Share. Independent variable is estimated ideological distance between legislator and his district.
118
Table C.10: Full Ideological Extremity Estimates: Part 3
Variable
Ideological Extremity
MA
0.038
(0.312)
-0.067
(0.088)
0.491*
(0.066)
1.373*
(0.278)
0.632
(1.34)
-1.328
(2.297)
41.161*
(5.218)
0.461
178
Presidential Approval
Inc. Party Presidential Vote
Incumbents Contributions Advantage
Freshman Incumbent
Democratic Party Dummy
Constant
R-Squared
N
∗p
ME
-0.171
(0.271)
0.139*
(0.058)
0.364*
(0.05)
1.424*
(0.231)
-0.869
(0.722)
-3.034*
(1.511)
35.274*
(3.629)
0.264
397
MI
-0.242
(0.163)
0.119*
(0.023)
0.665*
(0.023)
0.769*
(0.099)
0.276
(0.475)
0.405
(0.644)
19.914*
(1.89)
0.828
398
MN
-0.11
(0.211)
0.061*
(0.023)
0.485*
(0.027)
1.779*
(0.196)
-1.972*
(0.547)
0.477
(0.639)
30.274*
(2.217)
0.557
621
MO
-0.616*
(0.2)
0.059*
(0.029)
0.471*
(0.036)
1.667*
(0.149)
-0.155
(0.591)
2.091*
(0.831)
33.213*
(2.842)
0.557
373
MS
-0.809
(0.935)
0.035
(0.085)
0.217*
(0.064)
4.186*
(0.657)
1.068
(2.079)
7.186*
(3.127)
41.871*
(6.048)
0.602
46
MT
0.301
(0.296)
0.001
(0.035)
0.228*
(0.041)
1.908*
(0.266)
0.936
(0.803)
2.35*
(1.097)
43.25*
(3.419)
0.284
266
NC
0.397
(0.317)
0.067
(0.055)
0.388*
(0.048)
1.204*
(0.204)
-1.566
(0.979)
4.354*
(1.751)
28.365*
(3.941)
0.558
136
≤ .05; Standard Errors in Parentheses
OLS regressions where Dependent Variable is Contested Incumbent Two-Party Vote-Share. Independent variable is estimated ideological distance between legislator and his district.
Table C.11: Full Ideological Extremity Estimates: Part 4
Variable
Ideological Extremity
NM
0.585
(0.52)
0.132*
(0.056)
0.504*
(0.055)
1.256*
(0.285)
-2.246
(1.244)
0.851
(1.18)
21.043*
(4.889)
0.62
117
Presidential Approval
Inc. Party Presidential Vote
Incumbents Contributions Advantage
Freshman Incumbent
Democratic Party Dummy
Constant
R-Squared
N
∗p
NY
-0.145
(0.268)
0.095*
(0.023)
0.583*
(0.02)
1.047*
(0.097)
-1.065
(0.699)
-1.71*
(0.801)
27.607*
(1.67)
0.807
534
OH
-0.097
(0.294)
0.145*
(0.026)
0.506*
(0.031)
1.098*
(0.137)
-0.135
(0.609)
3.461*
(0.697)
23.645*
(2.457)
0.724
299
OK
-1.52*
(0.473)
0.038
(0.056)
0.295*
(0.054)
2.855*
(0.304)
-1.646
(1.251)
6.236*
(1.763)
46.187*
(5.026)
0.517
147
OR
-0.509*
(0.19)
0.126*
(0.032)
0.579*
(0.04)
1.191*
(0.142)
-1.344
(0.701)
-2.858*
(0.886)
23.665*
(2.961)
0.776
153
PA
-0.414
(0.264)
0.065*
(0.023)
0.432*
(0.021)
1.649*
(0.116)
-1.625*
(0.792)
-0.019
(0.64)
37.264*
(1.928)
0.633
538
RI
-1.653*
(0.496)
0.063
(0.052)
0.732*
(0.061)
1.364*
(0.248)
-2.921
(1.542)
-10.275*
(2.01)
29.741*
(4.036)
0.69
166
SC
-0.403
(0.615)
0.002
(0.055)
0.48*
(0.069)
1.405*
(0.346)
0.025
(1.534)
8.151*
(1.688)
29.409*
(5.1)
0.464
138
≤ .05; Standard Errors in Parentheses
OLS regressions where Dependent Variable is Contested Incumbent Two-Party Vote-Share. Independent variable is estimated ideological distance between legislator and his district.
119
Table C.12: Full Ideological Extremity Estimates: Part 5
Variable
Ideological Extremity
Presidential Approval
Inc. Party Presidential Vote
Incumbents Contributions Advantage
Freshman Incumbent
Democratic Party Dummy
Constant
R-Squared
N
∗p
TN
0.087
(0.633)
-0.018
(0.042)
0.279*
(0.051)
1.84*
(0.256)
-0.855
(1.375)
3.097*
(1.575)
41.908*
(4.524)
0.487
167
TX
-0.714*
(0.276)
0.096*
(0.028)
0.474*
(0.032)
0.485*
(0.123)
-0.827
(0.819)
4.309*
(0.943)
30.09*
(2.72)
0.556
262
UT
-0.997*
(0.432)
0.128*
(0.042)
0.487*
(0.038)
1.603*
(0.224)
0.736
(0.899)
6.898*
(1.267)
24.191*
(3.848)
0.63
185
VA
0.377
(0.528)
0.063*
(0.025)
0.354*
(0.052)
2.059*
(0.384)
0.638
(1.167)
1.252
(1.035)
33.256*
(4.146)
0.441
131
WI
-0.966*
(0.203)
0.142*
(0.029)
0.561*
(0.045)
1.602*
(0.187)
-1.533*
(0.754)
-0.854
(0.833)
31.467*
(3.361)
0.587
262
WY
0.04
(0.773)
0.078
(0.076)
0.375*
(0.079)
2.785*
(0.603)
1.441
(1.617)
8.535*
(3.008)
28.998*
(6.347)
0.473
80
≤ .05; Standard Errors in Parentheses
OLS regressions where Dependent Variable is Contested Incumbent Two-Party Vote-Share. Independent variable is estimated ideological distance between legislator and his district.
120
Table C.13: Cross-State Regression using NPAT Measures of Ideological Extremity
Variable
Ideological Extremity
Extremity
-0.155*
(0.048)
Extremity X Professionalism
Extremity X District Size
Incumbent Contrib. X Extremity
Challenger Contrib. X Extremity
Inc. Party Presidential Vote
Logged Incumbent Contributions
Logged Challenger Contributions
Incumbent Previous Vote Share
Incumbent Previously Contested Dummy
Q4 Presidential Approval
Presidential Approval X Pres Party
Professionalism
Logged District Size
Democratic Party Dummy
Member of Pres Party
Freshman Incumbent
Midterm Election Dummy
Constant
Log-Likelihood
Uncensored Observations
∗
0.288*
(0.006)
-0.090
(0.087)
-1.403*
(0.032)
0.245*
(0.008)
6.816*
(0.327)
-0.020
(0.025)
0.139*
(0.011)
-1.597
(2.778)
1.671*
(0.437)
1.33*
(0.164)
-7.79*
(0.613)
0.675*
(0.151)
2.866*
(0.777)
20.53
(4.24)
-27837.05
8851
Extremity w/ Controls
-0.496
(0.751)
-0.668
(0.496)
0.053
(0.087)
-0.017
(0.037)
0.013
(0.013)
0.288*
(0.006)
0.003
(0.219)
-1.477*
(0.078)
0.245*
(0.008)
6.802*
(0.328)
-0.018
(0.01)
0.139*
(0.011)
2.812
(4.248)
1.26*
(0.659)
1.338*
(0.164)
-7.809*
(0.616)
0.673*
(0.152)
2.95*
(0.778)
23.52
(5.91)
-27835.36
8851
p ≤ .05 Standard Errors in Parentheses, Dependent Variable: Contested Incumbent Vote Share
Results from a Heckman selection model where the selection equation follows the model
from chapter 2 of this dissertation controlling for Quarter 2 Presidential Approval interacted
with whether an incumbent was a member of the president’s party. There are a total of
17460 observations with 8609 censored and 8851 uncensored.
121
C.3
Alternative Measurement of Extremity
As described in the main text, prior work on Congressional elections at times measures
a legislator’s ideological extremity using the absolute value of a legislator’s ideal point. For
replication purposes Table C.14 presents comparable estimates of γ1 from Equation 4.4
using this absolute value measure.
Table C.14: Relationships between Absolute Value of Incumbent’s Ideal Point and
Vote Share
State
US
AL
AK
AR
CA
CT
CO
DE
FL
GA
HI
IA
ID
IL
IN
KS
KY
MA
ME
MI
MN
MO
MS
MT
NC
NM
NY
OH
OK
OR
PA
RI
SC
TN
TX
UT
VA
WI
WY
∗∗ p
Abs. Value of Ideal Point
-3.702** (0.481)
-3.714 (2.873)
-8.484** (3.407)
-3.738 (2.392)
-3.381** (0.922)
0.070 (1.177)
-0.822 (0.925)
4.949** (1.966)
-3.062** (1.495)
0.784 (1.055)
3.902 (3.353)
0.160 (1.472)
-4.928** (0.589)
-0.445 (0.825)
-1.938 (1.836)
-0.550 (0.978)
-0.528 (1.074)
3.385* (1.801)
-3.503** (0.983)
-6.914** (0.961)
-3.926** (0.916)
-1.926* (1.108)
-5.70** (2.679)
1.183 (1.212)
-0.912 (1.262)
-3.548** (1.322)
2.063** (0.705)
-4.895** (0.824)
4.620** (1.821)
-2.165** (1.009)
-2.398** (0.828)
-3.802** (1.242)
-0.492 (1.753)
3.468** (1.534)
-3.20** (0.850)
-3.774** (1.156)
-2.181 (1.603)
-2.357 (1.905)
-2.270 (1.529)
R-Squared
0.64
0.58
0.53
0.56
0.85
0.60
0.81
0.82
0.69
0.50
0.25
0.52
0.73
0.79
0.59
0.45
0.29
0.49
0.29
0.85
0.57
0.55
0.63
0.28
0.55
0.64
0.78
0.75
0.50
0.77
0.64
0.69
0.49
0.50
0.56
0.64
0.45
0.55
0.51
N
1931
72
83
35
239
466
132
97
152
275
188
293
429
298
285
290
160
178
397
398
621
373
46
266
136
117
534
299
147
153
538
166
138
167
262
185
131
262
80
≤ .05; ∗ p ≤ .10 Standard Errors in Parentheses
Dependent Variable: Contested Incumbent Vote Share.
122
An ideal point of zero does not imply a legislator is “perfectly moderate” (Ladewig
2010: 503). For example, Joe Democrat with an ideal point of -0.1 would be and be
classified as “extreme” as a Jane Democrat with an ideal point of 0.1, even though Joe is
more liberal than Jane. To investigate the extent to which liberal Democrats and conservative
Republicans receive lower vote shares, Table C.15 presents estimates of γ1 from Equation
4.4 when extremity is a legislator’s unadjusted ideal point. Estimates are from separate OLS
regressions on data subset by an incumbent’s political party. For the the Democratic subset,
positive coefficients indicate more liberal Democrats receive lower vote shares, and for the
Republican subset, negative coefficients indicate conservative Republicans receive lower
vote shares.
123
Table C.15: Relationships between Incumbent’s Ideal Point and Vote Share
State
US
AL
AK
AR
CA
CT
CO
DE
FL
GA
HI
IA
ID
IL
IN
KS
KY
MA
ME
MI
MN
MO
MS
MT
NC
NM
NY
OH
OK
OR
PA
RI
SC
TN
TX
UT
VA
WI
WY
Democrat Subset
2.725** (0.675)
5.202 (3.347)
-0.882 (3.666)
4.162* (2.426)
3.246** (1.094)
-1.179 (1.494)
2.019 (1.815)
-6.013** (2.968)
1.165 (1.433)
-0.865 (1.773)
-5.447** (2.501)
-0.643 (3.368)
1.514 (3.244)
-0.167 (0.648)
4.043 (2.683)
2.646 (1.672)
2.262* (1.276)
-5.388** (1.869)
2.442 (1.528)
8.631** (1.553)
6.641** (1.445)
3.219 (2.043)
1.776 (2.496)
0.952 (1.758)
1.822 (2.092)
1.629 (2.688)
-1.704** (0.572)
5.10** (1.945)
-5.971** (2.927)
-1.032 (1.655)
3.094** (1.383)
0.799 (1.128)
0.640 (1.604)
4.175 (2.723)
2.688** (1.187)
3.294* (1.946)
2.356 (2.930)
7.275** (3.549)
-3.014** (1.486)
∗∗ p
R-Squared
0.69
0.55
0.64
0.57
0.84
0.61
0.82
0.88
0.78
0.52
0.20
0.48
0.35
0.78
0.62
0.44
0.24
0.50
0.19
0.86
0.65
0.61
0.63
0.32
0.47
0.58
0.81
0.70
0.55
0.82
0.69
0.71
0.64
0.64
0.59
0.51
0.35
0.52
0.54
N
955
51
31
25
141
324
66
32
43
146
138
129
41
150
168
102
102
169
197
188
327
139
33
116
81
67
369
111
82
53
271
136
62
109
124
72
56
93
25
Republican Subset
-5.139** (0.676)
1.044 (5.459)
-19.441** (5.434)
-0.268 (3.319)
-3.642* (1.883)
0.211 (2.266)
-0.271 (0.951)
1.981 (2.549)
1.078 (1.827)
0.806 (1.103)
-7.480 (8.499)
0.009 (1.373)
-4.905** (0.504)
0.003 (1.274)
1.814 (2.496)
-0.155 (1.108)
1.567 (1.603)
-4.708 (3.425)
-4.004** (1.235)
-5.080** (1.180)
-1.944* (1.159)
-2.255* (1.273)
-11.411* (6.126)
1.265 (1.482)
0.002 (1.458)
-4.330** (1.286)
-5.901** (1.302)
-3.869** (0.80)
6.216** (1.977)
-3.918** (1.342)
-1.099 (0.976)
-0.812 (2.010)
4.454 (3.119)
5.008** (2.266)
-1.959 (1.331)
-20* (1.132)
-1.606 (1.693)
1.656 (2.245)
-5.884** (1.916)
R-Squared
0.49
0.67
0.55
0.77
0.71
0.29
0.84
0.57
0.64
0.51
0.26
0.56
0.76
0.61
0.61
0.48
0.51
0.58
0.35
0.67
0.25
0.54
0.65
0.29
0.62
0.78
0.39
0.65
0.57
0.74
0.47
0.14
0.42
0.34
0.53
0.76
0.59
0.56
0.58
≤ .05; ∗ p ≤ .10 Standard Errors in Parentheses
Dependent Variable: Contested Incumbent Vote Share.
124
N
989
23
53
13
98
142
66
66
111
135
50
166
388
149
117
192
59
45
201
212
295
236
15
150
55
50
171
191
65
100
275
37
81
58
140
116
75
169
57
C.4
Electoral Accountability for Party Loyalty
Some contend that “voters place greater weight on partisanship than ideology when
evaluating behavior in Congress” (Carson et. al 2010: 601-2). Carson and his coauthors
argue identifying legislators who break party ranks is likely easier for voters than pinpointing
their representative’s ideology. For example, Maine voters may recognize that Representative
Lajoie did not succumb to the party whip and reward him for his loyalty to district over
party. The extent to which a legislator votes with his party, furthermore, could approximate
their extremity and whether she represented her district, permitting an indirect application
of Downsian expectations.
P artyLoyalty =π0 + π1 [|AbsoluteV alueof IdealP oint|]
+ π2 [IncumbentP artyP residentialV ote]
+ π3 [IncumbentSpendingAdvantage] + π4 [F reshmanDummy]
+ π5 [P resident0 sP artyDummy] + π6 [P residentialApproval]
+ π7 [IncumbentP reviousV oteShare]
+ π8 [IncumbentP reviouslyContested] + π9 [P reviousP artyLoyalty] + (C.1)
ˆ
P artyLoyalty =α0 + α1 [P artyLoyalty]
+ α2 [IncumbentP artyP residentialV ote]
+ α3 [IncumbentSpendingAdvantage] + α4 [F reshmanDummy]
+ α5 [P resident0 sP artyDummy] + α6 [P residentialApproval]
+ α7 [IncumbentP reviousV oteShare]
+ α8 [IncumbentP reviouslyContested] + (C.2)
125
To replicate Carson’s analysis at the state legislative level, I evaluate the extent to which
voters punish party loyal state legislators. I again examine state house elections from 1998 2008. Similar to ideological representation analysis, the dependent variable is a contested
incumbent’s vote-share. The independent variable of interest is an incumbent’s party loyalty.
Carson et. al argue that “party loyalty is not exogenous to legislators’ reelection prospects”
and use a two-stage least squares panel data estimator where ideological extremity - the
absolute value of a legislator’s ideal point - instruments for party loyalty. To account for this
potential endogeneity, I calculate session specific estimates of party loyalty and extremity
for each state house member to use in a first state estimation of equation C.1. Similar to
Carson, I then conduct a second state estimation using equation C.2. Control variables
follow the measurement described in the main text.
The first column of Table C.16 presents the OLS coefficient on the party loyalty measure
from state-level estimations of Equation C.2. Similar to Carson et. al’s results, estimates
suggest more party loyal members of the US house receive lower vote shares. A 15% increase
in party loyalty results in approximately a 3.2% decrease in incumbent vote share. The same,
however, is not the case in most state legislatures. There is only a negative relationship
between party loyalty and incumbent vote share in 4 of 38 states. Statistical analyses again
provide evidence of district-level electoral connections in Michigan, but despite the findings
the Maine voters appear to hold their state representative accountable for individual roll-calls,
neither the ideological extremity nor party loyalty analysis provides similar evidence of
district level accountability. Results are relatively similar when controlling for legislators
ideological distance from their district (third column of Table C.16).
Figure C.1 presents differences across states by plotted the predicted average change in
incumbent vote-share if all legislators voted with the majority of their party 15% more of
the time. These estimates suggest only legislators in Missouri, Michigan, and Texas will
receive less votes on average for this increase in party loyalty, and it appears that legislators
126
Table C.16: Relationships between Incumbent’s Party Loyalty and Vote Share
State
US
AL
AK
AR
CA
CT
CO
DE
FL
GA
HI
IA
ID
IL
IN
KS
KY
MA
ME
MI
MN
MO
MS
MT
NC
NM
NY
OH
OK
OR
PA
RI
SC
TN
TX
UT
VA
WI
WY
Loyalty
-0.219** (0.041)
0.009 (0.016)
0.323** (0.138)
-0.115 (1.328)
-0.087 (0.077)
-0.012 (0.108)
-0.210 (0.248)
0.144 (0.121)
-0.146 (0.304)
0.035 (0.446)
0.257* (0.153)
-0.023 (0.126)
-0.157 (0.166)
0.223 (0.214)
0.017 (0.066)
-0.023 (0.119)
-0.275 (0.374)
0.157 (0.201)
-0.090 (0.077)
-0.263** (0.102)
-0.128 (0.101)
-0.353** (0.102)
-0.468 (0.478)
0.091 (0.115)
0.379** (0.113)
-0.052 (0.155)
-0.028 (0.113)
-0.159** (0.059)
-0.158 (0.130)
-0.004 (0.137)
-0.093 (0.10)
-0.047 (0.090)
-0.171 (0.112)
0.017 (0.112)
-0.907** (0.135)
0.021 (0.10)
-0.269 (0.243)
0.036 (0.051)
-0.40 (0.429)
∗∗ p
Loyalty - Control for Extremity
-0.195** (0.040)
0.010 (0.019)
0.301** (0.130)
-0.129 (1.405)
-0.093 (0.078)
-0.015 (0.109)
-0.224 (0.260)
0.157 (0.136)
-0.142 (0.306)
-0.038 (0.466)
0.211 (0.169)
0.073 (0.138)
-0.217 (0.198)
0.221 (0.214)
0.122* (0.072)
-0.017 (0.119)
-0.355 (0.390)
0.287 (0.216)
-0.094 (0.078)
-0.262** (0.102)
-0.153 (0.104)
-0.366** (0.102)
-0.583 (0.493)
0.070 (0.126)
0.390** (0.113)
-0.055 (0.156)
-0.035 (0.113)
-0.189** (0.060)
-0.281** (0.134)
0.014 (0.153)
-0.113 (0.103)
-0.121 (0.089)
-0.197* (0.114)
0.025 (0.113)
-0.704** (0.138)
-0.009 (0.105)
-0.291 (0.244)
0.039 (0.050)
-0.425 (0.430)
Extremity
-0.067 (0.115)
0.20 (2.745)
-3.230** (1.214)
-0.613 (1.580)
-0.130 (0.206)
-0.169 (0.282)
-0.092 (0.467)
0.102 (0.458)
-0.210 (0.318)
-0.225 (0.410)
-0.327 (0.496)
-0.30 (0.187)
-0.320 (0.574)
-0.228 (0.369)
0.398 (0.292)
0.290 (0.466)
-0.546 (0.741)
-0.051 (0.354)
-0.129 (0.434)
-0.036 (0.204)
0.115 (0.209)
-0.316 (0.30)
-1.262 (1.295)
-0.290 (0.382)
0.428 (0.408)
0.604 (0.650)
0.147 (0.293)
-0.492 (0.334)
-1.637** (0.585)
0.090 (0.324)
0.071 (0.307)
-1.347** (0.543)
0.368 (0.819)
0.475 (0.806)
-0.703** (0.273)
-0.508 (0.525)
0.599 (0.60)
-0.759** (0.226)
-0.986 (0.987)
N
1663
25
51
16
118
309
76
78
76
166
116
180
108
209
206
173
108
148
180
149
464
186
32
119
98
72
364
204
111
82
401
124
110
139
218
133
91
191
42
≤ .05; ∗ p ≤ .10 Standard Errors in Parentheses
Incumbent Vote Share as a function of Instrumented Party Loyalty
in Alaska and North Carolina would actually receive larger electoral margins if they vote
more often with the majority of their party.
127
Figure C.1: Predicted Changes in State House Incumbent Vote Share for Increased
Party Loyalty
Points represent average predicted vote-share loss associated with a 15% increase in an all
incumbents’ party loyalty scores for state house elections from 1998 - 2008. The point labeled
US represents the comparable estimate for the US House. Black and grey bars are 95% and 90%
bootstrapped confidence intervals.
128
Bibliography
Achen, Christopher H. 1978. “Measuring Representation.” American Journal of Political
Science 22(August): 475–510.
Aldrich, John H., and James S. Coleman Battista. 2002. “Conditional Party Government in
the States.” American Journal of Political Science 46(January): 164–172.
Ansolabehere, Stephen. 2010. “COOPERATIVE CONGRESSIONAL ELECTION STUDY,
2010: COMMON CONTENT.”.
Ansolabehere, Stephen. 2011. “COOPERATIVE CONGRESSIONAL ELECTION STUDY,
2008: COMMON CONTENT.”.
Ansolabehere, Stephen, and Philip Edward Jones. 2010. “Constituents Responses to Congressional Roll-Call Voting.” American Journal of Political Science 54(3): 583597.
Ansolabehere, Stephen, James M. Snyder, and Charles Stewart. 2001. “Candidate Positioning in U.S. House Elections.” American Journal of Political Science 45(January):
136–159.
Anzia, Sarah F., and Molly C. Jackman. 2013. “Legislative Organization and the Second
Face of Power: Evidence from U.S. State Legislatures.” The Journal of Politics 75(01):
210–224.
Arceneaux, Kevin. 2006. “The Federal Face of Voting: Are Elected Officials Held Accountable for the Functions Relevant to Their Office?” Political Psychology 27(October):
731–754.
Arnold, R. Douglas. 1992. The Logic of Congressional Action. Yale University Press.
Ashworth, Scott. 2012. “Electoral Accountability: Recent Theoretical and Empirical Work.”
Annual Review of Political Science 15(1): 183–201.
Atkeson, Lonna Rae, and Randall W. Partin. 1995. “Economic and Referendum Voting: A
Comparison of Gubernatorial and Senatorial Elections.” The American Political Science
Review 89(March): 99–107.
Bafumi, Joseph, and Michael C. Herron. 2010. “Leapfrog Representation and Extremism: A
Study of American Voters and Their Members in Congress.” American Political Science
Review 104(03): 519–542.
129
Bailey, Michael A., and Elliott B. Fullmer. 2011. “Balancing in the U.S. States, 1978 - 2009.”
State Politics & Policy Quarterly 11(June): 148–166.
Barnett, Jeffery. 2011. State and Local Government Finances Summary: 2009. Technical
Report G09-ALFIN U.S. Census Bureau.
Barro, R. 1973. “The Control Of Politicians An Economic Model.” Public Choice 14(September): 19–42.
Battista, James Coleman, Michael Peress, and Jesse Richman. 2013. “Common-Space Ideal
Points, Committee Assignments, and Financial Interests in the State Legislatures.” State
Politics & Policy Quarterly 13(March): 70–87.
Berry, William D., Michael B. Berkman, and Stuart Schneiderman. 2000. “Legislative
Professionalism and Incumbent Reelection: The Development of Institutional Boundaries.”
The American Political Science Review 94(December): 859–874.
Berry, William D., Richard C. Fording, Evan J. Ringquist, Russell L. Hanson, and Carl
Klarner. 2012. “A New Measure of State Government Ideology, and Evidence that Both
the New Measure and an Old Measure are Valid.” State Politics & Policy Quarterly
(December).
Besley, Timothy, and Anne Case. 2003. “Political Institutions and Policy Choices: Evidence
from the United States.” Journal of Economic Literature 41(March): 7–73.
Birkhead, Nathaniel. 2013. The Role of Ideology in State Legislative Elections. In State
Politics and Policy Conference. University of Iowa: .
Bishop, Bradford, and Rebecca Hatch. 2012. “Perception of State Parties and Voting in
Subnational Elections.”
Black, Duncan. 1958. The Theory of Committees and Elections. Cambridge University
Press.
Bond, Jon R., Cary Covington, and Richard Fleisher. 1985. “Explaining Challenger Quality
in Congressional Elections.” The Journal of Politics 47(June): 510–529.
Bonica, Adam. 2013. “Ideology and Interests in the Political Marketplace.” American
Journal of Political Science 57(2): 294–311.
Bovitz, Gregory L., and Jamie L. Carson. 2006. “Position-Taking and Electoral Accountability in the U.S. House of Representatives.” Political Research Quarterly 59(June):
297–312.
Bowen, Daniel. 2013. Small is Still Beautiful: Constituency Size and Citizen Attitudes
about American State Government. Chicago, IL: .
Brace, Paul, and Aubrey Jewett. 1995. “The State of State Politics Research.” Political
Research Quarterly 48(September): 643–681.
130
Brace, Paul, Kevin Arceneaux, Martin Johnson, and Stacy G. Ulbig. 2004. “Does State
Political Ideology Change over Time?” Political Research Quarterly 57(December):
529–540.
Broockman, David, and Christopher Skovron. 2013. What Politicians Believe About
Their Constituents: Asymmetric Misperceptions and Prospects for Constituency Contro.
Vanderbilt University: .
Cain, Bruce E., John Hanley, and Thad Kousser. 2006. “Term Limits: A Recipe for More
Competition?” In The marketplace of democracy: electoral competition and American
politics. Cato Institute.
Campbell, Angus, Philip E. Converse, and Donald E. Stokes. 1960. The American voter.
Chicago: The University of Chicago Press.
Campbell, James E. 1986. “Presidential Coattails and Midterm Losses in State Legislative
Elections.” The American Political Science Review 80(March): 45–63.
Canes-Wrone, Brandice, David W. Brady, and John F. Cogan. 2002. “Out of Step, Out
of Office: Electoral Accountability and House Members’ Voting.” American Political
Science Review 96(01): 127–140.
Canes-Wrone, Brandice, William Minozzi, and Jessica Bonney Reveley. 2011. “Issue
Accountability and the Mass Public.” Legislative Studies Quarterly 36(1): 535.
Canon, David T., and David J. Sousa. 1992. “Party System Change and Political Career
Structures in the U. S. Congress.” Legislative Studies Quarterly 17(3): 347–363.
Carsey, Thomas M., and Gerald C. Wright. 1998. “State and National Factors in Gubernatorial and Senatorial Elections.” American Journal of Political Science 42(July):
994–1002.
Carson, Jamie L., Gregory Koger, Matthew J. Lebo, and Everett Young. 2010. “The Electoral
Costs of Party Loyalty in Congress.” American Journal of Political Science 54(3): 598616.
Cass, Michael. 2012. “GOP expects to increase majorities in TN House, Senate.” The
Tennessean April 22(April): Online.
Chubb, John E. 1988. “Institutions, The Economy, and the Dynamics of State Elections.”
The American Political Science Review 82(March): 133–154.
Clark, John A. 1996. “Congressional Salaries and the Politics of Unpopular Votes.” American
Politics Research 24(April): 150–168.
Clinton, Joshua, Simon Jackman, and Douglas Rivers. 2004. “The Statistical Analysis of
Roll Call Data.” The American Political Science Review 98(May): 355–370.
Cohen, Jeffrey E. 1984. “Perceptions of Electoral Insecurity among Members Holding Safe
Seats in a U. S. State Legislature.” Legislative Studies Quarterly 9(May): 365–369.
131
Cox, Gary W., and Jonathan N. Katz. 2002. Elbridge Gerry’s Salamander: The Electoral
Consequences of the Reapportionment Revolution. Cambridge University Press.
Cox, Gary W., and Mathew D. McCubbins. 1993. Legislative Leviathan: Party Government
in the House. University of California Press.
Cox, Gary W., and Mathew D. McCubbins. 2005. Setting the Agenda: Responsible Party
Government in the U.S. House of Representatives. Cambridge University Press.
Cox, Gary W., Thad Kousser, and Mathew D. McCubbins. 2010. “Party Power or Preferences? Quasi-Experimental Evidence from American State Legislatures.” The Journal of
Politics 72(03): 799–811.
Cutler, Fred. 2004. “Accountability Payoffs in Federal Systems.” The Journal of Federalism
34(2): 19–38.
De Boef, Suzanna, and Paul M. Kellstedt. 2004. “The Political (and Economic) Origins of
Consumer Confidence.” American Journal of Political Science 48(4): 633–649.
Delli Carpini, Michael, Scott Keeter, and J. Kennamer. 1994. “Effects of the News Media
Environment on Citizen Knowledge of State Politics and Government.” Departmental
Papers (ASC) (July).
DiCamillo, Mark, and Mervin Field. 2008. Technical Report 2262 Field Research Corportation.
Downs, Anthony. 1957. An Economic Theory of Democracy. 1st ed. Harper and Row.
Downs, William M. 1999. “Accountability Payoffs in Federal Systems? Competing Logics
and Evidence from Europe’s Newest Federation.” Publius 29(January): 87–110.
Dubin, Michael J. 2007. Party Affiliations in the State Legislatures: A Year by Year Summary,
1796-2006. McFarland.
Dunk, Emily Van, and Ronald E. Weber. 1997. “Constituency-Level Competition in the
U. S. States, 1968-1988: A Pooled Analysis.” Legislative Studies Quarterly 22(May):
141–159.
Ebeid, Michael, and Jonathan Rodden. 2006. “Economic Geography and Economic Voting:
Evidence from the US States.” British Journal of Political Science 36: 527–547.
Engstrom, Erik, and Nathan Monroe. 2006. “Testing the Basis of Incumbency Advantage:
Strategic Candidates and Term Limits in the California Legislature.” State Politics &
Policy Quarterly 6(1): 1–20.
Erikson, Robert S., Gerald C. Wright, and John P. McIver. 1994. Statehouse Democracy:
Public Opinion and Policy in the American States. Cambridge University Press.
Erikson, Robert S., Michael B. Mackuen, and James A. Stimson. 2002. The Macro Polity.
Cambridge University Press.
132
Fearon, James. 1999. “Electoral Accountability and the Control of Politicians: Selecting
Good Types versus Sanctioning Poor Performance.” In Democracy, Accountability, and
Representation. New York: Cambridge University Press.
Fenno, Richard F. 1978. Home style: House Members in their districts. Little, Brown.
Ferejohn, John. 1986. “Incumbent Performance and Electoral Control.” Public Choice 50(1):
5–25.
Fiorina, Morris P. 1981. Retrospective Voting in American National Elections. Yale
University Press.
Fiorina, Morris P. 1994. “Divided Government in the American States: A Byproduct of
Legislative Professionalism?” The American Political Science Review 88(June): 304–316.
Fiorina, Morris P. 1995. Divided Government. First ed. Longman Publishing Group.
Folke, Olle, and James M. Snyder. 2012. “Gubernatorial Midterm Slumps.” American
Journal of Political Science p. np.
Forgette, Richard, Andrew Garner, and John Winkle. 2009. “Do Redistricting Principles and
Practices Affect U.S. State Legislative Electoral Competition?” State Politics & Policy
Quarterly 9(June): 151–175.
Geer, John G. 1996. From Tea Leaves to Opinion Polls. First ed. Columbia University Press.
Gerber, Elisabeth R. 1996. “Legislatures, Initiatives, and Representation: The Effects of
State Legislative Institutions on Policy.” Political Research Quarterly 49(June): 263–286.
Gerber, Elisabeth R., and Jeffery B. Lewis. 2004. “Beyond the Median: Voter Preferences,
District Heterogeneity, and Political Representation.” Journal of Political Economy 112(6):
1364–1382.
Gierzynski, Anthony, and David Breaux. 1996. “Legislative Elections and the Importance
of Money.” Legislative Studies Quarterly 21(August): 337–357.
Glass, Ira. 2012. “Red State Blue State.” National Public Radio: This American Life
(November).
Glineau, Franois, and ric Blanger. 2005. “Electoral Accountability in a Federal System:
National and Provincial Economic Voting in Canada.” Publius: The Journal of Federalism
35(June): 407–424.
Grau, Craig H. 1981. “Competition in State Legislative Primaries.” Legislative Studies
Quarterly 6(February): 35–54.
Grofman, Bernard. 2004. “Downs and Two-Party Convergence.” Annual Review of Political
Science 7(1): 25–46.
133
Groseclose, Tim, Steven D. Levitt, and James M. Snyder. 1999. “Comparing Interest Group
Scores across Time and Chambers: Adjusted ADA Scores for the U.S. Congress.” The
American Political Science Review 93(March): 33–50.
Grossman, Richard S., and Stephen A. Lee. 2008. “May Issue Versus Shall Issue: Explaining
the Pattern of Concealed-Carry Handgun Laws, 19602001.” Contemporary Economic
Policy 26(2): 198206.
Hall, Richard L., and Frank W. Wayman. 1990. “Buying Time: Moneyed Interests and
the Mobilization of Bias in Congressional Committees.” The American Political Science
Review 84(September): 797–820.
Hamm, Keith E, and Robert E Hogan. 2008. “Campaign Finance Laws and Candidacy
Decisions in State Legislative Elections.” Political Research Quarterly 61(September):
458–467.
Hogan, Robert E. 2003. “Sources of Competition in State Legislative Primary Elections.”
Legislative Studies Quarterly 28(February): 103–126.
Hogan, Robert E. 2004. “Challenger Emergence, Incumbent Success, and Electoral Accountability in State Legislative Elections.” The Journal of Politics 66(November): 1283–1303.
Hogan, Robert E. 2005. “Gubernatorial Coattail Effects in State Legislative Elections.”
Political Research Quarterly 58(December): 587 –597.
Hogan, Robert E. 2008. “Policy Responsiveness and Incumbent Reelection in State Legislatures.” American Journal of Political Science 52(October): 858–873.
Hotelling, Harold. 1929. “Stability in Competition.” The Economic Journal 39(March):
41–57.
Jacobson, Gary C. 1989. “Strategic Politicians and the Dynamics of U.S. House Elections,
1946-86.” The American Political Science Review 83(3): 773–793.
Jacobson, Gary C. 1993. “Deficit-Cutting Politics and Congressional Elections.” Political
Science Quarterly 108(October): 375–402.
Jacobson, Gary C., and Samuel Kernell. 1983. Strategy and Choice in Congressional
Elections. Yale University Press.
Javian, Kathy. 2011. Examining the Impact of the President’s Party on Gubernatorial
Elections. In Meeting of the Southern Political Science Association. New Orleans, LA:
p. np.
Jennings, M. Kent, and Richard G. Niemi. 1966. “Party Identification at Multiple Levels of
Government.” American Journal of Sociology 72(July): 86–101.
Jewell, Malcolm E. 1982a. “The Neglected World of State Politics.” The Journal of Politics
44(August): 638–657.
134
Jewell, Malcolm E. 1982b. Representation in State Legislatures. The University Press of
Kentucky.
Jewell, Malcolm E., and David Breaux. 1988. “The Effect of Incumbency on State Legislative Elections.” Legislative Studies Quarterly 13(November): 495–514.
Jewell, Malcolm E., and David Breaux. 1991. “Southern Primary and Electoral Competition
and Incumbent Success.” Legislative Studies Quarterly 16(February): 129–143.
Jones, David R., and Monika L. McDermott. 2004. “The Responsible Party Government
Model in House and Senate Elections.” American Journal of Political Science 48(January):
1–12.
Kahneman, Daniel, and Shane Frederick. 2002. “Representativeness revisted: Attribute
substitution in intuitive judgment.” In Heuristics and Biases: The Psychology of Intuitive
Judgment. New York: Cambridge University Press.
Kaplan, Martin, Kenneth R. Goldstein, and Matthew Hale. 2003. Local TV News Coverage
of the 2002 General Election. Technical report USC Annengerg School and University of
Wisconsin.
Kaplan, Martin, Kenneth R. Goldstein, and Matthew Hale. 2005. Local News Coverage
of the 2004 Campaigns. Technical report USC Annengerg School and University of
Wisconsin.
Kelleher, Christine A., and Jennifer Wolak. 2007. “Explaining Public Confidence in the
Branches of State Government.” Political Research Quarterly 60(December): 707 –721.
Kellerman, Michael. 2008. “Balancing or Signaling? Electoral punishment in sub-national
elections.”
Kernell, Samuel. 1978. “Explaining Presidential Popularity. How Ad Hoc Theorizing,
Misplaced Emphasis, and Insufficient Care in Measuring One’s Variables Refuted Common Sense and Led Conventional Wisdom Down the Path of Anomalies.” The American
Political Science Review 72(June): 506–522.
Key, Valdimer Orlando. 1949. Southern Politics in State and Nation. New edition ed. Alford
A. Knopf, Inc.
Key, V.O. 1964. Politics, Parties, and Pressure Groups. 5th ed. New York: Thomas Y.
Crowell Company.
Key, V.O. 1966. The Responsible Electorate. Belknap Press of Harvard University Press.
King, Gary. 1991. “Constituency Service and Incumbency Advantage.” British Journal of
Political Science 21: 119128.
Kingdon, John W. 1989. Congressmen’s Voting Decisions. Third edition ed. University of
Michigan Press.
135
Klarner, Carl. 2010. “Forecasting the 2010 State Legislative Elections.” PS: Political Science
& Politics 43(04): 643–648.
Klarner, Carl, William Berry, Thomas Carsey, Malcolm Jewell, Richard Niemi, Lynda
Powell, and James Snyder. 2013-01-11. “State Legislative Election Returns (1967-2010).”.
Konisky, David M, and Michiko Ueda. 2011. “The Effects of Uncontested Elections on
Legislator Performance.” Legislative Studies Quarterly 36(May): 199–229.
Ladewig, Jeffrey W. 2010. “Ideological Polarization and the Vanishing of Marginals:
Retrospective Roll-Call Voting in the U.S. Congress.” The Journal of Politics 72(02):
499–512.
Lascher, Edward L., Michael G. Hagen, and Steven A. Rochlin. 1996. “Gun Behind the
Door? Ballot Initiatives, State Policies and Public Opinion.” The Journal of Politics
58(03): 760–775.
Lax, Jeffrey, and Justin H. Phillips. 2009. “How Should We Estimate Public Opinion in the
State?” American Journal of Political Science 53(1): 107–121.
Layton, Charles. 1998. “Missing the Story at the Statehouse.” American Journalism Review
(July).
Lipset, Seymour Martin. 1997. American exceptionalism: a double-edged sword. New
York: W.W. Norton.
Lowry, Robert C., James E. Alt, and Karen E. Ferree. 1998. “Fiscal Policy Outcomes and
Electoral Accountability in American States.” The American Political Science Review
92(December): 759–774.
Maestas, Cherie D., Sarah Fulton, L. Sandy Maisel, and Walter J. Stone. 2006. “When
to Risk It? Institutions, Ambitions, and the Decision to Run for the U.S. House.” The
American Political Science Review 100(May): 195–208.
Malhotra, Neil. 2008. “The Impact of Public Financing on Electoral Competition: Evidence
from Arizona and Maine.” State Politics & Policy Quarterly 8(September): 263–281.
Masket, Seth E. 2011. No middle ground: how informal party organizations control
nominations and polarize legislatures. Ann Arbor: University Of Michigan Press.
Matsusaka, John G. 2001. “Problems with a Methodology Used to Evaluate the Voter
Initiative.” Journal of Politics 63(4): 1250–1256.
Mayer, Kenneth R., and John M. Wood. 1995. “The Impact of Public Financing on Electoral
Competitiveness: Evidence from Wisconsin, 1964-1990.” Legislative Studies Quarterly
20(February): 69–88.
Mayhew, David R. 1974. Congress: the electoral connection. New Haven, Conn.: Yale
University Press.
136
Mayo, Mike. 2004. The Relationship Between an Incumbent Governor’s Popularity and
State Legislative Election Outcomes: A Contemporary Assessment of the Coattails
Phenomenon PhD thesis Virginia Polytechnic Institute Blacksburg, VA.
McCarty, Nolan, Keith T. Poole, and Howard Rosenthal. 2008. Polarized America: The
Dance of Ideology and Unequal Riches. The MIT Press.
McCarty, Nolan, Keith T. Poole, and Howard Rosenthal. 2009. “Does Gerrymandering
Cause Polarization?” American Journal of Political Science 53(July): 666–680.
Miller, Warren E., and Donald E. Stokes. 1963. “Constituency Influence in Congress.” The
American Political Science Review 57(March): 45–56.
Moncrief, Gary F., Peverill Squire, and Malcolm Edwin Jewell. 2001. Who runs for the
legislature? Prentice Hall.
Mondak, Jeffery. 2000. “Reconsidering the Measurement of Political Knowledge.” Political
Analysis 8(1): 57–82.
Morehouse, Sarah M., and Malcolm E. Jewell. 2004. “STATES AS LABORATORIES: A
Reprise.” Annual Review of Political Science 7(1): 177–203.
Murdoch, Stephen. 2001. “The Debate Over Mandatory Minimums.” Washington Lawyer
(November).
National Conference of State Legislatures. 2012. http://www.ncsl.org.
National Institute on Money in State Politics. 2013. http://www.followthemoney.org.
New Jersey Constitution Convention Proceedings. 1947.
Nicholson, Stephen P., and Gary M. Segura. 1999. “Midterm Elections and Divided
Government: An Information-Driven Theory of Electoral Volatility.” Political Research
Quarterly 52(3): 609–629.
Nyhan, Brendan, Eric McGhee, John Sides, Seth Masket, and Steven Greene. 2012. “One
Vote Out of Step? The Effects of Salient Roll Call Votes in the 2010 Election.” American
Politics Research (March).
Oppenheimer, Bruce I., James A. Stimson, and Richard W. Waterman. 1986. “Interpreting
U. S. Congressional Elections: The Exposure Thesis.” Legislative Studies Quarterly
11(May): 227–247.
Piereson, James E. 1975. “Presidential Popularity and Midterm Voting at Different Electoral
Levels.” American Journal of Political Science 19(November): 683–694.
Piketty, Thomas. 2000. “Voting as Communicating.” The Review of Economic Studies
67(January): 169–191.
137
Powell, G. Bingham. 2000. Elections as Instruments of Democracy: Majoritarian and
Proportional Visions. Yale University Press.
Pritchard, Anita. 1992. “Strategic Considerations in the Decision to Challenge a State
Legislative Incumbent.” Legislative Studies Quarterly 17(3): 381–393.
Ray, David, and John Havick. 1981. “A Longitudinal Analysis of Party Competition in State
Legislative Elections.” American Journal of Political Science 25(February): 119–128.
Redlawsk, David. 2013. “Rutgers-Eagleton Poll.”.
Reif, Karlheinz, and Hermann Schmitt. 1980. “Nine Second-Order National Elections - A
Conceptual Framework for the Analysis of European Election Results.” European Journal
of Political Research 8(1): 344.
Remmer, K. L., and F. Glineau. 2003. “Subnational Electoral Choice: Economic and Referendum Voting in Argentina, 1983-1999.” Comparative Political Studies 36(September):
801–821.
Richardson, Lilliard, David Konisky, and Jeffrey Milyo. 2011. “Public Approval of State
Legislatures.”
Rodden, Jonathan, and Erik Wibbels. 2011. “Dual accountability and the nationalization
of party competition: Evidence from four federations.” Party Politics 17(September):
629–653.
Sanbonmatsu, Kira. 2006. “The Legislative Party and Candidate Recruitment in the American States.” Party Politics 12(March): 233–256.
Schattschneider, Elmer Eric. 1942. Party Government. Praeger.
Serra, George, and Neil Pinney. 2004. “Casework, Issues, and Voting in State Legislative
Elections.” The Journal of Legislative Studies 10(4): 32–46.
Shapiro, Catherine R., David W. Brady, Richard A. Brody, and John A. Ferejohn. 1990.
“Linking Constituency Opinion and Senate Voting Scores: A Hybrid Explanation.” Legislative Studies Quarterly 15(November): 599–621.
Shor, Boris, and Nolan McCarty. 2011. “The Ideological Mapping of American Legislatures.”
American Political Science Review 105(03): 530–551.
Simon, Dennis M., Charles W. Ostrom, and Robin F. Marra. 1991. “The President, Referendum Voting, and Subnational Elections in the United States.” The American Political
Science Review 85(December): 1177–1192.
Smith, Andrew. 2010. Obama Continues to Lose Support in New Hampshire. Poll University
of New Hampshire Survey Center.
138
Smith, Daniel A, and Caroline Tolbert. 2004. Educated by initiative the effects of direct
democracy on citizens and political organizations in the American states. Ann Arbor:
University of Michigan Press.
Snyder, James. 1996. “Constituency Preferences: California ballot Propositions, 1974-1990.”
Legislative Studies Quarterly 21(4): 463–488.
Snyder, James, and David Stromberg. 2008. Press Coverage and Political Accountability.
Working Paper 13878 National Bureau of Economic Research.
Squire, Peverill. 1993. “Professionalization and Public Opinion of State Legislatures.” The
Journal of Politics 55(May): 479–491.
Squire, Peverill. 2000. “Uncontested Seats in State Legislative Elections.” Legislative
Studies Quarterly 25(February): 131–146.
Squire, Peverill. 2007. “Measuring State Legislative Professionalism: The Squire Index
Revisited.” State Politics & Policy Quarterly 7(2): 211–227.
Stein, Robert M. 1990. “Economic Voting for Governor and U. S. Senator: The Electoral
Consequences of Federalism.” The Journal of Politics 52(February): 29–53.
Tausanovitch, Chris, and Christopher Warshaw. 2013. “Measuring Constituent Policy
Preferences in Congress, State Legislatures, and Cities.” The Journal of Politics 75(02):
330–342.
Tidmarch, Charles M., Edward Lonergan, and John Sciortino. 1986. “Interparty Competition
in the U. S. States: Legislative Elections, 1970-1978.” Legislative Studies Quarterly 11(3):
353–374.
Tufte, Edward R. 1975. “Determinants of the Outcomes of Midterm Congressional Elections.”
The American Political Science Review 69(3): 812–826.
Tversky, Amos, and Daniel Kahneman. 1974. “Judgment under Uncertainty: Heuristics and
Biases.” Science 185(September): 1124–1131.
U.S. Department of Justice. 2012. http://www.ucrdatatool.gov/.
Vogel, Ed. 2010. “’41 To Angle’.” Las Vegas Review-Journal (September).
Weber, Ronald E., Harvey J. Tucker, and Paul Brace. 1991. “Vanishing Marginals in State
Legislative Elections.” Legislative Studies Quarterly 16(February): 29–47.
Werner, Timothy, and Kenneth R. Mayer. 2007. “Public Election Funding, Competition,
and Candidate Gender.” PS: Political Science and Politics 40(October): 661–667.
Wolfson, Lewis W. 1985. The Untapped Power of the Press. Praeger.
139