Lame Ducks and Divided Government

Document technical information

Format pdf
Size 235.0 kB
First found May 22, 2018

Document content analysis

Category Also themed
Language
English
Type
not defined
Concepts
no text concepts found

Persons

Esther Duflo
Esther Duflo

wikipedia, lookup

Organizations

Places

Transcript

Lame Ducks and Divided Government: How Voters
Control the Unaccountable
Mark Schelker ∗
University of St. Gallen, CESifo and CREMA
Revised version: July 2013
Abstract
Divided government is not only the outcome of moderate voters’ electoral decision to balance
party ideology in government, but a more general reaction of voters to a systematic control
problem. Voters realize that term limited executives (“lame ducks”) cannot be held accountable
due to the missing re-election incentives. By dividing government control voters force a lame
duck to compromise on policies with an opposing legislature and restrict his ability to extract
rents. Based on US state data I present empirical evidence showing that the probability of
divided government is 8 to 12 percent higher when governors are lame ducks.
JEL Code: D72
Keywords: divided government, lame duck, term limit, accountability
∗
Corresponding Address: Mark Schelker, SIAW-HSG, University of St. Gallen, Bodanstrasse 8, CH-9000 St.
Gallen. Email: [email protected]
I would like to thank Jim Alt, Christine Benesch, Nels Christiansen, Reiner Eichenberger, Jürgen Eichberger,
Bruno Frey, Roland Hodler, Mario Jametti, Marcelin Joanis, Fabien Moizeau, Susanne Neckermann and Andrei
Shleifer for helpful comments and discussions. The paper also greatly benefited from comments by seminar
participants at the Universities of Fribourg, Lugano, Mannheim and St. Gallen and by conference participants of
the North American Summer Meeting of the Econometric Society in St. Louis, 2011, the International Society for
New Institutional Economics at Stanford, 2011, the World Congress of the International Economic Association in
Beijing, 2011, the International Institute of Public Finance at Ross School of Business, Ann Arbor, 2011, the
European Economic Association in Oslo, 2011, the European Public Choice Society in Rennes, 2011, the CESifo
Area Conference in Applied Microeconomics in Munich, 2012 and the European Political Science Association in
Berlin, 2012. The research is supported by the Swiss National Science Foundation Sinergia Project CRSI11130648.
1.
Introduction
Democracies put voters in charge of electing their governments assuming that voters are
capable of making informed electoral choices. The American political system is defined by its
strong separation of powers, whereby voters elect the members of the legislative and the
executive separately. This regularly leads to divided government in which the party holding the
executive power does not hold a majority in the legislature, requiring compromises on policy by
both parties. It has been argued that divided government is the result of rational electoral
decisions by moderate voters who balance political power between opposing party ideologies to
moderate policy outcomes (Fiorina 1992, Alesina and Rosenthal 1995, 1996). This requires
sophisticated voters who understand the institutional set-up, enabling them to make use of the
extensive checks and balances inherent to a system with such a strong separation of powers.
In this contribution I go one step further and argue that voters also use divided government
to mitigate control problems. Voters use divided government to control term-limited executives
(“lame ducks”) who are otherwise unaccountable due to their lack of re-election incentives.
Divided government forces lame ducks to compromise on policy with an opposing legislature. In
the empirical analysis, I attempt to determine whether there is a higher probability of divided
government in terms when the executive term limit is binding, that is, a U.S. state governor is a
lame duck. Note that the hypothesis requires voters to make rather sophisticated electoral
decisions. Nevertheless, I find robust evidence showing that lame ducks face an approximately 8
to 12 percent higher probability of divided government. This is (indirect) evidence of the ability
of voters to make coherent electoral decisions in a rather demanding situation: voters consider
the institutional setting when making electoral decisions.
In Section 2, I introduce the main theoretical arguments on which the empirical hypothesis
is based. I review the relevant insights obtained from the literature pertaining to divided
government and term limit legislation and formulate the testable hypothesis. In Section 3, I
present the data at the US state level and the empirical strategy. In Section 4, I provide empirical
results showing that the probability of divided government is significantly higher when US state
governors are lame ducks. This result is consistent with the view that voters balance a weakened
accountability mechanism by voting for a different party in the legislative branch. Section 5
summarizes and concludes the paper.
2
2.
Divided Government: The Reaction of Voters to a Systematic Control
Problem
This paper analyzes the interplay of term limits and divided government – two important
institutional characteristics of the American political system – and attributes such interplay to
rational voter behavior. To provide a foundation for the main argument that voters divide
government power to control an unaccountable lame duck, it seems useful to briefly discuss the
main arguments and mechanisms established in two relevant literatures.
2.1.
Causes of divided government
Divided government is a common phenomenon in presidential systems in which the
executive and legislative branches are elected separately by voters. A prime example is the
political system of the United States with its two main parties, the Republican and Democratic
Parties. At the federal level, divided government was the dominant form of government in the
US during the period from 1952 to 2010. Approximately 59 percent of all the governments were
divided in the sense that the Presidency and the US Congress – the Senate and/or the House of
Representatives – were dominated by opposing party majorities.
Fiorina (1992) and Alesina and Rosenthal (1995, 1996) analyze the causes of divided
government and argue that the division of government control is the result of rational voter
behavior. Voters at the center of the policy spectrum moderate policy by electing different
parties into the branches of government. 1 Alesina and Rosenthal (1995, 1996) build a formal
model that extends the standard spatial voting theory to include the separate election of the
executive and legislative branches with the possibility of dividing government control. In their
model, policy is viewed as a compromise between the executive and the legislative branch.
When both branches are held by one party – known as unified government – the party in power
can implement its preferred policy. When the branches of government are dominated by different
party majorities, the opposing parties in the different branches become veto players and are
forced to compromise on policy. Voters whose positions fall between the preferred party
positions take advantage of this legislative–executive interaction to moderate the policy
outcomes. Alesina and Rosenthal (1995) argue that divided government “[…] is not an undesired
1
Fiorina’s argument is not based on active moderating behavior. Moderate voters divide government power because
they offer policies closer to their bliss points compared to unified governments. In contrast, Alesina and Rosenthal
(1995) argument is based on an active strategic moderation effort of voters. The basic idea that voters actually
choose divided government is inspired by the well-documented phenomenon of split-ticket voting. For evidence of
split-ticket voting, see, for example, Fiorina (1992) and Garand and Glaslock Lichtl (2000). An alternative model of
split-ticket voting is provided by Chari, Jones and Marimon (1997).
3
result of a cumbersome electoral process, nor is it the result of a lack of rationality or of welldefined preferences of the electorate. Divided government occurs because moderate voters like
it, and they take advantage of “checks and balances” to achieve moderation. In dividing
government, the voters force the parties to compromise: divided government is a remedy of
political polarization” (Alesina and Rosenthal 1995: 44).
Their formal model is consistent with three important empirical observations: divided
government, split-ticket voting and the midterm cycle. The theory predicts divided government
as an outcome of voters’ electoral balancing of party ideologies in government, where moderate
voters split their tickets. In general elections, when simultaneously the executive and (partly) the
legislative are elected, voters are uncertain about which party will be holding the executive.
Voters who want to moderate policy by dividing government will therefore hedge the legislature
by dividing power between the legislative branches. Once the uncertainty about the incumbent
party in the executive is resolved, voters might want to moderate the government in midterm
elections further by shifting even more legislative power to the opposition party. This results in
the well-documented midterm cycle.
I extend the idea of balancing voter behavior beyond the ideological dimension. If voters
realize that providing veto power to different party majorities in the executive and the legislative
forces the different interests to compromise, they may also realize that they can use this
mechanism to mitigate other control problems. An intuitive example of a control problem is the
lack of electoral incentives of term limited executives potentially resulting in serious moral
hazard.
2.2.
Electoral accountability of term-limited executives
There has been much debate regarding the consequences of term limit legislation (for
example, Carey, Niemi and Powell 2000). The major disadvantage of term limit legislation stems
from the last period in office when the term limit is binding and the executive becomes a lame
duck. 2 Executives who care about maintaining a reputation for the purposes of being re-elected
must introduce policy that is in accordance with voter preferences. Being ineligible to run for reelection eliminates this powerful incentive. In most political agency models, re-election
incentives are the main channel through which to align the preferences of the office holder with
those of the voters. The lack of accountability to voters increases an agent’s incentives to engage
2
Note that in line with the previous literature I will generally refer to governors facing a binding term limit as “lame
ducks”. This is not meant as any particular qualification of term-limited governors’ behavior. It is conceivable that
unaccountable office holders invest more in acquiring private rents, but they might also become more independent
from party or special interests and invest more in the provision of public goods, and so on.
4
in opportunistic behavior, which could manifest in, for example, low levels of effort, activities
that favor specific interests or legacy building. By testing a political agency model, Besley and
Case (1995a) show that US state governors who are subject to a binding term limit implement
systematically different fiscal policies from governors who are eligible for re-election. They
suggest that governors who are eligible for re-election are concerned about reputation building
and hence adjust their economic policy choices according to this constraint, while lame ducks do
not have such incentives. In states with binding term limits, they find fiscal cycles with higher
taxes and expenditures in the last term when the governor is a Democrat. 3 List and Sturm (2006)
explore a political agency model in which policy makers decide on primary and secondary policy
issues and face binding term limits during the second period. They show that binding term limits
even influence secondary policy issues, such as environmental policy. 4 Incumbents implement
environmental policies to attract votes if they are eligible for re-election, but much less when a
term limit is binding. Ferraz and Finan (2011) find that re-election-ineligible second-term
mayors in Brazilian municipalities are significantly more corrupt than first-term mayors who are
eligible to run for re-election.
One obvious question is why voters actually re-elect a governor into a lame-duck term, if
moral hazard of the executive becomes an issue. By backward induction it becomes apparent that
if voters commit to not re-electing an incumbent into the last lame-duck term, the governor
actually becomes a lame duck already in the previous period. Hence, a term limit in whatever
future term would factually be reduced to a one-term limit (Alt, Bueno de Mesquita and Rose
2011). Alt, Bueno de Mesquita and Rose (2011) analyze an electoral model in which voters want
to select competent office holders and they want to hold them accountable. Elections provide the
instrument to select and retain good types and weed-out bad types. Intact electoral prospects
encourage incumbents to exert costly effort in order to become re-elected and signal competence.
The authors assume that only competent governors who exert effort have a positive probability
of being successful in fiscal and economic policy. They identify competence effects by
comparing newly elected and re-elected incumbents and accountability effects by comparing reelection-eligible with re-election-ineligible incumbents. An incumbent is re-elected if he is more
competent than the challenger in expectation. Voters benefit from the increasing competence of
an incumbent due to his or her accumulated experience in the previous term. However, due to the
3
Johnson and Crain (2004) extend the analysis to a cross-country setting. They find higher expenditures and taxes
and evidence of a fiscal cycle in which expenditures increase in lame-duck terms and decrease in terms in which an
executive is eligible for re-election.
4
Secondary policy issues are often believed to be less important for re-election than primary issues. Secondary
policy issues concern only smaller groups and voters remain uninformed due to the multitude of such policy issues
(List and Sturm 2006: 1249–50).
5
lack of re-election incentives in the lame-duck period the candidate has no incentives to exert
effort. In case of a binding term limit in the second term voters face a competent (selection) but
potentially lazy (moral hazard) lame-duck incumbent. The authors show that economic growth is
higher and tax and spending growth as well as borrowing cost are lower for incumbents who are
eligible to run for re-election as well as for re-elected governors.
Explaining why term limits have been introduced seems more controversial. Most
contributions base their arguments on the well-documented incumbency advantage (for example,
Gelman and King 1990, King and Gelman 1991, Ansolabehere and Snyder 2002, 2004), which is
reduced by introducing term limits. It has been argued that policy makers with more seniority
can more effectively transfer resources to their electoral districts (for example, Dick and Lott
1993, Buchanan and Congleton 1994, Friedman and Wittman 1995), that incumbents exploit
office benefits and take advantage of the higher level of television coverage (for example,
Ansolabehere, Snowberg and Snyder 2006, Prior 2006) and that incumbents are able to affect the
salience of policy issues to their advantage (Hodler, Loertscher and Rohner 2010). Daniel and
Lott (1997) show that the introduction of legislative term limits in California reduced campaign
expenditures and increased electoral competition. They attribute these effects to the reduced
returns to political careers, which cause new candidates to enter electoral races because
campaign expenditures and incumbency advantages are lower.
2.3.
How voters use divided government to control lame-duck executives
A straightforward extension of the ideas by Fiorina (1992) and Alesina and Rosenthal
(1995, 1996) is that the policy balancing behavior of moderate voters is not restricted to party
ideology, but is a more general reaction to systematic control problems. One such example is the
lack of electoral accountability of term-limited executives. Voters anticipate the weakened
incentives of a term-limited executive to pursue public rather than private interests. They use the
electoral mechanism by voting for the opposing party in the legislative branch in an attempt to
divide government control. Divided government constrains a lame-duck executive because it
forces him to compromise with the legislative branch on policy matters.
However, policy moderation by means of divided government may come at a cost if the
different parties in the executive and legislative branches cannot compromise on policy, which
may lead to gridlock of the policy-making process and obscured accountability. The alleged
gridlock of the policy-making process has been at the core of much scholarly debate (for
example, Cutler 1988, Sundquist 1988, Cox and McCubbins 1991, McCubbins 1991). Mayhew
(1991) argues that in terms of “significant” legislative enactments, there is no evidence of policy
6
stalemates in the United States. However, it has been shown that the evaluation of the effect of
divided government on legislative productivity depends heavily on the definition of “significant”
enactments, the definition of gridlock as well as additional factors, such as party polarization and
within-party ideological heterogeneity (for example, Krehbiel 1996, Binder 1999, Coleman
1999, Howell et al. 2000, Bowling and Ferguson 2001, Jones 2001, Rogers 2005, Saeki 2009).
With regard to fiscal policy, the evidence suggests that divided governments are less able to
respond to fiscal shocks and that responses are stronger on the spending than the revenue side
(Alt and Lowry 1994, Poterba 1994).
Hence, the decision to moderate policy by means of divided government depends on the
relative cost of divided government. Voters must weigh the cost of potential policy gridlock
against the cost of a lame duck wielding executive powers. If the expected costs of divided
government are lower than the expected costs of having a lame duck, then voters can opt for
policy moderation by dividing government control. While the deviation from previous-term
politics of lame-duck executives is well established, 5 the cost of policy making under divided
government remains ambiguous. In the empirical analysis I attempt to control for the alleged
drivers of the cost of divided government and I include party ideology and ideological
heterogeneity into the econometric framework. Failing to fully control for the cost of divided
government should result in downward bias of the coefficient of interest and make it more
difficult to find a systematic relationship between lame ducks and divided government.
The underlying mechanism leading to divided government is based on the number of seats
in the legislature. Therefore, it could also be interesting to analyze whether the anticipated
effects are also observable in the underlying distribution of legislative seats. We could expect
that the party of the governor loses seat shares in the lame duck term. However, the main
hypothesis of this contribution is based on a requirement of two opposing veto players to
compromise over policy in the two branches of government. This situation is effectively
achieved by divided government. The veto power leading to policy moderation is a
discontinuous function of the distribution of seats in the legislature. Hence, the main focus of the
empirical investigation is on divided government and the (consistent) empirical evidence based
on legislative seat shares is relegated to the Online Appendix.
3.
Data and Empirical Strategy
In line with the previous literature I use data for the 48 US mainland states from 1970 to
5
Note that for the main argument it suffices that lame ducks deviate from previous-term behavior and voters are
uncertain about the direction of the deviation, be it that they become particularly “public-spirited” or selfish.
7
2000. The US states are an ideal testing ground on which to assess the theoretical predictions.
First, many US states have implemented executive term limits. Out of the 50 states, 37 feature
binding executive term limits in 2000, many of which were introduced following voter
initiatives. During the period of this study, on average 26 percent of governors were lame ducks.
Table 1 provides an overview of the term limit legislation in each state. Second, the executive
and legislative branches are both elected directly by citizens. At the US state level, divided
government occurred 48 percent of the time during the period from 1970 to 2000.
[Table 1 about here]
3.1.
The data
The information pertaining to party majorities in the branches of government for the period
from 1970 to 2000 is primarily based on Alt, Lassen and Rose (2006). The data include
information regarding which party holds the executive and which party holds the majority in the
two legislative chambers. The dependent variable is an indicator variable that takes the value 1
when there is any form of divided government control (and parties need to compromise),
whether this control is divided between the executive branch and both chambers of the
legislature or the majorities are split in the legislative chambers. The bulk of the independent
variables stems from List and Sturm (2006). They provide information on state term limit
legislation (see Table 1), term-limited governors (lame ducks), and the electoral margin of
incumbent governors relative to challengers as well as economic and socio-demographic
characteristics. 6 Information pertaining to the timing of legislative elections in each state was
provided by Tim Storey from the National Conference of State Legislators (NCSL). Appendix
Tables A1 and A2 present summary statistics and data sources. 7
3.2.
Empirical strategy
The empirical strategy aims at identifying the causal effect of lame ducks on divided
government. The following empirical results start by establishing first baseline estimates that
focus on the main variables of interest only. I then extend the framework to eliminate potential
sources of bias such as gubernatorial experience due to electoral selection (Alt, Bueno de
Mesquita and Rose 2011) and general last-round effects that may also affect non-term-limited
governors. In a further step I concentrate on quasi-random variation in the empirical selection of
6
I cross-checked and, if necessary, corrected the data used in this study with the (original) data source by Besley
and Case (2003) and the relevant later studies by Alt, Bueno de Mesquita and Rose (2011) and Besley, Persson and
Sturm (2010).
7
Table OA.1 of the Online Appendix presents the yearly summary statistics of the main variables of interest.
8
governors and finally, I control for variables that are anticipated to have a potential influence on
divided government. The estimation approach is primarily based on difference-in-differences
estimation comparable to the related work by, for example, Besley and Case (1995a, 2003) and
List and Sturm (2006), who use similar data.
I estimate variants of the following general specification:
yit = β lame duckit + Iit ζ + Xit λ + μi + τt + εit
yit is a dummy variable capturing the form of government (1 if divided government, 0 if
unified government) in state i in year t. lame duckit is a dummy variable that takes the value 1 if
the executive is a lame duck and 0 otherwise. Furthermore, Iit is a vector that includes potentially
important institutional and gubernatorial characteristics (for example, term limit, vote margins,
number of consecutive terms) and Xit is a vector of additional (for example, political, economic,
socio-demographic) controls. β is the parameter of interest, ζ and λ are parameter vectors, μi and
τt are the state and year fixed effects and εit is the error term. The subscripts i = 1,…, n and t =
1,…,T indicate the state and year, respectively.
In a setup with a binary dependent variable natural specifications include linear probability
models or non-linear models such as logit or probit. The results in the text are primarily based on
linear probability models, which are typically good approximations, simple to interpret and
widely used in economic research. Conditional logit estimates are presented in the baseline
regressions and in the Online Appendix afterwards. 8 Because basic difference-in-differences
estimates might ignore autocorrelation in US state data (Bertrand, Duflo and Mullainathan
2004), I adjust the standard errors for clustering at the state level, which allows for arbitrary
correlations of the errors within states. 9
The main attention is on identifying the effect of lame duck governors on the occurrence of
divided government. Because lame ducks do not occur randomly I have to control for the two
main factors determining whether a governor can actually become a lame duck: whether there is
a term limit and a governor’s vote margin in the past election. Typically term limits have not
been introduced with the main intention to create an unaccountable last term, but to reduce the
perceived negative effects of long term incumbents. Therefore, term limit legislation per se is
likely to have an independent effect. For identification it is thus important to separate these
potentially divergent effects. The vote margin of an incumbent governor is important from two
8
The results from the logistic regressions are consistent with the results of the linear regressions. Note that due to
the incidental parameter problem fixed effects logit estimates are likely to be biased (Neyman and Scott 1948,
Lancaster 2000, Greene 2008) and, hence, I rely on conditional logit models.
9
The results also remain robust to the inclusion of state-specific time trends (not reported).
9
perspectives: First, it captures the popularity of an incumbent or candidate relative to the
popularity of a challenger in the electoral race. I expect that more popular candidates or
incumbents with higher vote margins should face a lower probability of confronting an opposing
party majority in the legislative branch. Second, the vote margin should be an unbiased ex ante
indicator of the predictability of the (re-)election of a candidate, which influences the ability of
voters to effectively moderate policy. From the theoretical discussion I expect the lame-duck
coefficient to be positive. For the term limit variable, I do not have an a priori hypothesis
regarding the direction of the effect.
Beyond the baseline specifications I control for differences in executive mandates across
states. Most governors serve four-year terms, and a few governors serve short terms of two
years. I exclude observations from the few states (for some period of time) that limit governors
to one term in office only, which makes all governors to be lame ducks at all times. Because not
all states follow identical electoral systems and electoral rhythms, I control for the electoral cycle
and I include an indicator identifying midterm congresses. Likewise, this allows me to evaluate
potential midterm dynamics (for example, Alesina and Rosenthal 1996).
The empirical strategy develops as follows: First, I present the simplest possible
specifications establishing the baseline results (4.1.). Second, I eliminate specific channels
introducing potential bias due to electoral selection and experience as well as last-round effects
(4.2.). Third, I restrict the sample to closely elected governors in order to introduce some quasirandom variation (4.3.). Fourth, I conduct a series of robustness checks by introducing additional
potentially influential factors such as political preferences, preference heterogeneity and party
affiliation as well as state economic and demographic characteristics (4.4.).
4.
4.1.
Empirical Results
Baseline results
Columns 1 to 3 of Table 2 report the regression results from the linear probability models,
and columns 4 to 6 report the results from (conditional) logit models. I first start with a random
effects specification (column 1 and 4) and only then introduce cross-section fixed effects
(columns 2 and 5). 10 In columns 3 and 6, I further control for differences in the gubernatorial
10
Unlike a conventional linear random-effects estimator, the random-effects logit estimator does not depend on the
assumption that the random effect is uncorrelated with the independent variables (Wooldridge 2002: 490). Thus, it is
possible to obtain a consistent estimator of the variable of interest without any assumption pertaining to the
relationship of the cross-section component with the independent variables. When comparing the estimates from the
10
term length (short term of 2 years) as well as for potential legislative shifts after midterm
elections. All presented specifications include year fixed effects.
[Table 2 about here]
Lame ducks
The lame-duck coefficient is in all specifications positive and statistically significant
indicating a 10 percent (12 percent based on the estimated marginal effect in the conditional logit
model) higher probability of divided government if the term limit is binding. The term limit
variable has an independent negative effect on divided government (more extensive discussion
below). The vote margin has a negative and significant impact on divided government
suggesting that more popular governors have a lower probability to confront an opposing
congress. 11
Overall, I consistently find that lame-duck governors are associated with a significantly
higher probability of divided government. This finding is in line with the prediction that the
impaired accountability of a lame duck increases the inclination of voters to counterbalance this
control problem by voting for divided government. In the linear regressions, the size of the
coefficients suggests a 10 percent and the marginal effects of the logistic regressions suggest a
12 percent higher probability of divided government when a governor is a lame duck.
Term limits
I find a robust negative correlation between term limit legislation and divided government.
Not surprisingly, this suggests that term limits have an effect that goes beyond just introducing a
last unaccountable term. The coefficient, however, is more difficult to interpret and several
interpretations are possible. First, the negative effect could originate from voter preferences.
Based on the literature pertaining to term limits, several interpretations could apply. For
example, term limits eliminate incumbency advantages after a few periods in office and the lack
random-effects and fixed-effects specifications, it is comforting that the estimated coefficients of the main variable
of interest are not dependent on this choice.
11
Subsamples: In Table OA.2 of the Online Appendix I reports regression results focusing on specific subsamples.
First, I restrict the sample to only include years after general or midterm elections. One concern could be that the full
sample of years between 1970 and 2000 could yield biased estimates, because all congress years are included in the
sample. The reason for the inclusion of all years in the baseline specification is primarily to maintain the panel
balanced, because the states follow different electoral cycles. Despite the reduced sample size the results based on
the restricted sample remain almost identical (Table OA.2, columns 1–3). Second, I use only the subsample of states
with term limit legislation (Table OA.2, columns 4–6). A potential concern could be that term limit states are
different from non-term limit states and that this difference is not controlled for by the term limit variable. I
consistently find positive and significant effects of the lame-duck variable, which indicate an 11 to 15 percent higher
probability of divided government. The magnitude of the effect is slightly higher but still comparable in size to the
estimates that include the full sample of states.
11
of such advantages increases electoral competition (for example, Daniel and Lott 1997). 12
Alternatively, term limits might enable voters to exchange long-term incumbents while keeping
the same party in the executive. This ability may suit the interests of voters if incumbents tend to
accumulate power over time and increasingly shirk or become corrupt with longer tenure.
Moreover, term limits may be favored because they enable voters to anticipate last-round effects
and coordinate their moderation efforts – namely, to counterbalance the missing accountability
by divided government control – to the specific lame-duck term. Without term limits, voters
remain uncertain regarding which term is a governor’s final term, in which re-election incentives
do not apply. Therefore, voters might be induced to hedge constantly against the risk of moral
hazard of the incumbent in a potential last-round term. Other reasonable interpretations of the
negative effect of term limits might apply, but I cannot discriminate between these possible
interpretations.
Second, the influence of term limits is net of last-period effects (due to controlling for lame
ducks), while last-period effects might also occur with non-term-limited executives. If electoral
defeat is anticipated or an incumbent decides not to seek re-election for other reasons, moral
hazard also becomes an issue for non-term-limited incumbents. Put differently, the selection of
governors in term limit states whose term limits are not yet binding are compared with the entire
group of governors in states without term limits. While the first group is potentially net of lastperiod effects, the second group includes governors anticipating retirement or expecting to be
voted out of office, which may also result in last-period effects. For all these reasons, I cannot
conclusively interpret the coefficient.
Electoral dynamics in midterm elections
When faced with greater uncertainty regarding who will be holding the executive office,
voters may find the task of moderating policy by means of divided government more difficult.
Alesina and Rosenthal (1996) argue that in general election years with great uncertainty about
which party will be holding the executive, voters who want to moderate policy by dividing
government will hedge the legislature by dividing power between the legislative branches. Once
the uncertainty about the incumbent party in the executive is resolved, voters might want to
moderate the government in midterm elections further by shifting even more legislative power to
the opposition party. This results in the well-documented midterm losses of the party holding the
executive. Folke and Snyder (2011) analyze gubernatorial midterm slump and show that in
12
I also investigate whether controlling for the intensity of electoral competition has an influence on my main
results. The measure on state electoral competition stems from Besley, Persson and Sturm (2010). All results remain
robust.
12
midterm elections the party of the governor experiences an average seat share loss of about 3.5
percent. Such dynamics might also be observable in our set-up. The results presented in Table 2
show that midterm congresses per se do not seem to have a significant influence on the
probability of divided government. I also investigate whether there is a specific midterm effect
for lame-duck governors and estimate an interaction effect of lame duck and the midterm
congress variable. The estimated interaction effect is not statistically significant. The baseline
lame-duck coefficient itself, however, remains robust to this change in the model specification.
This is consistent with the finding in Folke and Snyder (2011) who study the determinants of the
well documented midterm cycle. These authors do not find a significant effect of lame duck
governors on the legislative seat share of the incumbent governor’s party in midterm elections.
Because lame ducks are incumbent officeholders at the time of the relevant general election and
the size of the well-documented incumbency advantage is with about 8 to 10 percent relatively
large (for example, Ansolabehere and Snyder 2002, 2004), voters do not typically face a high
degree of uncertainty about the future officeholder. Therefore, these results are not entirely
surprising.
4.2.
Selection of governors
A major concern could be that lame-duck governors are of a particular selection of
incumbents. In that case it might not be the fact that the governor is a lame duck and that voters
anticipate a moral hazard problem but some characteristic not particularly related to lame-duck
governors that drives the results.
Consecutive terms and governor experience
So far, I have mainly suggested the moral hazard channel due to the missing accountability
of a term-limited executive without re-election incentives. However, lame-duck governors are reelected and hence, preselected and experienced executives. On average, lame ducks may be of
higher quality than newly elected governors. This concern relates to Alt, Bueno de Mesquita and
Rose (2011), who argue that lame ducks are more competent governors, because they have been
re-elected and elections weed out incompetent incumbents. Alternatively, voters might just
become weary of long-term governors and hence might want to restrict governors’ influence
over public policy. To identify the influence of the missing accountability (moral hazard), I
control for selection and competence effects (Table 3). I first control for the number of terms in
office (# governor terms) and then, more specifically, I compare lame ducks with re-elected and
experienced governors in states without term limits. Therefore, I control for incumbents in
13
consecutive terms in states without term limit legislation (no limit & term > 1). Finally, I
introduce the age of the governor as an alternative experience measure. This approach should
clarify the concern that any effect may be merely a result of electoral selection or competence
effects reflecting political experience or office tenure. This should be instructive of whether the
previously found lame-duck effect is specific to general experience or governors in their lameduck term.
Column 1 of Table 3 includes the number of consecutive governor terms (# governor
terms), column 2 contains the regressions controlling for governors who are not term limited and
are not serving their first term (no limit & term > 1), and column 3 presents regression results
when controlling for the age of the governor. The estimated coefficients of the lame-duck
variable are always statistically significant and almost identical in size to the baseline in Table 2.
The estimated coefficients of the number of office terms (# governor terms), the variable
capturing the competence effect of a non-term-limited office holder (no limit & term >1) as well
as the age of the governor (Governor age) are not statistically different from zero.
[Table 3 about here]
Last-period effects
Last-rounds and hence moral hazard are not restricted to term-limited governors. Any
governor who decides not to seek re-election or anticipates electoral defeat faces an expected last
term. Therefore, moral hazard could also be an issue for governors without term limit legislation.
Depending on the situation voters might be more or less able to accurately predict the last period
of non-term-limited governors. This influences their ability to divide government control as a
reaction to the missing re-election incentives. In a first step, I control for potential last-round
effects by including an indicator for all governors who did not run for re-election or suffered
subsequent electoral defeat either in the primary or general election. In a second step, I
concentrate on last-round effects that are (potentially) more easy to anticipate and I focus on
incumbents who did not run for re-election or who clearly lost the upcoming re-election (defeat
by a vote margin > 5%). The idea is that voters might be able to anticipate last-round effects of
governors with low re-election probabilities. If this is true, voters might divide government
control as a precaution for a potential last term. In a third step, I analyze whether the upcoming
electoral defeat becomes easier to anticipate over the course of the actual term. Then voters
might divide government control in midterm elections when the uncertainty over the electoral
chances of the incumbent governor further decreases.
14
[Table 4 about here]
Column 1 of Table 4 presents the regression results including the indicator of subsequent
electoral defeat. Column 2 includes an indicator for governors who did not run for re-election or
clearly lost subsequent re-election by more than a 5 percent vote margin (clear defeat (margin >
5)). The estimated coefficients of all last-round measures are not statistically significant. As can
be seen from Table 4, all the previous estimates of the influence of lame ducks on divided
government remain robust. The lame-duck coefficient is always statistically significant and it
varies between 0.083 and 0.111. Because the re-election prospects can change over time in
office, unsuccessful governors (and the respective voters) might anticipate electoral defeat only
in the course of time in office. In this case, last-round effects become more relevant in the last
years in office. Therefore, shortening the time horizon to the next gubernatorial election might
provide additional insights. When focusing on midterm congresses of governors who
subsequently (within 2 years) lost their re-election bid with a margin larger than 5 percent, I find
similar results (column 3). The estimated interaction effect of midterm congress and clear defeat
(margin > 5) does not reach statistical significance. This indicates that it might be very difficult
for voters to anticipate last-round effect. From this perspective term limits assist voters to
coordinate their moderation efforts.
4.3.
Closely elected governors
In the previous section I explicitly addressed channels of electoral selection of governors
by focusing on governor experience and last-round effects. Even though I eliminated various
sources of potential bias, in non-experimental setups there is always the concern that some
unobserved factor might still bias the results. Similar to, for example, Folke and Snyder (2011), I
include a low-order polynomial of the vote margin as a control function and I condition on
governors who were elected in close electoral races with vote margins equal to or smaller than 5,
4, 3, 2 or 1 percent. The fundamental idea is that closely elected governors are essentially
randomly assigned to hold office (Folke and Snyder 2011), which reduces the potential bias due
to unobserved heterogeneity. Such a design focuses on a subsample of officeholders to make
statistical inference. To the same degree as the statistical advantages of randomization are
acknowledged, one has to carefully evaluate if the selected subsample is useful for inference in
the context of the theory at hand.
In the present case, it seems highly problematic to condition on close races in which an
incumbent is running for re-election. Incumbents typically benefit from a sizable incumbency
15
advantage. 13 Taking a subsample of close races including lame-duck incumbents would focus on
incumbent governors who actually forfeited a sizable incumbency advantage. A design using
such a heavily selected sample of (incompetent) governors is likely to produce biased estimates.
Therefore, in the spirit of Lee, Moretti and Butler (2004) I focus on close elections for the initial
term of consecutive appointments. This should yield a “quasi-experimental” set-up because the
initial appointment is close to a random assignment. The subsample used to make statistical
inference is hence based on governors who stem from a quasi-random selection from the
underlying pool of candidates.
[Table 5 about here]
In order to flexibly control for the influence of the vote margin I first include low-order
polynomials of the governor vote margins (Table 5, columns 1 to 3) and then effectively restrict
the sample to governors who were initially elected with a vote margin ≤ 5, 4, 3, 2 or 1 percentage
points (columns 4 to 7). Note, however, that this reduces the sample size considerably. Overall,
the estimated coefficients are similar in size to the previous estimates, but more sensitive to
changes in the model specifications when focusing on the restricted samples. When restricting
the sample to races with margins ≤ 3 percentage points the estimated coefficients become
insignificant but remain similar in terms of the size of the coefficient for races with margins
larger than 1 percent.
4.4.
Extensions of the baseline model
As a further extension I examine political factors such as political preferences and political
heterogeneity, which might have a direct influence on the cost and occurrence of divided
government, and the party affiliation of the governor. Moreover, I include state economic and
demographic covariates.
Political preferences and party affiliation
As has been argued previously, the decision of voters to moderate policy by means of
divided government depends on the relative cost of potential policy gridlock versus the cost of a
lame duck wielding executive powers. Voters have to weigh these costs to make electoral
decisions. The cost of divided government is likely to depend on the distance between the policy
preferences of the leading parties. If the party positions are very polarized, moderate voters may
feel a greater need for moderation by dividing government control. Alternatively, the cost of
13
Ansolabehere and Snyder (2002, 2004) estimate the gubernatorial incumbency advantage to be around 8 to 10
percent.
16
divided government may increase because it becomes more difficult for the different parties in
government to compromise and agree on policy. This difficulty may result in a higher probability
of gridlock. Hence, the cost of policy moderation by means of divided government is likely to be
related to political preference heterogeneity and party polarization. 14 Political heterogeneity
could influence the main result if, for some reason, the less heterogeneous states re-elect termlimited executives more often in their lame-duck terms and simultaneously have a higher
probability of divided government due to the lower cost of policy gridlock.
When constructing measures of political preferences and heterogeneity, one faces the
problem that for the considered time period there is no standard measure of preferences at the
state level. As an approximation, I use the first dimension of the DW-Nominate scores proposed
by McCarty, Poole and Rosenthal (2006). The DW Nominate scores measure the liberal–
conservative attitudes from all of the roll-call votes of the state delegates in Federal Congress.
This approach has been applied by numerous authors, such as Hanssen (2004), Alt, Lassen and
Rose (2006) and Garand (2010). Typically, measures of political polarization represent the
absolute difference between the scores of Democratic and the scores of Republican delegates.
When calculating such a measure at the state level, one has to take into account that some states
do not have delegates from both parties in one or both chambers of Congress. This problem
causes the appropriate calculation of a polarization measure according to the mean (median)
distance of party representatives to be impossible without further assumptions. Moreover, it
seems that greater political preference heterogeneity, whether within a party or across parties,
would generally lead to a more difficult decision-making process (for example, Jones 2001 or
Saeki 2009). Therefore, I use the mean and standard deviation of the DW-Nominate score as
measures of political preferences and political heterogeneity, respectively. 15 I do not have an ex
ante hypothesis about the direction of the estimated effect of the political preference measure
because I have no theory regarding the influence of political ideology (liberal or conservative) on
divided government control. The hypothesis pertaining to the effect of political heterogeneity is
ambiguous. Greater political heterogeneity could lead to a higher probability of divided
government because more voters may feel the need for moderation. However, the potential for
policy gridlock as a result of divided government depends on the heterogeneity of policy
preferences. More heterogeneous political preferences could be associated with greater potential
14
For more information regarding party polarization in the US, see Poole and Rosenthal (1991, 1997), McCarty,
Poole and Rosenthal (2006) and Garand (2010).
15
I construct equivalent measures using adjusted ADA scores (Anderson and Habel 2009), which measure the
liberal–conservative attitudes of the members of Federal Congress according to selected roll-call votes by the
interest group Americans for Democratic Action (Groseclose, Levitt and Snyder 1999). The results are entirely
robust to the use of ADA scores rather than the more encompassing DW-Nominate scores that include all roll-call
votes.
17
for policy gridlock and thus greater costs of divided government control. I do not have an a priori
expectation regarding the direction of the net effect.
[Table 6 about here]
In column 1 of Table 6, I add the measures of political preferences and political
heterogeneity. Again, the inclusion of these two variables does not affect the main result.
Although I do not find a significant effect for the measure of political preferences, which is the
mean of the DW-Nominate score obtained by McCarty, Poole and Rosenthal (2006), I find that
political heterogeneity, as measured by the standard deviation of the same variable, has a
positive and significant effect on divided government. This positive net effect could indicate that
the policy moderation motive of voters is stronger than the negative effects of the expected cost
of divided government as a result of the increased potential for policy gridlock.
Column 2 controls for the party affiliation of a governor and includes a dummy variable
that takes the value of 1 if there is a democratic governor. I do not have an a priori hypothesis
pertaining to the effect of this control variable, but I want to ensure that the variable of interest
does not capture some unobserved party effect. The gubernatorial party affiliation does not have
a statistically significant impact on divided government, while the main variable of interest
remains positive and statistically significant.
Demographic and economic factors
Finally, columns 3 and 4 of Table 6 add potentially important sets of covariates to the
baseline specifications. These include the state population, real per capita income,
unemployment rate and finally the growth rates of income and unemployment. Larger states
might just be different from smaller states. They might be more heterogeneous, more difficult to
govern and the like, which may translate into different electoral behavior. The economic
variables and their growth rates may affect voter behavior in the ballot if they consider the
current economic situation and the economic development during the previous period when
making electoral decisions (economic voting).
The estimated effect of the state population and the real per capita income are both positive
and statistically significant, while all other variables do not reach standard levels of statistical
significance. I also included further covariates that reflect various dimensions of population
heterogeneity such as the population density, the fraction of inhabitants with a high-school
diploma, the fraction of colored population. None of the additional covariates affect the main
result that lame-duck governors face an approximately 8 to 12 percent higher probability of
18
divided government.
5.
Conclusions
The ability of voters to make informed and coherent decisions is a precondition for
functioning democracies. Economists typically assume that voters are capable of making
informed decisions. In the famous model by Alesina and Rosenthal (1996), moderate voters
rationally divide government control to achieve policy moderation. Such moderating voting
behavior requires a fair amount of sophistication on the part of voters. In this paper, I go one step
further and argue that voters use divided government to mitigate control problems more
generally. This again requires a considerable capacity of voters to process information in a rather
demanding environment. I focus on an electoral situation in which an incumbent executive is
facing a binding term limit. Such lame-duck executives cannot be incentivized by the electoral
process because they cannot run for re-election due to term limit legislation. I analyze whether
voters consider this impaired accountability of the incumbent executive and use divided
government as an alternative electoral instrument to counterbalance the weakened accountability
of a lame duck. In a presidential system in which the executive and legislative branches are
elected directly, voters can divide government control to moderate policy outcomes. Divided
government forces the opposing party majorities in both branches of government to compromise
on policy. The hypothesis predicts that lame-duck governors have a higher probability of being
confronted with an opposing party majority in the legislature than governors with intact reelection incentives. I test my hypothesis using US state data from 1970 to 2000. The results
document a clear pattern. Consistent with the theoretical arguments, I find that lame-duck
governors face an approximately 8 to 12 percent higher probability of divided government. The
effect remains robust to various model extensions and specification changes. Overall, the results
are consistent with the interpretation that voters use divided government to control
unaccountable executives who do not have re-election incentives. This is (indirect) evidence of
the considerable willingness and capacity of voters to process complex information.
At least three important consequences for future research arise: First, there has been ample
discussion about voters’ willingness and ability to make informed decisions, which show that
voters make relative performance comparisons to evaluate an incumbents performance, but are
not always perfectly capable of distinguishing signal from noise (for example, Besley and Case
1995b, Wolfers 2007). This paper adds to this discussion by showing that electoral decisions
depend on the institutional environment. In the context of this study voters (in the aggregate)
seem to be able to make informed electoral decisions, even in a rather demanding setup. More
19
research is needed to better understand the context of such choices. Second, most reduced-form
analyses of political institutions do not consider systematic voter reactions affecting the interplay
of different political institutions and thus policy outcomes. Hence, more research is due to
establish a more thorough understanding of voters’ electoral decisions taking institutional
interactions into account. Third, as a side product this study shows that analyses focusing on the
influence of term limits cannot merely concentrate on the lame duck term but must take into
account that term limits have an independent influence that goes beyond the last period effect.
20
References
Alesina, Alberto, and Howard Rosenthal (1995). Partisan Politics, Divided Government, and the
Economy. Cambridge University Press, Cambridge, UK
Alesina, Alberto, and Howard Rosenthal (1996). A Theory of Divided Government.
Econometrica 64 (6): 1311–1341
Alt, James E., and Robert C. Lowry (1994). Divided Government, Fiscal Institutions, and Budget
Deficits: Evidence from the States. American Political Science Review 88 (4): 811–828
Alt, James E., David Dreyer Lassen, and Shanna Rose (2006). The Causes of Fiscal
Transparency: Evidence from the American States. IMF Staff Papers
Alt, James E., Ethan Bueno de Mesquita, and Shanna Rose (2011). Disentangling Accountability
and Competence in Elections: Evidence from U.S. Term Limits. Journal of Politics 73 (1): 171–
186
Anderson, Sarah, and Philip Habel (2009). Revisiting Adjusted ADA Scores for the U.S.
Congress, 1947–2007. Political Analysis 17: 83–88
Ansolabehere, Stephen, and James M. Snyder, Jr. (2002). The Incumbency Advantage in US
Elections: An Analysis of State and Federal Offices, 1942–2000. Election Law Journal 1 (3):
315–338
Ansolabehere, Stephen, and James M. Snyder, Jr. (2004). Using Term Limits to Estimate
Incumbency Advantages When Officeholders Retire Strategically. Legislative Studies Quarterly
29 (4): 487–515
Ansolabehere, Stephen, Erik C. Snowberg, and James M. Snyder, Jr. (2006). Television and the
Incumbency Advantage in U.S. Elections. Legislative Studies Quarterly 31 (4): 469–490
Bertrand, Marianne, Esther Duflo, and Sendhil Mullainathan (2004). How Much Should We
Trust Differences-in-Differences Estimates? Quarterly Journal of Economics 119 (1): 249–275
Besley, Timothy, and Anne C. Case (1995a). Does Electoral Accountability Affect Economic
Policy Choices? Evidence from Gubernatorial Term Limits. Quarterly Journal of Economics 110
(3): 769–798
Besley, Timothy, and Anne C. Case (1995b). Incumbent Behavior: Vote-Seeking, Tax-Setting,
and Yardstick Competition. American Economic Review 85 (1): 25–45
Besley, Timothy, and Anne C. Case (2003). Political Institutions and Policy Choices: Evidence
from the United States. Journal of Economic Literature 41: 7–73
Besley, Timothy, Torsten Persson, and Daniel M. Sturm (2010). Political Competition, Policy
and Growth: Theory and Evidence from the US. Review of Economics Studies 77: 1329–1352
Binder, Sarah A. (1999). The Dynamics of Legislative Gridlock. American Political Science
Review 93 (3): 519–533
Bowling, Cynthia J., and Margaret R. Ferguson (2001). Divided Government, Interest
Representation, and Policy Differences: Competing Explanations of Gridlock in the Fifty States.
Journal of Politics 63 (1): 182–206
Buchanan, James M., and Roger D. Congleton (1994): The incumbency dilemma and rent
extraction by legislators. Public Choice 79: 47-60
Carey, John M., Richard G. Niemi, and Lynda W. Powell (2000). Term Limits in the State
Legislatures. University of Michigan Press, Ann Arbor (MI)
Chari, V.V., Larry E. Jones, and Ramon Marimon (1997). The Economics of Split-Ticket Voting
in Representative Democracies. American Economic Review 87 (5): 957–976
21
Coleman, John J. (1999). Unified Government, Divided Government, and Party Responsiveness.
American Political Science Review 93 (4): 821–835
Cox, Gary W., and Mathew D. McCubbins (1991). Divided control of fiscal policy. In: Cox,
Gary W., and Samuel Kernell (Eds.). The Politics of Divided Government. Westview Press,
Boulder
Cutler, Lloyd N. (1988). Some reflections about divided government. Presidential Studies
Quarterly 18: 485–492
Daniel, K., and J. Lott Jr. (1997). Term Limits and Electoral Competitiveness: Evidence from
California’s State Legislative Races. Public Choice 90: 165–184
Dick, Andrew R., and John R. Lott Jr. (1993) Reconciling voters’ behavior with legislative term
limits. Journal of Public Economics 50(1): 1-14
Ferraz, Claudio, and Frederico Finan (2011). Electoral Accountability and Corruption: Evidence
from the Audits of Local Governments. American Economic Review 101 (4): 1274–1311
Fiorina, Morris (1992). Divided Government. Macmillan Publishing Company, New York
Folke, Olle, and James M. Snyder, Jr. (2011). Gubernatorial Midterm Slumps. Columbia
University Working Paper
Friedman, Daniel, and Donald Wittman (1995). Why Voters Vote for Incumbents but against
Incumbency: A Rational Choice Explanation. Journal of Public Economics 57: 67–83
Garand, James C. (2010). Income Inequality, Party Polarization, and Roll-Call Voting in the US
Senate. Journal of Politics 72 (4): 1109–1128
Garand, James C., and Marci Glaslock Lichtl (2000). Explaining Divided Government in the
United States: Testing an Intentional Model of Split-Ticket Voting. British Journal of Political
Science 30 (1): 173–191
Gelman, Andrew, and Gary King (1990). Estimating Incumbency Advantage without Bias.
American Journal of Political Science 34 (4): 1142–1164
Greene, William H. (2008). Econometric Analysis. Pearson, New Jersey
Groseclose, Timothy, Steven D. Levitt, and James M. Snyder, Jr. (1999). Comparing Interest
Group Scores Across Time and Chambers: Adjusted ADA Score for the U.S. Congress.
American Political Science Review 93: 33–50
Hanssen, F. Andrew (2004). Is There a Politically Optimal Degree of Judicial Independence?
American Economic Review 94: 712–729
Hodler, Roland, Simon Loertscher, and Dominic Rohner (2010). Inefficient Policies and
Incumbency Advantage. Journal of Public Economics 94: 761–767
Howell, William, Scott Adler, Charles Cameron, and Charles Riemann (2000). Divided
Government and the Legislative Productivity of Congress, 1945–94. Legislative Studies
Quarterly 25 (2): 285–312
Johnson, Joseph M., and W. Mark Crain (2004). Effects of Term Limits on Fiscal Performance:
Evidence from Democratic Nations. Public Choice 119: 73–90
Jones, David R. (2001). Party Polarization and Legislative Gridlock. Political Research
Quarterly 54 (1): 125–141
King, Gary, and Andrew Gelman (1991). Systemic Consequences of Incumbency Advantage in
U.S. House Elections. American Journal of Political Science 35 (1): 110–138
Krehbiel, Keith (1996). Institutional and Partisan Sources of Gridlock: A Theory of Divided and
Unified Government. Journal of Theoretical Politics 8 (1): 7–40
22
Lancaster, Tony (2000). The Incidental Parameters Problem Since 1948. Journal of
Econometrics 95: 391–414
Lee, David S., Enrico Moretti, and Matthew J. Butler (2004). Do Voters Affect or Elect Policies:
Evidence from the U.S. House. Quarterly Journal of Economics 119 (3): 807–859
List, John A., and Daniel M. Sturm (2006). How Elections Matter: Theory and Evidence from
Environmental Policy. Quarterly Journal of Economics 121 (4): 1249–1281
Mayhew, David R. (1991). Divided We Govern. Yale University Press, New Haven
McCarty, Nolan, Keith T. Poole, and Howard Rosenthal (2006). Polarized America: The Dance
of Ideology and Unequal Riches. MIT Press, Cambridge, USA McCubbins, Mathew D. (1991).
Government on Lay-Away: Federal Spending and Deficits under Divided Party Control. In: Cox,
Gary W., and Samuel Kernell (Eds.). The Politics of Divided Government. Westview Press,
Boulder
Neyman, Jerzy, and Elisabeth Scott (1948). Consistent Estimates Based on Partially Consistent
Observations. Econometrica 16: 1–32
Niemi, Richard, Thad Beyle, and Lee Sigelman (2001). U.S. Officials’ Job Approval Ratings: A
Cooperative Project of the University of Rochester, University of North Carolina at Chapel Hill,
& George Washington University. http://www.unc.edu/~beyle/jars.html
Niemi, Richard, Thad Beyle, and Lee Sigelman (2010). Codebook: U.S. Officials’ Job Approval
Ratings: A Cooperative Project of the University of Rochester, University of North Carolina at
Chapel Hill, & George Washington University. http://www.unc.edu/~beyle/jars.html
Poole, Keith T., and Howard Rosenthal (1991). Patterns of Congressional Voting. American
Journal of Political Science 35 (1): 228–278
Poole, Keith T., and Howard Rosenthal (1997). Congress: A Political-Economic History of Roll
Call Voting. Oxford University Press, New York
Poterba, James M. (1994). State Responses to Fiscal Crisis: The Effects of Budgetary Institutions
and Politics. Journal of Political Economy 102 (4): 799–821
Prior, Markus (2006). The Incumbent in the Living Room: The Rise of Television and the
Incumbency Advantage in U.S. House Elections. Journal of Politics 68 (3): 657–673
Rogers, James R. (2005). The Impact of Divided Government on Legislative Production. Public
Choice 123: 217–233
Saeki, Manabu (2009). Gridlock in the Government of the United States: Influence of Divided
Government and Veto Players. British Journal of Political Science 39: 587–607
SPPQ (2005). State Politics & Policy Quarterly. The Practical Researcher. Data Resource.
http://academic.udayton.edu/sppq-TPR/index.htm
Sundquist, James L. (1988). Needed: A Political Theory for the New Era of Coalition
Government in the United States. Political Science Quarterly 103: 614–624
Wolfers, Justin (2007). Are Voters Rational? Evidence from Gubernatorial Elections. Working
Paper, Wharton School, University of Pennsylvania
Wooldridge, Jeffrey M. (2002). Econometric Analysis of Cross Section and Panel Data. MIT
Press, Cambridge, USA
23
Appendix A
Table A1: Summary statistics (1970–2000)
Variable
Mean
Std dev.
Min.
Max.
Divided government
Lame duck
Term limit
Vote margin
Short term
Midterm Congress
# governor terms
No limit & term > 1
Electoral defeat
Clear defeat margin > 5%
Political preferences
Political heterogeneity
Democratic governor
Real per capita income
State population
Unemployment
Income growth (%)
Unemployment growth (%)
0.48
0.26
0.61
8.4
0.09
0.4
1.49
0.19
0.12
0.07
-0.01
0.29
0.56
12'046.43
4.92E+06
6.08
1.9
2.45
0.5
0.44
0.49
7.74
0.28
0.49
0.73
0.4
0.33
0.25
0.2
0.11
0.5
2'606.23
5.15E+06
2.09
3.52
22.57
0
0
0
0
0
0
1
0
0
0
-0.49
0.02
0
6'405.47
3.34E+05
2
-20.41
-52.57
1
1
1
50
1
1
5
1
1
1
0.6
0.76
1
23'223.81
3.39E+07
18
45.07
215
24
Table A2: Variable descriptions
Variable
Description
Divided government
Divided government control: 1 (divided branch or divided legislature), unified
government control: 0. Main source: Alt, Lassen and Rose (2006)
Lame duck
Governor is a lame duck: 1, 0 otherwise. Main source: List and Sturm (2006)
Term limit
State with gubernatorial term limit: 1, 0 otherwise. Main source: List and Sturm
(2006)
Vote margin
Vote margin measured as the percentage vote share of governor in vote of top two
candidates - 50. Main source: List and Sturm (2006)
Short term (2 yrs)
States with 2-year governor terms: 1, 0 otherwise (4-year terms). Main source: List
and Sturm (2006)
General election
General election year (executive and legislative): 1, 0 otherwise. Source. List and
Sturm (2006)
Midterm election
Midterm election year: 1, 0 otherwise. Source: National Conference of State
Legislators (NCSL)
Midterm congress
Midterm congress: 1, 0 otherwise. Source: National Conference of State Legislators
(NCSL)
# governor terms
Number of consecutive gubernatorial terms. Main source: Alt, Lassen and Rose
(2006)
No limit & term >1
Governor does not face a binding term limit and is not serving his first term: 1, 0
otherwise. Source: own calculation
Political preferences
DW Nominate: Measure of political preferences on a liberal–conservative scale from
roll-call votes of members of the 94th to 106th US Congress. State mean of the first
dimension of DW-Nominate score of state representatives (House and Senate) in
Federal Congress. Negative values for Democrats, positive values for Republicans.
Source: own calculation based on McCarty, Poole and Rosenthal (2006)
Political heterogeneity
DW Nominate: State standard deviation of the first dimension of DW-Nominate
scores for state representatives (House and Senate) in the Federal Congress. Source:
own calculation based on McCarty, Poole and Rosenthal (2006)
Democratic governor
Governor is a democrat: 1, 0 otherwise. Main source: Besley, Persson and Sturm
(2010)
State population
Yearly state population. Main source: List and Sturm (2006)
R.p.c. income
Real per capita personal income in 1982–1984 dollars. Main source: List and Sturm
(2006)
Unemployment
Unemployment rate (%). Source: SPPQ (2005)
Income growth (%)
Percent income growth. Source: own calculation
Unemployment growth (%) Percent unemployment growth. Source: own calculation
25
Table 1: Governor term limits in the US states
Term limits for governors by state (1970–2000)
States with no term limits:
CT, IDa, IL, IA, MAb, MN, NH, NY, ND, TX, VT, WAc, WI
States limiting governors to one term in office:
VAd
States limiting governors to two terms in office:
AL, DE, FL, LA, MD, ME, MO, NE, NJ, NV, OH, OK, OR, PA, SD, WV
State law changed from no term limit to a three-term limit:
UT (1994)
State law changed from no term limit to a two-term limit:
AZ (1992), AR (1992), CA (1990), CO (1990), KS (1974), MI (1992), MT (1992), RI (1994),
WY (1992)
State law changed from a one-term limit to a two-term limit:
GA (1976), IN(1972), KY (1992), NM (1991), MS (1986), NC (1977), SC (1980), TN (1978)
Source: List and Sturm (2006).
Note: The year in brackets is the year in which the term limit legislation changed.
a. A two-term limit was passed in 1994 but repealed in 2002 by the Idaho State Legislature.
b. Term limits were enacted in 1994 but were declared unconstitutional by the Massachusetts
Supreme Court in 1997.
c. Enacted a two-term limit in 1992, which was declared unconstitutional by the Washington
Supreme Court in 1998.
d. Restrictions on terms enacted in 1954.
26
Table 2: Baseline results
Dependent Variable: Divided Government
(1)
(2)
(3)
(4)
(5)
(6)
GLS
OLS
OLS
LOGIT
CLOGIT
CLOGIT
Lame duck
0.106**
(0.045)
0.104**
(0.046)
0.102**
(0.046)
Term limit
-0.205***
(0.076)
-0.013***
(0.003)
-0.237*
(0.121)
-0.012***
(0.003)
-0.267**
(0.119)
-0.012***
(0.003)
-0.226
(0.182)
0.010
(0.025)
0.566***
(0.195)
[0.120]***
-1.117***
(0.245)
-0.073***
(0.011)
0.573**
(0.249)
[0.121]**
-1.223**
(0.583)
-0.073***
(0.018)
0.567**
(0.249)
[0.113]**
-1.368**
(0.570)
-0.071***
(0.018)
-1.133
(0.895)
0.045
(0.120)
random effect
yes
yes
yes
yes
yes
random effect
yes
yes
yes
yes
yes
1,326
0.087
47
1,326
0.084
47
1,326
0.104
47
1,326
1,263
0.077
44
1,263
0.081
44
Vote margin
Short term
Midterm
State FE
Year FE
Observations
(pseudo) R-squared
Number of states
47
Note: Standard errors are adjusted to within-state clustering and reported in parentheses. Marginal effects are
reported in brackets. Significance level: * 0.05 < p < 0.1, ** 0.01 < p < 0.05, *** p < 0.01.
27
Table 3: Gubernatorial selection & experience
Dependent Variable: Divided Government
Lame duck
Term limit
Vote margin
# governor terms
(1)
(2)
(3)
0.117**
(0.054)
-0.275**
(0.124)
-0.012***
(0.003)
-0.018
(0.047)
0.101**
(0.047)
-0.276**
(0.130)
-0.012***
(0.003)
0.102**
(0.047)
-0.268**
(0.120)
-0.012***
(0.003)
No limit & term >1
-0.015
(0.063)
Governor age
-0.000
(0.005)
Standard controls
State FE
Year FE
yes
yes
yes
yes
yes
yes
yes
yes
yes
Observations
R-squared
Number of states
1,326
0.103
47
1,326
0.103
47
1,326
0.104
47
Note: Linear probability models estimated by OLS. Standard errors are
adjusted to within-state clustering and reported in parentheses. Standard
controls: Short term, Midterm congress. Results of the conditional logit
estimation can be found in the Online Appendix. Significance level: * 0.05
< p < 0.1, ** 0.01 < p < 0.05, *** p < 0.01.
28
Table 4: Last-period governors
Dependent Variable: Divided Government
Lame duck
Term limit
Vote margin
Electoral defeat
(1)
(2)
(3)
0.083*
(0.046)
-0.257**
(0.118)
-0.013***
(0.003)
-0.134
(0.084)
0.111**
(0.047)
-0.200
(0.133)
-0.011***
(0.003)
0.110**
(0.047)
-0.200
(0.133)
-0.011***
(0.003)
-0.035
(0.111)
-0.008
(0.110)
-0.068
(0.142)
Clear defeat (margin > 5%)
Interaction: Midterm congress x
Clear defeat margin > 5%
Standard controls
State FE
Year FE
yes
yes
yes
yes
yes
yes
yes
yes
yes
Observations
R-squared
Number of states
1,326
0.111
47
1,218
0.101
47
1,218
0.101
47
Note: Linear probability models estimated by OLS. Standard errors are adjusted to
within-state clustering and reported in parentheses. Standard controls: Short term,
Midterm congress. Results of the conditional logit estimation can be found in the
Online Appendix. Significance level: * 0.05 < p < 0.1, ** 0.01 < p < 0.05, *** p <
0.01.
29
Table 5: Closely elected governors
Dependent Variable: Divided Government
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
2nd order
polynomial
3rd order
polynomial
4th order
polynomial
< 5%
< 4%
< 3%
< 2%
< 1%
0.091*
(0.047)
-0.250**
(0.118)
-0.008**
(0.003)
0.091*
(0.046)
-0.245**
(0.119)
-0.008**
(0.003)
0.092*
(0.046)
-0.237*
(0.121)
-0.008**
(0.003)
0.181**
(0.080)
-0.178
(0.176)
-0.007
(0.005)
0.168*
(0.096)
-0.169
(0.183)
-0.005
(0.005)
0.073
(0.097)
-0.093
(0.266)
-0.005
(0.006)
Std. controls
State FE
Year FE
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
Observations
R-squared
Nbr. of states
1,326
0.104
47
1,326
0.104
47
1,326
0.105
47
725
0.073
47
587
0.078
42
419
0.066
40
312
0.120
37
141
0.220
23
Initial
vote margin:
Lame duck
Term limit
Vote margin
0.101
-0.005
(0.108) (0.099)
-0.001
-0.020
(0.342) (0.148)
-0.001 0.017***
(0.006) (0.004)
Note: Linear probability models estimated by OLS. Standard errors are adjusted to within-state clustering and
reported in parentheses. Standard controls: Short term, Midterm congress. Results of the conditional logit estimation
can be found in the Online Appendix. Significance level: * 0.05 < p < 0.1, ** 0.01 < p < 0.05, *** p < 0.01.
30
Table 6: Political preferences, party, population & economic effects
Dependent Variable: Divided Government
Lame duck
Term limit
Vote margin
Political preferences
Political heterogeneity
(1)
(2)
(3)
(4)
Political
Preferences
Party
effects
Additional
Covariates
Economic
voting
0.092*
(0.046)
-0.255**
(0.117)
-0.011***
(0.003)
0.045
(0.246)
0.567*
(0.314)
0.081*
(0.048)
-0.330**
(0.125)
-0.010***
(0.003)
0.094**
(0.044)
-0.245**
(0.113)
-0.012***
(0.003)
0.094**
(0.045)
-0.246**
(0.115)
-0.012***
(0.003)
4.87e-8***
(1.30e-8)
6.02e-5*
(3.32e-5)
-0.002
(0.022)
4.88e-8***
(1.33e-8)
6.03e-5
(3.64e-5)
-0.003
(0.024)
-0.001
(0.004)
0.000
(0.001)
Democratic governor
-0.162
(0.106)
Population
Income (r.p.c.)
Unemployment (%)
Income growth (%)
Unemployment growth (%)
Standard controls
State FE
Year FE
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
Observations
R-squared
Number of states
1,326
0.110
47
1,310
0.126
47
1,279
0.095
47
1,279
0.095
47
Note: Linear probability models estimated by OLS. Standard errors are adjusted to withinstate clustering and reported in parentheses. Standard controls: Short term, Midterm
congress. Results of the conditional logit estimation can be found in the Online Appendix.
Significance level: * 0.05 < p < 0.1, ** 0.01 < p < 0.05, *** p < 0.01.
31

Similar documents

×

Report this document