Coalition government and electoral accountability
Stephen D. Fisher
, Sara B. Hobolt
Department of Sociology, University of Oxford, Manor Road, Oxford OX1 3UQ, UK
Department of Politics and International Relations, University of Oxford, Manor Road, Oxford OX1 3UQ, UK
Single-party governments are commonly thought to be more clearly responsible for
government policy than coalition governments. One particular problem for voters evalu-
ating coalition governments is how to assess whether all parties within a coalition should
be held equally responsible for past performance. As a result, it is generally argued that
voters are less likely to hold coalition governments to account for past performance. This
article uses data from the Comparative Study of Electoral Systems project to assess
whether and how the composition of coalition governments affects the way in which
people use their votes to hold governments to account, and which parties within coalitions
are more likely to be held to account for the government’s past performance.
Ó2010 Elsevier Ltd. All rights reserved.
The ability of citizens to hold governments to account for
their past actions is one of the pivotal functions of elections.
Yet, beginning with Downs (1957), it has been argued that
the complex political context of coalition governments
provides a hindrance to accountability. Electoral account-
ability is said to exist when citizens can discern whether
governments are acting in their best interest and can punish
or reward them accordingly in elections.
incumbents who perform well remain in ofﬁce and those
who do not are forced to leave (Key, 1966). According to this
view of elections as a sanctioning mechanism, ‘account-
ability is a retrospective mechanism, in that sense that the
actions of rulers are judged ex post by the effects they have’
(Cheibub and Przeworski, 1999: 225). Yet, electoral
accountability requires citizens to make attributions of
responsibility, and voters’ability to sanction governments
may be obscured by the blurred lines of responsibility
within multi-party governments. Indeed, a number of
studies have suggested that coalition governments create
difﬁculties for the rewardepunishment model of voting (see
e.g. Anderson, 1995a,b, 2000; Dorussen and Taylor, 2001;
Lewis-Beck, 1988; Powell, 2000; Powell and Whitten,
1993; Whitten and Palmer,1999). The general expectation is
therefore that ‘voters would be likely to hold single-party
governments more responsible for policies than multi-
party coalitions’(Powell and Whitten, 1993: 401).
This paper examines the relationship between coalition
governments and electoral accountability. We address two
questions: are voters less likely to hold coalition govern-
ments to account compared with single-party govern-
ments? And do voters hold all types of coalitions and all
parties within a coalition equally to account for past
performance? While the effect of political context on
electoral accountability has received a lot of scholarly
attention in recent years, this study differs from the extant
literature in a number of important ways. First, the issue of
electoral accountability has been examined almost exclu-
sively with respect to economic voting, that is the extent to
which the electorate rewards and punishes a government
for (perceived or actual) economic upswings and down-
turns. While the economy is undoubtedly a key element of
*Corresponding author. Tel.: þ44 1865 278 830; fax: þ44 1865 278 725.
E-mail addresses: stephen.ﬁsher@sociology.ox.ac.uk (S.D. Fisher),
firstname.lastname@example.org (S.B. Hobolt).
Tel.: þ44 1865 286 173; fax: þ44 1865 286 171.
There are other forms of accountability outside the electoral arena,
such as government accountability to legislatures and media scrutiny, but
we do not discuss them here.
Contents lists available at ScienceDirect
journal homepage: www.elsevier.com/locate/electstud
0261-3794/$ esee front matter Ó2010 Elsevier Ltd. All rights reserved.
Electoral Studies 29 (2010) 358e369
the government’s remit, it is not the only way to think of
electoral accountability, and in this paper we adopt a more
general approach to the notion of ‘government perfor-
mance’, namely voter assessments of how good a job the
government has done. Second, this paper not only exam-
ines how electoral accountability varies across single-party
and multi-party governments, but also how it varies within
coalition governments. One particular problem for voters
evaluating a coalition government is to assess whether all
parties within the coalition should be held equally
responsible. Hence, we address the question of whether all
parties in the coalition are treated the same way by the
electorate. We examine different factors that may explain
why voters are more likely to hold some coalition partners
responsible for the government’s action. We ﬁnd that the
most important factor explaining differences between
coalition partners in most of the countries under investi-
gation is the head-of-government’s party effect: in general,
voters are more likely to hold the head-of-government’s
party accountable for the government’s performance
compared with other parties in the coalition. In our ﬁnal
analysis we consider whether lower levels of electoral
accountability for coalitions as a whole can be explained by
lower levels of accountability for certain parties within
coalitions, particularly those who do not hold the ofﬁce of
head of government.
This paper proceeds as follows. First, we discuss the
existing literature on the effect of political context and
coalition governments on economic voting and we present
a series of testable hypotheses. Thereafter, these theoretical
propositions are tested using data from the second module
of the Comparative Study of Electoral Systems.
2. Coalition government and electoral accountability
The standard model of how the accountability mecha-
nism operates relies on ‘retrospective voting’. In this view,
voters evaluate the performance of the government on one
or more issues, such as economic management. They vote
to retain the incumbent only when the desired standard
has been met, and governments wanting to be re-elected
will make sure to satisfy voter expectations in anticipa-
tion of electoral sanction. The early models of electoral
accountability were developed for, and relied on assump-
tions speciﬁc to, the (American) case of two-party compe-
tition with a uniﬁed executive (see e.g. Key, 1966; Fiorina,
1981). Yet, in a global context, over 70 per cent of propor-
tional representation (PR) elections produce coalition
governments (Katz, 1997: 162; see Hobolt and Karp, 2010).
Even in the US context, divided government ewhere
opposing political parties control the executive and the
legislative branches ehas been the rule rather than the
exception in the post-war period. As a consequence
a number of studies have emphasized that responsibility
for performance is seldom as transparent as in the simple
model of retrospective voting (Anderson, 1995a,b, 2000;
Dorussen and Taylor, 2001; Lewis-Beck, 1988; Powell,
2000; Powell and Whitten, 1993; Whitten and Palmer,
1999). Indeed, clarity of responsibility is frequently
obscured by power sharing in coalition governments and
multiple levels of decision-making. The general contention
is that that when voters are unsure about which parties are
responsible for economic policymaking, their ability to use
the vote to sanction politicians is compromised (Powell and
Whitten, 1993; Whitten and Palmer, 1999; Anderson, 20 00;
Powell, 2000; Nadeau et al., 2002; Duch and Stevenson,
2005). In the context of economic voting models, this
implies that voters are less likely to use the vote to sanction
for past performance and the connection between the
economic performance and vote choice is weakened.
The empirical studies of clarity of responsibility have
often relied on general indices comprising a range offactors
that can be said to compromise clarity of responsibility.
Powell and Whitten (1993) and Whitten and Palmer (1999)
consider the weak voting cohesion of the governing parties,
a participatory and inclusive committee structure in which
the roles of chair are distributed proportionately between
the parties, an upper house controlled by the opposition,
minority government, and coalition government all to be
associated with lower clarity of responsibility.
et al. (2002) further expand the measure of clarity of
responsibility to include the proportion of the dominant
party seats in government, the ideological cohesion of
governing parties, the number of signiﬁcant parties and, in
some models, the age of the government. Division of
powers, whether vertical or horizontal, also reduces clarity
of responsibility, and hence both presidentialism and
federalism make it more difﬁcult for voters to attribute
responsibility (see Arceneaux, 2006; Gélineau and
Bélanger, 2005; Powell, 2000). The effects of these
various institutional features on electoral accountability are
often considered simultaneously in these empirical studies.
As Powell and Whitten (1993: 406) note:
[M]any of the factors that contribute to lower clarity of
responsibility go together. Systems with legislative
institutional arrangements that guarantee opposition
participation in policymaking tend to be those with
proportional representation and more multiparty and
minority governments. Thus we can fairly reasonably
distinguish systems by their average clarity of respon-
sibility, not having to worry too much about the
weighting of the individual variables.
While Powell and Whitten argue that many of these
institutional factors were highly correlated, this does not
seem to be the case for our data and we cannot form
a reliable scale for clarity of responsibility.
order to analyze how exactly coalition government medi-
ates electoral accountability we certainly need to separate
out at least that factor from the standard additive index of
Royed et al. (2000) have challenged the robustness of the Powell and
Whitten’s (1993) ﬁndings claiming that there is little support for the
clarity of responsibility argument. They even claim thateconomic voting is
higher for coalition as opposed to single-party governments. Their empir-
ical analyses, however, havebeen criticized byPalmer and Whitten (2003).
For our cases the average correlation between coalition government,
bicameralopposition, weakparty cohesion, opposition controlof committee
chairs and minority-government status was 0.02. No single correlation
betweenany pairof these variableswas greaterthan 0.26.Were we toform an
additive scale from the variables inthe way that Powell and Whitten (1993)
do, the Cronbach’s alpha would be 0.4 and so we could not have much
conﬁdence thatthe scale was measuring a single underlying phenomenon.
S.D. Fisher, S.B. Hobolt / Electoral Studies 29 (2010) 358e369 359
clarity of responsibility. In addition to separating the effects
of coalition government from institutional and other
effects, our aim in this paper is also to examine how the
strength of retrospective voting varies between and within
Following the literature, our ﬁrst testable hypothesis
simply concerns the extent to which multi-party govern-
ments are held to account compared with single-party
H1: Voters are more likely to hold single-party govern-
ments to account for their performance compared with coa-
lition governments, all other things being equal.
The core of the argument is simply that it is more
difﬁcult for citizens to discern responsibility for policy
arguments when power is shared among multiple actors in
a cabinet, and this will dilute the relationship between
evaluations of government performance and vote choice.
Several scholars have found weak or mixed effects of the
rewardepunishment models in multi-party systems with
coalition governments (see, for example, Lewis-Beck, 1988;
Anderson, 1995a, 2000; Bellucci, 1991; Nannestad, 1991;
Paldam, 1991), however, few explanations have been
offered as to why the strength of retrospective voting varies
across countries and time.
One possible explanation is that the ability of voters to
hold governments to account depends on the type of
coalition government. Lewis-Beck (1988) has argued that
economic voting is strong when governments are less
complex, since complex coalitions inhibit voters’ability to
clearly assign of responsibility for policy performance.
More speciﬁcally, it has been argued that party size may
inﬂuence the level of retrospective voting, referred to as
the ‘party target size’hypothesis (Lewis-Beck, 1988;
Anderson, 2000). When the coalition is dominated by
a strong party it is easier for voters to assign credit and
blame. The dominance of a single party in a coalition can
be operationalized in different ways. One way is to
measure the proportion of cabinet seats controlled by the
head-of-government’s party (see Anderson, 2000).
Another way is to simply measure the number of parties in
the coalition. Voters are likely to ﬁnd it more difﬁcult to
assign responsibility to a government that consists of
a large number of smaller parties.
Duch and Stevenson (2005, 2008) have also argued that
economic voting is weaker when responsibility is more
widely shared across parties. But in contrast to most of the
economic voting literature they build on a selection model,
rather than the traditional sanctioning model. They argue
that shared executive power reduces economic voting not
because voters cannot attribute responsibility over parties,
but because a more equal distribution of responsibility
weakens the signal that the previous economy provides
about the competence of the incumbent parties (Duch and
Stevenson, 2005, 2008).
In other words, whereas the sanctioning model assumes
that power-sharingmakes it more difﬁcult forvoters to assign
responsibility, Duch and Stevenson’smodelassumesthat
voters have perfect knowledge of the distribution of respon-
sibility, but argue that power-sharing leads voters to attribute
more weight to exogenous factors. Empirically, however, it is
difﬁcult to disentangle whether the conditioning effects of
power-sharing on economic voting that we ﬁndare due to the
difﬁculty for voters in attributing responsibility (as the
“sanctioning approach”would argue) or due to a muted
competency signal (as the “selection approach”would argue).
As Duch and Stevenson (2008: 31) note:
In general, the implications of the model in this regard
are the same as the implications from [the clarity-of-
responsibility literature] ethat is, as responsibility is
more widely shared across parties, economic voting will
be weaker. Further, as any individual incumbent party
has a greater share of responsibility, economic voting
will be a more important inﬂuence on that party’s
Electoral accountability may also be moderated by the
clarity of available alternatives, or what Lewis-Beck (1988)
has called ‘incumbent alternatives for dissent.’A more
clearly deﬁned set of viable alternatives to the incumbent
government should enable citizens to more readily express
content or discontent with the ruling party or parties
(Anderson, 2000; Bengtsson, 2004). The importance for
economic voting of an obvious alternative to the govern-
ment in power has been tested by Anderson (2000), who
argues that a larger number of effective parties competing
for power makes it more difﬁcult for voters to estimate what
an alternative government would look like. Interestinglythe
theoretical argument hinges on the fractionalization of non-
government parties, but Anderson’s measure of the clarity
of available alternatives is the effective number of parties
calculated across all parties in the system. Hence, party
system fragmentation is far from a perfect measure of how
clearly deﬁned the viable alternatives are. Nonetheless,
a fractionalized party system may make it harder for voters
to assign responsibility and compare performances of
governments with potential alternatives, and so it is
important to control for any sensitivity in retrospective
voting to the effective number of political parties in the
Our expectations concerning the effect of the nature of
coalition government on the strength of retrospective
voting are therefore as follows:
H2a: The more parties there are in a coalition the less
retrospective voting there is likely to be.
H2b: The greater the dominance of the head-of-
government’s party in the cabinet, the more likely voters are
to hold the government to account.
Another important question is whether retrospective
voting varies not only between different types of coalition
governments but also within governments. In other words:
who is the target that voters sanction in coalition govern-
ments? Whereas most studies of economic voting have gone
no further than to establish differences between single-party
and multi-party governments, this paper also seeks to
examine whether retrospective voting is stronger for certain
types of party within coalitions. A few scholars haveaddressed
this question. Most notably, Anderson (1995a,b) who in his
studies of Denmark, Germany and the Netherlands ﬁnds that
economic voting in systems with coalition governments is
modiﬁed by a complex mix of the ideological and strategic
positions that parties occupy within the coalition as well as
within the party system. His results conﬁrm that when
S.D. Fisher, S.B. Hobolt / Electoral Studies 29 (2010) 358e369360
coalitions are considered as a whole, economic effects are
relatively small (Anderson, 1995a,b).
One compelling suggestion relating to intra-coalition
variation is that all that voters need to know which party
has the most power within the government, and they will
assign credit or blame to this party accordingly (Anderson,
2000; Duch and Stevenson, 2008). Yet, the notion of ‘who
is most powerful’can be conceptualised in different ways.
One possibility is that voterswill primarily sanction the party
of the chief executive. Not only is the head-of-government’s
(HoG) party likely to be a powerful, if not the most powerful,
player within the coalition, but in an increasingly
personality-focused media environment, the government
performance may also be closely associated with the prime-
minister’s performance. Duch and Stevenson (2008) argue
that voters are aware of the distribution of administrative
responsibility between parties and that this information
alters the weight that they assign to changes in economic
performance in their vote calculus. Their comparative study
of economic voting broadly conﬁrms their expectation that
prime-ministerial parties have a larger economic vote than
other parties andthat cabinet partners have a largernegative
economic vote than opposition parties. In the American
context, studies have differed as to whether citizens resolve
the uncertainty in situations of divided government by
ascribing responsibility to the president and absolving
Congress (Norpoth, 2001) or whether both are seenas partly
responsible as a result of a more complex attribution process
(Rudolph, 2003). In presidential systems outside the US,
coalition governments frequently occur, and in this context
voters may be more likely to focus their attention, and
sanctioning, on the president’s party. Samuels’(2004) anal-
ysis of 23 presidential democracies has, for example, shown
that executives in presidential systems are subject to rela-
tively greater punishments or rewards than legislators. In
a parliamentary context, we might equally expect that the
prime-minister’s party is ‘liable’for the government’s
performance whereas other coalition partners are absolved.
This leads us to the following expectation:
H3: The head-of-government’s party is more likely to be
held responsible for the government’s performance than other
Another related argument is that voters assign responsi-
bility differentially depending on the size of the party (see
Anderson, 2000). According to this approach the strength of
the retrospective vote depends on the party’scabinetstrength,
that is the proportion of portfolios in the cabinet (Duch and
Stevenson, 2008). The greater representation a party has in
cabinet the more inﬂuence it should have over policy and so it
should take more responsibility for decisions. If this is right
then we would expect the following hypothesis to hold.
H4: Voters are more likely to hold a party responsible for
government performance the greater the proportion of seats
the party controls in the cabinet.
This ‘cabinet-strength hypothesis’is closely related to
the ‘head-of-government’s party effect’, but whereas the
former focuses on the relative strength of parties, including
the head-of-government’s party, within the cabinet, the
latter simply distinguishes between parties that control the
chief-executive ofﬁce and those that do not.
If we ﬁnd that voters’abilities and incentives to assign
credit and blame depend primarily on the perceived
‘power’of the party in question (whether it be the chief
executive post or their strength in the cabinet), this raises
the question of whether it is appropriate to think about
electoral accountability mainly in terms of the single party
versus multi-party distinction, or whether we should
instead focus on the type of party that voters tend to punish
and reward and the type that they tend to absolve of
responsibility. Hence, in the ﬁnal analysis we examine
whether there are differences in retrospective voting
between single-party governments and coalitions whenwe
take into account the power of the individual parties.
3. Data and methodology
This paper uses survey data from the 2nd Module of the
Comparative Study of Electoral Systems (CSES, 2007). This
project involved ﬁelding a common set of questions in
different countries shortly after national elections between
2001 and 2006, often as part of the national election study.
We used only surveys for countries which received
a FreedomHouse score of 1 or 2 onpolitical rights for theyear
of the survey (Freedom House, 2007). With this restriction,
there were 37 different surveys.
While Europe and North
America are over represented there are surveys from all
continents apart from Africa.
Of these 37 surveys, 25 were
conducted after elections for which the government imme-
diately before the election was a coalition, in the sense that
cabinet portfolios wereheld by members of differentpolitical
parties. So we have a sufﬁcient number of single-member
governments to compare with coalition governments, and
He also ﬁnds the patterns of economic sanctioning are mediated by
the ideological issue priority considerations of a government, but these
ﬁndings are very speciﬁc to the economic domain (and the assumed
unemploymentdinﬂation trade-off), and it is more difﬁcult to apply
a similar rational-expectations model to the general performance evalu-
ations used in this paper (see also Hibbs, 1977).
There are other approaches to measuring ‘who is most powerful’in
a government,such as measuring how pivotal a partyis within the coalition,
using the Banzhaf power index. This index captures whether a party is
critical to forminga winning coalition in parliament by computingthe share
of coalitions that a party is a pivotal player in (Banzhaf, 1965). Yet, for this
approach to be relevantrequires a rather more sophisticated electorate,and
has therefore been used mainly to understand elite behaviour rather than
citizen vote choices. We therefore do not test this approach in this paper.
Questions were ﬁelded following just one election in each country
except for Portugal (in which the questions were asked after both the
2002 and 2005 elections) and Taiwan (for which there are CSES module 2
surveys for 2001 and 2004).
The 37 surveys include two from Germany 2002 and an additional
internet survey for Britain 2005 that was part of the fourth advance
release (CSES, 2007;Clarke et al., 2006).
The countries and elections included were Australia 2004, Belgium
2003, Brazil 2002, Bulgaria 2001, Canada 2004, Chile 2005, Czech
Republic 2002, Denmark 2001, Finland 2003, France 2002, Germany
2002, Great Britain 2005, Hungary 2002, Iceland 203, Ireland 2002, Israel
2003, Italy 2006, Japan 2004, Korea 2004, Mexico 2003, Netherlands
2002, Norway 2001, New Zealand 2002, Philippines 2004, Poland 2001,
Portugal 2002 and 2005, Romania 2004, Slovenia 2004, Spain 2004,
Sweden 2002, Switzerland 2003, Taiwan 2001 and 2004, USA 2004. Note
that Peru 2006 does meet our criteria for inclusion, but the survey was
missing the key explanatory variable for our analysis.
S.D. Fisher, S.B. Hobolt / Electoral Studies 29 (2010) 358e369 361
an acceptablenumber of cases with which to assess variation
between coalitions of different kinds.
The CSES data have the distinct advantage that they
provide equivalent survey data across a range of countries,
and thus allow us to examine how the political context
conditions individual-level behaviour. Another important
reason for our choice of data is that the CSES surveys
include questions that enable us to measure the extent to
which citizens use their vote to hold the government to
account. The basis for this measurement is the association
between voting for or against the incumbent government
and the voter’s perceptions of how good or bad a job the
government has done. If there is a strong association we
will say that retrospective voting is strong and, equiva-
lently, voters are holding the government to account.
Perceptions of government performance are gauged
using the following question:
Now thinking about the performance of the govern-
ment/president, how good or bad a job do you think the
government/president has done over the past [number
of years]. Has it/he/she done a very good job? A good
job? A bad job? A very bad job? (B3011).
To measure the strength of retrospective voting, we
simply evaluate the coefﬁcient on this question, coded on
a scale of 1 (very bad) to 4 (very good),
logistic regression models for government voting. Multi-
level analysis explicitly models differences in voting
behaviour according to national context. While correcting
for the dependence of observations within countries, it also
makes adjustments to both within and between parameter
estimates that take into account the clustered nature of the
data (Snijders and Bosker, 1999). For some parts of the
analysis we will assess the strength of retrospective voting
for different governing parties separately
and so we will
use multinomial-probit models
(Greene, 2003: 725) in
The dependent variable in the binary government-
voting models is thus whether or not someone voted for
the government in ofﬁce immediately prior to the election,
measured on the basis of their vote for the lower house.
Where voters cast more than one vote for the chamber
that determines the composition of the government, we
used the vote that is most closely correlated with the party
composition of the chamber, essentially the list or PR vote
in mixed-member systems. Non-voters have been excluded
from the analysis. Details on the independent variables can
be found in Appendix 1.
One concern with this measure of retrospective voting is
that impressions of the government’s performance may be
inﬂuenced by whether it is composed of parties the
respondent likes. So the evaluations of government are not
exogenous to party support and therefore vote choice (see
Evans and Andersen, 2004, 2006) and thus the strength of
retrospective voting, as we measure it, is potentially
contaminated by problems of endogeneity bias. While this
is a problem for the interpretation of the main effect of
evaluations on support for government and we cannot
draw clear conclusions about the overall or average level of
retrospective voting, it is still possible to test our hypoth-
eses because they involve microemacro interaction terms.
If the association between government evaluations and
vote choice appears to be conditioned by a macro variable, it
either means, as we will suggest, that the strength of retro-
spective voting is affected by the macro variable, or that
people’s partisanship affects their evaluations of government
differently according to the level of the macro variable. Using
the example of coalition government as the macro variable
and positing the direction of the effect we expect to ﬁnd, this
would be to say that either retrospective voting is weaker
when thereis a coalition, or people’s partisanship affects their
evaluations of government less when there is a coalition.
Although in the reverse direction, this latter interpretation is
still broadly in line with the theory motivating our hypoth-
eses because it suggests voters are thinking about govern-
ment performance differently when there is a coalition. If
people’s evaluations of government are less sensitive to
whether they voted for the government where there is
a coalition, then it appears that respondents see lessneed for
theirevaluations of government tocorrespond as closelywith
their vote choice as they do with single-party governments.
This kind of process is in effect an indirect acknowledgement
by voters of the main hypothesised phenomenon that thereis
less reason to base your vote choice on your evaluation of
government performance where there is a coalition. Our
results need to be read with the possibility of these
two interpretations of the interaction terms in mind, but
for clarity of exposition and reasons of space we will discuss
the ﬁndings only in terms of the direct interpretation.
Similar statistical arguments can be made for the anal-
ysis of coalition members in which we test for conditioning
of the association between evaluations and vote by party
In this section, we address each of our research ques-
tions, and test the hypotheses outlined in the theory
section. Our ﬁrst question is: Are voters less likely to hold
coalition governments to account compared with single-party
The models in Table 1 suggest the answer, in accordance
with H1, is yes. The ﬁrst model includes a d ummy variable for
whether the government was a coalition or not, the
perceptions-of-government-performance score and the
There is a similar question to B3011 which asks for evaluations of
governmental performance on the respondent’s nominated ‘most
important issue’. Responses to the two questions are so highly correlated
that it does not matter which variable is used in the analysis.
For Hungary and the Philippines the coding of vote choice was too
crude to reliably assess voting for different coalition partners.
By using a multinomial probit, rather than a multinomial logit, we
avoid having to make an assumption about the independence of irrele-
There are some exceptions. For France 2002 and Taiwan 2004 the
legislative vote was unavailable and we used the presidential vote, while
the 2004 elections in Japan were for the upper house. For the USA, a vote
for the government in 2004 was a vote in the presidential election for
Bush. In Norway, the government evaluation question was clearly about
the Bondevik led coalition government comprising the Liberals, CPP and
Centre parties which formed after the 1997 election, rather than the
Stoltenberg Labour government that formed in 2000 and served up to the
2001 election, and so a vote for the government is coded accordingly.
S.D. Fisher, S.B. Hobolt / Electoral Studies 29 (2010) 358e369362
product of these two variables (the interaction term). As
expected, people were more likely tovote for the government
if they thought it had done a good job, and this association is
signiﬁcantly weaker when the incumbent government was
a coalition. Where government was controlled by a single
party, voting for the incumbent government went from 5%
among thosewho thought the governmentdid a very bad job,
to 86% among those who thought it did a very good job. The
equivalent ﬁgures where there was a coalition are 16% and
80%. So while evaluations of past performance are strongly
linked to government voting in both cases, the association is
weaker when there is a coalition.
The second model in Table 1 includes controls for various
features of the political system and their interactions with
government performance. These institutional features are:
majoritarian electoral system, presidential system, effective
number of parties in terms of seats (ENPS), bicameral
opposition, weak party cohesion, strong legislative
committees with opposition control, and minority govern-
ment (see Appendix 1 for details). As discussed above, these
features have all been posited as factorswhich might reduce
the clarity of responsibility for government and thus the
extent to which voters are willing and able to hold the
government to account through retrospective voting.
Except for federalism, the interaction between each of these
features and evaluations of government performance are
statistically signiﬁcant. However, some of the coefﬁcients do
not have the sign that theory and previous research
suggests. For instance, there is a negative effect of having
a majoritarian electoral system and a positive effectof weak
party cohesion. These ﬁndings should be interpreted with
caution since there are strong associations between the
different institutional features. While we cannot draw
conclusions about the effects of the control variables they do
still perform their main purpose of collectively providing
a rigorous set of controls for factors other than coalition
government that might affect the level of retrospective
With this in mind, after controlling for indicators of
clarity of responsibility, we can conclude that there is
signiﬁcantly less retrospective voting on the basis of general
government performance where there was a coalition than
where there was a single-party government.
Since government strength in the legislature is possibly
also relevant to the clarity of government responsibility,
Model 3 in Table 1 is the same as Model 2, but also includes
a dummy variable for minority-government status and its
interaction with perceptions of government performance.
This interaction term is not statistically signiﬁcant, and the
interaction between government evaluations and coalition
government remains signiﬁcant. So all three models suggest
that coalition governments are less likely to be held to
Multilevel logistic regression models for voting for the incumbent government.
Model 1 Model 2 Model 3
Coeff s.e. p-value Coeff s.e. p-value Coeff s.e. p-value
Intercept 5.68 0.23 0.00 10.30 0.45 0.00 10.03 0.45 0.00
Majoritarian electoral system 1.45 0.29 0.00 1.26 0.29 0.00
Presidential or semi-presidential 2.05 0.24 0.00 1.97 0.23 0.00
Federal 0.55 0.26 0.03 0.46 0.25 0.07
Effective number of parties on seats 0.83 0.09 0.00 0.85 0.09 0.00
Bicameral opposition 1.97 0.85 0.02 2.16 0.83 0.01
Weak party cohesion 0.80 0.26 0.00 0.91 0.26 0.00
Strong committees 0.72 0.24 0.00 0.76 0.24 0.00
Minority government 0.59 0.32 0.06
Coalition government 1.82 0.28 0.00 1.38 0.29 0.00 1.25 0.28 0.00
Perceptions of government
performance (Gvt perf)
2.00 0.04 0.00 3.67 0.09 0.00 3.67 0.09 0.00
Majoritarian 0.60 0.06 0.00 0.60 0.06 0.00
Presidential or semi-presidential 0.67 0.05 0.00 0.67 0.05 0.00
Federal 0.04 0.05 0.36 0.04 0.05 0.38
Bicameral opposition 0.82 0.20 0.00 0.81 0.20 0.00
Weak party cohesion 0.26 0.05 0.00 0.26 0.06 0.00
Strong committees 0.18 0.05 0.00 0.18 0.05 0.00
Minority government 0.01 0.07 0.83
ENPS L0.34 0.02 0.00 L0.33 0.02 0.00
Coalition government L0.56 0.05 0.00 L0.36 0.06 0.00 L0.35 0.06 0.00
Deviance 44,044 43,295 43,289
N(individuals; surveys): 40,678, 37. Source: CSES module 2.
We should also note that in only one of our surveys (Czech Republic)
did the incumbent government face opposition control of a strong second
chamber and so the coefﬁcients for this variable are entirely driven by this
one case. The effects of presidential and semi-presidential systems have
been represented together by one variable. Further analysis, not shown
here, suggests that the impact of presidentialism alone is slightly weaker,
but still signiﬁcant. We also considered the possibility that the strength of
retrospective voting depends on the age of democracy (with a binary
division between pre and post 1980) but found no evidence for such an
effect. There was, however, evidence that the average age of the political
parties in a system (more precisely the largesttwo government parties and
the opposition party, data taken from Bargsted (2007) and Beck et al.
(2001)) is positively and statistically signiﬁcantly associated with greater
retrospective voting.Moreover, in a fashion analogous to the ﬁndings of the
Vowles (2010) paper in this volume, we found some evidence that the
effect of coalition government on retrospective voting is greater in older
party systems, but not in old democracies generally. Our main results
regarding coalitions are robust to controls for all these effects.
S.D. Fisher, S.B. Hobolt / Electoral Studies 29 (2010) 358e369 363
account through retrospective voting, and thus corroborate
our ﬁrst hypothesis.
Having answered the ﬁrst of our research questions, we
now consider whether there are particular types of coali-
tion for which retrospective voting is particularly weak.
One possibility is that minority coalition governments
experience less retrospective voting than those with
a majority in the legislature, but as Model 3 of Table 1
shows, this does not appear to be the case.
Our two main hypotheses (H2a and H2b) are that larger
coalitions should experience less retrospective voting due to
lower clarity of responsibility, while coalitions dominated by
one large party have higher clarity of responsibility and
should be subject to correspondingly more accountability.
Although these variables refer to coalitions, they can be
extended to single-party governments and so the analysis in
Table 2 is based on thesame data as thatfor Ta ble 1.Model1in
Table 2 suggests that larger coalitions experience less
accountability but this does not fully explain the coalition
effect. Model 2 is the same as Model 1 but also includes the
share of the cabinet seats held by the head-of-government’s
party. Strangely this coefﬁcient of the interaction term
between this variable and government evaluations has the
wrong sign according to the theory that accountability is
more likely the more that the coalition is dominated by just
If the institutional variables and their interaction
terms are removed from the model, retrospective voting does
appear to be stronger for more dominant head-of-
government parties. This suggests a multicollinearity
problem, for whichwe do not havea clear explanation, andso
we cannot draw conclusions about the effect of major party
dominance within coalitions.
Nonetheless, it is clear, in
accordance with Hypothesis (H2a) that coalitions with more
members are less likely to be held to account on the basis of
general government performance evaluations, but that this
does not fully account for lower retrospective voting for
coalitions on average.
We also control for the possibility that a more frac-
tionalized party system will lead to lower levels of retro-
spective voting, and our ﬁndings corroborate this
suggestion since the interaction term between ENPS and
government performance is negative and signiﬁcant.
A further reason why there may be less retrospective
voting for coalition governments on average is that not all
members of coalitions are held equally to account for the
government’s actions. So we now consider whether retro-
spective voting is stronger for some coalition members than
The answer to this question is broadly yes, with the
head-of-government’s party experiencing more retrospec-
tive voting than other coalition members, as we hypoth-
esised (H3). Other than this there is no clear pattern. We
Multilevel logistic regression models for voting for the incumbent government.
Model 1 Model 2
Coeff s.e. p-value Coeff s.e. p-value
Intercept 9.91 0.41 0.00 12.28 0.92 0.00
Majoritarian electoral system 1.27 0.26 0.00 1.47 0.25 0.00
Presidential or semi-presidential 2.05 0.22 0.00 2.30 0.20 0.00
Federal 0.58 0.23 0.01 0.24 0.22 0.27
Effective number of parties on seats 0.50 0.10 0.00 0.73 0.10 0.00
Bicameral opposition 2.47 0.60 0.00 3.11 0.61 0.00
Weak party cohesion 0.52 0.24 0.03 1.08 0.24 0.00
Strong committees 0.99 0.22 0.00 0.76 0.21 0.00
Minority government 0.38 0.30 0.21 0.21 0.28 0.46
Coalition government 0.33 0.30 0.27 0.85 0.28 0.00
Number of coalition members 0.61 0.11 0.00 0.55 0.11 0.00
HoG’s party share of cabinet posts 0.02 0.01 0.02
Perceptions of government performance (Gvt perf) 3.64 0.09 0.00 5.20 0.23 0.00
Majoritarian 0.59 0.06 0.00 0.58 0.06 0.00
Presidential or semi-presidential 0.70 0.05 0.00 0.80 0.05 0.00
Federal 0.00 0.05 0.97 0.07 0.05 0.19
Bicameral opposition 0.84 0.20 0.00 0.83 0.20 0.00
Weak party cohesion 0.16 0.06 0.01 0.30 0.06 0.00
Strong committees 0.22 0.05 0.00 0.15 0.05 0.00
Minority government 0.05 0.07 0.50 0.01 0.07 0.88
ENPS L0.27 0.02 0.00 L0.38 0.02 0.00
Coalition government L0.17 0.07 0.01 L0.40 0.07 0.00
# coalition members L0.11 0.02 0.00 L0.14 0.02 0.00
HoG cabinet share (
100) L1.25 0.18 0.00
Deviance 43,250 42,173
Note: For Model 1 N¼40,678, for Model 2 N¼39,741. 37 surveys. Source: CSES module 2.
This negative association holds even if the model does not include the
number of parties in cabinet and its interaction with government
Similarly, we have considered the effective number of parties in
cabinet (according to cabinet seats) as a measure of government frag-
mentation and opacity of responsibility, but this seems to increase rather
than decrease electoral accountability when added to Model 2 of Table 2.
This is a result that makes little sense to us and is probably an artefact.
S.D. Fisher, S.B. Hobolt / Electoral Studies 29 (2010) 358e369364
will ﬁrst consider country-by-country analyses of retro-
spective voting before turning to a pooled analysis to test
our main hypothesis regarding the difference between the
head-of-government’s party and other parties.
To assess whether, within a given country, there was
a difference in the strength of retrospective voting between
the government parties, we estimated a multinomial-probit
model for each country separately with perceptions of
government performance as the single explanatory vari-
able. In each regression, the dependent variable grouped all
the opposition parties into a single category, but distin-
guished between the different governing parties. So for each
governing party we have a coefﬁcient that measures the
extent to which evaluations of government performance
serve as a good predictor of voting for that party instead of
an opposition party. Within each country, we then tested
whether these coefﬁcients were the same for all coalition
members. This hypothesis could be rejected in sixteenout of
This is a clear indication that in general
coalition members are not all equally held to account.
Of the sixteen cases where there were differences
between coalition members, nine were instances of coali-
tions with just two members and so only one possible
difference between parties.
In all of these nine cases the
head-of-government’s party was more strongly held to
account on the basis of evaluations of the government’s
past performance. There are a further three cases
which there were more than two coalition members and
retrospective voting was stronger for the head-of-
government’s party than for other coalition members,
between whom there was no signiﬁcant difference in the
strength of retrospective voting, i.e. the only difference was
that between the head-of-government’s party and the
others. The remaining four cases
show a more complex
pattern, with signiﬁcant differences in the level of retro-
spective voting between coalition partners other than the
head-of-governments’party. Even here, however, we can
say that retrospective voting was strongest for the head-of-
government’s party with only one exception (Finland). So
overall, the country-by-country multinomial-probit
models show that the head-of-government’s party is
subject to more electoral accountability than other coali-
tion members in ﬁfteen of the twenty coalitions.
While the country-by-country analysis provides
compelling evidence that retrospective voting is stronger
for head-of-government parties, it also suggests that there
are no other systematic patterns in the strength of retro-
spective voting between coalition members. Part of this is
Fig. 1. Retrospective voting for coalition parties. Notes: This ﬁgure depicts the predicted probabilities of voting for the head-of-government’s party and other
government parties, each relative to voting for the opposition. The predicted probabilities are based on estimates from two multilevel logistic regression models,
restricted to coalition governments only. N¼19,798 for the model of voting for the head-of-government’s party versus opposition, N¼15,821 for the model of
voting for an other government party versus opposition. Source: CSES module 2.
The four cases where there was no statistically signiﬁcant differences
were Brazil, Chile, Japan and Slovenia. One explanation for Brazil 2002
and Chile 2005 is that elections saw incumbent presidents not standing
for re-election. There are only twenty cases in this analysis because it was
impossible to distinguish vote choice between coalition members
adequately in Hungary and the Philippines. Switzerland has also been
excluded at this stage because it has a rotating head of government, and
so cannot sensibly be compared with the other countries. Unsurprisingly,
retrospective voting was similar for all the government members except
the Swiss People’s Party, who captured popular discontent and managed
to force a change after the election in the ‘magic formula’for government
composition. Full details of country-by-country multinomial-probit
models are not presented for reasons of space. Sample sizes for each
regression varied from 501 to 1849.
The nine cases are Australia, Germany, Denmark, Ireland, Iceland,
Mexico, New Zealand, Portugal and Romania.
Belgium, France and Norway.
Finland, Israel, Italy and the Netherlands.
S.D. Fisher, S.B. Hobolt / Electoral Studies 29 (2010) 358e369 365
structural, ten of the twenty coalitions had only two
members so only one signiﬁcant difference is possible, and
since it is consistently in the direction of stronger retro-
spective voting for the head-of-government’s party, little
more can be said for these cases on the basis of country-by-
country analysis. Of the ten coalitions with more than two
members, retrospective voting appears to be equally
strong for all members in three cases, and there are
a further four cases in which only the head-of-
government’s party can be identiﬁed as having
a different rate of retrospective voting from another party.
So, even where there is a possibility of identifying patterns
other than stronger retrospective voting for the head-of-
government’s party, signiﬁcant differences were only
present in four out of the ten cases. Furthermore, there is
no consistent pattern in these four.
So, our country-by-
country analyses suggest that there is only one system-
atic pattern of variation in retrospective voting between
different coalition members, and that is stronger
accountability for head-of-government parties (H3).
To provide an additional test of the hypothesis that
retrospective voting is stronger for the head-of-government's
party than for other members of a coalition, we ran two
multilevel binary logistic regression models using data from
all the coalition governments listed in the country-by-
country analysis. Both models are random-intercept models.
The ﬁrst model is of voting for the head-of-government’s
party versus the opposition, while the second is for voting
for any governing party other than the head-of-government’s
party versus voting for the opposition. So the former analysis
excludes voters for non-prime-ministerial governing parties
and the latter omits voters for the head-of-government’s
party. The results show very clearly that the coefﬁcient of
evaluations of government performance is signiﬁcantly and
substantially larger for voting for the head-of-government’s
party than it is for voting for other government parties.
Fig. 1 presents the predicted probabilities from the models
and it is clear that theline for the head-of-government’sparty
is considerablysteeper than that forother coalition members.
This conﬁrms the impression gained from the country-by-
country analysis that retrospective voting is stronger for the
head-of-government’s party than it is for other members of
In order to test whether party size is a relevant factor for
the rate of retrospective voting it is necessary to conduct
a two-step analysis (see Jusko and Shively, 2005; Duch and
Stevenson, 2005, 2008). We took the coefﬁcient of
government evaluation for each coalition member from the
multinomial-probit regressions in the country-by-country
analysis and analysed them as a dependent variable. The
results of two linear regression models of these coefﬁcients
for all the coalition members are shown in Table 3. The ﬁrst
model shows that there is a signiﬁcant positive effect of
being a head-of-government party on the strength of
retrospective voting, in line with Duch and Stevenson’s
(2008) analysis of economic voting. In contrast, there is
no evidence that party size, as measured by the share of
cabinet seats, affects the extent to which a coalition
member is subject to electoral accountability, when we
control for head-of-government status. This accords with
the results from the country-by-country analysis and Fig. 1.
The second model in Table 3 considers the possibility that
the extent to which retrospective voting is greater for the
head-of-government’s party itself depends on the level of
clarity of responsibility, as measured by Whitten and
Palmer’s (1999) index. Since this is a two-step analysis, we
have a limited number of degrees of freedom and we
therefore include the clarity-of-responsibility index as
a composite variable to capture the institutional effects. The
signiﬁcant negative interaction term between these two
variables means that where there is greater clarity of
responsibility retrospective voting for the head-of-
government’s party is stronger compared with other
members of the same coalition. Indeed, where there is low
clarity there is little discernable difference between the
head-of-government’s party and other coalition members in
the strength of retrospective voting.
Having discovered that retrospective voting is stronger
for the head-of-government’s party than it is for other
parties in a coalition, we now ask whether this is the reason
why retrospective voting is lower for coalitions on average.
Maybe, after controlling for institutional factors affecting
clarity of responsibility, retrospective voting is equally strong
for all head-of-government parties regardless of whether the
Regression models for the strength of retrospective voting across coalition
Model 1 Model 2
coeff. s.e. p-value coeff. s.e. p-value
Intercept 0.88 0.06 0.00 0.89 0.06 0.00
% of cabinet seats 0.00 0.00 0.11 0.00 0.00 0.16
0.28 0.10 0.01 0.73 0.20 0.00
0.21 0.08 0.01
Notes: The models include ﬁxed effects for survey but they are not shown.
Standard errors are robust to heteroskedasticity. N ¼62 coalition members,
in 21 surveys. These estimates represent the second step in a two-step
multilevel analysis. See text for details of the dependent variable.
There are two cases which bear some similarity. In both Israel and the
Netherlands retrospective voting the second largest party (Labour and the
Liberals (VVD) respectively) was signiﬁcantly lower than for any other
governing party. This is perhaps because these parties could be considered
rival prime-ministerial parties. This suggests a certain knowledge and
sophistication regarding coalition arrangements which is presumably
learnt from experience of the party system. But these are just two cases and
it is unclear how to devise a suitable general test with these data of the
effect of acquired knowledgeof the party system on retrospective voting. In
Italy, retrospective votingfor the tiny NPSI was signiﬁcantly lower than for
other parties and this might be partly due to a split and restructuring of the
party just before the election. Finally, in Finland the special interest
Swedish People’s Party had the strongest retrospective voting while the
other minor party in government (Left Alliance) had signiﬁcantly weaker
retrospective voting than the two main coalition members.
Note that the results are almost exactly the same if ﬁxed, rather than
random, effects for country are used.
This ﬁnding is the same if we remove the share of cabinet seats from
S.D. Fisher, S.B. Hobolt / Electoral Studies 29 (2010) 358e369366
government is a coalition or composed of just one party. If
so, when we model voting for government as a whole,
including any coalition partners, the strength of retrospec-
tive voting would inevitably be weaker for coalitions than it
would be if we looked at just voting for the head-of-
government’s party. To test this possibility, we repeated
the analyses in Tabl e 1, but excluded voters for government
parties that were not the head-of-government’s party. This
means that the models now regard the choice between the
opposition parties. The results are shown in Tabl e 4.
All three models in Table 4 suggest that head-of-
government accountability is lower where there is coali-
tion government, and so the weaker retrospective voting for
the non-head-of-government’s parties does not account for
the lower retrospective voting for coalitions that we
observed in Table 1.
The majority of governments are coalitions. Yet, much of
the literature on democratic accountability treats govern-
ments as unitary actors. As a consequence, we still have
only a limited understanding of how voters hold
governments to account when they consist of more than
a single party. The extant literature has argued that coali-
tions represent a hindrance to electoral accountability, but
has provided few conclusive answers to whether voters
approach different types of coalitions and different parties
within coalitions in different ways. These are the questions
that this paper set out to examine.
To summarize our main results, we have found that
retrospective voting (the association between evaluations
of government performance and vote choice) is weaker for
coalition, as opposed to single-party, governments, even
after controlling for various other factors that have been
thought to affect the clarity of government responsibility
and, in turn, the strength of retrospective (economic)
voting. Among coalitions, retrospective voting declines
with the number of members, but apparently not with the
cabinet strength of the largest partner. Moreover, control-
ling for coalition size does not fully explain the lower
retrospective voting for coalitions on average.
We also assessed differences in retrospective voting
within coalitions, starting with a series of country-by-
country models, which suggested a strong pattern
whereby the head-of-government’spartywassubjectto
greater retrospective voting than other coalition members,
in line with Duch and Stevenson’s (2008) study of economic
voting. This was conﬁrmed in a pooled individual-level
analysis and in a two-step multilevel analysis. This latter
approach showed that after controlling for the head-of-
government status, retrospective voting does not system-
atically decline with cabinet strength. But the extent to
which the head-of-government is held more accountable
than other coalition members depends on other institutional
factors affecting the general level of clarity of government
responsibility, with greater differencewhere there is greater
clarity. This is a very noteworthy ﬁnding since it suggests
Multilevel logistic regression models for voting for the head-of-government’s party versus an opposition party.
Model 1 Model 2 Model 3
Coeff s.e. p-value Coeff s.e. p-value Coeff s.e. p-value
Intercept 5.68 0.21 0.00 10.01 0.43 0.00 9.93 0.43 0.00
Majoritarian electoral system 1.75 0.27 0.00 1.71 0.26 0.00
Presidential or semi-presidential 1.90 0.23 0.00 1.95 0.21 0.00
Federal 0.09 0.24 0.73 0.07 0.24 0.77
Effective number of parties on seats 0.67 0.08 0.00 0.63 0.08 0.00
Bicameral opposition 1.79 0.76 0.02 1.59 0.75 0.03
Weak party cohesion 0.33 0.26 0.21 0.23 0.26 0.37
Strong committees 1.20 0.24 0.00 1.12 0.24 0.00
Minority government 0.34 0.30 0.25
Coalition government 0.52 0.26 0.05 0.60 0.26 0.02 0.61 0.26 0.02
Perceptions of government performance 2.00 0.04 0.00 3.60 0.10 0.00 3.64 0.10 0.00
Majoritarian 0.62 0.06 0.00 0.66 0.06 0.00
Presidential or semi-presidential 0.60 0.05 0.00 0.65 0.06 0.00
Federal 0.17 0.06 0.00 0.15 0.06 0.01
ENPS 0.31 0.02 0.00 0.29 0.02 0.00
Bicameral opposition 0.72 0.20 0.00 0.59 0.20 0.00
Weak party cohesion 0.10 0.07 0.14 0.04 0.07 0.57
Strong committees 0.32 0.06 0.00 0.28 0.06 0.00
Minority government 0.26 0.08 0.00
Coalition government L0.26 0.05 0.00 L0.16 0.06 0.01 L0.18 0.06 0.01
Deviance 35,364 34,969 34,956
N¼34,503, in 37 surveys. Source: CSES module 2.
Moreover, inclusion of the number of cabinet parties or the share of
cabinet seats help by the head-of-government’s party, and their inter-
actions with government evaluations, does not reduce the interaction
term for coalition government to statistical insigniﬁcance. Comparison of
the other interaction terms between Tables 1 and 4 suggests that different
conclusions can be reached as to the relevance of some of the indicators
of clarity of responsibility depending on whether government voting or
head-of-government’s party voting is of interest. Again we urge caution
with interpretation here given the correlation between variables and note
that these terms are in the models to act collectively as rigorous controls.
S.D. Fisher, S.B. Hobolt / Electoral Studies 29 (2010) 358e369 367
that clarity of responsibility conditions the way in which
voters sanction coalitions and their leaders.
Although the head-of-government’s party is more likely
to be subject to electoral accountability than other
members of a coalition, such parties are still less likely to be
subject to retrospective voting than single-party govern-
ments. So despite identifying various sources of heteroge-
neity in the strength of retrospective voting between
coalition governments and between members of coalition
governments, our analysis does not suggest that the lower
levels of electoral accountability of coalition governments
relative to their single-party counterparts can be explained
by retrospective voting being particularly low for certain
types of coalition or certain parties within coalitions.
Taken together, these ﬁndings provide an insight into
the way voters approach the sanctioning of coalition
governments at elections. Whereas previous empirical
studies have focused almost exclusively on economic
voting, this paper has adopted a broader approach to the
concept of electoral accountability and has identiﬁed
a distinct pattern of differences in retrospective voting both
between and within coalitions; most notably highlighting
the difference between how voters approach coalition
partners with and without the chief-executive’sofﬁce.
Appendix 1 Description of variables
We use the Lijphart (1999) four point scale: 1, Uni-
cameralism; 2, Weak bicameralism with asymmetrical and
congruent chambers; 3, Medium bicameralism with
asymmetric and incongruent chambers; and 4, Strong
bicameralism with symmetric and incongruent chambers
including Australia, Germany, Switzerland and the US.
Clarity of responsibility
Following Whitten and Palmer (1999: 56), we con-
structed an additive index of four political variables
expected to mute the relationship between the economy
and incumbent electoral fortunes. Each election study was
coded as having values between zero and four, with the
maximum value representing the joint existence of
bicameral opposition, opposition control of committee
chairs, a minority government, and weak party cohesion.
Identiﬁes when cabinet portfolios are held by members
of different political parties.
Effective number of parties on seats (ENPS), as speciﬁed
by Laakso and Taagepera (1979).
The federal states in our sample are Australia, Belgium,
Brazil, Canada, Germany, Mexico, Spain, Switzerland, and
HoG’s party share of cabinet posts
Identiﬁes the proportion of seats in the cabinet
controlled by the head-of-government’s party.
Majoritarian electoral system
Coded one for majoritarian and mixed-member major-
itarian (including Mexico) and zero for proportional and
mixed-member proportional (such as Germany) electoral
Identiﬁes when parties that belong to the cabinet do not
control a majority of the seats in the legislature.
Number of coalition members
Identiﬁes the number of different parties with a cabinet
Coded one for presidential and semi-presidential
systems and zero for parliamentary systems.
Identiﬁes cases where the opposition is given propor-
tional control of politically signiﬁcant committee chairs
within the legislature (see Whitten and Palmer, 1999).
Weak party cohesion
Identiﬁes weak internal cohesion of parties, measured
in terms of cohesion in legislative voting, e.g. Japan and USA
(see Powell, 2000: 60; Whitten and Palmer, 1999).
Anderson, C.J., 2000. Economic voting and political context: a compara-
tive perspective. Electoral Studies 19 (2e3), 151e170.
Anderson, C.J., 1995a. Blaming the Government: Citizens and the
Economy in Five European Democracies. ME Sharpe, Armonk, NY.
Anderson, C.J., 1995b. The dynamics of public support for coalition
governments. Comparative Political Studies 28 (3), 350e383.
Arceneaux, K., 2006. The federal face of voting: are elected ofﬁcials held
accountable for the functions relevant to their ofﬁce? Political
Psychology 27 (5), 731e754.
Banzhaf, J.F., 1965. Weighted voting doesn’t work: a mathematical anal-
ysis. Rutgers Law Review 19, 317e343.
Bargsted, M., 2007. Complementary Macro Level Data for CSES Modules 1
and 2. http://www.umich.edu/wcses/download/contributions/
Beck, T., Clarke, G., Groff, A., Keefer, P., Walsh, P., 2001. New tools in
comparative political economy: the database of political institutions.
World Bank Economic Review 15 (1), 165e176.
Bellucci, P.,1991. Italian economic voting: a deviant case of making a case
for better theory? In: Norporth, H., Lewis-Beck, M., Lafay, J.-D. (Eds.),
Economics and Politics: The Calculus of Support. University of
Michigan Press, Ann Arbor, pp. 63e81.
Bengtsson, A., 2004. Economic voting: the effect of political context,
volatility and turnout on voters’assignment of responsibility. Euro-
pean Journal of Political Research 43, 749e767.
S.D. Fisher, S.B. Hobolt / Electoral Studies 29 (2010) 358e369368
Clarke, H., Sanders, D., Stewart, M., Whiteley, P., 2006. British Election
Study, 2005: Comparative Study of Electoral Systems: SN 5495. UK
Data Archive, Essex.
Comparative Study of Electoral Systems, June 27, 2007. CSES Module 2
Full Release [dataset]. www.cses.org.
Cheibub, J.A., Przeworski, A., 1999. Democracy, elections, and account-
ability for economic outcomes. In: Przeworski, A., Stokes, S.C.,
Manin, B. (Eds.), Democracy, Accountability, and Representation.
Cambridge University Press, Cambridge, pp. 222e250.
Dorussen, H., Taylor, M., 2001. The political context of issue-priority
voting: coalitions and economic voting in the Netherlands,
1970 e1999. Electoral Studies 20, 399e426.
Downs, A., 1957. An Economic Theory of Democracy. Harper & Row, New
Duch, R., Stevenson, R., 2008. Voting in Context: How Political and
Economic Institutions Condition the Economic Vote. Draft Book
Duch, R., Stevenson, R., 2005. Context and the economic vote: a multilevel
analysis. Political Analysis 13 (4), 387e409.
Evans, G., Andersen, R., 2006. The political conditioning of economic
perceptions. Journal of Politics 68, 194e207.
Evans, G., Andersen, R., 2004. Do issues decide? Partisan conditioning and
perceptions of party issue positions across the electoral cycle. Journal
of Elections, Public Opinion and Parties 14, 18e39.
Fiorina, M.P., 1981. Retrospective Voting in American National Elections.
Yale University Press, New Haven.
Freedom House, 2007. Freedom in the World Comparative and Historical
Gélineau, F., Bélanger, E., 2005. Electoral accountability in a federal
system: national and provincial economic voting in Canada. Publius:
The Journal of Federalism 35 (3), 407e424.
Greene, W.H., 2003. Econometric Analysis, ﬁfth ed. Prentice Hall, Upper
Saddle River, NJ.
Hibbs, D., 1977. Political parties and macroeconomic policy’. American
Political Science Review 71, 1467e1487.
Hobolt, S.B., Karp, J.A., 2010. Voters and Coalition Governments. Electoral
Studies 29 (3).
Jusko, K.L.,Shively,W.P.,2005. Applying a two-stepstrategy tothe analysis of
cross-national public opinion data. Political Analysis 13 (4), 327e344.
Katz, R.S., 1997. Democracy and Elections. Oxford University Press, New
Key, V.O., 1966. The Responsible Electorate. Vintage Books, New York.
Laakso,M., Taagepera,R., 1979. “Effective”numberof parties: a measurewith
application to West Europe. Comparative Political Studies 12, 3e27.
Lewis-Beck, M.S., 1988. Economics and Elections: The Major Western
Democracies. University of Michigan Press, Ann Arbor.
Lijphart, A., 1999. Patterns of Democracy. Government Forms and Perfor-
mance in Thirty-Six Countries. Yale University Press, New Haven.
Nadeau, R., Niemi, R.G., Yoshinaka, A., 2002. A cross-national analysis of
economic voting: taking account of the political context across time
and nations. Electoral Studies 21 (3), 403e423.
Nannestad, P., 1991. Danish Design or British Disease? Aarhus University
Norpoth, H., 2001. Divided government and economic voting. Journal of
Politics 63 (2), 414e435.
Paldam, M., 1991. How robust is the vote function? A study of seventeen
nations over four decades. In: Norpoth, H., Lewis-Beck, M., Lafay, J.-D.
(Eds.), Economics and Politics: The Calculus of Support. University of
Michigan Press, Ann Arbor, pp. 9e32.
Palmer, H.D., Whitten, G.D., 2003. Questionable analyses with no theo-
retical innovation: a response to Royed, Leyden and Borrelli. British
Journal of Political Science 33, 139e149.
Powell Jr., G.B., 2000. Elections as Instruments of Democracy: Majori-
tarian and Proportional Visions. Yale University Press, New Haven.
Powell Jr., G.B., Whitten, G.D., 1993. A cross-national analysis of economic
voting: taking account of the political context. American Journal of
Political Science 37 (2), 391e414.
Royed, T.J., Leyden, J.M., Borrelli, S.A., 2000. Is ‘Clarity of Responsibility’
important for economic voting? Revisiting Powell and Whitten’s
hypothesis. British Journal of Political Science 30, 669e698.
Rudolph, T.J., 2003. Who’s responsible for the economy? the formation
and consequences of responsibility attributions. American Journal of
Political Science 47 (4), 698e713.
Samuels, D., 2004. Presidentialism and accountability for the economy in
comparative perspective. American Political Science Review 98 (3),
Snijders, T.A.B., Bosker, R.J., 1999. Multilevel Analysis. An Introduction to
Basic and Advanced Multilevel Modeling. Sage, London.
Vowles,J., 2010.Making a Difference? PublicPerceptions of Coalition, Single-
Party, and Minority Governments. Electoral Studies 29 (3), 370e380.
Whitten, G.D., Palmer, H.D., 1999. Cross-national analyses of economic
voting. Electoral Studies 18 (1), 49e67.
S.D. Fisher, S.B. Hobolt / Electoral Studies 29 (2010) 358e369 369