Theories of coalition formation: An empirical test using data from

European Journal of Political Research 46: 721–745, 2007
721
doi: 10.1111/j.1475-6765.2007.00709.x
Theories of coalition formation: An empirical test using data from
Danish local government
ASBJØRN SKJÆVELAND, SØREN SERRITZLEW &
JENS BLOM-HANSEN
Department of Political Science, University of Aarhus, Denmark
Abstract. Theories of coalition formation represent a diverse set of arguments about why
some government coalitions form while others do not. In this article, the authors present a
systematic empirical test of the relative importance of the various arguments. The test is
designed to avoid a circularity problem present in many coalition studies – namely that the
theories are tested on data of national government coalitions in postwar Europe: the very
data that gave rise to the theories in the first place. Instead, the authors focus on government
coalitions at the municipal level. They base their analysis on an expert survey of almost 3,000
local councillors from all municipalities in Denmark. They use conditional logit analysis to
model government formation as a discrete choice between all potential governments. The
analysis confirms some, but far from all, traditional explanations such as those based on
office and policy motives. At the same time, the analysis raises the question of whether actors
really seek minimal coalitions.
Introduction
Coalition formation is at the heart of politics in multi-party systems. This is
reflected in the overwhelming number of theories developed to account for
coalition formation (for overviews, see Lijphart 1999: Chapter 6; Laver 1998).
Empirical studies have followed a variety of strategies to investigate the
increasingly complicated theories: in-depth case studies of prominent
examples of coalitions (e.g., Damgaard 1969; Skjæveland 2003), detailed
accounts of coalition bargaining across a number of countries (e.g., Müller &
Strøm 2000), bivariate comparisons of large samples of cabinets (e.g., De
Swaan 1973) and multivariate statistical analysis of datasets of coalition formations across space and time (e.g., Martin & Stevenson 2001). Despite this
diversity of approaches and designs, these studies have an important thing in
common: they are all based on data on national government formation primarily in postwar Western Europe.1 This means that coalition theories have been
tested on basically the same data that gave rise to the theories in the first place.
Laver (1989: 16–17) describes this problem in the following way:
© 2007 The Author(s)
Journal compilation © 2007 (European Consortium for Political Research)
Published by Blackwell Publishing Ltd., 9600 Garsington Road, Oxford, OX4 2DQ, UK and 350 Main Street, Malden,
MA 02148, USA
722
asbjørn skjæveland, søren serritzlew & jens blom-hansen
[T]he main data set to which they [coalition theories] address themselves
– the universe of national governments in postwar Europe – is by now
one of the most thoroughly picked over in the entire social sciences. As a
consequence, the relationship between theory and data has become
extremely incestuous. It is simply no longer possible, for example, to
construct a general theory from a priori assumptions and then run off to
‘test’ it against the data, since the general properties of this data set are by
now very well known.
This circularity problem makes a new empirical testing ground desirable.
As suggested by Laver & Schofield (1990: 8; see also Laver 1989; Laver et al.
1987), local governments offer a promising new laboratory to test coalition
theories. However, this research strategy has only been used in a limited
number of studies. Although promising in many respects, these studies have
been suggestive rather than conclusive. This is either due to a lack of data on
the policy positions of local parties or to limited datasets that comprise a
non-random subset of a given country’s local governments. Denters (1985) and
Steunenberg (1992) analyze local coalitions in the Netherlands. Yet due to a
lack of data on local policy positions they need, first, to make the ‘heroic’
assumption that local parties do not differ from the policy positions of their
national parties and, second, to remove municipalities with large non-national
parties from their analyses. Laver et al. (1998) analyze coalitions in local councils in the United Kingdom in which no party commands an overall majority.
However, like the Dutch studies, they have no data on local policy positions,
which creates serious difficulties in relation to independent (i.e., non-party)
councillors. Bäck (2003) analyzes local coalitions in Sweden. She presents the
obvious solution to the problem of identifying local policy positions – namely
to investigate them empirically. This is done through a survey of local councillors. However, the survey only comprises 49 (out of Sweden’s 290) municipalities, all located in one region. A problem of representativeness is thus
involved.2
The purpose of this article is to bring the local government laboratory for
the study of coalition theories one step further. We investigate coalition theories within a Danish municipal setting. Local policy positions are measured by
an expert survey of local councillors in all Danish municipalities. To our knowledge, this is the first local government test of coalition theories that is based on
sound data on the policy positions of local parties as well as including all local
governments within a country. The article is structured in the following way.
The next section presents a review of the broad range of coalition theories. We
draw from them a set of hypotheses to be investigated empirically. Following
this review, we discuss a number of conceptual and methodological issues in
© 2007 The Author(s)
Journal compilation © 2007 (European Consortium for Political Research)
theories of coalition formation
723
relation to the choice of local governments as the empirical testing ground. We
then present our data and methods before moving on to the empirical analysis.
The analysis supports the proposition that actors seek office benefits and
policy influence when forming coalitions, but it questions whether actors necessarily seek minimal coalitions. Our study suggests that more attention should
be paid to oversized coalitions. Institutional characteristics are also found to be
important for coalition formation, not least the hitherto relatively unexplored
concept of a coalition’s office capacity. In the conclusion we discuss the potential of local governments as the future empirical testing ground for coalition
theories and the implications of our findings for future studies of coalition
theories.
Coalition formation theory
Our theoretical starting point is predictive coalition theory broadly conceived.
Coalition theory may be divided into at least two categories: those theories
that somehow take policy considerations into account and those that ignore
policy considerations and arguably, by implication, assume that parties care
only about obtaining government offices.3 Furthermore, while older coalition
theories are generally considered institution-free, more recent ones include
the effect of institutions.4 In the following, we discuss the classic theories as
well as a number of more recent ones. Office-oriented theories will be presented first, followed by theories that consider policy positions.5 Finally, two
institutional arguments are presented.
Office-oriented theories are generally considered to be the oldest. As noted
by Martin and Stevenson (2001: 34), there is no reason for parties that care
only about office to accept minority governments, and therefore a basic
hypothesis is that only majority cabinets will be formed. Von Neumann &
Morgenstern (1980 [1944]) suggested a refinement: that only minimal winning
coalitions will be formed. Parties unnecessary for the majority would simply
reduce the amount of spoils available for the other parties. Generally there will
be several minimal winning coalitions and various suggestions have been made
that reduce the number of possible coalitions in a particular instance of government formation. If the spoils of office are distributed in proportion to the
seats controlled by government parties in parliament, it makes sense to
hypothesize that the minimal winning coalition with the smallest legislative
weight (which is possible in a particular instance of government formation)
will be preferred. This was suggested by Gamson (1961) and also by Riker
(1962) according to a common interpretation (e.g., Laver & Schofield 1990).
Laver & Schofield (1990; cf. Riker 1962) call these coalitions ‘minimum
© 2007 The Author(s)
Journal compilation © 2007 (European Consortium for Political Research)
724
asbjørn skjæveland, søren serritzlew & jens blom-hansen
winning coalitions’.6 Instead, Leiserson restricted the predicted minimal coalitions to those with the smallest number of parties. The underlying assumption
is that a minimal winning coalition of, for example, three parties will reach an
agreement more easily than one of four parties. This model is therefore
referred to as the ‘bargaining proposition’ (Laver & Schofield 1990: 95). All
these models hypothesize that certain coalitions are more likely to be formed.
A more recent concept also ignores policy considerations, but points out a
particular party that is especially likely to be part of a government. This is the
‘dominant player’. The dominant player is included in a least one winning
coalition, which it can leave to form another winning coalition with new
parties, which the old partners cannot make winning. The dominant player
must be the largest party, but the largest party is not necessarily a dominant
player, in which case there is no dominant player (Van Roozendaal 1990).
From the coalition formation theories that ignore policy considerations, we
derive the following hypotheses:
H1: Potential governments are more likely to form if they control a
majority of the seats in the municipal council.
H2: Potential governments are more likely to form if they are minimal
winning coalitions.
H3: Potential governments are more likely to form if they are minimum
winning coalitions.
H4: Potential governments are more likely to form if they are minimal
winning coalitions with the smallest possible number of parties.
H5: Potential governments are more likely to form if they contain a
dominant player.
Turning to policy-oriented theories, Axelrod’s (1970) minimal connected
winning coalition theory is the most famous. It assumes that parties can be
placed on a left-right scale and that only parties adjacent to each other form
coalitions. Thus, coalitions will be connected as well as winning. They will not
include parties that are not necessary to meet these two demands, but they
would be minimal only in this sense, not necessarily in the sense mentioned
above. Inspired by De Swaan’s main text (but not the pertinent footnotes)
many authors have taken Axelrod one step further and hypothesized that the
minimal connected winning coalition with the smallest ideological range as
measured on an interval scale may be privileged (e.g., Laver & Schofield 1990:
111; cf. De Swaan 1973: 68–79).
As in the case of the office-oriented theories, it is also possible to point to
a specific party that should be particularly likely to participate in government
© 2007 The Author(s)
Journal compilation © 2007 (European Consortium for Political Research)
theories of coalition formation
725
when policy considerations are taken into account. Actually this line of thinking originates even earlier than the concept of minimal winning coalitions. The
median voter theorem was proposed by Galton (1907) and later elaborated by
Black (1958) and, among others, Laver & Schofield (1990). A median party
government should be impossible to bring down because no majority will find
another government more preferable. However, the size of the median party
may matter in several ways. If the median party is more likely to participate in
government only if it is the largest party, this could indicate that two policy
dimensions are important. Thus, the median party is more likely to occupy the
median position on a second dimension as well if it is the largest party (cf.
Laver & Shepsle 1996: 110–111). Furthermore, to be a structurally stable core
party in two dimensions, a party must not only be centrally located but also be
the largest party (Schofield 1993: 5–13; Laver & Benoit 2003: 221; Laver &
Shepsle 2000: 121–122).7 Furthermore, for national government formation,
Laver & Shepsle (1996, 2000) and Skjæveland (2003, 2004) have suggested that
the median party must be of a certain size to be able to cope with the practical
tasks of government – to have ‘office capacity’ (cf. Lees 2000). This might also
be the case in relation to municipal government formation; for instance, the
smaller the party, the greater the risk that it does not have a qualified candidate
for mayor. We set the minimum level for the variable Office Capacity in
Danish municipalities to four seats, a little less than the average number of
standing committees in the municipalities (which is 3.4, cf. Table 1 below) plus
the mayoralty.
The ‘office capacity’ concept may be seen as introducing an institutional
limitation. Some recent coalition formation theories focus on the formation
process and the institutions related to this, particularly the formateur.
However, these theories are difficult to test on the Danish municipalities
because the municipal government formation process is relatively uninstitutionalized with no formal or semi-formal formateur. However, another institution – the municipal electoral system – gives rise to a hypothesis. The
Danish municipal electoral system allows and rewards pre-electoral coalitions (it is based on the d’Hondt method), and as noted by Laver &
Schofield (1990: 206), such pre-electoral coalitions may also tend to enter
government together. The final institutional hypothesis to be put forward
here concerns an often-documented incumbency effect. Lupia & Strøm have
developed an interesting possible explanation of this involving transaction
costs: incumbent coalitions have developed procedures of cooperation and
communication as well as knowledge and trust of each other (Bäck 2003:
448).
From the coalition theories that include policy considerations and the
theories that include institutions we derive the following hypotheses.
© 2007 The Author(s)
Journal compilation © 2007 (European Consortium for Political Research)
726
asbjørn skjæveland, søren serritzlew & jens blom-hansen
H6: Potential governments are more likely to form if they are minimal
connected winning coalitions.
H7: Potential governments are more likely to form if they are minimal
connected winning coalitions with the smallest possible ideological range.
H8: Potential governments are more likely to form if they contain the
median party.
H9: Potential governments are more likely to form if they contain a party
that is both the largest and the median party.
H10: Potential governments are more likely to form if they contain a
median party that controls at least four seats in the municipal council.
H11: Potential governments are more likely to form if they are based on
pre-electoral coalitions.
At the operational level we apply two variants of H11:
H11A: Potential governments are more likely to form if they contain all
parties from one or more pre-electoral coalitions than if they contain only
some of the parties from one or more pre-electoral coalitions. In other
words, the coalition that forms contains no broken pre-electoral coalition
(but it may contain parties that were not part of an electoral coalition).
H11B: Potential governments are more likely to form if they contain all
parties from exactly one pre-electoral coalition and no other parties.
H12: Potential governments are more likely to form if they are incumbent
administrations.
Studying government coalitions at the local level
The attraction of investigating coalition theories at the local level is threefold.
First, it provides a new empirical testing ground and allows us to avoid the
circularity problem discussed in the introduction. Second, it increases the
number of cases for analysis. Most local government systems comprise several
hundred units while the number of comparable nations is rather limited. Third,
a study of local coalitions represents an attractive research design. A general
finding within coalition studies is that the performance of coalition theories
tends to be country-specific (Laver 1989: 29). This should be no surprise since
the institutional setting of coalition formation varies considerably across
nations. Focusing on the local level within one country means that the important variables of polity and time can be held constant while still retaining a
sufficient number of cases to conduct a sound analysis.
© 2007 The Author(s)
Journal compilation © 2007 (European Consortium for Political Research)
theories of coalition formation
727
However, the focus on the local level also raises conceptual and methodological questions that must be handled carefully. Coalition theories have not
been developed to explain local coalitions, but the formation of coalitions after
national elections in parliamentary democracies. Yet there is no compelling
reason why coalition theories cannot be applied at the local level. Local parties
fight elections on the basis of policy promises to voters and enact policies when
they take office. The enacted policies are important to local citizens and create
incentives to punish local politicians who renege on election promises. Furthermore, the local executive system typically involves political positions that
are both influential and financially rewarding. This means that both policy- and
office-oriented coalition theories are potentially relevant at the local level
(Laver et al. 1987; Laver 1989; Mellors 1989). In short, coalition theories should
work in a municipal context. If they do not, this would be a serious challenge
for the theories.
Coalition theories must be adapted to the specific local government
system in question before a meaningful analysis can be conducted. Four sets
of problems are relevant to consider. First, are both office and policy payoffs
relevant at the local level? Local governments may be so tightly regulated by
the central government that policy considerations play a minor role. For
example, Steunenberg (1992: 247) argues that office considerations dominate
in local coalition formations in the Netherlands for this very reason. In the
Danish case, local governments represent a case of ‘legal localism’ (Page
1991: 13–42) – that is a system in which local governments enjoy power over
wide areas of public policy. Danish local governments are the implementing
agencies of the welfare state. They are responsible for about 40 per cent of
all public expenditure and have considerable autonomy in the welfare areas
of child care, primary schooling and old age care (Blom-Hansen 1999). In the
Danish case, both office and policy considerations are clearly of potential
relevance for local coalition partners. In short, coalition theories should work
in a municipal context. If they do not, this would be a serious challenge for
the theories.
Second, the policy positions of local parties are not obvious. Local parties
are often branches of national parties, but their policy positions may differ.
Inferring local policy positions from national level data is tempting, but may
result in errors since local parties often enjoy some autonomy from their
national organization. To complicate matters, most local government systems
have local lists or independent councillors that are not organized on a national
basis. In the Danish case, such local lists occupy 10 per cent of all seats in local
councils, and in several councils they control an absolute majority. In the
following analysis we measure local policy positions directly by an expert
survey sent to almost 3,000 local councillors. The survey shows that the local
© 2007 The Author(s)
Journal compilation © 2007 (European Consortium for Political Research)
728
asbjørn skjæveland, søren serritzlew & jens blom-hansen
and the national left-right ranking of parties are identical in only 37 per cent of
the Danish municipalities (cf. Table 1).
Third, local coalitions may be formed in a national bargaining context, and
national attitudes to coalitions may influence local coalition politics. Colomer
& Martinez (1995) report an extreme example. After the local elections in
Spain in 1979, the national leaders of the Socialist and Communist parties
signed a general agreement to elect mayors together and form joint local
governments. If local parties have no autonomy to engage in coalition negotiations, they do not, of course, constitute a new empirical testing ground.
Although most Danish municipalities are governed by local branches of
national parties, such a pact has never been made at the national level. Municipal coalition formation is not centrally controlled in Denmark, and there are
many examples of local coalitions that would be unthinkable at the national
level.
Finally, the concept of a government coalition is relatively unambiguous at
the national level, but less clear-cut at the local level.8 There are local functional equivalents to a national government, but they depend on the specific
local government system under study. In alderman systems, the local council
forms an executive committee to which administrative authority is delegated.
This is a close functional equivalent to a national government and, for
example, has been used as such in studies of local coalition formation in the
Netherlands (Denters 1985; Steunenberg 1992). In committee systems, executive authority rests with the local councils’ standing committees. In these
systems, the committee chairmen have been used as the functional equivalents
of national governments in local coalition studies, for example in Sweden
(Bäck 2003) and the United Kingdom (Laver et al. 1998).
The Danish case is a committee system, but with a division of executive
powers between the mayor and the standing sectoral committees. The mayor is
elected by and among the council members on a simple majority basis. A
positive vote of investiture is required. The mayor is the formal head of the
municipal administration and chairs the council’s financial committee. The
members of the standing committees are also elected by and among the
council members, but on a proportional basis. The committees are responsible
for the day-to-day administration within the various policy sectors. The chairmen of the committees are elected by the committees on a simple majority
basis. The chairmen have few formal powers apart from organizing the meetings of the committees and deciding urgent matters, but in practice considerable influence is often delegated to them and besides the mayoralty they are
the only political positions with considerable remuneration (Le Maire & Preisler 2000). The positions as committee chairmen are normally decided after
intense negotiations on the night after the election (Pedersen 1997). Conse© 2007 The Author(s)
Journal compilation © 2007 (European Consortium for Political Research)
theories of coalition formation
729
quently, the mayor and the chairmen of the sectoral standing committees may
be considered the functional equivalent of a national government in the
Danish case.
In Table 1, we have summarized some relevant background information
about Danish municipalities in the election period 2001–2005. Block 1 shows
that Denmark has a proportional local election system with fixed 4-year terms.
Block 2 shows that Denmark has a multi-party system at the local level,
although this is not the case in all individual municipalities. A single party
commands a majority of the seats in the local council in about a third of the
municipalities. This row also documents the relation between local policy
positions and national-level politics. It shows that local and national policy
positions are not identical. This is the case in only 37 per cent of the municipalities. Finally, it shows the average policy positions of the local parties. Block
3 contains information about the three political institutions in Danish municipalities: the mayor, the local council and the standing committees. Block 4
shows that Danish municipalities vary considerably in size. Block 5 shows that
local welfare (social security and primary education) is the most important
municipal function. Finally, block 6 shows the number of parties in local
coalitions and their seat share.
Data and methods
The policy positions of Danish municipal parties are measured by a survey.The
local councillors are the best source of information about the relative positions
of the parties in their own council. They were asked to place all parties and
local lists in their municipality on an 11-point left-right scale, and hence used
not only as experts on their own party, but on all parties in their municipality.
We first conducted a pilot test of the questionnaire by sending it to 31 members
of the city council in Aarhus, which was one of four municipalities to be
excluded from the survey.9 After making the appropriate changes to the
wording of some questions, another pilot test with 24 respondents from a
random sample of municipalities was run with satisfying results. Finally, the
survey was sent in May 2004 to ten randomly selected local councillors in each
of the ordinary Danish municipalities. After eight days (and with a response
rate of 40 per cent), the nonrespondents were mailed a new questionnaire, and
12 days later (with a 59 per cent response rate), questionnaires were mailed for
the last time. The data collection was terminated after another 12 days. Of the
population of 4,459 local councillors, 2,669 were selected as respondents and
1,787 replied. The response rate was thus 67 per cent. The party shares in the
© 2007 The Author(s)
Journal compilation © 2007 (European Consortium for Political Research)
730
asbjørn skjæveland, søren serritzlew & jens blom-hansen
Table 1. Background information on Danish municipalities (2001–2005 election period)
Value
1. Local election system:
Election system: Proportional
–
Election district: The municipality
–
Election period: Four years (fixed term)
–
Election of mayor: By and among the local council (simple
majority)
–
Election of standing committee chairmen: By and among
the committees (simple majority)
–
2. Local party system:
Number of political parties and local lists in the individual municipalities:
– Average number
4.9
– Minimum number
2
– Maximum number
9
Percentage of municipalities in which a single party controls
a majority of seats in the local council
34.1
Percentage of municipalities in which the ranking of local
political parties on the left-right scale is identical to the
ranking of the national political parties*
36.7
Average position of parties on left-right scale (0 = extreme left; 10 = extreme right):
– Socialist People’s Party
2.8
– Social Democrats
4.1
– Social Liberal Party
4.8
– Centre Democrats
5.0
– Unity list and local parties
5.4
– Christian People’s Party
5.6
– Conservatives
7.0
– Liberal Party
7.2
– Danish People’s Party
7.7
– Progress Party
7.9
3. Local political institutions:
Mayor: Administrative head of municipality, chairman of financial committee:
– Percentage social democratic
30
– Percentage liberal
50
– Percentage other
20
Size of local council (overall responsible political body)
– Average number of members
17
– Minimum number of members
9
– Maximum number of members
31
© 2007 The Author(s)
Journal compilation © 2007 (European Consortium for Political Research)
731
theories of coalition formation
Table 1. Continued.
Value
Standing committees in the individual municipalities (responsible for day-to-day
administration)
– Average number of standing committees
3.4
– Minimum number of standing committees
2
– Maximum number of standing committees
10
4. Size of municipalities:
Average number of inhabitants
17,600
Minimum number of inhabitants
2,300
Maximum number of inhabitants
183,700
5. Functions of municipalities (budget share, 2003):
Urban renewal and environmental protection
2
Public utilities
6
Traffic and infrastructure
3
Education and culture
20
Social security
59
Administration
10
6. Local government coalitions:
Number of parties in the local coalition:
– Average number
2.4
– Minimum number
1
– Maximum number
5
Share of seats controlled by the coalition:
– Average share
0.71
– Minimum share
0.18
– Maximum share
1.00
Notes: The table comprises the 267 Danish municipalities. Four municipalities have been
excluded (see Note 9). The figures for party positions are due to missing cases in the survey
based on 264 municipalities. * Includes only local parties which are branches of national
parties. The ranking of national parties is taken from the recent international expert survey
by Benoit & Laver. The ranking of local parties is from the authors’ own expert survey
among local councillors (Blom-Hansen et al. 2004).
sample and in the answers match the population closely. This suggests that the
bias from nonresponse is insignificant.
A special version of the questionnaire was prepared for each municipality
such that only parties represented in each particular city council were listed.
Each councillor was asked to indicate on a scale from 0 to 10, where 0 was
marked ‘Left’ and 10 ‘Right’, where they ‘would generally place the parties in
© 2007 The Author(s)
Journal compilation © 2007 (European Consortium for Political Research)
732
asbjørn skjæveland, søren serritzlew & jens blom-hansen
the city council on the left/right-scale’. Each local party’s left-right score is
calculated as the average score given by the individual respondents. Table 1
contains the average positions across all municipalities. While the parties in
only 37 per cent of the municipalities are ranked similarly to the national
parties, these differences tend to cancel out so that the ranking of the overall
average of party positions corresponds closely to the ranking of the national
parties.
The returned questionnaires cover all municipalities, but the number of
responses for the individual municipalities varies from one to ten. In one
municipality, less than three councillors returned the questionnaire. In two
other municipalities, data are, due to problems with the questionnaire, missing
for the position of one party. In these municipalities we are unable to estimate
policy positions reliably, and we therefore excluded them and were left with
264 municipalities for the analyses. Data for the remaining variables in the 12
hypotheses listed above are available from the Danish Statistical Bureau’s
Internet databases (www.dst.dk) and Kommunal Aarbog, an annual municipal
handbook containing information on local councillors, parties, pre-electoral
coalitions and so on.
Whenever a coalition is formed, it is chosen among several other possibilities. For n parties, 2n – 1 potential coalitions exist. Many coalitions will never be
formed in practice. This is the case for coalitions that consist of only extreme
parties or control only a small share of the seats in the city council. According
to the hypotheses, we expect certain coalitions to be very likely. In order to test
whether these coalitions are in fact more likely to be formed, we need information about the coalition formed as well as on the potential coalitions. The
264 coalitions that were actually formed in the Danish municipalities in 2001
were picked from 11,036 potential coalitions. Table 2 describes these potential
coalitions and provides descriptive statistics. One hypothesis produces a
unique prediction, others imply that only a few coalitions can form, while most
hypotheses suggest that the coalition is likely to belong to a certain – sometimes quite large – group.
In the next section we test the hypotheses by analyzing which of the
variables in Table 2 best predict which of the potential coalitions are formed.
A coalition opportunity represents one case. In each of the cases, exactly one
coalition is chosen, but the number of potential coalitions varies by the number
of parties in the municipality. In a municipality with only two parties (the
observed minimum) a coalition is chosen from among only 22 – 1 = 3 potential
coalitions, while 29 – 1 = 511 potential coalitions exist in a municipality with
nine parties (the observed maximum). The estimation approach must take this
into account. The conditional logit model is designed to estimate how characteristics of alternatives affect the likelihood of one alternative being chosen. It
© 2007 The Author(s)
Journal compilation © 2007 (European Consortium for Political Research)
733
theories of coalition formation
Table 2. Characteristics of potential coalitions
Frequency
Share (percentage)
H1: Winning coalition (not necessarily minimal)
5,650
51.2
H2: Minimal winning coalition
1,193
10.8
H3: Minimum winning coalition (smallest number
of seats)
898
8.1
H4: Minimal winning coalition with smallest
number of parties
572
5.2
H5: Coalition including a dominant party
4,882
44.2
H6: Minimal connected winning coalition
496
4.5
H7: Minimal connected winning coalition with
smallest ideological range
271
2.5
H8: Coalition including the median party
5,950
53.9
H9: Coalition including the median party which is
also the largest
2,938
26.6
H10: Coalition including a median party which
controls at least four seats (office capacity)
3.322
30.1
H11A: Coalition not breaking pre-electoral
coalitions
3,484
31.6
H11B: Coalition identical to an electoral coalition
309
2.8
H12: Coalition identical to incumbent
administration
264
2.4
Connected coalition (not necessarily minimal)
4,132
37.4
All potential coalitions
11,036
100
allows the dependent and the independent variables to be dichotomous (Long
& Freese 2003: 235–245). Hence, this model, which has been used by Martin &
Stevenson (2001) and Bäck (2003) to test similar models, meets the requirements of the present problem.
Empirical analysis
In this section we test the hypotheses and scrutinize the results. We estimate
how the characteristics of the various potential coalitions specified by our
hypotheses affect the likelihood of the choice of the local coalition. Some of
the coalitions specified by our hypotheses are variants or subsets of each other.
For example, minimal winning coalitions with the smallest number of parties
are a subset of minimal winning coalitions. We have a total of 16 variables
to include in the analyses. Since some of them are highly correlated,
© 2007 The Author(s)
Journal compilation © 2007 (European Consortium for Political Research)
734
asbjørn skjæveland, søren serritzlew & jens blom-hansen
multicollinearity is a serious issue as in other quantitative studies of coalition
theories. We address this problem by using a series of models in which the
highly correlated variables are added and excluded in turn. In the first model,
the classic office- and policy-oriented theories are tested. In the second model
we include other versions of some of the variables. Instead of a variable
indicating whether a potential coalition is minimal winning, we use a variable
indicating whether it is minimum winning, and replace a variable indicating
whether the coalition is minimal connected winning with a variable indicating
whether it is a minimal connected winning coalition with smallest possible
ideological range. The problem is that we cannot use all variables simultaneously, and our strategy is therefore to include them in separate models and
then assess which model provides the best explanation of coalition formation.
In the third model the universe of minimal coalitions is left for the benefit of
variables indicating how many parties and seats are controlled by the coalitions, and in the fourth model we also include variables for institutional factors.
To gauge the relative effect of these variables, we compare their statistical
significance. In the models we report two measures of the effect of the variables. First, conditional logit coefficients indicate whether a given coalition
characteristic increases or decreases the likelihood of the potential coalition to
form. Since the conditional logit model is non-linear, the coefficients do not
indicate the substantive magnitude of the effects, only their direction and
statistical significance. For instance, a positive effect of a given coalition characteristic means that if a coalition has more of that characteristic, it increases
the likelihood that it is formed. Second, odds ratios are reported to indicate the
substantive effect of the coefficients. They describe the effect of a unit change
in a given coalition characteristic on the odds of this potential coalition being
formed. An odds ratio greater than 1 indicates an increased chance of a
potential coalition being chosen, while an odds ratio smaller than 1 indicates
the opposite. For example, an odds ratio of 1.5 for a given coalition characteristic indicates that it is 1.5 times more likely to be formed than one without this
characteristic, while an odds ratio of, say, 0.95 indicates that having this coalition characteristic reduces the coalition’s chances of being formed by 5 per
cent compared to one without the given characteristic.
All the reported models have been examined for interaction effects from
several factors that can potentially influence the coalition formation game.
Coalition formation may function differently in local government because of
special local conditions. First, one could argue that politics functions in a more
personal and less party ideological way in small communities, whereas local
politics in large municipalities is likely to resemble national politics. Danish
municipalities vary considerably in size. The smallest has about 2,000 inhabitants, and the largest about 300,000.The median size is about 10,000 inhabitants.
© 2007 The Author(s)
Journal compilation © 2007 (European Consortium for Political Research)
theories of coalition formation
735
We have examined whether our models work differently if we exclude small
municipalities. With very few exceptions this is not the case.10
Second, patterns of cooperation, tradition and confidence among certain
parties may affect coalition formation in the municipalities. In some councils,
groups of parties may have a tradition of cooperation, and confidence between
certain parties and individual politicians may also play an important role. We
have examined whether the hypotheses fare differently in municipalities
where the councillors indicate that coalition formation depends on: personal
chemistry, special confidence between some parties and traditions for cooperation between some parties. There are no systematic differences between
municipalities where the politicians indicate that these factors are important
and places where the politicians indicate that they are not, except that the
variable dominant party tends to be insignificant and the variable median
party to be significant only where these three factors are important. We conclude that personal chemistry, special confidence and traditions for cooperation do not have any important systematic effects on coalition formation. This
furthermore bolsters our claim that the results presented below are not special
for very small communities, but are also relevant for the understanding of the
workings of larger scale parliaments.
Third, in one of three of the Danish municipalities, a single party holds a
majority of municipal seats (cf. Table 1, row 2). Since, according to many
coalition theories, the incentive to form a multi-party coalition is removed in
this situation, many studies exclude majority situations. For instance, Martin &
Stevenson (2001: 39, Footnote 9) exclude the United Kingdom from their
cross-national analysis for this reason, and Laver et al. (1998: 344) exclude
majority situations from their study of British local governments for the same
reason. However, since (as noted by Lijphart 1999: 91–92) it is an empirical
question as to how the coalition formation game is played in this situation, we
include majority situations in our study, but we have examined whether our
models work differently if we exclude majority situations.With very few exceptions this is the not case.11 For these reasons, we do not include interaction
effects of the size of the municipality, special local conditions or majority
situations, and we report results for all Danish municipalities.
The results of the analysis are reported in Table 3. In the first model for the
classic office- and policy-oriented coalition theories, we test variables indicating whether the coalition is winning, whether it is minimal winning and
whether it includes the median party. We also include the dominant party (H5)
because this is the office-seeking parallel to the policy-seeking median party.
The model shows that office-oriented theories get considerable support, while
policy-oriented theories only have small effects. The classic office-oriented
theories predicting winning coalitions, minimal winning coalitions and
© 2007 The Author(s)
Journal compilation © 2007 (European Consortium for Political Research)
–
–
–
–
1.51***
0.04
H3: Minimum winning
coalition (smallest
number of seats)
H4: Minimal winning
coalition with smallest
number of parties
(the ‘bargaining
position’)
H5: Coalition including
a dominant party
© 2007 The Author(s)
Journal compilation © 2007 (European Consortium for Political Research)
H6: Minimal connected
winning coalition
1.04
4.52
1.74
0.56**
H2: Minimal winning
coalition
6.22
1.83***
Odds ratio
H1: Winning coalition
(not necessarily
minimal)
Coefficient
Model 1:
Classic office- and
policy-oriented theories
–
0.47
0.98***
-0.86***
–
1.88***
Coefficient
–
1.61
2.65
0.42
–
6.56
Odds ratio
Model 2:
Variations of office- and
policy-oriented theories
Table 3. Conditional logit analysis of coalition formation in Danish municipalities
–
-1.01*
–
–
–
1.96***
Coefficient
–
0.36
–
–
–
7.11
Odds ratio
Model 3:
Leaving the universe
of minimal coalitions
–
-0.72
–
–
–
1.77***
Coefficient
0.49
–
–
–
5.89
Odds ratio
Model 4:
Extending with
institutional factors
736
asbjørn skjæveland, søren serritzlew & jens blom-hansen
–
–
–
H11A: Coalition not
breaking pre-electoral
coalitions
H11B: Coalition
identical to a
pre-electoral coalition
H12: Coalition identical
to incumbent
administration
–
H9: Coalition including
the median party that
is also the largest
–
0.42*
H8: Coalition including
the median party
H10: Coalition
including a median
party that controls at
least four seats (office
capacity)
–
H7: Minimal connected
winning coalition with
smallest ideological
range
–
–
–
–
–
1.52
–
–
–
–
–
16.1
0.08
0.48
–
–
–
–
–
–
–
14.42
9.51 ¥ 106
–
0.52*
-0.28
1.08
1.61
–
–
–
–
1.84 ¥ 106
1.68
0.75
1.61***
0.27
0.13
2.31*
–
0.57*
-0.07
4.99
1.31
1.14
10.06
–
1.78
0.94
theories of coalition formation
737
© 2007 The Author(s)
Journal compilation © 2007 (European Consortium for Political Research)
–
Connected coalition
(not necessarily
minimal)
–
–
0.16
0.70
© 2007 The Author(s)
Journal compilation © 2007 (European Consortium for Political Research)
0.63
0.19
11,036
–
–
–
Coefficient
–
–
–
–
–
–
Odds ratio
Model 2:
Variations of office- and
policy-oriented theories
0.42
0.30
11,036
–
–
–
1.65
0.56
0.35
11,036
0.37*
-7.79***
8.46 ¥ 10-5
-9.38***
0.50**
7.36***
Coefficient
7.56 ¥ 103
Odds ratio
–
–
–
1.45
4.14 ¥ 10-4
1.58 ¥ 103
Odds ratio
Model 4:
Extending with
institutional factors
8.93***
Coefficient
Model 3:
Leaving the universe
of minimal coalitions
Notes: 1 The conditional logit analysis assumes the independence of irrelevant alternatives (IIA), which implies that the odds of choosing one
alternative over another do not depend on any other alternatives in the choice set (Long & Freese 2003: 207–210). We follow a strengthened
version of the procedure used by Martin & Stevenson (2001: 39, Note 8) to determine whether the IIA assumption is a problem in our analysis.
We drop a random 50 per cent of the alternatives from each formation opportunity and then apply a Haussman test. This procedure is repeated
50 times. In the table, we report the average p-value over the 50 replications for rejecting the null hypothesis that the IIA assumption holds. As
can be seen, we cannot reject the null hypothesis in any of the four models. We therefore conclude that IIA is not a problem. We are grateful to
Lanny Martin for providing us with the program to perform this test of the IIA assumption and for suggesting how to strengthen the test.
* p < 0.05; ** p < 0.01; *** p < 0.001. Dependent variable: Local coalition as measured by the party affiliation of the mayor and the committee
chairmen.
Pseudo R
Average p-value for
rejecting IIA1
–
–
–
–
Odds ratio
2
11,036
–
Coalition’s share of
parties
N
–
Coefficient
Model 1:
Classic office- and
policy-oriented theories
Coalition’s share of
seats
Table 3. Continued.
738
asbjørn skjæveland, søren serritzlew & jens blom-hansen
theories of coalition formation
739
coalitions containing a dominant party are all supported, although the last
turns out to have insignificant and even negative effects in the other models.
The characteristic of being a winning coalition are especially important. The
odds ratio shows that a winning coalition is 6.22 times more likely to be formed
than a non-winning coalition. In sum, this model shows that coalition behavior
is easier to explain by office than policy motives, but the model has a relatively
low explanatory power and the results are preliminary since primarily the
classic and somewhat crude coalition theories are examined.
In model 2 we examine other versions of the classic office- and policyoriented coalition theories by replacing some of the variables with more
sophisticated variants. We retain winning coalitions in the model, and again
this is a strong predictor. More recent policy-oriented theories are examined
by introducing minimal connected winning coalitions with the smallest ideological range at the expense of minimal connected winning coalitions (due to
multicollinearity). Although the coefficient is higher, it is not statistically significant. Policy considerations are also measured by including the median party
and the median and largest party, but these variables fail to obtain statistical
significance. More recent office-oriented theories are examined by removing
minimal winning coalitions (due to multicollinearity) and instead examining
two alternative versions of this coalition type – namely minimum winning
coalitions (i.e., minimal winning coalitions with the smallest number of municipal seats) and minimal winning coalitions with the smallest number of parties.
As for the minimal winning coalition in model 1, these two types of minimal
winning coalitions are strong predictors, but note an oddity: minimum winning
coalitions has a negative sign. We have examined this counter-intuitive result in
a number of different specifications of model 2. It is robust. The two versions
of minimal winning coalitions end up with statistically significant, but opposite
signs. This means that, given a winning coalition, actors seek to form coalitions
that contain as few parties as possible, but not as few municipal seats as
possible.
We interpret this result in the following way. It supports the idea that actors
take the number of municipal seats and the number of parties into consideration when forming coalitions. It does not support the idea that actors seek to
establish minimal coalitions. Since this characteristic is included in both the
minimum winning coalition and the minimal coalition with the smallest
number of parties, and since these two variables come out with different signs,
the concept of minimal coalitions is seriously undermined. It indicates that
actors care about office benefits, but do not necessarily seek to maximize them.
One is reminded of Herbert Simon’s (1985) concept of ‘satisficing behavior’. In
this interpretation, actors are satisfied once a certain level of office benefits is
achieved, after which other considerations may be taken into account. The
© 2007 The Author(s)
Journal compilation © 2007 (European Consortium for Political Research)
740
asbjørn skjæveland, søren serritzlew & jens blom-hansen
negative sign of minimum winning coalitions might indicate that actors seek to
reduce the vulnerability of the coalition from defection by including morethan-necessary seats. The positive sign of minimal winning coalitions with the
smallest number of parties may indicate that actors seek to keep the number
of parties low – albeit not necessarily at the minimal winning level – not only
to increase office benefits, but also because this reduces the transaction costs of
intra-coalition bargaining and makes the building up of trust within the coalition easier.
These are speculative thoughts, but to investigate them further we completely leave the universe of minimal coalitions in model 3. Instead we
examine whether the ‘seat-share’ and ‘party-share’ aspect – and not the
‘minimal coalition’ aspect – of the two versions of the minimal winning coalitions in model 2 are what drive the model. The seat-share and party-share
variables do not take into consideration whether the coalition is minimal. As is
evident from model 3, the two variables are surprisingly strong predictors. This
further undermines the idea that office-oriented politicians necessarily seek
minimal coalitions. To examine whether the same phenomenon holds in relation to policy considerations, we introduce a new variable that only measures
whether coalitions are connected, but disregards whether or not they are
minimal. As is evident from Table 3, this variable is also a strong predictor.
Compared to model 2, the explanatory power of the model jumps from 0.19 to
0.30 (i.e., by more than 50 per cent). Model 3 lends further credence to our
speculations that actors seek to gain influence, but not necessarily maximum
influence. Our conclusion, therefore, is that they do not seek minimal coalitions, but rather coalitions that are satisfactorily small.
In model 4 we extend our models with institutional characteristics of the
various coalition types. We introduce the concept of ‘office capacity’ (and, due
to multicollinearity, we leave out the median and largest party).As can be seen,
the hypothesis on office capacity and on the median party (which is not
necessarily the largest) is supported. The model also shows strong support for
an incumbency effect, whereas pre-electoral coalitions do not seem to matter
much for post-electoral coalitions. This model increases the explanatory power
to 0.35.
Conclusion: The defeat of the old coalition theories
Do actors seek minimal coalitions? This is a plausible theory, but the empirical
evidence is far from unequivocal. And evidence is mostly provided by the
well-known dataset consisting of national coalitions in postwar Europe – the
very dataset that gave rise to the theory in the first place (cf. the circularity
© 2007 The Author(s)
Journal compilation © 2007 (European Consortium for Political Research)
theories of coalition formation
741
problem discussed in the introduction to this article). Yet even in this dataset,
coalitions are not always minimal. About half of the coalitions in postwar
Europe are either oversized or undersized (see, e.g., Lijphart 1999: 98). Our
results indicate that we need to rethink the concept of minimal coalitions. Is
this what actors seek, or are minimal coalitions by-products of the pursuit of
other goals?
Our study confirms that actors prefer winning coalitions. However, given
that a positive vote of investiture is required in Danish municipalities, this
finding is not surprising. Given that coalitions are winning, do actors seek to
minimize them? We have tested a number of classic and more recent coalition
theories based on this assumption, and our results do not confirm it. The
theories are supported in the sense that actors do not establish all-embracing
coalitions. They seek coalitions that are sufficiently small to provide office
benefits. Yet this does not necessarily mean minimal coalitions. Our findings
also confirm that actors are policy-oriented and, consequently, seek to establish connected coalitions and include the median party – but again, such
coalitions need not be minimal.
The results suggest that once a coalition has reached a sufficiently small
size, actors let other considerations gain prominence. Our study was not
designed to explore what these considerations are, but we offer the thought
that actors care about things like their coalition’s vulnerability to defection
and the negotiation climate in the legislature. These goals induce actors to
establish larger-than-minimal coalitions. In short, our study suggests that we
need to direct more attention to the phenomenon of oversized coalitions.
The term ‘oversized’ suggests something irrational, but we are not convinced
that this is the case. Oversized coalitions are only irrational from the perspective of theories based on the assumption that actors want to maximize
office benefits or policy influence. If actors value other goals, rational behavior may lead to other coalition types. Oversized coalitions are slowly gaining
more scholarly attention (see, e.g., Volden & Carubba 2004). In subsequent
research we will investigate them more systematically within the Danish
municipal setting.
Turning to institutions, our study confirms that they also have an impact on
coalition behavior. We found support for an incumbency effect. This has also
been found by Martin & Stevenson (2001: 46) and Bäck (2003: 458–460).
Consequently, this must be considered a robust finding. Our results also
support the contention that parties need office capacity to enter coalitions. The
theory of office capacity is of recent origin and has so far primarily been
investigated within a Scandinavian setting (Skjæveland 2003, 2004). The results
are promising, and we think the theory deserves exploration in a broader
empirical setting.
© 2007 The Author(s)
Journal compilation © 2007 (European Consortium for Political Research)
742
asbjørn skjæveland, søren serritzlew & jens blom-hansen
Finally, our test of coalition theories has avoided the circularity problem
discussed in the introduction – that is, we have not used national coalitions in
postwar Europe as our empirical setting because the theories we wanted to
test were formulated with an intimate knowledge of this dataset. By focusing
on local governments, we have followed an empirical strategy suggested by
many, but followed by few. Our study does not allow us to gauge the extent to
which our findings are contingent on the Danish context, but, as we argue in
the article, there is no compelling reason why the theories should not work at
the municipal level in Denmark. We thus believe that our findings may have
value beyond the Danish case. We think that focusing on local governments in
coalition studies is a strategy worth exploring in other national settings. The
strength of a new empirical testing ground is that established theories may be
challenged and new thoughts and perspectives stimulated. We feel that our
study has shown some potential in this regard and demonstrated a need for
more local coalition studies.
Acknowledgments
The authors would like to thank anonymous referees of this journal, Jørgen
Elklit, Bernard Grofman, participants at the IPSA conference on Sub-State/
Sub-National Legislatures, Quebec City, Canada, 21–23 October 2004, and the
annual meeting of the Nordic Local Government Researchers, Oslo, Norway,
26–28 November 2004, and colleagues at the Department of Political Science,
University of Aarhus, for helpful comments to the questionnaire and earlier
drafts of this article.
Notes
1. Various definitions of a national government can be found in the literature. Laver &
Shepsle (1996: 28–30) want to distinguish between a government as such and the government’s legislative support coalition. We agree, and by implication define a ‘national
government party’ as a party with one or more government ministers. The ‘national
government coalition’ is the set of national government parties. Following De Swaan, for
convenience the term ‘government coalition’ is also used for a one-party government,
but opposed to De Swaan we do not take the empty coalition of no parties into
consideration (cf. De Swaan 1973: 30, Footnote*). Please refer to the section about
studying government coalitions at the local level for our corresponding definition of a
‘local government coalition’.
2. In addition to the studies of local coalitions mentioned in this paragraph, note should
also be taken of the volume edited by Mellors & Pinjenburg (1989), which contains
country studies of local coalitions in nine European countries. However, these country
© 2007 The Author(s)
Journal compilation © 2007 (European Consortium for Political Research)
theories of coalition formation
3.
4.
5.
6.
7.
8.
9.
10.
11.
743
studies are not supposed to be – and do not in fact constitute – systematic tests of
coalition theories. On the contrary, they are explorative studies examining the potential
of studying coalitions at the local level (and as such they have considerable merit).
We should emphasize that the distinction between office- and policy-oriented theories is
an analytical one, which aids the understanding of the theories. We do not claim that
actors cannot bear both objectives in mind when actually forming government coalitions.
Perhaps a more accurate way of putting the point is that more recent theories include the
effect of institutions other than the institutional demand for a parliamentary majority
behind the coalition.
We present and examine theories that consider party positions in one-dimensional
policy space, not multi-dimensional theories.
Gamson’s theory can be reduced to predicting minimum winning coalitions. His broader
theory actually includes policy considerations (Gamson 1961).
A ‘core party’ is one located in the core. The ‘core’ is the intersection of the compromise
sets for all winning coalitions. A party that is the median party on the left-right dimension and is also the largest party is not necessarily a (stable) core party.
Please refer to Note 1 above for our definition of a national government.
Four of Denmark’s then 271 municipalities were not included in the survey: Copenhagen, Frederiksberg, Bornholm and Aarhus. The first three function as both municipalities and county governments. The last has an alderman system, whereas all other
municipalities have a committee system (for coalition formation in the few municipalities with an alderman system, see O’Leary et al. 2005). The survey thus comprises all the
267 ordinary Danish municipalities. For a detailed description of the survey, see BlomHansen et al. (2004).
In the analyses that only include large municipalities, the variables Minimal Winning
Coalition and Median Party fail to obtain statistical significance at the 5 per cent level in
model 1. In model 3, the variable Median Party is insignificant, and in model 4 the
variables Connected Coalition, Median Party and Coalition including a Median Party
which Controls at Least Four Seats (Office Capacity) fail to obtain statistical significance
at the 5 per cent level. We judge these differences from the full analysis to be minimal.
Results are available from the authors upon request.
In the analyses that only include minority situations, the variables Median Party and
Connected Coalition are insignificant at the 5 per cent level. We judge these differences
from the full analysis to be minimal. Results are available from the authors upon
request.
References
Axelrod, R. (1970). Conflict of interest: A theory of divergent goals with applications to
politics. Chicago, IL: Markham.
Bäck, H. (2003). Explaining and predicting coalition outcomes: Conclusions from studying
data on local elections. European Journal of Political Research 42: 441–472.
Black, D. (1958). The theory of committees and elections. Cambridge: Cambridge University
Press.
© 2007 The Author(s)
Journal compilation © 2007 (European Consortium for Political Research)
744
asbjørn skjæveland, søren serritzlew & jens blom-hansen
Blom-Hansen, J. (1999). Policy making in central–local government relations: Balancing
local autonomy, macroeconomic control and sectoral policy goals. Journal of Public
Policy 19: 237–264.
Blom-Hansen, J, Serritzlew, S. & Skjæveland, A. (2004). Spørgeskemaundersøgelse om koalitionsdannelse i de danske kommune. Design og svarfordelinger. Aarhus: Aarhus University, Department of Political Science.
Colomer, J.M. & Martinez, F. (1995). The paradox of coalition trading. Journal of Theoretical
Politics 7: 41–63.
Damgaard, E. (1969). The parliamentary basis of Danish governments: The patterns of
coalition formation. Scandinavian Political Studies 4: 30–57.
Denters, B. (1985). Towards a conditional model of coalition behaviour. European Journal of
Political Research 13: 295–309.
De Swaan, A. (1973). Coalition theories and cabinet formations: A study of formal theories of
coalition formation applied to nine European parliaments after 1918. Amsterdam/
London/New York: Elsevier Scientific.
Galton, F. (1907). One vote, one value. Nature 75(1948): 414.
Gamson, W.A. (1961). A theory of coalition formation. American Sociological Review 26:
373–382.
Laver, M. (1989). Theories of coalition formation and local government coalitions. In C.
Mellors & B. Pinjenburg (eds), Political parties and coalitions in European local government. London: Routledge.
Laver, M. (1998). Models of government formation. Annual Review of Political Science 1:
1–25.
Laver, M. & Schofield, N. (1990). Multiparty government: The politics of coalition in Europe.
Oxford: Oxford University Press.
Laver, M. & Shepsle, K.A. (1996). Making and breaking governments: Cabinets and legislatures in parliamentary democracies. Cambridge: Cambridge University Press.
Laver, M. & Shepsle, K.A. (2000). Ministrables and government formation: Munchkins,
players and big beasts of the jungle. Journal of Theoretical Politics 12: 113–124.
Laver, M. & Benoit, K. (2003). The evolution of party systems between elections. American
Journal of Political Science 47: 215–233.
Laver, M., Rallings, C. & Thrasher, M. (1987). Coalition theory and local government:
Coalition payoffs in Britain. British Journal of Political Science 17: 501–509.
Laver, M., Rallings, C. & Thrasher, M. (1998). Policy payoffs in local government. British
Journal of Political Science 28: 333–353.
Le Maire, E. & Preisler, N. (2000). Lov om kommunernes styrelse. Med kommentarer.
Copenhagen: Jurist- og Økonomforbundets forlag.
Lees, C. (2000). The Red-Green coalition in Germany: Politics, personalities and power.
Manchester/New York: Manchester University Press.
Lijphart, A. (1999). Patterns of democracy. New Haven, CT: Yale University Press.
Long, J.S. & Freese, J. (2003). Regression models for categorical dependent variables using
Stata. College Station, TX: Stata Press.
Martin, L.W. & Stevenson, R.T. (2001). Government formation in parliamentary democracies. American Journal of Political Science 45: 33–50.
Mellors, C. (1989). Sub-national government: A new arena for the study of coalitions. In C.
Mellors & B. Pinjenburg (eds), Political parties and coalitions in European local government. London: Routledge.
© 2007 The Author(s)
Journal compilation © 2007 (European Consortium for Political Research)
theories of coalition formation
745
Mellors, C. & Pinjenburg, B. (eds). (1989). Political parties and coalitions in European local
government. London: Routledge.
Müller, W. & Strøm, K. (eds). (2000). Coalition governments in Western Europe. Oxford:
Oxford University Press.
O’Leary, B., Grofman, B. & Elklit, J. (2005). Divisor methods for sequential portfolio
allocation in multi-party executive bodies: Evidence from Northern Ireland and
Denmark. American Journal of Political Science 49(1): 198–211.
Page, E.C. (1991). Localism and centralism in Europe: The political and legal bases of local
self-government. Oxford: Oxford University Press.
Pedersen, M.N. (1997). Når kagen skal deles: Konstitueringens politik. In J. Elklit & R. Buch
Jensen (eds), Kommunalvalg. Odense: Odense Universitetsforlag.
Riker, W. (1962). The theory of political coalitions. New Haven, CT: Yale University Press.
Schofield, N. (1993). Political competition and multiparty coalition governments. European
Journal of Political Research 23: 1–33.
Simon, H.A. (1985). Human nature in politics: The dialogue of psychology with political
science. American Political Science Review 79: 293–304.
Skjæveland, A. (2003). Government formation in Denmark, 1953–1998. Aarhus: Politica.
Skjæveland, A. (2004). Modelling Government Formation in Denmark and Beyond. Unpublished working paper. Aarhus: Department of Political Science.
Steunenberg, B. (1992). Coalition theories: Empirical evidence for Dutch municipalities.
European Journal of Political Research 22: 245–278.
Van Roozendaal, P. (1990). Centre parties and coalition cabinet formations: A game theoretic approach. European Journal of Political Research 18: 325–348.
Volden, C. & Carubba, C.J. (2004). The formation of oversized coalitions in parliamentary
democracies. American Journal of Political Science 48: 521–537.
Von Neumann, J. & Morgenstern, O. (1980 [1944]). Theory of games and economic behavior.
Princeton, NJ: Princeton University Press.
Address for correspondence: Søren Serritzlew, Department of Political Science, University of
Aarhus, DK-8000 Aarhus C, Denmark. Tel.: +45 89421274; E-mail: [email protected]
© 2007 The Author(s)
Journal compilation © 2007 (European Consortium for Political Research)