Coalition Context, Voter Heuristics and the Coalition
Directed Vote
Raymond M. Duch
[email protected]
Nuffield College
University of Oxford
Jean-Robert Tyran
[email protected]
Department of Economics
University of Vienna ∗
February 18, 2013
∗
Paper prepared for presentation at the Annual meeting of the European Political Science Association,
Berlin, Germany, June 21-23, 2012. We thank the Carlsberg Foundation for financial support. We thank
Ulrik H. Nielsen for effective research assistance.
1
2
Abstract
In contexts with multi-party governing coalitions national electorates employ
heuristics that are adapted to the complexity of the political environment. Evidence
from Denmark and Germany, indicates that voters acquire the heuristics that allow
them to anticipate the types of coalitions that form after an election. Employing
experimental vingettes embedded in internet surveys we find that large numbers
of voters 1) understand the basic arithmetic of majority coalition formation; 2)
recognise that formateur parties likely enter the governing coalition; 3) anticipate
that parties that are proximate on the ideological scale will more likely agree to a
coalition; 4) anticipate coalitions of parties that are ideologically connected; and
5) anticipate the formation of minimal winning connected coalitions. The Danes
exhibit more sophisticated coalition heuristics than the Germans which may result
from differences in the complexity of the two countries’ coalition dynamics. Acquiring these heuristics increases the likelihood of exercising an informed coalition
directed vote.
3
1
Introduction
In most democratic elections voters are faced with a choice between an incumbent
coalition, composed of a number of parties, and an expectation that a similar or different
group of parties could form a new governing coalition after the election. But do voters
actually engage in the kind of coalition reasoning that would allow them to attribute
responsibility to the parties in a multi-party coalition and to anticipate the kinds of
coalitions that form after an election? This essay will focus on the latter aspect of
coalition reasoning – their ability to anticipate the kinds of coalitions that form after an
election.
Recently a number of comparative voting behaviour scholars, in particular Kedar
(2009), Duch and Stevenson (2008), Duch, May and Armstrong (2010), have argued for a
vote utility function that incorporates coalition reasoning. Common to all these models
(some of which have a policy voting orientation while others have a performance voting
perspective) is the notion that voters exercise a coalition-directed vote. Voters are not
simply assessing the party in isolation but rather thinking about the party’s contribution
to an outcome that is taken by a governing coalition, formed after the election and,
obviously, made up of multiple parties.
But these large N studies do not explore in any detailed fashion the micro-level assumptions underpinning the notion of a coalition-directed vote choice. One could imagine
voters who are uber-sophisticated and perfectly informed in their coalition calculations.
But more likely most voters employ heuristics for understanding, and making decisions
about, coalition governments. By heuristics we mean strategies that ‘guide information search and modify problem representations to facilitate solutions’ (Goldstein and
Gigerenzer 2002). Heuristics are used when information acquisition is costly and decision
making is cognitively challenging (Kahneman 2011, Simon 1955) – conditions clearly met
in complex coalition systems. We make the argument that the sophistication of these
4
heuristics is ecologically rational: coalition reasoning is relatively sophisticated in political context that have a history of complex coalition governance and formation while they
are much less well-developed in countries without a history of coalition governance.1
The essay is organised as follows: First we describe the coalition reasoning that is
associated with a coalition-directed vote. We argue that the incidence of coalition reasoning in the population is a function of context, specifically the complexity of coalition
governance and formation in the country. We then propose a strategy for recovering,
using vignettes, the heuristics that make up coalition reasoning at the individual voter
level. A subsequent section describes the specific coalition heuristics employed by the
Danish and German electorates and finds that large proportions of the two populations
exhibit relatively sophisticated coalition reasoning heuristics. We find a higher incidence
of coalition reasoning heuristics in the Danish population and speculate that this is a
result of ecological rationality. Finally we demonstrate that a more informed coalitiondirected vote is exercised by respondents identified as having more developed coalition
reasoning heuristics.
2
Coalition-directed voting
Coalitions form after elections as a result of bargaining amongst parties over the poli-
cies to be enacted by the government (Austen-Smith and Banks 1988). Policy outcomes in
coalition governments are expected to reflect the policy preferences of the parties forming
the governing coalition weighted by their legislative seats (Indridason 2011, Schofield and
Laver 1985). In multiparty contexts with coalition governments, Austen-Smith and Banks
(1988) argue, policy voting, directed simply at parties, is not rational. The implication
1
The insight regarding ecological rationality is from Gigerenzer and Todd (1999) who
pointed out that “A heuristic is ecologically rational to the degree that it is adapted to
the structure of the environment.”
5
of the Austen-Smith and Banks (1988) insight here is that voters anticipate the likely
coalition formation negotiations that occur after the election and they condition their
vote choices accordingly in order to maximise the likelihood that a coalition government
forms that best represents their preferences over government policies or performance.
Scholars have recently taken this theoretical proposition seriously and have suggested
modifications to the conventional vote utility function that incorporate this voter coalition
reasoning. When incumbent governments consist of a single party the standard vote
utility, building on Downs (1957), is simply expressed in terms of Euclidean distance:
(xi − pj )2 . In a context where the governing coalition consists of multi-parties, voters
should think of pj as the likely contribution of a party, pj , to a governing coalition’s policy
position.
Duch, May and Armstrong (2010) argue that a coalition-directed vote utility function
should incorporate information about likely coalition outcomes and the distribution of
administrative responsibility within those coalitions. The utility that the voter derives
from a party incorporates the probability of the party entering into a particular coalition
along with the expected influence of the party on the coalition’s policies. Voters are
expected to know which parties are likely to coalesce and also to recognise that the seat
share of coalition members translates into inflence over coalition policies.
Kedar (2005, 2009) has argued that the rational voter focuses on policy outcomes
and hence on the issue positions that are ultimately adopted by the coalition government
that forms after an election. And she demonstrates that in political systems with coalition governments this leads to “compensational voting” aimed at minimizing the policy
distance between the policy compromises negotiated by the governing coalition and the
voters ideal policy position.
Duch and Stevenson (2008) contend that the amount of administrative responsibility
that each party holds in the governing coalition, or is likely to hold after the election,
6
conditions the overall magnitude of the economic vote and how it is distributed across
parties. A key assumption in their model is that voters know, or anticipate, the distribution of administrative responsibility which could narrowly be defined as the relative
number of portfolios allocated to the parties within the coalition. Also, voters anticipate
the likely coalitions that form after an election and they assess the impact of their party
vote choice on the likelihood of different coalitions coming to power. And this information
regarding their pivotal role is used by voters to weight the importance of an economic
competency signal in their vote choice function.
These authors and others find that voters do respond in an instrumentally rational
fashion to the strategic incentives associated with post-election coalition formation possibilities (Bargsted and Kedar 2009, Gschwend 2007, Bowler, Karp and Donovan 2010,
Blais, Loewen and Bodet 2004) or they engage in vote discounting whereby voters support
more extreme candidates because they anticipate the moderating impact of the legislative
process on policy outcomes (Tomz and Houweling 2007, Adams, Bishin and Dow 2004,
Merrill and Groffman 1999, Alesina and Rosenthal 1995).
These empirical results are based on theories in which voters know about the likely
coalitions that form after an election; consider the location of parties in the ideological
space; and treat seats as a proxy for policy influence in a coalition. One might conclude
from this that voters essentially have a mental regression model that predicts the likelihood of different coalition outcomes. In effect, they could be replicating the calculations
in empirical coalition formation models such as Martin and Stevenson (2001) – a model
that identifies the most significant variables that predict coalition formation outcomes.
In fact, we have very little empirical insight into precisely what mental coalition
formation models inform the coalition directed vote of the average individual. Its unlikely
that these mental models are faithful to the full information calculus implied by these
theoretical models. This essay proposes a strategy for identifying some of the elements
7
that make up the mental coalition formation model of the average voter. First, we propose
a method for measuring coalition reasoning. Second, we argue that these reasoning
abilities should vary by political and institutional context. And we present some evidence
that this may be the case although we are careful to point out that the results are simply
that, suggestive. Finally, we examine the extent to which individual-level variation in
these coalition reasoning abilities conditions vote choice in a fashion consistent with
theories of the coalition vote.
3
Heuristics and the coalition directed vote
More likely, voters employ coalition reasoning heuristics that result in choices that
approximate those implied by models of the coalition directed vote. By heuristics we
mean strategies that ‘guide information search and modify problem representations to
facilitate solutions’ (Goldstein and Gigerenzer 2002). Heuristics are used when information acquisition is costly and decision making is cognitively challenging (Simon 1955). A
key finding of this research on heuristics (Gigerenzer and Todd 1999) is that, for a wide
variety of decision making problems, there are decision making short-cuts that result in
choices that approximate those of full information rational decision making. For example,
in assessing the likelihood of a particular coalition forming, voters may not consider all
significant variables (along with their weights) from the Martin and Stevenson (2001)
regression model. They might focus simply on whether the coalition included the formateur party and whether all parties were ideologically “connnected.” And such decision
making short-cuts may generate predictions that very closely approximate those of the
fully-specified model. Our goal in this essay is to identify the decision making short-cuts
that voters employ when exercising a coalition-directed vote.
Second, the heuristics individuals adopt are calibrated to the decision making tasks.2
2
For a contrasting view that heuristics can result in behavioural biases, for example
8
According to this line of reasoning, individuals employ heuristics that approximate optimal choices. Hence if the decision problem is extremely simple, individuals require, and
adopt, only very simple heuristics to approximate optimal decisions. And if the decision problem is informationally or cognitively more demanding individuals adopt more
complex heuristics that are at the same time less demanding than those assumed by
full-information rational models of decision making.
Voters face varying levels of complexity in anticipating coalition formations depending
on the political context. As an example, for many years German coalition formation dynamics were extremely simple – recognising the importance of the formateur rule allowed
voters to easily predict what coalition would form. Typically, the party with the largest
number of seats would coalesce with the FDP. Hence a very simple heuristic sufficed. In
other contexts, such as Denmark, voters would require more complex heuristics to predict
coalition outcomes. In such cases, formateur status might be important but the range of
potential coalition partners could be much more numerous in which case other heuristics
would be required to narrow down the parties likely to enter the coalition government.
Our conjecture is that this complexity is communicated to the average voter. Stromback and Aalberg (2008) provide evidence to this effect. They compare the content
of election coverage in Norway and Sweden – the former experiences multiparty coalitions with considerable bargaining while in the case of Sweden the Social Democrats
have dominated government formation. Their analysis of media coverage of election campaigns suggests quite significant differences in how the campaigns are framed – in Norway
in the context of assessing risky prospects, see (Tversky and Kahneman 1974). We of
course recognise that there is a long-standing debate as to whether and when heuristics
are ecologically rational and when they are misleading (and Tversky and Kahneman
emphasized that they are functional as a rule but can be systematically misleading in
well-specified circumstances).
9
what they call the governing frame, i.e., discussions of the coalition formation process, is
the dominant frame while it is much less important in Sweden. The point here is that
in contexts in which coalition bargaining is quite important, and typically complex, the
electorate will be exposed to considerable information about the process. In contexts
in which coalition bargaining is less complex voters will be exposed to much less media
narrartive about the bargaining process. Our expectation is that coalition bargaining
heuristics in the former case will be more sophisticated than in the latter.
This raises three empirical challenges that we try to address in this essay. First, how
do we measure coalition reasoning at the micro-level: How do we know whether voters
have the aptitude to do what our coalition-direct voting models say they should do?
Second, what is the appropriate strategy for demonstrating ecological rationality? Can we
demonstrate that individuals acquire coalition reasoning appitudes that are appropriate
to the complexity of the coalition governing context in which they find themselves? Third,
how do we empirically test the argument that voters condition their vote choice on the
characteristics of either the governing coalition or coalitions that are likely to form after
an election?
4
Coalition reasoning vignettes
Lets begin with the measurement issue. We define coalition reasoning as being a
general understanding of which parties are likely to enter into a governing coalition. It
is important to point out that it is not political knowledge in the sense of knowing the
names of the parties in the governing coalition or their porfolio allocations. We think of
it as a more general understanding of coalition formation dynamics.
Our expectation is that the heuristics adopted by voters will generally represent accurate characterisations of post-election coalition formations. They do so because important
features of the coalition formation process re-occur regularly (the role of the formateur,
10
for example) and they become prominent in media representations of the political process.
And in contexts with relatively complex coalition formation patterns we expect to see
more of these aspects of coalition formation featured in the national political discourse.
Theory and the empirical analysis of coalition formation patterns provide a good guide to
the kinds of heuristics voters are likely to acquire in these contexts (Laver and Schofield
1990, Martin and Stevenson 2001, Riker 1962). We have identified six features that, while
not exhaustive, capture aspects of coalition formation that the literature suggests should
weigh heavily in the voter’s calculations.
First, at the very minimum, a coalition-directed vote requires that voters understand
what constitutes a majority governing coalition (Duch, May and Armstrong 2010, Duch
and Stevenson 2008). A second dimension concerns the formateur advantage that is
typically accorded the party with the largest number of seats. Formateur status should
increase a party’s likelihood of entering a governing coalition (Ansolabehere and Ting
2005). A third dimension is the left-right spatial representations of electoral competition. Ideological proximity has been demonstrated to condition the likelihood of parties
agreeing to form a coalition (deSwaan 1973). Accordingly we expect voters to adopt
ideological proximity as a heuristic that signals the increasing likelihood of coalition
agreement amongst parties. A related heuristic is that coalitions tend to be constructed
from parties that are ideologically connected – an ideological neighbour is not ignored in
favour of a party that is ideologically more distant. A fifth dimension is whether coalitions are over-sized or minimal winning (Martin and Stevenson 2001, Riker 1962). The
empirical and theoretical literature suggest that a minimal winning connected heuristic
would generally contribute to predicting, correctly, coalition outcomes and hence is a useful voter heuristic. Finally, our expectation is that voters recognise that parties from the
incumbent coalition have an advantage in post-election coalition formation negotiations.
11
The experimental treatments described below are designed to determine the extent to
which these six heuristics are in fact prevalent in the voting population.
The experimental vignettes were administered in two online surveys. The German
CCAP study consisted of a baseline survey in June of 2009. Subsequent panel waves
took place three months after the first wave (August 2009) before the State elections in
Saxony, Saarland, and Thuringia. A third wave was carried a month later (September
2009) before the Federal election of September 27th. And a final post-election wave
conducted on October of 2009. A total of 4,301 respondents were interviewed in the
2009 baseline, with an extra 1,904 which entered new in the second wave for a total of
6,205 respondents. The Danish vignette experiment was run in summer 2011 as part of
iLEE4, the fourth wave of the Internet Laboratory for Experimental Economics project
at the University of Copenhagen. iLEE4 is an omnibus experiment (consisting of several
independent modules) carried out over the internet with participants drawn from the
adult Danish population in collaboration with Denmark Statistics. Of the 2,291 invited
to participate in the survey, 689 completed the whole survey.
Respondents are exposed to eight hypothetical coalition vignettes in total and in
each case are asked to identify the most likely majority governing coalition that would
form. Respondents are presented with a graph that summarises the seat allocations –
in percentages – for each of the parties. Respondents were asked to identify the most
likely majority governing coalition in each case. In the Danish experiments we added
two modifications to the vignettes. First, we introduced an incumbent cue treatment.
Half of the Danish subjects were randomly assigned to a treatment in which one of
the parties was identified as an incumbent from the pre-election coalition government.
The other half received no incumbent cue. This allows us to assess whether incumbent
cues represent an important part of the heuristics voters employ in predicting coalition
formation outcomes. The incumbent cue treatment results in different coalition choices
12
in only one case: Treatment 3A and 3B which is described in Figure 3 and Figure 4.
A second treatment varied the left-right spatial location of parties for scenarios that
clustered parties on the left or right. In these cases, half of the subjects see a clustering
to the left for a given number of parties, the other half see a clustering to the right. We
introduced this treatment in order to assess whether the left versus right clustering of
parties might have affected the parties subjects expected to form a coalition. There was no
difference in the coalition choices of subjects in the left versus right clustering treatments.
A detailed description of the respondent instructions is provided in the Online Appendix.
We begin with a description of the scenarios and of the basic choices made by respondents. Generalisations about the heuristics respondents employed will then follow.
Figure 1 and Figure 2 present the first treatments with more than two parties.3 In
Treatment 2A and 2B the parties have the same seat allocations – A has 45 percent of the
seats; B has 16 percent; and C has 39 percent. In Treatment 2A the parties are equidistant
and straddle the ideological centre. The respondents’ choices are recorded below the
figures. Interestingly, not only do most German and Danish respondents recognise that
a coalition is necessary to form a majority government, but they favour the AB majority
coalition over BC or ABC. Just under 50 percent of Danish respondents and almost 40
percent of German respondents expect that the party with the largest numbers of seats
in the legislature would enter into coalition with B. The coalition BC was the secondmost favoured choice by respondents in both samples. By favouring AB, respondents
seem to recognise the formateur advantage in coalition formation. Moreover there is
3
An initial treatment simply provided respondents with two parties – one receiving
49 percent of the seats and another receiving 51 percent of the seats. As in the other
vignettes they were asked to identify which party or parties would most likely form a
majoarity government. Over 80 percent of both German and Danish respondents selected
the majority party.
13
some evidence that formateur advantage, i.e., AB, trumps a minimum winning connected
heuristics since BC would be the choice associated with such an heuristic.4
Treatment 2B results in a large shift of the parties B and C to the Right extreme of the
ideological continuum leaving Party A far to the Left of these parties. Responses to the
increase in ideological distance between A and B are as we might expect. The respondents
are now more likely to choose BC (which does not include the largest party in terms of
seats) than AB. Again, one interpretation here is that respondents recognise that the
formateur advantage can be out-weighed by party considerations of ideological proximity
– although the ideological treatment in this case is quite extreme. And of course the BC
choice in Treatment 2B has the advantage of being majority, ideological proximate and
minimum winning connected. There is though a large difference between the Danes and
Germans here. Over seventy percent of the Danes opt for this choice while only forty-two
percent of the Germans do so. The difference may lie in the greater importance associated
with formateur advantage in the German institutional context than in the Danish one.
4
Both AB and BC are minimal winning coalitions in that they do not include excess
parties although BC is the minimum winning coalition because it includes fewer seats
than AB.
14
Figure 1: Treatment 2A
Figure 2: Treatment 2B
2A
Germany
Denmark
2B
Germany
Denmark
A
AB
ABC
AC
B
BC
C
N
17.61
9.51
37.4
46.54
1.79
1.44
11.39
2.74
4.35
3.6
23.6
31.99
3.85
4.18
4021
694
19.12
5.76
20.34
10.66
1.37
0.29
9.05
2.45
3.93
2.31
41.9
72.91
4.28
5.62
4021
694
15
In treatment 3A and 3B, parties A and B are the only two parties that can form a
minimal winning connected coalition. The results are reported in Figure 3 and Figure 4.
In treatment 3A the two parties are ideologically equidistant and straddle the ideological centre. Danish and German choices differ here. The Danes are overwhelmingly in
agreement on the AB coalition which is both minimal winning connected and includes
the party with the most seats – 65 percent of Danish respondents make this choice. Germans respond quite differently with only 25 percent selecting the AB minimal winning
coalition. The modal response (27 percent) in effect is BC which is just shy of enough
seats to form a majority governing coalition. Somewhat surprisingly, 19 percent chose an
unconnected coalition BD. A common theme though is that most of the German choices
include the largest party, B. The Danes, much more than the Germans, are in agreement
on a coalition outcome that is arguably the most likely one to occur.
In treatment 3B, Party A is ideologically isolated on the extreme Left of the continuum while Parties B,C, and D are ideologically cluster on the extreme Right. Assuming
respondents recognise that ideologically distant parties are unlikely to enter a coalition,
the percent selecting AB in Treatment 3B should fall dramatically. This clearly happens
in Denmark where the 65 percentage selecting AB in Treatment 3A drops to 4 percent in
Treatment 3B. It is particularly interesting that the effect of the widening of the ideological gulf between A and B increases, quite dramatically, the extent to which individuals
anticipate the inclusion of Party C in the coalition – one explanation here is that voters
anticipate that B (the formateur) will include ideologically proximate C in the oversized
coalition in order to balance the policy effect of ideologically distant Party A. The ideological treatment effect is less pronounced in Germany – the 24 percent choosing AB
in Treatment 3A drops to 10 percent in Treatment 3B; primarily to the benefit of BC
which is just shy of constituting a majority governing coalition. Like Treatment 3A, the
16
results to Treatment 3B suggest little agreement in the voter population and possibly less
sophisticated coalition reasoning than we find in the Danish context.
Treatment 3A is the one treatment in which we find a large difference between the
treatments with and without an incumbent cue. The results for the two treatments are
presented in the DK-Incumbent and DK-Non-Incumbent rows in Figure 3 and Figure 4.
In the incumbent cue treatment Party A is designated as an incumbent party. And the
incumbent cue in Treatment 3A results in fewer subjects predicting that Party A would
enter into the governing coalition. This is not the case in Treatment 3B although it is
likely that the incumbent cue here is overwhelmed by the ideological distance heuristic.
To the extent that incumbent cues matter we expected them to increase expectations that
a party would enter into the governing coalition. But in this one case we find that they
reduce expected participation in a coalition government. This is an isolated incumbent
cue result – it does not have an effect in all the other treatments – and therefore we are
not inclined to draw any general conclusions based on this one case.
17
Figure 3: Treatment 3A
Figure 4: Treatment 3B
Treat 3A
Germany
Denmark
DK - Non-Incumbent
DK - Incumbent
Treat 3B
Germany
Denmark
DK - Non-Incumbent
DK - Incumbent
AB
ABC
B
BC
BCD
BD
C
CD
D
N
24.27
64.55
69.74
59.36
0.1
8.93
9.22
8.65
13.93
7.78
6.05
9.51
26.59
2.31
2.31
2.31
1.17
3.03
1.44
4.61
18.9
4.76
4.03
5.48
1.64
0.29
0.29
0.29
4.4
1.3
1.15
1.44
2.64
1.01
0.58
1.44
4021
694
347
347
10.07
4.18
4.03
4.32
1.02
31.7
28.82
34.58
14.52
0.86
0.29
1.44
36.98
7.64
8.36
6.92
4.53
8.93
10.95
6.92
20.49
3.31
2.59
4.03
1.59
5.91
6.05
5.76
3.76
11.1
10.95
11.24
2.59
1.87
1.73
2.02
4021
694
347
347
18
In Treatment 4A (Figure 5) voters who anticipate minimal winning connected coalitions, who recognise the formateur advantage of being the largest party in the legislature,
and are sensitive to ideological proximity should anticipate that CD will likely form a
governing coalition. In both Denmark and Germany, 30 percent of the respondents select this as the most likely outcome. But large numbers of the Danes choose ABC and
CDE (24 and 21 percent, respectively) while about 24 percent of the German respondents
select a coalition of the two largest, although non-proximate, parties BD. Note that in
Treatment 4A the parties are ideological equidistance from each other and straddling the
ideological centre.
In Treatment 4B (Figure 6), Parties C and D move far apart from each other on
the ideological spectrum. The Danish response recalls the heuristics identified in the
treatments 3A and 3B. First, the formateur heuristics is clearly prevalent in the Danish
coalition choices in Treatment 4B: Respondents do not anticipate an ABC coalition which
is ideologically proximate and minimal winning connected. Rather over 60 percent of the
respondents anticipate the formation of a CDE coalition that includes the largest party.
But the inclusion of party E, which is not necessary for a majority coalition, suggests,
again, this heuristic of including parties for ideological re-balancing which we saw in
Treatments 3A and 3B. Interestingly, the German response is in some fashion just the
opposite to that of the Danes: In Treatment 4B, the modal choice is ABC which is
ideologically proximate and minimal winning connected (in terms of parties and seats)
but does not include the party with the largest number of seats. Unlike the Danes, the
Germans abandoned their expectation of formateur influence on coalition formation and
embraced the importance of ideological proximity and minimum winningness. And there
is no evidence in the German case of the ideological balance heuristic.
19
Figure 5: Treatment 4A
Figure 6: Treatment 4B
Treat 4A
Germany
Denmark
Treat 4B
Germany
Denmark
ABC
B
BCD
BD
CD
CDE
D
DE
N
6.09
24.06
2.64
1.01
3.93
4.32
24.42
3.31
30.02
32.13
7.31
21.33
12.78
5.62
3.61
1.44
4021
694
24.57
14.7
3.01
3.75
1.84
1.3
18.85
2.31
13.95
1.3
5.07
60.52
11.32
3.17
10.1
0.43
4021
694
20
5
Distribution of Heuristics in the Electorate
All told the administration of these experimental vignettes resulted in about 38,000
coalition formation predictions (the combined sample for the two countries was close
to 4,700). The analysis of these decisions resulted in the identification of five distinct
coalition reasoning heuristics:
1. Majority Heuristic. Subjects that consistently choose a group of parties that have
a majority of seats in the legislature are considered to employ a majority heuristic.
2. Formateur Heuristic. Subjects who consistently include the party with the largest
number of seats in the legislature.
3. Ideological Proximity Heuristic. One of the experimental treatments involved significantly increasing the ideological distance between potential coalition partners.
Subjects who responded by shifting to parties that were more ideologically proximate employ an ideological proximity heuristic.
4. Connected Heuristic. Those subjects who select parties that are ideologically connected, i.e., are immediately proximate on the left-right ideological continuum (parties A and B would be connected while A and C would not).
5. Minimal Winning Connected Coalition Heuristic. Subjects who consistently selected the minimal winning connected coalition outcome.
We find little evidence that respondents employed an incumbent party heuristic in their
coalition reasoning.
Voters are unlikely to deploy all of these heuristics when making a vote choice. More
likely they will rely on one, or some reduced set, of such coalition formation heuristics –
or quite likely none. Our goal is to identify the heuristics that are most widely employed
in the population. In order to do this we identify heuristics based on whether they are
21
employed consistently across coalition scenarios by sizeable numbers of the population.
We assume that relying on a specific heuristic consistently across coalition scenarios
indicates that the individual accords considerable weight to this feature of the coalition
formation process. This is a conservative approach since one can imagine that individuals
could accord considerable importance to one of these features while at the same time
ignoring it in certain coalition scenarios.
The incidence of the five coalition heuristics, recovered from the Danish and German
experiments described earlier (and based on this consistency criterion), is summarised in
Figure 7. These are within-subject analyses so they indicate the percentage of respondents
who consistently employ each of the five heuristics. The results confirm our contention
that the Danish electorate exhibits more sophisticated coalition reasoning than is the case
for the German electorate. About a quarter of the German sample consistently chose
a majority governing coalition; while over 60 percent of Danish respondents exhibited
such behaviour.5 Recall that respondents were instructed to identify a likely majority
governing coalition so those consistently selecting a majority coalition exhibit an ability
to sort out which combinations of parties constitute a majority of seats.
The largest party, which is typically the formateur party in most parliamentary systems, is rarely excluded from the governing coalition. In their sample of coalition governments spanning the period 1945-1987 Laver and Schofield (1990) find that 83 percent
of the coalition governments formed included the largest party. In their Danish and German cases the frequency was somewhat lower – about 60 percent in both cases (Laver
and Schofield 1990). Our experimental vignettes had about 25 percent of the Danes con5
Note that in calculating the Danish figures we excluded from the calculation Treat-
ment 3b because for reasons unclear to us there was a dramatic drop in the number of
respondents selecting a majority coalition. There was no similar outlier in the German
case.
22
sistently choosing a coalition that included the largest party while this was the case for
less than 10 percent of the German respondents. Certainly in the Danish context there
seems to be a sizeable percentage of the voting population that recognises the importance
of party size and formateur status in shaping coalitions.
Ideological distance between parties – which essentially captures the extent to which
parties are ideologically “adjacent” – affects the likelihood that parties will agree to form
a governing coalition. The early work of deSwaan (1973), and many others subsequently,
established the importance of ideological compactness. Our ideological treatment described earlier is designed to recover such a heuristic. Approximately 30 percent of the
German sample responded consistently to the ideological treatment by modifying their
predicted coalition outcome while almost 45 percent of the Danes responded in such a
fashion.
In their study of multiparty governments, Laver and Schofield (1990) find that about
40 percent of the coalitions formed were minimal winning. Coalitions also tend to include parties that are “connected” in the sense that they are adjacent to each other on
the Left-Right ideological dimension. As a result we tend to see, as Axelrod’s theory suggests (Axelrod 1970), minimal connected winning coalitions. In the Laver and Schofield
(1990) sample of coalition governments, 72 percent of minimal winning coalitions are
connected coalitions. Both the minimal winning and connected heuristics are prevalent
in the coalition choices made by our Danish and German respondents. While just under
20 percent of the German respondents consistently selected a connected governing coalition, over 60 percent of the Danish sample expected such a coalition type to form. A
relatively more modest percentage of respondents consistently selected a minimal winning
connected coalition as a likely majority governing coalition – 13 percent in Germany and
17 percent in Denmark.
23
In this analysis we adopt a conservative threshold for assessing the incidence of a coalition reasoning heuristic in the population – it has to be used consistently by respondents
across all of the four scenarios. In spite of this high threshold we find strong support for
the presence of coalition reasoning heuristics in the Danish and German populations. As
many as 60 percent of the Danish sample employ, again consistently, at least one of our
0
Percentage respondents
20
40
60
heuristics while this is the case for 30 percent of the German sample.
Majority
Connected
Formateur
Denmark
Ideology
Minimal
Germany
Figure 7: Distribution of Coalition Heuristics
Figure 7 confirms that reasonably large numbers of voters develop heuristics for reasoning about coalition formation. The evidence suggests considerable diversity in the
particular heuristics any individual voter deploys. It is not the case that the population is divided into those who deploy most, if not all, of these heuristics, when assessing
coalition formation possibilities, versus those who employ few if any, i.e., the sophisticates versus the non-sophisticates. In fact, in most cases the correlations are modest in
24
size.6 The voter population is heterogeneous in its use of coalition heuristics – different
heuristics are favoured systematically by groups of voters.
Figure 8 summarises the nature of the heterogeneity described in the previous section. It compares the frequency with which Danish and German respondents consistently
employ different combinations of these heuristics. Approximately 60 percent of Danish
respondents consistently selected a majority coalition compared to about 25 percent of
the Germans. About 60 percent of the Danes and 18 percent of Germans consistently
selected a majority connected coalition. The third category are those who consistently
anticipated a majority connected coalition that included the largest party: 20 percent of
the Danes and 8 percent of the Germans fell in this category. Finally, about 20 percent
of the Danes and 5 percent of the Germans consistently predicted a majority connected
coalition that included the formateur party and preferred a coalition with ideologically
proximate parties. On balance, in both countries, there is considerable heterogeneity in
the mix of heuristics employed for assessing the likelihood of different coalition formations. It appears though from Figure 8 that the Danes exhibit higher levels of coalition
reasoning than the Germans.7
6
7
The correlations vary between 0.0 and 0.83 and are available in Online Appendix.
Controlling for education does not eliminate the German and Danish differences in
the sophistication of coalition reasoning heuristics. The differences in the low education
remain very pronounced. And while the differences are lower in the high education
category, the country differences remain quite significant.
25
Percentage respondents
.6
.4
.2
Denmark
Germany
0
M
M+C
M+C+F
M+C+F+I
M:Majority C:Connected F:Formatteur I:Ideology
Figure 8: Distribution of Coalition Reasoning Types
26
These differences in frequencies of coalition reasoning between Germany and Denmark
speak to our second empirical conjecture that coalition reasoning heuristics are ecologically rational. We have only two national cases here so these results offer no proper
empirical test of the conjecture. Nevertheless, the results are suggestive. First, they
suggest that this measurement strategy may be a useful one for calibrating national differences in coalition reasoning apptitudes. This is suggested by the simple fact that the
heuristics employed consistently by respondents across the vignettes map to those that
actually shape the party composition of coalitions. Many voters seem to have mental
maps of coalition formation dynamics that are entirely consistent with what actually occurs in the coalition formation process. Moreover, across all of the different heuristics –
or combinations of heuristics – Denmark consistently ranks higher than Germany on our
measures of coalition reasoning.
One possible explanation for why Denmark ranks higher than Germany in coalition
reasoning is ecological rationality. Ecological rationality suggests that the differences
result because the Danish context is more complex; voters get more information about
this complexity; the complex information is useful for exercising a coalition directed vote;
and hence voters develop these coalition reasoning skills. With only two contexts to
compare here we present this explanation as speculative although certainly worth more
careful examination with larger numbers of national contexts.
6
The Coalition-Directed Vote
Voters develop coalition reasoning heuristics described above because they allow them
to exercise a coalition directed vote. But as the previous section suggests, not all voters
exhibit these coalition reasoning heuristics. Our conjecture is that voters with underdeveloped coalition reasoning heuristics, regardless of context, are less likely to approximate
27
the informed coalition directed vote that occurs, as discussed earlier, in contexts with
coalition governments (Duch, May and Armstrong 2010, Kedar 2009).
We test this argument by comparing the coalition directed vote of respondents we classify as having well-, versus, under-developed coalition reasoning heuristics. A coalitiondirected vote involves attributing responsibility to particular coalition parties for the
governing coalition’s performance. It should reflect the role a party plays in shaping
policy outcomes (“administrative responsibility”) and whether the party is in contention
for significant governing responsibility. In exercising a coalition-directed vote, voters are
expected to recognise how voting for different parties affects the likely composition of
a post-election coalition government. Voters with poorly developed coalition reasoning
heuristics are less likely to exercise a coalition-directed vote because they have difficulty
assessing a party’s administrative responsibility or anticipating the kinds of coalitions
that are likely to form after an election. Moreover, these coalition reasoning skills will
be more important in some elections versus others (Duch and Stevenson 2008); depending on the characteristics of the incumbent governing coalition and the likely alternative
coalitions that could form after the election.
Recall that the online surveys were conducted just prior to the 2011 Danish and
2009 German elections. As it turns out the characteristics of the incumbent coalitions
going into these two elections and the likely coalitions to form after the elections posed
quite different coalition reasoning challenges for the average voter. The 2009 German
election occurred after four years of a Grand Coalition consisting of the CDU/CSU and
SPD. The incumbent coalition consisted of these two major parties in which cabinet
portfolios were divided equally amongst the two parties (roughly eight ministries for
each party). This deviates from most voting situations because the two major parties,
that would most certainly dominate the post-election coalition government, were both
effectively “incumbent” parties. Classic responsibility attribution models of vote choice
28
presume, at a minimum, well-defined incumbent and opposition parties that will either
be rewarded or punished as a function of perceived governmental performance. A Grand
Coalition does not provide such a clear dichotomy and hence the ability to reward or
punish performance via party vote choice is muted in such contexts (Anderson 1995,
Powell and Whitten 1993).
In the case of the informed coalition directed vote, if one of the two parties in the
Grand Coalition is expected to govern on its own after the election, the equal distribution
of responsibility in the incumbent cabinet would lead to neither party being rewarded
or punished for government performance. Neither party is “in contention” to either reelect or replace the incumbent distribution of administrative responsibility. Hence, an
informed coalition directed vote for most parties in the German case is likely to be weakly
correlated with government performance. Voters without these coalition reasoning skills
would ignore information about administrative responsibility within the Grand Coalition.
The 2011 Danish election presented a much more familiar situation for the performance voter. The Danish incumbent minority coalition government was lead by the
Liberal Party (VE) and included the Conservative People’s Party (K). The minority
government was supported by the Danish People’s Party. And it was widely expected
that the Social Democrats were in a strong position to form a new coalition government.
In the case of performance voting by the Danes, we expect the Liberal Party (VE) to
be most rewarded (or punished) for governmental performance; the Conservative People
Party (K) less so; and the most likely opposition party to form the government, Social
Democrats (SD), should be rewarded for negative perceptions of governmental performance. The demands on coalition reasoning in this case are reasonable – the incumbent
and opposition coalitions were both obvious and distinct. Accordingly, we have no reason
to believe that coalition directed voting in this election should differ for those exhibiting
either developed or under-developed coalition reasoning heuristics.
29
Our expectations then are as follows: 1) sophistication of coalition reasoning should
have little impact on the magnitude of the performance vote in the 2011 Danish election
– the vote of both developed and under-developed coalition reasoning types should be
correlated with perceived governmental performance; 2) sophistication of coalition reasoning skills will matter in the 2009 German election – for those with developed coalition
reasoning heuristics we expect a weak correlation between perceived governmental performance and vote choice while the party vote of those with under-developed heuristics
will be correlated with perceived governmental performance. One naive heuristic might
be to punish or reward the PM party (CDU/CSU) for government performance.8
We test these hypotheses by estimating a classic vote choice model for the German
and Danish respondents. And in each country we estimate separate vote choice models
for those identified as low versus high coalition reasoning types in the previous section.9
High coalition reasoning types are defined as respondents who were classified as any one
of the three coalition types: formateur, ideology or connected types. Low reasoning types
were those who were not classified in any one of these reasoning categories. The detailed
results from the multinomial logit model are presented in the Online Appendix. Performance voting is measured by the correlation between evaluations of national economic
performance and vote choice.
8
Punishing and rewarding the PM party in fact may be a perfectly reasonable heuristic
in many coalition contexts – but there clearly are some in which it is not (Duch and
Stevenson 2008). The case in which the incumbent party is a Grand Coalition is clearly
one of these exceptions.
9
The independent variables in the Danish multinomial logistic model were: subjective
evaluation of the economy; left-right self-identification; education; and gender. Similarly the independent variables in the German multinomial logit model were: subjective
evaluation of the economy; left-right self-identification; education; gender; age; union
membership; evaluation of health care provision; evalution of crime levels in the country.
30
Figure 7 presents the simluated effects of the subjective economic evaluation variable
along with simulated effects of the left-right self identification.10 The German results
are very much consistent with our conjecture. Subjective economic evaluations are not
correlated with vote choice for the high coalition reasoning types and they are correlated
with CDU/CSU vote choice for the low coalition reasoning types.
For the most part, German voters with relatively sophisticated coalition reasoning
are not punishing or rewarding either of the Grand Coalition parties for their perceptions
of performance. This is consistent with theories of coalition directed voting. On the
other hand, those with less developed coalition reasoning exercise a performance vote
that is directed at the CDU/CSU which was the PM party in the Grand Coalition. As
a comparison, note that for both coalition reasoning types left-right self identification is
strongly correlated with vote choice.
The Danish results are also consistent with our conjectures. This is a context in which
we would expect conventional performance voting for the incumbent coalition parties and
it should not differ based on coalition reason skills. The economic evaluations of both the
high- and low-coalition reasoning types are significantly correlated with the vote for the
Liberal Party that headed the incumbent governing coalition. The economic evaluations
of neither type are correlated with their vote for the junior partner in the coalition. And
finally the economic evaluations of both types are significantly correlated in the expected
direction with their vote for the Social Democrats who were the leading opposition party
in the election. Hence there is a performance vote in this context and it is statistically
significant for both coalition reasoning types.
10
We use Clarify to estimate the simulated effects and their standard errors (Tomz,
Wittenberg and King 2003).
31
Coalition Reasoning: High
Coalition Reasoning: Low
●
CDU/CSU
●
●
●
Free Democratic Party
●
The Greens
●
●
●
The Left
●
●
●
●
CDU/CSU
●
Free Democratic Party
●
●
The Greens
●
●
The Left
Ideological Vote
Social Democratic Party
Economic Vote
Social Democratic Party
●
●
●
−0.2
0.0
0.2
0.4 −0.2
0.0
0.2
0.4
Simulated Effects
(a) Germany
Coalition Reasoning: High
Radical Left Party
●
●
Conservative People's Party
●
●
Socialist People's Party
●
●
●
Social Democrats
●
●
●
●
Radical Left Party
●
●
Conservative People's Party
●
Socialist People's Party
●
●
●
Liberal Party
●
−0.2
Ideological Vote
Liberal Party
Economic Vote
Social Democrats
Coalition Reasoning: Low
●
0.0
●
0.2
0.4 −0.2
Simulated Effects
(b) Denmark
Figure 9: The Coalition Directed Vote
0.0
0.2
0.4
32
The goal of this section is to demonstrate that individual-level models of vote choice
that ignore heterogeneity in coalition reasoning skills are under-specified. We demonstrate
this by comparing two election contexts in which the demands on the coaliton reasoning
skills of voters vary. In one case, Germany’s 2009 federal election, demands on coalition
reasoning are high because the two largest parties shared administrative responsibility in
the incumbent coalition. The exercise of a coalition directed vote in this election context
is more demanding because administrative responsiblity was effectively shared by the two
major parties competing in the election. The second case, Denmark’s 2011 election, is less
demanding in that it presented coalition-directed voters with a much more conventional
set of choices between the incumbent coalition parties lead by the Liberal Party and a
challenging coalition lead by the Social Democrat party that did not participate in the
incumbent coalition. If coalition reasoning skills in fact matter and our measurement
strategy is reasonable then the high and low coalition reasoning types in Germany should
exhibit some differences in voting behaviour while these differences are likely to be much
less evident in Denmark. And this is precisely what we see. German respondents, who we
identify as having high coalition reasoning skills, are more likely to exercise an informed
coalition vote than those who do not use this heuristic; while the high and low coalition
reasoning types in Denmark behave similarly.
A critical part of our argument regarding coalitioning reasoning and the coalition
directed vote is that political context matters. The complexity of coalition formation
dynamics varies not simply cross-nationally but also from one election to the next. When
the coalition context is quite simple – for example when distributions of administrative
responsibility make it easy for voters to punish and reward incumbents – the coalition
reasoning skills we identify here will not condition responsibility attribution in individuallevel vote choice models. On the other hand, as the coalition context becomes more
complex then exercising a coalition directed vote that incorporates responsiblity attribu-
33
tion becomes more demanding. In this case, responsibility attribution is conditioned on
coalition reasoning skills and excluding this specification from the vote utility function
will result in an underspecified model.
7
Conclusion
This essay uses experimental vignettes for recovering the heuristics that voters employ
in order to anticipate coalitions that are likely to form as a result of an election outcome. A
large sample of respondents in internet panel surveys conducted in Denmark and Germany
were asked to identify likely coalition formations given hypothetical election outcomes
that result in different hypothetical seat allocations to parties that are located along
different positions on the left-right ideological continuum.
We find that high proportions of the Danish and German electorate employ identifiable
coalition formation heuristics; the Danes exhibit higher levels than is the case for the
Germans. A very basic heuristic is the ability to understand the notion of a majority
coalition: In Germany about one-quarter of the respondents consistently choose majority
coalitions in the experimental vignettes while over 60 percent of the Danish respondent
exhibit such consistent behaviour. Formateur status consistently informed the choices of
about 25 percent of Danes and less than 10 percent of Germans. Ideological proximity
was a consistent heuristic for over 40 percent of the Danes and for about 30 percent of the
Germans. Just under 40 percent of German respondents, but over 60 percent of Danish
respondents, consistently selected a connected coalition. Surprisingly, the incumbent
status of parties does not seem to be an heuristic employed in anticipating coalition
formations.
Ecological rationality is a possible explanation for why Denmark ranks higher than
Germany in coalition reasoning. The complexity of coalition formation dynamics in
Denmark suggests that Danish voters get information, for example from the media, that
34
informs them about this complexity; the complex information is useful for exercising a
coalition directed vote in Denmark; and hence Danish voters develop more sophisticated
coalition reasoning skills than voters, for example, in Germany where coalition formation
dynamics have been less complex. With only two contexts to compare here we present
this explanation as speculative although certainly worth more careful examination with
larger numbers of national contexts.
We find that this variation in coalition reasoning within the population is correlated
with the extent to which voters exercise a coalition directed vote. Respondents who we
identify as having high levels of coalition reasoning are more likely to exercise an informed
coalition directed vote. This suggests that some voters employ these coalition reasoning
heuristics in order to facilitate informed coalition voting in contexts in which multi-party
coalitions typically govern.
This heterogeneity in coalition reasoning heuristics – both cross-sectionally within
countries but also cross-nationally – suggests that the vote utility function will vary
across political contexts and also across different types within the same political context.
A voter in Denmark is more likely than one in Germany to incorporate into her vote
utility calculations a rich set of factors that predict coalition outcomes after an election.
Nevertheless, in both Denmark and Germany there is heterogeneity – some voters have
the coalition reasoning aptitude to exercise a coalition directed vote while others do not.
Ignoring the fact that some voters are likely exercising a coalition directed vote (those
with well developed coalition reasoning heuristics) while others are most likely not (they
lack these heuristics) can only result in misspecified models of vote choice.
35
References
Adams, James, Benjamin G. Bishin and Jay K. Dow. 2004. “Representation in Congressional Campaigns: Evidence for Discounting/Directional Voting in U.S. Senate
Elections.” Journal of Politics 66:348–73.
Alesina, Alberto and Howard Rosenthal. 1995. Partisan Politics, Divided Government
and the Economy. Cambridge: Cambridge University Press.
Anderson, Christopher J. 1995. Blaming the Government: Citizens and the Economy in
Five European Democracies. Sharpe.
Ansolabehere, Steven, James Snyder Aaron Strauss and Michael Ting. 2005. “Voting
Weights and Formateur Advantage in the Formation of Coalition Governments.” American Journal of Political Science 49(3):550–563.
Austen-Smith, David and Jeffrey Banks. 1988. “Elections, Coalitions and Legislative
Outcomes.” American Political Science Review 82:405–22.
Axelrod, Robert. 1970. Conflict of Interest. Chicago: Markham.
Bargsted, Matias and Orit Kedar. 2009. “Coalition-Targeted Duvergerian Voting: How
Expectations Affect Voter Choice under Proportional Representation.” American Journal of Political Science 53:307–23.
Blais, Andre, Peter John Loewen and Marc-Andre Bodet. 2004. Voters’ Veto: How New
Zealand Voted in 2002. Aukland University Press chapter Strategic Voting, pp. 68–84.
Bowler, Shaun, Jeff Karp and Todd Donovan. 2010. “Strategic Coalition Voting: Evidence
from New Zealand.” Electoral Studies 3.
deSwaan, Abram. 1973. Coalition Theories and Cabinet Formation. Amsterdam: Elsevier.
Downs, Anthony. 1957. An Economic Theory of Democracy. New York: Harper and
Row.
Duch, Raymond M., Jeff May and David Armstrong. 2010. “Coalition-Directed Voting
in Multi-Party Democracies.” American Political Science Review 104(4):698–719.
Duch, Raymond M. and Randy Stevenson. 2008. The Economic Vote: How Political and
Economic Institutions Condition Election Results. Cambridge: Cambridge University
Press.
Gigerenzer, Gerd and Peter M. Todd. 1999. Simple Heuristics that make us Smart. Oxford
University Press chapter Fast and Frugal Heuristics: The Adaptive Toolbox, pp. 3–34.
Goldstein, Daniel G. and Gerd Gigerenzer. 2002. “Models of Ecological Rationality: The
Recognition Heuristic.” Psychological Review 109(1):75–90.
36
Gschwend, Thomas. 2007. “Ticket-Splitting and Strategic Voting under Mixed Electoral
Rules: Evidence from Germany.” European Journal of Political Research 46:1–23.
Indridason, Indridi H. 2011. “Proportional Representation, Majoritarian Legislatures,
and Coalitional Voting.” American Journal of Political Science 55(4).
Kahneman, Daniel. 2011. Thinking, Fast and Slow. Farrar, Straus and Giroux.
Kedar, Orit. 2005. “When Moderate Voters Prefer Exterme Parties: Policy Balancing in
Parliamentary Elections.” American Political Science Review 99(2):185–199.
Kedar, Orit. 2009. Voting for Policy, Not Parties: How Voters Compensate for Power
Sharing. Cambridge University Press.
Laver, Michael and Norman Schofield. 1990. Multiparty Government: The Politics of
Coalition in Europe. Oxford: Oxford University Press.
Martin, Lanny and Randy Stevenson. 2001. “Government Formation in Parliamentary
Democracies.” American Journal of Political Science 45:33–50.
Merrill, Samuel III and Bernard Groffman. 1999. A Unified Theory of Voting: Directional
and Proximity Spatial Models. Cambridge University Press.
Powell, G. Bingham and Guy Whitten. 1993. “A Cross-National Analysis of Economic
Voting: Taking Account of the Political Context.” American Journal of Political Science 37:391–414.
Riker, William H. 1962. Theory of Political Coalitions. Yale University Press.
Schofield, Norman and Michael Laver. 1985. “Bargain Theory and Portfolio Payoffs in
European Coalition Governments 1945-83.” British Journal of Political Science 15:143–
164.
Simon, Herbert. 1955. “A Behavioral Model of Rational Choice.” Quarterly Journal of
Economics 69:99–118.
Stromback, Jesper and Toril Aalberg. 2008. “Election News Coverage in Democratic
Corporatist Countries: A Comparative Study of Sweden and Norway.” Scandinavian
Political Studies 31:91–106.
Tomz, Michael, Jason Wittenberg and Gary King. 2003. CLARIFY: Software for Interpreting and Presenting Statistical Results, version 2.1. Stanford University, University
of Wisconsin, Harvard University.
Tomz, Michael and Robert P. Van Houweling. 2007. “The Microfoundations of Issue
Voting.” Manuscript, August 12, 2007.
Tversky, Amos and Daniel Kahneman. 1974. “Judgment under Uncertainty: Heuristics
and Biases.” Science 185:1124–1131.