DO NUMBERS
COUNT?
Political Science:
Parties, Parliaments and Democracy
A quantitative research on the relationship
between the number of parties in
coalition and political party placement on
the left-right dimension by the
electorate
SANNE TOEBES
S2692295
S.A.F.TOEBES@LEIDENUNIV.NL
SUPERVISOR: DR. M. NAGTZAAM
SECOND READER: PROF. DR. P. KOPECKY
DATE: JUNE 15TH, 2020
Despite the obvious disadvantages of coalitions with many partners, apparent already in the formation process, electorates in established democracies have been witnessing coalitions with many partners throughout the years. In many studies on coalition governments however, coalitions are often treated as black boxes. In this quantitative study, I examine whether there is a linear relationship between the number of parties in coalition and the ability of the electorate to identify the party positions of the coalition partners. Through the use of the Comparative Study of Electoral Systems (CSES) and the ParlGov.org dataset, I created a dataset with 70 coalitions in twenty established democracies between 1996 and 2017. I used the standard deviation of the score on the left-right dimension assigned by the electorates to government parties as a proxy for disagreement among voters. Only a rather inconsistent relationship between the predictor variable and the dependent variable was found.
Sections
1. Introduction
3
2. Theoretical framework
5
3. Research Design
11
4. Results and discussion
16
5. Conclusion
22
Literature 26
2
1. Introduction
Since World War I, multiparty government has increasingly become the norm in most parliamentary democracies (Strøm et al., 2008). A study found that from 1945 until 1999, only thirteen per cent of all cabinets in seventeen parliamentary democracies were a single-majority government as opposed to 63 per cent multi-party majority coalitions (Saalfeld, 2008, p. 171). In recent years, we see that most parliamentary democratic governments are comprised of multiple parties (Klüver & Spoon, 2019, p. 2).
Where most coalitions are a formation of two or three parties, the democratic electorate has also been witnessing several coalitions comprised of four, five or even six parties, especially in new democracies (Niikawa, 2018). This stands in stark contrast with the common wisdom that coalition formation is less challenging with low ideological conflict and few coalition partners (Spoon & Klüver, 2017, p. 116). The difficult task of forming a coalition government can be lessened by creating a government with few partners (Warwick, 1996, p. 474). This is not to say however that coalitions with many parties do not regularly occur in more ‘experienced democracies’. Although having coalition experience since 1945, the Netherlands witnessed four parties taking office as recent as 2017. Finland has welcomed five parties into government at the end of 2019. In recent years New Zealand has been governed by one big party that has been needing the support from five smaller parties.
Despite disadvantages already apparent during the formation process, big coalitions do occur in established democracies. The objective of this study is measuring the effect those coalitions have on voters’ perceptions of coalition partners’ policy positions. Much research has already been done on party coalitions in proportional systems. What however stands out in research on coalition governments, as also mentioned by Sagarzazu and Klüver (2017, p. 346) is that they are often treated as black boxes, however their internal characteristics may differ
3 (p. 2). There are studies that have tracked the ability of voters to assess what coalition membership does to the policy positions of a party (Fortunato & Stevenson, 2013; Ganghof & Bräuninger, 2006; Gerber et al., 2015), and studies that focused on the perceived policy position shift of parties concerning a dimension other than the left-right ideological scale (Adams et al., 2016).
One neglected factor that could be of influence on voters’ perceptions of party positions, are the numeric differences between coalitions. Coalition parties in general have more difficulties marketing their policy positions to voters than single-party or opposition parties (Spoon & Klüver, 2017, p. 126). Coalition parties experience a significant trade-off between keeping their policy promises and effective government (Van de Wardt et al., 2014, p. 997). While during elections parties try to differentiate themselves from the competing parties, they often have to speak with one voice while in coalition (Sagarzazu & Klüver, 201 p. 346). With more parties in coalition, there are more different ideological voices that have to merge into one. Following this argument, there is reason to believe that it is more difficult for a party that is in coalition with, for example, four other parties, to differentiate herself than for a party that is in coalition with only one other party. I will therefore assess whether the number of parties in coalition influences the ability of the electorate to place the coalition parties on the left-right dimension.
A comparison among different established democracies is needed in order to assess whether the number of parties in coalition matters for the ability of voters to identify the coalition parties’ positions. The question that will be addressed in this paper is: “to what extent
does the number of parties in coalition influence the degree to which the electorate is able to identify government parties’ policy positions?” In this paper I examine both the possible
4
Relevance
An answer to the research question could have implications for the way in which parties value majority cabinets. Often, the reason for a relative high number of parties in coalition, is to have a majority of seats in parliament. If, however, with the number of parties in coalition increasing, the electorate is less able to identify party positions, ultimately, it might not be the best option for parties to participate in such big coalitions. Party profile identification namely touches upon essential democratic processes, Spoon & Klüver (2017) argue (p. 115). For linkage and party competition to work, they emphasize that parties need to be clear in where they stand for, and the electorate needs to be able to identify these positions. As is shown by the same authors, voters have already more difficulty identifying the party positions of parties in coalitions (p. 115), and they have possibly even more difficulty when the number of parties in coalition is relatively high.
Party programs have already become more alike, Katz and Mair (1995) argue, and in such the authors have expressed criticism on the ‘toning down of competition’ in proportional democracies. Parties have to offer voters clear political choices, as otherwise voters are unable to cast a politically informed vote (Key, 1966, p. 2). When voters have difficulty identifying what parties stand for, they have more difficulty with assessing which party is close to his or her policy positions. This can induce indifference from the side of the electorate, and as a result they might abstain from voting on the whole (Caul & Gray, 2002, p. 237). When the number of parties in coalition causes the electorate to be less concerned about politics, the importance ascribed to majority cabinets might be re-evaluated. To convince them of voting for their proposals, coalition parties in minority cabinets need to make their positions abundantly clear to the opposition parties and the electorate as well. Minority cabinets have been called a solution to indifference before (Van der Meer, 2017), and this research could instigate further discussions about it.
5 Conversely, this research might indicate a positive relationship between the number of parties in coalition and how clear the party profiles are for the electorate. It could be the case that with the number of parties in a coalition increasing, the better parties try to market their differences. A mechanism noted by Strøm & Müller (2009) on party systems (p. 44), that could be at play within coalitions as well. This could justify parties’ consideration for joining a coalition with a relative high number of parties. It however depends on what the numbers tell us.
Outline of the paper
In order to answer the research question, I will set out the theoretical foundation on which my research is based in the next section. I will assess in what way voters identify party policy positions, what makes for a clear, and distinct party profile, and I will introduce the hypotheses. In the research design I discuss the concepts, methodology, the different datasets that are used and why certain cases are either included or excluded from the sample. Also, the time frame and the reason for choosing certain variables are addressed. The results are shown and discussed in section four. I will state my conclusions in section five, where I also will discuss the limitations of the research, and considerations for future research.
2. Theoretical framework
The importance of clear and distinct party profiles
G. Bingham Powell (1982) has argued that the identifying property of the contemporary democratic process are the several political parties organizing the alternatives that face the voters (p. 3). The foundational structure of the competitive model is that the party one votes for “will somehow be different from and preferable to the competition” (Caul & Gray, 2002, p. 218). The ‘Downsian’ concept of spatial modelling has opened up the way in which we perceive
6 party competition. Parties and voters are in this model aligned along a left-right continuum. Positioning themselves and parties on the left-right continuum, voters will choose a party that is closest to their own position (Dalton, 2008, p. 901). If the nearest party is too far away from the voter, the voter might abstain from voting on the whole. Also, when two certain parties are at the same perceived distance from the voter, this is likely to produce indifference. Voting then might seem unnecessary and could result in nonvoting as well.
Who is ultimately responsible for voters’ indifference to politics, the voter or the parties, is heavily debated. Peter Mair (2006) claims that many democracies in Western Europe have witnessed a process of ‘mutual withdrawal’ (p. 33). Where parties have withdrawn from society into institutions, voters increasingly choose private forms of representation over political party representation (p. 33). Mair seems to hold voters accountable for being increasingly indifferent to politics, using the rising levels of volatility as a proxy for indifference (p. 38).
As I measure the perceived party positions of the electorate in general, I focus on the importance of marketing a clear and distinct party profile identifiable to a broad range of voters. Many scholars have drawn attention in their research on the ‘misperceptions’ of voters (Klüver & Spoon, 2019; Spoon & Klüver, 2017; Dahlberg, 2013). Comparing scores between voters and experts seems to imply that when voters would invest more time and thought into it, they would be able to perceive the ‘real’ policy positions of parties. I however focus more on the responsibility parties have in marketing a clear and distinct party profiles in order for all voters to be able to perceive differences between those parties. Along this line of argument, is Key’s notion (1966) that the electorate serves as an ‘echo chamber’ that reflects the alternatives that political parties pose (p. 2). As parties define the political options from which voters can choose from, the differences in party positions provide a standard against which voters can locate their own preferred position (Gerber et al., p. 147). For voters to be able to do so, Caul and Gray
7 (2002) argue that clear, distinct and consistent partisan profiles are essential to structuring voters’ choices (p. 236).
Political parties find it increasingly difficult, however, to maintain distinct policy profiles, and this difficulty is exposed at the level of the voter (Caul and Gray, 2002, p. 235). Already in 1995, Katz and Mair expressed criticism on the ‘toning down of competition’ within proportional democracies (p. 23). They argued that party programs have increasingly become more alike (ibid.). If parties do not offer voters clear political choices, those voters “could not possibly cast a politically informed vote” (Key, 1966, p. 2). An ultimate consequence of unclear differences between parties, might be that voters are less likely to see much relevance in going out to vote (Caul & Gray, 2002, p. 237).
The electorate
Where some voters will be very competent in identifying the policy positions of parties, other voters have more difficulty doing so. Educational levels are said to be influencing the ability of voters to place parties (Lesschaeve, 2017, p. 358). In general, however, Costello et al. (2012) argue that voters have no problem placing themselves on the left-right dimension, and they also “have a clear perception of where the main political parties stand” (p. 1229). Additionally, Blais & Bodet show that there is a strong linkage between the mean citizen placement and expert placement of a party (2006, p. 1249). Voters have more difficulty with placing coalition partners, however. This is already shown by Adams et al. (2016), as they argue that voters are less able to make sense of more nuanced information, such as parliamentary debates, politicians’ speeches and interviews, and government policy outputs (p. 811).
Gerber et al. (2015) have shown that parties can make sure by adopting a clear position on an issue that also less informed voters won’t miss its position. Both competent and noncompetent voters will then be able to use that issue as a criterion, they argue (p. 161). The
8 question that will be answered in this study concerning government parties is whether the number of government parties influences the success those parties have in marketing their policy profile.
Party profile
In developing a theoretical framework around the research question, a definition of a distinct party profile has to be established. In this section I will address the factors that contribute to a clear and distinct party profile, while also introducing my two hypotheses.
Gerber et al. (2015) base their definition of the clarity of party profile on two elements: first, the distinctiveness of a party’s position on a given issue, and second, the cohesiveness of the party’s position (p. 148). The first element depends to a large extent on the party system. Majoritarian systems are more likely to result in party competition around the median voter (Strøm & Müller, 2009, p. 44). The two competing parties in a majoritarian system will therefore often adopt centrist positions. Proportional systems, however, also allow space for more extreme parties. The more parties there are, Sartori (1976) has argued, the more the parties use the full range of the left-right dimension (p. 128). Although some of the parties in proportional systems could gain from moderation, Strøm & Müller argue, “the very rationale is to take distinctive policy positions” (p. 44).
In other words, because in proportional systems there is no “winner takes all” mechanism at play, parties are allowed to be more extreme. With only a relative low percentage of votes, parties in proportional systems are still able to participate in government. In short, as argued by Strøm & Müller (2009) parties in proportional systems are more likely to have a clear and distinct party profile.
Among proportional systems, there are important differences in how parties are identified. Coalition partners by themselves are perceived as more ideologically similar than
9 non-coalition partners, Fortunato and Stevenson (2013) have shown. Within coalitions, Adams et al. (2016) found that in the context of positions on the European Union, junior coalition partners are perceived as shifting their policies in the same direction as the prime-minister’s party is perceived to be (p. 811). As governments are held responsible by voters for ‘getting things done’ (Ganghof & Bräuninger, 2006, p. 525), the two pillars of coalition governments, collective agency and shared responsibility, blur the distinction between the coalition partners.
Within a coalition, parties have thus in general more difficulty marketing a distinctive profile as they have to prioritize government effectiveness over their own policy programmes (De Wardt et al., 2014, p. 997). Part of the explanation for this difficulty is that there must be some dissent between political parties, for voters to base their party choice on (Gerber et al., 2015). Voters will have difficulty positioning parties when there is consensus among them (p. 148). For voters, consent might be fuelling the idea that most party programs are alike.
In a coalition with a relatively high number of parties, the likelihood of dissent between parties is higher. More parties mean that a higher range of the left-right dimension is used. As the left-right dimension represents ideological differences, more parties that are part of the coalition could mean more ideological differences and thus dissent. Along this line of argument, more parties mean more distinctive party profiles. The first hypothesis then is:
I. The higher the number of parties in coalition, the better the electorate is able to identify party policy positions on the left-right dimension.
The second factor that makes for a clear and distinct party profile, is party cohesion. Gerber, et al. (2015) are convinced that to offer clear alternatives, the internal unity of the party must be defended (p. 148). When politicians from a certain party go against the party line, party cohesion will be lower. It increases uncertainty for voters when they receive contradictory
10 signals from the party. Therefore, when party cohesion is low, it is difficult for voters to identify the party profile (Caul & Gray, 2002, p. 235).
With regard to the number of parties in a party system, Strøm & Müller (2009) expect that proportional systems tend to generate more cohesive parties than majoritarian systems. The greater the number of parties, they argue, the smaller the range of preferences within each party as politicians have a better chance to find a party that is close to their preferences (p. 35).
But, simultaneously, Strøm and Müller argue that maintaining party cohesion is more difficult for parties in coalition, as they are forced to make concessions to the coalition partners (p. 40). This is what Strøm and Müller call the twin tasks of coalition members: building inter-party agreement and maintaining intra-inter-party consensus (p. 40). The bigger the concessions are, the harder it is to maintain intra-party consensus. Saalfeld (2008) underlines the established norm of coalition partners of not voting against each other, which makes maintaining party cohesion more difficult for coalition parties (p. 173).
The difficulty of maintaining party cohesion in coalitions, hinders the ability of voters to identify party profiles. It infringes upon the distinctiveness of a party, as coalition parties are often perceived as converging on the left-right ideological dimension (Ganghof & Bräuninger, 2006). Caul and Gray (2002) have questioned whether parties really provide valuable information to voters if they are not consistent from one election to the next (p. 236). It touches upon party’s responsibility says Downs (1957), as a party “is responsible if its policies in one period are consistent with its actions in the preceding period” (in Mair, 2009, p. 11). Parties in coalition, cannot guarantee this consistency. In coalitions, consensus is in general prioritized to avoid inter party-conflict (Klüver & Spoon, 2019, p. 5). Therefore, coalition parties often have to make decisions that they would otherwise not have made. The inconsistency caused by such compromises, makes it more difficult for voters to identify the party profiles. Thus, there is a likelihood of a negative linear relationship between the number of government parties and a
11 clear, distinct party profile. The mechanism that serves as the basis for the second hypothesis is that the more coalition partners there are, the more those parties have to compromise, rather than market their own policy positions. Therefore, the second hypothesis is:
II. The higher the number of parties in coalition, the more difficult it is for the electorate to identify party policy positions on the left-right ideological dimension.
Research design
Sampling
The case-selection is based on established democracies. This is important because the commonality with coalitions influences the way in which political parties compete. Although in relatively new democracies, big coalitions are not uncommon (Niikawa, 2018), parties in new democracies might compete differently in general from established coalition governments and this will hamper drawing conclusions from the data. Countries that were part of the third wave of democratization were relatively new democracies when the datasets I used took their first surveys. As such, politicians in those countries were new to bargaining and compromising (Druckman & Roberts, 2005, p. 541). Also, because cleavage structures have not had the time to evolve yet in society, it is likely that the electorate in those countries were less able to identify their parties’ policy positions. Therefore, I only use countries that are considered to be established democracies already in the late 1990s.
To look into the electorates’ assessment of political competition, this research will make use of The Comparative Study of Electoral Systems (CSES) (Modules 1-5), which has asked respondents from many Western democracies to place, for them relevant, political parties on the left-right dimension. Which parties were relevant in a certain year, was determined by
12 country experts1. Each party in any survey year was assigned a left-right score by at least 1,000
respondents. Left-right scores were not always available for every country in the CSES dataset. For Italy, I used data on party placement from the Itanes (2006) dataset, that issued the left-right dimension as well. Ultimately, I ended up with as much as 20 countries.
The countries that will be part of the research are Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Great Britain, Iceland, Ireland, Italy, Luxembourg, the Netherlands, New Zealand, Norway, Portugal, Spain, and Sweden. With its first-past-the-post system, Great Britain is not known for its coalition governments. However, I have chosen to include the UK and also Canada and Spain for example so I can use one-party governments as a reference category for the coalition governments.
Some difficulties while working with the CSES dataset appeared at the surface. Belgium, for example, is only represented in the 1999 survey data of CSES and lacks comparable data from other datasets that cover more than one year. Also, as a result of its complicated federal structure, Belgium is in the CSES 1999 dataset divided in a separate Walloon and Flanders survey. Some of the parties in the Belgian government have a Walloon and Flemish division. In the survey, the respondents only answer the questions that concern the party active in their region and therefore, the different divisions are counted as one in this research.
1Before conducting the research, I considered adding left-right scores assigned to parties by country experts,
and then measure the difference of those scores from before and after a coalition was in place. Through this analysis I could assess whether the effect of increasing number of parties in coalition on policy positions was only perceived by the electorate or by country experts as well. I could not however find enough data to analyze these differences. Not in the CSES datasets and not in the Chapel Hill dataset (2020) was it possible to compare before and after coalition scores on the scale that this research is based on.
13
The dependent variable
The essence of this study boils down to whether the standard deviation of the scores assigned to parties on a left-right ideological dimension, decreases or increases when the number of parties in a coalition, decreases or increases. I took the standard deviation as a proxy for (dis)agreement among the electorate to where the parties stand for. The standard deviation represents the range of scores assigned to the coalition parties. A standard deviation close to zero shows that there is almost no disagreement among voters to where the party stands for. When the standard deviation increases as the number of parties in coalition increases, the electorate is thus less able to identify party positions when there are many coalition partners. The standard deviation of the electorates’ placement of government parties is the dependent variable in this research. For each party in a survey year, I have taken the mean score on the left-right dimension given by the respondents, and the standard deviation from that analysis serves as the dependent variable. I then took an ANOVA test to compare those means of standard deviations among the different coalition sizes.
In this research, only the left-right ideological dimension is considered for party placement. Empirical research seems to confirm the predominance of the left-right dimension. In most studies on party competition in Western European democracies, the playing field is reduced to this dimension (Costello et al., 2012, p. 1229). It is predominantly an economic dimension, where the left often stands for redistributive policies, and the right for a low-tax regime. Every country can however assign different meaning to it (Dalton, 2008, p. 904). Ronald Inglehart has described the scale as representing “whatever major conflicts are present in the political system” (in Dalton, 2008, p. 904). The scale then functions as a summary of the issues and cleavages that structure the political competition in a country (Dalton, 2008). Although the meaning of the left-right dimension might differ among countries, the value of the dimension is approximately the same for all countries. Methodologically this means that the
14 respondents know what they are asked about. Also, as opposed to other dimensions, the left-right dimension is included in almost all - for this research - relevant datasets. This makes the left-right dimension suitable for creating a generalizable dataset.
The independent variable
The number of coalition parties serves as the predictor variable in this study. The effect is first measured through a One-Way ANOVA Test and through a regression analysis. I used the ParlGov.org dataset to count the parties in every coalition between 1996 and 2017. Although I have tried to include as many cases as possible, comparability between cases over the years was crucial to reach the right conclusions. Thus, each country will be represented as often as possible, depending on the data available. When a party exited the coalition before the next elections took place, I decided for some cases to include the party when the difference between exit and election was only a few months. In that case, the electorate was still likely to base their party placement on the experience of that party being part of the coalition.
Also, parties that formed an alliance just before the elections, are counted separately. This is because of three reasons: first, the respondents had not been able to familiarize themselves with the alliance serving as ‘one party’. Secondly, an alliance is a sort of coalition to begin with, and in this study, there is no reason to distinguish those concepts. And thirdly, as a more practical argument, the alliance members are asked about separately in the surveys.
Ultimately, I ended up with as much as 70 coalitions that had data on voters’ party placement in established democracies. A problem in the dataset is that each case represents a party in government in a certain year. Therefore, for one coalition government, there are multiple scores. This also means that these scores are not independent from each other, as such cases were part of the same coalition. To counteract this problem, I included dummy variables for the countries. In my dataset with twenty countries, it means I created nineteen dummy
15 variables. I chose Spain as the reference category, because all its cases represented the first category, one party in government. I decided not to make dummy variables of the different coalitions, as the explanatory power would end up being fairly limited, with 69 dummy variables.
Controlling for number of parties in parliament
This research focuses on the effect that the number of parties in coalition have on placing those parties on the left-right dimension. It is possible however that the results that derive from this analysis, are due to a more systemic factor: the total number of parties in parliament. To which extent the electorate is in agreement over coalition partners’ score on the left-right dimension with a high number of parties in coalition might be the result of systems with a high number of parties in parliament.
Marketing a clear and distinct party profile is more difficult when there are many parties to compete with. With regard to airtime and the possibilities in general to market your policy positions, logically the more parties there are, the less opportunities parties get. Also, voters might find it more difficult to identify party positions when there are so many to choose from to begin with.
At the same time however, it could be argued that in proportional systems, parties are stimulated to use the full range of the ideological dimension (Sartori, 1976, p. 128). As opposed to majoritarian systems, there is no “winner takes all” mechanism at play (Strøm & Müller, 2009, p. 44). Therefore, parties can be more extreme and still win enough votes to participate in government. Positioning yourself more on the extreme side of the left-right dimension, is likely to result in a clear and distinct party profile for the electorate. Then, the more parties in an electoral system, the more clear and distinctive party profiles parties in that system have.
16 One of these mechanisms could possibly bias the results. The number of parties in parliament is therefore included in the regression analysis as a control variable. I counted the parties in parliament through the ParlGov.org dataset (2019). While using this dataset, I did not distinguish between party-sizes, nor did I apply the ‘effective number of parties’ as issued by Laakso and Taagepera (1979). I relied on the data provided by ParlGov.
Controlling for the kind of party in coalition
As a second control variable, I added whether the coalition party delivers the
government leader or not. According to Warwick (1996), the party of the government leader, also known as the senior partner, clearly has a more central role than her partners (p. 473). Klüver and Spoon (2019) pointed out that this factor influences the distinctiveness of parties. As they argue, the senior partner is more likely to have a clear position on policy issues. This would be the result of considerably more media attention, fostered by institutional structures. Through weekly press conferences for example, the government leader has more opportunities to make clear what her/his party stands for (p. 7). Junior coalition partners at their turn, find it hard to differentiate themselves from the senior partner (p. 5). This factor could also possibly bias the results from the bivariate analysis. Therefore, I added this dummy variable in the multiple regression analysis, by using data provided by the ParlGov dataset.
Results and discussion
The first part of the analysis intended to determine through a bivariate analysis whether there is a linear relationship between the number of parties in government, and the ability of the electorate to place those parties on the left-right dimension. Two hypotheses are tested through this analysis. The first hypothesis assumes a positive relationship between the number of parties in coalition and the ability of voters to place these parties in the left-right dimension.
17 Graphically, this means that with the number of parties increasing, the line presenting standard deviation among the electorate placing the parties on a left-right dimensional scale, would decrease. The second hypothesis assumes a negative relationship between the number of parties in coalition and the ability of voters to place those parties on the left-right dimension.
By looking at the ANOVA test for the predictor variable and the dependent variable, we do observe significant differences between the groups: F(4, 152) = 6.538, p = .000. This means that it is better using the group means than using the overall mean of the group sizes to predict the dependent variable (Field, 2013, p. 434). However, when we put this relationship into a graph, we can notice quite clearly it is not a linear relationship (see Figure 1).
Figure 1. ANOVA plot of the number of parties (x) and the standard deviation of left-right score (y)
The y-axis represents the mean standard deviation on the scores assigned to parties by the electorate, and the x-axis represents the number of parties in coalition. The group of “five parties or more” represents the coalition group of five parties (ten cases), six parties (three cases) and eight parties (six cases) (see Table 1). So instead of presenting these groups
18 separately with limited explanatory power, I choose to merge these together as a proxy for a “big coalition”.
Table 1. Frequencies of coalitions sizes Coalition size Frequency
One party 22 Two parties 50 Three parties 47 Four parties 20 Six parties 3 Eight parties 6 Total 158
First, what stands out is that considering the left-right dimension is a scale of one to ten, we do see a relatively high standard deviation present at all groups. With standard deviations ranging from 1.8 to almost 2.1, none of the government sizes is particularly good at stimulating clear, distinct party profiles.
19 The second aspect that stands out is that the cabinets consisting of only one party, have a particularly high variance in scores assigned to them. To account for such a high variance, we could turn to the notion proposed by Strøm & Müller (2009), that in majoritarian systems, party competition is usually organized around the vote of the median voter (p. 44). Because they compete around the median voter, the parties often adopt centrist positions and as a result, they lack a distinct party profile.
In the case of Great Britain, which displayed high standard deviations (see appendix), we get a mixed image. The Labour Party and the Liberal Democrats are perceived as moderate, as most respondents assigned a score of five to those parties after they were in government. But the Conservatives are assigned a median score of eight on the left-right dimension, which cannot be regarded a moderate score. Canada is in this sense comparable to Great Britain, in that one of the two relevant parties is in all cases assigned a moderate median score of five (Liberals), and the other party (Conservatives) an eight. Standard deviation for both parties is high: for all Canadian cabinets in the dataset it’s above 2.1. As for the limited amount of cases in these one-party cabinets (22), it is difficult to draw too harsh conclusions from these observations.
After a small drop from ‘one’- to ‘two parties in coalition’, the standard deviation decreases quite drastically between ‘two parties in coalition’ and ‘three parties in coalition’. This essentially means that the party profile of a government party is clearer in a coalition of three parties, than when the cabinet consists of two parties. The dependent variable goes up again when there are four parties in coalition and hits its lowest score with five or more parties. Especially considering my hypotheses and the theories they were based upon, there is not really a theoretical explanation for this inconsistent relationship between the number of parties in coalition and perceived party positions. All in all, the means plot demonstrates a fairly inconsistent image. Therefore, I can conclude that there is no linear relationship between the
20 number of parties in coalition and the standard deviation of the left-right score assigned to parties by the electorate.
The regression analysis
Although there is no linear relationship between the predictor and dependent variable, the differences between the number of parties in coalition are significant. To assess whether these differences are due to other factors than the number of parties in coalition, two models were included in the regression analysis. The first model represents the bivariate relationship between the predictor and dependent variable. As can be seen in Table 2, the predictor variable only explains about 12.4% of the variance in the dependent variable. None of the variables in this model are correlated, and the Variance Inflation Factors (VIF) are well below five.
The second model of the multiple regression analysis includes the predictor variable, the number of parties in parliament, the dummy for the senior partner, and dummies for the countries. This model accounts for 49.8% of the variances in the dependent variable (Table 2). The values of the residuals are independent with a Durban-Watson score of 1.92, and the variance of the residuals is constant and is normally distributed. There are no influential cases biasing the results. Although the ANOVA makes clear that the variables differ significantly from each other, F(25, 131) = 7.194, p = .000, neither one of the control variables ‘number of parties in parliament’, and the dummy ‘senior partner’ (ref. junior), are significant.
As laid out in the research design, there was a problem with dependency between cases in the bivariate analysis. Because the standard deviation scores were assigned to every individual coalition partner, these scores aren’t independent: 136 of the 157 standard deviations are part of a coalition. Although I could not counteract this problem by creating dummy variables for every coalition, as that would mean creating 69 of them, I created nineteen dummy variables for the countries.
21
Table 2. Multiple regression analysis with dependent variable standard deviation of left-right score
Model 1 Model 2
(Constant) 2.078 1.795
(.069) (.372)
Number of parties (ref. One)
Two parties in coalition -.028 -.087
(.083) (.093)
Three parties in coalition -.277* -.018
(.084) (.104)
Four parties in coalition .001 -.055
(.101) (.127)
Five or more parties in coalition -.304* .022
(.102) (.162)
Kind of coalition partner (ref.: Junior)
Senior partner .075
(.046)
Number of parties in parliament -.007
(.031)
R2 .146 .579
Adj. R2 .124 .498
N 157 157
***p < 0,001, **p < 0,01, *p < 0,05
22 There were five countries that had significant coefficients with regard to the reference country Spain. Australia (stnd. d. > 2.17, p = .029), Belgium (stnd. d. > 2.04, p = 0.05), Great Britain (for three out of four scores have stnd. d. > 2.32), Ireland (stnd. d. > 2.06, p = .005) and New Zealand (stnd. d. > 1.92, p = .009). The low standard deviations of Spain, (<1.9), can account for these significant coefficients.
For one of these significant scores, New Zealand, it also had a problematic VIF score (12.646). As can be seen in the appendix, three out of five of New Zealand’s coalitions had four parties, and all got a standard deviation above 2.1. The country is more or less regarded as ‘the same’ as ‘four parties in coalition’. It might therefore not be surprising that correlation is high. Although these observations can be a point of scrutinization for future research, there is not much more to do than speculate on what these results mean with regard to the research question. Only with more cases more drastic conclusions could be drawn but still then, these could be ecological errors – as no individual factors are taken into account.
Conclusion
Many authors have developed arguments about the increasing levels of indifference among voters. High volatility rates and low party-membership numbers are used as a proxy for this indifference, summarized by Peter Mair (2006) as a process of ‘mutual withdrawal’ (p. 33), as both the voter and the party distance themselves from each other. In this study, the focus is more on the role of the party in this process, where the voter merely serves as an echo-chamber of the alternatives parties pose (Key, 1966, p. 2). Along this notion, Caul and Gray (2002) argue that to involve the electorate, clear and distinct party profiles are essential (p. 236), which is the main focus of this research.
This study’s aim was to answer the following question: “to what extent does the number
23
government parties’ policy positions?”. As the dependent variable in this study I have taken
the standard deviation of the left-right score assigned by the electorate to political parties in coalition. The higher the standard deviation, the more disagreement among the electorate to where the coalition partners stand for. Conversely, the closer the standard deviation is to zero, the more the electorate is in agreement on where the parties stand for. The two hypotheses that were tested expected either a positive (Hypothesis 1) or a negative relationship (Hypothesis 2) between the number of parties and the agreement among voters to where the parties stand for.
First, I conducted an ANOVA test to compare the group means of coalition sizes. Second, I included two models into a multiple linear regression analysis. In the second model of the regression analysis, I controlled for: the number of parties in parliament, and what position the party has in coalition (senior or junior). Because each standard deviation score represents one party, not all scores are independent (as 136 out of 158 scores are part of coalitions). Because I would end up with a limited explanatory power when I would include dummies for the 70 coalitions, I created 19 dummy variables for the countries and included this in the second model as well.
The Comparative Study of Electoral Systems (CSES) features data on electorates among Western democracies. It asked respondents to place parties on the left-right dimension. For the number of parties in coalition and parliament, and whether the party is senior partner or not, I used the ParlGov (2019) dataset.
However straightforward the research question central in this study, the answer is not that straightforward. There is no linear relationship between the number of parties in coalition and the party policy positions perceived by the electorate, but the differences between the coalition sizes are significant. Following this, we can conclude that numbers do count, but not in the way that was expected. The parties from one-party governments fostered a mean standard deviation of almost 2.1; then the graph showed a small drop to the two-party coalitions, and a
24 big decrease to three-party coalitions. The line went firmly up again from three to four and showed a big decrease to the ‘five or more’-party category. There is no theoretical explanation for the inconsistent relationship between the effect of number of parties in coalition on party placement by the electorate. The explanatory power of the predictor value is fairly limited however (Adj. R2 = .124). In other words, we cannot simply say that the more parties there are
in coalition, the less clear party’s policy positions are for voters. Controlling for the number of parties in parliament and the kind of party in coalition, did not foster significant results.
It is important to note however that there are some clear limitations to this study. This dataset was built predominantly on the availability of data. The limited amount of cases (158), is something that could be improved in future research. It could be the case that with more cases, the predictor value of the variables increases as the groups of government sizes might display a more coherent image. Now, outliers are taken into the study as well to be able to analyse at all, although no influential cases are biasing the results. Also, the fact that the scores aren’t independent as several were of parties in the same coalition, is problematic. I have tried to counteract this problem by controlling for countries, but this hasn’t solved the problem of course. More than once, the dataset lacked data in the years a certain country had a coalition government. But it is also good to keep in mind that in the grand scheme of it all, coalitions with a high number of cases are just not that common.
What I did not control for, but what could have influenced the results, is the range of ‘used space’ on the left-right dimension. This would test the theoretical expectations of party competition in systems with a high number of parties in parliament, where parties are more inclined to use the full range of the left-right dimension. In a two-party coalition, the difference between the two scores would then be measured. However, for coalitions with more than two parties, the result is not as straightforward. For example, with parties A, B, and C, it could mean that party A and B are close, and C is on the other side of the range, or B and C are close, and
25 A is on the other side, and they could also be evenly distributed over the dimension. In this research it was difficult controlling for this, as coalitions with two parties could have the same range as coalitions with four, five, or even six parties. The explanatory power of this variable would therefore be limited. I was not able to conduct a mathematical calculation for this factor. For future research however, I would recommend looking into this.
In this research I focused on the electorate as a whole, because for linkage and party competition to work, the electorate needs to be able to identify party positions (Spoon & Klüver, 2017, p. 115). Both competent and noncompetent voters are part of the electorate, and therefore should parties aim to market their policy positions so that no one could miss its position. In this research, I found no linear relationship between the number of parties in coalition and the ability of the electorate to place those parties on the left-right dimension. However, what could be interesting for future research is to look whether the effect of the number of parties in coalition manifest itself on individual voters’ level.
All in all, there is no reason to state that from this research, it becomes clear that parties and people in general should be more in favour of minority governments, as proposed earlier in this study. What does become clear is that most parties could gain from clear and distinct profiles as we see a relatively high standard deviation among all sizes of cabinets.
26
Literature
Adams, J., Ezrow, L. & Wlezien, C. (2016). The company you keep: how voters infer party positions on European integration from governing coalition arrangements. American
Journal of Political Science, 60(4), pp. 811-823.
Blais, A. & Bodet, M. (2006). Does Proportional Representation Foster Closer Congruence Between Citizens and Policy Makers? Comparative Political Studies, 39(10), pp. 1243-1262.
Caul, M. L. & Gray, M. M. (2002). From Platform Declarations to Policy Outcomes: Changing party profiles and partisan influence of policy. In Russell J. Dalton & Martin P. Wattenberg (red.), Parties without Partisans: Political Change in Advanced Industrial
Democracies. Oxford: Oxford University Press.
Chapel Hill Expert Survey (2020). 1999-2014 Chapel Hill Expert Survey (CHES) trend file. Accessed from: https://www.chesdata.eu/1999-2014-chapel-hill-expert-survey-ches-trend-file.
Comparative Study of Electoral Systems (2020a). IMD (1996-2016). Retrieved from:
https://cses.org/data-download/cses-integrated-module-dataset-imd/.
Comparative Study of Electoral Systems (2020b). Module 1 (1996-2001). Retrieved from:
https://cses.org/data-download/cses-module-1-1996-2001/.
Comparative Study of Electoral Systems (2020c). Module 2 (2001-2006). Retrieved from:
https://cses.org/data-download/cses-module-2-2001-2006/.
Comparative Study of Electoral Systems (2020d). Module 3 (2006-2011). Retrieved from:
https://cses.org/data-download/cses-module-3-2006-2011-2/.
Comparative Study of Electoral Systems (2020e). Module 4 (2011-2016). Retrieved from:
https://cses.org/data-download/module-4-2011-2016/.
27
https://cses.org/data-download/cses-module-5-2016-2021/.
Costello, R. Thomassen, J. & Rosema, M. (2012). European Parliament Elections and Political Representation: Policy Congruence between Voters and Parties. West European
Politics, 35(6), pp. 1226-1248.
Dahlberg, S. (2013). Does context matter – the impact of electoral systems, political parties and individual characteristics on voters’ perceptions of party positions. Electoral Studies,
32, pp. 670-683.
Dalton, R. J. (2008). The quantity and the quality of party systems: Party system polarization, its measurement, and its consequences. Comparative Political Studies, 41(7), pp. 899-920.
Druckman, J. N. & Roberts, A. (2005). Context and Coalition-bargaining: comparing
portfolio allocation in Eastern and Western Europe. Party Politics, 11(5), pp. 535-555. Field, A. (2013). Discovering Statistics using IBM SPSS. Sage.
Fortunato, D. & Stevenson, M. R. T. (2013). Perceptions of Partisan Ideologies: The Effect of Coalition Participation. American Journal of Political Science, 57(2), pp. 459-477. Ganghof, S. & Bräuninger, T. (2006). Government status and legislative behavior: partisan veto
players in Australia, Denmark, Finland and Germany. Party Politics, 12(4), pp. 521-539.
Gerber, D., Nicolet, S. & Sciarini, P. (2015). Voters are not fools, or are they? Party profile, individual sophistication and party choice. European Political Science Review, 7(1), pp. 145-165.
Itanes (2006). Itanes 2006 Monitor. Retrieved from: http://www.itanes.org/en/.
Katz, R. S. & Mair, P. (1995). Changing models of party organization and party democracy: the emergence of the cartel party. Party Politics, 1(1), pp. 5-28.
28 Cambridge: Belknap Press of Harvard University Press.
Klüver, H. & Spoon, J. J. (2019). Helping or Hurting? How Governing as a Junior Coalition Partner Influences Electoral Outcomes. The Journal of Politics. DOI:
https://doi.org/10.1086/708239.
Laakso, M. & Taagepera, R. (1979). “Effective” number of parties: a measure with application to West Europe. Comparative Political Studies, 12(1), pp. 3-27. Lesschaeve, C. (2017). The predictive power of the left-right self-placement scale for the
policy positions of voters and parties. West European Politics, (40)2, pp. 357-377. Mair, P. (2006). Ruling the Void? The hollowing of Western democracy. New Left Review,
42, pp. 25-51.
Mair, P. (2009). “Representative versus Responsible Government”, MPIfG Working Paper 09/8. Cologne: Max Planck Institute for the Study of Societies.
Niikawa, S. (2018). Oversized coalitions in Central and Eastern Europe: A Qualitative Comparative Analysis. East European Politics, 34(4).
ParlGov (2020). Parliaments and governments database. Accessed on February 18th,
retrieved from http://www.parlgov.org/.
Powell, G. B. (1982). Contemporary democracies. Harvard University Press.
Saalfeld, T. (2008). Intra-Party conflict and cabinet survival in 17 West European democracies, 1945-1999. In Giannetti, D. & Benoit, K., Intra-Party politics and coalition
governments. London: Routledge.
Sagarzazu, I. & Klüver, H. (2017). Coalition Governments and Party Competition: Political Communication Strategies of Coalition Parties. Political Science Research and
Methods, 5(2), pp. 333-349.
Sartori, G. (1976). Parties and Party Systems: A Framework for Analysis. New York: Cambridge University Press.
29 Spoon, J. J. & Klüver, H. (2017). Does anybody notice? How policy positions of coalition
parties are perceived by voters. European Journal of Political Research, 56, pp. 115-132. DOI: 10.1111/1475-6765.12169.
Strøm, K. & Müller, W. C. (2009). Parliamentary democracy, agency problems and party politics in Daniela Giannetti and Kenneth Benoit (red.), Intra-Party Politics and
Coalition Governments, pp. 26-50. London: Routledge.
Strøm, K., Müller, W. C. & Bergman, T. (eds) (2008). Cabinets and Coalition Bargaining: The
Democratic Life Cycle in Western Europe. Oxford: Oxford University Press.
Van der Meer, T. (2017). Niet de kiezer is gek. Amsterdam: Spectrum.
Van der Wardt, M., De Vries, C. E. & Hobolt, S. B. (2014). Exploiting the Cracks: Wedge Issues in Multiparty Competition. Journal of Politics, 76(4), pp. 986-999.
Warwick, P. V. (1996). Coalition government membership in West European parliamentary democracies. British Journal of Political Science, 26(4), pp. 471-499.
30
Appendix
Country Year # parties in
coalition # parties in parliament Standard deviation on left-right dimension Australia 1996 1 3 2,336 Australia 2004 2 4 2,437 Australia 2004 2 4 2,475 Australia 2007 2 4 2,167 Australia 2007 2 4 2,573 Austria 2008 2 5 2,113 Austria 2008 2 5 2,057 Austria 2013 2 5 2,165 Austria 2013 2 5 2,177 BelgiumFLA 1999 2 9 2,499 BelgiumFLA 1999 2 9 2,226 BelgiumWAL 1999 2 9 2,042 BelgiumWAL 1999 2 9 2,286 Canada 1997 1 5 2,112 Canada 2004 1 5 2,227 Canada 2008 1 5 2,423 Canada 2011 1 4 2,544 Canada 2015 1 5 2,388 Denmark 1998 2 12 1,548 Denmark 1998 2 12 1,946 Denmark 2001 2 14 1,477 Denmark 2001 2 14 1,426 Denmark 2007 2 11 2,38 Denmark 2007 2 11 2,054 Finland 2003 5 10 2,014 Finland 2003 5 10 2,083 Finland 2003 5 10 1,831 Finland 2003 5 10 1,667 Finland 2003 5 10 2,168 Finland 2007 3 9 1,914 Finland 2007 3 9 1,55 Finland 2007 3 9 2,031 Finland 2011 4 9 1,613 Finland 2011 4 9 1,841 Finland 2011 4 9 1,967 Finland 2011 4 9 1,838
31 Finland 2015 5 9 1,941 Finland 2015 5 9 1,692 Finland 2015 5 9 1,707 Finland 2015 5 9 1,874 Finland 2015 5 9 1,759 France 2002 3 10 2,35 France 2002 3 10 2,262 France 2002 3 10 2,044 France 2007 1 11 1,869 France 2012 2 11 1,75 France 2012 2 11 1,409 Germany 1998 3 6 2,206 Germany 1998 3 6 2,371 Germany 1998 3 6 1,946 Germany 2002 2 6 1,955 Germany 2002 2 6 1,955 Germany 2005 2 6 2,098 Germany 2005 2 6 2,029 Germany 2009 3 6 1,572 Germany 2009 3 6 1,705 Germany 2009 3 6 1,627 Germany 2013 3 6 1,629 Germany 2013 3 6 1,811 Germany 2013 3 6 1,871 Germany 2017 3 5 1,763 Germany 2017 3 5 2,026 Germany 2017 3 5 1,417 Greece 2009 1 5 1,717 Greece 2012 3 5 2,443 Greece 2012 3 5 2,028 Greece 2015 2 7 1,576 Greece 2015 2 7 2,051 Great Britain 1997 1 9 2,626 Great Britain 2005 1 10 2,361 Great Britain 2015 2 10 2,321 Great Britain 2015 2 10 1,72 Iceland 1999 2 6 1,685 Iceland 1999 2 6 1,506 Iceland 2003 2 5 1,443 Iceland 2003 2 5 1,646 Iceland 2007 2 5 1,671
32 Iceland 2007 2 5 1,703 Iceland 2009 2 5 1,911 Iceland 2009 2 5 1,768 Iceland 2013 2 5 1,655 Iceland 2013 2 5 1,65 Ireland 2002 2 8 2,058 Ireland 2002 2 8 2,081 Ireland 2007 2 8 2,224 Ireland 2007 2 8 2,488 Ireland 2011 3 6 2,979 Ireland 2011 3 6 2,458 Ireland 2011 3 6 99 Ireland 2016 2 6 2,44 Ireland 2016 2 6 2,128 Italy 2001 8 11 1,342 Italy 2001 8 11 1,615 Italy 2001 8 11 1,605 Italy 2001 8 11 1,536 Italy 2001 8 11 1,517 Italy 2001 8 11 1,573 Italy 2006 6 12 2,142 Italy 2006 6 12 1,495 Italy 2006 6 12 2,155 Luxembourg 2014 3 6 1,942 Luxembourg 2014 3 6 2,238 Luxembourg 2014 3 6 1,941 Netherlands 1998 3 11 1,784 Netherlands 1998 3 11 1,747 Netherlands 1998 3 11 1,486 Netherlands 2002 3 9 1,593 Netherlands 2002 3 9 1,692 Netherlands 2002 3 9 1,542 Netherlands 2006 3 9 1,72 Netherlands 2006 3 9 1,722 Netherlands 2006 3 9 1,42 Netherlands 2010 3 10 1,913 Netherlands 2010 3 10 1,901 Netherlands 2010 3 10 2,115 New Zealand 1996 2 4 2,236 New Zealand 1996 2 4 1,928 New Zealand 2002 2 7 2,606
33 New Zealand 2002 2 7 2,189 New Zealand 2008 4 8 2,213 New Zealand 2008 4 8 2,452 New Zealand 2008 4 8 2,261 New Zealand 2008 4 8 2,207 New Zealand 2011 4 7 2,149 New Zealand 2011 4 7 3,135 New Zealand 2011 4 7 2,368 New Zealand 2011 4 7 2,207 New Zealand 2014 4 8 2,352 New Zealand 2014 4 8 2,648 New Zealand 2014 4 8 2,133 New Zealand 2014 4 8 2,07 Norway 1997 1 8 1,947 Norway 2001 1 8 1,779 Norway 2005 3 8 1,542 Norway 2005 3 8 1,64 Norway 2005 3 8 2,063 Norway 2009 3 7 1,699 Norway 2009 3 7 1,484 Norway 2009 3 7 1,314 Norway 2013 3 7 1,779 Norway 2013 3 7 1,446 Norway 2013 3 7 1,287 Portugal 2002 1 6 1,708 Portugal 2005 1 6 2,331 Portugal 2009 1 6 2,497 Portugal 2015 2 5 2,07 Portugal 2015 2 5 2,345 Spain 1996 1 11 1,852 Spain 2000 1 11 1,715 Spain 2004 1 12 1,9 Spain 2008 1 11 1,894 Sweden 1998 1 7 1,894 Sweden 2002 1 7 1,939 Sweden 2006 1 7 1,775 Sweden 2014 4 8 1,554 Sweden 2014 4 8 1,398 Sweden 2014 4 8 1,438 Sweden 2014 4 8 1,739
