The closer, the better – How perceived ideological proximity between parties affects the emergence of coalition preferences

ABSTRACT

While coalition governments have been studied extensively, there remains a lack of understanding of how coalition preferences emerge and what factors are the most influential. Utilising coalition formation theories, this study posits that voters tend to prefer coalitions with a narrower range of ideological differences between 1.) the parties involved and 2.) between the voter and the party perceived to have the greatest ideological distance from the individual. Previous studies have mainly concentrated on a general left-right ideological dimension or specific issue dimensions, despite the fact that some of the countries analysed are characterized by a two-dimensional political space. Using Germany as a case study that exemplifies this inherent two-dimensionality, this analysis adopts a nuanced approach by employing the calculation of Euclidean distance based on socio-economic and socio-cultural measures. This approach aims to explain the formation of coalition preferences leading up to the federal election in 2021. Our results show how strongly the perceived ideological distance of parties contributes to developing preferences for a specific coalition option, even when controlling for party evaluation and socio-demographics. For this reason, they have important implications for the understanding of the emergence of coalition preferences and party competition.

KEYWORDS:

• Coalition preferences
• coalition research
• coalition government
• German politics
• coalition theory

Why study coalition preferences and what we can learn about them

For a long time, the political science approach to the study of coalitions and coalition building was focused on whether parties would be compatible enough to form a government (Strøm and Müller 1999). However, little attention has been paid to the attitudes of those affected by the government, namely the citizens. Nowadays, though, coalition governments are the rule rather than the exception in multi-party systems. Against this background, it seems only natural that considerations of which coalition might form an upcoming government are also taken into account by voters in their electoral decision-making (Meyer and Strobl 2016). In the process, voters also make strategic decisions based on the consideration of whether a party they prefer will be involved in future government participation after an election (Harsgor, Itzkovitch-Malka, and Tuttnauer 2023). This decision is also strategic because it involves weighing up multiple preferences for or between parties (Blais, Guntermann, and Bodet 2017). Fundamentally, it is also evident that specific expectations and preferences of government formation influence how voters ultimately arrive at an electoral decision (Huber 2014; Bargsted and Kedar 2009). In this context, it is particularly important how the issue-specific agenda of a possible government is assessed, which would be expected within the framework of a future coalition government (Meffert and Gschwend 2010).

The extent to which parties behave towards these corresponding expectations and preferences regarding the formation of a government also influences how they receive political support in the future (Debus and Müller 2013). It is therefore doubtful that preferences for coalition formation, so-called coalition preferences, are not important in multi-party systems. Nevertheless, the factors that lead to the emergence of specific coalition preferences are still largely unexplored (Welz 2023). In recent years more and more contributions have dealt with the explicit question of how coalition preferences arise in the first place. This appears to be only logical against the backdrop of diversifying party systems, resulting in a steadily increasing heterogeneity observed in the formation of coalition models (Huber and Welz 2022). These studies show very impressively, for example, that self-placement on an ideological left-right axis, a general party affinity or familiarity with coalition models have a considerable influence on whether increasing preferences for a coalition model can be observed or not (see e.g. Falcó-Gimeno 2012; Debus and Müller 2014; Plescia and Aichholzer 2017). In addition, more general factors established in electoral research, such as general party affiliation or candidate orientation (Wurthmann 2022) or spatial distance based on specific issues, play a further role (Welz 2023). Although classically proximity to a party not only leads to a higher probability to vote for the party (Downs 1957) but also to a higher evaluation of a coalition in which a corresponding party would be involved, exceptions to this can also be found: if cooperation with a party is understood as contradictory to one’s canon of values, a higher identification with such a party can also lead to the rejection of a coalition (Wurthmann 2022). While the preferences for parties in the coalition matter, previous research also suggests that coalition options are entities themselves and that they are more than just a combination of the member parties (see e.g. Huber 2014; Plescia and Aichholzer 2017). In addition to these findings, which are essentially based on ideological components of party competition, entirely non-ideological reasons also influence whether preferences for a particular coalition model are developed. Nyhuis and Plescia (2018), for example, show that confidence in the ability of a government to act is also relevant as an explanatory variable, irrespective of ideological components. However, it is essential to delve further into the ideological elements. Although essential priorities of coalition preference formation can be derived from the more general left-right self-placement (Falcó-Gimeno 2012; Debus and Müller 2014; Plescia and Aichholzer 2017; Wurthmann, Marschall, and Billen 2019), it is questionable whether this one-dimensional mapping does justice at all to the rather two-dimensional party competition (Marks et al. 2006) and the parties competing and forming coalitions in it. This is especially relevant in light of decreasing precision in measuring the left-right dimension (Steiner 2023). Moreover, further studies indicate that citizens are well aware of the two-dimensionality of party competition (e.g. Bowler, Gschwend, and Indridason 2020), and a one-dimensional representation in such cases can lead to significant distortions in spatial determinants. In contrast, other approaches that use general issue positions (Wurthmann 2022) or the spatial distance between individuals and parties based on concrete issues (Welz 2023) appear to be too reductionist.

We add to the existing research by taking a point of view that is more focused on coalition formation theories. Research on coalition formation has clearly shown that parties prefer to form those coalition governments that are less ideologically diverse. In this paper, we argue that voters should also be interested in the ideological divisiveness of coalition options since this has important consequences for government effectiveness and durability. Accordingly, we empirically test whether the perceived ideological divisiveness of coalition options influences the voters’ coalition preferences. Another contribution of this paper is that we do not focus on the general left-right scale but include ideological divisiveness on two more fine-grained scales: the socio-economic and the socio-cultural scale. This will give us a more differentiated insight into which ideological conflicts, which also significantly structure European party competition (Dassonneville, Hooghe, and Marks 2023), are essential concerning the emergence of coalition preferences. Our findings suggest that both, the ideological range as well as the maximum distance between the individual and the party furthest away importantly predicts the evaluation of coalition options.

The paper is structured as follows. First, we present our theoretical framework, focusing on both parties and the coalition options as separate entities as important determinants of coalition preferences and deducting our hypotheses from this. Following this, we present our research design and empirical analysis. Lastly, we conclude with a discussion of our results and their implications.

Coalition preferences in the context of coalition research

Studying coalition preferences comes with an important question: who are the main actors that voters evaluate? The parties that form a coalition or the potential coalition itself? Do voters evaluate coalition options solely based on the parties that are members of the potential coalition, or is the coalition option such an independent entity for the voter to consider? Previous research indicates that there might be something to the assumption that coalition options are more than just the parties they consist of (Huber 2014; Plescia and Aichholzer 2017), and we also know from earlier studies that potential coalition options matter for voting decisions, especially with regard to strategic voting (see e.g. Bargsted and Kedar 2009; Meffert and Gschwend 2010). While the results are far from clear-cut, this paper focuses on coalition options as independent entities and how citizens’ perceptions of them influence their evaluation.

Voter preferences and coalition formation theory

If we aim to analyse the emergence of coalition preferences, we must delve into how voters develop these preferences. Following the model proposed by Downs (1957), to maximize utility, voters evaluate parties based on their positions and then vote for the party that is closest to the voter’s preferences. When assessing coalition options, voters should similarly evaluate the coalition best that maximises utility for them. However, doing so is more complicated since voters must consider the probabilities of each coalition forming and potential negotiation outcomes (Downs 1957).

The primary assumption in this paper is that citizens, when assessing coalition options, take into account aspects of a potential coalition that align with what parties consider during negotiations for coalition formation. Previous research on coalition formation has shown that governments are more likely to form the less ideologically diverse they are (see e.g. Martin and Stevenson 2001; Debus 2022). The formation process is easier and faster (Ecker and Meyer 2015), and the risk of early government termination is also lower the less ideologically diverse coalition governments are (see e.g. Saalfeld 2008). Even though voters may not have detailed knowledge of the specific similarities and differences between parties, studies have demonstrated that the left-right placement of parties by voters is not significantly different from party placements based on expert judgements or party positions derived from manifestos (Adams, Ezrow, and Somer-Topcu 2011). However, voters are capable of positioning parties, and previous research has also indicated that they possess an understanding of the ideological range within a potential coalition (Bowler, Gschwend, and Indridason 2020). Moreover, similar to parties, voters should also prefer those coalition options that are less ideologically diverse. This is attributed to the responsible party model, where citizens vote for parties based on the electoral programs presented to them by the parties before the election (APSA 1950; Thomson 2001). Additionally, voters also base their preferences for parties on their perceived level of competence in implementing the pledges they made during the electoral campaign (Fortunato 2021). If the parties make it into government, voters expect them to enact the promises they made during the electoral campaign. Research has shown that pledge fulfilment is less likely if there is no agreement between the coalition partners (see e.g. Thomson et al. 2017). Agreement between coalition partners is more likely, the smaller the ideological distance between the coalition partners. Additionally, compromises between ideologically close parties will not deviate as far from the parties’ ideal positions compared to compromises within more ideologically diverse coalitions (Martin and Vanberg 2008). The more likely agreement between coalition partners, the more likely parties can implement their pledges at least partially. Accordingly, since voters should be interested in coalitions to form that are capable enough to fulfil their electoral promises, they should be interested in coalition options with smaller ideological distances.¹

Political spaces’ two-dimensionality and coalition preferences

Although a certain familiarity has been shown to contribute to the preferential evaluation of specific coalition formats (Debus and Müller 2014), which have been handed down to a certain extent (Armstrong and Duch 2010), these are now only suitable as heuristics to a limited extent (Gross et al. 2023a). This is due to the fact that party competition in Western Europe has become increasingly fluid in recent decades and thus also significantly less predictable (Green-Pedersen 2019; Laver and Benoit 2015). Consequently, this leads then, for example, to an increasing swing voting behaviour and turning away from mainstream parties (Spoon and Klüver 2019). One of the consequences of this increased uncertainty is that previously acceptable certainties about which coalition would be formed from a set of possible options after an election (Meffert and Gschwend 2010; Switek 2010) are now challenging to maintain. In addition, the general coalition-building processes have also become significantly more difficult, which is why an ‘erosion of familiar patterns of coalition politics’ (Welz 2023, 21) has already been attested.

The heuristics that contribute to reducing the perceived complexity of an increasingly complex political reality are thus changing – also with regard to coalition-building processes and desirable coalition formats (Bowler, McElroy, and Müller 2022). This also impacts traditional indicators that might be employed to deduce the formation of coalitions, such as coalition signals. If parties send out signals that they want to form a coalition with a specific party after an election, voters take up the corresponding information, for example, for electoral decisions (Gschwend, Meffert, and Stoetzer 2017; Bahnsen, Gschwend, and Stoetzer 2020). Nevertheless, exceedingly party-devoted voters may be dissuaded from supporting their preferred party if there is an indication of seeking a coalition with a party that is not favoured as a potential government partner (Meffert and Gschwend 2011). Such information thus contributes to the fact that certain coalition formations are at least considered more likely, which leads to an informed understanding of possible coalition formations (Bowler, McElroy, and Müller 2022) and can also lead to an increased preference for such an alliance (Plescia and Aichholzer 2017; Wurthmann 2022). This is especially true in light of the fact that such mutually cooperative behaviour contributes to being perceived as more ideologically homogeneous (Adams, Weschle, and Wlezien 2021). At the same time, coalition building also leads to a re-evaluation of how parties are perceived ideologically (Fortunato and Stevenson 2013; Falcó-Gimeno and Fernandez-Vazquez 2020).

Undoubtedly, the perceived ideological position of a possible coalition plays a central role in how it is evaluated, although in the past, this has been done exclusively with one-dimensional approaches (Gschwend and Hooghe 2008; Debus and Müller 2014; Plescia and Aichholzer 2017). It is questionable, however, to what extent this is a contemporary approach when, for example, Western European multi-party systems are usually complicated to break down to such a unidimensional view (Marks et al. 2006; Dalton 2018). Instead of a one-dimensional left-right dimension, the distinction between a socio-economic and a socio-cultural conflict line is more adequate to define party competition (Kriesi et al. 2006; Debus 2022). Recent research has further provided evidence that party competition is now shifting from the economic left-right dimension to the GAL-TAN dimension (Dassonneville, Hooghe, and Marks 2023; Steiner 2023). While the left socio-economic position is associated with welfare state redistribution and interventions in market freedom, the right socio-economic position is associated with tenure-preserving policies and resistance to interventions in market freedom (Norris and Inglehart 2019). The socio-cultural line of conflict is again spanned between a liberal-progressive and a conservative-traditionalist pole. The former corresponds to liberal-progressive ideas in dealing with migration, sexual minorities, and issues such as European integration, whereas the latter describes more restrictive and negative attitudes towards liberal lifestyles (Marks et al. 2006; Rovny and Polk 2019).

However, this does not invalidate the assumption that spatial explanations can be used to structure party competition (Downs 1957; Stokes 1963). Instead, it leads to the necessity of adapting them to the corresponding multidimensionality (Sartori 1976). Concerning the emergence of coalition preferences, where a one-dimensional operationalisation has sometimes been used exclusively (Debus and Müller 2014; Plescia and Aichholzer 2017), an adjustment must be made accordingly that can distinguish between a socio-economic and a socio-cultural dimension. This holds even greater significance as voters can position themselves and political parties based on this multidimensionality (Bowler, McElroy, and Müller 2022), allowing them to anticipate spatial distances accordingly. Moreover, they develop specific substantive expectations for how parties could position themselves on issues and assert themselves within a coalition government’s framework (Fortunato et al. 2021). In line with the assumption of explanatory spatial models, it seems natural that this would increase or decrease preferences for specific coalitions. Nevertheless, with regard to coalitions formed between three or more parties, this is an increasingly complex task (Fortunato 2021), especially against the background of diverging issue priorities in the population (Welz 2023).

In summary, a spatially unidimensional approach explains the emergence of coalition preferences. However, this does little justice to the multidimensionality of multiparty systems in Western Europe. An approach focussing on many substantive issues is unsuitable as a complexity-reducing heuristic. It is, therefore, all the more useful to refer to the two lines of conflict that structure party competition in Western Europe, the socio-economic and the socio-cultural conflict dimensions (Dassonneville, Hooghe, and Marks 2023). If it is assumed that a more ideologically heterogeneous perceived coalition would be perceived as less effective in governance (Nyhuis and Plescia 2018), resulting in less conflict resolution, which is, however, considered desirable (Downs 1957), it can be deduced the following hypothesis:

H1: The larger the perceived ideological distance between the potential coalition parties, the lower the rating of the coalition option.

Our first hypothesis focuses on the perceived distance between potential coalition parties and originates from coalition formation theory. What has been neglected so far, however, is the distance between the citizens and coalition parties. Our argument relies on the core premise of veto player theory, where the consent of an individual or group is essential for instigating political transformation and reforms (Tsebelis 2002). Depending on the configuration of the potential coalition, parties that are not in line with the voters’ preferences might act as veto players in the coalition government. For instance, it should be quite obvious that for voters that place themselves on the extreme right of the political spectrum, an ideologically highly compact coalition between the social democrats and the green party might still be highly unfavourable. However, even a coalition government between a right-wing party and a social democratic party might be less optimal for these voters since this option still includes a party far away from their position on the left-right scale. When parties negotiate about forming a coalition, they need to find compromise policies that they want to pursue during their time in office. Accordingly, if a coalition includes a party that is far away from the preferences of the voter, it is quite likely that this will lead to compromise policies that are also further away from the voters’ preferences. Hence, the social democratic party could potentially act as an important veto player regarding the enactment of the pledges made by the right-wing party. There are two reasons that underscore the need to consider this determinant. On the one hand, Debus and Müller (2014) convincingly demonstrate that voters perceive members of coalition governments as veto players. If government parties are perceived as positioned very far from the respondents, this diminishes the overall evaluation of the coalition. At the same time, coalition parties, once they have formed a government, are perceived as significantly closer to each other, as they cooperate and seek compromises with each other (Fortunato and Stevenson 2013). As such, including only one party that is far away from a potential voter might be detrimental for coalition option evaluations. As mentioned previously, it is not only the coalition option as such that matters for coalition preferences but also individual parties. Party preferences matter for evaluating coalition preferences (e.g. Falcó-Gimeno 2012; Debus and Müller 2014; Plescia and Aichholzer 2017). Additionally, suppose parties might form a coalition with a party that clashes with the individual’s values. In that case, citizens might reject such a coalition option even though it strongly identifies with other member parties (Wurthmann 2022).

The multidimensionality of party competition is relevant not only for the overall perception of the coalition option but also for the preferences of individual parties. Because the socio-cultural dimension is increasingly important for party competition (Dassonneville, Hooghe, and Marks 2023; Steiner 2023), a unidimensional approach to the party-citizen-centric explanation of coalition preferences is inadequate. For instance, citizens’ preferences might overlap with a potential coalition on the socio-economic dimension, but there might be major differences regarding the socio-cultural dimension. If this is the case, only including the classic left-right dimension overlooks important ideological divisions and might lead to biased results.

Incorporating predictors independently into the model contradicts the fundamental assumption that citizens are conscious of the two-dimensional categorization, capable of situating themselves and political parties accordingly. As a result, no positioning is carried out independently from another (Bowler, Gschwend, and Indridason 2020; Stoetzer and Zittlau 2020). We, therefore, argue also to pursue a multidimensional approach for the party-citizen-centric logic. Accordingly, we hypothesize the following:

H2: The larger the perceived distance between the individual respondent and the party with the largest distance of the coalition option, the lower the rating of the coalition option.

Research design

Case selection and data

To test our hypotheses, we draw on original survey data collected in the run-up to the 2021 German federal election. Germany, in general, and this federal election in particular is a suitable use case for our assumptions for several reasons. First, Germany’s longstanding tradition of forming coalition governments played a significant role in shaping the country after the state’s founding in 1949. At the national level, even in 1957, when the Christian Democratic Union (CDU) and the Christian Social Union (CSU) won an absolute majority, a coalition was formed with the national conservative German Party (DP). Although it is by no means unknown that voters per se are not necessarily supporters of issue compromises (Fortunato 2021; Gross et al. 2023a), the respondents in Germany are nevertheless aware that this is also the essence of democratic government formation. In addition, the German party system is undoubtedly one whose party competition can be described along the two-dimensional conflict structure already described here (Franzmann, Giebler, and Poguntke 2020; Debus 2022). The socialist party The Left (Left), the Social Democratic Party of Germany (SPD) and Alliance 90/The Greens (Greens) are located in the socio-cultural and socio-economic left spectrum. The Free Democratic Party (FDP) is a business-friendly liberal party in the socio-economic right and socio-cultural left spectrum. The CDU and CSU, as well as the far-right Alternative for Germany (AfD), on the other hand, are located in the socio-economic and socio-cultural right quadrant (see Figure 1).

Figure 1. The German Party System. Author’s own calculation and presentation based on the Chapel Hill Expert Survey (Jolly et al. 2022). Note: the median positions are shown here.

Another factor that makes this federal election particularly interesting for the present analysis is the openness voters faced in the run-up to the election date. For a long time, it was not clear whether the to-be chancellor would be called Annalena Baerbock (Greens), Olaf Scholz (SPD) or Armin Laschet (CDU). The parties remained equally open with their explicit coalition statements, as only the AfD was ruled out as a coalition partner by all parties. However, the fact that a three-party coalition would govern after the election was considered certain even beforehand. Moreover, a continuation of the ‘grand’ coalition was clearly ruled out by all parties in the run-up to the election and did not play a role in political discussions (Faas and Klingelhöfer 2022). However, the exact coalition configuration remained uncertain until a considerable time after the election. Nevertheless, it was clear that the Greens and the FDP would play a central role here and that this would be a new type of coalition formation at the federal level (Gross et al. 2023a; Debus 2022).

Consequently, this meant, on the one hand, that the heuristic of traditional familiarity with specific coalition models (Debus and Müller 2014) was at least challenged. On the other hand, coalition-directed, strategic voting became almost impossible, given the uncertain outcome of the election and the subsequent coalition negotiations. For this paper, this means that we work with a conservative case in which many contextual factors that have high explanatory power for the emergence of coalition preferences are attenuated or completely dropped as heuristics. This refers to, for example, coalition signals that can influence voting behaviour (Bahnsen, Gschwend, and Stoetzer 2020). These were utterly absent in the context of the high degree of uncertainty – it seemed too risky for the parties to commit themselves publicly. This, in turn, posed a challenge at the individual level in terms of shaping expectations about a probable government formation. Such expectations have been demonstrated to be significant predictors of coalition preferences (Meffert et al. 2011; Wurthmann 2022). Indeed, one can even state that nearly every coalition option ‘seemed to be on the table, giving pundits as well as disoriented voters a tough time’ (Faas and Klingelhöfer 2022, 1513).

If there is no direct control on the parties’ part, correspondingly less polarising explanatory patterns are to be expected, which makes Germany a reliable case to test our assumptions. In conclusion, no institutions, characteristics or circumstances make Germany an outlier for most European countries. Indeed, Welz (2023) also points out that the formerly very stable German party system has recently become increasingly unstable in its configuration and coalition arithmetic – and is thus also following alignment processes with neighbouring Western European countries. Hence, we believe our results can also travel to other European countries with coalition governments.

The data² used for this paper were gathered within a research project right before the 2021 German federal election. In cooperation with the market research company Respondi, an online panel survey was designed to cover the German electorate adequately. For the aim of this paper, we make use of the data that were collected during the first survey wave, which took place between August 9th and August 14th – six weeks before the German federal election. The respondents, aged between 18 and 74 years at the time of the survey, were recruited through a quota sample based on the Respondi Access Panel and its 100,000 registered members. Only minor deviations can be found when comparing the socio-demographic characteristics between representative census data and the respondents (see Online Appendix, Table A.1). A total of 2,270 individuals completed the survey. Of these 2,270 persons, 64 must be excluded in the first step, as they are not eligible to vote in the federal election or not properly informed whether they may vote or not. Subsequently, a further 165 people are excluded from the analysis because they are identified as straightliners when answering specific item batteries. An additional 674 individuals were omitted from the analysis as they indicated an inability to categorise at least one party along a socio-economic or socio-cultural conflict line (see 3.1 for the detailed question formulation and Online Appendix Figure A1 for the distribution of how many party positions the respondents refused to position). This corresponding proportion of 33.02 per cent aligns with the conclusions of prior studies, which suggest that up to a third of respondents may struggle to adopt accurate party positions or may even position them incorrectly (Schultze 2014; Banducci, Giebler, and Kritzinger 2017). Thus, 1,367 respondents remain in this stage of the analysis. Furthermore, we only calculate our models based on respondents who have given a substantive answer to all the variables we use, i.e. have not refused to answer specific issues or questions. Our analyses are therefore based on a total number of 1,274 respondents.

Dependent variables

Our dependent variable in this analysis is the emergence of coalition preferences, meaning the evaluation of a potential coalition. We decided to restrict our analysis to the four coalition options that had the highest probability to form after the election: the so-called Black-Red coalition between the CDU, CSU and the Social Democrats (the incumbent coalition before the 2021 elections), the Jamaica coalition (CDU, CSU, FDP, Greens), the Traffic Light coalition (SPD, FDP, Greens) as well as the Red-Red-Green coalition (SPD, the Left, Greens).³ Respondents were asked the following question: ‘Regardless of the outcome of the federal election, how desirable do you personally consider the following coalition governments to be?’. To express their individual preferences, they could choose their position on a scale ranging from (1) −5 not at all desirable to (11) + 5 extremely desirable. This is a common approach to measuring coalition preferences in surveys as some national election studies, such as the Austrian National Election Study (AUTNES) or the German Longitudinal Election Study (GLES) provide these items in their survey programs for usage (see e.g. Meffert and Gschwend 2010; Plescia and Aichholzer 2017; Wurthmann 2022; Welz 2023).

Independent variables

In order to empirically test our hypotheses, we rely on a number of different explanatory variables. First, we include a variable that measures the ideological distance between the potential coalition parties as perceived by the respondents. We based this assessment on two different dimensions: a socio-economic and a socio-cultural one. To capture the socio-economic self-assessment and the assessment of party positions, the respondents were presented with a scale on which they were to situate themselves as well as the parties. This scale ranged from (1) The government should play an active role in the economy to (11) The government should not intervene in the market at all. The respondents were also presented with a socio-cultural oriented scale on which they were also asked to place themselves and the party positions. This, however, ranged between the extreme positions (1) progressive and (11) conservative. On this basis, we have made a calculation to determine the range between the respective extreme positions as a distance measure of those parties that would be part of a coalition option. Suppose we have three parties that would be involved in a coalition and whose positions could be described as position (p) = c(2,4,7). On this basis, the range is calculated as follows: $\begin{array}{c}Range=max\left(p\right)-min\left(p\right)\\ =7\text{–}2=5\end{array}$This results in two distance measures per respondent – one socio-economic and one socio-cultural – for each coalition option. For our explanatory variable, we make use of these two dimensions in order to acknowledge the two-dimensionality of the political space (Marks et al. 2006; Dalton 2018; Dassonneville, Hooghe, and Marks 2023). Hence, we calculate the Euclidean distance based on these two dimensions.

We made this decision based on the following considerations: First, the essential character of the German party system is defined by the two conflict lines mentioned earlier, along which parties compete (Debus 2022; Wurthmann 2022). Utilising Euclidean distances to analyse coalition preferences is essential in two-dimensional conflict spaces (see, for example, Hinich and Munger 1997; Bräuninger and Debus 2008; Armstrong and Duch 2010; Welz 2023; Backlund 2023). Consequently, a one-dimensional approach would fail to adequately capture the intricacies of party competition in Germany (see e.g. Debus 2022; Dassonneville, Hooghe, and Marks 2023). Furthermore, it is crucial to acknowledge that ‘a one-dimensional approach is likely to overlook crucial aspects of political competition’ (Welz 2023, 24). This becomes even more apparent since recent findings compellingly illustrate that attributions of content to the traditionally defined left-right dimension, often considered a super issue (Dalton 2008), are undergoing significant transformations regarding the associated issues. This is particularly true among the younger generation who significantly attribute this dimension of rather socio-cultural nature (Steiner 2023). The independent incorporation of each of the calculated distance measures, on the other hand, is to be dismissed, as they exhibit collinearity with one another (Jolly et al. 2022). Much more so, these are intertwined in parts to such an extent that they should not be used independently, but also not without each other for spatial models (Stoetzer and Zittlau 2020).

In addition to this calculation and in order to account for veto players within a coalition, we create a measure that takes into account the distance between the individual’s position and the party that is farthest away from the respondent on the dimensions above.⁴ To do so, we calculate the distances between the potential coalition partners and the individual for both dimensions and then take the maximum distance for every dimension and every coalition option. As before, we then use this to calculate the Euclidean distance between the individual and the corresponding maximum distant party in a corresponding coalition.

However, while Euclidean distance is considered crucial for assessing distances in two-dimensional political spaces (e.g. Hinich and Munger 1997; Bräuninger and Debus 2008; Armstrong and Duch 2010; Welz 2023; Backlund 2023), in our analysis, we cannot directly account for how important each dimension is to the respondents and, based on that, assign a weighting. Nonetheless, it is possible to consider this trade-off at least indirectly through the assessment of party sympathies, which we include as a control variable and introduce in the following. It can be assumed that if two parties, for example, are very similar economically but culturally diverse, a corresponding party evaluation is likely to reflect a preference.

Control variables

In order to control for party preferences, we include the scalometer of all relevant parties. It seems intuitive that voters would prefer those coalition options more, the more they like the parties that are included in the potential coalition. The variables enable an evaluation along the scale from (1) −5 to (11) + 5.

In addition to that, previous research has demonstrated that coalition familiarity has a positive influence on coalition evaluations (e.g. Debus and Müller 2014). In order to take that into account, we include coalition familiarity in our models. In this process, we proceed as follows: Familiarity with a preceding government constellation implies having experienced it oneself. Thus, the maximum temporal coverage to be considered is measured by the age of the youngest respondents in the dataset. This ensures that coalition models to be appropriately considered have governed a federal state during the respondents’ lifetimes. In a 2021 survey, this implies that, after subtracting the relevant 18 years corresponding to the minimum age of the respondents, only coalitions established since 2003 are regarded as familiar. In other areas of coalition research, temporal-based familiarities also play a role. Relevant findings suggest that the formation of coalitions is significantly dominated and determined by the familiarity of acting party leaders. This familiarity is also limited in time, given party leaders’ average tenure of eight years (see e.g. Martin and Stevenson 2010; Franklin and Mackie 1983).

We also choose familiarity as to refer to the state level because, in the literature, the German Länder have proven to be a form of testing ground for coalition formations. On the national level, no coalitions are formed for which sufficient experiences have yet to be gathered at the state level (Switek 2010; Wurthmann, Marschall, and Billen 2019). Such experiences are even attributed with much broader significance, as Gross et al. (2023b) observe that

‘government formation at the regional level is influenced by political considerations at the national level. Therefore, studying state elections and subsequent government formation processes in the 16 German states is indispensable for comprehending both regional and national politics within the German multi-level political system’.

However, Debus and Müller (2014) come to a similar conclusion with regard to their experiences of coalition familiarity.

We measure previous government familiarity, therefore, in the following way: Only if the corresponding coalition has governed at the level of a federal state within the period since 2003 was the coalition familiarity coded = 1, and if this was not the case = 0.

While no Jamaica or Red-Red-Green Coalition has governed before 2003, a Traffic Light coalition was formed in the state of Bremen in the early 1990s, which is, for this reason, not considered here. A Black-Red coalition has existed at the state level in ten federal states from 2003 to 2021 (see Online Appendix Table A.2). In addition, such a coalition ruled at the federal level in three legislative periods. If we were to consider those Black-Red Coalitions that had already governed at the state and national level before 2003, the total experience with this coalition would increase even further, making this indicator seem redundant, as it applies to all respondents.

Additionally, sex, age and the level of education have demonstrable effects that can be seen, for example, at the level of attitudes or on voting behaviour. Moreover, several studies have provided evidence for socio-demographics influencing whether coalition preferences are higher or less pronounced, independent of other explanatory factors such as party identification or candidate orientation (e.g. Plescia and Aichholzer 2017; Wurthmann, Marschall, and Billen 2019; Wurthmann 2022; Welz 2023). For this reason, the variables are included in the modelling for control purposes. For sex, we distinguish between male (0) and female (1) individuals. In terms of educational level, we distinguish between people who finished school without a diploma (1), people with a secondary school diploma (2), people with a secondary modern school diploma (3), people with a technical college entrance qualification (4) and people with a high school diploma (5). The age was recorded by a free-field statement of the respondents.

Analysis

In the following part, we will test our hypotheses by relying on survey data from Germany. Before presenting the results of our linear regression models explaining the emergence of preferences for the Black-Red, Jamaica, Red-Red-Green, and Traffic Light coalitions, we begin by providing some descriptive results.

Descriptive results

Looking at the assessment of the respective coalition options, several aspects become apparent. First, it should be noted that all coalitions presented here for evaluation are rated negatively on average and median. While the median for the Traffic Light coalition, the Jamaica coalition or the Red-Red-Green coalition is −2, the median for the black-red coalition in power at the time is still slightly better at −1. Also, concerning the average rating, only slight differences can be observed. On average, the Black-Red coalition is still rated best of all options (mean[m] = −1.17). This is followed by the Red-Red-Green coalition (m = −1.18), the Traffic Light coalition (m = −1.48) and the Jamaica coalition (m = −1.60). The fact that the Black-Red coalition is the favourite amongst the survey respondents is somewhat surprising, considering that both parties and public opinion voiced its rejection against a continuation of the Black-Red coalition after the 2021 elections (Faas and Klingelhöfer 2022). One potential explanation for this finding could be that the grand coalition has been the only option that has been in office at the national level so far among those analysed here. Previous work has already provided evidence for preferences arising from habit and familiarity (Debus and Müller 2014). As such, some respondents might still prefer to choose the unbeloved option rather than the unfamiliar one (see Figure 2).

Figure 2. Coalition Preferences in the German Electorate. Source: Authors’ own calculation and presentation. Note: The y-axis shows the scale values from −5 not at all desirable to +5 extremely desirable and how each coalition was rated by the respondents. The horizontal black lines describe the median position, while the white triangles with black borders represent the mean. The figure is based on 1,274 respondents.

One of our central assumptions, based on coalition theory findings, is that parties’ ideological proximity can not only be used as an explanatory factor for coalition-building processes but that the perceived distance also explains the emergence of coalition preferences.

We have used this assumption as a starting point to calculate the Euclidean distance of coalition options based on a socio-economic and a socio-cultural conflict line. In particular, the Black-Red coalition of CDU/CSU and SPD is assessed as comparatively homogeneous, with a median distance of 4 units (m = 4.10). This value is slightly higher for the Red-Red-Green coalition of socio-economically left-wing parties. The median here is 4.47 (m = 4.82). The Traffic Light coalition, with 6.40 units (m = 6.35) and the Jamaica coalition, with a median of 7.28 units (m = 7.38), are perceived as considerably more heterogeneous (see Figure 3).

Figure 3. Perceived Coalition Range – Euclidean Distance. Source: Authors’ own calculation and presentation. Note: The x-axis shows the coalition models of a Black-Red Coalition, a Jamaica Coalition, a Red-Red-Green Coalition and the Traffic Light Coalition. The y-axis, in turn, describes the perceived ideological distance of the corresponding coalition models. The horizontal black lines describe the median position, while the white triangles with black borders represent the mean. The figure is based on 1,274 respondents.

In this context, looking at the first quartile is also interesting, as it provides additional information about the perception of the most minor distance measures. Here, the homogeneity of a Black-Red and a Red-Red-Green coalition is also clearly perceived. In contrast, a Jamaica and a Traffic Light coalition are perceived as much more heterogeneous (see Table A.3 in the Online Appendix for further descriptive statistics on the main dependent and independent variables).

Multivariate analysis

After the descriptive analysis of our main variables, we now turn towards our multivariate analysis. The results can be seen in Figure 4 and, in more detail, in Table A.4 in the Online Appendix. Hypothesis 1 suggested that the evaluation of coalition options should be lower, the higher the perceived ideological heterogeneity. Indeed, previous studies have shown that the way parties interact with each other is also perceived by voters and received accordingly (e.g. Adams, Weschle, and Wlezien 2021). Concerning coalition preferences, we find that this effect is also reflected at the individual level. At least for the Black-Red coalition (p < 0.01), a Red-Red-Green coalition (p < 0.01) and a Traffic Light coalition (p < 0.01), there is a substantial influence of the perceived coalition range on the evaluation. The greater the assessed ideological range of a coalition, the greater the rejection of this coalition option. Concerning a Jamaica coalition, similar effects can only be found under the conditions of a greater probability of error (p < 0.1) (for a more detailed overview, see Online Appendix Table A.3).

Figure 4. Explaining coalition preferences. Source: Authors’ own calculation and presentation. Note: The graph shows the beta coefficients with 95% confidence intervals.

The explanation provided initially clarifies that a perceived narrow coalition range does not necessarily lead to a more favourable evaluation of a coalition. We have justified this by noting that, for instance, a Red-Red-Green coalition, despite its high homogeneity, is not automatically positively rated, as it may contradict one’s substantive ideas. Conversely, an assumed heterogeneity does not automatically result in a less favourable coalition evaluation. While the Traffic Light coalition is at least very similar in socio-cultural respects and socio-economically quite heterogeneous, the Jamaica coalition is heterogeneous in both dimensions. Welz (2023) rightly points out that not all issues and substantive questions are equally relevant to every citizen. Thus, socio-economic differences would be equally significant in both coalition models, the Traffic Light and the Jamaica coalition. However, if socio-cultural issues are deemed particularly relevant, noticeable differences are observed only in the context of the Jamaica coalition. Furthermore, among the coalition options we examine here, the Jamaica coalition is, on average, the least popular (see Figure 2) and has the most extensive perceived coalition range (see Figure 3). One possible explanation for the coalition range being only significant at the 10% level might be the influential veto player assumption, suggesting that notably apparent substantive contradictions in government action are anticipated in this case—more than in the case of the Traffic Light coalition, which would be agreeing at least in socio-cultural regards. It is, therefore, understandable why a perceived distance between the individual position and the party furthest away in the context of a coalition leads to a significantly more negative evaluation of the Jamaica coalition (p < 0.01). We also find a substantial effect for the Traffic Light coalition. However, this effect is only significant at the 10% level. Similar to the Jamaica coalition, this also leads to a more negative evaluation of the coalition option. These findings cannot be confirmed for a Black-Red or a Red-Red-Green coalition. Accordingly, the effects observed become more diverse with regard to more heterogeneous and, in addition to that, more complex alliances. Specifically, attitudes towards an ideologically diverse Jamaica coalition are significantly influenced by the individual’s perspective on maximum distance. In contrast, the other models examined here are judged more firmly based on ideological heterogeneity. Accordingly, we cannot confirm our hypothesis H2, in which we assumed a more fundamental effect without restrictions. Nevertheless, the results certainly indicate that the veto player assumption, expressed through the individual maximum distance, can play a significant role in evaluating heterogeneous coalition formats. However, additional research is essential in the future to challenge the current findings.

Spatial distances thus play a significant role in coalition formation or voting behaviour (Downs 1957; Stokes 1963) and in forming coalition preferences. This is especially true against the background that, in addition to socio-demographic factors, we have primarily resorted to party evaluations, so-called party scalometers, as control variables. Previous analyses indeed indicate that coalitions as objects of attitudes also enjoy certain independence from the parties from which they are formed. However, this assumption is by no means true for the majority of voters (Plescia and Aichholzer 2017). In fact, our models also show very clear effects regarding how strongly the individual evaluation of single parties contributes to evaluating specific coalitions, for better or worse. We do not find a single effect in which the positive assessment of a party that would be part of a corresponding coalition would contribute to evaluating it significantly more negatively – the opposite is the case. Accordingly, our approximation-based findings are relevant and robust. This applies additionally against the background that we have also considered coalition experience with a coalition option at the level of the German Länder in the modelling. The finding of Debus and Müller (2014) concerning familiarity with a coalition formation leads to evaluating it significantly better in the case of the Traffic Light coalition.

Methodologically, in our second hypothesis, we suggested that the evaluation of coalition options should be worse the higher the distance between the respondent and the party furthest away from them. Alternatively, one could also suggest that it is not just about one party but the whole coalition. Accordingly, we ran additional models where we operationalized the variable by adding up all distances between the respondent and the coalition option parties and then calculating the mean (see Online Appendix Table A.5). The results remain consistent insofar as that a higher coalition range with regard to the Black-Red coalition, the Red-Red-Green and the Traffic Light coalition significantly contributes to a more negative evaluation of the corresponding coalition option. Interestingly, however, this robustness test shows that the effect of a Jamaica coalition, which was previously only significant with a ten per cent probability of error, becomes a significant predictor (p < 0.01). Previous findings for the individual maximum distance turn insignificant in this form of modelling. Overall then, this robustness test once again lends support for hypothesis 1 while at the same time challenging hypothesis 2.

In addition, we conducted an examination to assess the extent to which candidate-related effects may have been overlooked. Since the dataset used did not include a specific inquiry about chancellor preferences but solely evaluated individual candidates on a scale from −5 (strongly disapprove) to +5 (strongly approve), we formed three dichotomous proxy variables for this purpose. We considered a preference to be present (1) only when the individual was 1. rated higher than the two competing chancellor candidates and 2. received a positive evaluation. If these conditions were not met, no corresponding preference was indicated in the respective variable (0). For the CDU/CSU chancellor candidate Laschet, this applied to 158 respondents, for the SPD chancellor candidate Scholz, it applied to 319 respondents, and for the Greens politician Baerbock, 406 individuals could be accordingly classified. Consequently, a preference for Laschet led to a more favourable assessment of the Black-Red coalition (p < 0.001) and the Jamaica coalition (p < 0.05). Positive effects in favour of the Black-Red coalition (p < 0.001) and the Traffic Light coalition (p < 0.05) are evident for the SPD candidate Scholz. A preference for Baerbock, on the other hand, strengthens preferences for a Red-Red-Green coalition (p < 0.001) and a Traffic Light coalition (p < 0.05). However, the presented findings on the perceived range and the perceived maximum distance to a person involved in such a coalition remain unaffected (see Online Appendix Table A.6).

Classification and conclusion of the results

Nowadays, at least in parliamentary systems, coalition governments have become the norm rather than the exception. While the processes of coalition formation, governance and termination have traditionally received a lot of attention, coalition preferences remain severely understudied. In the present case, we relied on the classical assumption of coalition research that coalition formation becomes more likely the more ideologically homogenous the coalition parties are. For this reason, we first calculated the Euclidean distance based on the perceived socio-economic and socio-cultural heterogeneity of corresponding coalition options. Derived from coalition research findings and based on the assumption that the population is quite aware of an ideological distance of parties (Fortunato and Stevenson 2013), we formulated the hypothesis that the perception of increased heterogeneity should contribute to having a less favourable preference for a corresponding coalition.

Indeed, this predictor significantly contributes to elucidating the formation of coalition preferences leading up to the 2021 federal election. The perceived ideological range of a coalition played a crucial role in shaping evaluations, notably affecting the assessments of the Traffic Light coalition (comprising Social Democrats, Greens, and Liberals), the Red-Red-Green coalition (comprising Social Democrats, The Left, and Greens), and the Black-Red coalition (comprising Christian Democrats and Social Democrats). On the contrary, the ideological range does not exert a significant influence on the evaluation of the so-called Jamaica coalition between Christian Democrats, Greens, and Liberals. However, in the context of the Jamaica coalition, it stands out as the only case where the perceived distance between the respondent’s position and the party ideologically furthest away serves as a significant predictor in explaining the emergence of preferences. The higher the distance, the more negative the evaluation of the respective coalition option. The observed effect could be explained by the notably high level of heterogeneity within the Jamaica coalition, evident in both socio-cultural and socio-economic dimensions of conflict. Unlike the Traffic Light coalition, where there is at least a considerable degree of socio-cultural proximity among potential partners, the Jamaica coalition represents an option with a significantly higher potential for veto players in day-to-day policy implementation. This is due to the government parties in the Jamaica coalition spanning three ideological quadrants, while a Traffic Light and a Black-Red coalition would involve two quadrants, and the Red-Red-Green coalition would consist solely of parties from a single quadrant.

In summary, despite controlling for party evaluations, we observe consistent and substantial effects of both the ideological range and the maximum distance on the assessment of coalition options. However, these findings and the provided explanation warrant further examination in the future.

Nonetheless, these findings have important implications for the study of coalition preferences but also for party competition more generally. The perceived ideological distance between specific coalition parties can contribute to explaining the emergence of coalition preferences for or against a coalition option. Therefore, it is not only general sympathies or apathy that structure corresponding preferences. Spatial approaches, which are used to explain the formation of coalitions, also provide an essential explanatory contribution at the individual level. However, based on our findings, it seems likely that for coalition options with high ideological heterogeneity different heuristics are used. For the Jamaica coalition, the respondents relied more on the distance of the coalition parties to their own position.

All in all, this paper has shown that perceptions of distances between parties as well as parties and voters matter. Not only for the left-right dimension as previous research has shown (see i.e. Plescia and Aichholzer 2017) or specific issues (see i.e. Welz 2023) but also for the Euclidean distance, which was calculated based on spatial distance on a socio-economic and a socio-cultural conflict dimension – those dimensions that essentially structure the party systems of Western Europe (Dassonneville, Hooghe, and Marks 2023) and Germany in specific (Debus 2022).

While we relied on data from the German election in 2021, there is reason to believe that these results also travel to other countries. Germany, like many other European countries, has a long tradition of coalition governments. Additionally, not only German voters are used to electoral campaigns during which they receive, sometimes conflicting, coalition signals. Potential coalition options also matter in other countries (such as Austria, see Plescia and Aichholzer 2017). Lastly, Germany is also no outlier with regard to other important characteristics such as turnout, the number of parties running in elections and government formation processes. We therefore believe that our results are also potentially applicable in other countries and contexts.

However, further studies are needed that provide corresponding evidence proving the multidimensionality of ideological distance in developing coalition preferences. This also includes that such an assumption should, on the one hand, be examined in a context in which different established models of coalition formation appear as a realistic option for forming a government after an election. In the present case, this was not the case concerning the 2021 Bundestag election, which is why, for example, a coalition familiarity (Debus and Müller 2014) may not have been able to unfold the full potential of its impact power. On the other hand, the present case is only a singular cross-sectional study. Further analyses that follow a similar approach could be dedicated to longitudinal observation to investigate possible fluctuations over time. Furthermore, future studies should explicitly consider the importance of ideological dimensions in respondents’ decision-making, especially in the case of two-dimensional systems. This would make it possible to depict more nuanced differences in their relative significance for the formation of coalition preferences.

Although this was not possible in the present case, it can nevertheless be concluded that, with a view to the emergence of coalition preferences, an assumption can be confirmed based on which coalition options are evaluated: the closer, the better – ideologically, at least.

Acknowledgements

The authors express their gratitude to the three reviewers and the editors for providing valuable feedback on this article. Their comments have significantly contributed to enhancing the overall quality of the paper. Finally, we would like to acknowledge and thank the participants of the conference of the DVPW 2022 Section on Government Systems and Governance in the Federal Republic of Germany and the panel on Coalitions and Minority Governments at EPSA 2022 for their insightful suggestions. The publication of this article was funded by the Open Access Fund of the Leibniz Association.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by Deutsche Forschungsgemeinschaft: [grant number 491156185]; Fritz Thyssen Stiftung: [grant number 20.21.0.008PO].

Notes

1 Of course this only holds for citizens that voted for one of the coalition parties.

2 The data collection for this article was financially supported by the Fritz Thyssen Stiftung (Az. 20.21.0.008PO). These data were collected as part of a project by Stefan Marschall and L. Constantin Wurthmann at the Heinrich Heine University in Düsseldorf.

3 While minority governments are quite common in other European countries, they are currently not a viable alternative in Germany and their formation is considered to be highly unlikely (Faas and Klingelhöfer 2022; Gross et al. 2023a; Gross et al. 2023b). As the formation of a minority government was considered correspondingly unlikely before the election, it was not queried as part of the survey.

4 The respondents position themselves on the liberal-conservative dimension, on average, in the centre (mean [m] = 5.43 [95% CI = 5.29, 5.57]) with a slight tendency towards the liberal spectrum. Similar patterns are observed on the socio-economic dimension (m = 5.55 [95% CI = 5.43, 5.69]). An average positioning, slightly to the left of centre, has been widely reported for Germany (e.g. Dalton 2008). See Appendix Figure A.2 for more details.