Party competition and the structuring of party preferences by the left-right dimension

Abstract

There is a wide selection of theoretical approaches to explain preferences citizens have for political parties, among them the spatial model of party competition in which voters choose based on proximity in a policy space, such as the left-right dimension. However, it has not ultimately prevailed against its competitors. Thus, a literature has emerged that allows for heterogeneity, asking whose preferences follow this logic and whose do not. However, research on how context affects spatial structuring is still sparse. Therefore, I combine CSES survey data with manifesto data in a sample of established democracies to examine the effects of party competition structure, measured by the “effective” number of parties and the polarization and dimensionality of party positions, on left-right structuration of party preferences in a single model. While I do not find significant context effects with a conventional measure of proximity voting, I propose a different operationalization which shows that while there are systematic effects of the party system, party preferences are mostly quite strongly structured by the left–right dimension.

Keywords

Party competition Voting behavior Proximity voting Left–Right dimension

Introduction

There is a plethora of reasons why a voter might choose one party over the other(s). Consequently, theoretical models that claim to explain voter behavior have proliferated over the decades. One of the major contestants in this arena is the spatial model of party competition (Downs, 1957), which posits one or more continua that map political discourse and on which political actors position themselves. Spatial voting, and other theories likewise, have been studied in great depth, and a lot of work has gone into examining which of them explains voting behavior best. However, despite all this effort, this question has not been answered conclusively. Recently, this has induced a shift of focus from whether the voting behavior of individuals can be predicted by a given theory to the question why some individuals behave in correspondence with a theory and some do not (Rivers, 1988; Lachat, 2008; Singh, 2010). The question is not anymore which theory works best, but under what conditions a given theory works.

This kind of research has produced a lot of insight on how individual traits are a part of these conditions: for instance, as regards spatial voting, political sophistication (i.e., knowledge about and understanding of politics) has been found to be an important correlate of whether party preferences correspond to party positions. How context affects this correspondence, however, is still in the process of being understood. There are relatively few cross-nationally comparative studies on the subject, and those that exist differ considerably in methodological approach and findings. They do, however, demonstrate the need for this kind of work. The explanatory power of left–right positions for party choice does vary over countries (Van der Eijk et al., 1999; Lachat, 2008). This brings the party system and the shape of the political space into focus (Kroh, 2009). The clarity of ideological patterns in a party system arguably has a direct impact on whether citizens can employ these patterns to form a preference (Facchini and Jaeck, 2019). Most importantly for this article, party systems differ with regard to the number of parties, how strongly their positions differ (i.e., how polarized they are, see Dalton 2008), and the number and composition of policy dimensions they compete on (Pennings, 2002; Stoll, 2004; Albright, 2010).

The potential consequences for how well the left-right dimension explains party preferences are straightforward: Having to accommodate more parties on it makes it harder to do so consistently (Wessels and Schmitt, 2008). So does being confronted with parties whose positions are not clearly discernible, that is, not very far apart from each other (Lachat, 2008; Pardos-Prado and Dinas, 2010). Also, as Singh (2010) argues, processing political information within the framework of the left-right dimension becomes more cognitively straining if the political space is not perfectly unidimensional. Thus, the complexity of political space should matter for how strongly it structures citizens’ party preferences. Empirical findings on this relationship, however, are mixed: while Singh does find an effect, Fortunato et al. (2016) do not.

To further elucidate on this matter, I propose a novel operationalization of the extent to which party preferences are structured by the left–right dimension. The agreement between the spatial model and individuals’ actual party preferences is measured as the rank order correlation between actual evaluations of the parties and their perceived proximity on the left–right dimension (which, according to spatial voting theory, is proportional to the level of utility derived from them). As I argue in greater detail below, this measure on the one hand displays more nuance than the commonly used dummy measure that focusses on the party eventually voted for, because it utilizes the entire preference order of a citizen, instead of only the information which party is on top of it. On the other hand, it makes for a more consistent and intuitive analysis than designs that use party–respondent dyads instead of individuals as units of observation, not least because it provides a continuous measure of how well the left–right dimension describes party preferences at the individual level. It also makes testing the influence of all context variables in a single model much easier.

As regards context effects, I take a consistently supply side focused perspective. I focus on the role that political parties play in shaping political discourse and communicating it to citizens. Parties are central actors in the formation of the political space because they transport issues into the political debate (Carmines and Stimson, 1986; Van de Wardt et al., 2014; Hobolt and De Vries, 2015). While in principle, there is a vast number of issues that could be politicized and dimensions these issues could form (Schattschneider, 1975; Robertson, 2006), parties deliberately and strategically pick up some of these issues and combine them in a specific manner, thus forming specific dimensions of political conflict. This makes certain thought patterns more accessible for citizens and increases their inclination to make up their minds about politics in a certain way (Zaller, 1992; Sniderman and Bullock, 2004). Programmatic competition hence plays an important role for the understanding that citizens have of political debate (De Vries et al., 2013). While left–right is a concept that is familiar to a vast majority of citizens in nearly all political systems (Mair, 2007), the actual positions that parties take do not always neatly align on a single dimension, for example, because their stance on a few specific issues does not “fit” their other positions within a general left-right pattern. Whether citizens’ party preferences are structured by the left–right dimension therefore is likely to be influenced by the degree to which the political space, as formed by the parties, clearly resembles this dimension.

I hence analyze the respective effects of the effective number of parties, their spread (i.e., polarization) on the left–right dimension, and their alignment with it (i.e., dimensionality) on left-right structuring of party preferences. I make a point of measuring these variables by use of party-based instead of voter-based data: often, studies in this area operationalize concepts that belong to the political supply side by use of survey data, that is, data gathered on the demand side. This is suboptimal from a conceptual point of view because it departs from the theoretical objective of exploring supply side effects. Moreover, it somewhat muddles genuine context effects and those of the sample composition. By using data measured on the party level, I introduce a clear distinction between the dependent and the independent variables.

I present my analysis and the reasoning underlying it as follows: the following section elaborates on the concepts and theories invoked here, presents the relevant literature, and derives empirical expectations from it. Empirically, I rely on data from the Comparative Study of Electoral Systems (CSES, 2007, 2013) and the Manifesto Project (Volkens et al., 2014). I explain how I process these data in the section after that before moving on to the empirical analysis. I use multilevel regression models that account for the clustering within these data to study the relationship between party system characteristics and the “spatiality” of individual party preferences. Curiously, using the conventional dummy variable operationalization, I do not find systematic context effects. With the correlation-based measure mentioned above, I find that in line with the existing literature, a higher effective number of parties diminishes left–right structuration, while polarization increases it. Dimensionality, if at all, appears to have a negative effect only on the very least politically informed. Generally, these findings notwithstanding, by this measure party preferences seem mostly quite strongly structured by the left–right dimension. The last section summarizes these findings and concludes with a few thoughts on what the results might mean for normative demands that we make on the process of vote choice.

Spatial structuring of party preferences, its correlates, and the shaping of political space

Accounts differ on whether and how issue preferences, and/or ideology, play into vote choice.¹ The classical conception of Downs (1957) sees the distance between policy positions as the sole driver of party-voter linkage. Voters are assumed to vote for the party that is closest to them on a single left-right dimension. This logic is labeled “proximity voting.” Other mechanisms are sociological, group-based electoral alignment (Lipset and Rokkan, 1990 [1967]; Hellwig, 2008) and the concept of partisan identification (Campbell et al., 1960; Green et al., 2002), which holds that rather than actually weighing parties’ programs against each other, voters conceive of supporting a particular party as part of their social identity. Especially partisan identification has long been presented as a counterpart to proximity voting, and the two “pitted” against each other in numerous studies (e.g., Inglehart and Klingemann, 1976; Huber, 1989; Greene, 2004; Abramowitz and Saunders, 2006; Medina, 2015). Last but not least, the spatial model has been criticized based on findings that most voters, except for the most knowledgeable and attentive ones, do not form structured, encompassing belief systems as is required for proximity voting (Converse, 1964; Kinder, 1983). However, given the empirical success of proximity voting and widespread familiarity of citizens with left–right (Mair, 2007), this critique arguably does ultimately not disprove the model.²

Since no single theoretical model has ultimately asserted itself, a more recent strand of literature has shifted the focus to whether there are subgroups of the electorate for whom proximity voting has more or less explanatory power than for others, effectively asking “Who are the spatial voting violators?” (Boatright, 2008). A lot of this research deals with the US case, focusing on individual characteristics. Spatial voting, it has been found, is for instance more prevalent among the politically knowledgeable (ibid.). Joesten and Stone (2014) have examined the role of other decision-making mechanisms in this context and distinguish two different kinds of them: The first are “facilitators”, meaning that they correlate positively with proximity voting.³ The other kind are mechanisms that are genuine alternatives to political positions, that is, some respondents follow these cues instead of party platforms. For these mechanisms, the correlation should be negative. There are examples of both kinds of mechanisms in the empirical literature: spatial voting is more prevalent among those with a partisan identification (Simas, 2013; Joesten and Stone, 2014) and weakened by alternative information cues such as personal information about candidates (Boudreau et al., 2013; Joesten and Stone, 2014).

While comparative work that looks at the effects of the decision context is less prevalent, a lot of work suggests that it does play an important role. The explanatory power of policy positions for vote choice, that is, the extent of proximity voting, has often been found to vary across countries (Granberg and Holmberg, 1988; Van der Eijk et al., 1999, 2005). As Wessels and Schmitt (2008) argue, this variation is primarily linked to structures of political supply that do or do not provide “meaningful choice sets,” that is, that contain a number of competitors that are “distinguishable in terms of ideology and/or in terms of competence” (20). The significance of political supply is also emphasized by the wider literature on voter behavior: most voters do not pay a lot of attention to politics and do not make up their mind about it in great detail. Through their programmatic platforms, parties serve as “information providers” (Lupia, 1994): by issuing and bundling policy statements (cf. Budge et al., 2001; De Vries et al., 2013), they set landmarks on the political map which anchor and make available certain ideas in voters’ minds (Zaller and Feldman, 1992; Zaller, 1992; Sniderman and Bullock, 2004). This enables voters to infer positional patterns that they can use as a heuristic or “information shortcut” in lieu of detailed political knowledge (Downs, 1957).

Accordingly, one context factor that has been studied is party system polarization. The general argument of the respective literature is that if parties present clearly discernible policy bundles, voters find it easier to cast their voting decision in accordance with left–right positions (Lachat, 2008; Kroh, 2009). Another recurrent finding of this literature is that a higher number of parties correlates negatively with proximity voting, presumably because it is cognitively more straining to consistently position a great number of parties on a left–right dimension. Lastly, while relatively little researched, the dimensionality of political competition should be an important factor: The yardstick of proximity voting, across the existing literature, is the left–right dimension (Downs, 1957; Fuchs and Klingemann, 1990; Huber and Powell, 1994). Thus, if parties play an important role in communicating the political space to voters, and the extent to which voters’ preferences follow the spatial paradigm is measured against a unidimensional political space, parties are implicitly assumed to adhere to this unidimensional conception.

On the party side, however, left and right is increasingly seen to be of varying empirical reach. It has repeatedly been found that a single dimension does not always suffice to describe party positions (Pennings, 2002; Warwick, 2002; Dalton, 2015), and that dimensionality varies across time (Albright, 2010) and space (Stoll, 2011). Even the meaning of left-right itself shows a lot of variation across systems (Franzmann and Kaiser, 2006; Rovny and Edwards, 2012). A rich literature shows the prominent, and proactive, role of parties and their competing with each other in establishing dimensions of the political space (Riker, 1982; Carmines and Stimson, 1986; Elias et al., 2015; Lee and Schutte, 2017). Specifically, parties strategically choose to emphasize or to remain silent on certain issues (Robertson, 1976; Budge and Farlie, 1983). These issues are “the smaller pieces from which ideological dimensions are constructed” (Warwick, 2002: 104), and, depending on their composition, give rise to specific kinds of dimensions. Because of this contingency however, the political space could in theory have all kinds of dimensions. On the basis of this reasoning, the left–right dimension and spatial voting will be much less functional as a decision-making mechanism for voters the less political discourse aligns to a pattern that is, in fact, unidimensional. However, empirical evidence on this relationship is both scarce and inconclusive. Two studies, to my knowledge, test for such an effect: while Singh (2010) finds that in party systems where voters’ evaluations of parties are harder to map unto a single dimension, voters have a lower probability to vote in line with the proximity paradigm, Fortunato et al. (2016) find no effect of the complexity of party competition on how strongly party preferences are guided by left–right positions. Among the context factors discussed here, dimensionality therefore is arguably the one that most calls for more research.

Of course, some voters are better prepared to find their way around the political space than others. A classical variable in this regard is political sophistication, that is, how well versed a person generally is in the realm of politics. In the case of the left–right dimension, we might hope that this relatively simple heuristic makes it easier for rather unsophisticated citizens to make political decisions. However, findings as early as those of Converse (1964) suggest otherwise. In fact, it might be that “[i]ronically, heuristics are most valuable to those who might in fact need them least” (Lau and Redlawsk, 2001: 967). This would mean that not only do politically more sophisticated voters have party preferences more strongly structured by the left–right dimension by themselves, but they also make use of the cues from their environment more efficiently. This should manifest in them being less sensitive to inhibitive context factors and more receptive to facilitating ones. Indeed, the effect of party system polarization on proximity voting seems to be stronger for political “experts” (Lachat, 2008). The number of parties and the dimensionality of party competition, on the other hand, presumably are not as straining for politically more adept individuals. Conversely, individuals who are less acquainted with politics would be less aided in structuring political debate by polarization, and more challenged by a higher number of parties and higher dimensionality. Below, I test this heterogeneity of effects through an interaction between the respective context variables and the level of political information. In terms of statistical findings, the arguments presented above imply a positive coefficient on the product term between each context variable and political information, such that higher information drives the effects of the number of parties and dimensionality towards zero and that of polarization away from it. The hypotheses to be tested thus are the following:

The higher the number of effective parties, the less strongly party preferences will be structured by the left–right dimension.

The more polarized the party system, the more strongly party preferences will be structured by the left–right dimension.

The more difficult it is to map party positions on a single dimension, the less strongly party preferences will be structured by the left–right dimension.

H4a

The more politically informed individuals are, the less strongly will the effective number of parties correlate with the degree to which their party preferences are structured by the left-right dimension.

H4b

The more politically informed individuals are, the more strongly will party system polarization correlate with the degree to which their party preferences are structured by the left-right dimension.

H4c

The more politically informed individuals are, the less strongly will the dimensionality of political space correlate with the degree to which their party preferences are structured by the left–right dimension.

Data and operationalization

To comprehensively examine the political space’s effect on the structuring capacity of the left-right dimension for individuals’ party preferences, I use the effective number of parties, polarization, and dimensionality as indicators, as discussed in the preceding section. They are operationalized through data derived from the Manifesto Project (MARPOR) data set (Volkens et al., 2014).⁴ The MARPOR data set classifies all the statements a party makes in its manifesto into 56 issue categories via quantitative content analysis and computes the salience score, that is, the importance of a given issue in that manifesto as the ratio between the number of statements made on that issue and the sum of all statements made in the manifesto. Because they are based on party manifestos, these data consistently capture political supply, unlike other data that are based on, for example, survey data aggregates and thus could be affected by characteristics of the political demand side.

I combine these data with survey data from the second and third module of the Comparative Study of Electoral Systems (CSES, 2007, 2013). They allow measuring the extent to which party preferences align with positions on the left–right dimension in a novel and, I would argue, more comprehensive fashion than before, which is presented in greater detail below. All in all, I arrive at a final sample of 50,562 respondents, nested in 47 election studies, which were carried out between 2001 and 2011. A list of the elections covered, as well as descriptive statistics of the variables in the analysis can be found in the appendix.

Some adjustments are made for theoretical reasons, but also because of considerations regarding data coverage. The first is predicated on the democratic experience of the countries in the sample. While there is important work on how the left–right dimension is handled in new and emerging democracies, this work also highlights that party-voter linkages work quite differently there than in established ones (Zechmeister, 2006; Ruth, 2016). Since the arguments above rest so much on the proposition of political parties freely strategizing and contesting each other, it seems fair to assume that this contest needs to have had some time to evolve before these effects can unfold (also see Singh, 2010). I therefore exclude countries which have not continuously been rated “free” in the Freedom House (2014) data base from at least 1995 on (with the second CSES module starting in 2001, this corresponds to at least 6 years of unhindered democratic competition). In the relevant data, this applies to Romania and Slovakia. Also, since the parties covered in the MARPOR and the CSES data are not the same in all instances, I harmonize the data sets by including only those parties contained in both of them. This primarily affects very small parties not included in the CSES, and specifically the election in Italy of 2006 (because there was a large number of parties forming two electoral alliances) and that in Spain of 2004 (because there are a lot of only regionally competing parties, which I exclude); a detailed list of the included parties, as well as any other coding decisions, can be found in the appendix. Finally, most of the MARPOR data points can directly be linked to the election studies in the CSES, except for Japan, where surveys were conducted on occasion of upper house elections, while the manifesto data cover lower house elections. Thus the survey data obtained in 2004 and 2007 are combined with party system data measured in 2003 and 2005, respectively.

Dependent variable: left–right structuring of party preferences

To study how much party preferences align with the spatial voting model, two concepts have to be related to each other: the preferences a citizen actually has and those she would have if she behaved completely according to the model. Earlier studies of left–right spatial voting have primarily solved this problem in two different ways. The first is to combine the two concepts in a dummy variable that indicates whether the actual party choice coincides with the one that the spatial voting paradigm predicts (Boatright, 2008; Singh, 2010; Joesten and Stone, 2014). Conversely, the other separates measured and implied preferences. Here, variables such as vote choice, propensities to vote or “electoral utilities” are employed as the dependent variable and the utility implied by spatial voting theory (i.e., proximity) as the independent variables, for instance, in a conditional logit model (Blais et al., 2001; Lachat, 2008; Wessels and Schmitt, 2008; Jessee, 2009, 2010; Pardos-Prado and Dinas, 2010; Boudreau et al., 2013; Simas, 2013). This specification requires that voter-party dyads, instead of voters, are chosen as units of analysis. Either of the two, however, leaves room for improvement in my opinion.

The dummy variable captures the essence of spatial voting theory, but loses a lot of information by only assessing its fit in reference to the party the voter eventually chooses. It may not even say much about to what extent left–right is the standard by which voters assess the political offer: It is both possible that the spatial voting rule coincides with vote choice, but does not structure the evaluation of parties as a whole, and that a voter who generally thinks very much in terms of proximity does not vote for the closest party (e.g., for strategic reasons, see Blais et al. 2001). I give an example of this possibility below. Moreover, working with vote choice censors the sample in a specific way: By design, this measure can only be used for respondents who cast a ballot. This would not necessarily be a problem for a model with a narrow focus on explaining (or forecasting) vote choice. However, if the focus is on the circumstances under which the proximity logic is more or less valid, it turns on the more general question on how individuals (voters or not) think about politics and how they form party preferences. Given that voters’ characteristics arguably differ vastly from those of non-voters, an answer to this question should include both groups, especially in times of declining turnout in many established democracies. As Adams et al. (2006) have found, turnout is susceptible to the format of party competition. As far as electoral participation is regarded to be normatively desirable, it is thus also an important question whether non-voters can rely on the left–right dimension as an information shortcut, which would reduce the costs of participation.

The dyadic regression approach includes more information, since it typically takes a citizen’s full preference profile into account (Lachat, 2008) and (if using propensity to vote) can also include non-voters. However, it makes examining the contextual correlates of conformity with the spatial model less intuitive precisely because the concept of interest is moved to the right-hand side of the equation: Transforming the data set into a “stacked data matrix” detaches the unit of observation (i.e., the respondent) from the unit of analysis. Likewise, the structuration of party preferences ceases to be a characteristic measured on the individual level and becomes a parameter to be estimated for the sample as a whole (i.e., the coefficient of the proximity variable).⁵ To analyze conditioning factors thus requires extensive use of interaction terms. For example, to model the context effects, and their interaction with political information, equivalently to as it is done here would require three-way interactions where now two-way interactions suffice, and thus 15 regressors instead of seven (cf. Brambor et al., 2006).

Because of these issues, I use a different measure for spatial party preferences that incorporates all the information from respondents’ preference profiles but at the same time retains the straight-forwardness of directly using the structuring of party preferences as the dependent variable, thus combining the advantages of either operationalization. To do so, I make use of the rich information available in the CSES data. I combine the actual party preference orderings of respondents with those that the distances to the parties imply and compute the variable proximity consistency as Spearman’s rank order correlation coefficient (ρ) between these orderings. This provides, for each respondent, a continuous measure of how close her actual party preferences are to predicted ones. For the actual preference profiles, I use party sympathy ratings, which can be seen as “super-generalizations” of different specific evaluations of parties (Wessels and Schmitt 2008, also see Blais et al. 2001; Singh 2010).⁶

This operationalization of course requires that respondents placed themselves on the left-right dimension and provided ratings and placements for a sufficient number of parties. I set the minimum for this number to three. If a respondent gave either the same placement or rating for all the parties, the measure is set to missing, since Spearman’s ρ cannot be computed in these situations. Being a non-parametric statistic, Spearman’s ρ captures how respondents order parties, but is insensitive to how the responses that underlie this order are distributed. That relaxes the assumptions required about how respondents interpret the scales used to answer the two items. It also makes it dispensable to specify a utility function: accounts differ on whether citizens should be assumed to have a quadratic or a linear utility function (Singh, 2014). Since this measure relies only on ordinal data, it can be agnostic in this regard, unlike the electoral utilities approach discussed above. I also recreate the commonly used dummy variable via this data basis to have a benchmark measure.⁷

To take a deeper look at the properties of the proximity consistency indicator, consider, as an example, a hypothetical voter v with a left-right-position lr = 4 on a scale from zero to 10 within a system of five parties with the following positions: A = 3, B = 6, C = 1, D = 8, E = 9. Table 1 explores the results the measure could yield assuming different rankings in terms of preference that v might come up with. In the upper row, the respective ranking is rather similar to the ranking by distance, leading to a high value of proximity consistency. In the bottom row, the preference ranking deviates quite a bit from the one based on left-right distance, which produces a low value. The columns showcase a phenomenon mentioned above: they illustrate how preference rankings can be constructed which for both spatial voting violators and conformers show the same proximity consistency.

Table 1.

Construction of the proximity consistency variable, illustrated for a hypothetical voter v with different preference profiles and parties A, B, C, D, and E.

Positions
	C		A	v		B		D	E
0	1	2	3	4	5	6	7	8	9	10
			Proximity conformer (d = 1)					Proximity violator (d = 0)
High			Scenario 1 (voted for party A)					Scenario 2 (voted for party B)
Proximity			Party ranking by…					Party ranking by…
Consistency			LR distance: A, B, C, D, E					LR distance: A, B, C, D, E
(ρ = 0.9)			Sympathy: A, C, B, D, E					Sympathy: B, A, C, D, E
Low			Scenario 3 (voted for party A)					Scenario 4 (voted for party C)
Proximity			Party ranking by…					Party ranking by…
Consistency			LR distance: A, B, C, D, E					LR distance: A, B, C, D, E
(ρ = 0.2)			Sympathy: A, D, E, B, C					Sympathy: C, D, A, B, E

Context variables: the structure of party competition

For two of the three context parameters, there are already fairly standard ways to measure them: I use the effective number of parties (ENP, Laakso and Taagepera 1979) and Taylor and Herman’s (1971) index of polarization, based on the left-right scores computed by Franzmann and Kaiser (2006) as party positions.⁸ Operationalizing the dimensionality of party competition is a more intricate task. Coming up with a measure of dimensionality not only includes many conceptual and methodological choices (for a discussion see Stoll 2011); another challenge is to devise a measure on the election level. Nearly all existing and available dimensionality measures (e.g. Lijphart 1999; Nyblade 2004; Stoll 2004; Bakker et al. 2012; Ganghof et al. 2014), to the best of my knowledge, are time-invariant. Singh’s (2010) approach avoids this problem by employing multidimensional scaling (MDS) to party sympathy ratings of the CSES respondents, thus arriving at a measure of how well these ratings can be represented by a single dimension. However, this technique measures voter-defined instead of party-defined dimensionality (Stoll, 2011). As I argued above, it is important to differentiate party-defined and voter-defined measures. This is especially important in the case of dimensionality, because otherwise, both the independent and the dependent variable eventually are based on the party preferences of the respondents (albeit on two different levels of aggregation). Of course, it is reasonable to assume that party-defined and voter-defined dimensionality corresponds to each other. But to test this assumption and the robustness of Singh’s findings, it is ultimately necessary to use a dimensionality measure that is based on data that map party positions instead of voter evaluations.

The approach I use here, apart from the harmonization of parties covered both in the CSES and the MARPOR data, is analogous to that of Reinermann and Barbet (2020). The measure is methodologically analogous to Singh’s in that it uses MDS, but based on the relative issue saliences within the parties’ manifestos as measured by the MARPOR data. MDS uses the differences in salience scores between parties as measures of “dissimilarity” between them and tries to reproduce these dissimilarities in a space of given dimensionality by an iterative placing algorithm. In the process, the method delivers an index called Stress, which indicates how much the salience differences as reproduced by an MDS solution deviate from the original differences. Like Reinermann and Barbet, I use the Stress resulting from a unidimensional MDS model as a proxy for the dimensionality of political space. As they point out, this also has a methodological advantage over many conventional approaches used to uncover latent dimensions: As Warwick (2002) and Stoll (2004) observe, factor analytic methods for example are not well suited for manifesto data, since these data often are not linearly correlated. As Van der Brug (2001) has pointed out, this is due to the salience-theoretic conception of the data and can be remedied by the use of MDS.

Individual-level variables: personal traits and alternative decision mechanisms

Hypotheses 4a-c posited that political sophistication moderates the impact of context factors on the structuration of voter preferences by the left-right dimension. Thus, on the micro level the central independent variable is political information. I make use of three items in the CSES probing factual knowledge about the political system a respondent lives in. I follow Singh (2015), who cautions that because the political knowledge questions asked in the CSES could differ in difficulty across countries, raw counts of correct answers might not be comparable between them. He proposes to standardize the variable by dividing the individual number of correct answers by the country-specific mean. I subtract one from the result so that a value of zero indicates the country-specific mean. Another variable of interest is whether or not a respondent voted in the last election. As mentioned above, including this variable is not possible using the dummy variable operationalization of proximity voting, since that variable can only be constructed for those who actually took part in that election. As Facchini and Jaeck (2019) argue, the structure of party competition and the decision to turn out are themselves intertwined, making the inclusion of non-voters in the analysis all the more important. Moreover, since casting a ballot arguably requires some interest in the political process, it is an important measure of how much attention citizens pay to and how well versed or experienced they are in the political sphere.

To cancel out election-sample composition effects, I control for the standard socio-demographics age, education and whether a respondent is female. Education is operationalized in the CSES by an eight-level ordinal variable, which I employ as provided. Mechanisms other than left-right proximity that might be available to respondents to establish a preference order are captured by whether or not the respondent identifies with a specific party⁹ and her evaluation of the government’s performance (to include retrospective voting).¹⁰ These variables are not only likely to be related to the spatiality of party preferences, as was discussed above, but also to characteristics of the party system: although partisan identification is an aspect of social identity much more than political positioning (Green et al., 2002), it has been found to be more intense in more polarized party systems (Lupu, 2015). Retrospective voting on the other hand can be assumed to be more pervasive in party systems where accountability for political outcomes can be more clearly assigned to a specific actor, as has for example been found in the economic voting literature (Anderson, 2000). On the individual level, it is conceivable that politically better informed respondents are also more likely to form partisan attachments and/or explicit opinions about the performance of their government.

Analysis

With regards to the dependent variable, I argued in favor of measuring the spatiality of party preferences using the correlation between actually measured favorability of parties and their utility implied by spatial theory. To give a first impression of the insights this operationalization has to offer, Figure 1 depicts its distribution in relation to the more traditional operationalization by dummy variable among those respondents for whom both measures are available, across the election studies used in this article. While it generally shows a lot of variation in proximity consistency, it also shows that proximity voting violators’ party preferences are much more structured by the left-right dimension than the term might lead one to assume. Their median proximity consistency lies above 0.25 in 43 of 47 election studies and above 0.5 in 16 (violators’ overall median proximity consistency is 0.46). This means even respondents whose vote choice itself does not conform to spatial voting mostly seem to entertain party preferences that by and large do so. Relatedly, as can be seen in the comparison of the variable’s distribution among spatial voting “violators” and “conformers,” respectively, the distinction between the two groups may be more clear-cut than the underlying party preference profiles warrant: in virtually all countries, there is considerable overlap between the two distributions. This underscores empirically a theoretical possibility pointed out in Table 1: individuals can be classified as conformers or violators although the underlying preference profiles are not that different. This is especially the case in the multiparty systems of Scandinavia (most pronounced in Finland), while the separation is stronger in, for example, Australia, Spain, Great Britain, and Italy.¹¹ Still, the distributions are sufficiently distinct in all cases to suggest that the variable measures what it is intended to.

Figure 1.

Distribution of proximity consistency for proximity voting violators (light) and conformers (dark), by election. Whiskers = median ±1.5 interquartile ranges. Compiled using the software package by Wickham (2016).

How well citizens’ preference patterns fit the left–right dimension is indeed “menu dependent” (Sniderman and Bullock, 2004) in that they are affected by characteristics of the party system, as can be seen in Figure 2: The three panes show how the mean of citizens’ proximity consistency in a specific election relates to each of the three context level variables discussed above. They show these relationships to be in line with the expectations developed above: the degree to which party preferences are structured by the left–right dimension appears to correlate positively with party system polarization, but negatively with the effective number of parties and the dimensionality of party competition. Interestingly, election studies tend to cluster by country (which seems to further increase the plausibility of the proximity consistency measure) but still display some within-country variation.

Figure 2.

Party competition structure as measured by ENP, Polarization and Dimensionality, against average levels of proximity consistency, per election. Compiled using the software package by Wickham (2016).

To back this up with a more rigorous analysis that takes into account micro-level characteristics, I estimate varying-intercept multilevel regressions. All models are specified with three levels (respondents nested in elections nested in countries). Table 2 shows the results of these analyses. Models 1–5 use the continuous, correlation-based measure of proximity consistency and therefore are estimated by linear regression; models 6–10 employ the more traditional dummy and thus are logistic regression models. Note that goodness-of-fit measures, such as information criteria or the log-likelihood, are not reported because the linear models are estimated using restricted maximum likelihood (REML).¹² Either of these groups proceeds in the same fashion, starting with an empty model to give an impression of how the variance in the dependent variables manifests itself within and across the different levels, and then including the micro level variables and each of the party system variables on its own in an individual estimation, before concluding with a full model that comprises all the party system variables. They are interacted with political information consistently in all models.

Table 2.

Varying-intercept multilevel models for proximity consistency of party preferences (1–5) and proximity dummy (6–10), with interaction effects for context variables and political information. [C] = grand mean centered. Political Information operationalized according to Singh (2015). Computed and compiled using software packages by Bates et al. (2015) and Leifeld (2013).

	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 8	Model 9	Model 10
Intercept	0.518***	0.367***	0.370***	0.368***	0.368***	0.095	−0.254**	−0.246**	−0.249**	−0.254**
	(0.020)	(0.020)	(0.020)	(0.019)	(0.018)	(0.078)	(0.087)	(0.080)	(0.080)	(0.092)
Age/100 [C]		0.019	0.018	0.018	0.019		0.055	0.054	0.053	0.055
		(0.013)	(0.013)	(0.013)	(0.013)		(0.061)	(0.061)	(0.061)	(0.061)
Education [C]		0.017***	0.017***	0.017***	0.017***		0.020**	0.020**	0.020**	0.020**
		(0.001)	(0.001)	(0.001)	(0.001)		(0.006)	(0.006)	(0.006)	(0.006)
Female		−0.000	−0.000	−0.000	−0.001		−0.005	−0.005	−0.005	−0.005
		(0.004)	(0.004)	(0.004)	(0.004)		(0.020)	(0.020)	(0.020)	(0.020)
Partisan identifier		0.129***	0.129***	0.129***	0.129***		0.614***	0.615***	0.614***	0.614***
		(0.004)	(0.004)	(0.004)	(0.004)		(0.020)	(0.020)	(0.020)	(0.020)
Perceived gov’t perform. [C]		−0.023***	−0.023***	−0.023***	−0.023***		0.002	0.002	0.002	0.002
		(0.003)	(0.003)	(0.003)	(0.003)		(0.014)	(0.014)	(0.014)	(0.014)
Voted		0.089***	0.090***	0.089***	0.089***
		(0.007)	(0.007)	(0.007)	(0.007)
Political information		0.050***	0.047***	0.049***	0.046***		0.099***	0.100***	0.098***	0.097***
		(0.004)	(0.004)	(0.004)	(0.004)		(0.019)	(0.019)	(0.018)	(0.019)
ENP [C]		−0.030*			−0.034*		−0.087			−0.113
		(0.013)			(0.014)		(0.067)			(0.080)
Political info. x ENP		−0.001			−0.008*		0.003			−0.011
		(0.003)			(0.003)		(0.013)			(0.016)
Polarization [C]			0.011		0.021*			0.005		0.054
			(0.011)		(0.010)			(0.046)		(0.058)
Political info. x Polarization			0.005*		0.008***			0.001		0.007
			(0.002)		(0.002)			(0.009)		(0.011)
Dimensionality [C]				−0.180⁺	−0.106				−0.592	−0.442
				(0.095)	(0.095)				(0.415)	(0.392)
Political info. x Dimensionality				0.056	0.119**				0.324⁺	0.395*
				(0.034)	(0.039)				(0.174)	(0.196)
Dependent variable	Cont	Cont	Cont	Cont	Cont	Dummy	Dummy	Dummy	Dummy	Dummy
Num. Obs	50 562	50 562	50 562	50 562	50 562	45 361	45 361	45 361	45 361	45 361
Num. Elections	47	47	47	47	47	47	47	47	47	47
Num. Countries	24	24	24	24	24	24	24	24	24	24
Variance: Election	0.002	0.002	0.002	0.002	0.002	0.042	0.031	0.041	0.037	0.024
Variance: Country	0.008	0.007	0.007	0.007	0.006	0.116	0.151	0.120	0.121	0.177
Variance: Residual	0.214	0.206	0.206	0.206	0.205	1.000	1.000	1.000	1.000	1.000

***p < 0.001, **p < 0.01, *p < 0.05, +p < 0.1.

As discussed above, contrary to the dummy measure, the continuous operationalization of preference structuration allows the inclusion of non-voters, which (albeit by coincidence of survey item response rates), increases sample size. At the same time, it retains the relative simplicity with regard to model specification of using the dummy measure as the dependent variable, as opposed to, for example, dyad data. Turning to the estimated effects, results on the micro level are plausible and mostly reproduce those arrived at in the literature. Respondents with higher levels of education and/or political information have party preferences that are more closely aligned with left–right placements, while gender and age have no or only scant effect. The structuring effect of left-right is stronger for partisan identifiers, while the findings with regard to respondents’ evaluation of government performance are inconsistent. In models 2–5, it has a significant, negative effect, as implied by the alternative linkage perspective discussed above. In models 7–10, the coefficient is essentially zero. This is likely because the dummy measure, for a given respondent, either is based on a government or an opposition party, whereas the continuous measure maps out deviations from left-right structuration brought about by the appreciation of the government parties. The newly introduced voted dummy has a statistically significant coefficient in models 2–5.

Taking into regard how left–right structuration differs across context, the continuous operationalization gives a much more conservative picture than the dummy variable operationalization, indicating that a share of 4.5% of its variance can be explained by the clusters as opposed to 13.6%. Strikingly, however, the effects of party system structure overall seem to turn up more clearly (in terms of statistical significance) in the linear models. The effects themselves mostly turn out as expected: on a purely descriptive level, a higher effective number of parties and increased dimensionality turn out to hinder left–right structuration of party preferences, while polarization aids it. Politically better informed respondents mostly appear to be better equipped to deal with these party system attributes. One curious exception to this pattern is the interaction effect between the effective number of parties and political information. While the evidence for the existence of such interaction effects is not particularly strong (see below), it might be worthwhile to explore this finding further, especially since this effect, to my knowledge, has not been examined in the literature. It may well be that this is a genuine effect and that politically more sophisticated citizens do make more exceptions from the proximity rule the more options they are presented with, controlling for other context factors.

Although the respective coefficients are, in part, statistically significant, graphical inspection of the interaction effects shows that this only manifests itself across the entire empirical range of political information (Figure 3), so that a statistically significant difference in effects arises only between the politically very uninformed and highly informed. Since being a non-voter might also moderate the context effects taken into perspective here, for instance because non-voters engage with politics to a smaller extent or not at all, I re-estimate models 2–5 with the cross-level interactions between the party system context variables and whether or not a respondent voted. However, as models 11–14 in Table 3 (as well as the right column of Figure 3) show, while the coefficients do indeed present a comparable pattern as those in models 2–5, they do not provide evidence of interaction effects.

Figure 3.

Effect of ENP, polarization, and dimensionality on proximity consistency, contingent on level of political information (left, based on model 5) an on whether a respondent voted (right, based on model 14). Shaded areas indicate 95% confidence intervals. Compiled using the software package by Solt and Hu (2018).

Table 3.

Varying-intercept multilevel models for proximity consistency of party preferences, with interaction effects for context variables and electoral participation. [C] = grand mean centered. Political Information operationalized according to Singh (2015). Computed and compiled using software packages by Bates et al. (2015) and Leifeld (2013).

	Model 11	Model 12	Model 13	Model 14
Intercept	0.368***	0.371***	0.368***	0.369***
	(0.020)	(0.020)	(0.019)	(0.018)
Age/100 [C]	0.019	0.018	0.018	0.018
	(0.013)	(0.013)	(0.013)	(0.013)
Education [C]	0.017***	0.017***	0.017***	0.017***
	(0.001)	(0.001)	(0.001)	(0.001)
Female	−0.000	−0.000	−0.000	−0.000
	(0.004)	(0.004)	(0.004)	(0.004)
Partisan identifier	0.129***	0.129***	0.129***	0.129***
	(0.004)	(0.004)	(0.004)	(0.004)
Perceived gov’t perform. [C]	−0.023***	−0.023***	−0.023***	−0.023***
	(0.003)	(0.003)	(0.003)	(0.003)
Political information	0.049***	0.049***	0.050***	0.049***
	(0.004)	(0.004)	(0.004)	(0.004)
Voted	0.089***	0.089***	0.089***	0.087***
	(0.007)	(0.007)	(0.007)	(0.007)
ENP [C]	−0.036**			−0.029⁺
	(0.014)			(0.015)
Voted x ENP	0.007			−0.004
	(0.006)			(0.007)
Polarization [C]		0.004		0.011
		(0.011)		(0.011)
Voted x polarization		0.008*		0.011*
		(0.003)		(0.004)
Dimensionality [C]			−0.195⁺	−0.188
			(0.110)	(0.117)
Voted x dimensionality			0.017	0.087
			(0.063)	(0.074)
Dependent variable	Cont	Cont	Cont	Cont
Num. Obs	50,562	50,562	50,562	50,562
Num. Elections	47	47	47	47
Num. Countries	24	24	24	24
Variance: Election	0.002	0.002	0.002	0.002
Variance: Country	0.007	0.007	0.007	0.006
Variance: Residual	0.206	0.206	0.206	0.206

***p < 0.001, **p < 0.01, *p < 0.05, +p < 0.1.

To complete the picture drawn by these analyses, I turn to models without interactions, which are presented in Table 4, arranged in the same fashion as in Table 2. While the effects in the logistic regression, again, turn out insignificant, the linear models (with proximity consistency as the dependent variable) return a similar configuration of context effects as their respective counterparts in Table 2, evidencing an effect of both ENP and polarization.

Table 4.

Varying-intercept multilevel models for proximity consistency of party preferences (15–18) and proximity dummy (19–22), without interaction effects. [C] = grand mean centered. Political Information operationalized according to Singh (2015). Computed and compiled using software packages by Bates et al. (2015) and Leifeld (2013).

	Model 15	Model 16	Model 17	Model 18	Model 19	Model 20	Model 21	Model 22
Intercept	0.367***	0.370***	0.368***	0.368***	−0.254**	−0.246**	−0.249**	−0.254**
	(0.020)	(0.020)	(0.019)	(0.018)	(0.087)	(0.080)	(0.080)	(0.092)
Age/100 [C]	0.019	0.018	0.018	0.018	0.055	0.054	0.054	0.054
	(0.013)	(0.013)	(0.013)	(0.013)	(0.061)	(0.061)	(0.061)	(0.061)
Education [C]	0.017***	0.017***	0.017***	0.017***	0.020**	0.020**	0.020**	0.020**
	(0.001)	(0.001)	(0.001)	(0.001)	(0.006)	(0.006)	(0.006)	(0.006)
Female	−0.000	−0.000	−0.000	−0.000	−0.005	−0.005	−0.005	−0.005
	(0.004)	(0.004)	(0.004)	(0.004)	(0.020)	(0.020)	(0.020)	(0.020)
Partisan identifier	0.129***	0.129***	0.129***	0.129***	0.614***	0.615***	0.614***	0.614***
	(0.004)	(0.004)	(0.004)	(0.004)	(0.020)	(0.020)	(0.020)	(0.020)
Perceived gov’t perform. [C]	−0.023***	−0.023***	−0.023***	−0.023***	0.002	0.002	0.002	0.002
	(0.003)	(0.003)	(0.003)	(0.003)	(0.014)	(0.014)	(0.014)	(0.014)
Voted	0.090***	0.090***	0.090***	0.089***
	(0.007)	(0.007)	(0.007)	(0.007)
Political information	0.050***	0.050***	0.050***	0.050***	0.100***	0.100***	0.100***	0.100***
	(0.004)	(0.004)	(0.004)	(0.004)	(0.018)	(0.018)	(0.018)	(0.018)
ENP [C]	−0.030*			−0.034*	−0.087			−0.113
	(0.013)			(0.014)	(0.067)			(0.080)
Polarization [C]		0.011		0.021*		0.005		0.054
		(0.011)		(0.010)		(0.046)		(0.058)
Dimensionality [C]			−0.180⁺	−0.106			−0.593	−0.443
			(0.095)	(0.095)			(0.415)	(0.391)
Dependent variable	Cont	Cont	Cont	Cont	Dummy	Dummy	Dummy	Dummy
Num. Obs	50,562	50,562	50,562	50,562	45,361	45,361	45,361	45,361
Num. Elections	47	47	47	47	47	47	47	47
Num. Countries	24	24	24	24	24	24	24	24
Variance: Election	0.002	0.002	0.002	0.002	0.031	0.041	0.037	0.024
Variance: Country	0.007	0.007	0.007	0.006	0.151	0.120	0.121	0.177
Variance: Residual	0.206	0.206	0.206	0.206	1.000	1.000	1.000	1.000

***p < 0.001, **p < 0.01, *p < 0.05, +p < 0.1.

A very noteworthy point that remains to be raised regarding all the findings presented here is the substantive significance of the context effects. While the structure of party competition exerts systematic, and, I would venture, important “menu dependence” effects, their scope appears somewhat limited; they are relatively small in size and/or only apply to specific groups. That large, sweeping effects are largely absent is also due to the fact that as pointed out above, the level of left–right structuration of party preferences is mostly high: the intercepts of models 1–5 can be interpreted as the conditional expected correlation between a respondent’s party evaluations and ratings of perceived closeness (setting the independent variables to zero). For instance, model five implies a value of about 0.37 for a male non-voter who is not a partisan identifier, with all the other variables at their respective means. To brand this (hypothetical) respondent a “spatial voting violator” would be quite a broad stroke; rather, he can be seen as “moderately conforming.”

Conclusion

The debate what makes voters opt for this or the other party has been and is ongoing. It is not only an interesting debate for prediction of electoral behavior, but also speaks to aspects that are important for the functioning of democracy itself, such as whether citizens’ preferences are adequately represented by the political system. From that vantage point, we might hope that they are presented with different “verbal images of the good society” (Downs, 1957)—that is, the party platforms—and weigh them against each other before they decide which of these images they buy into. But just as well, they might follow a much simpler conception of democracy, merely voting out of office politicians who in some way “offended” them (Riker, 1982). Lastly, it may be that vote choice is not much of a choosing after all, because it is driven by largely time-invariant characteristics like belonging to certain demographics or early developed personality traits.

The notion that all of these mechanisms may be at work in the same electorate is a relatively new one, more so the interest in the question which of a citizen’s attributes explain what mechanism her preferences conform with. Early advances on this topic like Rivers (1988) notwithstanding, empirical work on who does and does not vote in line with left–right mostly came up only very recent. This is all the more true for work that looks at the context a voter is situated in. In this paper, I attempted to offer a comprehensive account of the effects of political supply. Recognizing the important role that political parties play in “building” the political space, I tried to incorporate an encompassing description of the political space into my analysis, built on data measured at the party level. To do that I combined rather established concepts, that is, the effective number of parties and their polarization, with the concept of dimensionality, which has only very recently been taken into perspective (Singh, 2010; Fortunato et al., 2016). I then examined all these potential context effects in a comprehensive model.

This focus on party-based operationalizations of the context variables is one of two contributions of this article: discerning the political supply and demand side is important not only because the same variable can have different effects on either side (Pardos-Prado and Dinas, 2010), but also because it takes the complexities of the interaction between the two sides serious. I therefore strongly advocate for avoiding conceptually incongruent measures, which could lead to flawed inferences.

The second contribution regards the measurement of the dependent variable. Using the rank correlation coefficient between the policy distances a respondent sees between herself and the parties and her sympathy ratings for these parties as an operationalization of proximity consistency, I could show that beyond a crude division of respondents into “violators” and “conformers,” there is a lot of nuance to be found in their party preference profiles. Specifically, it became apparent that even violators’ preference profiles are often very much structured by the left–right dimension. This underscores Boatright’s (2008) conclusion that proximity voting theory is neither totally wrong nor totally right (124).

A very surprising finding regarding this specification is that while it varies much less across contexts than the more common dummy operationalization, the effects of variables situated on the context levels turn up much more systematically. At the same time, as regards the micro level, it reproduces common empirical findings from the existing literature to a large extent, for instance that respondents who know more about the political process have a way of making up their minds about parties that is more structured by left-right. Other mechanisms of party preference building appear to have heterogenous effects: partisan identification seems to reinforce spatial thinking. Retrospective voting, on the other hand, turns out to be an alternative pathway.

Still, the results suggest that context effects are rather small. This again underscores the strong structuring capacity of the left-right dimension and how robust it is (see Knutsen, 1995). From the perspective of the normative debate alluded to above however, this might actually be a positive finding: after all, if the proximity voting mechanism is robust to context even in more complex decision environments, and the left-right dimension is indeed more or less available as a preference-structuring device in all systems, this warrants optimism that indeed “citizens can manage the complexities of politics and make reasonable decisions given their political interests and positions” (Dalton and Klingemann, 2007: 6).

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Holger Reinermann

Notes

References

Abramowitz

Saunders

(2006) Exploring the bases of partisanship in the American electorate: Social identity vs. ideology. Political Research Quarterly 59(2): 175–187.

Adams

Dow

Merrill

(2006) The political consequences of alienation-based and indifference-based voter abstention: Applications to presidential elections. Political Behavior 28(1): 65–86.

Albright

(2010) The multidimensional nature of party competition. Party Politics 16(6): 699–719.

Anderson

(2000) Economic voting and political context: a comparative perspective. Electoral Studies 19(2–3): 151–170.

Bakker

de Vries

Edwards

, et al (2015) Measuring party positions in Europe. Party Politics 21(1): 143–152.

Bakker

Jolly

Polk

(2012) Complexity in the European party space: Exploring dimensionality with experts. European Union Politics 13(2): 219–245.

Bates

Mächler

Bolker

, et al (2015) Fitting Linear Mixed-Effects Models Using lme4. Journal of Statistical Software 67(1): 1–48.

Blais

Nadeau

Gidengil

, et al (2001) The formation of party preferences: Testing the proximity and directional models. European Journal of Political Research 40(1): 81–91.

Boatright

R. G.

(2008) Who are the spatial voting violators?. Electoral Studies 27(1): 116–125.

10.

Boudreau

Elmendorf

MacKenzie

(2013). Lost in space? shortcuts and spatial voting in low-information elections. UC Davis Legal Studies Research Paper Research Paper No. 328 .

11.

Brambor

Clark

Golder

(2006) Understanding interaction models: Improving empirical analyses. Political Analysis 14(1): 63–82.

12.

Budge

Farlie

(1983) Explaining and Predicting Elections: Issue Effects and Party Strategies in Twenty-Three Democracies. London: Allen & Unwin.

13.

Budge

Klingemann

Volkens

, et al (2001) Mapping Policy Preferences I: Estimates for Parties, Electors, and Governments, 1945-1998. Oxford: Oxford University Press.

14.

Campbell

Converse

Miller

, et al (1960) The American Voter. New York/London: Wiley.

15.

Carmines

Stimson

(1986) On the structure and sequence of issue evolution. American Political Science Review 80(3): 901–920.

16.

Converse

(1964) The nature of belief systems in mass publics. In: Apter

(ed), Ideology and Discontent. New York: Free Press, pp. 206–261.

17.

CSES (2007) Comparative Study of Electoral Systems Module 2 Full Release. DOI: 10.7804/cses.module2.2015-12-15 [dataset, december 15, 2015].

18.

CSES (2013) Comparative Study of Electoral Systems Module 3 Full Release. DOI: 10.7804/cses.module3.2015-12-15 [dataset, december 15, 2015].

19.

Dalton

(2008) The Quantity and the Quality of Party Systems. Comparative Political Studies 41(7): 899–920.

20.

Dalton

(2015) Party representation across multiple issue dimensions. Party Politics 23(6): 609–622. published online before print. DOI: 10.1177/1354068815614515

21.

Dalton

Klingemann

H-D

(2007) Citizens and political behavior. In: Dalton

Klingemann

H-D

(eds), Oxford Handbook of Political Behavior. Oxford: Oxford University Press, pp. 3–26.

22.

De Vries

Hakhverdian

Lancee

(2013) The Dynamics of Voters' Left/Right Identification: The Role of Economic and Cultural Attitudes. Political Science Research and Methods 1: 223–238.

23.

Downs

(1957) An Economic Theory of Democracy. New York: Harper & Row.

24.

Duch

Stevenson

(2005) Context and the economic vote: A multilevel analysis. Political Analysis 13(4): 387–409.

25.

Elff

Heisig

Schaeffer

, et al (2016) No need to turn Bayesian in multilevel analysis with few clusters: How frequentist methods provide unbiased estimates and accurate inference. Open Science Framework 20. DOI: 10.31219/osf.io/z65s4

26.

Elias

Szöcsik

Zuber

(2015) Position, selective emphasis and framing. Party Politics 21(6): 839–850.

27.

Facchini

Jaeck

(2019) Ideology and the rationality of non-voting. Rationality and Society 31(3): 265–286.

28.

Field

Miles

Field

(2012) Discovering Statistics Using R. SAGE Publications.

29.

Fortunato

Stevenson

Vonnahme

(2016) Context and Political Knowledge: Explaining Cross-National Variation in Partisan Left-Right Knowledge. The Journal of Politics 78(4): 1211–1228.

30.

Franzmann

Kaiser

(2006) Locating political parties in policy space. Party Politics 12(2): 163–188.

31.

Freedom House (2014) Freedom in the World 1973-2014. http://www.freedomhouse.org/report-types/freedom-world (accessed on 20 July 2014).

32.

Fuchs

Klingemann

H-D

(1990) 7 The Left-Right Schema. In: Jennings

van Deth

(eds), Continuities in Political Action: A Longitudinal Study of Political Orientations in Three Western Democracies, De Gruyter Studies on North America : Politics, Government, Society, Economy, and History. Berlin/New York: Walter de Gruyter, pp. 203–234.

33.

Ganghof

Eppner

Heess

(2014) Normative Balance and Electoral Reform: A Finnish Puzzle and a Comparative Analysis. West European Politics 38(1): 53–72.

34.

Granberg

Holmberg

(1988) The Political System Matters: Social Psychology and Voting Behavior in Sweden and the United States. European Monographs in Social Psychology. Cambridge University Press.

35.

Green

Palmquist

Schickler

(2002) Partisan Hearts and Minds: Political Parties and the Social Identities of Voters. New Haven/London: Yale ISPS SeriesYale University Press.

36.

Greene

(2004) Social Identity Theory and Party Identification*. Social Science Quarterly 85(1): 136–153.

37.

Hellwig

(2008) Explaining the salience of left-right ideology in postindustrial democracies: The role of structural economic change. European Journal of Political Research 47(6): 687–709.

38.

Hobolt

de Vries

(2015) Issue entrepreneurship and multiparty competition. Comparative Political Studies 48(9): 1159–1185.

39.

Huber

(1989) Values and partisanship in left-right orientations: measuring ideology. European Journal of Political Research 17(5): 599–621.

40.

Huber

Powell

(1994) Congruence between citizens and policymakers in two visions of liberal democracy. World Politics 46(3): 291–326.

41.

Inglehart

Klingemann

H-D

(1976) Party identification, ideological preference and the left-right dimension among Western publics. In: Budge

Crewe

Farlie

(eds), Party Identification and beyond. London: Wiley, pp. 243–273.

42.

Jessee

(2009) Spatial voting in the 2004 presidential election. American Political Science Review 103(1): 59–81.

43.

Jessee

(2010) Partisan bias, political information and spatial voting in the 2008 presidential election. The Journal of Politics 72(2): 327–340.

44.

Joesten

Stone

(2014) Reassessing proximity voting: Expertise, party, and choice in congressional elections. The Journal of Politics 76(3): 740–753.

45.

Kedar

(2005) When Moderate Voters Prefer Extreme Parties: Policy Balancingin Parliamentary Elections. American Political Science Review 99(2): 185–199.

46.

Kinder

(1983) Diversity and complexity in American public opinion. In: Finifter

A. W.

(ed), Political Science: The State of the Discipline. Washington: APSA Press, pp. 389–425.

47.

Knutsen

(1995) Value orientations, political conflicts and left-right identification: A comparative study. European Journal of Political Research 28(1): 63–93.

48.

Kroh

(2009) The ease of ideological voting: Voter sophistication and party system complexity. In: Klingemann

H-D

(ed), The Comparative Study of Electoral Systems. Oxford: Oxford University Press, pp. 220–236.

49.

Laakso

Taagepera

(1979) "Effective" Number of Parties. Comparative Political Studies 12(1): 3–27.

50.

Lachat

(2008) The impact of party polarization on ideological voting. Electoral Studies 27(4): 687–698.

51.

Lau

Redlawsk

(2001) Advantages and disadvantages of cognitive heuristics in political decision making. American Journal of Political Science 45(4): 951–971.

52.

Leader Maynard

Mildenberger

(2018) Convergence and divergence in the study of ideology: A critical review. British Journal of Political Science 48(2): 563–589.

53.

Lee

Schutte

(2017) Elite-level issue dynamics. Party Politics 23(3): 205–219.

54.

Leifeld

(2013) texreg: Conversion of Statistical Model Output in R to LATEX and HTML Tables. Journal of Statistical Software 55(8): 1–24.

55.

Lijphart

(1999) Patterns of Democracy: Government Forms and Performance in Thirty-Six Countries. New Haven: Yale University Press.

56.

Lipset

Rokkan

(1990) Cleavage structures, party systems, and voter alignments. In: Mair

(ed), The West European Party System. Oxford: Oxford University Press, pp. 91–138.

57.

Lupia

(1994) Shortcuts versus encyclopedias: Information and voting behavior in California insurance reform elections. American Political Science Review 88(1): 63–76.

58.

Lupu

(2015) Party polarization and mass partisanship: A comparative perspective. Political Behavior 37(2): 331–356.

59.

Mair

(2007) Left-right orientations. In: Dalton

Klingemann

H.-D.

(eds), Oxford Handbook of Political Behavior. Oxford: Oxford University Press, p. 206.

60.

Medina

(2015) Partisan supply and voters' positioning on the left-right scale in Europe. Party Politics 21(5): 775–790.

61.

Nyblade

(2004) The “effective” number of issue dimensions: A measure with application to west Europe. In: Annual Conference of the Midwest Political Science Association. Paper Prepared for delivery at the .

62.

Pardos-Prado

Dinas

(2010) Systemic polarisation and spatial voting. European Journal of Political Research 49(6): 759–786.

63.

Pennings

(2002) The Dimensionality of the EU Policy Space. European Union Politics 3(1): 59–80.

64.

Rabinowitz

MacDonald

(1989) A directional theory of issue voting. American Political Science Review 83(1): 93–121.

65.

Reinermann

Barbet

(2020) Party system dimensionality and perceived quality of representation. Party Politics 26(6): 834–846.

66.

Riker

(1982) Liberalism against Populism. A Confrontation between the Theory of Democracy and the Theory of Social Choice. Long Grove: Waveland Press.

67.

Rivers

(1988) Heterogeneity in models of electoral choice. American Journal of Political Science 32(3): 737–757.

68.

Robertson

(1976) A Theory of Party Competition. New York: Wiley.

69.

Robertson

(2006) On the dimensionality of political space and on its inhabitants. In: Bara

Weale

(eds), Democratic Politics and Party Competition: Essays in Honour of Ian Budge, ECPR Studies in European Political Science. New York: Routledge, pp. 159–178.

70.

Rovny

Edwards

(2012) Struggle over Dimensionality. East European Politics and Societies: And Cultures 26(1): 56–74.

71.

Ruth

(2016) Clientelism and the utility of the left-right dimension in latin america. Latin American Politics and Society 58(1): 72–97.

72.

Schattschneider

(1975) The Semi-sovereign People: A Realist’s View of Democracy in America. Boston: Dryden Press.

73.

Simas

(2013) Proximity voting in the 2010 U.S. House elections. Electoral Studies 32(4): 708–717.

74.

Singh

(2010) Contextual influences on the decision calculus: A cross-national examination of proximity voting. Electoral Studies 29(3): 425–434.

75.

Singh

(2014) Linear and quadratic utility loss functions in voting behavior research. Journal of Theoretical Politics 26(1): 35–58.

76.

Singh

(2015) Compulsory voting and the turnout decision calculus. Political Studies 63(3): 548–568.

77.

Sniderman

Bullock

(2004) CHAPTER 12: A Consistency Theory of Public Opinion and Political Choice: The Hypothesis of Menu Dependence. In: Saris

Sniderman

(eds), Studies in Public Opinion - Attitudes, Nonattitudes, Measurement Error, and Change. Princeton and Oxford: Princeton University Press, pp. 337–358.

78.

Solt

(2018) Interplot: Plot the Effects of Variables in Interaction Terms. R package version 0.2.1.

79.

Stoll

(2004) Social Cleavages, Political Institutions, and Party Systems: Putting Preferences Back into the Fundamental Equation of Politics. Stanford: PhD dissertation.

80.

Stoll

(2011) Dimensionality and the number of parties in legislative elections. Party Politics 17(3): 405–429.

81.

Taylor

Herman

(1971) Party systems and government stability. American Political Science Review 65(1): 28–37.

82.

Van de Wardt

de Vries

Hobolt

(2014) Exploiting the cracks: Wedge issues in multiparty competition. The Journal of Politics 76(4): 986–999.

83.

Van der Brug

(2001) Analysing party dynamics by taking partially overlapping snapshots. In: Laver

(ed), Estimating the Policy Positions of Political Actors. London: Routledge, pp. 115–132.

84.

Van der Eijk

Franklin

Van der Brug

(1999) Policy preferences and party choice. In: Thomassen

Schmitt

(eds), Political Representation and Legitimacy in the European Union. Oxford: Oxford University Press, pp. 161–181.

85.

Van der Eijk

Schmitt

Binder

(2005) Left-right orientations and party choice. In: Thomassen

(ed), The European Voter: A Comparative Study of Modern Democracies. Oxford: Oxford University Press, pp. 167–182.

86.

Volkens

Lehmann

Merz

, et al (2014) The Manifesto Data Collection. BerlinHenrike: Wissenschaftszentrum Berlin für Sozialforschung, with Schultze. Manifesto Project (MRG/CMP/MARPOR). Version 2014b .

87.

Warwick

(2002) Toward a common dimensionality in West European policy spaces. Party Politics 8(1): 101–122.

88.

Wessels

Schmitt

(2008) Meaningful choices, political supply, and institutional effectiveness. Electoral Studies 27(1): 19–30.

89.

Wickham

(2016) ggplot2: Elegant Graphics for Data Analysis. Springer.

90.

Zaller

(1992) The Nature and Origins of Mass Opinion. Cambridge University Press.

91.

Zaller

Feldman

(1992) A simple theory of the survey response: Answering questions versus revealing preferences. American Journal of Political Science 36(3): 579–616.

92.

Zechmeister

(2006) What's Left and Who's Right? A Q-method Study of Individual and Contextual Influences on the Meaning of Ideological Labels. Political Behavior 28(2): 151–173.