Introduction

Since the publication of Lipset and Rokkan’s (1967) influential Party Systems and Voter Alignments: Cross-National Perspectives, scholars studying the electoral behaviour and party competition have assumed that electoral dynamics relate to a set of potential conflicts that determine the formation of electoral coalitions. Later, the idea that electoral coalitions are not always fixed (Norris 1997, p. 52) encouraged scholars to analyse how those conflicts could evolve (Carmines and Stimson 1989; Knutsen and Kumlin 2005; Schofield et al. 2003). However, efforts to understand those changes have been limited primarily to major, well-known disruptions to the basis of competition (Carmines and Stimson 1986, p. 162), and consequently, little is known about subtler changes in issues addressed in political debates that occur during election season.

Efforts to understand the dynamics of voters, and parties agendas in elections other than critical ones have been made primarily by agenda-setting scholars who have tested how emphasizing certain issues over others affects the perceptions of voters (Behr and Iyengar 1985; Walgrave et al. 2009; Winter and Eyal 1981). However, their studies have tended to involve measuring short-term changes in voters’ perceptions, not whether such perceptions ultimately influenced voting behaviour. At the same time, scholars interested in issue-based voting have tested the extent to which parties’ emphasis or extremity on specific issues has corresponded to issues that have ultimately shaped the vote (de Vries and Hobolt 2012; Lachat 2008c, 2009). Nevertheless, by primarily using single cross-sectional analyses, their results have not clarified which part of the association between parties’ emphasis on issues and voting behaviour is structural and due to voters’ preference for parties that emphasize issues dearest to them and which part is dynamic and susceptible to change. Other researchers have combined both cross-sectional and cross-temporal variations in order to test the relationships between the agendas of parties and those of voters. Among them, Elff (2009) has shown that the lower levels of class-based voting are explained to a great extent by the lower levels of party polarization. Similarly, Freire (2008) found that left–right attitudes were more often shaped by party polarization than by sociological factors, while de Vries et al. (2010) observed that the extent to which European Union issues impacted voting for a party was moderated by the extent to which a party’s position on the issue was clear. However, because those scholars scrutinized specific issues and processes, their conclusions might not be generalizable to other cases.

To improve current understandings of how parties’ strategies influence the behaviour of voters, this paper investigates the relationship between parties’ platform and the alignment of voters along the different issues with a cross-sectional and dynamic approach that considers different cleavages within the electorate. By using data from different time points, the approach can test whether the observed relationship between parties’ strategies and voters’ behaviour holds when a change in them occurs. Furthermore, by using data from different countries, contextual variables can be included to control for other potential explanations. Further still, by using data about different cleavages in the electorate, it can consider differences among them. In the process, by measuring parties’ strategies independently of voters’ perceptions, the research minimizes endogeneity problems.

Examining whether changes in electoral alignments around cleavages are associated with the issues for which parties compete is not only relevant to understanding voting behaviour and party competition also poses implications for democratic theory because it affects interpretations of democratic mandates. Parties change the issues that they emphasize and the polarization that they create for strategic reasons (de Vries and Hobolt 2012; Klüver and Spoon 2014; Wagner 2012), and those changes should generate new coalitions that, at least in theory, create new forms of legitimacy for election winners (Riker 1982). However, Achen and Bartels (2016) have countered that changes in electoral alliances do not coincide with changes in the discourses of parties. If confirmed, that notion should change current understandings of the electoral process. After all, if voters do not respond to changes in the issues made salient by parties, then such shifts are likely less relevant than assumed, their use as electoral tools should be reconsidered, and changes in the discourses and strategies of parties should not be read as cues of the formation of new coalitions. More generally, the interpretation of electoral mandates would become more complicated than assumed.

In what follows, this paper presents an overview of how parties’ strategies can be theorized to affect voting and the corresponding hypothesis that will be tested. Later, it describes the empirical strategy and measurements employed, after which it discusses the results and their chief implications.

Expected association between issues debated by parties and voting

The overarching research question is whether the dynamic of adjusting to voters’ preferences found at the party level (Klüver and Spoon 2014; Spoon and Klüver 2014) is also matched by voters’ shifts in alliances according to issues debated by parties. Unlike major realignments of the electorate in critical elections, such shifts cannot be explained with reference to the politicization of issues not yet politicized (Sundquist 1983) but with reference to the dynamics of voters’ bounded rationality and the “menu dependence hypothesis” (Sniderman and Bullock 2004). Consequently, although the research question somewhat overlaps with the literature on the politicization of issues (Grande and Hutter 2016), its scope differs slightly. The interest is not only on the appearance of new issues on the agenda as a singular process, but also minor alterations and dynamic processes that might occur in the common process of electoral competition, in which parties choose to change the importance that they attach to different issues, are considered. The focus is not then on the politicization of issues but on the dependence that voters’ behaviour has on the choice offered by partisan platforms (Schattschneider 1975; Sniderman and Bullock 2004).

The lack of careful and systematic analysis of all considerations available to voters when deciding for whom to vote means that, depending upon how parties organize their platforms, voters’ understandings of the choice and the considerations that should be considered when deciding will differ, even regarding issues already on the agenda. As a result, if parties change the structure of the debate around the issues, then voters’ understandings of electoral choices should also change, and thus, the issues that structure their vote will differ. In theory, voters align more clearly along conflicts more saliently discussed by parties (Fazio and Williams 1986; Iyengar 1990). Such change in the alignment might occur because voters either change their preferred party choice according to the cleavage they consider relevant or because they adapt their positions on issues to their vote. This paper will not try to differentiate the two logics. As long as the level of association changes, parties and voters will be linked in a meaningful way. Future researchers should focus on understanding which dynamics are more common and when.

The salience of issues debated by parties has two components (Netjes and Binnema 2007). On the one hand, the emphasis on or space dedicated to a debated issue plays a role. If parties dedicate more space to an issue, theoretically this should translate into greater likelihood that voters will be exposed to messages about it and deem it relevant to their vote choice. Furthermore, if parties talk more about an issue, then voters will theoretically experience greater recall of the considerations linked to the issue when casting their votes (Zaller 1992). As a result, more voters will adjust their positions on the cleavage and their party of choice when the debate on the issue has occupied more space on the agenda.

H1

When parties dedicate more attention to a conflict, voters more strongly align their attitudes on it with their vote choice.

On the other hand, clarity of the positions and polarization on a conflict are also critical elements in defining the space of political competition (de Vries et al. 2011). By defending a clearly differentiated position on a conflict, parties can facilitate voters’ understanding of their positions on the issue in question (Lachat 2008b, p. 688) and increase their perceptions that, at the upcoming elections, something is at stake related to that cleavage (Orriols and Balcells 2012, p. 3). Ultimately, voters should therefore be more inclined to align their votes with positions on conflicts about which they perceive clear differences in parties’ stances.

H2

When parties differentiate themselves more clearly in relation to a conflict, voters more strongly align their attitudes with their party choice when they vote.

The effect of parties’ strategies might not be linear; two hypotheses are added to test for that possibility. First, polarization and emphasis on issues might exert an effect only when they achieve certain visibility; otherwise, voters might remain unaware of any change in the debate. If true, then the effect of parties’ strategies would not linear but dependent upon the values of the variable.

H3

The impact of changes in parties’ polarization or emphasis on voters’ alignments around issues is exponential—that is, greater at greater values than at lower ones.

Second, as Grande and Hutter’s (2016) politicization measure proposes, the two measures might be necessary for parties to shape a debate. Polarization and emphasis on issues might exert an effect only when combined, for only when both the polarization and visibility of the debate are salient can an issue be conceived as relevant and, in turn, expected to shape voters’ decisions. Consequently, the effect of parties’ strategies is not linear but dependent upon the values of the variable. Hypothesis 4 articulates that possibility more formally:

H4

The impact of changes in parties’ polarization or emphasis on voters’ alignments around issues depends upon the value of the other parameter.

H1–H4 uphold the idea that parties shape the conflicts that voters perceive to be relevant at each election. However, such an effect might not be as straightforward as it seems. For one, voters might find ways to minimize the impact of new issues to protect themselves from having to change their behaviour and attitudes frequently (Popkin 1991, pp. 15–16). The effect of parties’ strategies might therefore not always be direct or immediate. Two hypotheses are tested to account for that possibility. First, voters might need time to accept that certain issues have become important in voting before aligning around them. Accordingly, parties’ strategies might exert an effect only if they are pursued long enough for voters to realize that certain issues have become relevant to the relevant to the electoral choice (Popkin 1991, p. 100).

H5

When parties have placed more emphasis or showed greater disparity among themselves on an electoral cleavage in previous elections, voters more strongly align their attitudes towards it when voting.

Second, as the theory of affective judgement proposes (Marcus et al. 2000), voters are more likely to reconsider their electoral choices when they feel anxious about situations. Although such anxiety might arise solely due to parties’ discourses, it is more likely to stem from changes in the political, social and economic situations of a country. Contextual changes should increase the effect of parties’ actions by fuelling anxiety or dissatisfaction with those situations and thereby motivating voters to carefully consider their options and gather information about new cleavages.

H6

The impact of parties’ strategies on voters’ alignments around issues will be greater when elements of the political, social and economic context change; the more the context changes, the stronger the effects of the parties’ strategies.

Empirical strategy

To analyse the extent to which parties’ and voters’ behaviour around the different cleavages change congruently, the cleavages need to be defined. In established European democracies, scholars have typically identified two cleavages. First, there is an economic dimension, which structures attitudes towards the redistribution of wealth and the role of the state therein, and a cultural or social one (Heath et al. 1994; Hooghe et al. 2002). The second cleavage seems to have changed in recent years from one dimension—namely, the moral debate between religious and secular society, often related to traditional or religious views versus progressive or liberal ones—to a new conflict more closely related to globalization and the blurring of national borders and identities that has reshaped alliances in western European countries (Kriesi et al. 2008, pp. 13–15 & 270). Moreover, in some countries, a conflict concerning the structure of the state could exist that could relate to the distribution of power among territorial units or among democratic institutions (Hutter et al. 2017; Vidal 2017). That conflict seems to also have its own logics and to cut across alliances created around the cultural and economic cleavages (Polk et al. 2017; Toubeau and Wagner 2013).

In sum, four cleavages can be identified: an economic one, a moral or religious one, one about globalization and another about state structure. That list partly overlaps the four cleavages identified by Lipset and Rokkan (1967), although territorial cleavage has shifted from a rural–urban conflict to one about globalization.

The analysis is done at the level of elections or the party system and includes elections held in 18 western European countriesFootnote 1 since the early 1970s.Footnote 2 The western European context affords an opportunity to compare countries relatively close in geographic and political terms that are affected by fairly similar dynamics but have entirely distinct party and political systems. The selection of elections depends on the availability of data from the different countries and includes all years of data available. The dependent variable is the level of association between party choice and voters’ positions around cleavages at each election. The explicative variables are, on the one hand, the emphasis and polarization on issues presented by parties to voters at previous and current elections specified according to the hypotheses and, on the other, the political, economic and social changes liable to attract the attention of voters to certain conflicts. The next sections explain how these variables are operationalized.

Independent variable: parties’ behaviour

Parties’ behaviour and their strategies for stressing or not stressing cleavages are measured using data from the Comparative Manifestos Project, which is the sole data source that allows comparison across countries during the period under study (Gemenis 2013, p. 18).Footnote 3 Attention paid to the different cleavages by parties is calculated by totalling the weighted average space dedicated to the cleavage by all parties in the political system at each election (Stoll 2004, p. 104)Footnote 4:

$$S_{ie} = \mathop \sum \limits_{j = 1}^{J} w_{je} M_{iej}$$

in which \(S_{iem}\) is emphasis placed on the cleavage, i, at an election, e; J is the number of parties running in the election; \(M_{iej}\) is the emphasis of different parties on each cleavage; and \(w_{je}\) is the electoral weight of each party, calculated as the percentage of the vote received in the previous two elections. Applying weight according to the electoral strength of the party takes into account that not all parties have the same power to shape the agenda of the system. To explain, not only do some parties receive more attention than others from the media (Lefevere and Dandoy 2011), but also voters rely on the most visible parties to structure their understanding of their votes (Popkin 1991, pp. 91–93).

The level of polarization around cleavages is computed using the formula used by Lachat (2008b) that computes the sum of the squared weighted distances between each party’s position and the weighted mean of all parties in the system:

$$P_{ie} = \mathop \sum \limits_{j = 1}^{J} w_{je} \left( {p_{iej} - \bar{p}_{ie} } \right)^{2}$$

in which \(P_{ie}\) is the level of polarization of the system during the election, \(p_{iej}\) is the position taken by each party on each cleavage, and \(\bar{p}_{ie}\) is the weighted mean of the positions of the parties during the election calculated with the formula

$$\bar{p}_{ie} = \mathop \sum \limits_{j = 1}^{J} w_{je} p_{ie}$$

To test H3 and H4, the two quadratic and interactive terms are calculated. By contrast, to test H5, the mean value of the emphasis and polarization on each cleavage in the previous two elections is considered. As Fig. 1 shows, parties’ strategies in relation to cleavages vary greatly across countries and through time. Close scrutiny of the sources of such variation reveals that once between-country variation due to different cleavages’ having different mean values is considered, the variation between and within countries is quite similar in size.Footnote 5 The figure also shows that there are differences in the structure of the partisan debate along the decades considered, although these changes follow different logics depending on whether we observe polarization or emphasis measures.

Fig. 1
figure 1

From the comparative manifesto proejct

Distribution of the independent variable.

Full size image

Dependent variable: impact of cleavages on voting

Five sources are used to get indicators of voters’ positions on the cleavages and their choices of party: the European Social Survey, the International Social Survey Project, the European Value Study, several Eurobarometers and surveys from the European Election Studies. Any question asked on those surveys whose objective related to the Comparative Manifestos Project’s categories is included in the dataset.Footnote 6 Using as many questions as possible instead of focusing on the few frequently asked ones afforded two advantages. First, for some issues, none of the corresponding questions provide good coverage. Second, the use of more indicators can improve the validity of the scale (de Vaus 1985, p. 181). Regarding choice of party, voting intention is the primary measure used in the analysis. However, in cases in which voting choice is not available, but information about party preference or attachment is, the measure is considered to be a valid approximation of it (Knutsen 1995, p. 462).

The strength of the association between party position and the position defended by voters on questions related to each cleavage is calculated with the eta coefficient from a one-way analysis of variance (ANOVA) for each cleavage in each election considered.Footnote 7 Eta coefficients are “identical to Pearson’s correlation coefficients between party preferences and value orientation scales when the categories on the party choice variables are given their mean scores on the scales” (Knutsen 1995, p. 463). As a result, eta coefficients neatly capture the extent to which positions on an issue are strongly dependent on the party chosen by voters.Footnote 8

Typical controls, such as socio-economic characteristics or political views on different cleavages, that would ideally be added as controls for the association, are not consistently included in the sources used. Since adding them would create more problems than benefits due to the survey questions’ inconsistencies, the association is calculated without controls. As a result, voters are conceived to be allowed to adjust their alignments in relation to a cleavage by changing either their choice of party or their position in relation to the cleavage. Although analyses in the future should focus on differentiating the two logics because they have different implications (Lenz 2009), the goal of the study reported here is to understand the extent to which the association exists, regardless of the source of the variation.

To account for the fact that wordings of the questions used vary greatly across time and in different studies, Stimson’s (Stimson 1999, p. 51) algorithm was used.Footnote 9 Using the algorithm solves the problems created by the fact that changes in the wording of questions have likely not been random, a characteristic that would bias the dependent variable (de Vaus 1985). Figure 2 displays the evolution of the dependent variable by cleavage. Similar to parties’ strategies, the level of association between cleavages and vote varies considerably, and as shown in appendix, such variation is almost as common within countries as between them. Although the countries included vary generating some potential bias on the trends, the figure also shows that the association of the moral conflict with the vote has fallen, especially since the late 1980s. The globalization has become more aligned with party choice in the last years observed. Meanwhile, the association between vote and economic and state structure attitudes has fluctuated with less clear trends.

Fig. 2
figure 2

Several public opinion polls.

Distribution of the dependent variable.

Full size image

Controls: social and contextual changes beyond parties

Accounting for the context in which voters and parties interact is not easy, for the randomness present in the evolution of issues (Carmines and Stimson 1986) complicates anticipating when an issue will become relevant in the political space. That said, scholars have identified elements that are usually related to the appearance of certain issues on the agenda and that can be used as proxies of the extent to which voters might feel enough anxiety to reconsider their electoral alliances and behaviours.

The first elements hypothesized to affect issues that structure party competition are those related to economic performance of the country. In theory, if a national economy exhibits a worrying dynamic, then voters might be more willing to care about economic policies than when the economy is flourishing and no economic problems are anticipated. At the same time, those indicators should also influence the extent to which new post-materialist issues can emerge on the agenda (Inglehart 2009). By increasing the fulfilment of material needs, economic performance might leave more space for post-materialist concerns to appear with force and generate a new type of competition among parties. GDP per capita and the Gini index of inequalities are used to measure those dynamics (Stoll 2004, p. 136).Footnote 10

Beyond a lack of economic anxiety, social changes can also generate and drive social concerns. For example, the existence of a secularization process should affect the extent to which a moral conflict finds a fertile soil to progress or leaves space for other conflicts (Brooks and Manza 1997a, b).Footnote 11 Similarly, changes in the openness of a country to globalization, measured, for example, with migration and trade flows, should alter the possibility of a conflict around globalization occupying space in voters’ considerations (Kriesi et al. 2006; Lachat 2008a). The Freedom House Index of Democracy (Freedom House 2015) is included to account for changes in the structure of the democratic system of the country that could turn attention to such conflict.Footnote 12

Last, because anxiety is more likely to relate to changes in variables than their absolute values, control variables are recoded to measure change in their values from the year prior to the election to the year of the election. The use of a 1-year lag seems appropriate given the limited memory span that voters demonstrate when evaluating their governments (Achen and Bartels 2016, pp. 166–168) (Table 1).

Table 1 Summary of controls
Full size table

Results

The final dataset is similar to the second stage of a two-stage multi-level analysis (Duch and Stevenson 2005; Gelman 2005), with the sole difference that coefficients of the first stage are corrected with Stimson’s algorithm. Since the chief focus of the study is understanding how issue-based voting changes as parties’ strategies change, the models are estimated using issue- and country-related fixed effects. That is, each issue for each country serves as the subjects for which repeated observations are available. Variation is estimated within subjects, not across them. With that approach, structural elements (e.g. the cleavage’s traditional importance in some countries) are eliminated as potential explanations of the association. The coefficients thus reflect how parties and voters move together. The focus on variation within countries and within issues should render the model less affected by unobserved structural variables; however, the double causality problem remains (Steenbergen et al. 2007). Furthermore, since voters are conceived to be allowed to change their positions on an issue or their choice of party when adjusting their level of association, the causal mechanism is not tested, merely the association between parties and voters.

Since the strength of the association between votes and cleavages at each election is likely correlated because voters have to choose which issues structure their votes, country-clustered standard errors are used. White’s heteroscedastic consistent standard errors are computed to account for the uncertainty created by having an estimated dependent variable (de Vries 2010, p. 108). Moreover, since some of the four variables of parties’ strategies are heavily skewed to the right, the chief independent variables are log-transformed. This means that the associations between the dependent and the transformed independent variables estimated are not linear. Coefficients do not represent increases in the dependent variable with each unit change in the independent variable, but for each per cent point increase on it.Footnote 13 Or what is the same, the relationship estimated is slightly convex with higher effects on the lower values of the independent variable, something that will be corrected when quadratic terms are included.

To simplify the visualization of the results and given that some coefficients suffer from multicollinearity that makes them unreliable, Table 2 presents the goodness of fit of the models, including the different sets of variables, as a measure that quickly grasps the explicative power of different variables.Footnote 14 Results in the table present the measures of fit for a model with all observations and for the different cleavages separately to allow different cleavages to have different logics. The coefficients reveal that parties’ strategies on the election have weak explicative power in relation to changes in voters’ alignments along conflicts; the adjusted R2 values of those models always fall below 7%. This finding is relevant because the relationship between the spheres is estimated without any control at either level, which allows for different mechanisms to play a role, and even then, the results demonstrate limited explicative power.

Table 2 Explaining voters’ alignments: fit of the models
Full size table

Although including the interactive and quadratic terms does not increase the explicative power of the models in general, some differences emerged in terms of which specification has the strongest explicative power for the different cleavages. In the general model, including the lagged terms contributes most to explicative power, though the association between parties and voters in the economic model can be better explain when the variables are allowed to have exponential effects measured by the quadratic terms. In the case of the globalization conflict the best fit is the simple model with, polarization and emphasis on their own have more ability to explain alignments along the conflict than the other specifications, maybe due to its recent emergence in the political space of many of the countries included.

Including contextual changes in the model also increases model’s ability to predict alignments, especially for moral conflicts and conflicts related to state structure. However, models that included both contextual and party measures usually demonstrate better fits than models including only contextual variables. Although parties’ strategies might not be the strongest determinant of voters’ alliances, they seem to play some role. As found earlier, the best specification of parties’ strategies varies across cleavages even when context is taken into account. The model with the interactive component between polarization and emphasis seems to have a better fit for the general model, but for some issues (e.g. economic conflicts and conflicts related to state structure), the models with quadratic terms explain a greater proportion of the variation in voters’ alignment along cleavages. By contrast, for moral and globalization conflicts, the association is better explained by taking the past into account. H1–H5 thus find only partial and conditional support. Parties’ strategies seem to be associated with voters’ alignment along conflicts, albeit not in a deterministic way and not always with the same logic.

Last, the interaction between parties’ strategies and contextual changes predicted in H6 is tested with two types of model. The first type of model contains interactions among all contextual variables and the two primary components of parties’ strategies, whereas the second model is estimated only with interactions between parties’ strategies and contextual variables directly linked to the conflict. The adjusted R2 values revealed that, for all models but the moral one, the best R2 value surfaced in the model that includes all of the interactive terms. Despite the loss of degrees of freedom due to the abundance of coefficients, allowing the effects of parties’ strategies to depend on how different components of social and economic realities vary affords the best predictions of all.

To elucidate the impact of the different coefficients when separated, Table 3 presents the results for the models with current party measures and lagged, interactive and quadratic terms, respectively. The table shows that the coefficient of party polarization is usually positively associated with stronger alignments in relation to the cleavage, although it reaches statistical significance only in the general model. The more parties are polarized, the more voters’ opinions on the corresponding conflicts seem to become associated with their party choice.Footnote 15 By contrast, the coefficients related to emphasis are not consistently positive and, on average, smaller than the ones for polarization, with the sole exception of religious conflict.Footnote 16 Similar dynamics can be observed with the lagged coefficients. Polarization, both at the current and previous elections, seems to be more consistently correlated with voters’ alignments in relation to cleavages than with emphasis on a divisive issue, although some exceptions emerge, especially for moral conflicts.

Table 3 Models results
Full size table

Neither the interactive term nor the two quadratic terms reach statistical significance in any of the models, although this could be due to the multicollinearity problems typical of these coefficients. The interaction between polarization and saliency is only positive in the case of the economic and globalization conflicts, while negative for the other two conflicts. The quadratic terms, on its turn, are both positive for the economic and state structure conflict. Polarization also might have an incremental effect in the general model, and saliency might have increasing effects for the moral conflicts.

Results in Table 3 also show that improvements in the GDP per capita of the country or increasing levels of inequality indicated by the Gini index, are associated with a drop in the level of association between the vote and economic conflict. Higher international flows of migrants or goods and better performance on democratic indices are associated with dealignments from conflicts related to state structure. Finally, the negative impact of lagged party strategies on the association between moral issues and the vote can be explained through the loss of relevance that the issue has suffered in the period covered by the data observed in Fig. 2. A dynamic that would have made the relevance of the conflict on the political system more dependent on its past relevance in the country's dabate than to its decreasing importance in recent years.

Figure 3 shows the effects of the interactive terms between context and parties’ strategies that attained statistical significance. To simplify the analysis, the two measures of parties are combined into a single measure of politicization, which is the product of the results of the two variables (results with the two coefficients separated are shown in appendix). The graphs show that the politicization of conflicts usually moderates how contextual changes translate into dealignments in relation to cleavages. For example, good economic performance or the openness of the country to other countries only produces dealignments on the state structure conflict when the issue has low saliency on the agenda. When the conflict remains salient on the agenda, the levels of association do not fall.

Fig. 3
figure 3

Statistically significant interactions with emphasis (90% margins of error)

Full size image

Conclusions

The results presented here show that the relationship between, on the one hand, cleavages that parties make salient and relevant in their discourses and, on the other, those associated with voters’ choice of party is not as close and direct as assumed by the menu dependence hypothesis (Sniderman and Bullock 2004). Theoretically, the issues that parties prioritize should affect the extent to which voters align their positions in relation to cleavages when they vote; however, in reality, that mechanism is far from direct. Effects need time to appear, and context seems to play a highly relevant role on its own and in moderating the effects of parties’ behaviours. Voters might have their own ideas about which cleavages are the most relevant, or they might have mechanisms to avoid needing to change their alignment and attitudes every time parties change their discourse.

If confirmed, such findings have important implications for the study of electoral behaviour and party competition. They show that voters do not always follow the cues they receive from parties about which issues they should consider when thinking about their choice. This means that party should chose to strategically emphasize some issues over others, but they should also consider voters’ perceptions of the issues that are relevant as they do not have total leeway to shape it (Klüver and Spoon 2014). More importantly, results imply that parties need to pay attention to the context when deciding which issues to emphasize, as not only context moderates the effect of their strategies, their strategies also moderate the effect of context on voters’ perceptions.

The lack of direct connection between parties and voters also has important implications for understanding the democratic mandate. The fact that parties do not have absolute control over shaping the issues that drive voters implies that circles of voting are not easily created and that any equilibrium reached is more stable than assumed (Riker 1982). By extension, the populist ideal of democracy might not be so unrealistic, although it will not necessarily always coincide with public debate. For example, electoral winners’ legitimacy might not always arise from a coalition formed around the most discussed issues. Mandates would exist to some extent, but their interpretation would be more complicated than assumed.

Results highlight the need to clarify the different dynamics of emphasis and polarization on the different issues. Party polarization usually exerts a positive, albeit not always significant, effect on voters’ alignment in relation to cleavages, confirming previous cross-sectional studies (Lachat 2008b; Orriols and Balcells 2012), whereas the emphasis that parties place on cleavages has a less consistent effect. Furthermore, the extent to which is the combination of the two, the existence with strength of one of the two, or an additive effect, also changes depending on the cleavage considered. The finding that the strength of the relationship between the behaviour of parties and voters varies in relation to different cleavages implies that scholars should be cautious when using specific case studies to analyse the electoral process as a whole. Voters might react to party strategies differently depending on the issue they relate too and what functions in one case might not function in another. Scholars should attempt to elucidate those differences and test whether they are due to differences in the nature of the cleavages or to different logics of competition around them. At the same time, results also underscore the need of expanding current understandings of the ways in which voters respond to parties’ strategies in relation to the agenda. A great deal of research has been performed on how voters respond when parties change the positions that they defend (Adams et al. 2008; Fernandez-vazquez 2014), and similar efforts should be directed towards analysing how voters respond to changes on the issues that parties stress. More research is also needed to understand the conditions that make voters responsive or not to changes in issues debated. Furthermore, scholars of the party strategies should consider the limited impact of those strategic moves and their interaction with contextual changes when analysing how parties decide which issues to prioritize.

Before pursuing answers to those research questions, however, additional effort should be dedicated to testing whether the conclusions presented here ring true with different research designs. To that end, researchers should study the association by differentiating the part due to voters’ changing their preferences in relation to cleavages and the part due to them changing their electoral alliances, which might be a more meaningful dynamic. Perhaps more importantly, with new data, the analysis could be performed using different datasets and other contextual variables that might change some of the dynamics observed. Last, the comparative nature of the study allowed limiting risks of double causality and spurious relationships, although the identification strategy remained weak. In the future, scholars should seek to overcome those limitations, as well as to confirm the validity and reliability of the results.