The dimensionality of voting behaviour in two European parliaments

Transcription

1 The dimensionality of voting behaviour in two European parliaments Daphne van der Pas Political science department, University of Amsterdam Elmar Jansen Political science department, University of Amsterdam Draft. All comments are welcome. Please do not cite without permission from the authors. Abstract. Although parliamentary voting behaviour has been widely studied in the context of the US congress using roll call analysis, knowledge of voting behaviour in the multiparty systems of the European context is limited. With automated content analysis on parliamentary proceedings, we collected all votes cast by parties over the past decades in two West European parliaments - Denmark and the Netherlands. We analyse the dimensional structure of policy behaviour of the Danish and Dutch parties comparatively, using spatial modelling method of multidimensional scaling. In doing so we are able to create a relatively dense series of party positions and of the dimensionality of the political space through time. In the paper we focus on the development of the dimensional structure of the two party systems. Several authors have argued that political arenas in West Europe are structured by two dimensions: a socio-economic left-right dimension and a cultural dimension. While the latter dimension used to be characterized by a secular-religious divide, it is now said to be governed by issues like immigration, asylum, nationalism, European integration and cultural progressiveness. We find, however, that the behaviour of parties in the party systems under study predominately follows the logic of a single dimensional space, in which the new politics issues are -in timeintegrated. Additionally, the results indicate that the entry of new parties raises the number of dimensions necessary to adequately represent voting behaviour, but that this effect is temporary. Introduction Over the past decennia we have seen a growing scholarly interest in party ideologies from a spatial perspective, with studies mapping party positions and tracing their shifts through time. An essential part of the spatial model is the dimensionality of the political systems, i.e. over how many

2 dimensions political conflicts are structured. Most recently, there has been quite some debate over the question of how many issue dimensions are present in Western European party systems. Some authors, most prominently Kriesi et al (2008), argue that since the 1990s party systems as well as electorates are increasingly structured by a new dimension on which the losers and winners of globalization take the opposing positions on both poles. The present study will further investigate this claim by analyzing the dimensions that structure the voting behaviour of parliamentarians in two Western European countries through time. Central to the differences between the conclusions of several scholars regarding the dimensionality of party systems, is the fact that the concept of party positions can be measured and has been measured in many varying ways; among others by studying how issue positions of parties are presented in the media, how parties present their positions in party manifestos and how their positions are perceived by political experts. However, these methods have one thing in common: they are all based on how parties present themselves or, probably as a consequence of this presentation, how they are perceived by others. In this study however, we analyze party positions based on how parties vote in parliament, i.e. how they act in the policy making process. As we will argue, this approach can have clear advantages over other approaches, particularly in those cases where we are not so much interested in party positions as the input of the voting decision of individual citizens, as we are when studying voting behaviour, but in party positions as the input for the policy process, e.g. when we want to gain a better understanding of the representation of policy preferences of citizens in terms of policy outcomes. Dissatisfied voters often complain that politicians, who present themselves as having strongly opposing views during election campaigns, are actually indistinguishable when it comes to producing policy outputs. If we want to take such claims seriously, it means that we have to study positions on the supply side in terms of policy outcomes. In a parliamentary democracy, parties have the task to translate competing ideologies into coherent policy packages. How parties combine policy positions can be complex matter. But even though positions on issues can be combined in infinite ways, parties are inclined to do so in a predictable manner, reducing all issue positions to the position on no more than a few political dimensions. Obviously, in how many dimensions parties position themselves has a huge influence on the actual potential for voters to choose on the party they want. If the policy packages, in terms of policy outcomes, that parties offer are extremely similar, it should come as no surprise that citizens complain. By studying the voting behaviour of parliamentarians, we aim at understanding the political space as it is formed when policy decisions have to be taken and parties no longer only have to think about winning votes, but also about securing strategic cooperation with neighbouring parties or coalition partners. So, the party positions in this study do not come from the orderly world of clear oppositions in election campaigns, but from the world of the day-to-day policy bargaining. Although this means that we do not necessarily position parties as they would position themselves and as they are seen by voters during electoral campaigns, we place them according to their behaviour in the policy making process, which is what politics as a whole is eventually judged for by citizens. There is also a non-substantive advantage to the use of parliamentary voting behaviour as a means for studying party positions. Parties have been casting votes in parliamentary divisions for a very long time, and have done so frequently. Moreover, in most parliaments the voting behaviour is well

3 recorded in parliamentary minutes. Consequently, using this voting behaviour as the basis for measuring party positions allows us to reconstruct party positions over long periods and with a higher frequency than any of the other methods allow. In this paper, we contribute to a recent scholarly discussion on the dimensionality of the political space in Western Europe, by presenting data based on parliamentary voting behaviour in two Western European countries: the Netherlands and Denmark. As we will show, our data suggests that parties are inclined to line up in one-dimensional configurations, even though this singledimensionality can be temporarily disrupted by either the emergence of new cross cutting issues or the arrival of new parties which do not (yet) want to subject themselves to this single dimension. In what follows, we will in short present the current dimensionality debate, after which we will present the results of our analysis. Dimensionality of Western European political spaces In their important study on the effects of globalisation, Kriesi et al. (2006, 2008) argue that the division between its winners and losers in society has transformed the political landscape in Western European countries. National political spaces, according to these authors, are now shaped by two main cleavage lines: one economic and one cultural. While the economic dimension is a continuation of the traditionally class-based left-right dimension, the new cultural dimension has replaced the old religious-secular dimension of the mid 20 th century. Their theory and empirical evidence, therefore, combine elements of change with continuity: the number of dimensions is constant over time, but the meaning of the cultural dimension has been reinterpreted from an opposition between religious and secular ethical viewpoints, to one in which policies of cultural openness -for example pro migration and pro EU- are contrasted with policies of cultural demarcation (Kriesi et al. 2008: ). Kriesi and his colleagues are not the first to observe two cross-cutting cleavages in the policy preferences of electorates in Western European contexts. Inglehart (1977, 1990) has famously proposed that a materialist post-materialist conflict line has become dominant in most advanced, industrialized societies, relegating the classic left-right conflict to the place of second dimension. Likewise, Kitschelt (1995; 2004) draws a two-dimensional picture of the voter space, with a libertarian-authoritarian axis crossing the socialist-capitalist axis. In general, most scholars of public opinion find an economic left-right dimension and a second dimension which either combines issues related to the cultural openness of the society, the attachment to religious values or combinations of both 1 (for example also Middendorp 1992, Hooghe et al 2002, Knutsen 1988 and 1989). In their globalisation study, Kriesi et al. not only find a two-dimensional issue space among voters, they also assert that these two cleavages are mirrored in the political arena. However, not all students of Western European politics are convinced that political parties follow the twodimensional pattern. On the basis of expert surveys, van der Brug and van Spanje (2009) find that parties in Western European political systems tend to position themselves in one-dimensional 1 The exact interpretation of the second, cultural dimension varies widely from study to study, but a treatment of these differences falls beyond the scope of this paper. Our focus here is on the cultural dimension as understood by Kriesi et al., which is for the most part also compatible with Kitschelt s authoritarian-libertarian dimension.

4 configurations, even though voters are spread out over a two-dimensional space. In a similar vein, Kitschelt (1995, 2004) finds that until the early nineties parties are positioned on a single line, which runs diagonally through the two-dimensional voter space. There are good theoretical reasons to expect that in multi-party systems the policy space reduces to a single dimension, where new issues constantly align with existing oppositions. The main reason for this dynamic is that parties form coalitions with other parties. These coalitions are valuable assets: no party can achieve anything -whether it is in a coalition government or in parliament- without cooperation with other parties. As orthogonal issues might bring a party in strong opposition with its regular coalition partners and parties have a desire not to jeopardize existing coalitions, parties will in general try to steer clear from orthogonal issues either by ignoring the issue or by aligning their position with the position of the neighbouring parties on the main dimension. As a result, parties have a tendency to neatly line up in the issue space on a single dimension (Van der Brug and Van Spanje, 2009: 311). Yet, one can also think of situations in which parties are coerced to leave their fixed positions on the single dimension and, at least temporarily, position themselves on a second dimension. The principal reason for this to happen is that new issues might emerge, that are not readily integrated into the main dimension. In times when society splits along new cleavage lines or when these lines become increasingly salient, parties are expected to readjust their policy positions. However, when these issues do not easily fall within existing lines of conflict, parties cannot immediately react to these demands from society in predictable and structured ways. This means that, at times, parties will oppose their usual collaborators until the new issue has been accommodated completely into the existing left-right dimension. How these new issues are integrated is not necessarily a question of building coherent ideologies, as it can also be the result of specific path dependent processes, but it does in the end nonetheless lead to predictable party families with fixed policy packages. Not only new issues may force parties to move away from their comfortable positions on the main dimension. New parties are not yet tied to existing policy coalitions, and are therefore freer to explore unconventional, but electorally attractive, positions in the two dimensional voter space or they can try to politicize entirely new issues (Hobolt and de Vries, 2011). As there is an infinite amount of potential political issues, those who have no interest in maintaining the vested order can always find a political opposition that is orthogonal on the main dimension (Schattschneider 1960, Riker 1995). This is especially the case for new parties, who in general have nothing to gain from maintaining the vested order and can, either out of electoral or ideological motives, put the onedimensionality of the party system under pressure. Hypotheses Based on Kriesi s and on the competing empirical studies, we can formulate two sets of hypotheses about the dimensionality of party positions in Western European contexts. First, we can derive two opposing hypotheses from the existing literature regarding the number of dimensions on which parties take policy positions: H1a: Political space in West European democracies is continually two dimensional H1b: Political space in West European democracies is continually one dimensional

5 As an extension of these first competing hypotheses, we can also specify opposing expectations with respect to the new cultural issues that have entered politics in recent decades in many Western European countries: H2a: The second dimension has transformed from a religious dimension to a culturally inclusive-exclusive dimension H2b: New issues, such as those arising from globalisation, are integrated into the dominant, left-right dimension The two double hypotheses correspond with Kriesi et al s continuity hypothesis (H1a) and embedding hypothesis (H2a). In their comparative work the authors propose a third hypothesis, namely the convergence hypothesis. This hypothesis states that under the influence of globalisation in Western European countries, national political spaces and conflict structures have become more similar (Kriesi et al. 2008: 325). In this study, we empirically inspect hypotheses 1 and 2 in two West-European party systems, Denmark and the Netherlands, but with only two cases we do not have enough leverage to test the Kriesi s third hypothesis about the convergence of West European political spaces. Methods The Netherlands and Denmark as cases For our study of the dimensionality of parliamentary behaviour, we use data from two Western European countries: the Netherlands and Denmark. Both countries fall within the theoretical scope of Kriesi s argument, which concerns stable West European countries 2. Both are consolidated, economically developed, liberal democracies, both are members of the European Union, and both have also more generally experienced the pressures arising from globalisation. The Netherlands functions as one of the six case studies in Kriesi s book, providing evidence in accordance with the authors hypotheses, while the Scandinavian countries, fitting neatly into the theory, are only excluded from the study for practical reasons, according to the authors (Kriesi et al. 2008: 55). The Netherlands closely resembled the idealtype of a frozen party system until the 1960s (Andeweg 1982). However, from that time the institutionalized pluralism of the so called pillarisation began to fall apart (Bax 1988), and, according to many scholars, religion lost much of its structuring influence in politics. In the late 70s, three Christian parties merged into the CDA, which put far less emphasis on its Christian heritage than its predecessors (Van der Eijk and Niemöller, 1983; Van der Brug, 1997). New parties also rose, such as the radical liberal D66, social liberal DS70 and the mostly unsuccessful anti-immigrant parties CP and CD of Janmaat. Anti-immigrant parties finally made a successful entry in the Dutch national parliament in the form of the List Pim Fortuyn (LPF) in the 2002 elections, which are rated among the most volatile in post-war Europe (Mair 2008). Most recently, the anti-immigrant sentiment is represented by the anti-islam Freedom party (PVV) of Geert Wilders, which since the 2010 elections provides pivotal support to the centre-right minority government of the VVD en CDA. 2 To be precise, Kriesi et al. describe events in Western Europe, but allow that South European countries are less typical cases than the North European countries. The Scandinavian countries, however, fit neatly into the theory, and are only excluded from the study for practical reasons, according to the authors (Kriesi et al. 2008: 55).

6 Denmark is the second country we study here, albeit much less extensively. Like the Dutch, the Danish electorate follows a trend of increasing levels of volatility in recent years, while new political issues such as refugees and immigration have become more important for parties and voters alike (Green-Pedersen 2006). The globalisation-related issues of immigration were politicised relatively early, in the 1970s, by the Progress Party. Since the 1995, the populist radical right is represented by the Danish People s Party (DF), while since 2001 the minority government of the centre-right relies on parliamentary support by the DF in exchange for stricter immigration policies (Green-Pedersen and Krogstrup 2008, Bale 2008). Data collection The voting data were collected by automatically analyzing the minutes of the Dutch Tweede Kamer and Danish Folketing, the two respective parliaments. Specifically, we used computerized text recognition to construct a dataset containing all non-unanimous divisions held in the Dutch parliament between 1972 and 2010 (n=38105), and all divisions of the Danish parliament held between 1997 and 2010 (n=7980). In the Dutch parliament, a very small number of divisions (about 0.7%) were roll calls with the voting behaviour of each individual parliamentarian recorded instead of that of the party. As these votes did not translate easily into votes per parliamentary fraction we disregarded these. In two cases we treated fractions that were formally distinct as one single party to aid the interpretation of the results. First, we merged the Dutch Centrumdemocraten with the Centrumpartij, as both fractions consisted of the same parliamentarian. For similar reasons we merged the Freedom Party (PVV) with its predecessor: the parliamentarian fraction of its leader Geert Wilders. Multidimensional scaling The resulting dataset contains a record for each (non-unanimous) vote in parliament, with for each party a yea or nay verdict. As this does not tell us the subject matter of the vote, nor what exactly the reject or the accept options stand for, the nays and yeas of individual parliamentary fractions are in itself meaningless. However, by comparing which parties voted in agreement with each other in the same division, we can get valuable information on which parties are close to one another and which parties stand far apart. The percentage of agreement for each pair of parties, i.e. the percentage of divisions in which both parties casted equal votes (either nay or yea), is a direct indicator of the closeness of both parties: the more ideological approximate, the more they will tend to cast equal votes. To translate the percentages of agreement into party positions, we take two more steps. First, we compute the implied distance between each pair of parties as 1 percentage of agreement, so combinations of parties whose fractions always vote the same have a distance of 0 and parties whose fractions always cast opposite votes have the maximum distance of 1. Second, we use multidimensional scaling (MDS) to fit the parties into a lower-dimensional space in such a way that the Euclidian distance between each pair of parties corresponds as closely as possible with the distance as implied by the parliamentary voting behaviour. 3 3 Throughout the paper, we used unweighted regular metric multidimensional scaling. However, to test the robustness of our results we also used weighted modern multidimensional scaling, using the number of seats of each fraction as a weight. These analyses yielded no substantively different results, and are available with

7 To assess the number of dimensions in the voting behaviour, we used the squared Pearson correlation between the distances in the MDS solutions and the distances we computed on the basis of the voting behaviour as an R 2 -measure (see also Borg en Groenen 1997: 67, and van den Brug 1997). The higher this number, the better the MDS-procedure was capable of fitting the distances into the chosen number of dimensions. By comparing this R 2 for several dimensions, we can assess how many dimensions are needed to adequately fit the data. As an alternative we also looked at Mardia-values for goodness of fit of one- and two-dimensional solutions, but as these were in all cases concurrent with the R 2 measure, we present only R 2 s. In some of the presented plots of the two-dimensional solutions, we drew issue vectors as arrows indicating how specific issues are lined up in the two dimensional space. To create these vectors, we used expert survey data as external source for the position of parties on the issue. The direction of these vectors is based on the coefficients of both dimensions in a regression analysis with the expert survey score on the specific issue for each party as the dependent variable (see also van der Brug 1997). The length of the arrows depicts the predictive power (R 2 ) of the dimensions on the expert scores. In this way, these vectors can be interpreted as the direction over which the specific political conflict takes place in a two dimensional space. A wide collection of external sources was used for the purpose of cross-validation and to draw the issue vectors: Castles and Mair s (1984) expert survey, Huber and Inglehart s (1995) expert survey, the Chapel Hill expert survey (Hooghe et al., 2010), Laver and Benoit s expert survey (Laver and Benoit, 2005), Laver and Budge s (1992) codings of Comparative Manifesto Project (CMP) data. Results Dimensionality of the Dutch legislative space: We start by presenting an overview of the dimensionality of the Dutch political space through time. Figure 1 displays the R 2 measures of the one-dimensional and two-dimensional solutions for the years from 1972 to As discussed in the previous section, a higher R 2 indicates a better fit, so in years where the lines drop, more dimensions are needed to adequately represent the data. The line for the two dimensional solution remains consistently high throughout the entire period, confirming the notion that there were never more than two dimensions underlying Dutch parliamentary politics. The line for the one-dimensional solution, however, shows considerable fluctuation over time. Looking at these fluctuations, we can distinguish at least three different periods in the dimensionality of the political landscape: , and << Figure 1 about here >> During the first period, which ends around 1989, voting behaviour of Dutch parliamentarians can easily be predicted by party positions on a single left-right scale. The fit for the one dimensional solution is in fact most of the time as high as that for the two dimensional solution, meaning that a second dimension adds no predictive power. The early years of this period are marked by somewhat lower R 2 s, pointing to the possibility that our research period was preceded by a period of twothe authors on request. Additionally, we analyzed the data using Poole s method to create spatial maps (Poole, 2005), and as these analyses also yielded no different results, we present only the MDS findings.

8 dimensional, pillarised politics. The rising lines of would then be a sign of depillarisation and consequently the disappearance of the secular-religious dimension, but as our analysis does not include the period before 1972, this can only be said tentatively. In any case, by the mid seventies and lasting till the late eighties, Dutch politics was clearly dominated by a single dimension. In the second period, from the late 80s onwards, one dimension is no longer sufficient to describe parliamentary voting, implying that political conflict is structured along more than one line. Although the R 2 of one dimension does oscillate over the span of this decade, a second dimension is always needed for a good fit. Then early in the new millennium a third period begins, in which the overarching trend is less straightforward. A marked return to uni-dimensionality ( ) is followed by a phase of two-dimensionality, while most recently the importance of the second dimension seems to be declining again. From what we have seen so far, we can draw several conclusions regarding the dimensionality of the Dutch party system and more particularly the perspective of Kriesi and his colleagues. First of all, in the light of these results, it is hard to maintain that Dutch politics has always been multi-dimensional. Not only can the period starting from the early seventies until the late eighties easily be fitted on a single dimension, the early 2000s are also clearly one-dimensional. Second, the long nineties show the emergence of a second dimension, so at least for this period Kriesi s theory seems to hold true. However, whether this second dimension is indeed, as Kriesi indicates, a cultural integrationdemarcation dimension, remains a question open for further investigation. This is the topic we turn to in the next section, were we examine the substantive meaning of each dimension by looking at the issues that underlie it. The nature of the main dimensions in the Netherlands: Since for the whole period of 1972 to 1989 the MDS configurations are one dimensional, they can be visualized through time in a single plot. Figure 2 displays the positions of the main parties 4 on this single dimension, while the y-axis represents time. The positions of parties during this period are remarkably stable, even so much so that lines of different parties never cross. We assessed the validity of these MDS positions by comparing them with left-right party positions on various external data sources 5, and correlations were all between 0.91 and This confirms the idea that most voting in parliament in this period closely followed the pattern of what is generally called a left-right scale. << Figure 2 about here >> The nature of the main dimensions in the Netherlands: As discussed above, from 1989 to 2002 a second dimension was needed to explain voting in parliament. This means that we can now plot positions in a two dimensional space, and subsequently interpret these two main lines that shape legislative divisions. To help give substantive meaning to the dimensions, we use expert survey data to position issue vectors in the two-dimensional MDS 4 CDA predecessors KVP, CHU and ARP are combined, as well as GroenLinks predecessors PPR, PSP and CPN and Christenunie predecessors RPF and GPV. 5 Correlations were calculated between the MDS positions and positions on the left right scales of the Huber and Gabel expert survey, Laver and Budge manifesto coding and the Laver and Garry manifesto coding in 1972, and the Castles and Mair expert survey, Huber and Inglehart s expert survey, Eurobarometer voter perceptions, Huber and Gabel s manifesto codings and Laver and Budge s manifesto codings for 1977.

9 solutions. In Figures 3 and 4, the horizontal axes depict the first dimension, meaning that the horizontal positions of the parties are the best representation of parliamentary voting behaviour on a single dimension. The second, vertical, dimension is the additional dimension we can get when we instruct the MDS procedure to fit the model on two instead of one dimension. Whether this dimension indeed adds much to the model differs, as we have seen, widely over time. << Figures 3 and 4 about here >> Figure 4 shows the political space from May 1998 to March 2001, which is exemplary of the two dimensional phase. Figure 3, on the other hand, is based on the parliamentary votes in the term lasting from June 1986 to September 1989, the very last period before the two-dimensional phase began. So while Figure 4 gives insight into the meaning of the second dimensions at its peak, Figure 3 sheds light on the emergence of this dimension. As we see by the more or less horizontal arrows in Figure 3, party positions on the dimension that explains most voting in this period, most closely match the positions the experts give these parties on the issues of environment, public ownership and raising taxes to improve public services. Thus, during this period most of the actual parliamentary behaviour was shaped by party positions on what is described by Kriesi et al. as the economic dimension. As far as divisions take place over the second dimension and not the first, parties vote according to positions experts assign them on EU integration. This indicates that although it is not yet politicized, the upcoming second dimension already seems to take on the character of Kriesi s cultural demarcation-integration dimension. By the parliamentary arena is fully determined by two equally important dimensions that can readily be interpreted as a cultural one and an economic one, as Figure 4 illustrates. Three EU issue vectors are completely orthogonal to the Chapel Hill economic left-right dimension, while the asylum and Gal/Tan vectors are much more in line with the EU vector than with the economic dimension. This plot therefore fits very well with the picture Kriesi et al. draw of this period in Dutch politics. The nature of the main dimensions in the Netherlands: Right after the 2002 elections, the political space retracts to a single dimension. The party positions on this dimension are displayed through time in Figure 5. Again, these positions are stable throughout the period and correlate highly with what experts consider left-right positions. Both of the expert surveys that were held during this period, the 2003 survey by Benoit and Laver and the 2006 Chapel Hill survey, placed parties on a policy scale for the immigration issue. Remarkably, these positions also correlate highly with the MDS positions on one dimension. This implies that, by this time, parties had integrated the immigration issue into the dominant left-right dimension. << Figure 5 about here >> Yet this one-dimensional situation did not last long, as following the 2006 elections the onedimensional solution results in increasingly bad fits. The placement of parties in a two-dimensional space is shown in Figure 6 for the moment when the one-dimensional solution suffers the most stress, i.e. when the second dimension is most needed. Two points arise from this figure. First, the positions on the horizontal axis are for most parties not entirely unlike those of the preceding, onedimensional phase. The most notable change is the relocation of the PvdA and the Christenunie towards the right, which can be explained by their participation in a coalition with the CDA. Second,

10 on the vertical axis parties are relatively concentrated around the middle, except for the PVV, which is far below other parties, and -to a much lesser extent- the VVD. << Figure 6 about here >> This picture suggests that the two-dimensionality of the party space is mainly caused by the position of the PVV in the Dutch party landscape. In that case, the MDS procedure would have the most difficulty placing this party satisfactory on a single dimension. An analysis of the stress values for individual parties indeed confirms the conjecture that positioning the PVV requires a second dimension. The party stress lines are shown in Figure 7 starting from the entry of the PVV, with high values indicating more stress in a one-dimensional solution. << Figure 7 about here >> Finally, in 2010 the short period of two-dimensionality seems to come to an end. This happens at the moment when Wilders PVV agrees to support the governing VVD-CDA minority coalition, in that way (loosely) binding itself to this coalition s position. The PVV can be understood as telling case of a new party s ability to combine issues in unconventional ways. The PVVs point of departure when entering the political arena was - unsurprisingly- very close to that of the VVD, from which the PVV leader originated. It entered the political arena in 2004 with a focus on anti-immigration policies and on lowering taxes, but it gradually started adopting more leftist standpoints. In the 2010 manifesto the party pronounced itself against increasing the age of retirement, relaxing labour law with regards to layoffs, and lowering social benefits, while retaining the anti-immigrant standpoints and a tough-on-crime stance. In their parliamentary behaviour this repositioning was visible, as the PVV moved away from the leftright axis. After the 2010 elections however, the party was brought back into left-right discipline by their agreement with the governing coalition, and the party subsequently relinquished most of its economically left standpoints. Dimensionality of the Danish legislative space: In Figure 8 we plotted the R 2 -measure for the one-dimensional and two-dimensional MDS solutions for each year in the period for Denmark. Here too, a two-dimensional solution is more or less sufficient to explain parliamentary voting behaviour, and again we see a larger fluctuation in the fit of one-dimensional models. Remarkably parallel to what happened when the PVV entered the Dutch parliament, we see a strong increase in two-dimensionality the moment the Danish People s Party (DF) wins its first parliamentary seats in Yet again, this effect lasts only until the moment the DF binds itself to the coalition-agreement in << Figure 8 about here >> Conclusion We set out to examine the dimensional structure of two Western European national political spaces - the Netherlands and Denmark. Regarding the number of dimensions, the literature suggests two competing hypotheses, one stating that politics tends to be structured along a single line, and another stating that it is organized along two axes. For both countries the results are slightly in favour of the second statement, though not without exceptions. In the Netherlands the 90s indeed

11 seem to have been distinctly two dimensional, while the preceding period was clearly onedimensional, and uni-dimensionality returns for the most part of the 2000s. In Denmark, the party landscape also leans more toward a single dimensional space than a two dimensional space. In both countries, the arrival of successful anti-immigrant parties in parliament also marks the start of a period of two-dimensionality, that lasts until the party is integrated into the system when it starts supporting a government coalition. Parties in parliament, it seems therefore, have a clear tendency to one-dimensionality, but can -temporarily- be unhinged from this pattern to form a twodimensional configuration. Our second pair of hypotheses regards the nature of the political space. What is the substantive meaning that can be given to the axes that govern voting behaviour? According to Kriesi et al. the new issues and political conflicts caused by globalisation were slowly incorporated in a previously existing cultural dimension (embedding hypothesis), while van der Brug and van Spanje argue that they have become part of the classical left-right divisions. Globalisation related conflicts became relevant from the late 80s, and from the entrance of these issues in the political realm to the end of the 90s, they are part and parcel of the second, cultural dimension. So for this period, Kriesi et al. s contention seems to hold true. However, in time the most distinct globalisation issues became aligned with the first economic dimension, reducing the political space to one line. In conclusion, the evidence presented in this paper is mixed, but suggests that in the long run the left-right axis has a notable capacity to absorb new conflicts. Taken together, these findings suggest that legislative space tends to seek the structure of a single dimension, but can be disrupted from this tendency by two alternative mechanisms. First, as discussed in the literature section, new cleavages may rise within society that confront political parties with new issues. As new issues do not necessarily naturally fit into the existing dimension, and as new issues are often orthogonal to the dominant dimension for voters, they may demand an extra dimension. In time, however, parties are able to impose the familiar left-right structure on these issues. This structuring tendency is a result of the formation of coalitions and collaborations, which makes fighting battles over multiple lines costly and irrational from a policy-seeking perspective. Second, new parties are less restricted to a single line, as they have fewer vested interests in maintaining existing collaborations. They are therefore freer to combine policy positions in ways that suit many voters, but are not in line with the dominant dimension. Of each of the two mechanisms that expand the political space our data provide examples. The 90s typify a period when contestation around new issues is not yet structured along left-right lines, while the Dutch PVV and the Danish DF are examples of new parties that temporarily rock the boat of leftright politics. One puzzle still remains to be solved, and that is why our results differ so much from what was found previously by Kriesi et al. The opposing findings are particularly odd because the Netherlands was used as a case in both this study and in West European politics in the age of globalisation. The key difference is in the source that is used to base the parties policy positions on. Kriesi et al. relied on statements by party members in the media, which were coded within the relational approach of the Network Method (Kleijneijenhuis studie). As a result, the policy positions they found mostly reflect

12 the way parties present their policy preferences in the media 6. The present paper, by contrast, derives party positions not from what parties say, but from what they do once elected in parliament. In fact, Kriesi et al. mention the nature of our data, i.e. the often incoherent issue positions taken by parties during election campaigns as explanation for the high dimensionality (Kriesi et al. 2008: 269). Possibly parties do combine issues in more complex ways when they campaign for elections and present themselves to voters. As the electoral space is spread out over two dimensions, parties might address certain segments of the electorate by taking positions in this two-dimensional space. But once the campaign and elections are over and parties take their seats in the legislature, the parliamentary reality of forging coalitions and agreements compels them to seek conflicts along a single line. These findings raise new questions and can form a starting point for further research. First, our findings ask for more investigation of the apparent discrepancy between the presentation of parties in the media, and their actual policy making behaviour. One way to study this would be to inspect whether parties, although aligned on a single dimension, obtain their electorate from different parts of the two-dimensional voter space and whether their voters perceive them as positioned on one or rather two dimensions. A second direction for future research would be to deepen our understanding of the mechanisms that drive parties to form one-dimensional configurations. An important part of this is to deconstruct precisely the way in which coalitions and other forms of intra-party cooperation discipline party behaviour. Such studies could be either based on in-depth interviews with parliamentarians to see in what way their collaborations with other parties influences their voting behaviour, or on more formal approaches such as multi-agent modelling. Our third and last suggestion for further research is to extend the scope of the present study. Where this study is based on data from only two countries and a relatively limited time period for one of the two countries, there is room to expand the data to more countries and longer periods. In a related effort, we can increase the depth of our analysis by also taking the substance matter of the votes into account. As voting behaviour in multiparty systems is a relatively understudied research area, we feel that there are many interesting patterns as well as puzzles to be unravelled. 6 Two problems have been pointed out by scholars with regard to Kriesi s use of the Network method to measure policy positions. One is that the Network method measures, besides policy positions, the political issue agenda of a given election (van der Brug and van Spanje 2009, 328). Another, related, criticism is that instead of the policy positions that parties aim to put forth, the method mostly measures the selection of positions that the media choose to emphasize (Pellikaan and Walter 2010). The focus of the discussion here, however, is not on the validity of Kriesi s operationalisations, as the point raised stands even if the method measures what it intends to.

13 References Andeweg, Rudi B Dutch voters adrift: on explanations of electoral change ( ). (Accessed August 18, 2011). Bale, Tim Turning round the telescope. Centre-right parties and immigration and integration policy in Europe1. Journal of European Public Policy 15(3): (Accessed August 8, 2011). Bax, Erik Hanse Modernization and cleavage in Dutch society. A study of long term economic and social change. University of Groningen. Benoit, Kenneth, and Michael Laver Party Policy in Modern Democracies. London: Routledge. Borg, Ingwer, and Patrick J.F. Groenen Modern multidimensional scaling: Theory and applications. 7&dq=borg+groenen+multi+dimensional&ots=SCZp7oPJq- &sig=id4rjs5_rwiy31imnhgsfkjytxm (Accessed August 18, 2011). Van der Brug, Wouter Where s the party? Voters' perceptions of party positions. (Accessed August 18, 2011). Castles, Francis G., and Peter Mair Left-Right Political Scales: Some Expert Judgments. European Journal of Political Research 12(1): (Accessed August 19, 2011). Van Der Brug, Wouter, and Joost Van Spanje Immigration, Europe and the new cultural dimension. European Journal of Political Research 48(3): (Accessed October 27, 2010). Van der Eijk, Cees, and Kees Niemöller Electoral change in the Netherlands: Empirical results and methods of measurement. Green-Pedersen, Christoffer Long-term Changes in Danish Party Politics: The Rise and Importance of Issue Competition. Scandinavian Political Studies 29(3): (Accessed August 18, 2011). Green-Pedersen, Christoffer, and Jesper Krogstrup Immigration as a political issue in Denmark and Sweden. European Journal of Political Research 47(5): Hobolt, Sara B., and Catherine E. De Vries Issue entrepreneurship & multiparty competition. Hooghe, Liesbet et al Reliability and validity of the 2002 and 2006 Chapel Hill expert surveys on party positioning. European Journal of Political Research 49(5): (Accessed July 21, 2011). Huber, J., and R. Inglehart Expert Interpretations of Party Space and Party Locations in 42 Societies. Party Politics 1(1): (Accessed June 29, 2011). Inglehart, Ronald Culture shift in advanced industrial society. Princeton: Princeton University Press The silent revolution: Changing values and political styles among Western publics. Princeton: Princeton University Press Princeton.

14 Kitschelt, Herbert Diversifcation and Reconfiguration of Party Systems in Postindustrial Democracies. Kitschelt, Herbert, and Anthony J McGann The radical right in Western Europe: A comparative analysis. University of Michigan Press. Kleinnijenhuis, Jan, and Paul Pennings Measurement of party positions on the basis of party programmes, media coverage and voter perceptions. In Estimating the Policy Position of Political Actors, ed. Michael Laver. London: Routledge, p Knutsen, Oddbjørn Cleavage Dimensions in Ten West European Countries: A Comparative Empirical Analysis. Comparative Political Studies 21(4): (Accessed August 18, 2011) The Impact of Structural and Ideological Party Cleavages in West European Democracies: A Comparative Empirical Analysis. British Journal of Political Science 18(03): (Accessed August 18, 2011). Kriesi, Hanspeter et al Globalization and the transformation of the national political space: Six European countries compared. European Journal of Political Research 45(6): West European politics in the age of globalization. Cambridge University Press. de Lange, Sarah L A New Winning Formula? Party Politics 13(4): Laver, Michael, and Ian Budge, eds Party Policy and Government Coalitions. Houndmills: The MacMillan Press. Mair, Peter Electoral Volatility and the Dutch Party System: A Comparative Perspective. Acta Politica 43(2-3): (Accessed August 12, 2011). Middendorp, Cees P Left-right self-identification and (post)materialism in the ideological space; their effect on the vote in the Netherlands. Electoral Studies 11(3): /2/076da314b29d40e e2c Pellikaan, Huib, and Annemarie S. Walter New Cleavages in Western Europe: The Advantage of the Confrontational Approach. Pellikaan.pdf. Poole, Keith T Spatial Models of Parliamentary Voting. Cambridge: Cambridge University Press. Riker, W.H The strategy of rhetoric: Campaigning for the American Constitution. Yale University Press. Schattschneider, Elmer E The SemiSovereign People A Realists View of Democracy in America 143 The Semisovereign People: A Realist s View of Democracy in America. Holt, Rinehart and Winston.

15 1-dec-72 1-jul-74 1-feb-76 1-sep-77 1-apr-79 1-nov-80 1-jun-82 1-jan-84 1-aug-85 1-mrt-87 1-okt-88 1-mei nov jun jan aug mrt okt mei nov jun jan aug mrt-09 Appendix 1: Figures and tables 1 0,9 0,8 0,7 0,6 0,5 0,4 0,3 1D 2D 0,2 0,1 0 Figure 1. R 2 for the one- and two-dimensional MDS solutions, the Netherlands ,8 0,6 0,4 0,2 0-0,2-0,4-0,6 cda christenunie d66 groenlinks pvda pvv sgp sp vvd -0,8 Figure 2. Party positions on dominant dimension, the Netherlands Note: each data entry represents the average position over one parliamentary term, while the displayed date is the midpoint of the term.

16 Dimension 2 Dimension Gal-Tan European Parliament 0.5 Economy Left-Right Asylum EU Unification groenlinks sp pvda d66 EU Foreign Policy vvd 0.0 cda -0.5 rpf gpv sgp Dimension 1 Figure 3. MDS configuration of parties according to parliamentary voting behavior, the Netherlands June September Ethical 0.5 Left-Right Decentralisation groenlinks sp 0.0 pvda Environment d66 lpf Deregulation Taxes vs Spending vvd Immigration cda -0.5 christenunie sgp Dimension 1 Figure 4. MDS configuration of parties according to parliamentary voting behavior, the Netherlands May 1998 March 2001.

17 Dimension 2 0,8 0,6 0,4 0,2 0-0,2-0,4-0, cda christenunie d66 groenlinks pvda pvv sgp sp vvd -0,8 Figure 5. Party positions on dominant dimension, the Netherlands Note: each data entry represents the average position over one parliamentary term, while the displayed date is the midpoint of the term. Metric MDS with expert survey issue vectors groenlinks pvdd sp d66 christenunie pvda cda sgp -0.5 vvd pvv Dimension 1 Figure 6 MDS configuration of parties according to parliamentary voting behavior, the Netherlands September September 2009.

18 1-okt-97 1-apr-98 1-okt-98 1-apr-99 1-okt-99 1-apr-00 1-okt-00 1-apr-01 1-okt-01 1-apr-02 1-okt-02 1-apr-03 1-okt-03 1-apr-04 1-okt-04 1-apr-05 1-okt-05 1-apr-06 1-okt-06 1-apr-07 1-okt-07 1-apr-08 0,80 0,70 0,60 0,50 0,40 0,30 0,20 0,10 0, CDA SGP ChristenUnie LPF Leefbaar NL PVV VVD D66 PvdD PvdA GroenLinks SP Figure 7. Kruskal stress values per party for the one-dimensional MDS solution, the Netherlands ,9 0,8 0,7 0,6 0,5 0,4 0,3 1D 2D 0,2 0,1 0 Figure 8. R 2 for the one- and two-dimensional MDS solutions, Denmark