The Development of the Education Cleavage at the Electoral Level in Denmark: A Dynamic Analysis

Transcription

1 The Development of the Education Cleavage at the Electoral Level in Denmark: A Dynamic Analysis By Rune Stubager 1 Department of Political Science University of Aarhus Universitetsparken 8 Aarhus C Denmark stubager@ps.au.dk Paper presented at workshop on Politicising Socio-Cultural Structures: Elite and Mass Perspectives on Cleavages at the 35 th ECPR Joint Session of Workshops, Helsinki, 7-12 May 27 1 The author is grateful for helpful comments and suggestions provided by Lise Togeby and Christoffer Green- Pedersen. Remaining errors and omissions are solely my responsibility.

2 Over the past three decades the socio-structural anchoring of electoral behaviour in advanced industrial democracies has been debated. During the first two decades the prevailing argument was that this anchoring has been seriously weakened and, in some cases, has disappeared all-together (cf. e.g. Dalton et al. 1984; Franklin et al. 1992). Instead, it has been argued, values and issues without structural links have come to the forefront of the voters minds on election day thereby dissolving the previous type of cleavage politics (cf. e.g. Dalton 22; Inglehart 1997). More recently, however, this conclusion has been disputed, on the one hand, by scholars aiming to document the resilience of social class as an important influence on voting (cf. e.g. the debate in Evans 1999b) and, on the other hand, by scholars analysing the development as reflecting a transformation as much as a decline of the relationship between social structure and voting (Knutsen 24). Under the last heading we may also list authors such as Kitschelt (1994; 1995) and Kriesi (1998; Kriesi et al. 26) whose analyses point to the increasing importance of education as an independent electoral influence in societies characterised by a globalised knowledge economy. From a different perspective, authors analysing the formation of values have noted the strong connection between education and authoritarian-libertarian values where high education groups consistently tend towards the libertarian pole while low education groups tend towards the authoritarian pole thus creating an objective conflict between the groups (Brooks 26; Van de Werfhorst & de Graaf 24; Weakliem 22). Combining these observations the contours of a new cleavage emerge: Just as the class cleavage was reflected in the opposition between a socialist left and a capitalist right represented in the political arena by the respective parties, the basic idea pursued here is that education is giving rise to a new cleavage that lies beneath what has been seen as a purely ideological conflict over New Politics issues or authoritarian-libertarian values as reflected in such topics as immigration and environmentalism a conflict that is represented by parties giving voice to the positions of the conflicting groups. The aim of the paper, is first to test this hypothesis about the existence of an education cleavage at the electoral level and, second, on this basis to explore, in a longitudinal framework, the dynamic of the development of the alleged cleavage. This latter aim has implications for the choice of setting in which to conduct the analyses. Thus, as can be verified by inspecting the details of the abovementioned studies the developments in the individual countries are often quite different. Probably, this variability reflects both institutional differences in terms of e.g. electoral rules and the differential behaviour of the political actors not least the parties (cf. Knutsen & Kumlin 25: 161). The impact of such factors raises doubts about the explanatory gain from comparative analyses of especially new and emerging cleavages: Particularly in the initial phases of the establishment of a cleavage we should expect country specific factors (the electoral threshold and/or 1

3 the mobilisation activities of fringe parties, e.g.) to play relatively large roles (cf. Evans 1999a: 332). By restricting focus to one country, on the other hand, it is possible to take into account the idiosyncrasies of the national political and social context, e.g. the organisation of the educational system, the configuration of the party system, and the policy stances of taken by the parties. This means that we can optimise the possibilities for detecting new developments because we can finetune the analytical measures of both the social structure and the electoral choices available to the voters to fit the specific context. To capitalise on these advantages, the analyses will focus on only one country. The country chosen is Denmark where New Politics issues have been subject to intensified political polarisation over the past two-three decades and where the political system is sufficiently open for new developments to manifest themselves (the threshold for a party to gain representation in parliament is as low as 2% of the national vote). In relation to the first aim of the paper, thus, Denmark constitutes a kind of critical case: If the hypothesised cleavage cannot be found here, then it is difficult to argue for its existence in other contexts characterised by less focus on New Politics issues and/or more rigid political structures. The paper unfolds as follows: First, it is necessary to establish the cleavage model that underlies the analyses. Then, the ideological core of the new cleavage authoritarian-libertarian values has to be defined before turning to a short discussion of central aspects of the macro-social development. With these elements in place a set of hypotheses that can guide the analyses can be developed before turning to the actual analyses in the fourth section. The final section is devoted to a discussion of the dynamics behind the development of the proposed cleavage. 1 Theory, Background, and Hypotheses 1.1 The Cleavage Model Cleavages have been subject to scholarly interest for an extended period of time. Yet, their definition is contested. The central reference is Lipset and Rokkan s Cleavage Structures, Party Systems, and Voter Alignments: An Introduction from 1967 and most subsequent research follow leads contained in this seminal piece. The article itself does not contain a clear definition of the cleavage concept, however. Such a definition has, on the other hand, been presented by Bartolini and Mair in 199 and refined by Knutsen and Scarbrough in 1995 and it is their approach that forms the foundation for the use of the cleavage concept in the present context. Cleavages are defined, hence, as consisting of three interrelated elements: a socio-structural, a psychological (consisting of an interest and a consciousness aspect 2 ), and an organisational. In line 2 Due to data limitations, we have to disregard the consciousness aspect in the following. 2

4 with much existing research (cf. also Zuckerman 1975: 238), further, the definition revolves around the existence of conflict as the fundamental aspect of cleavages: Cleavages arise because sociostructurally defined groups disagree about certain matters. To be able to speak of a cleavage in accordance with the model presented here, thus, it is crucial to document that socio-structurally defined groups have developed opposing interests that may be either material or non-material (i.e. values) and that the members of the opposing groups support those political parties that articulate the interests of the groups in the political arena. The underlying conflicts can be seen as erupting due to macro-social developments that impact the (material or non-material) interests of the socio-structurally defined groups. This impact arises in so-called critical junctures (cf. Lipset & Rokkan 1967: 37) where new issues become salient due to macro-social changes that elevate them onto the societal agenda. Cleavages may develop, thus, where such newly salient issues are related to the interests of social groups in different ways where, in other words, the groups have conflicting views on how to respond to the issues. It should be noted that, in keeping with the tradition of cleavage analyses, the model proposed is not of a deterministic character. Thus, for cleavages to come into existence parties have to incorporate the conflicting positions into their platforms thereby articulating the positions and identities of the conflicting groups. All elements of the model need to be in place, that is. 1.2 Authoritarian-Libertarian Values Before proceeding to a discussion of the socio-structural background to the proposed cleavage it is relevant to briefly discuss its ideological content: New Politics or authoritarian-libertarian values. Such discussions are often marred by changing conceptualizations. To counter this problem, in the analyses that follow we shall apply a definition of authoritarian-libertarian values developed on the basis of the works of Flanagan (1987; Flanagan & Lee 23) and Kitschelt (1994). 3 Both authors use the term authoritarian-libertarian values to label a dimension which has the concept of hierarchy at its core: Authoritarians favour social hierarchy, that is the rank ordering of individuals in a system with a clear distinction between superior and inferior groups or persons (cf. e.g. Kitschelt 1994: 1; Flanagan & Lee 23: ). Such hierarchies are both natural and right to authoritarians (cf. also Jenssen 1993: 122). Libertarians, on the other hand, dislike social hierarchies and prefer the free and equal interaction of people without regard to social positions of any kind; in Jenssen s words libertarians favour parity of power in social interactions. In addition to the attitudes to hierarchy attitudes to tolerance for non-conformity form the second core aspect of the authoritarian-libertarian dimension. Entailed in a libertarian position, thus, 3 Although he uses slightly different labels this definition is also closely related to Feldman s (23) social conformityautonomy construct. 3

5 is a basic respect for and tolerance of other people including those who deviate from one s own norms or the norms of society. Variety among humans is, in fact, seen as something to be protected for its own sake. This is not so for authoritarians: Those who deviate from conventional norms or who stand outside society should first and foremost be made to comply with the norms of society. Their deviance (no matter at what level) is not accorded any value in itself; variety is potentially bad (cf. also Adorno et al. 195: ) and deserves no protection. In this way hierarchy and tolerance form the ideational figure underlying authoritarian-libertarian values. This is the core of the conflict which, as mentioned, sharply divides individuals according to their educational level: high education groups are generally libertarian in outlook while low education groups are authoritarian. 4 The concrete manifestation of the conflict may vary from one country to the other, however. The underlying values, that is, may play out in relation to different issues dependent on the sociopolitical and historical context. Thus, whereas the conflict e.g. has been reflected in the so-called culture wars over moral and religious issues in American politics (cf. Smith & Tatalovich 23), in the more secular countries of northern Europe (not least in Denmark) it has e.g. been reflected in the debate about immigration, criminal justice and environmental protection. 5 This point has to be kept in mind when constructing measures of the dimension (cf. below) as they need to tap into issues relevant to the dimension. 1.3 The Macro-Social Background As is clear from the discussion of the cleavage model above macro-social factors play an important role in the model by providing the impetus for cleavage development. Thus, to support the hypothesis regarding the existence of an education based cleavage it is necessary briefly to examine relevant aspects of the development in Denmark. The examination is entirely instrumental in the sense that it will focus narrowly on factors that are related to the proposed cleavage and may play the roles expected by the cleavage model. It is left to others, that is, to conduct more thorough sociological analyses of modern Danish society. The natural place to start is with education itself. On this point the evidence is clear: There is a marked and steady increase in the educational level among Danes. Thus, according to Jacobsen (24) the proportion with tertiary educations has increased from 12% in 1983 to 22% in 23; and during the same period the proportion having only primary or lower secondary education dropped 4 This result has been found to hold up in cross-national analyses that also include countries outside Western Europe and North America (Weakliem 22). 5 In relation to authoritarian-libertarian values the central aspect of environmental protection is the question of whether man is hierarchically superior to nature thus opening the way for untamed exploitation or, rather, nature has intrinsic value that is worthy of tolerance and protection in its own right (cf. e.g. Cotgrove & Duff 198: 341). 4

6 from 52% to 35%. In other words, while the low education group is rapidly diminishing the high education group has grown to a size that makes it politically relevant in the sense that parties may be able to base their existence on catering to the interests of the group. This constitutes an important precondition to the establishment of an education based cleavage. In accordance with the model presented, cleavages do not simply arise whenever different structural groups with conflicting interests can be identified. Rather, it is necessary that the conflict of interests is activated at the structural level by events that put the conflicting perspectives in play. According to the model, this is what happens in critical junctures where problems are put on the societal agenda; problems for which the groups have different preferred solutions. In this context we need to examine, therefore, whether it is possible to identify macro-social developments in relation to authoritarian-libertarian values that have served to heighten their political salience and thus the salience of the proposed underlying conflict of interests. An overview of Danish politics through the past two and a half decades reveals that two of the New Politics issues (in particular) have experienced marked changes in their salience: the environmental issues and the issue of immigration. As for the former, the environment first made its way onto the public agenda in the late 196s and early 197s. Throughout the 197s, however attention oscillated at a rather low level but the mid 198s brought renewed focus on problems related to the pollution of air, sea, and ground spurred partly by media reports on the state of the marine environment and partly by the accident at the Chernobyl nuclear power plant. The result can be seen in the first column in Table 1. From a fairly low level in the early 198s the environment jumped up as one of the most frequently spontaneously mentioned problems that voters thought politicians should care about in 1987 (cf. also Goul Andersen 23b: 138). Detailed analyses by Goul Andersen indicate, further, that this increase in salience was not driven by increasing media attention but rather that the media attention was driven by the increasing environmental consciousness among the voters (Goul Andersen 1988: 396). Thus, we can conclude that the environment was put firmly on the public agenda in the late 198s and although the salience has dropped since (cf. Table 1) it appears justified to assert that these events can be seen as a critical juncture in relation to the cleavage model. 6 Table 1 about here 6 It should be noted that the term critical juncture is used as referring to developments that may appear to be of a less sweeping character than the critical junctures in Lipset and Rokkan s seminal article (the Reformation, the French revolution, and the industrial revolution). The term is used, however, because the operative mechanism is the same then and now: A critical juncture is characterised by the arrival on the agenda of new issues that generate conflict between social groups. In this way the historical as well as the more contemporary developments serve the same function in relation to the generation of cleavages. 5

7 Another such critical juncture has manifested itself in relation to the issue of immigration although a certain amount of immigration is no new phenomenon in Denmark. The number of guest workers and refugees that have arrived in the country over the past four decades, however, far surpasses anything seen before. And even though regulations have been tightened over the past years the proportion of foreigners from outside the EU, North America and the Nordic countries has increased fairly rapidly from a few per mille in 1967 to about 6 per cent of the population in 22 (Togeby 23: 38-39). As we can see in the second column of Table 1 the influx of immigrants has not gone unnoticed among the voters: From the mid-198s where the salience of the immigration topic boomed in response to a surge in incoming refugees (the increase does not appear in the table as it took place in non-election years) the topic has been among those mentioned as the most important for politicians to care about at every election since While the issue did not initially attract more widespread attention the 199s brought a steady increase in salience and the development culminated in 21 where the issue was seen as second in importance to only the somewhat vaguer issue of welfare (Goul Andersen 23b: 138). Also for the immigration issue, then, it seems plausible to argue that the macro-social development (in the form of the arrival of immigrants in the country in this case) has contained a critical juncture that has served to activate an issue related to authoritarian-libertarian values. 7 Since, as argued above, the environmental and immigration issues are strongly related to the authoritarian-libertarian dimension we can say that the result of these junctures was an activation of the dimension in the minds of the voters. Although the salience boosts happened with some time in between them the combined impact of the two issues has been to activate the conflict between authoritarian and libertarian values. This shows how different issues at different times may act as crystallisation points for the conflict underlying a given cleavage. 1.4 Hypotheses On the basis of the model and the selected features of the macro development it is possible to develop a set of hypotheses about the development of the proposed education cleavage that can guide the subsequent analyses. The first set of hypotheses pertains to the first aim, mentioned in the introduction, of identifying the new cleavage. Thus, in accordance with the cleavage model we should expect to find a relationship of some strength between the voters length of education and their position on authoritarian-libertarian values (H1). The existence of a new education based 7 Green-Pedersen (26) notes that the development among the electorate in the saliency of the issues of immigration and the environment is mirrored in the parliamentary agenda. Part of the development among the voters, thus, may be driven by the activities of the parties. An exploration of this question, however, requires data that are only beginning to be available; hence I shall not go into this point in the present paper. 6

8 cleavage presupposes that any relationship found between education and the values is not caused by other variables. The application of a full set of control variables is, however, deemed excessive in the present context where focus is on the longitudinal perspective. 8 On the other hand, to test the independence of the proposed new cleavage from the previously dominant class cleavage a measure of social class has been included in the analyses. Similarly, to provide a standard by which to judge the results for the education cleavage a parallel set of analyses is (in most cases) conducted for the class cleavage which pits the social classes against each other in a conflict of interests reflected in the opposition between socialist and capitalist values (henceforth referred to as economic values). When commenting on the results, main attention is focused on the education cleavage, however. Looking at the organisational level we should expect to find parties that adopt the stances of the educational groups and articulate them in the political arena (H2); and when bringing the elements together, the third hypothesis entails a relationship between education and values on the one hand and party choice on the other a relationship that can be expected to vary in strength across parties in accordance with the incorporation of authoritarian-libertarian stances in their platforms (H3). With regard to the longitudinal aspect of the analysis the main expectation is that the education cleavage strengthens over time in comparison with the class cleavage (H4). 2 The Crystallisation of a Socio-Structural Foundation: Social Background and Value Positions The first set of analyses focus on the linkage between the socio-structural level and the values. In accordance with H1 the basic expectation at this point is that the education variable will have the stronger relationship with authoritarian-libertarian values both compared to the class variable and to its effect on economic values. We begin by setting out the foundations of the analysis in terms of analytical procedure, data, and measurement. 2.1 Data, Operationalisation, and Measures Data for the analyses come mainly from the Danish election studies for the period 1979 to 25 for which roughly comparable indicators of authoritarian-libertarian values are available. Unfortunately a suitable measure of vocational education was not included prior to 1984 meaning that the core analyses cannot be pushed further back. Whenever possible, however, the entire time span will be analysed (details on the surveys can be found in the appendix). 8 One set of controls have, however, been applied although the results are not shown in the following. Thus, to guard against spurious effects caused by the rise in the level of education over time the relationship between length of education and class on the one hand and authoritarian-libertarian values on the other has been controlled for the impact of both age and cohort. The effect of these controls was negligible; the effects of the structural variables are not driven by differences in education across cohorts or by the respondents age, that is. Therefore, the results of the more parsimonious model without these controls are presented below. 7

9 The respondents length of education is measured by a five-category variable: Only primary or lower secondary education, upper secondary vocational education, short-cycle tertiary education (including respondents with a general upper secondary education) medium-cycle tertiary education, and long-cycle tertiary education. The variable, thus, is roughly comparable to UNESCO s ISCED scheme for categorising education (cf. Braun & Müller 1997: ). To avoid having too small categories on the variables the class variable has been coded in four categories: workers, the self-employed, and the high and low salariat. Thus, students and retirees have been excluded from the analysis in order to focus on the core of the class conflict. The construction of the value dimensions requires a bit more comment. Both dimensions were constructed as simple additive scales of three items and were recoded to run from (libertarian or socialist) to 1 (authoritarian or capitalist). The extended time frame implies that it was impossible to use the exact same items in all years, however, and workable substitutes had to be found in a number of instances. On the other hand, the comparability of the resulting measures is quite high thus allowing us to regard them as tapping into the same underlying value dimensions. Details of the scales including the wording of the items can be found in the appendix. Means and standard deviations for the scales used in the various years can be found in Table 2. Table 2 about here The most important point to note in the table is the absence of any clear trend on the authoritarian-libertarian value dimension. Whatever changes we may find in the following analyses, thus, they do not seem to be driven by any substantial changes in the aggregate position of the electorate on this value dimension: The position only varies to a limited extent and the changes display no clear trend. For now we just note this finding; below, however, we shall see that it is a particularly interesting result. For economic values, Table 2 conveys a similar message at least from 1981 onwards; no major trend is discernible, in other words, at the aggregate level. With interval level dependent variables the analytical tool of choice is OLS regression. However, due to the large number of potential coefficients to report from such analyses a strategy is needed that can condense the complexity of the results into a manageable format without sacrificing too many details. Since the aim is to determine the degree to which the two socio-structural variables influence the voters positions on the value dimensions an obvious candidate as statistical measure is the adjusted R 2 from regressions of the value dimensions on the socio-structural variables. As pointed out by Müller (1997: ) in relation to comparable measures from logistic regression analyses, however, PRE-type measures such as the adjusted R 2 may reflect both the effect of one variable on another and the composition of the electorate. Thus, when large parts 8

10 of the electorate is found in categories with a homogenous electoral behaviour the adjusted R 2 will be higher than when the majority of the voters are placed in categories with heterogeneous behaviour (cf. also Elff 22: 282). It is clear that these two processes are both interesting, but in the analyses it is relevant to try to separate them. We cannot therefore, rely entirely on reporting PRE-measures. Instead it is necessary to also include measures that pertain more directly to the relationships between the variables. In a regression framework such measures are contained in the estimated coefficients. Changes in the coefficients between two elections, thus, would reveal the first kind of changes in which we are interested effect changes. As we are working with categorical variables, however, the analyses produce a fairly large number of coefficients all of which cannot be reported. Therefore, a new measure δ (delta) is introduced. δ is defined as the standard deviation of the linear regression coefficients the βs associated with a given categorical variable in accordance with the following formula: δ = K 1 ( β β ) k K 2 where the variable has k = 1, 2, K categories. Just as the κ-index (Hout et al. 1995) on which it is modelled, the δ-index allows us to condense the different coefficients associated with a categorical variable (including the associated with the reference category) into one number. The value of δ, thus, is the average deviation between the K categories and the mean of the electorate measured in the same units as the dependent variable (cf. Manza & Brooks 1999: 264). We shall be looking for changes in the value of the δ-index as an indication of a changing relationship between the variables, then. In the following, changes in both the PRE-based adjusted R 2 and the δ-index will be reported. We shall refer to δ as measuring the effect of the independent variables on the dependent and to the adjusted R 2 as measuring their impact (comprising, that is, both the effect and the composition of the electorate). The analyses will be based on five different measures: For both length of education and class we shall report the values of the adjusted R 2 for bivariate analyses with the socio-structural variable as independent and either of the value dimensions as the dependent variable. From these analyses we will also report the value of the δ-index summarising the effects of the socio-structural variables. In addition we shall report the adjusted R 2 from analyses including both socio-structural variables in the analysis together. By comparing these figures to the others it will be possible to gauge the 9

11 degree to which the impact of the socio-structural variables overlap in influencing the development in the value dimensions. 2.2 Analyses: Cleavages at the Value Level The results of the first set of analyses are displayed in Figure 1. They fall in fairly clear patterns. Looking first at the relationship between the socio structural variables and economic values in Panel A we observe a tendency towards a weakening of the relationship between the socio-structural variables and the value dimension as measured by the adjusted R 2 (depicted on the left hand scale) for the combined impact of both socio-structural variables. We can also note a very strong parallelism and short distance between the curves for the adjusted R 2 for the class variable and the overall adjusted R 2. When compared to the course of the curve for the adjusted R 2 for education this clearly indicates that the class variable has a larger impact on economic values than education as could be expected. At the same time, however, we can also observe a clear and persistent weakening of the effect of the class variable (cf. the development in the δ-index depicted on the right hand scale) although it still retains a level well above that of education. This indicates that the declining tendency in the overall relationship is caused both by an effect change and a (relatively modest) change in the composition of the electorate. Figure 1 about here Turning then to the results for authoritarian-libertarian values in Panel B we find an almost equally clear but different pattern: The first thing to notice is the strengthening of the overall relationship between the socio-structural variables and the value dimension over time. This strengthening has increased the adjusted R 2 from a 1984-level around that found in Panel A to a 25-level which is almost three times higher than the comparable figure for economic values (.17 against.6). As was the case with the class variable in Panel A we find a clear relationship between the overall adjusted R 2 and that for the effect of the education variable. In Panel B, further, the class variable seems to have a much weaker impact on the values and adds very little extra explanatory power to the model. This finding is also reflected in the levels of the δ-indices clearly showing that education has the stronger effect. Looking at the δ-curves we can observe the absence of clear trends in their development. The strength of the effects of the socio-structural variables, in other words, is unchanged. In itself this is an interesting finding when compared to the declining effect of the class variable found in Panel A of the figure. The weakening of the effects of the socio-structural variables, apparently, is not a ubiquitous phenomenon. However, the stability in these effects also indicates another thing: The 1

12 increase in the value of the adjusted R 2 for the overall model (and due to their strong connection the education variable) is, at least partly, driven by changes in the composition of the electorate. To a certain degree, thus, the increase harks from the growth in the high education groups which is a consequence of the increasing level of education in the population discussed above. For the class variable we can as for education in Panel A observe a high degree of parallelism between the R 2 and δ-curves, thus indicating that compositional changes play a minor role in relation to the effect of this variable. The conclusion that seems to follow from the results presented above is one of support for hypothesis H1. We did find the expected increase in the impact of education on authoritarianlibertarian values. The increase seemed to result in no small part from changes in the composition of the electorate. For economic values we found a decreasing impact of the class variable resulting in a downward tendency in the relationship between the socio-structural variables and the value dimension. 9 This indicates a weakening of the class cleavage at least at this level of analysis a development that may have repercussions at the electoral level as well (cf. below). A comparison of the results for the two value dimensions shows clear differences both in the course of the development of the relationship with the socio-structural variables and in relation to the question of which of these variables has the stronger impact. This strengthens the interpretation of the education-based authoritarian-libertarian cleavage as an increasingly important phenomenon qualitatively different from the class cleavage. 3 The Development of Cleavage Party Profiles The next step in the analysis is the examination of the stances of the parties on authoritarianlibertarian values. Thus, to be able to support H2 it should be possible to identify parties with a clear profile in relation to the values in focus here profiles that are expected to become increasingly clear over time that is. It should be pointed out that, in accordance with the focus of the paper on the electoral level, the analyses of this section merely aim at establishing the existence of a differentiation among the parties on the authoritarian-libertarian dimension. The factors underlying the dynamic are not analysed, in other words. Before launching into the analyses we shall briefly discuss the underlying operationalisations. 9 To complete the picture, it should be added that the relationships between the structural variables and the value dimensions are as could be expected with the higher salariat and the self-employed as the most capitalist and the medium and long-cycle tertiary education groups as the most libertarian. The typical difference on the authoritarianlibertarian scale between the highest and lowest educational categories is some 25 scale points. 11

13 3.1 Measuring Party Positions across Time: Operational Choices An obvious source of data with which to measure the positions of the parties over time is the Comparative Manifesto Project-data set (cf. Budge et al. 21). However, as pointed out by Hansen (26) there are a number of problems associated with the Danish part of the data. In the present context the most serious problem is the lack of a tradition, in Denmark, for producing comprehensive election manifestos covering (nearly) all policy areas. Thus, the manifestos that are coded are often fairly short and focus on the issues that the parties want to campaign on. This means that the Danish data may be characterised by a higher degree of volatility than that found in other countries. To counter this problem it is necessary to construct the measure as a moving average that calculates the positions of the parties at any give election based on their manifestos from that particular election as well as the previous one. The choice of weights is essentially arbitrary, but I have opted for a 75/25 split over the two elections. 1 The actual measure is based on those categories in the data set that pertain to authoritarianlibertarian values (thereby opening for a more focused analysis than when the overall left-right dimension is taken as point of departure, cf. Knutsen & Kumlin 25: 159). Thus, the libertarian pole is operationalised by the categories Internationalism: positive, Anti-Growth Economy, Environmental Protection, National Way of Life: negative, and Multiculturalism: positive ; the authoritarian pole is operationalised by the Internationalism: negative, Law and Order, National Way of Life: negative, and Multiculturalism: negative categories. The construction of the scale follows a procedure developed by Elff (22: 291) which calculates the position on the authoritarian-libertarian dimension (autlib) of party j by means of the following formula: 11 autlib j ( Internationalism : negative + Law and Order ) j j ( ) = ln Internationalism : positive j + Anti - Growth Economy j As pointed out by Elff this procedure has the advantage of removing the influence of the categories not included in the actual measure which constitutes a statistical problem in relation to measures constructed by simply summing the percentages for each category (cf. Budge 21: 88-89). Furthermore, the resulting scale can take on both positive and negative values and has an almost symmetrical distribution which makes it more appropriate for statistical analyses. The analyses will cover the period since the extension back to 1979, as we shall see below, permits a better evaluation of the development. As for the parties, the Christians, the Minority Party, and the Centre Democrats are excluded from the analysis as there are too few respondents in the election surveys to permit reliable analyses. Additionally, prior to 1994 the left 1 Substantially similar results were obtained with a 6/4 split or a 6/3/1 split over three elections. 11 The addition of 1 in both numerator and denominator is necessary to avoid the possibility of division by. 12

14 wing parties are represented as a combination of the positions of the Left Socialists, the Danish Communist Party (both 1979 to 1984) and Common Course (in 1987). In 1988 and 199 where no manifesto data is available for left wing parties I have interpolated their position on the basis of the 1987 data. 12 In the presentation, the left wing has been combined with the Socialist People s Party in the Socialist category based on an equally weighted average, thereby assuring comparability with the analyses at the voter level where left wing voters constitute a small group (cf. below). With these preliminaries we are ready to begin analysing the manifesto data. 3.2 Party Manifesto Profiles: The Establishment of Difference The results of the analysis presented in Figure 2 document the advantage of extending the analysis back to 1979: The increasing polarisation of the positions of the parties appears very clearly. We can observe, thus, how the parties, to varying degrees, drift away from the centre around over the period. The development is particularly noticeable from 199 to 1994 and the polarisation in 25 is markedly stronger than in 1979 (which, it should be added, marks the high point of differentiation for the 197s). Figure 2 about here The overall pattern in the positions of the parties is relatively stable (particularly from 1994 onwards) with a division into a libertarian group consisting of the socialists and the social liberals on the one hand, and an authoritarian group made up of the liberals, the conservatives, the Progress Party/Danish People s Party, on the other. For most of the period examined, the Social Democrats belonged to the libertarian group, but in later years they have moved closer to the authoritarian group. In that group, the Progress Party/Danish People s Party constitutes the authoritarian extreme (except for 1994). The two other parties in the authoritarian group the liberals and the conservatives display an interesting trajectory taking them from positions close to that of the Progress Party in 1979 over moderate stances in the (late) 198s and back to a position around that held by the Progress Party from 1994 onwards. Combined with the persistent movement of the first group of parties in a libertarian direction over the 198s it is this marked movement towards the authoritarian side by the liberals and the conservatives that establishes the clear differentiation in the parties profiles found in the 199s. One may speculate, thus, that whereas the two parties were dragged towards the libertarian side by the other parties (except the Progress Party) in the 198s, they changed course from 199 to The position of the parties is, simply, set equal to their 1987-position. For 1994 the position of the Unity List is based entirely on the data from this year without performing the usual weighting. 13

15 and turned to the authoritarian side of the spectrum in response to the mobilising activities of the Progress Party. The path of the Social Democrats seems to mirror this movement although with a time lag: While following the social liberals and the socialists towards the libertarian side in the 199s, they have been forced to turn towards the authoritarian side in later years possibly in an attempt to block the advances of the three parties in the authoritarian group. The overall development, then, entails an establishment of two clearly differentiated party groups within which certain parties can be seen standing out as exponents of the education-based authoritarian-libertarian cleavage: the social liberals and the socialists on the libertarian side and the Danish People s Party (/the Progress Party) on the authoritarian side. The latter occupy a marked (and in all years but one, the most) authoritarian position while the social liberals and the socialists oscillate a bit more although always clearly in a libertarian position and since 1984 always constituting the libertarian extreme. 13 On this background it seems justifiable tosee these parties as the main cleavage parties and the results show a fairly high degree of stability in their positions over the period analysed. 14 The increasing polarisation of the parties on the authoritarian-libertarian dimension, further, may be taken as an indication of an increasing importance of this dimension to party competition. By their differentiation on the dimension the parties enable themselves to attract voters from different value positions thereby opening the way for authoritarian-libertarian values to become an important factor influencing party choice. This is entirely in accordance with the expectation regarding an increasing importance of this value dimension over time and it also squares well with the strengthening in the connection between education and authoritarian-libertarian values discussed above. On this background it seems fair to say that Figure 2 shows the establishment of the cleavage at the party level, thus supporting H2. 4 Putting It All Together: Cleavages in Electoral Behaviour In the over-time analyses of the electoral behaviour of the Danish population I shall retain the focus on the central variables introduced in section 2 above. Within this framework, the analyses are designed to permit a test of the most important of the hypotheses formulated at the beginning of the chapter: H3 pertaining to the connection between the voters socio-structural background and values, on the one hand, and their voting choices on the other hand. 13 Analyses of the voters perceptions of the positions of the parties (not shown) document that the voters view the parties in an essentially similar way. The positions adopted by the parties translate to the electorate, in other words. 14 As noted above the role played by election manifestos in Danish politics makes it unrealistic to expect a high degree of stability in the parties positions on the dimension constructed here. Instead we have to focus on the overall pattern in the data as well as the existence of clear and persistent tendencies as is the case in the discussion here. 14

16 More precisely we should, first, expect to find an increasing importance of the length of the voters education to their voting decisions. Particularly we should expect to find this increase for the core cleavage parties the social liberals, the socialists, and the Progress Party/Danish People s Party. Likewise we should expect an increasing impact of authoritarian-libertarian values particularly in the case of these parties. Second, the impact of education and authoritarianlibertarian values is expected to increase relative to that of class and economic values although to varying degrees for the different parties (H4). We might expect, thus, that the latter two variables retain some higher degree of impact for the parties that have traditionally been the major opponents on the class cleavage: the Social Democrats, the liberals, and the conservatives parties that were found in the previous section to occupy less extreme positions on the authoritarian-libertarian dimension than the other parties. Before commencing on the analyses we turn to a discussion of the procedure. 4.1 Analysing Party Choice across Time: Analytical and Operational Choices The analyses will, basically, be conducted along the lines laid out in section 2; thus, estimates of both a PRE-type and a coefficient-based measure will be presented. However, with a categorical dependent variable party choice the analyse will be conducted by means of logistic regression; hence relying on Nagelkerke s R 2 as the PRE-type measure of the impact of the independent variables (for similar uses of the measure, see Berglund 23; Knutsen 24). 15 However, as discussed above PRE-type measures entail interpretational problems in analyses like the ones conducted here (cf. Müller 1997: ). For this reason we need to inspect coefficient-based measures for the effect of the variables as well. To summarise the coefficients associated with the categories on the class and education variables we may use the κ-index which was developed for this purpose (cf. Hout et al. 1995). For the two interval level variables, authoritarian-libertarian and economic values, we can use the estimated logistic regression coefficients that may be compared to each other since the two variables are measured on the same scale. Due to the categorical nature of some of the independent variables we are faced with the problem of perfect prediction (Long 1997: 59; Long & Freese 23: ). One solution to the problem would be to collapse some of the categories on the variables. However, this would substantially reduce the potential for arriving at interesting findings and has, therefore, been ruled out. Instead we have to deal with perfect prediction which creates problems in relation to the 15 Nagelkerke s R 2 is based on the proportional increase in the likelihood function by using the fitted model instead of the null model (Nagelkerke 1991). 15

17 estimation of coefficients for those combinations of the independent and dependent variable for which there are no observations in the samples: The estimates, simply, cannot be produced. In relation to Nagelkerke s R 2 this is less of a problem. Perfect prediction implies that the maximum likelihood procedure does not converge on a set of estimates or, in other words, that the maximum of the likelihood function cannot be found because of the indeterminable estimates. While attempting to derive these estimates, however, the procedure will end up with the best possible estimates for the parameters that can be determined and, after a few futile iterations trying to determine the remaining estimates, the procedure will come to a halt. The value of the likelihood function at this point will be the maximum attainable under the given data conditions and will, therefore, be a conservative (i.e. lower) estimate of the value that could be attained in the absence of perfect prediction. Hence, the value of Nagelkerke s R 2 based on the attainable likelihood value can be used as a conservative estimate of the value of the measure that could have been obtained with other data (typically a larger sample). With regard to the coefficients of the variables we can use those for variables without the perfect prediction problem while the coefficients for variables with the problem will have to be left out of the analysis. To minimise the influence of the problems related to perfect prediction and to be able to present a set of coefficients for each party the analyses will focus on the individual parties. For this purpose it is possible to use binomial logistic regression where the dependent variable is a dichotomisation of the respondents into those voting for a given party and those voting for all other parties. From these analyses I shall present Nagelkerke s R 2, the κ-index, as well as the logit coefficients for the value variables which directly provide information about the unique features of each party in relation to the overall party system. As in section 2, analyses of a number of different models are conducted. First, by means of Nagelkerke s R 2, measures of the bivariate impact of the two socio-structural variables on party choice will be presented. As above, the bivariate approach has been chosen in the interest of parsimony. Second, the analyses are extended to the value level by estimating models including both of the socio-structural variables and each of the value dimensions, thereby taking the causal sequencing of the variables into account (cf. the cleavage model). These models will allow for an evaluation of the structure-independent contribution of the value dimensions. Finally, the total impact of both socio-structural and value variables is analysed. A comparison of these results with those obtained in the second step will permit evaluations of the overlap between the impacts of the two value dimensions, thereby allowing us to determine which is the more important. The coefficient-based measures will come from, roughly, the same types of models. I shall, thus, present the bivariate effects at the socio-structural level while the coefficients for the value variables will come from the full models, i.e. where all four independent variables are included. 16

18 This will give us the unique effect of each value variable (i.e. that part of their effects which they do not have in common). 16 The coding of the socio-structural and value variables follows the procedure described in section 2 while party choice is coded with seven categories Depolarisation and Polarisation: Values and Vote As a preliminary to the multivariate analyses of the two cleavages we shall examine the development in the mean positions of the parties voters over the period from 1979 to 25 for which we have (fairly) comparable data. This analysis will provide a first indication of the level of support that can be found for the expectations about the development of the cleavages. Two methodological points deserve mentioning. For economic values, first, I am relying entirely upon Version 1 of the scale. This means that data for 1987 and 1994 cannot be presented. For 1984, instead of the class survey used for the other analyses (cf. the appendix) the election survey from that year can be used. This survey contains the version of the scale used in all other years (except 1987 and 1994), thus providing the information that is lacking in the class survey. Second, for authoritarian-libertarian values I am forced to rely on versions of the scale that are not strictly comparable if we want to push the analysis back before As mentioned above, however, there is a strong correlation between the two versions of the scale and I shall, thus, go forth with the analyses. Still, we saw in Table 2 that the level of the aggregate mean varies considerably between the scale versions (probably due to the inclusion of the fairly skewed criminal justice item in one version and not in the other). In order for this variation not to confound the analysis the positions of the parties are presented as deviations from the aggregate mean. This procedure will allow us to focus on the relative positioning of the parties which is of main interest. Further, the procedure also eliminates the confounding effects that may arise due to the influence of policy shifts on some of the items included in the scales (a so-called thermostat effect, cf. Wlezien, 1995). The positions of the voters for the different parties on the two value dimensions are presented in Figure 3. Figure 3 about here 16 Unfortunately, inferential statistics are not available for comparing Nagelkerke s R 2 and κ-values. For the value variables, however, significance tests of changes between the years can be conducted. 17 The categories are: Social Democrats, social liberals, conservatives, Socialist People s Party and left wing (the socialists), liberals, Danish People s Party and/or Progress Party, and Christian Democrats and Centre-Democrats (combined). I shall, however, refrain from commenting on the two last very small parties. Left wing parties are the Left Socialists, the Danish Communist Party, Common Course, the Greens, the Socialist Workers Party, the Communist Workers Party and the Unity List. Only the first and last of these parties received any noteworthy support in the samples (Common Course did, though, gain representation in the Folketing from 1987 to 1988). 17