Party Identification and the Vote. Six European Countries Compared

Party Identification and the Vote Six European Countries Compared by Frode Berglund, ISR Oslo Sören Holmberg, University of Gothenburg Hermann Schmitt, MZES, University of Mannheim Jacques Thomassen, University of Twente Paper prepared for presentation at the Second General Conference of the ECPR, University of Marburg, Germany, September 18-21, 2003. Authors are listed in alphabetical order. 1

1. Introduction Previously, we found a secular decline of the impact of the traditional cleavage structure on people s party choice. 1 As the strong and lasting relationship between the social structure on the one hand and people s political loyalties on the other hand used to be one of the most essential elements of in most European democracies, this development might be a threat to political stability. However, social roots are not the only possible basis for enduring party preferences. An alternative is the direct psychological attachment to a particular party. In the Michigan tradition of electoral research party identification has the function of ensuring people s lasting attachment to a political party. Party identification is a long-term, affective, psychological identification with one s preferred political party. At first, the concept of party identification was met quite critically in European electoral research. As in European elections party identification and party vote tended to coincide and to change in tandem, one of the most innovative elements of the use of party identification in electoral research, i.e. studying the impact of short term forces on the vote against the baseline of the normal vote, was hardly applicable in the European context. Paradoxically, one of the major reasons why party identification, at least the directional component of the concept, used to be less useful in the European compared to the American context, was the overriding importance of political parties. In European parliamentary systems political parties and not individual politicians are the principal actors in the interaction between voters and governmental institutions, leaving little lee-way for individual candidates to run their own campaign for office and offering little incentives for voters to deviate from their party preference in favour of an individual candidate from another party. This is quite different from Americans politics, in particular in presidential elections, where the policy stands and characteristics of the candidates have an important weight in addition to their party background. In a parliamentary democracy with its indirect election of the head of government a split of party identification and the actual vote is less likely. The functional model of party identification (Shively 1982) offers another possible explanation why party identification might be less important for European than American voters. According to this view party identification is a cost saving device, providing people 2

with a short cut to all kinds of decisions, including the decision for which party or candidate to vote. However, European voters might not have needed party identification as a cost saving device, because that function was already fulfilled by people s ties to a social class or religion, which in turn was strongly associated with a particular political party. Under these conditions expressed partisanship will be synonymous with the vote and parties as such will not serve as guides to organise behaviour. The logical extension of this argument might be that once the importance of the traditional cleavage structure will decline and voters are deprived of their traditional shortcutting device, they will develop party identification in the same way as their American counterparts. In other words party identification would not necessarily decline in the slipstream of the decline of the relationship between the social structure and party system. However, it is the very same functional theory that contradicts this prediction. The need of an information costs saving device is supposed to be a function of the political skills of the voters. Modernisation and in particular the cognitive mobilisation of citizens have increased these skills. According to Dalton, the dramatic spread of education in advanced industrial democracies has produced a qualitative change in the political sophistication of citizens. At the same time, these societies have experienced an information explosion through the mass media. Both developments have led to a substantial decrease of information costs. Because of this cognitive mobilisation, more voters now are able to deal with the complexities of politics and make their own political decisions. Thus, the functional need for partisan cues to guide voting behaviour is declining for a growing number of citizens (Dalton 1984). As a consequence, we should expect a decline of the level of party identification. In principle, testing this hypothesis requires not more than an assessment of the development of the level of party identification over time. The hypothesis predicts a secular decline of party identification in advanced industrial democracies. An advantage of the selection of countries in our study is that all six are advanced industrial democracies at about the same level of development. Modernisation theory therefore would predict a more or less similar secular decline of party identification in these six countries. This hypothesis will be tested in section two. However, even if the development of the level of party identification is consistent with what the functional theory as part of the broader theory of modernisation predicts, this does not 1 This is a draft chapter of a book that is forthcoming at Oxford University Press under the title The European 3

necessarily prove that this development is caused by the mechanisms supposed in functional theory. This theory is based on specific assumptions at the micro-level, which are anything but uncontested (Barton & Döring 1986). In section three we will elaborate these assumptions and test to what extent the assumed relations at the micro-level can be found in our data. The theory of modernisation tries to explain developments in politics by developments in society. As such one might call it a sociological approach, devoid of a political component in the explanans. As a reaction to this a-political approach an alternative explanation of the development of party identification has been proposed emphasising developments in the political context, in particular in the demand side of politics, rather than the developments in society. Schmitt and Holmberg (1995) for instance introduce and test the hypothesis that developments and fluctuations in the level of party identifications can be explained by simultaneous fluctuations in the level of polarisation and the degree of issue conflict between political parties. This approach does not necessarily lead to different predictions than the theory of modernisation. In the early end of ideology debate a gradual decline of the ideological political differences was predicted as a logical consequence of the decline of the importance of the cleavage structure. As far as the decline of the level of party identification is a consequence of such a gradual depolarisation, it should be a secular decline, just like predicted by the theory of modernisation. However, polarisation can fluctuate from one election to the other and differ across countries at the same level of modernisation. Therefore, other than the theory of modernisation, it can possibly explain non-linear fluctuations in the level of party identification. In section four we will further discuss this theoretical perspective and test it. Section five finally will be devoted to party identification as an independent variable: has modernisation and/or a possible decrease in ideological polarisation made it less closely tied to the vote? 2. The Evolution of Party Identification The question whether or not a secular decline in the level of party identification has occurred in advanced industrial democracies and whether modernisation is an underlying cause of decline, has been the subject of several previous comparative studies. Holmberg and Schmitt, having studied the developments in fourteen West-European countries and the United States Voter. The editor is Jacques Thomassen. 4

until 1992 came to the conclusion that If there is an overall tendency, it is of loosening party bonds. But specific developments, by country and party, are so varied that any general overall view disguises more than it discloses. (Schmitt & Holmberg 1995: 121). On the basis of a larger number of countries and a longer time span, regressing the year of the survey on trends in partisanship, Dalton (2000) comes to a less ambiguous conclusion. He finds negative trends in 17 of the 19 advanced industrial democracies represented in his study. In this chapter we will see to what extent the trends in partisanship in the six countries included in our study confirm these trends. Limiting ourselves to these six countries might have the disadvantage of a smaller database, but on the other hand it clearly has two major advantages. First, Dalton (2000) objects to the study by Holmberg and Schmitt that they include countries which have not yet become advanced industrial countries, and that many time-series are too short to reveal effects of cognitive mobilisation. An analysis based on national election surveys from Great Britain, Denmark, Germany, the Netherlands, Norway and Sweden, avoids these problems. All are old democracies, they were traditional industrial societies in the sixties and have transformed into advanced industrial societies in the nineties. Secondly, the limitation to six countries enables us to study the development in these countries in more detail. Figure 1 about here The development of the level of party identification for each of the six countries is presented in figure 1. In two countries at least the development is pretty much in line with what the modernisation thesis would predict. These countries are Great Britain and Sweden. Although the development in neither country is strictly linear, there can hardly be a dispute about the direction of change: down we go (Holmberg 1998). In Britain the decline started in the early 1970s, and continued for more than a decade. The decline came to a halt in the 1980s, i.e. in the Thatcher years, but went further in the 1990s. The pattern in Sweden is very similar, although the development over time is a little bit different. After an initial decline in the 1960s, the trend stabilised in the 1970s, to be followed by a continuous decline ever since. In the other three countries the development is more complex. In Norway a secular decline can be observed since 1980, but this decline was preceded by an increase of the level of party identification in the 1970s after an initial decline in the 1960s. Neither in Denmark nor in the Netherlands a clear trend can be observed. After a sudden decline in the early seventies, in 1973 to be precise when all of a sudden the party system was shaken up, the level of party 5

identification slowly clambered, but slowly declined again in the 1990s. In the Netherlands a sudden increase in the 1980s interrupts a trend that otherwise is a negative one. The German trend starts out at about thirty percent in 1961, reaches its highest level well above 50 percent in 1972, but then continuously declines afterwards, falling back to the 30 percent mark again with the major exception of the re-unification election in 1990. If we conduct a linear regression analysis on the numbers in figure 1, the technique used by Dalton in his analysis of changes in partisanship, we find compelling evidence that partisanship is declining with time, which is consistent with the modernisation thesis. In all countries partisanship is lower in the late nineties than in the sixties and the early seventies. However, it is disputable whether a linear technique is a proper technique in analysing these trends. It seems more likely that the fluctuations and deviations from a linear trend are due to periodical effects and country specific events. Therefore, we applied a non-linear approach to the time variable (year) in each country in order to test for periodical effects. Table 1 shows the results from logistic regressions with partisanship as dependent variable and time as explanatory variable. Partisanship is dichotomised partisan or independent, while time is treated as a nominal variable. The first election constitutes the reference category, positive signs indicate an increase of independents relatively to the first election, negative signs a decrease. Table 1 about here Table 1 confirms the impression from figure 1. As the signs of most of the coefficients in the table are positive, it is obvious that there is general trend towards a lower level of partisanship. In all countries but Denmark the percentage of partisans at the end of the trend line is lower than at the beginning. However, it is also clear that partisanship has developed in a more complex way than the modernisation thesis would predict. At least part of he development seems to be due to nation-specific characteristics, like Reiter (1989) suggested. It is hard to link the various patterns in the development of partisanship between countries to one general explanation associated with modernisation, at least if time is taken as an indicator for the process of modernisation. Not in a single case does time explain more than six percent of the variance (pseudo-r 2), as can be seen at the bottom line of the table. Given the limited explanatory power of the modernisation thesis we will continue our analysis by two consecutive steps. First, in the next section we will test the micro-theory underlying modernisation theory. Secondly, in section four and five we will in the spirit of Przeworski 6

and Teune see to what extent we can replace what seem to be nation specific causes by more general variables. 3. Dealignment and Cognitive Mobilisation In order to test the validity of the underlying assumptions of modernisation theory at the micro-level we first need to specify which aspects of modernisation are relevant. These can be derived from functional theory. The functional model argues that partisanship is becoming less relevant for electorates and predicts a lasting loss of loyalties to political parties in general. Theories of mass politics proposed that party identification had a functional value for citizens in industrial societies, as parties provided political cues that guided them through the complex world of politics (Campbell et al. 1960). Not just regarding the vote, but also in shaping public opinion and in evaluating political events. Party identification was of special importance for a certain segment of the electorate, as partisanship provides a clear and lowcost voting cue for the unsophisticated voter (Converse 1975:111). As political sophistication was not considered to be widespread in the industrial society (Converse 1964), the functional value of party identification was considered to be high in that kind of society. One consequence of the transition to advanced industrial societies is that the functional value of partisanship will vanish. The transition brings about a radical improvement of citizens political resources, because of a sharp rise in the level of education. At the same time, the media-explosion leads to an increase of the amount of (political) information available. Thus, the media-explosion is associated with citizens becoming more capable in processing the information they receive. As sophisticated citizens do not need political guidance from parties, this development should lead to a decline of partisanship. This process is usually referred to as the process of cognitive mobilisation. It implies that ever more citizens possess the skills and resources necessary to become politically engaged with little dependence on external cues. In addition, cognitive mobilization implies a psychological involvement in politics (Dalton 1984:267). If we accept this definition, cognitive mobilisation can only be a cause of a possible decline of party identification when the following is true. 1. Education and political involvement have been increasing over time. 2. There is a stable negative relationship between education and political involvement on the one hand and partisanship on the other hand, i.e. the higher the education and political involvement, the lower the level of partisanship 7

Nationally established measures of education are used, recoded into the trichotomy primary, secondary, and university education. Political involvement is measured here by questions measuring a general interest in politics. Table 2 about here In table 2 the effect of time on the two indicators of cognitive mobilisation is presented. The results are totally different for the two indicators. In all countries the level of education has been increasing over time. This, of course, is not really surprising. However, the effect of time on political involvement is negative in four of the six countries, meaning that political involvement has been decreasing rather than increasing. The effects are small but are pointing in the wrong direction. Therefore, of the two conditions for cognitive mobilisation being a cause of declining partisanship, the first one is met for education but not for political involvement. The second condition refers to the relationship between the indicators of cognitive mobilisation and partisanship. As the two indicators refer to different dimensions of the concept, skills and involvement respectively, it would make little sense to combine them in a single index. However, the least we should expect of two aspects of the same concept is that they point into the same direction, i.e if higher education leads to a lower level of partisanship, higher political interest should do so as well. Otherwise, the indicators will counterbalance each other. Table 3 about here As can be observed in table 3, education and political interest do not point into the same direction. The relationship between political interest and partisanship is relatively stable across time and countries. However, the sign of the relationship contradicts the cognitive mobilisation hypothesis. All relationships are positive, which means that political involvement leads to a higher rather than a lower level of partisanship. The effect of education in general is lower than that of political involvement, but at least points in the expected direction. However, the effect is not significant in Germany and the Netherlands. In Britain, Norway and Sweden it is clearly declining over time. Only Denmark shows a more or less stable and significant effect. As the two components of cognitive mobilisation show a different pattern with regard to the two conditions it is impossible to draw an unambiguous conclusion from these data. However, it might be argued that of the two components of cognitive mobilisation education is the more 8

important one. Cognitive mobilisation literally emphasises cognitive aspects more than psychological involvement, the second element of Dalton s definition of cognitive mobilisation. Also, the main argument of the functionalist model is that modernisation leads to a politically more sophisticated electorate that no longer needs the costs saving device of partisanship to find their way in the world of politics. In this argument political involvement might be considered as a possible consequence rather than an factor of cognitive mobilisation. Skills are more relevant than involvement. Moreover, it is most likely that political involvement will co-vary with the political context in the same way as predicted for partisanship in the alternative explanation of changes in the level of partisanship that will be tested in the second part of this chapter. In the case of education this alternative explanation is impossible. Therefore, avoiding a mix up of the two alternative explanations might be a second reason to focus on education as the major indicator of cognitive mobilisation. Tables 4a/f enable us to look at the relationship between education and partisanship in more detail. These tables confirm that at least in Britain, Denmark, Norway and Sweden the findings are consistent with the cognitive mobilisation hypothesis in the sense that there are relatively more independents among people with higher education than among less educated people. However, in the Netherlands and particularly in Germany the pattern contradicts the hypothesis. Also, in Britain, Norway and Sweden the differences between the different levels of education have become smaller. This is mainly due to a decline of partisanship among people with a low education. This again is the opposite of what the cognitive mobilisation hypothesis would predict. Table 4 about here It might be obvious that the level of education is a function of age. The level of education among the population at large has been increasing due to generation replacement. Younger cohorts are much better educated than older cohorts are. At the same time age is considered as the most important factor in the development of party identification, not because age in itself is important, but because it is a proxy for the length of time people have experience with a particular party or the length of time people have been able to confirm their identification with a particular party (Converse 1969, 1976). On the other hand, in the context of the party decline thesis, it has been claimed that the life-cycle thesis is no longer valid for the post-war generation. Being better educated they are less likely to develop partisanship and something that does not come into being cannot be confirmed either. Regardless which interpretation is correct, it is obvious that in order to assess the net effect of education on partisanship one 9

should control for age. The results of this analysis are presented on the bottom line of tables 4a/f, where the correlation between education and partisanship for the youngest generation is shown. As far as we were able to discern the predicted relationship in at least some countries, this is no longer the case among the youngest generation. At first, the correlations are lower than among the population at large and then over time turn into the wrong direction. This means that among the younger generations people with a lower instead of a higher level of education tend to turn their back to political parties. Taking all evidence together there is little support for the cognitive mobilisation hypothesis and thus for the modernisation thesis. Therefore, in the next section we will turn to the possible alternative explanation of changes in partisanship, focussing on the political context rather than changes in society. 4. Party Polarisation Schmitt and Holmberg (1995) tested four political factors as possible causal agents behind decreasing levels of partisanship the degree of party polarisation, the extent of ideological conflicts, the intensity of issue differences, and the evolution of new parties and found positive, yet weak relationships. The decline in the strength of party identification tended to be related to weakening party polarisation, declining ideological conflicts, diminished issue differences and increases in the number of political parties. We will not replicate their study in this context. We will, however, test three related hypotheses. In doing that, we will use a different set of data when measuring the relevant political variables than Schmitt and Holmberg did. The hypotheses we propose to test are the following. Hypothesis number one is tested on the aggregated macro level using elections as the unit of analysis. It specifies that: - partisanship will be higher in electoral systems with more polarised parties than in systems with less polarised parties. Hypothesis number two is also tested on the aggregate level. Here we will take advantage of the time-series aspect of our election study data. The hypothesis states that - within countries and across time, partisanship will change with the level of party polarisation. If polarisation goes up we expect partisanship to increase and if party conflict decreases we expect partisanship to decrease as well. This means that we do not view party identification as 10

something truly steadfast and almost unmoveable like in the original Michigan conceptualisation (Campbell et al. 1960). Our hypothesis presupposes that people s degrees of party attachment can vary somewhat between elections depending on the nature of party competition. The third and final hypothesis of this section is on the micro level and states that people s degree of party identification is related to the extent that they perceive ideological differences between relevant political parties. It states that: - people who experience larger differences between parties tend to develop stronger party attachments. If parties flock to the ideological centre and appear as look-a-likes, it is difficult for people to identify with particular parties. Why pick one if they are all the same? In order to determine how polarised a party system is, we consider two data sources. One is the familiar left-right differential derived from the content analyses of party manifestos (Budge, Klingemann, et al. 2001). The other relies on survey evidence and takes the distance between the mean left-right self-placement of party voters as a measure of polarisation. Moreover, we consider the left-right distance between the two major parties and that of the two polar parties (among those that reached parliamentary representation) as separate and distinct indicators of polarisation. With regard to partisanship, finally, we look at proportions of all identifiers and proportions of strong identifiers as two separate and distinct criteria. After this, we can turn to the business of hypothesis testing. Is it indeed the case that the proportion of party identifiers in national electorates covaries with left-right polarisation? Over 6 countries and about half a hundred elections, we find that the more polarised a systems is, the more numerous party identifiers are (see first row of table 5). There is clear evidence that party polarisation produces partisanship no matter what criterion we apply with one exception: overall partisanship seems to drop when the ideological distance between the voters of polar parties rises. All other coefficients are positive, two of them statistically significant despite the limited number of cases: large distances between the two polar system go along with high proportions of strong party identifiers both according to MRG data (r=.42) and survey evidence (r=.32). Table 5 about here The empirical evidence is less clear with regard to the second hypothesis: the within-country diachronic perspective produces a variety of different patterns. If we start at the bottom of table 5, Sweden is characterised by positive co-variations between polarisation and 11

partisanship: all but one coefficients seem to indicate that polarisation indeed produces partisanship. However, things are exactly the other way around just across the border in Norway: there, rising ideological polarisation decreases partisanship no matter what criterion we look at. Findings for the Netherlands are similar to the Swedish: the higher polarisation, the more identification. In the German findings, the particular importance of the two major parties is standing out: the farther apart SPD and CDU/CSU voters are ideologically, the more partisanship there is. In Denmark, manifesto-based measures of partisanship are hardly correlated with partisanship, while survey-based indicators are: the higher the ideological distance between party electorates is, the more partisanship we find. For Britain, finally, the funny finding is that manifesto- polarisation goes along with declining partisanship, while survey-polarisation coincides with rising partisanship. It should have become evident at this point that things are considerable less uniform in individual countries, than one would expect on the basis of the findings of the pooled analysis. Obviously, there are factors intervening in the association between polarisation and identification which cannot be controlled for in the bivariate design of the current analysis (see Schmitt 2002 for a more complex study design). In order to test the third and final hypothesis, in each of our six countries and across all elections, we created an individual level variable measuring how people perceived the ideological distance on the left-right scale between the Social Democratic party and the largest of the conservative parties (see table 6). This perceived polarisation variable turned out to be positively correlated, if only modestly, with people s strength of party attachment in each election studied in each of our six countries. Perceptions of large differences go along with stronger degrees of party identification. Table 6 about here Of course, correlations do not prove causality. In theory, our relationship could as well have been shaped by self-rationalisations, e.g. that people who are strong party identifiers tend to see, and perhaps want to see, large differences between parties. Lacking good panel data we can not disprove this counter-argument, but we have tried to strengthen our test by controlling in multiple regressions for a couple of other variables that we know are, or might be, related to the strength of party identification, for example people s age, but also level of education. These regression controls did not alter the conclusion. There is a significant relationship between people s perceptions of party conflicts and degrees of partisanship even after relevant controls have been applied. Voters who perceive large party differences tend to have, and maybe also tend to develop, stronger party identifications. 12

The conclusion of our endeavours in this section is fairly obvious. Ideological polarisation of the party system goes along with partisanship, not in each country and according to each indicator, and not very strongly, but on average and in the pooled analysis we find a moderate and positive correlation between the two constructs. What is more: we were able to demonstrate on the level of the individual voter that perceptions of ideological polarisation are conducive to partisanship (or caused by it as it may be the case as well). 5. Party Identification and the Vote In this last section of the paper, we finally change our analytical perspective and look at party identification as an independent rather than as the dependent variable. Is it still the overdeterministic predictor of vote choices that it was described to be in the early European analyses? Or is its impact on the vote declining over time? These are the research questions that we are pursuing in the following. As always, there is more than one way of doing it. 2 In the present paper, we will settle for a rather basic and yet straightforward and informative approach. We will determine, for each national election that has been covered by the NES programme in the six West-European countries under study, the proportion of loyalists, that is, identifiers who actually vote for their party. Those who don t are called defectors. Simple as this strategy may sound, it is still a matter of dispute what the appropriate percentage base should be. There are two basic option: all identifiers or only those who actually went to vote. The all identifier-option treats non-voting as a silent exit option; in the all voting-identifiers-option, these silent defectors are disregarded. We will pursue both paths and start with the former. In the six countries under study, during the 1990s, the proportion of party identifiers that actually support their party in an election varies between 70 and 80 percent (see figure 2 and table 7). A significant decline becomes apparent for three countries: Sweden, Norway and Germany. In these countries, proportions of identifiers supporting their party were up at close to 90 percent in the late sixties and seventies. In the remaining three countries Denmark, the Netherlands and the UK developments are less spectacular, mainly because our trend line there starts out from lower levels. 2 Among the alternatives, one option would be a series of block-recursive regressions as many as there are complete data matrices for specific elections; partisanship would always enter in the regression equation after social-structural factors and before more short-term factors like issue preferences and candidate evaluations. 13

Figure 2 and Table 7 about here In the 1990s, on average three in four West European citizens vote for the party they identify with. This is hardly a sign for a deterministic relationship: party identification and the vote are no longer one and the same thing (if this was ever the case). However, things look somewhat differently if we restrict our analysis to only those identifiers who actually turned out and went to vote. Compared to the easy way out into abstention, voting for another party is clearly a harder criterion for identifying defection rates. And as one would expect, loyalists in this perspective are more numerous, defectors less numerous. But still, in the 1990s, one in seven party identifiers who turn out and vote do vote for another party than their own (see table 8). Table 8 about here This proportion as well has risen over the past decades (in five of the six countries that we cover here Denmark being a clear exception). In the 1960s and early 1970s, on average only one in ten identifiers who did turn out failed to support his or her party. So in this perspective as well, partisanship and the vote have become more independent than they were a few decades ago. 6. In Lieu of Conclusion This chapter started out in a theoretical manner. We reviewed very briefly the history of the concept of part identification in European electoral systems, and identified reasons why it was at first rather critically received. Then, we discussed two alternative perspectives on the evolution of partisanship, one sociological, the other political. In the first perspective, which coincides in some way with the functional model of partisanship, a secular decline of the proportion of partisans is expected as the process of social modernisation proceeds. The simple reason is that modern citizens educated, politically skilful and involved as they are do not need anymore the crunch of party identification to meaningfully participate in an election. The other perspective is somewhat less linear and deterministic, and concentrates on political rather than social roots of party identification. It proceeds from the assumption that parties themselves have it in their hands whether they can rely on a stable basis of party identifiers. Adversary politics, to use Samuel Finer s term, is taken to produce partisanship Another option could be a structural equations model in which one could model the interactive effect of time even more elegantly. 14

while the opposite of it a broad political consensus across the major parties is expected to prevent partisanship from developing. According to the first explanatory perspective, proportions of party identifiers should decline, while the second is less explicit about diachronic developments here it all depends on the heat of party conflict and the degree of polarisation, i.e. on the political climate of a polity. What we actually find in our descriptive section on the evolution of partisanship in six Western European countries over the last three or four decades seems to conform quite nicely with the social modernisation perspective. Party identifiers are becoming less numerous over time. However, developments are all but linear in most countries. And when we test for the micro-foundations of modernisation theory, we find that cognitive mobilisation is not what actually drives the development. Something else is going on. This is why we went on and tested an alternative explanation of the evolution of partisanship, which centres around the notion of political polarisation. And indeed, there is some effect of left-right polarisation on partisanship. Both the analysis of aggregate and individual-level data confirms that politics matter. The more polarised a party system is, the more partisans do we find: this is what the pooled analysis (cross-country and over-time) revealed. Moreover, the more polarised a system is perceived to be by individual citizens, the more likely it is that they are partisans themselves. But again, these political correlates of the development of party identification are modest at best, and can not fully explain what is going on in the six West European countries under study. We have changed the perspective of looking at party identification in the final step of our analysis. Treating it as an independent variable, we considered the predictive power of it for vote choices. What we find is that an average of three in four partisans actually support the party they identify with, while one in four abstains from the election or votes for another party. If we restrict this analysis to voting identifiers only, the respective proportion rises to six in seven: one in seven voting identifiers votes for another party than the one he or she identifies with. In both perspectives, party identification has lost some of it s predictive power for vote choices in Western European settings over the past decades. Where does all of this lead us to? We believe there are a few insights that we have added to the body of knowledge in this domain. One is that there is indeed no linear decline in partisanship. While party identification is on the downturn in all the countries investigated, the patterns we identify in this evolution are very country-specific. Secondly: sociological explanations inspired by modernisation theory do not lead very far. Thirdly: political 15

polarisation is one factor in the evolution of partisanship, but certainly no the only one and perhaps not even a very powerful one. And fourth and finally: partisanship and the vote have become more independent over the past decades in Western Europe, which makes party identification a more powerful analytical instrument than it was found to be in the early analysis. References Barton, Terry & Herbert Döring (1986), «Weaking partisanship and the higher educated in Britain». European Journal of Political Research, 14:521-542. Campbell, Angus, Philip Converse, Warren Miller & Donald Stokes (1960), The American Voter. New York: John Wiley. Converse, Philip (1964), «The Nature of Belief System in Mass Publics». I: David Apter, red., Ideology and Discontent. New York: The Free Press. Converse, Philip (1969), «Of time and Partisan Stability». Comparative Political Studies, 2:139-171. Converse, Philip (1975), «Public Opinion and Voting Behavior». I: F. Greenstein & N. Polsby, red., Handbook of Political Science. Reading, Mass: Addison-Wesley. Converse, Philip (1976), The Dynamics of Party Support. Cohort-Analyzing Party Identification. Beverly Hills: Sage Publications. Dalton, Russel (1984), «Cognitive Mobilization and Partisan Dealignment in Advanced Industrial Democracies». The Journal of Politics, 46:264-284. Dalton, Russel (2000), «The Decline of Party Identifications». I: Russell Dalton & Martin Wattenberg, red., Parties without partisans: Political Change in Advanced Industrial Societies. Oxford: Oxford University Press. 16

Reiter, Howard (1989), «Party decline in the West. A sceptic's view». Journal of Theoretical Politics, 1:325-348. Schmitt, Hermann (2002), Partisanship in Western Europe and the US: Causes and Consequences. Paper prepared for presentation at the Annual Conference of the APSA in Boston, MA, August 27 to September 1. Schmitt, Hermann & Sören Holmberg (1995), «Political Parties in Decline?». I: Hans-Dieter Klingemann & Dieter Fuchs, red., Citizens and the State. Oxford: Oxford University Press. 17

Table 1. A secular development of non-partisanship? Pooled country files. Log-odds. Britain Denmark Germany Holland Norway Sweden Reference election 1964 1973 1972 1971 1965 1968 1 0,1 * 0,1 * 0,3 0 * 0,5 0 * 2 0,3-0,2 * 0,6 0,1 * 0,4 0,2 3 0,5-0,4 0,8-0,2 0,1 * 0,2 4 0,8 0,6 0,7 0,2 0,1 * 0,3 5 1-0,1 * 0,2 0,1 * 0,2 * 0,2 6 1,1 0 * 1,1 0,4 0,5 0,5 7 1 0,8 0,7 0,6 8 1,2 0,8 0,7 9 1,1 1,3 0,8 10 0,9 pseudo-r2,045,059,039,009,049,026 Asterisk marks entries that are not significant at the 1 % level. Coding for partisanship: Partisan 0, independent 1. Table 2. Time and cognitive mobilisation (skills and involvement component) Pearson s r. education political interest Great Britain,04 # -,08 Denmark,22 -,05 Germany,26,07 The Netherlands,20 -,03 Norway,19,04 Sweden,20 -,03 # check coding! 18

Cognitive mobilisation indicators and strength of party identification Table 3. Cognitive mobilisation indicators and strength of party identification. Raw regression coefficients. Great Britain 1974 1979 1997 2001 Political interest,25,26,23,23 Education -,21 -,15 -,06 -,09 Age,09,12,12,13 Denmark 1979 1984 1990 1994 1998 Political interest,38,19,17,21,33 Education -,15 -,10 -,08 -,09 -,16 Age,21,07,22,18,07 Germany 1972 1976 1983 1987 1990 1994 1998 Political interest,35,36,28,16,15,15,31 Education -,01 -,02 -,01 -,03 -,07,00,03 Age,14,00,06,03,00,06 -,03 The Netherlands 1971 1977 1981 1986 1994 1998 Political interest,33,11,32,36,32,42 Education -,04 -,04 -,05 -,06,03 -,06 Age,19,09,12,17,23,18 Norway 1965 1969 1973 1977 1981 1985 1989 1993 1997 2001 Political interest,35,30,38,33,34,32,38,28,36,33 Education -,13 -,20 -,16 -,16 -,12 -,16 -,15 -,03 -,01 -,08 Age,15,16,18,13,18,15,17,18,09,10 Sweden 1968 1970 1973 1976 1979 1982 1985 1988 1991 1994 1998 Political interest,22,21,22,25,23,26,25,23,22,22,24 Education -,27 -,22 -,23 -,23 -,17 -,14 -,18 -,15,07 -,15 -,13 Age,14,15,22,11,14,15,16,17,18,17,17 Bold types: Significant at the 1 % - level. Coding: Party identification 1. Independent 2. weak identification 3 strong identification. In 1997, 1. Independent 2. Party Identification. Highest Education, 1 primary, 2 high school, 3 university Political interest, 1. no interest at all, little interest 2. fairly interested, 3. very interested Age 1. 30 and below, 2. 31-50, 3. 51 and above +. 19

Partisanship and education Table 4a. Percent independents at different levels of education in Great Britain. 1964 1970 1974 1979 1983 1987 1992 1997 2001 primary 16 19 25 29 37 38 37 43 34 Britain secondary 19 25 28 34 41 40 39 40 42 high school 32 31 39 44 36 33 41 37 40 All Pearson r 0,06 0,08 0,06 0,08 0-0,01 0,02-0,04 0,06 Below 30 Pearson r 0,02 0,01-0,01 0,04-0,07-0,06-0,06-0,04 0,07 Table 4b. Percent independents at different levels of education in Denmark. 1973 1975 1977 1979 1984 1990 1994 primary 44 38 70 43 45 Denmark secondary n.a. n.a. 49 48 78 53 54 high school 54 48 75 53 60 All Pearson r 0,07 0,09 0,04 0,08 0,12 Below 30 Pearson r n.a. -0,09-0,06-0,05-0,05 Table 4c. Percent independents at different levels of education in Germany. 1972 1976 1980 1983 1987 1990 1994 1998 primary 39 52 58 60 56 48 69 65 Germany secondary 36 39 45 57 54 42 66 60 high school 34 34 43 43 52 39 61 46 All Pearson r -0,04-0,14-0,12-0,09-0,03-0,07-0,06-0,15 Below 30 Pearson r -0,01-0,12-0,17-0,1-0,04-0,02-0,12-0,16 Table 4d. Percent independents at different levels of education in the Netherlands. 1971 1972 1977 1981 1986 1989 1994 1998 primary 67 66 67 58 67 69 71 Netherlands secondary 60 n.a 71 66 61 74 68 75 high school 59 57 64 60 67 61 71 All Pearson r -0,06-0,02 0,02 0,02-0,06 0,01 Below 30 Pearson r -0,13-0,05-0,06 n.a. -0,18-0,08 20

Table 4e. Percent independents at different levels of education in Norway. 1965 1969 1973 1977 1981 1985 1989 1993 1997 Primary 22 32 33 23 24 24 29 37 45 Norway High school 32 42 37 34 31 33 40 46 50 University 33 53 49 36 34 37 46 44 42 All Pearson r 0,10 0,14 0,1 0,11 0,08 0,1 0,12 0,04-0,02 Below 30 Pearson r 0,04 0,08 0,08 0,1-0,01 0,01-0,03-0,12-0,12 Table 4f. Percent independents at different levels of education in Sweden.. 1968 1970 1973 1976 1979 1982 1985 1988 1991 1994 1998 primary 49 45 56 48 36 52 56 63 71 69 76 Sweden secondary 45 49 56 53 55 51 59 60 65 67 74 high school 56 52 61 52 54 54 66 65 67 66 68 All Pearson r 0,12 0,15 0,14 0,14 0,12 0,10 0,16 0,14 0,05 0,12 0,10 Below 30 Pearson r 0,02 0,05 0,03 0,03 0,02 0,01 0,08 0,03-0,02-0,02-0,07 21

Table 5. Left-Right Polarisation and Party Identification. Pearson s r (and n in parenthesis). MRG-based left-right distance two big parties MRG-based left-right distance two polar parties survey based left-right distance two big parties survey based left-right distance two polar parties all identifier strong identifier +.13 (61) +.21 (59) +.18 (61) +.42 (59)** +.24 (46) +.29 (44) -.27 (46) +.32 (44)* Great Britain all identifier strong identifier -.45 (10) -.49 (10) -.59 (10) -.60 (10) +.38 (7) +.41 (7) +.38 (7) +.41 (7) Denmark all identifier strong identifier +.07 (10) -.12 ( 9) +.03 (10) -.09 ( 9) +.49 (7) +.88 (6)* +.60 (7) +.67 (6) Germany all identifier strong identifier -.05 (11) -.42 ( 8) +.05 (11) -.44 ( 8) +.76 (8)* +.73 (8)* -.04 (8) -.24 (8) The Netherlands all identifier strong identifier -.18 (9) +.30 (9) +.30 (9) +.50 (9) +.53 (7) +.43 (7) +.74 (7) +.67 (7) Norway all identifier strong identifier -.51 (10) -.63 ( 9) -.07 (10) -.21 ( 9) -.44 (7) -.33 (6) -.20 (7) -.06 (6) Sweden all identifier strong identifier +.11 (11) +.69 (14)** -.52 (11) +.24 (14) +.20 (10) +.31 (10) +.33 (10) +.41 (10) Source: European Voters Data File; the MRG database (Budge, Klingemann et al. 2001); and the data collected and reported in the appendix tables of Schmitt 2002. Note: *p=.05 **p=.01 22

Table 6. Perception of Left-Right Polarisation Between the Two Major Parties and Party Identification. two major parties election perceived leftright distance by strength PID (Pearson s r) n of cases Great Britain Labour- Conservatives 2001 +.18 2221 Denmark S-Venstre 1994 +.10 1801 Germany SPD-CDU/CSU 1980 1983 1987 1990 1998 The Netherlands PvdA-CDA 1981 1982 1986 1989 1994 1998 Norway Labour- Conservatives 1973 1977 1981 1985 1989 1993 Sweden S-M 1979 1982 1985 1988 1991 1994 1998 +.13 +.15 +.20 +.06 +.12 +.12 +.05 +.06 +.13 +.06 +.09 +.08 +.08 +.13 +.06 +.15 +.09 +.25 +.25 +.18 +.23 +.23 +.19 +.20 1055 920 1215 765 1546 1468 674 1239 1429 516 1596 892 1344 1350 1923 1982 1870 2416 2592 2535 2395 2326 2154 1959 Source: European Voters Data File. 23

Table 7 Proportion of Loyalists in All Identifiers (i.e. including those who did not vote or can not tell which party; figures are %) Denmark Germany Netherlands Norway Sweden Great Britain 1960 1 2 3 4 81 1965 84 6 77 7 8 87 9 85 1970 86 74 1 68 2 88 3 78 85 85 4 75 1975 75 6 78 90 87 7 78 72 87 8 9 81 89 75 1980 88 1? 85 2? 86 3 87 74 4 1985 86 85 6? 7 86 76 8 82 9 72 83 1990 81 72 1 79 2 77 3 76 4 82 80 75 82 1995 6 7 81 70 8 81 77? 77 9 2000 1 68 Source: National Election Studies of these countries as available in the European Voters Data File. Note: Empty spaces indicate that the respective information is not available. A? indicates that the recoding of the data does not allow to produce the figure in question. 24

Table 8 Proportion of Loyalists in Voting Identifiers (i.e. excluding those who did not vote or can not tell which party; figures are %) Denmark Germany Netherlands Norway Sweden Great Britain 1960 1 2 3 4 91 1965 89 6 90 7 8 93 9 91 1970 95 90 1 91 2 92 3 81 92 93 4 86 1975 87 6 85 96 91 7 87 93 96 8 9 88 93 88 1980 94 1 82 94 2 93 92 3 93 87 4 1985 96 90 6 86 7 92 86 8 90 9 86 92 1990 90 86 1 86 2 87 3 89 4 88 88 82 89 1995 6 7 89 87 8 85 84 83 85 9 2000 1 87 Source: National Election Studies of these countries as available in the European Voters Data Base. 25

Figure 1 The Evolution of the Level of Partisanship in Six European Democracies 100 90 80 70 60 50 40 30 20 10 0 1950 1960 1970 1980 1990 2000 Germany Sweden UK Norway Netherlands Denmark 26

Figure 2 Proportion of Loyalists in All Identifiers 100 90 80 70 60 UK all N all S all D all NL all DK all 50 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 Source: National Election Studies. Note that all identifiers include those who do not report their vote and abstainers. 27

Appendix Education Table 3a. Education in Great Britain. Percent. 1964 1970 1974 1979 1983 1987 1992 1997 2001 primary 67 69 67 55 75 62 77 75 53 # Britain secondary 31 28 29 40 15 31 16 14 33 # check coding! high school 2 3 5 6 10 7 7 11 14 1740 1772 2454 1846 3935 3813 3527 4178 2303 Table 3b. Education in Denmark. Percent. 1973 1975 1977 1979 1984 1990 1994 primary 68 63 55 45 45 Denmark secondary n.a. n.a. 23 26 31 35 34 high school 9 12 13 20 21 Table 3c. Education in Germany. Percent. 1972 1976 1980 1983 1987 1990 1994 1998 primary 82 63 69 67 69 58 46 29 Germany secondary 9 23 20 22 19 26 34 39 high school 9 14 11 12 12 16 20 32 2039 2053 979 1622 1934 2046 2007 1993 Table 3d. Education in the Netherlands. Percent. 1971 1972 1977 1981 1986 1989 1994 1998 primary 78 79 81 63 59 47 45 51 Netherlands secondary 14 12 7 19 23 33 35 27 high school 8 10 13 17 19 19 23 22 Table 3e. Education in Norway. Percent. 1965 1969 1973 1977 1981 1985 1989 1993 1997 Primary 43 37 32 36 30 19 24 23 18 Norway High school 47 49 53 51 54 54 57 52 58 University 11 14 15 14 16 17 19 25 24 28

Table 3f. Education in Sweden. Percent. 1968 1970 1973 1976 1979 1982 1985 1988 1991 1994 1998 primary 49 45 56 48 36 52 56 63 71 69 76 Sweden secondary 45 49 56 53 55 51 59 60 65 67 74 high school 56 52 61 52 54 54 66 65 67 66 68 Political interest Table 4a. Political interest in Britain. Percent. 1964 1970 1974 1979 1983 1987 1992 1997 2001 * great deal 17 13 31 28 Britain some 45 48 36 33 not much 29 32 26 28 not at all 8 8 7 10 means 2,3 2,3 2,1 2,2 * In 2001, the alternative great deal was replaced with quite and quite a lot. These two are collapsed into great deal in table 4a. Table 4b. Political interest in Denmark. Percent. 1971 1973 1975 1977 1979 1981 1984 1987 1990 1994 1998 Very 12 19 21 21 20 20 18 15 19 20 24 Denmark Fairly 38 54 42 57 42 43 43 42 43 44 49 Little 50 24 32 18 32 31 31 35 32 32 24 Not at all 2 5 5 7 7 8 8 6 5 3 means 2,4 2,1 2,2 2,1 2,3 2,3 2,3 2,4 2,3 2,2 2,1 Table 4c. Political interest in Germany. Percent. 1972 1976 1980 1983 1987 1990 1994 1998 very 56 56 10 7 16 7 10 Germany strong 33 32 31 28 43 18 38 fairly 11 12 19 22 10 50 46 not 0 0 40 42 31 25 6 means 1,5 1,6 2,9 3,0 2,6 2,9 2,5 Different question in 1972 and 1976. 29