Jump to ContentJump to Main Navigation
Political Representation and Legitimacy in the European Union$

Hermann Schmitt and Jacques Thomassen

Print publication date: 1999

Print ISBN-13: 9780198296614

Published to Oxford Scholarship Online: November 2003

DOI: 10.1093/0198296614.001.0001

Show Summary Details
Page of

PRINTED FROM OXFORD SCHOLARSHIP ONLINE (www.oxfordscholarship.com). (c) Copyright Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a monograph in OSO for personal use (for details see www.oxfordscholarship.com/page/privacy-policy). Subscriber: null; date: 16 January 2019

Policy Preferences and Party Choice

Policy Preferences and Party Choice

Chapter:
(p.161) 8 Policy Preferences and Party Choice
Source:
Political Representation and Legitimacy in the European Union
Author(s):

Cees van der Eijk

Mark N. Franklin (Contributor Webpage)

Wouter van der Brug

Publisher:
Oxford University Press
DOI:10.1093/0198296614.003.0008

Abstract and Keywords

This chapter is the third of six on the question of political representation in the EU, and the third of four that put the five requirements of the Responsible Party Model (outlined in Ch. 6) to an empirical test. The questions investigated here are whether voters have preferences on the issues at stake, and whether those preferences motivate their electoral choices; the analysis is carried out using data from the European Election Study 1994. The data are found largely to support the condition that voters hold policy preferences. They also allow two contrasts to be made while studying determinants of party choice—nation‐specific issues can be contrasted with common issues, and position issues (as exemplified by the common European currency, border control, and unemployment) with valence issues; it can then be seen which of these types of issues are most important in explaining party preferences. In addition, degrees of policy voting can be contrasted in each of the political systems. The results are hardly supportive of the Responsible Party Model in that hardly any motivational basis is found in terms of specific issues and policies for voters’ party preferences; however, the model becomes more persuasive when the focus is not only on specific policy and issue concerns but more generally on substantive political concerns and voter orientations relating to the left–right ideological position of parties.

Keywords:   border control, common European currency, EU, left–right ideology, party choice, party preferences, policy preferences, political concerns, political representation, position issues, Responsible Party Model, unemployment, valence issues

Introduction

In Chapter 6 of this volume Schmitt and Thomassen list five requirements for political representation as conceived by the Responsible Party Model. The two preceding chapters assessed to what extent three of these requirements were met in the 1994 elections for the European Parliament: whether parties offer voters a choice between different programmes, whether they are sufficiently coherent to be able to carry out these programmes (Chapter 6), and whether voters know what parties stand for (Chapter 7). Here we focus on the two remaining conditions for effective representation: whether voters have preferences on the issues at stake, and whether those preferences motivate their electoral choices.

Policy voting is a necessary but not sufficient condition for political representation. When voters elect a party or candidate on the basis of their policy preferences while they are misinformed about the stands of the party or candidate in question, policy voting does not produce meaningful representation, but voters may have been motivated by policy nonetheless. In that case voters' preferences will not ‘fit’ the positions of the parties they voted for (Converse 1975). The opposite (p.162) situation is also conceivable. Voters may not care enough about certain issues to let these issues motivate their party choice, even though positions of voters and the parties they voted for correspond well. In such a case the respective issues are of little consequence for our evaluation of the extent to which preferences of voters are represented in the political domain. An example of this can be found in the 1989 European elections (van der Eijk and Franklin 1991 and 1996). Thus, for an overall evaluation of how well voters' preferences are represented in the European Parliament, one must also consider the evidence presented in the previous two chapters, which is a matter that will be discussed in the next chapter.

The previous chapters introduced various issue areas about which questions were asked in the surveys. First of all, the surveys include a number of valence issues where the focus of political conflict is not the goal to be reached, but where voters and parties may differ in terms of how important they find each of these generally desired objectives, or in terms of how (not whether) they are to be reached. According to the theory of issue‐ownership, voters evaluate parties largely on the basis of their perceptions of the priorities parties will give to issues or the competence of parties in certain policy areas (Budge and Farlie 1976). To the extent that this model correctly predicts electoral decisions, differences in the importance voters give to issues will be reflected in their party choice. Such differences were measured by asking respondents which of these valence issues they found most important, and which were the second and third most important issues.1

In addition to the valence issues, three position issue questions were asked where voters had to indicate both their own positions and those of the political parties in their country on a continuum defined by two poles reflecting different and opposite goals to be attained. These three position issues refer to policy options on matters of currencies, borders, and employment.2

Finally, respondents were asked to indicate their own position, as well as that of the various parties in their country, on the left–right dimension. Although this dimension does not refer to any single specific policy concern, the left–right continuum is known to be strongly related to the most important issues in most western democracies, reason for some to refer to it as a ‘super issue’ (for example, Inglehart 1984). Various studies have demonstrated that left‐right positions reflect to a large extent the positions of voters and parties on several concrete issues (p.163) (for example, Fuchs and Klingemann 1990; van der Brug 1996 and 1998). Moreover, it has been argued that the Responsible Party Model requires the behaviour of parties and voters alike to be constrained by the same uni‐dimensional ideology (Thomassen 1994: 254).

The information available in the European Election Study 1994 permits us to make two contrasts while studying determinants of party choice. We can contrast nation‐specific with common issues, and valence with position issues, seeing which of these types of issues are most important in explaining party preferences. In addition we can contrast degrees of policy voting in each of the political systems. First, however, we need to know whether voters in the various European countries have preferences on the various policy dimensions included in this study.

Do Voters Hold Policy Preferences?

The extent to which voters have preferences for different policies is indicated by the percentage that is or is not able to express such preferences when asked about them in surveys. Table 8.1 presents the percentage of respondents in each country (and over the EU as a whole) unable to express such a preference for the various issues in the European Election Study 1994.3 These percentages should be interpreted as lower‐bound estimates of the proportion of electorates in each country that have preferences.4

As we can see from Table 8.1, the number of respondents who do not provide a substantive answer to the different questions varies considerably, between countries as well as—for our present discussion more importantly—between the different kinds of questions. Naming which of the 11 valence issues are the three most important ones is the simplest task of all: in no country do we find more than five per cent of the respondents unable to do this; averaged across the various systems it amounts to no more than slightly over two per cent. When we turn to the three position issues and to left–right, we find rather more failures to respond, with as many as 15 to 20 per cent ‘don't knows’ on one or the other of these policy dimensions in some countries. Averaged over (p.164) the political systems of the Union, however, the group without preferences is smaller than ten per cent for each of these scales.5

Table 8.1. Percentage of Respondents Without Preferences for Various Issue Areas

European position issues

Valence issues

Currency

Employment

Borders

Left–right

Three most important

Flanders

8

13

8

11

3

Wallonia

15

15

18

12

5

Denmark

3

8

2

4

2

West Germany

6

12

4

7

2

Greece

10

14

4

5

2

Italy

9

9

9

11

0

Spain

14

10

12

6

3

France

4

7

1

5

1

Ireland

17

9

19

15

3

Northern Ireland

2

3

5

13

2

Luxembourg

6

14

1

10

4

Netherlands

5

6

3

6

2

Portugal

15

7

18

12

4

Great Britain

2

9

4

6

1

East Germany

7

9

5

10

1

EU

8

10

7

9

2

Source: European Election Study 1994.

Because of differences in the ways polling institutes in each of the countries conduct their interviews which affect the proportions non‐response (so‐called house effects), we must be very cautious with our interpretation of differences between countries. Nevertheless, differences in the proportions of non‐response between countries seem to indicate differences in the country‐specific significance of these policy concerns. For instance, left–right is not as important in Ireland as in Denmark or Greece.

Since large majorities of European voters have opinions on the issues they were questioned about, we may conclude that the condition that voters hold policy preferences is largely met. Although minorities (p.165) of voters seem not to have such preferences, more than 90% (on average) of the electorates in these 15 countries volunteer preferences for these policy concerns. Having assessed this, we now turn to the second condition of the Responsible Party Model to be addressed in this chapter: whether policy preferences motivate party choice.

Party Choice and How It Is Measured

Any attempt to evaluate the independent contributions made by policy preferences to the party choice of voters is always plagued by two problems: (1) policy positions and party preferences may both be the consequence of deep‐seated values generally encapsulated in ideological orientations such as voters' left–right position, or of affinities with social or religious groups, so that we run the risk of attributing effects to issues on the basis of spurious correlations; and (2) policy preferences of voters may be the consequence (rather than the cause) of party choice, where voters are educated into accepting the policies of the party they prefer (persuasion) or where they tend to overestimate the proximity between their own policy preferences and those of political parties that are preferred on other grounds (assimilation).

In this chapter we make a number of assumptions about these matters that allow us to arrive at tentative answers to our research questions. Different assumptions might well yield different answers. The assumptions we make, however, are not unique to this research. The same assumptions have been made by many researchers, including ourselves in past work. They are that (1) ideological dispositions and other group and religious norms derive from socializing processes (both in childhood and in later life); (2) that voters' preferences on specific issues (and on the relative importance of those issues) derive in part from ideological predispositions and socializing processes; and (3) that in explaining party preferences at any particular moment in time causal effects run from ideological and issue preference to party preferences and not the other way around. Thus, when looking for the effects of policies we only have to consider their additional effects over and above those of factors such as religion and class.6

In addition to these problems there is another matter which has serious repercussions for the research design, namely that party choice (the (p.166) dependent variable) is both a nominal and an ipsative variable, as voters in (almost) all political systems are only allowed to choose one party. From a substantive point of view, this causes two major problems. First of all, any explanatory statements about party choice imply an intra‐individual comparison of parties which, in multi‐party systems, cannot be observed when only analysing actual party choice. A theory that states, for example, that left–right proximity is an important determinant of party choice implies that voters evaluate all parties on the basis of such value orientations. When attributing actual party choice to this factor one should ideally be able to demonstrate empirically that (ceteris paribus) the second‐most proximal party is also the respondent's second choice, that the party most distant from the voter is also the least attractive to vote for. Such observations are impossible when analysing actual voting behaviour, since that relates generally only to a voter's first choice.7 Secondly, when analysing party choice we run into a problem that arises from the (sometimes exceedingly) small numbers of respondents who voted for small parties. If we measure the respondent's party choice in a specific election, we have no information about the electoral attractiveness of other parties. The specific context of our research into European elections causes an additional problem because party choice cannot be compared between countries that each have their own particular party system.

These problems—deriving from the ipsative character of the dependent variable, which does not reflect in any kind of detail the various degrees of electoral attractiveness of different parties, and the non‐comparable character of party choice—are dealt with by employing as the dependent variable the electoral attractiveness of a political party. This variable, which we will refer to as party preference, is a characteristic that can be measured for all parties, irrespective of their particular traits and irrespective of the political system in which they are located. Respondents to the 1994 European Election Study were asked, ‘Please tell me for each of the following how probable it is that you will ever vote for this party?’, after which they were presented with the names of practically all parties in their political system.8 These questions yielded a set of variables, one for each party contesting the European elections, measuring on a scale of 1 to 10 the likelihood that the respondent would vote for the party concerned.

The open‐ended reference to ‘ever’ in the question is intended to function only as a projective device encouraging respondents to express (p.167) the extent of their current preference for each party. In this way it avoids the restrictions of the more usually employed question about actual voting behaviour, which does not allow respondents to report the extent of their electoral preferences for other parties as well.9 Moreover, the 10‐point scale used in our question allows a continuous expression of preferences, rather than a merely dichotomous one. Because respondents were asked to evaluate all parties, even those they had not supported in the European elections, we have a score from each respondent for virtually every party that participated in the elections in that respondent's country.

In the analyses to be conducted in this chapter, this measure of party preference is the phenomenon to be explained, the dependent variable. In order to achieve comparability between political systems we need to use an innovative research strategy which will be described in detail below. At this point it is important to note that the use of party preference as a dependent variable avoids the problems referred to earlier. The responses to ‘probability to vote’ questions provide a much firmer empirical basis for explanatory propositions about determinants of party preference than does actual party choice.

A few objections to this approach might be raised. First of all, the ‘probability to vote’ question employed in the 1994 European election study asked about elections in general, even though it was posed in the context of a European election study. This is however a minor problem. Assessing a general vote intention at the time of a European election rather than at the time of a national election no doubt reduces election‐specific national forces that might colour the vote in particular countries at particular times; but this might be an advantage when our objective is to discover what determines vote choice in general terms, since European vote choice is in many cases quite similar to national vote choice.10 A more serious objection could be that we substitute the explanation of party preference (which may be seen as electoral attractiveness of parties) for what was originally to be explained, actual party choice. This objection would, however, only be valid if the party that respondents actually voted for were often different from the one to which they gave the highest score. In fact our whole approach rests on the assumption that respondents do vote for the party to which they give the highest probability‐to‐vote score. Only on this basis can we confidently interpret the second‐highest score as a voter's second choice, and so on. Table 8.2 demonstrates that this assumption is well‐founded (p.168) by showing the extent to which party choice can be deduced from the probability‐to‐vote variables. When asked which party they would vote for, far more than 90 per cent of the voters name the party they awarded the highest preference score. This suggests that party preference scores are very accurate reflections of actual vote intentions. As a consequence, by analysing the former we can arrive at valid conclusions about the latter.11

Table 8.2. Percentage of Respondents Supporting Party With Highest Vote Probability

Percentage support

Valid N*

Number of parties

Flanders

94

409

6

Wallonia

95

245

5

Denmark

98

799

8

West Germany

96

735

6

Greece

98

586

5

Italy

95

601

7

Spain

97

599

6

France

93

597

8

Ireland

93

597

8

Northern Ireland

93

305

8

Luxembourg

96

322

6

Netherlands

96

848

9

Portugal

96

607

5

Great Britain

95

923

6

East Germany

99

435

4**

Source: European Election Study 1994.

(*) All respondents who indicated that they intended to vote for one of the parties for which the probability‐of‐future‐vote (POV) question was asked, and who indicated their POV for at least one party.

(**) In East Germany the party lists included in the item measuring actual party choice do not correspond in two cases with the parties included in the battery of POV items (B‐90/Die Grünen, and Linke Liste/PDS). They are therefore not included in this table.

(p.169) Assessing the Impact of Policy Preferences on Party Choice

As indicated above, the dependent variable for our analyses is the electoral preference of voters for each of the political parties in their country, which is expressed as a score ranging from 1 (will never vote for that party) to 10 (will at some time certainly vote for it). These preferences would normally be represented in a data matrix as different variables, one for each party, which would not readily lend themselves to simultaneous analysis. Yet it is not sufficient to analyse these variables one by one. It would not do conceptually, because what we are looking for are determinants of party preference in general rather than a specific model for one party or another. It would also not do for another reason: we mentioned earlier that one of the major advantages of using these non‐exclusive party preferences is that for each voter they provide variation between parties in terms of their electoral attractiveness, which is needed for valid explanatory statements about the nature of party choice. Analysing these preferences one by one, however, would obscure this (individual‐level) inter‐party variation, as such a design focuses exclusively on the variation between individuals. An adequate analysis of these scores requires a research design in which inter‐party and inter‐individual variance is accounted for simultaneously. This can be realized by rearranging the original data into a so‐called stacked form (Brown and Halaby 1982; Stimson 1985): viewing each preference score given by a voter as a separate case to be explained. In this way, each respondent is represented by a number of cases in the stacked data set, as many as the number of parties for which he or she gave a preference score (see van der Eijk, Franklin, et al. 1996: 345). The stacked data set can be analysed in the same way as any normal rectangular data matrix. The dependent variable is the preference score; appropriate identifiers allow characteristics of individual respondents and of parties to be added as explanatory or control variables. The independent variables have to be defined in an appropriate manner before they can be included in the analysis (see below), but once this has been done, the stacked data matrix allows us to examine the dependent variable using familiar and straightforward methods of analysis, such as regression.

(p.170) This strategy can also be extended to encompass different political systems: we just combine the stacked data sets from various countries into one single data matrix, in which one single variable represents party preference for all respondents over all parties in all countries.12 But before any of this can be done, we first have to decide how to represent the independent variables in the regressions.

The Independent Variables

The policy variables available to us for inclusion in our analysis were already described in the introduction to this chapter. In order to employ these questions in an analysis that assesses their contribution to party choice, we derived for each of the three position issues a variable measuring the distance between the position preferred by each respondent and the position that respondent ascribed to each party. Thus for each combination of respondent and party there is a measure of distance. These can be added to the stacked data set described above, and the contribution made by each of these position issues to party preference can be assessed by calculating the impact of ‘distance from party’—a rather general concept. The overall importance of each of the three position issues can thus be assessed by means of an effect that has been calculated over all the parties in a country or, indeed, over all the parties in all the countries.

For the valence issues, a rather different procedure had to be adopted in order to be able to determine their contribution to general party preferences. From each of them we constructed two independent variables, representing two different views of the way these issues work. First of all we considered the logic of ‘issue‐ownership’. If both voters and parties are differentiated by the extent to which they give high priority to one issue or another, and voters were to react to this in their behaviour, then we would expect that voters would prefer those parties that share their own priorities among the valence issues. Even though no explicit information is available about voters' perceptions of parties' priorities, to the extent that voters are aware of which parties ‘own’ such issues, we would find that giving high priority to each specific valence issue would contribute to high preference for those parties. Therefore, we constructed 11 variables, one for each of the valence issues, to capture any such effects; the method of construction was the (p.171) same as for several other variables and is described below.13 The second way in which we constructed variables relating to the battery of valence issues took advantage of the presence of the answers to the question ‘which party is best at’ each of the three most important issues. Recall that the stacked data set has a case for each party for each respondent. For each of these records (i.e. party by respondent combinations) we injected a variable that indicated whether that particular party was named or not as best able to handle each of the three problems deemed ‘most important’ by that respondent.

In order to assess the relative contribution of these issues to party choice, we also had to take account of other influences. In particular, variables were considered that are assumed to partly determine issue preferences of voters: their religious and group identities and their left–right position.14

In the case of left–right position, the same procedure was followed as for the position issues. Respondents were asked about their own position on this scale and about the position they ascribed to each of the parties in their political systems. By subtraction we derived a distance of each respondent to each party in his or her political system that could be included in the stacked data set.

Other independent variables were handled somewhat differently. For class, religious denomination, and the valence issues introduced above, no measure of distance was obtainable. Instead, for each of the parties separately (and hence for each of the countries separately) an artificial measure of ‘closeness’ was derived from the predictions of party preference for each of these variables in turn.15 The same procedure was adopted for social class, using subjective social class and income as predictors, and for the valence issues, using whether or not the issue in question had been named as one of the three most important.16

An independent variable that has been added to our analyses without either referring to an issue in the proper sense of the word, or being a necessary control variable is ‘EU approval’. This variable is attitudinal in character and captures respondents' positive feelings (or the lack thereof) toward the process of European integration.17 The reason for including it in our analyses is twofold. On the one hand, the elections we investigate are European ones, providing room for the notion that attitudes and orientations toward the integration process itself may conceivably influence voters' electoral preferences for political parties measured in that context. Secondly, it allows us to replicate a similar (p.172) analysis of party preferences in the context of the 1989 European elections that has been reported elsewhere (van der Eijk, Franklin, and Oppenhuis 1996; Oppenhuis 1995).

As indicated above, certain other control variables can be added to the stacked data without any special transformation being applied. In the analyses conducted in this chapter, only a single control variable was included: party size. Although this variable plays no part in the causal processes underlying issue voting, other research has found it to be a strong predictor of party preferences, representing the strategic considerations employed by voters who, other things being equal, prefer to support a party that has a good chance of achieving its policy goals (van der Eijk, Franklin, et al. 1996, Chapter 20; Tillie 1995). Consequently, in order to arrive at a model specified as properly as our data allow, we included it in our analyses.

Missing Data and Weighting

In the analyses reported below a number of records in the stacked data set were excluded because of missing values for the dependent variable. As a consequence, not all respondents have equal weight in each analysis. If, for example, the ‘probability to vote’ question was asked for six parties, and a particular respondent failed to answer this question for two of these parties, (s)he is represented in the stacked data set by four records, whereas another respondent who gave valid answers for all six parties is represented by six records. Other missing data were handled by means of pairwise deletion, except for the y‐hat variables which were estimated from equations in which missing values for the predictor variables were replaced by their mean values.

When analysing the data for the various political systems separately we weighted them in such a way as to reflect the results of the European elections themselves (cf. van der Eijk, Franklin, et al. 1996, Appendix B). In the Euro‐wide analysis, these weights were multiplied by country‐specific constants so as to ensure, first, that each political system contributed equally to the analyses by being represented by the same number of effective cases, and, second, that the effective number of cases in the entire (stacked) data set was reduced to 15,000—approximately the real number of interviews obtained.

(p.173) Findings

We start by investigating the extent of variance in party preference (country‐by‐country and over the EU as a whole) that can be explained by the different (groups of) independent variables, taken one at a time. Because of their large number, Table 8.3 presents the joint effect of the set of 11 valence issues combined. As it turns out, voters' policy preferences in terms of the priorities they assign to various political problems have hardly any explanatory power over their party preferences. Taken all together, the highest R2 for the entire set is 0.08 in Flanders, and averaged across the 15 political systems only four per cent of variance is explained. At this low level of explanatory power it does not make much sense to consider the separate effects of the various issues, as these are minimal. This finding is somewhat surprising, because a very similar analysis of the data from the 1989 European election study yielded 12 per cent explained variance when using a set of 12 very similar valence issues (van der Eijk, Franklin, et al. 1996: 350).18 It is not very likely that these differences are the consequence of real changes in the calculus of voters. Much more likely is that the difference arises from a change in question format, such that the effects of the 11 valence issues presented in Table 8.3 underestimate the effect of voters' issue concerns on their party choice by some eight per cent.19

On average (see the bottom row of Table 8.3) the strongest explanation of party choice is provided by the questions about party competence—the parties that, according to respondents, are ‘best at’ the three valence issues they had indicated as the most important ones. However, this relationship is not entirely convincing as we fear that the causal relation behind it runs at least partly the other way around: from party preference to party ‘best at’.20 Ideological distance is more important than any one type of issue. The set of valence issues and each of the three position issues are all of roughly equal importance, each explaining approximately four per cent of the variance in party preferences. The strength of the effects of (distance to party in terms of) left–right, class, religion, and EU approval is almost identical to what a similar analysis of the data of the 1989 European election study yielded (van der Eijk, Franklin, et al. 1996: 350). Moreover, the relatively weak effects of class and religion are entirely in line with analyses of (p.174)

Table 8.3. Explained Variance in Party Preference by Various Independent Variables

Left–right distance

Currency distance

Employment distance

Borders distance

All valence issues

Class*

Religion

EU approval

Party competence

Party size

Flanders

.09

.03

.02

.04

.08

.02

.05

.02

.28

.03

Wallonia

.23

.07

.01

.02

.05

.08

.03

.01

.25

.07

Denmark

.21

.04

.05

.03

.06

.05

.02

.07

.30

.10

West Germany

.26

.02

.02

.08

.05

.01

.02

.01

.35

.18

East Germany

.19

.01

.03

.03

.02

.01

.03

.00

.17

.08

Greece

.33

.08

.09

.04

.03

.02

.02

.03

.35

.07

Italy

.28

.11

.06

.04

.05

.01

.04

.00

.30

.04

Spain

.20

.06

.07

.05

.03

.02

.04

.01

.44

.16

France

.25

.05

.04

.05

.07

.02

.04

.02

.18

.02

Ireland

.15

.02

.00

.03

.02

.02

.02

.01

.20

.06

Northern Ireland

.19

.04

.02

.13

.07

.04

.33

.02

.19

.02

Luxembourg

.19

.09

.12

.05

.04

.01

.03

.02

.28

.22

The Netherlands

.20

.04

.02

.03

.04

.02

.08

.00

.24

.21

Portugal

.26

.04

.02

.03

.03

.02

.03

.01

.31

.18

Great Britain

.14

.03

.05

.02

.06

.04

.02

.01

.41

.19

EU

.21

.04

.04

.04

.04

.03

.06

.02

.27

.10

Source: European Election Study 1994.

(*) The effect of social class on party choice is estimated with two predictors (y‐hats) of subjective social class and family income.

(p.175) the socio‐structural bases of party choice as reported in Franklin, Mackie, and Valen (1992).

Our findings here can thus be regarded as an independent confirmation of established insights, a fact that greatly increases our confidence in the results reported here. Incidentally, the sizeable differences between their respective R2s in Table 8.3 vindicate the decision to distinguish between Flanders and Wallonia, between the western and the eastern parts of Germany, and between Northern Ireland and Britain.

The results of Table 8.3 do not give a decisive perspective on the strength of issues in the explanation of party preferences because issues are not independent of other predictors, especially left–right. Table 8.4 takes a more causal view of the voter's decision‐making process by regarding issue preferences as being partly determined by variables earlier in the causal sequence, namely religious and class identity, left–right position, and party size.21 The table lists the explained variance for seven different models, starting with the simplest, which only incorporates religion and class, and adding other independent variables to this model in a series of steps. Models 2, 3, and 4 add first of all left–right ideology, then the three position issues, and next the set of 11 valence issue variables. This entire set of variables relates clearly to voters' policy considerations, and has a considerable impact on variance explained, raising it (EU‐wide) from 0.08 in Model 1 to 0.28 in Model 4. By far the largest part of this increase is effectuated by the ideological (left–right) distance between voters and parties, while the more specific issues combined add about three per cent to variance explained.22 As was the case in 1989 (and as was foreshadowed by the results of Table 8.3), adding EU approval (Model 5) does not yield any appreciable increase in R2. The next addition (Model 6) concerns the variable that reflects voters' strategic considerations, party size, which EU‐wide increases the share of explained variance considerably, from 0.28 to 0.36. Model 6 is the one that most closely resembles a similar model presented elsewhere for explaining party preferences in the 1989 European elections (van der Eijk, Franklin, and Oppenhuis 1996: 357). Yet its power is considerably less than in 1989, when the R2 reached almost 0.43. The cause for this difference is mainly the difference in independent variables that could be used. In 1989 a variable relating to respondents' approval of the government in power (reflecting either positively or negatively on the preferences of the parties holding government positions) added significantly to the explanation, while such (p.176) information was not available in the 1994 data. Furthermore, the change in question format for the set of valence issues (see above) robbed these of most of their explanatory power. Taking these considerations into account, we find our results for 1994 quite similar to those of 1989. Finally, to Model 6 we added party competence variables. This addition, however, gives rise to an important concern to which we now turn.

Table 8.4. R 2s of Seven Multivariate Models to Predict Party Choice

Model 1

Model 2

Model 3

Model 4

Model 5

Model 6

Model 7

Flanders

.07

.14

.17

.22

.23

.25

.39

Wallonia

.11

.29

.30

.32

.33

.35

.44

Denmark

.07

.24

.26

.28

.28

.39

.48

West Germany

.04

.27

.30

.31

.31

.42

.52

East Germany

.04

.21

.22

.23

.23

.28

.33

Greece

.04

.34

.36

.37

.37

.40

.53

Italy

.05

.30

.33

.34

.34

.37

.48

Spain

.06

.22

.26

.28

.28

.42

.57

France

.06

.27

.30

.31

.32

.34

.41

Ireland

.04

.17

.19

.20

.20

.24

.34

Northern Ireland

.34

.40

.41

.43

.44

.46

.50

Luxembourg

.05

.22

.27

.29

.29

.37

.47

The Netherlands

.10

.26

.27

.28

.28

.43

.49

Portugal

.05

.29

.29

.31

.31

.42

.51

Great Britain

.06

.18

.19

.22

.22

.43

.53

EU

.08

.25

.27

.28

.28

.36

.45

Source: European Election Study 1994. The regressions contain the following independent variables: Model 1: religion and class; Model 2: Model 1 plus left–right distance; Model 3: Model 2 plus three policy distances; Model 4: Model 3 plus rankings of 11 valence issues; Model 5: Model 4 plus EC approval; Model 6: Model 5 plus party size; Model 7: Model 6 plus party competence.

Earlier in this chapter we articulated our concern that these ‘best at’ or party competence variables might be circular, i.e., that they might be not the cause but the consequence of party preference. To assess the direction of the causal relationship between party preference (the dependent variable in the models displayed in Table 8.4) and these party competence variables, we used LISREL analyses to test two rivalling causal models, one founded on the premise that party competence is (p.177) one of the causes of party preferences, the other on the premise that it is the consequence. Neither one of these models fits the pattern of empirical relations between the variables satisfactorily, which indicates that a reciprocal relationship exists between them: to some extent the competence variable is the cause of preference, but to some extent it is also the result of those preferences. Therefore we added this variable to the regression models listed in Table 8.4 only after the more unambiguous causes were included. The addition of this variable increases (EU‐wide) the variance explained by nine per cent to 0.45.

Depending upon how one feels about the status of the different independent variables, one may derive from Table 8.4 different conclusions regarding the impact of issues. If one considers only the three position issues and the set of 11 valence issues as ‘proper’ issue variables, the contribution of issues to the explanation of party preferences amounts to a meagre three per cent. Considering left–right as sufficiently policy‐related to be of similar status as specific issue concerns in the Responsible Party Model, the picture changes to one where issues contribute a full 20 per cent to variance explained. If one is willing to over‐interpret the party competence variables, the result is that in Europe no less than 37 per cent of variance in party preferences can be explained by issues.

All these figures are to some extent underestimates of the results we would get if more and different issues had been included in the surveys. Yet all available evidence from other surveys—such as national election studies in various countries—points to a decreasing contribution of additional issue variables to a total explanation as the number of issue variables increases. The reason for this is simply that in most (stable) political systems, voters' and parties' positions on issues are structured in only a few different dimensions, the main effects of which are rapidly tapped by including variables relating to socio‐structural and ideological identifications.

Having evaluated different models in terms of their explanatory power for each of the political systems of the EU, as well as for the combined electorates from these 15 systems, we are still left with two questions. First, how do the effects of the independent variables compare with one another when we do not focus on stagewise increases in R2, but rather on regression effects when variables are viewed in conjunction with one another? Second, to what extend does it make sense to combine the data of all the different political systems (as we did in (p.178) the EU‐wide rows of Table 8.4) when the country‐specific analyses (the rows for each of the political systems) seem to indicate that the importance of particular explanatory factors differs considerably between them?

Table 8.5 presents the standardized regression coefficients for Models 6 and 7 when they are applied EU‐wide, i.e. on the data set in which the surveys from all the member‐states of the EU are combined. Model 6 shows clearly that only two factors have a strong impact on party preferences: the left–right distance between voters and parties (the larger the distance, the smaller the preference, hence the negative sign of the coefficient), and the size of a party, which represents that voters will, ceteris paribus, prefer large parties over small ones, presumably because they offer a better prospect for getting ideological or policy stands realized as actual government policies. All specific issue variables are weak, as is the effect of EU approval. Variables related to class are rather weak, as they have consistently been found to be in studies conducted since the late 1970s. Religious factors are of intermediate strength.

Yet to turn to the second question formulated above, how should we look upon these EU‐wide effects when we see in the country‐specific analyses that the explanatory power of, for example, socio‐structural factors varies between a high of 0.34 in Northern Ireland and a low of 0.04 in western and eastern Germany, Greece, and Ireland (see Table 8.4, Model 1)? Is it not necessary, one might wonder, to specify interaction effects that would tap differences between countries in the effects of the independent variables? Earlier research showed that indeed the regression effect of some variables is so different in the various countries that some interaction terms are warranted (van der Eijk, Franklin, and Oppenhuis 1996. The most important of these was an interaction between left–right distance on the one hand, and the perceptual agreement in a country about the position of the various parties on the left–right continuum on the other. Including this interaction term in Model 6 yields the results displayed in the middle column of Table 8.5. We find for 1994 the same value of this coefficient as in 1989: −0.04, which indicates that in countries where voters agree more with one another about parties' left–right positions, the effect of the left–right distance between a voter and a party has a stronger impact on voters' preferences for parties. A comprehensive scan was performed to assess whether or not any new and additional interaction terms involving the independent variables of Model 6 or 7 are necessary. The (p.179) independent variables of Model 6 or 7 are necessary. The results of this scan were overwhelmingly negative; in other words, there is no compelling need to complicate the models by including country‐specific interaction effects.

Table 8.5. Regression Coefficients (Betas) of the Three Most Encompassing Models of Party Choice (EU‐Wide)

Model 6

Model 6 + interaction

Model.7 + interaction

Religion

.16

.16

.13

Subjective social class

.09

.09

.07

Family income

.03

.03

.02

Left–right ‐ ideological distance

−.33

−.34

−.28

Currency ‐ policy distance

−.06

−.06

−.05

Employment – policy distance

−.06

−.06

−.04

Borders ‐ policy distance

−.04

−.04

−.04

National issue 1 – ranking

.03

.03

.03

Unemployment – ranking

.03

.03

.02

Stable prices – ranking

.04

.03

.03

National issue 2 – ranking

.04

.04

.02

European unification – ranking

.02

.02

.02

Immigration – ranking

.03

.03

.02

National issue 3 – ranking

.03

.03

.02

Agricultural surpluses ‐ ranking

.03

.03

.02

Environment – ranking

.06

.06

.03

National issue 4 – ranking

.03

.03

.02

Crime – ranking

.03

.03

.02

EU approval

.04

.04

.03

Party size

.27

.27

.16

Party competence (1st probl.)

.17

Party competence (2nd probl.)

.12

Party competence (3rd probl.)

.11

Interaction left–right distance‐perceptual agreement

−.04

−.04

R2

.36

.36

.45

Source: European Election Study 1994.

The differences in the effects of policy preferences on party choice in Tables 8.3 and 8.4 should therefore not be interpreted as differences between these countries in the process that generates preference (or (p.180) lack thereof) for political parties. They merely reflect differences in distributions of the respective policy preferences, not differences in how those preferences impinge on party choice. Having discovered this, we may indeed conclude that we are justified in combining the surveys from the different member‐states of the EU into a single analysis, because the mechanisms by which policy preferences affect party choice is basically the same everywhere in these countries.

Discussion and Conclusions

How do these findings help us to assess the Responsible Party Model? To the extent that the model focuses on specific programmatic stances of political parties, deriving for instance from their manifestos or their observable political behaviour, our findings are sobering. We find hardly any motivational basis, in terms of specific issues and policies, for voters' party preferences. One might wonder whether this result is caused either by a paucity of relevant issue and policy questions, or by the way in which these questions were phrased in our surveys. Although more and different questions would probably make some difference to the overall picture of our findings, we have no reason to think that they would really alter it. The experience of analysing other surveys (the European Election Study 1989, as well as national election studies in various countries) leads to the conclusion that even the most extensive set of questions on the most salient issues contributes relatively little to the explanation of party choice once other, causally precedent, factors such as socio‐structural factors and ideology have been taken into account.

On the other hand, if we are willing to look at the Responsible Party Model in a somewhat more relaxed way—along the lines suggested by Thomassen (1994)—by focusing not only on specific policy and issue concerns, but more generally on substantive political concerns and orientations of voters relating to how parties position themselves in politics, then the Responsible Party Model becomes more persuasive. The strength of ideological motivations is considerable and, in all countries of the European Union, it is amongst the most significant motivational factors in party choice. Of course, parties decide less directly about their ideological position than about their stance on specific issues, as (p.181) their left–right position is the product of inter‐party conflict on a host of specific issues. In spite of this, looking at left–right as akin to an issue, we find a strong basis for considering European elections from the perspective of the Responsible Party Model.

This conclusion is reinforced when we consider the accumulated evidence of Chapters 6, 7, and 8. Chapter 6 showed that party groups of different ideological colours compete in the European Parliament along a left–right dimension. Thus at the level of the European Parliament a party system develops, which offers voters a programmatic choice that is clearly structured along the left–right dimension. Voters have rather sophisticated perceptions of party positions in left–right terms (Chapter 7). Here we find that left–right positions are not only relatively powerful determinants of party choice, but that the effect of left–right is basically the same in all countries. With respect to policy preferences of voters in generic ideological terms, elections for the European Parliament thus seem to satisfy all requirements of the Responsible Party Model reasonably well. This is what we will discuss in greater detail in the next chapter.

Notes

(1) The wording of these questions was presented in Chapter 3 of this volume.

(2) The wording of these questions was presented in Chapter 6 of this volume.

(3) All analyses reported in this chapter were performed on the post‐election survey of the European Election Study 1994. For each country, the data have been weighted in such a manner as to reflect the actual outcome of these elections in that country. In the individual‐level analyses performed in this chapter on the combined data from all countries, these weights were further adapted so as to maintain on the one hand their capacity to reflect the actual election result in each country, and on the other hand to give each country as a whole an equal weight in the analysis. The weighting procedures used are identical to those used in van der Eijk, Franklin, et al. (1996, Appendix B). Moreover, as in the analyses reported in that volume, we distinguish more political systems than member‐states of the Union, by differentiating between the two main Belgian regions of Flanders and Wallonia, between Northern Ireland and Great Britain, and by distinguishing between the two former German states.

(4) These percentages are lower‐bound estimates of the percentages in the populations for two reasons. First, in our surveys we have no information about citizens who did not participate in the study, a group that consists of a larger proportion of people with little political interest than those who did (p.182) co‐operate with the survey. Secondly, there is an incentive in any interview to express a preference even if one does not really have an opinion.

(5) Elsewhere, non‐response was found to be lower for specific issues than for left–right (van der Brug 1996). Such differences are also known from the European Elections Studies (Schmitt 1995). The reason that no such differences materialize in Table 8.1 is due to the fact that only one of two non‐response categories is displayed—answering categories of the left–right scales provide a response ‘refused’ in addition to the usual ‘don't know’. Since refusal does not measure the absence of preference but rather an unwillingness to reveal it, we decided to disregard it in the present context.

(6) If we were concerned with the evolution of issue preferences or ideological positions over time, the causal connection might well be reversed to at least some degree, with party preferences influencing policy preferences, and the latter influencing values and ideological predispositions. At any one election, such as that of 1994, however, the much higher short‐ and medium‐term stability of ideological and value predispositions in comparison to specific policy and issue preferences ensures their causal priority (Campbell et al. 1960; Fiorina 1988; Granberg and Holmberg 1988). This assumption may also be tested by using more explicit causal models, as we shall do below. To the extent that we find reciprocal causation this will indicate that the effects of issue preferences have been inflated.

(7) Tillie (1995) demonstrates that it is indeed quite possible to commit causal mis‐attributions if only actual party choice is analysed. Of course, in strict two‐party or two‐candidate contests this argument does not hold, because then, but only then, the complete rank order of preferences is known by knowing which is first.

(8) ‘All parties’ has to be understood as all parties deemed sufficiently important to be included in the question. In general this includes all parties represented in the (national) parliaments, as well as those which, on the basis of contextual knowledge (such as opinion polls) could have been expected to gain such representation in a national election held on the same day. The number of parties about which this question (sometimes referred to as the ‘probability to vote’ question) was asked ranged from a low of five in Greece and Portugal to a high of nine in The Netherlands. The total number of parties included over all countries amounts to 98 (see the appendix to Chapter 7 for details).

(9) Some electoral systems allow the expression of multiple preferences. One can think of STV, the German dual ballot system, and the Luxembourg system of multiple votes with panachage. As the analysis in this chapter will show, however, the ‘probability to vote’ questions pose no special problems in these cases. To say whether one would ever vote for a party does not require the respondent to distinguish between the various methods by which he or she might do so.

(p.183)

(10) See van der Eijk, Franklin, and Mackie (1996) for details about percentages of voters voting identically or differently in European and national elections. In most cases where party choice was not the same as it would have been in national elections at the same time, the party actually chosen in the European election was one that was also given a high ‘probability to vote’ score for national elections. We may therefore assume that distortions caused by such differences in choice between national and European contexts are minimized. Moreover, analyses of such differences in party choice by Oppenhuis, van der Eijk, and Franklin (1996) suggest that they are not so much the result of different sets of substantive considerations that voters use for each, but rather of different (strategic) weighting of the elements of one and the same set of political considerations. This implies that analysis of these responses yields results that are also valid for party choice in European elections. An advantage of focusing the ‘probability to vote’ question on national rather than European elections is that this avoids the risk that the low‐stimulus character of European elections may cause high non‐response.

(11) Many other analyses can be performed to demonstrate that the replacement of actual party choice with this specific set of preference scores for each of the parties is justified. The most important of these is an unfolding analysis which demonstrates that the scores on the ‘probability to vote’ questions can be understood as emanating from the same latent factors for all parties, hence the origins of this score for the party which one actually voted for are the same as for all other parties. Such validating analyses have been reported in detail by Tillie (1995).

(12) Obviously, these ways of arranging data yield a matrix containing a very large number of ‘cases’: the sum over all countries of the number of respondents in each country times the number of parties for which the ‘probability to vote’ question was asked. After deleting missing data (but before weighting—see below), this data set contains 85,853 records.

(13) A number of conditions have to be fulfilled for these variables to have any appreciable effect on the explanation of party preferences. Consequently, to the extent that their explanatory power is low, this may be the result of a number of causes: parties not being sufficiently differentiated in terms of their issue priorities, voters being insufficiently cognizant of those differences, or voters not reacting to any such differences in their party preferences. Conversely, to the extent that these variables do contribute to the explanation of party preferences, the underlying logic of issue‐ownership and its consequences is empirically vindicated.

(14) Party identification measures were deliberately disregarded, as we consider the concept inappropriate for the analysis of electoral behaviour in the cleavage‐structured European multi‐party systems; see van der Eijk and Niemöller 1983.

(15) Taking religion as the example to explain this procedure, we conducted a set of regression analyses (one for each party in each country) in which (p.184) the preference score for that party was estimated from each voter's religious denomination and frequency of church attendance. The results indicate to what extent, in that specific country and for that specific party, religious differences between people determine their preference for that party. These effects of religion on preference are encapsulated in the regression predictions made for each respondent, which in statistical parlance are referred to as the y‐hats. These y‐hats are simply linear transformations of the original independent variable, and may therefore be used again as predictors of party choice. The y‐hats are saved and added to the stacked data set, yielding a variable that is comparable for all parties (and countries) and that can be referred to as the ‘predicted religious effect’, or, more briefly, ‘religion’. The actual variable added to the stacked data set is the deviation of the y‐hats from their mean. This still encapsulates the variance in party preference caused by religious differences but prevents the average of these y‐hats from being subsumed in this variable, an outcome to be avoided because in a stacked data set these averages cause correlations with party preference that cannot be attributed to religion and that would disturb subsequent analyses. See van der Eijk, Franklin, et al. (1996), Chapter 20, for a more detailed account of this procedure.

(16) In the procedure to construct these variables each of these issues was scored as follows: 0: not mentioned as one of the three most important problems; 1: mentioned as 3rd most important; 2: mentioned as 2nd most important, and 3: mentioned as the most important problem by the respondent.

(17) This variable was constructed using the same y‐hat procedure explained above for religion. As predictor the cumulative score on a scale consisting of four items was used, referring to whether or not one would feel ‘sorry’ were the EU dissolved, whether or not EU‐membership is regarded as beneficial for one's country, whether or not EU‐membership is ‘a good thing’, and whether or not one is ‘in favour’ of European unification. For details on the scale and its construction, see van der Eijk, Franklin, et al. 1996, Appendix B.

(18) The 12 issues in 1989 included four country‐specific issues and the following common issues: unemployment, stable prices, arms limitation, environmental protection, European unification, agricultural surpluses, EC membership of Turkey, and realization of the SEM. The difference in explanatory power between the 1989 and 1994 data cannot be attributed to the somewhat different selection of issues, as five of the common issues are identical in both years, while the differences in their respective explanatory power for those five is of the same magnitude as for the entire set (for details about these 1989 analyses, see Oppenhuis 1995, Appendix 8).

(19) In 1989 respondents were asked to indicate the importance of each of these issues in two different ways: by rating the importance of each, and—as in the 1994 survey analysed here—by indicating the three most important (p.185) ones (so‐called rankings). By far most of the explanatory power of these issues in 1989 derived from the rating information, while the rankings contributed only occasionally to the explanation of party preferences (Oppenhuis 1995, Appendix 8). When we analysed the 1989 data using only the ranking information, the average R2 value comes close to the four per cent reported in Table 8.3.

(20) In Chapter 7 of this volume (Table 7.2), van der Brug and van der Eijk showed that exceedingly low agreement exists between voters about which party is best at which problem, particularly when compared with the perceptual agreement of parties' left–right positions, and even in comparison with the perceptions of parties' positions on the three non‐salient position issues. This finding seems mainly explicable by interpreting the response to the ‘best at’ questions as preferential rather than perceptual.

(21) Although party size is a consequence of the number of people who support a party, and thus in the long term may be seen in part as the consequence of issue preferences, at any one election the sizes of parties can be regarded as exogenous. For a more detailed discussion of the interpretation of party size as reflecting voters' strategic considerations, see van der Eijk, Franklin, et al. 1996: 352–3, and Tillie 1995.

(22) One could argue that adding left–right to the equation after (rather than before) the position and valence issue variables would increase the contribution to the explanation of the latter. Still, such an alternative order of inclusion of variables in the equation does not seem very compatible with the notion that left–right orientations function as political schema (Conover and Feldman 1984; Kerlinger 1984), super issues (Inglehart 1984), or ideological identification (van der Eijk and Niemöller 1983).

Notes:

(1) The wording of these questions was presented in Chapter 3 of this volume.

(2) The wording of these questions was presented in Chapter 6 of this volume.

(3) All analyses reported in this chapter were performed on the post‐election survey of the European Election Study 1994. For each country, the data have been weighted in such a manner as to reflect the actual outcome of these elections in that country. In the individual‐level analyses performed in this chapter on the combined data from all countries, these weights were further adapted so as to maintain on the one hand their capacity to reflect the actual election result in each country, and on the other hand to give each country as a whole an equal weight in the analysis. The weighting procedures used are identical to those used in van der Eijk, Franklin, et al. (1996, Appendix B). Moreover, as in the analyses reported in that volume, we distinguish more political systems than member‐states of the Union, by differentiating between the two main Belgian regions of Flanders and Wallonia, between Northern Ireland and Great Britain, and by distinguishing between the two former German states.

(4) These percentages are lower‐bound estimates of the percentages in the populations for two reasons. First, in our surveys we have no information about citizens who did not participate in the study, a group that consists of a larger proportion of people with little political interest than those who did (p.182) co‐operate with the survey. Secondly, there is an incentive in any interview to express a preference even if one does not really have an opinion.

(5) Elsewhere, non‐response was found to be lower for specific issues than for left–right (van der Brug 1996). Such differences are also known from the European Elections Studies (Schmitt 1995). The reason that no such differences materialize in Table 8.1 is due to the fact that only one of two non‐response categories is displayed—answering categories of the left–right scales provide a response ‘refused’ in addition to the usual ‘don't know’. Since refusal does not measure the absence of preference but rather an unwillingness to reveal it, we decided to disregard it in the present context.

(6) If we were concerned with the evolution of issue preferences or ideological positions over time, the causal connection might well be reversed to at least some degree, with party preferences influencing policy preferences, and the latter influencing values and ideological predispositions. At any one election, such as that of 1994, however, the much higher short‐ and medium‐term stability of ideological and value predispositions in comparison to specific policy and issue preferences ensures their causal priority (Campbell et al. 1960; Fiorina 1988; Granberg and Holmberg 1988). This assumption may also be tested by using more explicit causal models, as we shall do below. To the extent that we find reciprocal causation this will indicate that the effects of issue preferences have been inflated.

(7) Tillie (1995) demonstrates that it is indeed quite possible to commit causal mis‐attributions if only actual party choice is analysed. Of course, in strict two‐party or two‐candidate contests this argument does not hold, because then, but only then, the complete rank order of preferences is known by knowing which is first.

(8) ‘All parties’ has to be understood as all parties deemed sufficiently important to be included in the question. In general this includes all parties represented in the (national) parliaments, as well as those which, on the basis of contextual knowledge (such as opinion polls) could have been expected to gain such representation in a national election held on the same day. The number of parties about which this question (sometimes referred to as the ‘probability to vote’ question) was asked ranged from a low of five in Greece and Portugal to a high of nine in The Netherlands. The total number of parties included over all countries amounts to 98 (see the appendix to Chapter 7 for details).

(9) Some electoral systems allow the expression of multiple preferences. One can think of STV, the German dual ballot system, and the Luxembourg system of multiple votes with panachage. As the analysis in this chapter will show, however, the ‘probability to vote’ questions pose no special problems in these cases. To say whether one would ever vote for a party does not require the respondent to distinguish between the various methods by which he or she might do so.

(10) See van der Eijk, Franklin, and Mackie (1996) for details about percentages of voters voting identically or differently in European and national elections. In most cases where party choice was not the same as it would have been in national elections at the same time, the party actually chosen in the European election was one that was also given a high ‘probability to vote’ score for national elections. We may therefore assume that distortions caused by such differences in choice between national and European contexts are minimized. Moreover, analyses of such differences in party choice by Oppenhuis, van der Eijk, and Franklin (1996) suggest that they are not so much the result of different sets of substantive considerations that voters use for each, but rather of different (strategic) weighting of the elements of one and the same set of political considerations. This implies that analysis of these responses yields results that are also valid for party choice in European elections. An advantage of focusing the ‘probability to vote’ question on national rather than European elections is that this avoids the risk that the low‐stimulus character of European elections may cause high non‐response.

(11) Many other analyses can be performed to demonstrate that the replacement of actual party choice with this specific set of preference scores for each of the parties is justified. The most important of these is an unfolding analysis which demonstrates that the scores on the ‘probability to vote’ questions can be understood as emanating from the same latent factors for all parties, hence the origins of this score for the party which one actually voted for are the same as for all other parties. Such validating analyses have been reported in detail by Tillie (1995).

(12) Obviously, these ways of arranging data yield a matrix containing a very large number of ‘cases’: the sum over all countries of the number of respondents in each country times the number of parties for which the ‘probability to vote’ question was asked. After deleting missing data (but before weighting—see below), this data set contains 85,853 records.

(13) A number of conditions have to be fulfilled for these variables to have any appreciable effect on the explanation of party preferences. Consequently, to the extent that their explanatory power is low, this may be the result of a number of causes: parties not being sufficiently differentiated in terms of their issue priorities, voters being insufficiently cognizant of those differences, or voters not reacting to any such differences in their party preferences. Conversely, to the extent that these variables do contribute to the explanation of party preferences, the underlying logic of issue‐ownership and its consequences is empirically vindicated.

(14) Party identification measures were deliberately disregarded, as we consider the concept inappropriate for the analysis of electoral behaviour in the cleavage‐structured European multi‐party systems; see van der Eijk and Niemöller 1983.

(15) Taking religion as the example to explain this procedure, we conducted a set of regression analyses (one for each party in each country) in which (p.184) the preference score for that party was estimated from each voter's religious denomination and frequency of church attendance. The results indicate to what extent, in that specific country and for that specific party, religious differences between people determine their preference for that party. These effects of religion on preference are encapsulated in the regression predictions made for each respondent, which in statistical parlance are referred to as the y‐hats. These y‐hats are simply linear transformations of the original independent variable, and may therefore be used again as predictors of party choice. The y‐hats are saved and added to the stacked data set, yielding a variable that is comparable for all parties (and countries) and that can be referred to as the ‘predicted religious effect’, or, more briefly, ‘religion’. The actual variable added to the stacked data set is the deviation of the y‐hats from their mean. This still encapsulates the variance in party preference caused by religious differences but prevents the average of these y‐hats from being subsumed in this variable, an outcome to be avoided because in a stacked data set these averages cause correlations with party preference that cannot be attributed to religion and that would disturb subsequent analyses. See van der Eijk, Franklin, et al. (1996), Chapter 20, for a more detailed account of this procedure.

(16) In the procedure to construct these variables each of these issues was scored as follows: 0: not mentioned as one of the three most important problems; 1: mentioned as 3rd most important; 2: mentioned as 2nd most important, and 3: mentioned as the most important problem by the respondent.

(17) This variable was constructed using the same y‐hat procedure explained above for religion. As predictor the cumulative score on a scale consisting of four items was used, referring to whether or not one would feel ‘sorry’ were the EU dissolved, whether or not EU‐membership is regarded as beneficial for one's country, whether or not EU‐membership is ‘a good thing’, and whether or not one is ‘in favour’ of European unification. For details on the scale and its construction, see van der Eijk, Franklin, et al. 1996, Appendix B.

(18) The 12 issues in 1989 included four country‐specific issues and the following common issues: unemployment, stable prices, arms limitation, environmental protection, European unification, agricultural surpluses, EC membership of Turkey, and realization of the SEM. The difference in explanatory power between the 1989 and 1994 data cannot be attributed to the somewhat different selection of issues, as five of the common issues are identical in both years, while the differences in their respective explanatory power for those five is of the same magnitude as for the entire set (for details about these 1989 analyses, see Oppenhuis 1995, Appendix 8).

(19) In 1989 respondents were asked to indicate the importance of each of these issues in two different ways: by rating the importance of each, and—as in the 1994 survey analysed here—by indicating the three most important (p.185) ones (so‐called rankings). By far most of the explanatory power of these issues in 1989 derived from the rating information, while the rankings contributed only occasionally to the explanation of party preferences (Oppenhuis 1995, Appendix 8). When we analysed the 1989 data using only the ranking information, the average R2 value comes close to the four per cent reported in Table 8.3.

(20) In Chapter 7 of this volume (Table 7.2), van der Brug and van der Eijk showed that exceedingly low agreement exists between voters about which party is best at which problem, particularly when compared with the perceptual agreement of parties' left–right positions, and even in comparison with the perceptions of parties' positions on the three non‐salient position issues. This finding seems mainly explicable by interpreting the response to the ‘best at’ questions as preferential rather than perceptual.

(21) Although party size is a consequence of the number of people who support a party, and thus in the long term may be seen in part as the consequence of issue preferences, at any one election the sizes of parties can be regarded as exogenous. For a more detailed discussion of the interpretation of party size as reflecting voters' strategic considerations, see van der Eijk, Franklin, et al. 1996: 352–3, and Tillie 1995.

(22) One could argue that adding left–right to the equation after (rather than before) the position and valence issue variables would increase the contribution to the explanation of the latter. Still, such an alternative order of inclusion of variables in the equation does not seem very compatible with the notion that left–right orientations function as political schema (Conover and Feldman 1984; Kerlinger 1984), super issues (Inglehart 1984), or ideological identification (van der Eijk and Niemöller 1983).