Campaign Effects 4 Campaign Effects in Referendums The electorate behaves about as rationally and responsibly as we would expect, given the clarity of the alternatives presented to it and the character of the information available to it.1 —V.O. Key The campaign leading up to the Danish referendum on the Euro on 28 September 2000 began half a year prior to the actual vote. During these six months of campaigning, the issues of the single currency and European integration became the main item on the news agenda. No one could be in doubt that the stakes were high: most lampposts were covered with campaign posters and national politicians used every opportunity to promote their side of the story. In the period from March to August, the campaign accounted for around 10 per cent of the main evening news and, in the final month of the campaign, 25 per cent of all television news covered the referendum (de Vreese 2004: 54). Daily newspapers carried on average one front-page story and three articles relating to the campaign each day during this period. By the time the referendum day came around, almost 88 per cent of Danish voters turned out to express their opinion. In contrast, the official campaign leading up to Irish referendum on the Nice Treaty on 7 June the following year lasted barely a month. With the next Irish general election just around the corner, most mainstream parties chose to save their campaigning funds for this occasion. As a consequence, there were few posters, little door-to-door canvassing, and relatively limited news coverage of the referendum campaign. After this lax and lethargic campaign, a record low of only 35 per cent of Irish voters turned out to give their say on the Treaty. In both of these referendums, the ballot proposal was rejected by a small margin of voters, but we would expect that the decision-making processes that led to these outcomes were different, given the very dissimilar political environments. In the former case, the campaign provided an abundance of information about the ballot proposal, the salience of the issue was high, and the political parties clearly expressed their recommendations. In the latter case, the campaign provided little information about the treaty, issue salience was low, and the parties gave ambiguous signals. Chapter 3 focused on the individual voter and variations in political awareness. Examining micro-level decision-making yielded a number of insights into how information influences vote choice. In particular, it was shown that voters who pay more attention to politics are more likely to vote on the basis of their opinions about Europe, whereas voters who receive less information about politics will tend to rely more on their feelings about domestic politics. However, information effects depend not only on the individual's ability to receive political messages, but also on the degree to which citizens are provided with the necessary information concerning the issue at stake. We would expect that vote choices are influenced by the context in which they are taken. An examination of the campaign environment is thus important to fully understand voters' decision-making. Based on the model presented in Chapter 2, our theoretical expectation is that when more information is available and issue salience is high, lower levels of uncertainty will increase the utility of voting and reduce abstention. Moreover, high-intensity campaigns provide more information about the location of the alternatives and therefore we expect that people become more capable of employing their issue preferences in their vote choice. This chapter examines these propositions by addressing the following question: How do variations in the intensity of the referendum campaign affect patterns of voting behaviour? Campaign effects are measured at two levels. First, campaign intensity is measured at the aggregate level and the effect on levels of turnout and 'issue voting' is evaluated. Second, the chapter analyses individual-level variation in exposure to campaign information and assesses the impact on knowledge and voting behaviour. These micro-, macro-, and multi-level analyses reveal that context matters: the campaign environment in EU referendums acts both as an informer and a mobilizer and influences patterns of behaviour. Turnout is higher and issue voting is more prevalent in high-intensity campaigns. The chapter proceeds as follows. First, it reviews the main theories of campaign and information effects. Second, it discusses the specific role of campaigns in EU referendums and presents testable hypotheses. Third, a measurement of campaign intensity, based on multiple indicators, is 84 8S Europe in Question applied to the nineteen referendum cases, and the impact of variations in campaign intensity on aggregate levels of turnout and issue voting is evaluated. Finally, the effect of individual-level variations in exposure to campaign information is analysed and the mediating effect of campaign intensity on vote choices is examined. 4.1. Theories of campaign effects The effect of election campaigns on voting behaviour has been studied extensively, especially in the American context (see Shaw 1999 for an overview). Studies of campaign effects have historically found little empirical evidence of election campaigns influencing vote behaviour. Instead, a number of studies has made the strong case that campaigns do little more than activate latent candidate or party preferences and can generally be viewed as having 'minimal consequences' (see Lazarsfeld et al. 1944; Rosenstone 1983; Abramowitz 1988; Gelman and King 1993). This literature typically emphasizes fixed social, demographic, and political determinants of voter behaviour, such as class or partisanship, and controlling for the effect of these stationary characteristics leaves little room for campaigns to make a difference. Yet, despite strong evidence for the 'minimal effects' thesis, the issue of campaign influence remains an open question in the literature. Recent studies have argued that the 'minimal effects' hypothesis is an artefact of conceptual and methodological problems in the literature (see Shaw 1999; Iyengar and Simon 2000; Hillygus 2005). In particular, it has been argued that the existing literature has focused too much on the impact of campaigns on preferences and has consequently overlooked other important effects. As Iyengar and Simon have noted: Perhaps the most fundamental obstacle to understanding the real-world role of political campaigns is a conceptual limitation of what effects arc deemed relevant. Traditional research has looked mainly at persuasion... The single-minded quest for persuasion effects has ignored the transmission of information, the setting of campaign agendas, and the alteration of the criteria by which candidates are judged (2000:151). Iyengar and Simon (2000) concede that campaigns might have little direct effect on voters' preferences, since they tend to activate voters' prevailing partisan sentiments. Yet, they argue that campaigns have pivotal effects in the areas of 'voter learning' and 'agenda control', which 86 Campaign Effects should be explored further. After all, campaigns decide the quantity and the nature of the information disseminated to voters. Other studies support this argument about the importance of campaigns as informers. Franklin's (1991) study of US Senate campaigns found that voters exposed to senatorial campaigns were more competent in their understanding of their incumbent senator's position on a liberal-conservative scale. Alvarez (1997) corroborates this result for a broad array of knowledge and attitude questions in analyses of contemporary presidential elections, and equally Barrels (1988) has found significant information gains during the presidential primaries of 1984. Several scholars have demonstrated that voters exposed to high-intensity campaigns, which make larger volumes of information available, are more engaged and cast more informed votes. According to Kahn and Kenney's (1997) study of US Senate elections, citizens experiencing an intense race use more sophisticated criteria when evaluating the contestants than citizens following less intense races. In his extensive study of Senate elections, Westlye (1991) equally demonstrates that 'issue voting, and voting on the basis of candidates' ideologies, takes place more often among voters who have information about both candidates as a result of exposure to a high-intensity campaign and less among those who have information only about the incumbent' (1991: 16). In short, these studies suggest that campaigns serve to inform voters and, in turn, help to structure vote choices. As Schmitt-Beck and Farrell (2002) point out, the effect of campaigns is highly contingent on the nature of the campaign and the individual voter. In line with the US literature, the comparative studies in Farrell and Schmitt-Beck's (2002) edited volume on campaign effects show that high-intensity campaigns have a greater effect than low-intensity campaigns and that the scope of campaign influence depends on the political interest of voters. Drawing on Zaller's work (1992), they conclude that the influence of campaigns is mediated by political awareness and predispositions of individuals. Recent studies have also focused on the campaign as a mobilizer. Kuklinski et al. (2001) have shown how the political environment affects the quality of citizens' political decision-making by not only acting as a source of information, but also by playing a motivational role of encouraging citizens to invest effort and time in making political judgements. Moreover, a growing body of research has argued that campaign efforts influence an individual's likelihood of voting (for an overview see Hillygus 2005). Experimental research has found consistent and substantial campaign effects on voting turnout (Ansolabehere and Iyengar 1996; Gerber and Green 2000; Lassen 2005). Similarly, some survey research has 87 Europe in Question Campaign Effects found a positive correlation between total campaign expenditures and competitiveness of the race on the one hand and turnout in elections on the other hand (Cox and Munger 1989; Franklin 2004). Building on this literature, this chapter examines the role of the campaign as an informer and a mobilizer in EU referendums. Rather than focusing on the broad effects of campaigns, including agenda-setting and persuasion, this chapter examines how variations in campaign intensity, that is, the quantity of information provided, affect patterns of voting behaviour. 4.2. Campaign effects in EU referendums Only recently has the literature on referendums started to discuss campaign effects in a systematic manner. Yet, the campaign context is crucial to the study of opinion formation in referendums since the degree of uncertainty associated with unfamiliarity with the issue and unclear elite cues make voters especially susceptible to campaign influence in referendums (Magleby 1984; Farrell and Schmitt-Beck 2002; de Vreese 2007). Leduc (2002) argues that since voting behaviour in referendums often exhibits greater volatility than is found in elections, campaigns are likely to play a greater role in the opinion formation process. In the Swiss context, several studies have shown that campaigns can make a difference (Btitzer and Marquis 2002; Christin, Hug, and Sciarini 2002; Kriesi 2005). Biitzer and Marquis (2002) have shown that, in accordance with Zaller's theory (1992), the division of political elites is of prime importance to opinion formation in referendum campaigns, and Christin, Hug, and Sciarini (2002) also find that Swiss voters can use campaign information, such as party endorsements, to vote competently. With regard to European referendums, the importance of the information environment during a campaign has been acknowledged in several studies (Siune and Svensson 1993; de Vreese and Semetko 2002, 2004; de Vreese 2004; Gerstle 2006), but no study has presented a systematic comparative study of its effect on voting behaviour. De Vreese and Semetko (2004) show that the information environment, in particular the media, served to crystallize individtial opinion on the vote choice in the Danish referendum on the single currency. In their study of the Dutch referendum on the Constitutional Treaty, Schuck and de Vreese (2008) also found that voters who were more exposed to the pro-Treaty news media were more likely to shift to the yes-side during the campaign. It is not surprising that these studies find that campaigns matter in referendums. Indeed, we may expect that they matter more than they do in elections. Referendums present a different choice to elections. No political parties or candidate names appear on the ballot and voters must choose amongst alternatives that are sometimes unfamiliar. If voters know little about the specific ballot proposal, it is mainly the various information sources available to them over the course of a campaign that provide the basis for their opinion on the ballot question. The campaign can therefore play a crucial role in informing voters about the choices and in disseminating relevant party cues. Hence, as discussed in Chapter 2, the vote choice will depend not only on the characteristics of the individual, but also on the political environment in which vote decisions are taken. People with high levels of political awareness may act differently from people with lower levels of awareness, as shown in Chapter 3. Moreover, electorates that are given extensive information on the ballot proposal can be expected to behave differently from electorates deciding in campaign settings where very little information is provided. More information will reduce uncertainty, and thereby increase the utility of the ballot proposal. However, sometimes campaign information may be misleading and deceptive. For example, political elites may deliberately set out to mislead the public about the consequences of the ballot or the reversion point. Following Lupia and McCubbins (1998), we can distinguish between two types of information effects: enlightenment (information will help voters to make more accurate predictions about the consequences of their actions) and deception (information will reduce the ability of voters to make accurate predictions). Voters are only likely to be deceived by political elites if they find them trustworthy and knowledgeable; so even if political elites set out to deceive, they are not always successful (Lupia and McCubbins 1998). Voters are thus more susceptible to persuasion by parties that they feel loyalty towards (more on this in Chapter 6). Moreover, attempts of deception can be costly for political elites if they are caught misleading the public. Political parties, in particular, may be reluctant to stray too far from the truth in referendum campaigns since this might damage their reputation more permanently. Deception is thus less likely in high-intensity and polarized campaigns. First, voters will receive more information from both sides, and this will make it easier to detect which messages are trustworthy and credible. Second, in a high-intensity two-sided campaign, the threat of verification and penalties for lying are likely to be higher, and this will make it less likely for elites to adopt a strategy of deception (Lupia and McCubbins 1998). To be sure, some deception 88 89 Europe in Question Campaign Effects will occur during referendum campaigns, but more intense campaigns will generally reduce, rather than increase, the likelihood of this. Building on the spatial model presented in Chapter 2 and the existing research on campaign effects, we would thus expect that the intensity of the campaign influences patterns of voting behaviour in EU referendums. The intensity of the campaign can be defined narrowly as 'the level of information disbursement' in a given election or referendum (Westlye 1991: 17). More specifically, the intensity of a referendum campaign can be conceptualized as the interaction between the cues emitted by competing political elite actors, media coverage, and the perceived closeness of the outcome (see Kahn and Kenney 1997; Farrell and Schmitt-Beck 2002). The interplay of these factors contributes not only to the quantity of the information available in the political environment, but also to the extent that the environment encourages individuals to process this information. The level of partisan polarization on the issue will influence the intensity and nature of competing elite cues. According to theories of issue evolution, it is mainly polarization among elites that raises the public's awareness of a new issue (see Riker 1982; Carmines and Stimson 1986, 1989). Equally, when parties take different stances on the ballot issues in a referendum, this tends to increase the salience of the referendum issue. This, in turn, may influence campaign strategies and spending, which affect the quantity and type of information available in the public sphere. The competitiveness of the contest is also likely to have an effect on the salience of the campaign and influence the strategies of both partisan and non-partisan groups, and, in turn, citizen participation (Franklin 2004). Finally, these factors will shape the way in which the news media decide to cover the referendum. Higher levels of polarization and a more competitive race will increase the 'newsworthiness' of the story and bring more coverage by the news media. The interaction of these forces determines the type of information available to voters and produces the environment in which they make their vote choices. This is illustrated in Figure 4.1. According to Figure 4.1, high-intensity campaigns are characterized by a combination of some or all of the following factors: partisan—and perhaps also non-partisan—competition on the issue, a high level of uncertainty about the outcome of the referendum, and extensive news coverage of the issue. High levels of intensity serve both to provide more information and to encourage people to invest time in obtaining information by increasing the salience of the issue at stake in the referendum. Campaign intensity is influenced by partisan polarization, which in turn Partisan polarization ■ I Perceived closeness of the race ^ News coverage Information and incentives available to voters Decision criteria of voters Figure 4.1. Elements of campaign intensity will affect the perceived closeness of the race and the coverage afforded to the issue by the news media. Following the model presented in Chapter 2, our expectation is that more information will increase the expected utility of a particular proposal to the voter (Equation 2.2). If campaigns reduce uncertainty and thus increase expected utility, we would also expect voters to have increased incentives to cast their vote (see Hypothesis 2.3). In the literature, formal models of turnout have predicted that informed citizens are more likely to vote, because information increases the voter's certainty that she is voting for the right candidate, that is, the candidate who yields the highest utility to the voter (see, e.g. Matsusaka 1995). This turnout effect is likely to exist at the aggregate level, where higher intensity campaigns will lead to higher levels of turnout, as well as at the individual level where more exposure to the campaign is likely to increase the individual's probability of voting. We can thus formulate the following hypothesis: Hypothesis 4.1: The higher the intensity of the campaign, the higher the level of turnout in the referendum. At the individual level, citizens exposed to more campaign information are more likely to vote than citizens exposed to less information. High-intensity campaigns are not only going to affect the likelihood of people voting in the first place, but they will also affect the way in which people make their specific vote choice. When campaign intensity is high, more information is available to individual voters and they are more motivated and more capable of making sophisticated judgements compatible with their underlying attitudes. Put differently, citizens will be more aware of the location of their 'ideal point' on the European integration- 90 91 Europe in Question Campaign Effects continuum and better equipped to relate this to a specific ballot proposition. In contrast, when campaign intensity is low, voters have limited access to easily available information and few incentives to make complicated decisions about the ballot proposal, and they are more likely to vote on the basis of non-issue-specific factors and depend more heavily on cognitive short cuts. High levels of campaign intensity should therefore produce effects similar to high issue salience, and as outlined in Hypothesis 2.5, we thus expect high-intensity campaigns to induce issue voting. To test these propositions at the aggregate level, the following section develops a measure of campaign intensity, which is applied to the nineteen referendum cases. 4.3. Measuring campaign intensity How do we create a measure of campaign intensity which can be used to compare campaigns across countries and over time? Most studies of campaign intensity employ a single measure to capture the level of intensity. In the American context, a frequently used measure is the summary reports about each election published in the Congressional Quarterly Weekly Reports (CQ rating) that classify campaigns dichotomously as intense or not intense (see, e.g. Kahn 1995; Westlye 1991). While the core of this measure, namely the 'competitiveness' of the race, is equally relevant in the context of EU referendums, there is little reason to believe that competitiveness can actually be classified as a binary variable. Another common measure is campaign spending by parties (see Kahn and Kenney 1997; Partin 2001). Campaign spending maybe considered a useful proxy of campaign activity, although less so in a cross-national analysis, since spending levels depends significantly on the specific national regulation of campaign funding. Moreover, it is almost impossible to get reliable data on campaign spending in the context of EU referendums, and it has therefore not been possible to use spending as a measure of campaign activity. Instead, this chapter follows the strategy of using multiple indicators, a method which has been recommended in several recent studies of campaign intensity. Both Kahn and Kenney (1997) and Sulkin (2001) recommend selecting measures that 'go back to the general conception of an intense race—that is, one that is "hard fought" in which both candidates run visible, vigorous campaigns, are successful at raising funds, and attract a good deal of media attention' (Sulkin 2001: 611). On this basis, these studies choose campaign spending, media coverage, and polling data as three indicators of campaign intensity. Adopting a similar approach, this chapter develops a measure of campaign intensity that seeks to capture the multidimensionality of the concept in the context of EU referendums. This measure relies on the following three indicators: partisan polarization, the perceived closeness of the race (uncertainty about the outcome), and media coverage. Partisan polarization on the ballot proposition is the first indicator of the intensity of the campaign, since competing partisan views are likely to lead to increased intensity of the campaign environment. In a situation where there is no partisan opposition to the ballot proposal, the referendum is also likely to be less salient to the news media and voters. Partisan polarization is measured as the percentage of parties who recommend a no-vote in the referendum weighted by these parties' share of the seats in parliament (see also Biitzer and Marquis 2002). This measure is then transformed into a 5-point scale depicting the level of polarization.2 The resulting measure provides an estimation of opposition to the ballot proposal in parliament.3 The perceived public stance on the issue—or the closeness of the race—is also likely to influence the campaign environment. If the public is perceived to be equally divided on the issue, the intensity of the campaign is likely to be higher than if there is a large majority in favour of the proposal. The perceived closeness of the race is measured as the average difference between intended yes- and no-voters in the opinion polls during the six weeks leading up to the referendum. This average figure is transformed into a 5-point scale4 to form a summated scale of all three indicators of intensity. Finally, media coverage of the referendum issue is a good indicator of the information available to the citizens. Information is mainly disseminated to the public through the media, and voter awareness of a ballot proposition is thus likely to grow with increasing media coverage (Biitzer and Marquis 2002; de Vreese and Semetko 2004; Schuck and de Vreese 2008). Moreover, media coverage can be seen as a general indicator of the saliency of an issue. In this study, news coverage has been analysed by measuring the average number of daily articles mentioning the referendum issue during the three months leading up to the referendum in two mainstream daily national newspapers with a high circulation. While not all voters will read the largest circulating newspapers, the news media will tend to respond in a similar fashion to the standard criteria of 'newsworthiness' (Kahn and 92 93 Europe in Question Kenney 1997: 1883). Hence, the referendum coverage in two large daily newspapers is a good indicator of the general level of media coverage of the campaign. In order to make cross-national comparisons possible, however, it is important that similar newspapers are examined in all countries. Newspapers were therefore selected on the basis of criteria of high circulation, quality (mainstream), and national scope (see also Kriesi et al. 1995: appendix). Applying these criteria, the following mainstream daily newspapers were chosen: Berlingske Tidende and Politiken in Denmark, Le Monde and Le Figaro in France, the Irish Independent and the Irish Times in Ireland, Luxemburger Wort in Luxembourg, Aftenposten (morning edition) and Dagbladet as the Norwegian sources, De Volkskrant and De Telegraaf in the Netherlands, and finally, El Mundo and El Pais in Spain. The coding was carried out by trained coders fluent in the respective languages and the reliability of the results is very high with an inter-coder reliability score of above 0.90. The news coverage measure was calculated as an average of daily coverage in the two national newspapers. As with the other indicators, this measure is transformed into a 5-point scale.5 Employing these indicators, the overall intensity of the referendum campaigns are shown in Figure 4.2. Figure 4.2 shows considerable variation in the level of campaign intensity on this scale which ranges from 3 to 14. The campaign environment was most intense in the second Norwegian referendum on membership (14), followed by the Danish referendum on the Euro (13), the first Norwegian referendum on membership (12), and the French referendum on the European Constitution (12). The campaigns leading up to the two Danish referendums on the ratification of the Maastricht Treaty were also highly salient. In general, the Irish referendums were the least intense, although we do find some variation. The campaign leading up to the Amsterdam Treaty has the lowest intensity; partly because it was overshadowed by the Good Friday Agreement ballot held on the same day. The salience of the issue of the Amsterdam Treaty was so low that the Irish newspapers did not even report the results of this referendum on the front-page in the days following the referendum. Equally, the campaign in the first Nice Treaty referendum was a lacklustre event, while the campaign leading up to the second referendum on the Nice Treaty was considerably more intense (as will be discussed in greater detail in Chapter 7), as was the campaign prior to the more recent Lisbon Treaty referendum. Interestingly, there was great variation in the four referendums on the European Constitution held in 2005. The votes in Spain and Luxembourg Campaign Effects Norway Accession 1994 Denmark Euro 2000 Norway Accession 1972 France ECT 2005 Denmark Maastricht 1992 Denmark Maastricht 1993 Denmark SEA 1986 Ireland Lisbon 2008 The Netherlands ECT 2005 Denmark Amsterdam 1998 Denmark Accession 1972 Ireland Nice 2002 Luxembourg ECT 2005 Ireland Accession 1972 Spain ECT 2005 Ireland Nice 2001 Ireland Maastricht 1992 Ireland SEA 1987 Ireland Amsterdam 1998 a Partisan polarization Q Closeness of the race ■ Media coverage 0 2 4 6 8 10 12 14 16 Figure 4.2. Campaign intensity in EU referendums Source: National newspapers, opinion polls, official parliamentary documents and elite interviews. were relatively lacklustre events when contrasted with the impassioned campaign in France (see Chapter 8). In the remainder of this chapter, our measure of campaign intensity will be employed to examine the impact on participation and vote choices. 4.4. Evaluating the impact of campaign intensity In this section, we examine the relationship between campaign intensity and voting behaviour at the aggregate level. In the first instance, we evaluate the theoretical propositions by presenting descriptive plots of aggregate patterns of voting and campaign intensity. While these descriptive statistics cannot be regarded as a formal statistical test of 94 95 Europe in Question Campaign Effects 100 90 30 - LU 05 © DK 72 • DK 93 e DK 00 • NO 94 m IE 72 ® DK98 DK86 ^ • a FR 05 • NO 72 • iE 98 iE 92 • IE 02 » !E87 rcn^ 9 ES 05 • NL.05 « iE 08 © IE 01 I-j-j-,- 2 4 6 i [ 8 10 12 ' 14 Campaign intensity Figure 4.3. Campaign intensity and turnout the hypotheses presented above, they nevertheless suggest the presence of a strong relationship between campaign intensity, turnout, and issue voting. 4.4.1. Campaign intensity and turnout According to Hypothesis 4.1, more intense campaigns would lead to higher levels of turnout. To evaluate this hypothesis, Figure 4.3 plots the relationship between turnout and the levels of campaign intensity. This clearly indicates a strong relationship between campaign intensity and turnout. The Pearson correlation between the two measures is 0.69. Generally, turnout is highest in referendums with high-intensity campaigns, such as the Norwegian vote on accession in 1994 and the Danish Euro referendum in 2000. Yet, there are also some outliers, such as the Luxembourgian referendum on the Constitutional Treaty where turnout is very high at 89 per cent despite low campaign intensity, but this is most likely due to compulsory voting laws in Luxembourg.6 Of course, we know that turnout varies significantly cross-nationally and this may partly drive the strong correlation, but in a simple bivariate 40- 35 15 TS 10- NO 72 • DK 00 ® FR 05 • NL 05 DK93 IE 08 NO 94 • IE 02 *DK 98 IE 01 LU05® • DK72 DK86 • • DK 92 & ES 05 a m IE 72 I -1-,-■ ■■.....-—n-1- 6 8 10 Campaign intensity 12 14 Figure 4.4. Campaign intensity and country-level issue voting analysis it is not possible to account for country-specific factors. In Chapter 7, we look in more detail at within-country variation in campaign intensity and turnout. The case study of the two Irish referendums on the Nice Treaty illustrates that the more intense campaign leading up to the second referendum vote most probably brought more voters to the polling stations. In support of Hypothesis 4.1, Figure 4.3 thus provides some suggestive evidence that high-intensity campaigns motivate people to go to the polls. 4.4.2. Campaign intensity and issue voting We would also expect that high-intensity campaigns activate attitudes, and that these attitudes play a greater role in the vote decision itself. According to Hypothesis 2.5, a highly salient campaign environment will encourage voters to rely more on their attitudes towards the ballot proposition when deciding. In order to allow easy comparison of the impact of attitudes in different referendums versus the campaign intensity, Figure 4.4 provides an overview of the predicted impact on the yes-vote of a change in the EU attitude variable in each of the referendums. When the 96 97 Europe in Question Campaign Effects impact of attitude changes is higher, it implies that issue voting is more pronounced in a particular referendum. This calculation of predicted probabilities is based on the logistic regression model of voting behaviour discussed in Chapter 3, and calculated by simulating a change in yes-votes calculated by simulating a half of a standard deviation change in the EU attitude variable, keeping all the other independent variables at their mean. Thereby, we can compare how much attitudes influence the probability of voting yes in different referendum contexts.7 Figure 4.4 shows that the impact of attitudes was greatest in the referen-dums with the most intense campaigns: the Danish Euro referendum, the Norwegian membership referendums, and the French ECT referendum. For example, a half standard deviation increase in EU attitudes in the France referendum (a move from 11 to 13 on a 20-point scale) would increase the probability of voting yes by an amazing 34 per cent, whereas the same change in attitudes in Spain would only increase the probability of voting yes by 5 per cent. The correlation between campaign intensity and the impact of attitudes on vote choice is 0.76. Hence, this figure corroborates the proposition that as campaigns become more intense, citizens will be more likely to draw connections between their own ideological and issue positions and the ballot proposal. While these descriptive statistics do not allow us to reach firm conclusions, the findings in this section do suggest that high-intensity campaigns increase turnout and enhance the impact of attitudes on vote choices. We can explore these relationships further by examining whether we find similar patterns at the individual level. 4.5. Testing the effect of campaign exposure To substantiate the link between campaign environments and patterns of voting behaviour illustrated above, this section examines how individuals respond to variations in campaign exposure and intensity. We expect that citizens in polities with high-intensity referendum campaigns are more likely to vote, more knowledgeable about the ballot issue, and more likely to vote on the basis of issue attitudes. To examine the impact of the campaign environment on individual-level participation and vote choices, we first analyse how exposure to campaign information affects participation. Campaign exposure is a measure of the degree of campaign information that the individual has received in the period leading up to the referendum. Campaign information includes campaign 98 posters, house visits from campaigners, official brochures and leaflets, and television programmes and newspaper articles covering the referendum. Our expectation is that variation in the degree of campaign exposure will have a similar effect to variation in the campaign environment. Second, we analyse the interaction between the campaign context and individual behaviour in a multi-level model that includes both individual-level (micro) and contextual (macro) variables. We expect high-intensity campaigns to induce issue voting. 4.5.1. Turnout in referendums As Downs (1957) noted, the decision to participate in any given election or referendum is affected by the information available to each individual. Information may increase the benefits of voting by helping citizens to make decisions that maximize their utility, but information is also inherently costly. Hence, according to the Downsian model of 'rational abstention', the acquisition of information plays an important role in the decision to vote, since rational citizens will only vote if the returns outweigh the costs. The informational demands of voting led Downs to predict that certain types of citizens are more likely to participate in elections than others. In particular, he argued that high-income (and highly educated) voters would be more likely to vote, because the voting and information demands would be less costly: There are two reasons to suspect that the proportion of low-income citizens who abstain is usually higher than the proportion of high-income citizens who do so. First, the cost of voting is hard for low-income citizens to bear; therefore, even if returns among high- and low-income groups are the same, fewer of the latter vote. Second, the cost of information is harder for low-income citizens to bear; hence more of them are likely to be uncertain because they lack information (Downs 19S7: 273). Downs' theoretical predictions have been corroborated by several studies of voter turnout in North America and Europe. These studies tend to show that better educated, wealthy, and older individuals participate in elections more frequently than others (Lijphart 1997; Wolfinger and Rosenstone 1980; Dee 2003). Though there are some notable exceptions, studies of voter turnout thus tend to emphasize relatively fixed demographic and socio-economic determinants. Yet, as discussed in Chapter 2 (Hypothesis 2.3), it seems equally reasonable to expect that exposure to campaign information will reduce the informational costs associated with 99 Europe in Question Campaign Effects voting and increase the utility of voting, thus making individual citizens less likely to abstain (see also Hillygus 2005). To evaluate whether campaign exposure actually influences turnout in referendums, we are interested in estimating the extent to which campaign exposure has an effect on the probability of voting. It is often argued that campaign research at the individual level suffers from an endogeneity problem, since a person's information acquisition may be endogenous to the outcomes that we are interested in examining, such as participation and knowledge (see, e.g. Lassen 2005). Yet, what is referred to as an endogeneity problem may also be seen as an omitted variable problem (Rosenstone and Hansen 1993; Hillygus 2005). In other words, while it is true that people who intend to vote are also more likely to follow the campaign coverage, this may not be because their vote intention causes them to acquire more campaign information, but rather because of some other factor that influences both, e.g. political interest or education level. Hence, while the problem of endogeneity is certainly a concern, it can be partially alleviated by estimating a multiple regression model and controlling for other factors that are likely to influence both campaign exposure and vote intention, such as education and income levels. Hence, in order to examine the effect of campaign exposure, we estimate a multiple logistic regression model with the turnout decision as the dependent variable. The key independent variable is the individual's level of exposure to the campaign. Campaign exposure is measured by analysing survey responses to question items on campaign information, such as whether the respondent had been exposed to advertisements, leaflets, canvassing, television and radio programmes about the referendum, newspaper articles, or posters.8 On the basis of these question items, a scale of individual campaign exposure was created.9 The control variables include a Likert scale of the individual's education level10 and social class,11 as well as age (see Appendix 2). Given the consensus in the literature, these variables are all expected to have a significant impact on turnout. Finally, we have included controls for region (urban or rural) and gender that may also have an influence on the propensity to vote.12 The results of estimating this model using a pooled dataset are shown in Table 4.1. As Table 4.1 illustrates, our key independent variables are all positive as we would expect, yet only campaign exposure, social class, gender, and age are significant. Older middle class citizens are most likely to turn out in EU referendums, just as we would expect in elections. Being female and living in a rural area is negatively correlated with turnout. Importantly, Table 4.1 also indicates that campaign exposure is positively Table 4.1. Predicting turnout in EU referendums Independent variables Logit Robust S.E. Campaign exposure 0.30" 0.11 Education 0.13 0.12 Social class 0.14* 0.06 Region (urban) -0.24 0.17 Gender (female) -0.24** 0.08 Age 0.21" 0.07 Constant -1.58» 0.43 % correctly predicted 78 McFodden R Squared 0.11 N 30,721 * Significant at 0.05 ** Significant at 0.01 Note: This logit model has been adjusted for clustering within each referendum sample. and significantly related to the likelihood of voting, even when we control for the relevant socio-economic determinants of voting. To get a more intuitive sense of the magnitude of the impact of campaign exposure on turnout, it is useful to present this relationship graphically. Since social class and age are two of the most powerful predictors of turnout, Figure 4.5 shows the predicted probability of voting in an EU referendum for individuals exposed to low, medium, and high levels of campaign information13 given their class and age. As we can see from the predicted probabilities plotted in Figure 4.5, the level of campaign exposure has a very considerable impact on the likelihood of voting. Working class respondents have a probability of 0.39 of turning out when their exposure to campaign information is limited, but this increases to a probability of 0.94 when they receive large amounts of information about the campaign (Figure 4.5a). Similarly, citizens in the 18-24 age group only have a 0.32 probability of voting when they receive little campaign information, but a 0.93 probability of voting when they are exposed to large amounts of information (Figure 4.5b). What is also noteworthy is that, whereas class and age clearly affect the probability of voting for people who are exposed to low and medium levels of campaign information, this effect disappears when individuals are exposed to very high levels of information (as the flatness of the 'high exposure' slope indicates). This is partly a 'ceiling effect' as the probability of voting yes approaches 1, but it could also suggest that campaign exposure tends to even out social and age differentials in the electorate. 100 i 101 Europe in Question Campaign Effects Working class Middle class Upper middle class -£- Low exposure ---&— Medium exposure -----»-— High exposure -®- Low exposure — -4>— Medium exposure -----^— j^jgh exposure Figure 4.5. Campaign exposure and turnout, (a) Social class and campaign exposure; (b) age and campaign exposure We should interpret these findings with some caution, since it is difficult to disentangle the direction of causality in these individual-level observational studies, where it is unclear whether exposure to campaigns leads people to vote, or whether individuals choose to be exposed to the campaign because they have an interest in voting. Notwithstanding the potential problem of endogeneity, these results suggest that exposure to campaign information makes it far more likely that an individual votes in a referendum, even when controlling for education levels. The results are robust across each of the referendums included in the analysis. In Chapter 5, we examine the effect of information tin vote intention in two experimental studies. The experimental design allows us to clearly disentangle cause and effect, and the results from these studies confirm the findings in this chapter: exposure to information makes individuals more certain about their attitudes and more likely to vote. 4.5.2. Ballot-specific knowledge One of the reasons that we expect campaign exposure to increase turnout is because campaigns inform voters about the available choices, thereby increasing the utility of voting. As stressed by Matsusaka (1995) and others, it is the voter's subjective belief about her information level that guides participation. Hence, formal models suggest that voters who feel more confident about their knowledge of the alternatives derive higher utility from voting. If referendum campaigns inform voters in a way that makes them feel more knowledgeable, we expect higher exposure to campaign information to be positively associated with voters' perception of their ballot-specific knowledge. To evaluate this proposition, we can estimate a model where subjective knowledge of the ballot is the dependent variable. The measure of subjective ballot model is a 4-point Likert scale where the respondent is asked to rate her understanding of the issue involved in the referendum (e.g. the Maastricht Treaty) from 'no understanding' to a 'very good understanding'. In addition to campaign exposure, the independent variables include education level, EU attitude scale (as described in Chapter 3), class, gender, and age. This model has been estimated using ordinary least squares,14 and the results are reported in Table 4.2. As expected, campaign exposure is positively and significantly related to perceived knowledge of the ballot. Perhaps surprisingly, the magnitude of this effect is greater than that of education level, which is insignificant in this pooled model. Women tend to report that they know less about the issue than men, as do younger people. Interestingly, attitudes towards the EU have no effect on the perceived knowledge of the ballot. We assume that these information effects have an impact not only on individuals' 102 103 Europe in Question Campaign Effects Table 4.2. Predicting ballot-specific knowledge Independent variables Coeff. Robust S.E, Campaign exposure 0.10" 0.01 EU attitudes 0.01 0.01 Education 0.06 0.07 Social class 0.06 0.04 Gender (female) -0.14' 0.05 Age 0.06' 0.02 Constant 1.47" 0.39 R Squared 0.13 — N 18,826 — * Significant at 0.05 ** Significant at 0.01 Note: This OLS model has been adjusted for clustering within each referendum sample. likelihood to vote, but also on the criteria they employ when they do vote. 4.5.3. Issue voting and campaign exposure As discussed above, we expect high campaign exposure to induce issue voting, because campaigns inform the voter about the available alternatives and increase the salience of the issue. Figure 4.4 illustrated a strong relationship between campaign intensity and issue voting at the aggregate level. We want to examine whether this relationship also exists at the individual level. In other words, are individuals in intense campaign environments more likely to vote on the basis of their attitudes? To examine this question, we estimate a multilevel logistic regression model with respondents at level one and referendums at level two.15 This allows us to examine how campaign intensity interacts with vote choices. Our dependent variable is the binary vote choice (1 = yes) and our key independent variables are EU attitudes, partisan recommendation, and government satisfaction (see descriptive statistics in Appendix 2). As in Chapter 3, EU attitudes is a summated rating scale (21-points) capturing citizens' attitudes towards European integration. Party recommendation measures whether the respondents' preferred party recommended a yes-or a no-vote, and government satisfaction is a 4-point scale ranging from 'very dissatisfied' with the government's performance to 'very satisfied'. We also include controls for education, class, age, and gender. Moreover, 104 Table 4.3. Issue voting and campaign intensity Independent variables togit S.E. EU attitudes 0.33" 0.02 Party recommendation 0.76" 0.03 Government satisfaction 0.62" 0.04 Education 0.05* 0.02 Social class 0.07" 0.02 Age 0.17" 0.02 Gender 0.03 0.04 Campaign intensity -0.10" 0.00 Campaign intensify * EU attitudes 0.02" 0.00 Constant -6.28" 0.45 Variance component 1.12" _ -2* Log likelihood 16716.05 — N (groups, individuals) 17,23,433 — * Significant at 0.05 " Significant at 0.01 Note: This is a random intercept logistic regression model where level 1 is individual and level 2 is referendums. the model includes campaign intensity (as in Figure 4.2) as a level 2 independent variable. To test for the mediating effect of the campaign context on issue voting, we include an interaction between EU attitudes and campaign intensity. The results are shown in Table 4.3. As also shown in Chapter 3, this table indicates that EU attitudes, party endorsements, and satisfaction with government performance are the most powerful predictors of vote choice. In this model, we can also see that campaign intensity has a negative effect on yes-vote. The direction of causality is not clear-cut, however. More intense campaigns may make it more likely that voters reject the ballot, or perhaps conversely, when the race is very close and the ballot is likely to be rejected, campaigns become more intense. As we know, the rejected ballots on the Euro in Denmark, the Constitutional Treaty in France, and accession in Norway were among the most intense, and this was at least partly because public opinion was so divided. More interesting is the positive and significant interaction between campaign intensity and EU attitudes. While the magnitude of the effect is small, it nevertheless suggests that voters rely more on opinions on Europe in intense campaigns than in less-intense campaigns. To get a better sense of the actual magnitude of the direct and conditioning effect of campaign intensity, we plot the predicted effect of attitudes on the probability of voting yes at three levels of campaign intensity. The plot is shown in Figure 4.6. 105 Europe in Question Campaign Effects 100 90 7 8 9 10 11 12 13 14 15 EU attitudes Figure 4.6. Issue voting and campaign intensity Figure 4.6 shows that vote choice overwhelmingly depends on voters' attitudes towards European integration, especially at the extremes, yet that the campaign environment also makes a difference. In more intense campaigns, the probability of voting yes is generally lower. But attitudes also matter more in these campaigns. A shift in attitudes in an intense campaign environment thus has a greater effect on vote choice than in more subdued campaigns. In sum, the evidence presented in this section suggests that the campaign environment is important when it comes to advancing issue voting and turnout. 4.6. Conclusion In response to the rejection of the Constitutional Treaty in France and the Netherlands in 2005, the European Commission launched Plan D for Democracy, Dialogue and Debate (European Commission 2005). At the heart of this proposal was the idea that more debate and dialogue about Europe would lead to greater public support for the project. It may be the case that an intensified debate on Europe could have a positive effect on attitudes. But the investigation of the effect of information on vote choices in European referendums in this book reveals that information effects are more subtle than policy-makers and politicians often assume. Chapter 3 showed that knowledge about the EU does not necessarily guarantee a positive vote in referendums on EU question. Rather, it implies that people are more likely to rely on their own convictions and less likely to follow the recommendations of national elites. This chapter has shown that campaigns with vigorous debate and arguments tend to induce higher levels of popular participation and more sophisticated decision-making, but they do not necessarily result in an affirmative vote. Exposure to campaign information also makes citizens feel more knowledgeable about the issue at stake. Moreover, there is some indication that more intense campaigns foster more issue voting. These findings support the theoretical propositions outlined in the spatial framework of Chapter 2: more information reduces uncertainty and thus increases the likelihood of people voting (Hypothesis 2.3). Higher campaign intensity makes voters more likely to rely on issue preferences (Hypothesis 2.5). Hence, the quantity of information is important when it comes to advancing issue voting and turnout, but it does not guarantee the outcome desired by national and European elites. It was not the lack of debate and information which led to the rejection of the Constitutional Treaty in France in 2005. It is perhaps the nature of the debate that may have affected vote choices as much as the quantity of debate. As Kuklinski et al. (2001: 412) have noted, 'rather than the volume, it is the diagnostic value of information that influences how well citizens are able to cope with policy choices. Information has a high diagnostic value when it clearly and fully conveys the central considerations relevant to a decision or a judgement task'. This chapter has focused on how the quantity of information influences vote choices. In the next chapter, we look more closely at how different types of information and 'frames' generate different responses from people. Using survey experiments we explore how particular types of campaign information—elite framing and elite endorsements—influence opinions on EU ballot questions. This allows us to present a much more detailed analysis of how citizens respond to different types of information and to more clearly distinguish between cause and effect. In Chapter 6, we focus on the 'quality' of information. When we refer to 'quality' in the context of vote choices, we move beyond the empirical investigation of patterns and explanations of political behaviour presented in the previous chapters and we address important normative questions: what is a competent vote choice in the context of EU 106 107 Europe in Question Campaign Effects referendums; how do we measure the 'quality' of voting behaviour; an which factors can influence levels of voter competence? These normative and empirical questions are addressed in the Chapter 6, which examines how different types of information may advance the competence of voters in EU referendums. Notes 1. Key (1966: 7). 2. The measure of polarization is calculated on the basis of the mean party position, weighted by the vote share of the party. The actual measure ranges from 0 (no opposition) to 52 per cent (majority opposition in Parliament). These numbers are transformed into a 5-point scale in the following manner: less than 10 per cent (weighted) opposition gives 1; 10-20 per cent gives 2; 20-30 per cent gives 3; 30-40 per cent gives 4; and more than 40 per cent opposition in parliament gives the score of 5 points. 3. This measure does not account for the level of polarization that can be found outside the parliamentary arena. Non-parliamentary organizations, such as Eurosceptic public movements, have played an important role in many of these referendums, notably in Denmark and Norway. This effect will be discussed in more detail in Chapter 7. The measure of media coverage is likely to capture high levels of extra-parliamentary activity during a campaign. 4. The difference between the yes- and the no-side is transformed into a 5-point scale in the following manner: an average lead of the yes-side of more than 15 per cent gives 1; a lead of more than 10 per cent gives 2; a lead of more than 5 per cent gives 3; a lead between 0 and 5 per cent is transformed to a score of 4, whereas a lead for the no-side (on average over the period) is given the score of 5 (maximum uncertainty for the yes-side). This scale reflects that the government as the agenda setter in these referendums always recommends a yes-vote, and hence any lead of the no-side is likely to enhance the intensity of the campaign. 5. The 5-point scale of media coverage is created as follows: less than half an article a day on average equals a 1-point score; 0.5-1 gives 2; 1-2 gives 3; 2-3 gives 4, and more than 3 gives a 5-point score. 6. The correlation between campaign intensity and turnout is 0.74 if Luxembourg is excluded from the sample. 7. The country-specific estimations are based on the same models shown in Chapter 3 (Tables 3.1-3.3). The same models, but without the interaction terms, are used to calculate the first differences shown in Figure 4.4. 8. In some of the surveys, the respondents were asked to rate the usefulness of these sources of information, but our measure merely recorded whether or not the respondent had been exposed to/was aware of the information sources. 9. The 13-point summated rating scale was based on a minimum of six question items and Cronbach alphas for each of the scales were in the range between 0.75 and 0.95. 10. Education level is measured as a 5-point scale ranging from 'leaving school before 15' to 'university education'. 11. This is a 5-point social class scale ranging from 'unskilled manual worker' to 'professional/manager'. 12. Given the numbeT and type of explanatory variables, one might expect that this model could suffer from multicollinearity, but the diagnostic tests showed that this is not the case for any of the results presented in this section. 13. These three levels of campaign exposure have been created on the basis of the scores on the campaign exposure scale: the quarter of respondents with the lowest scores (i.e. they were exposed to only very limited campaign information) were placed in the 'low' campaign exposure group and the quarter of respondents with the highest score were in the 'high' campaign exposure group, while the remaining half were allocated to the 'medium' campaign exposure group. 14. The dependent variable is an ordinal 4-point Likert scale, but I have chosen to report OLS estimates rather than ordered logistic regression estimates, since the results are very similar in this case and the former are easier to interpret. 15. Hierarchical (or multilevel) modelling is one method which allows us to explicitly model differences in voting behaviour according to the national context. Such models also conect for dependence of observations within a particular referendum (intra-class correlation) and make adjustments to both within and between parameter estimates for the clustered nature of the data (Snijders and Bosker 1999; Raudenbush and Bryk 2002). Since we are not interested in the individual referendums per se, but rather wish to draw general inferences about the effect of individual and contextual variables on voting behaviour, a random effects model would seem more appropriate (see Snijders and Bosker 1999: 43). When estimating the model as a three-level model (respondent, referendum, country), the estimates of the coefficients are virtually identical. We have also estimated the models as simple binary logistic regressions (correcting the standard errors for clustering within countries), and the results are very similar to the multilevel estimates. 108 109