West European Polí lies 11 Routledge Taylor &. Francis Group West European Politics ISSN: 0140-2382 (Print) 1743-9655 (Online) Journal homepage: https://www.tandfonline.com/loi/fwep20 First or Second Order Referendums? Understanding the Votes on the EU Constitutional Treaty in Four EU Member States Andrew Glencross & Alexander Trechsel To cite this article: Andrew Glencross & Alexander Trechsel (2011) First or Second Order Referendums? Understanding the Votes on the EU Constitutional Treaty in Four EU Member States, West European Politics, 34:4, 755-772, DOI: 10.1080/01402382.2011.572390 To link to this article: https://doi.Org/10.1080/01402382.2011.572390 Published online: 24Jun 2011. C2T Submit your article to this journal D? [JH. Article views: 1848 5>1 View related articles ß1 ffil Citing articles: 10 View citing articles 0" Full Terms & Conditions of access and use can be found at https://www.tandfonline.com/action/journal lnformation?journalCode=fwep20 West European Politics, Vol. 34, No. 4, 755-772, July 2011 Routledge Taylor & Francis Croup First or Second Order Referendums? Understanding the Votes on the EU Constitutional Treaty in Four EU Member States ANDREW GLENCROSS and ALEXANDER TRECHSEL This article uses post-referendum Flash-Eurobarometer surveys to analyse empirically voter attitudes towards the EU Constitution in four member states. The theoretical model used incorporates first and second order variables for voting to ascertain whether the outcome of the vote was a reflection of either first or second order voting behaviour. It is hypothesised that the cleavage politics over integration in the European arena had a major impact on the four votes, as captured by three first order variables: Europhile' and 'Constitution-phile' attitudes and Egocentric Europeanness', respectively. The quantitative analyses — controlling for a number of dimensions — strongly supports the hypothesis when compared with a model using solely second order party identification variables. These findings establish that how voters understood the EU polity, in particular whether membership is beneficial to one's own country, was a crucial factor in all the referendums. Implications for future research include the need to discover the cues or proxies influencing first order voting within domestic politics. Introduction The refusal of the French and Dutch electorates to ratify the Treaty Establishing a Constitution for Europe (TEC) was a seismic shock to the European political establishment, sparking an immediate flurry of anxious academic interpretation (Cuperus 2005; Laurent and Sauger 2005; Ricard-Nihoul 2005). Nevertheless, after an official 'period of reflection', the TEC -minus the constitutional rhetoric - remained the blueprint for the eventual signing of the Lisbon Treaty in 2007. The avowed purpose of this reflection period was to take stock of the reasons behind voters' disaffection with the project of European integration. It is in this context of introspection that several studies sought to understand the 'no' votes in France and the Netherlands, whether individually (Aarts and van der Kolk 2006; Brouard Correspondence Address: glencross@abdn.ac.uk; alexander.trechsel@eui.eu ISSN 0140-2382 Print/1743-9655 Online © 2011 Taylor & Francis DOI: 10.1080/01402382.2011.572390 756 A. Glencross and A. Trechsel and Tiberj 2006; Grunberg 2005; Hainsworth 2006; Ivaldi 2006; Perrineau 2005) or together (Dehousse 2006; Stefanova 2006). Some scholars also examined the two cases where the TEC was ratified by referendum in Spain and Luxembourg (Dumont and Poirier 2006; Marcet 2006; de Bruyn 2007). These studies variously describe and examine the factors explaining the respective referendum results and also question the validity of holding such votes in the first place. Yet so far there have only been two cross-national empirical studies of the votes in the four countries that held a referendum on the TEC in 2005 (Svensson 2005; Crum 2007). This paucity of comparative research means little of substance is known about the extent to which these votes reflected common European attitudes towards integration amongst citizens in these four countries or whether purely domestic factors influenced voting behaviour. Indeed, both the existing tendency to conduct single case studies and the preponderant focus on the two negative votes privilege domestic explanations. This article aims to broaden the field of inquiry into the four referendums on the TEC by assessing, somewhat counter-intuitively perhaps, the extent to which they reflect the development of a shared European political arena; in other words, the degree to which voters in the four countries took account of European concerns rather than domestic issues. To do so, the article uses Flash-Eurobarometer survey data from Spain, France, the Netherlands and Luxembourg to analyse empirically voters' decisions at the polls. Hence the article's primary goal is to theorise and measure potential common structural dimensions impacting on voters' decisions in the four TEC referendums. Prima facie, there are excellent reasons for considering that the political context surrounding EU referendums is markedly different from that of European Parliamentary elections, on which the second order voting thesis is traditionally based. This is because EU referendums, especially those over the proposed constitution, mobilise the electorate around a highly visible issue - as indicated by high turnout1 - that is subject to a simple decisionmaking procedure. Moreover, there is a clear understanding of what is at stake: citizens are in effect granted a potential veto on the course of integration. Such a situation contrasts markedly with elections to the European Parliament (EP), which engender second order voting (Reif and Schmitt 1980; Franklin et al. 1994; Marsh 1998). The dominance of domestic factors in EP elections is comprehensively explained by their low visibility, the perceived low significance of the parliament and the complex set of national rules governing the elections. Nonetheless, explanatory models in existing studies of the TEC referendums tend to follow one of two approaches, both of which assume the second order model pertains. One approach focuses on the migration to referendum arena of domestic political cleavages, such as the left/right dimension (Ivaldi 2006; cf. Perrineau 2005) and elite/people divergence (Milner 2006). The other, issue-driven, approach explores the infiltration First or Second Order Referendums? 151 into the referendum campaign of broader issues connected with integration, such as the economic effects of the euro, perceptions about immigration and potential Turkish accession (Aarts and van der Kolk 2006; Stefanova 2006). Hence these two types of study overlook general attitudes towards the EU and, more particularly, popular sentiment towards EU integration per se. To meet the challenge of identifying the possible European dimension evinced in voter preferences, this article uses a theoretical model that incorporates three specific integration-related dimensions: Europhile attitudes (towards EU institutions), Constitution-phile attitudes (towards the TEC) and Egocentric Europeanness (with regard to voters' perception of the benefits accruing from their country's membership of the EU). These variables are deployed to test the hypothesis that the notion of 'Europe' -rather than the usual question of the salience of European integration issues (Franklin 2002; Hobolt 2006b): whether qua polity in its own right or as an institutional order from which member states might benefit - has a major impact on voting behaviour. Our quantitative analyses - controlling for a number of background variables and dimensions related to national politics, strongly support our hypothesis. By identifying the potency of this first order European dimension relating to specific features of the integration project rather than general sentiments about issues associated with Europe, this article adds much-needed precision to the lively intellectual debate between 'attitudinal' and 'second order' interpretations of EU referendums (Hobolt 2006a). The article is organised as follows. Section one introduces the theoretical model. Section two presents the data and the methodological approach deployed for the purpose of the analysis. The third section presents the results of the quantitative analysis. A final section dwells on the implications of these findings, especially in relation to existing work on the politics of direct democracy in the EU. The Theoretical Model: Capturing Explanatory Variables for Voting Behaviour Traditional Explanations of EU Referendum Votes: Domestic vs European Factors The lineage of the dilemma over whether to explain the outcome of referendums on questions of European integration2 as a result of domestic or European factors can be traced back to earlier studies of elections to the European Parliament. In the wake of the first direct elections to the EP, a powerful thesis was proposed arguing that these votes were in fact 'second order elections' (Reif and Schmitt 1980). The theory posits that national issues dominate the campaign agenda as well as voters' preoccupations, meaning that EP elections do not reflect where voters stand on integration issues; supranational issues are thus relegated to second place. The 'second 758 A. Glencross and A. Trechsel order election' thesis continues to provide insights into EP election results (Franklin et al. 1994; Marsh 1998). Unsurprisingly, given its success in explaining EP elections, the second order theory was adopted by scholars in order to explain the results of direct democracy when used in the EU context (Ray 2003; Garry et al. 2005). When applied to direct democracy, the second order model demonstrates that voting behaviour is strongly influenced by national factors such as the levels of approval of incumbent government policy (Franklin et al. 1995) and voter identification with parties holding office (de Vreese 2004). The implication that the instrument of direct democracy, when used to settle questions of integration, fails - just like EP elections - to engender voter mobilisation on supranational issues is also supported by other research on domestic factors. Empirical analyses of referendums on EU matters have revealed the significance of socio-economic status (Gabel and Palmer 1995) as well as voters' perceptions of threats to the nation-state (Christin and Trechsel 2002; McLaren 2002). Notwithstanding this body of evidence for the second order nature of referendums on integration, certain scholars have put forward and tested the competing, first order hypothesis that referendum votes reflect citizens' preferences towards European integration (Siune et al. 1994; Svensson 2002). However, neither theory - second or first order - has comprehensively trumped the other, with both empirical and anecdotal evidence pointing to the simultaneous presence of EU issue voting and second order effects (Garry et al. 2005; Grunberg 2005; Ivaldi 2006). Indeed, these findings suggest that second and first order interpretations of referendums are less mutually exclusive than complementary (Hobolt 2006a). As a result of this mixed evidence, empirical studies have started to broaden the range of causal factors under investigation to understand the circumstances in which 'individual voters are more likely to rely on attitudes rather than second-order and vice versa' (ibid.: 155).This has, notably, meant the incorporation into this research agenda of the notion of campaign effectiveness (Le Due 2002; Garry et al. 2005) and party cues (Ray 2003; Hobolt 2006b; Crum 2007). The more recent focus on party positioning has in turn generated a new theoretical dichotomy concerning whether positioning is top-down or bottom-up (Hooghe 2007). Yet this more recent work still begs the question as to the relative importance of first and second order dimensions when voters participate in a referendum on integration. This paper seeks to provide answers to this fundamental question, which requires a more sophisticated cross-national comparison of the votes on the TEC than is found in the current literature. The Need for Cross-National Comparison One of the principal aims of this study is to overcome the limitations of single-country case studies, which continue to preponderate in this field of First or Second Order Referendums? 759 research (Hobolt 2006a: 154). Our approach is to apply two theoretical models to the pooled data from all four countries that held a referendum on the TEC. To date, statistical analysis of factors influencing voting behaviour in all four cases has been the exception (Svensson 2005), with comparative studies often largely limited to anecdotal evidence (Pfaff 2005; Stefanova 2006; Dehousse 2006). Dehousse (2006: 152) looks at the two referendum rejections, and argues that the Dutch and French votes were clearly the result of second order voting; however, no empirical evidence is adduced to this effect. One complex comparative analysis of the four votes studied the influence of party cues to show that the referendum outcomes were a product of governing parties' ability to mobilise the yes vote in the face of protest voting (Crum 2007). However, in elaborating what is in effect a second order conclusion about the TEC referendums, it did not consider other variables, leaving untested the question of the preponderance of first or second order factors. To remedy these lacunae, the twin first and second order model presented in this article not only compares a comprehensive set of variables, but pools the country data sets to examine causal significance as one single case as well as four separate ones. Thus the results from this comparative study should generate insights into the structural features of voting in EU referendums, thereby providing a measure of the extent to which voting behaviour reflects a concern for supranational, European-level issues. In this way, direct democracy may well tell a very different story about the extent of Europeanisation of national electorates as compared to conclusions derived from studying EP elections. Comparing First and Second Order Effects: The Hypotheses Testing the hypothesis about the prevalence of first order voting factors in referendums on the TEC depends on devising a theoretical model that allows for a direct comparison between first and second order dimensions in them. Two models are used to explain the vote in Spain, France, the Netherlands and Luxembourg. The second order model includes variables for domestic political attitudes as filtered by national parties. These consist of three dimensions: an issue dimension {Party lines on the TEC), a political dimension {Voter identification with government or opposition parties) and an ideological dimension {Voter identification on the left/right spectrum). Conversely, the first order model incorporates variables measuring citizens' unmediated attitudes towards integration - that is, attitudes considered unrelated to domestic party issues. Moreover, the first order variables chosen relate specifically to attitudes towards the construction of an EU polity. Such a twin model makes it possible to compare the explanatory power not only of both models but also allows for a comparative analysis of which variable has the most significant impact across the four countries that held referendums on the TEC. 760 A. Glencross and A. Trechsel The first hypothesis of this study is that the TEC referendums will show evidence of greater first order voting rather than second order effects. HI ('first order primacy'): referendums on EJJ institutional reform are first and foremost determined by attitudes towards EU integration, all other things being equal. A sub-hypothesis is also included regarding causality and first order voting, namely: Hl.l: positive attitudes towards EU integration have positive effects on the vote. Nevertheless, it is still expected that domestic factors will explain some of the referendum outcomes. However, the expectation is that these will be trumped in significance by voters' direct attitudes towards European integration. Thus the second main hypothesis is: H2 ('limited second order'): national party identification patterns add to the explanation of the vote, but this 'second order' model is less important than the first order' model. Similarly, two sub-hypotheses are used to test further second order voting effects: 7/2.7: identification with parties in government is positively related to voting in favour of ratifying the treaty. H2.2: identification with parties favouring the EJJ Constitution is positively related to voting in favour of ratifying the treaty. The Data and Methodology Approach The datasets used for the quantitative analyses are collected from the four 'Flash-Eurobarometer' surveys carried out in the aftermath of the four referendums in 2005. Table 1 presents a summary of the referendum results and the characteristics of the surveys. Unfortunately, the four post-referendum surveys were only partially coordinated by the European Commission's Eurostat office in Brussels. Hence it was necessary to harmonise the datasets by taking into account the relatively few identical questions in all four surveys. Because of the varying answer categories, most of the theoretically derived constructs are dummy variables. While a certain amount of information is lost by proceeding in this fashion, there is a substantial gain in terms of comparability across all cases. The result of this endeavour is a 'clean', 'poolable' dataset containing First or Second Order Referendums? 761 TABLE 1 THE FOUR REFERENDUM OUTCOMES AND THE SURVEY DATA Spain France Netherlands Luxembourg Date of referendum 20 February 2005 29 May 2005 1 June 2005 10 July 2005 Turnout (%) 42.3 69.3 62.8 Yes votes (%) 76.7 45.3 38.4 56.5 Verdict Accepted Rejected Rejected Accepted FlashEurobarometre 168 171 172 173 («)r Field work 21-22 February 31-31 May 2-4 June 2005 11-18 July 2005 2005 2005 N 2014 2015 2000 1001 harmonised data across all four cases. For the purposes of the analysis, only the answers of respondents having attained the voting age (>18 years old) and who declared that they had voted in their respective referendums were selected. The dependent variable for all subsequent analyses was a measurement of the voters' choice at the polls, i.e. their binomial answers to the question 'have you accepted or refused the Constitutional Treaty'. Blank votes were excluded from the analyses. Operationalising the Second Order Dimension As mentioned above, the second order dimension is primarily concerned with national party identification. The survey question was worded: 'To which of the following parties do you feel the closest to or the least furthest from?' Answers to this question were recoded according to the three sub-dimensions that were used for measuring the second order dimension: left-right partisan proximity (dummy), government-opposition (dummy) and pro- or contra-EU Constitution party recommendations (dummy). For every member state, therefore, the analysis provided three party-proximity based dummies. While in the pooled dataset the correlations between these dimensions do not exceed a 0.33 level, the left-right and government-opposition dummies highly correlate in the Spanish and, to a lesser extent, in the French sub-samples. This is a point to which we return in section three. Operationalising the First Order Dimension The main novelty of the analytical approach lies in the operationalisation of the three sub-dimensions to the first order dimension of voting behaviour: Europhile attitudes, Constitution-phile attitudes and Egocentric European-ness. These variables, therefore, reflect attitudes to the institutional construction of the EU, to which a the TEC contributed. The data made 762 A. Glencross and A. Trechsel it possible to measure these variables by using the following survey questions: • Europhile attitudes: 'the institutions of the European Union conjure up a good image to you'; • Constitution-phile attitudes: 'The European Constitution is essential in order to pursue the European Construction'; • Egocentric Europeanness: '[One's country's] Membership of the European Union is a good thing'. Each of these attitudinal variables was coded into dummy variables, with the value 1 indicating a positive response to the question. In neither the pooled dataset nor the individual country datasets did a correlation coefficient between these dimensions exceed 0.35, thereby warranting a high level of confidence in the theoretical and statistical independence of these dimensions. Background Variables The Eurobarometer data provides a series of background socio-economic and demographic variables that have to be controlled for in each model. In previous empirical studies, some of these socio-economic and demographic factors have proved to be of importance in previous instances of direct democracy in the EU (Anderson and Reichert 1996) as well as during the TEC referendums (Perrineau 2005). The six background variables included in the model were: • Age: a continuous variable coding the age of the respondent in years, ranging from 18 to 97; • Gender, dummy variable coded 1 (male) and 2 (female); • Education: measuring the duration of respondents' scholarly education. This variable has four values (0 = no education, 1 = only mandatory education, 2 = elective education level and 3 = higher education); • Economic activity: dummy variable measuring respondents' work activity (1 = working, 2 = non-working); • Rural-urban: dummy variable measuring respondents' living environment (1 = urban, 2 = rural); • EU political activity: dummy measuring the political activity of respondents based on the question of whether they participated in the last European Parliament elections of 2005 or if they abstained (1 = active, 2 = inactive). Method As the dependent variable (vote in the TEC referendums) is dichotomous the method deployed was logistic regressions in all the subsequent First or Second Order Referendums? 763 estimations. For all the analyses, a pooled dataset was used rather than relying on four individual, country-specific datasets because this permitted a more efficient analysis of potential interaction effects.3 The overall variance of the pooled dataset is greater than the individual variances of each country dataset. Results Table 2 contains the first two models for explaining voting behaviour in the 2005 TEC referendums. The first column shows the estimates of a partial model, including all background variables as well as the set of three 'second order' variables based on the party identifications of respondents. The second column represents the full model containing the additional three 'first order' variables measuring voters' attitudes towards Europe.4 An initial glance at the partial model reveals that four of the six background variables are significantly related to the voters' decisions at the polls. The older the respondents, the better educated they are, the more economically and politically active voters are, the higher their probabilities of voting in favour of the TEC at the ballot box. However, gender and living environment do not have a significant impact on voters' political behaviour. The 'second order' variables clearly produce significant results (statistically significant at 0.01) as originally hypothesised (H2.1; H2.2), albeit with the exception of the measure of ideological self-identification. As shown in Table 2, party identification with left- and right-wing parties, especially when coded into a simple dummy variable, correlates very highly with the variable measuring government-opposition party identification. Due to the possible multicollinearity resulting from this constellation of variables, the strength of the government-opposition coefficient may potentially be underestimated. Subsequent analyses for partial models in which the left-right measure is excluded confirm this. However, the increase of the coefficient at stake is minimal and none of the other results are more than cosmetically affected.5 Though not immediately interpretable, the coefficients among the three 'second order' dummies reveal the strong impact of the pro- or contra-EU Constitution party identification variable. The government-opposition dimension, though significant, is clearly less important to voters. Analyses of predicted probabilities for each variable (see below) confirm this observation. To demonstrate the improvement in model fit that comes from adding the three first order variables it is necessary to turn to the full model contained in column 2 of Table 2. Although an improved explanatory fit could be quite reasonably anticipated - after all, three independent variables were added to the model - the significance of this improvement is remarkable: the -2 log likelihood decreases by 30 per cent, with the Nagelkerke R2 more than doubling its value. This provides a first indication that taking into account European attitudinal dimensions relating to the construction of the EU can 764 A. Glencross and A. Trechsel TABLE 2 PARTIAL AND FULL MODEL OF VOTE CHOICE (2005 TEC REFERENDUMS) Variables Partial model Full model Gender 0.113 0.060 (0.070) (0.086) Age 0.008*** 0.008** (0.003) (0.003) Education 0.153*** 0.240*** (0.051) (0.063) Economic activity 0.381*** 0.303*** (0.080) (0.098) Rural-urban -0.001 -0.021 (0.075) (0.092) EU political activity -0.830*** -0.857*** (0.080) (0.099) Pro or contra TEC party identification 1.652*** 1.142*** (0.138) (0.165) Government-opposition party identification 0.902*** 0.832*** (0.079) (0.099) Left-right party identification 0.034 0.109 (0.077) (0.097) Europhile 1.331*** (0.086) Constitution-phile 1.690*** (0.096) Egocentric Europeanness 2.027*** (0.254) Constant — 1 937*** -5.201*** (0.300) (0.444) Valid cases 4238 3877 — 2 Log likelihood 4929.03 3486.69 Nagelkerke R2 0.23 0.48 Note: Standard errors in parentheses. ***p < 0.01; **p < 0.05; *p < 0.1. drastically increase the explanatory power of the model. A closer look at the results shows that all three attitudinal dimensions have a highly significant impact on the outcome variable. Moreover, all these effects follow the hypothesised direction of causality (Hl.l). Among these first order variables, Egocentric Europeanness clearly stands head and shoulders above the others. Evidently, the more a voter is convinced that EU integration so far has had a positive impact on his or her own country, the higher his or her probability of voting in favour of the TEC. To a lesser extent the same is true for those who sympathise with the object of the vote, the TEC itself, and for those who have positive attitudes towards European institutions in general. In a second analytical step, the country-dummy variables were entered in order to control for potential country-specific level differences in the overall model. However, the primary reason for proceeding in this fashion is to measure the stability of the estimates once country effects have been controlled for. The first column of Table 3 presents these results.6 First or Second Order Referendums? 765 What is striking in this attempt to control for country-specific effects is the minimal difference between the model controlling for country effects (first column in Table 3) and the initial model (second column in Table 2). No estimate changes its direction, their significance levels remain the same. Finally, with a few minor exceptions which do not affect the theoretical model, the coefficients remain almost identical. The models for Spain and France are significantly different from that for Luxembourg, but this difference pertains only to the level. In other words, there is good reason to trust the main results described above. Despite these level differences and the overall stability of the results, as long as the hypothesis that European first order variables impact differently on the referendums in the four countries is not tested there is a risk that the model is mis-specified. In order to test for this eventuality, a set of nine interaction effects between the three country dummies and the three attitudinal variables were created and analysed. Incorporating these nine potential interaction effects into the model demonstrates that the notion of a divergent impact of European first order attitudes partially holds true for the Constitution-phile variable (although the interactions of Europhile and Egocentric Europeanness dimensions are not presented in Table 3). However, seven of the nine interaction effects do not significantly improve the model fit. The second column in Table 3 contains the three Constitution-phile/country interaction effects. Interpreting the results contained in column 2 of Table 3, it becomes clear that the earlier estimates remain largely valid. While adding the Constitution-phile based interaction effects increases the strength of the Constitution-phile variable, the model is otherwise not substantially affected. The interaction effects are significant in the Dutch and French case, meaning that the impact of Constitution-phile attitudes on the vote in these countries was significantly lower than in Luxembourg. In other words, voters in the two countries that rejected the TEC did so less because of the Constitution itself but because of a generally pessimistic attitude about the European Union. In this sense, as a purely institutional reform, the TEC perhaps had a greater chance of being successfully ratified by referendum in France and the Netherlands than is often generally acknowledged. The analysis suggests that in these two countries the TEC bore the brunt of voters' frustrations with the wider EU institutional order and its perceived failure to benefit their country. This pessimism was not present in Spain and Luxembourg. However, if the estimates of interaction effects from the Constitution-phile estimate are subtracted, the coefficient remains positive, albeit at a lower level. Furthermore, and equally important, none of the other interaction effects between country dummies and attitudes prove to be of much relevance. Therefore, independently of whether voters are citizens of country x or y, these attitudes do have a significant impact on voters' choice at the polls. What do these results mean for the hypotheses proposed in section one? First, they confirm the three sub-hypotheses Hl.l, H2.1 and H2.2. In other words: 766 A. Glencross and A. Trechsel TABLE 3 FULL MODELS WITH COUNTRY DUMMIES AND INTERACTION EFFECTS Model with country Model with dummies and three country interaction effects Variables dummies (enter method) Gender 0.054 0.053 (0.088) 0.088 Age 0.012*** 0.012*** (0.003) 0.003 Education 0.423*** 0.437*** (0.068) 0.069 Economic activity 0.310*** 0.294*** (0.102) 0.102 Rural-urban 0.066 0.078 (0.101) 0.102 EU political activity -0.515*** -0.524*** (0.104) 0.104 Pro-/contra-TEC party identification 1.612*** 1.605*** (0.185) 0.186 Government-opposition party identification 0.736*** 0.773*** (0.118) 0.122 Left-right party identification 0.098 0.051 (0.114) 0.119 Europhile 1.174*** 1.162*** (0.089) 0.090 Constitution-phile 1.656*** 2.723*** (0.104) 0.251 Egocentric Europeanness 2.181*** 2 j7j*** (0.265) 0.267 Spain 1.550*** 2.078*** (0.190) 0.360 France -0.555*** 0.696** (0.136) 0.292 Netherlands -0.212 0.665*** (0.139) 0.245 Constitution-phile * France -1.610*** 0.323 Constitution-phile*Netherlands -1.287*** 0.291 Constitution-phile*Spain -0.724* 0.411 Constant -6.730*** -7.500*** (0.509) 0.548 Valid cases 3877 3877 — 2 Log likelihood 3294.882 3264.977 Nagelkerke R2 0.524 0.530 Note: Standard errors in parentheses. ***p < 0.01; **p < 0.05; *p < 0.1 • it is true that positive attitudes towards EU integration have positive effects on the vote (Hl.l); • it is also the case that identification with parties in government is positively related to voting in favour of ratifying the treaty (H2.1); First or Second Order Referendums? 161 • furthermore, identification with parties favouring the EU Constitution is positively related to voting in favour of ratification (H2.2) To complete the picture it is also necessary to assess the two main hypotheses: HI and H2. While H2 is partly verified by the results (it is true that second order factors had a significant impact on the vote) it is still an open question whether, as both HI and H2 posit, the first order effects are stronger than the second order ones - and if so by how much. In order to ascertain the veracity of these two hypotheses, the analysis computed the predicted probabilities for each first and second order variable (Table 4). Column 1 contains the predicted probability that a voter will accept the TEC if he or she, for example, identifies with a party that is not in government (value 0 on the government-opposition variable, all other variables held at the sample mean). Moving to column 2 of Table 4, the analysis yields the predicted probability of voting in favour of the TEC if our hypothetical voter had instead identified with a party in government. In this case, the voter's probability went up by 17 per cent, from 55.9 per cent to 73.3 per cent. More importantly, pro- or contra-TEC party identification has a much greater impact on voters. Here, the discrepancy reaches 38 per cent - going from a lower than 50 per cent probability to an almost 70 per cent probability of accepting the TEC at the polls. The left-right dimension, as could be expected from the regression estimates, is of no importance in this model. The results thus confirm that the second order dimension impacted significantly on the TEC vote. When looking more closely at the factors within this dimension 'pro-/contra-TEC party identification' clearly stands out. The government-opposition factor also has a significant impact on voters' choice at the polls, but to a much lesser extent than the pro-con TEC factor. Yet the predicted probabilities analysis reveals that first order considerations are of even greater importance for explaining the TEC referendum outcomes than their second order counterparts. Even the weakest predictor among the three variables, the 'Europhile' attitude, shows a change in the TABLE 4 PREDICTED PROBABILITIES IN % (2005 TEC REFERENDUMS) Dummy at Dummy at Variable value 0 value 1 Change Government-opposition party identification 55.9 73.3 17.4 Left-right party identification 64.3 65.4 1.2 Pro-/contra-TEC party identification 30.8 68.9 38.1 Europhile 48.5 75.1 26.6 Constitution-phile 35.4 74.9 39.5 Egocentric Europeanness 20.0 68.7 48.7 'Second order' 23.0 77.2 54.2 'First order' 3.7 85.3 81.7 768 A. Glencross and A. Trechsel probability of accepting the TEC of 26.6 per cent. Constitution-phile attitudes are not the strongest predictor, though it is even stronger than the strongest factor among the 'second order' group of variables: approving the TEC increases one's probability to vote for the latter by almost 40 per cent. This finding is consistent with the interpretation of the overall model presented in Table 3, especially when taking into account the interaction effects. As already demonstrated, the Constitution-phile variable significantly interacts with the French and Dutch cases, therefore reducing its explanatory character across all cases. The trump explanatory variable in the entire model, therefore, is the Egocentric Europeanness factor. Whether a respondent believes that European integration has had a positive or a negative effect on his or her country changes the probability that this person will accept or refuse, respectively, the TEC by 48.7 per cent. Final confirmation of the greater impact of first order variables as compared with second order ones comes by virtue of conducting a further analysis that groups the three sets of variables in each dimension. Moving from 0 to 1 on all second order items simultaneously increases one's likelihood of voting in favour of the TEC by 54.3 per cent. However, moving from 0 to 1 on all first order items simultaneously increases the probability of accepting the TEC by no less than 81.7 per cent. This truly extraordinary figure ultimately provides highly robust confirmation of the correctness of hypothesis HI. Clearly, these findings demonstrate the first order character of these referendums as measured by three variables covering voter attitudes towards the EU integration project. Indeed, the results of the analysis show that the first order aspect of the vote - voting on European integration without the distorting film of domestic party politics - was the primary dimension in these four votes. Moreover, the impact of such first order European, considerations was not the result of the overweening influence of a single case or two. Rather, the first order dimension was present across Spain, France, the Netherlands and Luxembourg. While EP elections may be above all second order events, EU referendums - even though they are nationally organised and debated largely in the context of a national public sphere - at this stage of integration are above all first order votes on European integration. Conclusions Given the political significance of the research question addressed in this article, it appears only appropriate to place the findings within the broader context of the enduring debate over the democratic legitimacy of EU integration. More specifically, the analysis demonstrated that direct democracy has fostered a high degree of politicisation of integration even if ultimately this turned against the TEC. In contrast with elections to the EP - now well into the third decade of their existence - the referendums on First or Second Order Referendums? 769 the TEC not only engaged public debate and solicited high participation. More fundamentally, these four votes also allowed citizens to express their opinion on the merits and demerits of the constitution as well as on the existing integration process - at least in so far as whether they approved its institutions and whether they considered membership a boon to their country. Hence this research confirms Mair's (2004) suggestion that a Europea-nisation of politics - namely the cleavage over different visions of integration - has already occurred but is poorly reflected in the study of EP elections. Furthermore, these first order findings in popular consultations that both approved and rejected the TEC also suggest that, pace Moravcsik (2006), the constitutive debate over the competences and decision-making rules of the EU is, as Bartolini (2006) also argues, far from foreclosed. Yet perhaps the most important consequence stemming from the first order nature of referendums revealed in this article concerns the use of direct democracy in the EU political system. The findings demonstrate that direct democracy is, in comparison with EP elections, a better method for allowing citizens both to participate in the political construction of Europe and hold their elites - at least minimally - accountable for the integration project these same elites have moulded. From a normative, democratic theory perspective, therefore, referendums are as legitimate as the decision-making outcomes of representative politics. This means the outcome of these votes -as displeasing as they may be - have to be taken seriously regardless of the inevitable second order effects that may lurk therein. Thus, if we intend to take seriously the requirement that integration ought, for the purposes of democratic legitimacy, to be endowed with popular sanction following proper debate about the EU qua polity, direct democracy should not be dismissed as an inadequate tool that excessively polarises debate as a result of second order effects (Dehousse 2006; Moravcsik 2006). This also implies that justifying the shielding of the consensus model of integration from direct democracy because of the infiltration of second order voting is not tenable. By implication, these results also bring to the fore new questions concerning the relationship between direct democracy and European integration. Most notably, given the first order nature of the referendums, it seems highly pertinent to examine the referendum campaigns in order to discover the information and reasoning that explain voters' first order evaluations of integration. Above all, this would seem to require research on campaign effectiveness and cues used by citizens when evaluating issues of integration. Presumably this will involve a return to analysing domestic politics - through qualitative as well as quantitative methods. It seems likely that such research would yield new evidence of Europeanisation, for if referendums show strong first order effects it appears unlikely that this emerges ex nihilo thanks purely to direct democracy. 770 A. Glencross and A. Trechsel Notes 1. In France, the TEC vote saw the largest ever turnout in a French referendum with 28.8 million votes cast (nearly 70 per cent), which compares favourably with the score of 32.8 million votes for the second round of the 2002 election. Shields (2006: 135). 2. Since 1972, there have been 41 referendums on questions of European integration (Hobolt 2006a: 154). For a discussion of the spread of EU-related referendums see Hug (2002). 3. It should also be noted that all models were estimated individually but the overall trends remain very similar to the pooled analyses, in particular when considering the effects of theoretically relevant interactions. 4. We also estimated identical models in which the country samples were weighted to identical «s. The results of the regression analyses were not affected by this test - hence the results are presented using the unweighted dataset. 5. In the partial model excluding the left-right dummy the regression coefficient for government-opposition party identification goes up from 0.902 to 0.917, while the respective standard errors go down from 0.079 to 0.072. 6. Our reference category is Luxembourg. References Aarts, Kees, and Henk van der Kolk (2006). 'Understanding the Dutch "No": The Euro, the East and the Elite', PS: Political Science and Politics, 39:2, 243-6. Anderson, Christopher, and Michael Reichert (1996). 'Economic Benefits and Support for Membership in the European Union: A Cross-National Analysis', Journal of Public Policy, 15:3, 231-49. Bartolini, Stefano (2006). 'Should the Union be "Politicised"? Prospects and Risks', Policy Paper No. 19. Paris: Notre Europe Etudes et Recherches. Brouard, Sylvain, and Vincent Tiberj (2006). 'The French Referendum: The Not So Simple Act of Saying Nay', PS: Political Science and Politics, 39:2, 261-8. Christin, Thomas, and Alexander H. Trechsel (2002). 'Joining the EU? Explaining Public Opinion in Switzerland', European Union Politics, 3:4, 415-43. Crum, Ben (2007). 'Party Stances in the Referendums on the EU Constitution: Causes and Consequences of Competition and Collusion', European Union Politics, 8:1, 61-82. Cuperus, René (2005). 'Why the Dutch Voted No. Anatomy of the New Euro scepticism in "Old" Europe', Prospective Politics, 4:2, 92-101. De Bruyn, Martyn (2007). 'Europe's Antifederalists: The Case Against the Constitution', International Studies Association Annual Convention, Chicago, 28 February-3 March. De Vreese, Claes (2004). 'Why European Citizens Will Reject the European Constitution', Harvard University Minda de Gunzburg Centre for European Studies, Working Paper No. 116. Dehousse, Renaud (2006). 'The Unmaking of a Constitution: Lessons from the European Referenda', Constellations, 13:2, 151-64. Dumont, Patrick, and P. Poirier (2006). 'Le referendum luxembourgeois sur le Traité établissant une Constitution pour l'Europe au Grand-Duché du Luxembourg', in Christian Franck and Sandra Boldrini (eds.), Annales ďétudes européennes de V Universitě catholique de Louvain. Brussels: Bruylant, 213-43. Franklin, Mark (2002). 'Learning from the Danish Case: A Comment on Palle Svensson's Critique of the "Franklin Thesis'", European Journal of Political Research, 41, 751-57. Franklin, Mark, Michael Marsh, and Lauren McLaren (1994). 'Uncorking the Bottle: Popular Opposition to European Unification in the Wake of Maastricht', Journal of Common Market Studies, 32:4, 455-72. First or Second Order Referendums? Ill Franklin, Mark, Cees van der Eijk, and Michael Marsh (1995). 'Referendum Outcomes and Trust in Government: Public Support for Europe in the Wake of Maastricht', West European Politics, 18:3, 101-17. Gabel, Matthew, and Harvey Palmer (1995). 'Understanding Variation in Public Support for European Integration', European Journal of Political Research, 27:1, 3-19. Garry, John, Michael Marsh, and Richard Sinnott (2005). '"Second-Order" versus "Issue-Voting" Effects in EU Referendums: Evidence from the Irish Nice Treaty Referendums', European Union Politics, 6:2, 201-21. Grunberg, Gerard (2005). 'Le referendum francais de ratification du traite constitutionnel europeen du 29 mai 2005', French Politics, Culture and Society, 23:3, 128-44. Hainsworth, Paul (2006). 'France Says No: The 29 may 2005 Referendum on the European Constitution', Parliamentary Affairs, 59:1, 98-117. Hobolt, Sara Binzer (2006a). 'Direct Democracy and European Integration', Journal of European Public Policy, 13:1, 153-66. Hobolt, Sara Binzer (2006b). 'When Europe Matters: The Impact of Political Information on Voting Behaviour in EU Referendums', Journal of Elections, Public Opinion and Parties, 15:1, 85-109. Hooghe, Liesbet (2007). 'What Drives Euroskepticism? Party-Public Cueing, Ideology and Strategic Opportunity', European Union Politics, 8:1, 5-12. Hug, Simon (2002). Voice of Europe: Citizens, Referendums and European Integration. Lanham, MD: Rowman and Littlefield. Ivaldi, Gilles (2006). 'Beyond France's 2005 Referendum on the European Constitutional Treaty: Second-Order Model, Anti-Establishment Attitudes and the End of the Alternative European Utopia', West European Politics, 29:1, 47-69. Laurent, Annie, and Nicholas Sauger (2005). 'Le referendum de ratification du Traite constitutionnel europeen: Comprendre le "non" francais', Les Cahiers du CEVIPOF, 42. Le Due, Lawrence (2002). 'Referendums and Elections: How Do Campaigns Differ?', in David Farrell and Rudiger Schmitt-Beck (eds.), Do Political Campaigns Matter? Campaign Effects in Elections and Referendums. London: Routledge, 45-162. Mair, Peter (2004). 'The Europeanization Dimension', Journal of European Public Policy, 11:2, 337-48. Mair, Peter (2007). 'Political Opposition and the EU', Government and Opposition, 42:1, 1-17. Majone, Giandomenico (2006). 'The Common Sense of European Integration', Journal of European Public Policy, 13:5, 607-26. Marcet, Joan, ed. (2006). La constitution europeenne ä referendum. Espagne et France 2005. Barcelona: Institut de Ciences Politiques I Socials. Marsh, Michael (1998). 'Testing the Second-Order Election Model after Four European Elections', British Journal of Political Science, 28, 591-607. McLaren, Lauren (2002). 'Public Support for the European Union: Cost/Benefit Analysis or Perceived Cultural Threat?', Journal of Politics, 64:2, 551-66. Milner, Henry (2006). '"YES to the Europe I Want; NO to This One": Some Reflections on France's Rejection of the EU Constitution', PS: Political Science and Politics, 39:2, 257-60. Moravcsik, Andrew (2006). 'What Can We Learn from the Collapse of the European Constitutional Project?', Politische Vierteljahresschrift, 47:2, 219-41. Perrineau, Pascal (2005). 'Le referendum francais du 29 mai 2005: L'irresistible nationalisation d'un vote europeen', in Pascal Perrineau (ed.), Le vote europeen, 2004-2005: de Velargissement au referendum franfais. Paris: Presses de Science Po, 229-44. Pfaff, William (2005). 'What's Left of the Union', New York Review of Books, 14 July. Ray, Leonard (2003). 'Reconsidering the Link between Incumbent Support and Pro-EU Opinion', European Union Politics, 4:3, 259-79. Reif, Karlheinz, and Hermann Schmitt (1980). 'Nine Second-Order National Elections - A Conceptual Framework for the Analysis of European Election Results', European Journal of Political Research, 8:1, 3-44. 772 A. Glencross and A. Trechsel Ricard-Nihoul, Gaetane (2005). 'Le "non" fran9ais du 29 mai 2005: comprendre, agir', Notre Europe Etudes et Recherches, no. 44. Shields, J.G. (2006). 'Political Representation in France: A Crisis of Democracy?', Parliamentary Affairs, 59:1, 118-37. Siune, Karen, Palle Svensson, and Ole Tonsgaard (1994). 'The EU: The Danes Said "No" in 1992, but "Yes" in 1993: How and Why?', Electoral Studies, 13:2, 251-5. Stefanova, Boyka (2006). 'The "No" Vote in the French and Dutch Referenda on the EU Constitution: A Spillover of Consequences of the Wider Europe', PS: Political Science and Politics, 39:2, 251-5. Svensson, Palle (2002). 'Five Danish Referendums on the European Community and European Union: A Critical Assessment of the Franklin Thesis', European Journal of Political Research, 41:6, 733-50. Svensson, Palle (2005). 'The European Referendum Experience in Western Europe', ECPR General Conference, Budapest, 8-10 September.