Voter Disenchantment in the Aftermath of the 2005 EU Constitutional Referendum in France

In a popular referendum in 2005, French voters rejected their country’s adoption of a proposed EU Constitution. Yet, in seeming defiance of the popular vote, the government subsequently proceeded to implement the core of the legislation without consulting the public again. This article empirically examines the electoral impacts of these events. We build a comprehensive fine‐grained dataset of nationwide election results for more than 36,000 metropolitan French municipali‐ ties. Employing cross‐sectional analysis for all national elections held in the decade after the referendum vote, we find that the strength of a municipality’s rejection of the EU Constitution in 2005 is associated with a lower voter turnout, higher shares of blank votes, and larger gains for anti‐system parties in subsequent elections. The findings are robust to various modelling choices and the inclusion of a large array of controls. The results indicate that bypassing a popular vote could entail protracted adverse effects on the quality of democratic participation and deliberation.


Introduction
On 29 May 2005, France voted on the adoption of the Treaty establishing a Constitution for Europe (TCE) in a popular referendum.A surprisingly clear majority of 55% of voters rejected the Constitution.Yet, despite the popular rejection, the government continued working on implementing the international agreement.In 2008, the revised version of the TCE, the Lisbon Treaty, was ratified without consulting French voters anew.
The purpose of this article is to examine the electoral impact of this bypassing of the popular vote on subsequent national elections.To do so, we compile a municipality-level dataset, containing more than 36,000 French communes, linking the voting outcomes of the 2005 TCE referendum to two cycles of European Parliament (EP), legislative, and presidential elections throughout the decade following the referendum vote.Applying cross-sectional analysis, we evaluate the impact of the bypassing of the referendum on three expressions of voter disillusion with the political process: turnout (passive disengagement), blank votes (active disengagement), and vote shares of anti-system parties on the political fringes.
Ceteris paribus, communes with higher no-vote shares at the referendum in 2005 registered a significantly lower voter turnout and higher shares of blank votes in subsequent elections (with the size of effects gradually waning out over time), while vote shares of fringe parties increased.Results are robust across various model specifications.
The findings uncover new empirical evidence that the bypassing of popular votes is detrimental to the quality of democratic deliberations.While our empirical analysis delves into one particular case, the findings are of broader relevance.For one, the referendum was an early manifestation of the strengthening political cleavages around globalization and international integration that continue to reshape electoral politics in early 21st-century liberal democracies (Bornschier, 2008;Ivaldi, 2018).Additionally, the bypassing of the French referendum vote represents a paradigmatic example of a more regular feature of contemporary democratic politics, emphasized by accounts of party cartelization (Katz & Mair, 2009), in which governing apparatuses (are perceived to) sidestep popular demands to push through a policy agenda favoured by political elites.Our analysis illustrates the potential electoral implications of such behaviour in fine-grained empirical detail.In this sense, the article speaks to deeper drivers of the causes and modes of EU disintegration that pre-date the Brexit vote.At the same time, looking forward, popular referendums on questions regarding international integration are becoming increasingly common, as the emerging literature on the new "mass politics of international disintegration" has highlighted (Walter, 2020).Thus, it is likely that more future governments will be facing similar dilemmas as the one confronted by French governing elites in the 2000s, forced to make a choice between respecting international obligations or the popular will.Revisiting the case can help us better understand what is at stake under such circumstances.

Voter Alienation and the Rise of Anti-System Politics
Established party systems in rich democracies on both sides of the Atlantic have experienced dramatic disruptions over the past decade.Turnout has fallen to record lows (Mair, 2013) and electoral support for centrist mainstream parties has declined rapidly (Trubowitz & Burgoon, 2022).Anti-system protest parties have thrived in many countries (Hopkin, 2020).In response to these developments, a rapidly growing field of literature in political science and economics has set out to examine the roots of this apparent electoral backlash against the political mainstream.Much of that literature has debated the deeper cleavages underlying the politicization of globalization and the extent to which these voter sentiments were driven by cultural or economic factors (Kriesi et al., 2006;Walter, 2021).
In particular, theories of party cartelization (Hopkin, 2020;Katz & Mair, 1995, 2009) have suggested that, in addition to economic grievances and cultural anxieties, the backlash against international integration is also fuelled by voters' growing disenchantment with mainstream parties and a loss in trust towards political elites.Patterns in the organization of political parties in advanced democracies emerging in the 1980s and 1990s, and firmly established in the 2000s, considerably widened the gap between large parts of the population and the political elite.Changes in party financing (e.g., reduced reliance on member contributions) and an associated professionalization of political elites (e.g., the rise of "career politicians" and political managerialism), coupled with an emerging consensus about the primacy of free markets and the desirability of deep international economic integration, gradually alienated mainstream parties from their voter bases (Blyth & Katz, 2005;Scarrow, 2006).Mainstream parties' discourses increasingly aligned with the views of socio-economic insiders (i.e., the highly educated in urban areas with relatively secure professions), which left little room for meaningful ideological competition (Blyth & Katz, 2005;Jacobs et al., 2021;Keman, 2014;Linsi et al., 2022).As a result, large segments of society, and in particular underprivileged groups, felt less and less represented in national politics and gradually lost trust in political institutions, generating political apathy and laying the ground for the rise of anti-system parties keen to exploit existing grievances (Hopkin, 2020).
Work on Euroscepticism has investigated related patterns at the EU level.Having emerged as an elite project, popular support for the European project has gradually grown more important for the legitimacy of the EU institutions over time (Hobolt & de Vries, 2016).But at the same time, in an attempt to deflect blame for the lack of more meaningful ideological competition, national politicians increasingly turned to paint the EU as a restrictive external institution that limits the available scope of national policymaking options (Turnbull-Dugarte, 2020).Deepening elite-mass divides and the depiction of the EU as a restricting force has turned the European project into an increasingly politicized phenomenon, a process Hooghe and Marks (2009) seminally described as a gradual shift from "permissive consensus" to "constraining dissensus.''Finally, a more recent stream of literature on the mass politics of international disintegration has drawn attention to popular referendums on international agreements as manifestations in which elite and popular interests increasingly frequently collide, generating difficult tensions for the multilateral international order built on the prefaces of liberalism (which seeks to combine interstate collaboration with respect of democracy at home).
Calls for popular referendums on international agreements-from Brexit or Switzerland's participation in the Schengen agreements to the Paris climate accords-have grown increasingly frequent over the past years (Malet, 2022;Walter, 2020).If they pit the people's vote against a country's international obligations, such referendums can put governments in a difficult bind.Breaching the latter can complicate a country's foreign policy.But bypassing the former could undermine the legitimacy of national institutions and, in the worst case, damage the domestic democratic process itself.
Against this background, this article proceeds to examine in depth the electoral consequences of one early controversial popular vote on international integration, the 2005 EU constitutional referendum in France.

Economic and Political Background
The politicization of globalization and the cleavages underlying them started to become salient in France earlier than in many other Western democracies.Having vocally resisted the full embrace of economic internationalization and visions of a US-led liberal order during its Gaullist period, the government irrevocably shifted towards market-based strategies in the 1980s.The liberalization strategy gradually transformed the economy and society by curtailing the government's power to intervene in markets and provide sustenance to its citizens, while expanding the free flow of people, capital, and goods (Schmidt, 1996).The imposition of these liberalizing reforms, adopted by nominally left-wing President François Mitterrand, was accompanied by structural changes in the political landscape.Political competition started to move away from traditional left-right, socio-economic, and religious lines towards divisions focused on national sovereignty, culture, and values.In particular, European integration became one of the most politically salient issues and contributed to an increasingly fragmented party system (Bornschier, 2008;Grunberg & Schweisguth, 2003).While established centrist parties converged in their approval of an everdeeper integration into the European and international free-market framework, fringe parties became successful by lumping the system together and offering a contrasting alternative (something that became apparent already in the presidential elections of 2002, in which a far-right candidate reached the second turn).It was in the context of this shifting political landscape that the referendum was held in 2005 (Bornschier, 2008).

Background of the Treaty Establishing a Constitution for Europe
After several years of consultation, the TCE was signed by representatives of all the 25 then-members of the EU, on 29 October 2004.The goal of the treaty was to integrate the various EU treaties into one single text and to somewhat strengthen political integration through some modifications of existing rules.After its signing, the treaty had to move through national ratification procedures in all member countries.Whereas 15 member states opted for parliamentary ratification, 10-including France-decided to hold a popular referendum on the adoption of the TCE (Crum, 2007).

The Position of French Parties Before the Vote
The French decision to hold a referendum was taken by then-President Jacques Chirac.On the one hand, since France had a history of putting important matters before a popular vote (Morel, 2007), Chirac faced outside pressure to act democratically.On the other hand, Chirac reportedly was confident that the treaty would easily pass the referendum, boost his image as a capable statesman, and weaken the opposition Socialist Party (PS), which was internally divided over the issue (Hobolt, 2006;Shields, 2006).Early polls indicated a comfortable margin of up to 20% in favour of the TCE.Chirac's own party Union pour un Mouvement Populaire (UMP) was clearly in favour of ratification.The main opposition party, the center-left PS, was split: While most of the leadership advocated supporting the treaty, a large part of the base opposed it, perceiving it as promoting a neoliberal version of the EU (Crespy, 2008;Wagner, 2008).An internal vote finally led the PS to officially endorse the yes campaign, but the margins were close.The parties on the far left (most notably the Lutte Ouvrière, but also the larger and better-known French Communist Party) and far right (most notably the Front National) opposed ratification of the TCE, which the left portrayed as an expression of (Anglo-Saxon) free market capitalism while the right framed the treaty as a threat to national sovereignty and the cultural heritage of France (Richard & Pabst, 2005).

The Referendum Vote in France
The first popular referendum on the TCE was held on 20 February 2005 in Spain.A large majority of more than 80% of Spanish voters approved ratification.On 29 May 2005, the second popular referendum was held in France.A clear majority of 55% of French voters ended up rejecting the treaty-voter turnout was around 70%, the largest ever for a referendum in modern France (Shields, 2006).The French rejection, together with a similarly clear no-vote in the Netherlands shortly after, derailed the ratification process and the constitutional project was postponed (Crum, 2007;Malet, 2022).
Why did French voters reject the referendum?Political analysts have highlighted both attitudinal and second-order forces (Brouard & Tiberj, 2006;Startin & Krouwel, 2013): the combination of different kinds of Euroscepticisms on the left (socio-economic) and right (cultural; van Elsas, 2017) and the referendum as an opportunity for voters to express discontentment with President Jacques Chirac, or with political elites more generally (Shields, 2006).Exit polls confirm the role of these factors and the salience of socio-economic concerns (see Startin & Krouwel, 2013, p. 72).

The Bypassing of the Referendum Vote
Although the popular rejection of the TCE in France severely interrupted the constitutional project, it did not ultimately halt it.After a "period of reflection," during which the French government made little attempts to address the referendum outcome (De Beer, 2006), the Berlin Declaration signed on 25 March 2007 set out a roadmap for the elaboration of a new version of the TCE.The new Lisbon Treaty simplified the structure of the TCE and omitted some of the dispositions about symbolical issues that proved particularly controversial (e.g., the EU flag and anthem), but in substance took over most of the content that had been rejected by French voters in 2005 (Cleppe, 2009).
The revival of the constitutional project had been boosted by the French presidential elections in May 2007, which were won by Nicolas Sarkozy who, in contrast to his challenger Ségolène Royal, had painted the eventual acceptance of a new constitutional treaty as inevitable.Despite strong protests by far left and far right parties as well as trade unions, which called on the government to respect the people's will, Sarkozy rushed the ratification of the Lisbon Treaty through parliament.While some abstained in protest, a majority of members of the PS ended up backing Sarkozy's strategy.On 7 February 2008, the Lower House and French Senate authorized ratification, paving the way to make France the second country (after Hungary) to formally ratify the Lisbon Treaty.
This bypassing of the popular referendum vote has been described as an important moment in the observed growth in voter apathy and Euroscepticism in France.Liberal activists decried politicians' concerted negligence of the popular vote as a betrayal of democratic principles (Cleppe, 2009) which exposed underlying democratic deficits (De Beer, 2006).Jean-Luc Mélenchon, one of today's most popular leftist politicians in France, accused elites "to despise" the people ("Le 'non' de gauche," 2005).The Communist Party leader Marie-Georges Buffet remarked that "the referendum marked a break.People voted, and there has been a refusal to accept the decision of the people.Disillusionment settled in" ("When France," 2016).Or, as the president of a progressive pro-EU movement (Sauvons l'Europe) phrased it: "It's not surprising that people don't care anymore" ("When France," 2016).

Research Design
In our empirical analysis, we test claims of voter frustration econometrically.Zooming in on French municipalities (communes), we evaluate the relationship between the strength of rejection of the TCE in a commune and its political behaviour in the aftermath of the referendum vote and its bypassing, focusing on three outcome variables.The first outcome variable is the turnout at elections as an indicator of the degree of political participation and civic engagement (Mair, 2013).The second is the number of blank or invalid votes cast at elections as a proxy of protest voting and expression of frustration with the political system (Driscoll & Nelson, 2014;Uggla, 2008).The third is the vote shares of mainstream vis-à-vis anti-system parties on the far right or left (Hopkin, 2020;Mair, 2013;Trubowitz & Burgoon, 2022).
As it became increasingly clear that the government would bypass the referendum outcome, our prior expec-tation is that political participation would decrease and support for anti-system parties would increase in municipalities where the referendum was rejected by a larger margin.Thus, we expect a negative relationship between the municipality-level no-vote share at the referendum in 2005 and turnout levels in subsequent national elections, a positive relationship with the number of blank votes, a negative relationship with vote shares of traditional centrist parties (PS and UMP), and a positive one with anti-system parties on the fringes.Since party vote shares can be strongly contingent on idiosyncratic election-specific factors (such as candidates' popularity or evolving party coalitions), and because France's political party landscape is known to be particularly volatile (Bornschier, 2008), we expect these last relationships to be less clear-cut than those for turnout and blank votes.Furthermore, in line with previous literature on voter "memory" (Bechtel & Hainmueller, 2011), we expect effects to be the strongest in elections taking place during, and soon after, the bypassing process and fading out after a few years.
Having no detailed and comprehensive data on individual voters' opinions, our unit of analysis is the municipality, and we focus on variation in electoral politics between municipalities.Although it is individuals, not municipalities as such, who vote, a substantial body of evidence indicates that local environments and cultural milieus influence the vote choices of individual voters (Ansolabehere et al., 2014;Harteveld et al., 2021;Rodríguez-Pose, 2018;Walsh, 2012).According to this literature, local conditions and interactions can serve as filters of perception that play a role in shaping political attitudes and voting behaviour.David et al. (2018) argue that the level of municipalities is optimal in identifying such effects statistically.Applied to our case, we assume that individual voters take cues from the (municipalitylevel) communities they are living in.

Data
Using election results published by the French Ministry of Interior and population census data from the French National Statistical Office (Insee), for our main analysis we build a municipality-level dataset containing the two EP elections in 2009 and 2014, and the two first rounds of the presidential and the legislative elections held in 2007 and 2012.Municipality-level outcomes from these six national elections are linked to the local outcomes of the 2005 referendum.
We limit our sample to mainland municipalities and to municipalities that are available throughout the sample period (including the pre-referendum years 2002 and 2004).Elections in Corsica and the overseas territories tend to follow somewhat different dynamics from the mainland and are more strongly influenced by local issues, such as the independence movement in Corsica or the legacy of colonialism in the overseas territories (De la Calle & Fazi, 2010;Hintjens et al., 1994).
Additionally, for many overseas departments, election and referendum results are not reported at the municipality level.Therefore, we exclude Corsica and overseas territories from our main analyses (additional analyses available in the Supplementary File show that results are not sensitive to the inclusion of the available data from Corsica and overseas territories).
To control for various socio-economic factors in our regression analyses, we match the data on election and referendum outcomes with the respective annual municipality-level information provided by Insee.This data-collection effort yields a database with complete information on about 33,000 French communes (out of a total of around 36,200 mainland communities in 2005, according to Insee) over the referendum vote and two subsequent national election cycles.

Variables
Our main dependent variables are voter turnout and blank votes as indicators of voter (dis)engagement.In further analyses, we also examine the impact of the refer-endum vote on party vote shares.Our primary independent variable is the municipality-level no-vote share at the constitutional referendum of 2005, which is depicted geographically in Figure 1.All electoral information is from the French Ministry of the Interior.
To control for relevant political, socio-economic, and demographic characteristics of communes we include a large array of relevant municipality control variables which we source from Insee (for a full list, see the notes beneath Table 1.We provide detailed descriptive statistics in the Supplementary File).

Empirical Strategy
We estimate our main results with standard linear regression analysis.In additional analyses, we check the robustness of our findings in a beta regression model.Our modelling choices are justified in greater detail in the Supplementary File.
To minimize the threat that unobserved variables are driving our results, we (a) include a large array of socioeconomic, demographic, and political control variables; (b) conduct Oster tests (OT; Oster, 2019) which evaluate treatment selection on unobservable variables and assess coefficient stability (test results above the value of one indicate that omitted variables would need to be at least as relevant as all included variables together to nullify the effect of the tested coefficient); (c) check that there is no effect on most pre-referendum outcomes, and (d) check our results with an alternative estimation technique (all tests provided in the Supplementary File).
We estimate cross-sectional regression analyses covering presidential, legislative, and EP elections from 2007 to 2014.For a randomly drawn commune i at election year t, we specify our full model as follows: The dependent variable y is either turnout, blank vote share, or vote shares of specific parties.The explanatory variable of interest is the intensity of the municipal rejection of the TCE, measured by the percentage share of no-votes among the valid votes in the referendum in France in 2005.The control matrix X represents a battery of political, economic, and sociodemographic municipality controls.To account for political cleavages around globalization and international integration that predate the referendum vote, and which might have also mattered for the referendum and subsequent elections, we include the variable y i, pre−ref which represents the respective outcome of the most recent election of the same kind (i.e., presidential, legislative, or European election) held before the referendum in 2005.The inclusion of this "lagged" outcome variable ensures that the coefficient of the no-vote variable measures the referendum's effect on current differences in turnout or party support, all else being equal (including distinct outcome levels in the election predating the referendum).Furthermore, it diminishes the threat of omitted variable bias since many unobserved factors that affect both municipal election outcomes after the referendum and its referendum no-vote are likely to be absorbed by previous election outcomes (Wooldridge, 2013).
To facilitate interpretation, all continuous variables are standardized.Finally, the regional fixed effects  i capture broad unobserved characteristics of 21 NUTS2 regions.We cluster our standard error over 95 departments to allow for potential correlation in the error term of municipalities located in the same department.

Main Results
We first show how the referendum vote affected electoral participation in the form of voter turnout and blank votes in subsequent elections.Then we turn to investigate the effects on vote shares of principal mainstream and anti-system parties.Our presentation prioritizes the outcomes of EP elections, which are of lower salience and thus arguably less influenced by idiosyncratic factors (such as strategic voting, individual candidates, changes in party landscapes, or particular local issues) than either presidential or legislative elections, which can make them more valid to assess broader underlying trends and revealed voter sentiments (Van der Brug et al., 2007, p. 5;Van der Eijk et al., 1996).Moreover, and specifically regarding party vote shares, the European elections might also be the type of elections most likely to capture rejection of further European integration.This being said, we investigate all types of national elections, namely French legislative, presidential, and European Parliament elections.Findings that are consistent across the different types of elections are particularly strong because they are robust to election-type-related idiosyncrasies.

Electoral Participation
Table 1 presents the regression results of the no-vote on the EP election turnout in 2009 and 2014.The first three columns show the impact of the no-vote on municipal turnout rates in 2009, gradually adding control variables to the model until arriving at the above-defined full model in column 3. Columns 4-6 repeat this exercise for the election in 2014.
Throughout all model specifications in both elections, the no-vote is consistently negatively associated with the municipal turnout.The first specification includes past political variables (namely the municipality turnout in the EP election in 2004 and participation in the referendum in 2005), as well as fixed effects for the 21 French NUTS2 regions.In the second column, municipality-level control variables are added to the equation.Finally, in column 3 municipality-level median income is included, restricting our sample size by excluding very small municipalities due to data availability.
Looking at the fully specified model results in columns 3 and 6, the control variables attenuate the impact of the no-vote, which indicates that disregarding socio-economic characteristics entails an upward bias in the coefficient of the no-votes.OT deltas exceed unity, making it highly unlikely that the correlation between the no-vote and the EP election turnout in 2009 and 2014 is spurious.A one standard deviation higher no-voteabout 10.1 percentage points (pp; see Supplementary File, Table A2)-is associated with a lower turnout of around 0.61 pp in 2009 and with 0.25 pp in 2014, ceteris paribus.To put the magnitude of these coefficients into perspective, in 2009 a one standard deviation higher unemployment rate (18.9 pp) decreases turnout by 0.17 pp (0.15 pp in 2014), and in a municipality with a one standard deviation higher share of university graduates (31.6 pp) turnout increases by 0.81 pp (0.80 in 2014), ceteris paribus (full regression results of every equation are available upon request).The effect of the no-vote for turnout is hence substantive, at least for the EP elections in 2009.
In the 2014 EP election, the impact is smaller, in line with our expectation of the referendum vote's waning importance over time, but its impact is still larger than Notes: Election controls include turnout in the previous EP election and turnout for the referendum of 2005; municipality controls contain (a) the size of the eligible voter base, (b) the population density in four categories (from very densely to very sparsely populated), (c) the share of the population with 60 years and older, (d) the share of residences that lay vacant, (e) the average household size, (f) the share of relatively lowly educated graduates (highest degree vocational studies or aptitude certificate), (g) the share of the population with a university degree, the share of workers engaged in (h) blue-collar labor, (i) agriculture, (j) education and science, and (k) artisans (such as craftsmen, tradesmen, and small business owners), (l) the population's immigration share, and (m) the unemployment rate; the median income is listed separately as it excludes municipalities with less than 50 households from the sample; all independent variables are standardized; also, included in all regressions are regional fixed effects; robust standard errors clustered over 94 departments are in parentheses; * p < 0.10; ** p < 0.05; *** p < 0.01.
a one standard deviation difference in unemployment.
In additional analyses, we further extend the timeframe to the EP elections held in 2019, where the association between the municipality-level no-vote and turnout disappears.This suggests that the referendum vote by itself ceases to be an important factor for electoral outcomes in France beyond the first decade after the referendum was held.This is in line with the literature on electoral myopia which argues that voter memory on policy rarely extends two election cycles (Bechtel & Hainmueller, 2011).Another consideration is that important changes over that period in the French political landscape, where former chief actors in the referendum process, such as the PS and UMP, had lost much of their electoral relevance by 2019.Estimation results on legislative and presidential elections corroborate this picture.Figure 2 graphically illustrates the results of applying the full model on all three election types (underlying regression tables available in the Supplementary File).
There is a significant and substantive effect for all immediate post-referendum elections, which is strongest for EP elections (which is the "most likely case" for the reasons previously discussed), followed by legislative, and presidential elections.Overall, the results suggest that the more strongly a municipality voted against the TCE, the more people abstained from voting in major national elections following the referendum's bypassing.At the same time, the participation-depressing effects appear to have largely dissipated in the second election cycle: There is no longer any statistically significant association in the legislative and presidential elections of 2012 and the size of the coefficient for the EP election in 2014 more than halves (and finally dissipates for the election in 2019).

Blank Votes
The share of blank votes is our second measure of voter disengagement.To the extent that casting a blank vote requires a greater effort than not voting, it can be considered a more active form of expressing voters' discontent with the political system (Driscoll & Nelson, 2014;Uggla, 2008).
The results for blank votes are presented in Figure 3.The findings mirror the turnout results: All else equal, the share of blank votes was significantly higher in municipalities that had a higher no-vote share immediately following the referendum's bypassing.As with turnout, municipality differences become null for the legislative and presidential elections in 2012 but remain statistically significant for the EP election in 2014 (dissipating again in 2019).
The results presented so far strongly suggest that the bypassing of the referendum outcome further fuelled voter disengagement in municipalities that had rejected the TCE.

Party Vote Shares
To evaluate the impact of the referendum's bypassing on municipality-level party support, we focus on the combined vote share of the two centrist mainstream parties at the time of the referendum, PS and UMP, and anti-system forces on the fringes, the Front National on the far right, and a group of parties associated to the far left (see the Supplementary File for a detailed listing).are from the fully specified model with all covariates included (as in columns 3 and 6 in Table 1); dots represent the estimated coefficients; whiskers show the 95% confidence interval around them.Results are from the fully specified model with all covariates included (as in columns 3 and 6 in Table 1); dots represent the estimated coefficients; whiskers show the 95% confidence interval around them.
Looking at party vote shares at all six elections (Figure 4), we observe some volatilities, as is to be expected given the different types of elections and the evolution of the French party system over the sample period.Yet, the general pattern is in line with our expectations.While the municipal vote share for anti-system parties correlates positively and strongly with the referendum no-vote, the established centrist mainstream parties lost support in municipalities that had more strongly rejected the TCE.The decline for the traditional mainstream parties is driven by vote share losses of the UMP, while both the far left and the far right experience gains in the aftermath of the referendum.
An important nuance of the party vote share analyses is that results are getting stronger over time (also beyond our sample period), which stands in contrast to turnout and blank voting results.We interpret this as an indication that in the aftermath of the referendum and its bypassing populist anti-system parties were successful in mobilizing apathetic, disaffected voters, particularly with eurosceptic and anti-establishment platforms (Ivaldi, 2018;Krouwel & Abts, 2007).At the same time, it is also plausible that over time factors other than the referendum itself started to play a greater role in the mobilization of disaffected voters by anti-system parties.In that sense, the referendum may have fed into-and further amplified-pre-existing cleavages.
In sum, our analyses uncover evidence for all three types of voter disengagement in response to the bypassing of the referendum.Municipalities that had more strongly rejected the referendum experienced lower turnout, more blank votes, and higher vote shares for anti-system parties in subsequent elections.Underlying regression tables for these results are provided in the Supplementary File.

Survey Data Analysis
In a final step, we assess whether the referendum and its bypassing can be linked to changes in individual attitudes.We do so by looking at trends in satisfaction with French and European democracy from 2000 to 2015 as expressed by French respondents in Eurobarometer surveys.
Given the central role played by the French government in the bypassing of the referendum, we would expect that French voters were more pessimistic about the democratic functioning of their state, especially so in regions that voted more strongly against the TCE.At the same time, to the extent that the electorate perceived the EU as an accomplice in, or even reason for, ignoring the popular demands voiced by the no-vote, there might have been a backlash against the EU as well.
The two Eurobarometer survey questions that we investigate in more depth have a scale from one to four and are: On the whole, are you very satisfied (4), fairly satisfied (3), not very satisfied (2), or not at all satisfied (1) with the way democracy works in France?And how about the way democracy works in the EU?
We split the sample into three groups according to the intensity of the no-vote of the respondents' residence (the Eurobarometer provides information only on the NUTS2 level) and perform event studies for each group.By regressing the Eurobarometer answers on results are from the fully specified model with all covariates included (as in columns 3 and 6 in Table 1); dots represent the estimated coefficients; whiskers show the 95% confidence interval around them.
regional and yearly fixed effects, these analyses show how satisfaction with democracy, either with respect to France or the EU, developed within these groups from 2000 to 2015.
Figure 5 depicts the estimated coefficients.Dots represent point estimates, i.e., relative differences to the respective group's satisfaction in the base year 2000, and the whiskers indicate the statistical significance of these differences within 95% confidence intervals.We highlight the referendum in May 2005 and the official ratification of the Lisbon Treaty by the French Parliament in February 2008 with dashed vertical grey lines.We interpret divergences between the three groups (accepted, rejected, and strongly rejected TCE) after the referendum and its bypassing as an indication that the regional no-vote mattered for individual sentiments towards the EU.
While results are necessarily cruder due to the higher level of geographical aggregation (NUTS2 level instead of municipalities), we observe some interesting patterns.The panel shows that satisfaction with French democracy is relatively stable initially-with the average for each group lying at around 2.6, which falls between not very satisfied (2) and fairly satisfied (3).However, starting with the referendum year 2005, average assessment drops and recovers to base levels only during short "honeymoon period" stints in the election years of 2007 and 2012.For every other year, the average assessment is relatively more pessimistic across all regional groups.This being said, there is a salient regional disparity in the years immediately following the referendum.In 2005 and 2006, years in which the media covered the referendum and its bypassing extensively, satisfaction with French democracy hit a low, especially so in regions which voted against the TCE: In 2006, the average satisfaction in regions that rejected the constitution dropped by 0.4 points, which was twice the drop in regions that accepted the TCE.
The bottom panel shows the same trend but concerning democracy in the EU.Dissatisfaction with the EU seems more muted and fewer obvious trends are discernible.In this sense, the analyses suggest that the French electorate does not seem to have blamed the bypassing of the referendum on European institutions in particular.The backlash from the bypassing appears to have more strongly affected views on domestic democratic processes.

Conclusions
Our analyses leverage the bypassing of the popular referendum on the TCE in France as a case study to empirically examine the electoral consequences of technocratic political engineering and the cartelization of mainstream political parties more broadly.Our analyses uncover evidence that turnout decreased, the share of blank votes increased, and support for anti-system fringe parties grew in municipalities that had more strongly rejected Notes: The markers show decreases and increase relative to the base year 2000; differences are statistically different from 0 when the dashed whiskers do not cross the grey horizontal line.
the referendum.The findings suggest that the decision of the French government to ratify the Lisbon Treaty led to both voice and exit behaviour among French voters, with adverse consequences for the quality of democratic participation and deliberation in France.In particular, it appears to have further fuelled the sustained decline of electoral participation and support for traditional centrist parties in France.Survey assessments corroborate that the referendum process seems to have contributed to decreased satisfaction with French democracy and its institutions.The behaviour and structure of political parties might thus have contributed to increasingly widespread perceptions among electorate to be living in a deficient democracy (Keman, 2014).
To strengthen the democratic legitimacy of international institutions, popular referendums on intergovernmental agreements have become more frequent in Europe in recent years (Walter, 2020).However, perceptions about the benefits of international cooperation often diverge between the nation's decision-makers and its population (Dellmuth et al., 2022).When voters reject an international treaty-as they did in the 1992 Danish Maastricht Treaty referendum, the French TCE referendum in 2005, the Irish vote on the Lisbon Treaty 2008, or the Brexit vote in 2016-they can create difficult dilemmas for governments, who can be left to choose between breaching its multilateral obligations and relations with other states (as in the case of Brexit), forcing a re-negotiation with their partners (as in the case of Denmark 1992 or Ireland 2008), or ignoring a democratically taken decision at the risk of fostering distrust towards elites and eroding electoral participation (as in the case of France 2005;Morel, 2007).Either way, the stakes are high.Understanding the causes and modes of EU disintegration makes it imperative to improve our understanding of the tensions between collective decisions taken at the international and the domestic level, and how they affect the quality of governance and the viability of international cooperation in the longer run.

Figure 1 .
Figure 1.Municipal voting in the French Referendum 2005.Notes: Municipalities are allocated into deciles according to the strength of their no-vote; the darker the shade of blue, the higher the municipal no-vote.

Figure 2 .
Figure 2. Association between municipal no-vote and turnout in the National French Elections 2007-2014.Notes: Results are from the fully specified model with all covariates included (as in columns 3 and 6 in Table1); dots represent the estimated coefficients; whiskers show the 95% confidence interval around them.

Figure 3 .
Figure 3. Association between municipal no-vote and blank vote share in the National French Elections 2007-2014.Notes:Results are from the fully specified model with all covariates included (as in columns 3 and 6 in Table1); dots represent the estimated coefficients; whiskers show the 95% confidence interval around them.

Figure 4 .
Figure 4. Association between municipal no-vote and aggregate vote shares in the National French Elections 2007-2014.Notes: Aggregate share of the established traditional parties on the left side, and for the far right and left on the right side; results are from the fully specified model with all covariates included (as in columns 3 and 6 in Table1); dots represent the estimated coefficients; whiskers show the 95% confidence interval around them.

Figure 5 .
Figure 5. Eurobarometer trend assessing average satisfaction with French and EU democracy in the years 2000-2015.Notes:The markers show decreases and increase relative to the base year 2000; differences are statistically different from 0 when the dashed whiskers do not cross the grey horizontal line.

Table 1 .
No-vote share and EP election turnout: Stepwise inclusion of control variables.