A New Concept of European Federalism
LSE ‘Europe in Question’ Discussion Paper Series
Spain’s Referendum on the European
Constitutional Treaty: A Quantitative
Analysis Within the Conceptual Framework
of First and Second Order Elections
Ozgur Erkan
LEQS Paper No. 25/2010
June 2010
Electronic copy available at: https://ssrn.com/abstract=1632308
All views expressed in this paper are those of the author and do not necessarily represent the
views of the editors or the LSE.
© Ozgur Erkan
Editorial Board
Dr. Joan Costa-i-Font
Dr. Vassilis Monastiriotis
Dr. Jonathan White
Ms. Katjana Gattermann
Electronic copy available at: https://ssrn.com/abstract=1632308
Ozgur Erkan
Spain’s Referendum on the European
Constitutional Treaty:
A Quantitative Analysis Within the
Conceptual Framework of First and
Second Order Elections
Ozgur Erkan*
Abstract
In contrast to the attention devoted to the rejection of the EU Constitutional Treaty at French
and Dutch referenda; the Spanish referendum, where this Treaty was ratified, remained
under-researched by political scientists. This paper analyses the voting behaviour at the
Spanish referendum on the EU Constitutional Treaty with the use of quantitative methods
and the concept of first and second-order elections. This paper finds that the Spanish
referendum was a second-order referendum, because the effects of domestic political issues in
Spain had a greater impact on the electoral behaviour of Spanish voters than had genuinely
European issues. This finding raises doubts over the suitability of using direct democracy in
the EU in order to raise the legitimacy and democratic accountability of the European project.
Keywords: Spain, EU, referendum, European Constitutional Treaty, first and
second-order elections
*London School of Economics
2008-09 graduate and award holder of the Paul Taylor prize for the best overall performance
in the degree “MSc. Politics and Government in the European Union”, European Institute,
Houghton St, London WC2A 2AE, UK
Email: ozer85@web.de
Electronic copy available at: https://ssrn.com/abstract=1632308
Spain’s Referendum
Table of Contents
Abstract
1. Introduction 1
2. Theoretical Explanations of Voting Behaviour at EU Referenda:
Domestic and European Factors in Interplay 4
3. Spain’s Referendum on the European Constitutional Treaty:
Hypotheses within the Framework of First and Second-Order
Elections 8
4. Data and Methodology 10
4.1. Dataset 10
4.2. Variables used in the Regression Analysis 12
4.3. Methodology 16
5. Results of the Data Analysis 17
5.1. Results of the Partial Model 21
5.2. Results of the Full Model 25
6. Conclusions 31
Endnotes
Bibliography
Acknowledgements
The author would like to express his gratitude to Prof. Simon Hix and Dr. Joan Costa-i-Font
for their guidance on finding the raw data sets and for providing constructive feedback at
different steps of this project, and to Dr. Sally Stares for her useful advice on the methodology
of this paper.
Electronic copy available at: https://ssrn.com/abstract=1632308
Ozgur Erkan
1
Spain’s Referendum on the European
Constitutional Treaty:
A Quantitative Analysis Within the
Conceptual Framework of First and
Second Order Elections
1. Introduction
Demise of the permissive consensus over the European project has placed
pressure on European political elites to consider alternative ways for
legitimizing the European political order by providing further legal and
political institutionalisation. The creation of a Constitution for Europe was
thus perceived as a panacea by the political elites to set out a new,
incrementally more ambitious European political order, and to increase the
loyalty of Europeans towards the EU (Castiglione in Castiglione et al. 2007:21-
22; Moravcsik in Meunier and McNamara 2007:37-38).
Once the European Constitutional Treaty (ECT hereafter) was completed, it
had to be ratified by all EU members, in order to come into force. While some
member states choose to steer the ECT through their legislative chambers;
four EU members, namely Spain, France, Luxembourg and the Netherlands
hold a national referendum on the ECT.1 Among these four member states,
Spain was the first country to hold a referendum on the ECT, possibly
because Spain has been a country with high attachment to Europe. The
Electronic copy available at: https://ssrn.com/abstract=1632308
Spain’s Referendum
2
referendum in Spain took place on 20 February 2005, only a year after a new
Socialist government was formed. The turnout rate in the referendum was of
42.32%, 76.73% of Spanish voters voted in favour of the ECT, whilst 17.24% of
voters expressed their opposition to the ECT (Torreblanca and Sorroza 2005:
2-3).
Unlike the referendum in Spain, subsequent referenda on the ECT in France
and the Netherlands did not yield positive results, and therefore, the question
that Spain raises is that of why Spain was different? What aspects of the
Spanish institutional and political order made Spain different than France, the
Netherlands and Luxembourg? Following this ratification crisis in France and
the Netherlands, several scholars devoted their attention to analysing the
electoral behaviour at the French and Dutch referenda and to investigating
whether, particularly, No votes were cast based on domestic cleavages or
European issues (e.g. Marthaler 2005, Harmsen 2005, Taggart 2006, Ivaldi
2006). On the other hand, cases of positive integration received less scholarly
attention. While few studies were conducted to explain the electoral
behaviour at the referendum in Luxembourg (Hausemer 2005, Dumont and
Poirier 2006a, 2006b); voting behaviour at the Spanish referendum remained
relatively under-researched within the concept of first and second-order
elections and formed rather part of a comparative study on electoral
behaviour at all four referenda on the ECT (Glencross and Trechsel 2007).
Initial descriptive analyses of the Spanish referendum (Malo de Molina and
Miguel de Elias 2005, Torreblanca 2005) did not also examine the electoral
behaviour at this referendum within the concept of first and second-order
elections.
The aim of this paper is to empirically explore the electoral behaviour at the
Spanish referendum as a case study within the framework of first and secondorder
elections. The main research question to answer in this study is whether
Electronic copy available at: https://ssrn.com/abstract=1632308
Ozgur Erkan
3
domestic cleavages or European issues had a greater impact on the positive
and negative votes cast at the Spanish referendum on the ECT.2 In other
words, this study investigates whether the Spanish referendum on the ECT
was a first or second-order referendum. Through a quantitative analysis, this
study finds that at the Spanish referendum on the ECT domestic cleavages
had a greater impact on the electoral behaviour compared to European issues,
which adds a predominantly second-order characteristic to this referendum.
In obtaining this result, a quantitative methodology, namely binary logistic
regression, was applied to the dataset of the post-referendum survey
conducted by Centro de Investigaciones Sociológicas (CIS). Appropriate
variables within this dataset, which pertain to first and second-order
characteristic of the referendum, were re-coded as appropriate and were
tested together with certain socioeconomic and demographic background
variables.
The remaining parts of this study are structured as follows: Section two
critically reviews the previous studies, which analysed electoral behaviour at
EU referenda within the conceptual framework of first and second-order
elections. In addition, how this study positions itself in the debate is
explained. The third section outlines the hypotheses on the Spanish
referendum, which will be tested through the available dataset. The fourth
section outlines the methodology used in this study and explains which
variables were used as well as how these variables were re-coded. The fifth
section runs the regression test and presents the results of this statistical
analysis. The concluding section has two aims. Firstly, it summarises the main
findings and discusses the implications of this study with regard to the
existing studies conducted on direct democracy in the EU. Secondly, it
demonstrates certain implications that this study may have for constitutional
politics and democracy in the EU.
Electronic copy available at: https://ssrn.com/abstract=1632308
Spain’s Referendum
4
2. Theoretical Explanations of Voting Behaviour at EU
Referenda: Domestic and European Factors in Interplay
Prior to the ECT, various member states held national referenda on their EU
membership as well as on previous EU treaties. French referendum on the
Maastricht Treaty, Danish referenda on its EU membership, the Single
European Act, and the Maastricht Treaty and Irish referenda on the Nice
Treaty are only a few examples of these.
The literature has devoted significant attention to explaining whether voters
cast their votes at EU referenda based on domestic cleavages or European
issues. In explaining the electoral behaviour at previous EU referenda,
scholars applied the concept of first and second-order elections to EU
referenda, which was originally coined by Reif and Schmitt (1980) for the
analysis of electoral behaviour at European Parliament (EP hereafter)
elections. In their analysis of the results of the first direct elections to the EP in
1979, Reif and Schmitt demonstrated that EP elections have a “second-order”
characteristic, meaning that at these elections domestic cleavages have a
greater influence on the votes than have European issues (Reif and Schmitt
1980:3).
Franklin et al. (1995) were among the first scholars to extend the framework of
first and second-order elections to the analysis of EU referenda. They used
this concept to analyse the electoral behaviour at national referenda on the
Maastricht Treaty. Franklin et al. (1995:111) found that the national referenda
on the Maastricht Treaty had a second-order characteristic, since voting
behaviour was predominantly influenced by domestic factors, such as
attitudes towards the government. This finding gave rise to the Franklin
thesis, that national referenda were rather reflections of attitudes towards the
Electronic copy available at: https://ssrn.com/abstract=1632308
Ozgur Erkan
5
government, rather than being a process of issue-voting (Svensson 2002:735).
Using the same concept, Svensson (2002) later analysed five Danish referenda
on EU issues, ranging from the referendum on Danish entry to the EU to the
one on the Amsterdam Treaty in 1998. Svensson’s findings show that,
contrary to the Franklin thesis above, the five Danish EU referenda were not
mainly second-order referenda, since ‘votes at these referenda also reflected
Danish voters’ attitudes towards the EU and the treaties in question’
(Svensson 2002:748). More recently, Garry et al. (2005) examined the two Irish
referenda on the Nice Treaty, again within the framework of first and secondorder
elections. Garry et al.’s finding refutes the Franklin thesis for the Irish
case, as they posit that the impact of “issue-voting” (i.e. attitudes towards the
EU and the Nice Treaty) outperforms the influence of “domestic factors” on
the votes at both referenda (Garry et al. 2005:214). Therefore, the two Irish
referenda on the Nice Treaty are shown to be “first-order” referenda, based
on the findings of Garry et al.
These rigorous analyses of electoral behaviour at EU referenda, however,
leave out one important element. These studies reveal that respective EU
referenda were first or second-order referenda for the entire country, yet they
do not investigate whether one can trace diverging first and second-order
characteristics in voting behaviour when measured against different
socioeconomic classes. Put differently, it would be interesting to know
whether at a first-order referendum all classes in the society vote on European
issues or whether some vote on still domestic issues. If such a diverging
pattern exists, it would be interesting to decipher, which classes cast secondorder
votes and which classes vote on European issues.
Although previous EU referenda garnered sufficient attention from scholars,
rejection of the ECT at French and Dutch referenda in 2005 has diverted the
Electronic copy available at: https://ssrn.com/abstract=1632308
Spain’s Referendum
6
attention of scholars to investigating the reasons of positive and negative
votes on the ECT in those countries.
Initial studies exploring the electoral behaviour at the French referendum on
the ECT demonstrated the existence of a second-order characteristic of the No
votes. A significant motivation behind the No votes was to punish the
government and express concerns for the domestic socioeconomic situation in
France (Taggart 2006:16-17). When measured against socioeconomic
indicators, the second-order characteristic of No votes in France was found to
be stronger among the unemployed in particular (Marthaler 2005:234). The
second-order characteristic of No votes in France was, however, a contentious
point among scholars, since one other study showed that No votes at the
French referendum also pertained to issue-voting, i.e. to European issues
(Ivaldi 2006:58). According to Ivaldi, these European issues, that add a firstorder
characteristic to the referendum, were mainly the absence of a social
Europe, neoliberal/free market policies of the EU, and possible Turkish EU
accession (Ibid:58-59). In reality, however, it is dubitable whether the
aforementioned issues are genuinely European issues, because as Moravcsik
posits, these three issues are predominantly reflections of national or global
concerns on European issues (Moravcsik in Meunier and McNamara 2007:39).
For example, while criticism of neoliberal policies in the EU actually stems
from the national opposition to globalisation; the fears over the Turkish EU
accession are fuelled by national concerns over considerable third-country
immigration to France and Europe. Thus, it is open to debate to what extent
the No votes in France carried a first-order characteristic.
With regard to the Yes votes at the French referendum, however, the picture
changes slightly, compared to No votes. The study conducted by Dehousse
(2006) demonstrates that the majority of Yes voters in France cast their votes
Electronic copy available at: https://ssrn.com/abstract=1632308
Ozgur Erkan
7
based on their attitudes towards the EU and the ECT, which adds a first-order
characteristic to the Yes votes at this referendum (Ibid:156).
For the Dutch referendum on the ECT, studies have found similar results.
Both Taggart (2006:19) and Harmsen (2005:10-11) demonstrated that No votes
in the Netherlands reflected both first and second-order characteristics. The
most frequently cited reasons for voting against the ECT in the Netherlands
were lack of information on the ECT, loss of national sovereignty, opposition
to national government and Europe being too expensive (Taggart 2006:19). On
the other hand, positive votes were found to possess a predominantly firstorder
characteristic at the Dutch referendum on the ECT (Harmsen 2005:9-10).
As seen from above, scholars have placed a significant emphasis on
examining the unsuccessful referenda for the ECT.3
Disproportionately less
attention was, however, devoted to the analysis of electoral behaviour at
successful referenda on the ECT in Luxembourg and Spain. So far Hausemer
(2005) and Dumont and Poirier (2006a; 2006b) have analysed the electoral
behaviour at the referendum in Luxembourg, whilst Glencross and Trechsel
(2007) have for the first time conducted a comparative analysis of the
referenda in France, the Netherlands, Spain and Luxembourg. Moreover,
while Torreblanca (2005) provided rather an analysis of the campaigns in the
run up to the Spanish referendum on the ECT; Malo de Molina and Miguel de
Elias (2005) only examined the effects of domestic issues on voting behaviour
at the Spanish referendum.
Despite these analyses of the successful referenda on the ECT, there is still
inadequate focus on explaining the electoral behaviour at the Spanish
referendum as a case study within the conceptual framework of first and
second-order elections, and this study aims to fill this gap in the literature.
Electronic copy available at: https://ssrn.com/abstract=1632308
Spain’s Referendum
8
Examining the Spanish referendum as a case study has two merits. First of all,
in a quantitative case study, potential cancelling out effects of different
country variables are eliminated. For example, in their quantitative
comparative study, Glencross and Trechsel (2007:14) find that the referenda in
all four aforementioned member states were first-order referenda. Glencross
and Trechsel, however, miss out on the possibility that in their pooled dataset,
domestic factors in one country may be cancelling out the effects of domestic
factors in another country, so that in the end domestic factors do not seem to
have a significant impact on the votes. Secondly, a case study has the
advantage of measuring variables that are specific to a country. In a
comparative study with pooled dataset, however, the researcher will not
necessarily incorporate certain explanatory variables into the analysis, which
may significantly matter for one country in the dataset but not for others.
3. Spain’s Referendum on the European Constitutional
Treaty: Hypotheses within the Framework of First and
Second-Order Elections
In examining the Spanish referendum within the concept of first and secondorder
elections, the hypotheses below will be tested with the use of a
quantitative method. These hypotheses were constructed based on actual
political events that took place in Spain prior to and during the campaigns on
the referendum for the ECT.
The first set of hypotheses tests the second-order characteristic of the Spanish
referendum on the ECT:
Electronic copy available at: https://ssrn.com/abstract=1632308
Ozgur Erkan
9
H1: Given the paucity of a genuine European debate concerning the
content of the ECT during the referendum campaigns in Spain
[Castiglione in Castiglione et al. 2007:24], it is expected that a
significant amount of Spaniards have voted based on domestic political
cleavages. As a result, the Spanish referendum on the ECT carries a
second-order characteristic.
H1.1: Previous studies have found that in Luxembourg, France and the
Netherlands pro or anti-government attitudes had an impact on the
outcome of referenda [Taggart 2006, Harmsen 2005, Hausemer 2005].
It will be thus hypothesized that in Spain attitudes towards the
incumbent Spanish Prime Minister Zapatero and his government also
had an impact on the voting behaviour of Spaniards at the referendum
on the ECT. This condition also adds a second-order characteristic to
the Spanish referendum.
The second set of hypotheses concerns the first-order characteristic of the
referendum:
H2: In line with the hypotheses posed by Glencross and Trechsel
(2007:6), it is expected that positive attitudes towards the EU and the
ECT in Spain have a positive impact on the vote at the polls. This
condition makes the Spanish referendum a first-order referendum, as
voters cast their votes based on European issues and/or on issues
pertaining to the content of the ECT.
H2.1: Medrano (1995:84) finds a statistically significant positive
correlation between possessing a European identity and supporting
European integration in Spain. By taking Medrano’s finding as an
anchor, it will be hypothesized that possessing a European identity
had a positive impact on the vote for the ECT in Spain, in order to
support further integration. Since the votes cast under a strong
European identity relate to attitudes towards the EU and/or the ECT,
this condition also adds a first-order characteristic to the referendum.
Electronic copy available at: https://ssrn.com/abstract=1632308
Spain’s Referendum
10
Finally, the positive and negative votes cast on the ECT at the Spanish
referendum will be tested together with certain socioeconomic and
demographic background variables. These background variables are: age,
gender, level of education, economic activity, religion of the respondents and the
autonomous community in which respondents live. Among these variables, the
following hypothesis was posed for religion:
H3: There was significant emphasis on religious affairs during the
campaigns on the ECT, both through the radically Catholic Aznar
government in Spain and the lobbying of Aznar government by the
Pope for including references to Christianity and God in the ECT (Closa
2004:335-337). Moreover, the conservative ABC newspaper, in
particular, defended the idea of referring to Christianity and God in the
constitutional text (Jiménez and Sampedro 2005:14). Despite these
demands, however, the ECT did not refer to Christianity and God.
Based on this fact, it will be hypothesized that Catholic voters in Spain
were more likely to vote against the ECT.4
These hypotheses will be tested in section 5. Before running the statistical
analysis, however, it is worth introducing the dataset and variables used in
this study to test the aforementioned hypotheses, which is the focal point of
the following section.
4. Data and Methodology
4.1. Dataset
The dataset used for testing the hypotheses in the previous section stems from
the post referendum survey of Centro de Investigaciones Sociológicas (CIS),
entitled Estudio CIS no.2595 – Postreferendum sobre la Constitución Europea.
Main features of this dataset are presented in Table 1 below.
Electronic copy available at: https://ssrn.com/abstract=1632308
Ozgur Erkan
11
Table 1: CIS Survey no. 2595 - Features
Sample size (N)
2487
Fieldwork date
26 February – 4 March 2005
Standard error of
the survey
± 2% at 95.5% confidence level
Area covered by the
survey
166 municipalities and 17 provinces in
17 Spanish autonomous communities
In addition to the CIS survey 2595, there is a second dataset available on the
Spanish referendum, namely Flash Euro-barometer 168 (Flash EB-168
hereafter). In this study, however, CIS survey was preferred to Flash EB-168
for two reasons: Firstly, the CIS survey places weight on the beliefs of
respondents, unlike Flash EB-168, which was an influential sociodemographic
factor during the referendum campaigns, due to the debate over
whether there should be any references to Christianity and God in the
European Constitution [Closa 2004:335-337]. Since religious factors mattered
during the campaigns, incorporating this variable into the analysis can have
an impact on the results. Secondly, compared to Flash EB-168, CIS dataset
includes more than one measure of attitudes towards the EU, both through
the evaluation of European identity and the reaction of Spaniards, in case the
EU dissolved tomorrow. The existence of more than one variable on attitudes
towards the EU gives the researcher ample opportunity to measure the firstorder
characteristic of the referendum from different perspectives.
Electronic copy available at: https://ssrn.com/abstract=1632308
Spain’s Referendum
12
4.2. Variables used in the Regression Analysis
4.2.a. Response Variable
The response variable used in this study reflects Spanish voters’ choice on the
ECT. The original survey question asks the respondents how they voted on
the ECT and it has four possible choices: Voting in favour of the ECT, voting
against the ECT, blank vote and don’t know. Since this analysis aims to find
out whether Spaniards voted in favour of or against the ECT based on
domestic or European issues, blank votes and don’t know answers were
excluded from the analysis. The response variable was named voting
constitution and recoded into a dichotomous variable with 1= voting in favour
of the ECT and 0= voting against the ECT.
4.2.b. Explanatory variables
Socioeconomic and Demographic Variables
The dataset includes socioeconomic and demographic variables, such as
education, gender, age, economic activity and religious affiliation of the
respondents, which were incorporated into this study. The reason for making
use of these variables in this study is the significance of some of these factors
in explaining EU-related votes, as propounded by Anderson and Reichert (in
Glencross and Trechsel 2007:8).
The socioeconomic and demographic factors used in this study are:
Age: continuous variable, with a range of values between 18 and 93
Electronic copy available at: https://ssrn.com/abstract=1632308
Ozgur Erkan
13
Gender: dichotomous variable, recoded as gender(1) = males and gender(0) =
females. The reference category was taken here as gender(0), which was thus
not demonstrated on Table 2 in section 5.
Education: categorical variable, demonstrating respondents’ completed level
of education. The original variable consists of eleven categories, yet to be able
to analyse this variable, it was clustered into five categories and recoded as
follows: Education(1) = less than five years of schooling, Education(2) = primary
school education, Education(3) = secondary school education, Education(4) =
occupational training, and Education(5) = tertiary education. “Don’t know”
and “no comment” answers were excluded and Education (5) was taken as the
reference category.
Religion: dichotomous variable, reflecting the Catholicity of respondents. The
original survey question asked the respondents which religion they belong to,
and comprised five categories. The variable was re-coded here into a dummy
variable with Religion (0)= the respondent is a Catholic, Religion (1)= the
respondent either belongs to another religion or does not believe in God.
Religion (0) was taken as the reference category here.5
Autonomous Community: categorical variable, indicating the autonomous
community a respondent resides in. The variable had originally seventeen
categories, i.e. each community being a category. To be able to analyse this
variable, however, it was clustered into the following three categories, based
on the level of positive and negative votes cast on the ECT at each
autonomous community: Autonomous Community (1)= Communities where
less than 70% of the residents voted in favour of the ECT. The communities in
the first category are Navarra, the Basque Country and Catalonia. Autonomous
Community (2)= Communities where more than 70% but less than 75% of the
residents voted in favour of the ECT, i.e. Madrid. Autonomous Community (3)=
Electronic copy available at: https://ssrn.com/abstract=1632308
Spain’s Referendum
14
all other communities with the highest level of votes in favour of the ECT,
where the level of Yes votes did not fall below 76.5%. The autonomous
communities in this category are Aragón, Andalucía, Asturias, Balearics,
Canary Islands, Cantabria, Castile-la-Mancha, Castile-and-León,
Extremadura, Galicia, Murcia, Rioja and Valencia. Autonomous Community(3)
was taken here as the reference category.
Economic activity: categorical variable, indicating the occupational status of
the respondents. The variable included initially nine categories, yet the “other
situation” and “no comment” categories were expunged from the analysis.
The variable was then recoded into the following five categories: Economic
Activity (1)= Employed; Economic Activity (2)= Retired; Economic Activity(3)=
Unemployed; Economic Activity (4)=Student; Economic Activity (5)= Housework
(unpaid). Economic Activity(5) was taken as the reference category.
Variables on the First-Order Dimension
Identity: categorical variable, measuring whether the respondents possess
only European, only Spanish or dual (i.e. both European and Spanish)
identity. The survey question encompassed six possible choices, yet “don’t
know” and “no comment” choices were excluded and the variable was recoded
into the following categories: Identity(1)= only European identity;
Identity(2)= only Spanish identity; Identity(3)= dual identity; Identity(4)= any
other identity not mentioned above. Identity (4) was taken as the reference
category here. Since the first and third categories of this variable pertain to
European issues; it was taken as a variable on the first-order dimension.
EUphile: categorical variable, indicating the level of support for European
integration in Spain. The question in the survey asks the respondents how
Electronic copy available at: https://ssrn.com/abstract=1632308
Ozgur Erkan
15
they would react if the EU dissolved tomorrow and did not exist anymore.
The variable had originally seven categories, yet “no comment” answers were
expunged. The categories of having “no opinion” and “being indifferent to
the dissolution of the EU” were clustered together as one category and the
variable was re-coded as follows: EUphile(1)= Being happy about the
dissolution of the EU tomorrow; EUphile(2)= Disapproving of the dissolution
of the EU tomorrow; EUphile(3)= Being indifferent on this issue. EUphile(3)
was accepted as the reference category.
Constitutionphile: categorical variable, measuring the respondents’ attitude
towards the ECT. The variable had initially seven categories, yet the
categories with “indifferent” and “no opinion” answers were clustered
together and the variable was re-coded into the following categories:
Constitutionphile(1)= The ECT is a bad thing for Europe; Constitutionphile(2)=
The ECT is a good thing for Europe; Constitutionphile(3)= Being indifferent to
the ECT, which is also the reference category here.
Variables on the Second-Order Dimension
Support for Zapatero: dichotomous variable, displaying confidence in the
Spanish Prime Minister Zapatero and his government. The variable had
originally six categories. The variable was re-coded into a dummy variable by
clustering the positive answers of “great confidence” and “enough
confidence” into one category and the negative responses of “little
confidence” and “very low confidence” into another category. As before,
“don’t know” and “no answer” options were left out. Of these two categories,
Support for Zapatero(1) shows low confidence in Zapatero and his government;
whilst Support for Zapatero(0) indicates high confidence in Zapatero and his
government, which was also defined as the reference category.
Electronic copy available at: https://ssrn.com/abstract=1632308
Spain’s Referendum
16
Party Politics: categorical variable, demonstrating the political party
respondents voted for at the general elections of 14 March 2004. The variable
initially comprised seventeen categories; yet “blank votes” at March 2004
elections, as well as “don’t know” and “no comment” answers were excluded.
The political parties were then recoded into the following three categories
based on their attitude towards the ECT: Party Politics(1)= parties strongly in
favour of the ECT, i.e. PSOE (Spanish Socialist Workers’ Party), CC (Canary
Coalition); Party Politics(2)= parties officially in favour of the ECT, albeit
experiencing internal fragmentations regarding the content of the ECT, i.e. PP
(People’s Party), CiU (Convergence and Union), PNV (Basque Nationalist
Party); Party Politics(3)= parties strongly against the ECT, i.e. IU (United Left),
ERC (Catalonian Republican Left), EA (Basque Solidarity), BNG (Galician
Nationalist Bloc), Naforra-Bai (Navarra Yes) and CHA-Aragon (Aragonese
Council) [Malo de Molina and Miguel de Elias 2005:78; Torreblanca and
Sorroza 2006:1]. Party Politics(3) was taken as the reference category.
Left-Right Politics: categorical variable, which measures the self-positioning
of voters on the left-right political space with regard to domestic politics. The
variable was re-coded into five categories as follows: Left-Right Politics(1)= farleft
voters; Left-Right Politics(2)= centre-left voters; Left-Right Politics(3)= voters
at the centre; Left-Right Politics(4)= centre-right voters; Left-Right Politics(5)=
far-right voters. Left-Right Politics(5) was taken as the reference category here.
4.3. Methodology
Given that the response variable (voting constitution) is a dichotomous
variable, a binary logistic regression was used in this analysis, which was
computed using the SPSS software. In line with other regression tests, firstly a
partial model was constructed, which encompasses the following
Electronic copy available at: https://ssrn.com/abstract=1632308
Ozgur Erkan
17
socioeconomic, demographic background variables as well as variables on the
second-order dimension: age, autonomous community, economic activity,
education level, gender, religion a respondent is attached to, left-right politics,
party politics, and support for Zapatero. The partial model was then developed
into a full model by adding the following explanatory variables on the firstorder
dimension: identity, constitutionphile, and EUphile.
The effects on the response variable of all aforementioned explanatory
variables were measured based on their respective reference categories and
by taking into account the statistically accepted significance levels. Moreover,
the Nagelkerke R2
and the -2log likelihood values were closely observed
throughout the analysis, to confirm the capability of both the partial and the
full model in explaining the variations of the response variable.
To also control for the effects of multicollinearity in the full model, bivariate
correlations for all explanatory variables were computed on a crosstab, which
were kept in sight throughout the analysis. The issues which may arise with
regard to the effects of multicollinearity will be addressed towards the end of
the next section.
5. Results of the Data Analysis
The results obtained through the test of the available dataset are presented in
Table 2 below. The data test was conducted in two steps, as explained at the
end of the last section. Firstly, a partial model was constructed, which
contains socioeconomic and demographic background factors as well as
second-order variables. Thereafter, this model was improved through the
addition of first-order variables, which leads to the full model.
Electronic copy available at: https://ssrn.com/abstract=1632308
Spain’s Referendum
18
Table 2: Partial and Full Models (with regression coefficients) for Explaining
Electoral Behaviour at the Spanish Referendum on the ECT
Variables Partial Model Full Model
Age 0.030*** 0.039***
(0.012) (0.017)
Autonomous Community (1) -1.203*** -1.120***
(0.289) (0.408)
Autonomous Community (2) 0.102 0.322
(0.378) (0.487)
Economic Activity (1) 0.200 -0.621
(0.508) (0.643)
Economic Activity (2) 0.022 -0.907
(0.617) (0.764)
Economic Activity (3) 0.048 0.262
(0.677) (0.904)
Economic Activity (4) -0.157 -0.685
(0.665) (0.910)
Education (1) 1.336 4.082
(1.273) (2.751)
Education (2) 0.728 0.973
(0.488) (0.614)
Education (3) 0.245 0.661*
(0.297) (0.393)
Education (4) -0.086 0.355
(0.425) (0.566)
Electronic copy available at: https://ssrn.com/abstract=1632308
Ozgur Erkan
19
Gender (1) -0.169 0.029
(0.275) (0.354)
Religion (1) -1.011*** -0.659*
(0.285) (0.391)
Left-Right Politics (1) 2.744*** 3.445***
(0.862) (1.214)
Left-Right Politics (2) 2.769*** 3.280***
(0.810) (1.145)
Left-Right Politics (3) 3.157*** 3.509***
(0.724) (1.051)
Left-Right Politics (4) 3.101*** 3.424***
(0.754) (1.061)
Party Politics (1) 2.900*** 2.823***
(0.341) (0.459)
Party Politics (2) 2.199*** 2.772***
(0.492) (0.680)
Support for Zapatero (1) -1.317*** -1.432***
(0.320) (0.454)
Identity (1) 0.124
(0.777)
Identity (2) 0.432
(0.538)
Identity (3) 0.580
Electronic copy available at: https://ssrn.com/abstract=1632308
Spain’s Referendum
20
(0.544)
Constitutionphile (1) -1.976**
(0.879)
Constitutionphile (2) 2.719***
(0.349)
EUphile (1) -1.228**
(0.560)
EUphile (2) 1.048***
(0.401)
Constant -3.572*** -6.935***
(1.087) (1.708)
Nagelkerke R2 0.500 (50%) 0.714 (71.4%)
-2Log Likelihood 471.930 297.091
Valid cases (N) 892 886
Standard errors are indicated in brackets below each coefficient.
* = p<0.1 (significant at 10% level); ** = p<0.05 (significant at 5% level); *** = p<0.01 (significant
at 1% level)
Electronic copy available at: https://ssrn.com/abstract=1632308
Ozgur Erkan
21
5.1. Results of the Partial Model
An initial review of the partial model shows that among the background
variables used; age, religion and the first category of Autonomous Community
have a statistically significant effect on the response variable, i.e. the vote cast
at the referendum for the ECT. Gender, economic activity, education, and the
second category of Autonomous Community, however, are not found to have a
statistically significant impact on the response variable under the partial
model.
Of the aforementioned background variables, the positive and statistically
significant coefficient of age yields that at the Spanish referendum older voters
were more likely to vote in favour of the ECT than younger voters, holding
other explanatory variables constant. This finding may seem surprising, as it
contradicts the well-established Inglehart thesis, that younger cohorts are
more supportive of EU integration than older people (Hix 2005:162).
However, similar studies done on analysing the electoral behaviour at
referenda on the ECT reached the same conclusion. Ivaldi (2006:61)
demonstrates that at the French referendum, voters aged 60 and over were
more likely to vote in favour of the ECT, compared to younger voters.
Similarly, Glencross and Trechsel (2007:9) found in all four EU members they
analysed that voting in favour of the ECT was likely to increase with rising
age. As will be elaborated in the conclusion, further research could shed light
on the reasons of this interesting pattern found at national referenda on the
ECT.
In addition to age, the coefficient of the first category of Autonomous
Communities also comes out as statistically significant. The statistical
significance of this coefficient bears two important interpretations: Firstly, the
coefficient of the first category of Autonomous Communities implies that ethnic
Electronic copy available at: https://ssrn.com/abstract=1632308
Spain’s Referendum
22
nationalism played a role at the referendum. As can be seen, after holding all
other variables constant, autonomous communities with the highest level of
ethnic nationalism, i.e. Navarra, the Basque Country and Catalonia were on
the whole more inclined to vote against the ECT, compared to autonomous
communities in the third category (i.e. the reference category).6 This finding
supports the argument of Torreblanca that the ‘territorial question pulled
down the level of Yes votes at the Spanish referendum on the ECT’
(Torreblanca 2005:1).
Secondly, a closer review of the third category of the variable Autonomous
Communities yields that, of the thirteen communities in that category except
Rioja, Aragón and Balearics, all were Objective 1 regions under the Structural
Fund Regulations of the EU Regional Policy at the time of the referendum
(European Union 2005). This tends to promote the fact that residing in an
autonomous community, which receives further financial support from the
EU through the EU Regional Policy, makes the residents of this autonomous
community more likely to support European integration and thus vote in
favour of the ECT. This conclusion can be drawn in comparison to the
autonomous communities in the first category of this variable, which were
also among the four richest Spanish autonomous communities in terms GDP
per capita by 2006 and which therefore do not qualify as Objective 1 regions.7
This interpretation also buttresses the ‘utilitarian support’ argument of
Medrano (1995). Medrano finds that support for further European integration
in Spain is conditional on direct financial transfers from the EU and the
economic development brought through European integration (Medrano
1995:51), making Spaniards’ endorsement for further integration a utilitarian
support.
The final statistically significant background variable under the partial model
is religion. It can be inferred from the coefficient of religion that, controlling for
Electronic copy available at: https://ssrn.com/abstract=1632308
Ozgur Erkan
23
all other explanatory variables, members of other religions or non-believers
were less likely to endorse the ECT, compared to Catholic voters. Contrary to
the expectations, this statistically significant finding refutes our hypothesis H3
(Catholic voters would be more inclined to reject the ECT, given the absence
of references to Christianity in the constitutional text).
The partial model also yields statistically significant results for variables on
the second-order dimension. A reassessment of the hypotheses on secondorder
variables demonstrates that the results under the partial model confirm
hypotheses H1 and H1.1.
Firstly, the statistically significant coefficients of left-right politics imply that
self-positioning of a voter in the left-right ideological space with regard to
domestic cleavages helps explain the electoral behaviour of that voter at the
Spanish referendum on the ECT. This finding provides evidence for
hypothesis H1 (the referendum had a second-order characteristic). Based on
the coefficients of this variable, it can be concluded that odds of voting in
favour of the ECT were 5.75 % higher for a Spanish voter at the centre of the
left-right dimension than a Spanish voter on the centre-right,8 after controlling
for other variables. Similarly, the odds of voting in favour of the ECT were
47.4 % higher for the same centrist voter compared to a Spanish voter on the
centre-left.9
In addition to left-right politics; party-politics is another explanatory variable on
the second-order dimension that helps confirm hypothesis H1. Given the
statistically significant coefficient of the first category of party politics, it can be
concluded that, after holding all other explanatory variables constant, having
voted for a party at March 2004 national elections, which strongly endorsed
the ECT, increases the likelihood of voting in favour of the ECT, compared to
the voters of anti-ECT parties.10
Similarly, the coefficient of the second
Electronic copy available at: https://ssrn.com/abstract=1632308
Spain’s Referendum
24
category of party politics asserts that having voted for a party at March 2004
general elections, which experienced internal fragmentations regarding the
content of the ECT, but which nevertheless endorsed the ECT, also increases
the likelihood of voting in favour of the ECT, compared to voters of anti-ECT
parties.11 This increase is, however, slightly less than the effect of the first
category of party politics. It should be noted that, the results from party politics
reveal one further crucial interpretation with regard to the variable
Autonomous Communities. The results of the latter variable should not be
interpreted as every resident in Catalonia and the Basque Country
particularly, having voted against the ECT. It should be considered that
supporters of PNV and CiU were more likely to vote in favour of the ECT,
despite living in the Basque Country or Catalonia, compared to other
residents of the Basque Country and Catalonia.
Finally, the last explanatory variable on the second-order dimension, support
for Zapatero, confirms hypothesis H1.1 (attitudes towards the government
have an impact on the votes at the referendum). The statistically significant
coefficient of this variable implies that expressing higher confidence for
Zapatero and his government increases the likelihood of voting in favour of
the ECT. The exponential of this coefficient [exp(-1.317)] yields that, holding
other explanatory variables constant, voters not trusting Zapatero and his
government were 73.21% more likely to vote against the ECT, compared to
voters trusting Zapatero and his government.12
As a result, the partial model confirms the hypotheses on the second-order
characteristic of the Spanish referendum on the ECT. With regard to the
background variables, as shown, the hypothesis on religion was not proved.
Finally, the partial model found statistically significant effects for age and the
first category of Autonomous Community of residence on the response variable,
Electronic copy available at: https://ssrn.com/abstract=1632308
Ozgur Erkan
25
yet not for the remaining background variables gender, education, economic
activity and the second category of Autonomous Communities.
5.2. Results of the Full Model
One initial observation with regard to the full model is that, moving from the
partial to the full model improves the model fit significantly. This tendency
can be observed through two different indicators. Firstly, the -2 log likelihood
ratio improves by 37.05% with the construction of the full model, as it falls
from 471.930 to 297.091. Secondly, the notable increase in Nagelkerke Rsquare
from 50% to 71.4% demonstrates the high capability of the full model
in explaining the variations of the response variable, compared to the partial
model.
The full model yields results with regard to background and second-order
variables that are virtually identical to the ones found under the partial
model. Just as under the partial model, the full model also reveals that the
likelihood of voting in favour of the ECT increases with rising age. Similarly,
the full model also finds evidence for the tendency that Spanish autonomous
communities receiving further financial benefits from the EU –through the
Regional Policy of the EU- were more likely to vote in favour of the ECT than
autonomous communities in the first category of the variable Autonomous
Communities. Moreover, the full model also confirms the existence of a
territorial dimension to the referendum through the same variable,
Autonomous Communities. The effect of religion on the votes remains
unchanged, although this variable has a slightly lower statistical significance
under the full model. In addition to that, statistically significant coefficients of
the explanatory variables left-right politics, party politics and support for Zapatero
reconfirm the second-order characteristic of the Spanish referendum under
Electronic copy available at: https://ssrn.com/abstract=1632308
Spain’s Referendum
26
the full model. The only exception under the full model is that, whilst under
the partial model all categories of education were statistically insignificant;
under the full model, the third category of education comes out statistically
significant at 10% significance level. Based on this finding, it can be concluded
only with regard to the third category of education that voters with secondary
education were more likely to endorse the ECT compared to voters with
tertiary education, after holding all other variables constant. It should be
noted, that this finding does not imply the existence of a linear relation. Put
differently, based on solely one statistically significant category of education, it
cannot be concluded that support for the ECT in Spain rises with falling level
of education.
As a final step, insertion of the first-order variables constitutionphile, EUphile
and identity into the full model proves the correctness of hypothesis H2
(positive attitudes towards the EU and the ECT have positive effects on the
vote for the ECT). Hypothesis H2.1 can, however, be neither confirmed nor
refuted as the coefficients for all categories of the variable identity are
statistically insignificant.
One of the explanatory variables that helps verify hypothesis H2 is the
variable constitutionphile. The statistically significant coefficient of the first
category of constitutionphile implies that, after controlling for other variables,
voters with a negative attitude towards the ECT were more likely to vote
against the ECT, compared to voters with a neutral attitude towards the ECT.
Similarly, the coefficient of the second category of constitutionphile asserts that
voters with a positive attitude towards the ECT were significantly more likely
to vote in favour of the ECT, compared to voters with a neutral attitude.
Another explanatory variable on the first-order dimension, EUphile, helps
confirm hypothesis H2. The first category of EUphile demonstrates that, after
Electronic copy available at: https://ssrn.com/abstract=1632308
Ozgur Erkan
27
holding other variables constant, voters with a negative attitude towards the
EU were more likely to reject the ECT, compared to voters with a neutral
attitude towards European integration. Moreover, the second category of
EUphile indicates that voters with a strongly positive sentiment towards the
EU tended to vote more in favour of the ECT, compared to voters with a
neutral attitude towards the EU. As can be seen from Table 2, both of these
aforementioned findings are highly statistically significant.
Reassessing the question whether the Spanish referendum had a first-order or
second-order characteristic, the findings under the full model demonstrate
that the Spanish referendum possessed both first and second-order
characteristics. Put differently, both domestic and European issues were
found to have an impact on the votes at the Spanish referendum for the ECT.
The more interesting question to ask at this stage is which of the two
aforementioned factors had a greater impact on the votes cast at this
referendum. In order to answer that question, fitted probabilities were
calculated for each of the first and second-order explanatory variables, i.e. for
EUphile, constitutionphile, identity, left-right politics, party politics, support for
Zapatero, as well as for aggregated first and second-order variables. The
calculated fitted probabilities for individual and aggregated first and secondorder
variables are presented in Table 3 below. The second column of Table 3
indicates, on the example of left-right politics, the predicted probability that a
Spanish voter on the far-right (which is the reference category here) votes in
favour of the ECT. The third column then yields the predicted probability of
voting in favour of the ECT, if the same voter identified himself on the centreleft
instead (which is the category with the highest frequency). The difference
between these two values shows as a percentage the effect of a change in the
explanatory variable left-right politics on voting in favour of the ECT. The
Electronic copy available at: https://ssrn.com/abstract=1632308
Spain’s Referendum
28
method used for calculating the fitted probabilities is described in the
endnotes.13
Table 3: Fitted Probabilities for First and Second-Order Variables
Variables
Value at dummy = 0, in
%
Value at dummy = 1, in %
Change, in
%
Support for Zapatero
14
99.42 97.61 1.81
Variables
Value at the reference
category, in %
Value at the category with the
highest frequency, in %
Change, in
%
Left-Right Politics 86.57 99.42 12.85
Party Politics 91.06 99.42 8.36
Constitutionphile 91.87 99.42 7.55
EUphile 98.36 99.42 1.06
Identity 98.96 99.42 0.46
General
Comparison
First-Order Variables 68.93 99.42 30.49
Second-Order
Variables
27.70 97.62 69.92
The fitted probabilities above indicate that most second-order variables had
individually a greater impact on the votes for the ECT than most first-order
variables. The calculations yield that the individual effects of second-order
variables left-right politics and party politics on voting in favour of the ECT
Electronic copy available at: https://ssrn.com/abstract=1632308
Ozgur Erkan
29
dominate the individual effects of all first-order variables. Among the firstorder
variables, only the individual effects of constitutionphile on the response
variable can dominate the individual effects of the second-order variable
support for Zapatero. As a result, even the individual effects imply that secondorder
variables had a greater impact on voting in favour of the ECT than firstorder
variables. In order to strengthen this conclusion, however, one needs to
compare the cumulative effect of all first-order variables on a positive vote at
the referendum with the cumulative effect of all second-order variables.
The final two rows of Table 3 above demonstrate the results of this
comparison under general comparison, which show the collective impact of first
and second-order variables on the response variable separately. A comparison
of the collective effect of all second-order variables on voting in favour of the
ECT with the collective effect of all first-order variables implies that domestic
cleavages had a far greater impact on a positive vote at the Spanish
referendum than had EU issues. As can be seen from the results in Table 3,
moving from the reference category to the category with the highest
frequency on all first-order variables simultaneously changes the likelihood of
voting in favour of the ECT by 30.49%. This effect, however, more than
doubles when all second-order variables move from their reference category
to their category with the highest frequency (69.92%). Consequently,
calculation of the fitted probabilities tends to promote that hypotheses H1 (the
Spanish referendum on the ECT has a second-order characteristic) and H1.1
(attitudes towards the government have an effect on the votes at the
referendum) provide the ultimate answer to the research question of this
study, whether the Spanish referendum on the ECT was a first or secondorder
referendum. This conclusion contradicts the findings of Glencross and
Trechsel (2007). As shown in section 2, Glencross and Trechsel find in their
comparative study of all four EU member states, which held a referendum on
Electronic copy available at: https://ssrn.com/abstract=1632308
Spain’s Referendum
30
the ECT, that in these countries, including Spain, the referenda had a firstorder
characteristic.
As mentioned at the end of the last section, the potential effects of
multicollinearity may be a cause for concern in concluding that second-order
variables had a greater impact on the response variable than first-order
variables. To clear any doubts about this question, bivariate correlations
between all explanatory variables were computed on a crosstab. The crosstab
demonstrates that the majority of the statistically significant bivariate
correlations were low, and oscillated between 0.048 and 0.25. Only three
bivariate correlations exceed the 0.30 level.15 These are the correlations
between support for Zapatero and left-right politics (-0.398), support for Zapatero
and party politics (-0.443) and education and age (-0.451). It may be assumed
that, since support for Zapatero is highly correlated with both left-right politics
and party politics, it may be inflating the values of these two latter variables.
To test this presumption, support for Zapatero was excluded from the full
model and the respective regression coefficients of left-right politics and party
politics were measured again. The result indicates that despite the high
correlations, support for Zapatero did not inflate the values of left-right politics
and party politics. Excluding the variable support for Zapatero from the full
model changes the coefficient of party politics (1), i.e. the category with the
highest frequency used in the calculation of the fitted probabilities, from 2.823
to 2.814 and its standard error from 0.459 to 0.438. Under the same conditions,
the coefficient of the category of left-right politics with the highest frequency,
left-right politics (2), changes from 3.280 to 3.144, whilst its standard error
changes from 1.145 to1.083. These are rather small changes, where the
variations do not exceed 5.4%. Moreover, even after the exclusion of support
for Zapatero from the full model, the coefficients of constitutionphile(2),
EUphile(2) and identity(3) are still smaller than the coefficients of left-right
Electronic copy available at: https://ssrn.com/abstract=1632308
Ozgur Erkan
31
politics (2) and party-politics (1). These findings imply that support for Zapatero
does not inflate the effects of the aforementioned two second-order variables
vis-à-vis the first-order variables in the calculation of fitted probabilities.
With regard to the potential multicollinearity between education and age, it
was assumed that the variable age could be pulling down the effect of
education and making this variable statistically insignificant. Therefore, as
done above, age was excluded from the full model and the coefficients of
education were measured again. The findings demonstrated that excluding age
from the full model changed the statistical significance only in the second
category of education, which became statistically significant. Moreover, the
coefficients of four categories of education vary only between 0.087% and 4.5%,
when age is excluded from the full model. As a result, age neither pulls down
the effect of education notably nor makes the entire variable insignificant.
The ensuing findings from this data analysis have certain implications for
previous studies done on direct democracy in the EU as well as for
constitutional politics and democracy in the EU. The concluding section will
discuss these implications.
6. Conclusion
Scholars analysing the voting behaviour at national referenda on the ECT
have left the Spanish referendum under-researched as a case study. This
paper aims at correcting this anomaly by filling this gap in the literature and
by bringing an insight into the question whether in Spain domestic or
European factors had a greater influence on the votes at the referendum for
the ECT. To answer this question, firstly the hypotheses were constructed
based on the political events that took place in Spain prior to and during the
Electronic copy available at: https://ssrn.com/abstract=1632308
Spain’s Referendum
32
referendum. Thereafter, these hypotheses were tested with an application of
binary logistic regression to the available dataset. The test results in particular
the fitted probabilities, implied that the Spanish referendum on the ECT
possesses predominantly a second-order characteristic.
The finding, that the Spanish referendum on the ECT was a second-order
referendum, carries certain implications for existing research done in the
context of direct democracy in the EU. This finding initially displays the
validity of the Franklin thesis (Franklin et al. 1995) for the Spanish case. In line
with the Franklin thesis, votes cast at the Spanish referendum on the ECT
reflected predominantly attitudes on domestic political issues, rather than
attitudes towards the EU and the ECT. Hence, reasons for Spanish
referendum outcome lies in the specific dynamics of its political cycle as
compared to the other three countries where a referendum was hold.
The findings of this study also lend support to the hypothesis of Hug and
Scarini. Hug and Scarini assert that referenda which are non-binding, i.e.
consultative, are more likely to demonstrate second-order effects (in Garry et
al. 2005:204). Given that the Spanish referendum on the ECT was merely a
non-binding referendum (Torreblanca 2005:3); the finding that the Spanish
referendum on the ECT possesses a second-order characteristic lends support
to the hypothesis of Hug and Scarini.
The second-order characteristic of the Spanish referendum, on the other hand,
provides evidence against the hypothesis of Crum (2007). Crum posits that in
a collusive party system, where all mainstream parties take a proximate stand
on the issue of the referendum, the referendum tends to possess a first-order
characteristic (Crum 2007: 67). In competitive party systems, however, where
the government and opposition take diverging stands on the referendum, one
would expect to observe a second-order referendum (Ibid 2007:67). This is
Electronic copy available at: https://ssrn.com/abstract=1632308
Ozgur Erkan
33
because the votes cast at the referendum will be regarded either as a reward
or punishment for the government (Ibid:67). In the case of the Spanish
referendum on the ECT, one finds a “collusive party system”, based on the
model of Crum. In Spain, official party lines of both the governing PSOE and
the main opposition party PP endorsed the ECT. According to the hypothesis
of Crum, therefore, one would anticipate the Spanish referendum to possess a
first-order characteristic. Contrary to the Crum hypothesis, however, the
Spanish referendum on the ECT turns out to be a second-order referendum,
despite the existence of a collusive party system endorsing the ECT.
This study also carries certain implications for constitutional politics and
democracy in the EU. Firstly, the second-order characteristic of the Spanish
referendum is closely linked to the democratic deficit in the EU. According to
Føllesdal and Hix (2006:552), one facet of the democratic deficit in the EU lies
in the second-order nature of electoral campaigns for the EP elections. The
campaigns for EP elections, which are run by national political parties, refer to
domestic political cleavages, rather than the European issues (Ibid:552).
Accordingly, voters elect Members of the European Parliament (MEPs
hereafter) at EP elections based on domestic cleavages. Thus voter preferences
on European integration are not reflected in the votes cast at EP elections
(Ibid:552). At the EP, however, MEPs make decisions on issues pertaining to
Europe and European integration, despite being elected on domestic
cleavages. Since MEPs are not fully aware of the preferences of their citizens
on European issues, their decisions at the EP will not necessarily reflect the
preferences of their citizens on European integration. As a result, MEPs will
not be representing the interests of European citizens, which engenders a
democratic deficit at EU level.
This facet of democratic deficit in the EU can similarly be applied to secondorder
EU referenda, including the Spanish referendum on the ECT. Given the
Electronic copy available at: https://ssrn.com/abstract=1632308
Spain’s Referendum
34
finding that the Spanish referendum possesses a second-order characteristic,
Spanish voters’ preferences on European integration were not predominantly
reflected in their votes at the referendum. Consequently, if the ECT were to
come into force, actors and institutions at EU level, deriving their power from
the European Constitution, could take actions, which would not necessarily
reflect the preferences of Spaniards on European issues. Therefore, at least for
Spanish citizens, this incongruence of preferences on European issues would
engender a democratic deficit at EU level.
One other implication of this study can be found in the issue of increasing the
legitimacy of the EU and citizens’ loyalty to the EU through a European
Constitution. As shown in the introduction, an important motive for the
European political elites in preparing a Constitution for Europe was to
legitimise the European political order and increase the loyalty of European
citizens towards the EU (Castiglione in Castiglione et al. 2007:21-22;
Moravcsik in Meunier and McNamara 2007:37-38). The ensuing second-order
characteristic of the Spanish referendum, however, casts a shadow on the
viability of this goal. Given that Spanish votes at the referendum reflected
predominantly domestic cleavages, rather than citizens’ stance on European
integration or the ECT itself, the Spanish referendum did not serve to
legitimize the European order for Spaniards nor did it increase the loyalty of
Spaniards towards the EU. Hence, it is dubitable whether a Constitution for
Europe can be an indispensable panacea to bring citizens closer to the EU,
without eliminating the pattern of voting on domestic cleavages at EU
referenda.
This study also bears two implications for Spain. Firstly, the results of this
study confirm the strength of ethnic nationalism in Spain, as ethnic
nationalism in Spain proved strong enough to pull down the level of Yes
votes at an EU-level referendum. This outcome can possibly be interpreted
Electronic copy available at: https://ssrn.com/abstract=1632308
Ozgur Erkan
35
such that because the ECT did not significantly extend the direct
representational rights of ethnic minorities at EU institutions; historic nations
in EU members were distanced to the ECT. Secondly, this study demonstrates
the existence of a utilitarian support for European integration in Spain. As
shown through the explanatory variable Autonomous Communities, Spanish
regions under Objective 1 of EU Regional Funds expressed a greater support
for the ECT, holding other variables constant, compared to other
communities. Put differently, obtaining additional financial support from the
EU was found to correspond to a greater endorsement for further European
integration in Spain.
Finally, the results of this study raise new questions for further research. As
mentioned in the previous section, this study helps delineate the limitations
of Inglehart’s thesis at national referenda on the ECT. It could be explored
through further research, why older people endorsed the ECT more than
younger voters, contrary to the conventional view that younger cohorts are
more supportive of further EU integration (Hix 2005:162).
As shown in section 2, there is still one overlooked dimension of EU
referenda. The application of first and second-order elections framework to
EU referenda has so far left out the question, whether EU referenda have a
first or second-order characteristic across all social classes. Put more simply,
the interesting question is whether all social classes vote on domestic issues at
an e.g. second-order referendum or whether some classes still vote on
European issues, whilst other social classes vote on domestic issues. For
example, it would be interesting to know whether there is an elite-mass gap at
EU referenda, such that the more educated, wealthier citizens feel themselves
closer to the EU and thus vote on European issues whereas the less educated,
and those with lower incomes vote predominantly on domestic cleavages.
Electronic copy available at: https://ssrn.com/abstract=1632308
Spain’s Referendum
36
Further research could identify the patterns of electoral behaviour by
socioeconomic class at EU referenda.
Finally, another interesting research question to analyse could be to what
extent the media reporting on the ECT as well as on referendum campaigns
has influenced the electoral behaviour at referenda on the ECT in Spain,
Luxembourg, France and the Netherlands.
Endnotes
1 Some other EU members also planned to hold a referendum on the ECT. In the Czech Republic,
the referendum required constitutional amendments. Since the political polarisation rendered
reaching a consensus on constitutional amendments difficult, the proposal to hold a referendum
was receded in the Czech Republic (Schönlau and Castiglione in Castiglione et al. 2007:240).
Poland also planned holding a referendum, yet a referendum did not eventually take place in
Poland (Ibid:241). Finally, the British and Portuguese referenda were put on hold after the
ratification crisis of the ECT in France and the Netherlands (Ibid:241-242).
2 Turnout at the Spanish referendum on the ECT was low compared to national elections in Spain,
albeit was close to the turnout at 2004 European Parliament elections in this country [45.14%]
(Torreblanca 2005:2). Although low turnouts may create concerns on legitimacy in democracies,
the causes and consequences of this low turnout at the Spanish referendum are beyond the scope
of this paper and could be addressed in a separate study.
3 This trend continued with the rejection of the Lisbon Treaty in Ireland. Following the
ratification crisis in Ireland; Holmes (2008), O’Brennan (2009) and Quinlan (2009) analysed the
electoral behaviour at the Irish referendum on the Lisbon Treaty, which particularly focused on
explaining the motives behind No votes.
4 Given the dominance of Roman Catholics in Spain, hypothesis H3 refers to Catholics, rather than
Christians in general. In selecting Catholics as a measure, the study of Montero was taken as an
anchor, who examines the influence of Catholicism, rather than Christianity, on the votes in
Spanish politics (Montero 1994).
5 The reason for recoding religion into a dummy variable is to test hypothesis H3 (Catholic voters
expected to vote against the ECT). It could be argued that measuring the frequency at which
religious voters practice could be a more appropriate way to evaluate hypothesis H3. Another
Electronic copy available at: https://ssrn.com/abstract=1632308
Ozgur Erkan
37
variable in the dataset, which included this measure, however, indicated the frequency of
practising both for Catholics and believers of other religions. Hence, it would not be the
appropriate variable to test hypothesis H3.
6 All subsequent comparisons in section 5 are done with regard to the respective reference
category of each explanatory variable.
7 In 2006, the top four Spanish regions with the highest GDP per capita were -in descending
order- 1.) Navarra 2.) the Basque Country 3.) Madrid 4.) Catalonia.
8 This figure was obtained through the calculation of odd ratios as follows: The ratio of a centrist
voter over a centre-right voter is Left-right politics (3)/left-right politics (4), which is equal to
left-right politics(3)/reference category * reference category/left-right politics(4). The first
expression is already given in SPSS output for the partial model through Exp(B) of left-right
politics(3), which is 23.499. The second expression is 1/ExpB [left-right politics(4)], which is
0.0450. The multiplication of these two factors gives 23.499*0.0450= 1.0575. As a percentage,
this means (1.0575-1)*100= 5.75%.
9 The odd ratio was calculated in an identical way as for the endnote 8 above. Only the formula
changes here, which is now: Left-right politics(3)/reference category * reference category/leftright
politics(2).
10 PSOE and CC had a strongly positive party line towards the ECT. In contrast, except one statewide
party, IU, mainly peripheral and nationalist parties had a strongly negative party line
towards the ECT. These parties were ERC, BNG, EA, Naforra-Bai and CHA-Aragon (Malo de Molina
and Miguel de Elias 2005:78; Torreblanca and Sorroza 2006:1).
11 These parties were PP, PNV and CiU. The PP had internal fragmentations regarding the content
of the ECT, as some party members bemoaned the absence of Christian values in the ECT as well
as the loss of Spanish votes at the Council compared to the formulae under the Nice Treaty (Malo
de Molina and Miguel de Elias 2005:80). Despite these fragmentations, the PP decided to endorse
the ECT, as it was the Spanish negotiator of the ECT at the time of Aznar government. PNV and
CiU did not welcome the ECT with great enthusiasm (Eschke and Malick 2006:12), as the ECT did
not promote new rights of representation for ethnic minorities at European institutions, yet the
party lines were nevertheless to support the ECT.
12 This figure was obtained through the exponential of the regression coefficient: Exp(-1.317)=
0.2679. To convert that into a percentage, one calculates (1-0.2679)*100= 73.21%.
13 The formula used for calculating the fitted probabilities is: Logit = Constant + 40*age +
Education(3) * 1 + Autonomous Community(1) * 0 + Autonomous Community(2) * 0 + Left-Right
Politics(2) * 1 + Party Politics(1) * 1 + Support for Zapatero(1) * 0 + Economic Activity(1) * 1 +
Electronic copy available at: https://ssrn.com/abstract=1632308
Spain’s Referendum
38
Religion(1)*0 + Gender(1) * 1 + EUphile(2) * 1 + Constitutionphile (2) *1 + Identity (3) * 1. The
obtained logit is fitted into the formula [1/1+exp(-logit)]*100, which gives the fitted probability
in percentages.
The formula above gives the fitted probability when all explanatory variables attain the value of
their category with the highest frequency. The reason for selecting the highest frequency for each
variable is to render the fitted probabilities representative of the entire population. Since this
formula incorporates both first and second-order variables, the calculation uses the regression
coefficients from the full model.
To calculate the individual impact of first and second-order variables on voting behaviour at the
Spanish referendum for the ECT, the variable in question was given the value at its reference
category and inserted into the formula above. Thereafter, this variable was given the value of its
category with the highest frequency, whilst holding all other variables constant at their category
with the highest frequency and the calculation was run again. The difference between these two
values gives the individual impact of this variable on voting behaviour.
To calculate the collective impact of aggregated first-order variables on voting behaviour at the
Spanish referendum, firstly all first-order variables EUphile, Constitutionphile and Identity were
allocated to their reference category and fitted into the formula above. Thereafter, all these
variables were given their value with the highest frequency, whilst holding all other variables
constant at their category with the highest frequency and the calculation was rerun. The
difference between these two values gives the impact of all first-order variables on voting in
favour of the ECT. This process was repeated to calculate the collective impact of aggregated
second-order variables on the response variable.
14 Since for Support for Zapatero the reference category is also the category with the highest
frequency, a special row was inserted for this variable into Table 3 with its value at 0 and 1 as a
dummy variable. Therefore, in calculating the effect of Support for Zapatero on the response
variable, exceptionally the change from its first category to its second category was calculated,
rather than the change from its reference category to the category with the highest frequency.
This exception, however, does not affect the statistical inference or interpretation.
15 This benchmark for measuring the level of bivariate correlations was borrowed from the
regression analysis of Glencross and Trechsel (2007:8).
Electronic copy available at: https://ssrn.com/abstract=1632308
Ozgur Erkan
39
Bibliography
Primary Source(s)
CIS Estudio no. 2595 – Postreferendum sobre la Constitución Europea, Febrero-Marzo 2005 [CIS
Study No 2595 – Post referendum Survey on the EU Constitutional Treaty, February-March
2005], conducted by Centro de Investigaciones Sociológicas (CIS), Madrid.
Data obtained from CIS, Montalbán 8, 28014 Madrid on 25 February 2009
Flash Eurobarometer-168,“European Constitution: Post-referendum Survey in Spain”, Gallup
Europe. Data obtained from GESIS, Bachemer Str. 40 D-50931, Cologne on 3 February 2009. Only
the ‘questionnaires’ attachment of Flash Eurobarometer-168 was used to compare the survey
questions of Flash Eurobarometer-168 with the survey questions in CIS survey 2595.
Secondary Sources
Castiglione, Dario (2007) “Constitutional Politics in the European Union” in Castiglione, Dario;
Schönlau, Justus; Longman, Chris; Lombardo, Emanuela; Pérez-Solórzano Borragán, Nieves; and
Aziz, Miriam “Constitutional Politics in the European Union: the Convention Moment and its
Aftermath”, Basingstoke ; New York : Palgrave Macmillan
Closa, Carlos (2004) “The Formation of Domestic Preferences on the EU Constitution in Spain”
Comparative European Politics, Vol. (2) pp.320–338
Crum, Ben (2007) “Party Stances in the Referendums on the EU Constitution: Causes and
Consequences of Competition and Collusion”, European Union Politics, 8 (1), pp. 61-82
Dehousse, Renaud (2006) “The Unmaking of a Constitution: Lessons from the European
Referenda”, Constellations, 13 (2), pp. 151-64
Dumont, Patrick and Poirier, Philippe (2006a) “Luxembourg” European Journal of Political
Research, 45 (7-8), pp. 1182-1197
Dumont, Patrick and Poirier, Philippe (2006b) “The July 10th Referendum on the EU
Constitutional Treaty in Luxembourg” in Garcia, J .M. Talero (ed.) “European Yearbook of Political
Campaigns 2005”, Vienna: European Association of Political Consultants
Eschke, Nina and Malick, Thomas (eds.) (2006) “The European Constitution and its
Ratification Crisis: Constitutional Debates in the EU Member States” ZEI Discussion Paper C156,
Zentrum für Europäische Integrationsforschung, Universität Bonn. Accessed on 28 June 2009.
Available at: http://www.zei.de/download/zei_dp/dp_c156_eschke_malick.pdf
European Union (2005) “Provisions and Instruments of Regional Policy - Objective 1”. Last
updated 14.07.2005. Accessed on 1 July 2009. Available at legislation summaries - EUROPA
portal of the European Union:
http://europa.eu/legislation_summaries/regional_policy/provisions_and_instruments/g24203
_en.htm#top#top
Føllesdal, Andreas and Hix, Simon (2006) “Why There is a Democratic Deficit in the EU: A
Response to Majone and Moravcsik”, Journal of Common Market Studies, 44 (3), pp.533-562
Franklin, Mark; van der Eijk, Cees and Marsh, Michael (1995) “Referendum Outcomes and
Trust in Government: Public Support for Europe in the Wake of Maastricht” West European
Politics, 18 (3), pp. 101-117
Electronic copy available at: https://ssrn.com/abstract=1632308
Spain’s Referendum
40
Garry, John; Marsh, Michael and Sinnott, Richard (2005) “Second-Order versus Issue-Voting
Effects in EU Referendums: Evidence from the Irish Nice Treaty Referendums”, European Union
Politics, 6 (2), pp. 201-221
Glencross, Andrew and Trechsel, Alexander H. (2007) “First or Second Order Referendums?
Understanding the Votes on the Constitutional Treaty in Four EU Member States”, paper
presented at the International Conference on Direct Democracy in Latin America, Buenos Aires,
14-15 March 2007. Accessed on 18 December 2008. Available at: http://www.dd-
la.ch/download/Glencross_Trechsel.pdf
Harmsen, Robert (2005) “The Dutch Referendum on the Ratification of the European
Constitutional Treaty, 1 June 2005”, European Parties Elections and Referendums Network
Referendum Briefing Paper, No.13, University of Sussex. Accessed on 31 July 2009. Available at:
http://www.sussex.ac.uk/sei/1-4-2-9.html
Hausemer, Piere (2005) “Luxembourg’s Referendum on the European Constitutional Treaty”,
10 June 2005, European Parties Elections and Referendums Network Referendum Briefing Paper,
No.14, University of Sussex. Accessed on 14 July 2009. Available at:
http://www.sussex.ac.uk/sei/1-4-2-9.html
Hix, Simon (2005) “The Political System of the European Union”, 2nd ed., Houndmills,
Basingstoke, Hampshire; New York: Palgrave Macmillan
Holmes, Michael (2008) “Referendum Briefing Paper No 16: The Referendum on the Treaty of
Lisbon in the Republic of Ireland, 12 June 2008” European Parties Elections and Referendums
Network Referendum Briefing Paper, No.16, University of Sussex. Accessed on 19 July 2009.
Available at: http://www.sussex.ac.uk/sei/1-4-2-9.html
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), pp. 47-69
Jiménez, Antonia María Ruiz and Sampedro, Víctor (2005) “La Constitución Europea en la
prensa española de ámbito nacional: un análisis comparativo de El País, El Mundo y ABC”,
Documento de Trabajo 1/2005, Real Instituto Elcano, Madrid. Accessed on 13 March 2009.
Available at: http://www.realinstitutoelcano.org/documentos/164/URJC%20pdf.pdf
Malo de Molina, Carlos and Miguel de Elías, José (2005) “Análisis del voto de los españoles en
el Referéndum sobre la Constitución Europea”, Revista Valenciana d’Estudis Autonomics, No:
43/44. Accessed on 10 March 2009. Available at:
http://www1.pre.gva.es/argos/fileadmin/argos/datos/RVEA/libro_43_44/74-43_44.pdf.
Marthaler, Sally (2005) “The French Referendum on Ratification of the EU Constitutional
Treaty, 29 May 2005”, Representation, 41 (3), pp. 230-239
Medrano, Juan Díez (1995) “Opiniónes y Actitudes Informes – La Opinión Pública Española y la
Integración Europea 1994: Resultados del Estudio 2084” Centro de Investigaciones Sociológicas
(CIS), Madrid
Montero, José Ramón (1994) “Religiosidad, Ideología y Voto en España”, Revista de Estudios
Políticos, No 83, pp. 77-112. Accessed on 1 August 2009. Available at:
http://dialnet.unirioja.es/servlet/articulo?codigo=27256
Moravcsik, Andrew (2007) “The European Constitutional Settlement” in Meunier, Sophie and
McNamara, Kathleen R. “Making History: European Integration and Institutional Change at Fifty”,
The State of the European Union Volume 8, Oxford, UK ; New York : Oxford University Press,
2007
Electronic copy available at: https://ssrn.com/abstract=1632308
Ozgur Erkan
41
O’Brennan, John (2009) “Ireland says No (again): the 12 June 2008 Referendum on the Lisbon
Treaty” Parliamentary Affairs 62 (2), pp. 258–277
Quinlan, Stephen (2009) “The Lisbon Treaty Referendum” Irish Political Studies 24 (1), pp. 107–
121
Reif, Karlheinz and Schmitt, Hermann (1980) “Nine Second-Order National Elections – A
Conceptual Framework for the Analysis of European Election Results”, European Journal of
Political Research, 8 (1), pp. 3-44
Schönlau, Justus and Castiglione, Dario (2007) “From the Convention to the Referendums and
Beyond” in Castiglione, Dario; Schönlau, Justus; Longman, Chris; Lombardo, Emanuela; PérezSolórzano
Borragán, Nieves; and Aziz, Miriam “Constitutional Politics in the European Union: the
Convention Moment and its Aftermath”, Basingstoke; New York: Palgrave Macmillan
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), pp. 733-750
Taggart, Paul (2006) “Questions of Europe – The Domestic Politics of the 2005 French and
Dutch Referendums and their Challenge for the Study of European Integration”, Journal of
Common Market Studies, 44 (Annual Review), pp. 7-25
Torreblanca, José. I. (2005) “Spain’s Referendum on EU Constitution: A Double
Disappointment” ARI No: 27/2005, 21/2/2005, Real Instituto Elcano. Accessed on 17 March 2009.
Available at: http://www.realinstitutoelcano.org/analisis/701/Torreblanca701.pdf.
Torreblanca, José I. and Sorroza, Alicia (2005) “Principales datos sobre el proceso de
ratificación de la Constitución Europea en España”, Documento de Trabajo, 8/2005, Real Instituto
Elcano. Available at: http://www.realinstitutoelcano.org/documentos/174.asp. Accessed on 9
March 2009.
Torreblanca, José I. and Sorroza, Alicia (2006) “Spain and the European Union: Country
Update for the EU/25 Watch” Europe- ARI 28/2006, 1/3/2006, Real Instituto Elcano. Accessed on
20 July 2009. Available at:
http://www.realinstitutoelcano.org/analisis/921/921_TorreblancaSorroza_Spain_EU_EU25Watc
h.pdf
Electronic copy available at: https://ssrn.com/abstract=1632308
Electronic copy available at: https://ssrn.com/abstract=1632308
Recent LEQS papers
White, Jonathan. ‘Left, Right and Beyond: The Pragmatics of Political Mapping.’ LEQS Paper No. 24,
June 2010
Monastiriotis, Vassilis & Zartaloudis, Sotirios. 'Beyond the crisis: EMU and labour market
reform pressures in good and bad times.' LEQS Paper No. 23, June 2010
Lütz, Susanne & Kranke, Matthias. ‘The European Rescue of the Washington Consensus? EU and IMF
Lending to Central and Eastern European Countries.’ LEQS Paper No. 22, May 2010
Hartlapp, Miriam; Metz, Julia & Rauh, Christian. 'The agenda set by the EU Commission: the result of
balanced or biased aggregation of positions?' LEQS Paper No. 21, April 2010
Costa-i-Font, Joan. 'Unveiling Vertical State Downscaling: Identity and/or the Economy?' LEQS Paper
No. 20, March 2010
Delanty, Gerard. 'The European Heritage from a Critical Cosmopolitan Perspective. LEQS Paper No. 19,
February 2010
Outhwaite, William. 'Europe at 21: Transitions and Transformations since 1989'. LEQS Paper No. 18,
January 2010
Lavdas, Kostas A..'Normative Evolution in Europe: Small States and Republican Peace'. LEQS Paper
No. 17, January 2010
Schelkle, Waltraud. ‘Good governance in crisis or a good crisis for governance? A comparison of the EU
and the US’. LEQS Paper No. 16, December 2009
Keating, Michael. ‘Second Round Reform. Devolution and constitutional reform in the United
Kingdom, Spain and Italy’. LEQS Paper No. 15, December 2009
Hyman, Richard. ‘Trade Unions and ‘Europe’: Are the Members out of Step?’. LEQS Paper No. 14,
November 2009
Dani, Marco. ‘Economic and social conflicts, integration and constitutionalism in contemporary
Europe’. LEQS Paper No. 13, November 2009
Rodríguez-Pose, Andrés & Krøijer, Anne. ‘Fiscal Decentralization and Economic Growth in Central and
Eastern Europe’. LEQS Paper No. 12, October 2009
Cheshire, Paul C. & Magrini, Stefano. ‘Urban Growth Drivers and Spatial Inequalities: Europe - a Case
with Geographically Sticky People. LEQS Paper No. 11, October 2009
McCrea, Ronan. ‘The Recognition of Religion within the Constitutional and Political Order of the
European Union’. LEQS Paper No. 10, September 2009
Bertola, Guiseppe & Mola, Lorenza. ‘Services Provision and Temporary Mobility: Freedoms and
Regulation in the EU’ LEQS Paper No. 9, September 2009
Walker, Neil. ‘Multilevel Constitutionalism: Looking Beyond the German Debate’. LEQS Paper No. 8,
June 2009
Frankel, Jeffrey. ‘The Estimated Trade Effects of the Euro: Why Are They Below Those From Historical
Monetary Unions Among Smaller Countries?’. LEQS Paper No. 7, June 2009
Electronic copy available at: https://ssrn.com/abstract=1632308
Spain’s Referendum
44
LEQS
European Institute
London School of Economics
Houghton Street
WC2A 2AE London
Email: euroinst.LEQS@lse.ac.uk
http://www2.lse.ac.uk/europeanInstitute/LEQS/Home.aspxElectronic copy available at: https://ssrn.com/abstract=1632308