Repeated Crisis Exposure, Euroskepticism & Political Behavior: An Econometric Analysis for European Countries (BestMasters) 3658392665, 9783658392666

Citation preview

BestMasters

Lukas Möller

Repeated Crisis Exposure, Euroskepticism & Political Behavior An Econometric Analysis for European Countries

BestMasters

Mit „BestMasters“ zeichnet Springer die besten Masterarbeiten aus, die an renommierten Hochschulen in Deutschland, Österreich und der Schweiz entstanden sind. Die mit Höchstnote ausgezeichneten Arbeiten wurden durch Gutachter zur Veröffentlichung empfohlen und behandeln aktuelle Themen aus unterschiedlichen Fachgebieten der Naturwissenschaften, Psychologie, Technik und Wirtschaftswissenschaften. Die Reihe wendet sich an Praktiker und Wissenschaftler gleichermaßen und soll insbesondere auch Nachwuchswissenschaftlern Orientierung geben. Springer awards “BestMasters” to the best master’s theses which have been completed at renowned Universities in Germany, Austria, and Switzerland. The studies received highest marks and were recommended for publication by supervisors. They address current issues from various fields of research in natural sciences, psychology, technology, and economics. The series addresses practitioners as well as scientists and, in particular, offers guidance for early stage researchers.

Lukas Möller

Repeated Crisis Exposure, Euroskepticism & Political Behavior An Econometric Analysis for European Countries

Lukas Möller Münster, Germany

ISSN 2625-3577 ISSN 2625-3615 (electronic) BestMasters ISBN 978-3-658-39266-6 ISBN 978-3-658-39267-3 (eBook) https://doi.org/10.1007/978-3-658-39267-3 © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2022 This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors, and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Responsible Editor: Marija Kojic This Springer Gabler imprint is published by the registered company Springer Fachmedien Wiesbaden GmbH, part of Springer Nature. The registered company address is: Abraham-Lincoln-Str. 46, 65189 Wiesbaden, Germany

Acknowledgments

The author wishes to express his deepest gratitude towards his supervisors, Prof. Dr. Thomas Apolte and Dr. Kim Leonie Kellermann. The scientific support, the reliability and the counseling during the complete research and writing process of the present study have been extraordinary. Furthermore, this study benefited a lot from all the advice given by the colleagues at the Chair of Political Economy, namely Dr. Lena Gerling, Helena Helfer, Julia Jänisch and Dr. Manuel Santos Silva, as well as the participants of the Chair’s research seminar and its student assistants. Many thanks to my family who influenced the study by their constant support. The biggest thanks, however, go to my fiancée: For ongoing encouragement, humor and enduring patience.

v

Contents

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1

2 Theoretical Foundation & Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Conceptual Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Crises & Countermeasures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3 3 5 7

3 Data & Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Dependent Variable: Euroskepticism . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Explanatory Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.1 Eurocrisis: Austerity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.2 Migration Crisis: Net Migration . . . . . . . . . . . . . . . . . . . . . . . 3.2.3 Corona Crisis: Infections & Casualties . . . . . . . . . . . . . . . . . 3.3 Control Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

9 9 12 13 14 16 17

4 Econometric Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1 Identification Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Composition of the Regressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

21 21 23

5 Regression Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1 Single Crisis Exposure Regressions . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1.1 Baseline Regressions (Single Crisis) . . . . . . . . . . . . . . . . . . . 5.1.2 Placebo Tests (Single Crisis) . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1.3 Robustness Checks & Driver Tests (Single Crisis) . . . . . . . 5.2 Repeated Crisis Exposure Regressions . . . . . . . . . . . . . . . . . . . . . . . . 5.2.1 Baseline & Placebo Regressions (Repeated Exposure) . . . 5.2.2 Driver Tests (Repeated Exposure) . . . . . . . . . . . . . . . . . . . . .

27 27 28 33 37 46 47 51

vii

viii

Contents

6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.1 Fulfillment of the Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2 Potential Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

55 55 57

7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

63

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

65

Abbreviations

AT BE BG CI CY CZ DE DiD dk DK DV EB EE EES EP ES ESS EU FE FI FR GDP GESIS

GIIPS

Austria Belgium Bulgaria Confidence interval(s) Cyprus Czech Republic Germany Differences-in-Differences Don’t know (Survey option) Denmark Dependent Variable(s) Eurobarometer(s) Estonia European Election Study European Parliament Spain European Social Survey European Union Fixed Effect(s) Finland France Gross Domestic Product Gesellschaft Sozialwissenschaftlicher Infrastruktureinrichtungen— Leibniz-Institut für Sozialwissenschaften (German Social Science Infrastructure Services—Leibniz Institute for the Social Sciences) Greece, Ireland, Italy, Portugal, Spain

ix

x

GIIPS+ GR HE HR HU IE IT JRC LT LU LV MT N NL NUTS OECD OLS PL PPP PT RO SE SI SK UK USD WCRB Y

Abbreviations

Greece, Ireland, Italy, Portugal, Spain + Cyprus, Hungary, Latvia, Romania Greece High Exposure Croatia Hungary Ireland Italy European Commission Joint Research Center Lithuania Luxembourg Latvia Malta No (Survey option) The Netherlands Nomenclature des Unités Territoriales Statistiques (Nomenclature of Territorial Units for Statistics) Organization for Economic Co-operation and Development Ordinary Least Squares Poland Purchasing Power Parity Portugal Romania Sweden Slovenia Slovakia United Kingdom United States Dollar Wild Cluster Restricted Bootstrap Yes (Survey option)

List of Figures

Figure 4.1 Figure 4.2 Figure 4.3 Figure 4.4 Figure 5.1 Figure 5.2

Equation for the baseline approach for the Eurocrisis . . . . . . . Equation for the baseline approach for the migration crisis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Equation for the baseline approach for the corona crisis . . . . Equation for the baseline approach for the repeated crisis exposure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Repeated crisis exposure: Average marginal effects . . . . . . . . Repeated crisis exposure (driver tests): Average marginal effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

24 24 25 25 49 52

xi

List of Tables

Table Table Table Table Table

5.1 5.2 5.3 5.4 5.5

Baseline regression results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Placebo tests results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Dummy robustness checks results (Single crisis) . . . . . . . . . . . Continuous robustness checks results . . . . . . . . . . . . . . . . . . . . . Driver tests (Single crisis) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

31 35 39 42 44

xiii

1

Introduction

The present study examines potential relationships between Euroskeptic attitudes and a single or repeated exposure to severe crises. During times of crisis, a growing number of people tends to blame the European Union (EU) for the crisis impact on their personal lives. Euroskeptic political behavior is therefore subject to current research: Several papers investigate the impact of crises on individual views concerning the EU (e.g., Baimbridge 2018; Flinders 2021; Göncz/Lengyel 2021; Taggart/Szczerbiak 2018). The present paper adds to this strand of literature, insofar as it looks not only at a single but a repeated crisis exposure. To this end, it analyzes the impact of an exposure to the Eurocrisis, the migration crisis and the coronavirus crisis. Harteveld et al. (2018) show that people differentiate between various responsibility levels in multi-level governance structures such as the EU. Therefore, we adopt the argumentation of the blame game literature that national level actors tend to shift the blame for negative crisis outcomes onto the European level (e.g., Giugni/Grasso 2019; Harteveld et al. 2018; Heinkelmann-Wild/Kriegmair, et al. 2020; Heinkelmann-Wild/Zangl 2019; Heinkelmann-Wild/Zangl, et al. 2021). This shift works enlarging on the partly preexisting Euroskeptic tendencies. The aim of this study is thus to understand both the single and joint impacts of these recent crises on contemporary political behavior. Accordingly, we develop

© The Author(s), under exclusive license to Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2022 L. Möller, Repeated Crisis Exposure, Euroskepticism & Political Behavior, BestMasters, https://doi.org/10.1007/978-3-658-39267-3_1

1

2

1

Introduction

strategies in order to investigate links between crisis-related factors and potential Euroskepticism. The underlying empirical analysis focuses on individuals from all member states of the EU1 . In addition to the novel simultaneous focus on single and repeated crisis exposure, this paper contributes to the literature through a manner, in which the way how Euroskepticism is defined and measured in the baseline case is not only based on two, as most papers do, but on three factors (cf. Vasilopoulou 2018). This aims at broadening the data base in order to achieve a more conservative estimation by attempting to capture more aspects which are relevant for the existence of Euroskepticism. The index combines the individual’s view on the EU membership benefit for their country, the overall EU image and their personal identification as Europeans. The findings support the view that Euroskepticism is exacerbated when people are exposed to a crisis. Concerning repeated crisis exposure, the findings are less clear: The further a crisis recedes into the past, the less impact it has on today’s political behavior; only having been exposed to both migration and corona crises can be seen as clearly increasing Euroskepticism. Additionally, people from crisisridden countries develop a resilience, preparing them for the next crisis exposure. Finally, we found a hysteresis-like phenomenon concerning Euroskepticism levels: After a crisis has been overcome, the Euroskepticism level does not return to the precrisis level but stays at the new high. As the econometric methods that we applied, such as bootstrapping and entropy balancing, are rather restrictive, the findings are conservative. This paper is structured as follows: First, the theoretical background is presented and the hypotheses stated, before the data set, based on the Eurobarometer (EB), and the applied econometric methods are displayed. Afterwards, the regression results are presented and discussed, before the paper is concluded and an outlook on further research is given. Finally, the list of the used literature follows. Furthermore, an appendix exists as electronic supplementary material containing additional tables and equations as well as a code book.

1

The following countries are included in this study: Austria (AT), Belgium (BE), Bulgaria (BG), Croatia (HR), Cyprus (CY), Czech Republic (CZ), Denmark (DK), Estonia (EE), Finland (FI), France (FR), Germany (DE), Greece (GR), Hungary (HU), Ireland (IE), Italy (IT), Lithuania (LT), Luxembourg (LU), Latvia (LV), Malta (MT), the Netherlands (NL), Poland (PL), Portugal (PT), Romania (RO), Slovakia (SK), Slovenia (SI), Spain (ES), Sweden (SE) and the United Kingdom (UK) prior to Brexit (cf. Publications Office of the EU 2021).

2

Theoretical Foundation & Hypotheses

In this section, the theoretical background of this study is presented. In a first step, the central theoretical concepts are classified, especially the phenomenon of Euroskepticism. In a second step, the crises which are under investigation are introduced and potential reasons why these crises may have led to an exacerbation of Euroskepticism are described. Additionally, governmental countermeasures to fight the impacts of the crises are laid out. Finally, the examined hypotheses of this study are stated.

2.1

Conceptual Framework

A clear distinction on how this study understands Euroskepticism is crucial, because a universal definition does not exist. Taggart (1998) was first to classify Euroskepticism as opposition towards the European integration and towards the coalescence of the European countries. It manifests itself in constructive or unsubstantiated criticisms (cf. Mudde 2012; Taggart 1998). Many scholars see a link between Euroskepticism and European populism, both in extreme left and extreme right parties (e.g., cf. Guriev/Papaioannou 2020; Mudde 2012; Rodrik 2018; Rooduijn et al. 2019; Taggart 2020). Views on the causes of Euroskepticism differ as well, yet mostly individual identification or perceived economic benefits from EU membership are indicated (cf. Loveless/Rohrschneider 2011). Euroskepticism developed especially in crisis-ridden countries which had to implement strict countermeasures (cf. Guiso et al. 2019; Taggart 1998; Vasilopoulou 2018).

Supplementary Information The online version contains supplementary material available at https://doi.org/10.1007/978-3-658-39267-3_2.

© The Author(s), under exclusive license to Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2022 L. Möller, Repeated Crisis Exposure, Euroskepticism & Political Behavior, BestMasters, https://doi.org/10.1007/978-3-658-39267-3_2

3

4

2 Theoretical Foundation & Hypotheses

The understanding of Euroskepticism in this study is a skepticism towards European institutions, first and foremost the EU, analogously to the views of Serricchio et al. (2013) and Vasilopoulou (2018). A simple criticism of the EU, a skepticism towards the European continent or towards the common European currency might be similar to the definition used here, but not necessarily (cf. Mudde 2012; Taggart 1998). The relationship between the EU and its citizens in the first decades of its existence has been described as “permissive consensus” (Inglehart 1970; Lindberg/Scheingold 1970): A majority of acting politicians has been convinced that the European integration is an uncontested aim of all citizens. Simultaneously, however, European topics have been seen as less relevant than national ones, both by most politicians and citizens (often still to this day) and national topics tend to be prioritized (cf. Ehin/Talving 2021; Hooghe/Marks 2009; Taggart 2020). Following from that, the electoral turnout in elections of the European Parliament (EP) tends to be lower than in national elections, as voters see it as secondary (cf. Ehin/Talving 2021). Nevertheless, since the 1990s a “constraining dissensus” (Hooghe/Marks 2009; Schimmelfennig 2020) can be found: Politicians have to check domestic public opinion when dealing with European topics. The salience of European topics not only in pan-European elections has grown and has especially been reinforced by the exposure to pan-European crises (cf. Ehin/Talving 2020, 2021). At the latest with the start of the Eurocrisis, all European citizens have been forced to develop their own opinion about pan-European subjects such as European integration because of its relevance also in national elections (cf. Grande/Hutter 2016; Mudde 2012; Schäfer/Debus 2018; Schäfer/Gross 2020). This led to more and more visible Euroskeptic parties (cf. Taggart 2020). The visibility of the EU in everyday life grows and this might affect respondents in various directions, while perceiving the pros and cons of the European integration (cf. Grande/Hutter 2016; Mudde 2012). The traditional concept of “economic voting” (Buchanan/Wagner 1977), describing that people reward incumbents in economically good times and punish them in adverse times, has to be taken into account, as well. Due to the multilevel governance when European topics are at stake, the responsibility and accountability is less easily visible for the citizens (cf. Talving 2018). Entangling the effects stemming from the crisis itself and those from the governmental countermeasures is challenging for the voters (cf. Duch/Stevenson 2010). If voters understand the urgency of the countermeasures, they accept even painful restrictions for a reasonable time span (cf. Alesina/Favero, et al. 2019; Alesina/Perotti, et al. 1998; Arias/Stasavage 2019; Talving 2017). The competence to distinguish crisis impacts from bad measures

2.2 Crises & Countermeasures

5

grows with the number of experienced crises (cf. Talving 2017). Voters evaluate the past performance, sanction if needed and try to vote for the most competent government for the future: All available information is used for that (cf. Duch/Stevenson 2010; Hobolt/Tilley 2016). Survey answers tend to be more radical than voting decisions, as respondents understand that their survey answers do not have an impact, and finally choose to vote in a less radical way: Research based on surveys thus could be biased, insofar as the respondents’ true political opinion does not show, at least not entirely (cf. Downs 1957; Lewis-Beck/Lobo 2017; Reinl 2020; Riker/Ordeshook 1968). This has to be kept in mind, as the present study concentrates on surveys that register how people intend to vote and what policies they favor. Crisis exposure is found to have a different impact on people’s attitudes depending on the time of life when the crisis happens (cf. Malmendier/Nagel 2011). Nevertheless, for all people, crisis exposure can alter the political positions, not only in the short term but also in the long term (cf. Cogley/Sargent 2008; Giuliano/Spilimbergo 2014). If—after the crisis is overcome—the triggered political attitude stays at a higher level than pre-crisis or even enters a new and potentially upward-sloping growth path, a hysteresis-like effect can be postulated (cf. Mota et al. 2020). Longer gone crises can thus have a long-term influence, as well (cf. Malmendier/Nagel 2011).

2.2

Crises & Countermeasures

The global financial crisis of 2008, which started in the United States of America, hit countries all over the world, in Europe especially ES, IE, IT and GR (cf. Jordà et al. 2015; Hobolt/Tilley 2016; Karanikolos et al. 2013; Serricchio et al. 2013). Nevertheless, from most citizens’ point of view, the national level was the one which had to take action and thus the direct impact on Euroskepticism of this crisis is found to be small (cf. Serricchio et al. 2013). For a crisis to be effective on Euroskepticism, whether upward- or downward-sloping, it has to be obvious for the citizens that not only their national governments but predominantly the European institutions are involved in designing countermeasures. After the global financial crisis, the Eurocrisis emerged in Europe (cf. Talving 2017). The worst hit countries have been ES, GR, IE, IT and PT, often taken together as GIIPS (cf. House et al. 2020). The countries of CY, HU, LV and RO have been suffering severely from the Eurocrisis, as well, therefore, several researchers explain that these countries belong to the group of the highly exposed countries, which is

6

2 Theoretical Foundation & Hypotheses

followed in several regressions of this paper (cf. Arias/Stasavage 2019; Baimbridge 2018; Hobolt/Tilley 2016). This expanded highly exposed group is called GIIPS+. It has to be taken into account, however, that the severity of the crisis differed between these countries (cf. Engler/Klein 2017; Natali/Stamati 2014). It is never an easy task to decide on the exact time span of a crisis, in the Eurocrisis case especially for its end: Several scholars locate the start in late 2009, and 2013 is seen as end, yet still in 2015 crisis-related events took place in GR (e.g., the electoral win of the radical left anti-austerity party Syriza; cf. Funke et al. 2016; Karyotis/Rüdig 2018; Talving 2017). This study defines the years 2012 and 2013 as peak time of the Eurocrisis, in order to get a fair trade-off between all highly exposed countries. There are several research papers dealing with the potential nexus between the Eurocrisis and the development of Euroskepticism, e.g., Baimbridge (2018), Crespy (2020), Hobolt/Tilley (2016) or Lewis-Beck/Lobo (2017): They found that the crisis triggered the growth of Euroskeptic anti-austerity parties and that the crisis accelerated economic voting patterns. Therefore, this crisis is taken as the first crisis which is investigated in this study on Euroskepticism. Austerity has been a central countermeasure of governments fighting the Eurocrisis, through spending less or taxing more (cf. Alesina/Favero, et al. 2019). As most governments tend to be risk averse, they tend to wait as long as possible before installing austerity measures; they implement them only if they are externally forced to do so and/or expect to survive the potentially following economic voting (cf. Alesina/Carloni, et al. 2011; Alesina/Favero, et al. 2019; Alesina/Perotti, et al. 1998; Arias/Stasavage 2019; Funke et al. 2016; Ponticelli/Voth 2020; Talving 2018). Especially when external shocks have fallen together with bad political decisions, austerity measures are used to reassure creditors of the country’s trustworthiness (cf. Alesina/Favero, et al. 2019). Funke et al. (2016) show that austerity measures influence political behavior for a time span of five to ten years, materializing in more and more fragmented parliaments, especially in the form of gains on the extreme right political spectrum. It is challenging to clearly distinguish effects from the crisis and from the taken countermeasures, leading to a potential endogeneity or reverse causality problem (cf. Alesina/Favero, et al. 2019). For the second relevant crisis for this study, the migration crisis in Europe, the peak time was 2015, when many of the highly exposed countries still suffered from the aftermath of the Eurocrisis (cf. Bandeira et al. 2019; Karyotis/Rüdig 2018). Mostly people fleeing from war in their home countries wanted to reach the EU (cf. Otto/Steinhardt 2017). For this crisis, the role of the EU was again central, because the EU member states try to organize their migration policies jointly, using the pan-European agency Frontex and the “Dublin system” (European Commission 2015; Frontex 2016; Hutter/Kries 2021). Papers by Pirro et al. (2018) or by

2.3 Hypotheses

7

Taggart/Szczerbiak (2018) are able to show a link between this crisis and Euroskepticism. Even at the beginning of the third central crisis, the corona crisis, in 2020, some researchers saw that for the GIIPS+ countries some negative impacts of the Eurocrisis have still been unsolved (cf. Crespy 2020). When writing this study, the corona crisis has not yet been completely overcome, but still the year 2020 is seen as the peak time of the crisis. This can be argued, because after a first shock, both the economy and the health system at least partly adapted to the new situation (cf. Mishra 2021). Furthermore, in 2020 almost no person already could get a vaccination. Thus, with keeping out data for 2021/2022, potential conflicting effects can be left out. Some studies already try to find a nexus between the corona crisis and Euroskepticism, e.g., Baute/de Ruijter (2021) or Göncz/Lengyel (2021). They find that also in this crisis, Euroskepticism plays a crucial role for political behavior. To sum up, this study adds to the literature by the simultaneous examination of an exposure to more than one crisis, what in the given approach has not been done yet, to the best of our knowledge. The stated peak times are central for the applied econometric methods.

2.3

Hypotheses

Given the presented background, five central hypotheses are of interest for this study: Hypothesis 1 Single crisis exposure: The single exposure to one of the three analyzed crises (Eurocrisis, migration crisis or corona crisis) leads to more Euroskepticism for respondents in highly exposed countries after the peak of the respective crisis has been reached. Hypothesis 2 Repeated crisis exposure: For respondents from countries that highly suffered from one crisis and which later are exposed to a second or a third crisis, a rise in Euroskepticism can be observed. Hypothesis 3 Resilience: Respondents from countries, which already suffered from one crisis, react less strongly when their country is hit again, because of the development of resilience: The exposure to a second or to a third crisis is less Euroskepticism-enhancing than the first exposure.

8

2 Theoretical Foundation & Hypotheses

Hypothesis 4 Antecedent crises: The longer gone a crisis is, the less impact it has on the respondents’ Euroskepticism. Hypothesis 5 Hysteresis: A hysteresis-like effect can be detected.

3

Data & Variables

In this chapter, the included variables, which are based on several raw data sets, are outlined. It is differentiated between those variables which constitute the baseline scenarios and those which are used for additional robustness checks. The DV and the explanatory variables as well as the control variables are introduced. For each variable, the selected definition out of different possible definitions is presented.

3.1

Dependent Variable: Euroskepticism

In the following, potential measurement approaches of Euroskepticism and the respective data is weighted and analyzed. Since Euroskepticism is the central concept to be explained by different explanatory variables, a thorough definition is crucial. A medium-term time range of data sets from 2004 until 2020 is investigated to identify existing changes1 . Data is needed which is available, consistent and comparable for all years on equal measure. To get a broader picture, the inclusion of each year and not only of each EP election year is beneficial. Therefore, only regular survey data and not pre- nor post-election surveys nor election results are taken into account. The inclusion of all EU member states is intended. This study focuses on the individual level and not on parties’ or party systems’ Euroskepticism, as, e.g., Schäfer/Gross (2020) do (see the discussion section). 1 Nevertheless,

the survey data from the years 2004–2007 has only been used in very basal regression models without crisis time dummies. The presented regression results are exclusively based on the time range 2008–2020.

Supplementary Information The online version contains supplementary material available at https://doi.org/10.1007/978-3-658-39267-3_3. © The Author(s), under exclusive license to Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2022 L. Möller, Repeated Crisis Exposure, Euroskepticism & Political Behavior, BestMasters, https://doi.org/10.1007/978-3-658-39267-3_3

9

10

3

Data & Variables

Four relevant ways to measure Euroskepticism can be identified in the literature: First, Baimbridge (2018) uses turnout rates in EP elections, motivated by the concept of “strategic non-voting”. A second way is to take election or survey results of all the 155 parties on both wings of the political spectrum that are classified as Euroskeptic by the PopuList2 (cf. Rooduijn et al. 2019); e.g., Treib (2021) follows a similar approach. Third, Werts et al. (2012) take survey questions concerning the view on the EP and on the process of the European unification from the European Social Survey (ESS). The high quality data of the ESS would have been a good choice also for the present study: On the one hand, it is conducted every two years, on the other hand it contains a suitable choice of countries, even enabling comparisons with non-EU countries. However, as the latest available data stems from 2018, the corona crisis could not be embedded in this study if it was based on ESS data (cf. European Social Survey 2021). Finally, Vasilopoulou (2018) points out that questions from the European Election Study (EES) and the EB3 surveys dealing with the EU membership, the national identification and the attitude towards the unification process have been used in many research papers for the measurement of Euroskepticism. Serricchio et al. (2013), e.g., base their study on EU membership questions from the EB in order to investigate a long time range. Therefore, the EB data is taken as basis for this study to be able to include all three crises of interest. The EB standard version is executed two times per year in all EU member states for a set of the same questions in order to enable long term comparisons; sporadically, it is also conducted in the EU candidate states (cf. Höpner/Jurczyk 2012). The EB data thus provides an annual base for the 28 EU member states including the UK prior to Brexit. Data for countries which have been candidate states at one point during the investigated time range, but became member states later, is always taken into the analysis (i.e., BG, HR and RO; cf. Publications Office of the EU 2021). Respondents from the Turkish Cypriot Community are left out, because this Turkish occupied territory is not under control of the EU member 2

The PopuList is a database initiative of Rooduijn et al. (2019). They try to give an order to extreme parties and to categorize them. Methodologically, they let more than thirty experts assess their opinions on those parties and aggregate the results (cf. Ruth-Lovell et al. 2019). 3 The EB is the central data source for the research in this study. Therefore, several editions from the investigated time range are used. For the sake of simplicity, only the first used edition (number 62.0 from the year 2004; cf. European Commission 2012) and the last one (number 94.2 from the year 2020; cf. European Commission/European Parliament 2021) are cited in the references, representing all 30 used EB editions. All editions and further information can be downloaded from the Gesellschaft Sozialwissenschaftlicher Infrastruktureinrichtungen— Leibniz-Institut für Sozialwissenschaften (German Social Science Infrastructure Services— Leibniz Institute for the Social Sciences (GESIS), https://www.gesis.org/en/eurobarometerdata-service/home).

3.1 Dependent Variable: Euroskepticism

11

state CY which claims the entire island (cf. Pieters 2021). The methodology of the EB is a repeated cross-section, as the interviewees are not always the same: The sample is chosen randomly in order to reproduce the population distribution in terms of living areas; the participants are interviewed in person at their houses (see the econometric methods section; cf. GESIS 2021a, b). The respondent’s country is alphabetically given in the variable isocntry and numerically in isocntry_num. The latter one is of use when controlling for Fixed Effects (FE). The baseline dependent variable (DV) is thus composed as follows: For each respondent, a binary Euroskepticism index states if he or she can be seen as Euroskeptic. For this purpose, the three EB questions concerning the respondent’s image of the EU (measured in five levels plus the don’t-know-option (dk)), the respondent’s own identification (three to five levels) and the respondents’ rating of the benefit of the EU membership for their country (two levels plus dk) are used simultaneously (see the code book in the electronic supplementary material for the whole way the questions are stated, incl. changes between rounds). These three questions have been used in a majority of papers as foundation of a definition of Euroskepticism; a novelty of the present study is to use all three questions simultaneously, most studies only include one or two components (cf. Vasilopoulou 2018). The three underlying variables are combined into one new binary variable; the aim is to get a picture of the concept of Euroskepticism that is as comprehensive as possible and covers a broad time range. The appendix table A.1 in the electronic supplementary material gives an overview of the EB editions from which the fundamental variables are taken. To provide an as up-to-date focus as possible, for the years 2013–2020 more than one edition per year is taken for a more in-depth picture. The amount of editions per year depends on the appearance of the necessary components of the Euroskepticism index, since the underlying questions come up regularly, yet not in each edition. Another advantage of the combined use of all three components is thus to always have at least two of them, simultaneously. The first step to build this index is to consolidate all responses for all rounds. Since not all three questions are asked in each round and the wording slightly changes, they are merged in order to get the most complete picture of the respondent’s opinion possible. That is why the continuous variables eu_ima and eu_memb4 are defined for the respondents’ EU image and their view on the membership benefit, respectively. 4

Concerning eu_memb, four editions in table A.1 are marked additionally: For EB 79.3, only data for HR is given, since this question has only been asked in then candidate countries; for EB 77.4, 82.4 and 92.2 a slightly different survey question has to be used, because the original one has not been asked during these editions. In this substitute version, the respondents are asked whether they assess their country’s membership in the EU as good, bad or as neither/nor. The answers are related correspondingly.

12

3

Data & Variables

In the next step, the binary variables ident_p, ident_s and ident_o are built; they state if the respondents see themselves at least partly as European, strongly as European or only as European. They stem from a consolidation of different ways how the question concerning the personal identification of the respondent is formulated. The wording and the amount of response options changes over the years: It is asked how attached people are towards Europe, how often or to which extent they feel as European or in which order they feel as their nationality and/or as European (see the code book in the electronic supplementary material). Finally, the overall binary variable euroskep is built from the combination of eu_ima, eu_memb and ident_p. The variable ident_p is chosen, because it symbolizes the broadest definition; ident_s and ident_o could be used for future research. The variable euroskep gives the answer whether the respondent may be seen as Euroskeptic. All respondents who answer that their countries did not benefit from the EU membership, have a negative image on the EU or do not even see themselves as partly European are considered as Euroskeptic. That gives in total 153,427 Euroskeptic respondents out of 806,606 respondents in total. If the respondents have a positive or neutral EU image, state that their country benefits from being an EU member, do not know how to answer or are at least partly identifying themselves as European, they are categorized as not being Euroskeptic (653,179 out of 806,606). This results in a ratio of 80.98 percent non-Euroskeptic to 19.02 percent Euroskeptic respondents. The or-condition is chosen to rule out more or less irrational combinations such as feeling only as European, but having a completely negative image of the EU. In this manner, all people who find themselves in at least one category of Euroskepticism are taken into the positive part of the binary indicator. To test the quality of the new built overall indicator of Euroskepticism, euroskep, the original variables (eu_ima, eu_memb and ident_p) are additionally regressed in the driver tests. In the discussion part, it is explained how in future research other approaches found in the literature could potentially serve as alternatives to the approach chosen in this study.

3.2

Explanatory Variables

In the following, the used independent variables will be outlined. For each of the three analyzed crises, one central way how to define and measure it for the baseline approach as well as the alternative versions for the placebo tests and the robustness checks and the driver tests will be presented. Where applicable, the data is taken at a regional level. Guiso et al. (2019) emphasize that populist attitudes are often subject to local aspects and therefore, the use of the finest possible regional segmentation

3.2 Explanatory Variables

13

is recommended for such an investigation. Due to the different sizes of the member states, the provided Nomenclature des Unitès Territoriales Statistiques (Nomenclature of Territorial Units for Statistics, NUTS)5 levels differentiate. Therefore, all respondents are assigned with their NUTS 1 level to enhance comparability. This is given as the dummy variable reg_1 which defines the affiliation to one out of 93 regions. Due to the fact that several smaller countries start being split only at a lower level than NUTS 1, an alternative regional indicator is elaborated where those countries6 are split in two in order to have more variation in the regressions. Only for MT this is impossible, because the lowest given level in the EB is the whole country (cf. Publications Office of the EU 2021). This alternative regional FE dummy is called reg_1alt which is of use to control for regional FE for the henceforth 106 regions.

3.2.1

Eurocrisis: Austerity

For the Eurocrisis case, austerity serves as proxy for the degree of how strongly a country was hit by the Eurocrisis. As seen above, austerity has been the central measure of countries to fight the Eurocrisis. Thus, how it is defined and measured is critical: In the majority of the existing literature, it was solved via defining a dummy for the highly exposed group of GIIPS or GIIPS+ countries (cf. Arias/Stasavage 2019; Baimbridge 2018; Hobolt/Tilley 2016; House et al. 2020). The dummy variable aust_giips assigns the GIIPS countries, the dummy variable aust_giips_x the GIIPS+ countries. Other approaches found in the literature use fiscal indicators for defining and measuring high exposure (HE), e.g., Talving (2017) focused on the change between governmental earning and spending per year. This approach is followed via the continuous variable aust_gov which expresses the difference of a country’s governmental revenue per capita minus its governmental expenditure per capita. The data comes from the Organization for Economic Co-operation and Development (OECD; 5

The NUTS standard is a statistical system of the EU to order its member states’ regional subdivisions: NUTS 1 is the broadest subnational classification (cf. Publications Office of the EU 2021). 6 The affected countries are CY, CZ, DK, EE, FI (it is already split in two subdivisions for the NUTS 1 level, namely Mainland FI and the Aaland Islands, yet as in the EB only observations from Mainland FI are given, this area was split into two parts as equally sized as possible), HR, IE, LT, LU, LV, PT (similar to FI, the NUTS 1 level is tripartite—Continental PT, the Azores and Madeira -, yet only data from Continental PT is given and thus split in two equal parts), SI (in the EB 93.1, the data for SI is given without any regional subdivision and is therefore completely assigned to the one more populated NUTS 2 region; in all other rounds the data is given at this two-tier partition) and SK (cf. ibid.).

14

3

Data & Variables

cf. OECD 2021)7 . Since national governments decide on most financial issues, a regional approach would not be useful, thus the data is matched via the respondent’s country. Unfortunately, data for some small countries and some years are missing (see discussion). A positive balance symbolizes a governmental budget surplus (i.e., contractionary policies, in the spirit of austerity), a negative balance stands for a budget deficit (i.e., expansionary policies; cf. Talving 2017). The data is measured in United States Dollar (USD) purchasing power parity (PPP) per capita: The local currencies have been calculated into USD; PPP outlines the needed amount of local currency of one country for buying the same services and goods in another country, related to the country’s population (cf. OECD 2021). Following the way of usage in the majority of the recent literature and due to the incomplete availability of the continuous data, the dummy approach with aust_giips_x serves as baseline scenario and the fiscal indicators approach aust_gov is taken as robustness check. Moreover, it is checked if changing aust_giips_x to aust_giips has an impact on the results. Furthermore, the binary treatment variable time_eur_crs is always presented and is additionally interacted with the main independent variable. It describes whether the respondent’s survey took place just before or just after the Eurocrisis had its peak, thus it restricts the sample to the affected time span. For the Eurocrisis, the years 2010 and 2011 are defined as before, the year 2012 and 2013 as after. This broader time range has been chosen due to the difference in duration between the concerned countries (see above; cf. Funke et al. 2016; Karyotis/Rüdig 2018). By this means, it is possible to show a potential impact of the crisis on the development of Euroskepticism. Via the interaction term (see below), it can be compared how the crisis potentially had a stronger impact in the most severely affected countries. For a placebo test, these results are compared with the results if instead the years 2010 and 2011 are considered as post-crisis years and the years 2008 and 2009 as pre-crisis years, via the variable time_eur_crs_rob. See the discussion section on how the conducted Eurocrisis robustness checks could be adapted in future research.

3.2.2

Migration Crisis: Net Migration

In the case of the migration crisis, the baseline identification is to take dum_mig as main independent variable. This binary dummy variable states whether a country 7

The OECD data stems from a biennial publication series, published in the years 2009–2021. As with EB, only the most recent and the oldest versions are cited in the references, for simplicity (cf. OECD 2021, 2009). All other editions can be found online (https://doi.org/10. 1787/22214399).

3.2 Explanatory Variables

15

is one of the ten countries with the highest exposure to migration, either in absolute numbers or per capita, on the national or the NUTS 1 level. The most exposed countries are DE, ES, FR, IT, SE and UK (highest absolute numbers of net migration) as well as AT, CY, LU and MT (highest ratio per capita). Contrary to several expectations, GR is not included (cf. Shutes/Ishkanian 2021). The underlying data comes from Eurostat (2021a), precisely the total population and the “net migration plus statistical adjustment”. The latter is defined as the statistically adjusted net migration, taking into account both population change via immigration and emigration and the natural change via births and deaths (cf. Eurostat 2021b). This mixed approach is used in order to take both the bigger countries into account which experienced more migration in absolute numbers, and the smaller ones, as well, which had a higher density. Together, this can proxy the exposure to and the severeness of the crisis: Taking both national and regional data serves as broader approach, since in several countries some regions have been highly exposed, whereas others only had little impact, e.g., Southern DE compared to Eastern DE (cf. Dustmann et al. 2018; Heider et al. 2020). Nevertheless, a phenomenon has been described that often regions which are less exposed to migrants, e.g., in rural areas, tend to still vote more anti-immigration, thus the control variable comm for being in a rural, a middle sized or a large town area is especially important to look at (cf. Dustmann et al. 2018). The alternative, continuous definition and measurement for conducting a robustness check is the variable rat_mig_1. This variable stores the original regional ratios, upon which the decision is based, which are the most exposed countries for the baseline dummy approach. Due to the fact that it shows net migration, it also looks upon the amount of people which have left the specific region. Therefore, it also can be negative. The NUTS 1 level was chosen to be as precise as possible, because the exposure differentiates strongly between regions, and populist attitudes are best investigated on the most local level, as said before (cf. Guiso et al. 2019; Heider et al. 2020). The time variable time_mig_crs splits the survey group into those who have been polled just before (2014) and those just after the migration crisis (2015). It is likely that 2014 respondents considered issues other than the migration crisis to be more pressing than 2015 respondents, since the 2015 rounds have been conducted in September and November when the migration crisis was at its peak (cf. Otto/Steinhardt 2017). This binary variable examines the impact the crisis had by itself. With an interaction term, it is possible to check the impact of the crisis in especially hard hit countries in comparison to less exposed ones. Again, a placebo test examines what happens when this time indicator is shifted (time_mig_crs_rob).

16

3

Data & Variables

Instead of dum_mig, the dummy variable dum_border for those countries which lay at the Southern EU border (BG, CY, ES, FR, GR, HR8 , HU, IT, MT, PT, RO and SI) and at the Eastern border (EE, FI, LT, LV, PL and SK) is used as a robustness check. The “Dublin System” (European Commission 2015; Frontex 2016; Hutter/Kries 2021) intends that the country where migrants entered the EU has to take care of them; thus, these are theoretically the countries which are most directly affected by the recent waves of migration. By additionally using this alternative dummy variable, potential effects are checked for stemming from the following fact: The border countries tend to initially have a higher number of migrants, yet mostly only being a station of the journey and not the final destination of the migrants. Especially in these areas, it could be witnessed that several locals changed their attitude from being very helpful towards the migrants to being adverse: They realized that the migrants had to stay longer than originally intended. Seeing this malfunction of the European schemes and also of their core values, it can be hypothesized that for those citizens the election of Euroskeptic parties became even more a realistic possibility than for the majority of Europeans living in other regions which have been less directly exposed (cf. Shutes/Ishkanian 2021). Additionally, many of these border countries suffered during the Eurocrisis as well, thus, they had to handle a new crisis without having completely resolved the former one (cf. Bandeira et al. 2019; Karyotis/Rüdig 2018). The countries CY, ES and IT are taken together in the dum_eur_mig-variable which shows whether a country suffered from both the Euro- and the migration crisis. It was created by querying whether a country is equally included in dum_mig and in aust_giips_x. See the discussion chapter on potential adaptations for the migration crisis robustness checks in future research.

3.2.3

Corona Crisis: Infections & Casualties

For the corona crisis, the baseline approach uses a dummy variable, analogously to the two antecedent crises. Here, the variable dum_cor displays those countries which suffered most from the crisis. This is measured via the number of infected and of casualties. The respective data is taken from the European Commission Joint

8

With the entrance of HR in the EU by mid-2013, SI has no direct external border anymore; yet, as HR is until today not part of the Schengen area, both countries are considered (cf. Bertaud et al. 2019, Publications Office of the EU 2021).

3.3 Control Variables

17

Research Center (2021)9 . To account for the different size of the countries studied, the ratio of patients to total population is used, based on population data from Eurostat (2021a). This analysis includes data from mid-2020 and end-2020. These countries are BE, CZ, ES, HR, IE, IT, LU, SE, SI, and UK. Analogously to the two other covered crises, via the time variable time_cor_crs it is investigated whether the crisis had an impact on both highly and less exposed countries. It differentiates between those respondents who have been asked closely before the beginning of the pandemic (last quarter of 2019) and those asked in 2020. The interaction term allows the impact of the crisis to be more accurately determined. For a placebo test, the results are compared when changing the time dummy to time_cor_crs_rob which is defined as pre-crisis for the year 2018 and post-crisis for the year 2019. For a continuous robustness check, cor_inf_1 is taken. In this alternative approach, the ratio of individuals testing positive for corona to the population is reported at the NUTS 1 level. Nevertheless, the handling of this continuous approach can be considered difficult, since this ratio is obviously not calculable for the pre-crisis respondents. In future research, an alternative approach could serve based on data on excess mortality (see discussion). Furthermore, dum5_cor_dead_nat_1 is taken to replace dum_cor for another robustness check in order to test if a pattern can be found. It reports the five most affected countries in terms of the ratio of casualties from Covid by mid-2020 to population (BE, ES, IT, SE and UK). As stated above, data for the UK is missing in the last used EB edition, due to Brexit. This may have an impact on the results (see discussion). The variables dum_eur_cor, dum_mig_cor and dum_eur_mig_cor show those countries which have been severely exposed to the corona crisis and to one or both of the other covered crises. For dum_eur_cor, ES, IE and IT are taken, as these three are both part of aust_giips_x and of dum_cor. For dum_mig_cor, it is ES, IT, LU, SE and UK which are all part of both dum_mig and dum_cor. If a country has been severely exposed to all three crises covered in the present study, it is taken into the dummy variable dum_eur_mig_cor: This is the case for ES and IT.

3.3

Control Variables

In econometric studies, it is crucial to add relevant and available control variables (cf. Angrist/Pischke 2009). For this study, it is first and foremost important to con9

The European Commission Joint Research Center (JRC) offers regional data for corona infections and Covid deceases on a daily basis. For this study, NUTS 1 data for 06/30/2020 and 12/31/2020 is used (cf. European Commission Joint Research Center 2021).

18

3

Data & Variables

trol for the respondents’ age and gender (gend); it can be hypothesized that crisis exposure has a different impact on a person depending on individual circumstances, especially the point in time of one’s life when the crisis happens (see the previously mentioned literature). The age-variable is measured continuously, gend as male, female or none of these two. In order to control for non-linear effects of age, also the squared age is added (age_sq). As the time range covered in this study runs from 2004 until 2020, the year-variable is used to contain possible time FE. To control for cluster effects that may stem from being polled during the same EB survey, the variable surv_num is added. The above-presented variables isocntry_num and reg_1alt serve as national respectively regional control variables, thus as country respectively regional FE10 . For most of the data, a look from the regional perspective is feasible, yet not applicable for all variables. To further check whether being part of the euro area has an impact or not, the binary variable euro_area is added. It can additionally be hypothesized that it makes a difference, in what kind of a community the respondent lives, whether it is in a rural area, in a middle-sized or in a large town, especially in the migration crisis case (see above). Therefore, the variable comm is added. Additionally, it shall be checked whether the marital status of the respondent has an impact: This is solved via the binary variable marit, stating whether the respondent is married or not. Also the education (educ) and current and historical employment (job respectively job_hist) are relevant to control for; an individual crisis reaction depends on own possibilities and resilience (see hypothesis 3): With a higher education, it might be easier to regain a job after a job loss and having experienced unemployment in the past might change the attitude towards a potential job loss (cf. Mishra 2021). The educ-variable is measured as the age, when the respondents stopped their full-time education. In job, the respondents state whether they are unemployed, retired or a student, or working in one of fifteen industries or positions11 . The scale used for job_hist consists of fourteen different industries/positions to which the respondent belongs or as never having worked in a paid position. The variable pol_scale measures the respondent’s political placement in ten steps, thus from leaning towards the left to leaning towards the right political spectrum. The three binary variables soc_place_w, soc_place_m and soc_place_u answer whether the respondents see themselves as being part of the working, the middle or the upper class, respectively. Another control variable is lifesat, stating the life satisfaction of the respondents. 10

As stated above, reg_1alt is the broader approach, thus it is chosen to be the regional FE variable in this study. Since isocntry_num is numerical and therefore works technically more smoothly, it is chosen as national FE variable. 11 For further information on this and all the other used variables, see the code book in the electronic supplementary material.

3.3 Control Variables

19

Summing up, the control variables which have been used in this study are the following: age, age_sq, gend, year, surv_num, isocntry_num, reg_1alt, euro_area, comm, marit, educ, job, job_hist, pol_scale, soc_place_w, soc_place_m, soc_ emphplace_u and lifesat. The availability of variables which are not always given can be seen in the appendix table A.1 in the electronic supplementary material12 . Furthermore, a correlation matrix is presented in table A.2 to show interconnections between the used control variables. Most interconnections are very small, a connection between age and marit, educ, job or job_hist is visible as well as between surv_num and year and between isocntry_num and reg_1alt. Several other personal characteristics would have been desirable to control for to get a broader picture of the respondents, yet the data availability is restricted. See the discussion chapter for further information on which other control variables could be fruitful.

12 Not all EB rounds include the following variables: pol_scale, soc_place_w, soc_place_m, soc_place_u and lifesat.

4

Econometric Methods

In the following chapter, the econometric methods applied in this study are presented. The statistic software program used for this study is Stata (Standard Edition, Version 17.0; cf. StataCorp 2021). Several statistical methods are checked concerning their benefit for this study, and the best suitable ones are chosen. For all chosen methods, an explanation is given why they fit the purpose of this study and why they are necessary to conduct. Due to refutations, some methods are rejected and their potential purpose and reason for rejection are presented. In the first subsection, the chosen identification strategy is shown, then the composition of the final regressions is given. Here, the equations of the regressed baseline models are displayed. For the alternative approaches, further equations are added in the electronic supplementary material.

4.1

Identification Strategy

The gold standard for an investigation of differences in political behavior before and after a crisis is a longitudinal setting in which the same group of respondents is surveyed over a long period of time, preferably with the same questions: This kind of design is known as a panel data design (cf. Angrist/Pischke 2009). In this case, a Differences-in-Differences-approach (DiD-approach) is possible to conduct in order to extract a time trend. The DiD-approach is based on the idea of comparing two highly similar subgroups before and after one subgroup, i.e., the treatment group, is treated: An exemplary treatment might be a payment of a higher wage or a Supplementary Information The online version contains supplementary material available at https://doi.org/10.1007/978-3-658-39267-3_4.

© The Author(s), under exclusive license to Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2022 L. Möller, Repeated Crisis Exposure, Euroskepticism & Political Behavior, BestMasters, https://doi.org/10.1007/978-3-658-39267-3_4

21

22

4

Econometric Methods

crisis exposure. Consequentially, the varying developments in both groups become visible: It is assumed that due to the high similarity of the subgroups, the same time trend shows in the control and the treatment group, while the latter one additionally experiences a treatment effect. The effect of the treatment is therefore equal to the difference between the trend observed in the treatment group and the trend observed in the control group. In this study, treatment is understood analogously to the average treatment effect of the treated (cf. Angrist/Pischke 2009). Nevertheless, real panel data is rarely conducted for the subject of the present study. Thus, for the purpose of this research, a slightly different approach is necessary. As stated above, the central part of the data set used in this study stems from the EB. Therefore, it is not possible to conduct a conventional DiD-approach, because EB data is not given as panel data but as a repeated cross-section (cf. GESIS 2021a). However, it is possible to design a version that is as close as possible to a DiD-design using the procedures applied by Gerling/Kellermann (2021) and Guiso et al. (2019)1 . A quasi-randomization in an event-study design is enabled (cf. Gerling/Kellermann 2021). For each of the analyzed crises, a time dummy is used in order to define two separate groups of the respondent sample: One which was surveyed shortly before the crisis happened (or when it had its peak) and one which was surveyed just afterwards (cf. Depetris-Chauvin et al. 2020; Gerling/Kellermann 2021; Giani/Méon 2021; Mikulaschek et al. 2020). By this manner, it can be assumed that mainly the respective crisis was responsible for a potential change in the measurable Euroskepticism of the surveyed respondents (cf. Angrist/Pischke 2009). Simultaneously, it is vital to control for FE, likewise to the procedure in DiD-approaches. Via the country dummy, it is possible to identify single effects for the highly exposed group of countries, prior to the crisis. Via the time dummy, the overall time trend for all countries after the crisis and via the interaction term, the additional effect after the crisis in the highly affected country group can be displayed separately (cf. ibid.). Interaction terms are products of two independent variables, mostly dummy variables (cf. ibid.). The regressed coefficient of this interaction term is an additional effect for the especially affected group; if one of the original independent variables equals zero, the product turns to zero and thus no coefficient is given. Therefore, interaction terms help to differentiate the found effects: The interaction term singles out the additional effect of the highly affected countries post-crisis compared to the less affected countries. Its effect is added to the other effects: On the one hand, the effect for being part of the most affected group of countries and on the other hand,

1

Other similar approaches can be found in Depetris-Chauvin et al. (2020), Giani/Méon (2021) or Mikulaschek et al. (2020).

4.2 Composition of the Regressions

23

the general time trend effect, which is valid for all countries after the crisis happened compared to before it happened (Angrist/Pischke 2009). The design of the EB is representatively chosen per country, so that the overall distribution is similar for the respondents surveyed before and those surveyed after the crisis (see above and in the technical specifications of the EB; cf. European Commission/European Parliament 2021). For further verification, entropy balancing is performed additionally in the most sophisticated regression versions: Entropy balancing is a process used to apply weights based on the distribution of the chosen control variables (cf. Hainmueller 2012). With these weights, a design is achieved where the treatment and the control group are as similar as possible (cf. Gerling/Kellermann 2021; Giani/Méon 2021; Hainmueller 2012). Thus, it comes close to using real panel data (cf. Angrist/Pischke 2009). The chosen entropy balancing mechanism is highly restrictive, as it is based on the respective time dummy and all FE (national, regional, temporal and per survey round), all demographics and socio-economics (age, age squared, gender, marital status, education, type of community, satisfaction with life, current and historic employment status) and all political attitudes control variables (self-positioning on left-right scale and on social placement) are included. The balancing is conducted for the first three moments of the distribution of the control variables (cf. Gerling/Kellermann 2021).

4.2

Composition of the Regressions

For testing the hypotheses of this study, several econometric methods are applied; the linear Ordinary Least Squares method (OLS-method) is the central one (cf. Angrist/Pischke 2009). This is one of the most straightforward and most common methods used in econometric studies. For the purpose of this study, it is highly suitable; this can be seen in the results. Apart from OLS, a non-linear method, which is frequently used in econometric studies, is the logit regression model. For this study, several logit-regressions have been conducted, yet no advanced gain in knowledge could be carved out. Therefore, in the results section of this paper, only OLS regressions are presented. For each result table, several varying versions are regressed, in which more and more FE as well as different control variables are added. The given standard errors are insofar robust, as the StataCorp (2021)-regression-command computes standard errors which are heteroskedasticity-consistent and as such they take a realistic distribution into account (cf. Angrist/Pischke 2009). This kind of standard errors is based on the work of Eicker (1967), Huber (1967) and White (1980).

24

4

Econometric Methods

Potentially, correlation problems might occur due to the fact that the 28 respondents’ countries may be seen as clusters as well as due to the fact that the data might be serially correlated over time (cf. Angrist/Pischke 2009; Moulton 1986). As one option to tackle this, bootstrapping is conducted in order to check for a potential bias in the robust standard errors and in order to get a better inference; bootstrapping works by drawing a new sample out of the used data, assuming that this is the whole population, in order to approximate the distribution of the used sample (cf. Angrist/Pischke 2009). In this study, the wild cluster restricted bootstrap (WCRB) following Gerling/Kellermann (2021) and Roodman et al. (2019) was employed via the boottest-command of StataCorp (2021). The WCRB including the six-point distribution of Webb (2013) is assumed to deliver p-values which are especially restrictive (cf. Cameron/Miller 2015; Gerling/Kellermann 2021; MacKinnon/Webb 2016). Due to the fact that in most models the robust p-values are very close to the ones generated with WCRB, the stars in the tables correspond to the robust ones; only, if the p-values differ from each other, the additional WCRB p-value is displayed and interpreted. In this study, 106 separate regions are used to control for regional FE. For this potential cluster, the stated standard errors can be assumed to be correct and sufficient, following Angrist/Pischke (2009). Therefore, the WCRB is not conducted for this case.

Figure 4.1 Equation for the baseline approach for the Eurocrisis

Figure 4.2 Equation for the baseline approach for the migration crisis

4.2 Composition of the Regressions

25

Figure 4.3 Equation for the baseline approach for the corona crisis

Figure 4.4 Equation for the baseline approach for the repeated crisis exposure

Adding several control variables is crucial, because some characteristics are respondent-specific; adding FE is important in order to tackle characteristics that are potentially country- and/or time-specific. With this design, fix yet unobserved factors can be accounted for and are not omitted. In the present study, only the effects of the actual crises are to be investigated, thus when using FE, everything else can be captured, in order to extract the crisis-only effects (cf. ibid.). These theoretic ways of identification are summed up in the following equations: Equation (1) is for the analysis of the Eurocrisis, equation (2) for the migration crisis and equation (3) for the corona crisis. The repeated exposure version is displayed in

26

4

Econometric Methods

equation (4). The equations describe the Euroskepticism of the individual i in country c in region r at time t. The term i,c,r ,t describes the error term at an individual level. In the following, only the baseline versions of the most sophisticated models are shown. All modified versions can be found in the electronic supplementary material.

5

Regression Results

The following section is the core of this study, since it presents, interprets and contextualizes its findings: In the first subsection, each of the three analyzed crises is looked at separately, in the second subsection, a combined exposure to one or more of these crises is presented. The results of the first subsection is displayed in tables. In all of them, the coefficients are rounded to four, the robust standard errors to five and the adjusted R2 to two decimal places. In the second subsection, the results are presented via figures, the underlying tables can be found in the electronic supplementary material. Unlike in the descriptive Data & Variables section, in the following, precise names—and not the previously presented technical ones—are used when describing the results.

5.1

Single Crisis Exposure Regressions

In this subsection, an exposure to the Eurocrisis, the migration crisis or the corona crisis is examined separately. The analysis is done for one baseline scenario, one placebo test, two robustness checks and one driver test for each of the crises. The baseline regressions use dummy variables as main independent variables and time dummies for the time when the crisis had its peak. For the placebo tests, the time dummies are changed to time ranges in which the respective crisis did not happen, in order to control, whether the baseline effects are random. The first robustness checks change the main country dummies to dummies representing different groups of countries. The second robustness checks change the main independent variables Supplementary Information The online version contains supplementary material available at https://doi.org/10.1007/978-3-658-39267-3_5.

© The Author(s), under exclusive license to Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2022 L. Möller, Repeated Crisis Exposure, Euroskepticism & Political Behavior, BestMasters, https://doi.org/10.1007/978-3-658-39267-3_5

27

28

5

Regression Results

to continuous variables. Finally, driver tests for each crisis with a split-up of the DV of Euroskepticism are fulfilled in order to potentially identify drivers of the results.

5.1.1

Baseline Regressions (Single Crisis)

In table 5.1, the results of the baseline regressions are depicted (see equation (1)– equation (3)). For each of the three analyzed crises, one separate panel is presented. In the following, the core results are interpreted. In each panel, five OLS-models are run which are insofar different, as they are calculated with different FE and control variables. The most sophisticated model includes in each case entropy balancing. Furthermore, the WCRB is used as elaborated above. The significance level of the resulting p-values is only reported additionally in the tables if it differs from the significance level of the standard p-values. In the most cases, they are in the same range. As stated above, the baseline approaches use a binary dummy variable as DV which states if the respondent is Euroskeptic or not. For each panel, three main explanatory variables are used: First, a binary country dummy stating if the respondent’s home country has been highly exposed to the respective crisis, second, a binary temporal dummy showing if a respondent has been surveyed closely after the respective crisis happened (compared to those respondents who have been surveyed closely prior the crisis) and third, an interaction term of these two. By this means, several comparisons are possible to conduct: The regression results present the difference in the impact of the highly exposed country group compared to the less exposed one, the overall time trend in all countries and—via the interaction term— the time trend for the highly exposed countries. The binary country variables are composed for the following countries: For the Eurocrisis-panel, it shows whether the respondent’s country is part of the GIIPS+ group (see above). For the migration crisis-panel, it states if the respondent lives in a country which was highly exposed to this crisis (see above, i.e., AT, CY, DE, ES, FR, IT, LU, MT, SE or UK). For the corona crisis-panel, it displays if the respondent lives in a country with HE to that crisis (see above, i.e., BE, CZ, ES, HR, IE, IT, LU, SE, SI or UK). Via the temporal dummies, it is possible to focus only on that part of the entire respondent sample which has been asked closely before or closely after the respective crisis took place. For the pre-Eurocrisis group this is 2010/2011, for the post-group 2012/2013. Concerning the second crisis, the year 2014 is considered as pre- and 2015 as post-migration crisis, concerning the third, surveys from the last quarter of the year 2019 are taken as pre- and those surveyed in the year 2020 as post-corona crisis.

5.1 Single Crisis Exposure Regressions

29

The results for the Eurocrisis panel are interpreted as follows: The HE dummy shows that prior to the crisis, for respondents from later highly exposed countries, all model specifications find a highly significant negative effect, meaning that in these countries the Euroskepticism tends to be lower than in the comparison group. This can be interpreted with the stance that this group of countries has benefited greatly from being part of the EU, prior to the occurrence of the Eurocrisis, e.g., from low interest rates enabling a rise in the overall standard of living (cf. Hale/Obstfeld 2016). The results of the time trend are less explicit, insofar as the less sophisticated models show a significant positive, the more sophisticated ones a significant negative effect. Thus, in the more sophisticated models the overall time trend for all countries, not only the highly affected, has a downward-sloping effect on Euroskepticism. It can be interpreted with European solidarity in the form of experiencing a supportive feeling during the crisis, at least for respondents from countries that offered support, e.g., to the GIIPS+ countries, which have been less exposed themselves (cf. Wallaschek 2020a, 2020b). Nevertheless, this contradicts in a way the pattern that people in many countries have been against far-reaching financial support for other EU member states: It might have exacerbated Euroskeptic feelings in the donor countries (cf. Gerbaudo 2017; Reinl 2020). The associated conditions led to anti-European protests in highly exposed countries (cf. Gerbaudo 2017; Reinl 2020). The interaction term is the additional time trend for the group of the highly exposed countries: Here, it is visible that it is significantly positive over all specifications. This means that after the Eurocrisis in those countries, which suffered severely from it, Euroskepticism rose significantly more than in the less affected countries. Thus, even with the support from other less affected EU member states, the respondents might have suffered more from the national countermeasures like austerity or the international conditionalities, and benefited less from the European solidarity (cf. Gerbaudo 2017; Karyotis/Rüdig 2018). This is in line with the expectations of the hypothesis 1 (single crisis exposure). The overall effect for the GIIPS+ countries post-crisis nevertheless stays negative in all more sophisticated models, when all three coefficients are summed, due to the very high value (according to amount) of the country effect. For the migration crisis panel, the results can be interpreted more straightforwardly than the Eurocrisis results: The HE country effect is significantly positive over all specifications, meaning that in all the highly affected countries, a more Euroskeptic tendency can be found compared to the less affected, even prior to the migration crisis. This might be at least partly explained by the overlapping of this group with the GIIPS+ countries, i.e., CY, ES and IT, which—as found in the first panel—became Euroskeptic over the Eurocrisis. Additionally, the traditional

30

5

Regression Results

Euroskeptic stance of the UK can be adduced, which is also part of the country group that was highly exposed to the migration crisis (cf. Mény 2020). The time trend is inconsistent over the different specifications and only significantly positive in the least sophisticated version. Thus, it is too weak for a thorough interpretation. The interaction term effect is significantly positive over all specifications, meaning that Euroskepticism rose in the highly affected countries after the migration crisis took place. This fulfills the hypothesis 1 (single crisis exposure). The overall effect for the highly exposed countries is then positive, too; thus, both the hypothesis 1 is fulfilled and, together with the reasoning of visible aftereffects of the Eurocrisis, also the hypothesis 2 (repeated crisis exposure) can be seen as supported. For the not affected countries, the crisis did not change the stance on Euroskepticism: A time trend is not visible, therefore this group seems relatively unaffected in terms of Euroskeptic attitudes. The corona crisis panel results can be analyzed as follows: The country variable effect is positive over all specifications, yet only significant in the least sophisticated specification. A more in-depth interpretation is thus not reasonable. The time trend is positive and significant for all specifications, meaning that the corona crisis had a Euroskepticism-raising effect in all countries, not only the highly affected. This seems plausible, as the corona crisis led to unilateralism with purely national health policies and border closures (cf. Bozorgmehr et al. 2020). The interaction effect is not easily interpretable, again, because the effects are only significant for the less sophisticated models and close to zero for the more sophisticated ones. It seemingly makes no difference whether a country was highly or less exposed to the corona crisis. This might be explained by the fact that respondents in all countries faced a totally new and insecure situation and that it made less of a difference how severely the exposure ultimately was. The overall effect in the more sophisticated versions is driven by the time trend and therefore it can be stated that the corona crisis rose Euroskepticism in all countries. The hypothesis 1 (single crisis exposure) is therefore not only valid for the highly exposed countries, but for all countries. Summing up, the results of the baseline approaches support the hypothesis 1 (single crisis exposure) and additionally give hints for supporting hypothesis 2 (repeated crisis exposure), as well. Nevertheless, the results are different for the crises, insofar as for the Eurocrisis and the migration crisis the country dummies and the interaction terms are as expected, whereas the time dummy is only reasonably interpretable for the corona crisis.

5.1 Single Crisis Exposure Regressions

31

Table 5.1 Baseline regression results DV: Euroskepticism

(1)

(2)

(3)

(4)

(5)

OLS

OLS

OLS

OLS

OLS

−0.0060**

−0.2663*** −0.3207*** −0.3228*** −0.3149***

Panel A: Eurocrisis HE

(0.00282) 0.0743***

Post-Crisis

(0.00224) HE × Post-Crisis

0.0667*** (0.00409)

(0.01248) 0.0800*** (0.00364) 0.0688*** (0.00396)

(0.01969)

(0.02621)

(0.02681)

0.0807*** −0.0670*** −0.0663*** (0.00512) 0.0583*** (0.00576)

(0.00528) 0.0300*** (0.00826)

(0.00529) 0.0309*** (0.00823)

Adjusted R2

0.02

0.09

0.11

0.10

0.09

Observations

191,013

191,013

84,089

42,543

42,543

Panel B: Migration crisis HE

0.0573*** (0.00340)

Post-Crisis

0.1138*** −0.0039 (0.00291)

HE × Post-Crisis

0.1395*** (0.01306)

0.0349*** (0.00538)

(0.00375) 0.0339*** (0.00522)

0.1839*** (0.02284) −0.0019 (0.00548) 0.0386*** (0.00876)

0.1902*** (0.02491) −0.0076 (0.00615) 0.0462*** (0.00981)

0.1912*** (0.02762) −0.0075 (0.00973) 0.0479*** (0.00978)

Adjusted R2

0.03

0.09

0.12

0.12

0.12

Observations

110,323

110,323

42,849

33,573

33,573

Panel C: Corona crisis HE

0.0295*** (0.00289)

Post-Crisis

0.0355*** (0.00240)

HE × Post-Crisis

0.0485

0.0010

0.0014

0.0092

(0.04165)

(0.05230)

(0.05111)

(0.05765)

0.0466*** (0.00261)

−0.0120*** −0.0101** (0.00430)

(0.00419)

0.1369*** (0.00469) −0.0007 (0.00830)

0.1377*** (0.00490) 0.0021 (0.00826)

0.1351*** (0.00501) −0.0002 (0.00872)

Adjusted R2

0.00

0.06

0.12

0.13

0.13

Observations

110,036

110,036

39,897

39,897

39,897

(Continued)

32

5

Regression Results

Table 5.1 (Continued) DV: Euroskepticism

(1)

(2)

(3)

(4)

(5)

OLS

OLS

OLS

OLS

OLS

National FE

Y

Y

Y

Y

Regional FE

Y

Y

Y

Y

Year FE

Y

Y

Y

Y

Survey Round FE

Y

Y

Y

Y

Y

Y

Y

Y

Y

Demogr. & soc.-econ. Political attitudes Entropy balancing

Y

Note: The DV is a binary dummy variable showing if a respondent is Euroskeptic. All models include intercepts (not presented in the table). In parentheses, robust standard errors are reported. For the Eurocrisis-panel, the main independent binary variable shows whether the respondent lives in a GIIPS+ country (CY, ES, GR, HU, IE, IT, LV, PT or RO). For the migration crisis-panel, the binary variable shows if the respondent lives in a country which is highly exposed to the migration crisis (AT, CY, DE, ES, FR, IT, LU, MT, SE or UK). For the corona crisis-panel, the binary variable shows if the respondent’s country is highly exposed to the corona crisis (BE, CZ, ES, HR, IE, IT, LU, SE, SI or UK). For each panel, a time dummy of the crisis and an interaction term of the time and the country dummy are shown. The time dummies compare respondents just around those points in time where the respective crisis took place (Eurocrisis time dummy: pre-crisis 2010/2011–post-crisis 2012/2013, migration crisis: pre-crisis 2014–post-crisis 2015, corona crisis: pre-crisis end-2019–post-crisis 2020). For all specifications, the boottest-command was conducted in StataCorp 2021, based on the six-point distribution by Webb 2013 for the time dummy using the WCRB. This aims at getting cluster-robust standard errors which are clustered by the respondent’s country. Their p-values are not specifically reported, as they are the same as those indicated by the stars. The FE are national, regional, temporal and for each included round of the EB. Demographics and socio-economics: age (15–99), age squared, gender (male, female, none), marital status (married or not), education (age when full-time education stopped; 13 cat.), type of community (4 cat.), satisfaction with life (5 cat.) as well as socio-economics (employment status (18 cat.), historical employment status (16 cat.)). Political attitudes: self-positioning on left-rightscale (0–10), self-positioning on social placement via dummy variables (working, middle or upper class). In model (5), entropy balancing is applied to define matching weights with the time variable as treatment variable, including the country dummy, the FE as well as the demographic, socio-economic and political attitude control variables. See code book in the electronic supplementary material for further details. Y = yes; *** p < 0.01, ** p < 0.05, * p < 0.1.

5.1 Single Crisis Exposure Regressions

5.1.2

33

Placebo Tests (Single Crisis)

For each baseline analysis, a placebo version is conducted where instead of the true time variable a different one is taken (see table 5.2). The equations for the underlying regressions can be found in the appendix (equation A.1—equation A.3) in the electronic supplementary material. The placebo time dummies are in the same manner composed as the time dummies in the baseline approach. There, a group of those respondents is taken out of the whole respondent sample which have been interviewed just around the respective crisis happened. In the placebo case nevertheless, time ranges are chosen during which the crisis did not happen in reality. The expectation is therefore, to find nothing significant—or at least only very little—to underpin that the baseline results are not spurious, but truly stem from the happening of the crises. The chosen placebo times are the years 2008/2009 for before the Eurocrisis and the years 2010/2011 for afterwards, the year 2013 for before the migration crisis and the year 2014 for post-migration crisis, as well as the year 2018 for pre- and the year 2019 for post-corona crisis. Concerning the Eurocrisis panel: Similar to the baseline results, the coefficients for the country dummy in all specifications are negative and significant and with similar amounts. It can be interpreted just as before, thus, that in this group of countries the Euroskepticism tends to be lower than in the comparison group. It shows that the effect of being part of this group is relatively time constant, it might—at this earlier point in time—even be stronger, because the negative impacts of the crisis still lay further ahead and the benefits are still mostly enjoyed (cf. Hale/Obstfeld 2016). Yet, this is not visible in the regression results, since the absolute values in the placebo version are not higher for all specifications. The placebo time dummy findings are all significantly positive. In the most sophisticated models, this is different to the baseline findings, where a negative effect was found: This supports the aim of this placebo test, as the placebo time trend effect is opposite to the real time trend. Concerning the interpretation of the central interaction effect, it is visible that only in the least sophisticated models, the effect is significantly positive. In the more sophisticated ones, it is insignificantly negative. In the baseline approach, all coefficients are significantly positive. This supports the expectation, that happenings at the placebo time are less relevant for Euroskeptic attitudes than those at the true point in time, thus the post-crisis time of the baseline approach. Together with the placebo time trend, this can be interpreted, insofar as the Eurocrisis has a real impact. Nevertheless, the detectable findings might at least partly be influenced by potential aftereffects of the then recent global financial crisis.

34

5

Regression Results

For the migration placebo panel, the time dummy and thereby also the interaction term are omitted in the most sophisticated specifications due to a restricted sample. Hence, only the complete models are shown in the table. The interpretation of the given models is straightforward, yet less restrictive and thus less comparable to the baseline findings: The country dummy is positive and significant over all specifications (same as in the baseline approach, also similar in terms of the amount). Again, this can be interpreted like before: This group of countries tends to be more Euroskeptic than the comparison group, also at least partly triggered by the then recent Eurocrisis. This finding and this argumentation also support the hypothesis 2 (repeated crisis exposure). For the included models, the time dummy and the interaction term are significantly negative. In the baseline approach, the time dummy is only for the least sophisticated model positive and significant, for the other ones close to zero, the interaction effect is positive and significant. Therefore, due to the given differences, it can be—with some caution—interpreted as backing the exposure to the migration crisis as truly causing an effect on Euroskepticism. Looking at the country dummy in the corona panel, results are visible which are partly similar to those from the baseline approach: They are positive for the less sophisticated models and slightly negative for the more sophisticated ones, but only significant for the least sophisticated version. An interpretation is therefore not expedient. The time dummies are all positive and significant, only the least sophisticated is significantly negative. In the baseline approach, all coefficients are positive and significant, yet with a higher amount. It can be interpreted as follows: A time trend is visible, yet the stronger visibility for the baseline version supports the reckoned connection; the crisis has an impact for all countries, potentially the visible results for the placebo come from aftereffects of prior crises. Concerning the interaction effect, the coefficients are positive and only significant for less sophisticated versions, thereby similar to the baseline results which are insignificant and close to zero. Thus, it can be assumed that the visible effects are offshoots of the then recent migration crisis and potentially still also of the Eurocrisis (see hypothesis 2 (repeated crisis exposure) and hypothesis 4 (antecedent crises)). Also for the corona crisis, the placebo test supports the findings of the baseline approach. Summing up, for all crisis panels, the placebo tests support the baseline findings and interpretations.

5.1 Single Crisis Exposure Regressions

35

Table 5.2 Placebo tests results DV: Euroskepticism

(1)

(2)

(3)

(4)

(5)

OLS

OLS

OLS

OLS

OLS

Panel A: Eurocrisis HE

−0.0244*** −0.2789*** −0.3331*** −0.3104*** −0.2946*** (0.00314)

Post-Crisis (Placebo)

0.0232*** (0.00243)

HE × Time-Placebo

0.0184*** (0.00422)

(0.01305) 0.0309*** (0.00333) 0.0181*** (0.00416)

(0.02024) 0.0350*** (0.00473) 0.0052 (0.00598)

(0.02703) 0.0279*** (0.00555) −0.0022 (0.00871)

(0.02808) 0.0273*** (0.00551) −0.0028 (0.00868)

Adjusted R2

0.00

0.05

0.10

0.10

0.10

Observations

137,699

137,699

69,462

42,920

42,920

−0.0341

−0.0309

Panel B: Migration crisis HE

0.0945*** (0.00350)

0.1540*** (0.01253)

0.1422*** (0.02182)

Post-Crisis (Placebo)

−0.1471*** −0.1780*** −0.1701***

HE × Time-Placebo

−0.0372*** −0.0358*** −0.0307***

(0.00265) (0.00488)

(0.00349) (0.00474)

(0.00500) (0.00720)

Adjusted R2

0.04

0.10

0.13

Observations

137,312

137,312

57,345

Panel C: Corona crisis HE

0.0279*** (0.00350)

Post-Crisis (Placebo)

−0.0425*** (0.00231)

HE × Time-Placebo

0.0422 (0.02879) 0.0651*** (0.00311)

0.0064

0.0082**

(0.00410)

(0.00397)

−0.0460 (0.04075) 0.0700*** (0.00453) 0.0158*** (0.00608)

(0.03953) 0.0757*** (0.00449)

(0.03972) 0.0748*** (0.00459)

0.0069

0.0084

(0.00589)

(0.00605)

Adjusted R2

0.01

0.07

0.12

0.12

Observations

161,585

161,585

65,314

64,381

0.12 64,381

(Continued)

36

5

Regression Results

Table 5.2 (Continued) DV: Euroskepticism

(1)

(2)

(3)

(4)

(5)

OLS

OLS

OLS

OLS

OLS

National FE

Y

Y

Y

Y

Regional FE

Y

Y

Y

Y

Year FE

Y

Y

Y

Y

Survey Round FE

Y

Y

Y

Y

Y

Y

Y

Y

Y

Demogr. & soc.-econ. Political attitudes Entropy balancing

Y

Note: The DV is a binary dummy variable showing if a respondent is Euroskeptic. All models include intercepts (not presented in the table). In parentheses, robust standard errors are reported. For the Eurocrisis-panel, the main independent binary variable shows whether the respondent lives in a GIIPS+ country (CY, ES, GR, HU, IE, IT, LV, PT or RO). For the migration crisis-panel, the binary variable shows if the respondent lives in a country which is highly exposed to the migration crisis (AT, CY, DE, ES, FR, IT, LU, MT, SE or UK). For the corona crisis-panel, the binary variable shows if the respondent’s country is highly exposed to the corona crisis (BE, CZ, ES, HR, IE, IT, LU, SE, SI or UK). For each panel, a time dummy of the crisis and an interaction term of the time and the country dummy are shown. For these placebo tests, the time dummies show points in time where the respective crisis did not happen in reality (Eurocrisis placebo time dummy: pre-crisis 2008/2009–post-crisis 2010/2011, migration crisis: pre-crisis 2013–post-crisis 2014, corona crisis: pre-crisis 2018– post-crisis 2019). For all specifications, the boottest-command was conducted in StataCorp 2021, based on the six-point distribution by Webb 2013 for the time dummy using the WCRB. This aims at getting cluster-robust standard errors which are clustered by the respondent’s country. Their p-values are not specifically reported, as they are the same as those indicated by the stars. The FE are national, regional, temporal and for each included round of the EB. Demographics and socio-economics: age (15–99), age squared, gender (male, female, none), marital status (married or not), education (age when full-time education stopped; 13 cat.), type of community (4 cat.), satisfaction with life (5 cat.) as well as socio-economics (employment status (18 cat.), historical employment status (16 cat.)). Political attitudes: self-positioning on left-right-scale (0–10), self-positioning on social placement via dummy variables (working, middle or upper class). In model (5), entropy balancing is applied to define matching weights based on the FE as well as the demographic, socio-economic and political attitude control variables. In panel B, the models (4) and (5) are omitted due to a restricted sample. See code book in the electronic supplementary material for further details. Y = yes; *** p < 0.01, ** p < 0.05, * p < 0.1.

5.1 Single Crisis Exposure Regressions

5.1.3

37

Robustness Checks & Driver Tests (Single Crisis)

In the following, the conducted robustness checks and driver tests are presented. The first of the presented approaches is different to the baseline approach, insofar as the country dummy variables are changed in their definition, so that different countries are investigated (see table 5.3 and equation A.4—equation A.6 of the appendix in the electronic supplementary material). The second robustness approach is to change the baseline country dummies into continuous variables (see table 5.4 and equation A.7—equation A.9 of the appendix). The third way splits the Euroskepticism index into its three components in order to investigate the drivers of the results (see table 5.5 and equation A.10—equation A.18 of the appendix). In the Data & Variables section, it is explained how the variables are defined. In table 5.3, all baseline approaches are repeated with a different definition for the highly exposed countries for each crisis. In the Eurocrisis panel, only the GIIPS and not the GIIPS+ countries are taken into account. For the most sophisticated version, the results are very similar to the baseline results, only the country effect is smaller in absolute terms. This means that adding the four countries which are not part of the GIIPS group, thus CY, HU, LV and RO, moves the coefficient to a more Euroskepticism-diminishing area. Living in the GIIPS countries prior to the crisis had thus less a positive effect than living in one of the four countries added in GIIPS+. A potential reason might be that those countries could enjoy even more being part of the EU, as they entered it more recently than the GIIPS countries (cf. Baas/Brücker 2011; Campos et al. 2014). Nevertheless, Serricchio et al. (2013) find for the time range of 2007–2011 a steeper growth in Euroskepticism in the new member states, reversing the “accession effect” (Down/Wilson 2008; Guerra 2013; Verney 2011) of turning less Euroskeptic when incorporated in the union. The time dummy results are significantly positive for the less sophisticated versions, but significantly negative for the more sophisticated ones. This is the same situation as in the baseline approach, also in terms of the amount. Looking at the interaction term coefficients, their positive significance is visible, in the same manner as in the baseline approach. Thus, hypothesis 1 (single crisis exposure) is fulfilled again. For the migration panel, the alternative country dummy definition includes eighteen instead of ten countries, chosen due to their EU border location instead of the mere numbers of exposure. The results are more different to the baseline ones than those in the first panel, especially for the most sophisticated version: The country coefficients are significantly negative (significantly positive in the baseline), the time trend significantly positive (insignificantly negative in the baseline) and the interaction term significantly negative (significantly positive in the baseline). Furthermore, the WCRB results differ for most of the specifications, insofar as they become less

38

5

Regression Results

significant. A potential reason for these findings might be that now a geographic measure was used and therefore, many different countries are combined. Most of them are only connected through their external location. Even though migratory movements reach almost all borders, exposure varies, depending mainly on time. The results in this robustness check can therefore be interpreted in the direction that a border location both before and after the migration crisis had less an impact on Euroskeptic tendencies than just the mere time trend. The border country group seems to be too heterogeneous to allow a clear judgment. However, due to the significantly positive time trend it can be stated that this panel validates the hypothesis 1 (single crisis exposure) for all countries, likewise to the corona crisis in the baseline approach. For the corona crisis, the modified country dummy consists of five instead of ten countries. It is visible that the interaction term is omitted over all specifications due to sample restrictions. Hence, it is not shown in the table. The country effect is positive, yet only significant in the least sophisticated version, and the time trend is significantly positive; both results are therefore equal to the results of the baseline approach. These results again support the findings that the exposure to the corona crisis had an impact on all countries and not to an especially higher extent on the more exposed ones; thus, it underlines the validity of the hypothesis 1 (single crisis exposure) for all countries. The second robustness check uses continuous variables, as outlined above. In the Euro- and migration crisis setting, these absolute values can be negative, thus the coefficient values are very different to the ones from the baseline approach. For the Eurocrisis, a variable was chosen which states how the respective country keeps house in the respective year, insofar as it builds a balance deficit or surplus; the latter can be interpreted as following austerity measures, in accordance with Talving (2017). The range goes from −48,650 to 3,696 USD PPP per capita, therefore the resulting regression coefficients are very small in absolute terms. A negative balance means having a governmental budget deficit, thus following expansionary policies, a positive one means having a budget surplus, in other words contractionary policies, which is in the interest of austerity (cf. ibid.). For the migration crisis, a ratio of the net migration on the NUTS 1 level to the original population is taken. Negative values mean in this case that the respective region lost more people than it gained from migration. For the corona crisis, a ratio of positive corona cases to the population on the NUTS 1 level is taken. Only the migration crisis panel can be completely presented in the table 5.4, since only there, the sample was not restricted in any specification, so that no variable was omitted. In the two most sophisticated versions of the Eurocrisis panel, the time variable was omitted, hence only the first

5.1 Single Crisis Exposure Regressions

39

three versions are shown. For the corona crisis, the time variable was omitted in all specifications, thus it is completely left out. In the Eurocrisis panel, the country’s austerity coefficients have a very small amount; for the less sophisticated versions, the coefficients are significantly positive, only for the least sophisticated one, it is significantly negative. A thorough analysis is thus hindered. The time dummy is significantly positive and similar to the baseline

Table 5.3 Dummy robustness checks results (Single crisis) DV: Euroskepticism

(1)

(2)

(3)

(4)

(5)

OLS

OLS

OLS

OLS

OLS

0.0046

−0.0814***

−0.1440***

−0.1681***

−0.1489***

(0.00342)

(0.01250)

(0.02005)

(0.02803)

(0.03165)

−0.0643***

−0.0643***

(0.00496)

(0.00498)

Panel A: Eurocrisis HE (Alt.) Post-Crisis

0.0793*** (0.00206)

HE (Alt.) × Post-Crisis

0.0863*** (0.00501)

0.0855*** (0.00353) 0.0871*** (0.00485)

0.0849*** (0.00493) 0.0778*** (0.00711)

0.0356*** (0.01038)

0.0395*** (0.01038)

Adjusted R2

0.02

0.09

0.11

0.10

0.09

Observations

191,013

191,013

84,089

42,543

42,543

−0.0386***

−0.1553***

−0.1867***

−0.1927***

−0.1961***

(0.00326)

(0.01304)

(0.02283)

(0.02489)

(0.02767)

Panel B: Migration crisis HE (Alt.) Post-Crisis

0.1263*** (0.00417)

HE (Alt.) × Post-Crisis

−0.0011

(0.00517)

0.0083*/ (0.00483) −0.0010

(0.00502)

0.0303***/*

0.0307***

0.0317***/

(0.00755)

(0.00801)

(0.00800)

−0.0296***

−0.0376***

−0.0384***

(0.00857)

(0.00935)

(0.00932)

Adjusted R2

0.03

0.09

0.12

0.12

0.12

Observations

110,323

110,323

42,849

33,573

33,573

Panel C: Corona crisis HE (Alt.)

0.1118*** (0.00611)

Post-Crisis

0.0204***/** (0.00202)

0.0099

0.0110

0.0130

0.0142

(0.00714)

(0.01034)

(0.01026)

(0.01092)

0.0430*** (0.00223)

0.1345*** (0.00427)

0.1358*** (0.00450)

Adjusted R2

0.01

0.06

0.12

0.13

Observations

110,036

110,036

39,897

39,897

0.0700*** (0.00449) 0.09 39,897

(Continued)

40

5

Regression Results

Table 5.3 (Continued) DV: Euroskepticism

(1)

(2)

(3)

(4)

(5)

OLS

OLS

OLS

OLS

OLS

National FE

Y

Y

Y

Y

Regional FE

Y

Y

Y

Y

Year FE

Y

Y

Y

Y

Survey Round FE

Y

Y

Y

Y

Y

Y

Y

Y

Y

Demogr. & soc.-econ. Political attitudes Entropy balancing

Y

Note: The DV is a binary dummy variable showing if a respondent is Euroskeptic. All models include intercepts (not presented in the table). In parentheses, robust standard errors are reported. For the Eurocrisis-panel, the main independent binary variable shows whether the respondent lives in a GIIPS country (ES, GR, IE, IT or PT). For the migration crisis-panel, the binary variable shows if the respondent lives in a country which is situated at an external border of the EU (BG, CY, EE, ES, FI, FR, GR, HR, HU, IT, LT, LV, MT, PL, PT, RO, SI or SK). For the corona crisis-panel, the binary variable shows if the respondent’s country has the highest Covid casualty rate by mid-2020 (BE, ES, IT, SE or UK). For each panel, a time dummy of the crisis and an interaction term of the time and the country dummy are shown. The time dummies compare respondents just around those points in time where the respective crisis took place (Eurocrisis time dummy: pre-crisis 2010/2011–post-crisis 2012/2013, migration crisis: pre-crisis 2014–post-crisis 2015, corona crisis: pre-crisis end-2019–post-crisis 2020). For all specifications, the boottest-command was conducted in StataCorp 2021, based on the six-point distribution by Webb 2013 for the time dummy using the WCRB. This aims at getting cluster-robust standard errors which are clustered by the respondent’s country. Their p-values are reported only if they differ from the regular ones indicated by the stars. In that case, the stars after the slash symbolize the p-values using WCRB. The FE are national, regional, temporal and for each included round of the EB. Demographics and socio-economics: age (15–99), age squared, gender (male, female, none), marital status (married or not), education (age when full-time education stopped; 13 cat.), type of community (4 cat.), satisfaction with life (5 cat.) as well as socio-economics (employment status (18 cat.), historical employment status (16 cat.)). Political attitudes: self-positioning on left-right-scale (0–10), self-positioning on social placement via dummy variables (working, middle or upper class). In model (5), entropy balancing is applied to define matching weights with the time variable as treatment variable, and including the country dummy, the FE as well as the demographic, socio-economic and political attitude control variables. In panel C, the interaction term is omitted due to sample restrictions. See code book in the electronic supplementary material for further details. Y = yes; *** p < 0.01, ** p < 0.05, * p < 0.1.

5.1 Single Crisis Exposure Regressions

41

results. The interaction effect has a very small amount, is significantly negative and therefore different to the baseline results. The whole panel is thus not as clearly interpretable as was hoped for. The migration crisis panel is more straightforward, at least concerning the interaction effect: It is significantly positive for the more sophisticated models, exactly as in the baseline approach. Thus it shows that being exposed to a higher influx of newly arrived people let the Euroskeptic stance rise in those countries after the crisis peak. The time trend and the net migration by themselves are less easily interpretable: Both are only significant for the less sophisticated versions and have different algebraic signs between specifications. The p-values for the WCRB are even insignificant for all time trend coefficients. However, the tendency shows that the most sophisticated versions go in the same direction as the baseline versions, with a negative time trend and a positive effect for net migration. Thus, the hypothesis 1 (single crisis exposure) and hypothesis 2 (repeated crisis exposure) are again at least partly supported. For this robustness check, no corona crisis panel is presentable, due to sample restrictions. This points to the said problem of comparability of points in time concerning corona cases, because prior to the crisis, the cases are obviously equaling zero. For future research, the expectedly higher excess mortality would be advisable to use (see discussion). Overall, it is visible that the effects for the continuous approach are less lucid, yet mostly pointing in the same direction as the baseline dummy approach. In table 5.5, the binary Euroskepticism index was split into its three components (feeling not even partially as European, the EU image and believing to have no benefit from the EU membership) as elaborated above (see equation A.10—equation A.18 of the appendix in the electronic supplementary material). Each of the three serves as DV in the three crisis panels and only the most sophisticated version including entropy balancing is shown. The absolute amounts are only comparable for the first and the third specification, because the EU image was measured in five categories and not in a binary way. Nevertheless, the membership regression can only be shown for the first panel, as the time dummy was omitted in the other panels due to sample restrictions. For the Eurocrisis panel, all HE coefficients are negative, for the image and the membership versions even significantly. The EU image and the view on the membership are thus more driving the less Euroskeptic attitude in the highly exposed countries. This is exactly in line with the reasoning that these countries have profited greatly from EU membership, prior to the Eurocrisis (cf. Hale/Obstfeld 2016). The time trend is significant over all specifications, yet positive for the first two ones and only negative (as in the baseline approach) for the membership. This means the

42

5

Regression Results

found less Euroskeptic stance after the crisis for all countries stems from feeling a membership benefit. After the crisis, people had less the opinion of not having benefited from the membership and this suits the baseline reasoning of European solidarity (cf. Wallaschek 2020a, 2020b). The interaction effect is not really interpretable, as its algebraic signs are inconclusive and insignificant. The HE coefficients are significantly positive for both presented migration crisis specifications (identification and image), exactly as in the baseline approach. The interpretation is therefore the same as above (higher Euroskepticism, inter alia, potentially due to an overlapping of the affected countries). The time trend and the interaction effect are nevertheless less straightforwardly interpretable, as they are significant for the given specifications, but inconclusive. When looking at the WCRB, no significance at all can be found for the time trend. The inconclusiveness of the time trend is similar to the baseline approach, whereas there, the interaction

Table 5.4 Continuous robustness checks results DV: Euroskepticism

(1)

(2)

(3)

(4)

(5)

OLS

OLS

OLS

OLS

OLS

Panel A: Eurocrisis Austerity

−0.000006***

(0.000002) Post-Crisis

0.1090*** (0.00440)

Austerity × Post-Crisis

0.000009*** (0.000003) 0.0904*** (0.00522)

0.00002*** (0.000005) 0.0840*** (0.00753)

−0.00003***

−0.00002***

−0.00002***

(0.000003)

(0.000003)

(0.000004)

Adjusted R2

0.03

0.07

0.11

Observations

88,423

88,423

33,766

5.4485***

3.4622**

Panel B: Migration crisis Net migration (NUTS 1)

−0.7889***

(0.26413) Post-Crisis

0.1212***/ (0.00260)

Amount × Post-Crisis

2.0278*** (0.38794)

(0.94012)

(1.55202)

0.0034

0.0022

(0.00354)

(0.00501)

−0.6322

(0.39365)

2.1728*** (0.64704)

2.2093 (1.87016) −0.0036

(0.00564) 3.5665*** (0.78518)

2.4444 (1.86790) −0.0033

(0.00563) 3.6312*** (0.78123)

Adjusted R2

0.02

0.09

0.12

0.12

0.12

Observations

110,323

110,323

42,849

33,573

33,573

(Continued)

5.1 Single Crisis Exposure Regressions

43

Table 5.4 (Continued) DV: Euroskepticism

(1)

(2)

(3)

(4)

(5)

OLS

OLS

OLS

OLS

OLS

National FE

Y

Y

Y

Y

Regional FE

Y

Y

Y

Y

Year FE

Y

Y

Y

Y

Survey Round FE

Y

Y

Y

Y

Y

Y

Y

Y

Y

Demogr. & soc.-econ. Political attitudes Entropy balancing

Y

Note: The DV is a binary dummy variable showing if a respondent is Euroskeptic. All models include intercepts (not presented in the table). In parentheses, robust standard errors are reported. For the Eurocrisis-panel, the main independent variable states the stance of the respondent’s country on austerity, measured via its household deficit or surplus. For the migration crisis-panel, the ratio of net migration on the NUTS 1 level to the population is taken. For the corona crisis-panel, it is the ratio of positive corona cases to the population on the NUTS 1 level. For each panel, the interaction term of the continuous variable and the time dummy is shown. For all specifications, the boottest-command was conducted in StataCorp 2021, based on the six-point distribution by Webb 2013 for the time dummy using the WCRB. This aims at getting cluster-robust standard errors which are clustered by the respondent’s country. Their p-values are reported only if they differ from the regular ones indicated by the stars. In that case, the stars after the slash symbolize the p-values using WCRB. The FE are national, regional, temporal and for each included round of the EB. Demographics and socioeconomics: age (15–99), age squared, gender (male, female, none), marital status (married or not), education (age when full-time education stopped; 13 cat.), type of community (4 cat.), satisfaction with life (5 cat.) as well as socio-economics (employment status (18 cat.), historical employment status (16 cat.)). Political attitudes: self-positioning on left-right-scale (0–10), self-positioning on social placement via dummy variables (working, middle or upper class). In model (5), entropy balancing is applied to define matching weights based on the FE as well as the demographic, socio-economic and political attitude control variables. Due to sample restrictions, in panel A only models (1) though (3) are displayed. The corona crisispanel is completely left out for the same reason. See code book in the electronic supplementary material for further details. Y = yes; *** p < 0.01, ** p < 0.05, * p < 0.1.

effect is positive and highly significant (i.e., Euroskepticism rose for the highly exposed countries after the crisis). This latter effect is seemingly driven by the coefficient for the EU image. For the corona panel, the country coefficients are significantly positive, the time trend coefficients positive and highly significant. This is similar to the baseline approach, thus it can be thought of that these two components (identification and image) drive the effect. The interaction term is insignificant and around zero and

44

5

Regression Results

thus not interpretable (similar to the baseline results). Summing up, the conducted driver tests help to understand the composition of the measured baseline effects.

Table 5.5 Driver tests (Single crisis) OLS DV: Panel A: Eurocrisis HE

(1) Not ev. part. Europ.

−0.0302 (0.04779) 0.1602*** Post-Crisis (0.01532) −0.0126 HE × Post-Crisis (0.01799) 0.13 Adjusted R2 Observations 30,742 Panel B: Migration crisis HE 0.1450*** (0.04130) Post-Crisis −0.0309**/ (0.01321) HE × Post-Crisis 0.0039 (0.01694) Adjusted R2 0.11 Observations 22,322 Panel C: Corona crisis HE 0.1553** (0.07846) Post-Crisis 0.0588*** (0.01290) HE × Post-Crisis 0.0127 (0.01638) Adjusted R2 0.12 Observations 26,266

(2) EU im.

(3) No memb. ben.

−1.0613*** (0.07896) 0.1404*** (0.02892) 0.0494) (0.03276) 0.13 42,066

−0.3999*** (0.03901) −0.1121***/** (0.01613) 0.0281 (0.01870) 0.16 35,725

0.3291*** (0.05858) 0.0428*/ (0.02191) 0.0481* (0.02738) 0.13 33,341 0.2605* (0.13568) 0.0868*** (0.02152) −0.0403 (0.02555) 0.14 39,381 (Continued)

5.1 Single Crisis Exposure Regressions

45

Table 5.5 (Continued) OLS DV:

(1) Not ev. part. Europ.

FE Y Dem., soc.-econ. & pol. att. Y Entropy balancing Y

(2) EU im.

(3) No memb. ben.

Y Y Y

Y Y Y

Note: The DV of the first model in all three panels is a binary variable stating if the respondents do not even feel partly as European, the second one constitutes the respondents’ image of the EU on a scale from one (very positive) to five (very negative) and the third model is based on a binary variable stating if the respondents say that their country did not benefit from the EU membership. Due to sample restrictions, the third model is omitted in panel B and C. All models include intercepts (not presented in the table). In parentheses, robust standard errors are reported. For the Eurocrisis-panel, the main independent binary variable shows whether the respondent lives in a GIIPS+ country (CY, ES, GR, HU, IE, IT, LV, PT or RO). For the migration crisis-panel, the binary variable shows if the respondent lives in a country which is highly exposed to the migration crisis (AT, CY, DE, ES, FR, IT, LU, MT, SE or UK). For the corona crisis-panel, the binary variable shows if the respondent’s country is highly exposed to the corona crisis (BE, CZ, ES, HR, IE, IT, LU, SE, SI or UK). For each panel, a time dummy of the crisis and an interaction term of the time and the country dummy are shown. The time dummies compare respondents just around those points in time where the respective crisis took place (Eurocrisis time dummy: pre-crisis 2010/2011–postcrisis 2012/2013, migration crisis: pre-crisis 2014–post-crisis 2015, corona crisis: pre-crisis end-2019–post-crisis 2020). For all specifications, the boottest-command was conducted in StataCorp 2021, based on the six-point distribution by Webb 2013 for the time dummy using the WCRB. This aims at getting cluster-robust standard errors which are clustered by the respondent’s country. Their p-values are reported only if they differ from the regular ones indicated by the stars. In that case, the stars after the slash symbolize the p-values using WCRB. The FE are national, regional, temporal and for each included round of the EB. Control variables: Demographics (age (15–99), age squared, gender (male, female, none), marital status (married or not), education (age when full-time education stopped; 13 cat.), type of community (4 cat.), satisfaction with life (5 cat.)), socio-economics (employment status (18 cat.), historical employment status (16 cat.)), political attitudes (self-positioning on left-right-scale (0–10), self-positioning on social placement via dummy variables (working, middle or upper class)). In all models, entropy balancing is applied to define matching weights based on the FE as well as the demographic, socio-economic and political attitude control variables. See code book in the electronic supplementary material for further details. Y = yes; *** p < 0.01, ** p < 0.05, * p < 0.1.

46

5.2

5

Regression Results

Repeated Crisis Exposure Regressions

This subsection aims at giving a clearer view if—and if so, how—being exposed to more than one crisis within few years has an impact on the Euroskeptic attitudes of respondents from these countries. Therefore, in the baseline approach all the dummies from the single crisis baseline approaches are taken again and four further dummies are added, next to the corona crisis time dummy (end of 2019 as prior, 2020 as after). These four new dummies show if a country has been exposed to more than one crisis. The twofold exposed countries are thus those which suffered from both the Eurocrisis and the migration crisis (i.e., CY, ES and IT), the Eurocrisis and the corona crisis (i.e., ES, IE and IT) or the migration crisis and the corona crisis (i.e., ES, IT, LU, SE and UK). ES and IT have been exposed to all three crises. The countries are allocated as follows: If a country is both part of the HE group for one crisis and for another, it is part of the respective twofold exposed group. If it was even highly exposed to all three crises, it belongs to the threefold exposed group. Thus, there are seven HE dummies and each of them is interacted with the said time dummy. This most current time dummy has to be used, as only by then, all three crises have already happened. This gives many explanatory variables, therefore, the way of representing the results is different to the single crisis exposure results: For a better visibility, the results are presented in figures which are generated with the marginsplotcommand of StataCorp (2021). These figures show the average marginal effects, thus the derivative changes with respect to the main explanatory variables (cf. Angrist/Pischke 2009). The effects and their 95% confidence intervals (CI) are therefore presented in a more lucid way, their significance level is more directly graspable. In the appendix in the electronic supplementary material, the complete regression results are shown for the most sophisticated versions. Again, placebo tests with the corona placebo time dummy and a driver test via the split-up of the Euroskepticism index are conducted. The underlying regression results for the figures of the baseline and the placebo approaches can be found in the table A.3 and the underlying equations are equation 4 and equation A.19. The underlying results of the driver test figures are stored in table A.4, based on equation A.20—equation A.22. Due to sample restrictions, the time dummy is omitted for the membership version of the driver tests, thus it is neither shown as figure nor in the table. The interpretation of the interfaction terms is more demanding for the repeated exposure because of the different composition of the country groups. For example, for ES and IT, the two countries which have been exposed to all three crises, all coefficients have to be summed up to get their effect.

5.2 Repeated Crisis Exposure Regressions

5.2.1

47

Baseline & Placebo Regressions (Repeated Exposure)

To begin with the time dummy effect: For all specifications, it is positive and highly significant, yet less for the placebo (which is potentially positive, as well, due to offshoots of the then recent migration crisis). This was expected from hypothesis 1 and is analogous to the findings for the corona crisis in the single crisis approach: For all included countries, the corona crisis seems to have a Euroskepticism-enlarging effect. Concerning the interpretation of the exposure to one crisis, it is visible that analogously to the findings for the single crisis cases, a downward-sloping effect on Euroskepticism of having been highly exposed to the Eurocrisis can be found, both prior—due to the significant HE coefficient—and after the corona crisis— due to the significant interaction effect. This supports hypothesis 3 (resilience) and hypothesis 4 (antecedent crises), yet contradicts partly hypothesis 5 (hysteresis). For the migration crisis, the found effect is upward-sloping, yet only beforehand; afterwards, it is inconclusive due to missing or low significance. For being highly exposed to the corona crisis alone, no clear effect can be shown, neither before nor after the crisis. The placebo results mostly go in the same direction, yet are less significant. The significantly positive placebo finding for the corona crisis can be explained again with the then recent migration crisis. In the following, the country dummies for twofold repeated exposure experiences are interpreted. It is visible that for those countries that suffered from Eurocrisis and migration crisis, but have not extraordinarily been exposed to the corona crisis, this repeated crisis exposure does not seem to matter, at least not closely prior or after the corona crisis. This can be interpreted from the insignificant values in both the baseline and the placebo approaches for the country and the interaction effect. This finding supports both the antecedent-crises-hypothesis 4, insofar as the Eurocrisis is seemingly too long gone to have a broad impact as well as the resilience-hypothesis 3, insofar as this longer gone twofold exposure is less Euroskepticism-enhancing. Concerning those that suffered severely from both the Eurocrisis and the corona crisis, a Euroskepticism-enlarging effect is visible, yet only prior to the corona crisis, as only the country effect but not the interaction effect is significant. This is the other way around for the placebo models, thus supporting the findings and showing that at this earlier point in time, the Eurocrisis exposure still has an impact. A potential explanation might be that the three countries in this group (i.e., ES, IE and IT) had huge impacts from both crises, and when the corona crisis started, the aftermath of

48

5

Regression Results

the Eurocrisis was not completely solved, especially in IT and ES (cf. Bozorgmehr et al. 2020). A potential nexus might also be the restrictions to the health sector after the Eurocrisis which was highly needed during the corona crisis and due to that shortage, a potential Euroskeptic feeling might have grown (cf. ibid.). Thus, it supports the resilience-hypothesis 3, insofar as respondents from these countries are crisis-tested and the hysteresis-hypothesis 5, insofar as they have built a certain amount of Euroskepticism and keep it at that level. For countries which suffered from migration and corona crisis, no difference to the comparison group can be found prior to the corona crisis, as the country effect is insignificant, yet afterwards, a significantly positive effect is visible. That means, countries which have been double exposed to migration and corona crisis became more Euroskeptic than the comparison group, thus supporting the repeatedcrisis-exposure-hypothesis 2. This speaks again in favor of the resilience-hypothesis 3, because respondents from these countries seemingly reacted quicker in their political behavior after the second crisis happened. The already previously existing Euroskepticism has been reinforced, supporting the hysteresis-hypothesis 5. Being exposed to all three crises, i.e., ES and IT, has in the baseline versions a significantly negative effect, prior to the corona crisis. This could be interpreted as resilient, supporting the homonymous resilience-hypothesis 3: Having known the prior crises and the pan-European handling, the affected became less Euroskeptic compared to others. However, as the interaction term is insignificant, a threefold exposure does not have an impact on Euroskepticism at the point in time post-crisis. Regarding the twofold exposure, the placebo findings support the baseline results, because they are all insignificant, except for the placebo interaction term of highly Eurocrisis and corona crisis exposed countries. The overall interpretation of the results is that, as expected, a hysteresis-like effect becomes visible. Thus, people tend to partly forget their former crisis exposure, the longer gone it is, but once a certain level of Euroskepticism is reached, it is kept and not dismantled again (see hypothesis 4 (antecedent crises) and hypothesis 5 (hysteresis)). Probably, the new exposure experiences superimpose the older experiences. Hypothesis 2 (repeated crisis exposure) is only fulfilled for the double exposure to the migration crisis and the corona crisis (Figure 5.1).

Figure 5.1 Repeated crisis exposure: Average marginal effects. (Note: The DV is a binary dummy variable showing if a respondent is Euroskeptic. All marginsplot-figures display the average marginal effects with 95 % CI. Figures a) and b) are the baseline approaches with the true corona time dummy (pre-crisis end-2019, post-crisis 2020), c) and d) use the placebo corona time dummy (pre-crisis 2018, post-crisis 2019). Figures a) and c) exclude, b) and d) include entropy balancing. HE (Eurocrisis): CY, ES, GR, HU, IE, IT, LV, PT, RO; HE (Migration crisis): AT, CY, DE, ES, FR, IT, LU, MT, SE, UK; HE (Corona crisis): BE, CZ, ES, HR, IE, IT, LU, SE, SI, UK; HE (Euroand migr. crisis): CY, ES, IT; HE (Euro- and cor.-crisis): ES, IE, IT; HE (Migr. and cor. crisis): ES, IT, LU, SE, UK; HE (Euro-, migr. and cor. crisis): ES, IT. See the appendix A.3 and code book in the electronic supplementary material for further details)

5.2 Repeated Crisis Exposure Regressions 49

Figure 5.1 (continued)

50 5 Regression Results

5.2 Repeated Crisis Exposure Regressions

5.2.2

51

Driver Tests (Repeated Exposure)

As in the single crisis case, also for the repeated exposure case a driver test of the baseline results is conducted via splitting the Euroskepticism index into its three underlying components: Feeling not even partially as European, the EU image and having no membership benefit. The latter is not shown in the results, because the time variable has been omitted and therefore no meaningful estimation is possible. For the two remaining components, the most sophisticated version including entropy balancing was calculated. The results are shown in the appendix table A.4 in the electronic supplementary material. Again, the results are presented via figures which are built with the marginsplot-command in StataCorp (2021). The underlying equations are equation A.20 and equation A.21. The time trend is significantly positive for both models, just as in the single and the repeated crisis interpretations. Also, the single crisis exposure variables point in the same directions as before: Eurocrisis downward- (for both models), migration crisis upward-sloping (for both models), corona crisis ambiguous (for the identification only). Interestingly (because differently to the baseline measure), the effect of the highly corona crisis exposed countries on the EU image is both prior and after this crisis significantly negative and thus has a downward-sloping effect. All other single exposure interaction terms with time, however, are insignificant for both models. Regarding the twofold exposure: The Eurocrisis and migration crisis exposed countries have a significant effect on the identification and the image. The overall insignificant effect in the baseline approach is thus driven by the membership benefit part of the index. Following the identification and the image results, the interpretation would be different, insofar as then these countries had a higher Euroskepticism, at least prior to the crisis and thus supporting the hysteresis-hypothesis 5. After the crisis peak, a clear interpretation is not possible, because the interaction terms are insignificant. It is similar for those countries which have been highly exposed to Eurocrisis and corona crisis: The effect is strongly significant and positive for the identification and the image and these two thus drive the significantly positive baseline finding. Post-crisis, it is significantly negative, in the baseline approach it was not significant. This tendency supports the resilience-hypothesis 3. The effect for the highly exposed countries to migration and corona crisis is insignificantly positive for the baseline approach, but significantly positive for the identification and the image, thus driven

5

Figure 5.2 Repeated crisis exposure (driver tests): Average marginal effects. (Note: Both figures display the average marginal effects with 95% CI. Figure a) uses a binary variable as DV which states if the respondent does not even partially feel as European. Figure b) uses a variable as DV which is on a scale from one (very positive) to five (very negative) and it constitutes the respondent’s image of the EU. The third component of the baseline binary variable, a binary variable stating the respondents view on the benefit from the EU membership, is omitted due to sample restrictions. All models include entropy balancing. HE (Eurocrisis): CY, ES, GR, HU, IE, IT, LV, PT, RO; HE (Migration crisis): AT, CY, DE, ES, FR, IT, LU, MT, SE, UK; HE (Corona crisis): BE, CZ, ES, HR, IE, IT, LU, SE, SI, UK; HE (Euro- and migr. crisis): CY, ES, IT; HE (Euro- and cor.-crisis): ES, IE, IT; HE (Migr. and cor. crisis): ES, IT, LU, SE, UK; HE (Euro-, migr. and cor. crisis): ES, IT. See the appendix table A.3 and code book in the electronic supplementary material for further details)

52 Regression Results

5.2 Repeated Crisis Exposure Regressions

53

into insignificance by the membership benefit. Post-crisis, a significantly positive effect is only given for the identification, thus it is the component which drives the likewise significantly positive baseline effect. Being exposed to all three crises, has a highly significant and negative effect for identification and image, analogously to the baseline findings. These two components thus drive the overall effect. Postcrisis, likewise to the baseline results, the effect is insignificant in the driver tests (Figure 5.2).

6

Discussion

In the following section, the shown results are critically examined in terms of their overall fulfillment of the hypotheses. Afterwards, potential limitations concerning the data set or alternative ways to define and measure some of the variables are presented. The identification of potential shortcomings or methodological difficulties will help to ensure that appropriate adjustments are made in future studies.

6.1

Fulfillment of the Hypotheses

Subsequently, the five postulated hypotheses are checked for their overall fulfillment. Furthermore, potential recommendations for the handling of future crises are derived. For all hypotheses, corroborating factors have been shown, yet some verifications are stronger than others. The single crisis exposure-hypothesis 1 is verified for the Eurocrisis, the migration crisis and the corona crisis, both in the single crisis exposure regressions as well as in the repeated crisis exposure regressions. Also, the placebo and driver tests as well as the robustness checks support this finding. Respondents from highly exposed countries develop more Euroskepticism after the respective crisis happened than people from less exposed countries; evident at least for the Eurocrisis and the migration crisis. For the corona crisis, it is visible that the overall time trend is Euroskepticism-enhancing, thus not only the highly affected countries but all EU member states experience a growing Euroskepticism in their people. This last finding can be explained with the exceptionality of the corona crisis for everyone. In future research, when the crisis will be overcome, potential differences between the highly and less exposed countries might be visible. Nevertheless, as more pandemic waves still lay ahead and the situation regarding the vaccination evolves, the assessment of which countries to include in the highest exposure group will vary. © The Author(s), under exclusive license to Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2022 L. Möller, Repeated Crisis Exposure, Euroskepticism & Political Behavior, BestMasters, https://doi.org/10.1007/978-3-658-39267-3_6

55

56

6

Discussion

For future crises, it is important to derive political strategies and interventions in order to counterbalance the rising Euroskepticism. In between crises, it is crucial to ameliorate EU policies, procedures and international relations: This implies reacting to criticisms and tackling skepticism, also indirectly, e.g., via cohesion policy (cf. Rodríguez-Pose/Dijkstra 2021). For the repeated crisis exposure-hypothesis 2, first confirmatory hints are found in the single crisis exposure regressions: In the baseline and the placebo approach of a single exposure to the migration crisis, effects are visible that can be attributed to aftereffects of the Eurocrisis. In the repeated crisis exposure regressions, however, the findings have been of a mixed nature: Only for the twofold exposition to the migration and the corona crisis, a clear effect after the corona crisis can be seen. For the other twofold or threefold exposures, the found effects are not sufficiently different from zero (see figure 5.1). This finding shows the relevance for policy makers to bear in mind not only current crises, but also former ones, which may still be partly unsolved. Euroskepticism is seen as a “disease” (Harmsen 2010; Ozlem Ultan/Ornek 2015) or a “malaise” (Hooghe/Marks 2007), yet once infected, the cure takes time. Concerning hypothesis 3 (resilience), hypothesis 4 (antecedent crises) and hypothesis 5 (hysteresis), their partial overlap impedes a clear differentiation. Nevertheless, the finding that in the repeated exposure setting the Eurocrisis—as longest gone crisis in this study—has a less significant impact than the migration crisis, supports the antecedent-crises-hypothesis 4. The notion of Funke et al. (2016) that austerity measures have an impact on political behavior for up to ten years is somewhat limited by this finding. In addition, it can be interpreted as the twofold exposed individuals developing resilience, meaning that they react less intensely than they did during their first exposure (cf. Essers 2013; Paul/Roos 2019). Finally, this can be attributed to the hysteresis-effect, insofar as building up Euroscepticism is quicker than dismantling it. It is obvious that these three hypotheses need more specified research in the future in order to get more extensive insights. On a more global level, it seems to be important to establish a successful pan-European crisis management in order to stop an acceleration of Euroskeptic tendencies (cf. Harteveld et al. 2018; Hobolt/Tilley 2014). Often people and national politicians tend to blame the EU for something, even if it clearly does not fall within the EU’s area of responsibility (cf. Harteveld et al. 2018; Hobolt/Tilley 2014). In this context, prospective heterogeneity analyses could account for effects of one’s level of education, one’s employment status, or one’s left-right-placement.

6.2 Potential Limitations

6.2

57

Potential Limitations

When using survey samples, the values of the adjusted R2 are often relatively low, identical to the values of the adjusted R2 in the present study. This consideration points to a central limitation: Via surveys, it is nearly impossible to explore all potential reasons for which a respondent’s view may be classified as Euroskeptic (cf. King 1990). For the present study, it follows that the definition of the concept of Euroskepticism in particular is crucial, but that, as mentioned above, there is no exclusive way to define it. Therefore, alternative ways how to measure and define Euroskepticism are discussed in the following. First, Baimbridge (2018) uses turnout in EP elections to measure Euroskepticism. He justifies his choice with economic voting reasoning: People who are less interested in European politics tend to abstain from their vote (“strategic non-voting”; ibid.). This is a simple approach to measure the phenomenon of Euroskepticism, yet it is criticized: The fact that overall turnout in the EP is rising and that more and more people are voting for Eurosceptic parties are brought forward, as are structural and personal difficulties (cf. Franklin 2001; Hernández/Kriesi 2016; Schäfer 2017). This approach is therefore not followed in the present study, but could represent a robustness check in future research. Second, Werts et al. (2012) combine distrust in the EP and the respondents’ stance on the process of the European unification as their definition of Euroskepticism. They stress the difficulties of measuring said concept and therefore combine these two aspects. Their approach is based exclusively on ESS data. For the present study, it was however problematic that due to the corona crisis the 2020-ESS-round had to be postponed and will be available only by 2022 (cf. European Social Survey 2021). That is why the usage of ESS data has been completely renounced in this study, but could be included in future research in order to verify the findings. Third, in several studies, questions from the post-election survey EES have been used to define Euroskepticism (cf. Vasilopoulou 2018). The focus of the EES lies on electoral consequences, not on general political behavior on which the present study focuses, as stated above, yet it could serve as future extension or validation. A further alternative measure and definition would have been the party-based view on Eurospekticism, e.g., via the PopuList, where both left- and right-wing Euroskeptic parties are listed (cf. Rooduijn et al. 2019; Vasilopoulou 2018). Concerning these Euroskeptic parties, parliamentary seats or voting intentions of respondents have been looked at, e.g., by Treib (2021). In contrast, the present study concentrates on individual behavior. Therefore, also expert surveys such as the Chapel Hill expert survey are not taken into account, because there the whole parties’ programs are screened and not individual decisions (cf. Jolly et al. 2022; Vasilopoulou 2018).

58

6

Discussion

These expert surveys are not able to take all existing Euroskeptic parties into account, some parties are too small, new parties emerge and older ones might disappear; the sample selection thus might suffer from a selection bias (cf. Angrist/Pischke 2009). Therefore, the exclusive use of the EB as base for this study enables a thorough analysis. Nevertheless, some researchers consider using EB data as problematic, insofar as the European institutions are criticized for being both the institutions that finance the study and that are evaluated in it (cf. Höpner/Jurczyk 2012). Since these institutions are interested in presenting the best EU image possible, this potential bias could lead to even less reported Euroskepticism; the true Euroskepticism of the respondents might thus be stronger than the detected one. Concerning the EB methodology, a potential social desirability bias might be given which may similarly bias the found effects: The respondents are interviewed personally and not anonymously, thus they could tend to dampen their true opinions in order to better fulfill estimated expectations of the interviewers (cf. Gerling/Kellermann 2021; GESIS 2021a; Krumpal 2013). Another potential limitation might be that the EB data is not given as panel data, but as a repeated cross-section (see above; cf. GESIS 2021a). It would have been preferable to use panel data, especially to better monitor the development of the political behavior of individuals from crisis to crisis and also concerning the age when a respondent was exposed to a crisis (cf. Angrist/Pischke 2009). Thus, a limitation of this study may be that not the same sample has been observed over the whole time range. Nevertheless, the described statistical measures, especially the entropy balancing, help to dampen possible restricting effects for the analysis. For future research, a panel study would be preferable, yet the availability of suitable data seems questionable. Concerning the components of the here applied Euroskepticism-index, Serricchio et al. (2013) emphasize that inquiring the respondents’ view on their country’s membership benefit is often used regarding the existence of long-term Euroskepticism, next to the respondents’ view on the velocity of the European unification. But, such usage of the membership benefit is criticized by Guerra (2013) and Leconte (2010) due to many people’s little knowledge about the EU and the relatively high volatility of this measure. Still Serricchio et al. (2013) see the membership benefit as a central factor of individual Euroskepticism, displaying “scepticism, rather than criticism”: Skepticism mostly is directed on a complete withdrawal, whereas criticism aims at ameliorating existing structures. Additionally to the applied driver tests for single and for repeated crisis exposure, further indices could deliver another source of robustness check, e.g., comparing the use of only two of the components as Euroskepticism index. As another robustness check, Euroskepticism could be changed to satisfaction with national governments or with the EU in its entirety, respectively with the EU solidarity during the crises (cf. Reinl 2020).

6.2 Potential Limitations

59

It is shown in this study that—clearly for the Eurocrisis and the migration crisis, indirectly for the corona crisis—after the respective crisis peak, people from HE countries have been leaning more towards Euroskepticism. This might also be a sheer indicator for the overall time trend that people become more Euroskeptic and more nationalistic over the years, e.g., as the strengthening of populist parties all over Europe shows (cf. Guriev/Papaioannou 2020; Rodrik 2018; Taggart 2020). Prospectively, the clear distinction between the overall time trend and the impact of the crises themselves should be inspected more in-depth. The choice of the crises peak times might be worth discussing: When investigating, e.g., elections, as Gerling/Kellermann (2021) do, the time is exactly determined. In the present case, the opinions differ when a crisis started, had its peak or ended (cf. Funke et al. 2016). This might even differ from country to country, or from region to region. Furthermore, a nexus between the starting time and the kind of the ruling parties and its leaders might be found, e.g., if during a crisis a populist party had to implement unpopular measures. The approach followed in this study can thus be understood as an approximation concerning several aspects: In future studies, these issues should be further elaborated. In terms of missing data, no EB data have been conducted in HR for the year 2012, because at that time HR was still a candidate state and only member states have been surveyed. This can be considered as a minor problem because of HR’s small size and the resulting lack of only few data. The absence of data for the UK in the second round of 2020 due to Brexit, however, might be seen more problematic: This data could have been helpful in studying the impact of the corona crisis, especially since the UK is considered as one of the especially exposed countries. Concerning the governmental expenditures and revenues, data for BG, CY, HR and MT is missing (see the continuous robustness check of the Eurocrisis panel). Therefore, the presented regression results might slightly differ if the data was available. Due to the relative small size of these countries, their impact would assumingly be very little. Furthermore, BG, CY and HR can be seen as part of the group of countries which have been hit harder by the Eurocrisis than others and thus had to enforce austerity measures (cf. Kopri´c et al. 2019; Koutsampelas et al. 2021; Petkov 2014). MT did not install any austerity instrument (cf. Michael/Christofides 2020). Since the found effects in this robustness check are already small and rather difficult to interpret, an inclusion of this additional data would probably have no consequences. Another potential limitation is the fact, that not for the whole included time range data of governmental expenditures and revenues is available. Nevertheless, the used annual data does stem from those years which are of the highest relevance concern-

60

6

Discussion

ing the examined crises (2006, 2007, 2009, 2011, 2013–2020)1 . In future research, it would be preferable to add all missing countries and years to potentially get a cleaner effect. Also concerning the explanatory variables, different approaches might be prospectively taken into consideration. An alternative approach for defining and measuring austerity in the Eurocrisis panels via fiscal indicators stems from House et al. (2020) who looked at the difference between the prediction of the amount a government wanted to spend and how much it actually did spend. Due to the fact that such data is less available, this study follows the closely related approach of Talving (2017), focusing on governmental expenditures and revenues. A further alternative approach measures a country’s public debt in relation to its amount of Gross Domestic Product (GDP; cf. Engler/Klein 2017; House et al. 2020; Mota et al. 2020). The rationale is that austerity aims at lowering the public debt, yet tries to affect the GDP as little as possible (cf. Engler/Klein 2017; House et al. 2020). Since there is not one sustainable ratio for all countries due to country-specific characteristics, this approach might have led to further problems and has not been followed (cf. Engler/Klein 2017). For verification reasons, these other approaches of defining austerity might potentially be examined. For the continuous migration panel robustness check in this study, the used data is “net migration plus statistical adjustment” (Eurostat 2021a). Exclusive data on incoming migrants or on refugees could be a fruitful alternative. Furthermore, the EB data is not sufficient regarding potential personal migration experiences of the respondents: It would be worth checking if these experiences might have consequences on the evaluation of crises (cf. Batista et al. 2019; Careja/Emmenegger 2012; Maxwell 2010). People with this background might be influenced concerning a potential Euroskepticism: On the one hand, they might be more open towards Europe, because they have experienced a reality in other, less democratic parts of the world. On the other hand, their cultural background may reinforce a Euroscepticism: Some studies have found that people with migratory experiences tend to be in particular critical towards next generations of incoming migrants (cf. Gans 1992; Groenewold 2008). It would be desirable to outline this potential nexus more thoroughly. An alternative approach for examining the corona crisis relies on the excess mortality rate: This continuous approach would base further regressions on data of the amount of deaths before and during the crisis (cf. Backhaus 2020). Concerning all corona data, it is important to acknowledge that a comparison of numbers is not 1

For LU and CZ, data for 2014, for SI and EE 2006, for LT 2006, 2011, 2013, 2014, for LV 2006 and 2011 are missing, for RO only 2007 and 2019 are given.

6.2 Potential Limitations

61

completely possible, because the testing infrastructures have been vastly different over time and between regions. Due to two reasons, the number of infections are only taken until the end of 2020: First, suitable data has not been completely available. Second, this also makes sense contentwise: At that time, only very few Europeans already had gotten their vaccinations (cf. Della Polla et al. 2021). The potential effect from behaving differently because of being vaccinated is thus left out and cannot interfere with the results. Concerning potential interconnections of the corona crisis with other examined crises, several further questions open up. In several countries, Eurocrisis-related austerity measures negatively affected the health systems: People might have been healthier to cope with Covid without these prior measures and the hospitals might have been better functioning during the corona crisis (cf. Bozorgmehr et al. 2020; Karanikolos et al. 2013; Reeves et al. 2014; Sherpa 2020). Furthermore, it can be thought of the possibility that several individuals who have already been Euroskeptic became Covid deniers, or Covid deniers additionally developed a Euroskepticism (cf. Makarychev/Crothers 2020). Moreover, medical topics might have become more relevant for the people’s political attitudes (cf. Hellwig 2008; Magalhães 2014). Governmental decisions related to the corona crisis might have had different impacts at different points in time, e.g., the EU vaccination strategy was viewed as meaningful in the beginning due to the joint purchase power, but was later considered as too slow in comparison with other countries (cf. Vanhuysse et al. 2021; Warren/Lofstedt 2021). All these questions could be answered in future research, against the background of the results found in this study. In studies to come, the single effects of the control variables via heterogeneity analyses could be examined, e.g., whether the results presented in this study differ between gender, age or job. Concerning the choice of control variables, several additional ones would have been desirable to include, e.g., the respondents’ religion and whether they attend their house of prayer regularly (cf. Werts et al. 2012). Also, how respondents perceive their own economic status and whether they see an ethnic or a criminal threat might have had an influence, yet unfortunately, no suitable information is given in the used EB data. Additionally, it might have proven beneficial to know whether the respondents have been living in their area for several years or moved there just recently and where they came from originally, e.g., from which former part of the reunited DE. For comparative reasons, it would be helpful to additionally take non-EU member states into account; the EB focuses on EU member and candidate states (cf. Höpner/Jurczyk 2012). When the ESS publishes its 2020 round in the end of the year 2022, a new analysis will be possible where also countries such as Norway or Switzerland may be taken into consideration (cf. European Social Survey 2021).

62

6

Discussion

Furthermore, checking for potential differences within countries, e.g., via the addition of a border location dummy for regions, might add insights. The present study aimed at getting an overview for the EU. Future research might be able to enlarge and deepen findings of this study for specific countries or regions only. As shown above, the highly affected countries still differed in the intensity how the crises affected them. When exclusively using the baseline dummy approach, potential differences within this group can not be found. Therefore, in order to find more distinctions, a consistent continuous measure is advisable in prospective research. A potential variation could be not to use a binary index, but a scale version for the intensity of Euroskepticism. Concerning the repeated crisis exposure, a separate view on two of the three examined crises, e.g., on the Eurocrisis and the migration crisis only, might have an additional enlightening effect. Thus, the separate effect of having been highly exposed to the Eurocrisis when the migration crisis happened could be compared to the findings from the present study. Also a combination of the continuous approach with the repeated exposure dummies can be thought of. Furthermore, using the alternative country definitions for the highly exposed countries as a dummy robustness check, and accordingly adapted twofold respectively threefold exposure dummies could be tested for. In summary, this discussion shows that this study contributes to the existing body of research, but also identifies many research gaps that need to be addressed further.

7

Conclusion

This study shows the impact of single and of repeated crisis exposure, within the last fifteen years, on the political behavior of European individuals. To this end, multifaceted econometric methods have been applied in order to provide evidence and causality: Through the use of restrictive measures, such as entropy balancing as well as through the conducted placebo and the driver tests plus the robustness checks, the following findings are supported: Each crisis exposure by itself has in highly exposed countries an impact on the respondents’ political behavior. A novel contribution to the recent strand of literature which is accomplished by this study, is the focus on the repeated exposure of crises. Over time, people tend to forget prior crisis exposure in their political behavior. A hysteresis-effect becomes visible. An effect of the migration crisis is still distinguishable during the corona crisis, while direct effects from the Eurocrisis are more or less gone. It might have been different, if a greater time span separated the first, second and the third crisis, but in today’s interconnected world, one crisis promptly follows another (cf. Bozorgmehr et al. 2020). People know how to adapt, but some experiences have longer-lasting effects than others. Respondents from crisis-tested countries tend to develop resilience. Therefore, prospective studies need to keep on researching on the phenomenon of Euroskepticism in order to understand the rising criticisms towards the EU. This should be done for a different time range (i.e., considering only two crises or comparing more than three at the same time), a different set of countries (i.e., comparing the EU member states with neighboring non-EU countries or with other world regions) or for a different database. Zooming in on the situation in just one country or one region, also in comparison to its neighbors, might yield further insights. These additional insights might help to understand and to counter potential skepticism towards the EU more effectively. If everybody enjoys the advantages of a life in the EU, but nobody reacts to unsubstantiated skepticism, a further growing Euroskepticism is

© The Author(s), under exclusive license to Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2022 L. Möller, Repeated Crisis Exposure, Euroskepticism & Political Behavior, BestMasters, https://doi.org/10.1007/978-3-658-39267-3_7

63

64

7

Conclusion

not surprising (cf. Taggart/Szczerbiak 2018). Having said this, substantiated criticism needs to be addressed, in order to cope with future crises. For instance, a renewed look on the still ongoing corona crisis might lead to different results, if reflected upon with some distance, once its critical stage has been fully vanquished. The found results and the comprehension of political behavior in past crises may help to develop and evaluate procedures for future crises. One central example may be the handling of the climate crisis that may affect each and every citizen. A pan-European coordinated process would proof beneficial and ensure high acceptance of the transformative steps to fight the climate crisis. The more thoroughly the analyses are thought through in advance, the more successful an implementation can be. The present study makes a contribution for this purpose.

References

Alesina, A.,/Carloni, D., & Lecce, G. (2011). The Electoral Consequences of Large Fiscal Adjustments. NBER Working Paper Series, W17655. https://doi.org/10.3386/w17655 Alesina, A.,/Favero, C., & Giavazzi, F. (2019). Austerity. Princeton: Princeton University Press. https://doi.org/10.1515/9780691185019 Alesina, A.,/Perotti, R.,/Tavares, J.,/Obstfeld,M., & Eichengreen, B. (1998). The Political Economy of Fiscal Adjustments. Brookings Papers on Economic Activity, 1998 (1), 197– 266. https://doi.org/10.2307/2534672 Angrist, J. D., & Pischke, J.-S. (2009). Mostly Harmless Econometrics: An Empiricist’s Companion. Princeton: Princeton University Press. https://doi.org/10.1515/9781400829828 Arias, E., & Stasavage, D. (2019). How Large Are the Political Costs of Fiscal Austerity? The Journal of Politics, 81 (4), 1517–1522. https://doi.org/10.1086/704781 Baas, T., & Brücker, H. (2011). EU Eastern Enlargement: The Benefits from Integration and Free Labour Movement. CESifo DICE Report, 09 (2), 44–51. http://hdl.handle.net/10419/ 167037 Backhaus, A. (2020). Common Pitfalls in the Interpretation of COVID-19 Data and Statistics. Intereconomics, 55 (3), 162–166. https://doi.org/10.1007/s10272-020-0893-1 Baimbridge, M. (2018). The impact of the economic crisis on Euroscepticism. In B. Leruth/N. Startin, & S. Usherwood (Eds.), The Routledge Handbook of Euroscepticism (pp. 427–443). London: Routledge. https://doi.org/10.4324/9781315464015-39 Bandeira, G.,/Caballe, J., & Vella, E. (2019). Fiscal Austerity and Migration: A Missing Link. Sheffield Economic Research Paper Series, 2019009. http://eprints.whiterose.ac.uk/ 145829/ Batista, C.,/Seither, J., & Vicente, P. C. (2019). Do migrant social networks shape political attitudes and behavior at home? World Development, 117, 328–343. https://doi.org/10. 1016/j.worlddev.2019.01.019 Baute, S., & de Ruijter, A. (2021). EU health solidarity in times of crisis: explaining public preferences towards EU risk pooling for medicines. Journal of European Public Policy, 1–23. https://doi.org/10.1080/13501763.2021.1936129 Bertaud, N.,/Ernst, T., & Kolanko, K. (2019). European Commission—Press release: Schengen accession: Croatia on the way to join the Schengen Area [Retrieved online 11/21/2021]. https://ec.europa.eu/commission/presscorner/detail/en/ip_19_6140

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2022 L. Möller, Repeated Crisis Exposure, Euroskepticism & Political Behavior, BestMasters, https://doi.org/10.1007/978-3-658-39267-3

65

66

References

Bozorgmehr, K.,/Saint, V.,/Kaasch, A.,/Stuckler, D., & Kentikelenis, A. (2020). COVID and the convergence of three crises in Europe. The Lancet Public Health, 5 (5), e247–e248. https://doi.org/10.1016/s2468-2667(20)30078-5 Buchanan, J. M., & Wagner, R. E. (1977). Democracy in Deficit: The Political Legacy of Lord Keynes New York: Academic Press. Cameron, A. C., & Miller, D. L. (2015). A Practitioner’s Guide to Cluster-Robust Inference. The Journal of Human Resources, 50 (2), 317–372. https://doi.org/10.3368/jhr.50.2.317 Campos, N. F.,/Coricelli, F., & Moretti, L. (2014). Economic Growth and Political Integration: Estimating the Benefits from Membership in the European Union Using the Synthetic Counterfactuals Method. IZA Discussion Papers, 8162. http://hdl.handle.net/10419/98976 Careja, R., & Emmenegger, P. (2012). Making Democratic Citizens: The Effects of Migration Experience on Political Attitudes in Central and Eastern Europe. Comparative Political Studies, 45 (7), 875–902. https://doi.org/10.1177/0010414011428591 Cogley, T., & Sargent, T. J. (2008). The market price of risk and the equity premium: A legacy of the Great Depression? Journal of Monetary Economics, 55 (3), 454–476. https://doi.org/ 10.1016/j.jmoneco.2008.01.006 Crespy, A. (2020). The EU’s Socioeconomic Governance 10 Years after the Crisis: Muddling through and the Revolt against Austerity. JCMS: Journal of Common Market Studies, 58 (S1), 133–146. https://doi.org/10.1111/jcms.13083 Della Polla, G.,/Pelullo, C. P.,/Di Giuseppe, G., & Angelillo, I. F. (2021). Changes in Behaviors and Attitudes in Response to COVID-19 Pandemic and Vaccination in Healthcare Workers and University Students in Italy. Vaccines, 9 (127611), 1–11. https://doi.org/10. 3390/vaccines9111276 Depetris-Chauvin, E.,/Durante, R., & Campante, F. (2020). Building Nations through Shared Experiences: Evidence from African Football. American Economic Review, 110 (5), 1572– 1602. https://doi.org/10.1257/aer.20180805 Down, I., & Wilson, C. J. (2008). From ‘Permissive Consensus’ to ‘Constraining Dissensus’: A Polarizing Union? Acta Politica, 43 (1), 26–49. https://doi.org/10.1057/palgrave.ap. 5500206 Downs, A. (1957). An Economic Theory of Democracy. New York City: Harper Brothers. Duch, R. M., & Stevenson, R. (2010). The Global Economy, Competency, and the Economic Vote. The Journal of Politics, 72 (1), 105–123. https://doi.org/10.1017/s0022381609990508 Dustmann, C.,/Vasiljeva, K., & Damm, A. P. (2018). Refugee Migration and Electoral Outcomes. The Review of Economic Studies, 86 (5), 2035–2091. https://doi.org/10.1093/restud/ rdy047 Ehin, P., & Talving, L. (2020). Second-order effects or ideational rifts? Explaining outcomes of European elections in an era of populist politics. Italian Political Science Review/Rivista Italiana di Scienza Politica, 50 (3), 350–367. https://doi.org/10.1017/ipo.2020.27 Ehin, P., & Talving, L. (2021). Still second-order? European elections in the era of populism, extremism, and euroskepticism. Politics, 41 (4), 467–485. https://doi.org/10.1177/ 0263395720986026 Eicker, F. (1967). Limit Theorems for Regressions with Unequal and Dependent Errors. In L. M. Le Cam & J. Neyman (Eds.), Proceedings of the Fifth Berkeley Symposium on Mathematical Statistics and Probability (pp. 59–82). Berkeley: University of California Press. https://digitalassets.lib.berkeley.edu/math/ucb/text/math_s5_v1_article-06.pdf

References

67

Engler, P., & Klein, M. (2017). Austerity measures amplified crisis in Spain, Portugal, and Italy. DIW Economic Bulletin, 7 (8), 89–93. http://hdl.handle.net/10419/155378 Essers, D. (2013). Developing country vulnerability in light of the global financial crisis: Shock therapy? Review of Development Finance, 3 (2), 61–83. https://doi.org/10.1016/j. rdf.2013.02.001 European Commission. (2012). Eurobarometer 62.0 (Oct-Nov 2004). https://doi.org/10.4232/ 1.10962 European Commission. (2015). A European Agenda on Migration [Retrieved online 11/17/2021]. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX: 52015DC0240 European Commission & European Parliament. (2021). Eurobarometer 94.2 (2020). https:// doi.org/10.4232/1.13722 European Commission Joint Research Center. (2021). COVID-19 data-by-region. https:// github.com/ec-jrc/COVID-19/tree/master/data-by-region European Social Survey. (2021). Round 10 fieldwork update [Retrieved online 11/18/2021]. https://www.europeansocialsurvey.org/about/singlenew.html?a=/about/ news/essnews0115.html Eurostat. (2021a). Population change—Demographic balance and crude rates at regional level (NUTS 3) (demo_r_gind3). https://ec.europa.eu/eurostat/databrowser/view/DEMO_ R_GIND3__custom_1524999/default/table?lang=en Eurostat. (2021b). Population change—Demographic balance and crude rates at regional level (NUTS 3) (demo_r_gind3) Reference Metadata in Euro SDMX Metadata Structure (ESMS) [Retrieved online 11/06/2021]. https://ec.europa.eu/eurostat/cache/metadata/en/ demo_r_gind3_esms.htm Flinders, M. (2021). Democracy and the politics of coronavirus: Trust, blame and understanding. Parliamentary Affairs, 74 (2), 483–502. https://doi.org/10.1093/pa/gsaa013 Franklin, M. N. (2001). How Structural Factors Cause Turnout Variations at European Parliament Elections. European Union Politics, 2 (3), 309–328. https://doi.org/10.1177/ 1465116501002003003 Frontex. (2016). European Border and Coast Guard Agency launches today [Retrieved online 11/17/2021]. https://frontex.europa.eu/media-centre/news/news-release/europeanborder-and-coast-guard-agency-launchestoday-CHIYAp Funke, M.,/Schularick, M., & Trebesch, C. (2016). Going to extremes: Politics after financial crises, 1870–2014. European Economic Review, 88, 227–260. https://doi.org/10.1016/j. euroecorev.2016.03.006 Gans, H. J. (1992). Second-generation decline: Scenarios for the economic and ethnic futures of the post-1965 American immigrants. Ethnic and Racial Studies, 15 (2), 173–192. https:// doi.org/10.1080/01419870.1992.9993740 Gerbaudo, P. (2017). The indignant citizen: anti-austerity movements in southern Europe and the anti-oligarchic reclaiming of citizenship. Social Movement Studies, 16 (1), 36–50. https://doi.org/10.1080/14742837.2016.1194749 Gerling, L., & Kellermann, K. L. (2021). Contagious populists: The impact of election information shocks on populist party preferences in Germany. European Journal of Political Economy, (102098), 1–34. https://doi.org/10.1016/j.ejpoleco.2021.102098

68

References

GESIS. (2021a). Sampling and Fieldwork: The sampling [Retrieved online 11/21/2021]. https://www.gesis.org/en/eurobarometer-data-service/survey-series/standard-special-eb/ sampling-and-fieldwork GESIS. (2021b). Standard and Special Topic Eurobarometer: The longest running regular cross-national and cross-temporal opinion poll program [Retrieved online 11/21/2021]. https://www.gesis.org/en/eurobarometer-data-service/survey-series/standard-special-eb Giani, M., & Méon, P.-G. (2021). Global Racist Contagion Following Donald Trump’s Election. British Journal of Political Science, 51 (3), 1332–1339. https://doi.org/10.1017/ s0007123419000449 Giugni, M., & Grasso, M. T. (2019). Blame and contention: How perceptions of the government’s role in the economic crisis shape patterns of political action. Acta Politica, 54 (3), 346–370. https://doi.org/10.1057/s41269-017-0073-x Giuliano, P., & Spilimbergo, A. (2014). Growing up in a Recession. Review of Economic Studies, 81 (2), 787–817. https://doi.org/10.1093/restud/rdt040 Göncz, B., & Lengyel, G. (2021). Europhile Public vs Eurosceptic Governing Elite in Hungary? Intereconomics, 56 (2), 86–90. https://doi.org/10.1007/s10272-021-0959-8 Grande, E., & Hutter, S. (2016). Introduction: European integration and the challenge of politicisation. In S. Hutter/E. Grande, & H. Kriesi (Eds.), Politicising Europe: Integration and Mass Politics (pp. 3–31). Cambridge: Cambridge University Press. https://doi.org/10. 1017/cbo9781316422991.002 Groenewold, G. (2008). Identities and intercultural relations. In M. Crul & L. Heering (Eds.), The Position of the Turkish and Moroccan Second Generation in Amsterdam and Rotterdam: The TIES study in the Netherlands (pp. 105–128). Amsterdam: Amsterdam University Press. Guerra, S. (2013). Does Familiarity Breed Contempt? Determinants of Public Support for European Integration and Opposition to it before and after Accession. JCMS: Journal of Common Market Studies, 51 (1), 38–50. https://doi.org/10.1111/j.1468-5965.2012. 02300.x Guiso, L.,/Herrera, H.,/Morelli, M., & Sonno, T. (2019). Global crises and populism: the role of Eurozone institutions. Economic Policy, 34 (97), 95–139. https://doi.org/10.1093/epolic/ eiy018 Guriev, S., & Papaioannou, E. (2020). The Political Economy of Populism. Journal of Economic Literature, (forthcoming). https://doi.org/10.2139/ssrn.3542052 Hainmueller, J. (2012). Entropy Balancing for Causal Effects: A Multivariate Reweighting Method to Produce Balanced Samples in Observational Studies. Political Analysis, 20 (1), 25–46. https://doi.org/10.1093/pan/mpr025 Hale, G., & Obstfeld, M. (2016). The Euro and the Geography of International Debt Flows. Journal of the European Economic Association, 14 (1), 115–144. https://doi.org/10.1111/ jeea.12160 Harmsen, R. (2010). Concluding Comment: On Understanding the Relationship between Populism and Euroscepticism. Perspectives on European Politics and Society, 11 (3), 333– 341. https://doi.org/10.1080/15705854.2010.503036 Harteveld, E.,/Schaper, J.,/Lange, S. L. D., & Brug, W. V. D. (2018). Blaming brussels? the impact of (news about) the refugee crisis on attitudes towards the EU and national politics. JCMS: Journal of Common Market Studies, 56 (1), 157–177. https://doi.org/10.1111/jcms. 12664

References

69

Heider, B.,/Stroms, P.,/Koch, J., & Siedentop, S. (2020). Where do immigrants move in Germany? The role of international migration in regional disparities in population development. Population, Space and Place 26 (8). https://doi.org/10.1002/psp.2363 Heinkelmann-Wild, T.,/Kriegmair, L., & Rittberger, B. (2020). The EU multilevel system and the europeanization of domestic blame games. Politics and Governance, 8 (1), 85–94. https://doi.org/10.17645/pag.v8i1.2522 Heinkelmann-Wild, T., & Zangl, B. (2019). Multilevel blame games: Blameshifting in the european union. Governance, 33 (4), 953–969. https://doi.org/10.1111/gove.12459 Heinkelmann-Wild, T.,/Zangl, B.,/Rittberger, B., & Kriegmair, L. (2021). Blame shifting and blame obfuscation: The blame avoidance effects of delegation in the european union. European Journal of Political Research. https://doi.org/10.1111/1475-6765.12503 Hellwig, T. (2008). Globalization, Policy Constraints, and Vote Choice. The Journal of Politics, 70 (4), 1128–1141. https://doi.org/10.1017/s0022381608081103 Hernández, E., & Kriesi, H. (2016). Turning your back on the EU. The role of Eurosceptic parties in the 2014 European Parliament elections. Electoral Studies, 44, 515–524. https:// doi.org/10.1016/j.electstud.2016.04.013 Hobolt, S. B., & Tilley, J. (2014). Blaming Europe?: Responsibility Without Accountability in the European Union. Oxford: Oxford University Press. https://www.ebook.de/de/ product/21513167/sara_b_hobolt_james_tilley_blaming_europe_responsibility_without_ accountability_in_the_european_union.html Hobolt, S. B., & Tilley, J. (2016). Fleeing the centre: The rise of challenger parties in the aftermath of the euro crisis. West European Politics, 39 (5), 971–991. https://doi.org/10. 1080/01402382.2016.1181871 Hooghe, L., & Marks, G. (2007). Sources of Euroscepticism. Acta Politica, 42 (2–3), 119–127. https://doi.org/10.1057/palgrave.ap.5500192 Hooghe, L., & Marks, G. (2009). A postfunctionalist theory of european integration: From permissive consensus to constraining dissensus. British Journal of Political Science, 39 (1), 1–23. https://doi.org/10.1017/s0007123408000409 Höpner, M., & Jurczyk, B. (2012). Kritik des Eurobarometers: Über die Verwischung der Grenze zwischen seriöser Demoskopie und interessengeleiteter Propaganda. Leviathan, 40 (3), 326–349. https://doi.org/10.5771/0340-0425-2012-3-326 House, C. L.,/Proebsting, C., & Tesar, L. L. (2020). Austerity in the aftermath of the great recession. Journal of Monetary Economics, 115, 37–63. https://doi.org/10.1016/j.jmoneco. 2019.05.004 Huber, P. J. (1967). The behavior of maximum likelihood estimates under nonstandard conditions. In L. M. Le Cam & J. Neyman (Eds.), Proceedings of the Fifth Berkeley Symposium on Mathematical Statistics and Probability (pp. 221–233). Berkeley: University of California Press. https://projecteuclid.org/ebooks/berkeleysymposium-on-mathematical-statistics-and-probability/Proceedings%20of%20the %20Fifth%20Berkeley%20Symposium%20on%20Mathematical%20Statistics%20and %20Probability,%20Volume%201:%20Statistics/chapter/The%20behavior%20of %20maximum%20likelihood%20estimates%20under%20nonstandard%20conditions/ bsmsp/1200512988 Hutter, S., & Kriesi, H. (2021). Politicising immigration in times of crisis. Journal of Ethnic and Migration Studies, 1–25. https://doi.org/10.1080/1369183x.2020.1853902

70

References

Inglehart, R. (1970). Public Opinion and Regional Integration. International Organization, 24 (4), 764–795. https://www.jstor.org/stable/2706152 Jolly, S.,/Bakker, R.,/Hooghe, L.,/Marks, G.,/Polk, J.,/Rovny, J.,/Steenbergen, M., & Vachudova, M. A. (2022). Chapel Hill Expert Survey trend file, 1999–2019. Electoral Studies, 75 (102420, forthcoming). https://doi.org/10.1016/j.electstud.2021.102420 Jordà, Ò.,/Schularick, M., & Taylor, A. M. (2015). Sovereigns Versus Banks: Credit, Crises, and Consequences. Journal of the European Economic Association, 14 (1), 45–79. https:// doi.org/10.1111/jeea.12144 Karanikolos, M.,/Mladovsky, P.,/Cylus, J.,/Thomson, S.,/Basu, S.,/Stuckler, D.,/Mackenbach, J. P., & McKee, M. (2013). Financial crisis, austerity, and health in Europe. The Lancet, 381 (9874), 1323–1331. https://doi.org/10.1016/s0140-6736(13)60102-6 Karyotis, G., & Rüdig, W. (2018). The Three Waves of Anti-Austerity Protest in Greece, 2010–2015., 16 (2), 158–169. https://doi.org/10.1177/1478929916685728 King, G. (1990). Stochastic Variation: A Comment on Lewis-Beck and Skalaban’s “The RSquared”. Political Analysis, 2, 185–200. https://www.jstor.org/stable/23317771 Kopri´c, I.,/Škarica, M., & Toma´n, R. M. (2019). Local Public Services and Austerity Measures in Croatia: Adaptations of the Resilient Social Model Through the Central-Local Power and Blame Games. In A. Lippi & T. N. Tsekos (Eds.), Local Public Services in Times of Austerity across Mediterranean Europe (pp. 167–191). Cham: Palgrave Macmillan. https:// doi.org/10.1007/978-3-319-76225-8_8 Koutsampelas, C.,/Andreou, S. N.,/Papaloi, E., & Dimopoulos, K. (2021). Austerity, economic crisis and children. In S. M. Hall/H. Pimlott-Wilson, & J. Horton (Eds.), Austerity Across Europe: Lived Experiences of Economic Crises (pp. 96–109). Abingdon: Routledge. https:// doi.org/10.4324/9780429201332 Krumpal, I. (2013). Determinants of social desirability bias in sensitive surveys: a literature review. Quality & Quantity, 47 (4), 2025–2047. https://doi.org/10.1007/s11135011-9640-9 Leconte, C. (2010). Understanding Euroskepticism. Houndmills: Palgrave Macmillan. Lewis-Beck, M. S., & Lobo, M. C. (2017). The Economic Vote: Ordinary vs. Extraordinary Times. In K. Arzheimer/J. Evans, & M. S. Lewis-Beck (Eds.), The SAGE Handbook of Electoral Behaviour (pp. 606–630). London: SAGE Publications. https://doi.org/10.4135/ 9781473957978.n26 Lindberg, L. N., & Scheingold, S. A. (1970). Europe’s Would-Be polity: Patterns in the European Community. Englewood Cliffs: Prentice-Hall. Loveless, M., & Rohrschneider, R. (2011). Public perceptions of the EU as a system of governance. Living Reviews in European Governance, 6 (2), 5–38. http://www.livingreviews. org/lreg%202011-%202 MacKinnon, J. G., & Webb, M. D. (2016). Wild Bootstrap Inference for Wildly Different Cluster Sizes. Journal of Applied Econometrics, 32 (2), 233–254. https://doi.org/10.1002/ jae.2508 Magalhães, P. C. (2014). Introduction—Financial Crisis, Austerity, and Electoral Politics. Journal of Elections, Public Opinion and Parties, 24 (2), 125–133. https://doi.org/10.1080/ 17457289.2014.887090 Makarychev, A., & Crothers, L. (2020). Assessing Populism at Europe’s Margins: Pervasive, Performative, Persistent. Populism, 3 (2), 145–147. https://doi.org/10.1163/2588807202021044

References

71

Malmendier, U., & Nagel, S. (2011). Depression Babies: Do Macroeconomic Experiences Affect Risk Taking? The Quarterly Journal of Economics, 126 (1), 373–416. https://doi. org/10.1093/qje/qjq004 Maxwell, R. (2010). Evaluating Migrant Integration: Political Attitudes across Generations in Europe. International Migration Review, 44 (1), 25–52. https://doi.org/10.1111/j.17477379.2009.00797.x Mény, Y. (2020). Foreword: Against Europe: The Changing Patterns of Euroscepticism Over Time. In M. Baldassari/E. Castelli/M. Truffelli, & G. Vezzani (Eds.), Anti-Europeanism: Critical Perspectives Towards the European Union (pp. V–IX). Cham: Springer International Publishing. https://doi.org/10.1007/978-3-030-24428-6_6 Michael, M., & Christofides, L. N. (2020). The impact of austerity measures on the public— private sector wage gap in Europe. Labour Economics, 63 (101796), 1–20. https://doi.org/ 10.1016/j.labeco.2020.101796 Mikulaschek, C.,/Pant, S., & Tesfaye, B. (2020). Winning Hearts and Minds in Civil Wars: Governance, Leadership Change, and Support for Violent Groups in Iraq. American Journal of Political Science, 64 (4), 773–790. https://doi.org/10.1111/ajps.12527 Mishra, D. P. (2021). COVID-19 Disruption and Service Firms’ Adaptation Strategies: Institutional Theory Perspectives. In J. Lee & S. H. Han (Eds.), The Future of Service PostCOVID-19 Pandemic, Volume 2, The ICT and Evolution of Work (pp. 175–203). Singapore: Springer Nature. https://doi.org/10.1007/978-981-33-4134-0_8 Mota, P. R.,/Fernandes, A. L. C., & Vasconcelos, P. B. (2020). Employment hysteresis: An argument for avoiding front-loaded fiscal consolidations in the eurozone. Review of Keynesian Economics, 8 (4), 536–559. https://doi.org/10.4337/roke.2020.04.05 Moulton, B. R. (1986). Random group effects and the precision of regression estimates. Journal of Econometrics, 32 (3), 385–397. https://doi.org/10.1016/0304-4076(86)900217 Mudde, C. (2012). The Comparative Study of Party-Based Euroscepticism: The Sussex versus the North Carolina School. East European Politics, 28 (2), 193–202. https://doi.org/10. 1080/21599165.2012.669735 Natali, D., & Stamati, F. (2014). Reassessing South European Pensions after the Crisis: Evidence from Two Decades of Reforms. South European Society and Politics, 19 (3), 309–330. https://doi.org/10.1080/13608746.2014.951515 OECD. (2009). Government at a Glance 2009. Paris: Organization for Economic Co-operation and Development. https://doi.org/10.1787/9789264075061-en OECD. (2021). Government at a Glance 2021. Paris: Organization for Economic Co-operation and Development. https://doi.org/10.1787/1c258f55-en Otto, A. H., & Steinhardt, M. F. (2017). The Relationship between Immigration and the Success of Far-right Political Parties in Germany. ifo DICE Report, 15 (4), 20–23. http:// hdl.handle.net/10419/181256 Ozlem Ultan, M., & Ornek, S. (2015). Euroscepticism in the European Union. International Journal of Social Sciences, 4 (2), 49–57. https://www.eurrec.org/ijoss-article-156? download=6 Paul, R., & Roos, C. (2019). Towards a new ontology of crisis? Resilience in EU migration governance. European Security, 28 (4), 393–412. https://doi.org/10.1080/09662839.2019. 1637340

72

References

Petkov, V. S. (2014). Advantages and Disadvantages of Fiscal Discipline in Bulgaria in Times of Crisis. Contemporary Economics, 8 (1), 47–56. https://doi.org/10.5709/ce.1897-9254. 130 Pieters, D. (2021). Cyprus. In D. Pieters (Ed.), The World of Small States: Social Security Law in Small Jurisdictions (pp. 85–94). Cham: Springer International Publishing. https:// doi.org/10.1007/978-3-030-78247-4_6 Pirro, A. L. P.,/Taggart, P., & van Kessel, S. (2018). The populist politics of Euroscepticism in times of crisis: Comparative conclusions. Politics, 38 (3), 378–390. https://doi.org/10. 1177/0263395718784704 Ponticelli, J., & Voth, H.-J. (2020). Austerity and anarchy: Budget cuts and social unrest in Europe, 1919–2008. Journal of Comparative Economics, 48 (1), 1–19. https://doi.org/10. 1016/j.jce.2019.09.007 Publications Office of the EU. (2021). Interinstitutional style guide. Luxembourg: Publications Office of the European Union. https://doi.org/10.2830/400520 Reeves, A.,/McKee, M.,/Basu, S., & Stuckler, D. (2014). The political economy of austerity and healthcare: Cross-national analysis of expenditure changes in 27 European nations 1995–2011. Health Policy, 115 (1), 1–8. https://doi.org/10.1016/j.healthpol.2013.11.008 Reinl, A.-K. (2020). Euroscepticism in Times of European Crises: The Role of Solidarity. In M. Baldassari/E. Castelli/M. Truffelli, & G. Vezzani (Eds.), Anti-Europeanism: Critical Perspectives Towards the European Union (pp. 95–113). Cham: Springer International Publishing. https://doi.org/10.1007/978-3-030-24428-6_6 Riker, W. H., & Ordeshook, P. C. (1968). A Theory of the Calculus of Voting. American Political Science Review, 62 (1), 25–42. https://doi.org/10.2307/1953324 Rodríguez-Pose, A., & Dijkstra, L. (2021). Does Cohesion Policy reduce EU discontent and Euroscepticism? Regional Studies, 55 (2), 354–369. https://doi.org/10.1080/00343404. 2020.1826040 Rodrik, D. (2018). Populism and the economics of globalization. Journal of International Business Policy, 1 (1–2), 12–33. https://doi.org/10.1057/s42214-018-0001-4 Roodman, D.,/Ørregaard Nielsen, M.,/MacKinnon, J. G., & Webb, M. D. (2019). Fast and wild: Bootstrap inference in Stata using boottest. The Stata Journal, 19 (1), 4–60. https:// doi.org/10.1177/1536867x19830877 Rooduijn, M.,/Van Kessel, S.,/Froio, C.,/Pirro, A.,/De Lange, S.,/Halikiopoulou, D.,/Lewis, P.,/Mudde, C., & Taggart, P. (2019). The PopuList: An Overview of Populist, Far Right, Far Left and Eurosceptic Parties in Europe. www.popu-list.org Ruth-Lovell, S. P.,/Lührmann, A., & Grahn, S. (2019). Democracy and Populism: Testing a Contentious Relationship. V-Dem Institute Working Paper Series, 2019:91. https://doi.org/ 10.2139/ssrn.3471132 Schäfer, C. (2017). Euroskeptizismus und Wahlenthaltung. Zeitschrift für Vergleichende Politikwissenschaft, 11 (1), 50–80. https://doi.org/10.1007/s12286-017-0327-z Schäfer, C., & Debus, M. (2018). No participation without representation: Policy distances and abstention in European Parliament elections. Journal of European Public Policy, 25 (12), 1835–1854. https://doi.org/10.1080/13501763.2017.1363806 Schäfer, C., & Gross, M. (2020). Euroscepticism in Times of Crisis: A Macro-Level Analysis of the Euro Crisis’ Effects on Public Opinion and Party Competition on European Integration. In M. Baldassari/E. Castelli/M. Truffelli, & G. Vezzani (Eds.), Anti-Europeanism:

References

73

Critical Perspectives Towards the European Union (pp. 31–55). Cham: Springer International Publishing. https://doi.org/10.1007/978-3-030-24428-6_3 Schimmelfennig, F. (2020). Politicisation management in the european union. Journal of European Public Policy, 27 (3), 342–361. https://doi.org/10.1080/13501763.2020.1712458 Serricchio, F.,/Tsakatika, M., & Quaglia, L. (2013). Euroscepticism and the Global Financial Crisis. JCMS: Journal of Common Market Studies, 51 (1), 51–64. https://doi.org/10.1111/ j.1468-5965.2012.02299.x Sherpa, D. (2020). Estimating Impact of Austerity policies in COVID-19 fatality rates: Examining the dynamics of Economic policy and Case Fatality Rates (CFR) of COVID-19 in OECD countries. https://doi.org/10.1101/2020.04.03.20047530 Shutes, I., & Ishkanian, A. (2021). Transnational welfare within and beyond the nation-state: civil society responses to the migration crisis in Greece. Journal of Ethnic and Migration Studies, 1–18. https://doi.org/10.1080/1369183x.2021.1892479 StataCorp. (2021). Stata Statistical Software: Release 17. Taggart, P. (1998). A touchstone of dissent: Euroscepticism in contemporary Western European party systems. European Journal of Political Research, 33 (3), 363–388. https://doi. org/10.1111/1475-6765.00387 Taggart, P. (2020). Failing the European Rorschach Test? European Integration and Euroscepticisms. In M. Gilbert & D. Pasquinucci (Eds.), Euroscepticisms: The Historical Roots of a Political Challenge (pp. 222–229). Leiden: Koninklijke Brill. https://doi.org/10.1163/ 9789004421257_013 Taggart, P., & Szczerbiak, A. (2018). Putting Brexit into perspective: the effect of the Eurozone and migration crises and Brexit on Euroscepticism in European states. Journal of European Public Policy, 25 (8), 1194–1214. https://doi.org/10.1080/13501763.2018.1467955 Talving, L. (2017). The electoral consequences of austerity: Economic policy voting in Europe in times of crisis. West European Politics, 40 (3), 560–583. https://doi.org/10.1080/ 01402382.2016.1271600 Talving, L. (2018). Economic voting in Europe: Did the crisis matter? Comparative European Politics, 16 (4), 695–723. https://doi.org/10.1057/s41295-017-0092-z Treib, O. (2021). Euroscepticism is here to stay: what cleavage theory can teach us about the 2019 European Parliament elections. Journal of European Public Policy, 28 (2), 174–189. https://doi.org/10.1080/13501763.2020.1737881 Vanhuysse, P.,/Jankowski, M., & Tepe, M. (2021). Vaccine alliance building blocks: a conjoint experiment on popular support for international COVID-19 cooperation formats. Policy Sciences, 54 (3), 493–506. https://doi.org/10.1007/s11077-021-09435-1 Vasilopoulou, S. (2018). Theory, concepts and research design in the study of Euroscepticism. In B. Leruth/N. Startin, & S. Usherwood (Eds.), The Routledge Handbook of Euroscepticism (pp. 22–35). London: Routledge. https://doi.org/10.4324/9781315464015-4 Verney, S. (2011). Euroscepticism in Southern Europe: A Diachronic Perspective. South European Society and Politics, 16 (1), 1–29. https://doi.org/10.1080/13608746.2010.570124 Wallaschek, S. (2020a). Framing Solidarity in the Euro Crisis: A Comparison of the German and Irish Media Discourse. New Political Economy, 25 (2), 231–247. https://doi.org/10. 1080/13563467.2019.1586864 Wallaschek, S. (2020b). Contested solidarity in the Euro crisis and Europe’s migration crisis: a discourse network analysis. Journal of European Public Policy, 27 (7), 1034–1053. https:// doi.org/10.1080/13501763.2019.1659844

74

References

Warren, G. W., & Lofstedt, R. (2021). COVID-19 vaccine rollout risk communication strategies in Europe: a rapid response. Journal of Risk Research, 24 (3–4), 369–379. https://doi. org/10.1080/13669877.2020.1870533 Webb, M. D. (2013). Reworking Wild Bootstrap Based Inference for Clustered Errors. Queen’s Economics Department Working Paper, (No. 1315), 1–21. https://www.econstor.eu/handle/ 10419/97480 Werts, H.,/Scheepers, P., & Lubbers, M. (2012). Euro-scepticism and radical right-wing voting in Europe, 2002–2008: Social cleavages, socio-political attitudes and contextual characteristics determining voting for the radical right. European Union Politics, 14 (2), 183–205. https://doi.org/10.1177/1465116512469287 White, H. (1980). A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroskedasticity. Econometrica, 48 (4), 817–838. https://doi.org/10.2307/ 1912934