The Economic Determinants of Populism: The Role of Labor Market Protection Erosion
Master Thesis Faculty of Economics and Business
18
thJune 2019
Double Degree Master in Economic Development and Growth (MEDEG)
Master in International Economics and Business, University of Groningen
Master in Economic Development and Growth, University of Lund
Marco Lovato
m.lovato.1@student.rug.nl S3729648
Supervisor Dr. Anna Minasyan, University of Groningen
Abstract
The present study addresses the role of ‘labor market protection erosion’ as an economic determinant of the renewed populist sentiment currently interesting the European continent. The analysis is carried out on 28 European states between 2002 and 2017. The expectation is that a lower expenditure on welfare programmes intended to diminish the risk and cost associated with unemployment is related with a higher support for populist parties. The results, however, are confirmed only relative to the cost channel.
Keywords: populism, labor market, welfare.
Table of contents
Introduction ... 6
1. Defining Populism ... 8
2. The Reasons Beneath Populism’s Success ... 9
3. Data ... 13
3.1 Election data ... 13
3.2 Labor market data ... 14
3.3 Control variables ... 15
3.4. Descriptive statistics ... 16
4. Methodology ... 19
5. Results ... 22
Conclusion ... 29
References ... 31
Appendix ... 34
6
Introduction
In recent years, many Western societies have been beset by a dramatic political polarization.
The rise of right-wing nationalistic political parties has shown the fragility of supranational entities such as the European Union. The current time, however, differs from previous epochs characterized by a mounting political polarization due to a particular feature with which far- right and far-left political parties are labeled nowadays: populism. The concept, while not entirely new, has experienced a massive diffusion in the last decade, finally leaving the academic world to become a commonly used expression. While its exact definition is still heavily debated among scholars, the word has been widely used to define the rhetoric employed by, for example, Donald Trump. Many political analysts considered it the determining factor allowing his victory in the US 2016 presidential elections. In Europe, on the other hand, populist parties have also begun to flourish, proposing themselves in a clear discontinuity with the traditional parties, labeled as the ‘establishment’ and considered responsible for the economic uncertainty many national states have been experiencing in past years.
The populist rhetoric distinguishes itself as a ‘thin centered ideology’ which contrasts the people, for whom the populist leader claims to speak, to the elite, who is deemed responsible for the current issues the country is facing (Mudde, 2004). The vagueness of concepts such as people and elite, together with the lack of distinguished ideological features, allows populist movements to flourish on both sides of the political spectrum. As a matter of fact, in the United States, if president Trump is commonly associated with far-right populism, Sanders is considered as the leader of a far-left populist movement. According to Rodrik (2018), populism turns to the left or the right of the political spectrum depending on the situations faced by the countries in which it originates. Due to the massive migration flows towards the Old Continent after the outbreak of the Arab Spring in 2011, most of the European populist parties embraced far-right ideologies, e.g. Front National in France, Lega in Italy, PVV in the Netherlands or AfD in Germany.
Nevertheless, while certainly being the minority, there are also European populist parties that embrace far-left ideologies (e.g. Syriza in Greece, Podemos in Spain, SP in the Netherlands) or maintain moderate positions which do not classify as far right or left.
If the recent surge in populist sentiment is to be interpreted as a general discontent towards the
‘establishment’ and a desire to change, it becomes important to understand what are the main determinants of such phenomenon. By doing so, it would be possible to get to the root of the success of these often divisive political parties and preserve social cohesion. The academics who addressed the factors underlying the recent populist wave have traditionally focused on the import penetration produced by globalization (Autor, 2016), cultural features
7
(Inglehart and Norris, 2016) or the 2008 Great Recession (Algan et al., 2017; Dustmann et al., 2017). Moreover, given their prominence, the literature targeting the European diffusion of populist parties has typically narrowed down the analysis to far right populist parties. On this matter, the present study takes a different approach by including all populist parties, regardless of their ideology.
Most importantly, previous literature tended to overlook the role covered by the labor market among the economic determinants of populist support. What Rodrik (2018) defines “labor market protection erosion”, i.e. the increasing labor flexibility and the lower rights guaranteed to workers which point to a gradual dismantling of the welfare states, is believed to be one of the reasons for populism’s recent success. Nevertheless, there has been a systematic lack of empirical research focusing on this particular economic determinant. By using two variables to represent mechanisms which proxy for labor market protection, this study attempts to address the following research question:
Does labor market protection erosion increase the support to populist parties across Europe?
By trying to answer this question, the present paper aims to target the progressive erosion of welfare programs dedicated to protect the labor market and the role such erosion plays in fostering populist support. If a negative and significant relationship were to be found between those mechanisms and populist support, it would confirm that the progressive dismantling of those programs fosters populist emergence. By contrast, it would also imply that a greater investment on those mechanisms might help lowering populist support, thus, presenting consistent policy implications.
The study is structured as follows. Section 1 provides a brief discussion over the much debated concept of populism with links to previous literature and the definition employed throughout the paper. Section 2 reviews previous research addressing the economic determinants of populism and formulates the hypotheses the study attempts to test. Section 3 describes the process of data collection and the summary statistics of the variables included in the model;
Section 4 defines the regression equation, providing a critical comparison of the different models employed. Finally, Section 5 presents and discusses the results. After Section 5, which includes both results and robustness checks, conclusions are drawn.
8
1. Defining Populism
“To each his own definition of populism, according to the academic axe he grinds”
Peter Wiles (1969: 166)
In the last decade, the word populism has pervaded the political talks, media, and as a consequence, has finally become widespread among the population. The great diffusion it experienced, however, does not represent the emergence of a new phenomenon, but rather the comeback of a notion that was previously constrained to the academic world.
As a matter of fact, the history of the concept can be traced back as early as the 19th century with the US People’s Party formation, considered to be one of the ‘defining populist movements’
(p.548, Mudde, 2004). If on the one hand, Latin America is characterized by a longstanding presence of left-wing populism, Europe has seen its reemergence begin only around the 1990s (Funke et al., 2015). Such event was fostered, according to some, by the crisis of the “Washington Consensus” (Aytac and Onis, 2014) and a mounting mistrust towards trade and FDI liberalization.
While populism is certainly not a new phenomenon, its interpretation is still highly controversial in the academic world. Scholars have attempted to find a common definition for decades, but the divergences led to the inevitable emergence of different schools of thought (for an extensive literature review see Gidron and Bonikowski, 2013). In this study, I will build upon the definition offered by Cas Mudde, considered as one of the most prominent voices on the matter. The scholar defines populism as “a ‘thin centered ideology’ that considers society to be ultimately separated into two homogenous and antagonist groups, ‘the pure people’ versus ‘the corrupt elite’ and which argues that politics should be an expression of the volonté générale (general will) of the people.” (Mudde, 2004, p.543). Mudde makes use of the expression ‘thin centered ideology’, first coined by Freeden (1996), to underline the ease with which different ideologies can be attached to populist movements, resulting in far-left or far-right populism. As a consequence, populism is not recognized by the economic policies it supports, but rather by the crucial contrast between ‘the pure people’ against ‘the corrupt elite’.
Ultimately, the populist discourse aims at the ‘gut feelings’ of the people, promising popular policies in exchange for the voters’ support in upcoming elections (Mudde, 2004). Such a feature, however, is argued by many to be the basis of political expression (Kazin, 1995), rather than a prerogative of populist parties alone. When one reckons that this characteristic is inherent in every political party, to a greater or lesser extent, “our assessments [shift] from binary opposition –a party is populist or not– to a matter of degree –a party has more populist characteristics or fewer.” (Deegan-Krause and Haughton, 2009, p.822).
9
The definition of populism used in the present paper will thus unify the two positions discussed, describing populism as a ‘thin centered ideology’, inherent, to a greater or lesser extent, to all political parties, that suggests society to be ultimately separated into two homogenous and antagonist groups, ‘the pure people’ versus ‘the corrupt elite’ and which argues that politics should be an expression of the volonté générale (general will) of the people.
2. The Reasons Behind Populism’s Success
The bulk of the literature investigating the factors underlying populism’s comeback has produced two well-distinguished branches focusing on the demand side, each posing a greater emphasis on different determinants: the cultural backlash and the economic insecurity theses.
The former argues how “populist support can be explained primarily as a social psychological phenomenon” (Inglehart and Norris, 2016, p.13) opposing progressive values such as the redefinition of gender roles and the greater importance given to social tolerance and minorities’
rights. The cultural change these progressive values caused was perceived positively in times of economic prosperity, but has then ignited a powerful counter-reaction with the 2008 economic crisis. Nonetheless, such thesis, while certainly presenting a credible explanation for the increasing support received by far-right nationalistic parties, is inadequate when addressing the rise of left-wing populism (which is typically characterized by the defense of progressive values).
The economic insecurity perspective, on the other hand, emphasizes the profound changes that occurred in post-industrial economies and the impact such great transformation had on voting behavior. The disruptive phenomena that caused the crisis of the two-party system and a rising inequality are argued to be “the rise of knowledge economy, technological automation, and the collapse of manufacturing industry, global flows of labor, goods, peoples, and capital, the erosion of organized labor, shrinking welfare safety-nets, and neo-liberal austerity policies.”
(Inglehart and Norris, 2016, p.2).
A number of scholars have addressed the role of globalization (relative to both goods and people) with regard to the US (Autor, 2016, 2017), Europe (Malgouyres, 2014; Dippel et al., 2015) and the Brexit referendum in the UK (Colantone and Stanig, 2016a). The results typically show a positive relationship between far-right populist support, import penetration and immigration flows. On a similar matter, a research by Swank and Betz (2003) tests the relationship between globalization and far-right populist parties, finding a positive relationship between the two variables. In the present study, globalization is defined according to the book from Baldwin (2016). The scholar defines globalization as the progressive reversal of bundling between
10
production and consumption. Such reversal has been made possible by what Baldwin defines as two major unbundlings caused by the dramatic decline of shipping and communication costs, respectively.
When introducing a variable characterizing the generosity of a country’s welfare state, the findings, however, change: the relationship remains positive only for countries characterized by a liberal welfare structure, whereas it turns insignificant in countries with a more comprehensive welfare coverage. The results from the scholars suggest how globalization is not the driver of populist support per se, but rather the lack of a welfare state employed to correct the inevitable redistributive effects (Rodrik, 2018) that greater economic integration generates. While offering a comprehensive analysis on the economic determinants behind populism reemergence, the studies discussed have lacked to address the European left-wing populist movements growing support, focusing exclusively on right wing populist parties.
Another stream of literature has stressed the role played by the 2008 Great Recession. Stemming from the analysis by Funke et al. (2015) which shows how far-right parties are historically the biggest beneficiaries of systemic crises, Dustmann et al. (2017) address the economic determinants driving the current mistrust in European institutions, analyzing the results in the European Parliament’s elections. They find unemployment to be positively associated with mistrust in national and European institutions, which then leads to increasing support for populist parties. On the same matter, making use of regional data on national elections, Algan et al. (2017) implement a difference-in-difference analysis, showing how support for populist parties is positively associated with the change in the unemployment rate before and in the aftermath of the 2008 financial crisis. Other studies (Swank and Betz, 2003), however, when testing the effect of the rise of unemployment on right-wing populist support, do not find any significant relationship. The present study will test the link between unemployment and support for all populist parties (left- and right-wing), broadening the focus of previous research. The first hypothesis will therefore be:
H1: A higher unemployment level is positively associated with populist support.
While the studies discussed help to understand the role that the 2008 crisis played in increasing populist support through unemployment level or change, the latter is solely used as a proxy for the 2008 Great Recession and not to specifically represent the labor market.
Previous research focusing on the link between populist success and the labor market has typically addressed Latin America, due to the abrupt trade openness that affected the continent.
The consequent pervasive presence of MNEs in the strategic sectors of the national economies has led to a surge in the support for left-wing populism, which advocates for a greater labor market protection (Rodrik, 2018). With regard to Western societies, however, ‘labor market
11
protection erosion’ has not received the same attention. As discussed above, the prominence of right-wing populism in Europe has led most researchers to focus on immigration, globalization or euro-skepticism, i.e. the core points of right-wing populist narrative. The present paper aims to shed light on the role played by the dismantling of labor market protection with regard to populist support across European countries.
Previous research (Carr and Chung, 2014) has provided evidence of a negative relationship between employment insecurity and life satisfaction, while finding at the same time how such relationship is weaker in countries that employ more generous Labor Market Policies (LMPs), i.e.
policies that mitigate the economic insecurity associated with unemployment. LMPs are defined as either active or passive. The former refer to “training, job rotation schemes, employment incentives, supported employment and rehabilitation, job search, direct job creation and start- up incentives”, i.e. those interventions that are aimed at increasing the beneficiaries’
employability (p.386, Carr and Chung, 2014). The latter, instead, refer to the workers who are already unemployed and consist primarily in “unemployment benefits, redundancy and bankruptcy compensation or programmes for early retirement”.
The relationship between Labor Market Policies and populist support has been then explored by Halikiopoulou and Vlandas (2016). The scholars contributed to the existing literature by addressing the relationship between labor market protection mechanisms’ effect and the support for far-right populist parties in the European Parliament elections. As done by Carr and Chung (2014), the authors distinguish two main channels through which unemployment leads to economic insecurity: risk and cost. The former refers to the risk for a person already employed to lose the job, whereas the latter is relative to the cost of being unemployed, i.e. the drop in the disposable income following the loss of a job. If Carr and Chung (2014) referred to such channels through the use of active and passive LMPs respectively, the scholars adopt two different variables. With regard to the risk, they include the Employment Protection Legislation index (EPL) developed by the OECD, which measures the degree of protection of currently employed people. Relatively to the cost, they adopt the unemployment benefit replacement rate, which captures ‘the size of the income loss upon becoming unemployed’ (p.645, Halikiopoulou and Vlandas, 2016).
While the authors find that in countries where the two variables present high values unemployment has no effect on the support for far-right populist parties, with regard to the EPL index this occurs only when the share of foreign-born population in the country is taken into account. As argued by the scholars themselves, the variable has some key limitations: the index might also act as a deterrent for firms to hire new workers, given the high market rigidity which would then not allow to implement layoffs with the same ease.
12
Given the unclear direction of the effect of the EPL index, with regard to the risk channel I will employ the active LMPs. Regarding the cost channel, I will adopt the net replacement rates. The two newly added explanatory variables will be the basis for my second and third hypothesis.
H2: A higher risk of unemployment is positively associated with populist support.
H3: A higher cost of unemployment is positively associated with populist support.
The economic reasoning behind these hypotheses is that if the state’s commitment to provide a comprehensive national welfare fails, and, as a consequence, labor market protection becomes less extensive, populist support will experience a surge in response, due to the discontent caused by the higher economic uncertainty. Another way to interpret the hypotheses reported above is that countries characterized by a higher level of LMP measures’ expenditure and net replacement rates will be relatively less exposed to the increasing populist support witnessed across Western societies. If the country is committed to increase the employability of current employees through constant trainings while also providing the unemployed with economic benefits until they are reabsorbed in the labor market, then economic uncertainty would be mitigated, and with it, populist support should come out weakened.
The main contribution of the current study is to address the relationship between ‘labor market protection erosion’ and populism. As a consequence, the paper also attempts to target the role played by the unemployment welfare programs in mitigating populist support. Hence, the research not only contributes to fill a consistent gap in the literature over populism’s recent reemergence, but investigates the possible policies a country might implement in order to contain the populist wave. Such approach can be seen as a contribution to the study by Carr and Chung (2014), which tested the relationship between economic insecurity and labor market policies. Moreover, the use of active LMPs against the EPL index and the decision to rely on national elections (rather than on European Parliament elections) provides a positive complementarity to the research by Halikiopoulou and Vlandas (2016).
Finally, the study addresses all European populist parties across the political spectrum, whereas previous literature (Colantone and Stanig, 2016a; Algan et al., 2017; Dustmann et al. 2017; Guiso et al., 2017) has typically targeted far-right populism alone. As argued before, the reason behind such decision stems from the expectation that, beneath the different narrative different populisms adopt, they all leverage on widespread discontent originating from economic insecurity.
13
3. Data
The present study addresses the relationship between economic variables related to the labor market and the support for populist parties across a group of European countries in the parliamentary elections occurred between 2002 and 2017. It is important to stress how not all of the countries in the sample belong to the European Union, nor the countries belonging to the European Union were part of it for the entire time period. The sample consists of 144 national parliamentary elections held in 28 European countries. However, due to missing values for LMPs expenditure and net replacement rate (the variables representing the risk and cost channel, respectively) the observations had to be narrowed down to 102. I will now proceed to explain the construction of my dependent and independent variables.
/ 3.1 Election data
The national election data used in order to construct my dependent variable is retrieved from the ParlGov database, which provides information for a number of countries over parliamentary and European parliament elections. The current study adopts only the results from national parliamentary elections and thus, presidential and European parliament elections are excluded from the sample. In order to define a party as populist, I relied on PopuList, a classification of populist, far right, far left and/or Eurosceptic European parties that obtained at least 2 percent of the vote in at least one national parliamentary election since 1998.
The list, result of a close cooperation between journalists and academics, employs the definition formulated by Mudde (2004). The definition adopted in the current study also follows the one by Mudde; nonetheless adding how populism is, to a greater or lesser extent, an inherent feature of all political parties. Such extension does not lead to any practical difference, but rather recognizes that a ‘populist’ feature is a matter of degree, rather than a mere binary classification.
This allows my definition to be consistent with the classification employed by PopuList. The classification employed by such list, however, might present some caveats. For instance, in Italy the political party by Berlusconi –Forza Italia (previously PdL)– is considered populist, whereas in recent times it has been seen as a moderate force mitigating the increasing support experienced by populist parties such as Lega Nord and Movimento 5 Stelle. It is therefore important to understand that not all the parties listed as populists in the current study can be ascribed to the far-left or far-right ideologies. The presence of populist parties advocating moderate economic propositions might affect the estimates, since they received high shares of votes also in times of relatively positive economic performance.
14
As stated previously, the classification employed by PopuList defines parties according to four main dimensions: populist, far-left, far-right, and Eurosceptic. Since in the current study I am interested in the share of populist votes overall, I include all parties classified as populist, irrespective of their ideological position. Hence, the votes’ share received by different populist parties in the same country for a certain elections are added up to one value. Looking at the case of Greece, for example, Syriza’s votes’ share, a far-left populist party, will be added to the one for LAOS, a far-right one. As stressed before, such decision stems from the belief that both forms of populisms originate from the widespread economic insecurity, which is then blamed on globalization or economic inequality, depending on the narrative adopted by the political party (Rodrik, 2018). When data is missing for a certain election, I complement the dataset with data from national databases.
In one country, Estonia, the populist party Res Publica Party (ERP) has merged with the party Pro Patria Union (I) in 2006, becoming Union of Pro Patria and Res Publica (IRL). While the classification in use does not list IRL among the populist parties, I include it as such in my analysis.
In the case in which parties listed in the PopuList are not retraceable in the election database, after controlling through the use of external sources whether they are present with a different acronym, I decide to exclude them from the dataset. Finally, if more than one election was held in a certain year, I include only the first one. Typically, another election is held shortly after the first only in cases in which there is a profound political instability which does not allow for the formation of a government. As a consequence, the second election will reflect less accurately the voters’ preferences, since the population will tend to shift from smaller to bigger parties to guarantee political stability. The list of populist parties divided by country and political orientation (far right, far left, none) can be found in the appendix Table I.
/ 3.2 Labor market data
The data on unemployment rate is retrieved from Eurostat, the statistical office of the European Union. The data refers to the unemployment rate over the total active population, i.e. the fraction of the population that is either employed or actively seeking employment. Since voting preferences are based on the previous economic performance of a country, the unemployment rate used refers to the year before the election. Thus, the unemployment rate in use will refer to the year t-1, with t being the year of the election. The choice to include lagged versions of the variables, seeming appropriate from a theoretical perspective, allows also to account for possible endogeneity problems (e.g. reverse causality) (Bellemare et al., 2017).
15
With regard to the labor market protection, I will rely on two different variables capturing the risk and the cost of the unemployment: LMP measures’ expenditure and net replacement rates.
Data on the former is provided by the Directorate-General for Employment, Social Affairs and Inclusion of the European Commission and consists in what Carr and Chung (2014) define ‘active labor market policies’. The variable is defined as a percentage of GDP, in order to allow the comparison across countries which do not use the same currency nor present the same purchasing power. Carr and Chung (p.388, 2014) argue how LMP spending ‘increases in-line with unemployment’, and, in order to avoid the issues stemming from such relationship, the authors standardize the values by the national unemployment rate (total LMP measures as a percentage of GDP x 100 divided by the national unemployment rate). In contrast with the authors’ study, however, unemployment is not my dependent variable but rather one of my regressors. As a consequence, the only issue such relationship could cause is the presence of multicollinearity, for which I will control.
Finally, the net replacement rate data is retrieved by the OECD database, and represents the
‘proportion of previous in-work income that is maintained after 1, 2, …, T months of unemployment’ (OECD, 2019) for different categories of individuals (e.g. single without children, single with two children, etc.). For my analysis, I choose to use a short-term net replacement rate in unemployment, i.e. the proportion of the average wage maintained after 2 months of unemployment for a couple with 2 children with the partner being out of work. The OECD’s calculations refer to a jobseeker aged 40 with an uninterrupted employment record since the age of 19 until the job loss. The choice of using a couple with two children was determined by the expectation that the cost of being unemployed would be greater for those individuals, and thus, the welfare provided by the state more necessary.
/ 3.3 Control variables
The model employed will also include a set of control variables such as GDP per capita growth, import growth, education and age. Such variables have been found to have a significant relationship with right-wing populist support in a number of studies (Dippel et al., 2015, Dustmann et al., 2017, Inglehart and Norris, 2016). Moreover, since they are both time- and country-variant, they cannot be canceled out with the use of a fixed effects estimation, posing a possible case of omitted variables’ problem to the model. The data with regard to GDP per capita growth and import of goods and services growth is retrieved from the World Bank. The data is relative to the value of the year before the elections for the same reasons argued above
16
with regard to the labor market measures. The choice to include the two variables as a change is justified by the little variation in GDP per capita across the years and the much greater magnitude of the volume of imports of goods and services relative to the other variables. The inclusion of values of such different scale would have produced excessively small coefficients, affecting the results.
The data on education and age are both retrieved from Eurostat. The former is represented by the percentage of individuals between 25 and 64 years who detain tertiary education, whereas the data on the latter is exemplified by the median age of the population. With regard to these two variables I make use of the values relative to the year of the election, since they do not belong to the economic determinants on which individuals might base their voting preference, but they are rather individuals’ features which have been found to be related with populist support.
/ 3.4. Descriptive statistics
The descriptive statistics of the variables employed in the model are reported in Table 1 below.
The dependent variable, 𝑃𝑉𝑆𝑖𝑡, assumes a minimum value of 0 in those cases when no populist party participated in a certain election. The maximum value of 69.40, on the other hand, is the result of the 2010 parliamentary elections in Hungary where the far-right populist political alliance Fidesz-KDNP received an incredibly high share of votes (52.73), complemented by the far-right populist party Jobbik’s result (16.67). The high standard deviation reported (16.65) points to the fact that the dependent variable is quite dispersed, rather than concentrated around the mean value.
17
Table 1: Descriptive statistics
Variables N Mean SD Min Max
pvs 102 18.40 16.65 0 69.40
unemp 102 8.834 4.623 2.700 26.50
lmp 102 0.469 0.325 0.0200 1.450
netrate 102 70.87 13.53 42 133
Δimport 102 4.791 7.945 -31.71 26.02
Δgdppc 102 1.808 4.198 -12.98 24.38
educ 102 27.28 8.057 9.400 45.1
age 102 40.35 2.245 32.80 45.90
On the other hand, the variable unemp exhibits a mean value of 8.834, while peaking to a maximum of 26.50 reached in Greece in 2015. The minimum value of 2.7, instead, is relative to Norway in the year 2009, a rather surprising value given the outbreak of the Great Recession in those same years. The opposite geographical collocation of the two countries exhibiting the maximum and minimum value suggests the possible presence of different patterns among European countries. The second main explanatory variables, lmp, presents rather low values both for the mean and the standard deviation, suggesting a small variance of the variable throughout the time period observed. Once again, the maximum and minimum values point to the existence of divergences between different geographical area, with Denmark and Romania on opposite ends. Finally, netrate is characterized by a quite high standard deviation, suggesting a great variance of the variable across countries, whereas minimum and maximum values are relative respectively to Greece and Denmark.
With regard to the control variables, import growth (Δimport) presents a high mean and standard deviation, suggesting how the variable’s values are quite dispersed, whereas GDP per capita growth (Δgdppc) assumes lower values, indicating a greater concentration around the
18
mean. Finally, the variable educ is also quite dispersed, as hinted by its minimum and maximum value (and confirmed by the standard deviation), whereas age presents a much more concentrated distribution.
In order to better visualize the relationship between the dependent variable and the three main explanatory variables the graphs reporting the scatterplots are displayed below. Due to the presence of many values for each country which would make difficult a clear reading of the graphs, a single average value for each country has been calculated. While the scatterplots report only the one value for each country, the linear fits refer to the total number of observations in the sample to allow a correct visualization of the relationship between populist support and the main explanatory variables.
Graph 1: Scatterplot between populist support Graph 2: Scatterplot between populist support and unemployment rate (averages). and LMP measures (averages).
The variable unemp shows a positive linear fit, whereas both lmp and netrate, the two variables representing the forces mitigating the risk and the cost channels respectively, present a negative linear relationship, in line with the hypotheses formulated in the previous section.
The relationships exhibited by the graphs are confirmed by the correlation matrix in Appendix Table II: the correlation coefficient (and thus, slope of the linear fit in the graphs) are equal to 0.123, -0.208 and -0.234 for unemployment, Labor Market Policies and net replacement rate respectively.
The scatterplots allow also to easily visualize whether there are any outliers which might affect the estimates of the model. In all three graphs, Hungary, Slovakia, Bulgaria, Poland, and Italy stand out from the sample, due to the high votes’ share received by populist parties in those countries’ national elections.
19
Graph 3: Scatterplot between populist support and unemployment net replacement rate (averages).
Since, excluding Italy, all countries belong to Eastern Europe (as defined by the Statistics Division of the United Nations), it might be a sensible choice to present a regression model without Eastern Europe along with the models reporting the full sample.
4. Methodology
In the present section I will define the regression equation I will employ in the study. The model will be characterized by three main explanatory variables and a set of control variables, as previously discussed.
𝑃𝑉𝑆𝑖𝑡 = 𝛽0 + 𝛽1𝑖𝑡𝑈𝑁𝐸𝑀𝑃𝑖𝑡−1+ 𝛽2𝑖𝑡𝐿𝑀𝑃𝑖𝑡−1+ 𝛽3𝑖𝑡𝑁𝐸𝑇𝑅𝐴𝑇𝐸𝑖𝑡−1+ 𝛽𝑖𝑡𝑋𝑖𝑡+ 𝜀𝑖𝑡
𝑃𝑉𝑆𝑖𝑡 is the sum of all votes’ share received by populist parties in the country i in the year of the election t. 𝑈𝑁𝐸𝑀𝑃𝑖𝑡−1 is the unemployment rate over the active population, 𝐿𝑀𝑃𝑖𝑡−1 is the LMP measures’ expenditure as a percentage of GDP and, finally, 𝑁𝐸𝑇𝑅𝐴𝑇𝐸𝑖𝑡−1 is the net unemployment replacement rate for a couple with two children where the partner is out of work after 2 months of unemployment. All variables are relative to the year t-1 for country i. 𝑋𝑖𝑡 represents the set of control variables used in the analysis, including the median age of the population (AGE), the percentage of individuals between 25 and 64 years who detain tertiary education (EDUC), the growth of imports of goods and services (∆IMPORT), and the GDP per
20
capita growth (∆CGDPG). As stressed before, the first two variables refer to the year of the elections (t), whereas the second two are relatively to the year before the election (t-1).
Table 2 below reports the expected signs of the coefficients. The signs of the three main explanatory variables refer to the three hypotheses previously formulated, whereas the control variables’ signs are in line with findings from previous research (Autor, 2016; Dippel, 2015;
Inglehart and Norris, 2016).
Table 2: Expected signs of the explanatory variables
Variable Expected sign
Δunemp +
Δlmp -
Δnetrate -
Δimport +
Δgdppc -
educ -
age +
Given the possibility of time-invariant unobservable or unobserved country-specific features that might affect the dependent variable, I will include country fixed effects in order to cancel them out from the regression. Furthermore, the period of time analyzed is characterized by the outburst of the 2008 Great Recession. Such event deeply affected the support for populist parties across Europe, as proven by a number of studies (Funke et al., 2015; Algan et al. 2017;
Dustmann et al., 2017). Since such variable is time-variant, the country fixed effects cannot account for it. Nonetheless, all European countries in the sample were hit by the crisis. Being it a common feature to all countries in the sample, the use of year fixed effects will allow to control for the event and cancel it out from the regression. Year fixed effects will also allow to account for any other common exogenous shock which has affected all countries, e.g. the entrance of Eastern countries in the European Union or the entrance of China into the WTO.
The OLS fixed effects estimator, while presenting some clear advantages, might raise significant problems with the dataset in use. The dependent variable, 𝑃𝑉𝑆𝑖𝑡 , belongs to Time Use Data
21
(TUD). Such data is characterized by non-negative and right skewed values. Furthermore, for certain observations the dependent variable assumes the value of zero. In Spain, for instance, the far-left populist party Podemos has participated only in the 2015 and 2016 elections, since it has been founded only recent years, while Portugal does not present any populist party in the time period observed. To avoid selection biases, the elections with no populist party are included in the dataset as zeros. As a result, “the least squares of the regression parameters are biased and inconsistent” (p.617, Hill et al., 2011) due to the violation of the assumption of normality.
While linear regression models can tolerate moderate violations of the assumptions of normality (Brown and Dunn, 2011), for TUD containing many zeros and greatly skewed to the right alternatives such as the Tobit or the Poisson Gamma model are recommended. The former
“assumes the zeros represent censored values of an underlying normally distributed latent variable that theoretically includes negative values” (p.512, Brown and Dunn, 2011). Such assumption has been criticized by some scholars (Maddala, 1992), who argue how it might be correct when the null values represent structural zeros (e.g. countries who do not have any populist party), but risks to negatively affect the significance tests when such values are sampling zeros (e.g. countries who do not have any populist party only in the time period observed).
Finally, the Tobit model assumes constant variance, leading to the risk of inconsistent and biased parameter estimates in the presence of heteroskedasticity (Hurd, 1979). Nonetheless, the Tobit model was successfully employed by Swank and Betz (2003) in a study on the increasing support of right-wing populist parties, a research question closely related to the present paper.
The Poisson-gamma model, also known as the negative binomial regression, on the other hand, has also been used to address TUD. Contrary to the Tobit model, this model “copes with genuine (non-censored) zeros that represent zero episodes, as well as a continuous component where activity episodes have occurred.” (p.531, Brown and Dunn, 2011). Furthermore, the negative binomial regression does not assume variance to be constant, thus, allowing to account for heteroskedasticity. Compared to the normal Poisson model, this model allows to deal with overdispersion, i.e. the presence of a much greater variance than the mean in the dataset in use.
Since I suspect such feature in my sample, I run the negative binomial regression which reports also a likelihood ratio test to control for overdispersion (Hilbe, 2011). The value presented by the test confirms the overdispersion of the data, which justifies the use of the Poisson-gamma model rather than the simple Poisson model. While much of the previous literature has traditionally employed the model exclusively in the presence of count data, it has been recently adopted also in the presence of continuous dependent variables (Hilbe, 2011). Given the fewer assumptions required by the model and its recent diffusion with regard to continuous variables, I will include it among the models employed in the study.
22
Moreover, given the presence of a limited number of zeros in the dataset in use, there is the possibility that OLS might tolerate such moderate violation of normality. On the other hand, since such possibility is not certain, it might be helpful to employ also the Poisson-gamma and Tobit models. I will therefore employ all three models and proceed to compare the different estimates obtained, which will also serve as a robustness check of the results obtained. Finally, I will also include another Fixed Effects estimator model. This model will differ from the one previously discussed in the way the observations with a null dependent variable are treated. In the second OLS Fixed Effects model, the observations with the null value for the dependent variable are substituted with missing values, and therefore, dropped. Such methodological step might lead to selection biases, but it is justified by the fact that in the dataset in use the number of zeros does not exceed 14% of the total number of observations. Such observations are equally divided between sampling and structural zeros i.e. zeros result of the absence of a populist party in the sample in use and zeros reflecting the complete absence of a populist party in the country.
5. Results
In the present section I will discuss the results of the regression analyses. To avoid heteroskedasticity issues, the standard errors reported are robust. The first column refers to the OLS fixed effects estimator with the dependent variable assuming the value of 0 in the absence of populist parties participating in certain elections. The model employed in the second column is the Poisson-gamma (or negative binomial regression) with fixed effects, whereas the third column reports the Tobit estimates. Finally, the fourth column reports the estimates obtained with the OLS fixed effects estimator by excluding the observations where the dependent variable assumes a null value.
As argued in the “Labor market data” section (3.2) LMP measures’ expenditure increases in-line with unemployment. Since the variable is expressed as a percentage of GDP, the presence of unemployment and GDP per capita growth among my explanatory variables might lead to multicollinearity issues. I control for this possibility through the correlation matrix reported in the appendix Table II. The correlation matrix does not report any coefficient exceeding 0.8, except the one relative to Δimport and Δgdppc. In order to rule out any possible problem, I calculate the Variance Inflation Factor (VIF), an indicator commonly used to measure the degree of multicollinearity. When such indicator is higher than 10, the relationship between the explanatory variables affects the reliability of the model (O’Brien, 2007).
23
The VIF values are reported in appendix Table III. No VIF exceeds the value of 10, meaning that the model does not present a multicollinearity issue.
Among the models reported in Table 3, the Tobit model is the one I consider the least reliable due to the many underlying assumptions it requires. Such assumptions have been often harshly criticized in recent times, with less and less scholars employing the model in their studies.
Moreover, the OLS estimator in column 4 might also present some significant limits given the exclusion of election years with no populist parties, which leads to a narrower sample and possible selection biases.
Three out of the four models reported in Table 3 show a significant and positive coefficient for the unemployment rate, in line with the hypothesis previously formulated. According to the estimates, an increase in the unemployment rate leads to an increase in populist support. The result reinforces the same findings from the studies by Algan et al. (2017) and Dustmann et al.
(2017) against the contradicting outcomes obtained by other researches (Knigge, 1998; Lubbers et al, 2002; Swank and Betz, 2003).
The variable lmp, on the other hand, is expected to lower the risk of becoming unemployed, and thus, to entertain a negative relationship with the dependent variable. Nevertheless, the coefficient presents a positive and significant value in two out of four models, whereas in the remaining two it is insignificant. Such ambiguous estimates do not allow to infer unequivocally whether LMPs’ measures do have an actual effect on populist support. The absence of an effect might be due to the fact that the variable does not cover any role in mitigating the populist support, i.e. the risk of becoming unemployed might not be a relevant factor in determining populist votes’ share.
24
Table 3: General model results
(1) (2) (3) (4)
Variables OLS FE Gamma-Poisson FE Tobit OLS FE
unemp 0.595** 0.0276 0.698*** 0.695**
(0.270) (0.0196) (0.244) (0.280)
lmp 14.00 0.689 17.40** 13.86**
(8.350) (0.473) (6.775) (5.958)
netrate -0.217* -0.0274** -0.365*** -0.353*
(0.118) (0.0111) (0.132) (0.173)
Δimport 0.270 0.0343* 0.358* 0.0339
(0.197) (0.0175) (0.187) (0.244)
Δgdppc 0.447 0.0149 0.436 0.565
(0.288) (0.0246) (0.287) (0.340)
educ -1.124** -0.0386 -1.201*** -0.831*
(0.470) (0.0290) (0.420) (0.405)
age 1.067 -0.00261 1.919 3.645*
(2.420) (0.115) (1.645) (2.119)
Constant -4.390 3.604 -29.32 -97.22
(94.11) (4.611) (69.38) (83.47)
Observations 102 94 102 88
R-squared 0.540 0.617
Number of country 28 26 26
Country FE YES YES YES YES
Year FE YES YES YES YES
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Note: Column 1, 2, and 3 report the estimates from the three different models including zeros for those observations in which there was no populist party, whereas column 4 excludes those observations.
unemp is the unemployment rate over the active population in the year t-1, lmp is the Labor Market Policies’
expenditure as a percentage of GDP in the year before the elections, netrate is the proportion of average wage received by a member of a couple with two children whose partner is out of work after two months of unemployment. With regard to the control variables, Δimport is the annual growth of imports of goods and services and Δgdppc is the GDP per capita growth, both relative to the year t-1. Age is the median age of the population and educ is the percentage of individuals between 25 and 64 years who detain tertiary education in year t.
25
While columns 3 and 4 represent the least reliable models, they present a significant positive coefficient for a variable (lmp) that was expected to negatively affect populist support. For this reason, a series of explanations is offered to account for such outcome.
First, the positive relationship might be the result of populist political pressure. For instance, it might be the case that traditional parties, feeling the increasing support towards populist parties as a threat for their own existence, would attempt to mitigate the general discontent by increasing the expenditure in welfare programs such as LMPs. As a consequence, populist parties might gain even more support, since voters would feel that the political pressure is producing results. Alternatively, if the populist party was part of the government before the election, then the increase in LMP measures’ expenditure could be the result of policies advocated by the party itself. Given the mitigating role such policies have on the economic uncertainty felt by the population, the voters would feel more motivated to vote for the populist party in the following election.
Nonetheless, lmp might also be simply inadequate as a variable compensating the risk of becoming unemployed. The resources’ allocation might be inefficient or the initiatives proposed by the government might not be effective in lowering the economic uncertainty perceived by the population. Moreover, the variable is characterized by low, concentrated values. Such features explain the magnitude of the coefficient, since an increase of one percentage point on the GDP would be more than a 100% increase for most countries.
The variable netrate is expected to lower the cost of becoming unemployed, and, thus, to have a negative relationship with populist support. The higher the value of netrate the greater the proportion of the previous wage provided by the state to a person who has become unemployed. The coefficient is negative and significant in all columns, although the level of significance changes from one to the other. The hypothesis is therefore confirmed and is robust across the different models employed. The confirmation of the third hypothesis openly contradicts the possible reasons for lmp’s positive and significant coefficient previously enlisted.
If the political pressure or the economic policies advocated by populist parties would lead to an increase in Labor Market Policies’ expenditure, the same reasoning should apply to the unemployment net replacement rate. Ergo, the different outcome for such variable than the one observed for lmp seems to suggest how the cost channel might be a greater determinant of economic insecurity than the risk channel instead. The mechanism mitigating the cost of being unemployed might be more effective in lowering populist sentiment across the population, whereas the mechanism intervening on the risk channel seems to not have a significant effect.
26
Finally, with regard to the control variables, educ is significant in three out of four models, suggesting how more educated people tend to avoid voting for populist parties. The magnitude of the coefficient indicates that an increase of 1 percentage point in the proportion of the population detaining tertiary education has a considerable negative impact on populist support.
Moreover, Δimport is significant in two out of four models. Nevertheless, the variable’s coefficient is not significant in the two OLS Fixed Effects models, making the significant estimates rather unreliable. The questionable results with regard to the variable might be caused by the limits of the measure employed. As a matter of fact, import growth can reflect also input necessary to the national production which is then destined to the domestic market or to be exported abroad. Consequently, import growth does not reflect exactly import penetration, i.e.
the import that raise the competitiveness for domestic firms, thus, leading to job losses.
The remaining control variables do not offer any satisfying estimate either. On the one hand, age is significant only in one out of four models, i.e. the OLS model excluding observations with no populist party in the election. This seems to suggest how the narrowing down of the sample might have produced selection biases. On the other hand, Δgdppc is insignificant across all models employed, suggesting how the variable does not affect populist support.
The results reported in Table 3 show how unemp, netrate and educ are the only results that remain robust across the majority of the models. In the Descriptive Statistics’ section (3.4), however, the scatterplots suggested the presence of outliers which might affect the estimates.
Since most of the outliers were Eastern European countries (e.g. Bulgaria, Hungary, Poland) I now proceed to drop the countries that belong to Eastern Europe according to the Statistics Division of the United Nations. Table 4 presents the new estimates after narrowing down the sample. As shown by the table, the number of country has now dropped to 22 (20 for models which exclude countries presenting only zeros for the dependent variable).
27
Table 4: General model without Eastern European countries
(1) (2) (3) (4)
Variables OLS FE Gamma-Poisson FE Tobit OLS FE
unemp 0.917** 0.0347 1.027*** 0.894**
(0.333) (0.0232) (0.233) (0.357)
lmp 8.791 0.353 11.81* 11.96**
(7.764) (0.556) (6.453) (5.323)
netrate -0.220* -0.0402*** -0.489*** -0.498***
(0.117) (0.0136) (0.133) (0.124)
Δimport 0.235 0.0320** 0.270* -0.0784
(0.203) (0.0163) (0.155) (0.233)
Δgdppc 0.580* 0.0411* 0.712*** 0.861**
(0.296) (0.0241) (0.253) (0.301)
educ -1.310*** -0.0923*** -1.429*** -1.079***
(0.389) (0.0310) (0.356) (0.367)
age 0.220 0.0138 0.973 2.923
(2.573) (0.146) (1.475) (1.894)
Constant 30.95 6.310 22.50 -56.70
(98.57) (5.934) (61.88) (73.79)
Observations 82 74 82 69
R-squared 0.626 0.725
Number of country 22 20 22 20
Country FE YES YES YES YES
Year FE YES YES YES YES
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Note: the table refers to the original model with the exclusion of Eastern European countries according to the definition of Eastern Europe provided by the Statistics Division of the United Nations. The countries excluded from the analysis are the following: Bulgaria, Czech Republic, Hungary, Poland, Romania, Slovakia.
28
Table 4 presents a similar picture to Table 3, although with a few significant differences.
Unemployment is still significant in three out of four models, but the coefficient is now much higher than the one exhibited in the previous table. The same goes for the unemployment net replacement rate, which is still significant across all models (at a higher level than before) and whose magnitude has increased relative to the previous estimates. Moreover, education’s coefficient is also consistent with Table 3, and, as for the other variables, the value is now higher.
The change observed in educ seems to suggest how the countries dropped from the sample were also outliers relative to this variable.
Finally, and most importantly, Δgdppc is now significant and positive at different levels across all models. The positive relationship between GDP per capita growth and populist votes’ share contradicts the variable’s expected sign and previous findings (Swank and Betz, 2003). A greater economic growth should reduce current economic uncertainty, and thus, lowering the support for populist movements who leverage on the population’s discontent. A possible interpretation for such outcome is that GDP per capita growth is a highly volatile value, and thus, the value relative to a single year does not affect the perceived economic uncertainty nor voters’ choice.
Moreover, caution should be used whenever GDP per capita growth is interpreted as a measure of a population’s well-being. In the last two decades, scholars have progressively scaled down the indicator’s importance, proposing alternative measures to grasp the myriad of factors affecting the economic welfare of a country (Arrow et al., 2004; Jones and Klenow, 2010).
29
Conclusion
In recent years, Western societies have witnessed the resurgence of populist sentiment. In the US, the far-right populist rhetoric was of fundamental importance in allowing Trump to win the 2016 US presidential election, while Front National in France, Lega in Italy, FPÖ in Austria, have experienced a dramatic increase in their votes’ share, deeply influencing the economic policies of the national governments. The re-emergence of populism has posed a significant threat to the unity and cohesion of supranational entities such as the European Union. For the profound impacts such phenomenon might entail, the academic world has attempted to define the determinants of its success. While most of the previous literature has traditionally focused on the role played by globalization (one of the milestones of the right-wing populist rhetoric), the present paper has targeted the labor market. In the words of Rodrik (2018), the ‘labor market protection erosion’ is one of the factors which determined populism’s recent surge. However, scholars have often overlooked said factor, addressing import penetration instead.
The present study attempted to measure the role of labor market by using three different variables: unemployment, LMP measures’ expenditure and unemployment net replacement rate.
The first variable was previously tested as a proxy for the 2008 Great Recession by a number of studies (Algan et al., 2017, Dustmann et al., 2017) and was expected to present a positive relationship with populist support. The second and third variables, on the other hand, represent mechanisms that should act as mitigating forces relative to the risk and cost of unemployment, respectively. By testing whether those two variables entertained a negative relationship with populist support, the paper addressed the ‘labor market protection erosion’ described by Rodrik, i.e. whether a lower level of those variables had a positive impact on populist votes’ share.
The results, however, confirmed only two hypotheses, i.e. that a higher level of unemployment is associated with a higher endorsement for populist parties, and that a higher level of unemployment net replacement rate is negatively related to populist support. The hypothesis relative to the variable representing the unemployment’s risk channel did not find confirmation.
The reasons for such outcome can be at least fourfold. First, from a methodological perspective, the sample observed in the present study suffered from the lack of data, which negatively affected the number of observations. On the same matter, it is important to stress how, to this date, there is not an unequivocal definition of populism yet. Hence, the lack of significant results might be classification-specific, meaning that if a different classification was employed, the same sample might produce different results.
30
Second, there might be better suited variables to represent the risk channel of unemployment than the one used in the current study. LMPs’ expenditure measured as a percentage of GDP is characterized by small values relatively concentrated. Hence, a variable with a greater variance might be more accurate to represent the risk of becoming unemployed.
Third, the risk channel might simply not affect populist votes’ share, thus, not figuring among the economic determinants of the phenomenon. The significant effect found with regard to the cost channel seems consistent with such explanation: if the risk is a matter of perception, and thus, can be felt more or less intensely depending on the person, the cost immediately affects the material conditions of the unemployed person, and hence, its well-being. Finally, the ‘labor market protection erosion’ might be advocated more strongly by far-left populism rather than far-right. While the present research addresses populism of any form (far-left, far-right, or none), the presence of far-left populism is rather small across Europe, and thus, the topics omnipresent in the far-right populist rhetoric (e.g. globalization, immigrants, euroskepticism) might be more accurate determinants of such phenomenon.
The research’s findings regarding unemployment’s cost channel entail significant policy implications. A greater coverage and expenditure on welfare programmes mitigating the cost associated with unemployment might be effective in reducing the discontent among the population, negatively affecting populist rhetoric and leverage. As a consequence, governments interested in avoiding an increasing populist support should increment their commitment in lowering the economic uncertainty caused by unemployment.
The present study has the merit of approaching the recent populist surge from a different perspective than the one typically assumed by the literature on the topic. Being the ‘labor market protection erosion’ and the role of mechanisms mitigating populist support relatively unexplored, further research is certainly needed on the topic. In order to allow a deeper analysis, more extensive data relative to unemployment’s risk and cost channels should be gathered, taking into consideration also the employment of different measures and classifications.
31
References
Algan, Y., Guriev, S., Papaioannou E., Passari E. 2017. The European Trust Crisis and the Rise of Populism. Brookings Papers on Economic Activity.
Arrow, K., Dasgupta, P., Goulder, L., Daily, G., Ehrlich, P., Heal, G., Levin, S., Mäler, K., Schneider, S., Starrett, D., and Walker, B. 2004. Are We Consuming Too Much? Journal of Economic Perspectives, 18 (3): 147-172.
Autor, D., Dorn, D., Hanson, G., & Majlesi, K. 2016. Importing political polarization? The electoral consequences of rising trade exposure. Working Papers 2016:21, Lund University, Department of Economics.
Aytaç, S. and Onis, Z., Varieties of Populism in a Changing Global Context: The Divergent Paths of Erdogan and Kirchnerismo (May 6, 2013). Comparative Politics, 47(1): 41-59. Available at SSRN: https://ssrn.com/abstract=2261178 or http://dx.doi.org/10.2139/ssrn.2261178 Baldwin, R. 2016. The Great Convergence: Information Technology and the New Globalization.
Belknap Press of Harvard University Press.
Bellemare, Marc F., Takaaki Masaki, and Thomas B. Pepinsky. 2017. Lagged Explanatory Variables and the Estimation of Causal Effect. The Journal of Politics 79:3, 949-963 Brown, J. E., & Dunn, P. K. 2011. Comparisons of Tobit, Linear, and Poisson-Gamma Regression
Models: An Application of Time Use Data. Sociological Methods & Research, 40(3), 511–
535. https://doi.org/10.1177/0049124111415370
Carr, E. and Chung, H. 2014. ‘Employment Insecurity and Life Satisfaction: The Moderating Influence of Labour Market Policies across Europe’. Journal of European Social Policy, Vol.
24, pp. 383–99.
Colantone, Italo, and Piero Stanig. 2016a. Global Competition and Brexit. BAFFI CAREFIN Centre Research Paper No. 2016-44, September.
Deegan Krause, K., Haughton, T. 2009. Toward a more useful conceptualization of populism:
Types and degrees of populist appeals in the case of Slovakia. Politics and Policy, Vol.
37(4), pp.821-841.
Dippel, Christian, Robert Gold, and Stephan Heblich. 2015. Globalization and Its (Dis-)Content:
Trade Shocks and Voting Behavior. NBER Working Paper No. 21812.
32
Dustmann, C., Eichengreen B., Otten S., Sapir A., Tabellini G. and Zoega G. 2017. Europe’s Trust Deficit: Causes and Remedies, Monitoring International Integration. CEPR Press.
Eurostat. 2019. [database] Retrieved from https://ec.europa.eu/eurostat/data/database
Freeden, Michael. 1996. Ideologies and Political Theory: A Conceptual Approach. Oxford: Oxford University Press.
Funke, M., Schularick M., and Trebesch C. 2015. Going to extremes: Politics after financial crises, 1870-2014. CEPR, Discussion Paper No. 10884.
Gidron N, Bonikowski B., 2013. Varieties of Populism: Literature Review and Research Agenda, in Weatherhead Working Paper Series, No. 13-0004.
Guiso, L., Herrera, H., Morelli, M., & Sonno, T. 2017. Demand and supply of populism. CEPR Discussion Paper DP11871.
Halikiopoulou, D. and Vlandas, T. 2016. Risks, costs and labour markets: explaining cross-national patterns of far right party success in European Parliament elections. Journal of Common Market Studies, 54 (3). pp. 636-655. ISSN 1468-5965
Hurd, Michael. 1979. Estimation in Truncated Samples When There Is Heteroscedasticity. Journal of Econometrics 11:247-58.
Hilbe, J. M. 2011. Negative binomial regression. Cambridge, UK, Cambridge University Press.
Inglehart, Ronald F. and Norris, P. 2016 Trump, Brexit, and the Rise of Populism: Economic Have- Nots and Cultural Backlash. HKS Working Paper No. RWP16-026. Available at SSRN:
https://ssrn.com/abstract=2818659 or http://dx.doi.org/10.2139/ssrn.2818659
Ionescu, Ghita, and Ernst Gellner. 1969. Populism: Its Meanings and National Characteristics. New York: Macmillan.
Jones, C. I. and Klenow, P, J. 2010. Beyond GDP? Welfare Across Countries and Time. NBER Working Paper No. w16352. Available at SSRN: https://ssrn.com/abstract=1674796 Kazin, M. 1995. The Populist Persuasion: An American History. Itacha: Cornell University Press.
Knigge, P. 1998. The ecological correlates of right-wing extremism in Western Europe, European Journal of Political Research 34, 249-279.
33
Lubbers, M., Gijsberts, M., & Scheepers, P. 2002. Extreme right‐wing voting in Western Europe, European Journal of Political Research, 41(3), 345-378.
Maddala, G. 1992. ‘‘Censored Data Models‘‘ In The New Palgrave Econometrics, edited by J.
Eatwell, M. Milgate, and P. Newman. London: Macmillan.
Malgouyres, C. 2014. The Impact of Exposure to Low-Wage Country Competition on Votes for the Far-Right: Evidence from French Presidential Elections. Mimeo, European University Institute.
Mudde, C. 2004. The Populist Zeitgeist. Government and Opposition, 39, pp.542-563.
O’Brien, R.M. 2007. A Caution Regarding Rules of Thumb for Variance Inflation Factors, Quality and Quantity 41: 673. https://doi.org/10.1007/s11135-006-9018-6
OECD. 2019. [database] OECD Social Protection and Well-Being. Retrieved from https://data.oecd.org/
Rodrik, D., 2018. Populism and the economics of globalization. Journal of International Business Policy, June 2018, Volume 1, Issue 1–2, pp 12–33.
Stanley, B. 2008. The Thin Ideology of Populism. Journal of Political Ideologies, Vol.13 (1), pp.95- 110.
Swank, D. and Betz, H.G. (2003) ‘Globalization, the Welfare State and Right-Wing Populism in Western Europe’. Socio-Economic Review, Vol. 1, pp. 215–45.
World Bank. 2019. [database] Global Development Indicators. Retrieved from https://data.worldbank.org/
