Populism and Social Polarization in European Democracies

Abstract

1. Introduction

During the last 15 years, two parallel evolutions could be observed: mounting populist waves and widening social divides. We show that these evolutions were a matter of perception much more than of reality, a perception that was conveyed by resounding elections of populist characters (Bolsonaro in Brazil, Trump in the US, Orban in Hungary, and others), accompanied by upsurges of social protests such as the Gilets Jaunes in France, which received a wide coverage in the media. When looking at hard facts, we can observe that indeed populist attitudes as well as social divides prevail in most countries and that both phenomena are closely related, though their extent reveals to be quite stable over time but varies across countries.

We do not rely on voting behavior to capture the populist wave, but rather on values that are traditionally viewed as populist, such as distrust of institutions and neighbors, rejection of migrations, and strong preferences for law and order. We collected these values for 25 European countries from 2004 to 2018. Social divides are measured by the index of polarization introduced by Esteban and Ray (1991, 1994, 2011), a measure that is often applied in political science to study conflicts. We match aggregated indicators of populist values and social polarization using, respectively, European Social Survey (ESS) and Survey on Income and Living Conditions (SILC) survey micro-data.

A couple of words on these choices are in order. First, we explain populist attitudes rather than populist votes, which are neither consistent nor homogeneous: they come from very different types of elections (local, regional, national, and even European) and the definition of populist parties is often questionable and changing. Second, traditional measures of inequality and poverty do not capture well the complex reality of social divides and the tensions between the middle-class and both the lower and the upper tail of the income distribution. Polarization indicators do reflect such reality.

Our approach is at odds with the usual one that relates the populist vote to factors such as globalization, cost of living increase, immigration, or unemployment. It is closer to the work by Norris and Inglehart (2018) who use the ESS data to explain populist vote and authoritarian/libertarian values with generational differences and cultural backlash, as main factors. We use the same data, but focus on social polarization as possible argument for populist attitudes.¹

To be clear, there are four possible variables (or groups of variables) that are important: (a) populist votes, (b) populist attitudes and values, (c) social divides, and (d) a number of economic factors. In this article, we look at the effect of (c) on (b), while most of the literature focuses on the effect of one or several variables listed in (d) on (a).

This literature explains populist votes by factors such as an extended lack of income growth, combined with massive growth at the top of the income distribution; an education system that limits the opportunities of children from modest parental backgrounds; outsourcing of ‘good’ jobs to China and other emerging economies; technological change that has made many ‘routine’ middle-class jobs redundant. A couple of examples follow.²

Using 17 years of the German Socio-Economic Panel, Geishecker and Siedler (2012) find strong and robust evidence that subjective job-loss fears foster affinity for parties at the far right-wing of the political spectrum. Nikolka and Poutvaara (2016) and Becker et al. (2017) show that the 2016 Brexit referendum result in the UK is strongly correlated with various characteristics of voters across districts. Individuals over 45 years old, having little or no qualification, living in areas with a strong tradition of manufacturing employment or with a high number of immigrants from the twelve 2004 EU accession countries, were identified as strong predictors of the Brexit vote. Autor et al. (2020) study the populist vote in the 2016 American presidential elections. They show that growing import competition from China contributed to a shift in congressional voting toward ideological extremes. In a study based on regional data for 14 Western European Countries over the period 1993– to 2016, Anelli et al. (2019) find that higher robotization exposure may lead to increasing support for national and radical right parties. In his analysis of populism across Europe and America, Rodrik (2017) shows that throughout history waves of globalization are prone to populist backlashes. Algan et al. (2017a) try to explain the Front National (FN) vote during the recent French presidential election. They show that a sense of deteriorating wellbeing is one of the main explanations of the rising support for the FN, cutting across most boundaries of age, education, or economic status.

Edo et al. (2019) analyze presidential elections results in France between 1998 and 2017. They show that extreme right and extreme left votes are associated with immigration. This association is positive for the extreme right and negative for the extreme left. However, it is stronger for the former than for the later. In our paper, we follow Edo et al. (2019), who use an instrumental variable (IV) approach to address the potential bias due to endogenous immigration.³Bordignon et al. (2019) also analyze municipalities’ data but in a Northern Italian region, Lombardia, and conclude that ‘the share of immigrants follows a U-shaped curve, which exhibits a tipping-like behavior around a share of immigrants equal to 3.35%’. Finally, Algan et al. (2017b) study the relation between unemployment and the rise of populism, vote, and values, using ESS microdata aggregated at the regional level. Their estimates, which include the great recession period, confirm a strong positive relation between the growth of unemployment and the increasing populist vote in European countries.

More in line with our study, another path of the literature is interested by noneconomic factors driving populist attitudes. Guriev et al. (2020) analyze the link between the growth of mobile internet, 3G broadband, over the period 2008–2017 around the world and individuals’ government approval. Using Gallup World Poll surveys, they show that confidence in government increases with mobile expansion, at least if the access to the internet is not censored. This is also in line with experimental studies which contradict the view that economic factors account for a wide portion in populist attitudes and vote (Margalit 2019).

Anticipating the results of our paper, we show that social polarization varies dramatically across European regions though, at the same time, some convergence is observed. Our estimations often confirm a positive and statistically significant relation between social polarization and populist attitudes.

We also show that, in most cases, our results are relatively invariant at the level of regions (NUTS1) instead of countries. These results are also consistent, with some rare exceptions, when we run our model separately for three different age-cohorts: born before 1946, born in 1946–1974, and born after 1974.

This article is organized as follows. In Section 2, we discuss our main variables, namely populist attitudes and polarization and show how they differ across countries and over time. In Section 3, we test the relation between populist attitudes, the degree of social polarization and covariates, particularly immigration inflows, using OLS and IV-2SLS, to address potential endogeneity problems. In Section 4, we present robustness checks, using regional and age-cohorts panel data. We conclude in Section 5.

2. Social Divides and Populist Attitudes

2.1. Data

To study the link between social divides and populist attitudes, or behavior, we rely on two large European surveys, SILC (Eurostat, 2020) and ESS (European Social Survey, 2020).

SILC microdata are used to compute polarization indices that describe the possible fractures of living conditions across income classes. SILC is an annual survey that started in 2004 and includes data on economic and living conditions, in particular on the disposable income of households. The sample is representative of the population aged 16 years plus.

ESS data come in to compute several indicators related to individual populist attitudes and voting behavior. The surveys started in 2002, but ESS collects data during even years only. Some parts (or modules) are repeated every even year, others are more occasional, or even unique. ESS aims at collecting changes in individual ‘attitudes, beliefs and behavior patterns’ including political orientation and parties for which individuals voted in the last election that preceded the year in which the survey was taken. The sample is representative of the population aged 18 years plus.

We combine the information available from both surveys, and aggregate it at the level of countries and years, implicitly assuming that both samples, randomly chosen from the same population, are representative. Table A1, in Appendix A, describes which data are available by country and years. Given that ESS runs its surveys every even year, while SILC started in most countries for every year in 2005, the final panel includes 25 countries, 8 years (2004–2018, even years only), but only 157 data points⁴ instead of 25 × 8 = 200, since some countries are missing, especially in 2004.⁵ With the exception of Greece, Italy, and Iceland, all countries are present at least five times out of eight, but the panel is obviously unbalanced. As a robustness check, we also estimate the same equations using a panel data of regions (NUTS 1 level), instead of countries. Given that NUTS information is not available in ESS before the 2010 wave, the final panel is composed of 279 observations.⁶

2.2. Populist attitudes

Political scientists use two types of variables to address populism: (i) Votes or membership participation to populist parties or (ii) values and attitudes, which can be considered supporting or related to populist behavior. We chose option (ii) for the reason that in surveys evidence on populist values is much more reliable than voting behavior: Interviewees often do not remember the vote they casted, or feel uncomfortable to confess they voted for a populist party or personality, and others simply did not vote (some 30% in many countries), as we can see from ESS data reported in Appendix Table A2.

We follow Norris and Inglehart (2018), who introduced indicators of ‘distrust of institutions,’ ‘anti-immigration feelings’, and ‘leaning for law and orders’ (authoritarianism) in their work. These are computed using ESS individual answers to specific questions. We added a fourth indicator reflecting ‘distrust of people,’ which also prevails in populist behavior (Olivera 2015). Appendix Table A3 displays the questions used. For each indicator and each individual, we added the scores given as answers to the corresponding questions, and normalized them between 0 and 100. Table 1 contains a summary by country over three periods (2004–2008, 2010–2014, and 2016–2018). Countries are classified in four homogenous groups. As can be observed, there are large variations with and within each subgroup depending on the indicator at hand. However, more importantly and contrary to what is usually assumed, there are no clear time trends.

As already mentioned, voting data are not consistent. Nevertheless, we used the ESS sample to obtain some evidence that extreme right voting and abstentions are positively related to distrust of institutions, anti-immigration attitudes, authoritarianism, and distrust of people, by running logistic regressions in which the dependent variable (extreme right voting results or abstentions) values 0 or 1. We also include gender, education, and year of birth (age cohort) as control variables. Results are shown in Table 2.

Votes for extreme right parties and abstentions may both reflect a rejection of the political system. They are positively related to each populist attitude, particularly to anti-immigration attitudes, in the first case, and to distrust in institutions, in the second.⁷ The only puzzling result is the negative effect of authoritarianism on abstentions in the last two columns of Table 2. In each case, the same analysis was run controlling for heterogeneity across countries and years and the results are relatively stable. In most cases, the estimated parameters vary but their sign and statistical significance do not change. Gender, education, and age cohorts’ effects are also relatively stable and, in most cases, with the expected sign. The results confirm huge vote abstention rates among younger generations of less educated people and among those who voted, females, high educated, and aged people’s votes go less often to extreme-right parties.⁸ The same age and education gradients are reported by Nikolka and Poutvaara (2016) in the case of leave votes in the 2014 British Brexit referendum.

2.3. Social polarization

Income inequality and poverty measures are often used to obtain a representation of the state of a society.⁹ However, it has become increasingly clear that these measures do not reflect the ‘feeling of being left behind’ that characterizes the lower middle class and that fuels political and social instability in a number of ways. First, increasing ‘social barriers’ between groups implies that individuals feel less familiar with and connect less to other people. Second, it is difficult to develop trust in others if they are seen to have unfair advantages. Finally, unequal communities may disagree over how to share (and finance) public goods, and those disagreements can turn breaking social ties and lessen social cohesion. Broken trust leads to intolerance and discrimination. To assess these social divides, we assumed that simple poverty measures are insufficient, and we resorted to polarization measures that have been widely used to analyze ethnic conflicts and linguistic differences. Based on a recent study by OECD (2019), we postulate that a polarization process is at work in European societies and that it is possible to measure it making the distinction between three main income categories within the population, low-income, middle-income, and upper-income classes, on the basis of their relative position to a country’s median income. As shown in this OECD study, the middle-class is under pressure, particularly in Western industrialized countries, where the increasing cost of life and job insecurity simultaneously impact their living conditions and reduce their social mobility opportunities.¹⁰

The idea of polarization can be described using a certain number of steps. Assume that we have a given exogenous partition into income groups j and k, $ni,j$ and $ni,k$ in country i where $ni,j$ and $ni,k$are the shares of income group j and k in country i. Income ‘diversity’ can then be defined by $∑j∑kni,j.ni,k,j≠k$⁠. If, in addition, one can estimate distances $δi,jk$measured by, say the ‘ability to make ends meet’ between groups j and k, then ‘distance weighted diversity’ can be written as $∑j∑kni,j.ni,k.δi,jk$⁠.¹¹ Esteban and Ray’s (2011) formulation of polarization is very close to diversity, but they include a parameter (1+ $θ)$⁠, $θ∈0,1.6$⁠, which expresses the fact that members of group i put a larger weight (for instance in terms of confidence) on those of the same group than on other ones, or are more antagonistic to those who do not belong to their own group: $pi=∑j∑kni,j1+θ.ni,k.δi,jk$⁠.

Using SILC microdata and following OECD (2019) definitions, we segment populations in each year and country in three groups. Middle-income group involves individuals in households with disposable income (standardized using the OECD equivalence scale) between 0.75 and twice the median disposable income. The lower-income group contains those individuals in households with disposable income less than three-fourth of the median disposable income. The upper-income group contains those whose income is larger than twice the median disposable income.

SILC microdata also allow to compute two types of distances which reflect living conditions gaps across income classes. First, a subjective measure based on the answer given by the households’ reference individual, on her/his ‘ability to make ends meet.’ The answer is qualitative and goes from ‘with great difficulty’ to ‘very easily’ on a scale from 1 to 6. For each income group, we take the percentage of individuals with score 1 (‘with great difficulty’) or 2 (with ‘difficulty’). Secondly, we use a so-called ‘material deprivation index’ (Townsend 1979) which corresponds to the addition of binary answers (yes or no) to an array of SILC questions. Their precise formulation can be found in Appendix Table A5, which also contains questions that make it possible to compute Townsend’s deprivation index (see also Verbunt and Guio 2019). For each income group, we take the average percentage of individuals who cannot afford two or more of these items.

Table 3 provides the distances between the lowest and the highest income classes for the ‘ability to make ends meet’ and for the index of ‘deprivation’. The difference between Central-Western and Northern Europe, on the one hand, and Southern and Eastern Europe, on the other, is, as expected, striking. Across the period, however, the distance between income classes, particularly the gap measured by the index of material deprivation, diminished dramatically in a majority of Eastern countries.¹² At the same time, it increased in other European countries whose population were more affected by the financial crisis: Greece and Spain, but also Ireland and the UK.

Regarding identification, Esteban and Ray (1991, 1994) show that, under given axioms, the parameter $θ$ can take values ranging from 0 to 1.6. In what follows, we have chosen the most commonly used value, that is $θ$ = 1.¹³

Using the above distances and the identification value of 1, we compute polarization indexes with distances based on subjective or material deprivation scores, listed in Table 4.

Focusing first on the indices based on the subjective ‘ability to make ends meet’ we observe, as Wang et al. (2018) did, that average polarization is higher in Southern and Eastern countries and lower in Northern countries.¹⁴ We observe however also that, between 2004 and 2018, increases are particularly important in Nordic and Central-Western countries. Only Finland experiences a decline in polarization. For the two other subgroups, results are mixed. In Southern countries, polarization increases dramatically in Cyprus and Spain while in Greece it decreases. In Eastern Europe, Czechia, Estonia, and Poland also happen to become less polarized.

The pattern is somewhat different for ‘deprivation scores.’ On average, our indices are quite stable over time in all regions, with the exception of Southern Europe. They also tend to be higher in Southern and Eastern countries than in the two other subgroups. There are, however, some countries (Austria, France, Denmark, Finland, the Czech Republic, and Estonia) in which polarization decreases and others in which it increases (Ireland, UK, Spain, the Netherlands, Sweden, and Hungary). There is no clear pattern here, but this is also so for populist attitudes. We now turn to the main part of this article, namely the relation between those two sets of variables.

3. Polarization and Populist Attitudes

To compute the immigration stock, we rely on detailed cross-country migration data from the United Nations Population Division (UN 2020). For each of the 25 European countries, we compute the net immigration stock of people who were born abroad. Given that the information is available every 5 years only, we interpolated the immigration stock for missing years. We use the ratio between the immigration stock and total population (WDI 2020) in our estimations. GDP per capita and the rate of urbanization are also taken from WDI (2020).

The OLS estimates are displayed in Table 5.¹⁵ In this table, the polarization index relies either on the ‘ability to make ends meet’ distances, or on the ‘deprivation index’ distances. In all four regressions (where the left-hand side variables are distrust in institutions, anti-immigration, authoritarianism, and distrust of people), the polarization index picks a positive sign that is significantly statistically different from 0 (in some at the 10 percent level). The increase in immigration stocks picks a significant and positive effect for authoritarianism and distrust of people, only.