Economic correlates of populist attitudes: an analysis of nine european countries in the aftermath of the great recession

This paper analyzes individuals’ adoption of populist attitudes in nine European countries in the wake of the Great Recession. We assess the consequences of three different, interrelated aspects of economic hardship that are expected to foster the development of populist attitudes at the individual level: vulnerability, grievances, and perceptions of the national economic situation. Using comparative survey data, we find effects of all three of these individual aspects. Our analysis suggests that the main explanation for populist attitudes is neither the vulnerability nor the economic hardship suffered by the people, but rather the perceptions that citizens have about the economic situation in their country. Using panel data from Spain, we address concerns about the presence of endogeneity in the relationship between economic perceptions and populism and conclude that the effect goes mostly from economic perceptions to populist attitudes, not the other way around.


Introduction
The last decade in Spain was characterized by two extraordinary circumstances: an economic crisis that has produced rocketing levels of unemployment (up to 26%) and substantively reduced citizens' purchasing power (-17% on average), and the rise of a new left-wing populist party, Podemos. Founded only recently in early 2014, Podemos obtained 21% of the popular vote in the 2015 general elections, becoming the third largest party in the Spanish political landscape. Although Spain has thus far been immune to the pervasive growth of populist radical right parties affecting other European countries, the main political consequence of the Great Recession has been the growth of support for a party whose discourse is structured around the constant opposition between la gente (the people) and la casta (the caste) (Gómez-Reino and Llamazares 2016).
This example, which could be extended to other peripheral countries such as Greece (with Syriza) or Italy (with the Five Star Movement), serves to illustrate the question addressed in this paper: to what extent is populism facilitated by economic hardship? In order to answer, we analyze individuals' adoption of populist attitudes in Europe in the wake of the Great Recession, drawing on individual data from a nine-country study (Study 1) and a two-wave panel survey from Spain (Study 2).
The contribution of this paper is threefold. First, the study focuses on explaining individual variation in populist attitudes, rather than support for populist parties as most previous works have done. Understanding how populist attitudes are forged is fundamental, as these constitute the breeding ground from which electoral support for populist parties may (or may not) spring. While most existing works focus on the electoral support of populist parties, not much comparative empirical research has yet been done on the antecedents of populist attitudes among voters, and the question of the specific role of economic factors remains open. This paper aims to fill this gap by examining the correlates of populist attitudes based on the analysis of comparative survey evidence from nine European countries (N = 18,368), which were affected by the economic crisis in varying degrees. Some have been heavily affected by the crisis with several consecutive years of negative growth (Spain, Greece, Italy), some only mildly affected with one or two years of negative growth (France, Germany, United Kingdom, Sweden, Switzerland), and one mostly unaffected (Poland). Hence, the effect of individual economic hardship is assessed in a variety of economic circumstances, which enhances the external validity of our findings.
Second, the paper assesses the consequences of three different, interrelated aspects of economic hardship that are expected to foster the development of populist attitudes at the individual level: vulnerability, grievances or personal experience with the crisis, and perceptions of the national economic situation. Using comparative survey data, we find effects of these three individual aspects. Our analysis yet suggests that the main predictor for populist attitudes is not the vulnerability or the economic hardship suffered by the people, but rather the perceptions that citizens have about the economic situation.
Third, the paper addresses the potential endogeneity problem that may arise between perceptions of the economic situation and populism. A bad economic situation may enhance populist attitudes, but populist attitudes may also condition how people evaluate the economic context. Using panel survey data from Spain, we conclude that the effect goes mostly from economic perceptions to populist attitudes, rather than the other way around.
The paper is structured as follows. First, we discuss the choice of populist attitudes as a dependent variable within the literature on populism and populist party support. Next, we review the debate over the role of the economy as a source of populist resentment and present our hypotheses. We then describe the data and methods used in the cross-national and longitudinal analyses and present the corresponding results. In the last section, we summarize our findings and discuss the implications thereof.

Populism and populist attitudes
One of the major hurdles in the quest for explaining the rise of populism has certainly been the many, sometimes divergent, and often hardly specified meanings that the concept has been given, not only within the academia but also by the media and among political commentators and politicians themselves. Even if authors still disagree on whether it should be thought of as an ideology, a discourse, a communicational style, or even as an organizational strategy (see Bonikowski and Gidron 2016), a growing consensus is recently emerging around a minimal set of core features that define populism. These have been succinctly conveyed by Mudde (2004, p. 543) when arguing that populism ''considers society to be separated into two relatively homogeneous and antagonistic 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'' (see also Abts and Rummens 2007;Albertazzi and McDonnell 2008;Mény and Surel 2002;Rooduijn 2014;Stanley 2008). Accordingly, populism is conceived of as a Manichean view that sees politics as the struggle between the worthy people's commonsense and the harmful, selfserving power elite-a view that is deeply suspicious of any constitutional restraints to the democratic principle and hence advocates for the absolute primacy of popular sovereignty.
Such a minimal conceptual core renders populism ideologically ubiquitous (Taggart 2000). Lacking any true programmatic content, populism does not provide an internally coherent set of specific solutions to the major conflicts present in modern societies. Instead, populist rhetoric is easily attached to different fullfledged ideologies on both sides of the left-right spectrum. Although the populist radical right has proven to be one of the most prolific party families in Europe over the last three decades, other distinct families have recently earned the same qualifier, such as neo-liberal populist parties or social-populist parties, not to mention the diverse crowd of contemporary leftist-populist movements in Latin America (such as those led by Chávez in Venezuela, Morales in Bolivia, and Correa in Ecuador).
Relatedly, populism is not to be understood as a quality confined to a precise set of allegedly populist parties. Rather, populist rhetoric can be adopted in different degrees by any actor, not only political parties and leaders but also journalists and voters-provided that their discourse complies with the minimal definition. Also mainstream parties might occasionally or even consistently voice populist appeals (Deegan-Krause and Haughton 2009;Jagers and Walgrave 2007;Pauwels 2011). Populism can vary in degree across actors and over time. It is not an ''either-or'' concept (Pauwels 2011), and hence it is best used as a ''descriptor'' rather than as a ''classifier'' ( van Kessel 2014).
This ideational conceptualization of populism-that is, the understanding of populism as a set of ideas conforming an ideology or a worldview (Mudde and Rovira Kaltwasser 2017)-has significant implications in terms of research strategy. Most of the existing research on the origins of populism as a mass phenomenon uses vote choice as the dependent variable, taking as a starting point some categorization of parties as populist and nonpopulist (e.g., Agerberg 2017;Bakker et al. 2016;Bowler et al. 2017;Dunn 2015;Harteveld et al. 2015;Ivarsflaten 2005Ivarsflaten , 2008Oesch 2008;Pauwels 2014;Ramiro and Gomez 2016;Schumacher and Rooduijn 2013;Van Kessel 2015). Leaving aside the debates on how to classify parties as populist, individual populism in the analysis of vote choice is equated with support for populist parties, which in addition adopt very diverse ideological stances that further condition their electoral support.
Using populist attitudes as the dependent variable, as compared to support for populist parties, helps alleviate at least some of the methodological barriers associated with the study of the breeding ground of populism. First, it allows populism to vary across individuals regardless of their vote, thus avoiding an artificial dichotomization that overlooks differences of degree in the levels of populism in both the discourse of parties and the attitudes of their voters. Second, it helps discern the determinants of populism from those of the other (and often radical) ideological stances that individual populist movements happen to embrace-that is, to ''separate populism from features that might regularly occur together with it, but are not part of it'' (Mudde and Rovira Kaltwasser 2012, p. 2). Finally, it allows us to better focus on demand-side factors-such as the degree to which people are in a vulnerable economic position, have experienced economic hardship, or how they perceive of the country's economy-rather than on explaining the electoral performance of specific individual parties or party families, a question for which supply-side factors-such as the characteristics of the parties and institutional constraints-have been shown to be overwhelmingly relevant (Mudde 2007; van Kessel 2013).

Economic sources of populist resentment
The role of the crisis as a trigger of populist upsurge figures prominently in a number of works inquiring into the nature and origins of populism. In Taggart's (2004, p. 275) words, ''populism is a reaction to a sense of extreme crisis'' that ''spills over into a critique of politics and into the sense that politics as usual cannot deal with the unusual conditions of crisis.'' According to Laclau (2005, pp. 37-38), populism is the result of ''a situation in which a plurality of unsatisfied demands and an increasing inability of the institutional system to absorb them differentially coexist.'' In a similar vein, Panizza (2005, p. 11) argues that populism typically emerges out of critical circumstances that produce ''a breakdown of social order and the loss of confidence in the political system's ability to restore it.'' The rationale behind such theoretical accounts is thus that the perception of persistent unresponsiveness to popular demands undermines the public's confidence in the political establishment to the point that it calls into question the whole institutional system's capacity to satisfactorily handle the situation.
Large-scale economic crises, such as the Great Recession and the sovereign-debt crisis recently experienced by several Eurozone member states, clearly provide the conditions for feelings of dissatisfaction and a perception of unresponsiveness of the political elite to spread among citizens. It is worth noting, however, that populist upsurge is not inevitably restricted to times of crisis and structural transformation, as most strongly argued in studies of the Latin American experience (de la Torre 2000; Knight 1998). Populist attitudes must be conceived as the result of the interplay of various factors at multiple levels-the economy being but one of them.
Even if economic hardship is clearly not a necessary or even a sufficient condition for the emergence of populism, populist attitudes may arguably be nourished by economic difficulties. That voters turn against governments in times of economic strain has been long established by the economic voting literature (Lewis-Beck and Stegmaier 2000). While the economy plays an important role in ordinary times, its impact has been found to intensify in times of crisis (Lewis-Beck and Lobo 2017). Further, continued poor economic performance, often spanning across different governments and incumbent parties and/or concerning decisions taken by previous governments and parties, may end up damaging the public's confidence in the entire political establishment and its ability or even its willingness to redress the situation (see, e.g., Stimson 2004). In line with this argument, recent research has found that the electoral implications of the Great Recession in Western Europe go well beyond the incumbent; such punishment shifts to mainstream parties in general, while populist parties appear to benefit from economic failure (Hernández and Kriesi 2016). Although disentangling the conditions under which economic crisis produces dissatisfaction with the whole system beyond the government goes beyond the scope of this paper, we are interested in disentangling the microfoundations of these macroprocesses. To the extent that crises negatively affect the living conditions and perceptions of citizens, economic crisis may breed dissatisfaction with the elites who are seen as being responsible for governing the affairs of the country and ultimately enhances the perceived antagonism between the ''people'' and the ''elite,'' however defined.
Indeed, this is one of the working hypotheses guiding Kriesi and Pappas' (2015) efforts to examine the impact of the financial crisis on the performance of populist parties in Europe. They find that electoral support for populist formations experienced a moderate but nonnegligible increase during the Great Recession. The evolution of populism was far from uniform in the countries under consideration, the growth being particularly strong in Southern and Central-Eastern Europe and almost nonexistent in Western Europe. Their analyses provide partial empirical support for the hypothesis that populism benefits from economic crisis, as attested by the fact that populist formations tended to perform better in countries more seriously affected by the global economic downturn-yet with remarkable exceptions to the general trend (Pappas and Kriesi 2015). The crisis-breedspopulism thesis has gained additional, albeit indirect, empirical backing from other recent work examining the influence of the Great Recession on related but distinct attitudes, such as satisfaction with democracy and trust in political institutions, that experienced dramatic declines over the last decade (Armingeon and Guthmann 2014).
Overall, however, extant empirical work concerning the relationship between economic hardship and mass populism remains rather unsystematic, is hardly comprehensive, and shows markedly mixed results. A case in point is the effect of unemployment on support for the populist radical right. Whereas several studies have found a positive relationship between unemployment rates and populist electoral performance (Anderson 1996;Arzheimer 2009;Givens 2005;Jackman and Volpert 1996), some other macrolevel studies show no significant association Swank and Betz 2003) or even a weak negative correlation (Arzheimer and Carter 2006;Knigge 1998).
The apparent contradiction raised by the negative associations found in some studies has been explained by virtue of bad economic conditions lending greater salience to socioeconomic issues, which have been traditionally owned by mainstream parties, over immigration and other debates wherein populist radical right parties play a more visible role (Mudde 2007;Rydgren 2007;Bornschier 2010). Accordingly, the combination of issue salience and issue-specific perceptions of party competence would offset or even reverse the influence of economic hardship on citizens' adherence to populist discourse.
Another alternative approach criticizes the conventional understanding on the basis of its tendency to conceive crisis as a purely exogenous factor. To the extent that crises are socially constructed, populism may become a trigger of crisis rather than crisis being a precondition of populism, as the crisis discourse is a key instrument for populist leaders to convey to the public the Manichean worldview that sees politics as an antagonism between the people and the elite (Moffitt 2015). It is in the interests of populist challengers to fuel the perception of crisis regardless of actual conditions, since their appeal stems from their self-proclaimed ability to fix that very problem. Indeed, the denunciation of a crisis has been found to be one of the most recurrent themes of populist rhetoric (Rooduijn 2014).
In our view, this interpretation qualifies but does not detract from the analysis of the influence of crisis on populism, once the definition of crisis has been expanded beyond objective indicators to include subjective perceptions and such perceptions are acknowledged to be affected by factors other than actual conditions. Crises may be real or imaginary, but the sense of threat and emergency they give rise to may be vividly perceived by a significant portion of the people (Taggart 2004), and its effects merit inquiry. If a major concern when explaining populist party support, the distinction becomes less of a problem when the focus is on populist attitudes, and particularly when the analysis is set in the aftermath of the unquestionably critical context of the Great Recession.
The aim of this paper is to disentangle to what extent different dimensions of economic hardship enhance populist attitudes at the individual level. For this purpose, we distinguish three facets of economic hardship: (1) vulnerability, (2) grievances, and (3) evaluation of the national economic situation. The first concerns the likelihood that an individual is affected by the crisis, by virtue of her position in the social structure and her membership in given social categories. The second relates to the retrospective deterioration of the objective economic conditions of the individual during, and likely as a direct consequence of, the crisis. The third dimension involves the individual's subjective perceptions of the country's economy. 1 Our expectations regarding the role of a person's vulnerability are chiefly informed by the ''economic interests thesis'' as first applied to the radical right to account for electoral support coming from ''the losers in the competition over scarce resources and/or those who suffered from some form of relative deprivation'' (Eatwell 2003, p. 53). These typically emphasize the impact of large scale socioeconomic changes and the fears and uncertainties these bring about (Betz 1994;Kitschelt 1995;Kriesi et al. 2008;see Mudde 2010). Major changes such as globalization, massive immigration flows, cultural diversity, or European integration have been interpreted to be giving rise to a new conflict opposing winners and losers of modernization, the latter typically comprising groups with lower socioeconomic status (Kriesi et al. 2006(Kriesi et al. , 2008. According to Kriesi (2014), the Great Recession only would have exacerbated the emerging cleavage, further fueling the populist radical right, while likely leading to the breakthrough of more overtly class-based, left-wing populist movements.
The influence of crisis through an individual's socioeconomic position does not only derive from her economic disadvantage but also, and perhaps more importantly, from the threat of deprivation. The threat might not eventually be realized but is nevertheless consequential, via the feelings of fear and insecurity it brings about. Indeed, empirical research has quite consistently found that populist parties, and the radical right in particular, draw disproportionate support from persons with lower income, lower education, and lower occupational status .
Consequently, we expect that: Hypothesis 1 The more vulnerable the socioeconomic position of the individual (lower education, manual occupation, lower income, unemployed), the higher the level of populist attitudes On the other hand, individuals may vary in the extent to which they have suffered economic grievances, i.e., a deterioration of their economic and working conditions. This retrospective assessment captures the depth and pace of the consequences of the economic crisis among those actually affected by it. Material strain is typically experienced as the inability to keep up with once affordable payments, the reduction in consumption of basic goods and services, and the loss of a job or worsening working conditions.

Hypothesis 2
The more an individual suffers from grievances derived from the crisis (having to reduce consumption, having suffered deteriorating working conditions), the higher the level of populist attitudes Next, we expect any effect the personal economy may have on populism to be largely mediated and outweighed by group-based judgments of economic conditions. A large body of economic voting literature has confirmed the prevalence of national concerns over personal, or self-interested, considerations (Kinder and Kiewiet 1981). People-centrism is one of the necessary components of the aforementioned definition of populism. If populism actually is the product of a ''plurality of unsatisfied demands,'' it is the perception of grievances shared by the community identified as the people, rather than individual economic hardship, which should play a paramount role in explaining citizens' degree of endorsement of populist attitudes. Indeed, Elchardus and Spruyt (2016) find in their analysis of the Flemish case that an individual's economic vulnerability only indirectly affects populism, via its influence on feelings of relative group deprivation and other groupbased considerations. Furthermore, as just noted above, crises may be real or artificially constructed by populist leaders, but it is the subjective perceptions of crisis, rather than the picture conveyed by standard macroeconomic figures, that ultimately fuel populism.
Hypothesis 3 The more negative the perceptions of national economic conditions, the higher the level of populist attitudes.
Hypothesis 4 The effect of perceptions of national economic conditions over populist attitudes will exceed that of vulnerability and grievances.
Finally, we explore the direction of causality between populist attitudes and perceptions of the national economy. As noted, it has been argued that economic evaluations may be endogenous to support for populism, which would challenge the effect assumed in two preceding hypotheses. Contrarily, we expect perceptions of the economy to shape populist attitudes, rather than being shaped by these.
Hypothesis 5 Perceptions of the country's economy predict changes in populist attitudes, rather than vice versa.
This study utilizes two datasets to test these hypotheses. We first examine comparative cross-sectional data (Study 1) in order to test hypotheses 1-4 at a correlational level. We then test hypothesis 5 using panel survey data from Spain (Study 2), in order to address legitimate concerns about the direction of causality in the relationship between populist attitudes and perceptions of the national economy.

Study 1: cross-national evidence Data and measures
The cross-national analysis is based on the Livewhat survey, which was conducted online in June of 2015 in nine European countries: France, Germany, Greece, Italy, Poland, Spain, Sweden, Switzerland, and the United Kingdom. This selection of countries provides fair coverage of European diversity in terms of effects of the crisis, political cultures, and range and performance of populist parties. The samples, recruited by YouGov using the methodologies available in each country, are quota balanced in order to match national population statistics in terms of sex, age, and education level. This cross-country study allows us to examine political attitudes as a function of individuals' position of vulnerability, grievances due to the economic crisis, and perceptions of the state of the national economy in a variety of contexts.
Respondents' populist attitudes were measured using the scale proposed by Akkerman et al. (2014), which includes the following six statements 2 : 1. The politicians in [country] need to follow the will of the people 2. The people, and not politicians, should make our most important policy decisions 3. The political differences between the elite and the people are larger than the differences among the people 4. I would rather be represented by a citizen than by a specialized politician 5. Elected officials talk too much and take too little action 6. What people call ''compromise'' in politics is really just selling out on one's principles Respondents' agreement with each of the statements was measured using a fivepoint Likert scale, from strongly disagree to strongly agree (see the Appendix for details on question wording and coding). The internal consistency of the resulting composite scale (mean of scores) is satisfactory for the whole sample, with a Cronbach's alpha of 0.83, and across all countries in the survey, with alphas varying between 0.77 (Greece) and 0.87 (France).
The specification of the models takes into account three groups of economic factors available in the survey, along with a number of control variables. We first consider variables related to vulnerability: low education (defined as less than secondary), employment status (unemployed), occupation (manual), and household income (coded in deciles of the national income distribution and adjusted for the size of household using the OECD-modified equivalence scale). These individual characteristics are expected to affect the extent to which people are vulnerable to the consequences of the economic crisis, and hence we expect them to affect their level of populism.
In themselves, however, these variables reflecting vulnerability are not indicators of having necessarily suffered economic hardship as a consequence of the Great Recession. A second group of economic factors concerns individual characteristics that reflect precisely the extent to which people have personally suffered negative economic consequences of the crisis. We used two indicators of these grievances, tapping into their financial and occupational dimensions. First, we include a composite index of ''reduced consumption'' calculated as the number of measures, out of a list of ten, which the respondent or anyone in the respondent's household has had to take for economic or financial reasons during the last 5 years. We also included an index of ''worsening job conditions,'' obtained from an 11-item battery of experiences that denote a deterioration of the respondent's working conditions during the last 5 years.
The third of the economic factors to be considered, perceptions of the national economy, is measured using a single variable. Following the rationale of longestablished survey instruments such as the University of Michigan's Index of Consumer Sentiment, we combined both retrospective and prospective assessments of the state of the economy. Specifically, we averaged respondents' evaluations about the country's economic situation over the last year and their expectations about its likely prospects one year ahead, both measured on an 11-point scale running from ''much worse'' to ''much better.'' All models include a set of standard sociodemographic controls: gender (female), age and its square (to account for the potential curvilinear effect of age), and citizenship (nonnational). We eventually control for a set of additional variables, namely: political knowledge, a measure of closeness to the party or parties in government, self-placement on the left-right scale, and attitudes toward immigrants. These additional controls might be considered as potential factors that influence both perceptions of the economy and populist attitudes; their inclusion thus provides a robustness check of the effects of economic perceptions. But, as discussed in the next section, it also provides valuable insights regarding the formation of populist attitudes among citizens.

Results
Before turning to the results of the analysis, it is worth looking at how our dependent variable, populist attitudes, is distributed in our nine countries. Figure 1 displays boxplots and mean values of the composite index of populism for each of the country samples. All countries show mean values above the neutral position (neither agree nor disagree, reflected by value 2 on the 0-4 scale). 3 Our data thus appear to lend support to the claim that ''a large pool of potential followers always exists'' (Hawkins and Riding 2010, p. 20) which populist movements can draw on.
However, although differences are modest in size, countries might be grouped in two blocks according to their overall level of populist attitudes. Sweden, Switzerland, the United Kingdom, and Germany show relatively lower levels (around 2.6), while Poland, France, Spain, Greece and Italy show higher levels (between 2.8 and 2.9).
We first estimate four models on the pooled dataset, with country-level fixed effects and robust standard errors clustered by country. Our dependent variable, the index of populism, is scaled from 0 to 4, with higher numbers denoting higher levels of populism. All independent variables except age (in years) have been rescaled to run from 0 to 1. The key predictors are included in sequential steps, following the scheme defined above. Thus, Model 1 includes the variables that denote a position of economic vulnerability, Model 2 adds variables related to grievances, Model 3 adds individual perceptions of the country's economic conditions, and Model 4 adds the attitudinal controls. Given the nonnegligible proportion of missing data due to item nonresponse for household income and some of the attitudinal variables, all the models are estimated using full information maximum likelihood (FIML), which employs all the available information to provide a maximum likelihood estimation. 4 Table 1 presents the results.
Among the vulnerability factors, occupation and household income are significantly related to populism throughout the analysis, whereas education and unemployment do not appear to make a difference at any point. Manual workers are slightly more likely to display higher levels of populist attitudes. Yet, it is The proportion of missing data on income is 15%. Missing values are also present on left-right placement (15%), economic perceptions (11%), and attitudes toward immigrants (3%). Results using OLS with list wise deletion of missing data show no substantial deviations from the FIML estimates reported in Table 1. individuals living in less affluent households who tend to be more supportive of populism. The effect of income is notably diminished once the experience of the crisis is controlled for (Model 2), and it shrinks further when we account for perceptions of the national economy (Models 3 and 4). This pattern suggests that the effect of material vulnerability is in large part the result of actually being more affected by the crisis and forming a more negative assessment of the state of the country's economy. As expected, grievances are found to be significantly related to individual levels of populism, particularly through its financial dimension. Respondents reporting having been forced to make sacrifices as a consequence of economic difficulties show more support for populism. There is some evidence that, albeit to a lesser extent, also workers who saw their job conditions worsen during recent years are more likely to support populism. The effects of both these variables seem to be partially mediated by assessments of the country's economy, as shown in Models 3 and 4.
In line with expectations, perceptions of the national economic conditions are by far the strongest predictor of individuals' levels of populism. According to the estimates, people holding the most positive views of the national economy score on average nearly one point lower on the 0-4 populist scale than those with the most negative view. The difference decreases substantially, to 0.73 points, when attitudinal controls are added in Model 4, but this variable still remains as the most decisive factor of support for populism.
We conducted a series of tests to check the robustness of these results. First, we examined alternative specifications of the models outlined above, and in all cases the overall pattern remained unchanged. Among other tests, we explored the influence of perceptions of the personal economic situation, measured in the same fashion as perceptions of the national economy. This variable partially overlaps with our measures of personal experience of the crisis (which are focused on specific events and thus provide a less subjective account) and perceptions of the country's Unstandardized FIML coefficients with cluster-robust standard errors in parentheses. The dependent variable, populist attitudes, is coded from 0 (lowest) to 4 (highest). All independent variables are coded to run from 0 to 1, except for age (in years) ? p \ 0.1, * p \ 0.05, ** p \ 0.01 economy (which we argued should be more directly concerned with populist attitudes). As shown in Table A2 of the online appendix, personal economic perceptions do have a strong impact on populism when included along with vulnerability factors and personal experience of the crisis (Model 1) and indeed appear to carry some of the effects of the latter. However, perceptions of the personal situation cease to be significant once we control for national perceptions (Models 2 and 3) which wholly capture their influence, while income and both reduced consumption and worsened job conditions retain part of theirs.
In addition, we re-estimated the four models in Table 1 on each of the country samples independently, thus allowing the relationships to vary by country. The full results are presented in Tables A3 to A11 of the online appendix. To summarize, Fig. 2 shows the coefficients of the key independent variables as estimated using the specification of the model in which they are first introduced (i.e., Model 1 for vulnerability factors, Model 2 for experience of the crisis, Model 3 for perceptions of the national economy) and as estimated with all controls (Model 4). The close, country-by-country inspection of the results across all models reveals interesting variations and some remarkable regularities. Using the baseline specification, the influence of household income is consistently negative and statistically significant in all countries except Italy and Poland. However, the effect fades as variables in subsequent blocks are introduced, such that it only remains significant in Greece, Sweden, and Switzerland-this appears to confirm that income has an influence on populism mainly through its effects on the experience of the crisis and national economic perceptions. On the other hand, manual workers tend to be somewhat more supportive of populism than those in other types of occupations in all countries except France, Greece, and Italy-a difference that becomes nonsignificant only in the United Kingdom when all other variables are controlled for. Low education is consistently associated with populism in Italy and Sweden, while for Germany and the United Kingdom, its impact vanishes in the full model. In line with the mixed findings of previous studies, unemployment appears to have a modest positive effect only in Greece and Italy, but a negative one in Germany and Poland.
The baseline effects of grievances are also pervasive, if not uniform. These are mainly driven by the financial dimension, with Greece as the clearest exception. Reduced consumption is significantly related to populism in all countries and worsened job conditions is in most of them. Again, the inclusion of economic perceptions captures a substantial proportion of the effects of experiences of the crisis, but in all cases at least one of these two factors remains significant when all controls are included.
The most persistent finding of the analysis, however, is the close association between individual levels of populism and negative perceptions of the country's economy. The coefficient for national economic perceptions is statistically significant across all countries and models, varying between -0.66 (Greece) and -1.14 (Italy) in the baseline model, and between -0.31 (Sweden) and -1.07 (France) in the fully specified model, which shows that the effect is robust to the influence of additional political attitudes.
Even if they are not the central concern of this paper, the results of the control variables deserve some attention. According to the overall estimates in Table 1, gender and nationality appear to be weakly related to populist attitudes, although women and, particularly, nonnationals are consistently less likely to support populism in a few countries (e.g., France and Greece). Age shows a curvilinear effect, the level of populism gradually increasing until it peaks at age 60 and then remaining stable or declining slightly. In the country-wise analysis, the effect of age tends to fade as new predictors are added to the models.
The results for the attitudinal controls show a number of interesting patterns. Political knowledge has a sizeable positive influence on levels of populism in the pooled estimates (see Table 1), an association that also emerges in the national samples, achieving statistical significance in all countries (Tables A3-A11). It is particularly the highly sophisticated citizens that appear to be more critical about elites and more confident about the people's own ability to deal with political issues. This pattern is consistent with the view of the process of cognitive mobilization (of which knowledge may be capturing the attitudinal consequences) as one of the longterm developments explaining why voters in Western democracies have become more receptive to populism over the last few decades (Mudde 2004).
Closeness to the parties in government is closely related to support for populism, such that the closer respondents feel toward the national incumbent the less likely they are to adhere to the populist discourse. Interestingly, populist parties happen to be in the executive in the only two countries where this does not occur (Greece and Switzerland)-although support for the incumbent does play a role in Poland, despite the government also being controlled by populists.
Somewhat surprisingly, left-right orientations appear to be negatively associated with populist attitudes in the pooled model, even if the coefficient is only marginally significant. The country-by-country analyses show that populist attitudes are significantly more prevalent on the left everywhere but in Italy, Poland, and the UK, where the effect of ideological orientation does not achieve statistical significance. That is, despite the predominantly right-wing character of populist parties in Europe, there is no country were populist attitudes are more prevalent among citizens self-placed on the right.
In view of these results, it is even more interesting to find that attitudes toward immigrants-which have been found to be a common predictor of support for the populist radical right in Western Europe (Ivarsflaten 2008)-are consistently related to individuals' levels of populism. Anti-immigrant sentiments are significantly associated with populist attitudes in eight of the nine countries, and most effectively in Sweden, the United Kingdom, and Germany. Spain, where populist radical right parties are irrelevant (Alonso and Kaltwasser 2015), is the only exception to the rule.
The effects of some of the factors identified with economic vulnerability, experience of the crisis, and economic perceptions hold in spite of the strong effects displayed by our control variables, at least in the pooled analysis. However, of all the variables in the analysis, only perceptions of the national economy is significantly related to populist attitudes in each and every country. Furthermore, economic perceptions have a stronger or equally strong effect as any other variable in all countries except Sweden, where attitudes toward immigration clearly outweigh their influence.
Economic mood thus emerges as a robust correlate of populist attitudes in European countries. Cross-sectional analysis, however, does not allow us to assess whether and to what extent negative perceptions of the economy drive support for populism or if, alternatively, it is populism itself that promotes negative economic perceptions. Societal crises may well prompt populist reactions among citizens, but, as noted above, populist rhetoric has been shown to promote perceptions of crisis. In order to test for the reciprocal effects of national economic judgments and populist attitudes, in the next section we turn to longitudinal data from Spain-one of the most recent democracies in Europe to experience a populist upsurge.

Study 2: longitudinal evidence
The discourse articulated by the nascent populist movement in Spain was inexorably linked to the denunciation of the economic crisis and the adoption of austerity policies (Gómez-Reino and Llamazares 2016). If there were a situation of reverse causality in which populism influences perceptions of the economy, this should likely be apparent in the Spanish case. On the other hand, the results of our individual analysis presented in Fig. 2 suggest that Spain does not particularly stand out in the comparative analysis. As far as the relationship between economic perceptions and populist attitudes is concerned, the relationship is similar to that of France, Italy, or the United Kingdom.
One possible strategy to deal with issues of reverse causality is panel data. By gauging the temporal precedence among the variables, survey panel data allow us to better address the issue of causal order. In addition, with multiple repeated measures we can correct for measurement error in the variables of interest by treating these as latent variables.

Data and measures
Our data come from an online panel survey of young and middle-aged Spanish residents (Anduiza et al. 2017). The sample was selected from an online pool set up through active recruitment of potential subjects in commercial online services and websites. Quotas were used to ensure a balanced representation in terms of gender, education, size of municipality, and region.
Specifically, we draw on the waves consecutively conducted in May 2014 and May 2015, the only two for which the required measurements were available, yielding a final sample of 726 individuals who completed the survey in both waves. Hence, our analysis coincidentally focuses on a critical stage in the mobilization of Podemos' support. The first panel wave under consideration was conducted just 4 months after the party was founded and right before the 2014 European Parliament election, where Podemos received nearly 8% of the national vote. At the time when the second wave data was collected, the party had already come out as a major contender, with opinion polls placing it at least third in popular support-but occasionally ahead of the Socialist Party and even the incumbent Popular Party.
Populist attitudes were measured using the same 6-item battery included in the Livewhat study and employing a 7-point Likert scale ranging from ''strongly disagree'' to ''strongly agree.'' We combined three items to measure perceptions of the country's economic situation: an assessment of the current situation (on a 5-point scale from ''very bad'' to ''very good'') and 1-year retrospective and prospective assessments (''worse,'' ''same,'' or ''better''). 5

Results
Taking advantage of the longitudinal nature of the data and the availability of multiple indicators for each of our key variables, we estimate a latent-variable cross-lagged structural equation model to gauge the reciprocal relationship between perceptions of the country's economy and populist attitudes over time (Finkel 1995).
Structural equation modeling uses observed indicators to estimate latent constructs and then estimates relationships between latent variables. In each wave, we estimate the latent construct for respondents' degree of populism using the 6-item battery and the latent construct for respondents' national economic perceptions using our three economic assessments (current, retrospective, and prospective). 6 Then, each of the latent variables is modeled at time t (i.e., 2015) as a function of its lagged score and the lagged score of the other latent variable at time t -1 (2014). Hence, two regression models are estimated in a single step: The autoregressive effects, k 1 and k 2 in Eqs. 1 and 2, reflect the stability in individuals' relative standings on the corresponding variables controlling for the influence of the other variable. The reciprocal cross-lagged effects, b 1 and b 2 , reflect the unique effect of one variable at time t -1 on the other variable at time t, controlling for the latter's autoregressive effect. It is thus the cross-lagged coefficients that are of main concern here, as they indicate the extent to which interindividual differences in one variable predict subsequent changes in the other variable. As far as b 1 is different from zero, it implies that changes in the rank ordering of populist attitudes are related to individual differences in perceptions of the national economy measured one year before, while accounting for initial values of populism. Likewise, the extent to which b 2 differs from zero implies that relative changes in economic assessments can be ascribed to differences in prior levels of populism, net of preexisting differences in economic assessments. Table 2 presents the results of the structural portion of the model. The autoregressive coefficients for populist attitudes and economic perceptions are both significant, but the latter clearly exceeds the former in size. This suggests that, in the context under consideration, inter-individual differences in economic perceptions were actually more stable than were differences in populist attitudes. On the other hand, the estimated cross-lagged effect of economic perceptions on populist attitudes is negative and statistically significant, denoting that previous assessments of the country's economy affected later changes in levels of populism. The negative sign indicates that individuals who held positive views about the economy in 2014 showed a tendency to reduce their relative levels of populism, while those with negative views tended to change theirs upwardly. In contrast, changes in economic perceptions between 2014 and 2015 appear to be unrelated to prior standings in populism, as the coefficient estimate of the cross-lagged effect of populist attitudes on economic perceptions is negligible and not significantly different from zero. The results of the longitudinal analysis thus confirm that, at least for Spanish voters in the period of study, effects flow mainly from perceptions of the national economy to populist attitudes rather than vice versa, which is consistent with the hypothesized causal relationship.

Conclusion
This paper examined the extent to which economic difficulties create a breeding ground for populist movements. We argued that the analysis of populist attitudes helps to overcome some of the methodological hurdles related to the study of populism among citizens while remaining faithful to the intricacies of the ideational definition that has recently attracted a growing academic consensus.
Our analyses have shown that economic hardship matters for populist attitudes but also suggest that most of its impact is driven by subjective perceptions of national conditions. Informed by extant literature on the origins of populist movements and anecdotal accounts of their recent upsurge, we posited three pathways to the influence of the economy on the formation of populist attitudes among individuals. The results of the cross-national analyses provide partial evidence in support of the hypothesis (1) that populist attitudes are more prevalent among citizens in a position of economic vulnerability, as measured by low income and, to a much lesser degree, manual occupations. Low education and unemployment, however, do not appear to have a relevant effect on populism, which renders the interpretation of the influence of vulnerability not entirely consistent with the ''losers of modernization'' explanation. While the null effect of unemployment continues the inconclusive results obtained by previous research, the lack of association between populism and education is more intriguing. Part of the explanation may lie in the fact that educational attainment is linked to elements that influence citizens' populist attitudes in opposite directions. Although the highly educated are more likely to follow prevailing social norms, they are also more likely to display higher levels of internal political efficacy, which is arguably a precondition for people to challenge the established political elite. Indeed, our finding that populism is positively associated with political knowledge is consistent with the political irreverence brought about by the process of cognitive mobilization and research suggesting that less-efficacious citizens are more likely to stick with mainstream parties (Mudde 2007).
We also found support for the hypothesis (2) that populist attitudes are associated with citizens' grievances related to the economic crisis. It is mainly the need to cut the household expenses that appears to be more consequential on this dimension, although worsening job conditions had far-reaching effects in some countries, particularly Greece. That the inclusion of grievances reduces the effect of income indicates that, as one would expect, individuals in a vulnerable economic position were more likely to be hit by the crisis and, as a consequence, were also more likely to show support for populism. Coupled with Hypothesis 1, these results suggest that material deprivation likely played a relevant-albeit limited in size-role in the forging of populist attitudes in European societies in the wake of the Great Recession.
The role of grievances, however, pales in comparison with the paramount influence of perceptions of the national economy (Hypotheses 3 and 4). The effect of these was found to be sizeable and significant for each of the nine countries and remained markedly robust to alternative specifications and to the inclusion of other attitudinal factors. Moreover, the results of the longitudinal analysis seem to rule out the possibility that the observed association is merely the product of perceptions of the economy being shaped by populist attitudes themselves (Hypothesis 5). Rather, we found economic perceptions to predict later changes in levels of populism, which supports the causal precedence of economic perceptions over populist attitudes. While panel data do not solve all challenges to causal inference (Finkel 1995), it is at least reassuring that the results of the cross-lagged model are consistent with the direction of causality assumed in the analysis of the crosssectional data.
Taken together, these findings highlight the importance of economic concerns as a key predictor of populist attitudes in Europe in the aftermath of the crisis. Widespread economic hardship is certainly not a necessary condition for the rise of populism movements, as attested by numerous examples of populism's success during good times. However, our study provides evidence in support of the idea that hardship paves the way for that success by enhancing populist attitudes among deprived citizens and, above all, those holding negative views of the national economy. It has been demonstrated elsewhere that populist attitudes are strongly associated with support for populist parties across a variety of contexts (Anduiza et al. 2016). The electoral performance of populist leaders may ultimately depend on supply-side factors, but we have shown here that economic difficulties, or perceptions thereof, enhance the appeal of populist rhetoric.
Our results thus support the idea that it is not so much an individual's objective economic situation that matters for the development of populist attitudes, but rather the subjective perception that there is indeed a critical economic situation in the country. This is in line with the spirit of populism as an ideology of the people, and as such characterized (at least professedly) by people-centric concerns, rather than by self-interest or restricted in-group motives. Although we found evidence that personal economic hardship matters as well, the overwhelming influence of national perceptions further substantiates populism's depiction as ''a reaction to a societal diagnosis'' (Elchardus and Spruyt 2016, p. 125).
Importantly, the relationship holds not only in countries that have recently gone through serious economic troubles, such as Greece and Spain, but also in other countries, like Switzerland or Poland, which were hardly touched by the crisis. This indicates that, regardless of the actual economic conditions, perceptions are paramount. Even if perceptions cannot be immune to the real economy, they can also be swayed by political discourse, and the case has been made that populist entrepreneurs might be particularly interested in conveying an impression of crisis. In the analysis of panel survey data from Spain, populist attitudes were found to be less stable than economic perceptions, and our estimates suggested that the causal relationship goes mainly in one direction-from perceptions of the country's economy to populist attitudes. However, the fact that we did not find the reverse effect in this specific case does not preclude an influence in a different context, presumably under conditions of economic uncertainty, where perceptions might be more malleable by political discourse. In any case, our results should not be viewed as ruling out the agency of populist politicians by, for example, influencing perceptions about the responsibility of the established elite for the economic circumstances (Vasilopoulou et al. 2014). The role of populist actors does not necessarily lie in their ability to shape citizens' perceptions of the country's conditions; the economy itself can provide a favorable discursive opportunity for them to advance their support. Disentangling the interplay between perceptions of the economy and populist predispositions requires future research examining the use of populist rhetoric by political parties and social movements across economic circumstances while taking political and discursive opportunity structures into account.

Appendix: Question wording
Cross-national survey

Populism
The index of populism is the average score of the 5-point agreement scales of the following statements: (1) The politicians in [country] need to follow the will of the people; (2) The people, and not politicians, should make our most important policy decisions; (3) The political differences between the elite and the people are larger than the differences among the people; (4) I would rather be represented by a citizen than by a specialized politician; (5) Elected officials talk too much and take too little action; (6) What people call ''compromise'' in politics is really just selling out on one's principles.

Occupation
''Please tell us which one of the following options best describes the sort of paid work you do. If you are not in paid work now, please tell us what you did in your last paid employment.'' Respondents answering ''Skilled Manual Work (e.g. plumber, electrician, fitter)'' or ''Semi-Skilled or Unskilled Manual Work (e.g. machine operator, assembler, postman, waitress, cleaner, laborer, driver, barworker, call-center worker)'' are coded as manual workers. The occupation of the household's chief income earner was used when the respondent's was not available.

Household income
''What is your household's MONTHLY income, after tax and compulsory deductions, from all sources? If you don't know the exact figure, please give your best estimate.'' Coded in deciles of the income distribution in the given country, and adjusted for the size of the household using the OECD-modified equivalence scale.

Reduced consumption
''In the past 5 years, have you or anyone else in your household had to take any of the following measures for financial/economic reasons?'' The listed items were (1) reduced consumption of staple foods; (2) reduced recreational activities; (3) reduced use of own car; (4) delayed payments on utilities; (5) moved home; (6) delayed or defaulted on a loan installment; (7) sold an asset; (8) cut TV/phone/internet service; (9) did not go on holiday; (10) reduced or postponed buying medicines/visiting the doctor. Additive index (alpha = 0.86).

Worsened job conditions
''Please select those of the following has happened to you in the last 5 years.'' The listed items were (1) I took a reduction in pay; (2) I had to take a job for which I was overqualified; (3) I had to work extra unpaid overtime hours; (4) I had to work shorter hours; (5) I had to take or look for an additional job; (6) My work load increased; (7) The working environment deteriorated; (8) I had less security in her job; (9) I had to accept less convenient working hours; (10) Employees were dismissed in the organization for which I work; (11) I was forced to take undeclared payments. These questions were asked only to those that are employed or have been in the past; all other respondents are assigned the lowest value in the resulting scale. Additive index (alpha = 0.83). from other countries? Please state your answer on this scale where 0 means 'Bad' and 10 means 'Good.''' (ii) Would you say that the [respondent's country]'s cultural life is generally undermined or enriched by people coming to live here from other countries? Please state your answer on this scale where 0 means 'Undermined' and 10 means 'Enriched.' Both measured on 11-point scales. Additive index (alpha = 0.86).

Spanish panel survey
Populism Agreement (using seven-point scales) with the same six statements used in the cross-national survey.

Perceptions of the national economic situation
Three questions load on this latent dimension: (i) ''Referring to the general economic situation in Spain, would you say it is…?'' (1) Very good; (2) Good; (3) Neither good nor bad; (4) Bad; (5) Very bad. (ii) ''Would you say that the current economic situation of the country is better, the same, or worse than 1 year ago?'' (1) Better; (2) Same; (3) Worse. (iii) ''Would you say that over the next year the economic situation of the country will be better, the same, or worse than it is now'' (1) Better; (2) Same; (3) Worse. Responses are recoded to run from negative to positive assessments.