Discontent in the ‘peripheries’: an investigation of the rise of populism in Italy

ABSTRACT The concept of periphery, going beyond the mere interpretation in geographical terms, incorporates a ‘relational’ character that is implicitly characterized by connotations of power and inequality. Peripheries can be actively created as an outcome of shifts in economic and political decision-making at various scales. The recent populist wave has brought to the fore the issue of peripheries as ‘left behind’ places, striking back in the ballot boxes. In this paper, we investigate the role of different dimensions of peripherality and their changing geographies on populist voting patterns by analysing the growth of non-traditional parties between the general elections held in Italy in 2013 and 2018. First, we consider a spatial dimension of peripherality, that is, the geography of access to services of general interest. Second, we explore some factors that might be associated with a condition of peripherality, meant as marginality. We find that the growth of discontent in Italy has different explanatory factors in core and peripheral areas. Irrespective of geographical peripherality, where, however, populist votes are more concentrated, our findings reveal that a condition of marginality may feed a sense of revenge connected to the feeling of ‘not mattering’ also in urban areas.


INTRODUCTION
The concept of periphery is not an absolute one. It is well acknowledged that peripherality goes beyond the mere interpretation of geographical distance from a centre and of location on the fringes of a region. It incorporates a relative dimension, depending on the centre being considered and the phenomenon under scrutiny. In most instances, it is associated with remoteness or 'edgeness' (De Souza, 2018;Herrschel, 2012;Pezzi & Urso, 2016, 2017, hence with distance decay from a presumed core. Attempts to conceptualize and operationalize peripherality have been made in regional economics and economic geography over many decades. The dichotomy core versus periphery is present (either explicitly or implicitly) in classical models of economic growth and land use (Crone, 2012). However, the notion of periphery has recently broadened up to better account for its multifaceted nature.
The 'relational' character of core and periphery is implicitly characterized by connotations of power and/or inequality (Crone, 2012). Peripheral areas are not linked to fixed geographical is thus crucial to delve into the determinants of populist voting to better understand whether, and how, it is connected to the perception of being 'peripheral'.
In this paper, we analyse the rise in populism in Italy and the specific factors associated with the geography of peripherality. We do that attempting to go beyond the urban-rural analytical frame which has dominated the academic and public discourse on the topic by adopting the perspective (and in some way operationalizing the reflections developed) by Yiftachel and Rokem (2021) on the double polarization as an important political-geographical dimension of right-wing political populism (in the form of neonationalism). According to the authors, horizontal polarization is conceptualized along a 'metropolitan-non-metropolitan' spectrum in which the latter is also partially urbanized, disrupting the traditional urban-rural dichotomy. Besides (and interacting with) these horizontal tensions, the two scholars acknowledge the relevance of what they call 'vertical polarization', reflecting stratifications and inequalities within the cities.
We start from exploring voting patterns of the national senate elections held in 2013 and 2018 in the light of a peripherality gradientfrom urban cores to ultra-peripheral areasas expression of the geography of access to services of general interest (SGIs). We think this well serves the purpose of our research. In fact, as stressed by some authors (Essletzbichler et al., 2018;Martin et al., 2018;Rodríguez-Pose, 2020), rising depopulation and a consequent loss in quantity and quality of basic services has caused certain rural or remote areas to become demographic semi-deserts, deprived of public and private facilities. According to some interpretations, residents are reacting to this reality at the ballot box via populist voting. Shrinking regions experiencing limited access to essential services, public transport options dwindling, schools closing down and health services delocalizing could have been pushed to shake the system so as to make their voices heard (Rodríguez-Pose, 2020).
In order to have a widely accepted defining element for this dimension of peripherality, we exploit the classification produced within the Italian National Strategy for Inner Areas (SNAI) which groups all municipalities based on the distance to essential services. We then consider other individual, social and contextual dimensions that proxy a condition of peripherality (or marginality/exclusion) which does not pertain only to rural or, more generally, non-core spaces, but may operate within urban spaces as well (i.e., living in degraded areas/neighbourhoods, deprivation, overcrowding). This goes in the direction of the call made by Lizotte (2019, p. 140) of keeping on working towards 'a geographically-grounded approach to populism', which would make us ultimately understand which material and symbolic grievances populist politics mobilize.
Our results show that the rise in populist votes experienced in between the two electoral rounds is driven not only by individual economic conditions, such as income and employment status, but also by structural factors such as access to public services, network bandwidth and overcrowded dwellings. Importantly, when analysing the geography of voting patterns, we observe a large heterogeneity across geographical areas in terms of drivers of discontent, which was larger in remote municipalities far from service provisions centres. Therefore, geographical peripherality is a key factor in explaining the growth of the recent Italian discontent. However, in another periphery, within cores, or more precisely in the immediate periphery of cities, we found different factors fuelling resentment: some dimensions of social or material vulnerability, likely to be related to a condition of feeling 'peripheral', are strong predictors of populist voting.
The remainder of the paper is structured as follows. Section 2 reviews the relevant scientific literature on the debate on the rise of populism. Section 3 briefly outlines the Italian political framework and the entry of the two populist parties in the national electoral arena. Section 4 describes the data and methods used in the empirical analysis. Section 5 presents and discusses the results. Section 6 concludes.

A REVIEW OF THE EXPLANATIONS FOR POPULIST VOTING
Against the backdrop of a growing pressure towards globalization and of development policies mainly aiming at sustaining agglomeration economies, the challenge of peripherality has been a recurring theme in academic and political debates, where concerns about 'inequalities and the implication for the legitimacy of democratic governments' (Herrschel, 2012, p. 40), at different scales, were raised. Scholars and decision-makers have started to anticipate that such unevenness in policy attention and scope and, by consequence, in opportunities, might have provoked, at some point, a potential response from those feeling marginalized by living in the peripheries (Rodrik, 2018). Herrschel (2012), for instance, highlights that different strategies may be put in place by people required to overcome exclusion from, and marginality to, networks and wonders about the repercussions of any unevenness in relevance and representativeness (in the sense of 'having a say') on the degree of acceptance of, and support for, policymakers. He calls for further research to gain a better understanding of response strategies, mechanisms, and roles of peripheral areas. Rodrik (2018) argues that the fact that advanced stages of economic globalization would produce a political backlash was predictable, although the specific form it took was probably not. Hence, populism has not come out of the blue, but has been hiding in the background for a while.
Recent upheavals in the geography of voting patterns in Europe and United States have brought to the fore the longstanding issue of peripheries as places 'left behind' by globalization and by city-first policies. The diffusion and magnitude of the phenomenon call for a deeper investigation of its determinants and in a way, by implication, require revisiting existing concepts and presumptions of peripherality based on mere distance and on clear-cut categories such as remoteness and centrality. The literature on the geographical roots of populism in Europe (Dijkstra et al., 2020;McCann, 2020;Rodríguez-Pose, 2018;Rodríguez-Pose et al., 2020), though recent, is rich and continuously expanding, and it has not come yet to an overarching shared explanation.
Following one of the latest reviews of the academic knowledge hitherto produced on the topic (Rodríguez-Pose, 2020), three main groups of explanations have been provided for the rise of populism.
The first, so far dominant, line is the one that moves along the cleavage between cultural and economic explanations. On the one side, increasing multiculturalism, cosmopolitanism and globalism are alienating a part of our societies who do not recognize themselves with the new, hegemonic cultural values, 'pitching the elderly against the young, as well as cities against small towns and rural areas' (Rodríguez-Pose, 2020, p. 5;Rodden, 2019). On the other side, globalization and its pervasive effects (Figus et al., 2018) on the labour market and the related widening of interpersonal inequalities have rendered more and more individuals economically vulnerable, eliciting their resentment.
The third group of studies interprets the phenomenon of populism through the dualism individual versus territorial drivers of the vote (Crescenzi et al., 2018). The factors identified at the individual level include age, education, occupation, income, personality traits and political views (Goodwin & Heath, 2016;Los et al., 2017). The ones identified at the regional level include demography, exposure to trade shocks, immigration, austerity policies and cultural values (Abreu & Öner, 2020), which are often associated with socio-spatial patterns (e.g., the urbanrural divide).
The study at hand mostly engages with, and contributes to, the literature falling into the comparatively less investigated second and third streams of research. Within them, it especially exploits the elements that might be related to some dimensions or dynamics of peripherality in its broader sense. It innovates by advancing knowledge on less analysed aspects of peripherality, as conceptualized above, which could have a role in explaining the populist vote. If this has been mostly investigated so far in its spatial dimension, often through the lenses of the urban-rural dichotomy (Mitsch et al., 2021), our analysis aims to make the most of a more comprehensive understanding which goes beyond the mere 'geodesic' interpretation of the notion, which enables us to capture peripherality in cores as well. We think, in fact, that this notion, which allows keeping together both individual and territorial factors, might be empirically tested providing strong explicative power and overcoming the analytical cleavage along which most studies have moved.
In our analysis of the rise of populism in Italy, we therefore operationalize peripherality accounting for the two (horizontal and vertical) interrelated axes of polarization of the phenomenon (Yiftachel & Rokem, 2021). This is proxied by the difference in the share of votes obtained by the two populist parties of the country, that is, Lega and Movimento Cinque Stelle (M5S) in 2018 and 2013 political elections at the municipal level. Following Goetz et al. (2019), Agnew and Shin (2019) and Rodríguez-Pose et al. (2020), we use the difference in share instead of overall share of votes as we assume that this margin better signifies the increase in populist vote between the two national elections.
We look at peripherality from two perspectives. First, besides using some common individual variables, we explore voting patterns considering an 'objective' distance-defined peripherality (De Souza, 2018), using a distance-decay, time-related official indicator: the SNAI classification (see below). This allows us to answer the call, made by Essletzbichler et al. (2018), for further, detailed investigation on the effect of the interaction between the supply of and demand for public services on the populist vote. To better account for this aspect, we also include specific additional variables: internet capacity, the number of hospitals and high schools and the use of public transport for study-and work-related commuting. Second, we consider some factors that might be associated with a condition of vulnerability and/or exclusion. To this purpose, among salient elements such as lack of employment opportunities, female participation in the labour market, low-skilled employment rate and foreign population, we explore the effect of some less scrutinized elements within the faster and faster growing literature on populism, mainly related to housing conditions and neighbourhood environment. We thus include indicators for physical decay of buildings and adequacy of services available in dwellings, presence of non-compliant dwellings and household crowding. Through this set of variables, which account for a dimension of social and material vulnerability, we intend to ascertain whether the housing and neighbourhood spheres play a role in the feeling of being left behind or abandoned, going beyond the mere urban-rural dichotomy perspective.

ITALIAN POPULISMS
Two 'shades' of populism have emerged in Italy in recent times: the anti-EU and anti-immigration versionembodied by Legaand the anti-establishment onerepresented by M5S. In line with a general trend which can be observed in Europe and beyond (e.g., the United States), between 2013 and 2018, two Italian populist parties, jointly considered, have seen an increase in their votes of 80%. Moreover they ended up governing the country together following the 2018 election, with 17% and 32% of total preferences, respectively.
Until the early 1990s, the Italian party system has been characterized by stable geographical differences since the end of the Second World War and across the so-called First Republic , with the two main parties (the Christian Democrats and the Communist Party) dominating in different macro-regions (Mancosu & Vezzoni, 2018). After that time, as the old political system collapsed leading to a massive party replacement, the electoral geography of the country remained quite stable. The new centre-left parties which came out from the Communist Party (the PDS/DS and the smaller ally Communist Refoundation) stably established themselves in the centre of the country. On the other side, new right-wing parties -Forza Italia, Alleanza Nazionale and the autonomist party Lega Nordwere embedded in the areas traditionally supporting the Christian Democrats (Mancosu & Vezzoni, 2018).
That is why it is intriguing to look at the new geographies brought about by the entry of a brand new self-declared 'non-party' movement, M5S, not historically building on a political legacy, coupled by the simultaneous progressive strengthening of the deeply politically embedded right-wing party, Lega, and the enlargement of its local electoral base. This is interesting also because any political shift might have implication for peripheries, as mentioned in the introduction with reference to their active production through policy. As sharply pointed out by Mancosu and Vezzoni (2018, p. 6), in fact, 'parties … can be seen as institutions that connect the state with civil society groups committed, for instance, to channelling economic resources from the centre to the periphery (usually, at expenses of other peripheries)'.
The 'Lega Nord' ('Northern League', the original name of Lega) was born in the late 1980s: the party's electoral breakthrough came in the wake of the political and economic crisis that upset Italy in the late 1980s and the beginning of the 1990s. Starting as a regionalist movement, Lega's populism has broadened its appeal in an attempt to become a national party. The importance of territorial identity has undergone tactical shifts, while the emphasis on opposition to the political system, to immigration and to globalization (with an 'Italians first' message attached) has endured (Passarelli, 2015).
The 2013 general elections saw the emergence of M5S and have introduced a third-party system (Mershon, 2015). The movement earned more than a quarter of the preferences (25.6%) and was the most voted national party. The issues proposed by M5S revolve around claims of 'common' people subjugated by the elite, the banks, the politicians/government and the technocracy by and large (Passarelli, 2015). M5S is an openly non-ideological party and the opposition between the people and the corrupt elites does not move along ideological lines (Manucci & Amsler, 2018).
We agree with Bloise et al. (2021) that Lega and M5Sas is also clear from their histories and valuesare very different in nature, which implies that specific drivers are at play in their underlying voting behaviours. Lega has had a long sedimented electoral constituency in the peripheral parts of the far North of Italy, partly because it appealed to anti-southern sentiments. M5S seemed to gain much support in the peripheral fringes of large cities across Italy and in the South. Faggian et al. (2021) find that the spatial distribution of their votes is associated with territorial socio-economic and institutional differences (other than with the individual characteristics of their electoral constituencies), with Lega capitalizing its traditional support in the North, while M5S diffusing in the South. Beyond these expected regional patterns, the study also highlights the existence of an urban-rural contrast in the case of Lega, which calls for a deeper investigation.
Our aim is to look at the driving factors (including 'socio-economic' peripherality) explaining populist voting in different spatial contexts (moving along the peripherality gradient) and across the two kinds of populisms, the nationalist/protectionist and the anti-system one. This exploratory study stems from the reflections developed within the debate on left-behind places interpreted as (variously meant) peripheral territories that do not matter and that, for this, strike back in the ballot boxes. Our hypothesis is that living in peripheral areas/contexts or living in conditions of peripherality might imply the role of specific drivers in explaining the resentment expressed through the protest vote.
This in fact seems to be a missing gap in the literature of the field, especially in the Italian one, as stressed by Giovannini and Vampa (2020) in their study on the roots and outcomes of autonomy referendums in the country. They observe a significant internal variation of regions in their territorial mobilization of autonomy requests, which seems to replicate a 'centre-periphery' cleavage, so far often neglected in the academic debate, according to the authors. And yet, della Porta et al. (2021), in their insightful lead essay introducing a recent special issue on inequalities, territorial politics and nationalism, claim that the crises in capitalism, the economy and the state have contributed to reopen ancient (Rokkanian) cleavages that seemed to have attenuated (e.g., class divide) and to produce increasing fractures along the urban versus rural as well as the centre-periphery lines, which deserve more academic attention. To overcome a widespread interpretation of 'regions as political "monoliths"' (Giovannini & Vampa, 2020, p. 16), scholars call for a more nuanced approach, accounting for tensions within them. This is further evidence that centre versus periphery dichotomies are potential extremely meaningful analytical lenses for several territorial issues, including politics or political attitudes, which are instead often overarched by macro-regional narratives (e.g., North-South divide in Italy). The increase in spatial unevenness in spite of cohesion policies for territorial equity, given the wide-ranging effects this may have in terms of 'social and territorial conflicts, changes in electoral behaviour, a reshaping of political systems' (della Porta et al., 2021, p. 326), has strongly brought back the reflection that inequalities and discontent need fine-grained spatial analyses to well inform policies (della Porta et al., 2021). This is particularly urgent for Italy, where, as claimed by Asso (2021), regional inequalities, which have been further exacerbated by the Great Recession, are more substantial, persistent and spatially polarized (along the North/South dualism and within macro-regions) than in any other European country and have brought about relevant political changes with an unprecedented majority of political seats conquered by populist parties. Precisely this political impact is investigated in Bloise et al. (2021). The paper explores the role of inequality and other economic conditions in Italy's general elections from 1994 to 2018 at the regional level. More specifically, the study examines 'protest votes', that is, the shares of non-voters, of votes for Lega and for the M5S, which the authors define as 'challenger' parties. The results of quantitative analyses show that inequality, lack of wealth and precarization are closely associated with the regional patterns of Italy's electoral change, with some specificities. Namely, higher abstention is associated with greater income inequality; the vote for Lega is connected to the downward pressure on the income of the middle classes in northern and central regions; support for the M5S is driven by youth's precarious employment and poverty of the South of the country. Inequality and economic conditions emerge therefore as powerful explanatory factors of non-voting or voting for non-traditional parties.
Both these recent studies on Italy are performed at the macro-regional and regional levels. Though adding a highly valuable contribution to these topics, as pointed out by many, within-region dynamics need to be addressed academically and policy-wise. And this is precisely what our paper aims to contribute to with an analysis at the most granular possible scale: shedding light on the role of peripheralityboth meant as purely geographical and in terms of marginality (beyond the consideration of economic condition only)on the recent transformations in the constructions of collective identifications and interests and in the related territorial bases of Italian politics.

DATA AND METHOD
Following the theoretical discussion in section 2 and our research objectives, our chosen variables capture socio-economic territorial characteristics, as well as different dimensions of peripherality.

Data
Our analysis combines administrative electoral outcomes, income and census data at the municipal level. This section describes the datasets and the method. For additional details and data sources, see Table A1 in the supplemental data online.

Electoral outcomes
We build our outcome variables by collecting the total valid votes jointly ('Growth of discontent') and separately obtained in each municipality by the two populist parties, Lega and M5S in the national senate elections held on 24-25 February 2013 and 4 March 2018. We calculate the percentage growth rate between 2013 and 2018. 1 This measure provides a direct interpretation of the dynamic of the two populist parties across the two electoral rounds at a granular geographical level. Summary statistics for the outcome variables reported in Table 1 show a substantial growth of populist parties: on average, between 2013 and 2018, jointly considered, they increased the number of votes by approximately 80%.
The high standard deviation signals a largely heterogenous growth pattern. Preliminary evidence shows that most of this heterogeneity emerges when clustering municipalities along a core-periphery line according to SNAI classification. Figure 1 indicates indeed that the growth of populist parties is higher in municipalities belonging to peripheral and ultraperipheral areas. This is true for both individually considered, and especially for M5S (for summary statistics disaggregated by spatial classes, see Table A1 in the supplemental data online). This evidence paves the way for an exploration uncovering the socio-economic factors that contributed to this heterogenous pattern.

Income data
We collect income data at the municipal level from tax declarations as received by the Italian Ministry of Finance each year. From these data, we calculate the average municipality income, which represents one of the key factors in explaining economic inequality. To reduce potential endogeneity deriving from governmental actions between the two electoral rounds, we consider real per capita income data for 2013. Income statistics show an average of €16,800. Also, in this case a large difference exists, approximately €3500, when splitting the sample between core areas (SNAI A + B + C) and peripheral areas (SNAI D + E + F) (see the supplemental data online).

Census data
We also collect several census variables to capture additional socio-economic dimensions. Data derive from the 2011 Italian National Population Census (ISTAT). To begin with, we include a series of structural condition indicators, available for 2011, to measure potential gaps in internet capacity, number of high schools (to measure access to education), number of hospitals (to measure access to healthcare) and seismic risk index as an indicator of exposure to natural disasters. The population age structure is captured by the old age-dependency ratio. Moreover, we control for the presence of foreign resident population.
Gender inequalities are detected by three variables measuring, respectively, gender gap in higher education, female occupation rate and female unemployment rate. Labour market dynamics are captured by the youth unemployment rate, share of foreign employment and lowskilled employment rate. We then include an indicator to consider public mobility of commuters.
Neighbourhood environment and household living conditions are measured by a series of indicators: the presence of non-compliant dwellings, number of overcrowding of dwellings, of adequately served dwellings and of dwellings in bad status, and household crowding.
Considering that some of these variables are not available for all Italian municipalities, our final dataset consists of 6759 out of 8084 total municipalities according to the 2013 ISTAT classification. Our sample thus represents approximately 84% of total Italian municipalities.

Econometric model
We estimate the following ordinary least square (OLS) model: is the percentage growth rate of populist votes (total votes, votes for Lega, votes for M5S) between 2013 and 2018 in the municipality i, S is a vector including a set of municipality-level controls referred to 2011, such as digital divide, number of high schools, number of hospitals and seismic risk. I is the real per capita income expressed in logarithms, while D consists in a vector of census variables as described in section 4.1 including: old age-dependency ratio, share of adequately served dwellings, of dwellings in bad status, of overcrowded dwellings, of noncompliant dwellings and household crowding, an indicator for gender gap in higher education, female employment and unemployment rate, youth unemployment rate and low-skilled employment rate, an indicator for public mobility, the share of foreign resident population and the foreign employment rate. 1 represents an idiosyncratic error term.
Since our outcome and regressors are expressed as municipality averages, we weight our estimates by municipality population. Standard errors are clustered to account for within-province error correlation. Table 2 summarizes the results of the OLS regression explaining the growth of populist parties, jointly considered, of Lega and of M5S between 2013 and 2018.

RESULTS AND DISCUSSION: PERIPHERALITY AND THE ITALIAN POPULIST WAVE
We thoroughly discuss the results for the drivers of the general growth of discontent (1), highlighting, when relevant, the differences between the two parties in driving it.
We start from what should be, based on the empirical literature on the topic, one of the most powerful individual-level predictors of populist voting: income. As extensively found elsewhere  (Antonucci et al., 2017;Becker et al., 2017;Goodwin & Heath, 2016), income plays a crucial role in explaining the growth of discontent in Italy as well, especially for Lega. Increase in preferences for the two populist parties and income are negatively correlated, which means that the higher the income the smaller the growth in the share of resentment. Concerning Brexit referendum, the precursor of the recent populist wave together with the 2016 US presidential elections, income is frequently coupled with age and education to form the 'holy trinity' of the populist voter (Becker et al., 2017;Hobolt, 2016;Los et al., 2017). The average anti-system and anti-EU voter is not only poorer, but also older and less well-educated (Dijkstra et al., 2020). Since these characteristics were over-explored, we were more interested in trying to capture these dimensions at a more territorial level in terms of population structure, hence as a feature of the place more than of the individual. We therefore include in our model the old agedependency ratio and the number of high schools. The former does not have any effect in aggregate terms, but when looking at the two parties individually, it shows a strong negative correlation for M5S, that in fact has mostly appealed to young and more educated people, as the result on the latter seems to suggest. Low-skilled employment and youth unemployment rate, both proxying a condition of marginality in the job market, are important factors explaining the rise in populism. Both results are driven by M5S, that championed the idea of a 'citizenship minimum income' that has made it gain support among out-of-work young people. As stressed by Goodwin and Heath (2016, p. 325) populist supporters are often 'citizens with few qualifications, who live on low incomes and lack the skills that are required to adapt and prosper amid the modern, post-industrial economy'. Noteworthy, the higher women's participation in the labour market, the lower the support for both Lega and M5S. Our result on female employment rate seems to reveal that a gender dimension may be at play in explaining populist voting.
Given the wide use of social networks and online platforms used by the two Italian non-traditional parties, it is important to look at the effect of broadband internet availability. In aggregate terms, the higher the internet capacity, the higher the growth of populist votes. However, when disentangling by party, the opposite is true for Lega. This is confirmed in Table A2 in the supplemental data online, where we see that this result is consistent across centres versus inner areas. Inequality in internet access seems thus to increase people's resentment.
Interestingly, exploring the effect of the supply for public services on the populist vote (Essletzbichler et al., 2018), we see that a higher number of hospitals in a municipality is significantly and negatively associated with discontent. Thus, access to welfare mitigates the resentment. Moving to other dimensions related to housing and neighbourhood conditions and, more generally, to social and material vulnerability (as interpreted by ISTAT 3 ), which we assume that might influence the sentiment of being 'cut out', it is noteworthy to observe that a high percentage of overcrowded dwellings strongly predicts the increase in populism, while the state of conservation of buildings and household crowding seem to generally go in the opposite dimension, reducing the rise of populism. The presence of non-compliant dwellings, which is meant to proxy disadvantaged/degraded areas, ceteris paribus, explains little for the aggregate model, but it is highly and positively correlated with the increase in votes for Lega. The presence of non-authorized camps or similar accommodations, a manifest sign of an insufficient control or concern from national and subnational authorities, often related to the presence of foreign communities, has significantly increased the vote for Lega, which is consistent with its anti-immigration and anti-central government positions. As for the presence of immigrant population, the higher its presence, the lower the growth of support for M5S, showing that the new political force did not appeal to it. Further, this suggests that the contact between Italian and foreign population mitigates the resentment among citizens. To sum up, based on our first general model, the main drivers of the growth of populist voting in Italy are income, low-skilled employment and unemployment on the individual level, internet capacity and lack or limited access to healthcare services, on the territorial level, and finally overcrowding of dwellings, on the social level. Hence, both a peripherality as 'distance' from essential services and one as 'marginalization' seem to play a role in the Italian discontent, with some difference across the two Italian 'challenger' parties.
We are now interested in delving more into these findings and exploring them considering the opposition core/inner areas based on the SNAI classification, our spatial measure of peripherality. Results are shown in Table 3.
Very interestingly, we notice that the negative effect of income on discontent holds both in centres and inner areas, where its magnitude is more than double. This association is therefore stronger in peripheral territories; however, as shown by the result of the Chow test, this difference is not statistically significant. 4 Acknowledging the different nature of the two shades of populisms under scrutiny, we perform the analysis for the two parties individually considered as well (see Table A2 in the supplemental data online). In the following, we will mainly comment on the results of Table 3 for centres versus inner areas, by reporting the differences across the two parties when relevant.
As for income, we find that in the centres it has a negative, though weak, effect only on the votes for Lega, while in inner areas it is highly significant for M5S and less for Lega.
The number of hospitals in a municipality and the use of public transport mobility are positively associated with an increase in discontent in peripheral areas. One possible reason for this may be linked to the quality of these essential services: people's dissatisfaction with the services provided by local healthcare facilities, when in place, and with a more frequent use of public transport for study/work reasons could have led to a greater discontent. In light of the attention that was given to this topic in both the academic debate and within the claims and discourses made by populist movements, the result on the percentage of foreign population, which is highly and negatively significant in inner areas, is particularly salient. A consensus has not yet emerged on the direction of the influence of immigration on voting behaviour because it is unclear what the actual effect of increased contact with immigrants is on the perception of them (Essletzbichler et al., 2018;Matejskova & Leitner, 2011). In our analysis, a larger share of foreign resident population mitigates the growth of populist votes (more specifically, for M5S) in inner areas, where probably the chances to interact more closely with immigrant people are higher than in urban centres. However, when disentangling the results for two parties, the same effect is found in the centres for M5S, meaning that this is probably more related to M5S supporters' attitudes and beliefs than a proper spatial effect. Beyond these differences, this result seems to suggest that increasing contact with foreign residents could reduce the fear and misperceptions about them, hence corroborating the 'contact hypothesis' also in areas traditionally coping with unemployment and decline of essential services, where foreigners could be associated with increased job competition and pressure on welfare services (Backman et al., 2021) as in the interpretation of the 'contrast hypothesis' explaining anti-immigration attitudes. However, Table  A2 in the supplemental data online shows how the unfolding of a 'conflict' dynamic related to a competition in the labour market might be glimpsed: in inner areas, higher levels of foreign-born employees increase the votes for Lega. Moreover, as stressed above discussing a potential gender dimension in populist voting, women's participation in the job market has a strong effect in reducing the growth of M5S in urban centres. Disaggregating by centrality versus peripherality allows us to further interpret the findings on low-skilled employment and youth unemployment, two powerful predictors of populist voting, respectively in centres and inner areas. The former is highly relevant in core areas, where less educated workforce is a strong populism voter base (more specifically for M5S). It is still significant, though weakly, in peripheral areas. Concerning youth unemployment, it is worth noting that, as stressed by Rodríguez-Pose (2018), the lack of job opportunities and the frustration and discouragement this produces are a large source of resentment in declining and lagging-behind areas which feel they have poor future prospects. M5S proved to successfully target these people in inner areas.
The result related to 'Non-compliant dwellings' corroborates the assumption that the perception of being somehow 'abandoned' may be a driver of the rise of populism. This has a strong effect in the increase in populist votes in inner areas, where probably non-authorized settlements, if present, are more visible. However, by disaggregating by parties, we see that in urban contexts it is associated with an increase of consensus for Lega. A powerful predictor of the rise of populist voting in core areas is the percentage ratio of over-occupied dwellings: the higher it is in a municipality, the higher the growth of discontent (channelled into voting for M5S). This is testament of the existence of a peripheral dimension, connected to undesirable housing conditions, in urban areas as well that pushes people to strike back against the status quo. The other dimensions meant to proxy a condition of material and social vulnerability, that is, poor neighbouring and household conditions (i.e., Dwellings in bad status and Household crowding), are strongly but negatively associated with the rise in populism in core areas, but they explain little when looking at the two parties individually.
In short, looking at the Italian increase in populism through the centre-periphery lenses, its main drivers in urban cores are the presence of a large low-skilled workforce, a condition of marginality related to over-occupation of dwellings and, to a lesser extent, female unemployment.
As far as inner areas are concerned, the main factors explaining the growth of populist preferences are a high number of young people struggling to access the labour market, living in a degraded environment with low public control (as evidenced by the effect of the presence of makeshift dwellings, that is, more or less temporary informal settlements) and the use of essential services (healthcare and transport), likely because of their poor quality.  (1) and (2): F-statistic = 0.251, p-value = 0.77.
As stated in the Introduction, from a theoretical point of view, 'periphery' is a relative notion and, partly because of this, it could be extremely heterogeneous. Furthermore, empirically wise, the SNAI classification we use, in its identification of 'inner areas', encompasses different degrees of remoteness, from intermediate to ultra-peripheral areas. What we call 'centres', on the other hand, also incorporate the urban belt. This does not allow to properly disentangle the 'peripherality' effect in general and within each macro-category. To this end, we isolate the last class (C) of the 'centres' and the first one (E) of 'inner areas' to better account for their potential differences. Few studies have moved along this peripherality gradient in their interpretation of populist voting patterns, also due to the dominance of the analytical framework of the urban versus rural dichotomy within the existing academic debate. Thus, we deem essential to deepen our analysis exploring different levels of spatial periphery further disaggregating the SNAI classes. This allows us to better define which periphery is driving which result among the ones we find in our previous estimates. Put it differently, we are interested in identifying the drivers of the growth of resentment in each type of periphery, ultimately providing insights on which dimension of a broadly meant peripherality is relevant to which geographically defined periphery. Table 4 reports the results of the regression by the four classes we consider. Table A3 in the supplemental data online reports the results by individual parties. First, we see that income has a negative and highly significant effect on the total growth of populist parties (more specifically for M5S) both in intermediate and peripheral and ultra-peripheral areas. The differences between these two latter classes are significant (see the Chow test value), while the ones between the other pair combinations are not.
Interestingly, a higher presence of foreign population has a mitigating effect on discontent (expressed through the support for M5S) in the belt areas surrounding urban cores, while its participation in the labour market is positively associated with the growth of Lega in more remote areas. High rates of overcrowded dwellings and low-skilled employees are very salient in core and outlying areas. It is in the urban peripheries and in urban belts where, due to the dynamics of suburbanization, these elements are concentrated and have produced the rise in resentment which manifested in the ballot boxes through the increased support to populist movements. The other variables relating to housing conditions and material vulnerability are significant as well, but they do not show the expected sign, making them difficult to interpret. However, disentangling the two parties, we see that higher shares of adequately served dwellings are negatively associated with the increase of votes for Lega in intermediate areas, and higher shares of overcrowded dwellings are associated with a higher growth of M5S in outlying areas.
Youth unemployment is a driving factor for populist voting (channelled through M5S voting) in the two aggregations of inner areas considered, pointing to one of the main challenges to the policy attempt to revert their population decline. Among the critical issues of living in these areas, the quality and quantity of essential services (chiefly tackled by SNAI) is one of the more pressing. It is what the result on public transport mobility seems to suggest: higher shares of commuters who daily use public transport for study-or work-related reasons in peripheral areas, and who experience their paucity and poor quality, significantly contribute to foster the growth of discontent (for both parties). As a further testament of the importance of essential services in understanding this phenomenon, a better internet access is negatively associated with the increase in votes for Lega in intermediate and more peripheral areas. This thus probably goes hand in hand with people's dissatisfaction with their experience as public service users and their feeling of being disadvantaged and neglected.
More generally, Table 4 shows peculiar factors explaining the Italian populist wave when further isolating municipalities along the peripherality gradient.
Taken together, our findings seem to suggest that different 'peripheries', and their different inherent dimensions, have driven the upsurge of populism in Italy. On the one hand, inner areas, coping with all the issues connected to their condition of remoteness, where youth Discontent in the 'peripheries': an investigation of the rise of populism in Italy unemployment, the deterioration of essential services and the perception of being abandoned by (or of lack of control from) local authorities have produced a greater and greater dissatisfaction with mainstream parties. On the other, if we find that discontent is mainly concentrated in peripheral and ultra-peripheral areas, and less, comparatively, in outlying and intermediate areas ( Figure 1)that is, the areas which have best reacted to the latest recessionary shock(s) (Urso et al., 2019) urban and belt areas show very peculiar drivers of populism which call for more attention from scholars and policymakers. Some factors that can be associated with some kind of 'peripherality', meant as marginality/exclusion, such as poor housing conditions and the lack of qualified skills among workforce, are proxies for a material and social vulnerability which might further fuel resentment.

CONCLUSIONS
Peripherality is not merely a matter of physical distance, of a general disconnectedness or remoteness related to the fact of being located on the fringes of central cores. Nonetheless, in order to face peripherality, which was interpreted in its narrowest sense as mere 'edgeness', improving accessibility through the provision of better physical infrastructure has been one of the main policy responses by governments 'concerned with reducing inequalities between different parts of a country in the interest of greater homogeneity' (Herrschel, 2012, p. 34). These investments have at some point shown their limits in displaying effective impacts in terms of reducing spatial unevenness and, hence, fostering cohesion. The feeling of being left behind and disconnected from the general development trajectory of the country has likely accumulated over time also because of decades-old city-centric policies aiming at boosting (urban) competitiveness. The recent populist upsurge has in a way also urged us to recognize peripherality as being more than mere geographical distance and not solely about infrastructure alone. If it has been increasingly acknowledged that peripherality rather implies a difference in terms of welfare, skills, opportunities, prospects and quality of life, the wave of discontent expressed through the ballot boxes has unequivocally put forward the issue of the 'perception' of some distance from institutions, policymakers and traditional parties. They have started to be increasingly pointed to  (1) and (2): F-statistic = 0.94, p-value = 0.39; between (1) and (3): F-statistic = 0.23, p-value = 0.79; between (1) and (4): F-statistic = 2.24, p-value = 0.11; between (2) and (3): F-statistic = 0.93, p-value = 0.39; between (2) and (4): F-statistic = 2.07, p-value = 0.13; between (3) and (4): F-statistic = 3.54, p-value = 0.03.
as an aloof elite, who is deemed to be primarily focused on keeping the status quo, with its related benefits and corresponding unevenness, pretty much unaltered. This reminds of Tarrow's (1977) powerful image about Italy in the 1970s that is still effective to describe today's attitude towards politicians and that has been further exacerbated by recent events at the supra-national scale, as testified by the diffusion of populist voting across whole Europe and beyond. Tarrow claims that 'to citizens at the periphery the ideological drama at the top may seem like a fresco of baroque armies locked in futile combat above a stage in which the low comedy of political exchange is being played' (pp. 253-254). On top of recent broader tendencies which are likely to shape the Italian geography of discontent, the importance of accounting for a longitudinal approach should be acknowledged, though being beyond the scope of this work. In fact, today's voting patterns may also interact with (and may be interpreted in light of) historical macro-regional dynamics that used to dominate understandings of Italian politics with different ways of thinking about voting (e.g., ideological, clientelist, party affiliation) (Putnam et al., 1994;Tarrow, 1977) .
In our paper, we show how the growth of discontent in Italy between the two previous general elections (2013 and 2018) has different explanatory factors in core and peripheral areas. Irrespective of geographical peripherality, where, however, discontent is more concentrated, our findings reveal that a condition of marginality or vulnerability may feed a sense of revenge connected to the feeling of 'not mattering' also in urban areas. This poses a great challenge for policymakers. Indeed, whereas geographical remoteness cannot be modifiedthough it can certainly be mitigated through investments in physical infrastructure and servicesthe perception of distance from institutions could be even more difficult to observe and unpredictable in its effects, as was the case with the unexpected populist shock. If peripherality has to be tackled, hard infrastructures are therefore not enough. Soft ones, aimed at fostering participation and inclusiveness to reduce the feeling of exclusion from key networks where decisions are taken, are essential. On the other hand, from an academic perspective, research is more than ever needed to understand where (and how) old divisions or marginalities are reinforced and where (and how) new ones are created.

DISCLOSURE STATEMENT
No potential conflict of interest was reported by the authors.

NOTES
1. The growth rate is calculated as follows: (Valid Votes 2018 − Valid Votes 2013 /Total Valid Votes 2013 ) * 100. 2. It is worth specifying that electoral outcomes provided by the Ministry of Interior do not include the ISTAT code for municipalities, but only their names as a string variable. We therefore carry out a fuzzy matching to align electoral data to our covariates. Considering that the official list of municipalities changed several times between 2013 and 2018 (some municipalities have been merged with larger cites, others changed their name), and because of the fuzziness of the matching procedure based on strings, of 8084 municipalities in 2013, we could automatically match only 7235 municipalities (about 89% of the initial 2013 sample) in order to keep the list of municipalities balanced between 2013 and 2018. We further lose an additional 476 municipalities because of missing values in our set of covariates collected from ISTAT. More specifically, we missed observations for foreign employment rate, youth employment rate and seismic risk. This procedure leaves 6759 observations in the final sample. However, municipalities excluded from the estimation sample represent minor municipalities that are not spatially sorted in specific areas. Excluded municipalities have a median population of 857 people (population density of 24 people/km 2 ), compared with a median population of 4763 people for in-sample municipalities (population density of 127 people/km 2 ). The use of population weights in our estimates accounts for size effects, so that the small population size of excluded municipalities should not significantly alter our results.