A capacity approach to territorial resilience: the case of European regions

Abstract

Regional resilience is a new paradigm to explain the local system ability to cope with a negative event, tolerating the effect produced by the perturbing action. The first objective of the paper is to analyze the complex concept of regional resilience, adopting a systemic and holistic approach. Using a multidimensional methodology, regional resilience is described by outcome and driver variables, with focus on sustainability of local systems, broken down into the three pillars of economy, society and environment, whereby the holistic approach means that each dimension of territorial sustainable development is partly determined by its relations with the other dimensions. The second aim of the paper is then to test the relations between determinants and outcome of regional resilience. This framework is different compared to previous empirical studies, which primarily focus on economic performance in terms of income or employment dynamics. The model is applied to the case of European regions, to get a map of regional resilience in its different dimensions.

1 Introduction

Sustainability of territorial development, as the future capacity of local systems to support human well-being, is closely associated with the concept of resilience. The more the systems are resilient, the less they likely shift into configurations that involve a reduction in the quantity or quality of territorial assets and resources, as effect of stressors of various nature. The inclusion of resilience in any comprehensive measure of sustainability is necessary to take into account risks that significant losses in well-being may occur (Dasgupta and Mäler 2003).

The scale and impact of shocks and stresses which affect the development of territorial systems increase with the growing of urban areas and the urban population. OECD (2016) classified these stresses into several groups. Industrial structural change (for example relocation or closure of a city’s key firms) affects employment in that specific industry and in related ones. Economic crises, such as the financial crisis of 2008 and sovereign debt crises which affect European Union since 2009, have a global impact. The responses of cities depend on characteristics such as the structure of their economy, the proximity to the capital city (OECD 2014) and the internationalization of the local economy (Turcu et al. 2015). Population inflow and outflow influence the employment rate, taxable income, and the need for public services: migration has an important impact on societies and economies, because it lead to social friction, and social integrations constitutes a big challenge for local communities, in particular when they accept asylum seekers. Violence, crime, terrorism may represent critical shocks for a city. In the same way natural disasters (for example earthquakes, floods and hurricanes) have a critical impact not only on environment, but also on the economy and society of territorial system, in particular when they can cause severe disruptions of the energy supply. Leadership change and any discontinuity of policies represent other stressors, which could affect the economic base of a territory and the social structure. Any sort of shock to complex systems such as a territorial system have significant economic, social, environmental and institutional repercussions.

The economic crisis of 2008 constituted the most severe economic downturn in the history of the European Union, which is the investigated area of this study. Not all regions experienced economic decline, and rates of recovery have largely varied. While some places experienced a swift return to pre-crisis levels of employment and economic output, other entered a period of long lasting stagnation. The effects of the economic crisis were relevant across most territories, at a national level Poland and Switzerland were the only states that had not experienced a fall in levels of GDP output. Whereas for employment these two States together with Germany, Luxembourg and Belgium showed the ability to retain pre-crisis employment levels. From 2010 the first signs of recovering regions are visible: Polish regions continued to maintain their stronger GDP performance, and some regions in Germany, Austria, France, Belgium and the Netherlands also managed to contain the declining levels of GDP. In terms of employment, the stronger performance of Germany, Southern France, Belgium, Luxembourg, Malta and of some areas in the UK, Portugal and the Netherlands is evident. GDP decline remains a strong feature of the peripheral economies, but growth has been maintained in regions across a broad group of Poland, Germany, Switzerland, Austria, Slovakia, Netherlands, Belgium and southern France territories (ESPON 2014).

Most of the natural disasters in Europe are due to climatological and hydro-meteorological events. While proactive measures have minimized the loss of human life from disasters, economic losses due to disasters continue to rise in Europe (UNISDR 2016). In a study performed by Munich Re, the effect of 160 natural disasters in Europe in 2014 have been analyzed, highlighting that they caused losses of US$18 billion and an amount of 350 deaths. In particular, it is the case of floods in the Bosnia and Herzegovina, Bulgaria, Croatia and Serbia, the intense storms across France, Belgium and Germany and UK. As a consequence of both development from an anthropic point of view and the increasing impacts of climate change, the trend of economic losses and social costs tends to challenge both sustainability and economic growth.

The economic crisis of 2008, the growing of social inequality, the political and financial instability, the worsening of ecological imbalance and natural disasters represent big challenges for European regions. They lead to a reflection on the need for a definition model of resilience capacity that catch the economic, social and environmental dimensions and effects of the crisis in order to provide adequate responses to the demand for sustainable development policies.

The first aim of this work is to analyze the complex concept of regional resilience, referring to the theoretical background of sustainability. Hence, a systemic and multidimensional approach is adopted and regional resilience is represented as a process and described by outcome and driver variables, connecting them to the three pillars of sustainability (economy, society and environment).

The relationships between outcome and driver variables are complex and characterized by cross-linkages that are poorly investigated in literature. The second aim of this work is to attempt an initial test of the linkages between variables within each dimension and among dimensions of territorial resilience, through the application to the case of European regions.

The work is articulated as follows. The next section focuses on literature review, highlighting the connection between the concepts of competitiveness, sustainability and resilience and the differences between the approach of resilience performance and the one of resilience capacity. The third section introduces the conceptual framework designed by the authors in order to represent territorial resilience from a multidimensional point of view. Outcome and driver variables are defined and connected to the economic, social and environmental dimensions of the phenomenon: GDP per capita, life expectancy and $\hbox {CO}_2$ emissions for the representation of resilience outcome, 23 variables for the representation resilience capacity, identified through the previous literature review. Unit of analysis, data and multivariate techniques are specified for the construction of resilience composite indicators as well as for the analysis of the relationships among resilience outcomes and drivers: principal component analysis (PCA) and spatial econometric model (SAR). The fourth section presents the results of PCA and the maps of economic, social and environmental resilience of European regions. The last section shows the results of SAR, outlining some differences from OLS regressions, and highlights the effects of economic, social and environmental drivers on GDP pc, life expectancy and $\hbox {CO}_2$ emissions. In a specific paragraph, an analysis of the economic resilience outcome before and during the global crisis is presented. In the conclusions, paper synthesis and some policy recommendations have been included.

2 Literature review: resilience and sustainability

The notion of resilience has been first elaborated in material sciences: it is in fact the physical property of a material to return to its original shape or position after a deformation that does not exceed its elastic limits. From this meaning, the term was used in different disciplines, but early studies on the topic of resilience are attributable to the research on the environmental phenomena. According to Odum (1985) resilience is the ability of a system to recover when it is affected by a disturbance, with a definition that is very similar to the one used in the field of engineering. The theme of ecological resilience has been studied by Holling (1973) since the early seventies: starting from the analysis of complex systems and adaptive behaviors, he introduced an articulate definition of resilience in the social-ecological systems. According to this perspective resilience is the ability to cope with a negative event, tolerating the effect produced by the perturbing action. A resilient system is able to evolve into multiple states of equilibrium different from the previous one, after the perturbing action, without entering into functional crisis. In this approach, the resilience of a system is its ability to tolerate disturbance, counteracting the increase of entropy produced.¹ The components that feed resilience are those that promote the availability of resources and facilitate the ability to adapt, recover and regenerate (Paton 2001; Resilience Alliance 2007; World Bank 2014).

There is a strong connection between resilience and sustainability. Sustainability captures the aspiration for persistent and equitable well-being in the long run, which is summarized in the dimensions of resilience: the ability to persist and the ability to adapt. Sustainable development has the aim to create and to maintain prosperous social, economic and ecological systems from a co-evolutive point of view. Both sustainability and resilience recognize the need for precautionary action on resource use and on emerging risks aimed at promoting the integrity of well-being into the future.

Resilience is not about promoting growth or change for its own sake. It is about promoting the ability to absorb shocks and stresses and still maintain the functioning of society and the integrity of the ecological systems.... Most important of all, resilience requires societies to have the capacity to adapt to unforeseen circumstances and risks. These objectives give generic guidance on how to promote sustainability at different scales (Adger 2000).

The theoretical frameworks proposed in regional sciences for the description of resilience (Vale and Campanella 2005; Foster 2007; Pendall et al. 2010; Simmie and Martin 2010; Martin 2012; Graziano and Rizzi 2016; Boschma 2015) offer interesting insights into the analysis of territories capabilities to respond or use the negative event as an opportunity for change and development. One of the charm of resilience notion is its efficacy for understanding, managing and governing complex linked systems of people and nature (Folke et al. 2004). It is a goal that should be promoted, not for ecosystems or for social institutions per se, but for social and ecological system interactions. Resilience thinking promotes the understanding of the co-evolution of socio-economic and ecological systems, which describe territorial development from a multidimensional point of view (Graziano 2014). Resilience facilitates the analysis of the dynamics of social and ecological systems and the definition how those evolutionary cycles enable urban systems to reorganize themselves (Eraydin and Tasan-Kok 2013). Resilience is the ability of maintaining information that organize the territorial system and of introducing variations that can constitute important innovations, in an evolutionary perspective (Boschma 2004; Sotarauta 2005; Martin and Sunley 2007). The applications of these concepts in economic literature (Briguglio et al. 2009; Naudé et al. 2009; Chapple and Lester 2007; Foster 2007; Sotarauta 2005; Liou and Ding 2004), social sciences (Zimmerman and Arunkumar 1994; Cutter and Finch 2007) and ecological and socio-ecological studies (Carpenter et al. 2001; Walker et al. 2004; Folke 2006) have contributed to enrich the notion of resilience with different interpretations.

In regional science the approaches to regional resilience can be summarized into two research lines. The first one identifies resilience as a performance measure (Bailey and Turok 2016; Martin 2012, 2016; Martin and Sunley 2015), the second one identifies resilience as a capacity measure (Cutter and Finch 2007; Foster 2007; Graziano 2014; Graziano and Rizzi 2016; Walker et al. 2004; World Bank 2014).

The first approach is often used to represent the economic dimensions of the phenomenon, analyzing the dynamics of local systems in terms of employment or value added during and after cyclical crises. It is an hazard-specific approach, based on the conceptualization of resilience as a result of a path, which involves the identification of the phases of regional economic recession (in particular the shock and the phase of recovery). According to this approach indicators are calculated to represent resistance during the crisis and recovery after recession. These measures compare the change in regional employment or value added with the national average change, during and after the shock. In this approach the focus of the analysis is on resilience as a performance measure and on its determinants.²

The second approach of holistic nature focuses on sustainable development from anthropic and ecological points of view identifying resilience as a complex input, a multidimensional ability which determines the development of territory. It focuses on the capacity to improve autopoietic mechanisms of territorial systems, considering resilience as driver factor. The territory is an open system, characterized by interconnected components and feedbacks that cause nonlinear processes. When the probability of specific shocks is unknown, a holistic approach is useful to provide an initial informative framework of all systemic features and resources that could determine fragility or could influence the adaptation capacity (Walker et al. 2004). This approach avoids underestimation of unexpected aspects, focusing on factors observed on long periods of time, rather than on resources needed to tackle a specific critical event (Paton 2001). An holistic approach complies with the need for preventive strategies of risk management which needs to combine the capacity to prepare for risk with the ability to cope afterward: preparation should include a combination of actions such as gaining knowledge, acquiring protection and obtaining insurance.³ The growing interest in this approach to the study of resilience highlights the transition from a culture of emergency and reconstruction, following the occurrence of a specific traumatic event, toward a preventive approach to the problem of risk. It constitutes an attempt to meet the expectations of a territorial planning that has to face the problems arising from the uncertainty of the local systems (Federal Emergency Management Agency 1998; United Nations Commission on Sustainable Development 2001; World Bank 2014).

Table 1 The drivers of regional resilience: pillars, themes and basic literature sources

Table 1 summarizes the elements of description of territorial resilience drivers, emerging from the analysis of theoretical and empirical literature on economic, social and environmental resilience. The characteristics of a resilient system are: availability of resources, low vulnerability and strategic adaptation capacity (Resilience Alliance 2007). Resources availability comes from the endowment of natural, human and economic capital and necessary assets for responding to, withstanding and recovering from adverse situations (Foster 2007; World Bank 2014). Vulnerability is connected to structural homogeneity of the system (Gunderson et al. 1995; Young et al. 2006), and it determines the propensity for damage (Adger 2000). The adaptation capacity refers to adjustments in ecological, social or economic systems in response to present or expected external stimuli, i.e., policies or strategies to moderate vulnerabilities, respond to shocks or reinforce strengths and opportunities (Boschma 2004; Sotarauta 2005).

This specific themes are connected to the three dimensions of sustainability and then to the semantic pillars “resource availability”, “vulnerability” and “strategic adaptation” according to the analyzed literature.

3 A multidimensional approach to regional resilience

The present work focuses on resilience capacity and on the relationships between this ability and territorial well-being, broken down into the three dimensions of sustainability. A vision of territorial well-being as the ability of a given territory to ensure its inhabitants a proper sustainable development in economic, social and environmental terms, is adopted. This is a refocusing of outcome variables that more directly measure standard of living well-being and quality of life: the Gross domestic product per capita in purchasing power parity, as synthetic variable of economic well-being (Huggins and Thompson 2012; Kitson et al. 2004); the life expectancy as indicator of social well-being (Aiginger et al. 2013; Dallara and Rizzi 2012); the $\hbox {CO}_2$ emissions, as negative indicator of environmental sustainability (Rizzi et al. 2015; UNDP 2014). These measures will be considered and analyzed separately, in order to better understand the specific relationships between drivers and single outcome variables in the three spheres of sustainability (Table 9 in “Appendix”). Figure 1 shows the research design which highlights outcomes and the drivers of regional resilience, which is represented as a process.

Fig. 1

The regional resilience approach: research design

The drivers of resilience are the combination of multiple tangible and intangible elements that are fundamental for maintaining of territorial self-renewal capacity (Graziano and Rizzi 2016). A group of variables has been identified according to the analysis of theoretical and empirical studies on the theme of resilience and vulnerability of economic, social and environmental systems (Table 2).

Table 2 The drivers of regional resilience: variables and basic literature sources

3.1 Data and methodological aspects

The conceptual framework of regional resilience has been applied to the case of European regions, in order to measure the outcome variables and the resilience drivers in their different dimensions and to highlight the linkages among them. The analysis focuses on 248 European regions⁴ at NUTS-2 level, observed over the period 2000–2013. EUROSTAT and OECD have been the sources for the collection of 26 variables, used as temporal average: GDP per capita, life expectancy and $\hbox {CO}_2$ emissions for the representation of resilience outcome, 23 variables for the representation resilience capacity, in the economic, social and environmental dimensions (Table 2).

In order to overcome the difficulties of reading the local systems characteristics through a system of individual indicators and to deal with collinearity problems which could affect the econometric analysis, a method of statistical aggregation of variables has been adopted (Annoni and Kozovska 2010; Dallara and Rizzi 2012; Graziano 2014).

Starting from the initial dataset collected, a group of variables has been selected, according to the result of structure analysis, based on asymmetry and kurtosis checking. When appropriate, the elementary variables have been transformed and then standardized.

Coefficients of Pearson correlation were considered for each pair of variables in order to exclude redundant or too little correlated variables, analyzing the corresponding correlation matrix. Among the couples of variables which displayed correlation coefficients greater than 0.8 or less than 0.3 as absolute value, one of them has been eliminated. This type of approach for the selection of the individual variables and for the construction of the aggregated measures has been adopted to avoid the duplication of information as well as to ensure the possibility to get composite indicators at the same time (Dallara 2008; Jolliffe 2002).

For measuring the resilience drivers, the principal component analysis (PCA) has been applied to the selected variables. The composite indicators which represent resilience drivers are the first component that explains more than 50 percent of the total variance. The loading factors resulting from principal components analysis are the measure of correlation that links the elementary variables to the composite indicator. Through the loading factors it is possible to identify elementary variables that could better describe the phenomenon investigated. The eigenvector associated to the first eigenvalue represents the row vector of the coefficients, which multiply the original variables in the linear combination that generates the new variable, that is the composite indicator.

For the analysis of the relationships among resilience outcomes and drivers, two instruments have been used: scatter plots and econometric models. The econometric models use as regressors both the elementary variables and the three composite indicators obtained by the principal component analysis. The elementary variables are standardized, and they maintain their natural orientation, as well as the three composite indicators are oriented in order to express positive meaning. The aim is to highlight whether and to what extent the elements that describe the resilience drivers act on the outcome variables.

For each dimension of resilience, an OLS regression is first of all proposed using composite indicators as regressors. Skewness–Kurtosis and Shapiro–Wilk tests for normality of residual distribution are reported as well as Breusch–Pagan test for Heteroscedasticity. The choice of use composite indicators depends on the need for deal with collinearity problem (Chatterjee and Hadi 2006). In order to validate this choice, an OLS regression using elementary variables is also proposed, reporting the value of variance inflation factors for collinearity checking.

It is necessary to point out that all the used variables are cross-sectional among the European regions and they are characterized by high spatial autocorrelation (significant Moran’s I). The distribution of individual observations (each single region) does not exhibit a constant variance (LeSage 1999). Values of a given variable observed at one region (ith observation) depend on values of neighboring observations at nearby regions, indicating spatial dependence (LeSage and Pace 2009). Spatial interactions, diffusion effects, hierarchies of place, spatial spillovers are causes of spatial dependence among the observations. Moreover, cross-sectional variables present usually a second problem: the spatial heterogeneity in the relationships among variables, which violates the assumption of the classic linear econometric model. In spatial contexts, every region in space holds a different relationship among given variables. In order to take into account these two typical problems of spatial cross-sectional data, the spatial autoregressive model (SAR) introduced in the literature by Whittle and systematized by Cliff and Ord (1973, 1981) is implemented.⁵ For each regressions spatial lag and spatial error are reported, using maximum likelihood estimators (ML) and generalized two-stage least square estimators (GL2SLS) as parameters and both elementary variables and composite indicators as regressors.

4 Measuring resilience drivers in European regions

For measuring the resilience drivers, the principal component analysis (PCA) has been applied to the variables that have been selected, according to the result of structure analysis, based on asymmetry and kurtosis checking. In particular, 13 variables define resilience drivers for 248 European regions: 4 variables describing the economic sphere, 5 the social and 4 the environmental dimensions.

Economic resilience drivers can be attributed to physical and human capital, innovation and entrepreneurship. Gross fixed capital per employee is a proxy of resources availability for the economic dimension (Briguglio et al. 2009) the proportion of graduates on population represent the dimension education attainment, as an indicator of resources availability in terms of human capital (Lung et al. 2013; World Bank 2014) research and development expenditure, employees in science-technology sector describe the local innovation system, as fundamental drivers of territorial resilience (Chapple and Lester 2007; Sotarauta 2005).

Social resilience drivers are related basically to social vulnerability. The indicator of death rate for circulatory disease is a proxy of health state (Cutter and Finch 2007; Glatron and Beck 2008), whereas disease for accidents, long-term unemployment rate and the proportion of Neet on population describe the dimension of social hardship which affects negatively the ability of social systems to cope with a negative event (Glatron and Beck 2008; Rizzi et al. 2015). Lifelong learning is a proxy of the self-renewal capacity of local social system (Rajib 2009).

Table 3 The PCA results for the resilience drivers (248 European regions Nuts 2)

Fig. 2

The map of economic resilience drivers in European regions

Environmental resilience drivers are connected to the theme of eco-systemic qualities and pressures. Biodiversity is an indicator of heterogeneity of ecological structure, which intensely impacts on territorial resilience. In this work, this measure is referred to land cover (Schneiderbauer et al. 2013; Costanza et al. 2007). Wood land represent the endowment of natural capital, which increases the availability of necessary resources for territorial capability to regenerate. Artificial infrastructures and population growth rate are proxies of anthropic negative forces on nature which negatively affect the resilience capacity of local environmental system.

Table 2 reproduces the obtained results of PCA for the three dimensions, communalities, loading factors, coefficient scores, explained variance and Kmo tests. Table 9 in “Appendix” reproduces the description, the units of measurement and the sources of the final set of 13 variables.

The first component for economic resilience drivers (64% of total variance) is positively correlated with all single variables. For the social resilience drivers the first component that explains 53% of total variance highlights positive correlations with the five vulnerability variables and negative with lifelong learning: so the sign of the regional scores should be reversed when used in the following statistical and econometric analysis to obtain a composite indicator with positive meaning of resilience capacity. Finally the first component of environmental drivers (41% of total variance) displays positive correlation with biodiversity and woodland and negative with urbanized areas and demographic balance as pressure indicators, giving the expected logic meaning to the environmental regional scores (Table 3).

Economic resilience drivers favor the metropolitan regions of the great European capitals and industrialized areas. This is the case of the top 10 capital regions of Hovedstaden (1st place in the ranking), Helsinki-Uusimaa (2nd), Stockholm (3th), Inner London (4th), Île de France (7th), Région de Bruxelles-Capitale (8th), Luxembourg (10th), followed by and Wien (13th) and Praha (14th), which highlight strong economies of urbanization leading to positive demographic flows and to attraction of investments and high-skilled human capital (Fig. 2). In the group of the most resilient regions several areas of Germany, Belgium and UK emerge: some with a strong presence of high tech discricts (Oberbayern, Berkshire, Buckinghamshire and Oxfordshire, respectively, at the 6th and the 18th place in the ranking), with a remarkable trade vocation (Hamburg and Antwerpen, at the 9th and 23th positions) and with the highest density of scientific, academic and research organizations (Walloon Brabant and Stuttgard, respectively, at the 5th, and 17th places), because of their high propensity to private and public investment in innovation and a widespread quality of human capital.

In contrast Swietokrzyskie, Kujawsko-Pomorskie, Lubuskie and Opolskie in Poland, Ionia Nisia and Notio Aigaio in Grece, Východné Slovensko and Západné Slovensko in Slovakia, and Észak-Magyarország in Hungaria highlight the lowest levels of economic resilience drivers in Europe. These regions are located in the European areas characterized by the small presence of economic infrastructures, they reach the lowest level of Gross Fixed Capital and are characterized by poor private and public investments in innovation. In addition to the known dichotomy between regions of the Blue Banana and Northern Europe on one hand and the Mediterranean (South of Spain, Italy and Greece) and Eastern Europe (Hungary and Poland) on the other hand, it is possible to point out the good performance of some wealthy areas but peripherals such as Midi-Pyrénées in France and País Vasco in Spain.

These data are confirmed by several regional resilience studies in terms of employment or Gdp pc in countries such as Italy (Di Caro 2015; Lagravinese 2015), Spain (Cuadrado-Roura and Maroto 2016), Hungary (Benke et al. 2016), Slovakia (Hudec et al. 2016), Poland (Masik 2016) and in general from comparative studies at European level (ESPON 2014; Sensier et al. 2016; Crescenzi et al. 2016; Giannakis and Bruggeman 2017). In the case of Greek regions there are different results regarding regional levels of resilience: Giannakis and Bruggeman (2015) found that the coastal and rural regions of Greece (Ionia Nisia, Notio Aigaio and Voreio Aigaio) were more resilient to the recent economic crisis than the continental urban regions for the sectoral composition of these economies and the positive role of agriculture, agro-industry and tourism.

Social resilience drivers are linked with the economic ones, with the best results of Scandinavian regions bathed by the North Sea and the Norwegian Sea (Fig. 3). In the top 10 group, the high level of social resilience drivers of Midtjylland, Syddanmark in Denmark, Hampshire and Isle of Wight in the English Channel emerge. The map confirms the excellent performance of capital regions of Stockholm (1st place), Helsinki (4th), Outer and Inner London (17th and 22th) and the good result of Madrid (66th), Île de France (69th) and Praha (72th), because of low level of long-term unemployment and death rate for accident as well as high level of self-renewal capacity, represented by population involved in lifelong learning programs.

Fig. 3

The map of social resilience drivers in European regions

It is also possible to observe the excellent positioning of some Duch areas such as Utrecht (3th) in Randstad’s polycentric urban region due to a low level of social hardship highlighted by the lowest rate of Neet. The performance of some Southern English regions as Surrey, East and West Sussex (10th), Berkshire, Buckinghamshire and Oxfordshire (11th), Gloucestershire, Wiltshire and Bristol (13th) is explained by the high rate of participation in lifelong learning programs reflected in a very low rate of long-term unemployment. It is possible to notice some regions in the group of ones that record a good level of social resilience drivers which are characterized also by a low level of social hardship, such as in the Austro-German area: Vorarlberg, Tirol, Salzburg, Freiburg, Oberbayern, Stuttgart, Tübingen and, in particular, Hamburg with the lowest death rate for accident in Europe. The performance of Alsace and Rhône–Alpes French regions is noticeable, as well as some peripheral ones such as Bretagne and Pays de la Loire due to the best performance in health dimension represented by low death rate for circulatory disease. This is also the case of the Spanish Comunidad Foral de Navarra and País Vasco, which demonstrate how even poorly urbanized areas are able to reach a social balance and quality of life more than acceptable.

On the contrary, it is possible to notice the worst performance of some Greek regions such as Sterea Ellada, Peloponnisos, Anatoliki Makedonia, Ipeiros, Dytiki Makedonia, and Thessalia, but also of Sicily in Italy as marginal area characterized by high level of all components of social hardship and low of self-renewal abilities. In the bottom 10 the Slovakian regions of Západné Slovensko, Stredné Slovensko and Východné Slovensko record a very low level of social resilience drivers due to the highest death rate and social hardship.

Environmental resilience drivers reverse the picture emerged in terms of territorial economic and social factors: the richest regions of the UK and the Netherlands are in fact the most penalized where the northern European areas of Sweden (Norra Mellansverige, Småland med öarna, Mellersta Norrland and Övre Norrland) and Finland (Pohjois-ja Itä-Suomi, Etelä-Suomi), some Austrian (Steiermark and Kärnten), and German regions (Oberfranken and Gießen) are awarded thanks to the large extensions of wood land and the low urbanization (Fig. 4). But in the top of the ranking, also some poor regions of Portugal (Alentejo, Centro, Açores), Italy (Calabria, Molise, Basilicata) and Slovenija (Vzhodna and Zahodna Slovenija) show positive ecological balances thanks to lower environmental pressures because of minor industrial and infrastructural sites; the same evidence in some East Europe regions such as Stredné Slovensko in Slovakia and Moravskoslezsko in Czech Republic.

On the contrary, UK regions as Inner and Outer London, West Midlands, Merseyside, Manchester and West Yorkshire, German regions as Hamburg and Bremen, or Dutch and Belgian regions such as Utrecht, Flevoland and Bruxelles, show low level of environmental resilience drivers for their diseconomies of urbanization due to production and transport congestions.

Fig. 4

The map of environmental resilience drivers in European regions

5 The relations between drivers and outcome of regional resilience

In this section, the relationships between the composite indicators defined for measuring the drivers of regional resilience and the single outcome variables are presented. It resorts to scatter plots that display the relative positioning of each European region and the relationships between the variables, and correlation matrix.

Among the economic outcome and the economic resilience drivers (Fig. 5), a marked positive relationship is observed (Fig. 5): European regions that are characterized by the best levels of outcome variable (GDP per capita) seem to be so driven by higher economic resilience capacity, as measured by levels of innovation and research, technology and capital investment.

It is the cases of Inner London, Luxembourg, Bruxelles Capitale, Hamburg, Helsinki-Uusimaa and Hovedstaden, regions with high levels of virtuosity and good performance of economic resilience drivers. Some regions, such as Ionia Nisia and Notio Aigaio (Greece), Lubuskie (Poland), Észak-Magyarország and Dél-Dunántúl (Hungary), are penalized in terms of economic outcome and unsatisfactory results also in terms of economic drivers.

Fig. 5

The relations between drivers and outcome of economic regional resilience

The composite indicator of social resilience drivers affects the social performance of regions in terms of life expectancy (Fig. 6). The correlation is still positive, albeit to a lesser extent than the economic dimension. The comparison between outcome and determinants rewards some rich areas (Stockholm and Utrecht) but also some Spanish regions (Madrid and Comunidad Foral de Navarra), virtuous regions in terms of life expectancy and good endowment of social resilience drivers as quality of human capital. It is possible to observe some exceptions such as Ipeiros in Greece or Molise and Abruzzo in Italy (high level of outcome, low level of social drivers) and the English regions of Greater Manchester, Eastern and South Western Scotland (penalized in terms of life expectancy despite the excellent social resilience drivers). It can be said that the response capacity of local systems at social level does not seem to be for all regions a good driver for the social welfare dimension.

Fig. 6

The relations between drivers and outcome of social regional resilience

Environmental resilience drivers instead are negatively correlated with ecological outcome ($\hbox {CO}_2$ emissions), with the best “twin” performance in Swedish regions of Ovre Norrland, Mellersta Norrland and Norra Mellan Sverige and in Alentejo (lower right quadrant) while the worst “twin” performance in the metropolitan areas of Berlin, London and Bruxelles (left upper quadrant). But it is possible to observe some mixed case of high levels of ecological vulnerability associated with high values of resilience drivers, linked to active policies of land use (German regions of Saarland, Arnsberg and Darmstadt). In these cases, a possible explanation is because resilience drivers develop in response to vulnerabilities but do not solve them. Such considerations lead to more emphasis on the role of governance and policy measures that could be characterized either as preventive or as palliative (Fig. 7).

Fig. 7

The relations between drivers and outcome of environmental regional resilience

In general the correlation matrix between outcome variables and drivers of regional resilience (Table 4) shows different linkages among economic, societal and environmental dimensions, setting the stage for an in-depth analysis of these relations. The study of links in the following section could be helpful to clarify which factors contribute most to explain the resilience in European regions.

Table 4 The correlation matrix

The estimated relations between GDP pc and the variables that describe resilience drivers are reproduced in Table 5. When using elementary variables as regressors (first column) in the OLS, residuals do not have a normal distribution and are heteroscedastic. Furthermore, the mean variance inflation factors greater than unity reveal that there is some evidence of collinearity (Chatterjee and Hadi 2006). The correlation matrix of the economic resilience drivers highlights significant bivariate linear relationships, and the same occurs for social resilience and environmental resilience variables.

Table 5 Economic regional resilience: outcome and drivers (dependent variable: GDP per capita ppp mean 2003–2012)

Because of these correlations and the value of mean VIF, a new OLS is estimated, in which the elementary variables are substituted by the unique principal component with the eigenvalue greater than unity and these first components are inserted in the OLS as regressors. The same approach is applied to the variables of economic, social and environmental resilience drivers (second column). In this case, the GDP pc is positively linked with economic resilience drivers, but negatively with environmental ones, highlighting a clear trade-off between economy and environment. The linkage between GDP pc and social resilience drivers is not significant. In this second OLS, the mean VIF decreases, but residuals are non-normal distributed and heteroscedastic.

In order to deal with spatial cross-sectional data (spatial dependence and spatial heterogeneity), the spatial autoregressive model (SAR) is implemented. The third column of Table 5 presents the results: the GDP pc is regressed with the elementary variables that make up the economic, social and environmental resilience drivers. The parameters are maximum likelihood estimators (ML), they retain the same signs obtained with the previous OLS, and the spatial lag is significant (p-value of $\lambda$ is less than 0.05) while the spatial error ($\rho )$ is not significant.

In the fourth column of Table 5 generalized two-stage least square (GL2SLS) estimators for the parameters of the spatial autoregressive model with spatial autoregressive disturbances are presented.⁶ The parameters retain the same signs of the previous regressions and the spatial lag is significant, indicating spatial autoregressive dependence in GDP pc.

When the GDP pc is regressed with the composite indicators of economic, social and environmental resilience drivers, using heteroscedasticity consistent estimators (GL2SLS), the parameter signs result positive for economic drivers but negative for environmental ones, confirming the presence of trade-off between economic and environmental dimensions, just observed in OLS equations (last column of Table 5). The linkage between GDP pc and social resilience drivers is negative but not significant. The spatial error is significant, indicating spatial autoregressive dependence in error term.

In general, the results confirm findings from other regional resilience studies, in particular the positive role on Gdp pc of gross fixed capital formation, share of employment in science and technology, research and development expenditures (Giannakis and Bruggeman 2017; Crescenzi et al. 2016), with the exception of graduate and lifelong learning effect, and the negative role of unemployment.

The relationships between the life expectancy and the drivers of economic, social and environmental resilience are presented in Table 6. The ordinary least squares present the same problems discussed for the GDP: multicollinearity, non-normality of residuals, heteroscedasticity. The spatial autoregressive models retain the same parameters signs obtained with the OLS except for employment in S&T sectors, woodland and population growth rate and the spatial and error lag are significant. Then the SAR model which implements maximum likelihood estimators to the synthetic indicators shows that the life expectancy is positively related with economic and social resilience drivers while the linkage with environmental resilience is not significant significant. So the social outcome, represented by life expectancy, seems to largely depend on the economic and social dimensions.

Table 7 presents the relationships between the ecological territorial outcome ($\hbox {CO}_2$ emissions) and the economic, social and environmental drivers of resilience. Considering the composite indicators, the inverse relation between $\hbox {CO}_2$ emission and economic resilience drivers emerges, as well as the direct one with the social and environmental dimensions. In order to overcome the problems of non-normality and multicollinearity, the spatial autoregressive models are used, displaying the same parameters signs obtained with the OLS. The last column of the table presents the spatial autoregressive model which implements generalized two-stage least square estimators to the synthetic indicators.

Table 6 Social regional resilience: outcome and drivers (dependent variable: life expectancy mean 2003–2012)

Table 7 Environmental regional resilience: outcome and drivers (dependent variable: $\hbox {CO}_2$ emissions mean 2003–2012)

5.1 The economic resilience outcome before and during the global crisis

As in the empirical analysis presented in previous paragraphs, also in this one some relationships between the GDP per capita, interpreted as an economic outcome of resilience, and inputs of economic and social resilience are analyzed.⁷ As in some recent papers (Crescenzi et al. 2016; Giannakis and Bruggeman 2017), the entire time period analyzed previously is subdivided in two parts: the pre-crisis period (2005–2007) that immediately preceded the global economic and financial shock and the post-2008 period (2008–2013). In each of these two periods the arithmetic average of GDP pc and of resilience indicators are the values inserted in the relations analyzed. The regression method based on the ordinary least squares (OLS) is the benchmark, and a spatial autoregressive model is also implemented

In both periods the ordinary least squares (Table 8) shows that the GDP pc is positively related with the three inputs of economic resilience: R&D expenditures, gross fixed capital per employee and graduates. The sign of these three parameters is as expected, but the R&D expenditures are not significant in statistical terms.

The GDP pc is negatively related with five input of social resilience: long-term unemployment, circulatory system death rate, transport accidents death rate, Neet, lifelong learning. The coefficient signs are the expected ones, with the exception of life long learning.

These OLS regressions are robust respect with residuals’ heteroscedasticity that are not normally distributed. The variance inflation factor (VIF) of each variable is acceptable according the prevalent rule of thumb proposed in empirical literature (Chatterjee and Hadi 2006).

As the cross-sectional observations under investigation are not independent of one another, spatial autoregressive models are applied. In these models, regressors with a statistically non-significant parameter (R&D expenditures, long-term unemployment) and that one with an unexpected sign (lifelong learning) are deleted.

In both periods physical and human capital are the most relevant drivers of resilience: these factors, which express the territorial availability of resources, grow in importance during the period when the crisis has manifested its effects, more than the adaptation capacity drivers, as the increasing magnitude of the coefficients indicates.

Table 8 Economic regional resilience: outcome and drivers (dependent variable: GDP per capita ppp, mean 2005–2007 and 2008–2013)

6 Conclusions

The paper analyzes the complex concept of regional resilience, adopting an holistic approach which distinguishes the three dimensions of sustainability (economy, society and environment) and outcome-driver variables. This approach is different than most of previous empirical studies, which primarily focus on economic performance in terms of income or employment dynamics during and after crisis cycles. The model is applied to the case of European regions, to get a geographical map of territorial resilience in its different dimensions.

Applying the principal component analysis, the drivers are synthesized in composite indicators and used for mapping the European regional resilience factors. Then they are used as regressors in both the OLS and the Spatial Autoregressive Model to deal with the problems of spatial lags, multicollinearity and non-normality of residuals.

The descriptive and econometric results show something expected and some new evidence.

The first expected evidence in the descriptive analysis is the well-known map of European economic and social well-being, which results more intensive in the metropolitan regions of the capitals and industrialized areas, penalizing conversely the Mediterranean regions of Spain, Italy and Greece, Portugal and Eastern Europe. But the costs of good economic and social performances are found in poor environmental resilience drivers in the Blue Banana regions, with the exception of Scandinavian area, capable of positive ecological balance.

The second expected result is the positive relation between territorial outcome and regional resilience drivers: so the economic drivers, such as innovation, investment and human capital, help to explain the level of economic well-being measured by gross domestic product per capita; the same for life expectancy, considered as proxy of social well-being, which is related with social resilience drivers such as low mortality rate, low unemployment level or social hardship; finally the drivers of environmental resilience, such as high level of biodiversity and low level of artificial areas, explain good ecological outcome, summarized in low level of emissions.

But the model also analyzes the cross and cumulative relationships among the three spheres of resilience in their outcome and drivers elements, with new emerging evidence through some econometric models. In the economic dimension, the social and environmental drivers of regional resilience are negatively related to economic outcome, highlighting a double trade-off, as if to say that the components of social and ecological welfare are in some cases obstacles or brakes to the wealth measured by income per capita. The same result is obtained in the environmental sphere, where the economic resilience drivers can worsen the state and the environmental balance. Only in the social dimension, the economic and social resilience drivers are cumulative and strengthen the welfare of society.

Starting from this analysis clear trade-offs emerge among the economic, social and ecological dimensions, highlighting the need for policies pointed to sustainability and equity. Territorial planning has to take into account two important goals, in order to achieve the conditions for sustainable development of local systems: the improvement of the ability to recover to adverse shocks and the containment of actions that could affect this capacity, causing a strong anthropic pressure on territorial assets. Empirical studies on vulnerability and resilience of regions represent useful instruments for the design of more effective hazard mitigation policies, oriented not to palliative but preventive measures.

First of all, place-based development policies are needed to address interventions on regional specificities, regarding the evidence of strong diversity across European regions. In this direction, the EU cohesion policies become essential to reduce the differences between countries and regions, especially in terms of infrastructure, education and green economy.

In particular, from the principal component analysis and the econometric test, the more relevant policy suggestions are the strengthening of innovation (R&D expenditures and employment in S&T sectors) and infrastructures for economic dimension, labor market interventions to contain unemployment and reinforce education for social dimension, and actions to defend biodiversity and woodland in environmental sphere in order to support economic, social and ecological resilience. But the emergence of trade-off between the three dimensions of sustainability requires integrated policies that aim at reaching an evolutionary balance of the various spheres of regional development. The emphasis on economic and industrial policies could actually facilitate economic recovery and resilience, but this would be at the expense of ecological and social balances.

From this point of view, an holistic representation of resilience capacity of territorial system and multi-hazard approach could represent a tool for policy making in order to achieve the system changes necessary for a sustainable territorial development (ICLEI 2016). Local governments and partners from the private sector have a great responsibility to invest in more resilient infrastructure, services and risk reduction strategies.

The next empirical work will explore linkages and reverse causality between resilience drivers and outcomes that will be represented by a broader batch of variables, connected with new themes such as social and institutional capital, inequality, environmental protection. The new steps of research would explore these cross-links with simultaneous equation models and latent variables, in order to better clarify the complex relationships and balances of territorial systems in the perspective of more targeted and balanced policies.

Appendix

See Tables 9 and 10.

Table 9 The selected variables of resilience in European Regions

Table 10 Economic, social, environmental Resilience Drivers Composite Indicator (rank)