Introduction
As the world heats up from political fragmentation, economic restructuring, climate change, and technological advancements, we must continue to centre the most marginalized and oppressed communities within and between nations in our discourse and analysis. The concept of social inclusion offers a salient opportunity to do so. Social inclusion has been the focus of many social policy regimes since the late 20th century, such as in Australia and Europe. This moment of increasing global pressure demands the revisitation of the theoretical and material realization of social inclusion, which may offer insight and opportunity to promote the welfare of the world’s excluded and left behind.
Yet, little is known about how social inclusion manifests in disparate corners of the globe and how social inclusion policy mitigates or perpetuates exclusion. In this study, we seek to build on recent theoretical developments related to the concept of social inclusion by considering global social inclusion through the lens of welfare regime theory. While Aspalter (Reference Aspalter2023) proposed a 10-fold typology of welfare regimes, the present study identifies a more parsimonious six-fold typology of social inclusion regimes – Low, Mid, and High Social Inclusion and Low, Mid, and High Social Exclusion – when accounting for theoretically relevant social, economic, and political social forces. The six-fold typology strikes a balance between capturing the diversity of social inclusion outcomes across countries and maintaining a more manageable and interpretable set of categories (Everitt et al. Reference Everitt, Landau, Leese and Stahl2011). This parsimony offers advantages such as facilitating straightforward comparisons, easier application in policy and practice contexts, and alignment with the principles of scientific simplicity (Baily Reference Baily1994; Esping-Andersen Reference Esping-Andersen1999). However, it is important to acknowledge the trade-offs between parsimony and complexity. While the six-fold typology offers greater simplicity, it may not capture the full nuance of social inclusion regimes to the same extent as Aspalter’s (Reference Aspalter2023) 10-fold typology. Future research could explore the relative merits of different levels of complexity in social inclusion regime typologies and their implications for theory, research, and policy (Emmenegger et al. Reference Emmenegger, Kvist and Marx2015; Ferragina and Seeleib-Kaiser Reference Ferragina and Seeleib-Kaiser2011).
This effort has both scholarly and practical applications. In this study, we problematize and advance welfare regime theory by adding further nuance to existing typologies. We take a data-centred, inductive approach to typologizing that also allows us to test existing Weberian ideal welfare regimes, which refers to the application of Max Weber’s concept of ideal types to the study of welfare systems, involving the construction of theoretical models that represent the essential characteristics of different welfare regimes. However, as noted by Ferragina and Seeleib-Kaiser (Reference Ferragina and Seeleib-Kaiser2011) and Aspalter (Reference Aspalter2019), Esping-Andersen implicitly referred to Weber’s methodology in his work, although his approach to constructing welfare regime typologies can be seen as an application of Weberian ideal types as explicitly referenced earlier in Esping-Andersen’s work (Esping-Andersen Reference Esping-Andersen, Rein and Esping-Andersen1987a; Esping-Andersen Reference Esping-Andersen, Rein and Esping-Andersen1987b). Aspalter (Reference Aspalter2020) provides a comprehensive guide on how to apply ideal types in research on comparative social policy. The use of Weberian ideal types in welfare regime research has been subject to debate and criticism, with some scholars arguing that it oversimplifies the complexity and diversity of real-world welfare systems and may not adequately capture the dynamics of change over time (Arts and Gelissen Reference Arts and Gelissen2002; Powell and Barrientos Reference Powell and Barrientos2011). Nonetheless, ideal types remain a useful analytical tool for comparative research, providing a framework for understanding the key dimensions along which welfare systems vary (Aspalter Reference Aspalter2019; Kuhnle and Sander Reference Kuhnle, Sander, Castles, Leibfried, Lewis, Obinger and Pierson2010). While not without its limitations, this approach has been influential in shaping comparative welfare state research and continues to inform debates about the nature and diversity of welfare systems across countries (Aspalter Reference Aspalter2019; Ferragina and Seeleib-Kaiser Reference Ferragina and Seeleib-Kaiser2011).
We build on the work of Abu Sharkh and Gough (Reference Abu Sharkh and Gough2010) by including as many countries as possible in our analysis. Following the recommendation of the United Nations Economic Commission for Europe (UNECE) (2022), we used the United Nation’s Sustainable Development Goals (UNSDG) as a guiding framework for measuring social inclusion and constructed a social inclusion index for 225 countries between roughly 2000 and 2020 using the UNSDG Global Database. Moreover, we connect social inclusion with welfare regime theory, adding a novel layer of analysis to both. This effort can allow nations and international organizations to understand the extent and distribution of social inclusion more clearly across the globe, which may allow for the more effective amelioration of social exclusion and further explanation for why the incidence of social inclusion varies so dramatically between nations.
As yet, there is no systematic ideal-typical classification of social inclusion, despite the successful categorization of welfare state systems into families of nations (Aspalter Reference Aspalter2006, Reference Aspalter2011, Reference Aspalter2012; Esping-Andersen Reference Esping-Andersen1990, Reference Esping-Andersen1999). This study focuses on social inclusion as a distinct research object, conceptualized as a multidimensional concept encompassing access to resources, participation in social and economic life, and a sense of belonging (MacLeod et al. Reference MacLeod, Ross, Sacker, Netuveli and Windle2019). We develop a theoretically informed social inclusion index that captures a broader set of societal factors beyond specific welfare arrangements, moving the concept of analysis beyond poverty to a more encompassing social process (Aspalter Reference Aspalter2014). Drawing on Luhmann’s (Reference Luhmann, Bednarz and Baecker1995) systems theory perspective, we view social inclusion as an emergent property arising from the complex interplay between various functional subsystems in society (Stichweh Reference Stichweh2021). To investigate social inclusion across countries, we employ a two-step methodology. First, we operationalize social inclusion by constructing a novel index based on 10 indicators selected from the UNSDG framework (United Nations Department of Economic and Social Affairs (UNDESA), 2023). We then develop this social inclusion index using principal component analysis (PCA) to reduce the dimensionality of the 10 selected indicators into a single composite measure. This index allows us to quantify and compare levels of social inclusion across 225 countries from 2000 to 2020. Second, we use two clustering techniques to identify distinct groups of countries based on their social inclusion index scores and other theoretically relevant variables (social expenditure, government effectiveness, and control of corruption). These clustering methods enable us to uncover patterns of similarity and difference in social inclusion outcomes across countries and to explore how these patterns relate to welfare regime typologies and other societal factors. By combining index construction and cluster analysis, our study aims to provide a comprehensive and nuanced understanding of social inclusion as a multidimensional phenomenon shaped by the complex interactions of various societal subsystems. This approach moves beyond the focus on specific welfare regimes to capture the broader set of factors that influence inclusion outcomes in different country contexts, ultimately contributing to the development of a systematic ideal-typical classification of social inclusion regimes.
Through the lens of social inclusion
Social inclusion is a multidimensional, multilevel, dynamic, and relational concept constituting both a process and a distinct outcome. With roots in the French Republican ideology of the 1970s (Lenoir Reference Lenoir1974; Silver Reference Silver1994; Silver Reference Silver2015), the concept of social inclusion has evolved from restricted labour force participation and material deprivation to fuller participation in society. Following the conceptual framework advanced by MacLeod et al. (Reference MacLeod, Ross, Sacker, Netuveli and Windle2019), we conceptualize social inclusion across multiple domains – including environment and neighbourhood, civic and cultural, economic, social relations and resources, service provision and access, and health and well-being – spanning multiple systematic levels.
Where social inclusion represents access and participation across societal domains, social exclusion represents the negative outcomes related to health, education, economic inequality, poverty, violence, and well-being (Khan et al. Reference Khan, Combaz and McAslan Fraser2015); the abandonment of mainstream norms (LaFree Reference LaFree1998; Liebow Reference Liebow2003); the development of subcultures (Hagan and McCarthy Reference Hagan and McCarthy1998); non-participation in the labour market (Atkinson and Hills Reference Atkinson and Hills1998); and withdrawal from social and political life (Putnam Reference Putnam2000) associated with barriers to access and participation. Social exclusion seems to harm socially marginalized groups, such as older adults (Nyqvist et al. Reference Nyqvist, Häkkinen, Renaud, Bouchard and Prys2021), people in workfare or public works programs (Girardi et al. Reference Girardi, Pulignano and Maas2019), children (Gross-Manos Reference Gross-Manos2017; Koller et al. Reference Koller, Pouesard and Rummens2018), people who live in rural areas (Walsh et al. Reference Walsh, O’Shea and Scharf2020), and those with developmental disabilities or mental illness (Koller and Stoddart Reference Koller and Stoddart2021; Wright and Stickley Reference Wright and Stickley2013), the most.
Social inclusion also seems to be relative, leading many scholars to develop national typologies of social inclusion. Silver (Reference Silver1994) posited three paradigms of social exclusion – solidarity, specialization, and monopoly. The solidarity approach draws on French Republican thought and attributes social exclusion to the breakdown of social solidarity, the specialization paradigm draws on Anglo-Saxon notions of social differentiation because of individual specialization in the labour market, and the monopoly paradigm views exclusion because of hierarchical group monopolies and their exertion of power through class and status that are remedied through social citizenship. Silver (Reference Silver2015) developed these ideas further by advancing refined typologies where Liberal thought conceptualized social inclusion as guaranteed rights to individual freedoms; Republican thought, social solidarity; Social Democratic thought, rights to a minimum standard of living; Conservative thought, natural hierarchy; and Confucian thought, social harmony over individual freedoms. Gidley et al. (Reference Gidley, Hampson, Wheeler and Bereded-Samuel2010) offered another approach framed by ideologies of neoliberalism, social justice, and human potential (Gidley et al. Reference Gidley, Hampson, Wheeler and Bereded-Samuel2010). Lyons and Huegler (Reference Lyons, Huegler, Healy and Link2012) developed another framework where societies take either the Moral Underclass (MUD) approach, Social Integrationist (SID), or Redistributive (RED). MUD highlights the moral and cultural failings of the individual rather than structural issues in society; SID emphasizes social exclusion as exclusion from paid work; RED prioritizes the redistribution of power and wealth. Scholars like Chau et al. (Reference Chau, Yu and Boxall2018) have also added collective production as a fourth approach, characterized by co-ownership of the means of production, agenda, and production process.
While these scholars have made insightful theoretical contributions to understanding how social inclusion is conceptualized in various national contexts, less is known about how relative understandings of the concept relate to differences in tangible social inclusion policy and welfare regimes. These scholars suggest that social inclusion and social inclusion policy in each national context can generally be characterized by one of these regime types, distinguished by approaches such as workforce activation or redistribution, which share considerable conceptual territory with theories of welfare regime typologies. Regardless of context, social inclusion policy aims to promote the social inclusion of the general population or a specific subpopulation, though occasionally not explicitly. As such, we consider social inclusion policy to incorporate many types of social policy.
Perspectives on welfare regime typologies
Esping-Andersen (Reference Esping-Andersen1990) popularized welfare regime typologies through his three regimes of welfare capitalism based, in part, on the decommodification of labour: the liberal, conservative, and social democratic welfare regimes. In this framework, liberal regimes are characterized by modest, means-tested assistance that emphasizes market-based solutions to social issues; conservative regimes, by family-based assistance that emphasizes traditional family values where government steps in as a last resort; and social democratic regimes, by universalistic assistance that emphasizes the decommodification services.
In their overview of the field of comparative social policy, Kennett (Reference Kennett2013) argued that globalization would change the focus of the field, claiming that the field has evolved from focusing on the welfare programs of nation-states to more broadly examining the role of culture, institutions, informal support, and ideas, as well as critical perspectives on social policy. Despite these burgeoning streams in the literature, Esping-Andersen’s (Reference Esping-Andersen1990) ideal-type typologies continue to dominate the field.
The literature on welfare regimes has evolved significantly since Esping-Andersen’s (Reference Esping-Andersen1990) seminal work, with numerous studies proposing substantive innovations and critical perspectives (Arts and Gelissen Reference Arts and Gelissen2002; Ferragina and Seeleib-Kaiser Reference Ferragina and Seeleib-Kaiser2011). Castles (Reference Castles1993) introduced the concept of “family of nations,” arguing that countries with shared cultural, historical, and geographic characteristics tend to develop similar welfare regimes. Gallie and Paugam (Reference Gallie and Paugam2000) examined the relationship between welfare regimes and the experience of unemployment in Europe, highlighting how different welfare arrangements shape the social and economic consequences of job loss.
Goodin et al. (Reference Goodin, Headey, Muffels and Dirven1999) provided a comparative analysis of the “real worlds of welfare capitalism,” moving beyond Esping-Andersen’s ideal types to explore the complex realities of welfare provision across different countries. They emphasized the importance of considering the interplay between welfare states, labour markets, and family structures in shaping social outcomes. Critical perspectives on welfare regime typologies have also emerged. Kasza (Reference Kasza2002) argued that the concept of welfare regimes creates an “illusion” of coherence and homogeneity within countries, obscuring the complexity and diversity of welfare arrangements across different policy domains. Bambra (Reference Bambra2006) and Scruggs (Reference Scruggs2006) raised methodological concerns about the operationalization and measurement of welfare regimes, calling for greater transparency and robustness in the construction of typologies, while Shaleve (Reference Shaleve2007) proposed an alternative approach to welfare regime classification based on the concept of social stratification, emphasizing the need to consider how welfare states shape and reproduce social inequalities.
Where a myriad of scholars has made incremental additions to Esping-Andersen’s (Reference Esping-Andersen1990) typologies, such as Ferrera’s (Reference Ferrera1996) model of southern European countries or Lee and Ku’s (Reference Lee and Ku2007) model of East Asian countries, Aspalter (Reference Aspalter and Aspalter2017); Aspalter Reference Aspalter2023) has advanced a synthesizing global framework of 10 worlds of welfare capitalism, building on the foundation of Esping-Andersen’s (Reference Esping-Andersen1990) framework. These include the Social Democratic regime in Scandinavia, the Christian Democratic regime in most of Continental Europe, the Neoliberal regime in Anglo-Saxon countries, the Pro-welfare Conservative regime in East Asia, the Anti-welfare Conservative regime in Latin America, the Slightly Universal Rudimentary regime in South Asia, the Ultra Rudimentary regime in Most of Africa, the Exclusion-Based regime in the Oil-Exporting Gulf States, the Selective Rudimentary regime in Northern/Central Asia and the Far East of Europe, and the Communist/Socialist Universal regime in Cuba (for a more thorough description of these regimes, see Aspalter’s (Reference Aspalter2023) Table 3.1 The Ten Ideal–Typical Worlds of Welfare Regimes (part 1).
Interestingly, Aspalter (Reference Aspalter2023) seems to largely neglect other factors that may influence social inclusion and welfare regimes like countries’ demographic, social, economic, and political influences. Drawing on theories of welfare state development, other social forces likely influence both social inclusion and welfare regimes. Three major groups of theories of welfare state development exist in the literature – industrialization, state-centric, and class struggle (also known as the power resources approach) (Huber and Stephens Reference Huber, Stephens, Smelser and Swedberg2010). Advanced by foundational thinker Wilensky (Reference Wilensky1975) and historian Katz (Reference Katz1996), the industrialization thesis posits that countries develop welfare states as industrialization, economic growth, and demographic changes upend social norms and lead to government intervention. On the other hand, state-centric thinkers like Skocpol (Reference Skocpol1995) and Cruikshank (Reference Cruikshank1999) advance a perspective that emphasizes national and institutional politics, the balance of power between state institutions, and the relationship between governance and the welfare state. Finally, another group of theorists (e.g., Castles Reference Castles1982; Korpi Reference Korpi1989) posits that welfare states are influenced by the balance of power in society between political parties, labour groups, and the lower class.
Social exclusion and welfare regimes
The relationship between social exclusion and welfare regimes has been a subject of scholarly inquiry, with various studies exploring the complex interplay between these two concepts. Tsakloglou and Papadopoulos (Reference Tsakloglou and Papadopoulos2002) demonstrated that country and welfare regime effects significantly influence the probability of social exclusion, emphasizing the role of institutional and policy factors in shaping social inclusion outcomes. This finding is further supported by Ogg (Reference Ogg2005), who found that developed welfare regimes are associated with lower rates of social exclusion among older Europeans, highlighting the protective function of comprehensive social policies. Moreover, the impact of welfare regimes on social exclusion is not limited to specific age groups, as evidenced by Whelan et al. (Whelan and Maître Reference Whelan and Maître2010), who discovered systematic variations in economic vulnerability levels across different welfare regimes. This study underscores the broader implications of welfare arrangements for poverty and vulnerability, suggesting that the design and generosity of social policies can either mitigate or exacerbate social exclusion. In addition to objective measures of social exclusion, welfare regimes also shape public attitudes and opinions regarding social welfare. Jakobsen (Reference Jakobsen2011) explored the relationship between welfare attitudes, social expenditure, and welfare regimes, revealing how regime types influence public support for social policies. This research highlights the importance of considering the normative and ideological dimensions of welfare regimes in understanding their impact on social exclusion.
The limitations of defining social deprivation as exclusion from social norms, as discussed by Bowring (Reference Bowring2000), further emphasize the need for a more nuanced and contextualized understanding of social exclusion. This critique suggests that the relationship between welfare regimes and social exclusion cannot be reduced to a simple dichotomy of inclusion versus exclusion based on adherence to dominant social norms. The complexity of the relationship between social exclusion and welfare regimes is further highlighted by studies that examine the role of local policies and practices. Political and economic factors also play a significant role in shaping welfare regimes and their impact on social exclusion. Eibl (Reference Eibl2020) discussed the relationship between intra-elite conflict and social policies and spending, highlighting how external threats can influence social expenditure. This research suggests that the dynamics of power and resource allocation within societies can have profound implications for the design and implementation of welfare policies. The socioeconomic context of countries also influences attitudes towards social inclusion and exclusion. Fernandez-Barutell (Reference Fernandez-Barutell2021) examined the association between poverty/social exclusion and support for excluding immigrants from social services in Eurozone countries, revealing the impact of socioeconomic factors on welfare nationalism and exclusionary practices.
The examination of social inclusion typologies through the prism welfare regimes has been instrumental in shedding light on the dynamics of social welfare systems and their significant impact on the processes of social inclusion and exclusion. However, a critical review of the recent contributions to this field, particularly by Aspalter (Reference Aspalter2023), reveals a notable oversight of the broader array of factors – ranging from demographic shifts to economic dynamics and political structures – that profoundly influence the configurations and outcomes of welfare regimes. This critique points towards the necessity of adopting a more comprehensive analytical lens to fully capture the array of determinants that shape welfare policies and their implications for social inclusion.
The present study
In the present study, we test the premise of Aspalter’s (Reference Aspalter2023) 10 worlds of welfare capitalism and their theory against the concept of social inclusion, a previously unexamined indicator to the authors’ knowledge. Aspalter (Reference Aspalter2023), in fact, has tested their typologies with metrics of povertization and inequality, but while we find these concepts useful, they inadequately address our conception of social inclusion.
Aspalter (Reference Aspalter2023) selected measures of access to clean water, hygiene conditions, mortality rates, the incidence of certain illnesses, homicide rate, and corruption perception to represent povertization, which seems to mean the theft of a “decent” life as associated with the lack of resources and opportunities. Aspalter (Reference Aspalter2023) also includes a measure of overall economic inequality to capture relative poverty. While, collectively, these variables constitute an interesting take on multidimensional poverty, they stop short of representing the concept of social inclusion by measuring only certain aspects of poverty consequently neglecting other domains of social inclusion like social relations and service provision.
While poverty and inequality are valid metrics, they only represent a component of the larger concept of social inclusion. In turn, this study develops a theoretically informed, novel social inclusion index to that captures a broader set of societal factors beyond specific welfare arrangements. Our index moves the concept of analysis beyond poverty to a more encompassing social process, covering multiple domains of social inclusion, such as environment and neighbourhood, civic and cultural, economic, social relations and resources, service provision and access, and health and well-being. This multidimensional approach allows for a more nuanced examination of social inclusion outcomes across countries, distinct from the focus on welfare regimes alone. This study aims to answer the following questions: Does social inclusion constitute an appropriate indicator for grouping welfare regime typologies?; And in terms of social inclusion, are there ways to classify countries based on social inclusion other than the 10 worlds of welfare capitalism?
These questions problematize the existing paradigm of welfare state typologizing by moving the concept of analysis beyond poverty to a more encompassing social process that may lead to poverty. Further, by using an index of social inclusion to inductively build typologies, we are testing Aspatler (Reference Aspalter2023) and other welfare regime typologists’ theories of ideal classifications, which may lead to new typologies or the refinement of current regimes, following Ferragina et al.’s (Reference Ferragina, Seeleib-Kaiser and Spreckelsen2015) approach of examining the “reality” behind the ideal constructs.
To answer our questions, we employ various clustering algorithms to inductively group our social inclusion index. We build on previous work by including as many countries as possible across as many time points as possible as available, diverging from previous studies (Abu Sharkh and Gough Reference Abu Sharkh and Gough2010).
Practically speaking, this study adds a new social inclusion index to our repertoire for analysing global trends and differences in social inclusion and social inclusion policy, which can allow for a clearer understanding of its material manifestations and more effective resource redistribution among nations and international organizations. Moreover, the possibility of new or refined welfare regime typologies may further this purpose by augmenting our understanding of which countries experience the most social exclusion and building towards explanations of why certain countries experience more than others.
Methods
Analytic approach
We first developed a social inclusion index by employing PCA to reduce dimensions of social inclusion to a single index for 225 countries across roughly the years 2000 through 2020. PCA and other forms of cluster analysis have already been established in previous research in comparative social policy (e.g., Ferragina et al. Reference Ferragina, Seeleib-Kaiser and Spreckelsen2015; Ferragina et al. Reference Ferragina, Seeleib-Kaiser and Tomlinson2013). This approach is intentionally descriptive rather than prescriptive, pointing to the multidimensional and fuzzy nature of the conceptual boundaries scholars ascribe to states and welfare regimes (Shaleve Reference Shaleve2007). We then used a one-way analysis of variance (ANOVA) to test whether average index values varied between welfare regimes. We used univariate k-means clustering to determine optimal clustering within the index alone and then model-based cluster analysis to develop optimal clustering when accounting for the other theoretically informed social forces. To our knowledge, these two clustering approaches have yet to be utilized in the welfare regime literature, though the precedent was established with other clustering methods such as k-means and hierarchical clustering (Abu Sharkh and Gough Reference Abu Sharkh and Gough2010). All data management, analysis, and visualization were conducted in R and are freely available on the authors’ GitHub repo.
Measures of the social inclusion index
In the absence of a national measure of social inclusion, we followed the recommendations of UNECE (2022) by using the UNSDG as a guiding framework for measuring social inclusion. We began by reviewing the more than 210 UNSDG indicators in the United Nation’s Sustainable Development Goals Databse (UNDESA 2023) and comparing them to the theoretical framework for social inclusion posited above along MacLeod et al.’s (Reference MacLeod, Ross, Sacker, Netuveli and Windle2019) multiple dimensions. These data come from various international organizations including the World Bank, Organization for Economic Co-operation and Development (OECD), and the International Labour Organization.
We selected 10 measures based on their representation of at least one domain of social inclusion. While other measures may be tested in future studies, these 10 were selected because, collectively, they covered all the expressed domains of social inclusion, most other variables would only serve as proxies to these variables, and because many more countries reported values for them. Data were collected for each country (of which there were 225 unique countries) for the interval from 2000 to 2022 as available. Each country was then matched with its corresponding welfare regime as described by Aspalter (Reference Aspalter2023). A full description of each indicator and our selection criteria can be found in Appendix A.
Additionally, a measure of social expenditure as a percentage of gross domestic product (GDP) from 2000 to 2019 was developed to test the industrialization theory following Hong and Ngok (Reference Hong and Ngok2022) and Wilensky (Reference Wilensky1975). Social safety net expenditures were collected from The World Bank (2023a) and the OECD (2023), representing total social safety net and social assistance program expenditures per person including benefits and administrative costs as a percentage of GDP. Instead of using two variables for country GDP and population size like Hong and Ngok (Reference Hong and Ngok2022), we combined them into a single GDP per capita variable adjusted to real GDP for the US dollar in 2023 (The World Bank 2023b). This variable measures the average per capita post-transfer from all social protection and labour across all available years between 2000 and 2020.
Measures of government effectiveness and control of corruption were also added for each country from 2000 to 2021 as proxies for the State-centric and Class Struggle theories (Worldwide Governance Indicators 2022). In the absence of standard measures for these theories that can be applied to nearly every country, we chose a measure of government effectiveness to represent state capacity (Weir et al. Reference Weir, Orloff and Skocpol1988; Cruikshank Reference Cruikshank1999). Barring a suitable measure for the class struggle theory that would apply to all countries, we elected for a measure of government corruption instead. While measures of government corruption have not traditionally been used in this line of theory, we argue that it serves as a proxy by representing the balance of power between capital and the public in government, particularly because this specific variable measures perceptions of the “capture" of the state by elites and private interests. These variables can be seen below in Table 1.
Table 1. Variable descriptions.
Principal component analysis
A reduced measure of social inclusion was then developed using PCA (Labrín and Urdinez Reference Labrín and Urdinez2021). First, means were calculated for each country and variable by averaging the values of each variable by country and year. This resulted in a mean estimate for each variable and country that aggregated across however many years of data were collected from UNSDG (2023), leaving values for the 225 countries in the dataset. Each variable was then normalized and centred. We then exclude Social Expenditure, Government Effectiveness, and Control of Corruption from the analysis. The optimal number of PCA dimensions was found using the estim_ncpPCA function from the missMDA package in R using k-fold cross-validation, which resulted in a single optimal dimension. We then used the imputePCA function from the missMDA package in R to impute missing values using the regularized iterative PCA algorithm using one component. While imputation invariably introduces additional bias to the data, complete data is needed for PCA. The regularized iterative PCA algorithm works by imputing missing values with initial values from the variable’s mean, then performing PCA on the imputed dataset, and finally inputting the missing values through iteration based on the fitted matrix until convergence (Josse and Husson Reference Josse and Husson2013, Reference Josse and Husson2016). With the complete dataset, we then conducted PCA using the FactoMineR package in R, matching the first component of the PCA (which explained 54% of the total variance) with their respective countries. Figure 1 and Figure 2 in Appendix A illustrate the eigenvalues of each component and the quality of representation of each variable to the first dimension, respectively.
Figure 1. Optimal univariate k-means clustering by country.
Figure 2. Model-based cluster densities.
Cluster analysis
It is essential to recognize that the clustering techniques used in this study (univariate k-means and model-based clustering) are reductionist methods that group data points based on similarities, simplifying some nuances to identify broad patterns (Everitt et al. Reference Everitt, Landau, Leese and Stahl2011; Kassambara Reference Kassambara2017). Cluster analysis is a data-driven approach that partitions a heterogeneous population into more homogeneous subgroups, focusing on similarities between observations and minimizing within-cluster variation while maximizing between-cluster variation (James et al. Reference James, Witten, Hastie and Tibshirani2013; Kaufman and Rousseeuw Reference Kaufman and Rousseeuw2009). The reductionist nature of cluster analysis can lead to a loss of information and oversimplification of the complexity of social reality (Castellani Reference Castellani2018). The resulting clusters are not definitive categories but rather a way of organizing and interpreting the data based on selected variables and clustering algorithms (Fraley and Raftery Reference Fraley and Raftery1998). Different methods may yield different groupings, and the interpretation of clusters requires domain knowledge and theoretical insight (Han et al. Reference Han, Kamber and Pei2012; Kassambara Reference Kassambara2017). Cluster analysis differs from the use of Weberian ideal types, which are conceptual constructs designed to inform theoretical insights and guide comparative analysis (Weber Reference Weber, Shils and Finch1949; Rosenberg Reference Rosenberg2016). While cluster analysis is data-driven and seeks to uncover empirical groupings, ideal types are abstract, hypothetical constructs that accentuate certain elements of reality to facilitate comparative analysis and theory development (Swedberg Reference Swedberg2018). In this study, we aim to identify data-driven patterns of social inclusion that can complement and problematize existing welfare regime frameworks, recognizing the limitations of reducing complex social realities to discrete categories. The clusters identified should be viewed as exploratory and provisional groupings that can inform further theoretical development and empirical investigation. Future research should validate and refine these clusters using different methods and data sources and explore the underlying mechanisms and processes that give rise to these patterns of social inclusion (Olsen Reference Olsen and Holborn2004).
We took two approaches to cluster analysis – optimal univariate k-means clustering and model-based clustering – which allowed for the exploration and comparison of regime types as they relate to the social inclusion index. Since both clustering methods rely on a Gaussian distribution, we verified that the index was distributed normally.
We used optimal univariate k-means clustering to determine the optimal clustering of the index using Wang and Song’s (Reference Wang and Song2011) dynamic programming Ckmeans.1d.dp function in R. This algorithm works by optimally assigning values to clusters by minimizing their within-cluster sum of squared distance. The optimal number of clusters was selected by the algorithm using Bayesian information criteria. Essentially, this mode of clustering identifies potential subgroups within the distribution of a variable. Two clusters were selected and the threshold between the clusters was calculated at −0.159. Figure 1 illustrates the division between the clusters by country.
We also used model-based clustering to develop optimal clustering of the index along with Social Expenditure, Government Effectiveness, and Control of Corruption (McNicholas Reference McNicholas2016). Social Expenditure was logarithmically transformed and then all variables were normalized and centred so that each variable had a normal distribution. Because this clustering method requires no missing data, we inputted missing values for the other variables based on each variable’s mean (18 values for Social Expenditure, 18 for Government Effectiveness, and 18 for Control of Corruptoin), which again introduced bias and reduced sensitivity. We then used the Mclust function in R to estimate optimal clustering between the index and other variables based on parameterized finite Gaussian mixture models (Scrucca et al. Reference Scrucca, Fop, Murphy and Raftery2016). The clusters were estimated using the EM algorithm and the optimal model was selected according to Bayesian information criterion. Three clusters were selected, and Figure 2 illustrates their classification across the variables. We follow the work of Ferragina et al. (Reference Ferragina, Seeleib-Kaiser and Spreckelsen2015) in Appendix B by presenting a sensitivity analysis of the model-based clustering method by iteratively randomly selecting countries and indicators in the model 10,000 times.
We then used a one-way ANOVA to determine whether mean index values varied across welfare regimes. Although normality could be assumed, homogeneity of group variance was violated. In addition to accounting for this in the ANOVA, we used the Games–Howell post hoc for this reason.
Findings
The social inclusion index seems to have performed well as a dimensional reduction of the 10 variables selected as indicators of social inclusion. As verified by manually comparing countries’ index values with the values of the other 10 variables, the higher the PCA value, the greater the social inclusion, and the lower the PCA value, the greater the social exclusion. Table 2 reports the mean, standard error, and standard deviation of each of the 10 social inclusion indicators before missing data were imputed. Table 3 shows the mean social inclusion index, and other variables by welfare regime after missing data were imputed, which seem to vary significantly between regimes. The Socialist/Communist regime did not report standard errors of standard deviations across the variables because the regime only included one country – Cuba.
Table 2. Descriptive statistics by welfare regime.
* Missing value because of too few data.
Table 3. Social inclusion index and social forces by welfare regime.
* Missing value because of too few data.
The one-way ANOVA without equal variances assumed found that there was a statistically significant difference between the mean estimates of the social inclusion index between welfare regimes (F(9, 32.611) = 84.115, p < .001). The Games–Howell post hoc test revealed statistically significant differences between regimes whose standard errors did not overlap (see Table 2 and Figure 3).
Figure 3. PCA by univariate k-means cluster.
In terms of the index, there appears to be six groups among the welfare regimes demarcated by the relative threshold generated by the univariate k-means clustering – low social inclusion, mid social inclusion, and high social inclusion, and low social exclusion, mid social exclusion, and high social exclusion. Generally, regime types with greater values for Social Expenditure, Government Effectiveness, and Control of Corruption seem to have greater social inclusion in terms of the index, but with the exceptions of the Exclusion-Based, Selective Rudimentary, and Unclassified regime types. Despite high values for the other variables, the Exclusion-Based regime had low-to-mid social inclusion. On the other hand, despite low values for the other variables, the Selective Rudimentary regime had mid-to-high social inclusion. As an amalgamation of dozens of disparate countries, the Unclassified regime type averaged to relatively high values of the other variables but low social inclusion. Table 4 helps evince these findings by showing the means, standard errors, and standard deviation for each cluster from both clustering methods, showing that the group averages for both clustering methods reveal distinct groups with non-overlapping standard errors.
Table 4. Social inclusion index and social forces by cluster.
* Unsure because of missing data.
As shown in Figure 3, the optimal univariate k-means clustering algorithm split the index into two groups at the −0.159 index threshold. We can interpret this as representing low and high social inclusion. On average, the Extremely Rudimentary and Slightly Universal regimes fall into the low social inclusion group, and the Anti-Welfare Conservative and Unclassified regimes are on the group’s margin. All other regime types fall into the high social inclusion group. However, there are exceptions to this average where some countries within a regime classified as low or high social inclusion fall into the other category.
Where the optimal univariate k-means clustering algorithm optimized for two groups, the model-based clustering algorithm selected six groups when factoring in Social Expenditure, Government Effectiveness, and Control of Corruption. We can interpret these groups as low social inclusion or low social exclusion with low values of covariates, mid social inclusion or mid social exclusion with mid values of covariates, and high social inclusion or high social exclusion with high values of covariates. Figure 4 illustrates the overlap between regimes and model-based clusters. Most countries in Cluster 1 were classified as Extremely Rudimentary, Selective Rudimentary, or Slightly Universal regimes, though the Extremely Rudimentary and Selective Rudimentary regimes lean slightly towards Cluster 2; Cluster 2 primarily consists of countries in the Anti-Welfare Conservative and Pro-Welfare Conservative regimes, though the Anti-Welfare Conservative regime leans slightly towards Cluster 1 and the Pro-Welfare Conservative regime leans slightly towards both Cluster 1 and Cluster 3; and Cluster 3 is made up of the Exclusion-Based and Socialist/Communist regimes. On the other hand, the Christian Democratic, Neoliberal, and Social Democratic regimes are primarily in Cluster 4. Cluster 5 seems to consist mainly of unclassified countries and Saudi Arabia, which has been classified as an Exclusion-based regime, and Cluster 6 has a few countries from the Extremely Rudimentary regime and other unclassified countries.
Figure 4. Welfare regime by model-based cluster.
We should note that the cluster names do not appropriately correspond with gradations in the social inclusion index. Table 5 orders the clusters in terms of the social inclusion index and simplifies the variable averages with the distinction of Low, Mid, or High.
Table 5. Social inclusion index by social inclusion regime.
* Unsure because of missing data.
Discussion
While it has many limitations, our aggregated PCA indicator offers a pragmatic method of comparing social inclusion between countries. In terms of validity, our social inclusion index correctly matched values across the 10 contributing variables to create a measure where the lower the value, the greater the social exclusion, and the higher the value, the greater the social inclusion. Moreover, our social inclusion index accurately classified welfare regimes so that the Social Democratic regime had the highest mean social inclusion, and the Extremely Rudimentary regime had the lowest mean social inclusion. When grouped by welfare regime, we saw noticeable differences in group averages, which lend support to the notion that welfare regimes may not only be an appropriate ideal distinction between countries but also represent real differences in material social inclusion.
However, because we used PCA, we had to impute missing values for countries that did not report values for one of the 10 contributing variables. While these values were derived iteratively through PCA based on normal distributions, imputation undoubtedly introduced bias. We attempted to mitigate this bias by aggregating individual-level country values at the group mean, whether that was grouped by welfare regime or cluster.
Our social inclusion index allowed us to use univariate k-means clustering and model-based clustering algorithms to develop an optimal number of groups, irrespective of the ideal guideposts of welfare regime type. When only referencing the social inclusion index, two groups were elucidated, representing a low social inclusion group and a high social inclusion group. When referencing both the social inclusion index and social expenditure by GDP per capita, perceptions of government effectiveness, and perceptions of government corruption, the model-based clustering algorithm optimized for six groups, representing Low, Mid, and High Social Inclusion and Low, Mid, and High Social Exclusion. Importantly, while there were cases of countries deviating from these three groups, we found a general trend where countries with higher social expenditure by GDP per capita, higher perceptions of government effectiveness, and lower perceptions of government corruption had higher social inclusion.
While our findings support Aspalter’s (Reference Aspalter2023) notion of the 10 worlds of welfare capitalism and the theory of welfare regime typologies broadly, our findings reveal far fewer optimal typologies than previously hypothesized when drawn inductively from our social inclusion index. While we found differences in our social inclusion indicator between some welfare regimes, many regime types saw no generalizable differences. In effect, the 10 welfare regimes advanced by Aspalter (Reference Aspalter2023) were reduced to two groups when only considering the social inclusion index (based on low or high social inclusion) and six groups when considering the social inclusion index alongside social expenditure by GDP per capita, perceptions of government effectiveness, and perceptions of government corruption. Moreover, these covariates consistently showed statistically significant relationships with the social inclusion index.
In turn, in terms of social inclusion as an outcome of social policy, our findings advance six typologies that add nuance to extant welfare regime theory: Low Social Inclusion, Mid Social Inclusion, and High Social Inclusion regimes and Low Social Exclusion, Mid Social Exclusion, and High Social Exclusion regimes. As Table 5 illustrates, based on our cluster analysis, we reclassify the Christian Democratic, Neoliberal, and Social Democratic as High Social Inclusion; an unclassified group of countries (see Table 1 in Appendix B for a complete list of countries) as Mid Social Inclusion; and Exclusion-Based and Socialist-Communist regimes as Low Social Inclusion. On the other hand, we reclassify an unclassified group of countries (again, see Table 1 in Appendix B for a complete list of countries) as Low Social Exclusion; Extremely Rudimentary, Selective Rudimentary, and Slightly Universal regimes as Mid Social Exclusion; and Anti-Welfare Conservative regimes as High Social Exclusion. These typologies align with our theoretical understanding of material social inclusion as a spectrum between social inclusion and social exclusion. Moreover, our findings reveal a striking contradiction in social inclusion outcomes between the Extremely Rudimentary Welfare Regime in Africa and the Anti-Conservative Regime in Latin America. Despite having far less developed formal welfare systems, countries in the Extremely Rudimentary regime, on average, rank higher in social inclusion than those in the Anti-Conservative regime. This unexpected result underscores the complex and multifaceted nature of social inclusion, which cannot be reduced to the presence or absence of specific welfare policies alone (Atkinson and Marlier Reference Atkinson and Marlier2010; Silver Reference Silver and Orum2019). One possible explanation for this finding lies in the role of informal social support networks and cultural values in promoting social cohesion and inclusion. In many African countries, traditional kinship structures, community-based organizations, and informal social safety nets play a crucial role in mitigating the effects of poverty and social exclusion (Adésínà Reference Adésínà2015; Awortwi and Aiyede Reference Awortwi and Aiyede2017; Oduro Reference Oduro2008). These informal support systems, rooted in cultural norms of reciprocity and solidarity, may help to foster a sense of belonging and shared identity, even in the absence of comprehensive state-provided welfare (Aboderin and Hoffman Reference Aboderin and Hoffman2015; Apt Reference Apt2012). Moreover, scholars (e.g., Gyekye Reference Gyekye and Zalta2010; Metz Reference Metz2018) have argued that African societies place a strong emphasis on communal values and social harmony, which may contribute to greater social inclusion at the interpersonal and community level. In contrast, many Latin American countries, despite having more developed formal welfare systems, are characterized by high levels of social inequality and stratification (Hoffman and Centeno Reference Hoffman and Centeno2003; Torche Reference Torche2014; Amarante et al. Reference Amarante, Galván and Mancero2016). The legacy of colonialism, combined with persistent racial and ethnic disparities, has created deeply entrenched social hierarchies that limit opportunities for social mobility and inclusion (De Ferranti Reference De Ferranti2004). Moreover, the neoliberal economic policies adopted by many Latin American countries in recent decades have exacerbated income inequality and social polarization, undermining the inclusionary potential of existing welfare institutions (Huber and Stephens Reference Huber and Stephens2012; Pribble Reference Pribble2014; Martínez Franzoni and Sánchez-Ancochea Reference Martínez Franzoni, Sánchez-Ancochea, Huber and Stephens2019). These structural barriers to social inclusion may help to explain why countries in the Anti-Conservative regime fare worse than those in the Extremely Rudimentary regime, despite having more expansive welfare provisions. It is important to note, however, that these explanations are necessarily tentative and require further empirical investigation. The relationship between welfare regimes, informal social support systems, cultural values, and social inclusion outcomes is likely to be highly context-specific and may vary considerably both within and between regions (Gough and Wood Reference Gough and Wood2004; Wood and Gough Reference Wood and Gough2006). Moreover, our analysis is limited by the available data on social inclusion indicators, which may not fully capture the complexities of social exclusion and marginalization across different domains and levels of analysis (Babajanian and Hagen-Zanker Reference Babajanian and Hagen-Zanker2012; Popay et al. Reference Popay, Escorel, Hernández, Johnston, Mathieson and Rispel2008; Mathieson et al. Reference Mathieson, Popay, Enoch, Escorel, Hernandez, Johnston and Rispel2018). Nonetheless, our findings underscore the need for a more contextualized understanding of the factors that shape social inclusion outcomes, beyond the narrow focus on formal welfare institutions. They suggest that the promotion of social inclusion may require not only the expansion of state-provided welfare but also the strengthening of informal social support networks and the cultivation of inclusive cultural values and practices (Dani and de Haan Reference Dani, de Haan, Dani and de Haan2008; Hickey Reference Hickey2020). At the same time, they highlight the importance of addressing structural inequalities and power imbalances that can undermine the inclusionary potential of even relatively generous welfare provisions (Osei-Assibey Reference Osei-Assibey, Aryeetey and Kanbur2021). Further research is needed to unpack the complex interplay between welfare regimes, societal contexts, and social inclusion outcomes across different regions and countries. This may involve the use of more fine-grained and context-specific measures of social inclusion, as well as qualitative and ethnographic approaches that can provide a deeper understanding of the lived experiences and perspectives of marginalized groups (Picherit Reference Picherit2018). By combining macro-level comparative analysis with micro-level empirical investigation, future studies can help to shed light on the diverse pathways through which social inclusion is promoted or hindered in different settings and inform more effective and tailored policy interventions to support inclusive development (Molyneux et al. Reference Molyneux, Jones, Samuels, Molyneux, Jones and Samuels2017).
We find both the 10 welfare regimes and our social inclusion typologies fruitful modes of conceptualizing social inclusion and social inclusion policy. Instead of refuting Aspalter’s (Reference Aspalter2023) classifications, our findings support their claims and add further refinement in the context of social inclusion.
Our social inclusion typologies not only help us classify countries and welfare regime types, which can inform national and international intervention related to social inclusion but also further our theoretical understanding of the societally generative process of social inclusion. Incorporating social expenditure by GDP per capita, perceptions of government effectiveness, and perceptions of government corruption as additional dimensions of social inclusion and welfare regimes appear to be not only conceptually appropriate but also an empirical tool to clarify our understanding of social inclusion. While Silver (Reference Silver1994; 2015) and Lyons and Huegler (Reference Lyons, Huegler, Healy and Link2012) have raised theoretical classifications of conceptions of social inclusion, our index shows that countries and welfare regimes with dissimilar philosophies of social welfare may have, on average, relatively similar material social inclusion, such as the case between countries classified as Christian Democratic, Neoliberal, and Social Democratic countries. What our social inclusion typologies add to the theoretical discussion is that like welfare state development theory, countries’ demographics, economies, and political factors play an integral role in social inclusion and demark categorical variations between countries and welfare regimes.
On a theoretical level, our findings also extend the question as to why certain countries have lower social inclusion, lower social expenditure, and greater state instability, while others do not. While our social inclusion typologies do not map to the Global North and South strictly, we do see trends where formerly colonized and conquered countries have lower rates of social inclusion along with lower social inclusion by GDP per capita and greater state instability. Although outside the scope of this investigation, future research should explore this line of inquiry using our findings. Additionally, our social inclusion regimes have identified two previously unidentified welfare typologies (Cluster Five/Mid Social Inclusion and Cluster Six/High Social Exclusion. While a list of these countries can be found in Table 1 of Appendix B, future research should explore the extent to which these countries are similar in terms of welfare policy and social inclusion.
This study has many additional limitations. As is often the nature of comparative data, our data were incomplete and offered only a high-level view of the phenomena at best. Because we used PCA and cluster analysis, we were forced to impute a handful of country averages. While we used some of the most advanced methods available for this imputation, any form of imputation likely introduces bias to the result and reduces the study’s validity, especially at the country level. Moreover, while aggregating countries’ values across time adds stability to our findings and is appropriate for this level of analysis, the approach loses out on understanding how these phenomena interact with the dimension of time. Lastly, while our selection of variables was theoretically informed, our social inclusion index does not entirely represent the concept of social inclusion, in our view, because of the need to use existing indicators instead of creating our own.
Future research should not only overcome the limitations of this study but also leverage this study’s insights to further inquire into global social inclusion and social inclusion policy. While we now understand that there are different social inclusion typologies within the 10 worlds of welfare, we are left wondering how these countries’ social inclusion policy may differ and why we see varying rates of social inclusion between them. Further, while we have not made a causal claim between welfare regime, social inclusion typology, and material social inclusion, we are left wondering if such a causal claim might be defended or if there may be other causal explanations at play.
Conclusion
This study embarked on an exploration of the confluence of social inclusion and welfare regime typologies, revealing nuanced interplays that enrich the discourse on global social policy. In an era rife with political volatility, environmental exigencies, and technological metamorphosis, the imperative to uplift marginalized communities assumes paramount importance. Amidst these pressures, social inclusion emerges as a holistic concept, encompassing economic metrics alongside environment, culture, economics, healthcare, and social relations, pointing to the multidimensionality of well-being.
Extending the foundation laid by Aspalter’s (Reference Aspalter2023) 10 welfare typologies, our empirical analysis showed that distinctions between welfare regimes do not uniformly dictate material social inclusion. Further, socioeconomic contexts, historical trajectories, and global power dynamics conspire to shape social inclusion realities. Welfare regimes served as compass points rather than rigid determinants, revealing six novel worlds of social inclusion – Low, Mid, and High Social Inclusion and Low, Mid, and High Social Exclusion – that offer a contextually granular framework for the relationship between social inclusion and welfare regimes.
Beyond a scholarly contribution, our typologies provide a pragmatic toolset for policymakers and international entities to organize efforts to promote social inclusion. By acknowledging the limitations of conventional welfare regime classifications in the context of social inclusion, practitioners can tailor interventions more precisely. Moreover, our social inclusion typologies deepen the understanding of societal nuances, enabling targeted strategies that address the unique combination of factors perpetuating social exclusion for a given nation.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/ics.2024.14.
Funding statement
This study was funded by the Qatar Foundation.
Competing interest
The authors declare no competing interests.
