1 Introduction

Demographic transition theory explains the reproductive pattern of societies in which high birth and death rates decline over time, resulting in changes in the size and distribution of the population (Yuan & Gao, 2020). Demographic changes and the speed at which they have occurred in several countries, combined with political, social, and economic transformations, have affected both young and older people. In this context, one of the issues currently being addressed is vulnerability concerned with ageing. This is because "[w]hile growing old is not a new human experience, the way people are ageing is new" (Conboy, 2021, p. 362).

In Europe, life expectancy has been increasing linearly for 150 years now. Should this trend continue, the number of older adults in Europe will likely increase by 45% by 2030 and exceed 30% of the population by 2060 (Ince, 2015). Therefore, it is important to ask to what extent changes in the population pyramid create vulnerability and affect the quality of life of different population groups.

Vulnerability to social exclusion plays a central role in the quality of life of older adults, especially in ageing societies. Changes in family and marital structures can potentially influence the probability of social exclusion (Gray et al., 2011). In addition to psychological and physical challenges, ageing may be associated with loss of independence and a reduced ability to participate in social and cultural activities (Van Regenmortel et al., 2018). In other words, age per se is not a determinant of social exclusion but individual characteristics and the accumulation of disadvantages in various areas of life mean that the risks of social exclusion may be higher and endure longer within the older population group than in other age groups (Jose & Cherayi, 2017; Scharf & Keating, 2012). Hence, this is a topic that requires special attention in research in this field.

The concept of social exclusion has been used by politicians and/or activists since the 1960s, but it was not until the study by Townsend (1979) on the effects of relative poverty and resource scarcity in the United Kingdom that it took on substantial importance within the social sciences, especially among gerontological researchers (Van Regenmortel et al., 2016).

Therefore this paper addresses two key research questions, namely: Which micro and macroeconomic factors influence whether an older person is excluded? What is the pattern of social exclusion in Europe?

In answering these questions, the paper makes two main contributions to the literature. The first is to establish a measure of social exclusion adapted to the data and theory available to date. The second is to estimate a multilevel linear regression model with longitudinal data, which can analyse the impact of social exclusion on older adults and the associations with macro and microeconomic factors over time. The goal is to create an instrument for the early detection of social exclusion in older adults in Europe.

2 Background

2.1 Theoretical Framework of Social Exclusion

Defining social exclusion among older people is complicated, as the concept is open to ambiguity. Originally, the concept of social exclusion emerged as a response to the limitations of using poverty alone as a measure of disadvantage. Over time, it has evolved to encompass a broader range of inequalities, considering political, social, and cultural factors (Walsh et al., 2017).

Sen (1998) highlights capabilities and functionings, while Room (1995) emphasises employability, contributing to a comprehensive understanding of this complex phenomenon. Additionally, the breakdown of essential social systems, crucial for ensuring full citizenship (Room, 1995; Silver, 1994), further deepens the exploration of social exclusion. However, an alternative perspective posits that social exclusion is a collective outcome arising from various social disadvantages that result in both income poverty and deprivation (Berghman, 1995; Raileanu Szeles & Tache, 2008). Poverty, deprivation and social exclusion are interconnected, but it is crucial to recognise the nuanced distinctions between them (Raileanu Szeles & Tache, 2008). The concept of poverty primarily addresses distributional matters, while social exclusion revolves around relational issues, placing emphasis on inadequate participation in society (Raileanu Szeles & Tache, 2008; Room, 1995).

Walsh et al. (2017) significantly contribute to this discourse by adapting the concept of social exclusion for older adults. Drawing on the framework of Levitas et al. (2007), they define social exclusion in the context of older individuals as follows:

Social exclusion of older persons is a complex process that involves the lack or denial of resources, rights, goods and services as people age, and the inability to participate in the normal relationships and activities, available to most people across the varied and multiple dimensions of society. It affects both the quality of life of older individuals and the equity and cohesion of an ageing society (Walsh et al., 2017, p. 83).

The existing literature on social exclusion among older adults contributes to this multifaceted analysis by exploring dimensions such as relativity, dynamism, cumulative characteristics, and multidimensionality (Van Regenmortel et al., 2016). This research sheds light on exclusionary mechanisms that can lead older individuals into precarious situations and limit their opportunities for escape (Scharf et al., 2005a). Moreover, older people are identified as particularly vulnerable to the detrimental associations of intersecting exclusionary processes, emphasizing the need for a nuanced understanding (Jehoel-Gijsbers & Vrooman, 2008; Walsh et al., 2017).

It is seen as relative because it is tied to activities that are considered typical within a particular society, highlighting that individuals are excluded in comparison to other groups or societal norms rather than solely based on their individual circumstances (Dahlberg et al., 2020; Walsh et al., 2017; Prattley et al., 2020).

The dynamic nature of social exclusion is understood by Paugam (1995) as a "spiral of precariousness", where individuals transition from a state of social integration across multiple dimensions to a zone of vulnerability or precariousness where one or more forms of exclusion occur (Raileanu Szeles & Tache 2008).This suggests that exclusion in one area may be associated with a heightened risk of exclusion in others, and the interaction can occur during major life transitions or gradually over time, resulting in a cumulative process (Scharf et al., 2005a; Heap & Fors 2015).It implies the accumulation of economic and social disadvantages across at least two dimensions for at least three consecutive years (Raileanu Szeles & Tache 2008).

Recognising the multidimensional nature of social exclusion, individuals can experience exclusion in multiple aspects of their lives (Burchardt et al., 2002). Based on the concept put forward by Scharf et al. (2000), social exclusion comprises five dimensions: exclusion from material resources, from social relations, from civic activities, from basic services and from neighbourhoods (Van Regenmortel et al., 2016; Walsh et al., 2017).

In general, a lack of financial resources throughout life can trigger monetary problems that are accentuated in old age, contributing to social exclusion in older adults and affecting their ability to participate in society (Scharf et al., 2005a). Exclusion from material resources pertains to the economic disadvantages faced by older individuals, specifically concerning their financial status and ownership of consumer goods and housing (Prattley et al., 2020). Mack and Lansley (1985) introduce a perspective known as the "consensual view of needs," suggesting that social exclusion occurs when individuals lack "socially perceived" items needed to access adequate living conditions. In an adaptation of this dimension, authors such as Keogh et al. (2021) introduce a variable that reflects a household’s financial difficulties and its inability to meet basic needs. Additionally, Tong et al. (2011) adapt to the data available in the study and consider income adequacy as an indicator of this dimension.

Exclusion from social relations reflects the importance of building relationships with others in order to ensure integration in society. For this reason, this dimension is characterised by indicators of loneliness, social isolation and lack of participation in social activities (Scharf et al., 2005b). Loneliness has serious consequences at all ages, but the opportunities to escape loneliness decrease in later life as social networks fragment. Those consequences are reflected in mental and physical health and in social and health costs (Morgan et al., 2021). For example, Myck et al. (2021) base exclusion from social relations, understood as loneliness, on the 3 items available in the Survey of Health, Ageing and Retirement in Europe (SHARE) related to lack of companionship, exclusion and isolation.

Participation in civic activities refers to the ability of people to participate in society and influence decisions that affect their lives (Scharf et al., 2005b). In general, it is oriented towards cultural, sports, social and volunteering participation, which can influence the improvement of older adults’ health (Van Regenmortel et al., 2018). Serrat et al. (2019) also consider the importance of participation in political decisions. However, the activities deemed to form part of exclusion from civic activity vary from one survey to another and according to the data available.

The basic services dimension can be represented by access to and use of different services in and outside the home, i.e. access to basic services such as water, electricity, gas, food and doctors and proximity to pharmacies, bus stations, etc. (Scharf et al., 2005b). This can be adapted to the data available. For example, in their study Dahlberg et al. (2020) consider variables related to visits to the doctor or dentist in constructing this dimension.

Neighbourhood exclusion may have a greater impact on older adults than on younger people. Older adults have a greater preference for staying in or near their homes, and many of them may have spent their entire adulthood or a long period in the same place, so they tend to prefer to have the right conditions to remain in the neighbourhood. In addition, an emotional attachment develops over the years (Scharf et al., 2005b). For example, Scharf et al. (2005b) consider indicators related to individuals’ perceptions of their neighbourhoods and feelings of safety when walking around their neighbourhood within this dimension.

Considering these dimensions, the concept of social exclusion goes beyond a binary perspective of disadvantage in old age and embraces the idea of simultaneous exclusions and inclusions (Keogh et al., 2021). Currently, there is no standard measure of exclusion that has been accepted by the research community. However, with the current qualitative and quantitative inputs an approximate measure can be created. The measurement systems currently used to quantify social exclusion differ in their methodologies and analysis techniques, so they are difficult to compare (Van Regenmortel et al., 2016; Parodi & Sciulli, 2019; Poggi, 2007). In parallel to the lack of social exclusion, studies that address this issue from the perspective of older people (Van Regenmortel et al., 2018) mention that the methods most often used to approximate a measure of social exclusion include non-linear canonical correlation analysis (Van Bergen et al., 2014), Schwarz’s Bayesian criterion (Ogg, 2005) and creation of categories (Saito et al., 2012).

Each approach possesses unique strengths in capturing the multidimensional nature of the phenomenon and developing an instrument that is empirically tailored to a specific dataset. Furthermore, during the construction of measures, each approach faces two fundamental decisions: how to weight indicators and how to aggregate them (Keogh et al., 2021). It is evident that exclusion in later life is not only multidimensional but also exhibits layered multidimensionality, as discussed by MacLeod et al. (2019). This makes assigning weights within an index significantly more complex. It is necessary to allocate weights at various levels, including the dimension, sub-dimension, and indicator levels, to comprehensively capture the nature of this multidimensional construct. Therefore, it is necessary to establish a prior weighting technique (Nardo et al., 2005a, 2005b). Methods can be "equal” or "unequal” and are applied using different forms of aggregation (Greco et al., 2019). The equal weighting method is the most widely used. Researchers justify its use when the associations between indicators are not exactly known and on grounds of simplicity or lack of sufficient theoretical foundation or objectivity (Greco et al., 2019). Once indicators are weighted, variables must be aggregated using linear (e.g. by adding up indicators), geometric (by multiplying indicators) or multi-criterion aggregation (using non-linear techniques). This last method tends to be the least used (Van Regenmortel et al., 2018; Nardo et al., 2005a).

An example of a method of linear aggregation and equal weighting in studies of social exclusion in older adults can be found in Prattley et al. (2020): the authors create a measure of exclusion from the first seven waves of data of the English Longitudinal Study of Ageing (ELSA). Similarly, Pellegrini et al. (2021) create a continuous measure of exclusion by summing binarised indicators that represent this measure, using European Social Survey 8.

2.2 Trends and Associations of Individual and Aggregate Variables

As life circumstances change with time and age, so does the way in which exclusion is conceptualised. For example, income level, education and/or labour market areas may play greater roles for children and young people, whereas for older people the focus may be more on savings and possession of current material resources (MacLeod et al., 2019). Although age itself is not a dimension of social exclusion, being old implies an accumulation of events over a lifetime that can generate exclusion in some dimensions. These events are determined by individual, institutional and/or group factors over time (Walsh et al., 2021; Morgan et al., 2021). In other words “[a]s people’s needs change with age, how we conceptualise exclusion also changes, and whilst the overall framework of exclusion may remain constant, the specific indicators used in the operationalisation are modified across the life course” (Walsh et al., 2021, p. 82).

Changes in marital status and family structure also influence the likelihood of exclusion (Gray et al., 2011), which is increased by being separated or widowed (De Jong Gierveld et al., 2009). Old age living conditions may also differ for men and women (Dahlberg et al., 2020). Jose and Cherayi (2017) show that being a woman makes social exclusion significantly more likely. Furthermore, although both older men and women who have higher levels of educational attainment are less likely to be excluded, the prevalence is greater for men.

Moreover, labour market participation plays a key role for working-age adults but the situation is more complex for older adults. There are several pros and cons to this stage of life. The post-retirement period can be positive in terms of increased availability of free time, but others may see it as a form of isolation, boredom and exclusion (De Boissieu et al., 2021). After retirement, it is important for older adults to maintain their active participation in society, as this can influence their well-being and health. As people grow older, health and functional limitations typically worsen, contributing to the risk of exclusion (Prattley et al., 2020). Furthermore, social exclusion is related not just to physical but also to mental health. Depression is negatively correlated with reduced social relations, poor housing conditions and insufficient income resources, which also reduce well-being (Tong et al., 2011).

Another factor related to social exclusion among older adults is the frequency of contact with adult children and grandchildren, if any. Research shows that a lack of participation in the lives of children and grandchildren is core to social exclusion in some cultures (Wethington et al., 2016). However, having poor health, a bad personal situation or increased dependency often means having a stronger relationship with children and grandchildren, as they are usually the caregivers (Morgan et al., 2021).

The concept of social exclusion should be seen as affected by the historical, socio-cultural and political context of the region/country in which people live (Sunwoo, 2021). Social relationships and family contact may be more or less important depending on the demographic environment in which older adults find themselves. Not having pre-conceived expectancies of strong family support may decrease the risk of isolation (Borboudaki et al., 2020).

In Europe, the welfare state plays a crucial role in the well-being of older individuals, as it influences the availability of services that facilitate their active participation in society. The Nordic regime is recognised for its extensive state involvement in social welfare, which fosters the development of social capital and promotes participation among older adults through an empowering welfare system. In Eastern European countries, the welfare state and societal changes following the fall of communism may have directly impacted well-being, especially in countries that have transitioned towards liberalism. Conversely, in predominantly Catholic countries social participation tends to be lower. Within the continental regime, where the role of the state is vital, feelings of loneliness tend to increase when there are stronger family ties. Lastly, the prevalence of loneliness depends heavily on social and family connections in the Southern European regime, characterised by a strong emphasis on familism and a socially disabling welfare state (Nyqvist et al., 2019).

In general, being from Northern or Western Europe reduces the risk of social exclusion (Ogg, 2005; Borboudaki et al., 2020). Europeans currently over 65 are an age group which, as mentioned by Van Herk and Poortinga (2012), has experienced several political and social changes during their lifetime due to the transition from World War II to the formation of the European Union. Due to the support received, economic recovery was faster and more effective for Western countries and those that remained neutral, (such as Denmark, Sweden, Norway and Switzerland) than for post-communist countries (Sunwoo, 2021). “That is, diverse historical background and socio-economic gap of many years\('\) standing between Western/Nordic and East Bloc nations are an underlying cause in determining old age social exclusion in Europe today” (Sunwoo, 2021, p. 426).

Seeking to examine the impact of aggregate variables on social exclusion in later life, several studies, including Hansen et al. (2021), have demonstrated significant associations with factors such as trust in government. Trust in governmental institutions is crucial for effective interaction between institutions and citizens, as low trust in institutions can have negative consequences on social connections, leading to reduced social trust, social unrest, crime, and diminished social unity. Moreover, some countries emphasise cultural or political values of solidarity and relational practices that foster cohesion and socialisation. This emphasis not only enhances social resources but also shapes expectations regarding relationships. Another relevant variable explored in earlier research is income inequality, as investigated by Jehoel-Gijsbers and Vrooman (2008). Income inequality consistently exhibits significant individual associations in their models. As the income gap between different groups widens, individuals with lower incomes often encounter greater challenges in accessing the necessary resources and services essential for maintaining a dignified life. This, in turn, increases social exclusion in later life.

In line with the above, Prattley et al. (2020) and Vela-Jiménez and Sianes (2021) also show that social exclusion is more prevalent in urban areas. As urbanisation continues to expand, it is forecast that around 70% of the global population will reside in cities by 2050. This rapid urban growth exacerbates social inequalities, with older adults being particularly affected due to population ageing (Vela-Jiménez & Sianes, 2021). One potential explanation for this is that urban environments can pose challenges for individuals with limited mobility, hindering their access to essential services. Additionally, the continuous influx of migrants into urban areas can contribute to a sense of constant transition, resulting in a diminished sense of belonging for certain individuals (Van Regenmortel et al., 2018).

Having emigrated from one’s home country to one’s current country of residence at any time influences social exclusion in adulthood, as it is more difficult to form new social networks in the destination country for various reasons, including language (De Jong Gierveld et al., 2015).

3 Data and Methods

3.1 Data and Sample

This study employs data from the Survey of Health, Ageing and Retirement in Europe (SHARE). This is a longitudinal data survey that includes information from 2004 to 2019 on health, socio-economic, employment and other statuses.

The sample is based on information from waves 5, 6 and 7 of the surveyFootnote 1, which were conducted in 2013, 2015 and 2017 respectively. These waves contain modules with topics related to social exclusion. SHARELIFE respondents from wave 7 who completed a questionnaire that focused on their retrospective life histories are excluded; only those who completed a regular interview and did not live in a nursing home are considered.

To focus the field of study, only older people, defined as individuals aged over 60, are considered. Finally, the analysis covers only the 10 European countries that participated in all three waves of SHARE data (5, 6, 7) (n= 63,116): Austria, Belgium, the Czech Republic, Denmark, France, Germany, Italy, Spain, Sweden, and Switzerland.

3.2 Developing a Social Exclusion Measure

A social exclusion measure has been constructed with 19 answers representing five underlying dimensions (For more details, see Table 4 in the Appendix).

For the “Material and Financial Resources” dimension, as defined in Scharf et al. (2005b), two variables are used related to the inability to make ends meet and the inability to do things due to lack of money. In “Social Relations” the study by Myck et al. (2021) is taken as a reference and variables expressing a lack of companionship and feelings of exclusion and isolation are used. In the “Civic Activities” dimension the reference taken is Van Regenmortel et al. (2018), and variables oriented towards cultural, sports, social and volunteering participation are taken into account. For “Basic Services” Dahlberg et al. (2020) is the reference and variables related to visits to the doctor or dentist are considered. We also decided to include 2 more variables that fit the definition of the dimension, i.e. not being able to afford to eat meat or fish and feeling cold because of the cost of food and/or heating. For neighbourhood exclusion, we adapt Scharf et al. (2005b) and consider variables that denote the respondent’s perception of crime in the area, cleanliness, feeling part of the area and being helped.

After selecting each variable, we calculate the score for each dimension and the multiple exclusion measure. As indicated above, there are different ways of calculating this measure, but the structure of the data available has led us here to build on Prattley et al. (2020) and Pellegrini et al. (2021) to create a measure of exclusion in older adults in Europe.

First, each variable is reinvested and rescaled with values of 0 or 1 so that a higher score reflects a higher degree of social exclusion. The binary variables of each dimension are integrated to create a scale of values. The value of each dimension for each individual is divided by the total number of variables in each dimension. The same is performed for the total measure, but now considering all the variables that make up the dimensions of social exclusion. Finally, the values obtained are multiplied by 100 to create a scale from 0 to 100. This gives a continuous measure in which a higher value represents greater social exclusion in older adults (0 = not excluded and 100 = totally excluded). The results for each wave of the survey are shown in Table 1.

Table 1 Social exclusion measure
Full size table

3.3 Independent Variables

Based on the literature review and the data set described above, the individual attributes and aggregate variables listed in Table 2 are included (For more details, see Table 5 in the Appendix).

Table 2 Descriptive statistic of characteristics of social exclusion (Waves 5–7)
Full size table

3.4 Statistical Methodology

The second aim of our research is to study social exclusion among older adults taking into consideration disparities between European Union countries over time. We examine the hypothesis that not only individual but also community factors affect the social exclusion of older adults.

The nested structure of the data is handled using three-level hierarchical linear modelling (HLM) techniques, also known as mixed-effects or multilevel modelling. The multilevel methodology is the most appropriate in this case, as it enables the models to integrate different levels of data in their structure, provided that a hierarchical structure can be determined between them (Steenbergen & Jones, 2002).

This methodology enables us to observe whether the associations of variables at individual level depend on factors at higher levels of the hierarchy, and the probability of misspecification is reduced by integrating several levels of variables into a single model (Steenbergen & Jones, 2002). The method also explains the results of all levels simultaneously.

As mentioned above, longitudinal data from three consecutive waves (waves 5, 6 and 7) are used. The hierarchical structure of the data is country-individuals-observations. Consequently, at level 1 we consider repeated measurements of individuals over the 3 waves (three observations if they participated in all three waves of the survey); these are nested within individuals (respondents) at level 2, and individuals are nested within different European countries (10 countries) at level 3.

Initially, the three-level variance components model (or unconditional model) is estimated, in which the predictor variables are not specified. The model can be written as in Raudenbush and Bryk (2002):

$$\begin{gathered} y_{{ijk}} = \gamma _{{000}} + \mu _{{00k}} + r_{{0jk}} + e_{{ijk}} , \hfill \\ \mu _{{00k}} \sim N\left( {0,\sigma _{u} ^{2} } \right);r_{{0jk}} \sim N\left( {0,\sigma _{r} ^{2} } \right);\,e_{{ijk}} \sim N\left( {0,\sigma ^{2} } \right) \hfill \\ \end{gathered}$$
(1)

The indices i, j and k denote observations, respondents and countries, respectively. \(y_{ijk}\) represents the social exclusion measure for wave i, in individual j, living in country k. It is modelled as a linear combination of the overall mean of social exclusion in older adults \((\gamma _{000})\) averaged over all occasions for all individuals. There is also \(\mu _{00k}\), a random country effect (level 3) which measures the gap between country k’s mean and the overall mean (between-country). \(r_{0jk}\) is a level 2 random effect that quantifies the deviation of the mean of an individual jk from the mean score for individuals in country k (between-individual), and \(e_{ijk}\) is a random effect that measures the deviation in the repeated social exclusion score around the mean score for individual j in country k (within-individual).

This type of model enables us to estimate the intraclass correlation coefficient (ICC), which estimates the proportion of total variance that is attributable to differences within individuals and countriesFootnote 2 (Raudenbush & Bryk 2002).

The country-level ICC is:

$$\begin{aligned} {ICC}_u=\frac{{\sigma _u}^2}{\sigma ^2+{\sigma _u}^2+{\sigma _r}^2} \end{aligned}$$
(2)

The individual-level ICC is:

$$\begin{aligned} {ICC}_r=\frac{{\sigma _r}^2+{\sigma _u}^2}{\sigma ^2+{\sigma _u}^2+{\sigma _r}^2} \end{aligned}$$
(3)

Subsequently, several covariates are included and the overall model is defined as follows:

$$\begin{aligned} y_{ijk}=\gamma _{000}+\gamma _{100}({wave}_{ijk})+\sum _{q=1}^{Q}{\gamma _{q}Z_{qijk}}+\sum _{s=1}^{S}{\gamma _{s}W_ {sjk}+\mu _{00k}+r_{0jk}+e_{ijk}} \end{aligned}$$
(4)

where \(\gamma _{000}\) is the average social exclusion measure for individual j in the reference country k with other covariates set to zero; \(\gamma _{100}\) is the average change in exclusion for the same individual over time. \(\gamma _q\) and \(\gamma _s\) denotes fixed-effects estimates of all q time-varying covariates and s time-constant variables. \(\mu _{00k}\) , \(r_{0jk}\), \(e_{ijk}\), are the level 3, 2 and 1 random-effects respectively. Empirical analysis of the growth plots indicates that a linear trend could better capture the individual-specific pathways of change in the dependent variable. Therefore, there is no need to use more complex polynomial functions.

4 Results

Before a multilevel analysis is conducted, it is conventional to run some preliminary analyses to ascertain whether such an analysis is necessary based on the data structure. Starting from the null or empty model (Equation 1), we calculate the ICC at individual level and for countries (Equations 2 and 3). This to determine whether there is variability in social exclusion between individuals in the same country and between the countries considered in the study.

In Model 1 (Table 3) the ICC is 0.17 at country level and 0.77 at individual level, which means that 17% of the total variance in social exclusion of older adults is attributable to differences between countries, while 77% is attributable to differences between individuals within countries. In regard to the statistical significance of the variances in the null Model, the fact that the estimated values of \(\mu _{00k}\), \(r_{0jk}\), \(e_{ijk}\) are larger than their respective standard errors suggests that there is significant variation in the social exclusion of older adults across individuals and countries. This, again, supports the choice of a hierarchical model in this study.

Once the null model has been estimated, the explanatory variables for the three levels are incorporated. In Model 1, the wave variable (wave = 0,1,2) is introduced, which refers to changes over time in social exclusion in older adults. Then age and age squared are included in Model 2, followed in Model 3 by covariates related to the rest of the individual characteristics (demographic and socio-economic). Finally, Model 4 introduces aggregate country variables such as Gini and trust in government. Two dummy variables referring to Eastern (Czech Republic) and Southern Europe (Italy, Spain) are also added, taking Northern (Denmark, Sweden) and Western European countries (Austria, Belgium, Germany, France, Switzerland) as the reference categories, to determine their influence on the variability between countries.

Table 3 Multilevel regression analysis of individual and country-level factors
Full size table

To observe whether the changes over time across waves, individual and aggregate variables have any association with the variance in the level of social exclusion of older adults, the random part of each estimated model must be considered and compared with the initial estimate (null model). If this variance decreases, then the explanatory variables are exposing the differences.

As shown in Table 3, when individual variables (Model 3) are added the country-level ICC decreases from 17.05% to 15.51%. Finally, when the aggregate and dummy variables referring to Eastern and Southern European countries are included the country-level ICC drops to 4.65% in Model 4. The individual-level ICC also decreases from 76.59% in the null model to 68.81% in Model 4.

Confirming an increase in the explanatory power of the model, the variance predictions for all 3 levels decrease in Model 4 (compared to the null model). Model 4 explains 79.58% of between-countries variance \((R^{2} = (30.08-6.14) / 30.08 = 0.7958)\), 19.25% of between-individuals variance \((R^{2} = (105-84.79)/105 = 0.1925)\) and 0.22% of within-individuals variance \((R^{2} = (41.3-41.21)/41.3 = 0.0022)\).

The goodness of fit of the model can also be assessed by considering only the difference in the log-likelihood ratio statistic (Log-Likelihood) of Model 4, which has the highest value of any of the models estimated.

Additionally, the Bayesian Information Criterion (BIC) and the Akaike Information Criterion (AIC) suggest that a model with a lower AIC and BIC will have a better fit. This is borne out here in the case of Model 4. Therefore, in regard to Model 4, it can be said that part of the differences in social exclusion among older adults over time across countries and individuals can be attributed to geographical location and to the macroeconomic variables corresponding to the respondent’s place of residence.

Now it is time to interpret our findings. The results of Models 1–4 reveal significant factors associated with social exclusion among older adults. Over time, there has been a decrease in social exclusion \((\beta =-0.19, \rho <0.05)\). This supports the idea that the social exclusion of older adults over time statistically follows a linear trend, which in turn indicates that social exclusion scores on average decrease by that amount over intervals of approximately two years if all other variables remain unchanged.

Age has a significant, negative association with the social exclusion of older adults \((\beta =-0.28, \rho <0.05)\), while age squared has a positive, significant association \((\beta =0.00, \rho <0.05)\), indicating a U-shaped relationship. This is in line with Walsh et al. (2021), who mention that age in itself is not a determinant of social exclusion, but rather events that accumulate over a lifetime make this population group more vulnerable to exclusion.

Lower risk of social exclusion in older adults is also associated with individuals who are married (rather than single, widowed or divorced) \((\beta =-4.40, \rho <0.05)\), and with those who are retired \((\beta =-0.99, \rho <0.05)\), which is consistent with Gray et al. (2011) and De Boissieu et al. (2021) respectively.However, functional limitations due to poor health increase the risk \((\beta = 3.10, \rho <0.05)\) (Prattley et al., 2020).

As mentioned in the background review above, living conditions in later life are different for men and women (Dahlberg et al., 2020). Model 4 shows that being a woman has a positive, significant association with social exclusion \((\beta =0.86, \rho <0.05)\). Also, at higher levels of education (measured in years) the severity of social exclusion decreases significantly. Furthermore, migrant status \((\beta =3.92, \rho <0.05)\) (De Jong Gierveld et al., 2015) and living in an urban environment \((\beta =0.69, \rho <0.05)\) (Van Regenmortel et al., 2018; Prattley et al., 2020), also have significant, positive associations with the severity of social exclusion.

Predictors associated with frequency of contact with children among those with no children of their own indicate that having daily or weekly contact reduces the risk of social exclusion \((\beta =-0.43, \rho >0.05)\), while seldom or never having contact increases the risk \((\beta =1.40, \rho <0.05)\) (Hansen et al., 2021).

The estimation in Model 4 makes it clear that social exclusion among older adults differs from one country to another. Income inequality \((\beta =0.24, \rho <0.05)\) and lower trust in government \((\beta =-0.03, \rho <0.05)\) are associated with higher levels of social exclusion. This is in line with the findings of Jehoel-Gijsbers and Vrooman (2008) and Hansen et al. (2021) respectively.

Finally, when Eastern \((\beta =8.62, \rho <0.05)\) and Southern \((\beta =7.00, \rho <0.05)\) European countries are compared to those in Northern and Western Europe, these factors are found to have a positive, significant association with social exclusion, as shown also by Ogg (2005) and Borboudaki et al. (2020) respectively.

5 Discussion and Conclusions

Demographic ageing is a major challenge facing European societies, with a number of economic and social implications that are typical of several advanced societies. This paper contributes to research on social exclusion among older adults in Europe by studying and helping to understand this concept. Rather than viewing old age solely in terms of age-related changes or societal responses to them, it is important to recognise that older adults are more susceptible to social exclusion due to the prevalence and accumulation of various individual and collective factors throughout their lives. These factors have both short-term and long-term impacts, making it more difficult for older people to overcome exclusion (Walsh et al., 2017).

Building upon this understanding, our study seeks to provide a comprehensive understanding of complex social exclusion in later life. Specifically, we address two key research questions: (1) what micro and macroeconomic factors are associated with the social exclusion among older individuals?; and (2) what is the pattern of social exclusion in Europe?. By addressing these questions, this paper makes two main contributions to the literature. The first is to develop an approximate measure of social exclusion among older adults in 10 European countries; the second is to estimate a multilevel linear regression model using longitudinal data to analyse the impact of social exclusion on older adults and its associations with macro and microeconomic factors.

Our proposed measure encompasses five dimensions: social relations, material and financial resources, access to basic services, participation in civic activities, and neighbourhoods. Operationalising social exclusion poses a key challenge, and the adoption of a multidimensional approach in this study enables us to capture various dimensions and integrate them into a single measure. However, it is important to note that social exclusion per se cannot be directly measured, at least not at present. Instead, its existence is inferred from other phenomena that act as indicators. Nonetheless, we believe that our measure can provide useful tools for addressing social exclusion among older adults from a life course perspective, or at least serve as a starting point for further studies.

Based on the results of our analyses, our study draws several conclusions. First, we find that of the dimensions examined, participation in civic activities had the highest level of exclusion on average (83.32). This indicates that older adults may prefer activities that can be conducted within their homes, indicating a need for tailored approaches and support systems to promote social inclusion. These findings highlight the importance of creating opportunities for older adults to engage in meaningful activities beyond their immediate environment. By contrast, the dimension measuring participation in basic services (4.65) demonstrates the European system’s effectiveness in ensuring accessible, available essential services for the general population, including older adults. This finding implies that the European system has made progress in guaranteeing access to basic services, thus helping to mitigate social exclusion among older adults.

The average social exclusion score of approximately 29 points indicates a moderate level of social exclusion among older adults. However, it is noteworthy that the highest average scores are concentrated mainly in Southern and Eastern Europe (see Fig. 1 in the Appendix). This highlights the existence of regional disparities in social exclusion across Europe. A one-size-fits-all approach is insufficient when addressing social exclusion. Understanding these regional differences is crucial for informing targeted strategies and interventions to effectively address social exclusion among older adults. Our country-by-country analysis (see Fig. 2 in the Appendix) reveals the complexity of the issue, further emphasising the need for tailored strategies to suit the specific needs and challenges faced by each group. Heterogeneous associations of microeconomic covariates are evident across different countries, indicating that the determinants of social exclusion operate in different ways depending on the specific national context.

In our pursuit of a more comprehensive understanding of social exclusion, we extended our analysis, drawing on insights from the study conducted by Raileanu Szeles and Tache (2008). The additional analysis, presented in Table 6 in the Appendix, explores whether social exclusion among older adults can be characterised as an accumulation of economic and social disadvantages over time. Using a categorisation approach where individuals are assigned a ‘1’ based on dimension values exceeding the average, we assess the percentage of the population excluded in 0–5 dimensions. Approximately 13.51–15.1% of older adults are not excluded in any dimension, while 27.68–28.1% are excluded in two dimensions, and 19.21–18.6% experience exclusion in more than two dimensions.

This analysis supplements our understanding of social exclusion by providing insights into its multidimensional nature and the varying degrees to which individuals are affected across different dimensions. While our study suggests that social exclusion does not exhibit a strong cumulative nature across two dimensions, and despite the apparent stability between waves 5 and 7, the proportion of individuals excluded in more than two dimensions decreases slightly over time. This nuanced understanding underscores the importance of prolonged investigations to capture the complexities of social exclusion.

Under this approach, our research develops the second goal of analysing social exclusion factors among individuals followed through three waves conducted at two-year intervals (2013, 2015, 2017). The study reveals meaningful variations in the risk of social exclusion across countries and individuals and within individuals over time, emphasising the complexity of social exclusion. This highlights the importance of considering the interplay between individual and collective factors as influential associations. Additionally, a separate analysis of the dimensions of social exclusion, presented in Table 7 in the Appendix, helps to disentangle the different effects of the covariates.

Our findings indicate a decrease in social exclusion over the three periods. This short-term decrease in exclusion may be attributed to factors such as the development of feelings of "survival" among older adults and a sense of security in terms of housing and amenities, which partially mitigate the perception and experience of social exclusion (Paine et al., 2022). However, it is important to note that a relatively small short-term decrease does not imply the absence of persistently high social exclusion scores or justify accepting such levels. Therefore, extending this study over a longer period is crucial to gain a more comprehensive understanding of social exclusion trends.

Consistent with Walsh et al. (2021), who argue that age itself is not a determinant of social exclusion but that it is due to the accumulation of events over a lifetime that make this population group more vulnerable, our study finds a U-shaped relationship between age and social exclusion. This suggests a decrease in social exclusion as individuals grow older, followed by an increase in later stages of ageing. This pattern may be influenced by age-related health limitations and changes in social dynamics.

Moreover, several factors are identified as significant predictors of increased social exclusion, including gender (being a woman), education (low levels of education), social contact (seldom or never having contact with children), urban environment, migration status and functional limitations due to poor health. Conversely, being married, being retired and having regular contact with children are associated with a lower risk of social exclusion.

Macro-level factors such as income inequality (measured by the Gini coefficient) and trust in the government are found to be associated with higher and lower levels of social exclusion, respectively. These findings underscore the importance of addressing socioeconomic disparities and promoting inclusive governance to support the social inclusion of older adults.

The factors identified in our study offer valuable insights for policy-makers and stakeholders in developing targeted interventions and policies to reduce social exclusion and improve the well-being of older individuals. Additionally, measuring social exclusion in later life can serve as a tool for monitoring progress over time, enabling changes in exclusion rates and the effectiveness of implemented policies to be measured. This facilitates informed decision-making and the allocation of appropriate resources. However, it is crucial to acknowledge that social exclusion is a complex issue influenced by interconnected factors, which necessitates a comprehensive, holistic approach for effective intervention. These findings have significant implications for identifying the specific needs of each country. Initiatives aimed at strengthening trust in governments and reducing income inequalities can serve as initial steps. Taking a household perspective, allocating resources to individuals living alone or those dependent on healthcare would have a profound impact in terms of addressing social exclusion.

Adapting to demographic ageing is crucial in ensuring the well-being of populations, and the social exclusion of older adults is thus a matter of utmost importance. Nevertheless, it is essential to acknowledge the conceptual and theoretical limitations associated with studying social exclusion among older adults. Firstly, our study focuses solely on 10 European countries, thus limiting the generalisability of our findings to other regions or countries outside Europe. The interpretation of our estimates might be constrained in a broader global context, as patterns of exclusion in non-European settings might differ depending on various contexts, whether economic, social, or cultural. However, the measure developed in our study can be adapted to other countries based on data availability. Future research should continue to explore social exclusion among older adults and identify effective strategies to promote social inclusion and well-being among older adults in diverse societal contexts. Adopting a life course perspective and including a more diverse range of countries and cultures are essential for capturing the variability in social exclusion experiences among older adults. Additionally, the inclusion of other macroeconomic variables, such as public support, is crucial for elucidating these associations. Secondly, our study relies on longitudinal data from three waves conducted over a relatively short period (2013, 2015, 2017). This limited timeframe may not capture long-term trends and fluctuations in social exclusion, potentially underestimating or overestimating trends and changes in social exclusion among older adults. Extending the study over a longer time frame, particularly considering the impact of the COVID-19 pandemic, would provide a more comprehensive understanding of the long-term patterns of social exclusion and its associations with older individuals, especially those living in urban areas. Finally, the measurement and operationalisation of social exclusion pose complex, multidimensional challenges. The conceptual ambiguity surrounding the definition and dimensions of social exclusion, as emphasised by Van Regenmortel et al. (2016), introduces a potential limitation to our study. The absence of clear distinctions between various dimensions, risk factors, and outcomes of social exclusion adds nuance to our findings. This inherent complexity in the conceptual framework may influence our results, given that interpretations of social exclusion can vary across individuals and contexts. Consequently, it is crucial to consider the nature of the social exclusion construct when interpreting our results. Moving forward, future research could explore alternative measurement approaches and additional dimensions to further enhance our understanding of this intricate phenomenon.

Furthermore, while there is a broad consensus in gerontological research that social exclusion is a multidimensional concept that includes diverse aspects beyond poverty, it is important to consider that the concept of social exclusion is sometimes closely linked to income poverty and deprivation, although it is not synonymous with either. Expanding on the insights of Raileanu Szeles and Tache (2008), who highlight the need to distinguish between these concepts for effective policy purposes, recognising the distinct characteristics, determinants, and vulnerable groups associated with each of them enables policies to address them as separate stages in the social exclusion process. Conflating these concepts might cause common misinterpretations of the results and, correspondingly, a failure to accurately estimate the prevalence and degree of social exclusion. This, in turn, may result in policy recommendations that might only partially address the stages of this process.

In summary, this research underscores the significance of addressing social exclusion among older adults and offers valuable insights into its associations with other factors. Further research that includes a wider range of countries and cultures, longer time frames and more refined measurement approaches is necessary to extend our understanding of social exclusion. Such research can provide the knowledge needed to develop targeted interventions aimed at promoting social inclusion and enhancing the well-being of older adults.