Self-rated health and social meetings

Self-rated health (SRH) (also referred to as “self-assessed health” or “self-perceived health” refers to a single-item health measure in which individuals rate the status of their own health on a four- or five-point scale, from excellent to poor. It is popular for its simplicity and has been extensively studied in Western populations. A series of national and international analyses has consistently shown that SRH is a good predictor of mortality resulting from various diseases (Benjamins et al., Reference Benjamins, Hummer, Eberstein and Nam2004). Despite the apparent non-specific nature of SRH, it has been shown to be an unusually strong predictor of mortality. SRH is the most widely used comprehensive health measurement recommended by the World Health Organisation (WHO) (DeSalvo et al., Reference DeSalvo, Bloser, Reynolds, He and Muntner2006).

There are several indicators that can be used when studying what factors influence SRH. Many studies have shown a strong association between health and socioeconomic determinants (Kaikkonen et al., Reference Kaikkonen, Rahkonen, Lallukka and Lahelma2009; Mackenbach, Reference Mackenbach2012). Individuals with lower socioeconomic status can expect to enjoy fewer years in good health, have higher rates of mortality, and die younger (Mackenbach, Reference Mackenbach2012). Earlier studies have also looked at the significance of variables that measure the significance of what one might define as the social dimension of SRH. This dimension has a weaker connection to material conditions, and can include the extent to which an individual participates in regular social activity, the degree of perceived social support, and so forth. Even if these variables capture social factors influencing SRH, they differ from the variable “social meetings”, which specifically measures the frequency of social meetings.

Social meetings is a commonly used variable in research, and several studies incorporate the frequency and characters of social meetings into scales of social support, social participation, and social network. The Lubben Social Network Scale (LSNS) (Lubben, Reference Lubben1988) is one among the scales that has explored social meetings as an indicator. This scale has been used to measure size, closeness, and frequency of contacts in a respondent’s social network. Into this scale there are variables that measure the same dimension as social meeting that are associated with health outcomes, such as personal network size, consistency. For example, it has been shown that larger networks have been associated with a lower risk of depression (Noteboom et al., Reference Noteboom, Beekman, Vogelzangs and Penninx2016). Buckley et al. (Reference Buckley, Becker and Burnette2022) have examined the direct effect of social isolation on SRH and the potential mediating effect on the psychological sense of community. Their study shows how social isolation affects SRH among older adults in Puerto Rico two years after a major natural disaster. This provides insights into the mediating role of psychological sense of community in the association of social isolation and SRH.

Even if many studies have identified indicators of social meeting and SRH, there is a lack of research using this variable to measure the moderating role that welfare regimes play in the association between social meetings and SRH. Our natural point of departure is research on variables that measure the influence of social dimensions on SRH. We therefore build our theoretical argument on prior research that has used variables that are most similar to the ones we test. Social activity is one variable that has been commonly used in research on SRH, and even if social activity can take different forms, the frequency of social meetings can be considered one indicator of social activity. It is widely accepted that an active social lifestyle is associated with better health outcomes and increased longevity. Several studies performed in the USA and Europe since the 1970s indicated an association between social networks and mortality. Social and community ties were shown to be associated with reduced mortality (Berkman and Syme, Reference Berkman and Syme1979), while reduced levels of social interaction were associated with increased mortality (Morgan et al., Reference Morgan, Schuster and Butler1991).

A consequence of fewer social meetings may also lead to increased feelings of loneliness. Several studies have shown that loneliness is associated with individual-level characteristics such as gender, age, marital status, and socioeconomic status (Gierveld, Reference Gierveld1998; Pinquart and Sörensen, Reference Pinquart and Sörensen2001). For example, older people report more loneliness than younger individuals; divorced persons are more lonely than married people, while a higher economic status is associated with less loneliness. Loneliness has also been associated with cultural factors at the societal level, with older people in more individualistic societies reporting lower levels of loneliness (Lykes and Kemmelmeier, Reference Lykes and Kemmelmeier2014). This may seem counterintuitive, but one interpretation is that older people in the more individualistic Northern European societies may feel less lonely since these types of societies tend to have stronger welfare institutions, which could play a moderating role in reducing such feelings. When identifying the social mechanisms, we make a connection between social meetings, experiences of loneliness, and the effect it has on SRH across different welfare regimes. We build upon the model of Perlman and Peplau (Reference Perlman, Peplau, Peplau and Perlman1982), which states that the prevalence of loneliness comes from unmet needs from having few social contacts. Secondly, they point out that the prevalence of loneliness can be associated with poor living conditions, such as low socio-economic status, poor health, or a deprived living environment.

Even if it is possible to draw an analogy between the impacts of social meetings on loneliness and SRH, these factors are more limited to the social factors leading to SRH. However, SRH is measured by more than just the psychological or mental aspect, and one advantage of using the regime categories as a moderating variable is that it takes into consideration both the material and the social conditions. Accordingly, how the welfare regimes perform should be an indicator of how the individual is taken care of both socially, mentally, and physically. Finally, the welfare state might also influence loneliness by supporting services that enable older people to interact socially and to engage in social activities. The welfare state influences individual social expectations, quality of living conditions, and level of social integration, which are thus formed in the exchange with the welfare state context of a person (Nyqvist et al., Reference Nyqvist, Nygård and Scharf2019). For the purpose of our argument, the stronger welfare states may fill unmet social needs through the health and social services.

Comparison of welfare regimes

Our data is based on Esping-Andersen’s (Reference Esping-Andersen1990) differentiation of welfare states into regime categories (see Table 2 for classification of countries). Welfare regime types have been accepted as an important factor determining health inequalities (Craveiro, Reference Craveiro2017). Many attempts have been made to create a classification scheme that is applicable to most welfare states. Esping-Andersen’s (Reference Esping-Andersen1990) classification elaborated in his book The Three Worlds of Welfare Capitalism is probably the most well-known. He categorized Western states into three main regime categories: Social democratic, Liberal, and Conservative, and shows how they differ regarding the organization of the welfare state as well as how extensive the standard of welfare services the state provides for its citizens is.

The social-democratic welfare type is characterized by a high level of universalism and equality of the highest standards. In other words, the workers were guaranteed full participation in the quality of rights by the better-off. It crowds out the market and constructs an essentially universal solidarity in favor of the welfare state. (Esping-Andersen Reference Esping-Andersen1990: 28). In line with these criteria, we have grouped the five Nordic countries into this category.

Liberal welfare states can be observed in most Anglo-Saxon countries like the United States and the United Kingdom. These welfare states are characterized by individuality and the primacy of markets. The liberal welfare state is based on means-tested assistance, modest universal transfers, and modest social insurance plans. (Esping-Andersen, Reference Esping-Andersen1990). Since we have limited our empirical study to European countries, we have grouped the UK and Ireland into this category.

The conservative-corporatist welfare regime is characterized by a moderate level of decommodification. The model is based on a historical corporatist-statist legacy in which social rights and equality were never a significant issue. (Esping-Andersen, Reference Esping-Andersen1990). We have added seven countries from the surveys into this category.

Esping-Andersen’s model has received praise but also criticism, for example, that it does not take into account a Southern European dimension. Among others, Ferrera (Reference Ferrera1996) and Trifiletti (Reference Trifiletti1999) have argued that the Southern European countries should be seen as a separate cluster, and that they represent a separate “Southern model” of social policy. Esping-Andersen’s lack of specification of the Mediterranean dimension has led to a debate about the existence of a “Southern model.” As such, we have included the Mediterranean category in our analysis to address this potential limitation of the original model. The Mediterranean regime is characterized by a fragmented system of welfare provision, which consists of diverse income maintenance schemes that range from the meagre to the generous, and a health care system that provides only limited and partial coverage. There is also a strong reliance on the family and charitable sector (Ferrera, Reference Ferrera1996).

Lastly, it has been debated whether the post-communist countries of Central and Eastern Europe fit into Esping-Andersen’s typology. Deacon (Reference Deacon and Jones1993) suggested a “probably temporary” classification of most of these countries as an additional type, that of a “post-communist conservative corporatist welfare regime.” According to Esping-Andersen, the Eastern post-communist countries are clearly the most under-defined and understudied region in terms of welfare state development (1999). These countries have experienced economic growth and have also undertaken social reform throughout the 1990s (Kovacs, Reference Kovács2002). After the collapse of Communism in 1989 a very rapid transition to a market economy was compounded by economic liberalization and the elimination of subsidies and price controls (Ferrera and Rhodes, Reference Ferrera and Rhodes2000).

We have therefore included the post-communist category in our analysis, where we present data from 15 former Eastern European communist countries. There are certainly substantial variations among countries within this category. However, they all share a similar historic transition from centrally planned economies to market-based systems, which in turn has significantly impacted their welfare systems.

We have operationalized the welfare regime types by grouping countries into the five mentioned categories commonly used in comparative European research, which also includes later updates of Esping-Andersen’s (Reference Esping-Andersen1990) model: (1) the social-democratic (Denmark, Finland, Iceland, Norway, Sweden), (2) the liberal (Ireland, United Kingdom), and (3) the conservative (Austria, Belgium, France, Germany, Luxembourg, Netherlands, Switzerland), (4) the Mediterranean (Cyprus, Greece, Italy, Portugal, Spain, Turkey), and (5) the post-communist (Albania, Bulgaria, Croatia, Czechia, Estonia, Hungary, Latvia, Lithuania, Poland, Romania, Russia, Serbia, Slovakia, Slovenia, Ukraine). We use these categories as measures of Esping-Andersen’s ideal types. More specifically, the social-democratic, the liberal, and the conservative countries represent an operalization of Esping-Andersen’s original typology, while we also build on later studies that extended his typology, including countries characterized by other distinct directions, such as the Southern or Mediterranean countries (Leibfried 1993; Ferrera 1996). The Southern/Mediterranean category, in addition to the post-communist, contributed to capture empirically distinct welfare state routes in Southern and Eastern Europe (see Deacon 1993 and Ferrera, Reference Ferrera1996).

To sum up, our operationalization measures variations in welfare state design by grouping the countries into empirical categories that broadly reflect Esping-Andersen’s (Reference Esping-Andersen1990) ideal types. Additionally, we included later expansions of his model so that we could extend the analysis to a broader European context.

Chung and Muntaner (Reference Chung and Muntaner2007) have conducted a multilevel analysis of health indicators clustered in welfare state regime types. Their results have shown that 20% of the differences in the infant mortality rates among countries can be explained by the type of welfare state. Bambra and Eikemo (Reference Bambra and Eikemo2009) found that the welfare state regime accounts for approximately half of the variation in the health status at the national level. Cross-national differences in SRH health have been discussed in relation to welfare state systems, where the Nordic welfare system (often referred to as the Scandinavian or social-democratic regime) with an extensive welfare state, is often held in high regard. Previous research also implies that the extent of welfare state regime provision plays an important indirect role in the prevalence of loneliness in later life (Nyqvist et al., Reference Nyqvist, Nygård and Scharf2019). Studies show that general well-being and happiness are largely due to the ability of the Nordic model to create a framework for “the good life,” a safety net that creates security, free education, and a sensible balance of work and leisure time, allowing people to enjoy both their work and their family life (OECD, 2017). Finally, there might be a combination of material and cultural factors leading to health outcome. Since values associated with familism are stronger in both southern and eastern European countries compared to the Nordic European countries, it is possible that this might influence SRH in a positive direction. One assumption is that a culture with more family-oriented values may lead to less loneliness.

To sum up, access to social meetings can be driven by a combination of cultural factors, health services, and material resources. Since research has shown that social meetings have an overall positive effect on health, we expect that social meetings have a positive effect in all the welfare regimes, despite cultural variations and how extent of the services are. Our first hypothesis is postulated as follows:

Nyqvist et al. (Reference Nyqvist, Nygård and Scharf2019) found that the Nordic as well as Anglo-Saxon and Continental welfare regimes performed better than the Southern and Eastern regimes concerning the absence of loneliness. They showed that the elderly in the Nordic regime, characterized as a more socially enabling regime, are less dependent on individual resources for loneliness compared to regimes where loneliness to a greater extent is conditioned by family and other social ties. This goes against the assumption that familism in Southern and Eastern Europe leads to lower levels of loneliness. One interpretation is that, despite the fact that the Nordic countries are characterized by individualism, which could lead to more loneliness, it could also be that the social security net provided by the welfare state has a preventive effect on loneliness. Another speculation is whether the welfare services can play the equivalent role in individual societies, in the same manner as the family does in less extensive welfare states. We assume that there might be some variation in the effect social meetings have in different regime types. As such, we test both the direct effect of social meetings, in addition to examining the direct effect of being conditioned by regime type, by including cross-level interactions. We thus present the following hypothesis:

Data and methods

We ran two-level hierarchical models using individual-level data taken from nine rounds of the European Social Survey (ESS, 2020), covering the period 2002–2019.Footnote ¹ Our analysis includes 35 countries which gives us 212 country-survey-years (level-2), and approximately 3,90,000 respondents (level-1). The ESS is a biennial repeated cross-sectional study based on personal interviews.

Individual-level variables

All our variables are presented in Table 1. Our dependent variable is a five-point ordinal scale measuring self-rated health, measured using the question: “How is your health in general?” with the response categories being “very bad” (1.55%), “bad” (7.28%), “fair” (27.18%), “good” (42.10%), and “very good” (21.90%). High values mean that the respondent reports good health, and we argue that this variable captures physical health as well as the mental and social aspects of the health dimension.

Table 1. Descriptive statistics

Our main independent variable, social meetings (1–7) is operationalized through the following question: “How often do you socially meet with friends, relatives or colleagues?” with the answer categories (with percentage of our sample) being “never” (2.13%), “less than once a month” (8.86%), “once a month” (9.62%), “several times a month” (19.43%), “once a week” (17.68%), “several times a week” (26.89%) and “every day” (15.38%).

One of the strengths of our analysis is the large N at both levels. As such, we have been cautious when it comes to including control variables, including only those with few missing values. At the individual-level we have included woman (0–1), age, education years, and partner (0–1), the latter denoting whether the responded is married/cohabitant, as well as child (0–1) denoting whether the respondent have a child or children (of any given age) as older children can function as a support network and younger children can create social contact to other adults.

Higher-level variables

We included two controls at the country-survey-year level: health expenditure per capita and GDP per capita, both gathered from the World Bank (2022), and divided by 1,000 for easier interpretation. Our expectation is that both these control variables should have a positive effect on individual SRH. It follows the same strain of logic as to why having higher education or belonging to a privileged social class would be associated with better health. At the country-level, this standard of living can be measured by GDP per capita (Eikemo et al., Reference Eikemo, Bambra, Judge and Ringdal2008). Olsen and Dahl (Reference Olsen and Dahl2007) point out that some of the differences in health between the Mediterranean countries and the former communist countries on the one hand, and the liberal, conservative, and social-democratic regimes on the other hand, are due to the relative poverty of the former countries. Health expenditure follows somewhat the same logic. If a country is able to provide good health care for its citizens, we expect this to be positively associated with self-rated health at the individual level. This expectation is supported by a review of the literature on public health spending and population health, where the author finds that increases in spending on public health is associated with a similar improvement in population health (Singh, Reference Singh2014).

Several studies also show that there is a relationship between SRH and government spending (see, for instance, Mueller et al., Reference Mueller, Park and Mowen2019), as well as the economic situation and SRH (see for instance Leão et al., Reference Leão, Campos-Matos, Bambra, Russo and Perelman2018; Copeland et al., Reference Copeland, Bambra, Nylén, Kasim, Riva, Curtis and Burström2015). Our main independent variable is regime type and is measured at the country level. This is operationalized as a set of dummies where we have chosen to set the Social Democratic type as the reference group. We thus include dummies (0–1) for the liberal, conservative, Mediterranean, and post-communist countries. For a list of the regime categories, see Table 2. For our main hypothesis, this dummy set is multiplied by level-1 variable, social meetings. For the results of the individual-level variables, we employ sampling theory, thus generalizing from the sample to the population. Regarding the upper-level variables, our aim is to investigate the full population of European countries. As such, we are here generalizing within stochastic model theory, from the observation made to the process that brings about the data (Gold, Reference Gold1969; Henkel, Reference Henkel1976). In our models, the standard errors of the variables are calculated based on the N for the corresponding level. Regarding the cross-level interactions, the standard errors are calculated using the N of the lowest level (level-1).

Table 2. Regime classification of countries included in analysis

Models

Since we are investigating nested data, we employ multilevel modeling to account for variance in the dependent variable measured at the lowest level, using information from all levels of analysis (Mehmetoglu and Jakobsen, Reference Mehmetoglu and Jakobsen2017). Multilevel analysis allows us to estimate the residual terms at all levels simultaneously with the linear coefficients (Ringdal, Reference Ringdal1992). As our dependent variable is situated at the ordinal level, we present to 2-level ordered logisticFootnote ² models, one general (to test the direct effect of social meetings) and one including cross-level interactions (to see if the direct effect is conditioned by regime type). We thus assume that our observed dependent variable provides incomplete information about a latent variable (Long and Freese, Reference Long and Freese2014). $ {Y}_{ij} $ tells us the value on self-rated health for an individual i from a specific survey-country-year j (e.g., Sweden-2011); $ {\tau}_m $ represents our four cut-points (analogous to intercepts, one for each answer category of Y minus the baseline group), and eij and u0j are the level-1 and level-2 residual terms. Equation (1) shows the model testing the direct effect of social meetings on self-rated health, while Equation (2) includes the moderating effect of regime categories:

(1) $$ {\displaystyle \begin{array}{l}\log \left\{ pr\left({Y}_{ij}\le m\right)\right\}={\tau}_m+{\beta}_1 womanij+{\beta}_2 ageij+{\beta}_3 edyrsij+{\beta}_4 partnerij\\ {}+{\beta}_5 childrenij+{\beta}_6 socialmeetij+{\beta}_7 healthexpj+{\beta}_8 GDPpcj\\ {}+{\beta}_9 liberalj+{\beta}_{10} conservativej+{\beta}_{11} mediterraneanj\\ {}+{\beta}_{12}{postcommunist}_j+{e}_{ij}+{u}_0j\end{array}} $$

(2) $$ {\displaystyle \begin{array}{l}\hskip-2em \log \left\{ pr\left( Yij\le m\right)\right\}={\tau}_m+{\beta}_1 womanij+{\beta}_2 ageij+{\beta}_3 edyrsij+\\ {}\hskip11.3em {\beta}_4 partnerij+{\beta}_5 childrenij+{\beta}_6\mathrm{socialmeetij}+{\beta}_7 healthexpj+\\ {}\hskip11.3em {\beta}_8 GDPpcj+{\beta}_9 liberalj+{\beta}_{10} conservativej+{\beta}_{11} mediterraneanj+\\ {}\hskip11.3em {\beta}_{12} postcommunistj+{\beta}_{13} socialmeetij\ast liberalj+\\ {}\hskip11.32em {\beta}_{14} socialmeetij\ast conservativej+{\beta}_{15} socialmeetij\ast mediterraneanj+\\ {}\hskip11.3em {\beta}_{16} socialmeetij\ast postcommunistj+ eij+{u}_0j\end{array}} $$

In our model, we actually multiply each regime dummy (minus the reference category) with the level-1 variable social meetings, thus producing four interactions (though substantially, only one).

By running an empty model, we calculated the intraclass correlation coefficients for our dependent variable, finding around 91% of its variance to be situated at level-1 and 9% at the upper levels. If the upper-level variation is above 5% it should not be ignored (Mehmetoglu and Jakobsen, Reference Mehmetoglu and Jakobsen2017) and is also a reason for our choice of modeling. We wished to test a regime hypothesis and present the operationalization of the regime variables in Table 1. The categorization is based on Esping-Andersen’s original categories (1990), as well as the updated Mediterranean and Post-communist categories.

We present two models in Table 3, the first showing the direct effect of regime-belonging while the second test the interaction with social meetings. All the variables tested at level-1 are statistically significant. Women score lower than men on the dependent variable: the older one is, the poorer their health is reported to be. Education has a positive effect on SRH, and respondents who are married or living with a partner report better health than those living alone. Having children has a negative impact on the dependent, and both country-year-level controls, GDP per capita and health expenditure, are positive and significant. Not surprisingly, the effect of our main independent variable, social meetings, is positive and significant. The more regularly one meets friends, relatives, or colleagues, the better the score on self-rated health is. Looking at the substantive effect, age is by far the strongest in our models, followed by education years and social meetings. Looking at the direct effect of the regime categories, the liberal is positive and significant, while the conservative and post-communist are negative and close to 10% significance (which we should mention as the standard errors for these variables are calculated using the level-2 N). The Mediterranean category is positive but not significant.

Table 3. The effect of social meetings on self-rated health (1–5), 2002–2019

Looking more specifically at Model 2, we see that all the cross-level interaction terms are positive and statistically significant except for the conservative regime (i.e., positive and close to being significant at the 5% level). First, the effect of social meetings among respondents belonging to the social-democratic regime category is both positive and significant (seen from the variable social meetings). The other regime categories receive this positive effect plus their corresponding interaction effects, implying that all other regime categories exhibit a significantly stronger (or nearly significant) effect of social meetings compared to the social-democratic regime category (see Figure 1). The total effect is strongest for the post-communist countries, followed by the Mediterranean, liberal, and conservative categories.

Figure 1. The effect of social meetings (1–7) on self-rated health (1–5), moderated by regime category.

In Figure 2, we report the mean scores of the countries included in the analysis of the dependent variable, self-rated health and social meetings. Here, we see a cluster of social-democratic countries in the upper-right corner of the scatter plot. The conservative countries are also concentrated on the upper-right, though to a lesser degree. The two liberal countries score relatively highly on self-rated health but around the middle on social meetings. Mediterranean countries are spread across the scatter plot, showing large within-group variation - from Greece in the upper-left corner to Portugal in the lower-right. Post-communist countries generally score low on both variables.

Figure 2. Scatterplot of social meetings on self-rated health, country means.