Introduction

Despite sharing common philanthropic traditions and institutions, contemporary European countries display profound differences in their volunteer rates—that is, the share of a country’s population involved in volunteering. Altruism and charity in Europe are rooted in shared medieval institutions and practices, especially religious ones (e.g. the Catholic Church). The modern voluntary sector emerged in most European countries during the eighteenth and nineteenth centuries in the wake of political transformations, including the rise of democratic institutions and popular movements (Harris et al., Reference Harris, Morris, Ascoygh, Chicoto, Elson, McLoughlin, Muukonen, Pospíšilová, Roka, Smith, Soteri-Proctor, Tumanova, Yu, Smith, Stebbins and Grotz2016). However, the continent today appears quite heterogeneous and diversified regarding the share of the population volunteering in a formal setting. Formal volunteering in Europe—measured according to the International Labour Organization (ILO, 2011) definition that emphasizes its organizational, unpaid, non-compulsory, and outside the family and kin characteristics—ranges from high rates in Nordic countries (50–30% of the population volunteers) to low rates in Southern post-Communist countries (8–3% of the population volunteers). How can such variations in volunteer rates be explained?

Researchers have emphasized that the social and institutional contexts in which people live are significantly associated with volunteering (Curtis et al., Reference Curtis, Baer and Grabb2001; Ruiter & De Graaf, Reference Ruiter and De Graaf2006). However, a multidimensional theoretical explanation for how contextual factors of volunteering affect individual volunteering behaviours is lacking. The aim of this research is to advance the comparative study of determinants of volunteering by proposing a theoretical explanation for the variation in volunteering across Europe that integrates both contextual and individual factors—namely the capability approach. To test this approach by modeling contextual effects, a two-step strategy was adopted that involved estimating two samples separately: the micro-level sample estimation produces micro-level effects, and the dependence of those effects on macro-level factors emerges from a second macro-level estimation.

Theoretical Perspectives

Figure 1 shows how greatly formal volunteer rates vary across Europe. How can these cross-national variations be explained? Answering this question will make a crucial contribution to our understanding of the forces underpinning the dynamism of civil society.

Fig. 1

Formal volunteering across Europe in percent of countries’ population

(Source: author based on EU-SICL-2015 data)

Looking beyond individual factors, the ways in which the larger sociocultural context or contextual factors impact volunteering is a matter that has received increasing attention over time. In their seminal contribution, Salamon and Sokolowski (Reference Salamon, Sokolowski, Dekker and Halman2003) pointed out that the micro-structural approach to volunteering, which focuses on social capital and personal value systems, does not provide a good explanation for the differences in volunteer rates across countries. Macro-structural explanations for cross-national variations in volunteering are necessary to account for national patterns of individual behaviour. However, despite the multidimensional character of determinants of volunteering, the literature on comparative volunteering lacks an integrative multidimensional theoretical framework to inform empirical research. This literature has provided theoretical explanations that either are limited in terms of explanatory factors (mainly economic factors are central to the market/government failure, resource mobilization, and voluntary failure approaches) or, as in the case of the social origin approach, are difficult to test empirically with cross-sectional data, since regime theories track historical institutional features to make causal inferences about the present, ignoring potential co-evolutionary patterns (Anheier et al., Reference Anheier, Lang, Toepler, Powell and Bromley2020). Thus, the purpose of the present study is to develop and test a multidimensional explanatory framework of cross-national differences in volunteering.

Salamon and Sokolowski (Reference Salamon, Sokolowski, Dekker and Halman2003) identified four possible explanations in their examination of theories explaining cross-national variations in volunteering. The market failure/government failure theory (Weisbrod, Reference Weisbrod1977) sees volunteer action as a necessary response to the incapacity of the market and the government to satisfy heterogeneous demands for collective goods. Voluntary failure/complementarity theory (Salamon, Reference Salamon and Powell1987) emphasizes interdependence and cooperative relations between the government and the voluntary sector as a result of the limitations inherent in voluntarism. The resource mobilization hypothesis underscores the availability of social and organizational resources for voluntary mobilization (McCarthy & Zald, Reference McCarthy and Zald1977) and articulates an entrepreneurial theory of the formation of civil society organizations where the major factor is the availability of resources, especially cadres and organizing facilities. The last explanation, privileged by Salamon and Sokolowski (Reference Salamon, Sokolowski, Dekker and Halman2003) and labeled the social origins explanation (Salamon & Anheier, Reference Salamon and Anheier1998), emphasizes the institutionalization of power relations between social classes that have led to different types of welfare states and voluntary sector roles. The social origins theory explains the voluntary sector’s size in terms of the factors determining the size of its workforce and its revenue base. These determining factors are, in turn, traced back to institutional features reflecting power compromises that once shaped different civil society models (liberal, social democratic, corporatist, and statist) and are characterized by the division of responsibility for the provision of quasi-public goods and services between the state, the market, and civil society. Each explanation provides a set of hypotheses relative to the relationship between government spending and the size and nature of the voluntary sector across countries that the above-cited authors tested empirically. However, these theories, as Steinberg and Young (Reference Steinberg and Young1998) underlined, are not rival theories but complementary theories that explain more jointly than they do separately. These theories also have different explanatory scopes. As shown in Table 1, whereas social origins theory and the heterogeneity hypothesis aim at explaining why different institutional configurations structuring the opportunity structure of the voluntary sector appeared in the first place, the voluntary failure and resource mobilization approaches articulate some of the mechanisms through which these institutional configurations are sustained over time—that is, how, whether through resource mobilization or according to the degree of cooperation between the government and voluntary organizations, the division of labor between the market, the government, and the voluntary sector is maintained. The market failure/government failure explanation emphasizes economic mechanisms, while the social origins explanation relies on institutional mechanisms.

Table 1

The four theoretical explanations of cross-country variation in volunteer rates compared

	Why different institutional configurations?	How different institutional configurations are sustained?
Economic factors	Market/government failure	Resource mobilization
Institutional factors	Social origins	Voluntary failure/complementarity

Additionally, as Ragin (Reference Ragin1998) noted, social origins explanations involve a large variety of historical factors that combine in a unique fashion in each country and are, therefore, better suited to a qualitative approach for testing the theory—a fact that Salamon et al. (Reference Salamon, Sokolowski and Haddock2017) seemed to acknowledge in their most recent attempt to provide support for the theory. Regime theories—as far as regimes are specific institutional arrangements—look at whole configurations, emphasizing the conjunctions and interactions between civil society, the welfare state, and political and economic institutions. Whereas regression analysis requires that relationships between variables obey a common structure across countries, regime theory allows for different relationship patterns within each country, making it poorly suited for a quantitative analysis.

Recent empirical literature has further explored a range of mechanisms whose outcomes might contribute to different volunteer rates. These studies have investigated the cross-national macro-structural factors of volunteering, characterized by the adoption of quantitative methods for multilevel models that allow for investigating macro-structural factors and controlling for compositional differences due to the inclusion of individual-level variables. In a review of this literature, Baer et al. (Reference Baer, Prouteau, Swindell, Savicka, Smith, Tai, Smith, Stebbins and Grotz2016) listed eight factors that have been mobilized to explain cross-country differences: welfare regime types, state involvement in the economy, ethnic heterogeneity, economic development, democracy, religiosity, cultural values, and trust.

European studies have adopted different theoretical perspectives and modeling strategies. Plagnol and Huppert (Reference Plagnol and Huppert2009) concluded from their analysis of 2006 European Social Survey data that national differences in volunteer rates cannot be fully explained by differences in individual social, psychological, or cultural factors associated with volunteering, and thus, it is likely that contextual factors largely determine volunteer rates. However, how much such contextual factors account for this difference remains an open question. Gil-Lacruz et al. (Reference Gil-Lacruz, Marcuello-Servos and Saz-Gil2016) sought to identify the underlying reasons for differences in youth volunteer rates across Europe. They combined individual with national contextual data from the European Values Study in a cross-sectional study of 20 European countries and estimated a logistic regression including the individual and contextual variables with dummies for each type of welfare regime. They concluded that macro-structural factors, such as welfare systems, by influencing government expenditures and employment, impact youth volunteering decisions. In another publication, the same authors (Gil-Lacruz et al., Reference Gil-Lacruz, Marcuello-Servos and Saz-Gil2017) used three waves of the European Values Study and found that differences in volunteering participation rates could not be fully explained by individual characteristics, and they emphasized the significance of contextual factors. However, they employed as contextual variables the size of a residential area, the mean of other people’s attitudes, and the mean of trust in institutions without providing a clear theoretical rationale for this selection of variables. Using four waves from the European Values Study, Damian (Reference Damian2018) examined cross-sectional and longitudinal effects of contextual cultural and economic characteristics of individual formal volunteering. The study considered cultural and economic explanations and assessed the effects of the following contextual factors on volunteering: religiosity, religious denomination, economic development, and income inequality. The results showed that church attendance had a negative effect on formal volunteering and lower inequality was positively associated with formal volunteering, indicating that people in secular and equal countries engage more in formal volunteering. In sum, the results of this literature have been mixed and often dependent on empirical specifications of macro-structural variables, as far as each contribution has focused on certain explanatory factors to the exclusion of others. Common to these studies, however, is that they tested for contextual economic, cultural, and institutional predictors of volunteering but did not build an integrative theoretical framework relating these explanatory factors to each other. Explaining cross-national differences in volunteering requires a general theoretical framework—testable with quantitative means and enabling the integration of multiple factors into a coherent explanatory model—to guide the empirical investigation. The next section outlines such a framework based on the idea of capability.

The Capability Approach

Clearly, as noted in the social origins approach, institutional factors play a central role in explaining variations of volunteer rate across countries. However, this theoretical approach is difficult to test empirically because it lacks parsimony and because the long timeframe and complexity of factors that the approach identifies make it difficult to disentangle historical from present factors (Anheier et al., Reference Anheier, Lang, Toepler, Powell and Bromley2020). Additionally, economic, political, social, and cultural factors play a role in determining variations of volunteer rates. The theoretical understanding advanced here, first, operates takes into account institutional variations across civil society regimes (as identified by the social origins approach). Different institutional regimes can be expected (as they relate to different patterns of redistribution and social security, different qualities of civil liberties, pluralism, and the rule of law) to generate different sets of capabilities that, in turn, impede the volunteer rate. Second, the capability approach aims to provide an encompassing framework integrating different factors—economic, political, cultural, and social—that diverse existing theories emphasize separately. Third, the capability approach allows for bridging the micro–macro-link and agency-structure relation within a coherent theoretical framework. Indeed, volunteering is seen as the result of individual agency (choice and action) conditioned on the capability to volunteer (agency-structure relation) and of individual as well of contextual factors (micro–macro-link). This capacity to transform resources into functioning is, in turn, influenced by contextual structures—namely “material and ideal constraints to the context that provides the conditions of existence of specific interactions” (Layder, Reference Layder1981, p. 96).

A way to integrate multiple explanatory factors at the micro- and macro-structural levels is to consider the amount of volunteering in each country as reflecting citizens’ propensity to voluntarily engage in the creation (entrepreneurship), maintenance, and operation of voluntary organizations. Thus, the total amount of volunteering is the combined result of the demand for volunteering (i.e. existing voluntary organizations and entrepreneurship initiatives, both individual and collective) and the supply of volunteer work (i.e. individuals willing to do non-paid work in different capacities for given purposes).

From this perspective, cross-country differences might be attributed to factors that affect the supply and demand of volunteer work. Concerning the supply of volunteering, these factors have been thoroughly studied at the micro-individual level (Wilson & Musick, Reference Wilson and Musick2008). At the micro-level, determinants of volunteering seem to be conditioned on individual levels of human, economic, social, and cultural capital, because volunteer work is productive work that requires human and economic capital, collective behaviour that requires social capital, and ethically guided action that requires cultural capital (Wilson & Musick, Reference Wilson and Musick1997). However, at the micro-individual level, these forms of capital are influenced by macro-structural features that impact their availability and distribution in each country. In contrast, the demand for formal volunteering is influenced by the size and sustainability of the population of voluntary organizations enrolling individuals in activities that require volunteer work as well as the dynamics of creating new organizations. The size of the existing population of voluntary organizations, their activity levels, and the dynamics of their creation are, in turn, determined by a range of economic, political, social, and cultural factors that set the conditions for collective and individual civil society entrepreneurs and leaders to create and sustain voluntary organizations.

The claim is that the supply of voluntary work can be expected to vary accordingly to individual capabilities to engage in volunteering. The concept of capability introduced by Sen (1980/Reference Sen1982, Reference Sen1985, Reference Sen1992) is defined as “set of vectors of functionings, reflecting the person’s freedom to lead one type of life or another” (Sen, Reference Sen1992, p. 40). Functionings, in turn, consist of beings and doings, constitute the basis of living, and range from basic things, such as good health, to complex achievements, such as self-respect and happiness. Closely related is the concept of capability to function—a set of functionings—reflecting a person’s freedom to choose among valuable functionings to achieve well-being. Nussbaum (Reference Nussbaum2000) provided a list of central human capabilities, including life, bodily health, bodily integrity, sense of imagination and thought, emotions, practical reason, affiliation, other species, play, and control over one’s environment (political and material). Thus, the concept of capability is very broad—including all dimensions of life—and provides an alternative method to utility or basic goods for assessing well-being and inequality. More importantly, for our purpose, the concept of capability relates the availability of resources and structural conditions (the means of freedom) to the ideas of choice and freedom. Volunteering is, as its etymology indicates, a free or voluntary action, but integral to freedom of action are the means of freedom—that is, the individual and collective resources and conditions that allow free choice and, in the case of volunteering, social cooperation. Voluntary action may be considered a capability that is strongly related to three capabilities identified by Nussbaum (Reference Nussbaum2000): control over one’s political and material environments and affiliation.

Then, the concept of capability, although broad, is related to the concept of capital insofar as it includes the idea that the availability of individual and collective resources (capital) is decisive in explaining different forms of inequality, including inequality in engagement in volunteer work. In economic terms, capital is a stock that has value as a source of current and future flows of output and income. The principles of capital theory were extended first to human capital (Becker, 1963/Reference Becker1993)—the stock of skills and productive knowledge embodied in people—and then to cultural and social capital (Bourdieu, Reference Bourdieu and Richardson1986). According to Bourdieu (Reference Bourdieu and Richardson1986), capital exists in three fundamental forms: “as economic capital convertible into money and institutionalized in the form of property rights, as cultural capital institutionalized in the form of educational qualification and as social capital made up of social obligations (connections)” (p. 16). These forms of capital not only benefit the individual who “owns” the capital; as Coleman (Reference Coleman1988) indicated, social capital, economic capital, and cultural capital all have a public good aspect, meaning that they benefit everyone who is part of the social structure in which these forms of capital function. The concept of capability, thus, encompasses the concept of capital insofar as it is concerned with the means of freedom, or the resources that enable individual action, but add to it a dimension of agency, as the resources or forms of capital seen as enabling the individual’s free choice are the conditions for freedom and choice. As Sen (Reference Sen1997) emphasized, if notions of human and social capital are related to the idea of capability, the concept of capability is not an alternative to that of capital but a broader perspective that focuses “on the expansion of human freedom to live the kind of lives that people have reason to value” (p. 1960). In short, the difference between capability and capital is homologous to the difference between ends and means. A capability is “a person’s ability to do valuable acts or reach valuable states of being; [it] represents the alternative combinations of things a person is able to do or be” (Deneulin et al., Reference Deneulin, Nebel and Sagovsky2006, p. 2). Thus, capabilities encompass resources and forms of capital, as they enable individuals to freely choose the pursuit of valuable finalities, as well as the dimension of choice and freedom.

By emphasizing freedom and choice, the capability approach allows for bridging the micro–macro-gap, linking individual choices to macro-level conditions. Indeed, individual capabilities are not only dependent on individual resources and forms of capital but also on macro-structural features that enable or limit both choice and action. Thus, the capability of volunteering—the individual choice and functioning actualized in doing volunteer work—is dependent on individual resources or forms of capital (converted into functionings) and macro-structural features that enable or limit this choice and functioning.

The capability approach does not contradict but complements the social origins theory of the voluntary sector. Institutions and capabilities are related, since institutions are both constraints to, and enablers of, individual agency—in Sen’s terminology, the conversion of capabilities into functionings, or beings and doings (Nambiar, Reference Nambiar2013). Different civil society regimes can be regarded as producing different conditions, enabling more or less individual capability and functioning for volunteering, thereby influencing the supply and demand of volunteering. Whereas the social origins theory examines the institutional-structural characteristics of each country-specific civil society and assesses how these characteristics generate different patterns of volunteering, the capability approach to volunteering is located at a lower explanatory level and focuses on the factors that, given the institutional-structural features of civil society, influence the supply and demand of volunteering. Different welfare state and civil society regimes can be expected to generate different sets of capabilities, enabling individuals to engage more or less in the voluntary sector. Furthermore, different forms of individual capital (human, economic, and social capital) and contextual features in the economic, political, cultural, and social realms, relating to different institutional welfare state and civil society regimes, can be expected to condition individuals’ capability to engage in volunteering.

Volunteering is a functioning, and we observed only functionings. The capability approach to volunteering asserts that volunteering (functioning) is dependent on the capability to volunteer—that is, an individual’s ability to convert resources into functioning according to freely chosen valued finalities. The availability of both micro-level resources and macro-structural features shapes this capability. This perspective allowed us to formulate the following hypotheses linking resources, macro-structural features, and functionings.

In sum, the idea of capability has a subjective dimension (the free choice of ends) and an objective dimension (the conditions enabling this free choice). Capabilities are transformed into functionings i.e. actions. What we observe is not the capability per se, but the action (functioning) of volunteering. Institutional factors (including the individual distribution of different forms of capital) influence the objective dimension of the capability to volunteer (the conditions of free choice) and consequently the ability to act (functioning). Compared to existing approaches, the capability approach provides a coherent theoretical framework for linking individual-level resources and macro-level institutional features that shape the capability to volunteer.

Data, Measurements, and Method

In this work, an empirical strategy was designed to assess the effect of macro-structural factors given individual-level differences. For this purpose, a multilevel model was estimated. The capability approach entails that individual and contextual factors influencing individuals’ means of freedom and, consequently, capabilities to volunteer will be correlated with a country’s volunteer rate. These factors are related to economic, human, and social capital as well as political factors that shape individuals’ ability to participate in civil society. However, there are many ways to empirically operationalize such concepts as economic, human, and social capital and political opportunity structures at the individual and contextual levels. Additionally, the choice of empirical indicators is constrained by the availability of comparative data and the necessary parsimony of the model at level two of analysis. Furthermore, the number of variables included at the two levels of analysis is constrained by the size of the country sample. With data available for 23 countries, only a few contextual variables could be used in each model. Therefore, imperfect indicators of the main capability dimensions—economic, social, and political—were selected for the two levels of estimates. In what follows, the data and variables used at both levels are described.

Data and Samples

The analysis of factors influencing volunteering levels in Europe relied on two data sources. At the individual level, we used data from the 2015 EU Survey on Income and Living Conditions (EU-SILC) including an ad hoc module on volunteering. These data were documented in Eurostat (2016) and constitute the EU reference source for comparative statistics on income distribution and social exclusion in Europe. Each year, the survey includes different specialized modules, and in 2015, one ad hoc module was devoted to social and cultural participation, including volunteering. The reference population of EU-SILC is private households and their current members residing in the territory of the member states at the time of data collection. Persons living in collective households (e.g. boarding houses, dormitories at educational institutions, etc.) and institutions are generally excluded from the target population. The cross-sectional component has a minimum sample size of 106,000 households representing EU member states plus Iceland and Norway. All persons over 16 years old were interviewed in each household, giving a sample of 310,150 persons in the 26 EU countries in 2015 (23 EU member states plus Iceland, Norway, and the Former Yugoslav Republic of Macedonia [FYROM]). Individual sample weights provided by Eurostat were used to obtain a representative sample of the population in each country.

At the macro-level, we used data collected by the Quality of Government Institute (QoG) and documented by Teorell et al. (Reference Teorell, Kumlin, Dahlberg, Holmberg, Rothstein, Alvarado Pachon and Svensson2019) and the PEW Research Center. The QoG data set is composed of data compiled at the country level (including aggregated individual data), drawing on several freely available data sources and comprises more than 2,000 variables. The 2019 version of the QoG data includes data from and around 2015. If no data for 2015 existed, data for 2014 were included, and so on, up to a maximum of + / − 3 years. The cross-country data include 35 countries that were members of the Organisation for Economic Co-operation and Development (OECD) at the end of 2018. The data on religion come from two surveys administered by the PEW Research Center in 2015 and 2017, respectively, in Eastern Europe (survey conducted between June 2015 and July 2016) and Western Europe. Matching the EU-SILC data with the QoG data and PEW data yielded a data set of 23 countries available for the two-level analysis, as not all 25 countries included in the EU-SILC data set are in the QoG data set.

Variables at the Individual Level

The outcome variable of interest is a measure of formal volunteering—that is, volunteering performed within the framework of an organization. The EU-SILC survey asks respondents whether they have participated in formal volunteer work during the previous 12 months. Possible answers are Yes; No, lack of interest; No, lack of time; and No, other reason. The variable was coded into a dichotomous variable, Yes/No. The main advantage of using the 2015 EU-SILC survey for measuring volunteering is, first, the quality of data in terms of cross-country comparability, common methodology implemented by 25 European countries under the auspice of Eurostat, and a large sample size for each country. However, the formulation of the question concerning volunteering (as a Yes/No answer referring to the previous 12 months) presents some shortcomings compared to other data sources. In contrast to the recommendations of the ILO (2011), the EU-SILC survey does not implement prompting to assist respondents. The high “buffering” prompting recommended by the ILO manual consists of an extensive series of Yes/No questions about specific volunteer activities in which the respondent may have participated and then asks the respondent for details about each activity. Additionally, the EU-SILC survey employs a 12-month reference period during which the respondent may have volunteered, with the risk of lower accuracy of recall compared to the four-week reference period recommended by the ILO manual. However, there are also some benefits of using a 12-month reference period. For instance, the four-week period can capture time-of-year effects, such as specific (religious) holidays, during which people volunteer more often.

We used two variables to capture differences in human capital: Education, measured in terms of the highest International Standard Classification of Education (ISCED) level attained (ranging from 0 = pre-primary education to 6 = the second level of tertiary education), and self-assessed General health (ranging from 1 = very bad to 5 = very good).

Number of hours usually worked per week in the main job was used for considering, in combination with income level, the substitution effect between free time and working time. Additionally, different forms of economic capital, and especially income levels, were considered indicators of social status that are positively associated with volunteering, as wealthy individuals are three times more likely to be asked to volunteer (Hodgkinson, Reference Hodgkinson, Schervish, Hodgkinson and Gates1995). Therefore, income was measured by Household income rather than the respondent’s personal income, because household income is considered a better indicator of social status. Another indicator of social status was the individual’s Employment status, measured by a dummy variable indicating whether the individual was full-time employed (1) or had another status (0), including inactive, retired, unemployed, student, and part-time employed.

At the individual level, social capital was measured by two indicators—Frequency of contact with friends and Number of children in the household—assuming that people who report frequent contact with friends and parents of children still living in the household have more social contacts and are more likely to volunteer.

At the individual level, three demographic factors that might affect individual volunteering rates were considered: Marital status (never married, married, separated, widowed, divorced, or consensual union), Gender, and Age. Marital status might influence the social integration of individuals within the community and, consequently, the likelihood of volunteering. We expected gender differences in volunteering especially in countries with higher gender inequality, because traditional gender roles socialize men and women into different patterns of participation in the public sphere. Age was expected to affect the likelihood of volunteering inasmuch as obligations and responsibilities vary across the lifetime.

A logistic regression model at the individual level (step 1) of analysis was estimated for each country with the dichotomous variable of participation in formal volunteering as the dependent variable and the following independent variables: marital status, education, gender, health, economic status, weekly hours worked, household disposable income, age, number of children living in the household, participation, and frequency of contact with friends. We now turn to the description of variables at the contextual level (step 2) of analysis.

Variables at the Contextual Level

At the contextual level (second level), the variables used in the analysis were chosen for their ability to operationalize relevant dimensions of the two theoretical approaches to be tested. The capability approach predicts positive associations between cross-national volunteer rates and indicators of democracy, social equality, and trust, whereas the failure approach predicts a positive association between heterogeneous preferences and volunteering. These indicators are described in detail as follows.

Statistical Method

Multilevel (or mixed effects) models are an extension of simple linear models that allow both fixed and random effects, and they are particularly used when there is non-independence in the data, which can arise from a hierarchical or multilevel structure. Multilevel models can be thought of as a trade-off between (i) aggregating individual observations at the unit (country) level and (ii) analyzing data from one unit at a time. The individual regressions have many estimates and considerable data, but they are prone to high margins of error (statistical “noise” or estimation error). The aggregate is less characterized by estimation error, but may lose important differences when all samples within each unit are averaged. Additionally, multilevel modeling is suitable for investigating contextual effects. Multilevel modeling with two levels (individual and country levels) was implemented in this study in a two-step estimation procedure to investigate macro-contextual influences on individual volunteering. This two-step approach consists of one regression at the individual level and one regression at the country level (Borjas & Sueyoshi, Reference Borjas and Sueyoshi1994; Card, Reference Card1995; Hanushek, Reference Hanushek1974; Saxonhouse, Reference Saxonhouse1976; a special issue of Political Analysis introduced by Kedar & Shively, Reference Kedar and Shively2005). In the first step, one fits, for each country, the individual-level model using a logistic regression. Country-level effects are ascertained from the second step in which the fitted country intercepts from step 1 are regressed on the country-level predictors. A two-step procedure presents several advantages compared to a simultaneous estimation. First, it offers more flexibility in modeling individual-level specifications for each country without loss of efficiency (Jusko & Shively, Reference Jusko and Shively2005). Second, simultaneous multilevel models that assume a multivariate normal distribution of error terms do not converge when the data structure is complex (Franzese, Reference Franzese2005). Finally, the results of the estimation are easy to display graphically, facilitating the interpretation of results.

As Bryan and Jenkins (Reference Bryan and Jenkins2016a) emphasized, the two-step procedure gives unbiased estimates of individual- and country-level coefficients, and the procedure produces correct standard errors. Moreover, the ordinary least square in step 2 provides an unbiased estimate of the variance of country effects (σ ²). These properties apply even if there are few countries (Bryan & Jenkins, Reference Bryan and Jenkins2016a). An issue related to the two-step method, which applies more generally to estimation using clustered data, is that the estimation of country-level predictors is based on a limited number of observations (number of countries), entailing that the variance of the country-specific effect (σ ²) is likely to be estimated imprecisely with a small number of countries. Monte Carlo simulations based on empirical distributions of variables in the 2007 EU-SILC data indicated that the two-step model performs well with 25 countries (Bryan & Jenkins, Reference Bryan and Jenkins2016b) and that country-level fixed parameters are likely to be estimated imprecisely with a small number of countries (around 10).

Additionally, as estimate parameters from the first step were used as dependent variables in the second step, an estimated dependent variable (EDV) correction with a feasible generalized least square, as suggested by Lewis and Linzer (Reference Lewis and Linzer2005), was implemented using the R script that fits the feasible generalized least square (FGLS) models described in Lewis and Linzer (Reference Lewis and Linzer2005). This EDV correction accounts for uncertainties stemming from the first-step estimation and the macro-level error term from the second-step regression.

The model estimated here, for individuals i = 1….n, and countries j = 1….J, can be written as follows:

Level 1

(1) $y_{\mathrm{ij}}={\beta }_{0j}+{\beta }_{1j}{x}_{\mathrm{ij}}+{\epsilon }_{\mathrm{ij}}.$

Level 2: Estimated intercept (varying intercept by country)

(2) ${\beta }_{0j}={\gamma }_{00}+{\gamma }_{01}{Z}_j+{\mu }_{0j}.$

Level 2: Estimated coefficients as dependent variable in level 2 (varying slope with constant intercept)

(3) ${\beta }_{1j}={\gamma }_{10}+{\gamma }_{11}{Z}_j+{\mu }_{1j}.$

Full model

We substitute to obtain a single-equation form:

(4) $y_{\mathrm{ij}}=\left({\gamma }_{00}+{\gamma }_{01}{Z}_j+{\mu }_{0j}\right)+\left({\gamma }_{10}+{\gamma }_{11}{Z}_j+{\mu }_{1j}\right)x_{\mathrm{ij}}+{\epsilon }_{\mathrm{ij}}.$

Substituting (3) in (4) yields the full model (varying intercept, varying slope model, where both the intercept and the slope vary by country):

(5) $y_{\mathrm{ij}}={\gamma }_{00}+{\gamma }_{01}{Z}_j+{\beta }_{1j}{x}_{\mathrm{ij}}+{\mu }_{0j}+{\epsilon }_{\mathrm{ij}}.$

The varying slope model (Eq. 3) estimates the effects of macro-indicators on the estimated coefficients of individual-level variables, thus allowing for investigating the pathways linking macro-indicators to the individual capability to volunteer.