Measuring the immeasurable? Operationalising social capital in health research ()
1. INTRODUCTION
In the last decades, the concept social capital has gained growing interest in diverse scientific disciplines [1]. Social capital has been explored in relation to schooling and education, economy, governance, criminology and other fields [2]. However, in none of these disciplines the importance of the concept is as explicitly emphasized as in the literature on health and well-being [3]. Social capital has been related to mortality, morbidity, self-rated health, physical and mental health [2,4-8]. Even though the popularity of the concept has increased, its rise has not been welcomed uncritically. An important reason for this criticism is the lack of clarity that surrounds the concept [5]. This might be explained by the numerous research traditions in which the concept has been studied as well as the different backgrounds of researchers interested in the subject [9,10].
An important point of disagreement in social capital research concerns the level on which the concept is operationalised; can social capital be the property of individuals or is it a community characteristic [11,12]? One of the founding fathers of the theory on social capital, Robert Putnam, defines the concept as “features of social organization such as networks, norms, and social trust that facilitate coordination and cooperation for mutual benefit” [13]. This definition reflects Putnam’s opinion of social capital as a community characteristic. By contrast, others consider social capital to be an individual characteristic and therefore define the concept as the ability of individuals to command scarce resources by virtue of their membership in networks or broader social structures [9,14].
Another frequently mentioned point of criticism is that the all-encompassing interpretation of the concept used in research limits its practical applicability [15]. The term “social capital” is used to refer to an extensive number of different social characteristics at the individual, group, institutional and state level [9,16]. The lack of congruence in ways of defining social capital prevents researchers from drawing firm conclusions about the concept, its effects and the pathways through which it acts [16]. It also hinders the comparability between different studies [9,11,17] and may hamper focused strategic implementation of the concept in practice.
Consequently, this conceptual lack of clarity is reflected in the ways used to measure social capital. Researchers have used a wide range of indicators to operationalise the concept [17,18], starting from different theoretical backgrounds. The number of survey questions used to measure the concept varies, ranging from one question to a large number of questions used to calculate a scale or overall sum score.
Deciding on how to include social capital in health research can be challenging. This study highlights common, yet different, ways to operationalise social capital. It focuses on the pragmatic reduction of social capital to generalized trust, the measurement of social networks, the measurement of bonding, bridging and linking social capital as well as a multidimensional and complementary measurement of different components of social capital.
1.1. Generalized Trust
For pragmatic reasons, researchers usually try to capture the essence of the broad concept social capital in a limited number of indicators. In this approach, scholars often use the indicator “generalized trust” [12,19], which reflects the trust one has in people in general. This indicator is included in various questionnaires, such as the Social Capital Community Benchmark Survey1, the Social Capital Questionnaire by Onyx and Bullen [20], the European Social Survey2, the General Household Survey3 and the World Values Survey4.
1.2. Social Networks
The belief that social networks act upon practically all aspect of people’s lives has increasingly found its place in social sciences [21]. With his pioneering work “Le Suicide” [22], Emile Durkheim was among the first to empirically explore the influence of social networks on health by contributing rising suicide rates to a lack of social cohesion and increasing feelings of anomie. Since then, numerous studies have underlined the influence of social networks on different aspects of health [4,9, 23-24].
Today, the concept “social networks” is a central concept in social capital literature, especially in the literature that defines social capital as an individual characteristic. One of the social capital “schools” defines social capital as the “resources embedded within people’s social networks” [17]. As described by Henk Flap, “people will invest in relations with others in view of the perceived future value of the social resources made available by these relations” [21].
Different approaches can be followed to measure social capital within this theoretical tradition. One possibility is to measure the resources embedded in one’s social networks directly. This is the aim of instruments such as the “position generator” [25] which measures the social prestige associated with network members’ social position or the “resource generator” [26] which measures specific resources available in the respondent’s network. These specifically developed social capital instruments however have rarely been used in health sciences. Another possibility to measure social capital, more frequently used in relation to health, is to make use of instruments which focus on characteristics of the respondent’s social network, such as network size, network composition, the strength of ties and the feelings of trust and reciprocity present in the networks. An example of such a questionnaire is the Social Network Index [24], which inquires the respondents’ participation in 12 distinct social networks by investigating the number of people one has regular contact with.
1.3. Distinction between Bonding, Bridging and Linking Social Capital
In 2000, Robert Putnam described the distinction between bonding and bridging social capital. Recently, Szreter and Woolcock [10] elaborated on this distinction by describing a third form of social capital: linking social capital. They define bonding social capital as “trusting and co-operative relations between members of a network who see themselves as being similar”. Bridging social capital, on the other hand, refers to trusting and reciprocal relationships between people who do not consider themselves alike on specific socio-demographic characteristics. The added dimension, linking social capital, is defined as “norms of respect and networks of trusting relationships between people who are interacting across explicit, formal or institutionalized power or authority gradients in society” [10]. Although some researchers have criticised this theoretical development [27-29], others underline the added value that this conceptual elaboration could offer [30]. It is believed that the distinction between bonding, bridging and linking social capital could contribute to more clear and detailed evidence on the relationship between social capital and health [5] and thus to the public health research [12,31].
However, this conceptual contribution is not reflected in the empirical research field; to our knowledge, no published health study has measured bonding and bridging as well as linking social capital according to the definitions formulated by Szreter and Woolcock [10].
1.4. The Complementary Measurement of Different Components of Social Capital
Social capital is a broad “umbrella term”, used to refer to a wide variety of social factors. There is a growing consensus that social capital has to be measured in a way that matches its multidimensionality, by collecting information on different dimensions of the concept [32]. However, agreement on the precise composition of the components that make up the broad concept social capital is lacking.
1.5. Aims
The aim of this study is to assess to what extent five manners to measure social capital at the individual level can explain the variance in two health outcomes: selfrated health and pain. More specifically, we focus on generalized trust, a social network index, an expanded social network index, the distinction between bonding, bridging and linking social capital as well as a multidimensional social capital index.
2. METHODS
2.1. Design, Population and Sample
A cross-sectional study was set up in eight neighbourhoods in the region of Ghent, a densely populated area in the northern part of Belgium. The neighbourhoods were purposively selected, aiming for a maximal variance in socio-economic and geographic characteristics (level of deprivation, level and type of urbanisation). The inhabitants of the selected neighbourhoods were invited to participate in the study if they met specific criteria: they had to be older than 18, officially reside in the neighbourhood, have the Belgian nationality and have sufficient language skills to complete the survey. People living in residential settings (e.g. nursing homes) were excluded.
2.2. Instrumentation
The self-administered questionnaire is composed of questions on the respondents’ health, their social capital, socio-economic and demographic characteristics. Cognitive interviews are used to assess the comprehensibility and acceptability of the questions and the meaning respondents give to the used key concepts. Thirteen individual cognitive interviews were conducted with people with a low degree of schooling, as they can be considered as the “lower limit” regarding comprehensibility. These participants were purposely sampled based on age and mother tongue. Based on the findings of these pretesting interviews, minor changes were made to the questionnaire before further use in the study, mainly on specific terms and expressions used in the questionnaire and on the questionnaire’s structure.
2.2.1. Independent Variables: Social Capital
A set of questions was composed to assess the ability of five approaches to operationalise social capital (further referred to as “five models”). Where possible, existing and validated questionnaires were used.
In general, the questionnaire covers six components of the broad concept “social capital”. The presence of social networks, the magnitude of social networks and the behaviour of the respondent in these networks are explored to measure the structural components of social capital. Furthermore trust, the strength of ties (proxy for reciprocity) and the available resources in the network are included to measure the cognitive components of social capital.
The questionnaire focuses on twelve different social networks, namely relationships within the neighbourhood, relationships with schoolmates, workmates, members of religious groups, the partner, children, parents, in-laws, other close family members, friends, fellow volunteers and fellow members in groups without religious affiliation (e.g. social or recreational groups). For all twelve networks, the availability and size of networks are inventoried, as is the number of people one talks to on a regular basis, the number of people one trusts and the number of people one feels close to. For four specific networks (i.e. partner, close family members, friends and fellow members in groups without religious affiliation), information on the composition of the networks is also gathered. For these four networks, information is inquired at the level of personal income, the educational level, the religious adherence and the ethnicity of the network members. The number of network members that work in the health care sector was also investigated for these specific four networks.
One of the included cognitive components of social capital is trust. Three dimensions of trust are distinguished in the questionnaire: personalized, generalized and institutional trust. Personalized trust is reflected by the number of network members one trusts. Generalized trust refers to trust one has in people in general. The trust one has in his/her general practitioner on the one hand and doctors in general on the other hand are considered indicators of institutional trust.
In total, 83 social capital-related questions were included in this questionnaire. Based on these questions, five different models to operationalise social capital are composed. The number of questions needed to form each of these models varies from 1 to 83. Table 1 gives an overview of the theoretical basis of each of the models, the manner in which the theoretical concepts are operationalised and the number of questions used in each model.
2.2.1.1. Model 1: Generalized Trust
To be able to assess the value of a delimited social capital model, the first model comprises one single variable. This measure is based on the question “Would you say, in general, that most people can be trusted?” A four-point Likert scale was offered to answer this ques-
Table 1. Overview of theoretical concepts that the models encompass, used operationalisations and the number of questions in each of the five models to measure social capital.
tion ranging from “everybody” to “nobody”. This measure was dichotomized, with a score of 0 representing trust in no one or a distrust in most people, and a score of 1 representing trust in everyone or most people.
2.2.1.2. Model 2: Original Social Network Index
The second model is based on an existing questionnaire on social networks, the Social Network Index (SNI). The SNI was initially used to investigate the influence of social networks on the susceptibility to the common cold (Cohen et al., 1997). The original SNI contains information about the availability of twelve social networks for the respondent and the number of people one regularly talks to within these networks. The original questions from the SNI are included in the developed questionnaire. Based on the guidelines published by Cohen5 three original SNI-scores could be calculated. The number of high-contact roles (=network diversity) represents the number of social networks in which the respondent has regular contact with at least one other person. “Regular contact” is defined as contact once every two weeks or more often. The number of people counts the total number of people with whom the respondent has regular contact.
The number of embedded networks describes the number of a total of eight different social networks in which a respondent is active. These include family, friends, religious group, school, work, neighbours, fellow-volunteers, and fellow-members in non-religious groups. Being “active in a network” is defined as having at least five people within that social network one talks to on a regular basis (i.e. minimum once every two weeks). To calculate this measure, the five family roles (partner, children, parents, in-laws and extended family) are added together into one overall network. This overall family network is considered as an embedded network if the following two premises are fulfilled: the respondent has to talk to at least one network member on a regular basis in at least three different family networks and has to talk regularly to at least five people within the overall family network.
2.2.1.3. Model 3: Expanded Social Network Index: Social Networks, Trust and Reciprocity
Social capital is often defined as having trustful and reciprocal relationships [33]. Following this definition by Robert Putnam, questions to capture feelings of trust and reciprocity are added to the original SNI (= model 2). Feeling close to others (i.e. strong ties) is considered to be related to feelings of reciprocity [34]. Trusting relationships in which one feels close to the other are expected to be high in social capital [33].
Corresponding to the procedure to determine the original SNI scores (see model 2), scores are calculated taking the information on trust and strong ties into account. This results in the following scores: the number of high-contact, trustful and strong ties, the number of people one trusts, feels close to and talks to on a regular basis and the number of trustful, close and embedded networks.
2.2.1.4. Model 4: Bonding, Bridging and Linking Social Capital
In accordance with the definitions of Szreter and Woolcock [30], measures for bonding, bridging and linking social capital are calculated based on the sociodemographic characteristics of the respondent’s network members. Bonding social capital takes place between people with similar socio-demographic backgrounds [10]. Consequently, a score for bonding social capital was calculated, based on the homogeneity in the level of education, religion and ethnicity between the respondent and his/her family, partner, friends and the most important “other group”. The more people in these networks share the same level of education, ethnicity and religion as the respondent, the higher the score for bonding social capital.
Bridging social capital flourishes between people that differ on certain socio-demographic measures [10]. The score for this variable is based on the heterogeneity between the respondent and his/her family, partner, friends and the most important “other group” regarding level of education, religion and ethnicity. The more people in the network differ with the respondent on those sociodemographic characteristics, the higher the score for bridging social capital.
The score for linking social capital is based on the level of trust in professional caregivers expressed by the respondent. This measure is assessed by questions on the extent of the respondent’s trust in doctors in general as well as trust in his/her general practitioner. Respondents can express their opinion on the statements “To what extent do you trust doctors in general/your general practitioner” by means of a four-point Likert scale ranging from “completely” to “absolutely not”. The higher the level of trust, the higher the score for linking social capital. The original score is dichotomised with a score of 0 representing low (mean score of 3 or lower) and a score of 1 representing high linking social capital (mean score higher than 3).
2.2.1.5. Model 5: Theoretical Social Capital Index
The fifth model builds on a broad theoretical basis and distinguishes seven different components of social capital, grasping the multidimensional nature of the concept. These components are the number of available networks, the magnitude of networks, the behaviour of the respondent towards the network, personalized trust, reciprocity, the number of resources available in social networks and generalized trust.
The number of available networks represents how many of the twelve networks are available for the respondent. The magnitude of networks is calculated by adding the number of members in the available networks. The third component, the behaviour of the respondent towards the network, is operationalised by summing up the number of people with whom the respondent has regular contact. To measure personalized trust, the number of network members one trusts is determined. Reciprocity is inventoried by a proxy measure, namely the number of people to whom the respondent feels close. The number of resources available for the respondent is measured by adding the number of people with a higher educational level, a higher personal income or working in the health care sector within the networks “partner”, “friends”, “family” and “most important other group”. The last available component of social capital, generalized trust, is identical to the question on the basis of model 1.
2.2.2. Dependent Variables: Health Outcomes
Two health-related outcomes are included in the analysis, namely self-rated health and pain. To assess self-rated health, the most widely used single item to assess overall self-rated health [35] is used: respondents are asked to evaluate their general health status, using a five-point Likert scale ranging from “very good” to “very poor”. The original score is dichotomized for the analysis. A score of 1 represents good or very good perceived health and 0 represents very poor, poor or fair perceived health. To evaluate the impact of pain, respondents are asked how much pain they experienced during the past 4 weeks using a six-point Likert scale ranging from “none” to “very severe”. The original score for this variable is also dichotomized with 0 representing no or very light pain and 1 representing light, moderate, severe and very severe pain.
2.2.3. Background Variables
The socio-demographic variables included in the questionnaire are age, gender, country of birth of the respondent and his/her parents and educational level. Respondents are considered as being of Belgian origin if both they and their parents are born in Belgium. Educational level was divided into three categories: low educational level (i.e. no degree, finished primary school or the first years of secondary school), middle educational level (i.e. finished secondary school) and high educational level (completed higher education).
2.3. Procedure
In each neighbourhood, 50 adults were randomly selected from the municipal register, stratified by sex and age, representative for the neighbourhood’s sociodemographic composition. These were visited at home by an interviewer and invited to participate. Respondents who could not be reached or refused to participate were replaced by a randomly selected respondent from the same age and gender stratum until a total of 50 participants per neighbourhood was reached. After signing the informed consent, the interviewer handed over the selfadministered questionnaire and an appointment was made to collect the questionnaire later at the respondent’s home.
This study is approved by the Ethics Committee of Ghent University Hospital (project number EC/2008/355 and registration number B67020084311).
2.4. Data Analysis
The explanatory value of the five social capital models was compared for each health-related outcome separately. For this comparison, multiple logistic regression models were used. The values of the Nagelkerke pseudo R2 were used to estimate the variance in the outcome measure explained by each respective model.
Due to skew distribution, some of the independent variables (i.e. the number of people in model 2, the number of people one trusts, feels close to and talks to on a regular basis in model 3 and behaviour of the respondent towards the network, reciprocity and resources in model 5) were transformed by adding the constant value “1” to the original score and subsequently performing a log transformation. The transformed variables show less striking outliers and a more linear relationship to the outcome variables.
For most questions, respondents were offered five answering categories to estimate the number of people for which the question was applicable (i.e. “0”, “1”, “2 until 4”, “5 until 7”, “8 until 15” and “more than 15”). In the analyses, the mathematical mean of the intervals was used to estimate the respondent’s answer. The highest possible answer, “more than 15”, was given the value 15. The data are analyzed using the SPSS Software version 17 for Windows.
3. RESULTS
3.1. Response Rate
Of the 712 contacted inhabitants, 202 could not be reached (i.e. nonexistent address or not at home at three different points in time) and 131 refused participation. For the analysis, 379 respondents provided useful data, resulting in an overall response rate of 53%.
3.2. Description of the Sample
Characteristics of the respondent are described in Table 2. The number of men and women is analogous, with respectively 48.4% men and 51.6% women. The age of the participants ranges from 18 to 91, with a mean of 48.
Most of the respondents (N = 328) are of Belgian origin (i.e. both they and their parents are born in Belgium). One hundred and thirteen respondents (30.3%) have a low educational level (i.e. no degree, finished primary school or the first years of secondary school). One hundred and fourty-six respondents (39.2%) finished secondary school, while 114 respondents (30.5%) finished higher education.
Most of the respondents (298 versus 76) rate their general health as good to very good and about half of the sample reports having experienced no or very light pain in the last four weeks.
3.3. Comparison of the Model’s Explanatory Power
Table 3 gives an overview of the models’ ability to explain the observed variance in self-rated health and pain.