Social Capital and Health: Empirical Evidence from Indonesia

Through the Indonesian Family Life Survey (IFLS4 and IFLS5), this paper examined effects of individuals’ social capital on various health outcomes, analyzing specifically whether generalized trust, bridging social capital (trust in many ethnic groups, trust in many religious beliefs), and social isolation had effects on various health outcomes. Estimated results for fixed-effects models did not show significance in any health outcome variables, except for social isolation, which had a positive effect on self-perceived health. Moreover, estimated results for random-effect probit models showed that generalized trust had weak negative effect on self-perceived health, while trust in many ethnic groups had negative association. Meanwhile, trust in many religions had no effect on health outcomes. In addition, social isolation had negative effects on most self-rated health and mental health outcomes. Modal Sosial dan Kesehatan: Bukti penelitian dari Indonesia


Introduction
Originating from sociology, the concept of social capital has grown rapidly through various disciplines, for instance economics, education, politics, and so on, and therefore, social capital's definition is multi-dimensional. With emphasis on trust, Guiso, Sapienza, and Zingales (2008) argued that social capital is a set of beliefs and values that facilitate cooperation among members of a community. This research, then, focuses on a micro dimension of the social capital concept to which Guiso et al. (2008) refer. Additionally, in both public health and economics, social capital's effects on health have recently attracted increased attention. Some studies have investigated roles of various aspects of bonding, bridging, and linking social capital. Examining effects of these three dimensions of social capital in building community resilience and health, for example, Poortinga (2012) found in secondary analysis of 2007 and 2009 Citizenship Surveys that most indicators of bonding, bridging, and linking social capital were December 2018 ½Vol. 22 ½ No. 2 significantly associated with better community health.
Thus social capital seems to benefit health. As economic growth becomes strong, health outcomes seem to indicate improvement in welfare, while low income has been argued to contribute to poor health, especially in developing countries. Pritchett and Summers (1996) found that higher income inequality was often associated with poorer population health. Moreover, Kawachi, Kennedy, and Wilkinson (1999) stated that income inequality is a reason for society not to communicate well across income barriers, and thus it reduces opportunities to build social capital. However, social capital is needed for promoting economic growth through accessing external assets and sharing information (Putnam, 2000). Poor health, then, links to poverty because of lower social capital. Therefore, this paper examined social capital's impact on individual health in Indonesia.
Meanwhile, with its diversity of cultures and religions, Indonesia has many activities reflecting cooperation, which could stimulate social capital as measured by trust. Unfortunately, social cohesiveness is not likely to exist in Indonesia these days because people have become distrustful of each other and created conflict across ethnic and religious groups. Baum's (1999) observations about the negative impact of bonding social capital has shown that the Indonesian people still have strong bonds within the same ethnicity and religion (bonding social capital), which might not be healthy for those who disagree with the majority. Therefore, this study focused on bridging social capital since it seems appropriate to Indonesian characteristics.
Individuals better connected with their community, i.e., having high social capital, have positive health outcomes because of their sense of belonging. For people with chronic illnesses, social capital can be a potential supportive resource that encourages them to cope with frail conditions (Waverijn, Wolfe, Mohnen, Rijken, Spreeuwenberg, & Groenewegen, 2014). However, mechanisms by which social capital affect health differ, depending on the specific dimension of social capital.
Previous literature on the relationship between social capital and health showed an estimation of potential endogeneity that stems from unobservable heterogeneity and reverse causality (Rocco, Fumagalli, & Suhrcke, 2014). For example, Ho (2016) estimated the model using a number of friends as a proxy of social capital and warned of the possibility of unobservable factors that affect individuals' decisions in selecting their friends and perhaps influencing their health. Previous literature used an instrumental variable estimate to tackle the endogeneity problem (Hollard & Sene, 2016). Following Rocco and colleagues (2014), however, this research estimated by using individual fixed effects in health production to control unobservable heterogeneity.
Empirical studies on social capital and health are still limited to two in Indonesia (Miller, Scheffler, Lam, Rosenberg & Rupp, 2006;Sujarwoto & Tampubolon, 2013). However, these studies were cross-sectional and did not account for unobserved heterogeneity among individuals. Thus, this study estimated a fixed-effects model using the Indonesian Family Life Survey (IFLS) 2007 and 2014 to examine social capital's effect on health. Moreover, social capital variables in this study were generalized trust and trust in different ethnic groups and in different religions. To the author's knowledge, these proxies have not been discussed in the social capital field. For health outcome variables, two aspects of self-reported health were examined, self-rated health and mental health.
This study's aim was twofold: first, after controlling sociodemographic determinants, to investigate strength association (ordinary least squares [OLS] and fixed effects) among four proxies (generalized trust, trust in different ethnicities, trust in different religions, and social isolation) as bridging social capital on individual self-perceived health; second, to examine how these proxies relate to self-rated health based on occurrence of five chronic diseases and psychological health, together with other health outcomes, using pooled probit and random-effects probit methods.
Estimated results for fixed-effects models have not shown significance in any health outcome variables except for social isolation, which had a positive effect. Moreover, estimated results for random-effect probit models showed that generalized trust had weak positive impacts on self-rated health, while trust in many ethnic groups had negative association. Meanwhile, trust in many religions did not have an effect on different health outcomes.
The remainder of this paper is structured as follows. The second section reviews previous literature on social capital and health. The third section describes the methodology and data, and then the fourth section presents empirical results and the discussion. The final section provides concluding remarks and some policy implications.  (Boehm, Eisenberg, & Lampel, 2011). In addition, social capital's third dimension is linking social capital, suggested by Szreter and Woolcock (2004) as relationships related to differences in power or resource bases and status, for instance, the government and the police. This distinction is crucial since people live in heterogeneous societies.

Social
Meanwhile, Guiso et al. (2008) defined social capital as an individual belief about others' willingness to cooperate. Charles and Kline (2006) also mentioned that trust makes cooperation easier, with more trust network is associated with bigger return for individuals. When defined in this way, social capital can be considered a form of capital, achieved from durable social relations among individuals, which they can accumulate, transfer, and access for their particular interests. However, social capital is not the same as human capital because its returns depend on other community members' norms and beliefs. Production of capital requires opportunity costs, and, clearly, social capital has opportunity cost, for example, in time and effort, to maintain relationships and to build trust with community members (Folland & Rocco, 2014).
Moreover, selection of social capital's proxy is based on the study area's characteristics (Grootaert & Bastelaer, 2001). Since Indonesia has many traditional activities involving communities rich in social ties, with a diversity of ethnic groups and religious beliefs, bridging social capital is the most suitable for Indonesia's characteristics, but with focus on the individual through cognitive assessments.

Social Capital and Health. Villalonga-Olives and
Kawachi (2015) explained how bridging social capital is linked to health, by: (1) improving access to valuable resources such as health-relevant information; (2) improving access to instrumental and emotional social support that have been widely reported to affect health behaviors and outcomes; (3) social reinforcement of identity and mutual influence; and (4) fostering stronger connections across social divisions within the community-thereby strengthening collective ability to voice demands or to undertake coordinated action. These mechanisms are then pictured as "opportunity based," rather than having a psychological and behavioral nature.
Bridging social capital can be illustrated by a poor man obtaining a cash loan from a person who possesses valued resources and trusts the poor man. Trust helps the disadvantaged man gain access to a resource (cash loan) so that he also gains from his relationship with the other person. Bridging social capital might generate health benefits because actors can access valuable assets and information related to health via connections with heterogeneous individuals. Therefore, bridging social capital can play an important role in overcoming health inequalities in Indonesia, since people trapped in poverty often lack bridging social capital (Villalonga-Olives & Kawachi, 2015).

Methods
Empirical Model. Using panel data from 2007 to 2014, this study employed a simple specification following Ronconi, Brown, and Scheffler (2012) to model the relationship between individual-level social capital and health: H_it= β〖SC〗_it+X_it δ +u_it (1) where u_it=α_it+ε_it "# is self-perceived health for individual i, a categorical variable that takes values from 1 to 4 (1 = very healthy, 4 = unhealthy); "# is a set social capital for individual I; "# is a set of exogenous controls including age, education, marital status, household income, micro criminality, area, homeownership, and children under 17 years old; is the parameter of interest and captures the effect of trust (generalized trust, trust in many ethnic groups, and trust in many religions) and social isolation from health problems. "# contains time invariant fixed effects "# and "# , which follows independent and identically distributed random variables. Self-assessed health was treated as a continuous variable and estimated with Eq. (1) using pooled OLS and fixedeffects estimation.
Moreover, self-rated health was observed based on occurrence of five chronic diseases as a binary choice that had the value of 1 if the respondent had no health problems, and 0 otherwise. This study also did the same for mental health. Since the dependent variable was limited, the following equation was estimated by pooled probit and random-effect probit.
where Φ(. ) is the cumulative distribution function of a normal standard. Dependent variables. Self-perceived health was measured by asking "In general, how is your health?" This variable was considered continuous. Both time periods had the same pattern.
Self-rated health was assessed based on occurrence or absence of five chronic conditions (hypertension, diabetes, respiratory disease, heart disease, neurological disease). Respondents who answered that they did not suffer from any of the diseases had a value of 1; those with at least one chronic condition had a value of 0. The dataset for 2007 and 2014 were the same.
Mental health was established by asking respondents "Have you felt happy, anxious, restless, bothered, fearful, or depressed in the past week?" For 2007, the answer was: "Yes" or "No" (discrete variable). For 2014, the answer was respondents' frequency of feelings. It was a categorical variable. Responses were: "Rarely or none (≤1 day)," "Some days (1-2 days)," "Occasionally (3-4 days)," or "Most of the time (5-7 days)." For 2007, mental health had a value of 1 if respondents reported they did not feel any of those conditions, except happy, and 0 otherwise. Those who felt happy had a value of 1, and 0 otherwise. Moreover, for 2014, mental health had a value of 1 if respondents reported "none" or "rarely" for all those feelings, except happy; the rest had a value of 0, except for happy. Moreover, respondents who reported being happy on "some days," "occasionally," or "most of the time" equaled 1, and the rest, 0.

Independent variable: "Bridging social capital"
variables. Generalized trust was measured by respondents' ranking of the statement "In this village, I have to be alert, or someone is likely to take advantage of me." This variable was treated as continuous.
Trust in many ethnicities was assessed by respondents' ranking of the statement "Taking into account the diversity of ethnicities in the village, I trust people with the same ethnicity as mine more." This was also treated as continuous.
The question about trust in many religions was similar, and questions and answers for the independent variable above were the same for both periods.
Social isolation was measured by respondents' answer to the question "Do you feel lonely," and the answers for 2007 and 2014 differed, but they were the same as those for mental health. For 2007, social isolation had the value of 1 if respondents felt lonely, and 0 otherwise. For 2014, social isolation had the value of 1 if they answered, "some days," "occasionally," or "most of the time"; the rest were zero.

Control Variables.
To account for other phenomena that might influence health and social capital, following Fiorillo and Sabatini (2015), a set of individual and household control variables were included in the analysis.
At the individual level, gender (female), marital status (married), age, education (higher education), and quality of the surrounding social environment according to an indicator of subjective perception of its safety (how safe respondents considered the village) were considered.
At the household level, following previous literature, for household income (sum of labor income, capital income, and pensions) were considered. In addition, the number of children less than 17 years old, home ownership (binary variable) and area (urban) were considered. Table A1 (presents a statistical summary of all variables. On average, the final sample's average age was 43. About 79% of respondents reported good health, whereas only 38% of people did not suffer from mental health problems. Moreover, only 8% of Indonesian people felt sad. Almost half the respondents were female, and more than half were married. Almost half reported higher education. Only 21% of respondents had children less than 17 years old, and over half were homeowners.  Notes.

Determinants of Self-perceived Health
(1) Robust standard errors in brackets; (2) *** p<0.01, ** p<0.05, * p<0.1 Unexpectedly, social isolation was significantly and positively correlated with good health, with isolated individuals having self-perceived good health at 9.7%. In contrast, many Indonesian people like to interact with each other, but sometimes they enjoy talking about negative things; then, they become worried and rate themselves as "not in good condition." Furthermore, news in Indonesia has recently focused on negativity, to include corruption, crime and so on. As a result, the news audience becomes stressed and subsequently rates their health as poor. Table 2 shows results from estimation of social capital's effect on health from the nonlinear equation (2) in the empirical model. Column 1 reports estimation of the pooled probit and randomeffects probit model for self-rated health based on occurrence of five chronic illnesses; Column 2 reports estimation of the pooled probit estimation and randomeffect probit for mental health. Results, presented in Table  3, are transformed to probability derivatives.

Self-rated Health and Mental Health: Pooled Probit and Random Effects Probit.
The two results give the same direction of coefficients except for the dummy year 2007, although it is not significant. Interestingly, generalized trust becomes significant when using random-effect probit and has the expected sign. Thus, social capital, represented by generalized trust, increases the probability of reporting good health by 1%. Although this is weak evidence, the result aligns with our hypothesis. Similarly, trust of different religious beliefs reports a positive, but insignificant, coefficient. On the other hand, trust in many ethnic groups is strong and negatively associated with self-rated good health. Individuals who trust many ethnic groups are 1.6% more likely to report poor health. This finding supports Baum's (1999) suggestion that Indonesians might have a strong bond with the same ethnicity, thus causing them to be exclusionary and distrustful with dissimilar others, so this condition may not be healthy for outsiders. Furthermore, socially isolated individuals are about 5% less likely to consider themselves in self-rated good health. This result holds for both estimations. Isolated people appear to have less support and mutual respect for mitigating stress, and this circumstance affects their health. This also confirms d 'Hombres, Rocco, Suhrcke, and McKee (2010) that loneliness has a significantly negative impact. Lastly, no significant events affected self-rated health in 2007.
Furthermore, Column 2 presents the effect of bridging social capital on mental health. Both estimates also give the same direction on coefficients. Interpretation of results based on random-effect probit estimates shows that all trust variables were not significantly correlated with good mental health. Social isolation had a strong negative impact on the likelihood of good mental health.
Interestingly, the time dummy displays a significant association with good mental health. Indonesia faced a financial crisis, which was part of the Asian financial crisis, in 2007 and 2008. This 2007 -2008 crisis was the second economic crisis since the first crisis in 1997 and 1998. Although Indonesian people were hit by the crisis, they were not greatly affected. Indonesia was more resilient than in the first crisis in 1997 and 1998. Not only had the Indonesian economy performed well from 2005 to 2008 but also consumer prices remained stable because the Indonesian people learned from the first crisis; as a result, mental health was reported as good.
In summary, our findings provide little support for the hypothesis that social capital positively correlates with self-rated health. However, trust in many ethnic groups had significant negative effects for each mental health outcome. Sujarwoto and Tampubolon (2013) argued that individuals who belong to the same ethnic group were more likely to have social relationships due to similarities of custom or language. Moreover, Alesina and La Ferara (2002) found that, in more heterogeneous communities, trust was lower because interracial contacts were more frequent. For this reason, even though Indonesian people trust everyone (generalized trust), thereby leading to positive health effects, this does not mean trust in many ethnic groups will succeed in Indonesia because it is more sensitive and deeper than generalized trust, especially considering Indonesian characteristics. From Putnam's (2007) insight, we can conclude that building trust and creating bonds might require more time among varied ethnic groups and might then affect their health. Therefore, further research is needed on this topic.
In addition, even trust in many religious beliefs did not significantly correlate with all health outcomes. The joint test of significance for four indicators of social capital strongly rejected the null hypothesis with p-value<0.0000 (OLS, probit, random-effect probit estimations) and p-value<0.097 (fixed-effect estimation).

Conclusion
This paper presented the first examination in Indonesia (2007Indonesia ( & 2014 of the effect of bridging social capital, represented by trust in many ethnic groups and in many religious beliefs along with two other measures, generalized trust and social isolation, on different health outcomes. The paper's contribution is threefold: first, providing evidence on the role of bridging social capital in differing health outcomes-proxies never previously assessed in Indonesia; second, examination of the relationship between two variables using panel data; and third, more objective measures of self-rated health based on occurrence or absence of five chronic diseases.
The fixed-effect estimate did not reveal any significance for the estimated parameter of interests, except for social isolation on self-perceived health. However, since some unobserved individual effects correlated with social capital variables, using fixed-effect methods was appropriate for this estimation to rule out potential omitted variable bias. Moreover, random-effect probit suggested that generalized trust is associated with small and statistically significant increases in the likelihood of good self-rated health. Even though it is weak evidence, this social capital plays an important role in individual health conditions. However, trust in many ethnic groups was strongly and negatively associated with self-rated good health. Therefore, further research is needed for this topic and, particularly, better data, preferably longitudinal, are necessary for better identification, both theoretical and empirical. In addition, social isolation showed a negative effect on most self-rated health and mental health outcomes.
This study has a number of limitations. Because there are no appropriate instruments, the possibility of reverse causality was not considered, and thus, a fixed-effects instrumental variable method would be more appropriate. Second, we used only two time periods over a 7-year span. This potentially reduced the estimate's precision. Third, this study relied on only one of two item scales of social capital; hence, a weak or inconsistent connection with health was found. Moreover, this study examined only bridging social capital; however, future studies should consider other dimensions, i.e., bonding social capital and linking social capital, to discover which is the most appropriate for Indonesia. Finally, due to incidental parameter problems, the fixed-effects model for the limited dependent variable was not estimated. Further robustness checks are needed if our results for the random-effects model hold when controlled for fixed effects.
In sum, this study has shown empirically that, in social capital's effect on health, generalized trust has positive impact, even though it is small. Meanwhile, trust in other religions and in ethnic groups did not produce a satisfactory result. Likely, Indonesians still have strong bonds within ethnic and religious groups. If, as Putnam (2000) states, bridging social capital is important for achieving economic growth, it would be advantageous for Indonesia, with its rich diversity, if this dimension could be attained and utilized. For this reason, the government, as policymaker, should stimulate development of social capital, especially bridging social capital, by funding projects that allow it to be easily formed and by triggering enthusiasm among communities and their leaders to develop social capital.

Declaration of Interest
The author reports no conflicts of interest in this work.