1 Introduction

Even after decades of studies on food security, the role of social capital on alleviating food insecurity has been largely overlooked. This under-emphasis of enhancing social capital as a potential mechanism for the mitigation of food poverty has to do with two factors: the lack of a solid theoretical framework that establishes the social capital-food security link, and an inconsistent and vague treatment of the concept of social capital. This paper, built on the premise that strong social cohesion exists among agrarian households in much of the developing world, hypothesizes that social capital can act as an affordable coping strategy to overcome many food security challenges. Cohesion may be motivated by the economic need for cooperation and mutual assistance, but in the specific case of Nepal it is chiefly born out of cultural-traditional roots. Here, we postulate that social capital can have multifaceted roles depending on economic needs. While non-vulnerable households may have proportionately more socio-psychological uses for social capital, for vulnerable households, it plays a cushioning role against potential covariate and idiosyncratic shocks.

As development resilience is being increasingly adopted by food security (FS) studies as an analytical framework, some focus has begun shifting towards the multidimensional, dynamic and sporadic nature of food insecurity (Barrett and Constas 2014; Upton et al. 2016). Such conceptualization allows the possibility of bringing forward other crucial determinants that have largely remained obscure owing to the lack of a proper conceptual framework. For instance, Barrett and Constas (2014) portray development resilience as a state variable representing some measure of wellbeing that becomes depleted or enhanced according to various dynamics: exposure to exogenous negative shocks reduces resilience whereas adaptive mechanisms add to the resilience stock. The aim of this paper is not to formalize this mechanism; nor is it to justifiably translate it into an empirical framework. In that sense, our goal is rather modest: we remain merely suggestive in that the development resilience framework can be a viable theoretical alternative that may provide a unique vantage point for discovering unexplored determinants of food security. Within this framework, social capital may be interpreted as an effective adaptive mechanism that can add to the resilience stock, which in our empirical demonstration represents the food security aspect of wellbeing. Admittedly, a faithful adherence to the resilience framework requires conceptualization of a socio-economic system as a dynamic entity with moving parts. Therefore, as a cautionary note, it should be borne in mind that the static, cross-sectional nature of the data used in this analysis precludes the tracking of movement of the relevant parts across time.

This paper diverges from extant studies in two ways. First, we account for the endogenous nature of individual level social capital, which, although seeming apparent, has not been the norm in most empirical studies. Second, we depart from the conventional categorization of social capital into bonding, bridging, and linking types, and instead classify them according to levels of operation. Doing so allows us to scrutinize both the “compositional” and “environmental” impacts of social capital on food security, on which we shall elaborate later. In line with Putnam (1995a, b), Coleman (1990), and Kawachi et al. (1997) among others, we use individual participation in voluntary groups and density of community groups as measures of individual level and community level social capital respectively. Our empirical findings confirm the link that exists between social capital and food security. We find that participation in different social groups can be an effective strategy to cope with severe food insecurity. However, we also show that not all forms of social capital have uniform impacts on all food security measures, and that generalizing social capital as a panacea for overall wellbeing improvement can be a misguiding principle.

The remainder of this paper is organized as follows. Section 2 succinctly outlines previous literature. Section 3 discusses data and measures used in the paper. Section 4 presents a conceptual and econometric framework employed for our analysis. Section 5 presents results based on empirical estimation and Section 6 concludes.

2 Literature review

Modern studies on the relationship between social capital and health can be traced back to Durkheim (1951), who argues that higher suicide rates can be explained by the extent of social disintegration and the consequential constraints that it imposes on moral forces of collective life. Although the term ‘social capital’ was not used until Bourdieu (1977), Durkheim’s work motivated a barrage of studies on various dimensions of what can be identified today as social capital. Among the first studies to ground the otherwise abstract, symbolic notion of social capital into an empirically testable framework is that of Coleman (1988), who presents social capital as “paralleling the concepts of financial capital, physical capital, and human capital” but one that was “embodied in relations among persons.” Since its systematic conceptualization by Coleman (1988) and popularization by Putnam (1993; 1995a, b), social capital has continued to garner generous attention from researchers across disciplines: economics (Becker and Murphy 2009; Dasgupta 2000; Murgai et al. 2002; Ostrom and Ahn 2008), sociology (e.g. Portes 1998; Sampson et al. 1997), psychology (e.g. Brown and Harris 1978; Kawachi and Berkman 2001), medicine and health (e.g. Rose 2000; Runyan et al. 1998), public health (e.g. Folland 2007; Whitley 2008), and disaster studies (e.g. Aldrich 2012a, b; 2012; Nakagawa and Shaw 2004), among others.

In recent decades, there has been an upsurge of empirical studies connecting social capital to health outcomes. Social capital positively affects self-rated health (Baron-Epel et al. 2008; Chen and Meng 2015; Kim et al. 2011; Poortinga 2006; Sirven 2006), mental health (Beaudoin 2009; Caughy et al. 2003; Fone et al. 2007; Harpham et al. 2004; Steptoe and Feldman 2001), and mortality rates (Berkman and Syme 1979; Lochner et al. 2003; Wilkinson et al. 1998). However, in areas specific to nutrition and food security, it remains relatively underemphasized. A handful of studies conducted in the United States (Dean and Sharkey 2011; Martin et al. 2004; Walker et al. 2007), by examining associations between social capital and food security among rural and/or low-income households, depict social capital as a support mechanism to improve access to food and/or its usage. Even fewer studies have explored the social capital-food security nexus outside the United States. In a study conducted in South Africa, Tibesigwa et al. (2016) suggest that the informal social capital can counteract agriculture-related shocks and sustain dietary requirements. In another study in South Africa, Misselhorn (2009) argues that social capital related failures can be linked to food insecurity. Sseguya’s (2009) findings in Uganda are also in line with those of the former two studies. No prior studies have explored this nexus in the context of agrarian households in Asia.

While social capital has a significant impact on health, the reverse is also true; that is, individuals with good health are better equipped to cultivate more social capital (Younsi and Chakroun 2016). A very few social capital studies pertinent to various health outcomes address this endogeneity concern using instrumental variables approach (e.g. d’Hombres et al. 2010; Folland 2007; Sirven 2006). However, among the studies that establish associations between social capital and food security, none has explicitly addressed endogeneity concerns. Therefore, extant papers do not provide adequate evidence to establish a causal mechanism by which social capital leads to better food security outcomes. This paper attempts to fill that gap by endogenizing social capital (SC) as an outcome of SC-specific investments of time and public speaking variables as proxies for individuals’ latent capabilities.

A major challenge that can preclude generalizability of social capital impacts lies in its contextual, often inarticulate treatment. A significant divide in the treatment of social capital is whether to consider it as a community characteristic (Kawachi et al. 1997; Varughese and Ostrom 2001) or an individual or household level characteristic (Rose 2000; Runyan et al. 1998). To circumvent these shortcomings and to retain policy-relevance, we include both community and individual level variables for social capital in our analysis. Consistent with Coleman (1990), Kawachi et al. (1999) and Putnam (Putnam et al. 1993; Putnam 1995a), we employ Woolcock’s (2001) definition of SC as “resources available to individuals through their social behaviors and memberships in community networks” for our analysis. We use community group density and individual participation in formal/informal groups as proxies for community and individual social capital.Footnote 1 Doing so allows us to examine both: i.) the “compositional effectFootnote 2”(Berkman et al. 2000), and ii.) the “environmental effectFootnote 3” of SC (Wilkinson 1992, 1996).

3 Data and measures

The data for this study comes from the baseline population-based survey (PBS) for Feed the Future (FTF) initiative in Nepal, a project led by the United States Agency for International Development (USAID). The baseline survey was conducted by the Feed the Future FEEDBACK (FTF FEEDBACK), a project jointly implemented by Westat, TANGO International, the International Food Policy Research Institute (IFPRI), and the Carolina Population Center (CPC) of the University of North Carolina at Chapel Hill. The survey, conducted in 2013, represents the geographic areas targeted by Feed the Future interventions, and is meant to serve in the assessment of FTF intervention impacts. In order to track post-intervention progress towards the proposed goals, information collected from the PBS-FTF survey was used to calculate indicators that measure women’s empowerment in agriculture, prevalence of households with moderate and severe hunger, and women’s dietary diversity.

The intervention-targeted geographic areas, named by FTF as Zones of Influence (ZOI), constitute 20 districts across the western, mid-western and far-western development regions in Nepal. These three development regions of Nepal are among the most food impoverished in the already “severely food deficient” country with a per capita GDP of less than $750 (2016 estimate) (The World Bank 2017). A total of 2000 households spread across 100 clusters within the 20 districts in the ZOI were interviewed during the data collection process. Unlike other population surveys in Nepal (Nepal Demographic and Health Survey-NDHS, Nepal Living Standard Survey-NLSS), unique to the Nepal Baseline PBS (2013) questionnaire is its inclusion of special modules on prevalence of hunger within households, women’s dietary diversity, and women’s empowerment index. Furthermore, questions on food items for women’s dietary diversity are adapted to fit the local context.

The primary unit of analysis in this paper is at the individual level, albeit only women respondents are considered because information on dietary diversity is limited to women. Accounting for missing observations for relevant variables, the final data for our analysis includes 3211 observations (all women from 15 to 59 years of age). The variables used for this study, along with descriptive statistics, are compiled in Table 1. Further details are provided in the succeeding sub-section:

Table 1 Description of variables
Full size table

3.1 Variables

Our dependent variable is hunger scale, a measure of degrees of food insecurity, which has four ordered categories: no food insecurity (0), low food insecurity (1), medium food insecurity (2), and high food insecurity (3). This scale, developed based on a set of survey questions that are meant to elicit the frequency and intensity of extant hunger, represents the prevalence and rate of food insecurity within the household. The household member responsible for food preparation was asked the following questions: 1) In the past [4 weeks/30 days] was there ever no food to eat of any kind in your house because of lack of resources to get food? 2) How often did this happen in the past [4 weeks/30 days]?3) In the past [4 weeks/30 days] did you or any household member go to sleep at night hungry because there was not enough food?4) How often did this happen in the past [4 weeks/30 days]? 5) In the past [4 weeks/30 days] did you or any household member go a whole day and night without eating anything at all because there was not enough food? 6) How often did this happen in the past [4 weeks/30 days]? The first, third, and fifth questions had binary responses (yes or no), whereas the second, fourth, and sixth allowed for a third alternative to account for frequency (never, rarely, sometimes, often). Following (Ballard et al. 2011), these frequencies were collapsed into three responses: never (0), rarely or sometimes (1), often (2). A composite household hunger scale was created by summing the collapsed measures, producing a raw hunger scale (HHS) ranging from 0 to 6. Subsequently, we used FTF conventions to categorize the raw hunger scale into four bins to indicate severities of hunger: no hunger (HHS = 0), low hunger (HHS = 1), moderate hunger (HHS = 2–3), and severe hunger (HHS 4–6), thus creating the variable hunger scale that we used in our analysis.Footnote 4

The other variable used to measure the qualitative dimension of food security is Dietary Diversity (DD), which is a validated measure of micronutrient adequacy of diets (Feed the Future FEEDBACK 2013). DD, which is the mean number of food groups consumed, was generated using questions on food consumption the previous day (that is, “yesterday during the day or night”). Adapted to fit the nutritional context of Western Nepal, FTF categorizes all consumed foods into nine groups: (1) grains, roots, and tubers; (2) legumes and nuts; (3) dairy products; (4) organ meat; (5) eggs; (6) flesh foods and other small animal protein; (7) Vitamin A dark green leafy vegetables; (8) other Vitamin A-rich vegetables and fruits; and (9) other fruits and vegetables. Note that only women of reproductive age (15–49 years) were asked the module containing these questions.

Internalizing a barrage of criticisms on the context-dependency and multifaceted nature of social capital, we used a “deconstructive” approach in that we broke down the notion of social capital into its constituent levels that best illustrate the mechanisms by which it affects food security outcomes. We acknowledge that not all forms of social capital can be lumped together into a generalizable indicator of social capital in order to force a coherent narrative on the social capital-economic outcomes nexus. At an individual level, participation in social networks, groups and associations can provide members with resources and information that may lead to positive outcomes, which is referred to as the “compositional effect” (Sirven 2006). At an aggregate level, social capital may have “environmental effects” through buttressing social cohesion and engendering collective endeavors, which in turn could have positive behavioral outcomes (Sirven 2006; Wilkinson 1992, 1996). In order to account for both compositional and environmental effects, we used association variables based on individual participation in community groups and the density of such groups in the community in our model. To further understand the varying impacts of different types of association, we categorized community groups into three types: finance-related, informational, and other associations. The guiding hypothesis is that, while all forms of associations can be helpful in improving other wellbeing measures, the type that best targets food security issues is finance-related. Finance related groups include: credit or microfinance groups, trade or business associations, and mutual insurance groups, whereas informational associations include agriculture, water, and forest groups. Participation in all other forms of associations (civic, charitable, and religious groups) are lumped into the third category. Each of these three variables take values that range from 0 to 2, where 0 indicates no participation, 1 represents participation in one voluntary association, and 2 represents participation in two or more voluntary associations. While the former variables represent individual level participation in community associations, the “environmental” impact of social capital is elicited using a community level variable, namely Community group density, which captures the number of such associations present in the locality.

Other variables used in the model are enlisted in Table 1.

4 The empirical model

The conceptual framework employed in the empirical analysis is represented using a two-equation system in a recursive modeling set up, where we allow for contemptuous correlation across equations. The equations employed for empirical evaluation are:

$$ {FS}_i={\beta}_0+{\beta}_1{ISC}_i+{\beta}_2{CSC}_{loc}+{\beta}_3{X}_i+{u}_i $$
(1)
$$ {ISC}_i={\upgamma}_0+{\upgamma}_1{Z}_i+{\upgamma}_2{X}_i+{v}_i $$
(2)

In the first equation, FS represents two food security measures: first, the prevalence of hunger as reported by the respondent and second, women’s dietary diversity. Prevalence of hunger (HS) is reported in a scale ranging from 0 to 3 with 3 referring to severe hunger. HS is determined by individual level social capital (ISCi), community level social capital (CSCloc), and household characteristics (Xi). We postulate that individual participation in voluntary associations is endogenously determined, as confirmed by many studies before ours (e.g. Glaeser et al. 2002). To account for this, we instrument it with variables indicating individual investment in social capital (variable: time allocated for social activities) and social skills (variable: comfort in public speaking). These two variables are represented in equation-2 by vector Zi. First, we establish the relevance and exclusion criteria to justify the choices of our instruments, and proceed to further examine the strength of these instrumental variables using LR tests. These processes will be discussed in the succeeding results section. βs and γs are parameters to be estimated. It should be noted that Eq. (2) in the above model represents a set of up to three equations representing different categories of individual social capital depending on the model specification. However, for representational simplicity, we depict them as a unit. As iterated previously, the empirical framework employed for this analysis allows for contemporaneous correlation across equations, estimating Eqs. (1) and (2) simultaneously. We assume that error terms follow a multivariate normal distribution such that:

$$ \in =\left[{u}_i,{v}_i\ \right]\sim N\left(0,\Sigma \right)\kern0.24em \mathrm{where},\sum =\left[\begin{array}{cc}1& \sqrt{{}_{22}}\\ {}\sqrt{{}_{22}}& {}_{22}\end{array}\right]\kern0.5em \left(\mathrm{normalizing}\ {\sigma}_{11}=1\right) $$
(3)

5 Results

Prior to proceeding to model estimates, we first tested the appropriateness of our modelling approach, and examine the reliability of instruments employed for the analysis. To confirm the suspected case of endogeneity, we evaluated the Fisher’s z-transformed correlation parameters (inverse hyperbolic tangent of rho, tanh−1ρ) of our full model. We rejected the null hypothesis that they are equal to zero in 11 out of 12 equation match ups (Table 4 in appendix). Note that although the instrumental variables were carefully selected from among available variables based on the established convention in the extant literature (e.g. Glaeser et al. 2002),Footnote 5 no single instrument employed is sufficiently strongFootnote 6; that is, the F-test of the excluded instruments generated a value<10, which is less than the ‘rule of thumb’ value of 10 (Staiger and Stock 1997). When one (or few) instrument(s) is (are) not strong enough and the variance of the two-stage least squares is high, a natural solution is to add more instruments to reduce the variance. However, that has its costs: that is, adding instruments that add little to R-square increases the finite-sample bias even in large samples (Murray 2006). In consideration of these issues, we employed the full-information maximum likelihood (FIML) approach that allows for contemporaneous correlation as it has better finite-sample properties and addresses these issues. Also, we chose FIML over limited-information maximum likelihood (LIML) because FIML generates standard errors that are moderately smaller than when LIML is used (West 1986).

As in the linear simultaneous-equation model, the order condition for identification requires that the number of excluded exogenous variables (that is, the additional instruments) be at least as great as the number of included endogenous variables. This was achieved by including social capital investment and comfort in public speaking variables in the ISC equation (s) that were excluded in the FS equation (s). The strength of these variables was tested using likelihood ratio (LR) test in the first stage equations comparing the restricted model with no instruments against the unrestricted model with instruments. In each case, LR (chi-squared) value was significantly large, indicating that additional variables in the unrestricted model were jointly significant (Table 5 in appendix). Alternatively, LM and CM tests were also conducted to verify LR test results (not reported in the paper).

5.1 Social capital and hunger

Table 2 reports recursive estimates of the impact of social capital on hunger. For robustness purposes, we tested different model specifications of Eq. (1). Based on the comparison of Akaike Information Criteria (AIC) values and relative gain/loss of explanatory power, we deemed that the third column (Model 2) is the preferred model. In Model 1, we only report the impact of individual participation in finance-related individual social capital, while controlling for socioeconomic and household characteristics. Results indicate that participation in finance-related associations has negative and significant impact on the prevalence of hunger. Coefficients for control variables show that agricultural land, literacy, and residential status (urban) all play positive roles in hunger mitigation. On the other hand, single-parent families with female household heads are more vulnerable to episodes of hunger. In Model 2, we added community social capital (CSC) variable, represented by the density of community groups in the household’s immediate locality, to the base model (1). We found the presence of “environmental effects” of community social capital, regardless of their participation in the respective groups/associations. Model 3 includes individual SC variables of two types, financial and informational, while excluding CSC. Results for finance-related ISC remained unchanged as compared to Model 1, but we found that participation in informational groups had no significant impact on hunger mitigation. In Model 4, we expanded on Model 3 to also include CSC. Results for finance-related ISC and CSC remained steady, whereas, once again, we found no significant impact of informational ISC. Model 5 excludes CSC, but includes all three forms of ISC: finance-related, informational, and others. Once again, we found that only finance-related ISC had significant impact. Model 6 expands on Model 5 by adding CSC. Results for three forms of ISC remained unchanged, and we also found steady (significant) impact of CSC.

Table 2 Recursive model estimates for hunger-scale
Full size table

We found that finance-related individual social capital and community social capital had consistent impacts on hunger mitigation across all model specifications. So, based on the evidence from Table 2, we can safely assert that, for households on the cusp of extreme food poverty, only finance-related associations played significant roles in hunger mitigation. We found that involvement in other forms of associations that do not directly enhance households’ financial capital had no direct impact on hunger mitigation. On the other hand, community-level social capital (density of community groups) had a positive and significant role in hunger mitigation. This suggests that a community’s social capital endowment can have a public good nature in that it benefits all its members, regardless of their participation in voluntary associations. Across all model specifications (Models 1–6), we found that agricultural land, literacy, and residential setting (urban vs rural) play hunger mitigation roles, whereas family type (household head: female only) seems to show inconsistent impact.

5.2 Social capital and dietary diversity (nutrition)

Switching our focus to the qualitative dimension of food security indicated by the nutritional indicator, dietary diversity, results in Table 3 paint a slightly different but complementary picture of social capital impacts. Consistent with results from Table 2, the role of finance-related ISC on food diversity was robust across all model specifications (columns 2 through 7 in Table 3). That is to say that finance-related ISC not only helps with hunger mitigation but also plays a vital role in increasing nutritional quality. What is distinguishable in this analysis as compared to the results from hunger scale analysis is that informational ISC also has positive and significant impacts on nutritional quality. This reveals an interesting dimension of the causal mechanism by which different participatory associations impact food security. Consistent with the “compositional effects” hypothesis that was discussed earlier, our results support the postulate that voluntary associations can provide their members with social support, information, and appropriately incentivize them to adopt healthy behavior. While informational ISC may not have direct hunger mitigating roles, it contributes to food security through indirect channels such as knowledge-sharing, behavioral adjustment in dietary habits, and so on.

Table 3 Recursive model estimates for food diversity
Full size table

Based on an evaluation of AIC-BIC values for different model specifications and their corresponding tradeoffs in terms of interpretability, we deemed that model 5 best concurs with the narrative of this paper. Nonetheless, to check for the sensitivity of our findings, Table 3 presents results across different model specifications. Models 1–6 include socioeconomic and household controls. In Model 1, only finance-related ISC variable is included; Model 2 adds informational-ISC to the specifications in Model 1. We found that the impact of both ISCs, finance-related and informational, remain highly significant. Model 3 adds another ISC variable (other associations). This time, we found no significant impact of other ISCs, but those that are finance-related and informational remained unaltered. Model 4 only includes finance-related ISC and CSC; Model 5 includes both finance-related and informational ISC along with CSC, and model 6 includes all three ISCs and CSC. Results for finance-related and informational ISC remained robust across all specifications. However, the impact of CSC on dietary diversity was sensitive to model specification. When only finance-related ISC and CSC are included (Model 10), we find that CSC coefficients were significant at the 90% confidence level, but this dissipates once we included other ISC variables, so our results precluded a generalizable claim regarding the impact of CSC on dietary diversity. As far as the controls were concerned, results indicated that literacy, urban-rural divide (urban = 1), age, and livestock assets positively contributed to dietary diversity. Contrasting coefficients for controls in Table 3 with those for Table 2, we see that livestock possession seems to add to dietary diversity. This is presumably due to the increased access to certain food groups such as meat, eggs, and/or milk associated with owning more livestock. On the other hand, while having more agricultural land has positive impacts on hunger mitigation, we see that it has no role on dietary diversity.

6 Conclusion

In this paper, we attempt to empirically establish social capital effects on food security. However, the goal was not merely to link the abstract notion of social capital to improvements in food security status, but rather to explain the mechanisms by which different forms of social capital can have different roles depending on the outcome measure at hand. We did so by dichotomizing SC impacts into two levels: individual and community. We further split individual SC into financial, informational, and others. This allows us to identify how SC can have multifaceted roles in different aspects of individual lives. To further bolster a causal SC-FS relational claim, we accounted for the endogenous nature of individual social capital in a recursive modeling set up. In general, our findings support the assertion that social capital positively impacts food security. This complements a barrage of prior studies that overwhelmingly demonstrate a positive relationship between SC and various health outcomes. Participation in finance-related associations leads to hunger mitigation, whereas participation in informational and other associations do not do so. However, informational associations do have positive effects in improving the qualitative aspect of food security—nutrition (food diversity). At the community level, we found consistent evidence of positive “environmental effects” that the density of formal and informal groups in a locality can have across all food security measures.

In the first section of the paper, we speculated on potential reasons why only a dearth of SC-FS studies exists. Our prime suspect was the lack of a systematic theoretical framework that can be used to formulate the SC-FS nexus. The suggested remedy was to advance development resilience as a viable framework to study food security issues, especially so in an agriculture-based developing country setting where episodes of extreme hunger are sporadic and severe. Such conceptualization helps the understanding of extreme food poverty as systemic or idiosyncratic shocks, and various coping mechanisms, including social capital enhancement, as contributing to strengthening adaptive capacity of households to overcome such shocks. Our findings provide a strong signal of the viability of development resilience as a possible research track that deserves greater attention.

Voluntary participation in community associations should not be confused as a panacea for combatting all disparities. Instead, social capital should simply be conceived as a cheap and accessible coping strategy that can boost households’ adaptive capacity in the face of dire food insecurity. This paper remains silent regarding the precise mechanism dictating social capital enhancement, for that requires further analysis, accounting for the significant socio-cultural heterogeneity that is prevalent in western Nepal. Moreover, of the four known dimensions of food security, − availability, access, utilization, and stability – our study directly addresses only two: access and utilization. Therefore, the findings of this study are not conclusive and generalizable to all aspects of food security. That said, we successfully establish that social capital is an important determinant of food security that should not be overlooked. Taking into account the traditional norms, institutions, and deep ties to eco-system services in the region, we advocate a customized approach to addressing food security challenges and contend that a one-size-fits-all approach to food policy, which does not acknowledge the rich social fabric that connects households in these food-impoverished regions is sub-optimal at best.