Résumé
Social capital is critical for economic development, especially in settings where institutions are absent or weak (Cardenas & Carpenter, 2008). Social capital, defined as "the social networks and the norms of reciprocity and trustworthiness that arise from (individuals)" (Putnam, 2000), is a theoretical construct that cannot be measured directly in its entirety, but rather its various elementary components need to be assessed. Recent literature has focused on two main components: structural and cognitive. The structural component encompasses roles, rules, procedures, and social networks, while the cognitive component includes pro-social attitudes such as trust, truthfulness, trustworthiness, reciprocity, and willingness to contribute to public goods. These two components are interconnected and mutually reinforcing, making it difficult to disentangle them. (Krishna, 2007; Avdeenko and Gilligan, 2015; Pisani, 2017; Uphoff and Wijayaratna, 2000). Social capital is beneficial for economic and community development as it allows for information sharing, pooling of resources, reduction of transaction costs and facilitate informal contracts (Cameron et al., 2015; Heß et al., 2021; Peralta & Shupp, 2017). More generally, communities with high social capital are better suited for collective action which in turn positively affects their economic outcomes. The development community has increasingly turned to community-driven development (CDD), a bottom-up approach in which the targeted population becomes the primary driver of change. Such initiatives both rely on existing social capital and aim to enhance it (Avdeenko & Gilligan, 2015; Ban et al., 2020; Mansuri, 2004). The success of CDD is closely linked to pre-existing levels of trust, trustworthiness, as well as community interactions (Peralta & Shupp, 2017). Trust and trustworthiness are intrinsic values that depend on individual preferences, making it challenging for external actors to build social capital. However, development projects can increase community interactions through meetings, trainings, and joint activities, which in the long run could increase trust and cooperation within the community. The success of development projects relies on pre-existing levels of social capital and the selection of beneficiary communities and individuals. Three primary biases can affect this selection process. Site selection bias can occur when the implementation agency chooses where to carry out their projects. Self-selection bias arises when beneficiaries choose to participate voluntarily, creating problems with program evaluation and targeting. The identification bias concerns the selection of beneficiaries and the delegation of service delivery to local groups or agents, which can lead to rent-seeking behavior and social incentives (Bandiera et al., 2022; Basurto et al., 2020; Gugerty et al., 2019; Roy, 2022). These selection biases can favor communities with pre-existing social capital, which may have better compliance with agency criteria (literacy level, willingness to engage in a longterm program, etc.), be quicker to respond to solicitations (as their stronger social links reduce the transaction costs and their risk aversion), and represent an easier and more direct target for delivery agents, which often are members of those communities. This can result in targeting communities with higher structural and cognitive social capital, potentially exacerbating inequalities. This paper aims to investigate selection bias with regards to social capital in the context of a development project implemented by a local NGO in the Murehwa district of Zimbabwe. The project focuses on supporting communities in setting up Village Saving and Lending Groups (VSLA), which are community-based microfinance groups that provide poor, rural households with a safe place to save, earn interest, and access manageable loans. VSLA are widely used across 77 countries, with over 20 million participants.(1) Social capital is particularly important in the context of saving and loaning groups as these groups are selfregulated, in the sense that there is no formal institution ensuring that the rules are followed, that the money is safe guarded and that loans are repaid. Members of VSLAs need to trust each other in order for the mechanism to work. Concerning our case study, this aspect is emphasized by the NGO. They insist that when creating a VSLA, targeted beneficiaries should to some extent know each other, be similar in their socio-economic characteristics and trust each other. Additionally, social pressure induces higher saving and repayment rates (Ban et al., 2020). This study is built around two main hypotheses. First, targeted beneficiaries (treatment group) have a higher level of social capital than non-beneficiaries (H1). According to this hypothesis, targeted beneficiaries are more likely than non-beneficiaries to have had social interactions among them before the implementation of the program. Intuitively, it makes sense as the NGO and delivery agents encourage targeted beneficiaries to create VSLA with people they already know and trust. The second hypothesis (H2) relates solely to the treatment group: targeted beneficiaries that have previously participated in projects or initiatives similar to the project at stake exhibit higher social capital than targeted beneficiaries that have not. More specifically, looking at the beneficiary pool, we aim at looking if individuals that participated to saving and/or lending groups possess more social capital than the ones that did not. Our methodology proposes a first step into a two-stages analysis (ex-ante and ex-post development program) of social capital in communities that benefit of development programs. Social capital is multidimensional, hence we implemented different methodologies to measure it (Avdeenko & Gilligan, 2015; Ban et al., 2020; Fitzsimons et al., 2019). First, to assess cognitive social capital, we implemented incentivized experimental games: a dictator game, a trust game and a public good game. Second, a questionnaire to gather information on (1) stated preferences of risk-tolerance, patience and trust was administrated. Third, to measure structural social capital, we implemented a questionnaire adapted from Grootaert et al. (2004) as well as a grid adapted from Avdeenko & Gilligan (2015), the latter aiming at eliciting the stated level of relationships among the participants of a session. Finally, subjects completed a questionnaire allowing us to get the main socio-demographic variables. These different measures allowed us to estimate the level of social capital in both targeted beneficiaries of the VSLA project (treated group) and non-targeted beneficiaries (control group). We use Ordinary Least Squares (OLS) for each of our estimations. The main results are the following. From the experimental games results, we find that targeted beneficiaries are more altruist, trusting and trustworthy than non-beneficiaries. Indeed, the amount sent by the dictator, as well as both the amounts sent and sent back in the trust game (indicating trust and reciprocity) are significantly higher in the treated group. When looking at the measures of structural capital, we find that the density of relations in the treated group is higher than the one in the control group. This means that among the participants in the experimental session, treated subjects (targeted beneficiaries) have more relationships (i.e. friends, family, belong to the same producer groups etc.) than subjects of the control group. This is coherent with the fact that the NGO and the delivery agents promote that targeted beneficiaries should create VSLA associations with people they know and already trust. Another interesting result is that treated individuals tend to be members of more groups (e.g. farmers/producer groups) and gather socially with people more often than control individuals. These three results confirm our first hypothesis: targeted beneficiaries tend to have more social capital than non-beneficiaries. However, our second hypothesis which states that targeted beneficiaries that were previously in saving/lending groups should have more social capital is not confirmed in our context. Indeed, when looking at both cognitive and structural measures of social capital we find no differences between subjects that were previously members of saving/lending groups and subjects that were not.