Score contribution per author:
α: calibrated so average coauthorship-adjusted count equals average raw count
Do all aspects of social capital improve repayment behavior in group lending programs? The group lending literature typically uses one or few measures of social capital in a linear form, and systematically understates the uncertainty of results and model specifications. As a result, many papers conclude that specific measures of social capital do not matter. This paper introduces Bayesian Model Averaging to the group lending literature and simultaneously tests the validity of six different measures of social capital. While the initial analysis suggests that only few measures matter, this result becomes invalid when using model selection criteria and Bayesian Model Averaging while allowing for nonlinearity and interactions between different aspects of social capital. We find geographical proximity, trust, friendship, group homogeneity, and acquaintanceship to be important factors in explaining group repayment with mixed evidence for relatives. Our results suggest that microfinance institutions should target borrowing groups with more social capital.