Competing theories increasingly support the positive role of social capital in small loan default costs of group lending; at the same time, potential group collusion may increase loan delinquencies. Findings from the available literature are mixed on the role of the various attributes of group lending. But past studies suffer from estimation bias due to the unobserved sorting behavior of group members and their other attributes. This paper attempts to resolve that estimation bias by utilizing longitudinal data from 297 Grameen Bank groups since their inceptions. A dynamic lagged dependent model with correction for time-varying heterogeneity of group and individual behavior is applied to estimate the effect of group liability in the Grameen Bank. The results suggest that group liability matters in both loan disbursement and repayment, with women less of a credit risk than men and women's groups more homogeneous than men's. Finally, the benefits of social capital outweigh the costs of group collusion, especially for women's groups, thereby reducing overall default rates. The risk-pooling behavior of diverse men's groups increases men's repayment behavior. Overall, group lending as practiced by Grameen Bank appears to increase repayment rates.