Previous analyses of the formation and composition of community-based organizations (CBOs) have used cross section data. So, causal inference has been compromised. We obviate this problem by using data from a quasi-experiment in which villages were formed by government officials selecting and clustering households. Our findings are as follows: CBO co-memberships are more likely between geographically proximate households and less likely between early and late settlers, members of female-headed households are not excluded, in poorer villages CBO co-membership networks are denser and, while wealthier households may have been instrumental in setting up CBOs, poorer households engage shortly afterward.