We study long-run selection and treatment effects of a health insurance subsidy in Ghana, where mandates are not enforceable. We randomly provide different levels of subsidy (1/3, 2/3, and full), with follow-up surveys seven months and three years after the initial intervention. We find that a one-time subsidy promotes and sustains insurance enrollment for all treatment groups, but long-run health care service utilization increases only for the partial subsidy groups. We find evidence that selection explains this pattern: those who were enrolled due to the subsidy, especially the partial subsidy, are more ill and have greater health care utilization.