In a novel experiment with residents of 48 villages in Tanzania, I test how cooperation in a public goods dilemma is affected by the possibility that free riding can have consequences for individuals' future social interactions and reputations outside the experiment. This is accomplished by manipulating the observability of contribution decisions in a public goods game among residents of the same village and in a placebo experiment with strangers. I find that observability reduces free riding among strangers, who are unlikely to ever interact in person, but appears to increase free riding among covillagers who are engaged in repeated social interaction. Unlike in laboratory environments, where the possibility of reciprocity and altruistic punishment causes effective decentral enforcement cooperation in public goods games, this experiment suggests real-world social incentives can exacerbate free riding among covillagers. This could be a consequence of egalitarian norms and redistributive practices in the villages, which propagate selective tolerance for free riding and cause individuals to be concerned not to appear exploitable, not in need, or suspiciously generous. This hypothesis is supported by data on study participants' self-reported reasoning and by ethnographic literatures on sharing norms in village communities. Moreover, the effect of social incentives on free riding varies across villages. A more negative effect of village-level social incentives on contributions in the experiment is associated with lower actual revenue mobilization in the village and greater prevalence of centralized enforcement structures.