"We analyze the dynamics of neighbourhood watch programs in a local interaction framework. Agents can watch their neighbours' houses and thus deter burglars from breaking in.nAt the same time, agents also try to recruit their neighbours to join the neighbourhood watchnprogram. The probability of an agent joining the neighbourhood watch program depends on the success of the program, i.e., whether burglaries continue to occur. We show that the punishment of burglars plays a dual role in this context. On the one hand, punishmentndeters burglaries if the level of punishment is sufficiently high. On the other hand, it alsonaffects the probability of an agent joining the neighbourhood watch program. In particular,nwe show that if recruitment is harder when burglaries do not occur, a legal policy attemptingnto improve deterrence using more severe punishment is suboptimal. In a second part, we extend our model to the study of norm enforcement in public goods dilemmas and show that our results remain valid if agents can punish each other (instead of burglars) for not contributing to the public good. Our paper thus provides a first analysis of the evolution of ""altruistic punishment"" in large populations with local interaction."