Why do armed groups sometimes coerce and sometimes not? Civilian suffering due to coercion in conflicts is larger; yet, anecdotal evidence suggests that armed groups often choose not to coerce. To explain the observed variation in coercive practices, I combine a two-sector specific-factos trade model with a model of violence. Armed tgroups operating in the resourc esector and allocate military reosurces between conflict and coercion, which captures more land and labour respectively. The model shows that coercion depends, not only on economic factors, but also the military landscape and the interactin between the two. First, coercion is higher if labour scare or extraction labour-intensive. Second, coercion is high if one group is dominant, relative to the others. Third, the impact of the prcie of the commodity depends on the distribution of military strength: coercion increases with price if one group is dominant, but this effect is reversed if military power is highly decentralised. The first result is consistent with historical accounts of the re-emergence of serfdom in 16th century Russia, and the prevalence of slavery in West Africa. The second result explains why coercion decreased in the Kivu privinces after 2002: the Rwandan Army, by far the most powerful group, evacuated. The third result explains why the rubber boom in late 19th lead to a highly coercive regime in the Congo Free State, but less so in Amazonia. The Congo Free State had a monopoly, but conflict between Spanish and Portuguese colonies escalated during the boom, reducing their coercive power. It further explains why, during the protracted Civil War in Sierra Leone, coercion was common in the rice plantations, but not the diamond mines. The number of battles were higher in the diamond-rich areas, but level of civilian victimisation less. With land the valueable factor of productions, violence was allocated to conflict, not coercion.
