Does military intervention affect civilian death tolls? Existing research has focused on international actors’ ability to limit ongoing slaughter, but has not examined their ability to prevent the emergence or escalation of such killing. I develop a theory of government killing that accounts not only for the government’s decision to kill civilians but also for the transference of the killing order from leader to perpetrator, and for the perpetrator’s implementation of that order. Focusing on the principal-agent relationship produces new expectations about the effects of military intervention on government killing. I find that international actors are well-equipped to limit civilian slaughter: Intervention supporting the government decreases the likelihood that a government orders civilians killed. Intervention against the government leads to a decrease in death tolls when killing occurs. Ultimately, supportive intervention is a useful means of preventing government killing, while oppositional intervention limits its escalation once it begins.
That’s the abstract from a really interesting paper by Jacqueline DeMerritt that’s forthcoming in the Journal of Conflict Resolution. It’s great to see someone pull together these seemingly disparate literatures—theories of political violence on the one hand and organization theory on the other—in search of better ways to think about addressing a profound and very real problem. Props to Erica Chenoweth and Daniel Solomon for calling it out on Twitter.
In hopes of seeing this project grow, whether in DeMerritt’s hands or someone else’s, here are two thoughts I had after reading the paper about directions in which this model could and should be stretched.
1. State agents are capable of initiating atrocities without orders from their leaders. For simplification’s sake, DeMerritt construes atrocities as something state leaders initiate and state agents may or or may not carry out. In fact, it’s not always clear who is leading whom in these situations, or if hierarchy is even a relevant concept. It would be useful to try to incorporate into the model the idea that committing atrocities can also be a form of “shirking” and see where that leads.
2. Time matters.The interactions represented in DeMerritt’s model occur repeatedly, and the outcomes of previous iterations shape incentives at later junctures. For example, once certain agents have committed some atrocities, we might expect the threat of future punishment to have a weaker effect on those agents’ later actions. In the miserably cold language of economics, the marginal expected cost of additional atrocities will often be lower than the expected cost of that initial leap. If the model is going to address the dynamics of violence against civilians and not just the question of whether or not they will occur, we’ll want to extend it to incorporate these kinds of path-dependent processes.