The empirical studies that Erica Chenoweth has highlighted in the past few days give ample reason for pause about the likelihood that a U.S. led military intervention will reduce civilian killings or end the war in Syria. That said, I am somewhat skeptical that we can generalize from these studies to the case of Syria.
All of these studies investigate whether the likelihood of some outcome (high civilian killings, ending a civil war, etcetera) becomes more likely when an intervention has taken place than when it has not. The answer for the most part is that on average few good things happen post military intervention. There is the obvious problem that one can only uncover correlations this way. For instance, it may be that interventions only take place in the worst conflicts.
There is an even more basic problem. As Jon Western points out, there is tremendous variation in the scope, size, and purpose of interventions. Conflict situations similarly differ enormously. The correlations are thus average associations between a very heterogeneous “treatment” (interventions) among very heterogeneous units of analysis (cases of civil war/conflict). Think of it as trying to estimate what effect medicine has on your health when we group together patients with different diseases and thus different medicines. We may find a positive effect of “medicine” but we don’t know if this was because all medicines on average improve health or if there are some that work exceptionally well while others leave patients worse off.