We study a bilateral negotiation setup where, at a bargaining impasse, the disadvantaged party chooses whether to escalate the conflict or not. Escalation is costly for both parties, and it results in a random draw of the winner of the escalated conflict. We derive the behavioral predictions of a simple social utility function, which is convex in disadvantageous inequality, thus connecting the inequity aversion and the prospect theory models. Our causal laboratory evidence is, to a large extent, consistent with the predicted effects. Among other things, the model predicts that the escalation rate is higher when escalation outcomes are riskier, and that the disagreement rate is lower when the cost of escalating the conflict is higher.