This chapter investigates the empirical data on alliances and conflict to determine whether a moral hazard effect exists with alliance agreements. One key result derived from Chapter 3 is that leaders will bargain more aggressively with adversaries when they anticipate that a third-party defender will assist them if bargaining breaks down and war results. This finding informs us how world leaders will likely respond when they receive promises of military assistance such as those formalized in military alliances. Compared with states that do not have any external assurances, countries that have military alliances in crises might be more likely to reject adversaries’ challenges and make aggressive demands from adversaries. These results imply good news and bad news for international conflict: alliances may both enhance deterrence and lead to conflict. That is, moral hazard generated by an alliance commitment might improve deterrence, because adversaries are less likely to challenge countries when they expect those countries to be unwilling to concede to their demands. On the other hand, an alliance may also embolden protégés to challenge their adversaries in crisis bargaining. According to the theory, the key to understanding how military alliances affect conflict lies in the design of the commitment. Commitments affect how the protégé bargains with the adversary and then the adversary modifies its bargaining behavior accordingly.

This chapter examines the broader intuition, established in Chapter 3, that expectations of third-party intervention, such as we might expect in wars involving an alliance member, can give rise to a moral hazard effect observable in the alliance member's decision to initiate a conflict against a non-alliance member who is targeted in the terms of the alliance agreement. The chapter begins with a discussion about the empirical implications derived from Chapter 3. The following section in this chapter explains the research design and introduces a dataset of alliance agreements, which is generated from the categories in the typology from Chapter 2. I then estimate statistical models to determine whether different types of deterrent alliances affect the likelihood that a revisionist leader of the state holding the alliance will initiate a conflict against a targeted adversary. The chapter closes with a section that summarizes the evidence and major findings.