Measuring alliance reliability – current practices and future research

Defensive alliances commit states to aid each other in the event of conflict. Their goal is often twofold: to increase the ability to deter armed conflicts with an adversary and to increase the probability of prevailing in conflict should deterrence fail. Thus, the prospects of allied support in the event of war constitute a cornerstone in defense planning for states across the globe. Yet, not all alliances are honored. In fact, recent empirical studies of alliance reliability, testing how often alliance commitments are honored, conclude with a low fulfillment rate. After the Second World War, the U.S. entered into a range of alliances. A recent study has found that in this same period, the U.S. did not fulfill any of its alliance commitments. This is remarkable not only because so many states depend on the U.S. as an ally, but because it seems incompatible with the fact that most U.S. alliances have endured throughout the period. This report explores this puzzle by revisiting the assumptions of the prevailing research design in the literature on alliance reliability. It explores four such assumptions and applies them to the Second Taiwan Strait Crisis as an illustrative example of the restrictions within the prevailing approach in the alliance reliability literature and the implications for future research. It finds that while existing research has produced invaluable contributions to our understanding of alliance dynamics, the research design is restricted in scope, including only certain alliances, and producing results that may not be precise. In the case of the U.S., which is the focus of this report, the method and operationalization used to assess alliance reliability are particularly unsuited. Specifically, only alliances confronted by interstate war are tested, despite interstate wars occurring relatively infrequently after 1945. This leads to a small and skewed sample of alliances being ‘put to the test’, as alliances that successfully deter conflicts from occurring in the first place are not included in the sample. Moreover, the language and context of each individual treaty are seldom taken sufficiently into account in big data research. Lastly, the operationalization of alliance fulfillment requires a state to enter into war alongside its ally. Such a dichotomous definition of allied support disregards military contributions beneath the conflict threshold. This report suggests that alliances may be more reliable than previous research would allow. Much in the same way as related concepts like deterrence, alliance reliability is a complex phenomenon, difficult to measure. In studying alliance reliability, we should take into account each treaty’s language and provisions, the allied relation in question, and the context of the conflict. Thus, there is a need for a broader concept of alliance reliability. In sum, we should be careful drawing generalizations about alliance reliability from existing big data research, and rather apply this research as a fruitful foundation for future case-by-case studies.