In international conflict, foreign policy similarity helps us control for the willingness to fight. States with more similar foreign policies, we assume, have a shared vision for world politics and so should cooperate more and fight less in pursuing that shared vision. Measuring alignment, though, has troubled researchers for decades.
In the international conflict literature, two approaches to measuring foreign policy similarity exist.
Alliances and foreign policy similarity
The first, developed by Bruce Bueno de Mesquita, relies on alliance portfolio similarity. Using alliance portfolio similarity to measure foreign policy similarity essentially assumes that states that share a similar vision of world politics align with each other, and the more allies we have in common the closer is our shared vision. Bueno de Mesquita operationalized the concept by calculating the correlations of how many allies pairs of states shared with each other.
Two problems with Bueno de Mesquita’s approach emerged. First, alliances, a realist prescription for maintaining a balance of power in the international system, reflect short-term security interests, not long-term shared visions of global order. Second, his measure of correlation was not a good one to use.
Fixing foreign policy similarity
To correct the first issue, Erik Gartzke replaced alliance portfolios with voting records in the United Nations General Assembly. UNGA votes don’t really bind states to anything, but they do produce public records of shared opinions on global issues.
To correct the second issue, scholars—using both alliance and UNGA voting data—have produced alternative similarity measures. These take various considerations into account such as weighting relative power of states and the expected versus actual level of similarity among states. The three measures we focus on here include S by Signorino and Ritter (1999), Scott’s π, and Cohen’s κ.
Comparing foreign policy similarity measures
In a paper with Isaac Castellano on the effects of international military interventions (IMIs) on foreign policy similarity, we generate ten measures of alliance portfolio similarity and four of UNGA voting similarity. We generated the measures based on impressive work by Frank Häge (2011) at the University of Limerick.
As you can see, each measure produces a very different outcome. In general, S produces similarity scores biased toward 1, so much so that several of the minima range above zero. Weighting S by states’ relative power produces seemingly better results, but this complicates both the calculation and interpretation of these measures.
Based on these comparisons, we join Häge in preferring either κ or π based on these summary statistics and encourage researchers to incorporate these alternative measures.