Vito D'Orazio
University of Texas at Dallas
School of Economic, Political, and Policy Sciences
Political Science Program
  • Home
  • Research
  • Teaching
  • Data
  • Software

Advancing Measurement of Foreign Policy Similarity

Extant theoretical literature in International Relations has suggested that the similarity of foreign policy preferences is a meaningful and important concept. The commonly used measures of this concept -- tau-b and, more recently, the S-score -- are problematic in two ways. First, as algorithms, neither operationalize the concept in a manner consistent with how students of IR theorize foreign policy similarity. Specifically, tau-b is not a measure of similarity---it is a measure of concordance or rank-order similarity. Although S is a measure of similarity, it does not account for the degree of third-party relationships---only the symmetry of such relationships. A new algorithm, D, is introduced which incorporates the degree of mutual relationships into the measurement of foreign policy similarity. Furthermore, the resulting D-score is standardized, as opposed to the S-score which has four commonly used versions in addition to several others that may be subjectively defined. Second, the input data for measures of foreign policy similarity are typically data on formal alliances. Generally speaking, alliances are a granular indicator of an underlying, military-strategic relationship between states, more generally referred to as military cooperation. As such, the policy portfolio used to calculate D can include additional indicators of military cooperation. The five derived indicators include alliances, joint military exercises (of which new data has been collected), arms transfers, and multinational combat and peacekeeping operations. The indicators are combined using a graded response model to estimate military cooperation as a latent trait. The latent trait is used as input data to calculate the D-score, which is directly compared with the S-score and shown to be consistent with intuitive notions of policy similarity while the S-score is not. In a replication study, the new measure is shown to not be substitutable with the S-score.

  • Chapter Five PDF