In the era of strategic competition, the U.S. government is intensely focused on building stronger alliances and partnerships around the world. And in the new era of data science, governments and industry are gaining competitive advantages by employing big data and machine learning techniques to identify, measure, and predict patterns.
But until now, no research organization has harnessed the tools of data science for the purpose of building stronger U.S. alliances and partnerships. CNA experts in data science and in national security strategy have come together to develop a more rigorous and quantified approach to U.S. alliances and partnerships. The result is a statistical index that identifies countries with the strongest and weakest alignment with the United States. Such an index might highlight cracks in relationships that are being taken for granted. Machine learning can also signal opportunities for stronger alliances and partnerships and predict conditions that could strengthen or weaken alliances.
What data science brings to alliance management
Past efforts to quantify alliance strength have chosen individual measures, such as numbers of treaties. Some have merged multiple measures, subjectively deciding how much to weight each factor. CNA took a data science approach, developing a machine learning algorithm to analyze 150 nations on a dataset of nine variables signifying engagement or hostility:
- International agreements with U.S.
- Defense agreement with U.S.
- U.S. diplomatic representation abroad
- U.S. arms exports
- Foreign assistance
- Militarized interstate disputes, same side as U.S.
- Militarized interstate disputes, different side
- U.S. sanctions
- Cyberattacks against the U.S.
With no other information about which countries are considered friend or foe, and no guidance on weighting, the algorithm found the correlations among these variables. It assigned them weights according to how close those correlations are. And it came up with an index value for each country’s alignment with the United States.
Note that while this kind of “unsupervised” machine learning algorithm independently decides how to weight and correlate each variable, the human decision of which variables to include is a subjective one that can influence the results.Download full report
- Document Number: DMM-2022-U-032921-Final
- Publication Date: 6/30/2022