Estimated effect of partnership treaties

Having specified the content of partnership treaties, we will now move on to discuss how to identify China’s bilateral relations in terms of partnerships. Rather than classifying China’s partnerships using the wording in titles, we argue that content is a better approach. To verify the effect of commitments made in a partnership treaty, we adapt the topic modeling results from the previous sections—the number of paragraphs mentioned under each topic by each country—as variables for К-means clustering to group countries with identical features. After that, we proceed with the one-way analysis of variance (ANOVA) to test if the mean is significantly different for all groups.

Just like topic modeling, the К-means algorithm must identify the number of clusters to classify in advance. We use the silhouette evaluation to measure the quality of a clustering. The higher the average silhouette width, the better the cluster results, meaning that most of the countries are assigned to the proper cluster. Figure 1.17 computes the average silhouette method for clusters of two to 20, showing that three clusters maximize average silhouette values.

In this section. Topic 1 (Industrial and Infrastructural Investment), Topic 6 (Trade and Economy), and Topic 9 (Security) were selected to assess the pragmatic cooperative relationship with China. This analysis concludes with a classification of the 78 countries into three groups that capture the degree of cooperative ties with China. Cluster 1, which has the most substantial ties with China among all countries, consists of solely one country—namely, Russia. As we can see in Figure 1.18, China has made far more commitments with Russia than with other country; consequently, Russia is classified into

Figure 1.17 Average silhouette width for different numbers of К

Figure 1.18 Numbers of commitments made in each topic for three country clusters

a separate category on its own. Cluster 2, comprising 17 countries, includes countries that have relatively close ties to China apart from Russia. Cluster 2 includes the United Arab Emirates, Belarus, Brazil, Egypt, France, Greece, Indonesia, Kazakhstan, Kyrgyzstan, South Korea, Mongolia, Pakistan, the Philippines, Tajikistan, Turkmenistan, Uzbekistan, and South Africa. Cluster 3, the group of countries least connected to China, covers the rest of 60 partner countries in our sample.

Next, we use the ICEWS data, an event dataset available from 1995 to 2019, to measure China’s bilateral cooperation as the dependent variable and perform the ANOVA test. The ICEWS data covers a time period that largely overlaps wuth China’s practice of partnerships and provides detailed categories of cooperation. That is, in addition to other measures of bilateral relations (e.g., trade flows, international organization memberships, and historical relations), it includes cooperation related to political, economic, and cultural issues. These issues include cooperating economically and providing financial assistance. By including specific cooperative events, this variable presents a comprehensive picture of cooperative relations with China under the practice of partnerships. We aggregate the assigned intensity scores of the following two cooperation events, including “Engage in Material Cooperation” and “Provide Aid” at the country level with lagged and raw values. Supporting economic activities and participating in judicial assistance are examples in the former category; providing economic aid and providing humanitarian aid are representative of the latter case. In addition to the other eight cooperative

Table 1.3 ANOVA test with logged material cooperation as the dependent variable

Table 1.4 ANOVA test with material cooperation in raw values as the dependent variable

events, the two categories are consistent with practices outlined by China’s partnerships.

The results are displayed in Tables 1.3 and 1.4 with different measures of cooperation in lagged and raw values, respectively. The significance values for clusters based on content are both smaller than 0.10, indicating that the concept of content classification to understand China’s partnership relations is more meaningful and has statistical differences among distinct groups. Countries whose treaties outline greater commitments and consensus, such as Clusters 1 and 2, have a more intimate level of cooperation with China. On the contrary, the significance values for groups as identified by titles alone are both larger than 0.10. This shows that the wording in titles of partnerships does not consistently describe the level of China’s bilateral relations. More specifically, a partnership with “comprehensive” or “strategic” in the title does not refer to a stronger connection with China than an SP (for a CSP) or a P (for a CSP or an SP). Indeed, the commitments and consensus stated in the content of partnership treaties can explain the disparity of cooperative status and provide a preliminary and operational way to decipher China’s partnership network.