Bringing Cooperation Back In: A Dynamic Model of Interstate Interaction

Abstract

In an earlier article (Crescenzi & Enterline, 2001), we developed a formal, dynamic model of the cooperative and conflictual dimensions central to interstate relationships. However, the empirical data employed as inputs into the original model informed only the model's conflictual dimension. Here, we operationalize the conflictual and cooperative dimensions of the model, with the latter derived by inputting information on joint participation in intergovernmental organizations (IGOs) for the period 1965–2000. Doing so enables us to trace the joint cooperative–conflictual temporal trajectories of interstate dyads, in addition to capturing the degree and dynamism of these relationships. We demonstrate the flexibility and practicality of the model-derived empirical indicators of interstate interaction with an analysis of dyadic interstate conflict. Our dynamic approach to studying interstate relationships promises to facilitate fruitful contributions to several research agendas in comparative politics and international relations.

The authors are grateful to the COW2 Project in the Department of Political Science, The Pennsylvania State University. Data assembly was conducted with the assistance of EUGene 3.0.3 (Bennett & Stam, 2000). The statistical analysis executed herein is accomplished in STATA 9.0. The data and replication files are available from the authors upon request. This research was conducted with the support of NSF grant #SES-0450111. All remaining errors are the responsibility of the authors.

Notes

¹ Hensel (1996) suggests that rivalries can be nonmilitarized and can emerge when pairs of states compete in, for example, trade. Also, see arguments by Thompson (1995) on the centrality of strategic perceptions in the emergence of rivalries

² Important diplomatic breakpoints and cooperation appear in the relationship between the United States and the Chinese (Goldstein & Freeman, 1990; Goldstein, 1991), as well as the often militarily intense rivalry between North and South Korea following the Korean War

³ This erosion is clearly illustrated in Goldstein and Freeman (1990: 42, 65)

⁴ We owe a special debt of gratitude to Chad Atkinson, Dina Zinnes, and Robert Muncaster of the Merriam Lab, University of Illinois, Champagne–Urbana, for their contributions and encouragement during the development of this model

⁵ This parameter weights the impact of the shock on i _t . It might be the case that researchers will have a theoretical motivation for adjusting β

⁶ Indeed, one could build multiple shocks for each phenomenon. These shocks are then weighted to equilibrate their maximum and minimum values. While this does not completely assuage the problems related to the ordinal character of such information, it constrains the behavior of both shocks to the same range

⁷ This assumption places our approach in contrast to the punctuated equilibrium models discussed in Cioffi-Revilla (1998) and Diehl (1998)

⁸ Similar approaches to modeling the decay properties of conflict history are employed by Hegre et al. (2001) in their study of regime changes and civil war, by Partell (1997) in his study of dispute escalation, and by Hegre (1997) in their study of the hazard of interstate war

⁹ The denominator is adjusted by adding a constant, λ, so that it never assumes a value of zero. We hold the value of α to be positive, in order to ensure a decay toward neutrality

¹⁰ γ > 0

¹¹ The decision to use the bounding function resides with the researcher, based upon how the researcher conceptualizes the accruement of interstate interaction

¹² Goldstein (1992) offers the best effort in this regard, scaling the cooperative and events in the WEIS data set onto one dimension. But the events data to which this scale applies do not match the scope of the MID data. We do think, however, that Goldstein's scale can be useful in operationalizing the IIS measure with event data generally

¹³ Recall that our formal model of the Interaction Level requires at minimum two time points from which to calculate a single score (e.g., a data input for the year 1965 is used to create an interaction score for the year 1966). Therefore, in Figure 1a, Figure 1b, Figure 1c, Figure 1d the plotted values for the conflictual and cooperative interaction scores in the year 1965 are each zero, or neutral

∗∗ p = 0.001

∗ p = 0.05

¹⁴ These figures also illustrate the left-censoring limitations of the model. In this analysis, we assume that dyads start at neutrality until their behavior shocks them out of it. This is less of an issue when the data underpinning the model begin at a natural starting point like 1815, but here the 1965 left-bound for the annual IGO data leads to an obvious issue. In the analysis below, this limitation should bias against support for the hypotheses. We also ran the analysis with data truncated to a 1970 starting point, giving the IIS scores time to correct for the initial lack of information. The results were consistent with those reported in Table 1

¹⁵ See the discussion of relevant dyads in Oneal and Russett (1997) and Maoz and Russett (1993)

¹⁶ The bounding function used for both components of the IIS in the presented models employs a value of 1 in the denominator

¹⁷ We find that Model 1 fails both the linktest and Schoenfeld residuals global tests for violation of the proportional hazards assumption, but further investigation reveals that all of the independent variables pass their individual tests when they are interacted with the natural log of time (none of the interactions terms are significant). While we cannot explain the model's failure on the global tests, the next step of investigation commonly pursued shows no clear problem, so we do not believe there is a serious problem indicated by the tests taken as a group given the lack of agreement

¹⁸ The results for the controls are also similar to those in Model 1

¹⁹ We performed a number of robustness checks on both models. While variables indicating the presence of an alliance in the dyad and the presence of joint democracy in the dyad are missing from the presented results, we did run models that included these variables. We dropped the variables because we believe that the alliance variable shares substantial common ground with the data that we use to operationalize cooperation, joint entrance into IGOs, since alliances are included as IGOs. We also drop joint democracy here because we believe that democracies are more likely to engage in cooperative behavior (Leeds & Davis, 1999) and thus joint democracy shares common ground with our measure of dyadic cooperation. When we include joint democracy and alliance variables, we find that all three variables (alliance, joint democracy and Cooperative IIS) become insignificant, which supports our argument that they are related. When we include joint democracy but not alliance, we find that the significance of Cooperative IIS is weakened and joint democracy is insignificant; this also supports our understanding of the connection between these variables. Both alternative models also result in the variable Major Powers becoming statistically insignificant. Furthermore, Model 2 fails both the linktest and Schoenfeld residuals global tests for violation of the proportional hazards assumption, and in this case, not all of the independent variables pass their individual tests when they are interacted with the natural log of time. In Model 2, Cooperative IIS, Conflictual IIS, and Contiguity each fail their individual tests, while Capabilities and Major Powers do not

²⁰ We lag this variable one year in the analysis, which is consistent with the other operationalizations

²¹ While we do not recommend including both the Cooperative IIS score and IGO Membership Change in the same model (it almost certainly violates the notion of independent variables), we did run these models as a final robustness check. In the PRD sample, the Cooperative IIS coefficient retains its sign and significance while the control IGO membership coefficient is not statistically significant. In the larger sample, neither coefficient is statistically significant

²² The United States and the Soviet Union were each members of the United Nations during the Cold War (as we measure it above, an indication of dyadic cooperation), but their cooperative interactions in terms of resolutions in the General Assembly and Security Council might reveal the low levels of overt cooperation transpiring between the two states during this period