Multiparty Disputes and the Probability of War, 1816–1992

Abstract

Previous theory and research have suggested that multiparty disputes might be significantly more likely to escalate to war than bilateral disputes, because of the difficulty of reaching a mutually acceptable agreement as the number of parties increases. This study presents a systematic test of this hypothesis. Efforts to provide such a test have been hampered by the absence of data that distinguish the number of participants in a militarized interstate dispute prior to the outbreak of war from the number of participants after the war breaks out. We find that multiparty disputes do have an increased probability to escalate to war. In addition, we find that the issue over which the disputants contend has an important effect on the probability that the dispute will escalate to war; multiparty disputes that are over territory have a higher probability of escalating to war than multiparty disputes in general. Lastly, it is found that the effects related to the number of parties in a dispute and to whether the dispute is over territory are independent, and one does not eliminate the effect of the other. In order to contribute to future scholarship on this topic, the data for the new classification scheme of multiparty disputes are published in the appendices.

Keywords:

• multiparty disputes
• territory
• MID
• war

The authors wish to thank anonymous referees for valuable comments. Research reported in this article has been supported in part by NSF Grant #SES-9818557 (John Vasquez and Paul Senese, principal investigators).

Notes

¹ Leeds did find empirical support for this claim. Using different data and a shorter temporal domain than employed herein, Leeds found that 5.2% of bilateral disputes escalate to war, as opposed to 12.5% of multilateral disputes.

² The business dispute literature also addresses the particularly difficult nature of multiparty dispute resolution and the importance of underlying issues; see Menkel-Meadow (2001).

³ “Satisficing” behavior is an alternative to maximizing (or optimizing) behavior. Under conditions of bounded rationality, decision makers will attempt to achieve at least some minimally acceptable goal as opposed to the rational actor striving for utility maximization (Simon, 1957). In situations with multiple participants negotiating an agreement, satisficing may be the best possible outcome given that it is unlikely that all participants could be satisfied by an agreement. Still, from our theoretical perspective, satisficing is less probable with multiple parties than it is with two.

⁴ The following disputes are excluded from all of the analyses below: MID number 1129 is a war according to COW war data, but not according to MID data, so it is excluded; MID numbers 320, 324, 397, 518, 3813, and 3826 all have only one participant that reached level 5 hostility and for that reason do not appear in the war data, so they are excluded. Additionally, two disputes have more than one revisionist actor with more than one revision type coded (MID numbers 21 and 3564). Any disputes where all revisionist actors are coded as nonapplicable are excluded from all of the analyses. Of the non-applicable disputes, five go to war (three are bilateral disputes).

⁵ In all cases (both with MIDs and wars) the start year and start month were available. In a few cases (with MIDs), the start day was unknown. For those cases, we simply substituted “1.” We could have used any other day for these cases without impacting our classification. See Appendix A for a list of all multiparty disputes that escalate to a multiparty war and Appendix B for a list of bilateral MIDs that escalate to a multiparty war.

⁶ Alternatively, we used two other coding schemes. The first required more than two participants in the MID at least three days prior to the outbreak of war, reducing the number of multiparty MIDs from 260 to 256. The second required that more than two participants be in the MID at least five days prior to the outbreak of war, reducing the number of multiparty MIDs from 260 to 253. Neither of these more restrictive coding schemes changed the outcomes of the analyses presented herein appreciably, so results using the least restrictive rule are presented.

⁷ A war can still have multiple participants and not arise from a multiparty MID if states join the war after the first day. There are 10 cases of such war-joining behavior. Given our coding scheme, these cases are considered bilateral MIDs that escalate to war (listed in Appendix B).

⁸ All exclusions listed in footnote 2 apply to this description of the data. Two hundred sixty of the 1,663 disputes are classified as multiparty under our guidelines.

⁹ The Z-score compares the base and conditional probability of going to war. The Z-statistic is calculated as follows: Z = (Pc-Pb)/√Pb(1–Pb)/Nc, where Pc = conditional probability; Pb = base or unconditional probability; Nc = total number of cases having the condition (multiparty or bilateral), and √ indicates square root of the entire denominator. Our use follows Bremer (1992, 327). Significance levels are based on a two-tailed test.

^a All exclusions from text footnote 2 apply to this and the following analyses.

¹⁰ We found that if entry dates are not taken into consideration, 42 of 270 (about 16%) multiparty disputes go to war and 49 of 1,393 (about 3.5%) bilateral disputes go to war for the time period 1816–1992.

¹¹ Policy disputes have a lower conditional probability (.082) than the base probability (.123), and that difference barely misses statistical significance at the .05 level. While multiparty policy disputes are less war-prone than multiparty disputes in general, it should not be assumed that multiparty policy disputes are not more war-prone than bilateral policy disputes (see below). If the latter were to be the case, then size of the dispute would not matter.

¹² There are four wars in the MID data with a revision type coding of “other”: the Franco-Prussian War, the Football War, the Spanish-Peruvian/Chilean War of 1866, and the Lopez War. The Franco-Prussian War is the one case of a multiparty war with a revision type (issue) coding of “other.” In the case of the Franco-Prussian war, Prussia is coded as seeking a revision of type “other,” involving a dispute over succession rights. The revision type is coded based only on explicit claims made publicly by official representatives of the state before the outbreak of war. However, Vasquez and Henehan (2001, 129, fn 9) discussed this question in more detail, with particular emphasis on the territorial elements of the four “other” wars, and claimed that, “None of these cases indicates that there is some ‘fourth’ factor that makes more war probable than territory; in fact three of them suggest that territory played a role in the war, even if it was not explicit in the claims made before the war. Therefore, they should not be treated as a distinct type of issue with their own level of probability.”

^a We also ran this model with interaction terms for revision type and multiparty (not shown). None of the interaction terms were statistically significant and the individual coefficients for the multiparty and revision variables remained significant in all cases. The model with the interaction terms provided a slight (.01) increase in the Pseudo R².

^b N = 1663; Wald Chi² = 47.08; Prob. > Chi² = 0.000; Pseudo R² = 0.066; 94.53% of cases correctly classified. See Long and Freese (2001) or http://www.indiana.edu/~jsl650/stata/ for more information.

¹³ Policy disputes are used as the reference category because they are the modal dispute.

¹⁴ We also ran the same model in Table 3 to compare territorial disputes and regime disputes (as the reference category), and we found that territorial disputes are more war-prone, but the relationship, as with regime and policy disputes, is not statistically significant.

¹⁵ CLARIFY is an ado file written for use with STATA in order to make data interpretation and presentation more accessible and reader-friendly. CLARIFY employs simulation to approximate features of the probability distribution (King, Tomz, Wittenberg, 2000, 349). For more information, see Tomz, Wittenberg, and King (1999), http://gking.harvard.edu/.

¹⁶ In order to subject Proposition 3 to further scrutiny, we collapsed policy, regime, and other into a “nonterritorial” category, thereby creating a dichotomous measure of issue type (territory/nonterritory), and tested the proposition using the multiparty variable and this dichotomous measure. The results (not shown) were supportive of Proposition 3. Multiparty territorial disputes have a predicted probability of escalation equal to 0.21; multiparty nonterritorial disputes, 0.09; bilateral territorial disputes, 0.07; bilateral non-territorial disputes, 0.03. This model of escalation to war also has a significant Chi²; however, the model presented in Table 3 is better suited to demonstrating the varying effects of the different types of issues as opposed to just classifying the issue at stake as territory or non-territory.

¹⁷ Regime disputes are not reported because the coefficient is not significant and the number of cases is small.

¹⁸All exclusions from text footnote 2 apply in the appendices.