One-Sided Crises in World Politics: A Study of Oxymoron, Violence and Outcomes

Abstract

This paper focuses on crises and seeks to extend understanding of escalation processes, outcomes, and legacy. We go beyond Hewitt and Wilkenfeld's (1999) initial study of one-sided crises, which emphasized crisis type as an explanation for violence levels, in three ways: We (1) pursue an explanation for why some crises remain one-sided; (2) include two additional crisis attributes, protractedness of conflict and ethnicity, which are expected to impact upon the role of violence; and (3) link outcomes and subsequent tension levels for adversaries with crisis type (i.e., one-sided versus others) to expand the potential explanatory range of one-sidedness. To achieve these goals, the paper unfolds in four parts. First, the study is placed in the context of ongoing research on crises in world politics, most notably as carried out by the International Crisis Behavior (ICB) Project. The second part presents a theoretical overview of the factors that might distinguish crisis type, along with those deemed important in determining violence, outcomes, and subsequent tension. Explicit hypotheses are derived as well. The third part conveys data, variables, data analysis for crisis type (Stage 1) and violence, outcomes, and subsequent tension levels (Stage 2), and a comparison of results for the two stages. The fourth and final part summarizes the paper's accomplishments. Key findings are that (1) we can distinguish crisis type on the basis of characteristics such as contiguity, gravity of threat, and civil war involvement; and (2) the Hewitt and Wilkenfeld model is most successful in explaining violence as opposed to outcome and legacy, which seem especially difficult to account for, even with the addition of theoretically important factors like ethnicity and protracted conflict.

KEYWORDS:

• asymmetry
• crisis
• ICB
• one-sided
• outcomes
• violence

Acknowledgments

We are grateful to J. Joseph Hewitt for a valuable commentary and data.

Authors names appear in alphabetical order for this collaborative work.

Notes

¹ Version 4.1 of the ICB data includes 434 international crises from 1918 to 2001. Limitations on data pertaining to crisis type from Hewitt and Wilkenfeld (1999) restrict the analysis in this study to 1918–1994.

² The list of one-sided crises can be found in the Appendix. It includes the ICB Project's crisis number, crisis name, triggering entity and crisis actor. ‘Triggering entity’ is the party that unintentionally initiates a one-sided crisis and does not perceive itself in crisis. The ‘crisis actor’ considers itself as the target and can respond to the trigger in various ways.

³ Case descriptions are taken from the ICB project's webpage, available at http://www.cidcm.umd.edu/icb/ (03/25/2005).

⁴ An example is Rousseau et al.'s (1996, p. 532) study of the democratic peace, which excluded many one-sided crises from the data analysis “because there was no strong evidence that the challenger considered using force in the cases” (Hewitt and Wilkenfeld, 1999, p. 315, note 8).

⁵ One factor from Hensel and Diehl (1994, p. 494), designated as “learning”, did not prove useful in distinguishing non-military responses and therefore is excluded from the present research design.

⁷ We are aware that our research design is not fully isomorphic with that of Hensel and Diehl (1994). They attempt to explain why the target responds in a militarized way to a nonmilitary stimulus from the initiator, while our version of their model focuses on why the target's response does not induce a reciprocal perception of crisis for the initiator. In spite of such differences, we believe that the configuration of variables from their model will be able to explain both of the preceding asymmetries.

⁶ All of the variable descriptions are taken from the ICB system-level codebook (Brecher and Wilkenfeld, CD-ROM 2000). This CD-ROM itself is attached to the 2000 version of the book. Operationalizations from Hewitt and Wilkenfeld (1999) are applied directly to enhance consistency and comparability.

⁸ The ICB data/codebook and one-sided crisis data/codebook are available at http://www.icbnet.org/Data/index.html (10/11/2004). Version 4.1 of data from ICB covers the period between 1918 and 2001, but the coding of one-sided crises currently stops at 1994. Therefore, the previous version of the ICB dataset is used in the analysis.

⁹ Further details about one-sided crises appear at http://www.icbnet.org/Data/1sidecod.pdf (10/11/2004).

¹⁰ Given the multiple and related dependent variables, some thought might be given to setting up simultaneous equations. However, all of the dependent variables are categorical and estimation techniques are not sufficiently advanced to permit such modeling in the present study (personal communication with Michael Alvarez (2003)).

¹¹ Bivariate correlations between the independent variables reveal a maximum of 0.67 for ethnicity with the interaction term ethnicity x protractedness, with most correlations much lower. Thus the results reported in the stages of data analysis that follow are unlikely to be affected by multicollinearity. We also have used Achen's (2002) “rule of three,” running each of our independent variables in subsets of three, to further check on the reliability of the results. No discrepancies result from this additional series of runs.

¹² ICB's power discrepancy variable departs somewhat from the emphasis placed on exploring different levels of power preponderance by Hensel and Diehl (1994, p. 494), who implement capability ratios of three and ten times, respectively, as a check on the possibility that only a very high discrepancy will be sufficient to produce a statistical connection.

¹³ Within the ICB data, the “ethnic component” variable is used instead of the “ethnic driven” variable. The ethnic component variable answers the question “was ethnicity a salient aspect of any component or phase of ethnic conflict?” (Brecher and Wilkenfeld, 1997a, p. 175). By comparison, the ethnic driven variable is more narrowly focused on cases in which ethnicity is a central aspect (Brecher and Wilkenfeld, 1997a). The main reason for this choice is to recognize the primordial element within ethnicity. Since identities are fixed by birth, the ethnic aspect of a crisis will keep its salience even if not a central issue within the case at hand.

¹⁴ While the period of two months might appear to be arbitrary, some lag time must be incorporated in order to probe for potential diversionary effects. The results, however, are unaffected by marginal changes in measurement for either this variable or the one that follows.

¹⁵ All of the results reported from this data analysis have been checked for sensitivity to geographic region, following Brecher and James's (1988) ICB-based investigation of Middle East location versus protractedness vis-à-vis causal impact on violence. As with that study, geographic location does not affect the results.

¹⁶ A measure analogous to the R² statistic in regression does not exist in logit. Thus the pseudo-R² is reported for completeness and the more informative “correctly classified” indicator is given greater prominence (Agresti, 2002, p. 226). Correctly classified indicates the percentage of observations that the model classifies correctly among the total. In other words, it could be seen as a measure of the probability that this model will correctly predict the dependent variable on the basis of the explanatory model. In the models reported here, the correctly classified percentage ranges from 62% to 87%.The modal category also is reported. It shows the improvement represented by the model over a prediction based strictly on the relative frequency of cases. It should be noted that improvement becomes more difficult when dependent variables are skewed, so relatively small increments upward in predictive power may occur even when a model includes several variables with statistical significance.

Alvarez, Michael (2003). Personal communication with the author(s).