Abstract
Crisis escalation to war is a subject of longstanding interest. Case studies, formal models and statistical analysis offer compelling explanations for why some crises escalate to war while others do not. Much less can be said in answer to the following question: where do crises come from in the first place? In this paper, we first introduce the concept of a near crisis following the approach taken over the course of more than four decades by the International Crisis Behavior (ICB) Project. A near crisis is just below a crisis as defined by ICB with regard to intensity, as it lacks one essential condition for a crisis—higher likelihood of military hostilities. Second, we present a newly developed dataset that contains information on 86 cases in which actors perceived a threat to one or more basic values, along with an awareness of finite time for response to the value threat. We also present simple statistical models comparing (a) near crisis to crisis and (b) crisis to war that show that analyses based on the Near Crisis dataset will contribute to advancement of knowledge.
Introduction
Crisis escalation to war is a subject of longstanding interest in International Relations. Case studies, formal models and statistical analysis offer compelling explanations for why some crises escalate to war while others do not (Asal and Beardsley, 2007; Ben-Yehuda et al., 2013a, b; Brecher et al., 2000; Gartzke and Hewitt, 2010; James, 2018). Since its inception in 1975, the International Crisis Behavior (ICB) Project has provided researchers with theoretical models and datasets that have facilitated a vast amount of research on causes, processes and consequences of crises in world politics. Existing studies use ICB data to explore the role of domestic politico-economic processes (Kisangani and Pickering, 2007, 2009, 2011; Pickering and Kisangani, 2005, 2010); economic factors (Gartzke and Hewitt, 2010); justice, legitimacy and norms (Butler, 2005; Gelpi, 2010); leadership (Chiozza and Goemans, 2004); media (Ben-Yehuda et al., 2013a, b); and other factors affecting the likelihood of crisis escalation to war. In contrast, much less can be said in answer to the following question: where do crises come from in the first place? And are there any substantial differences between the process of crisis escalation to war and transition from near crisis to crisis?
A principal reason for incomplete answers to these questions lies in the lack of data on periods prior to crisis onset. Since the ICB Project dataset only includes cases that escalated to a crisis level, it was previously impossible to develop and test models of crisis emergence and explore possible differences in the processes of crisis emergence and crisis escalation to war. In addition, ignoring the processes prior to crisis could result in sample selection bias. How can we assess the implications of a theory on the process of escalation using only a set of cases that already experienced significant escalation? In order to continue to advance our understanding of conflict processes, new data are needed. A new dataset on Near Crisis cases—cases that did not escalate to full-fledged crises—has been developed for precisely this reason. This dataset makes it possible to compare situations that (a) have been on the “knife’s edge” and tipped toward crisis with (b) those that did not rise to such a level of conflict.
Inspiration for further study of multiple stages of rising conflict comes from the analysis of Braithwaite and Lemke (2011), who focused on this matter in the context of Militarized Interstate Disputes (MIDs). Braithwaite and Lemke (2011: 111) observe that “very few investigations permit their causes to vary across different types of escalation”. As it turns out, the conventional wisdom about escalation with regard to MIDs is honored in some ways more than others. For example, the expectation that joint democracy will make “both onset and escalation less likely is roundly discredited by our analyses” (Braithwaite and Lemke, 2011: 119). Given the findings about MIDs, an assessment of whether escalation processes work the same way across stages in the context of crisis is, if anything, overdue.
This study unfolds in four sections. The first section focuses on the purpose of the study and introduces the idea of a near crisis. The next section, on data collection, describes the data and includes some detail about case identification and coding of near crises. The third section presents some simple theoretical models for (a) near crisis in relation to crisis and (b) crisis in connection to war, aiming to demonstrate the practical utility of the Near Crisis dataset. Results from the two models reveal at least some differences regarding variables that matter at each stage of escalation and point to the need for further investigation of these two processes. The fourth and final section sums up what has been accomplished and suggests ideas for future research. Our aim is to introduce the Near Crisis data and to persuade the reader that this dataset is useful for testing many hypotheses regarding conflict processes.
The concept of near crisis
Long remembered as an experience at the brink of World War III, the Cuban Missile Crisis impacted significantly upon scholarship in International Relations. The concept of an international crisis, in and of itself, attracted attention from scholars who hoped to carry out systematic research (Hermann, 1969; Young, 1969). Hermann (1969), for example, combined threat, short time and surprise into an influential exposition on the meaning of crisis. Data collection on conflict processes continues to be influenced by the concept formation from Hermann (1969), which provided a foundation for later elaboration. One salient example is the ICB Project, which through extensive empirical research confirmed the centrality of finite time to the dynamics of crisis (Brecher, 1977; Brecher and Wilkenfeld, 1997).
Near crisis, crisis and war all are subsets within the more encompassing set of international conflicts, broadly defined (James, 1988). The concepts of crisis and war have been identified as key tipping points or breakpoints already in the vast literature on conflict processes. This study adds the concept of near crisis to that list.
This study defines near crisis following the approach taken over the course of more than four decades by the ICB Project. Since its inception in 1975, the ICB Project has collected and analyzed data on crises since World War I. The reason for this time frame regarding data analysis is straightforward. The Project focuses on perceptions of decision-makers at the state level, which becomes much more feasible to study and code in the form of variables in the era after World War I. ICB Project data holdings on international crises and foreign policy crises, each to be explained in a moment, span the period from December 1918 through 2015. Identification and coding of new cases is an ongoing task, with data available at the Project’s website (https://sites.duke.edu/icbdata/).
The ICB Project has collected data on 1052 foreign policy crises, encompassed within 476 international crises, from December 1918 through 2015. Two conditions define an international crisis: “(1) a change in type and/or an increase in intensity of disruptive—that is, hostile verbal or physical interactions between two or more states, with a heightened probability of military hostilities, which in turn (2) destabilizes their relationship and challenges the structure of an international system—global, dominant, or subsystem” (Brecher et al., 2000: 39). A foreign policy crisis, according to ICB, occurs when the highest leadership of a given state perceives “a threat to one or more basic values, along with an awareness of finite time for response to the value threat, and a heightened probability of involvement in military hostilities” (Brecher and Wilkenfeld, 1997). Once the conditions for a foreign policy crisis are in place, the state in question is referred to as a crisis actor.
A near crisis is analogous to an ICB foreign policy crisis except that it excludes the third of the preceding three perceptions: these are cases in which a state perceives a threat from another state and finite time for response, but not yet a heightened probability of military escalation. These cases—some of which escalate to become crises, while others never do—entail situations on the knife’s edge of becoming a full crisis. Near crises combine into an overall case, just as foreign policy crises experienced by individual actors are linked together to form an international crisis. In the Cuban Missile Crisis, for example, the USSR, US, and Cuba experienced foreign policy crises and their collective experiences are assembled into the case known by that name. Case structure for near crises parallels that of crises. For example, within the near crisis context, the case identified as Upper Silesian Uprising encompasses the near crises in foreign policy of Poland, Germany, France and Britain.
One natural point of curiosity concerns the designation of threat and time as the perceptions that identify a near crisis. Why not some other combination of the three conditions? The reason is that basic properties rule out other possibilities. If the probability of escalation is perceived by leaders, it is certain that threat and finite time already are believed to be in place. Among the three conditions that make up a foreign policy crisis—threat to basic values, finite time for response and potential escalation of military hostilities—the only one that can be deleted on its own is the final one on that list. (In addition, threat is the anchor for both near crises and crises; in other words, how can finite time for response or risk of escalation in military hostilities be perceived in the absence of a threat to basic values?) Thus the near crisis combination of perceptions of threat to basic values and finite time for response, minus the belief in a possibility of escalation to military hostilities, stands as the natural event to focus on along the ladder of escalation just below the level of crisis.
Near crisis, international crisis, and war combine together to cover the most salient steps in escalation. A near crisis encompasses the experiences together of states with leadership that perceives a threat to basic values and finite time for response. An international crisis is the totality of experiences for states who perceive all three conditions for a foreign policy crisis. Thus an international crisis is said to begin when at least one actor perceives all three conditions of a foreign policy crisis and end when the last actor no longer perceives all of those conditions. International war is a proper subset of crisis—those cases that escalate to extreme violence assessed in terms of casualties.
The data
The Near Crisis dataset includes 86 cases in which a state perceives a threat to one or more basic values, along with an awareness of finite time for response to the value threat. The current version of the dataset covers the full range of the ICB Project dataset, from December 1918 through to the end of 2015. Since this is the first study to include data on near crises, things get underway with a presentation of the procedures used to identify such cases. Near crisis data collection occurred in two phases: (a) case identification; and (b) coding. The time interval-based approach to finding cases will be conveyed first at this point. The coding-related discussion will be brief because the procedures of the ICB Project have been refined over the course of decades, with very high reliability in inter-coder comparison as the norm.
Time-based case identification consisted of primary, secondary, and tertiary stages. Case identification at the primary stage involved, for a given month, a search through Lexis-Nexis and other mechanisms available via the internet for news items that reveal near crisis potential. 1 Two team members were assigned each month in primary research. If a potential case identified at the primary stage was deemed viable after the principal investigator surveyed the full team of research assistants, secondary research—based on extensive review of academic source materials (e.g. articles from JSTOR and university press books)—then ensued. 2
The formal procedure of case identification required every team member at both primary and secondary research stages to produce a document delineating how a would-be case in question satisfies each criterion of a near crisis: a threat to one or more basic values and a finite time for response to the value threat. In addition, and quite important in practice, is the requirement that a case is international in character. The justification to include a case in the collection had to be supported by the references from primary or secondary sources depending on the research stage. During bi-weekly team meetings, the resulting documents were compared to make sure that the investigators reached a similar conclusion without much “coaching” from a PI.
If a consensus could be reached, the case was included or excluded from the dataset. Significant disagreement among team members about the case at the secondary stage produced a tertiary stage of research in which an area expert was asked to render an opinion. Experience with this set of procedures produced the following pattern, with approximate percentage values: 50% of potential cases were eliminated at the primary stage; about 25% of those that reach the secondary stage went into the dataset, with 10% undecided; and the small number of tertiary instances entered the dataset at a rate of approximately 75%.
With regard to the effort of identifying periods of near crisis that never became a crisis, 86 near crises have been identified and coded. Table 1 conveys the most basic information about the cases in this new dataset: name, actors involved and time span. The relatively small number of cases may come as a surprise given that (a) near crises have more permissive conditions than crises; and (b) there are many more ICB crises over the same period. This relative scarcity confirms a point in theory from earlier on, namely, that the near crisis definition effectively identifies knife-edge events of special interest to the study of escalation.
Near crisis cases, 1918–2015.
Note that the list in Table 1 includes only those near crises that did not evolve into crises (the so-called standalone near crises). For any given international crisis, it is possible and even probable that, for one or more of the states involved, a near crisis came into being in close proximity beforehand. 3
The table shows that some of the cases involve only one near crisis actor, for example, Bulgarian Exit from World War II, Amethyst Incident, Deportation of Turks from Bulgaria, and a few others. This happens when only one state satisfies the conditions for a near crisis actor, i.e. one state perceives a threat from another state and finite time for response while another state does not satisfy one or both of these conditions. For example, in the case of Deportation of Turks from Bulgaria, the Turkish government perceived a threat of a potential refugee crisis (i.e. economic burden and potential national security threat) as a result of Bulgaria’s actions. Turkey was also acting under substantial time pressure as Bulgaria issued a decree to deport all Turks living in the country within three months. In contrast, Bulgaria’s actions were taken out of a desire to antagonize Turkey and receive material benefits from the deportations rather than in response to a threat or time pressure. Thus, this near crisis involves only one actor—Turkey—that satisfies the conditions for a near crisis actor.
It is also important to mention that the exact starting or ending time for some of the near crises could not be determined. Such cases have asterisks in the start/end date column. For example, in the case of the killing of two Albanian Soldiers, neither primary nor secondary sources provide a clear identification of finite time, rather indicating that the tensions simply faded over time. In other cases, we were not able to identify a termination date specifically down to an exact day, for example, only a month or (rarely) a year is known. In such cases the unknown information is marked by two asterisks.
Section 1.1 of the Online Appendix contains the basic orientation material provided to research assistants at the time they joined the near crisis research team to engage in case identification. Orientation also included meeting with the faculty member supervising the work and the opportunity to review accepted and rejected near crisis cases from previous work.
Sections 1.2 and 1.3 of the Online Appendix contain the instructions provided to research assistants at the primary and secondary stages of research for case identification. Research assistants at the primary stage were assigned a month (e.g. December 1990) and followed the procedures in Section 1.2 of the Online Appendix to locate candidate cases on an initial round. Each month was assigned at this primary stage to two team members, who worked independently to identify candidate near crises for assessment by the team as a whole. A debriefing session, led by the faculty member supervising the work, brought together all team members to discuss their research reports for months assigned on a given cycle. Potential cases were debated and those with significant support for further investigation—defined as at least one-third of team members—moved on to the secondary stage. (Records have been kept for all potential cases ruled out at the primary stage.)
After secondary research, following the procedures described in Section 1.3 of the Online Appendix, team members assembled once again for adjudication. If a clear consensus emerged in favor of including or excluding a case, a decision was made and coding began for the cases that had been approved. If not, a tertiary stage—reached for only a small fraction of potential cases—ensued. An area expert was consulted to offer additional analysis and a recommendation about the case; Section 1.4 of the Online Appendix contains the generic version of an invitation to provide a tertiary analysis. In all instances so far, the faculty member supervising the project’s research had concurred with the expert opinion and therefore been able to reach a final decision on the case.
Section 1.5 of the Online Appendix contains three sample case summaries, which are representative of those that exist for near crisis cases. 5 The case summaries offer enough detail and references to be at the level of those included for ICB cases and therefore suitable for more intensive analysis. The standard is set by the summaries from ICB (http://sites.duke.edu/icbdata/data-collections/), which provide a solid platform for further research.
While it is beyond the scope of this study to recreate the process of selection for a given case, providing a few details about events excluded from the dataset could help in grasping the criteria more effectively. Two examples of cases that received significant attention, but did not gain inclusion in the dataset, will be summarized: Iraq/Egypt Syria (March 1959) and China/USSR (December 1963).
The primary and secondary stages of research reveal the first of these two cases to be intrastate, rather than international, in character. Although the regime of President Gamal Nasser advocated for an immediate union between his United Arab Republic (UAR) and Iraq, he viewed the conflict between the government and rival nationalists in Iraq as something internal to that state. Nasser, put simply, wanted to stay out of that strife (Ismael, 2008: 86). The leader of the UAR also had assured assistance to the rebels, but that did not materialize. Instead, Nasser continued to give his usual radio speeches with a pan-Arab flavor (Farouk-Sluglett and Sluglett, 2001: 68). In sum, given the absence of evidence for involvement of the UAR as an actor, this potential case falls short owing to a lack of international character.
With regard to the second failed case, a lack of threat to basic values is the reason for exclusion after primary and secondary research. Tensions had been increasing rapidly between the PRC and USSR during 1963. Riots had taken place in the Sinkiang region of China and 100,000 refugees crossed the border into the USSR. The PRC accused the USSR of making things worse for them by supplying weapons to insurgents. On the other side of the issue, the Soviet Union did not want any more refugees, because Sinkiang Muslims would be very likely to be both anti-communist and trained as guerrilla fighters. Research reveals, however, that the border dispute and attendant refugee crisis did not rise to the level of a threat to basic values for either state involved. Border tensions had been longstanding and the refugee-related issues, according to the sources consulted, merely added to the Sino-Soviet split that had reached a point of culmination earlier in the year when the USSR signed the Partial Nuclear Test Ban Treaty, which the PRC regarded as a challenge to its nuclear program (Mastroianni, 1991; Robinson, 1972). 6 The refugee crossing dissipated rapidly and the issue faded away and became a minor adjunct to the overall rising conflict between the PRC and USSR.
This completes the overview of case-finding procedures for near crises. This is a much more detailed description than what is required for identification of crises. Location of such events is at an inherently lower level of difficulty in comparison with near crises, all other things being equal, given the higher prominence of these events. Material to identify cases of international crisis is more plentiful in both media and academic sources for that reason in particular. 7
The second stage of the data collection process involved coding of the identified near crisis cases following the ICB Project codebook. 8 The list of variables was divided into two broad types: actor-level variables and system-level variables. Actor-level variables pertain to specific actors’ attributes such as age of state, territorial size, political regime, alliance capability, and others. System-level variables pertain to contextual variables of a crisis, such as number of states in system, system polarity, superpower involvement in crisis, and others. For further details on the ICB variables see the Project’s website (https://sites.duke.edu/icbdata/). It is also important to note that owing to the particular nature of near crises, some variables were truncated; for example, the variable Intensity of Violence could not take value “4” indicating a full-scale war.
Although partially constructed prior to Salehyan’s (2015) epitomizing article on data collection best practices, coding of near crises is situated within those standards. ICB, for instance, has been very attentive to the need for “unambiguous” coding rules and near-crisis research adopts these procedures to maximize rigor and comparability. The Project endeavors to (a) be transparent about the sources; (b) consider omissions from them; and (c) address their potential biases. ICB is just as aware of the need for greater vigilance about reliability. Put forward by Ruggeri et al. (2011), a state-of-the-art approach toward reliability is adopted in ongoing work. Procedures go beyond the basic percentage assessment of intercoder reliability, already implemented in the pilot project for identifying near crises, and are found to be quite favorable—over 85% agreement prior to debriefing. Best practices on reliability involve some additional coding of cases already completed to permit assessment of stochastic and systematic errors (Ruggeri et al., 2011).
Near crisis is a new concept that might even be regarded as overdue for investigation. After all, the emergence of a crisis event, which by definition contains the risk of war, obviously is an important item for the agenda of both academe and the real world of politics. However, events that almost made it to the level of a crisis, but in the end did not, are much more challenging to identify. This is because an international crisis, even if it does not involve great powers, still is virtually certain to attract media attention. The same cannot be guaranteed for near crisis events. These quite easily can “fly under the radar”. Thus the near crisis data is especially valuable because it creates the opportunity for verification of what is taken to be known about higher rungs of the ladder of escalation. What if it turns out that there are important differences between how crises come into being vs. the processes through which these events end up in war? Given the relative difficulty of locating less intense events, bias may exist in what is believed about escalation. The two-stage analysis of escalation—impossible in the crisis domain prior to collection of the near crisis data—is the initial step forward to a more comprehensive treatment of the subject matter.
Ironically, this conclusion about the value of the data leads into a self-critique with regard to its completeness. Near crises that stand alone and do not move up to the level of a crisis are relatively difficult events to locate precisely because of their comparatively mild nature. Near crises that do not escalate, for instance, are very unlikely to generate huge newspaper headlines or show a sustained media presence. Consider this in contrast to international crises, which are much easier to identify. While near crises may be important to find, the risk of missing them becomes a concern. This point is relevant especially for the early years of the dataset, during which it is even more difficult to be confident that all relevant events have been located.
This limitation should be acknowledged as research moves forward on near crises. At present, further efforts to locate cases, with a region-specific and inductive character, are in progress. Initial efforts focus on Africa because it is at the greatest risk for missing cases owing to a lack of media coverage. Work on other regions, tailored to circumstances in each location, will follow on soon. Constraints on space prevent enumeration of the entire case-finding regime, but a few observations are offered to give a sense of the work being carried out for Africa. Given the particular possibility that internal instability may result in interstate conflict processes, various datasets are being consulted. The idea is to locate additional near crises, if they exist, through data collections that can go beyond the previous approach based on keyword identification to locate possible spillover effects from subnational strife. In addition, a team of area experts on Africa is being consulted about potential missing cases. 9
Simple factors of escalation
Models
In this section, we present some simple associations intended to show that the Near Crisis dataset can be helpful in testing hypotheses that could not be considered easily before. Specifically, we want to demonstrate that the Near Crisis dataset could help us explore the differences between the events that were on the brink of crisis but never escalated into full-fledged violence and actual crises. Using this new dataset we can finally begin exploring what differences, if any, emerge when variables already found to be significant in comparison of crises and wars are used for comparison between near crises and crises.
We want to emphasize that new theorizing about crises is not the priority for the current study. We do not conduct extensive diagnostics, examine the robustness of our results, or consider different operationalizations of the variables; therefore, the results presented here should be taken as suggestive and illustrative. Our goal is not to build comprehensive theoretical models of crisis escalation, but to suggest that the Near Crisis dataset could be helpful in exploring theories and testing hypotheses related to conflict processes. The approach parallels that of Braithwaite and Lemke (2011: 112) when comparing processes of escalation in the context of MIDs.
For this illustrative analysis, modeling is restricted to variables directly from within the ICB Project data holdings. We have selected factors that have particular interest in the literature and revealed sustained statistical associations with crisis escalation to war. The first three factors correspond to what might be called pre-existing conditions. These variables focus on the context for a given set of interactions, along with the traits of the parties involved. All of these variables are quite familiar within the lexicon of the ICB Project and have been shown as significant predictors of crisis escalation in prior data analysis (Brecher, 1999; Brecher and Wilkenfeld, 1997; Brecher et al., 2000; James, 2018). Thus only a brief rationale is provided for each proposition.
Regional Location is a nod to insight from area studies, which when transplanted into the world of conflict processes, tends to identify the Middle East as possessing a logic of its own. The region frequently is singled out in research designs because of its presumed vulnerability regarding escalation of conflict processes. For the purposes of our analysis, the region is coded dichotomously as 1 if the primary region of the international (near) crisis is Middle East and 0 if otherwise. This coding is adopted to follow from previous findings that conflicts occurring in the Middle East have higher probability of escalating into a full-scale war (Brecher, 1993).
Proximity corresponds to opportunity for interaction, whether cooperation or conflict (Most and Starr, 2015). Contiguous and nearby states, as opposed to those more distant from each other, possess greater inherent opportunity for contact of all kinds. Previous research confirms that contiguous states are more likely to engage in a dispute, and the farther apart dyadic states are, the less likely they are to be involved in a dispute (Choi and James, 2007; Most and Starr, 2015; Oneal and Russett, 1999). The distance factor is a standard in research designs that focus on interstate conflict. Proximity is coded categorically with three levels: contiguous actors, near neighbors, and distant actors.
Conflict Setting is a factor well-supported via ICB Project research. Prior conflict as an enabling condition for renewed strife is familiar from research even antedating the ICB Project (Wilkenfeld, 1973). Protracted conflict between adversaries establishes a context within which escalation becomes the norm rather than the exception (Brecher and Wilkenfeld, 1997; Brecher, 1999; Brecher et al., 2000). The protracted conflict variable identifies whether the states are involved in a certain type of a protracted conflict. The variable is coded through three categories: non-protracted conflict, non-long-war protracted conflict, and long-war protracted conflict.
Members of the second group of variables pertain to events already in progress and, like those preceding, are present because of a long and successful track record from ICB-based testing.
Breakpoint refers to the nature of the transition into international (near) crisis. The breakpoint is an event, act, or situational change that catalyzes a (near) crisis for the earliest actor. The role of the nature of breakpoint is among the most strongly verified generalizations in the history of ICB Project data analysis. A violent starting point is likely to produce more of the same as interactions continue. In line with previous research, we adopt a dichotomous coding of this variable (Brecher, 1993). Breakpoint is coded as 1 if (near) crisis is triggered by an indirect violent act or violent act, and 0 if triggered by any other events (such as verbal, political, economic or other non-violent acts).
Number of actors identifies how many states are perceived by the crisis actors to be involved in an international (near) crisis (including the actors themselves). Previous research indicates that, the larger number of actors involved, the more likely crisis escalation becomes (Brecher, 1993; James and Wilkenfeld, 1984). With a greater number of parties involved, all other things being equal, management of the situation becomes more difficult. The variable is coded categorically as “small” for conflict involving 1–2 actors, “medium” for conflict with 3–4 actors, “extensive” for 5–6 actors, and “large” for over 6 actors. 10
Finally, heterogeneity focuses on the diversity, as opposed to the sheer number, of participants in an event. Greater heterogeneity among those involved is expected to complicate matters and be associated, therefore, with the likelihood of escalation (Brecher and Wilkenfeld, 1997). Heterogeneity is measured by the number of attribute differences between the most heterogeneous pair of adversaries within an international (near) crisis. It is identified along four key attributes: military capability, political regime, economic development, and culture. Heterogeneity is coded as a categorical variable with five levels (no differences, one difference, two differences, three differences, and all four attributes are different). 11
We use simple difference of means tests to check whether crises are different from near-crises along these six substantively important characteristics. In order to provide the comparison, the Near Crisis dataset is combined with 470 crisis cases collected and coded by the ICB Project. Thus, the combined dataset contains two types of cases used for the analysis: near crisis cases (86) and crises (470).
In addition, we fit the same variables into a logit model to test whether the same factors explain the differences between crises and wars. The model of war emergence includes 470 cases from ICB. The dependent variable here is a familiar specification that identifies the extent of violence in an international crisis. It is coded 1 if a crisis escalated into a full-scale war and 0 otherwise. 12
One obvious critique comes to mind at this point: what, collectively speaking, does this analysis represent? The purpose of this study, as noted earlier, is not to produce new theory. Instead, a sample of variables with prior success in accounting for crisis escalation to war is included to provide a starting point for analysis of how crises come about in the first place. Thus the answer to the question above is that the collective identity of the variables tested in this study is not deemed to be a significant matter. Instead, the idea is to start the process of looking through the inventory of hypotheses about escalation, with a priority on those already “battle tested” through ICB Project research. Given the vast scope of ICB Project findings already in place, it is easy to put forward any number of alternative variables that might be included in (a) further study of how near crises develop into crises or (b) networks of variables within which the transition from near crisis to crisis plays some type of role that is not necessarily a purely dependent variable. 13 Taken together, items (a) and (b) just noted provide the foundation for a significant research agenda that is well worth carrying out.
Results
First, we will compare characteristics of near crises from the new Near Crisis dataset with characteristics of international crises from ICB. In Table 2 we present variables pertaining to pre-existing conditions: region, proximity, and protracted conflict. First, we see that region does not have a statistically significant relationship with the type of crisis. Both near crises and crises mostly occur in the regions other than the Middle East (83% of near crises and 85% of crises). In contrast, proximity of actors emerges as a highly significant differentiator between near crises and crises: 38% of near crises include contiguous actors while only 20% of crises have this characteristic. Conversely, a much higher percentage of crises emerge between distant actors—68% vs 48% of near crises. This finding is slightly surprising considering previous research that shows that proximity facilitates conflict escalation (Brecher, 1993; Brecher and Wilkenfeld, 1997) and is worthy of more in-depth investigation.
Pre-existing conditions.
The last factor based on external conditions—protracted conflict—shows a statistically significant relationship in the expected direction. An overwhelming number of near crises emerge in a non-protracted conflict setting (66%) with only 8% of near crises being a part of a long war protracted conflict. At the same time, 58% of all crises occur within some type of a protracted conflict. This finding is largely consistent with prior ICB Project research that has shown that prolonged strife between actors establishes a highly volatile context within which crises and escalations are frequent (Brecher and Wilkenfeld, 1997; Brecher, 1999).
Now, let us move to case-specific factors, such as nature of trigger, number of actors, and heterogeneity, presented in Table 3. First, as we can see, the nature of trigger has a significant relationship with the type of crisis. Only 22% of near crises are triggered by some sort of a violent act compared with 46% of crises. This result is not surprising considering the definition of a near crisis and is consistent with previous findings that a violent starting point is likely to produce more violent outcomes (Brecher, 1993).
Case-specific factors.
Moving to the number of actors, this variable emerges as a highly significant differentiator of crises and near crises as well. The table indicates that crises typically feature a higher number of actors than near crises—an overwhelming 64% of all near crises occur between 1 and 2 actors—and the number of actors barely exceeds 5. In contrast, one-half of all crises involve over 5 actors. This finding is consistent with an established result—that the greater number of parties involved, the harder it is to manage the conflictual situation (James and Wilkenfeld, 1984). Finally, the table shows that heterogeneity does not have a statistically significant relationship, as the majority of crises and near crises occur among actors with 3–4 different attributes.
We have thus far shown that near crises are distinct from crises on some theoretically important characteristics. Now, we can check whether a similar set of covariates can explain differences between crises and full-fledged wars. Table 4 presents the effect of the above-mentioned variables on war. 14 First, as we can see, while region does not emerge as a significant differentiation between near crisis and crises, it has a significant effect on the probability of crisis escalation to war. Changing region from 0 (not in Middle East) to 1 (Middle East) and holding other variables at their means or modes, the probability of crisis escalation increases from 3.9 to 8.1% (4.2 percentage points increase). 15 This finding might indicate the specific nature of conflict processes in the Middle East: the region might be highly vulnerable to interstate war once confrontation gets above a certain level; however, it is not more vulnerable to interstate crises than other regions. Besides, these differences might indicate that crisis initiation and crisis escalation, at least to some degree, are two separate processes that need to be explained through overlapping but still somewhat different theoretical frameworks.
Predictors of crisis escalation into war.
p < 0.05; **p < 0.01.
Notes: The table displays logit coefficients, standard errors in parentheses. The reference categories are: distant, non-protracted, small number of actors, no attribute differences.
Going further, the proximity of actors does not emerge as a significant predictor of crisis escalation to war. The coefficients for near neighbors and contiguity are positive, but do not achieve the required level of statistical significance. The last external conditions factor—protracted conflict—seems to affect escalation to war the same way it impacts upon escalation into a crisis. Cases of long-war protracted conflict, understandably, have higher propensity of moving into a full-scale war. Keeping other variables at their means and modes, the difference of war likelihood between long-war protracted conflict and non-protracted conflict is 23%.
Moving to the case-specific factors, breakpoint and the number of actors achieve the required level of statistical significance, indicating that crises and wars are different on these characteristics in the same way that near crises and crises are. The positive sign of the coefficient for breakpoint indicates that cases triggered by violent events have higher likelihood of escalating from a crisis into a full-scale war. Substantively, changing the breakpoint from non-violent to violent increases the probability of war by 5.5 percentage points. As for the number of actors, Table 4 indicates that a higher number of actors is associated with higher probability of war. When moving from small to large conflict, the probability of war increases from 2.6 to 17.5%. Finally, identically to our investigation of differences between near crises and crises, heterogeneity is not a consistent predictor of crisis escalation.
Overall, our results indicate that near crises and crises are distinct events that might converge in some characteristics but diverge in others. Besides, the characteristics that differentiate near crises and crises might be different from the characteristics that differentiate crises and wars. Using the novel Near Crisis dataset introduced in the paper, we show that, while some factors seem to differentiate these events in similar fashion (for example, protracted conflict, nature of trigger, and number of actors), some other factors might play different roles. For example, our results show that cases located in the Middle East have higher likelihood of escalating into a war, but near crises and crises are equally likely to emerge in this region.
Of course, the comparison of near crises vs. crises and crises vs. wars presented here is somewhat limited by the nature of data available. A more comprehensive analysis would require comparison between two escalation models: the model of escalation from near crises to crises and the model of escalation from crises to war. However, the modeling of escalation from near crisis to crisis will only be possible once we have the complete universe of all near crisis cases, some of which escalated to crisis and some of which did not. As mentioned above, the existing dataset includes only those near crises that did not evolve into crises. The next stage of data collection and coding will focus on near crises that precede existing ICB crises, thus allowing for more complete analysis of escalation from near crises into crises. 16 Despite this limitation, we believe that the Near Crisis dataset presented and described in this paper opens new opportunities for further investigation and comparison of conflict processes.
Final thoughts
The main goal of this paper has been to introduce a new concept of near crises in world politics and present the resulting Near Crisis dataset. We believe that this new dataset will prove to be a useful resource in the study of conflict processes and that it will enhance our understanding of how international crises emerge and develop. Existing models of escalation might suffer from sample selection bias as they use only a set of cases that already experienced significant escalation. Cases that reached a certain level of escalation are likely to have particular characteristics that produced escalation in the first place (e.g. certain leadership structure, economic system, etc.), that, in turn, might make escalation into a full-fledged war more likely. These particular characteristics make it difficult to assess the implications of a theory. For instance, if we wanted to examine the validity of nuclear deterrence theory, we would expect that crisis actors that have nuclear weapons would be less likely to escalate (Gartzke and Jo, 2009; Huth and Russett, 1990). However, the process by which actors already entered into crisis might have selected actors who are not deterred by the prospect of nuclear escalation. Thus, the deterrent effect of nuclear weapons might not be observed when studying only sets of opponents who self-selected into a hostile engagement. Valid analysis of the process of crisis escalation requires events on the knife’s edge—the cases in which crisis initiation could have occurred but did not, that we call near crises. The Near Crisis dataset could be especially useful in combination with the ICB Project dataset as the combined data allow for careful studying of key stages of crisis escalation.
In addition to addressing issues of selection bias that pervade models of escalation, coding of near crises allows for analysis of the initial steps of escalation that can be compared with crisis escalation to war. The results presented above are more exploratory than conclusive. However, they do point to one important conclusion: crisis initiation and crisis escalation are two separate, albeit related, processes that need to be theorized accordingly. While some factors differentiate near crises/crises and crises/wars in similar fashion, at least one of six well-established predictors for crisis escalation does not differentiate between near crises and crises (i.e geographic region). This preliminary finding questions the existing assumption that interactions occurring in the Middle East are more conflict-prone. While cases located in the Middle East do indeed have a higher likelihood of escalating into a war, they are not more vulnerable to interstate crises than other regions.
Further research is needed to assess the nuances of findings so far in this new, two-equation “world” regarding crisis escalation. Some ideas are obvious already. As mentioned earlier, it will be interesting to code all existing international crises for their near crisis phase, if any, and engage in further data analysis. Once we collect an additional dataset of near crises that preceded ICB crises, we will be able to combine these two datasets to conduct a comprehensive comparison of the processes of crisis initiation and crisis escalation. It is plausible that crisis initiation and crisis escalation are two separate processes that need to be explained through overlapping but still somewhat different theoretical frameworks.
Moreover, analysis presented in this study includes only a few readily accessible and well-verified factors from prior ICB Project data analysis on crisis escalation. So much more can and should be included in models of near crisis, and crisis, escalation. A new and challenging priority is to explore the distinct characteristics of near crises. Finally, results from this study and others of its variety should be used to guide process tracing through case studies to obtain greater in-depth knowledge about escalation-related causal mechanisms. We invite all interested researchers to start using the data and contribute to this new wave of research on conflict processes.
Supplemental Material
CMP855610_rep_data – Supplemental material for Near crises in world politics: A new dataset
Supplemental material, CMP855610_rep_data for Near crises in world politics: A new dataset by Evgeniia Iakhnis and Patrick James in Conflict Management and Peace Science
Footnotes
Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article. This work was partially supported by a grant: MINERVA #12432771 “Power Projection, Deterrence Strategies, and Escalation Dynamics in an Era of Challenging Near Peers, Rogue States, and Terrorist and Insurgent Organizations,” under the auspices of the Office of Naval Research. We thank Adam Badawy, Tom Jamieson, and James Lo for helpful comments and suggestions to previous versions of the paper. We also gratefully acknowledge financial support from the Center for International Studies at USC, the USC Undergraduate Research Associates Program, and the USC Summer Undergraduate Research Fund.
Supplemental material
Notes
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
