Trade Relationships and Asymmetric Crisis Perception1

Timothy M. Peterson Oklahoma State University Jerome F. Venteicher II

Abstract In this paper, we demonstrate that dependence on trade influences asymmetric crisis perception. Unilateral crisis perception is more likely to persist when the initiator of the crisis does not depend on trade with the target because in this case the target lacks capability to harm the initiator. Conversely, when the initiator is dependent on trade with the target, mutual crisis perception occurs sooner. Additionally, a state is more likely first to perceive a threat from another state – beginning a crisis as a target – when its trade dependence on that state is high. We find support for these expectations in survival time regressions and probit models spanning the period from 1919 to 2001. Right running head: Timothy M. Peterson and Jerome F. Venteicher II Left running head: Trade and Crisis Perception Word Count: 9,922

1

A supplemental appendix containing additional empirical tests is available at: https://sites.google.com/site/timothympetersonosu/research.

1

Because each side in an interstate dispute represents a potentially varied set of capabilities, traits, and beliefs, it follows that perceptions among actors in international disputes are not necessarily uniform. While one state may immediately interpret another's behavior as decidedly threatening to its interests, there is considerable variation in the time it takes for reciprocation of crisis perception. In fact, the crisis initiator itself may never perceive a crisis as defined by the International Crisis Behavior Project (ICB). Despite the fact that such cases, termed one-sided crises, represent approximately one third of all crises identified in the ICB,2 we have only begun to understand how asymmetric perceptions arise throughout the crisis process. Prior research isolates several factors associated with the existence of one-sided crises (e.g., Akbaba, James, and Taydas 2006). However, typically not explored is how trade relationships affect states’ potentially asymmetric perceptions of disputes. This lack of attention is somewhat surprising, given that many of the theories linking international trade to militarized conflict assume, either explicitly or implicitly, that trade affects the perception and decision calculus of foreign policy decision-makers. Yet, empirical tests of these theories focus almost exclusively on conflict outcomes rather than the perception of these foreign policy actors. However, because the ICB records the timing of crisis perceptions, there is opportunity to examine some of the fundamental foreign policy claims on which theories linking trade to conflict are founded.3

2

Hewitt and Wilkensfeld (1999) show that 109 of 325, or 33%, of non-intra-war crises, are one-sided. However, looking at the dyadic level, one-sidedness is even more common, occurring in 306 of 677 (45%) non-intra-war crises. 3 Indeed, assessments of the underlying assumptions on which trade-conflict theories are based have led to important advances in the field. For example, Barbieri and Levy (1999) contest the claim that conflict leads to lost trade – the fundamental assumption of the opportunity cost argument linking trade to peace. Their study has invigorated inquiry into the direction of causation in the trade-conflict relationship (e.g., Anderton and Carter 2001; Kastner 2007; Long 2008; Keshk, Pollins, and Reuveny 2004; Kim and Rousseau 2005).

2

In this paper, we examine how trade relationships affect the process through which policy-makers perceive, or fail to perceive, threats to their state's security. We argue that reliance on trade with another state provokes a perception of vulnerability to trade partners. Applying this argument to the occurrence of one-sided crises, we find that when one state triggers crisis perception in another state, its trade reliance on that state is associated with reciprocal crisis perception. Stated differently, a crisis initiator that is highly trade dependent on the crisis target will more quickly perceive a crisis than one that is not trade dependent. Conversely, when the initiator is less dependent on trade with the target, the initiator will be slower to perceive a crisis itself, and, therefore, a one-sided crisis will endure longer. Starting from the same assumptions, we also demonstrate that a state's higher trade dependence on a given trade partner is associated with a higher probability that it perceives an initial threat from that trade partner, leading to the onset of a crisis. The paper proceeds as follows. First, we review prior studies on one-sided crises, with a focus on the capability of targets to reciprocate disputes and the restraint by dispute initiators as two potential determinants of asymmetric crisis perception. Next, we explore the foreign policy implications of theories relating trade to conflict and develop a theoretical framework linking trade dependence – specifically, initiators' relatively low levels thereof – to the persistence of asymmetric crisis perception. Then, we present our research design, in which we use survival time regression and probit models to capture the effect of trade dependence on the duration of one-sided crisis and crisis onset from 1919-2001. Finally, we present our results and discuss the implications of our findings for future studies.

3

Asymmetric Perception in International Crises Snyder and Diesing (1977: 6; quoted in Hewitt and Wilkenfeld 1999; emphasis added) state that “[a]n international crisis is a sequence of interactions between the governments of two or more sovereign states in severe conflict, short of actual war, but involving the perception of a dangerously high probability of war”. As such, an understanding of the determinants of state perception is essential when studying these disputes, particularly given that these perceptions are not always equal.4 Examples of asymmetric perception include the relatively common yet rarely studied one-sided crisis, defined by the ICB as a case in which one state (the initiator) triggers a crisis for another state (the target) by some foreign policy action, but does not perceive itself to be in a crisis (Hewitt and Wilkenfeld 1999).5 For example, while the Cuban Missile crisis is a wellknown (two-sided) crisis in which Cuba (as a proxy for the Soviet Union) triggered threat perception for the United States, Cuba also sparked perception of a threat for the US in the leadup to the US invasion of Grenada in 1983, without itself ever experiencing crisis perception.6 In a one-sided crisis, one state perceives a disagreement to be much more serious and threatening to national wellbeing than does its adversary. One-sided crises appear most likely to occur if the underlying dispute is not threatening to both states' territory, core values, or to their existence in 4

A focus on state perception suggests that domestic sources of international disputes are crucial explanatory variables. Regime type is the most commonly cited domestic factor of crisis onset and reciprocation; however, these studies tend to examine regime type in terms of its impact on stability (e.g., Bennett and Stam 2005; Prins 2003), the influence of audience costs (e.g., Partell and Palmer 1999; Schultz 2001; Fearon 1994), or the existence of domestic opposition (e.g., Bueno de Mesquita and Lalman 1992), rather than in terms of how regime type affects policymakers' perceptions regarding the existence of threats from other states. 5 It is important to distinguish between reciprocal perception, dispute reciprocation, and escalation, as these are separate, although related concepts. Reciprocal perception does not necessarily require reciprocation of a militarized dispute (or even the initiation thereof), although it is intuitive that a reciprocated MID would likely invoke reciprocal perception. Similarly, reciprocation is not synonymous with escalation (which is usually described in terms of hostility levels within disputes), although Braithwaite and Lemke (2011; see also Schultz 2001) note that reciprocation of the same level of force can be viewed as one type of escalation. 6 Specifically, the US perceived a threat from Cuba due to Cuban influence (again, as proxies for the Soviets) in Grenada. Cuban workers were involved in the construction of an airport that US policy-makers feared could be used to as a Soviet military base. These construction workers later supported the resistance to the US invasion.

4

general (Hewitt and Wilkenfeld 1999; Akbaba et al. 2006). For instance, if a dispute emerges over low-level diplomatic disagreements or a state's influence, as was the case for the US and Cuba during the Grenada crisis, it is more likely that the actors will have asymmetric perceptions regarding the salience of the situation, relative to, for example, a territorial dispute. However, the explanation of why asymmetry in crisis perception can endure (in some cases, for years) is less clear. We focus on two possible explanations advanced by extant research. First, one-sided perception of crisis could follow from a disparity in capability to harm. For example, if two states are engaged in a political dispute, a disparity in their military capabilities could lead one side to consider the issue as more salient, potentially to the extent that it perceives a crisis. Intuitively, one would expect that if State A enjoys a large advantage over State B in relative military capabilities, then in a heated political conflict, leaders in B would be more likely than those in A to perceive the existence of a crisis, ceteris paribus. Simultaneously, B's relative weakness could preclude it even from desiring to instill in A a similar perception of foreign policy crisis. Looking at reciprocation of militarized interstate disputes (MIDs), Hensel and Diehl (1994) find evidence that a state facing a stronger adversary will be less likely to respond aggressively to provocation because it seeks to avoid escalating a dispute in which it suffers from a military disadvantage.7 Similarly, Prins (2003) finds evidence that major power initiators deter reciprocation of MIDs. Although Akbaba et al. (2006) find that asymmetric capabilities appear not to influences crisis type, they also find evidence that distance – another factor likely reducing capability to harm – is associated with a higher likelihood of one-sided crises.

7

However, Bueno de Mesquita, Morrow, and Zorick (1997) suggest that there may be a non-monotonic effect wherein an increase in the observable military advantage leads first to a decrease and then an increase in the probability of violence reciprocation.

5

Second, one-sided perception of international crises may follow from restraint by the initiator of the crisis, which in turn reduces the willingness of the target to reciprocate with enough force to trigger crisis perception for the target (Hewitt and Wilkenfeld 1999; Akbaba et al. 2006). This explanation follows from observation that asymmetric crisis perception is most likely when the initial crisis trigger is not severe. For example, a non-violent act, although triggering the perception of crisis, could lead the target to eschew escalation. Given perception of a crisis due to a non-violent act, all else equal, the gravity of the threat experienced by the target should be lower, leading it also to exercise restraint in its response. This expectation mirrors an argument by Leng (1993), who suggests that a crisis triggered by a violent event is likely to escalate to higher levels of violence due to norms of reciprocity. Restraint in the initial crisis trigger may also facilitate one-sided crises because the initiator presumes that the triggering event is not threatening to the target. Believing that the target will not perceive the situation as a crisis, the initiator will not expect violent reprisal and thus be less likely to perceive a crisis. Given the lack of evidence presented by Hewitt and Wilkenfeld (1999) and Akbaba et al. (2006) that power disparity affects crisis type, the restraint explanation appears to carry more weight than does the ability to harm explanation. However, both of these causal mechanisms deserve further exploration. Capability to harm has traditionally been operationalized primarily in terms of military capabilities, yet international trade is increasingly leveraged as a nonmilitary tool of coercion in world politics (e.g., Baldwin 1999). The determinants of foreign policy restraint potentially leading to one-sided crises are even less understood. However, studies linking interdependence to peace have suggested that economic ties serve to restrain policymakers from aggressive behavior (e.g., Polachek 1980). The consequences of trade for asymmetric perception are also implied by Bennet and Stam (2005), who find a disparity in the

6

effect of trade dependence between initiator and target during crises. While the influence in the former's trade dependence are unclear, target trade dependence demonstrates a positive relationship with crisis onset. Although the authors caveat this finding with the warning that data limitations preclude any generalizations, the trends Bennet and Stam present suggest that trade dependence does affect initiators and targets differently.8 We turn next to a discussion of how trade can influence the perceptions and actions of foreign policy decision-makers. The Foreign Policy Implications of Trade-Conflict Theories The trade-conflict literature is often described in terms of a theoretical battle between realists and liberals. Although this simplification obscures the fact that there are actually many, nuanced theoretical mechanisms purportedly explaining the trade-conflict relationship, most, if not all, of these theories assume that trade relationships affect the perception of foreign policy decision-makers. However, this perception is typically consigned to a “black box” in empirical research that tends to focus attention primarily on the relationship between trade levels and conflict outcomes.9 We contend that it is useful to examine the impact of trade on asymmetric crisis perception because the various theories of trade and conflict hold (often divergent) implications for both the capability to harm among trade partners and the likelihood that trade partners exercise restraint in foreign policy, both of which of have been suggested as determinants of one-sided crises. Theories typically associated with realism often view trade as potentially aggravating because reliance on trade with another state suggests vulnerability, which could lead to coercion attempts and violent actions to preclude such coercion (e.g., Hirschman 1945; Keohane and Nye 8

Bennet and Stam (2005: 142) also contend the more general point that, “trade correlates with lower levels of conflict,” further suggesting that asymmetric levels of trade dependence may be correlated with less severe disputes. 9 Dyadic conflict has been operationalized in a variety of ways, most commonly in terms of the probability of militarized interstate dispute (MID) onset (particularly for uses of force and cases in which there are battle deaths) and the prevalence of conflictual dyadic events.

7

1977; Barbieri 1996). The implications of this argument for the perception of threat by foreign policy decision-makers are clear: a state is more likely to perceive a threat to its interests from a trade partner on which it depends for its prosperity, either due to an explicit hostile act by its trade partner or simply due to the possibility of future threats. Another strand of realism suggests that trade relationships are epiphenomenal of state interests, yet potentially associated with higher levels of conflict because increased interaction equates with more opportunities for conflicts of interest to arise (e.g., Waltz 1970). If true, this claim implies that policy-makers will be more likely to perceive a threat to the state's interests from a trade partner if either its own trade dependence or its partner's dependence increase, given that both of these cases represent a deepening of economic ties that could lead to economic and political disputes. Theories of trade and conflict typically associated with liberalism have undergone considerably more evolution than their realist counterparts in recent years. Early liberal theories focused on the costly consequence of lost trade as a deterrent to conflict (Polachek 1980). These theories suggest that dependence on trade should lead policy-makers to exercise restraint in foreign policy in order to retain valuable trade gains. Recent studies have emphasized the role of trade in increasing the value of the status quo (e.g., Oneal and Russett 1999; Russett and Oneal 2001) and promoting common interests and preferences, such that fewer conflicts of interest arise (e.g., Dorussen and Ward 2010). All of these liberal theories imply that foreign policy actors should be less likely to perceive a crisis with trade partners. Some recent studies argue that the pacifying influence of trade arises from its facilitation of credible signaling rather than its function as an opportunity cost to conflict (e.g., Morrow 1999; Gartzke 2003; Reed 2003). Similarly, research on the capitalist peace also emphasizes the role of common preferences and

8

market signaling associated with commerce (e.g., Gartzke 1998; 2007; McDonald 2009).10 However, the common implication of these arguments is the same: crisis perception by foreign policy decision-makers should be less likely as either trade partner's dependence on trade increases. Modeling Trade Dependence and One-sided Crisis One of the challenges associated with uncovering the impact of trade on crisis perception is the fact that trade likely has its influence at the margins. Crises caused explicitly by trade vulnerability are rare; yet we contend that increased sensitivity to trade partners' actions resulting from dependence has a measurable impact on the endurance of asymmetric crisis perception when controlling for other common determinants thereof. One-sided crises present us with an opportunity to test empirically the expectations regarding the impact of trade on the perception of foreign policy decision-makers. We focus on the concept of trade vulnerability, which Keohane and Nye (1977) conceptualize as the opportunity costs State A stands to lose if State B chooses to restrict trade on which A is reliant either for income or for strategic resources. Put differently, vulnerability is complementary with capability to harm, capturing the economic advantage, and the accompanying political leverage, that B enjoys over A when the latter relies more on dyadic trade than does the former. Prior work demonstrates that a given flow of trade between two countries does not necessarily result in equivalent dependence on that trade (e.g., Hirschman 1945; Keohane and Nye 1977). For example, a state importing fuel or minerals on which its industries depend may be more vulnerable to interrupted trade than a state importing luxury goods. Although, sophisticated measures of vulnerability are generally unavailable, a simple measure of trade as a 10

However, recent studies of the capitalist peace suggest that financial openness may be more important than trade in promoting peace (Gartzke 2007; Gartzke and Hewitt 2010).

9

proportion of each state's GDP is used commonly to capture the salience of trade for a state's prosperity.11 We argue that as a state's prosperity is derived more from trade with a given trade partner, it will be more sensitive to reductions of that trade stream and, as such, will be more likely to perceive a foreign policy crisis were this trade partner to engage in potentially provocative behavior. Given that a state has triggered the perception of crisis for another state (that is, it is the initiator in a crisis),12 its own dependence on trade with its dyadic partner (the crisis target) will result in a quicker reciprocal perception of crisis by foreign policy decisionmakers, either because the target explicitly reciprocates the dispute by threatening or enacting a termination of trade ties on which the initiator relies for income or strategic commodities, or simply because policy-makers expect that such a reprisal is forthcoming.13 Conversely, an initiator with lower trade dependence on the target should be slower to perceive a crisis, given the lack of potential loss due to terminated trade.14 In this case, the target has less ability to harm the initiator.

11

See Gartzke and Li (2003) for a discussion of this and other measures of trade. Importantly, there is potential for a selection effect when examining the impact of the initiator’s trade dependence given that potential initiators are likely to consider the consequences of their actions prior to engaging in an action that might spark crises for the target (e.g., Reed 2000). In the supplemental appendix, we present robustness tests of our primary statistical models, demonstrating that our results do not suffer bias from non-random selection into crisis onset. 13 Although the target would also lose if trade is interrupted, prior work suggests that, during disputes where a termination of economic ties are possible, states focus on inflicting economic costs on their adversaries rather than avoiding economic costs themselves (Tsebelis 1990). Indeed, sanction threats are most likely to be used when states are prone to experience conflict (Drezner 1998). 14 We also test for an interaction effect in which the impact of one state’s trade dependence is conditional on the dependence of the other state. This follows from the notion that vulnerability is a function of relative trade dependence (Keohane and Nye 1977). For example, a moderate level of dependence for State A on State B is a disadvantage for A when B faces no reciprocal trade dependence on A (because only A is faces opportunity costs if trade is terminated), but an advantage when B’s trade dependence on A is very high (given that B faces higher opportunity costs in this case). All of our results are consistent when we model an interactive specification of each state's trade dependence. 12

10

The ICB dyadic crisis data demonstrate this pattern. For example, after the US began imposing increasingly tough sanctions on Libya in 1978 and US-Libya trade decreased accordingly (Hufbauer, Schott, Elliot, and Oegg 2007), the US sparked a series of crises for Libya – including the 1981 Gulf of Syrte conflict and the US military support campaigns for Sudan and Egypt in 1983 and 1984 – which failed to illicit a similar perception of threat in the US (Brecher and Wilkenfeld 2000). Arguably, if the US had greater trade ties to Libya, it may have been concerned with the high cost associated with terminating such a relationship, thus increasing the likelihood that it would perceive a crisis. Similarly, the United Kingdom initiated several crises for Iraq beginning in 1961. The earlier crises (regarding Kuwait's independence in 1961 and the Gulf War in 1990) also triggered crisis perception for the UK, while latter crises (regarding no-fly zones in 1992 and Iraqi expulsion of weapons inspectors in 1997) did not. Notably, UK dependence on Iraqi trade was more than 140 times higher in 1961 than in 1992 and 1997. Even in 1990, UK dependence was nearly 70 times higher than it was in 1992. In these cases, it is likely that UN sanctions taking effect after Iraq's invasion of Kuwait in 1990 led to decreased UK vulnerability to Iraq, on which it had previously depended, primarily for fuel imports. There is also variation within crises composed of multiple crisis dyads. For example, the 1990 Gulf War crisis is composed of 32 dyadic crises, of which 12 are one-sided. The average initiator trade dependence in these 12 one-sided Gulf War dyadic crises is approximately onetenth of the average exhibited in its two-sided crises. As such, the argument above leads to our first hypothesis. H1: Higher initiator trade dependence on the target is associated with a shorter duration of onesided crisis.

11

This argument may run counter to the expectations of scholars advocating the “liberal” view of trade and conflict, who might claim that a greater dependence on trade makes a potential initiator exercise restraint in order avoid lost trade gains (e.g., Polachek 1980; Oneal and Russett 1997, 1999; Polachek and Xiang 2010). However, these studies tend to expect a natural symmetry to dependence on trade. We contend that the potential for one state to rely on trade more (in this case in terms of their income) could promote fear and lead to divergent beliefs and preferences; as such, we advocate the capability to harm explanation of one-sided crises stemming from realist trade-conflict theories. However, if the restraint effect following from implications of the opportunity cost theory does operate, then higher initiator trade dependence might actually increase the duration of one-sided crises because the crisis trigger would, on average, be less severe. As such, by modeling of the impact of initiator trade dependence, we can compare the target capacity explanation of one-sided crises to the initiator restraint explanation, given the opposite results expected for initiator trade dependence. Similarly, the targets of crises may also exercise restraint in their response to provocation when their dependence on trade with the initiator is high if they seek to avoid losing trade with the initiator. This argument mirrors findings by Hensel and Diehl (1994) that targets avoid escalation when their adversaries have clearly superior military capabilities.15 If so, then the initiator may be slower to perceive a crisis in return. An alternative hypothesis follows: H1A: Higher initiator or target trade dependence is associated with a longer duration of onesided crisis.

15

Again, our primary expectation that this pattern does not occur with regard to trade dependence follows from insights of the sanctions literature, which demonstrates an apparent lack of restraint within conflict-prone dyads (Tselbelis 1990; Drezner 1998).

12

Our first hypothesis focuses on initiator vulnerability stemming from dependence on trade with the target (i.e., the target's capability to harm). However, we can also derive important expectations by examining the opposite case: the target's vulnerability stemming from trade dependence on the initiator. The capability to harm of the (potential) initiator is particularly important at the crisis onset stage. Given that exposure to dyadic trade for income invokes perceptions of vulnerability in policy-makers, we contend that initial crisis perception by the target will be more likely as the potential target's trade dependence on the potential initiator increases. This perception follows because foreign policy actors in potential targets will be sensitive to possibly hostile actions by potential initiators. Indeed, initiators may explicitly threaten the target with termination of trade if the target does not acquiesce to some demand, or the target may simply interpret trade partner behavior as threatening despite the lack of an explicit demand. At first glance, the data again support this claim. On average, crisis targets are more than 460 times as trade dependent on crisis initiators than are potential targets not perceiving a crisis from their dyadic partner. This argument leads to our second hypothesis: H2: Higher trade dependence by the potential target is associated with a higher likelihood of initial crisis perception. Again, this expectation regarding the influence of trade dependence on the perceptions of policy-makers diverges from that expected by advocates of liberal trade-conflict theories, particularly those emphasizing the role of trade in facilitating common interests and preferences (e.g., Dorussen and Ward 2010; see also Gartzke 1998). If trade facilitated common interests, we would expect more trade dependence for either the potential initiator or the potential target to be

13

associated with a lower likelihood of initial crisis perception.16 As such, we can test empirically the directly opposed foreign policy implications of realist and liberal trade-conflict theories on crisis onset as well as asymmetric crisis perception. An alternative hypothesis follows: H2A: Higher trade dependence by the either trade partner is associated with a lower likelihood of initial crisis perception. Research Design We examine the influence of trade dependence on crisis perception using data spanning the period from 1919 to 2001.17 In an important departure from recent studies of one-sided crises (e.g., Akbaba et al. 2006), we emphasize the utility of the directed dyad as the unit of analysis.18 The directed dyad allows for a more nuanced examination of the actors on each side of a dispute. Unlike a crisis-level approach, it is able to capture the “direction” of the initial threat perception, trade flows, relative military capabilities, and other explanatory factors capturing pre-existing relationships between states.19 This directional precision is available for crises because the ICB records the sequence in which states become crisis actors.20

16

However, prior work often notes that dependence on trade must be mutual in order for the pacifying impact to hold. Many empirical studies include variables for the less trade dependent state in a dyad, given that this state is the “weak link” with respect to the expected pacifying impact of trade (e.g., Oneal et al. 1996; Oneal and Russett 1997). Also, Morrow (1999), in his discussion of trade as having an unclear impact on trade partners' relative resolve, notes that asymmetric dependence could lead to asymmetric reduction in the resolve to fight, possibly increasing the probability of conflict. Again, our results are consistent when we model an interaction of each state's trade dependence. 17 Although the ICB crisis data extend beyond 2001, the dyadic level trade data limit the time span of our data. 18 Bennett and Stam (2000a) cite the potential danger associated with the use of directed dyads given possible violation of the statistical assumption of independence across observations. If A initiates a crisis against B, there may be less opportunity for B to initiate against A. All of our results are robust when we exclude all B-A dyads when A initiates a crisis against B (except when B initiates a crisis as well). Our clustered standard errors by non-directed dyad also mitigate this concern for non-independence. 19 Additionally, Morrow (1999) notes the necessity of directed dyads when examining the influence of relative trade dependence on crisis bargaining. 20 A focus on the timing of perception is central to our theory. Our conceptualization can be contrasted with an examination of crises in terms of challengers and defenders, a framework that focuses on desire to change or preserve the status quo rather than timing of actions (and perceptions).

14

We begin with Hewitt’s (2003) non-directed dyad dataset, reordering the dyads such that the state first causing another state to perceive a threat, i.e., the initiator of the crisis, is listed first. We merge this directed data with a directed dyad-year dataset produced in EUGene (Bennett and Stam 2000b). We exclude intra-war crises, as well as crises ongoing from the previous year, because we are interested only in cases with no existing dyadic conflict. Of the remaining 594 dyadic crises, missing data with respect to our explanatory variables reduce the sample of crises to 425 observations. The ICB codes one-sided crises as those in which the crisis ends before the initiator perceives a threat. However, even in two-sided crises, there is variation in the length of time between the target’s initial threat perception and the point at which the initiator likewise perceives a foreign policy crisis. This span of time in days, which can be interpreted as the duration of one-sidedness, is the dependent variable in tests of hypothesis 1 and 1A.21 Accordingly, we use a survival-time regression model, specifying a lognormal distribution of one-sided crisis duration,22 and coding one observation per crisis.23 Cases in which mutual crisis perception never occurs before the crisis ends – that is, one-sided crises as defined by the ICB – are right-censored in our data. There are 139 such occurrences among the 425 crises within our data. These cases provide information regarding the survival of one-sided crises but not on failure: that is, not on mutual crisis perception.

21

Some crises (67 in our data) become two-sided within a day of the target’s first threat perception. Recording the duration of one-sidedness as 0 in these cases would cause them to drop from the duration analysis. As such, we add 1 to the duration of one-sidedness in all cases. 22 We chose a lognormal distribution after comparison to alternate specifications of (1) a Weibull distribution and (2) an exponential distribution. A lognormal distribution fits the data best, as demonstrated by a comparison of log likelihoods for each model. The supplemental appendix includes a figure that illustrates the goodness of fit provided by a lognormal specification. Additionally, we do not present a Cox model because we are unable to test for nonrandom sample selection with this model. However, all results are consistent using a Cox model. 23 There are only 10 crises that last long enough (longer than 1 year) such that a multiple observations per case would be possible, given that our variables are available on a yearly basis. Additionally, our tests for selection bias are possible only when specifying one observation per case.

15

To test hypothesis 2 and 2A, we use data on directed dyad years between 1919 and 2001, composed of 1,082,087 observations. Our dependent variable, crisis onset, is dichotomous. Accordingly, we use probit models in these empirical tests.24 We tested for selection bias in our duration models by examining one-sided duration and selection into crisis simultaneously, employing a statistical technique developed by Boehmke, Morey, and Shannon (2006).25 However, all such tests demonstrate that our results do not suffer from selection bias.26 As such, we present these models in a supplemental appendix. In all models, we cluster standard errors by the non-directed dyad to account for heteroscedasticty resulting from non-independence by country pair. We also lag most independent variables by one year to mitigate simultaneity bias. Finally, in our crisis onset models, we include variables for years since crisis and cubic polynomials in order to mitigate duration dependence (Carter and Signorino 2010). Operationalizing Trade Dependence Our primary independent variables capture trade dependence between the initiator and target of the crisis. We present models using two different sources of trade data in order to test for the robustness of our results. First, we use the Oneal and Russett's (2005) dataset, which covers nearly the entire span for which we have crisis data (specifically, 1919-2001). As the

24

The 425 crisis initiations are rare events among the more than one million directed dyad years that span 1919 to 2001. However, results are consistent using rare events logit models (King and Zeng 2001). Additionally, results are consistent in replications in which the sample is constrained to dyads with opportunity for conflict However, given Benson's (2005) criticism of this type of sample selection variable in studies of trade and conflict, we present only the results using all dyads (all results are available by request from the authors). 25 Specifically, we use the DURSEL statistical model in Stata 11 to specify these equations (Boehmke 2005). These models are analogous to using a Heckman probit model on crisis onset and a binary one-sided crisis indicator. Additionally, in robustness checks presented in the supplemental appendix, we do specify a Heckman probit models with binary DVs in each equation, finding equivalent results. All results are consistent in these models. 26 Specifically the rho term in these models are far from significance, suggesting that there is no correlation between the error terms of our duration equation and crisis onset equation. As expected given the lack of finding for correlated error terms, substantive results of each equation in the simultaneous models are near exact matches for those presented in this paper.

16

Oneal and Russett dataset suffers from missing observations, we also use Gleditsch’s (2002) expanded trade data (version 4.1), which includes imputed values in order to fill in data gaps and reduce bias associated with missing data, but which covers a shorter time frame (1950-2000).27 From each data source, we code each state’s dyadic trade dependence, defined as bilateral trade flow divided by GDP.28 For example, to create the initiator’s dependence on the target (A’s dependence on B), the initiator’s exports to the target and imports from the target are summed and the total then divided by the initiator’s total GDP. This measure captures the importance of trade to each state’s national economy. High dependence suggests that a state earning a significant portion of its income from trade with a given partner would be hurt more if that trade were to be suspended.29 Additional Explanatory Variables In both equations, we control for the dyadic capability ratio, democracy within the dyad, distance, and preference similarity. We code the capability ratio using the Composite Index of National Capabilities (CINC) from the Correlates of War project (Singer, Bremer, and Stuckey 1972). Specifically, we take the natural log (plus one) of the ratio between the initiator's CINC score and the target's CINC score. We expect that a preponderance of capabilities for the initiator

27

Oneal and Russett (2005) use Gleditsch's (2002) data for the 1950-2000 period; however, they supplement it with pre-1950 data from Russett and Oneal (2001). Additionally, in the supplemental appendix, we present models using trade data from the Correlates of War (Barbieri et al. 2009). Although the COW trade data cover the entire period for which we have crisis data (and, in fact, are available through 2006), the only commonly used GDP data available before 1950 come from Oneal and Russett (2005). 28 Although “dependence” is the term commonly given to this measure, Crescenzi (2003) points out that this measure of trade interaction ignores the availability of alternate markets for each state. Crescenzi suggests using measures of exit costs, derived from import demand elasticities, thus capturing the ability of states to adjust down their demand if dyadic trade is terminated. However, exit cost data are too limited both temporally and spatially to be useful in this study. Accordingly, trade as a proportion of GDP serves as a proxy for exit costs. 29 We found no evidence of an interaction effect between each state’s trade dependence. Specifically, in models in which we added an interaction term of each state’s trade dependence, the interaction term was not significant and component variables looked nearly identical to those presented. A full examination of interaction effects confirms that the impact of each state’s trade dependence remains statistically unchanged over a wide range of the other state’s trade dependence (e.g., Braumoeller 2004; Brambor et al. 2006; Kam and Franseze 2007). Models with interactive specifications are available by request from the authors.

17

should make it more likely that the initiator fails to perceive a threat that it sparks for the target. Similarly, however, this power preponderance could spark initial crisis perception by the target, given that states should, all else equal, be more likely fear others with higher military capabilities. Weaker targets might also be less likely to reciprocate disputes (e.g., Hensel and Diehl 1994). Given that trade and power tend to correlate, controlling for relative capabilities should prevent potentially spurious results given that more powerful states may be less dependent on trade with a less powerful trade partner (e.g., Hegre 2004). Additionally, we code joint democracy as a dummy variable equal to one for cases in which both states score above six on the polity IV combined democracy-autocracy index (Marshall and Jaggers 2009). Democracy has been shown to have substantial effects on the ways in which crises are initiated and the manner in which they are fought (e.g., Bennett and Stam 2005; Prins 2003; Partell and Palmer 1999; Fearon 1994). Given that many advocates of the democratic peace point to trade as a mechanism bringing democracies together in peace (e.g., Oneal and Russett 1999), it is important to distinguish between the impact of regime type and that of economic ties. We take the natural log of the distance (plus one) between the states; contiguous states therefore receive a value of zero. Distance is a crucial control variable in studies of trade and conflict given that closer states tend to trade and fight more owing simply to increased opportunity for interaction afforded by proximity; the omission of a distance variable could produce spurious results in which trade appears aggravating (e.g., Oneal and Russett 1997). Finally, we include a variable for preference similarity, using Signorino and Ritter’s (1999) global, weighted S score. This variable, which captures the similarity of alliance portfolios, in theory ranges between -1 (most different) and 1 (identical); however, the lowest value present in

18

our data is approximately -0.6. We expect increasingly similar foreign policy preferences to facilitate information flow, preventing the onset of crises and precluding asymmetry of perceptions and, therefore, possibly reducing the period in which a crisis remains one-sided. In addition to baseline models, we present duration equation models including characteristics of the crisis for the target that may influence whether the crisis remain one-sided. Consistent with the extant literature, we expect one-sided crises to follow from instances of relatively low-level hostility (Hewitt and Wilkenfeld 1999) and not to involve severe threats to the target (Akbaba et al. 2006). Accordingly, we include variables for gravity of the threat (to the target) and violent trigger (both taken from the ICB). Gravity is coded as a dummy variable equal to one if the crisis represents a threat to the target's existence, or if the target fears “grave damage” (Brecher and Wilkenfeld 1989, 1997).30 We code violent trigger as a dummy variable equal to one when the initiator threatens, displays, or uses violence against the target, and zero otherwise.31 These variables are included to account for the fact that more severe initial threats are likely to lead to reciprocation of disputes (Leng 1993) and, therefore, of crisis perception, regardless of trade levels. Analysis We find support for our two primary hypotheses (1 and 2) in statistical tests of one-sided crisis duration and crisis onset. Specifically, we find that the initiator's trade dependence is associated with a shorter duration of one-sided crises and that initial crisis perception by the target is most likely when target trade dependence is higher. Importantly, we find no evidence

30

Although the highest values of the raw gravity variable represent the most grave threats, this gravity variable is not truly ordinal because it is difficult to rank lower-level threats (e.g., economic, political, and influence threats). As such, we code a dichotomous variable capturing the most severe threats to the target. 31 Essentially, this variable takes the value of 1 if the crisis trigger could be considered a militarized interstate dispute (MID) (Ghosn and Palmer 2003).

19

for alternative hypotheses (1A and 2A) linking trade dependence to restraint both before and during crises. Table 1 presents coefficients and robust standard errors for survival time regressions examining the duration of one-sided crises. We present an accelerated failure time interpretation of coefficients in all four models in Table 1.32 Models 1 and 2 utilize data from Oneal and Russett (2005), while Models 3 and 4 replicate 1 and 2 substituting Gleditsch (2002) data, which contain relatively fewer missing observations but cover a shorter time span. Models 1 and 3 present limited specifications, excluding gravity of the threat and violent trigger, while Models 2 and 4 include these additional explanatory variables. In each of Models 1 through 4, the coefficient for the initiator’s trade dependence is negative and significant. As such, higher initiator dependence appears associated with a shorter duration between target perception of crisis and reciprocal perception by the initiator. This result supports hypothesis 1. Conversely, the coefficient for target dependence is positive but not significant with regard to the duration of one-sided crises in any of the four models presented in Table 1. As such, there is little evidence that crisis targets exercise restraint in their response when they are dependent on trade with the initiator.33 [Table 1 about here]

32

An accelerated failure time (AFT) model is required when specifying a lognormal distribution of analysis time, unlike when using the Weibull distribution, with which one can specify an AFT model or a proportional hazards (PH) model. One benefit of AFT models is a relative robustness to unobserved heterogeneity when compared to PH models (Keiding et al. 1998). 33 Our supplemental appendix contains several robustness check models, the results of which are considerably similar to those presented in Table 1. For example, in Table A-1, which replicates our primary models using COW data, we find that the coefficient for initiator trade dependence is negative and significant at the 0.001 level, stronger than in the presented models (however these models suffer from considerably more missing observations). Table A2 presents probit models in which the DV is a binary indicator of one-sided crisis. With all three trade data sources, we find that our primary results are robust. Table A-3 contains DURSEL models examining one-sided duration and crisis onset simultaneously. Again, our results are robust with all three trade data sources. Finally, Table A-4 presents Heckman probit models estimating the probability of one-sidedness (a binary DV) and crisis onset simultaneously. Again, all results are robust.

20

Figure 1, derived from Model 2, presents the substantive impact of initiator trade dependence on the survival of one-sided crises.34 Specifically, the graph displays the probability that crises remain one-sided over time since the target first perceived a crisis. The solid line represents the probability of survival when the initiator does not depend on trade with the target for any of its GDP (that is, bilateral trade divided by the initiator’s GDP is equal to 0). The dashed line represents the probability that one-sided crises survive when the initiator’s trade dependence is held at the maximum value it takes in our sample of crises (when initiator dependence is equal to 0.38: i.e., when the initiator depends on trade with the target for 38% of its GDP). [Figure 1 about here] Although the survival of one-sided crises is more likely when the initiator is not trade dependent over the entire range of analysis time, the absolute difference is greatest early in the crisis. At day one, there is a 0.96 probability that the crisis remains one-sided when the initiator is not at all dependent on the target for trade. This probability falls to 0.77 when initiator trade dependence is held at its maximum. Both of these probabilities decline quickly within a few days in crisis. For example, after 10 days in crisis, with no initiator trade dependence, the probability that the crisis remains one-sided is 0.68; with maximum initiator trade dependence, however, the probability that the crisis remains one-sided is approximately 0.33. After 50 days in crisis, there is a nearly 0.4 probability that the crisis remains one sided if the initiator is not at all trade dependent on the target, while this probability is only 0.14 when initiator trade dependence is at its highest value. The marginal decline in survival probabilities diminishes after 50 days, signifying that crises that do not become reciprocated within this time are unlikely ever to do so

34

We use the stcurve command in Stata to create the survival graph.

21

before the target's crisis perception ends. The fact that the probability of one-sided crisis survival never reaches 0 reflects the fact that approximately one third of our cases never experience reciprocal crisis perception. Another method of assessing the impact of trade relationships on crisis perception is to compare the median survival time, i.e., the number of days at which 50% of crises have become two-sided. In each of Models 1 through 4, the median survival time decreases considerably as initiator trade dependence increases. For example, in Model 2 (from which Figure 1 is derived), median survival time falls from approximately 33 days when initiator trade dependence is held at the minimum, to approximately 4 days when initiator trade dependence is held at its maximum. In short, trade dependence appears to have a considerable effect on initiator perception. [Table 2 about here] Turning to empirical analysis of hypothesis 2, Table 2 presents Models 5 and 6, probit models examining the probability of crisis onset (using Oneal and Russett [2005] and Gleditsch [2002] data, respectively). In both of these models, higher trade dependence for potential target is associated with a higher likelihood that it perceives a crisis.35 Although all substantive probabilities are quite small, min-to-max changes in target trade dependence are associated with meaningful increases in the probability of crisis onset in both models. For example, in Model 5, this change is associated with approximately a 707% increase in the likelihood that the target perceives a crisis.36

35

As with tests of hypothesis 1, we present several robustness checks of hypothesis 2 in the supplemental appendix. All results are consistent both in single equation models of crisis onset and in simultaneous models of crisis onset and one-sidedness. 36 Specifically, the probability of crisis onset increases from 0.0000915 to 0.0007387 in Model 5. Probability changes are calculated as the new probability minus the old probability, divided by the old probably, times 100. As noted above, all results are consistent using rare events logit.

22

With regard to control variables, we find that relative capabilities appear unrelated to crisis onset and the duration of one-sided crises. This result supports previous findings that military capabilities appear unrelated to crisis type (Hewitt and Wilkenfeld 1999; Akbaba et al. 2006). However, in robustness check models in which we look at a binary indicator of one-sided crises, we find some evidence that relative capabilities are associated with a higher probability that a crisis is one-sided. These models are presented in the supplemental appendix. Remaining control variables generally look as expected. Distance is associated with a lower likelihood of crisis onset and a longer duration of one-sidedness. Joint democracy is associated both with a lower likelihood of crisis onset, and with a shorter duration of one-sidedness, suggesting that democracies are less likely to face conflicts of interest and, when they do, are prone to similar perceptions. This result also supports a finding by Prins (2003) that democracies are more likely to reciprocate disputes. Preference similarity is associated with a lower likelihood of crisis onset, yet there is little evidence for its impact on the duration of one-sided crises (specifically, it appears associated with a longer duration of one-sidedness only in Model 1). Although previous research at the crisis level found that grave threats and violent triggers were associated with onesided crises (Akbaba et al. 2006), we find no such relationship in dyadic crises with respect to the duration of one-sidedness. Finally, in the crisis onset equation, the variable for years since crisis and cubic polynomials suggest a decreasing baseline hazard of crisis onset over time. Discussion and Conclusion This study speaks to research on perception and the process by which nations develop assessments of their international environment. Myriad factors influence how states perceive themselves in relation to others, and as we argue in the preceding pages, international economic relationships are a key element in the shaping of a state’s perceived context. Specifically, we find

23

that once a crisis has begun, the more trade-dependent the crisis initiator is on the target, the more quickly the former will mirror the latter's crisis perception, while the potential target's trade dependence suggests higher likelihood of crisis onset. We argue that this relationship stems from trade as an indicator of capability to harm, which invokes threat perceptions. Although previous research has focused considerable attention on the link between trade and conflict outcomes, by looking at the crisis process, we demonstrate that vulnerability stemming from trade affects the very perception of whether conflict exists. Our results suggest that dependence on trade for economic wellbeing results in higher sensitivity to the loss of that prosperity. Foreign policy-makers will be quicker to perceive a crisis resulting from the actions of another state if they fear losing valuable trade gains. At first glance, this result appears contrary to liberal theories that trade reduces conflict by instilling common interests (e.g., Russett and Oneal 2001; Dorussen and Ward 2010) and improving information flows, facilitating costly signaling and less noisy bargaining (Morrow 1999; Gartzke 20003; Reed 2003). Given the improved logical consistently associated with these recent tradeconflict theories, the question arises of why policy-maker behavior appears to match expectations stemming from realist theories that have lost popularity among academics since the end of the Cold War. Future research should examine this puzzle. Future research can also benefit from examining how this impact of trade on perception translates to crisis escalation. Liberal theories connecting trade to peace – whether based on opportunity costs or improved information flows – could come into play after both sides have already perceived a crisis. For example, it seems logical that for opportunity costs to encourage policy-makers to avoid severe conflict, they must first be sensitive to the threat of lost trade. However, if only one trade partner actually perceives a crisis, then perhaps the opportunity cost

24

or bargaining impact of trade may not function. Many recent theories of trade and conflict assume that trade promotes common beliefs and preferences among policy-makers. Yet, if trade can sometimes lead to a divergence in these preferences (e.g., when one state depends on trade to a greater extent), then attention should be paid to the conditions in which trade might, alternatively, promote perceptions of vulnerability or camaraderie by foreign policy decisionmakers.

25

References AKBABA, YASEMIN, PATRICK JAMES, AND ZEYNEP TAYDAS. (2006) One-Sided Crises in World Politics: A Study of Oxymoron, Violence and Outcomes. International Interactions 32: 229-260. ANDERTON, CHARLES H., AND JOHN R, CARTER. (2001) The impact of War on Trade: An Interrupted Time-Series Study. Journal of Peace Research 38: 445-457. BALDWIN, DAVID A. (1999) The Sanctions Debate and the Logic of Choice. International Security 24: 80-107. BARBIERI, KATHERINE. (1996) Economic Interdependence: A Path to Peace or a Source of Interstate Conflict? Journal of Peace Research 33: 29-49. BARBIERI, KATHERINE, AND JACK LEVY. (1999) Sleeping with the Enemy: The Impact of War on Trade. Journal of Peace Research 36: 463-479. BENNETT, SCOTT D., AND ALLAN C. STAM. (2000a) Design and Estimator Choices in the Analysis of Interstate Dyads: When Decisions Matter. Journal of Conflict Resolution 44: 653-685. BENNETT, SCOTT D., AND ALLAN C. STAM. (2000b) EUGene: A Conceptual Manual. Interaction Interactions 26: 179-204. BENNETT, SCOTT D. AND ALLAN C. STAM. (2005) The Behavioral Origins of War. Ann Arbor: University of Michigan Press. BENSON, MICHELLE A. (2005) The Relevance of Politically Relevant Dyads in the Study of Interdependence and Dyadic Disputes. Conflict Management and Peace Science 22: 113133. BOEHMKE, FREDERICK J. (2005) DURSEL: A Program for Duration Models with Sample Selection (Stata version 2.0). Iowa City: University of Iowa. http://myweb.uiowa.edu/fboehmke/methods. BRAUMOELLER, BEAR. (2004) Hypothesis Testing and Multiplicative Interaction Terms. International Organization 58: 807-820. BRAITHWAITE, ALEX, AND DOUGLAS LEMKE. (2011) Unpacking Escalation. Conflict Management and Peace Science 28: 113-123. BRECHER, MICHAEL AND JONATHAN WILKENFELD. (1989) Crisis, Conflict, and Instability. Oxford: Pergamon. BRECHER, MICHAEL AND JONATHAN WILKENFELD. (1997) A Study of Crisis. Ann Arbor: University of Michigan Press. BREMER, STUART A. (1992) Dangerous Dyads: Conditions Affecting the Likelihood of Interstate War, 1816-1965. Journal of Conflict Resolution 36: 309-341. BUENO DE MESQUITA, BRUCE AND DAVID LALMAN. (1992) War and Reason: Domestic and International Imperatives. New Haven: Yale University Press.

26

BUENO DE MESQUITA, BRUCE, JAMES D. MORROW, AND ETHAN R. ZORICK. (1997) Capabilities, Perception, and Escalation. The American Political Science Review 91: 15-27. CARTER, DAVID AND CURTIS SIGNORINO. (2010) Back to the Future: Modeling Time Dependence in Binary Data. Political Analysis 18: 271–292. CRESCENZI, MARK J. C. (2003) Economic Exit, Interdependence, and Conflict. The Journal of Politics 65: 809-832. DORUSSEN, HAN, AND HUGH WARD. (2010) Trade Networks and the Kantian Peace. Journal of Peace Research 47: 29-42. FEARON, JAMES D. (1994) Signaling versus the Balance of Power and Interests: An Empirical Test of a Crisis Bargaining Model. Journal of Conflict Resolution 38: 236-269. GARTZKE, ERIC. (1998) Kant we All Just Get Along? Opportunity, Willingness, and the Origins of the Democratic Peace. American Journal of Political Science 42:1-27. GARTZKE, ERIC. (2003) The Classical Liberals Were Just Lucky: A Few Thoughts about Interdependence and Peace. In Interdependence and International Conflict: New Perspectives on an Enduring Debate, edited by Edward Mansfield and Brian Pollins. Ann Arbor: University of Michigan Press. GARTZKE, ERIC A., AND J. JOSEPH HEWITT. (2010) International Crises and the Capitalist Peace. International Interactions 36: 115-145. GLEDITSCH, KRISTIAN SKREDE. (2002) Expanded Trade and GDP Data. The Journal of Conflict Resolution 46: 712-724. HECKMAN, JAMES J. (1979) Sample Selection Bias as a Specification Error. Econometrica 47: 153–62. HENSEL, PAUL R., AND PAUL F. DIEHL. (1994) It Takes Two to Tango: Nonmilitarized Response in Interstate Disputes. The Journal of Conflict Resolution 38: 479-506. HEWITT, J. JOSEPH. (2003) Dyadic Processes and International Crises. Journal of Conflict Resolution 47:669-692. HEWITT, J. JOSEPH AND JONATHAN WILKENFELD. (1999) One-Sided Crises in the International System. Journal of Peace Research 36: 309-323. HUFBAUER, GARY C., JEFFREY J. SCHOTT, KIMBERLY A. ELLIOTT, AND BARBARA OEGG. (2007) Economic Sanctions Reconsidered, 3rd Edition. Washington, D.C.: Peterson Institute for International Economics. KAM, CINDY D. & ROBERT J. FRANZESE JR. (2007) Modeling and Interpreting Interactive Hypotheses in Regression Analysis. Ann Arbor: The University of Michigan Press. KASTNER, SCOTT L. (2007) When do Conflicting Political Relations Affect International Trade? Journal of Conflict Resolution 51: 664-688. KEOHANE, ROBERT O., AND JOSEPH S. NYE. (1977) Power and Interdependence: World Politics in Transition. Boston: Little, Brown.

27

KESHK, OMAR M. G., BRIAN M. POLLINS, AND RAFAEL REUVENY. (2004) Trade Still Follows the Flag: The Primacy of Politics in a Simultaneous Model of Interdependence and Armed Conflict. Journal of Politics 66: 1155-1179. KIM, YUNG MIN, AND DAVID L. ROUSSEAU. (2005) The Classical Liberals Were Half Right (or Half Wrong): New Tests of the “Liberal Peace,” 1960-88. Journal of Peace Research 42: 523-543. LENG, RUSSELL J. (1993) Interstate Crisis Behavior, 1816-1980: Realism vs. Reciprocity. Cambridge: Cambridge University Press. LONG, ANDREW G. (2008) Bilateral Trade in the Shadow of Armed Conflict. International Studies Quarterly 52: 81-101. ONEAL, JOHN R, FRANCES H. ONEAL, ZEEV MAOZ, AND BRUCE RUSSETT. (1996) The Liberal Peace: Interdependence, Democracy, and International Conflict, 1950–85. Journal of Peace Research 33: 11–28. ONEAL, JOHN R., AND BRUCE M. RUSSETT. (1997) The Classical Liberals Were Right: Democracy, Interdependence, and Conflict, 1950-1985. International Studies Quarterly 41: 267-293. ONEAL, JOHN R., AND BRUCE M. RUSSETT. (2001) Clear and Clean: The Fixed Effects of the Liberal Peace. International Organization 55: 469-485. ONEAL, JOHN R., AND BRUCE M. RUSSETT. (2005) Rule of Three, Let it Be? When More Really is Better. Conflict Management and Peace Science 22: 293-310. PARTELL, PETER J., AND GLENN PALMER. (1999) Audience Costs and Interstate Crises: an Empirical Assessment of Fearon's Model of Dispute Outcomes. International Studies Quarterly 43: 389-405. POLACHEK, SOLOMON W. (1980) Conflict and Trade. Journal of Conflict Resolution 24: 55-78. POLACHEK, SOLOMON W., AND JUN XIANG. (2010) How Opportunity Costs Decrease the Probability of War in an Incomplete Information Game. International Organization 64: 133-44. PRINS, BRANDON. (2003) Democratic Politics and Dispute Challenges: Examining the Effects of Regime Type on Conflict Reciprocation, 1816-1992. International Journal of Peace Studies 8: 61-84. REED, WILLIAM. (2000) A Unified Statistical Model of Conflict Onset and Escalation. American Journal of Political Science 44: 84–93. REED, WILLIAM. (2003) Information and Economic Interdependence. Journal of Conflict Resolution 47: 54-71. RUSSETT, BRUCE, AND JOHN ONEAL. (2001) Triangulating Peace: Democracy, Interdependence, and International Organizations. New York: W. W. Norton & Company. SCHULTZ, KENNETH A. (2001) Democracy and Coercive Diplomacy. Cambridge: Cambridge University Press.

28

SINGER, J. DAVID, STUART BREMER, AND JOHN STUCKEY. (1972) Capability Distribution, Uncertainty, and Major Power War, 1820-1965. In Peace, War, and Numbers, edited by Bruce Russett. Beverley Hills: Sage. SNYDER, GLENN H. AND PAUL DIESING. (1977) Conflict Among Nations. Princeton: Princeton University Press. WALTZ, KENNETH. (1970) The Myth of National Interdependence. In The International Corporation, edited by Charles B Kindleberger. Cambridge, MA: MIT Press.

29

Table 1. Trade relationships and the duration of one-sided crises. Oneal and Russett (2005) data: 1919-2001 Model 1 Model 2 Initiator trade dependence -4.857* -4.886* (2.574) (2.558) Target trade dependence 2.991 3.106 (3.595) (3.601) Capability ratio 0.150 0.135 (0.114) (0.118) ln Distance 0.128*** 0.130*** (0.039) (0.039) Joint democracy -0.965** -0.947** (0.421) (0.429) Preference similarity 0.466* 0.479 (0.281) (0.295) Gravity of the threat 0.535 (0.518) Violent trigger -0.046 (0.244) Constant 2.266*** 2.267*** (0.297) (0.300) Observations 425 425 σ 2.166*** 2.162*** Log likelihood -729.807 -729.433 χ2 (df) 25.31 (6)*** 28.08 (8)*** Lognormal distribution; accelerated failure time interpretation Standard errors clustered on the dyad *** p<0.01, ** p<0.05, * p<0.1; two-tailed tests

30

Gleditsch (2002) data: 1950-2001 Model 3 Model 4 -9.424* -9.322* (5.619) (5.609) 10.605 10.673 (11.369) (11.259) 0.049 0.044 (0.120) (0.120) 0.125*** 0.128*** (0.043) (0.043) -1.115*** -1.092** (0.424) (0.436) 0.319 0.310 (0.300) (0.310) 0.489 (0.582) 0.010 (0.252) 2.463*** 2.445*** (0.325) (0.332) 379 379 2.148*** 2.145*** -654.667 -654.444 22.88 (6)*** 24.46 (8)***

Table 2. Trade relationships and crisis onset. Model 5 Model 6 Oneal and Gleditsch Russett (2005) (2002) data: data: 1919-2001 1950-2001 Initiator trade dependence 0.033 0.068 (0.166) (0.277) Target trade dependence 0.242* 0.439* (0.110) (0.218) Capability ratio -0.016 -0.019 (0.009) (0.010) ln Distance -0.166*** -0.166*** (0.006) (0.006) Joint democracy -0.442** -0.433** (0.141) (0.148) Preference similarity -0.645*** -0.621*** (0.100) (0.109) Years since crisis -0.041*** -0.042*** (0.007) (0.008) Years since crisis2 0.001*** 0.001*** (0.000) (0.000) Years since crisis3 -0.000** -0.000** (0.000) (0.000) Constant -1.432*** -1.456*** (0.115) (0.125) Observations 1,082,087 1,0415,97 Log likelihood -3034.902 -2747.976 χ2 (df) 959.52 (9)*** 853.81 (9)*** Standard errors clustered on the dyad *** p<0.001, ** p<0.01, * p<0.05; two-tailed tests

31

Figure 1 (from Model 2).

32