Questioning the Effect of Nuclear Weapons on Conflict

Abstract

We examine the effect of nuclear weapons on interstate conflict. Using more appropriate methodologies than have previously been used, we find that dyads in which both states possess nuclear weapons are not significantly less likely to fight wars, nor are they significantly more or less belligerent at low levels of conflict. This stands in contrast to previous work, which suggests nuclear dyads are some 2.7 million times less likely to fight wars. We additionally find that dyads in which one state possesses nuclear weapons are more prone to low-level conflict (but not more prone to war). This appears to be because nuclear-armed states expand their interests after nuclear acquisition rather than because nuclear weapons provide a shield behind which states can aggress against more powerful conventional-armed states. This calls into question conventional wisdom on the impact of nuclear weapons and has policy implications for the impact of nuclear proliferation.

Keywords

nuclear weapons nuclear proliferation international conflict

What effect do nuclear weapons have on interstate conflict behavior? Do nuclear weapons bolster the defense, deterring aggression and making states more secure? Or do nuclear weapons embolden the states that possess them, leading to conflicts that would not have occurred in their absence? Do nuclear weapons have different effects on conflict depending on the nuclear capabilities of the adversary? These questions, central to policy debates during the cold war, have only grown in importance in the last few decades with the emergence of new regional nuclear powers such as Pakistan, India, and North Korea. Today, debates over US policy toward the Iranian nuclear program largely hinge on differing opinions about the answers to these questions, as pessimists worry nuclear weapons will increase Iran’s belligerence (e.g., Kroenig 2012), while optimists argue nuclear weapons will either dampen Iranian aggression (as Iran grows more secure) or have no effect on Iranian behavior due to the United States' and Israel’s ability to deter Iran with nuclear weapons of their own (e.g., Waltz 2012).

Despite the importance of these questions, we lack clear answers. Empirical work has generally failed to keep pace with the extensive theoretical literature on the topic and findings from the few existing empirical studies are mixed. We identify and improve upon three shortcomings in existing work to seek more conclusive answers to these questions: (1) omitted variable bias stemming from the fact that conflict is one of the core drivers of proliferation, as well as one of its possible consequences; (2) miscoding of cases, specifically the 1999 Kargil War; and (3) the failure to properly deal with the methodological problem of separation.

We first review the existing literature on the connection between nuclear weapons and interstate conflict. Next, we analyze a dyad-year data set (used by Rauchhaus 2009) to examine whether existing findings on the effect of symmetric nuclear weapons possession on conflict are robust to the improvements noted above. We find that once pre-nuclear dyadic conflict is controlled for, symmetric nuclear dyads are not more likely to experience low-level conflict. Nor are they significantly more or less likely to experience full-scale war once we confront the problems of separation and the miscoding of the Kargil War, a finding at odds with the literature suggesting that nuclear weapons are a powerful force for peace. However, we find that asymmetric nuclear dyads (dyads in which one state possesses nuclear weapons) are more prone to low-level conflict than nonnuclear dyads, even after controlling for past conflict. To explore the finding on asymmetric nuclear dyads more closely, we use a directed dyad-year data set that allows us to distinguish between dispute initiators and targets and thus probe the underlying mechanisms that may connect asymmetric nuclear weapons possession and conflict. In contrast to previous work (e.g., Gartzke and Jo 2009), we find that nuclear-armed states are more likely to initiate disputes, even after controlling for the determinants of proliferation, but that this effect is limited and conditional. In particular, there is little evidence that states use nuclear arsenals as a shield behind which to aggress against more powerful conventional armed states. However, there is evidence that nuclear-armed states are more likely to initiate disputes against new opponents, a finding consistent with the idea that nuclear weapons lead states to expand their interests in world politics. In short, important elements of the conventional wisdom on the consequences of nuclear weapons on international conflict appear to lack empirical support. We conclude by outlining the significance of our findings and their policy relevance.

Existing Literature and Hypotheses

Symmetric Nuclear Dyads

Whether nuclear weapons spur or dampen conventional interstate conflict has long been debated (Sagan and Waltz 1995) and is perhaps the most fundamental theoretical question relating to nuclear weapons. Optimists have typically believed that nuclear weapons make states more secure, reduce the risks of misperception and false optimism, deter aggression, and thus reduce conflict (Waltz 1981; Mearsheimer 1984, 1990, 1993; Jervis 1989). Pessimists, by contrast, have generally accepted the logic of nuclear deterrence while noting the unique risks of interstate conflict that nuclear weapons create. These include the possibility of preventive war, accidental or unauthorized nuclear use, or escalatory risk taking (Feaver 1992–1993; Sagan 1994; Blair 1993; Kapur 2005). A third option is that nuclear weapons have a more nuanced effect on different levels of conflict. For example, the “stability–instability paradox” suggests that nuclear weapons might deter all-out nuclear war but exacerbate conflict at lower levels (Snyder 1965), a possibility Waltz recognized in his early work (Waltz 1959, 236).

Originally suggested by Snyder (1965, 198–199), the stability–instability paradox suggests that “the greater the stability of the ‘strategic’ balance of terror, the lower the stability of the overall balance at its lower levels of violence … firm stability in the strategic nuclear balance tends to destabilize the conventional balance.” In short, nuclear powers might not fight all-out nuclear wars against each other, but may experience more conventional conflict. Or, as Jervis (1984, 31) puts it, “to the extent that the military balance is stable at the level of all-out nuclear war, it will become less stable at lower levels of violence.” The stability–instability paradox implies the following hypothesis for the conflict behavior of nuclear dyads:
Hypothesis 1: Symmetric nuclear dyads are more likely to experience conventional war and low-level conflict than are nonnuclear dyads.¹
The most sophisticated empirical work on the topic, Rauchhaus (2009, reprinted in Rauchhaus, Kroenig, and Gartzke 2011), purports to find statistical support for the stability–instability paradox while controlling for the standard predictors of international conflict. He finds that dyads in which both states possess nuclear weapons are more likely to become involved in low-level disputes and conflicts, but much less likely to fight wars.² However, as noted earlier, the stability–instability paradox in fact implies that conventional war should be exacerbated (not reduced) by a stable nuclear balance; in other words, Rauchhaus’s (2009) findings actually contradict the stability–instability paradox, which only suggests that all-out nuclear war should be deterred. Indeed, Snyder (1965, 199) suggests that even tactical nuclear exchanges might be included among those levels of conflict that might be exacerbated by strategic stability—a level of conflict beyond any that the world has yet seen.

However, Snyder also acknowledges that “one could argue precisely the opposite [to the stability-instability paradox].” This provides further motivation for our analysis. It is possible that a stable strategic balance reduces all levels of conflict, because the fear of inadvertent or mistaken escalation to the nuclear level is sufficient to provide a significant deterrent to belligerence even at low levels of conflict. For example, Mearsheimer (2001, 129) argues that “nuclear weapons make states more cautious about using military force of any kind against each other” (emphasis added). The possibility of nuclear war—even if small—may make states more cautious instead of encouraging brinksmanship at low levels of conflict. This alternative nuclear optimist position is the one that Waltz and Jervis (in particular 1989, 21) would eventually adopt. This suggests Hypothesis 2:
Hypothesis 2: Symmetric nuclear dyads are less likely to experience all levels of conflict than are nonnuclear dyads.
Finally, it is possible that symmetric nuclear dyads behave no differently than do nonnuclear dyads. For example, existing theoretical works suggest that military technologies are not major causes of war or peace (Lieber 2005) and that the prohibitively high costs of conventional war in the modern era make nuclear weapons irrelevant as a cause of peace (Mueller 1988, 2010). This suggests Hypothesis 3, essentially a null hypothesis:
Hypothesis 3: Symmetric nuclear dyads are no more likely to experience war and low-level conflict than are nonnuclear dyads.
In addition to the competing theoretical claims about the effects of symmetric nuclear possession on conflict, there are strong methodological reasons for reexamining existing empirical work. First, while existing work finds that nuclear symmetry is associated with a greatly reduced risk of war, this is suspect due to the failure to address the problem of separation. Separation is a thorny problem that exists when a linear combination of independent variables perfectly predicts a value on the dependent variable and is often dealt with poorly in political science data (Zorn 2005). In this case, the problem is that no two nuclear-armed states have fought a war.³ As we discuss in more detail in the following, dealing inappropriately with the problem of separation has led existing work to wrongly inflate the effect of nuclear weapons by a factor of some 2.7 million.

Second, the association between nuclear symmetry and increased low-level conflict could be spurious since conflict is an important cause of nuclear proliferation. Extant work has largely failed to consider this serious threat to inference. To obtain unbiased estimates, one must control for all variables that are causally prior to the treatment variable (possession of nuclear weapons), affect the outcome (conflict), and are correlated with the treatment. Past conflict meets these criteria, since it is likely associated with variables such as mistrust that affect the probability of future conflict (Jervis 1976) and is correlated with the pursuit and acquisition of nuclear weapons (Singh and Way 2004). It may be that symmetric nuclear dyads are associated with low-level conflict because it is low-level conflict that leads them to acquire nuclear weapons in the first place, not because nuclear symmetry causes low-level conflict.

Asymmetric Dyads

Less work has been done theoretically on the effect of nuclear weapons on the likelihood of conflict between nuclear and nonnuclear states. Some scholars have worried that nuclear weapons make conventional conflict more likely (Betts 1988; Van Creveld 1993), perhaps as nuclear-armed states “bully or attack their non-nuclear neighbors and then use their nuclear arsenals to intimidate foreign powers from intervening” (Lavoy 1995, 737). This is in line with the long-standing fear that nuclear-armed states may “blackmail” nonnuclear adversaries with the threat of the offensive use of nuclear weapons, a concern that led China to develop nuclear weapons in response to American coercion in the 1950s (Lewis and Litai 1988), and that alarmed Indian, Japanese, Taiwanese, and American leaders following the Chinese nuclear test in 1964 (Campbell and Sunohara 2004; Ganguly 1999; Mitchell 2004; Gavin 2005; Miller, 2014). The belief that nuclear weapons make states more aggressive, both against nuclear and nonnuclear states, underlies much current concern over the Iranian nuclear program (e.g., Kroenig 2012).

Others argue that nuclear weapons may reduce the odds of aggression against nonnuclear states. According to Waltz (1981), nuclear-armed states will not need to engage in aggression, because their nuclear arsenal ensures their security. This rests on a defensive realist theory of international politics in which states seek security and are easily sated. If states seek prestige, to maximize relative power, or ideological or nationalist goals, however, there is no deductive reason to believe nuclear weapons would reduce aggression by the states that possess them (Lavoy 1995, 738).

There has been little empirical work on these questions.⁴ One exception is Gartzke and Jo (2009), which finds nuclear-armed states gain diplomatic status and are able to settle disputes in their favor peacefully, but are no more likely to initiate militarized disputes. Sobek, Foster, and Robison (2012) find that nuclear acquisition is associated with higher likelihood of being targeted in a militarized interstate dispute (MID), although this is lower than the likelihood of being targeted while in the process of acquiring nuclear weapons.

However, these findings are worth revisiting given concern that conflict may be a potential driver of nuclear acquisition. Sobek, Foster, and Robison (2012) acknowledge this as a concern and in robustness checks add a control variable that counts the number of times a state was targeted with an MID in the five years preceding an exploration of nuclear weapons. They acknowledge that when they do so that the coefficients on half of their variables of interest become statistically insignificant.⁵ This indicates that past conflict is a powerful confounder and that controlling for only a portion of it reduces the hypothesized effect of nuclear weapons on conflict. Gartzke and Jo (2009) acknowledge and explicitly seek to model the selection into nuclear weapons possession, but do not explicitly control for past conflict.⁶ These conflicting theoretical claims and empirical findings are reflected in the following hypotheses:
Hypothesis 4: Nuclear-armed states are more likely than other states to initiate disputes against nonnuclear states.

Hypothesis 5: Nuclear-armed states are less likely than other states to initiate disputes against nonnuclear states.
Moving beyond existing work, we introduce additional hypotheses to distinguish between two potential mechanisms that may cause nuclear-armed states to be more aggressive against nonnuclear states. First, possessing a nuclear arsenal may convince leaders that the downside risk of conflict is lower (since conquest can be deterred with the threat of nuclear retaliation), allowing them to initiate disputes against stronger conventionally armed enemies that would have been too risky in the absence of a nuclear arsenal. For example, North Korea’s nuclear arsenal may make its leadership more willing to initiate disputes with South Korea since its nuclear capability provides a shield against large-scale retaliation. Second, the perceived increase in power associated with the acquisition of nuclear weapons may cause states to take a more expansive view of their global or regional interests, leading them to be more aggressive and assertive with conventionally armed states that they may not have bothered with in the past. For example, American acquisition of nuclear weapons may have led American leaders to expand their conception of American interests in the post-1945 era, resulting in disputes against new nonnuclear states.

While identifying and testing theoretical mechanisms is normally a task reserved for qualitative research, we provide an initial assessment of the plausibility of these mechanisms by inferring and testing their observable implications. If the first mechanism is right, and nuclear weapons make states more likely to fight by lowering the downside risk of war, this suggests (1) that the nuclear-armed state should initiate disputes and (2) that the disputes should be disproportionately targeted at relatively powerful states (those that conceivably could have severely punished the nuclear state in the absence of its nuclear arsenal). If the second mechanism is correct, and nuclear weapons lead states to expand their global interests and assertiveness, we would expect (1) that nuclear-armed states initiate disputes, but (2) that these disputes should be aimed disproportionately at states with no history of conflict with the nuclear state. These two mechanisms suggest the following hypotheses:
Hypothesis 6: Nuclear-armed states are more likely to initiate disputes against nonnuclear states, and this association should be magnified against relatively powerful nonnuclear states.

Hypothesis 7: Nuclear-armed states are more likely to initiate disputes against nonnuclear states, and this association should be magnified against nonnuclear states with whom the state has no history of conflict.

Data, Method, and Results

In order to reexamine the effect of nuclear weapons in symmetric and asymmetric nuclear dyads, we use Rauchhaus’s (2009) data. We use Gartzke and Jo’s (2009) data to examine in more detail the effect of nuclear weapons on asymmetric dyads. These studies are among the most recent and empirically sophisticated existing work, and thus represent a strong starting point from which to begin our analysis.

Symmetric Dyads

To assess the conflict behavior of symmetric nuclear dyads, we start with a dyad-year data set (used by Rauchhaus 2009). However, we make important methodological adjustments to previous work. First, we address the problem of separation discussed briefly earlier. Rauchhaus's (2009, 268) analysis uses a logit generalized estimating equation (GEE) partly because GEE “allows us to deal with the problem of perfect separation…the data does [sic] not include any instances of two nuclear states engaging in war with one another. Thus, there is a perfect correlation between the absence of war and dyads in which both states possess nuclear weapons.”⁷ However, while GEE does allow one to deal with the autocorrelation that occurs in panel data where observations in one period are highly dependent on those in previous years, it does not get around the problem of separation—separation leads to nonidentification with GEE just as with many other generalized linear models for binary dependent variables such as logit or probit.⁸ Given the small number of nuclear-armed states, the limited period in which they have interacted, and the low baseline rate of war in general, it is plausible that the lack of major wars between nuclear powers is the result of chance, or is overdetermined by the presence of other conflict-reducing factors in nuclear dyads.

Because different software programs deal differently with separation, the results obtained depend strongly on the software program and convergence criterion used (Zorn 2005). In fact, Rauchhaus is only able to obtain results because the software he uses fails to detect the separation in the data.⁹ Because GEE coefficients cannot be easily interpreted substantively, we use Rauchhaus’ results to simulate the ratio of the probability of war in a nonnuclear dyad to that in a nuclear dyad (King, Tomz, and Wittenberg 2000).¹⁰ This ratio describes how many times more or less likely, on average, a nonnuclear dyad is to experience a given level of conflict compared to a nuclear dyad (holding other variables constant). Because the probability of conflict between any two states in a given year is low, this provides a straightforward and intuitive way of assessing the impact of nuclear weapons on conflict. A ratio of 1 implies that a dyad in which both states have nuclear weapons are equally likely to experience conflict as are dyads without nuclear weapons. A ratio of, say, 2 would imply that nonnuclear dyads are twice as likely to experience conflict as nuclear dyads. A ratio of 0.5 would imply that nonnuclear dyads are half as likely to experience conflict as nuclear dyads.

These ratios demonstrate that failing to account for separation leads to implausible results. As Table 1 shows, Rauchhaus's results imply that a nonnuclear dyad would be 2.7 million times more likely to go to war than the same dyad in which both countries possessed nuclear weapons. To correctly estimate the effect of nuclear weapons on war in the presence of separation, we follow Zorn (2005) and use a penalized maximum likelihood estimator developed by Firth (1993) as an alternative to the maximum likelihood estimation employed by GEE. By using the Jeffreys’s (1946) prior, this method, Firth logit, allows for the consistent estimation of parameters even in the presence of separation (Heinze and Schemper 2002). We utilize this method for the model where full-scale war is the dependent variable.

Table 1.
The Ratio of the Probability of War in a Nonnuclear Dyad to That in a Nuclear Dyad According to (1) GEE and (2) Firth Logit, Including and Excluding the Kargil War, along with 95 Percent Confidence Intervals.^a

(1) GEE (2) Firth logit

Estimate 95 percent confidence interval Estimate 95 percent confidence interval

Kargil excluded 2,717,000 [893,000, 8,531,000] 1.606 [0.088, 30.079]

Kargil included 0.693 [0.545, 9.563] 0.471 [0.077, 2.985]

Note: GEE = generalized estimating equation. All other covariates are held constant at their median values. An estimate of 1 indicates equal probability of war in nuclear and nonnuclear dyads. Ratios are simulated as recommended by King, Tomz, and Wittenberg (2000) and are based on 1,000 simulations.

^aAll models use the same battery of control variables used by Rauchhaus: contiguity, distance between states, state capabilities, presence of an alliance, a dummy for major powers, democracy, economic interdependence, and intergovernmental organization membership. Full regression tables for all models run are included in the online appendix.

Second, we make a small coding adjustment. In good social scientific research, coding rules should be determined ex ante and followed consistently. The data set used in Rauchhaus (2009) excludes the 1999 Kargil war between nuclear-armed India and Pakistan. This is justified in a footnote stating that the number of casualties was insufficient to warrant inclusion.¹¹ However, four wars with lower casualties than the Kargil war are retained in the data set. Further, the Correlates of War data set (whose coding is accepted for every other conflict) codes Kargil as a war: total losses of 1,172 are more than the 1,000-casualty threshold necessary for classification as a war, and the conflict featured two months of intense fighting in Kashmir.¹² We reinstate the Kargil war, and use Rauchhaus's original logit GEE model on the new data. Reinstating the Kargil war means there is no longer separation in the data, allowing us to assess the robustness of the war result both to the choice of estimator (logit GEE vs. Firth logit) and to small changes in coding the dependent variable.

Table 1 displays the results of the models where war is the dependent variable. Simply switching from logit GEE to Firth logit drastically changes the finding: instead of being 2.7 million times more likely to go to war, nonnuclear dyads are in fact no more likely to fight wars than nuclear dyads. The same finding holds if Kargil is included and the logit GEE model is maintained. Thus, regardless of whether we adopt a more appropriate estimator or change the coding of one observation in a defensible manner, we reach the same conclusion: symmetric nuclear dyads are not significantly less likely to go to war than are other dyads. Table 1 also shows that the Firth logit produces more plausible and stable results than does GEE: including or excluding Kargil has a smaller impact on the results.¹³ Regardless of whether Kargil is included or excluded, the confidence intervals on the estimate of the relative risk include 1, indicating that symmetric nuclear dyads are not significantly less likely to go to war than are nonnuclear dyads.

We now examine the effect of nuclear weapons on low-level conflict (fatal MIDs, MIDs that escalate to the use of force, and all MIDs). Before controlling for past conflict, Table 2 shows that nonnuclear dyads had only around 12 to 15 percent of the probability of engaging in conflict as symmetric nuclear dyads, suggesting that nuclear dyads are more prone to low-level conflict, with this difference being highly statistically significant. However, as discussed earlier, these estimates are likely to be biased because they do not control for past conflict. We remove this source of bias by controlling for the total number of MIDs a dyad has experienced prior to the year in question.¹⁴ When we do so, the difference between nonnuclear dyads and nuclear dyads is significantly reduced, and at each level of conflict the 95 percent confidence interval includes 1. In short, there appears to be little evidence that symmetric nuclear possession is associated with significantly higher levels of conflict at any level.¹⁵

Table 2.
The Ratio of the Probability of Each Level of Conflict in a Nonnuclear Dyad to That in a Nuclear Dyad According to (1) GEE (as used in Rauchhaus 2009) and (2) GEE but Controlling for Pre-nuclear Cumulative MIDs, along with 95 Percent Confidence Intervals.

(1) GEE (2) GEE controlling for past MIDs

Estimate 95 percent confidence interval Estimate 95 percent confidence interval

Fatal MID 0.128 [0.058, 0.289] 0.462 [0.067, 3.373]

Use of force 0.126 [0.055, 0.299] 0.344 [0.054, 2.420]

MID 0.145 [0.055, 0.384] 0.278 [0.051, 1.653]

Note: GEE = generalized estimating equation; MID = militarized interstate dispute. An estimate 1 of indicates equal probability of conflict in nuclear and nonnuclear dyads. All other variables are held at their median value. Ratios are simulated as in table 1.

To probe these results more deeply, we examine the conflict behavior of individual dyads. If dual nuclearization is associated with an increase in dyadic conflict, we might expect to see dyads engaging in more MIDs after the states within them acquire nuclear weapons. Examining the twenty-five symmetric nuclear dyads with at least five years of experience both pre- and post-dual nuclearization, there are only five dyads that experienced a higher rate of MIDs in the period after nuclearization and eight where the rate of conflict decreased (the remaining twelve saw no change in conflict rates).¹⁶ The US-Soviet Union, UK–USSR, and US–China dyads saw the largest increases in conflict rates post-nuclearization, with seventeen, six, and five more MIDs, respectively, in the twenty years post-nuclearization compared to the twenty years before. Conversely, France–USSR, France–China, and China–India experienced three, five, and eight fewer MIDs in the period post-nuclearization. This lack of a consistent effect of dual nuclearization in either direction runs counter to the expectations of both optimists and pessimists.

That the US-Soviet Union dyad would provide the strongest evidence of the stability–instability paradox is unsurprising since analysts observing the dynamics of the cold war developed the concept. Thus, if there is evidence for the stability–instability paradox anywhere, it ought to be found in the US-Soviet Union dyad. It is certainly true that the number of MIDs increased significantly once both the Soviet Union and the United States obtained nuclear weapons. However, attributing this change to nuclear weapons would be unwarranted. Nuclear acquisition by both states coincided with the start of the cold war and the corresponding shift in alliance patterns. The rise in MIDs could be equally attributed to this shift in geopolitics as to the acquisition of nuclear weapons. It is hard to disentangle the independent effect of nuclear weapons on the behavior of the United States and USSR during the cold war.

Our results, therefore, do not support the stability–instability paradox nor the positions of nuclear pessimists or optimists. Nuclear dyads do not appear to behave differently to nonnuclear dyads—neither more or less likely to fight wars nor significantly more or less likely to engage in low-level conflict than nonnuclear dyads. The finding that nuclear dyads are not less likely to experience war is surprising, given the influential theoretical literature arguing that nuclear arsenals drastically reduce the odds of war (e.g., Waltz in Sagan and Waltz 2003). There are several possible explanations for this null finding. First, it may be that states are more likely to initiate conflict when they have a material advantage, but that on average, symmetric nuclear dyads are relatively evenly matched, as are nonnuclear dyads, making each type of dyad similarly likely to experience conflict.¹⁷ Second, nuclear weapons may have different effects depending on the specific nuclear posture adopted, an argument advanced by Narang (2009–2010, 2013). Narang outlines three nuclear postures (assured retaliation, catalytic, and asymmetric escalation) and argues that only an asymmetric escalation posture—which threatens the first use of nuclear weapons in response to conventional attacks—should theoretically be expected to have a strong deterrent effect on conventional conflict. Third, it could simply be that the high costs of conventional war given modern technology make nuclear weapons superfluous as a deterrent, as suggested by Mueller (1988, 2010).

Our results also suggest that prior findings that symmetric nuclear dyads are more likely to experience low-level conflict result largely from selection effects: that past and anticipated military conflict drives the pursuit and acquisition of nuclear weapons (a conclusion reached in both quantitative and qualitative studies of nuclear proliferation, e.g., Sagan 1996–1997; Singh and Way 2004; Meyer 1986). Our finding that nuclear dyads are not significantly associated with increased rates of low-level conflict when past militarized disputes are controlled for supports this selection effect argument, and suggests that nuclear weapons may be less a cause than a consequence of conflict, at least in symmetric nuclear dyads.

Asymmetric Dyads

Although symmetric nuclear dyads do not appear more belligerent at any level of conflict once past conflict is controlled for, asymmetric nuclear dyads do appear more prone to low levels of conflict (MIDs and uses of force).¹⁸ As Table 3 shows, even once past conflict is controlled for, nonnuclear dyads have approximately two-thirds the probability of experiencing low-level conflict that asymmetric nuclear dyads do.

Table 3.
The Ratio of the Probability of Each Level of Conflict in a Nonnuclear Dyad to That in an Asymmetric Nuclear Dyad, along with 95 percent Confidence Intervals.

Estimate 95 percent confidence interval

War 0.730 [0.294, 1.857]

Fatal MID 0.644 [0.407, 1.024]

Use of force 0.644 [0.443, 0.945]

MID 0.633 [0.461, 0.873]

Note: MID = militarized interstate dispute. An estimate of 1 indicates equal probability of conflict in asymmetric nuclear and nonnuclear dyads. All other variables are held at their median value. Ratios are simulated as in the preceding tables.

To further probe the low-level conflict behavior of asymmetric nuclear dyads, we use a directed dyad-year data set (as used by Gartzke and Jo 2009). Unlike the dyad-year data set used above, the directed dyad-year data set allows us to identify the states which initiate conflict in a given MID and thus evaluate Hypotheses 6 and 7. We use a rare events logit estimator (King and Zeng 2001),¹⁹ controlling for the same variables as Gartzke and Jo (2009), and for temporal dependence using a variable measuring the number of years since the last conflict, along with the squared and cubed term of this variable (Carter and Signorino 2010). The results in Table 4 reinforce those obtained using the undirected dyad-year data set. Once the determinants of proliferation are controlled for, nuclear weapons states are not more prone to initiate MIDs against other nuclear states, but are more likely to initiate MIDs against nonnuclear states.²⁰

Table 4.
The Ratio of the Probability of MID Initiation in a Nonnuclear Dyad Compared to Asymmetric and Symmetric Nuclear Dyads Using Rare Events Logit, along with 95% Confidence Intervals.

Estimate 95 percent confidence interval

Asymmetric nuclear dyad 0.598 [0.397, 0.904]

Symmetric nuclear dyad 0.792 [0.409, 1.528]

Note: MID = militarized interstate dispute. An estimate of 1 indicates equal probability of conflict in nuclear and nonnuclear dyads. All other variables are held at their median values. Ratios are simulated as in the preceding tables.

We now seek to provide some traction on the question of whether this association exists because nuclear weapons reduce the downside risk of war and thus cause nuclear states to take on more powerful opponents than they would otherwise; or because nuclear weapons lead states to expand their interests and thus initiate MIDs with states who they would otherwise not engage in conflict with. We test the first mechanism using an interaction between the nuclear possession dummy variable for the initiating state (state A) and the ratio of state A to state B’s capability. We test the second mechanism using an interaction between the nuclear possession dummy for state A and a dummy variable indicating whether state A and state B had ever experienced an MID in the past.²¹ While this approach can only indirectly test the relevant theoretical mechanisms (Imai et al. 2011), it allows for a tentative initial assessment of the relative plausibility of the two mechanisms. We also control for the full battery of variables used by Gartzke and Jo (2009), including the determinants of proliferation.

The results in Table 5 favor the second mechanism. Nuclear weapon states are more likely to initiate MIDs against states with whom they have no previous MIDs, but no more likely to initiate MIDs against countries with whom they have a history of conflict. However, the interaction between nuclear possession and capabilities ratio is insignificant—the level of relative capabilities does not affect the association between nuclear weapons and conflict.

Table 5.
The Ratio of the Probability of MID Initiation in a Nonnuclear Dyad to an Asymmetric Nuclear Dyad Using Rare Events Logit.

Estimate 95 percent confidence interval.

Had previous MID 0.780 [0.546, 1.122]

No history of MIDs 0.353 [0.222, 0.567]

Capabilities ratio = 2 0.619 [0.410, 0.940]

Capabilities ratio = 0.5 0.620 [0.410, 0.941]

Note: MID = militarized interstate dispute. An estimate of 1 indicates equal probability of conflict in asymmetric nuclear and nonnuclear dyads. All other variables are held at their median values. Ratios are simulated as in the preceding tables above.

The “expanding interests” finding is not driven by one or a few nuclear powers. In fact, the data show that eight different nuclear states (the United States, the United Kingdom, France, Russia, South Africa, Israel, China, and India) initiated MIDs after nuclearization against states with whom they had no history of conflict.²² These results hold when the dependent variable is changed from simply initiating MIDs to MIDs that escalate to fatalities, regardless of initiator. While it is possible that this effect could be due to nuclear states initiating MIDs against powerful states that had previously deterred them with conventional superiority, and thus could in fact provide evidence for the first mechanism, the dyads involved suggest this is not the case. Of the forty-four cases where a nuclear state initiated an MID against a new, nonnuclear opponent, in only one of these cases was the target conventionally more powerful than the initiator (measured by composite index of national capability scores). The mean ratio of capabilities of initiator to target is 389:1, while the median is 47:1. It seems unlikely that states this conventionally inferior were able to deter the initiator pre-nuclearization. Rather, the evidence suggests that nuclear states find new—and often weak—opponents.

Conclusion

Our results call into question several aspects of the conventional wisdom on nuclear weapons and conflict. Using more appropriate methodological approaches than previous studies, we find that symmetric nuclear dyads are not less likely to fight wars, nor significantly more likely to engage in low-level conflict than nonnuclear dyads. This pattern holds up even when one examines individual dyads. Thus, for all its theoretical plausibility, this component of the stability–instability paradox appears to lack strong empirical support, as do the predictions of nuclear optimists and pessimists: the conflict behavior of symmetric nuclear dyads appears much the same as that of nonnuclear dyads. It is important to note, however, that we have only examined one of the predictions of the nuclear pessimists—that the spread of nuclear weapons increase conventional conflict post-nuclearization. We have not tested the view of the nuclear pessimists that nuclear proliferation may cause incentives for preventive war prior to nuclearization or raise the possibility of nuclear war or accidental nuclear use.

With respect to asymmetric nuclear dyads, even after controlling for the factors that lead to proliferation, nuclear-armed states are more likely to initiate low-level conflict against nonnuclear states, but this effect is limited and conditional. Rather than aggressing against more conventionally powerful states, the evidence suggests that states possessing nuclear weapons tend to initiate disputes against new, weak adversaries.

These findings have important policy implications. At least with regard to its consequences for conventional conflict, the findings suggest that nuclear proliferation is neither as menacing or stabilizing as many have believed. The pessimists’ prediction that nuclear weapons exacerbate international conflict after nuclear acquisition seems to lack support, except in the case of nuclear-armed states’ propensity to initiate disputes against new adversaries. Meanwhile, the optimists’ hope that nuclear weapons significantly reduce the likelihood of conflict also appears largely illusory. Nonetheless, contrary to Mueller (2010), our findings do not imply that policy makers should be unconcerned about nuclear proliferation. Even if nuclear dyads behave similarly to nonnuclear dyads, if the mere presence of nuclear weapons generates some non-zero probability of nuclear war, policy makers still have reason to worry about proliferation given the severe consequences of any nuclear use.²³ Moreover, as noted above, we do find that nuclear states are more likely to initiate low-level conflicts against new nonnuclear opponents, suggesting that nuclear weapons may not be completely irrelevant, and may lead states to expand their interests in world politics. This suggests at least a note of caution for those who see little danger in future proliferation.

	(1) GEE	(2) Firth logit
Kargil excluded	2,717,000	[893,000, 8,531,000]	1.606	[0.088, 30.079]
Kargil included	0.693	[0.545, 9.563]	0.471	[0.077, 2.985]

	(1) GEE	(2) GEE controlling for past MIDs
Fatal MID	0.128	[0.058, 0.289]	0.462	[0.067, 3.373]
Use of force	0.126	[0.055, 0.299]	0.344	[0.054, 2.420]
MID	0.145	[0.055, 0.384]	0.278	[0.051, 1.653]

	Estimate	95 percent confidence interval
War	0.730	[0.294, 1.857]
Fatal MID	0.644	[0.407, 1.024]
Use of force	0.644	[0.443, 0.945]
MID	0.633	[0.461, 0.873]

	Estimate	95 percent confidence interval
Asymmetric nuclear dyad	0.598	[0.397, 0.904]
Symmetric nuclear dyad	0.792	[0.409, 1.528]

	Estimate	95 percent confidence interval.
Had previous MID	0.780	[0.546, 1.122]
No history of MIDs	0.353	[0.222, 0.567]
Capabilities ratio = 2	0.619	[0.410, 0.940]
Capabilities ratio = 0.5	0.620	[0.410, 0.941]

Footnotes

Acknowledgments

We thank Robert Rauchhaus, Erik Gartzke, and Dong-Joon Jo for making their data available. We also thank Christopher Clary, James Conran, Jeremy Ferwerda, Brian Haggerty, Yue Hou, David Jae, Gary King, Vipin Narang, Molly Roberts, Brandon Stewart, Christopher Zorn, the panelists and discussant at MPSA 2013, and the Journal of Conflict Resolution editors and two anonymous reviewers for helpful suggestions and comments.

Authors’ Note

Regression tables for all models and robustness checks run are included in an online appendix, along with the materials necessary to replicate all findings.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

References

Asal

Victor

Beardsley

Kyle

. 2007. “Proliferation and International Crisis Behavior.” Journal of Peace Research 44 (2): 139–55.

Beardsley

Kyle

Asal

Victor

. 2009. “Winning with the Bomb.” Journal of Conflict Resolution 53 (2): 278–301.

Betts

Richard

. 1988. “Nuclear Peace and Conventional War.” Journal of Strategic Studies 11 (1): 79–95.

Blair

Bruce

. 1993. The Logic of Accidental Nuclear War. Washington, DC: Brookings Institution.

Bueno de Mesquita

Bruce

Riker

William

. 1982. “An Assessment of the Merits of Selective Nuclear Proliferation.” Journal of Conflict Resolution 26 (2): 283–306.

Campbell

Kurt

Sunohara

Tsuyoshi

. 2004. “Japan: Thinking the Unthinkable.” In The Nuclear Tipping Point, edited by Campbell

Kurt

Einhorn

Robert

Reiss

Mitchell

, 218–53. Washington, DC: Brookings Institution.

Carter

David B.

Signorino

Curtis S.

. 2010. “Back to the Future: Modeling Time Dependence in Binary Data.” Political Analysis 18 (3): 271–92.

Feaver

Peter

1992–1993. “Command and Control in Emerging Nuclear Nations.” International Security 17 (3): 160–87.

Firth

David

. 1993. “Bias Reduction of Maximum Likelihood Estimates.” Biometrika 80 (1): 27–38.

10.

Fuhrmann

Matthew

. 2009. “Spreading Temptation: Proliferation and Peaceful Nuclear Cooperation Agreements.” International Security 34 (1): 7–41.

11.

Ganguly

Sumit

. 1999. “India’s Pathway to Pokhran II: The Prospects and Sources of New Delhi’s Nuclear Weapons Program.” International Security 23 (4): 148–77.

12.

Gartzke

Erik

Dong-Joon

. 2009. “Bargaining, Nuclear Proliferation, and Interstate Disputes.” Journal of Conflict Resolution 53 (2): 209–33.

13.

Gavin

Francis

. 2005. “Blasts from the Past: Proliferation Lessons from the 1960s.” International Security 29 (3): 100–35.

14.

Geller

Daniel

. 1990. “Nuclear Weapons, Deterrence, and Crisis Escalation.” Journal of Conflict Resolution 34 (2): 291–310.

15.

Heinze

Georg

Schemper

Michael

. 2002. “A Solution to the Problem of Separation in Logistic Regression.” Statistics in Medicine 21:2409–19.

16.

Horowitz

Michael

. 2009. “The Spread of Nuclear Weapons and International Conflict: Does Experience Matter?” Journal of Conflict Resolution 53 (2): 234–57.

17.

Huth

Paul

. 1988. “Extended Deterrence and the Outbreak of War.” American Political Science Review 82 (2): 423–43.

18.

Huth

Paul

Russett

Bruce

. 1984. “What Makes Deterrence Work? Cases from 1900 to 1980.” World Politics 36 (4): 496–525.

19.

Huth

Paul

Russett

Bruce

. 1988. “Deterrence Failure and Crisis Escalation.” International Studies Quarterly 32 (1): 29–45.

20.

Imai

Kosuke

Keele

Luke

Tingley

Dustin

Yamamoto

Teppei

. 2011. “Unpacking the Black Box of Causality: Learning about Causal Mechanisms from Experimental and Observational Studies.” American Political Science Review 105 (4): 765–89.

21.

Jeffreys

Harold

. 1946. “An Invariant Form of the Prior Probability in Estimation Problems.” Proceedings of the Royal Society of London Series A 186:453–61.

22.

Jervis

Robert

. 1976. Perception and Misperception in International Politics. Princeton, NJ: Princeton University Press.

23.

Jervis

Robert

. 1984. The Illogic of American Nuclear Strategy. Ithaca, NY: Cornell University Press.

24.

Jervis

Robert

. 1989. The Meaning of the Nuclear Revolution: Statecraft and the Prospect of Armageddon. Ithaca, NY: Cornell University Press.

25.

Kapur

S. Paul

. 2005. “India and Pakistan’s Unstable Peace: Why Nuclear South Asia Is Not Like Cold War Europe.” International Security 30 (2): 127–52.

26.

King

Gary

Tomz

Michael

Wittenberg

Jason

. 2000. “Making the Most of Statistical Analyses: Improving Interpretation and Presentation.” American Journal of Political Science 44 (2): 341–55.

27.

King

Gary

Zeng

Langche

. 2001. “Explaining Rare Events in International Relations.” International Organization 55 (3): 693–715.

28.

Kroenig

Matthew

. 2009. “Importing the Bomb: Sensitive Nuclear Assistance and Nuclear Proliferation.” Journal of Conflict Resolution 53 (2): 161–80.

29.

Kroenig

Matthew

. 2012. “Time to Attack Iran: Why a Strike Is the Least Bad Option.” Foreign Affairs 91 (1): 76–86.

30.

Kroenig

Matthew

. 2013. “Nuclear Superiority and the Balance of Resolve: Explaining Nuclear Crisis Outcomes.” International Organization 67 (1): 141–71.

31.

Lavoy

Peter

. 1995. “The Strategic Consequences of Nuclear Proliferation: A Review Essay.” Security Studies 4 (4): 695–753.

32.

Lavoy

Peter

ed. 2009. Asymmetric Warfare in South Asia: The Causes and Consequences of the Kargil Conflict. Cambridge: Cambridge University Press.

33.

Lewis

John

Litai

Xue

. 1988. China Builds the Bomb. Palo Alto, CA: Stanford University Press.

34.

Lieber

Kier

. 2005. War and the Engineers: The Primacy of Politics over Technology. Ithaca, NY: Cornell University Press.

35.

Mearsheimer

John J.

1984. “Nuclear Weapons and Deterrence in Europe.” International Security 9 (3): 19–46.

36.

Mearsheimer

John J.

1990. “Back to the Future: Instability in Europe after the Cold War.” International Security 15 (1): 5–56.

37.

Mearsheimer

John J.

1993. “The Case for a Ukrainian Nuclear Deterrent.” Foreign Affairs 72 (3): 50–66.

38.

Mearsheimer

John J.

2001. The Tragedy of Great Power Politics. New York: W. W. Norton.

39.

Meyer

Stephen M.

1986. The Dynamics of Nuclear Proliferation. Chicago: University of Chicago Press.

40.

Miller

Nicholas L

. 2014. “Nuclear Dominos: A Self-defeating Prophecy?” Security Studies 23 (1): 33–73.

41.

Mitchell

Derek

. 2004. “Taiwan’s Hsin Chu Project: Deterrence, Abandonment, and Honor.” In The Nuclear Tipping Point, edited by Campbell

Kurt

Einhorn

Robert

Reiss

Mitchell

, 293–313. Washington, DC: Brookings Institution.

42.

Mueller

John

. 1988. “The Essential Irrelevance of Nuclear Weapons: Stability in the Postwar World.” International Security 13 (2): 55–79.

43.

Mueller

John

. 2010. Atomic Obsession. Oxford, UK: Oxford University Press.

44.

Narang

Vipin

. 2009–2010. “Posturing for Peace: Pakistan’s Nuclear Postures and South Asian Stability.” International Security 34 (3): 38–78.

45.

Narang

Vipin

. 2013. “What Does It Take to Deter? Regional Power Nuclear Postures and International Conflict.” Journal of Conflict Resolution 57 (3): 478–508.

46.

Rauchhaus

Robert

. 2009. “Evaluating the Nuclear Peace Hypothesis: A Quantitative Approach.” Journal of Conflict Resolution 53 (2): 258–78.

47.

Robert

Rauchhaus,

Kroenig

Matthew

Gartzke

Erik

. eds. 2011. Causes and Consequences of Nuclear Proliferation: A Quantitative Analysis Approach. New York: Routledge.

48.

Sagan

Scott D.

1994. “The Perils of Proliferation.” International Security 18 (4): 66–107.

49.

Sagan

Scott D

. 1996–1997. “Why do States Build Nuclear Weapons? Three Models in Search of a Bomb.” International Security 21 (3): 54–86.

50.

Sagan

Scott D.

Waltz

Kenneth N.

. 1995. The Spread of Nuclear Weapons: A Debate. New York: W. W. Norton.

51.

Sagan

Scott D.

Waltz

Kenneth N.

. 2003. The Spread of Nuclear Weapons: A Debate Renewed. New York: W. W. Norton.

52.

Sechser

Todd S.

Fuhrmann

Matthew

. 2013. “Crisis Bargaining and Nuclear Blackmail.” International Organization 67 (1): 173–95.

53.

Signorino

Curtis

Tarar

Ahmer

. 2006. “A Unified Theory and Test of Extended Immediate Deterrence.” American Journal of Political Science 50 (3): 586–605.

54.

Simmons

Beth

Hopkins

Daniel

. 2005. “The Constraining Power of International Treaties: Theory and Methods.” American Political Science Review 99 (4): 623–31.

55.

Singh

Sonali

Way

Christopher R.

. 2004. “The Correlates of Nuclear Proliferation: A Quantitative Test.” Journal of Conflict Resolution 48 (6): 859–85.

56.

Sobek

David

Foster

Dennis

Robison

Samuel

. 2012. “Conventional Wisdom? The Effect of Nuclear Proliferation on Armed Conflict, 1945-2001.” International Studies Quarterly 56:149–62.

57.

Snyder

Glenn

. 1965. “The Balance of Power and the Balance of Terror.” In Balance of Power, edited by Seabury

Paul

, 184–201. San Francisco, CA: Chandler.

58.

Van Creveld

Martin

1993. Nuclear Proliferation and the Future of Conflict. New York: Free Press.

59.

Waltz

Kenneth N.

1959. Man, the State, and War: A Theoretical Analysis. New York: Columbia University Press.

60.

Waltz

Kenneth N.

1981. “The Spread of Nuclear Weapons: More May Be Better.” Adelphi Papers No. 171, International Institute for Strategic Studies, London.

61.

Waltz

Kenneth N.

2012. “Why Iran Should Get the Bomb: Nuclear Balancing Would Mean Stability.” Foreign Affairs 91 (2): 2–5.

62.

Zorn

Christopher

. 2005. “A Solution to Separation in Binary Response Models.” Political Analysis 13 (2): 157–70.