When personalism matters: Nuclear latency and conflict propensity

Abstract

While some scholars conclude that there is a positive relationship between the possession of nuclear latency and the initiation of international conflict, others conclude that there is no relationship between them. As such, these findings are in need of reconciliation, but there have been almost no scholarly efforts to do so. In this article, I argue and find that latent nuclear states with personalist authoritarian regimes are more likely to initiate international conflict than those with other types of regimes (i.e. both democratic regimes and nonpersonalist authoritarian regimes). My finding indicates that not only personalist regimes drive the positive relationship between nuclear latency and conflict initiation, but also inconclusive findings in previous studies stem from an oversight of the variation in the level of personalism across latent states.

Keywords

international conflict nuclear latency personalism

Introduction

How does nuclear latency affect international conflict? Does nuclear latency—often defined as either the indigenous technical capacity to build nuclear weapons relatively quickly, particularly when a security environment deteriorates (Fuhrmann and Tkach, 2015), or, more simply, the level/status below actual nuclear weapons acquisition (Hiim, 2022)—provide states with deterrence benefits? Given such benefits deriving from “virtual nuclear arsenals” (Mazarr, 1995: 7–26), are latent nuclear states as conflict-prone as nuclear weapons states? If so, what might explain their conflict-proneness (i.e. their tendency to initiate militarized interstate disputes)? Since the advent of nuclear weapons, nuclear latency has been more common than nuclear proliferation (e.g. Fuhrmann and Tkach, 2015; Mehta and Whitlark, 2017). Mehta and Whitlark (2017), in fact, point out that “nearly thirty states have, at some point in their history, possessed dual-use technologies that provide nuclear latency, though only nine have ultimately acquired the bomb” (p. 517). But in part because of “data unavailability” (Fuhrmann and Tkach, 2015: 444), research on nuclear latency has lagged behind that of nuclear proliferation, and we still lack answers to these important questions.

Although inconclusive, few large-N studies that exist on nuclear latency, including Fuhrmann and Tkach (2015) and Mehta and Whitlark (2017), suggest that latent nuclear states, similar to nuclear weapons states, are more likely to initiate international conflict.¹ While these studies help us better understand the effects of nuclear latency on state behavior, they all rely on the rather simple logic of emboldenment alone to explain their outcome.² In this regard, Mehta and Whitlark (2017) emphasize that the presence of—or the belief in—deterrence benefits could embolden the foreign policies of latent nuclear states, as such benefits enable them to engage in “military actions they would otherwise refrain from taking” (p. 519). As such, I argue and find in this article that latent nuclear states featuring personalist authoritarian regimes are more likely to initiate international conflict than those featuring other types of regimes (i.e. both democratic regimes and nonpersonalist authoritarian regimes). My findings suggest that (1) personalist regimes drive the positive relationship between nuclear latency and international conflict and (2) inconclusive findings in previous studies stem from an oversight of the variation in the level of personalism across latent nuclear states.

Unlike democrats and nonpersonalist dictators, personalist dictators do not face a powerful domestic audience (i.e. either ordinary voters or regime elites) and, at the same time, are likely to favor the use of military force as an effective means of dispute resolution, particularly because of how they rose to power (Weeks, 2012, 2014). In the absence of a powerful domestic audience that could impose ex ante and ex post constraints in personalist regimes, and combined with their propensity to “select for highly violent and ambitious leaders” (Weeks, 2012: 335), deterrence benefits that derive from nuclear latency are likely to accommodate the already profound conflict-proneness of personalist dictators through further emboldenment. When emboldened further by their belief in deterrence benefits, personalist dictators are likely to perceive the costs of using force to be even lower than before, which, in turn, increases their frequency of conflict initiation.³

The rest of the article proceeds as follows. First, I review existing studies on nuclear latency, with a focus on those examining its impacts on state behavior. Second, I theorize the ways in which personalism shapes the effect of nuclear latency on conflict initiation. In this section, prior to deriving my main hypothesis, I derive two other hypotheses that serve as a stepping stone to—or a justification for—my main hypothesis. While the first hypothesis is for testing and confirming the positive relationship between nuclear latency and the initiation of international conflict, particularly given its inconclusiveness in previous studies, the second hypothesis is for testing and confirming the positive relationship between personalist regimes and the pursuit of nuclear latency.⁴ Third, I elaborate on my empirical approach. Finally, I present my findings and conclude by discussing broader implications and avenues for future research.

Existing scholarship on nuclear latency

When it comes to existing scholarship on nuclear latency, it should be noted that there are two major ongoing debates. While the first debate focuses specifically on whether nuclear latency increases or decreases “the security and bargaining power of those that possess it” (Mehta and Whitlark, 2017: 517), the second debate focuses specifically on how to properly define and measure nuclear latency.⁵ In this section, I focus on the former, as it is more pertinent to the research question at hand.

At the core of the debate on how nuclear latency impacts state behavior, as Mehta and Whitlark (2017: 518–521) point out, are two competing perspectives: virtual deterrence theory and latency provocation theory.⁶ According to virtual deterrence theory, it is possible for latent nuclear states to “derive many of the benefits of an operational nuclear arsenal without actually constructing nuclear devices” (Mehta and Whitlark, 2017: 519). Its advocates emphasize that nuclear latency—simply to the extent that it can fulfill the role of an actual nuclear arsenal—provides possessors with certain advantages in deterrence, coercion, and conflict dynamics. The possession of “a ‘standby’ or ‘virtual’ nuclear deterrent” (Mehta and Whitlark, 2017: 519) helps states deter aggression, compel others, and get emboldened. Latency provocation theory, on the contrary, challenges virtual deterrence theory, positing that nuclear latency is more of a burden than an advantage for possessors. As proponents of latency provocation theory stress, nuclear latency is unlikely to help states deter aggression, compel others, and get emboldened, particularly because of its tendency to invite not only preventive strikes, but also nonproliferation sanctions. The lack of an actual nuclear arsenal, therefore, undermines the security and prosperity of latent nuclear states.

The two opposing camps mentioned above are composed of various studies analyzing the effects of nuclear latency on deterrent capability, coercive leverage, and conflict initiation. What should be noted here is that those aligning with virtual deterrence theory and those falling under latency provocation theory reach contradictory conclusions. When it comes to the relationship between nuclear latency and deterrent capability, Fuhrmann and Tkach (2015), as well as Spaniel (2019),⁷ find strong evidence that nuclear latency enhances possessors’ deterrent capability. In other words, latent states are less likely to be targeted in militarized interstate disputes than non-latent states. Both Jones et al. (2023) and Mehta and Whitlark (2017), however, find that nuclear latency does not affect the likelihood of being targeted.⁸ At least to some degree, Fuhrmann (2017) reconciles such conflicting findings by specifying three conditions in which what he calls “latent nuclear deterrence” (p. 2) is likely to succeed. These three conditions are (1) that a potential aggressor must oppose the acquisition of nuclear weapons by a latent state, (2) that a potential aggressor must believe that attacking would not lower the probability of nuclear proliferation,⁹ and (3) that a latent state must have higher stakes than a potential aggressor.^10,11

With respect to the relationship between nuclear latency and coercive leverage,¹² Reiss (1988) finds that latent states are more likely to achieve coercive success than non-latent states. In his seminal study, Levite (2003) also finds that nuclear latency helps possessors reinforce their “coercive diplomacy strategy, particularly against the United States” (p. 72). These findings, in fact, played a key role in the emergence of something close to a conventional wisdom as to how nuclear latency would affect coercive leverage. Consistent with such “conventional wisdom,” Juneau (2015) argues and finds that nuclear latency (i.e. threatening to go nuclear) enabled Iran to blackmail the international community. However, Fuhrmann (2019) acknowledges that “[c]ompellence based on nuclear latency may be more difficult than deterrence, since the stakes involved are smaller for the latent nuclear power” (p. 301). At the same time, Mehta and Whitlark (2017: 523–524), through their empirical assessments, find not only that nuclear latency does not affect the likelihood of achieving compellent success, but also that it significantly increases the likelihood of inviting economic sanctions from the United States.¹³ Fortunately, Volpe (2017, 2023), similar to Fuhrmann (2017) mentioned above, serves as a “reconciler” by bringing levels of nuclear latency to the fore. According to Volpe (2023: 10–13), possessing a modest amount of nuclear latency—or hitting what he refers to as either the “fissile material sweet spot” or the “Goldilocks zone”—increases the probability of achieving compellent success, particularly against great powers, as doing so helps latent states make both their proliferation threats and their nonproliferation assurances credible.¹⁴

Finally, extant findings on the relationship between nuclear latency and conflict initiation remain inconclusive. For example, whereas Fuhrmann and Tkach (2015) fail to find robust evidence that latent states are more likely to initiate international conflict than non-latent states,¹⁵ Jones et al. (2023), along with Mehta and Whitlark (2017), find robust evidence that such is indeed the case. As such, these findings require reconciliation, but there have been almost no scholarly efforts to do so. In other words, among those exploring the relationship between nuclear latency and conflict initiation, there have been simply no “reconcilers” comparable to Fuhrmann (2017) and Volpe (2017, 2023). Perhaps the only exception in this regard are Jones et al. (2023), as they investigate whether leaders’ hawkishness shapes the relationship between nuclear latency and conflict initiation. Jones et al. (2023), however, do not find evidence of such a conditional effect, and thus conclude that “leader characteristics do not affect dispute initiation for latent states” (pp. 2, 12). To be clear, I am not arguing in any way that their recent findings are unimportant. Rather, the aim here is to stress that we are yet to identify a factor—or a predictor—that better reconciles the aforementioned findings related to the relationship between nuclear latency and conflict initiation.¹⁶

Prior to moving on to the next section, what should be noted is that personalism has not been examined in the context of nuclear latency, which is certainly an important phenomenon in international politics, to date. This is quite surprising for two reasons. First, personalism has been already examined in the context of nuclear proliferation. For example, Way and Weeks (2014) find that personalist regimes are significantly more likely to pursue nuclear weapons than other regime types. Second, we already know that personalism is linked to other important phenomena in international politics, particularly the initiation of military conflict. Weeks (2012, 2014), for instance, finds that personalist regimes, on average, initiate more international military conflicts than other regime types. In the next section, I tackle this apparent lacuna by examining the interaction between personalism and nuclear latency and its impact on conflict initiation.

Linking personalism and nuclear latency to the initiation of international conflict

As mentioned above, I derive a total of three testable hypotheses in this section, but the specific purpose of the first two hypotheses is to lay the foundation for the third main hypothesis, which states that latent states with personalist regimes are more likely to initiate international conflict than those with other types of regimes. Given the inconclusiveness of extant findings, the aim of the first hypothesis is to contribute directly to the ongoing debate on the effect of nuclear latency on conflict initiation by providing additional empirical assessments that further tip the balance in favor of one side over the other.¹⁷ The aim of the second hypothesis, on the contrary, is to confirm that there is a positive and statistically significant relationship between personalism and nuclear latency.¹⁸ Confirming the presence of such a generalizable pattern validates—perhaps naturally—a scholarly need to systematically analyze the potential synergy between personalism and nuclear latency, particularly when it comes to conflict initiation, as both are linked closely to conflict-proneness.¹⁹

I start by hypothesizing that latent states are more likely to initiate international conflict than non-latent states. Then, why are latent states more conflict-prone than non-latent states? In order to explain the conflict-proneness of latent states, previous studies (e.g. Fuhrmann and Tkach, 2015; Horowitz, 2013; Jones et al., 2023; Mehta and Whitlark, 2017) rely on the logic of emboldenment. Given that nuclear latency is an advanced capability that enhances the capacity of a state, it is likely to encourage a latent state “to pursue a more aggressive approach to international relations” (Jones et al., 2023: 4).²⁰ Put differently, the possession of nuclear latency, as Mehta and Whitlark (2017: 519) point out, helps embolden latent states’ foreign policies. At the heart of the emboldenment process, however, are the firm beliefs of latent states in deterrence benefits that derive from possessing nuclear latency or, more simply, what is known as latent nuclear deterrence. Similar to an actual nuclear arsenal, nuclear latency could reduce the costs of taking military actions, specifically by turning into “a shield behind which aggression can be undertaken” (Bell, 2015: 93). When equipped with such deterrent power, it becomes not only possible, but also easier for latent states to engage in “actions that would otherwise result in military retaliation” (Mehta and Whitlark, 2017: 519). For example, there are evidence that suggest that nuclear latency played at least some role in emboldening Iraq to invade Kuwait back in 1990 (e.g. Cigar, 2011; Fuhrmann, 2017).

According to Fuhrmann and Tkach (2015) and Fuhrmann (2019), there are two ways through which nuclear latency could deter potential aggressors. Both of them are likely to force potential aggressors to exercise caution. First, prior to launching an attack against a latent state, a potential aggressor must consider the possibility of that particular latent state carrying out a delayed nuclear counterattack. In this regard, Fuhrmann and Tkach (2015) emphasize that “the prospect of nuclear punishment could deter potential aggressors even if the target does not possess a nuclear arsenal at the onset of a dispute” (p. 453). Second, prior to launching an attack against a latent state, a potential aggressor must consider the possibility of that particular latent state going nuclear after being attacked. “[E]ven if there is no possibility of nuclear use in the context of the initial conflict” (Fuhrmann, 2019: 299), the prospect of nuclear proliferation could deter potential aggressors, as it is likely to undermine their own security over time.²¹ The logic laid out above leads to the following hypothesis:

H1: Latent states are more likely to initiate military conflicts than non-latent states.

States with personalist regimes are known for their keen appetite for pursuing nuclear weapons. In this regard, Way and Weeks (2014) find that states with personalist regimes are substantially more likely to pursue nuclear weapons than those with other types of regimes. According to Way and Weeks (2014), there are three reasons as to why they are likely to “find nuclear weapons to be particularly tempting” (p. 709). The first two reasons are related to personalist dictators ensuring their firm grip on power from external and internal threats, respectively, whereas the third reason is related to them being much less constrained when it comes to achieving their nuclear ambitions.²² First, personalist dictators are quite vulnerable to foreign intervention, particularly because of not only their tendency to commit violence against civilians or, more simply, human rights abuses (Davenport, 2007), but also their tendency to initiate international conflict (Weeks, 2012, 2014). Possessing nuclear weapons, therefore, helps personalist dictators shield themselves from outside interference (i.e. both overt military attacks and covert military operations) that could jeopardize their security (Way and Weeks, 2014: 710). Second, personalist dictators often have to worry about potential military coups against them.²³ That being the case, possessing nuclear weapons helps personalist dictators “coup-proof” their regimes, specifically by reducing—or obviating—their need to build and maintain an effective conventional army that could turn against them in the future. Third, unlike leaders of other types of regimes, personalist dictators enjoy a unique institutional setting in which domestic veto players, as well as domestic checks and balances, are largely absent (e.g. Way and Weeks, 2014; Weeks, 2012, 2014). As such, once they are determined to acquire nuclear weapons, stopping them from doing so becomes nearly impossible.²⁴ For instance, nobody was capable of curbing Kim Jong-il’s expensive nuclear ambitions even when the majority of the North Korean population was suffering from a massive starvation in the early 1990s (Way and Weeks, 2014: 710).

Based on the logic above, states with personalist regimes also should be significantly more likely to pursue nuclear latency than those with other types of regimes, particularly given that it typically comes before the successful acquisition of nuclear weapons; it often serves as a stepping stone to nuclear weapons.²⁵ In short, for the specific purpose of demonstrating a generalizable pattern discussed earlier, I derive and test the following hypothesis:

H2: States with personalist regimes are more likely to pursue nuclear latency than states with other types of regimes.

How does personalism shape the effect of nuclear latency on conflict initiation? Similar to Jones et al. (2023), I start by challenging the assumption that all latent states are equally likely to initiate international conflict. However, I differ from them in that I go beyond leaders’ hawkishness and argue that what also matters is a specific institutional setting that synergizes with their hawkishness. Undoubtedly, in the absence of such a “favorable” institutional setting, leaders’ hawkishness is less likely to be translated fully into policies, particularly because of what are known as veto points.²⁶ It should be noted that personalist dictators in personalist dictatorships become quite important in this regard, as the former are known for harboring extreme hawkishness and the latter are known for lacking institutional constraints (e.g. Way and Weeks, 2014; Weeks, 2012, 2014).²⁷ Combining them together, I contend that when personalist dictators become more hawkish through further emboldenment by possessing nuclear latency, primarily because of their firm belief in deterrence benefits that derive from doing so, no meaningful domestic audiences (i.e. those comparable to ordinary voters in democracies and regime elites in nonpersonalist dictatorships) would be present to impose constraints on—or alter the “deterrence belief” held by—them.²⁸ This particular mechanism, in turn, makes latent states with personalist regimes more conflict-prone than those with other types of regimes.

Personalist dictators are commonly characterized by their extreme hawkishness. Mattes and Weeks (2019) define being hawkish as “favoring military solutions over diplomatic ones” (p. 58). Given that personalist dictators often rise to power violently (i.e. through revolutions, civil wars, or violent coups), they are particularly likely to consider the use of military force an effective and even necessary strategy for resolving disputes (e.g. Colgan, 2010; Gurr, 1988; Horowitz and Stam, 2014; Weeks, 2012, 2014). The perceived costs of using force, therefore, are lower for them. In fact, Weeks (2012, 2014), consistent with such expectations, finds that personalist dictators are more likely to initiate international conflict than leaders of other types of regimes. Moreover, because of their “unusually ‘tyrannical’ personalities” (Weeks, 2014: 29), personalist dictators are particularly likely to be driven by not only grand international ambitions, but also strong desire for absolute domination. Under these circumstances, both nuclear latency and deterrence benefits that derive from possessing it are likely to stimulate personalist dictators’ preexisting hawkishness—or conflict-proneness—specifically by emboldening them further.²⁹ As long as they maintain their firm belief in latent nuclear deterrence, personalist dictators are likely to perceive the costs of using force to be even lower than before.

It is important to note that personalist dictators are uniquely free of ex ante and ex post constraints on their rule, as personalist dictatorships are marked by “a near-total absence of institutionalized veto players” (Way and Weeks, 2014: 709). As such, when it comes to government decisions, they exercise massive personal discretion. Both Way and Weeks (2014) and Weeks (2012, 2014), in fact, stress that this particular institutional setting is the “other side of the equation” that allows personalist dictators to easily—or rashly—engage in nuclear weapons pursuit and conflict initiation, respectively. Similarly, in personalist dictatorships, there would be no effective domestic audiences that could constrain the (emerging) interplay between personalist dictators and nuclear latency, particularly by casting doubt on the deterrent value of nuclear latency; doing so would be challenging the “deterrence belief” mentioned above. Arguably, intelligence elites are in the best position to steer such efforts. However, out of fear of reprisal, they often avoid relaying unpleasant information to personalist dictators (Frantz and Ezrow, 2009).³⁰ The logic laid out above yields the following hypothesis:

H3: Latent states with personalist regimes are more likely to initiate military conflicts than latent states with other types of regimes.

Data and research design

In order to test the hypotheses derived above, I construct two separate data sets. Specifically, the first data set is for testing both H1 and H3, whereas the second data set is for testing H2. For constructing the first data set, I rely primarily on data from Fuhrmann and Tkach (2015) and Weeks (2014). As the relevant measures from the Nuclear Latency (NL) data set (Fuhrmann and Tkach, 2015) are already included in Fuhrmann and Tkach’s (2015) first-ever empirical application of the NL data set, I focus on extracting the ones related to latency.³¹ On the contrary, I obtain measures related to personalism from Weeks (2014). The first data set, which covers the years from 1945 to 2000, is in a dyadic data structure, and thus the unit of analysis is the directed dyad-year. Therefore, it allows me to distinguish between initiators and targets of militarized disputes. At the same time, as Weeks (2014) points out, a directed dyad-year setup, unlike its country-year counterpart, is quite useful for accounting for “some of the most important correlates of military conflict, such as the balance of military power, alliance relationships, trading relations, and geographic proximity between a country and potential targets of force” (p. 41). Simply put, conducting a directed-dyad analysis helps reduce the likelihood of omitted variable bias, particularly when modeling conflict initiation.

For constructing the second data set, I use not only the NL data set, but also data from Way and Weeks (2014). I secure measures related to latency from the former and those related to personalism from the latter.³² The second data set, contrary to the first data set, has the country-year as the unit of analysis. However, similar to the first one, it covers the time period from 1945 to 2000. It should be noted that latency becomes the dependent variable in H2. Here, I employ a monadic approach because H2 does not require modeling conflict initiation.

Given that my dependent variables are dichotomous, I estimate the models using logistic regression. When testing H1 and H3, I, similar to Weeks (2014), follow Carter and Signorino (2010) and account for temporal dependence by adding cubic polynomials of time since last conflict initiation. Then, when testing H2, I include cubic polynomials of time passed without the pursuit of latency. According to Beck et al. (1998), failure to account for temporal dependence can be very consequential; in fact, doing so “can result in underestimates of standard errors, leading to unduly optimistic inferences” (Way and Weeks, 2014: 712). Moreover, I estimate the models with fixed effects for the purpose of controlling for both unobserved effects specific to directed dyads and those specific to countries.³³

Dependent variables

In H1 and H3, my outcome of interest is the onset of militarized interstate disputes (i.e. conflict initiation). As it is dichotomous, it takes the value of either 0 or 1. If State A (i.e. the challenger) in a directed dyad initiates a militarized dispute against State B (i.e. the target) in a given year, it is coded 1 and 0 otherwise. Following the bulk of the literature, I consider State A as initiating a militarized dispute against State B “if it is the first to threaten or use military force” (Weeks, 2014: 41).³⁴ There is one observation per directed dyad-year. I use data from Fuhrmann and Tkach (2015) to construct this particular variable.³⁵ However, the dependent variable, as mentioned above, changes from conflict initiation to latency in H2. As latency also serves as one of my independent variables, it is described in greater detail below.

Independent variables

As the first independent variable is latency, I use measures related to latency extracted from the NL data set.³⁶ In a directed dyad, it is the latency status of the challenger. It is dichotomous, and thus takes the value of either 0 or 1. Unlike earlier data sets that have relied on proxies to operationalize and measure latency, the NL data set directly captures enrichment and reprocessing (ENR) technology, which is necessary for producing fissile material (i.e. weapons-grade highly enriched uranium or plutonium).³⁷ Fuhrmann and Tkach (2015: 444, 446) stress that ENR technology is considered “the most important feature of nuclear latency” because it is “the most important indicator of a state’s potential to build nuclear weapons.” From 1939 to 2012, the NL data set identifies all of the years in which states in the international system operated ENR facilities. The NL data set provides separate measures for laboratory-scale and pilot-scale ENR facilities. In this regard, I consider a state to be latent as long as it operates the former. If a state operates at least laboratory-scale ENR facilities in a given year, it is coded 1 and 0 otherwise.³⁸ If a state gets rid of its ENR facilities, it is coded 0. However, if that particular state rebuilds and reoperates its ENR facilities, it is coded 1 again. Similarly, if a latent state crosses the line and turns into a nuclear state by acquiring nuclear weapons, it is coded 0. For whatever reason, if a nuclear state abandons its nuclear weapons and reverts to being latent, it is coded 1 again.³⁹

The second independent variable is personalism and it comes from Way and Weeks’ (2014) data.⁴⁰ Using an index of eight yes-no questions capturing “the extent to which the leader is free of constraints on his personal rule” (Way and Weeks, 2014: 712), they, based on the proportion of “yes” answers, create a dichotomous variable;⁴¹ a state is coded 1 (i.e. personalist) if its total index score goes over their somewhat arbitrary cutoff of .5 and 0 otherwise.⁴² According to Way and Weeks (2014), “very few regimes come close to this [.5] threshold” (p. 712). Instead, most regimes are either very close to 0 or very close to 1.

The third independent variable is an interaction term between latency and personalism. Here, latency is a dichotomous variable indicating State A’s latency status, whereas personalism is a continuous variable representing State A’s raw score on the personalist index mentioned above. The latter comes from Weeks’ (2014) data, and it ranges between 0 and 1. Undoubtedly, the advantage of employing the raw personalism scores is that they do not “force one to define a particular cutoff between ‘personalist’ and ‘nonpersonalist’” (Weeks, 2014: 47).

Control variables

In order to control for potential confounders, when testing H1 and H3, I add covariates identical to those in Fuhrmann and Tkach (2015).⁴³ When testing H2, I include covariates similar to those in Way and Weeks (2014).⁴⁴ The following is the description of control variables in the first data set. Table 1 reports descriptive statistics for some of the most important ones.

Table 1.

Descriptive statistics for key variables (first set).

Variable	N	Mean	Std. dev.	Min.	Max.
MID Initiation	1,310,330	0.002	0.043	0	1
Nuclear Latency A	1,386,702	0.057	0.232	0	1
Nuclear Latency B	1,386,702	0.057	0.232	0	1
Personalism Index A	364,217	0.545	0.404	0	1
Dyadic Rivalry	1,386,702	0.009	0.094	0	1
Contiguity	1,386,702	0.038	0.192	0	1
Dyadic Alliance	1,313,740	0.067	0.250	0	1
Dyadic Alliance Portfolio Similarity	1,079,328	0.666	0.289	−0.608	1
Cumulative Past Dyadic Militarized Conflicts	1,067,647	0.143	1.096	0	39

Nuclear Latency: Although I only control for the latency status of State B in H1, I control for both the latency status of State A and the latency status of State B in H3. As mentioned above, I use data from Fuhrmann and Tkach (2015).

Nuclear Weapons Pursuit: Fuhrmann and Tkach (2015) point out that the presence of nuclear weapons programs in either State A or State B could affect conflict initiation. Accordingly, following them, I account for State A’s and State B’s nuclear weapons programs. As both are dichotomous variables, a state is coded 1 if it has ongoing nuclear weapons programs in a given year and 0 otherwise.

Nuclear Weapons (Possession): Fuhrmann and Tkach (2015) also emphasize that the possession of nuclear weapons by either State A or State B could affect conflict initiation. Again, following them, I control for both the possession status of State A and the possession status of State B. Given that these are dichotomous variables, a state is coded 1 if it possesses nuclear weapons in a given year and 0 otherwise. At the same time, an interaction term between them is included.

Democracy: In order to control for the regime types of State A and State B, I employ the Polity2 variable of the Polity IV data (Marshall et al., 2010). Although it ranges from −10 to 10, following Gartzke and Jo (2009), it is rescaled to an 11-point measure ranging from 0 to 10. In this regard, Gartzke and Jo (2009) “take the difference between Polity democ and autoc variables, add ten, and then divide by two” (p. 219). I also add an interaction term between them.

Rivalry Status: According to Gartzke and Jo (2009), both monadic rivalry status and dyadic rivalry status could affect conflict initiation. While I employ data from Fuhrmann and Tkach (2015) to create a variable related to the former, they rely on data from Gartzke and Jo (2009). As it is a dichotomous variable, a state is coded 1 if it has any ongoing rivalry outside its dyad in a given year and 0 otherwise. Then, I use Klein et al.’s (2006) updated data set on international rivalries to generate a variable related to the latter. Again, it is a dichotomous variable; a dyad is coded 1 if the members of it are considered rivals with one another in a given year and 0 otherwise.

Contiguity and Distance: Neighbors are more likely to fight than non-neighbors (Bremer, 1992). In order to account for such possibility, I use version 3.0 of the Correlates of War (COW) direct contiguity data (Stinnett et al., 2002), which covers the time period from 1816 to 2000. Stinnett et al. (2002) provide an ordinal measure of contiguity, but I dichotomize it for the purpose of simplification. A dyad is coded 1 if the members of it are separated either by land or by less than 150 miles of water and 0 otherwise. Moreover, following Gartzke and Jo (2009), I add a distance variable, which is “coded as the log transformed great circle distance between capital cities of countries in a given directed dyad year” (p. 219).

Alliance: Allies are less likely to fight each other (Kimball, 2006). I use data from Fuhrmann and Tkach (2015) to construct this particular variable. It is a dichotomous variable. As such, a dyad is coded 1 if the members of it share an alliance in a given year and 0 otherwise. Following Weeks’ (2014) empirical procedures, I also control for not only dyadic alliance portfolio similarity, but also each member’s respective alliance portfolio similarity with the United States (i.e. the systemic leader).

Capabilities: As more powerful states are more likely to engage in conflict initiation than less powerful states (Weeks, 2014), I control for State A’s and State B’s respective military power. I employ the COW Composite Indicator of National Capabilities (CINC) data (Singer et al., 1972). Then, I include an interaction term between them.

Dispute History: Given that the presence of the dispute history in a dyad could affect conflict initiation (Bell and Miller, 2015), I control for it by adding a variable that captures the cumulative number of previous dyadic disputes. It comes from Bell and Miller’s (2015) data.

Trade Dependence: Finally, as the relationship between trade interdependence and the use of military force is widely known in scholarship, I follow Weeks (2014) and include a variable that measures the trade dependence of the less dependent state. I rely on data from Weeks (2014), but she uses data from Gleditsch (2002).

Now, the following is the description of control variables in the second data set. Table 2 reports their descriptive statistics.

Table 2.

Descriptive statistics for all variables (second set).

Variable	N	Mean	Std. dev.	Min.	Max.
Nuclear Latency	8,682	0.068	0.252	0	1
Personalism	5,659	0.201	0.401	0	1
Number of Land Borders	8,680	3.180	2.424	0	20
Population (ln)	8,616	8.726	1.814	2.565	14.097
Capabilities	8,617	0.006	0.100	−9	0.364
GDP per Capita (ln)	8,484	8.239	1.121	5.139	11.343

Security Environment: Existing scholarship suggests that states feeling insecurity are more likely to pursue nuclear weapons than those that do not. For the purpose of accounting for the security environment, I add a proxy that measures the number of shared land borders that a state has with other states. In this regard, I employ the COW direct contiguity data.

Population: In order to control for “the possibility that more populous countries are better able to marshal the resources necessary for a nuclear weapons program” (Way and Weeks, 2014: 713), I include a variable for population size that is in the form of natural log.

Capabilities: Given that the capabilities of a state could affect its decision whether to pursue nuclear weapons, I control for a state’s capabilities in a given year. I create this particular variable using the COW CINC data.

Gross Domestic Product (GDP) per Capita: Finally, more developed states are more likely to engage in nuclear weapons pursuit than those less developed. They most likely will have the ability to overcome various technological hurdles (Way and Weeks, 2014). Accordingly, I add a variable for GDP per capita, which is in the form of natural log.

Results

Table 3 presents the full regression results and shows the estimated relationship between the latency status of State A and the likelihood of State A initiating conflict against State B. Both Models 1 and 2 correspond with H1. In Model 1, my first independent variable, Nuclear Latency A, is in the expected direction and statistically significant, indicating that latent states are more likely to initiate military conflicts. Undoubtedly, this provides support for not only H1, but also the logic of emboldenment, which, as mentioned above, has at its core the firm beliefs of latent states in latent nuclear deterrence. It should be noted that my finding lends support to the “conflict-proneness side,” thereby further tipping the balance in its favor. In Model 1, Nuclear Latency B is negative and statistically significant. It means that the latency status of State B decreases the likelihood of State A initiating conflict against State B. Simply put, it demonstrates that latent nuclear deterrence works and that latent states’ firm beliefs in it are not unwarranted.

Table 3.

Logit analysis of State A’s conflict initiation.

	Model 1	Model 2 (FE)
Nuclear Latency A	0.299** (0.132)	0.227* (0.137)
Nuclear Latency B	−0.302** (0.139)	−0.534*** (0.160)
Nuclear Weapons Pursuit A	0.256* (0.155)	0.295** (0.149)
Nuclear Weapons Pursuit B	0.082 (0.170)	−0.013 (0.181)
Nuclear Weapons A	0.841*** (0.217)	0.460 (0.291)
Nuclear Weapons B	0.029 (0.225)	−0.788*** (0.247)
Nuclear Weapons A × Nuclear Weapons B	−0.263 (0.380)	0.614 (0.456)
Democracy A	0.037** (0.018)	0.036 (0.028)
Democracy B	0.106*** (0.017)	0.094*** (0.029)
Democracy A × Democracy B	−0.012*** (0.003)	−0.006 (0.004)
Rivalry Status A	0.894*** (0.109)	0.426*** (0.146)
Rivalry Status B	0.365*** (0.103)	0.180 (0.152)
Dyadic Rivalry	2.579*** (0.180)	2.22*** (0.166)
Contiguity	2.171*** (0.308)
Distance (ln)	−0.107*** (0.032)
Dyadic Alliance	0.175 (0.111)	−0.035 (0.191)
CINC A	−1.300 (2.107)	−12.524** (5.051)
CINC B	1.129 (2.433)	−6.792 (4.612)
CINC A × CINC B	−1.210 (30.882)	120.168 (91.841)
Alliance Portfolio Similarity with the US A	0.045 (0.175)	−0.603* (0.348)
Alliance Portfolio Similarity with the US B	−0.049 (0.175)	0.002 (0.331)
Dyadic Alliance Portfolio Similarity	−0.387** (0.182)	−0.378 (0.303)
Cumulative Past Dyadic Militarized Conflicts	−0.003 (0.010)	−0.068*** (0.018)
Lower Trade Dependence in Dyad	11.248* (6.418)	10.739 (7.086)
t	−0.075*** (0.008)	0.012 (0.010)
t2	0.001*** (0.000)	0.000 (0.000)
t3	0.000*** (0.000)	0.000 (0.000)
Constant	−7.379*** (0.325)
Observations	1,007,239	21,115

Robust standard errors in parentheses. Two control variables dropped in Model 2.

Statistical significance: ***p < 0.01; **p < 0.05; *p < 0.10.

In Model 2, I employ a more demanding fixed-effects analysis to control for “the possibility that omitted variables specific to directed dyads could be affecting the results” (Weeks, 2014: 46; cf. Beck and Katz, 2001). When estimating Model 2, I drop both Contiguity and Distance (ln), as these variables do not vary much within directed dyads. Even after employing fixed effects, Nuclear Latency A remains in the expected direction and statistically significant (although at the 0.10 level now).⁴⁵ At the same time, Nuclear Latency B continues to be negative and statistically significant. These results indicate that the effect of latency “is not due primarily to omitted variables that happen to be correlated with” latency (Weeks, 2014: 46).

In order to examine the substantive significance of these findings, I look at the effect of State A’s latency on the predicted probability of conflict initiation. Figure 1 shows what is known as an “average” simulation scenario (based on Model 1). With all other binary, continuous, and discrete control variables held either at their means or at their medians, when State A is not a latent state, the predicted probability of initiating conflict is about 0.00015. However, when State A turns into a latent state, the predicted probability increases to about 0.00021. It is a notable rise of about 40%. The effect of possessing latency, therefore, is substantively significant.

Figure 1.

Impact of State A’s nuclear latency on probability of conflict initiation.

Table 4 contains two regressions (Models 3 and 4) of the effects of personalism and other variables on the pursuit of latency, thereby corresponding with H2. In Model 3, my second independent variable, Personalism, is in the expected direction and statistically significant, indicating that states with personalist regimes are more likely to pursue latency. This provides support for H2. As mentioned above, this particular finding helps establish a generalizable pattern between personalism and latency. Moreover, it is logically compatible with Way and Weeks’ (2014) finding that states with personalist regimes are significantly more likely to pursue nuclear weapons than those with other types of regimes. Latency is most likely serving as a stepping stone to nuclear weapons.

Table 4.

Logit analysis of nuclear latency pursuit.

	Model 3	Model 4 (FE)
Personalism	1.608*** (0.477)	2.333*** (0.678)
Number of Land Borders	0.075 (0.052)
Population (ln)	0.886*** (0.112)	0.767 (0.859)
Capabilities	−25.085*** (4.710)	0.849 (34.466)
GDP per Capita (ln)	0.915*** (0.171)	0.053 (0.504)
t	−1.704*** (0.155)	−1.370*** (0.116)
t2	0.097*** (0.012)	0.075*** (0.009)
t3	−0.002*** (0.000)	−0.001*** (0.000)
Constant	−15.379*** (2.287)
Observations	5,501	1,406

Robust standard errors in parentheses. One control variable dropped in Model 4.

Statistical significance: ***p < 0.01; **p < 0.05; *p < 0.10.

All other covariates in Model 3 perform as expected, with the exception of Number of Land Borders, which is “a suitable pretreatment proxy for intensity of the security environment” of a state (Way and Weeks, 2014: 713). While Number of Land Borders is in the expected direction, it is statistically insignificant. This is somewhat surprising. Both Population (ln) and GDP per Capita (ln) are in the expected direction and statistically significant: more developed and populous states are more likely to pursue latency. Capabilities is also in the expected direction and highly significant, suggesting that weaker states are more likely to pursue latency.

Model 4 shows an analysis with fixed effects. I drop Number of Land Borders, as it does not vary much within countries. As can be seen, only Personalism remains highly significant in the expected direction. Population (ln) and GDP per Capita (ln) are no longer statistically significant. Capabilities is in the opposite direction and, at the same time, loses its statistical significance. Indeed, these results increase my confidence that the effect of personalism does not come from omitted variables that are correlated with personalism.⁴⁶

For the purpose of examining the substantive significance of these findings, I look at the effect of personalism on the predicted probability of latency pursuit. An “average” simulation scenario (based on Model 3) is shown in Figure 2. With both continuous and discrete control variables held at their means and medians, respectively, when a state has other type of regime (i.e. a nonpersonalist regime), the predicted probability of pursuing latency is about 0.053. On the contrary, when a state has a personalist regime, the predicted probability increases to about 0.099. It is a considerable rise of about 86.8%. Accordingly, the effect of personalism is substantively significant.

Figure 2.

Impact of personalism on probability of nuclear latency pursuit.

Table 5 presents two regressions (Models 5 and 6) that correspond with H3: they report the effects of an interaction term between State A’s latency and State A’s personalism index score on the likelihood of State A initiating conflict against State B. In Model 5, my third independent variable, Nuclear Latency A × Personalism Index A, is in the expected direction and statistically significant, indicating that latent states with personalist regimes are more likely to initiate military conflicts.⁴⁷ This provides support for H3, which is my main hypothesis.

Table 5.

Logit analysis of interaction between State A’s latency status and personalism index score.

	Model 5	Model 6 (FE)
Nuclear Latency A × Personalism Index A	0.900* (0.534)	1.275* (0.754)
Nuclear Latency A	−0.445 (0.439)	−0.624 (0.595)
Personalism Index A	0.087 (0.159)	0.298 (0.326)
Nuclear Latency B	−0.381* (0.211)	−0.519** (0.209)
Nuclear Weapons Pursuit A	−0.402* (0.219)	0.209 (0.274)
Nuclear Weapons Pursuit B	0.062 (0.267)	−0.281 (0.263)
Nuclear Weapons A	−0.762* (0.441)	0.200 (0.588)
Nuclear Weapons B	−0.165 (0.293)	−1.261*** (0.415)
Nuclear Weapons A × Nuclear Weapons B	0.390 (0.486)	1.612** (0.730)
Democracy A	0.028 (0.055)	0.033 (0.069)
Democracy B	0.104*** (0.025)	0.081* (0.042)
Democracy A × Democracy B	−0.002 (0.009)	0.010 (0.010)
Rivalry Status A	0.694*** (0.174)	0.142 (0.260)
Rivalry Status B	0.258 (0.161)	−0.176 (0.237)
Dyadic Rivalry	2.594*** (0.231)	2.474*** (0.245)
Contiguity	2.125*** (0.366)
Distance (ln)	−0.104*** (0.035)
Dyadic Alliance	0.242 (0.172)	−0.060 (0.306)
CINC A	8.422*** (3.135)	12.340 (14.806)
CINC B	9.836*** (3.006)	20.752 (16.398)
CINC A × CINC B	−87.055** (39.085)	17.401 (202.293)
Alliance Portfolio Similarity with the U.S. A	0.059 (0.315)	−1.319** (0.662)
Alliance Portfolio Similarity with the U.S. B	0.065 (0.273)	0.976* (0.585)
Dyadic Alliance Portfolio Similarity	−0.383 (0.333)	0.068 (0.657)
Cumulative Past Dyadic Militarized Conflicts	0.009 (0.014)	−0.127*** (0.025)
Lower Trade Dependence in Dyad	20.761*** (5.513)	28.084*** (9.523)
t	−0.063*** (0.012)	0.027 (0.020)
t2	0.001*** (0.000)	0.000 (0.001)
t3	−0.000* (0.000)	0.000 (0.000)
Constant	−7.294*** (0.464)
Observations	349,911	7,476

Robust standard errors in parentheses. Two control variables dropped in Model 6.

Statistical significance: ***p < 0.01; **p < 0.05; *p < 0.10.

Similar to Model 2, Model 6 shows an analysis with more demanding fixed effects.⁴⁸ Again, as both Contiguity and Distance (ln) do not vary much within directed dyads, I drop them when I estimate Model 6. Even with the inclusion of fixed effects, Nuclear Latency A × Personalism Index A continues to be statistically significant in the expected direction.⁴⁹ One thing should be noted from Models 5 and 6: Nuclear Latency B remains not only negative, but also statistically significant. Interestingly, it becomes more significant with fixed effects (by reaching the 0.05 level).

Figure 3 (based on Model 5) displays the marginal effects of State A’s latency on the likelihood of State A initiating conflict against State B across different scores of State A’s personalism index. Clearly, when State A’s score increases, the marginal effects of State A’s latency on State A’s conflict-proneness increase as well. Moreover, the interaction coefficient indicates that there is a positive interactive relationship between latency and personalism. However, there are two important caveats in this regard. First, only when State A’s score goes higher than about 0.66, the marginal effects become statistically significant (at the 0.10 level). Second, as shown in the histogram in the same figure, many of the observations cluster around 0 and 1.

Figure 3.

Impact of interaction between State A’s latency status and personalism index score on conflict initiation.

In sum, I find empirical support for all of my hypotheses. To start off, as H1 suggests, I find that latent states are more likely to initiate international conflict than non-latent states. In other words, the former are more conflict-prone—or belligerent—than the latter. Then, as H2 suggests, I find that states with personalist regimes are more likely to pursue latency than states with other types of regimes. Finally, as my main hypothesis (i.e. H3) suggests, I find that latent states with personalist regimes are more likely to initiate international conflict than latent states with other types of regimes. It should be noted that I report the results of robustness checks in the appendices; none of the results cause me to question my core findings.

Conclusion

Extant findings on the relationship between the possession of nuclear latency and the initiation of international conflict remain inconclusive. There are findings that suggest a positive relationship between nuclear latency and conflict initiation, but there are also findings that suggest no relationship between them. In this article, I argue and find that latent nuclear states that feature personalist authoritarian regimes are more likely to initiate international conflict than their counterparts that feature other types of regimes.⁵⁰ By demonstrating that personalism shapes the effect of nuclear latency on conflict initiation, my contribution is in that I successfully reconcile the aforementioned inconclusiveness.⁵¹ At the heart of my argument is the inadequacy of the logic of emboldenment as a standalone explanation for the conflict-proneness of latent nuclear states. Instead, what also has to be considered is the presence of a specific institutional setting that facilitates the emboldenment process. In personalist dictatorships, when already-hawkish personalist dictators are emboldened further by their firm belief in deterrence benefits that derive from possessing nuclear latency,⁵² no powerful domestic audiences—or institutionalized veto players—would be present to constrain them.⁵³ Under such circumstances, their frequency of conflict initiation is highly likely to increase.

In addition to my main finding above, I also find that nuclear latent states are more likely to initiate international conflict than non-nuclear latent states and that states with personalist authoritarian regimes are more likely to pursue nuclear latency than states with other types of regimes. These two findings, as discussed earlier, help me set the stage for the main finding. However, their contributions do not stop there. For example, the former finding contributes to the ongoing debate on the effect of nuclear latency on conflict initiation, specifically by tipping the balance further in favor of one side over the other, but future research should continue testing the relationship empirically not only with either different or more demanding specifications, but also with either new or refined measures that become available over time. The latter finding, on the contrary, contributes to both studies of personalism and nuclear latency, particularly by establishing a generalizable pattern between them for the first time.⁵⁴ Indeed, future research should focus on theorizing this particular empirical pattern further. For personalist dictators, nuclear latency could be either a means to an end or an end in itself.

When it comes to the main finding, there are three important avenues for future research. First, in both theory and empirics, I do not distinguish between personalist dictatorships that are considered “bosses” and personalist dictatorships that are considered “strongmen.” According to Weeks (2012, 2014), the former are civilian personalist regimes, whereas the latter are military personalist regimes. As such, “strongman” personalist dictators, contrary to their “boss” counterparts, have military experience (e.g. formal military training).⁵⁵ Future research should examine whether distinguishing them affects the main finding. If so, it would require further theorization. Second, the scope of the main finding does not go beyond conflict initiation. Future research, therefore, should investigate whether the main finding here could be extended to other relevant state behaviors. Third, future research should explore the exact timing of the emboldenment process. For instance, at what level of nuclear latency (i.e. with a particular focus on “medium” vs “high,” as both feature ENR technology) does the belief of personalist dictators in latent nuclear deterrence become most pronounced? In this regard, building on Volpe’s (2017, 2023) “fissile material sweet spot” framework would be helpful.⁵⁶

Finally, with respect to the US counterproliferation strategy, my main finding has two important policy implications.⁵⁷ First, to the extent possible, the United States should continue relying on nuclear cooperation agreements, also known as “123 agreements” (Volpe, 2023: 183), for managing ENR technology in other countries, particularly those that have personalist authoritarian regimes. However, standard 123 agreements should be prioritized over “Gold Standard” 123 agreements, as target countries are often more willing to accept the former than the latter. While “Gold Standard” 123 agreements completely ban the development of ENR technology, standard 123 agreements require instead strict adherence to “a long list of nonproliferation obligations, including a pledge to refrain from enriching or reprocessing US-origin nuclear material without prior consent from Washington” (Volpe, 2023: 183–184). These standard 123 agreements provide the United States with a firm legal basis for not only conducting stringent inspections, but also withholding critical nuclear materials and services when violations occur; in this way, the United States is “buying out” target countries’ ambitions to develop ENR technology (Volpe, 2023: 183). For example, as Riyadh recently has been quite hesitant to be deprived of its right to develop ENR technology (i.e. uranium enrichment), the United States should work with Saudi Arabia to sign a standard 123 agreement.⁵⁸ Second, in order to maintain the attractiveness of nuclear cooperation agreements mentioned above, the United States should beef up the American nuclear industry.⁵⁹ If American nuclear firms (e.g. Westinghouse) lose their competitiveness on the global market, target countries (e.g. Saudi Arabia) could turn their eyes to other major suppliers (e.g. China and Russia) that offer laxer nuclear cooperation agreements.

Footnotes

Appendix 1

Appendix 1 provides robustness checks for the results from Table 3.

Appendix 2

Appendix 2 provides robustness checks for the results from Table 4.

Appendix 3

Appendix 3 provides robustness checks for the results from Table 5. Note that with a p-value of 0.104, Nuclear Latency A × Personalism Index A in Model 1 barely misses being statistically significant at the 90% confidence level.

Acknowledgements

I would like to thank Bryce W. Reeder, Stephen L. Quackenbush, Heather-Leigh K. Ba, Sheena Chestnut Greitens, Edward Goldring, Dan Altman, Kyungwon Suh, Connor Kopchick, Gidong Kim, and Jinho Kim for their valuable guidance and feedback on earlier versions of this article. I would also like to thank the editors of Cooperation and Conflict and the anonymous reviewers for their helpful suggestions. All errors are mine alone.

Funding

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

ORCID iD

Yu Bin Kim

Notes

Author biography

Yu Bin Kim is a Ph.D. Candidate in Political Science at the Truman School of Government and Public Affairs at the University of Missouri and Graduate Fellow at MU’s Institute for Korean Studies. Starting in September 2024, he will be an Assistant Professor of Political Science and International Relations at Knox College, which is located in Galesburg, IL. His work focuses on both interstate and intrastate conflicts. Further information about his research is available on his personal website.

References

Angrist

Pischke

J-S

(2008) Mostly Harmless Econometrics: An Empiricist’s Companion (1st edn). Princeton, NJ: Princeton University Press.

Beck

Katz

(2001) Throwing out the baby with the bath water: A comment on Green, Kim, and Yoon. International Organization 55(2): 487–495.

Beck

Katz

Tucker

(1998) Taking time seriously: Time-series-cross-section analysis with a binary dependent variable. American Journal of Political Science 42(4): 1260–1288.

Bell

(2015) Beyond emboldenment: How acquiring nuclear weapons can change foreign policy. International Security 40(1): 87–119.

Bell

Miller

(2015) Questioning the effect of nuclear weapons on conflict. Journal of Conflict Resolution 59(1): 74–92.

Bowen

(2020) Why the United States Should Remain Engaged on Nuclear Power: Geopolitical and National Security Considerations. New York: Columbia University; SIPA: Center on Global Energy Policy.

Bremer

(1992) Dangerous dyads: Conditions affecting the likelihood of interstate war, 1816–1965. Journal of Conflict Resolution 36(2): 309–341.

Carter

Signorino

(2010) Back to the future: Modeling time dependence in binary data. Political Analysis 18(3): 271–292.

Cigar

(2011) Saddam Hussein’s Nuclear Vision: An Atomic Shield and Sword for Conquest. Quantico, VA: Marine Corps University Press.

10.

Colgan

(2010) Oil and revolutionary governments: Fuel for international conflict. International Organization 64(4): 661–694.

11.

Davenport

(2007) State repression and the tyrannical peace. Journal of Peace Research 44(4): 485–504.

12.

Frantz

Ezrow

(2009) “Yes men” and the likelihood of foreign policy mistakes across dictatorships. In: Presented at the annual meeting of the American political science association, Toronto, ON, Canada. Available at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=1450542

13.

Fuhrmann

(2017) The logic of latent nuclear deterrence. Available at: https://ssrn.com/abstract=3052231

14.

Fuhrmann

(2019) Explaining the proliferation of latent nuclear capabilities. In: Pilat

(ed.) Nuclear Latency and Hedging: Concepts, History, and Issues. Washington, DC: Woodrow Wilson Center Press, pp. 289–314.

15.

Fuhrmann

Horowitz

(2015) When leaders matter: Rebel experience and nuclear proliferation. The Journal of Politics 77(1): 72–87.

16.

Fuhrmann

Tkach

(2015) Almost nuclear: Introducing the nuclear latency dataset. Conflict Management and Peace Science 32(4): 443–461.

17.

Gartzke

D-J

(2009) Bargaining, nuclear proliferation, and interstate disputes. Journal of Conflict Resolution 53(2): 209–233.

18.

Gelman

Hill

(2006) Data Analysis Using Regression and Multilevel/Hierarchical Models (1st edn). Cambridge: Cambridge University Press.

19.

Ghosn

Palmer

Bremer

(2004) The MID3 data set, 1993–2001: Procedures, coding rules, and description. Conflict Management and Peace Science 21(2): 133–154.

20.

Gleditsch

(2002) Expanded trade and GDP data. Journal of Conflict Resolution 46(5): 712–724.

21.

Gleditsch

Salehyan

Schultz

(2008) Fighting at home, fighting abroad: How civil wars lead to international disputes. Journal of Conflict Resolution 52(4): 479–506.

22.

Gurr

(1988) War, revolution, and the growth of the coercive state. Comparative Political Studies 21(1): 45–65.

23.

Herzog

(2020) The nuclear fuel cycle and the proliferation “danger zone.” Journal for Peace and Nuclear Disarmament 3(1): 60–86.

24.

Hiim

(2022) Revisiting nuclear hedging: Ballistic missiles and the Iranian example. International Affairs 98(4): 1367–1384.

25.

Horowitz

(2013) Nuclear power and militarized conflict: Is there a link? In: Stulberg

Fuhrmann

(eds) The Nuclear Renaissance and International Security. Stanford, CA: Stanford University Press, pp. 288–312.

26.

Horowitz

Stam

(2014) How prior military experience influences the future militarized behavior of leaders. International Organization 68(3): 527–559.

27.

Jones

Mattiacci

Nordstrom

(2023) How leader’s type shapes the effect of nuclear latency on dispute involvement. Conflict Management and Peace Science 41: 177–193.

28.

Juneau

(2015) Squandered Opportunity: Neoclassical Realism and Iranian Foreign Policy. Stanford, CA: Stanford University Press.

29.

Kimball

(2006) Alliance formation and conflict initiation: The missing link. Journal of Peace Research 43(4): 371–389.

30.

Klein

Goertz

Diehl

(2006) The new rivalry dataset: Procedures and patterns. Journal of Peace Research 43(3): 331–348.

31.

Levite

(2003) Never say never again: Nuclear reversal revisited. International Security 27(3): 59–88.

32.

Marshall

Gurr

Jaggers

(2010) Polity IV Project: Political Regime Characteristics and Transitions, 1800–2009: Dataset Users’ Manual. Vienna, VA: Center for Systemic Peace.

33.

Mattes

Weeks

JLP

(2019) Hawks, doves, and peace: An experimental approach. American Journal of Political Science 63(1): 53–66.

34.

Mattiacci

Mehta

Whitlark

(2022) Atomic ambiguity: Event data evidence on nuclear latency and international cooperation. Journal of Conflict Resolution 66(2): 272–296.

35.

Mazarr

(1995) Virtual nuclear arsenals. Survival 37(3): 7–26.

36.

Mehta

Whitlark

(2017) The benefits and burdens of nuclear latency. International Studies Quarterly 61(3): 517–528.

37.

Meier

Vieluf

(2021) Upsetting the nuclear order: How the rise of nationalist populism increases nuclear dangers. The Nonproliferation Review 28(1–3): 13–35.

38.

Narang

(2013) What does it take to deter? Regional power nuclear postures and international conflict. Journal of Conflict Resolution 57(3): 478–508.

39.

Narang

(2014) Nuclear Strategy in the Modern Era: Regional Powers and International Conflict. Princeton, NJ: Princeton University Press.

40.

Narang

(2017) Strategies of nuclear proliferation: How states pursue the bomb. International Security 41(3): 110–150.

41.

O’Reilly

Murphy

(2020) A new measure of state capacity, 1789–2018. Available at: https://ssrn.com/abstract=3643637

42.

Persbo

(2019) Latent nuclear power, hedging, and irreversibility. In: Pilat

(ed.) Nuclear Latency and Hedging: Concepts, History, and Issues. Washington, DC: Woodrow Wilson Center Press, pp. 43–72.

43.

Reiss

(1988) Without the Bomb: The Politics of Nuclear Nonproliferation. New York: Columbia University Press.

44.

Sagan

(2010) Nuclear latency and nuclear proliferation. In: Potter

Mukhatzhanova

(eds) Forecasting Nuclear Proliferation in the 21st Century: The Role of Theory. Stanford, CA: Stanford University Press, pp. 80–101.

45.

Sechser

Fuhrmann

(2017) Nuclear Weapons and Coercive Diplomacy. Cambridge: Cambridge University Press.

46.

Singer

Bremer

Stuckey

(1972) Capability distribution, uncertainty, and major power war, 1820–1965. In: Russett

(ed.) Peace, War, and Numbers. Beverly Hills, CA: Sage, pp. 19–48.

47.

Spaniel

(2019) Bargaining Over the Bomb: The Successes and Failures of Nuclear Negotiations. Cambridge: Cambridge University Press.

48.

Stinnett

Tir

Diehl

, et al. (2002) The correlates of war (COW) project direct contiguity data, version 3.0. Conflict Management and Peace Science 19(2): 59–67.

49.

Volpe

(2017) Atomic leverage: Compellence with nuclear latency. Security Studies 26(3): 517–544.

50.

Volpe

(2023) Leveraging Latency: How the Weak Compel the Strong with Nuclear Technology. New York: Oxford University Press.

51.

Way

Weeks

JLP

(2014) Making it personal: Regime type and nuclear proliferation. American Journal of Political Science 58(3): 705–719.

52.

Weeks

(2008) Autocratic audience costs: Regime type and signaling resolve. International Organization 62(1): 35–64.

53.

Weeks

(2012) Strongmen and straw men: Authoritarian regime and the initiation of international conflict. American Political Science Review 106(2): 326–347.

54.

Weeks

JLP

(2014) Dictators at War and Peace. Ithaca, NY: Cornell University Press.