The ORE of Politics will be available for subscription in late September. Speak to your Oxford representative or contact us to find out more.

Dismiss
Show Summary Details

Page of

date: 16 August 2017

# Democratization and Conflict

## Summary and Keywords

Is democratization good for peace? The question of whether democratization results in violence has led to a spirited and productive debate in empirical conflict studies over the past two decades. The debate, sparked by Mansfield and Snyder’s foundational work, raised a challenge to the notion of a universal democratic peace and elicited numerous critical responses within the literature. One set of such responses has emphasized issues of replicability, mismatches between the research design and directionality of the proposed causal mechanism, the role of outliers, and model specification. In addition, two issues have not been discussed sufficiently in the existing literature. First, conceptually, is the issue of concept stretching, specifically the form Sartori labeled the “cat-dog” problem. While past criticisms were mainly about model specification, we debate whether Mansfield and Snyder’s findings can be seen as a product of concept misformation. Second, quantitatively, are conceptual and empirical issues that Mansfield and Snyder use to capture state strength in their most recent attempts to provide ongoing evidence for their theory. The most optimistic estimates show that even when democratization has a statistically significant association with war onset at lower levels of institutional strength, the effect is substantively insignificant.

# Introduction

The relationship between democratization and conflict has evolved into an important debate in empirical international relations because of an influential and sustained line of research by Edward Mansfield and Jack Snyder. They have argued that countries undergoing democratic transition are more prone to engage in several forms of conflict. This line of research began with articles published over two decades ago (Mansfield & Snyder, 1995a, 1995b), which argued that democratizing countries were more likely to engage in war, and culminated in the book Electing to Fight (Mansfield & Snyder, 2005). Excerpts from these works have been included in best-selling textbooks (Betts, 2012) and discussed in the popular press (Bass, 2006). In other articles, Mansfield and Snyder (2012a, 2012b) argued and provided evidence that the process of democratization promoted conflict, both within and between states, and that democratizing states were more prone to engage in militarized interstate disputes (MIDs) (2002b). Overall, their substantial output represents the major framing contribution for the literature on regime change and conflict, including civil wars.

In one sense their argument represents a challenge to the notion of a universal democratic peace by arguing that states undergoing democratic regime change do not behave like established democracies. Yet at the same time, once they capture the impact of democratization, their tests show that established democracies are still less likely to engage in conflict. This duality remains consistent with their theory as they argue that the democratic peace is a product of international behavior produced by states with stable institutions based on the rule of law and representative government, which many democratizing states do not yet have (2002b).

The essence of Mansfield and Snyder’s argument on the special properties of democratizing countries echoes the classic work by Huntington (1968) on modernization and stability. They claim that transitions to democracy increase political participation before necessary institutions of democratic control are fully in place. Under such circumstances, leaders substitute populist appeals in the face of weak institutions to exercise authority over the citizenry. Most common forms of populist ideology are nationalist in content and include the stigmatization of “enemies of the nation.” Elites will favor this form of control because it allows them to avoid full accountability to citizens (Mansfield & Snyder, 2002b, pp. 531–532). In their later work, Mansfield and Snyder (2002a, 2005) extend and reinforce their thesis by explicitly operationalizing the conditions under which they argue that democratization induces hawkish behavior by examining the impact of weak institutions in democratizing states. In this sense, dangerous democratization is a product of regime change and weak state institutions.

# Critical Responses

Next we turn to the responses of other researchers to Mansfield and Snyder. Over the years, many studies have been critical of their findings on interstate conflict. These criticisms tend to concentrate on issues of research design and data. Mansfield and Snyder have not only responded to the criticisms but have also argued that their argument on dangerous democratization continues to have implications for events such as the Arab Spring. We then discuss the democratization literature in intrastate conflict.

## Democratization and Interstate Conflict

Mansfield and Snyder have to be commended for sparking a debate on whether the study of regimes in international conflict goes beyond and attempts to modify the findings on the democratic peace. Their foundational work has contributed to an abiding interest in the effects of regime change on interstate conflict. For example, Clare (2007) considers the implications of prior democratic and autocratic spells when examining democratizers and reports some differentiation in dispute onset among them. Tan (2013) finds that, among rivals, regime change among autocratic, mixed, and democratic regimes results in reduced risk of dispute occurrence and that particularly the change from mixed to democratic regimes increases the time between the occurrence of disputes. Others have focused on the system level to understand the democratization–conflict relationship (Kadera, Crescenzi, & Shannon, 2003; Mitchell, Gates, & Hegre, 1999).

From the onset this line of research has provoked controversy and debate. The publication of an article in International Security provoked a spirited exchange of letters (Wolf, Weede, Enterline, Mansfield, & Snyder, 1996). Early on, Thompson and Tucker (1997, p. 450) closely examined the stability of results across different time frames and model specifications and ultimately concluded that “regime change and war involvement are independent.” More recently, the dangerous democracy argument has been criticized on the basis of its conceptualization and operationalization of democracy (Bogaards, 2010; McFaul, 2007).

One set of critical responses has emphasized issues of research design and replicability, directionality, and the role of outlying cases in producing findings (Daxecker, 2007; Enterline, 1996, p. 197; Gleditsch & Ward, 1997; Narang & Nelson, 2009; Thompson & Tucker, 1997, p. 450; Ward & Gleditsch, 1998). Enterline (1996, p. 196) was among the first to address the directional component of the analysis by using a directed dyad approach and presenting a finding that democratization has a “negative impact on the likelihood of a state being on the initiating side of a dispute.” Such concerns are echoed by Narang and Nelson (2009), who have criticized Mansfield and Snyder for having a theory of diversionary war that can only be verified by a directed-dyadic analysis.

A number of critics have also talked about how the statistically significant findings on the bellicose behavior of democratizers is driven by only a few special cases. Ward and Gleditsch (1998, 2000) argue that the result is driven by a few special sequences of events and find that democratizing countries are, on the contrary, less likely to be engaged in conflict, except for those that have particularly rocky transitions and those that lead to democratic reversals rather than consolidated democracies (Ward & Gleditsch, 1998). This is similar to Daxecker’s (2007) finding that autocratizing states as well as states that liberalize and revert to authoritarian rule promote interstate conflict. Once this is controlled for, she finds that states that democratize have a lesser propensity to engage in conflict. Narang and Nelson (2009) have pointed to the inordinate effect the dissolution of the Ottoman Empire had on the results produced by Mansfield and Snyder. Concerns have also been raised about endogeneity and reverse causality—that conflict’s effect on regime change may be stronger than the effect of democratization on conflict (Rasler & Thompson, 2004). Thus, the theory has been subjected to extensive criticism of the sort often encountered in contemporary political science.

In response, Mansfield and Snyder (2005) offered several brief case studies from different eras and regions that arguably support their thesis (we discuss these case studies further below). Moreover, they have argued that recent belligerents such as Ethiopia, Eritrea, and Pakistan “suffered significant institutional deficits in the period leading up to the outbreak of [their] wars” (Mansfield & Snyder, 2009, p. 383). And more recently, in considering the implications of their argument for the Arab Spring uprisings of 2011, Mansfield and Snyder (2012b, p. 731) have maintained that outside of “Egypt and Tunisia which have already made headway toward democracy,” weak administrative institutions and deep cleavages mean that democratization would be dangerous. They also note that other states, like Morocco and Jordan, might also have “some of the institutional preconditions” to avoid such a fate (Mansfield & Snyder, 2012b, p. 731). Yet five years later it is difficult to speak of democratization following the Arab Spring, let alone of interstate wars initiated by these countries. These reflections shift the causal underpinnings of their argument to include more than just democratization. Even if we leave aside the criticism that their theory now conflates state power and regime (McFaul, 2007), their choice of variable to measure institutional strength is potentially problematic as it relies upon Polity II’s domestic concentration measure (a variable that Polity has dropped in its recent releases).1

Beyond statistical analyses, case studies drawn from a variety of world regions (East Asia, Western Europe, the Caucasus, and the Middle East) maintain that democratization by itself fails to explain the outbreak of conflict (see Acharya, 2010; Lind, 2011; Miller, 2012) and that the causal logic specified by Mansfield and Snyder is simply not present (Montgomery & Pettyjohn, 2010; Siroky & Aprasidze, 2011, p. 528). In direct contravention of Mansfield and Snyder’s logic, Alcañiz (2012) argues that democratizing states have a strong incentive to avoid the kind of aggressive nationalist rhetoric that leads to aggressive international behavior because they are anxious to develop a positive international reputation. She also shows that they outpace both established democracies and autocracies in committing to multilateral security treaties.

Among the studies that challenge the causal logic specified by Mansfield and Snyder, some support is provided by Siroky and Aprasidze’s (2011) study of Georgian democratization where President Saakashvili’s promises of national unity may have contributed to Georgia’s war with Russia. However, the authors modify a crucial dimension of Mansfield and Snyder’s current argument such that the emphasis is not on the strength of institutions but rather on the type of institution (administrative or representative). Elsewhere, Montgomery and Pettyjohn’s (2010, p. 528) study on Palestinian democratization in the context of the 2006 conflict argues in contravention to Mansfield and Snyder that aggressive nationalist rhetoric was lacking and rather that “democratization significantly increased Israel's perception of threat by empowering radical non-state groups already hostile to Israel.” Other case studies have been much more skeptical. Miller (2012) finds that democratization by itself fails to explain the war-proneness in Germany and Iraq from World War I to the present. In East Asia, Acharya (2010) and Lind (2011) find that there is no case of a newly democratic country going to war and question Mansfield and Snyder’s argument on nationalism following democratization.

## Democratization and Intrastate Conflict

Building on the interstate democratic peace literature, there has been considerable interest in regimes and civil wars (Krain & Myers, 1997; Mason, 2003). In fact, Mansfield and Snyder (2012a) as well as Snyder (2000) have argued that a similar mechanism also operates for internal wars. In fact, democratization is a major component of interest in the civil war literature. For example, democratization emerges as a key dimension in the role of good governance discussions (Hegre & Nygard, 2015; Walter, 2015). Moreover, when the discussion turns to the role of international factors in civil wars, again democratization plays major role. For example, studies on peace-building after conflict have relied upon democratization in the post–civil-war era as a measure of success (Doyle & Sambanis, 2000). Others focusing on foreign aid have concluded that aid during civil wars from interstate rivals decreases the chances of democratization (Colaresi, 2014), but one specific form of foreign aid, democracy aid, has been reported to decrease the risk of a civil war in democratizing countries (Savun & Tirone, 2011). Overall, it is clear that democratization occupies a central position in civil war studies.

Mansfield and Snyder (2012a) report that complete democratization, autocratization, and stable regimes decrease the risk of civil wars through the examination of Polity data with two different civil war data sets. It is noteworthy that they do not explicitly include an incomplete democratization variable in their models but rather maintain that it is the reference category for the other main regime variables of interest that differs from the strategy they use in their interstate work. Most of the evidence for a potential domestic dangerous democratization argument has been provided by other authors who also relied on Polity to capture regime change. For instance, Hegre, Ellingsen, Gates, and Gleditsch (2001) report that the risk of civil wars is double in the year following a regime change. This would suggest that democratization is dangerous for domestic peace. However, this finding is subject to a number of qualifications. Cederman, Hug, and Krebs (2010) find that democratization increases the risk of conflicts over government but not over territory or separatism (see also Sunde & Cervellati, 2014). Moreover, Hegre et al. (2001, p. 42) report that two years after democratization, the effect on civil war onset is negligible and that “the risk of civil war is increased the most by changes that lead to a semi-democracy, in particular if the shift is a large autocratization”; this suggests that transitions away from democracy have the greatest impact. Fearon and Laitin (2003) also reported that not democratization but moves away from democracy increase the risk of civil wars. These two findings suggest that the effect of regime change depends on the type of conflict as well as on time. While there have been calls for using a greater variety of measures and for further disentangling the various relationships (Arı, Gleditsch, Hegre, & Wig, 2016; Gleditsch & Hegre, 2014), it is difficult to say that this has yet been fulfilled.

The finding that there is an inverted U-shaped relationship between regimes (using the Polity data set) and civil wars is also pertinent to the debate here (Gleditsch, Hegre, & Strand, 2009; Hegre et al., 2001; Muller & Weede, 1990) as semi-democracies or anocracies are seen “often, but not always, in transition and more likely to have recent institutions under pressure for further political change” (Arı et al., 2016, p. 4). As we discuss further below, not only is labeling regimes “incoherent” problematic, but it is difficult to think of many of the countries in this range as being in a state of transition. It has also been shown that fractionalized competition in the Polity Index captures violent conflict, and when this is controlled for, the curvilinear relationship disappears (Vreeland, 2008).

A wider literature has incorporated how actors, i.e., political parties (Birnir, 2006; Manning & Smith, 2016; Söderberg Kovacs & Hatz, 2016) or voters (Letsa, 2017), the process and outcome of elections, particularly first elections (Collier, Hoeffler, & Söderbom, 2008; Flores & Nooruddin, 2012), influence civil war (recurrence). Brancati and Snyder (2011) show that there is a higher risk of violence during the first democratic election, but such risks disappear in later elections. Cederman, Gleditsch, and Hug (2013) find that elections will result in civil wars in newly democratic countries if there is a history of ethnic division and exclusion. Moreover, work in this vein has continued to examine the question of subnational electoral violence (Fjelde & Höglund, 2016; Wilkinson, 2006) and is likely to lead to further explorations in the future.

The related post–civil war (Joshi, 2010; Leonard, 2004) democratization literature has reported that the conditions of democratization are not tied to the outcome of civil wars but rather to a similar set of factors seen in non-conflict countries (Fortna & Huang, 2012). This is disputed by others who have reported that military victory diminishes the prospects for democratization but that negotiated settlements increase the level of democracy (Gurses & Mason, 2008; Nilsson, 2012) and that it is linked to power distributions (Joshi, 2010). One way to enhance the chances for democratization appears to be through post-conflict power sharing (Hartzell & Hoddie, 2015), but debates on the level of democracy and the effectiveness of power-sharing arrangements remain (Hippler, 2008; Jarstad & Sisk, 2008; Jung, 2012; Strasheim & Fjelde, 2014).

Having to this point summarized Mansfield and Snyder’s work on the potential dangers of democratization and the critical responses to their theory and findings, we turn to two additional areas that have not received much in-depth attention in the critical responses to their interstate conflict work, one qualitative and one quantitative. The first concerns the conceptualization and measurement of regime change in their work. Here we raise the issue of whether their operationalization creates a classic problem in concept misformation, the “cat-dog problem” (Sartori, 1991). Second, the critical literature has yet to fully address Mansfield and Snyder’s contention on weak institutions. Using their most recent work we replicate their results and closely analyze the marginal effects from the interaction of democratization and institutional strength.

# Concept Misformation?

While Mansfield and Snyder have sparked a vibrant debate, we believe that the potential problems with their argument go beyond issues of model specification and sample. A number of authors have already raised the issue of their operationalization of democratization (Bogaards, 2010; Cederman et al., 2010; Daxecker, 2007; McFaul, 2007; Thompson & Tucker, 1997). Following these reservations, we dug deeper into the data to explore the way in which democratization is operationalized in conflict studies (Bernhard, Örsün, & Bayer, 2017). Here we build on this by inquiring whether their findings also could be a product of a cat-dog problem in operationalization—specifically a failure to capture processes of democratization by agglomerating several processes of regime change into their measure.

The cat-dog problem occurs when two different phenomena are conflated into one. Such errors can lead to one of two problems. The first is a false finding for the null hypotheses. In Sartori’s example (1991), the humorous hypothesis “all cat-dogs go bow-wow” is falsified when the doctoral candidate finds that a high percentage of them say “meow.” The second is when conflation of two phenomena provides enough evidence to confirm that the conflated category is consistent with the hypothesis. It is on the second set of grounds that Mansfield and Snyder are potentially vulnerable with regard to how they operationalize different forms of regime change. Mansfield and Snyder observe whether a transition from one regime type to another has occurred in year t by looking at the Polity score at t - 1 and t - 6 to see if a state has moved from one regime range to another.2 Regimes with scores below -6 on the joint scale are labeled “coherent autocracies” and those above +6 as “coherent democracies.” States falling between these two are seen as incoherent regimes. This allows Mansfield and Snyder to test the impact of several different varieties of regime transition.

Their notion of democratization includes two different processes: from coherent autocracy to an incoherent regime, and in turn then to a coherent democracy. In the first stage, “old elites threatened by the transition may still be powerful and the institutions needed to regulate mass political participation tend to be very weak” whereas in the second stage “a system of unfettered political competition and full governmental accountability to a broad electorate” is finally institutionalized (Mansfield & Snyder, 2005, p. 70). They maintain that because institutions are generally weak until the second transition is complete, violence is a likely outcome during democratic transitions. But this logic betrays a teleological bias (Carothers, 2002). Positive movement on the Polity scale does not mean that a country has achieved or will achieve democracy. Sartori (1991, pp. 248–249) would criticize this as “degreeism,” conflation of movement along a scale with change in kind. Thus, if democratization is a process that ends in a democratic regime, then it is hasty to judge any upward movement on Polity as democratization unless that crosses the threshold for democracy.

We have reservations about Mansfield and Snyder’s findings because their conceptualization of transition seems driven by a convenient measure rather than careful thought about what the concept means and how to effectively capture it by measurement. Research should move from theory to inferential validation by taking concepts embedded in the theory and operationalizing them as measures (Goertz, 2005). Concept misformation occurs when the measure fails to grasp the referent and the resultant inference is made on the basis of something else. This pitfall leads to the potential cat-dog problem: Mansfield and Snyder have a theory about cats (democratization) but a measure that captures cat-dogs (democratization and authoritarian regime change).

So why do we think that the operationalization of democratization in this literature could be a cat-dog? To start, the very concept of regime in comparative politics is typological. One needs only look at the definition of the various types of regimes to understand that they are defined in a multidimensional way. There is no one thing we can point to and say that makes a country democratic. If we take Dahl’s (1971) notion of polyarchy as a starting point, its most abstract formulation combines two dimensions: contestation and participation. Similarly, Linz and Stepan’s (1996) definitions of totalitarianism, post-totalitarianism, authoritarianism, Sultanism, and democracy are distinguished along four dimensions. More recent codings of dictatorship are also typological (Geddes, Wright, & Frantz, 2014; Wahman, Teorell, & Hadenius, 2013).

Polity was originally created to understand patterns of authority across societies (Eckstein, 1973). It is composed of six different component parts that are simply added together and weighted on the size of their scales with no justification of this as an aggregation procedure (Gleditsch & Ward, 1997; Goertz, 2005; Munck & Verkuilan, 2002; Trier & Jackman, 2008; Vreeland, 2008). In practice, the combined score does a good job of capturing polities at the top end of the scale. These scores are comparable to other democratization measures.

However, in the middle and bottom range of the scale, Polity does a poor job of capturing the conventional ways regimes are conceptualized in the comparative politics literature. Coherent autocracies include both totalitarian regimes like Stalinist Russia (-9 on Polity) and Nazi Germany (-9), kleptocratic personalist dictatorships like Mobutu’s Zaire (-9) and Amin’s Uganda (-7), as well as contemporary sultanistic regimes such as Saudi Arabia (-10). The incoherent part of the scale includes constitutional monarchies such as Wilhelmine Germany (-4 to 2), the competitive oligarchies that preceded democracy in Western Europe (e.g., Netherlands -4 to -2 from 1848 to 1916), authoritarian developmental states like Singapore (-2), and contemporary monarchies like Jordan (-4 to -2 since 1988). It also includes new post–Cold War regime types such as electoral authoritarian regimes (Schedler, 2006) like Mubarak’s Egypt (-6 to -3), where elections have almost no chance of displacing autocrats, and competitive authoritarian regimes (Levitsky & Way, 2010), where the prospects of incumbents losing power is small but plausible, such as Putin’s first two terms in Russia (2000–2008, range of 4 to 6). Immediate problems suggest themselves in the terminology of coherence and incoherence. Were Germany under the Kaiser or Singapore under Lee Kwan Yew incoherent forms of rule? Was Amin’s rule in Uganda coherent?

Even more importantly, if these ranges include several types of historically distinct kinds of regimes, how can a scale that purports to capture democracy as a single dimension capture regime change as a process? Why is the way in which the Polity measures democracy prone to degreeism? It reduces the six component parts of the measure into a single scale. It ignores the dimensionality of a concept better understood as a type and folds its dimensions (six nominal scales) into a unidimensional scale through an additive process. This turns differences in kind into differences of degree and, as the examples above illustrate, addles the very distinction of types it is supposed to measure.

If we understand regime change in line with mainstream comparative politics, it is a process by which one form of identifiable rule replaces another. There are several forms of regime change that Mansfield and Snyder’s use of Polity would fail to correctly characterize. Consider the following hypothetical situation. The Saudi monarchy comes to be replaced by a fundamentalist Salafist regime. This would be the replacement of one form of non-democratic rule with another. If it occurred quickly it might not even register in Polity as a regime change. If there was a period of instability and as a result of reduced repression the Polity score increased from -10 to some score in the middle range before the full institutionalization of Salafist rule, then this would be coded as democratization.

Or we could change the names in this example from Saud to Pahlevi, and Salafist to Shia’ fundamentalist (moving from -10 under the Shah to 0 from 1978 to 1979). Neither the hypothetical example nor the real one is recognizable to comparative politics scholars as democratization. This is the breakdown of one form of authoritarian rule and its replacement by another. Yet Iran is highlighted by Mansfield and Snyder (2002b, p. 4) as just such a case of a country “beginning transitions towards democracy … that experienced military disputes.” Was it beginning a democratic transition or a neo-authoritarian one? If where it ended up matters, then the answer is clear.

Bogaards (2010, p. 484) points out several other problematic examples of “democratizers” who engaged in interstate war in his critical discussion of Mansfield and Snyder, including Cambodia and China in 1979. Cambodia moves from a -5 under the last years of the Lon Nol military dictatorship to a zero in the interregnum year of 1975 and then drops to a -7 under the subsequent period of Khmer Rouge rule. It is difficult to weave any plausible story that would construe this episode as a process of democratization, though it obviously is regime change from one very nasty dictatorship to an even nastier one.

The potential problem with their design is that many “coherent autocracies” that move into the range between -5 to +6 fail to become democratic in any recognizable sense. Mansfield and Snyder’s operationalization captures not only countries that eventually become coherent democracies but a large number of countries that make a transition from one form of authoritarianism to another. Thus their operationalization sometimes misses the mark; it tests whether the failure of established authoritarian regimes and their replacement either by neo-authoritarianism or democracy (artificially consolidated into one via concept misformation) has a statistically significant relationship with various forms of conflict.

This illustrates how degreeism can lead to poor operationalization of concepts—it conflates movement on a scale with a process of change to which it may or may not correspond. Take the example of the contemporary Jordanian monarchy—the king introduced reforms in 1988, including liberalization of the party system and elections, which moved the combined score from -9 to -4. While Polity would characterize this as regime change, it was continuity of regime with liberalization. Liberalization (reduction of repression and control), as Przeworski (1991) and subsequently many others have pointed out, is a tactic used frequently by authoritarian incumbents to maintain and strengthen their rule (Brownlee, 2007; Levitsky & Way, 2010; Magaloni, 2008). When it succeeds, Przeworski (1991, p. 61) refers to the outcome as “broadened dictatorship.” It makes no sense to categorize processes of authoritarian reform and continuity when they cross a particular threshold in Polity as democratization.

Thus far we have outlined two ways in which the operationalization chosen by Mansfield and Snyder includes events that should not be seen as democratization. First, it can conflate neo-authoritarian transitions, the change from one authoritarian regime to another, with democratization if that change results in a movement from one range on the scale to another. This is particularly true of movement from the coherent authoritarian range to incoherent regimes. Mansfield and Snyder (2005, p. 77) themselves are critical of Polity for the lack of justification for the various thresholds used in differentiating between coherent democracies, incoherent regimes, and coherent autocracies. Second, it is not always clear that movement across these thresholds corresponds to regime change. In particular, a movement of one or a few points across a threshold may not represent regime change but merely a change in policy of an existing regime, particularly in the move from a -7 to a nearby negative value on the Polity scale.

Finally, when one looks at the Polity scores of the participants of the MIDs that Mansfield and Snyder characterize as involving democratizing states, other conceptual issues arise. A number of MIDs involve countries that are experiencing regime instability, and this is reflected in Polity scores that frequently move around in very short periods of time yet produce little enduring democratic regime change. The question remains whether it is correct to characterize such states as experiencing democratization. Cederman, Hug, and Krebs (2010) explicitly model regime change to differentiate periods of minor fluctuation from major change in their exploration of the effects of regime change on civil war. Periods where Polity’s coding repeatedly crosses thresholds on the basis of small changes may be capturing the instability or breakdown of authority between dictatorships or prolonged periods of state weakness or even failure. One such example is afforded by the MID between Russia and Afghanistan in 1993. According to Mansfield and Snyder’s coding, Russia experienced incomplete democratization when its polity score moved from -7 in 1988 to -4 in 1989, then complete democratization when it moved from 0 in 1991 to 6 in 1992, and finally an incomplete authoritarian transition when it moved to a 4 in 1993.

A similar issue is raised by the MIDs provoked by border tensions between Tanzania and Burundi in 1995 and 1997. Burundi according to Polity moved from a coherent autocracy (-7) to an incoherent polity in 1992 (-3), when presidential elections were held. The newly elected president, Melchior Ndadaye, served for barely three months before he was assassinated. This led to a 12-year period of genocidal civil war in which an estimated 300,000 people perished. During this period the Polity score changes eight times but only transitions from incoherent regime to democracy (6) in 2005.

Similarly, Mansfield and Snyder’s (2005) mini-case studies point to general problems in identifying cases that provide face validity to their theoretical contention that regime change promotes conflict. They are motivated by the desire to verify the underlying causal mechanism by process tracing. However, most of the cases (14 out of 20), in fact, are chosen based on only one of the dimensions of the Polity score, such as executive recruitment, rather than on a change in the composite score. While a change on executive recruitment might have an effect on conflict, this requires a separate theory to link such change to it. This is but one dimension of the composite score which Mansfield and Snyder proxy regime change. It is unclear whether changes on this scale alone constitute a form of regime change. Bismarckian Prussia, Guatemala in 1885 or 1906, or Peru in 1995, despite remaining constant or regressing in the composite score, are put forward as cases of incomplete democratization due to an improved score on one dimension. This is highly problematic because while there might be improvement in one dimension, in any of the other dimensions there can be regression or several movements in a short period, which suggests regime instability, not wholesale regime change. Moreover, we did not find support in the Polity codings for the dangerous democratization argument for 8 out of 20 wars even when we focused on the component dimensions instead of the composite score: Argentina (1982), Armenia (1994), Azerbaijan (1994), Ecuador (1995), Eritrea (1998), India (1999), Pakistan (1999), and Rwanda (1990–1995). Furthermore, in several other cases Polity reports missing values for the countries of interest, and again we believe that such cases can be considered as examples of instability rather than coherent change, e.g., Burundi (1990–1995), Ethiopia (1998), Serbia (1999), and Thailand (1925–1940).

Having now summarized our concerns about the potential mismatch between the concept and the measurement Mansfield and Snyder use, we turn to investigating the most recent test of their theory, explicitly linking regime change and institutional strength.

# Does Institutional Weakness Matter?

The last issue we address is the most recent iteration of Mansfield and Snyder’s attempt to buttress their findings by incorporating institutional weakness, a dimension not measured in their earliest efforts. In an additional series of works, Mansfield and Snyder (2002a, 2005) extend and further refine their original thesis by identifying the scope conditions under which democratization induces conflict.

They posit that incomplete democratization leads to a heightened risk of war especially when governmental institutions are weak. They test this by interacting regime change with the domestic concentration variable from Polity II (Gurr, Jaggers, & Moore, 1989), which they treat as a proxy for weak institutions. They report that under conditions of weaker institutions, transitions from an autocratic to a partly democratic regime lead to an 11.3% increase in the probability of war onset (domestic concentration = 2). The effect decreases to 1% if the institutions are stronger (domestic concentration = 7) (Mansfield & Snyder, 2005, p. 112).

Beyond Braumoeller’s (2004) critical study on the interpretation of interactions, other responses to Mansfield and Snyder have not taken into consideration their institutional strength argument. Here we raise two issues, one conceptual and one empirical. With regard to the former, Mansfield and Snyder treat the domestic concentration variable as a black box, as an independent measure of the strength of institutions. It too is an index ranging from 0 to 10, composed of six Polity subcomponents. However, domestic concentration is compiled from many of the same individual components used in Polity itself—four of the six components overlap, including “regulation of participation” (regulated +2, restricted +1), “regulation of executive recruitment” (regulated +1), “competitiveness of executive recruitment” (selection +1, election +1), and “constraints on the chief executive” (none +3, intermediate category +2, slight to moderate limits +1). Two additional components are added—“monocraticism” defined as a “pure individual executive” (+1), and “unitary state” (as opposed to federal, +2) (Gurr, Jaggers, & Moore, 1989). Like Polity, this measure is overwhelmingly composed of attributes of the executive (Gleditsch & Ward, 1997). In this case it is 6 of the possible 10 points.

In terms of the research design, we see two problems. First, how does one effectively interact terms that are in large part drawn from the same components? Certainly this needs to be carefully considered and justified. To their credit, Mansfield and Snyder did run correlations between domestic concentration and their regime change variables. They find moderate statistical correlations, but admit that “[w]hile we treat domestic concentration and democratization as independent, it is obviously possible that they are related. It is beyond the scope of this article to conduct a detailed analysis …” (Mansfield & Snyder, 2002a, p. 316).

Second, we believe that there is an important issue that should be addressed, one foreshadowed by the authors of Polity itself: “While the highest concentrations of institutional power are to be found in highly autocratic polities, high power concentrations are not uncommon among modern democracies” (Gurr, Jaggers, & Moore, 1989, p. 40). Past research demonstrates that countries which are very low or very high on the Polity scale are very unlikely to engage in conflict (Bennet, 2006; Peceny, Beer, & Sanchez-Terry, 2002). Neither of these two groups of states is likely to have low values on the domestic concentration index. Thus, the concentration index at its lowest levels should be likely to capture conflict-prone states, making this a test where it should be easy to find confirming results. To convince ourselves that this was the case, we took the absolute value of the original Polity composite index to rank country years according to coherence (recall that Polity considers the high and low ends of the scale to be coherent regimes whether democratic and autocratic) and correlated this to the domestic concentration index. The value of the pairwise correlation was 0.42, much higher than the low correlations reported by Mansfield and Snyder on their transition variables.

Given that this interaction should be conducive to confirmatory results, Mansfield and Snyder’s findings are not unexpected, but there are important empirical concerns that call their findings into question. Braumoeller (2004) has shown that Mansfield and Snyder’s (2002a) first finding on the domestic concentration interaction was questionable because of a misinterpretation of their multiplicative interaction term. Mansfield and Snyder (2005) have published additional analyses in response, so we precisely replicate their main empirical analyses on wars (Model 1) from which they draw their substantive inferences. They conclude that “the likelihood of interstate war rises markedly if either state in a dyad undergoes an incomplete democratic transition when its political institutions are weak … [T]he relative risk of interstate war due to incomplete democratization increases by roughly four-fold to forty-five-fold when domestic concentration dips from 5 to 2” (Mansfield and Snyder, 2005, pp. 130, 155). We replicate their analysis of the effect of the regime transitions on the outbreak of a war between two states, calculating the effect of incomplete democratization at different values of domestic concentration with two quantities of interest: relative risk (RR) as adopted in their original study and first difference (FD) including the uncertainty around these estimates.3 Mansfield and Snyder estimate the RR of war given domestic concentration with a value of n and find very dramatic estimates using the following quantity of interest:

$Display mathematics$

Here $E(YnB)$ is the expected probability of war onset when there is no regime transition (B) at a domestic concentration level of (n), and $E(YnIDT)$ is the expected probability of war onset at the same level of domestic concentration (n) but under conditions where one or both of the states in the dyad have undergone an incomplete democratic transition (IDT). One important consideration, independent of the relative risk ratio, is the magnitude of $E(YnIDT)$ and $E(YnB)$ as well as the first difference ($FDn=E(YnIDT)−E(YnB)$). These are important to calculate because the relative risk ($RRn$) approaches infinity as $E(YnB)$ approaches zero, and Mansfield and Snyder (2005) do not provide information on the baseline expected probability $E(YnB)$ or the first difference. In addition, they do not examine the uncertainty around these estimates.

This raises concerns about whether their main conclusion is substantively and/or statistically significant. We thus examine both the statistical and substantive significance of incomplete democratization on the outbreak of war and calculate both the relative risk ratio as reported by Mansfield and Snyder and the first differences and the uncertainty around these estimates through simulation (King, Tomz, & Wittenberg, 2000). Denoting the simulated $Y..$ as $Y˜..$, we estimate the following two quantities of interest:

$Display mathematics$

$Display mathematics$

and repeat this process 1,000 times for each simulated coefficient to approximate the distribution of relative risks and first differences. The results of the empirical analysis are presented in Figure 1, where we show the relative risk of war as well as the first difference for the effect of incomplete democratization for side i and side j, respectively. We find that when institutions are very weak (domestic concentration = 0), the incomplete democratization side i leads to a 48-fold increase in the relative risk of war using the RR framework.4 However, even if $RR˜0=48.8$ is within the 95% confidence interval of 3.4 and 216.2, the corresponding first difference estimate is 0.0004 within the 95% confidence interval of 0.00002 and 0.002.5 Hence, the effect of incomplete democratization, despite its statistical significance across some range of values of domestic concentration, is so small even within their own sample that it does not matter at all even when the institutions are very weak.6

Click to view larger

Figure 1 Incomplete Democratization: First Differences vs. Relative Risk

Table 1 Institutional Strength and Dangerous Democratizationa

Original Model 1

Re-Logit Model 2

β‎

SE

β‎

SE

Incomplete Democratizationi

3.30

(1.07)

***

2.86

(0.87)

***

Incomplete Democratizationj

4.44

(1.67)

***

4.45

(1.53)

***

Complete Autocratizationi

-13.49

(1.92)

***

-7.63

(4.14)

*

Complete Autocratizationj

3.55

(0.84)

***

3.83

(0.93)

***

Domestic Concentrationi

0.08

(0.06)

0.07

(0.07)

Domestic Concentrationj

0.16

(0.07)

**

0.19

(0.08)

**

Incomplete Democratizationi× Domestic Concentrationi

-0.58

(0.24)

**

-0.50

(0.18)

***

Incomplete Democratizationj× Domestic Concentrationj

-0.94

(0.45)

**

-0.88

(0.43)

**

Complete Autocratizationi× Domestic Concentrationi

1.72

(0.23)

***

1.18

(0.57)

**

Complete Autocratizationj× Domestic Concentrationj

-0.57

(0.10)

***

-0.56

(0.12)

***

Democracyi

0.05

(0.02)

**

0.05

(0.02)

**

Democracyj

-0.09

(0.03)

***

-0.07

(0.04)

**

Intercept

-7.02

(0.89)

***

-6.99

(1.12)

***

Observations

321,255

15,220

a Robust standard errors are presented in parentheses. Control variables, peace years, and splines are not presented to save space. Model 1 duplicates Mansfield and Snyder (2005) Table 6.2, Model 1.

*p < 0.10,

**p < 0.05,

***p < 0.01.

A potential caveat is that the ratio of the war to non-war cases could mean that zero-inflation in the data biases the location of the cut-off for distinguishing $Yi=1|x$ and $Yi=0|x$, thereby potentially leading to these small marginal effects (King & Zeng, 2001). Because the bias in the predicted probabilities increases proportional to the rarity of the event, we employ an endogenous choice-based sample of all dyads, where we include all the war cases and draw a random sample of peaceful dyads and drop 95% of all non-war cases.7 Then we apply the prior correction to the estimated model intercept as suggested by King and Zeng (2001). The results of this analysis are shown in Table 1, Model 2, and in Figure 2. This bias correction yields even smaller point estimates for both side i and side j than the original estimates. The estimated marginal effect of incomplete democratization side i is significant when domestic concentration is smaller than 5, where $FD˜0$ reduces to 0.0001 with a 95% confidence interval of 0.00001 and 0.0008. The first difference estimates for side j are also similar.

Click to view larger

Figure 2 Zero-Inflation Corrected Estimates

Even in a scenario highly conducive to confirmatory results, we do not find a substantively significant effect of incomplete democratic transition on conflict behavior. These findings indicate that the dangerous democratization hypothesis does not stand up to basic statistical scrutiny even if we accept the potentially problematic conceptual and statistical choices that are made to confirm it.

# Conclusion

Mansfield and Snyder raised an important question about the democratic peace: Were the effects of democratic regime change different from the impact of well-institutionalized democracy on conflict behavior? This provoked a long, spirited, and productive debate on the question that they raised, a debate on the effect of regime change on conflict that continues productively to this day (Daxecker, 2011). Given the centrality of democratization to other discussions in international relations, such as on the role of public, legislative, executive, and other constraints as well as the role of democratization in peacemaking, the continued attention to it is certainly understandable and necessary (Reiter & Tillman, 2002; Ripsman, 2012).

However, strong reservations about Mansfield and Snyder’s findings were raised on the grounds that they could be a product of sample bias, operationalization, reverse causality, and model specification. In this spirit, our last section here also questions the evidence they provide in support of their latest argument concerning incomplete democratization and institutional strength. We also raise questions about the operationalization of their key independent variable, regime change, by delving deeply into the data they use, and raise another potentially overlooked problem, that their operationalization creates a cat-dog, conflating authoritarian regime replacement and authoritarian liberalization with democratization.

More basically and, perhaps, importantly, our results highlight the importance of keeping in mind basic social science principles in research. While scholars might have an interesting theoretical idea, they must be cautious in operationalizing their concepts using “off-the-shelf” data. If nothing else, the first step is to cross-check findings with alternate measurements, samples, and model specifications where possible (Bayer & Bernhard, 2010). In other work, we have explored seven different operationalizations of regime change based on three different conceptualizations of democracy (Bernhard et al., 2017). We maintain that whether we conceptualize democracy as a bounded whole, as a property, or as a hybrid has implications for understanding democratization. We show that Mansfield and Snyder offer one modeling strategy for democratization but that it is the not only possible one. Overall, our examination of different modeling strategies with various data sets, including new and innovative democracy data sets, did not lend support to the dangerous democratization thesis.

## References

Acharya, A. (2010). Democracy or death? Will democratisation bring greater regional instability to East Asia? The Pacific Review, 23(3), 335–358.Find this resource:

Alcañiz, I. (2012). Democratization and multilateral security. World Politics, 64(2), 306–340.Find this resource:

Arı, B., Gleditsch, K. S., Hegre, H., & Wig, T. (2016). Democratization and civil conflict. APSA Comparative Democratization, 14(2), 1, 4–7.Find this resource:

Bass, G. J. (2006, January 1). Are democracies really more peaceful? The New York Times. Retrieved from http://www.nytimes.com/2006/01/01/magazine/are-democracies-really-more-peaceful.htmlFind this resource:

Bayer, R., & Bernhard, M. (2010). The operationalization of democracy and the strength of the democratic peace: A test of the relative utility of scalar and dichotomous measures. Conflict Management and Peace Science, 27(1), 85–101.Find this resource:

Bennet, S. (2006). Toward a continuous specification of the democracy–autocracy connection. International Studies Quarterly, 50(2), 313–338.Find this resource:

Bernhard, M., Örsün, Ö. F., & Bayer, R. (2017). Democratization in conflict research: How conceptualization affects operationalization and testing outcomes. International Interactions.Find this resource:

Betts, R. K. (Ed.). (2012). Conflict after the Cold War: Arguments on causes of war and peace (4th ed.). Boston: Pearson.Find this resource:

Birnir, J. K. (2006). Ethnicity and electoral politics. Cambridge, U.K.: Cambridge University Press.Find this resource:

Bogaards, M. (2010). Measures of democratization: From degree to type to war. Political Research Quarterly, 63(2), 475–488.Find this resource:

Brancati, D., & Snyder, J. L. (2011). Rushing to the polls: The causes of premature postconflict elections. Journal of Conflict Resolution, 55(3), 469–492.Find this resource:

Braumoeller, B. F. (2004). Hypothesis testing and multiplicative interaction terms. International Organization, 58(4), 807–820.Find this resource:

Brownlee, J. (2007). Authoritarianism in the age of democracy. Cambridge, U.K.: Cambridge University Press.Find this resource:

Carothers, T. (2002). The end of the transitions paradigm. Journal of Democracy, 13(1), 5–21.Find this resource:

Cederman, L.-E., Gleditsch, K. S., & Hug, S. (2013). Elections and ethnic civil war. Comparative Political Studies, 46(3), 387–417.Find this resource:

Cederman, L.-E., Hug, S., & Krebs, L. F. (2010). Democratization and civil war: Empirical evidence. Journal of Peace Research, 47(4), 377–394.Find this resource:

Clare, J. (2007). Democratization and international conflict: the impact of institutional legacies. Journal of Peace Research, 44(3), 259–276.Find this resource:

Colaresi, M. (2014). With friends like these, who needs democracy? The effect of transnational support from rivals on post-conflict democratization. Journal of Peace Research, 51(1), 65–79.Find this resource:

Collier, P., Hoeffler, A., & Söderbom, M. (2008). Post-conflict risks. Journal of Peace Research, 45(4), 461–478.Find this resource:

Dahl, R. (1971). Polyarchy. New Haven, CT: Yale University Press.Find this resource:

Daxecker, U. (2007). Perilous polities? An assessment of the democratization-conflict linkage. European Journal of International Relations, 13(4), 527–554.Find this resource:

Daxecker, U. (2011). Rivalry, instability, and the probability of international conflict. Conflict Management and Peace Science, 28(5), 543–565.Find this resource:

Doyle, M. W., & Sambanis, N. (2000). International peacebuilding: A theoretical and quantitative analysis. American Political Science Review, 94(4), 779–801.Find this resource:

Eckstein, H. (1973). Authority patterns: A structural basis for political inquiry. The American Political Science Review, 67, 1142–1161.Find this resource:

Enterline, A. (1996). Driving while democratizing. International Security, 20, 183–196.Find this resource:

Fearon, J. D., & Laitin, D. D. (2003). Ethnicity, insurgency, and civil war. American Political Science Review, 97(1), 75–90.Find this resource:

Fjelde, H., & Höglund, K. (2016). Electoral violence: The emergence of a research field. APSA Comparative Democratization Newsletter, 14(2), 8–11.Find this resource:

Flores, T. E., & Nooruddin, I. (2012). The effect of elections on postconflict peace and reconstruction. The Journal of Politics, 74(2), 558–570.Find this resource:

Fortna, V. P., & Huang, R. (2012). Democratization after civil war: A brush-clearing exercise. International Studies Quarterly, 56(4), 801–808.Find this resource:

Geddes, B., Wright, J., & Frantz, E. (2014). Autocratic breakdown and regime transitions: A new data set. Perspectives on Politics, 12(2), 313–331.Find this resource:

Gleditsch, K. S., & Hegre, H. (2014). Regime type and political transition in civil war. In K. DeRouen Jr. & E. Newman (Eds.), Routledge handbook of civil wars (pp. 145–156). New York: Routledge.Find this resource:

Gleditsch, K. S., Hegre, H., & Strand, H. (2009). Democracy and civil war. In M. I. Midlarsky (Ed.), Handbook of war studies III: The intrastate dimension (pp. 155–192). Ann Arbor: University of Michigan Press.Find this resource:

Gleditsch, K. S., & Ward, M. D. (1997). Double take: A reexamination of democracy and autocracy in modern polities. The Journal of Conflict Resolution, 41(3), 361–383.Find this resource:

Gleditsch, K. S., & Ward, M. D. (2000). War and peace in space and time: The role of democratization. International Studies Quarterly, 44(1), 1–29.Find this resource:

Goertz, G. (2005). Social science concepts. Princeton, NJ: Princeton University Press.Find this resource:

Gurr, T. (1997). Polity II: Political structures and regime change, 1800–1986 (ICPSR 9263). Retrieved from http://privatewww.essex.ac.uk/~ksg/data/polity/P2cdebk.PDF

Gurr, T., Jaggers, K., & Moore, W. H. (1989). Polity II Codebook. Boulder, CO: Dept. of Political Science.Find this resource:

Gurses, M., & Mason, T. D. (2008). Democracy out of anarchy: The prospects for post-civil-war democracy. Social Science Quarterly, 89(2), 315–336.Find this resource:

Hartzell, C. A., & Hoddie, M. (2015). The art of the possible: Power sharing and post–civil war democracy. World Politics, 67(1), 37–71.Find this resource:

Hegre, H., Ellingsen, T., Gates, S., & Gleditsch, N. P. (2001). Toward a democratic civil peace? Democracy, political change, and civil war, 1816–1992. American Political Science Review, 95(1), 33–48.Find this resource:

Hegre, H., & Nygard, H. M. (2015). Governance and conflict relapse. Journal of Conflict Resolution, 59(6), 984–1016.Find this resource:

Hippler, J. (2008). Democratization after civil wars—Key problems and experiences. Democratisation, 15(3), 550–569.Find this resource:

Huntington, S. P. (1968). Political order in changing societies. New Haven, CT: Yale University Press.Find this resource:

Jarstad, A. K., & Sisk, T. D. (2008). From war to democracy: Dilemmas of peacebuilding. Cambridge, U.K.: Cambridge University Press.Find this resource:

Joshi, M. (2010). Post-civil war democratization: Promotion of democracy in post-civil war states, 1946–2005. Democratization, 17(5), 826–855.Find this resource:

Jung, J. K. (2012). Power-sharing and democracy promotion in post-civil war peace-building. Democratization, 19(3), 486–506.Find this resource:

Kadera, K. M., Crescenzi, M. J., & Shannon, M. L. (2003). Democratic survival, peace, and war in the international system. American Journal of Political Science, 47(2), 234–247.Find this resource:

King, G., Tomz, M., & Wittenberg, J. (2000). Making the most of statistical analyses: Improving interpretation and presentation. American Journal of Political Science, 44, 341–355.Find this resource:

King, G., & Zeng, L. (2001). Logistic regression in rare events data. Political Analysis, 9, 137–163.Find this resource:

Krain, M., & Myers, M. E. (1997). Democracy and civil war: A note on the democratic peace proposition. International Interactions, 23(1), 109–118.Find this resource:

Leonard, W. (2004). The paradox of “warlord” democracy: A theoretical investigation. American Political Science Review, 98(1), 17–33.Find this resource:

Letsa, N. W. (2017). Voting for peace, mobilizing for war: Post-conflict voter turnout and civil war recurrence. Democratization, 24(3), 425–443.Find this resource:

Levitsky, S., & Way, L. (2010). Competitive authoritarianism: Hybrid regimes after the Cold War. Cambridge, U.K.: Cambridge University Press.Find this resource:

Lind, J. (2011). Democratization and stability in East Asia. International Studies Quarterly, 55(2), 409–436.Find this resource:

Linz, J., & Stepan, A. (1996). Problems of democratic transition and consolidation: Southern Europe, South America, and post-Communist Europe. Baltimore: Johns Hopkins University Press.Find this resource:

Magaloni, B. (2008). Voting for autocracy: Hegemonic party survival and its demise in Mexico. Cambridge, U.K.: Cambridge University Press.Find this resource:

Manning, C., & Smith, I. (2016). Political party formation by former armed opposition groups after civil war. Democratization, 23(6), 1–21.Find this resource:

Mansfield, E. D., & Snyder, J. (1995a). Democratization and the danger of war. International Security, 54, 775–808.Find this resource:

Mansfield, E. D., & Snyder, J. (1995b). Democratization and war. Foreign Affairs, 74(3), 79–97.Find this resource:

Mansfield, E. D., & Snyder, J. (2002a). Democratic transitions, institutional strength, and war. International Organization, 56, 297–337.Find this resource:

Mansfield, E. D., & Snyder, J. (2002b). Incomplete democratization and the outbreak of military disputes. International Studies Quarterly, 46, 529–549.Find this resource:

Mansfield, E. D., & Snyder, J. (2005). Electing to fight: Why emerging democracies go to war. Cambridge, MA: MIT Press.Find this resource:

Mansfield, E. D., & Snyder, J. (2009). Pathways to war in democratic transitions. International Organization, 63(2), 381–390.Find this resource:

Mansfield, E. D., & Snyder, J. (2012a). Democratization and civil war. In J. L. Snyder (Ed.), Power and progress: International politics in transition (pp. 175–200). New York: Routledge.Find this resource:

Mansfield, E. D., & Snyder, J. (2012b). Democratization and the Arab Spring. International Interactions, 38(5), 722–733.Find this resource:

Mason, T. D. (2003). Globalization, democratization, and the prospects for civil war in the new millennium. International Studies Review, 5(4), 19–35.Find this resource:

McFaul, M. (2007). Are new democracies war-prone? Journal of Democracy, 18(2), 160–167.Find this resource:

Miller, B. (2012). Does democratization pacify the state? The cases of Germany and Iraq. International Studies Quarterly, 56(3), 455–469.Find this resource:

Mitchell, S. M., Gates, S., & Hegre, H. (1999). Evolution in democracy-war dynamics. Journal of Conflict Resolution, 43(6), 771–792.Find this resource:

Montgomery, E. B., & Pettyjohn, S. L. (2010). Democratization, instability, and war: Israel’s 2006 conflicts with Hamas and Hezbollah. Security Studies, 19(3), 521–554.Find this resource:

Muller, E. N., & Weede, E. (1990). Cross-national variation in political violence: A rational action approach. Journal of Conflict Resolution, 34(4), 624–651.Find this resource:

Munck, G. L., & Verkuilan, J. (2002). Conceptualizing and measuring democracy: Evaluating alternative indices. Comparative Political Studies, 35(1), 5–34.Find this resource:

Narang, V., & Nelson, R. M. (2009). Who are these belligerent democratizers? Reassessing the impact of democratization on war. International Organization, 63(2), 357–379.Find this resource:

Nilsson, M. (2012). Reaping what was sown: Conflict outcome and post-civil war democratization. Cooperation and Conflict, 47(3), 350–367.Find this resource:

Peceny, M., Beer, C. C., & Sanchez-Terry, S. (2002). Dictatorial peace? American Political Science Review, 96(1), 15–26.Find this resource:

Przeworski, A. (1991). Democracy and the market. Cambridge, U.K.: Cambridge University Press.Find this resource:

Rasler, K., & Thompson, W. R. (2004). The democratic peace and a sequential, reciprocal, causal arrow hypothesis. Comparative Political Studies, 37(8), 879–908.Find this resource:

Reiter, D., & Tillman, E. R. (2002). Public, legislative, and executive constraints on the democratic initiation of conflict. Journal of Politics, 64(3), 810–826.Find this resource:

Ripsman, N. M. (2012). Promoting regional peacemaking. International Journal, 67(2), 431–436.Find this resource:

Sartori, G. (1991). Comparing and miscomparing. Journal of Theoretical Politics, 3(3), 243–257.Find this resource:

Savun, B., & Tirone, D. C. (2011). Foreign aid, democratization, and civil conflict: How does democracy aid affect civil conflict? American Journal of Political Science, 55(2), 233–246.Find this resource:

Schedler, A. (2006). The logic of electoral authoritarianism. In A. Schedler (Ed.), Electoral authoritarianism: The dynamics of unfree competition (pp. 1–23). Boulder, CO: Lynne Rienner.Find this resource:

Siroky, D. S., & Aprasidze, D. (2011). Guns, roses and democratization: Huntington’s secret admirer in the Caucasus. Democratization, 18(6), 1227–1245.Find this resource:

Snyder, J. L. (2000). From voting to violence: Democratization and nationalist conflict. New York: Norton.Find this resource:

Söderberg Kovacs, M., & Hatz, S. (2016). Rebel-to-party transformations in civil war peace processes 1975–2011. Democratization, 23(6), 990–1008.Find this resource:

Strasheim, J., & Fjelde, H. (2014). Pre-designing democracy: Institutional design of interim governments and democratization in 15 post-conflict societies. Democratization, 21(2), 335–358.Find this resource:

Sunde, U., & Cervellati, M. (2014). Democratizing for peace? The effect of democratization on civil conflicts. Oxford Economic Papers, 66(3), 774–797.Find this resource:

Tan, B. S. (2013). Regime change and conflict recidivism within rivalry: Interludes in disputes. Swiss Political Science Review, 19(2), 189–211.Find this resource:

Thompson, W. R., & Tucker, R. M. (1997). A tale of two democratic peace critiques: The hypothesized bellicosity of democratic dyads and new democratizing states. Journal of Conflict Resolution, 41, 428–451.Find this resource:

Trier, S., & Jackman, S. (2008). Democracy as a latent variable. American Journal of Political Science, 52(1), 201–217.Find this resource:

Vreeland, J. R. (2008). The effect of political regime on civil war unpacking anocracy. Journal of Conflict Resolution, 52(3), 401–425.Find this resource:

Wahman, M., Teorell, J., & Hadenius, A. (2013). Authoritarian regime types revisited: Updated data in comparative perspective. Contemporary Politics, 19(1), 19–34.Find this resource:

Walter, B. F. (2015). Why bad governance leads to repeat civil war. Journal of Conflict Resolution, 59(7), 1242–1272.Find this resource:

Ward, M. D., & Gleditsch, K. (1998). Democratizing for peace. American Political Science Review, 92(1), 51–62.Find this resource:

Wilkinson, S. I. (2006). Votes and violence: Electoral competition and ethnic riots in India. Cambridge, U.K.: Cambridge University Press.Find this resource:

Wolf, R., Weede, E., Enterline, A. J., Mansfield, E. D., & Snyder, J. (1996). Correspondence: Democratization and the danger of war. International Security, 20(4), 176–207.Find this resource:

## Notes:

(1.) We inquired into why Polity dropped this index. According to Ted Gurr (personal communication), it was hard to obtain reliable information on all the components, and the team thought that existing data on state spending as a share of GDP was a better variable to capture the strength of the state.

(2.) While Polity has served the literature on conflict well, where alternative indicators exist, testing for robustness should become a more regular practice to insure findings are not a product of a particular coding. Work we have done suggests that among competing operationalizations of regime change, the Polity-based operationalization used by Mansfield and Snyder is most conducive to not rejecting their hypotheses (Bernhard, Örsün, & Bayer, 2017)

(3.) Like Mansfield and Snyder, we set all the values of the explanatory variables at their means for continuous ones and at zero for dichotomous ones.

(4.) This is an out-of-sample prediction and the combination is chosen to illustrate the best case scenario for dangerous democratization thesis.

(5.) The results for incomplete democratization side j are also similar.

(6.) Given that there are 124 observations with highly weak institutions experiencing an incomplete democratization, we can expect 0.05 wars to have occurred due to weak institutions during incomplete democratization.

(7.) The results are very close when we vary the size of the choice-based sample by dropping 90 to 95% of all non-war cases.