The Reciprocal Relationship Between Military Conflict and Democracy

In sum, the Extraction School posits that raising revenues and armies requires political compromises and triggers social change.3 Mobilization for large-scale war can, therefore, have th

The Reciprocal Relationship Between Military Conflict and Democracy

1 August 2005

Abstract:

Does democracy cause peace, or is democracy a consequence of peace? The burgeoning

democratic peace literature has provided strong empirical evidence for the claim that democraciesare a cause of peace However, several skeptics of the democratic peace have suggested that the statistical findings are spurious They contend that democracy is a consequence of peace That

is, democratic institutions and norms grow more rapidly in a peaceful environment In this case, some other factor (e.g., geographic isolation) is the root cause of peace among democratic sates

We test these competing claims using a simultaneous equation model Using a unique data set of all international disputes from 1960-1988, we find strong support for reciprocal causation As thedemocratic peace theorists claim, democracy causes peace even after controlling for military conflict in the system and region Conversely, peace in the region appears to encourage the development of democratic polities

David L Rousseau Department of Political ScienceRockefeller College of Public Affairs and Policy

University at Albany (SUNY)

135 Western AvenueAlbany, NY 12222

andHyung Min KimDepartment of Political ScienceUniversity of North Carolina at Chapel Hill

361 Hamilton Hall, CB #3265Chapel Hill, NC 27599-3265

Paper prepared for the annual meeting of the American Political Science Association, September

1st – 4th, 2005, Washington, DC Please send comments to the first author

Does democracy cause peace, or is democracy a consequence of peace? The enormous democratic peace literature claims democracy is a cause of peace Over the last decade a strong consensus has emerged supporting the claim that democracies do not use large-scale violence against other democracies (i.e., the existence of a dyadic democratic peace) More recently, a number of scholars have demonstrated that democracies are often more peaceful in general (i.e., the existence of the monadic democratic peace) From the perspective of this literature,

democratic institutions, norms or identities reduce the probably that democratic polities resort to military force to resolve international disputes

But several skeptics of the democratic peace argue that democracy is a consequence of peace That is, democratic institutions and norms grow more rapidly in a peaceful environment

In this case, some other factor (e.g., geographic isolation) is the root cause of peace among democratic sates Thus, skeptics such as Thompson (1996) and Layne (1994) argue that the statistical correlation between democracy and peace in the democratic peace literature is spurious

The purpose of this paper is to sort out these completing claims Using a simultaneous equation model, we test both arguments in the same statistical model We find strong support for reciprocal causation As the democratic peace theorists claim, democracy causes peace even aftercontrolling for military conflict in the system and region Conversely, the skeptics are correct in their narrow claim that peace in the region encourages the development of democratic polities However, the broader skeptical claim that the democratic peace will evaporate after controlling for conflict in the environment is not supported by our simultaneous equation model

Our model contributes to the literature in five ways First, it is one of only a handful of models that directly tests for reciprocal causation Second, it employs a data set of international disputes that is broader in scope than analyses that are restricted to wars or militarized crises Third, unlike previous simultaneous equation models, the data set employs a directed dyad structure that allows us to test monadic and dyadic arguments simultaneously Fourth, the model directly compares explanations from the systemic, regional, dyadic, and monadic levels Fifth, the model employs a unique conflict initiation variable that allows a more theoretically precise test of the democratic peace side of the equation

The remainder of the paper is divided into six sections The following section explores the causal mechanisms (or micro-foundations) of both halves the reciprocal relationship The second section explains the procedure and rationale for our simultaneous equation model The third section presents our hypotheses and the fourth section describes the data set The fifth section describes the results of the model and the sixth section summarizes our findings

CAUSAL MECHANISMS IN THE LITERATURE

Part I: Military Conflict Decreases (or Increases) Democracy

Reflecting on the formation of national states in Western Europe, Tilly (1975) argues thatwar (and the tax systems created to pay for war) played a vital role in the development of strong states “War made the state and the state made war” (Tilly 1975, 42) Beginning with almost 500independent political entities in 1500, Europe gradually consolidated into approximately twenty sovereign states by the start of the 20th century Many alternatives to the sovereign state, rangingfrom empires and city states to trading leagues and federations, fell by the wayside during the consolidation process (Spruyt 1994)

Tilly argues that the costs of war increased rapidly due to three revolutionary factors: technical innovations in weapons (e.g., cannons and firearms), tactical changes in warfare (e.g., the rise of infantry), and an expansion in the size of armies Prior to these changes, rulers in the largely non-monetized feudal economic and military system had relatively few expenses and could typically manage affairs using the revenue produced from crown lands In the feudal system, land was exchanged for military service and the vassals were responsible for equipping

and supplying their own troops After the technical and tactical innovations, the larger mercenaryarmies and their advanced weapons became horrifically expensive Desperate for additional revenue, rulers in Europe expanded the tax base and institutionalized the collection process The first bureaucratic structures were created for revenue collection The need for ever larger armies led to the extensive use of mercenaries from the 15th to the 18th century, and then to the use of conscripts during the Napoleonic Wars and the total wars of the 20th century Tilly also argues that the mobilization of fiscal and human resources triggered resistance He contends that states responded to this threat by increasing the coercive apparatus (e.g., police) to minimize

disruptions For the region as a whole, the process was autocatalytic As rulers fielded even larger armies or equipped their artillery with even more accurate cannons, neighbors were forced

to raise revenues in order to respond in kind The upward spiral continued in Europe for at least

600 years

If war did in fact contribute to the state building process, a second question emerges: justwhat sort of sovereign state emerged from this violent environment? Hintze (1906), like Tilly,argues that state development is inextricably linked to military development While domesticfactors such as the distribution of power between interest groups played an important role in statedevelopment, the level of external threat together with the state’s position in the internationalpolitical hierarchy could be decisive "All state organization was military organization,organization for war" (1906, 181) Hintze also argues that external threats can lead to the

development of autocratic institutions.1England, with her insular security, was not directly exposed to the danger of these wars She needed no standing army, at least not one of Continental proportions, but only a navy which served commercial interests as much as war aims In consequence, she developed no absolutism Absolutism and militarism go together on the Continent just as self government and militia in England The main explanation for the difference in the way political and military organizations developed between England and the Continent one which became more and more distinct after the middle of the seventeenth century lies in the difference in the foreign situations (Hinze

1906, 199)

England did not become democratic because it was either Protestant (Weber 1958), or

industrialized (Smith 1965; Dahrendorf 1967), or agriculturally commercialized (Moore 1966) Rather, the lack of a persistent external threat allowed interest groups to maintain and expand political and economic privileges For Hintze, this path was simply impossible for a state such as Prussia, which had to centralize power and eliminate domestic opposition in order to efficiently mobilize the resources required to maintain independence in an anarchic world

Lasswell (1941), writing at a time in which the world (but not yet the United States) had plunged into the Second World War, takes a similar position "The purpose of this article is to consider the possibility that we are moving toward a world of "garrison states" a world in which the specialists of violence are the most powerful group in society" (1941, 455) While not arguing the outcome was inevitable, Lasswell feared that mobilization for total war encouraged the authoritative allocation of resources at the expense of market allocation and the emphasis of the collective or the public at the expense of the individual or the private During war "decisions will be more dictatorial than democratic, and institutional practices long connected with modern democracy will disappear (1941, 461)." Throughout his essay, Lasswell emphasizes the role of propaganda in mobilizing resources, increasing morale, repressing dissent, socializing the young, and ritualizing democracy In sum, Lasswell echoes Hintze's contention that a hostile external environment leads to autocratic institutions but he focuses on the demise of existing modern democracies rather than the rise of embryonic autocratic states

External threats can undermine pluralist institutions and norms in a number of ways First, the existence of a severe external threat can lead to the elimination of free and fair elections (e.g., the U.K suspends national elections during both World Wars) Second, during war

decision-making power constitutionally vested in the legislature is often shifted to the executive

branch (e.g., Germany during the First World War) Third, the hostile external environment can lead to a suspension of the rule of law in the hope of efficiently mobilizing resources (e.g., Lincoln's suspension of habeas corpus during the American Civil War (Rehnquist 1998)) Fourth,threats often lead to the expansion of the power of the military over economic and political decision making (e.g., Ludendorff's military dominated government in Germany during the First World War) Fifth, threats often lead to the barring (or removal) of individuals of suspect

loyalties from holding public office (e.g., the removal of Jews from the provincial government in Vichy France in 1940) Sixth, in the face of an intense threat, even short of war, states often restrict the right of its citizens to form political parties and compete in the political arena (e.g., Communists in the United States during the early Cold War era) Seventh, severe threats often lead to restriction in the right to vote in national and local elections (e.g., Japanese-Americans during internment in the Second World War (Daniels 1981)) Finally, in the hope of minimizing opposition to extraction policies, governments often repress free speech (e.g., Espionage Act of

1917 in the United States)

While threats and wars often result in a restriction of political rights, Downing (1992) argues that under certain conditions war may expand rather than contract political rights

Downing focuses on the medieval origins of constitutional government in Europe Medieval customs and institutions (e.g., reciprocal rights between peasants and lords; the balance of power between the crown, the Catholic Church, nobles and burghers; prototype parliaments; and the rule

of law) laid the foundation for participatory government As with Tilly, the military revolution plays a decisive role in Downing's story States that are forced to extract resources domestically are apt to develop autocratic structures capable of squeezing every last cent out of reluctant lords, merchants, and peasants In contrast, states that are capable of raising capital in markets or fighting on the territory of other states (where they can extract resources from the victims of aggression) are less likely to develop autocratic institutions In sum, the lack of a medieval legacy, the presence of severe security threats, and the reliance on domestic resources to finance military campaigns all increase the probability of an autocratic state

Downing emphasizes that external threats and the need for wartime revenue often

triggered an intense battle between the crown and its subjects New taxes were not dictated by rulers from above; they were negotiated with interest groups from below Feudal customs and institutions limited the crown’s ability to raise revenue When the crown "asked" for

extraordinary revenue, it was often forced to reward interest groups for compliance

[E]states were essential to finance and consensus building: they debated matters

of war, foreign policy, trade, and justice … [E]states often took advantage of any

upper hand they might have had by enhancing their privileges and liberties, and

by expanding their role in the machinery of government In exchange for

financial support, more often than not in time of war, estates assumed increasing

control of law making (Downing 1992, 31)

Moreover, this expansion of rights related to the extraction of human resources as well

As the size of armies grew, the royal treasury simply could not afford to pay the mercenaries for their services The solution was to enlist volunteers and conscripts But why would a young peasant from Languedoc or Hanover volunteer for the hardship of military service or honor a conscription notice?2 They complied because they were citizens members of a community that provided rights and demanded obligations The Napoleonic Wars are considered a historical turning point because large “citizen” armies began to take the place of “mercenary” armies that had dominated in Europe since the end of the Middle Ages (Dupuy 1984, 156) Although

Napoleon's Grand Armée assembled for the invasion of Russia in 1812 was dominated by foreign

troops, it contained some 300,000 French conscripts and volunteers (Finer 1975, 146) "Citizen” armies would play a central role in the next global wars: World War I and World War II

In sum, the Extraction School posits that raising revenues and armies requires political compromises and triggers social change.3 Mobilization for large-scale war can, therefore, have the unintended consequence of expanding political rights in the long run.4

Part II: Democracy Decreases Military Conflict

The enormous democratic peace literature focuses on whether more inclusive and

competitive polities are less likely to use military force to resolve international conflicts (Kant 1795) At least three distinct causal mechanisms have been proposed in the literature: 1) a structural explanation, 2) a political norms explanation, and 3) and an identity explanation

The institutional structures school focuses on the relationship between political structuresand the domestic political costs of using force (Morgan and Campbell 1991; Morgan and

Schwebach 1992) According to this school, decisions to use military force are choices made by

political leaders based on domestic and international cost/benefit calculations Foreign policy decisions can have costly domestic political repercussions The expenditure of resources and loss

of human life can mobilize opposition groups or fracture a ruling coalition Relative to other political systems, democratic decision makers must be more sensitive to these potential domestic costs This constrains their behavior in comparison with leaders of non-democratic states Immanuel Kant, the first proponent of the democratic peace, uses this argument to support his claim that oligarchies are more likely to initiate war than republics (Kant 1795).5

The political norms school emphasizes the socialization of political leaders within their domestic political environment (Maoz and Russett 1993; Russett 1993; Dixon 1993, 1994; Huth and Allee 2002) This argument has two parts First, democratic political leaders are socialized within a system that emphasizes compromise and non-violence Leaders typically resolve

political conflicts in democracies through negotiation and log-rolling Losing a political battle does not result in the loss of political rights or exclusion from future battles Moreover, coercion and violence are not seen as legitimate means for resolving a dispute Conversely, non-

democratic political leaders are socialized in an environment in which politics is more akin to a zero sum game in which rivals and losers are regularly removed from the game Coercion and violence are more widely accepted as legitimate means for resolving political conflicts In general, political leaders are more likely to impose decisions rather than compromise when dealing with the opposition Second, the norms school assumes that domestic political norms are naturally externalized by decision makers when they confront international disputes Thus, democratic polities are expected to be less likely to use military force to resolve disputes

The shared identity school argues that peace between democracies is a function of a common social identity (Hopf 1998; Kahl 1998/99) Social identities are bundles of shared values, beliefs, attitudes, norms, and roles that are used to draw a boundary between the "in-group" and the "out-group" (Rousseau, forthcoming) Members of the in-group are viewed as less threatening than members of the out-group If democratic polities use democratic norms or values to define the in-group, then they will view the actions and capabilities of the other

democracies as less threatening The shared identity will reduce the likelihood that either party quickly resorts to violence to resolve a political dispute

The structural and political norms arguments imply that democratic polities should be

less likely to use force regardless of the regime type of the opponent This is a purely monadic

structural argument: the regime type of the state alone determines its behavior in international disputes However, the literature has proposed a dyadic explanation: due to the expectation that other democracies will be similarly constrained (by institutions or norms), democracies are only

peaceful with other democracies This is referred to as the dyadic democratic peace in the

literature While empirical evidence for the dyadic democratic peace is extensive, only recently has strong evidence emerged in support of the monadic democratic peace.6 Finally, the shared identity argument is generally stated as a dyadic argument: if you are a member of the in-group, I

am less likely to view your capabilities and actions as threatening In general, the construction of

an identity requires an “other.”7

Part II: Reciprocal Causation

Skeptics of the power of the democratic peace often suggest that reciprocal causation could explain the results.8 Case studies by Layne (1994) and Thompson (1996) suggest that the peaceful nature of the environment rather than the peaceful disposition of democracies best explains the democratic peace Over time, a handful of studies have attempted to explore the reciprocal relationship using a variety of data sets and estimation methods (see Table 1) In general, modeling the reciprocal relationship has been slowed by the lack of appropriate

estimation tools that are able to simultaneously address all the problems with the data (e.g., one dichotomous and one continuous endogenous variable, time dependency, lag structures, weak instruments) As with any complicated statistical analysis, fixing one problem tends to raise a host of new and difficult issues

Mousseau and Shi (1999) argue that the reciprocal relationship can be probed by focusing

on “anterior” regime changes – that is, regime changes that occur as states prepare for war (as opposed to regime changes during war (concurrent changes) or after a war (posterior changes)) Using democracy data from Polity III and war data from the COW interstate war data base for theperiod from 1816 to 1992, they employ an interrupted time series analysis with a separate

regression for each case Overall, they find only limited support for the anterior effect of war preparation decreasing democracy and conclude that the democratic peace is robust However, our ability to draw firm conclusions from the analysis is hampered by a number of problems: 1) only modeling one half of the reciprocal relationship; 2) extensive missing data; 3) short time series for each separate regression; and 4) a lack of control variables

Crescenzi and Enterline (1999) examine the reciprocal relationship between democracy, democratization, and interstate conflict at the systemic level of analysis Using democracy data from Polity III and war data from the COW interstate war data base for the period from 1816 to

1992, they employ Granger causality and vector auto-regression techniques Overall, they find little support for the triangular relationship across time and space For example, systemic warfaredoes decrease the proportion democratic in the system at particular times (e.g., Europe from 1936-1992), but not at other times (e.g., Europe 1816-75) or in other places (e.g., Asia from 1945-92 or Africa from 1960-92) The results do not conclusively resolve the debate because the systemic analysis does not shed light on the dyadic or monadic democratic peace and the lack of control variables raises the issue of omitted variables

James et al (1999) attempt to explore the relationship between MID involvement and democracy using a simultaneous equation model However, they found the results “implausible” due to the rarity of conflict involvement and their multiple categorical dependent variable Therefore, James et al employ a standard two-step procedure to estimate one half of the

reciprocal relationship (MID involvement decreases democracy) and multinomial probit model for the other half (democracy decreases MID involvement) Although James et al conclude that democracy is not an important cause of peace, their findings are severely criticized by Oneal and Russett (2000).9 Problems with the analysis include 1) the failure to model reciprocal causation directly, 2) the use of an extremely weak instrument for the two-step equation, 3) the use of a dyadic data structure; 4) the employment of a highly skewed multinomial dependent variable; 5) the absence of important control variables; and 6) the selection of a flawed level of democracy variable

Reuveny and Li (2003) examine the reciprocal relationship using a three equation

simultaneous equation model Using democracy data from Polity 98 and militarized dispute data from the COW Militarized Interstate Dispute (MID) data base for the period from 1950 to 1992, they find support for both sides of the reciprocal relationship: democratic dyads decrease MID involvement and MID involvement decreases democratic dyads While the analysis is an

important contribution to the field, it cannot resolve the debate due to 1) the use of an

involvement conflict variable, 2) the nondirected-dyad data structure, 3) the employment of lagged dependent variables (Achen 2000); 4) the use of the imprecise “weak link” assumption, 5) the neglect of the time series problem, and 6) an extremely complicated three equation design

Rasler and Thompson (2004) examine the reverse casual arrow from military conflict to democracy using three separate equations modeling domestic resource inequality, democracy, andmilitary disputes The data set consists of nine great powers (Austria-Hungary, Britain, China, France, Germany, Italy, Japan, Russia, and the United States) from 1816 to 1992 Using data from Vanhanen (1997), Polity III, and the MID, they find that wars and threats decrease

democracy and that democracy (using the Vanhanen but not Polity III) decreases MID dispute involvement As with the previous analyses, the article has a number of important limitations including 1) the limited number of countries in the data set; 2) the failure to model reciprocal causation directly; 3) the lack of control variables; 4) the use of lagged dependent variables; 5) the use of an involvement conflict variable; and 6) the averaging of variables over ten year periods

In sum, a handful of authors have attempted to tackle the very difficult problem of reciprocal causation While each paper has important strengths and have added to our

understanding of the problem, no paper has been able to resolve the debate We hope to

contribute to this on-going discussion by presenting a new test which eliminates some (but certainly not all) of the problems arising in these earlier studies

THE SIMULTANEOUS EQUATION MODEL

Most studies of military conflict and democracy have restricted their focus to one half of the reciprocal relationship This is due both to subfield interests (e.g., comparative politics specialists are more interested in the cause of democracy and international relations specialists aremore interested in the cause of military conflict) and the technical difficulty associated with modeling reciprocal causation Unfortunately, the failure to address simultaneity can lead to erroneous conclusions Gujarati (1995: 647) explains that simultaneity problems occur when the endogenous variable in one equation (e.g., military conflict) appears as an explanatory variable inanother equation of the system (e.g., democracy) As a result, such an endogenous explanatory

variable becomes stochastic and is correlated with the disturbance term of the equation where it

appears as an explanatory variable In this situation, Gujarati argues that the classical OLS will produce estimators that are not consistent regardless of the size of the sample (see also Greene, 1997: chap 16) The rationale for avoiding the classical OLS (ordinary least squares) method in the presence of the simultaneity problem also applies to the case of the classical logit or probit method because key assumptions of the OLS method (such as nonstochasticity and the

independent distribution of the explanatory variables) are also incorporated into logit and probit analyses Thus, employing a standard probit model in a situation with reciprocal causation will result in biased coefficients

The ‘simultaneous equation’ model used to test our hypotheses is an instrumental

variables, limited information two-stage probit least squares estimation method The model requires two steps First, we regress the endogenous explanatory variables (use of force and democracy) on all of the predetermined variables in the whole system to eliminate the likely correlation between those endogenous explanatory variables and the stochastic disturbance terms

in each equation, which violates the assumptions of the classical OLS and probit methods This first step provides us a ‘proxy’ for each endogenous explanatory variable—called an instrumentalvariable—that is uncorrelated with the disturbance term in each equation Second, we regress ourtwo original endogenous variables on these proxies (or instrumental variables) plus the other independent variables in each equation (Gujarati, 1995: 686-88) This two-stage probit least squares method gives us an unbiased and efficient estimator of each parameter in the equations (Amemiya, 1978; Heckman, 1978; Maddala, 1983)

In general, the simultaneity problem has been explored with one of two estimation approaches: 1) simultaneous equation models or 2) distributed lag models As just discussed, the simultaneous equation approach involves estimating two (or more) equations with each

endogenous variable appearing as an exogenous variable in the other equation(s) In the

democracy and military conflict literature, Reuveny and Li (2003) employ a simultaneous

equation model using a two-stage least squares method In the interdependence and war

literature, Keshk, Pollins, & Reuveny (2004), Mansfield (1994), Polachek (1997), and Kim (1998) use simultaneous equations estimation methods In contrast, distributed lag and related estimation models are time series methods that allow the researcher to relax assumptions such as right hand variable exogeneity and fixed lag structures In the democracy and military conflict literature, Crescenzi and Enterline (1999) employ a distributed lag model at the system level In the interdependence and war literature, Oneal, Russett, & Berbaum (2003), Reuveny (2001), and Reuveny & Kang (1996, 1998) use the distributed lags or other related estimation methods

Although the distributed lag approach has a number of advantages (e.g., permits a richer modeling of temporal dependence), we have selected the simultaneous approach for three

reasons First, distributed lag and related models tend to be very sensitive to the length of the lag selected by the modeler (Geweke, 1984) Second, the conclusion drawn from the bilateral Granger causality models may depend on the inclusion of a third variable (Granger, 1980) Third, detrending a series may either change its dynamic properties or lead to different causality conclusions (Kang, 1985) Thus, in this paper we will restrict our analysis to a simultaneous equation model

HYPOTHESES

Figure 1 illustrates the hypotheses tested in this paper according to their level of analysis

The two endogenous variables are the Level of Democracy and Aggressive Use of Force The Level of Democracy variable was constructed by subtracting the Polity IV autocracy index from

the democracy index to produce a variable which ranges from -10 to +10 To ease interpretation

of the statistical results, this variable was rescaled from 0 to 20 The Polity democracy index is comprised of four components: openness of executive recruitment, competitiveness of executive recruitment, competitiveness of participation, and legislative constraints on the executive The autocracy index contains all the elements in the democracy index plus a fifth component, the regulation of participation For example, using the Polity data and this method, the United States receives a score of 20 for all years in the data set while the Soviet Union receives a score of 6 for all years after the death of Stalin

Aggressive Use of Force is defined as the use of military force on the territory of another

sovereign state The variable codes for both uses of force by a state’s regular armed forces and through third parties on the territory of another state (e.g., the American support for the

Mujahideen in Afghanistan during the 1980s).10 The Aggressive Use of Force variable differs

from a conflict involvement variable in that the target of aggression is not coded as having used force For example, while autocratic Germany used aggressive force in 1914, democratic

Belgium did not Aggressive Use of Force is also superior to a conflict initiation variable for our

particular data set because the rarity of initiation undermines our ability to derive meaningful results from the statistical analysis

Systemic Level Variables

The relationship between democracy and military conflict has been explored at the systemic level in several articles (McLaughlin 1996; Crescenzi and Enterline 1999) The causal mechanism in these systemic variables involves demonstration effects For example, the more violence there is in the international system, the more likely a state leader is to view violence as a legitimate means for resolving a dispute Similarly, the greater the proportion of democracies in the system, the more likely a state is to view democracy as a legitimate and effective form of

government This demonstration effect should encourage the development of democratic

institutions and norms.11 In our model, Systemic War, measured as the total number of wars in the international system in a given year, is expected to decrease the Level of Democracy and increase the Aggressive Use Force Data on wars are from the MID data set System Democracy Average, measured as the average level of democracy using the 0-20 Polity scale in a given year,

is expected to increase the Level of Democracy and decrease the Aggressive Use Force Finally, Systemic Trade Openness, measured as the average level of trade interdependence in the

international system in a given year, is expected to increase the Level of Democracy and decrease the Aggressive Use Force Interdependence is measured as exports plus imports divided by gross

national product using the Expanded Trade and GDP Data Version 3.0 by Gleditsch (2002a)

Regional Level Variables

Gleditsch (2002b) argues that powerful regional effects have been neglected in the literature Therefore, we have created a parallel set of regional variables for our conflict,

democracy, and interdependence variables Regional coding relied on the Correlates of War region variable which divides the world into five regions (Europe, Middle East, Africa, Asia, and North/South America) The operationalization of the regional variables is identical to the

systemic variables

Dyadic Variables

The dyadic variables capture the interaction of states (e.g., shared alliance) or relative measures (e.g., the balance of power) In our model, all the dyadic variables are on the right-handside of the figure and only impact the state’s propensity to use military force First, the shared alliance hypothesis predicts that if two states are members of an alliance against a common foe, they will be less willing to use military violence to resolve a dispute (Gowa 1999) This dummy variable takes the value of 1 when the two states in the dispute share a defense pact, neutrality pact, or an entente Otherwise the value is 0 The sources for this variable is the Correlates of War (COW) Formal Alliances data set, Version 3.03 (Gibler & Sarkees, 2004) Second, the balance of power hypothesis predicts that the stronger state in the dyad will be more likely to resort to military violence (Huth 1996) Each state’s military capability is the average of three components—number of troops, military expenditures, and military expenditures per soldier—from the COW National Material Capabilities data set (ICPSR 9903, Singer & Small, 1993) Thefinal product ranges from 0 to 1 A value of more than 0.50 indicates that the state’s military capability is superior to its opponent, while a value of less than 0.50 indicates military inferiority Third, the preferential trading area (PTA) hypothesis predicts that if both states are members of the same trading organization they will be less likely to use violence because of the costs

associated with disrupting the economic relationship The dummy Shared PTA variable is coded

as 1 if both states in a dispute share at least one preferential trading agreement identified by Pevehouse and Mansfield (2003) Fourth, the joint democracy hypothesis predicts that the more democratic a state, the less likely it is to use military force to resolve the dispute against other democracies (Russett and Oneal, 2001) This hypothesis tests the dyadic version of the

democratic peace In order to isolate the impact of level of democracy when facing a democratic

opponent, we introduce an interactive term which multiplies the state’s Level of Democracy by a

dummy variable indicating whether or not the opponent is a democratic state If a state’s

opponent scores 17 or greater on the democracy scale, the dummy variable is coded 1 Otherwisethe variable is coded 0 Thus when the opposing state is not democratic, this variable takes on a value of 0 When the opposing state is a democracy, however, this variable is equal to the state’s

Level of Democracy score Including this interaction as a separate variable in the analysis will allow an identification of the additional effect that the Level of Democracy score has on a state’s

behavior because the opposing state is democratic The final two dyadic variables control for

Contiguity and Distance In both cases, we predict that geographic proximity increases the

probability of using force because it is easer to project power across short distances If the two states in a dispute share a boundary on land or are separated by less than 150 miles of water either

directly or through their colonies or other dependencies, the variable Contiguity is coded 1; otherwise, it is coded 0 The variable Distance is the natural logarithm of the great circle distance

between the two states in a dispute

“trading state strategy” (versus the military expansion strategy) facilitated transitions to

democracy (Rosecrance 1985) Second, the per capita income hypothesis predicts that rising income will increase the level of democracy The creation of a strong middle class, the expansion

of education, and the desire to protect private property all contribute to demands for a more inclusive and competitive political system Third, the greater the percentage of arable land in a country will be associated with higher levels of democracy Arable land tends to be positively correlated with a greater percentage of independent farmers in society; the independent land owners are often the backbone of a democratic polity Fourth, the openness hypothesis predicts that the more a state trades, the more likely it is to have a high level of democracy Historically, economic liberalism is highly correlated with political liberalism (Russett and Oneal 2001; Rousseau, forthcoming) Fifth, the growth hypothesis predicts that positive economic growth in terms of gross domestic product will be positively associated with the level of democracy

(Reuveny and Li 2003) Conversely, economic decline should decrease democracy Sixth, the inflation hypothesis predicts that the economic turmoil caused by high inflation will undermine democracy (Reuveny and Li 2003) Finally, the model includes a lagged level of democracy dependent variable and a “years” variable in order to address temporal dependency The data for

Per Capita Income and Openness are from the Expanded Trade and GDP Data Version 3.0 by Gleditsch (2002a) The data for Member of GATT are from Reinhardt (1999); the data for Arable Land are from Central Intelligence Agency (2004); the data for Growth Rate and Inflation are

from Heston, Summers, and Aten (2002)

Finally, Figure 1 indicates that two monadic variables influence a state’s propensity to use force aggressively First, the satisfaction hypothesis predicts that states which are satisfied with the status quo are less likely to use aggressive force This variable is coded 1 if a state is satisfied with the status quo regarding the issue at stake in the dispute at the time the crisis begins.Otherwise it is coded 0 For example, in a territorial dispute, if the defending state is content withthe current distribution of territory, it is coded as satisfied In several disputes, both states are dissatisfied (e.g., India versus China in 1962) Second, we predict that major powers will be more likely to resort to military violence to resolve a dispute The variable is coded as 1 if a state

is a major power identified by the COW Project: for the entire time period of our analysis, the U.S., France, Great Britain, the Soviet Union, and China qualify as major power The data for

Major Power are taken from EUGene (Expected Utility Generation and Data Management program) Version 2.40 by Bennett & Stam (2000a) The data for Satisfaction are from Rousseau

(2005)

Equations (1) and (2) summarize the model that will be tested below.13

9*Level of Democracy,Balance of Forces, t + 10*Level of Democracy,Shared PTA, t + 11*Level of Democracy,Joint Democracy, t + 12*Level of Democracy,Contiguity, t +

13*Level of Democracy,Distance, t + 14*Level of Democracy,Major Power, t + 15*Level of Democracy,Satisfaction with the Status Quo, t + + 16*Level of Democracy,Peace

14*Level of Democracy,Lagged Level of Democracy, t-1 + 15*Level of Democracy,Years, t + e

Although the model may appear unnecessarily complex, simultaneous equation models must meet several conditions.14 In practice, each equation must contain one or more strong predictors of the endogenous variable that do not appear in the other equation This is often particularly difficult in international relations (Kim and Rousseau 2005) James et al (1999), for example, employ an extremely weak instrument that explains very little of the variance of the endogenous variable While they explicitly recognize this in the article, they do not appear to question whether such a weak instrument can tell us anything at all about the reciprocal

relationship between democracy and war In contrast, our somewhat lengthy equations provide very strong instruments for the two-step probit model

THE DATA SET

We test our hypotheses using a set of international disputes from 1960 to 1988 developed

by Rousseau (2005) The primary source for the identification of each international dispute is a data set developed by Sherman (1994) that identifies all domestic quarrels and international disputes from 1945 to 1988 Rousseau has modified Sherman’s data set in a number of ways First, he has restricted his data set to the period of 1960-88 Second, he has removed all domesticquarrels because the main focus of his research is on a state’s decision to use military force against other states Third, he has eliminated several categories of dispute cases in order to focus

on political-security conflicts that have some probability of escalating to military conflict

between internationally recognized sovereign states Fourth, he has aggregated types of disputes (e.g., U.S vs Cuba territorial dispute in 1962 and U.S vs Cuba regime type dispute in 1962) into a single dyadic dispute because these disputes are not independent events

The final data set consists of 223 international disputes between pairs of countries.15 The disputes vary in length from one to twenty-nine years The directed dyad data structure includes observations for both states in order to isolate the behavior of each party in the dispute As Bennett and Stam explain, ‘a directed dyad study easily allows for behavioral choices and dyadic outcomes to be different in the two directions and hence allows simultaneous testing of varied theories and hypotheses’ (2000b: 655) In our model, testing monadic and dyadic hypotheses using a conflict initiation dependent variable requires the use of a directed dyad structure The cross sectional pooled time series directed dyad data set contains 5770 observations 16

The use of this dispute data set differentiates our research from most other scholarly work

on the reciprocal relationship between democracy and military conflict First, most international disputes do not escalate into crises in which one or both parties threaten or use military force and,

by the same logic, most international crises do not escalate into wars in which one or both parties use large military forces to resolve the crisis.17 Therefore, most previous empirical studies that have focused on crises or wars capture only a small sub-set of the population of international conflicts The issue is important with respect to the liberal peace because if democracies choose not to escalate a political dispute into a militarized crisis or war, then analyses restricted to the

subset of crises and wars will underestimate the pacifying impact of each pillar of the liberal peace

Second, many other scholarly work on the conflict and democracy relationship use a state’s conflict involvement rather than a state’s conflict initiation as their dependent variable (Reuveny and Li 2003) Conflict involvement is an acceptable dependent variable for the war causes autocratization side of the reciprocal relationship because target states often have to raise huge armies and restrict political rights However, conflict involvement is a poor dependent variable for the democracy causes peace side of the equation because it groups victims of

aggression (e.g., Belgium 1914) with the aggressors (e.g., Germany 1914) Given the absence of

a theoretically ideal variable, we employ a conflict initiation variable in our primary analysis and

a conflict involvement variable in our sensitivity analysis

Third, much of the work on the liberal peace in general (e.g., Russett & Oneal, 2001) and the conflict-democracy relationship in particular (e.g., Reuveny and Li 2003) employ non-

directed dyad structures This data structure makes it impossible to test monadic arguments and encourages the use of conflict involvement variables In order to distinguish the monadic effect

of democracies from the dyadic effect of two democracies, we employ the directed dyad design (Bennett & Stam, 2004)

model The results in Model 3 clearly indicate a reciprocal relationship between conflict and

democracy The Aggressive Use of Force Instrument in the top half of Model 3 is negative and

statistically significant (B=-0.190, SE=0.07, t-statistic=2.64) As Lasswell and Hintze expected,

using military force can lead to a decrease in a state’s level of democracy Conversely, the Level

of Democracy Instrument in the lower half of Model 3 is negative and statistically significant (B=-0.020, SE=0.01, t-statistic=2.61) In addition, the dyadic Joint Democracy variable remains

negative and strongly significant in the lower half of Model 3 The model refutes the skeptics’ claim that controlling for conflict in the environment will cause the democratic peace to

evaporate

Given our focus on the reciprocal model, we will restrict our discussion to Model 3 The marginal impacts of the variables in the top half of the table can be interpreted just like any OLS

regression The marginal impacts for the variable in the Aggressive Use of Force equation can be

found in Table 3 The top half of Model 3 provides strong support for several hypotheses

purporting to explain the level of democracy in a state As expected, Systemic War decreases the level of democracy and System Democratic Average and Systemic Trade Openness increase the level of democracy At the regional level, the results are more mixed While the Regional Democratic Average variable is positive as expected, the Regional War variable is significant in

the wrong direction: at the regional level violence seems to increase democracy However, the marginal impact of all these systemic and regional variables are quite small compared to other levels of analysis For example, a one standard deviation increase in the amount of regional war only increases the level of democracy by 0.33 on the 0 to 20 Polity scale At the monadic level, trade openness, high per capita income, membership in GATT, and the level of democracy last

year all increase the level of democracy in the current year Finally, Years and Growth have a

statistically significant negative but substantively unimportant impact on the level of democracy

While the inclusion of a lagged dependent variable is common in simultaneous equation models in international relations (e.g., Reuveny and Li 2003), Achen (2000) argues that the inclusion of a lagged dependent variable can depress the size of the estimated coefficients for the

other independent variables In our model, while the removal of the lagged dependent variable still produces support for the reciprocal model, the size of the coefficients of many (but not all)

variables increases dramatically (e.g., Systemic War by 2.1 times, monadic Openness by 3.5 times, Membership in GATT by 3.8 times) While we have included the lagged dependent

variable in order to address dependency across time and to significantly increase the power of the instrument, we recognize that it limits our ability to draw strong conclusions about the marginal impacts of other variables in the democracy equation

The lower half Model 3 in Table 2 provides the simultaneous equation results for the

Aggressive Use of Force half of the reciprocal relationship As expected, the amount of Systemic War has a positive impact on the use of aggressive force by a state However, unexpectedly, Systemic Trade Openness also is positively associated with military violence At the regional level, both the Regional Democratic Average and the Regional Trade Openness are negatively associated with the use of force At the dyadic level, Joint Democracy and Satisfaction with the Status Quo decrease the probability of military force For example, the marginal analysis in

Table 3 indicates that a shift from autocracy (0) to full democracy (20) when facing a democratic opponent decreases the predicted probability of using force from 6.10 to 1.18 percent Similarly,

a shift from dissatisfied to satisfied with the status quo decreases the predicted probability of

using force from 5.31 to 1.58 percent Finally, Balance of Forces, at the dyadic level, increases

the probability of military force Increasing the level of balance of forces from one standard deviation below the mean (0.21) to one standard deviation above the mean (0.79) increases the probability of military force by 1.94 percent (i.e., a shift in the predicted probability of using aggressive force from 4.41 to 6.35 percent)

We conducted a number of sensitivity analysis in order ensure our findings are robust The primary results (i.e., statistical significance in the expected direction for both halves of the reciprocal relationship) are stable when (1) replacing MIDs for wars in the systemic and regional conflict measures, (2) dropping the lagged democracy variable from the model, (3) employing alternative operationalizations for the economic interdependence variables, and (4) dropping the temporal corrections (i.e., peace years and splines) from the model When using conflict

involvement in place of aggressive use of force, the Level of Democracy Instrument decreases the Aggressive Use of Force (B=-0.19, SE=0.005, t-statistic=3.53) but the Aggressive Use of Force Instrument falls just short of statistical significance Thus, the monadic and dyadic democratic

peace variables remain robust in all analyses and the conflict decreases democracy hypothesis in most analyses

As with any two-step model, a strong instrument is vital In our model, the instruments are quite powerful for both the democracy equation (R2 of 0.74) and the aggressive use of force equation (pseudo R2 of 46) Although the findings are stable using a wide variety of independentvariables, removing a large number of statistically significant variables from the instrument can

in some specifications cause the monadic democracy variable in the use of force equation to drop below statistical significance Unfortunately, most researchers employing simultaneous equation models in international relations fail to report the goodness of fit for the instruments

CONCLUSIONS

Skeptics of the democratic peace have argued that the empirical literature has discovered

a spurious correlation Democratic polities are not less likely to use force (against other

democracies or in general) Rather, the skeptics claim that democratic states tend to be nurtured

in peaceful environments While this causal story may be true for some democratic states, the simultaneous equation model presented in this paper refutes the skeptics’ argument Even after controlling for the amount of violence in the area, democracies were less likely to use force against other democracies (i.e., support for the dyadic democratic peace) and less likely to use force against other states in general (i.e., support for the monadic democratic peace)