The Democratic Peace Revisited It is Veto Players

The Democratic Peace Revisited: It is Veto Players

George Tsebelis
Anatol Rapoport Collegiate Professor of Political Science
University of Michigan
6759 Haven Hall
505 S. State Street
Ann Arbor, MI 48109-1045
Phone: 734-647-7974
Fax: 734-764-3522
Email:
Seung-Whan Choi
Assistant Professor of Political Science
University of Illinois at Chicago
1007 West Harrison Street (M/C 276)
Chicago, IL 60607-7137
Phone: 312-413-3280
Fax: 312-413-0440
Email:

February 2008


Abstract
This study introduces a variable reflecting veto players into international conflict models. As our
title suggests, the inclusion of this variable dissolves the standard measure of democracy which
currently dominate the literature on the democratic peace. Through the development of a new
conceptual approach, and an empirical study strengthened by two methodological improvements,
we provide an alternative explanation of peace than those which currently exist in the
international relations literature. A cross-sectional, time-series dyadic data analysis during the
period 1885 to 2001 shows that, while controlling for several standard conflict-related variables,

the veto players variable outweighs the democracy variable in explaining the absence of
interstate disputes.

2


One of the major themes of the international relations literature has been the logic of the
“democratic peace” theory. This logic was first formulated by Kant (1795/1957) in Perpetual
Peace. Kant presented the three well-known definitive articles about republic constitutions,
commercial relations, and international organizations and argued that international peace could
be achieved through the establishment of these three elements. During the past three decades,
many studies of international conflict have paid attention to the effect of regime type by arguing
that two democratic states rarely fight each other. In fact, students of the democratic peace
theory initially discovered a strong empirical relationship between democracy and peace and
then began to formulate conceptual explanations by referring to Kant’s treatise on perpetual
peace as their intellectual foundation or introducing formal modeling (e.g., Fearon 1994). While
there were a series of criticisms about particular data points, variables in datasets,
methodological issues in statistical analysis or particular assumptions in formal models, the
evidence has been considered overwhelming: democracies are understood to be less likely to go
to war against each other (whether as an empirical regularity or a validated theory).
Levy (1988, 662) argues that the state of the democratic peace studies is “as close as
anything we have to an empirical law in international relations.” Similarly Gleditsch (1992, 372)
contends that “the perfect correlation between democracy and nonwar in dyads implies that most
behavioral research on conditions for war and peace in the modern world can now be thrown on
the scrap-heap of history, and researchers can start all over again on a new basis.” Finally, Ray
(1998, 27) summarizes that the evidence and the results “that support the democratic peace
proposition warrant confidence in its validity.” Moreover, the multiple streams of arguments and
evidence supporting the democratic peace proposition are highly diverse in character:
epistemological (Rummel 1975), philosophical (Doyle 1986), formal (Bueno de Mesquita and

3


Lalman 1992; Fearon 1994; Bueno de Mesquita and Siverson 1995), historical (Weart 1994; Ray
1995;Owen 1994), experimental (Mintz and Geva 1993), anthropological (Ember, Ember, and
Russett 1992; Crawford 1994), psychological (Kegley and Hermann 1995), economic (Brawley
1993; Weede 1996), political (Gaubatz 1991), and statistical (Ray and Russett 1996). Perhaps,
then, the more defensible of the two possible definitive answers to the question, does democracy
cause peace? is .
In this study we offer an alternative line of reasoning; we contend that deviations from
the peaceful status quo are more difficult when countries have multiple veto players. In doing
this, we step outside the current conceptual framework which takes the characteristics of the
whole political system and particularly the electoral connection between elites and masses as the
main focus of study.1 In contrast, we contend that a more accurate theoretical reasoning should
direct to the function of the “veto players” which are determined by the institutions regulating
the political game in each country. For empirical testing, we first replicate a most standard
democratic peace model and then introduce a veto players variable into the empirical equation.
In addition, by implementing two significant methodological improvements, we re-estimate both
the democratic peace model and the improved model with veto players. First, we control for
repetition of the same potential dyads over time, which have (inaccurately) been considered
independent observations in previous studies. Second, we introduce a second prediction into the
model: we argue that the variance of the predictions is a variable itself (the model is and should
be estimated as heteroskedastic). We argue that with these modifications in place, the reliability
of the empirical model should increase significantly. Our findings suggest that the variable
accounting for the type of regime (more or less democratic) loses any statistical significance in
1

Many empirical studies introduce various statistical models in which the effect of democracy is

measured by Polity datasets (on Polity, see Gurr, Jaggers, and Moore 1989 and 1991).

4


competition with veto players, while the significance of the control variables remains the same.
In addition, due to the two methodological improvements, the estimated coefficients become
more consistent. This does not merely confirm the expectations of Choi (2007) who has argued
that more veto players will reduce international conflict whether one considers democratic dyads,
autocratic dyads, or mixed dyads. Our empirical analysis goes one step further, suggesting that
in competition with the veto players variable in the same model, the effect of the democracy
variable fades away. Veto players are a contributing factor in peaceful coexistence between two
states.
This study is organized in four sections. First, we review the democratic peace literature
with an eye towards the theoretical, empirical, and methodological questions which remain open.
Second, we present our alternative argument for the eruption of war between two countries in a
dyad. Third, we present our research design and empirical results. We conclude by arguing that
our findings are only one of many arguments which converge toward an alternative view of war
and peace.
The Democratic Peace Literature
Many empirical studies have grouped democratic peace arguments into two streams: normative
(cultural) and institutional (structural). Proponents of the normative argument (e.g., Sobek,
Abouharb, Ingram 2006) see the primary causal mechanism engendered by democracy being a
notion of “externalization.” In this view, democratic states externalize the civic and peaceful
norms associated with internal democratic practice to settle claims; democracies are behooved by
their internal commitments to approach external problems in a comparably peaceful manner. We
want to reiterate an important criticism that Rosato (2003) makes to this line of argument. The
democratic peace argument concerns interstate disputes, that is interactions between two states in
dyad. Consequently the empirical tests exclude a number of different violent interactions—
5

colonial wars where one of the parties is not a state, military coups organized by a democracy
(US in Chile), and assassinations of leaders (Partic Lumumba)—which are not “interstate
disputes” strictly speaking. These incidents are significant, however, as they cast serious doubt
on the normative argument. Violent, but non-warring, interactions involving democracies either
show that democracies do not actually externalize their democratic norms, or it shows that an
externalization of norms does not preclude state violence. The fact of violence on the part of
democracies indicates a serious flaw in the reasoning of the normative argument.
In a way the normative argument falls victim to the “one hundred foot woman problem.”
Suppose that there is a theory predicting that no woman would be more than one hundred foot
tall. This theory would be corroborated by empirical evidence all over the world. Suppose now
that in a remote corner somewhere in the planet a ninety nine foot tall woman is discovered. The
proponents of the theory could claim that this is one more case corroborating the theory.
However, for most observers, the one hundred foot limit for female height would become a
doubtable proposition, because in the next village from the new discovery one could imagine that
another female would exceed the limit. In this sense, the fact of democratic violence is like the
ninety-nine foot tall woman; we must question whether the theory barring the occurrence of
democratic war is truly sound. No matter how compelling the normative arguments may seem,
they are like the advice that Machiavelli gave to the Prince: they are not self enforcing and can
be ignored. Normative arguments might aim to sway future foreign policy, and they might
render moral judgments on past or current practices. However, normative theory cannot, by
definition, provide an explanation for what tends and has tended to happen.
In contrast to the normative arguments, there have been a variety of institutional
explanations. The primary causal mechanism underlying the various institutional arguments is

6


“accountability.” This mechanism “derives from the fact that political elites want to remain in
office, that there are opposition parties ready to capitalize on unpopular policies, and that there
are regular opportunities for democratic publics to remove elites who have not acted in their best

interests” (Rosato 2003, 587). The institutional strain of democratic peace theory then primarily
sees democracy as creating a certain class of constraints which, by keeping elected leaders in
check, restrains the war-making ability of democratic states. The institutional argument of the
democratic peace theory has been further refined by formal theorists. According to Fearon
(1994) and Schultz (1998), democratic leaders face tremendous audience costs from concerned
citizens or opposition parties compared to those faced by autocratic leaders; hence democratic
leaders are likely to be cautious about waging war. If two democracies are in a crisis, the level of
uncertainty between them is lower because of the ability to demonstrate resolve efficiently,
making it easier to negotiate peaceful solutions.
It is worth noting that the selectorate theory of Bueno de Mesquita et al. (2003) contends
that regardless of regime type, political survival is the primary motivation and concern for all
political leaders, and that the primary objective of every leader is to maintain their position of
power and to be reselected as a leader when reselection takes place. All political leaders in any
regime are beholden to the same situation—the responsibility of pleasing the specific group of
individuals (the winning coalition) who helped the leader attain his or her post in the first place
and whose support is necessary for the leader to maintain their position.
Further, it has been argued that the normative and institutional arguments of the
democratic peace do not necessarily stand up to historical scrutiny. The Spanish-American war
of 1895–1898, Israel’s invasion of Lebanon in 1982, the India-Pakistan conflict of 1999, and the
Russia-Georgia war of 2008, for example, can be viewed as wars between two democratic

7


countries (see Spiro 1994; Owen 1997). In addition, it appears that democratic leaders often
disregard dissenting voices and escalate the level of conflict. According to Senese’s (1997)
study, once democratic dyads have engaged in disputes, they frequently resort to uses of military
force rather than seek diplomatic solutions. Ironically, despite the fact that democratic oversight
is put into practice to ensure proper foreign policy choices, covert military operations by a
democracy are often carried out against another democracy, producing “hidden victims of the

democratic peace” (James and Mitchell 1995). The bottom line is that there are sufficient
counter-examples of the presumed ubiquitous phenomenon of democratic peace to warrant
critical study; democratic leaders have demonstrated their ability to launch foreign military
assaults despite domestic political constraints.
Aside from explanations and theoretical claims, the datasets used in empirical testing
have been points of contention. Because the datasets in most existing studies only include wars
which have occurred since 1800, there are some wars that have been excluded from analysis.
One immediate example is the omission of the Sicilian Expedition, where Athenians attacked
other democratic cities far away from their own and ignored the advice of Nicias who was
insisting that the enemy was the oligarchic Sparta (excuse this excursion into the classics; one of
the authors is Greek). But this is not the only case of omission of relevant wars among
democracies. One should include the American Revolution, and the war of the US against
France among others, all of which happened before the beginning of the relevant datasets.
In addition, the empirical models should be screened for their implicit assumptions. Most
recent models correct for time dependence in analyzing time-series, cross-sectional data with a
dichotomous dependent variable by employing logit splines, a technique designed by Beck, Katz,
and Tucker (1998), or general estimating equations (GEEs), a method advocated by Zorn (2001).

8


However we want to raise two important problematic features of the empirical models that have
not been corrected for in existing studies. The first implicit assumption has been that
observations are considered to be independent from each other. To illustrate the problem of this
assumption, consider the fact that one datapoint in 1900 contains the dyad of Australia and
Switzerland, and another datapoint contains the same dyad in 1950. We think that these
observations involving the same dyad of countries should not be considered as independent, and
that this assumption biases the tests in favor of statistical significance, independent from other
observations. This correction should decrease the reliability of the existing statistical models.
since every dyad is considered over one hundred observations in terms of time, AND dyads

including a country low in the democratic scale are considered identical (for example, Iran vs
UK and Iran vs France are identical).
The second implicit assumption is that many existing studies measure “democracy” in a
numerical term, and focus on the score of the LESS democratic country. This is done on the
reasonable assumption that the less democratic country would be the weak link of the peace
chain, and thus the main determinant of war. But that means that all the potential conflicts
between, say, Iran and democratic countries will be based on one’s assessment of Iran. In doing
this, the study ignores countries which may be more or less democratic, more or less belligerent,
etc. As a result, when the level of democracy increases in the dataset, the set of potential
adversaries decreases (all dyads that involve less democratic countries have already been counted
as the adversaries of these countries). This observation indicates that the estimated models are
inherently heteroskedastic. One can argue that if the models demonstrate a high level of
reliability despite their heteroskedasticity, then their predictions can be considered more accurate
than what we thought. However, given that heteroskedasticity is predicted, the models make two

9


different predictions at the same time. Thus, if heteroskedasticity is detected in the empirical
estimations, it should count in favor of the model, not against it.
An Alternative Explanation of War
We will follow Tsebelis (2002) in his definition of veto players and policy stability, and apply the
reasoning to peace and war. Tsebelis has argued that every political system can be understood as
combining a number of institutional and partisan veto players, that is individual or collective
actors whose agreement is required for a change of the status quo. As an immediate result of the
definition, it follows that as the number of veto players increases, the likelihood of changing the
status quo diminishes. Tsebelis refers to this difficulty of changing the status quo as “policy
stability”. Tsebelis further analyses the impact of the ideological distance of veto players, and
finds that as the distance between them increases, policy stability increases. He then presents an
argument which assesses the relative importance of number of veto players and ideological

distance. He argues that if the preferences of a set of veto players (set B) is contained within
another (set A), then change in the first system is easier than in the second, regardless of the
number of their respective veto players (see Figure 1).
(Figure 1 about here)
The existence of multiple distinct veto players is a sufficient, but not necessary, condition
for policy stability. Tsebelis’ argument is that multiple veto players with ideological distances
among them make changes to the status quo very unlikely or incremental, while a small number
of veto players make the change possible. Whether this possible change will be materialized or
not, depends on the position of the status quo with respect to the veto players. If they want to
change it, they will, but if the status quo is located in the area surrounding them, they may
choose not to. In short, while the presence of veto players inhibits policy change, the absence of
veto players does not mean that change is probable, just that it is possible.
10


Figure 2 replicates the argument about sufficient but not necessary conditions, and
indicates that the expectation of the veto players theory is twofold: first, the mean of policy
changes will decline with the number of veto players, and second, the variance of the number of
policy changes will decline also with the number of veto players. This is a feature common to
lots of theoretical arguments in the social sciences (no armies, no war; no bourgeois no
democracy) who identify either necessary or sufficient conditions only, and should be tested as
producing joint predictions concerning both the mean and the variance of the phenomenon under
study.
(Figure 2 about here)
Let us now transpose the arguments about policy stability to the issue of war. Two
countries are in peaceful coexistence (whether in amicable or in tense relationship with each
other) until they decide to go to war (when their disagreements increase (***reference), when
there is incomplete information (Fearon 1994), or when there is rapid change in relative power of
one of them (Powell 1999)). The reason for the conflict is immaterial; what is important is that
conflict represents a significant departure from the (peaceful) status quo, which has to be

approved by the veto players of these countries. If we apply Tsebelis’ theory, we can hypothesize
that this change is unlikely with the more veto players each one of these countries has. This
argument should be true regardless of the type of regime (i.e., whether the democratic peace
argument is true or not), or whether other variables (e.g. capabilities, contiguity, superpower
status play, etc.) have any role in the interaction.
Research Design
This section explains model building and discusses the conceptual distinction between veto
players and democracy.
1) Model Building
11


For the empirical testing of the effect of veto players, we first look for a most standard
democratic peace study and then introduce a veto players variable into the research design.
Oneal and Russett’s (2005) study is used as the frame of reference to compare veto players with
democracy while controlling for conflict-related variables, such as economic interdependence
and national capability.2 Since their statistical model has been widely replicated for years and
proven to be highly reliable, the possibility of coding errors or faulty model building on our part
should be reduced. The empirical equation is expressed as follows:
Onset of International Conflictit = α + β1 (Veto Playersit-1) + β2 (Democracyit-1) + β3 (Economic
Interdependenceit-1) + β4 (Capability Ratioit-1) + β5 (Alliesit-1) + β6 (Contiguousit-1) + β7
(Geographic Distanceit-1) + β8 (Major Powerit-1) + β9 (System Sizeit-1) + ε
The equation includes a variable for veto players, a variable for democracy, and seven
independent control variables which commonly appear in studies of international conflict. These
seven control variables are economic interdependence, national capability ratio, allies, contiguity,
geographic distance, major power involvement, and the number of states in the international
system. The choice of these control variables primarily reflects a desire to maintain consistency
with existing studies of international conflict.3 To mitigate problems of reverse causality, all
independent variables are lagged one year.

2

The data set is publicly available at />
3

For a more detailed discussion on a similar model specification and certain debates such as the

inclusion of contiguity and geographic distance in the same model, see Oneal and Russett’s
(2005) recent compelling arguments and findings. We do not include IGOs as a control,
following Oneal and Russett’s (p. 298) approach that “we focus on the effects of the liberal
variables democracy and interdependence because…research on the role of IGOs is rapidly
evolving.”
12


The onset of international conflict is measured in three ways. The first dependent
variable is dichotomized for the onset of a militarized interstate dispute (MID) of any severity.
An MID is here defined as “a set of interactions between or among states involving threats to use
military force, displays of military force, or actual uses of military force” (Gochman and Maoz,
1984, 587; see also Jones, Bremer, and Singer, 1996; Sarkees, 2000). The second dependent
variable is dichotomized for the onset of a fatal MID where at least one soldier is killed per dyadyear. It should be noted that how national leaders respond to military fatalities continues to
stimulate scholarship, so the onset of a fatal militarized dispute is implemented as an additional
check for robustness (e.g., Oneal, Russett, and Berbaum, 2003; Oneal and Russett, 2005). The
third dependent variable is dichotomized for the onset of an interstate war, which is defined as
“a) sustained combat, involving b) regular armed forces on both sides and c) 1,000 battle
fatalities among all of the system members involved” (Sarkees 2000: 125).
We utilize the Henisz (2000) dataset of political constraints as an approximation of veto
players. It is a continuous measure on a scale of 0 (least constrained) to 1 (most constrained).
Tsebelis has noted that this dataset is based to a large extent on institutional veto players, and
does not take into account the ideological distance of the different actors. For this reason

Tsebelis (2002) has created a veto player dataset that measures the different actors in regards to
their policy positions. However, this dataset includes only advanced industrialized countries for
the post war period, so, for our purposes is extremely limiting.
The variable for testing the veto players hypothesis assumes the weak link (see Dixon,
1994), which holds that the state with weaker veto constraints is the stronger determinant of how
things will proceed. The stronger veto constraints that state has, the more constrained it will be
from resorting to military force and, in turn, the dyad will be more peaceful.

13


Democracy also assumes the weak link: the score for the less democratic state in a dyad
is taken to be the stronger determinant of how interactions will proceed. Hence, the more
democratic that state is, the more constrained it will be from initiating a dispute and therefore the
more peaceful the dyad. Economic interdependence also assumes the “weak link”: the score for
the less interdependent state in a dyad is taken to be the stronger determinant of conflict. Hence,
the more interdependent that state is, the more constrained it will be from initiating a dispute and
therefore the more peaceful the dyad. The national capability ratio variable is introduced to
control for power preponderance theory. It is expected that an asymmetric power relationship
creates a favorable environment for peaceful coexistence. The allies variable is included to
account for the argument that military alliance has a dampening effect on conflict, especially
within the bipolar international system. It has been argued that the likelihood of conflict
increases if a major power is involved in a dispute or if dyadic states share a border. Students of
realism view these two factors as important due to their highly detrimental effects on the stability
of the international system (e.g., Russett and Oneal, 2001). To consider such realist expectations,
both major power involvement and contiguity variables are controlled for. Since certain studies
argue that geographical proximity increases dyadic states’ opportunity and willingness to pursue
military adventures, the geographic distance variable is included. To control for the increasing
number of states in the international system, especially during the past 50 years, the system size
variable is introduced.4

4

According to Raknerud and Hegre’s (1997) study, membership in the international system has

expanded over years, so the probability of a dispute for any given pair of non-relevant states has
declined over time. In this context, the dramatic growth in the number of sovereign countries
since World War II should be considered.
14


For the replication analysis, we rely on logistic regression models with peace-years
correction (see Beck, Katz, and Tucker, 1998), as implemented by Oneal and Russett’s (2005)
study. For the robust data analysis, we employ generalized linear latent and mixed models
(GLLAMMs) and maximum-likelihood heteroskedastic probit models. As noted earlier, the
former corrects the dependence of the same dyad of countries, while the latter accounts for
heteroskedasticity.
2) Is the Veto Players Variable related to the Democracy Variable?
As we will demonstrate, when a measure of veto players is incorporated into Oneal and Russett’s
empirical models, the variable of “democracy” loses any statistical significance. As a
preliminary discussion we want to compare the two variables (democracy and veto players), in
order to eliminate the possibility that the two variables are conceptually measuring the same or
similar phenomena.
The standard measure of democracy in empirical models comes from the Polity IV
datasets. Polity IV has an eleven point, additive democracy score, and an eleven point additive
autocracy score. Each ranges from 0 to 10, and subtracting the autocracy score from the
democracy score gives an overall Polity score that can range from full democracy (+10) to full
autocracy (-10). Democracy and autocracy are thus assumed to be opposites on a single
dimension. Although democracy is conceptualized as having five attributes – competitiveness of
participation, regulation of participation, competitiveness of executive recruitment, openness of

executive recruitment, and constraints on the executive, it can be re-categorized into two
(Marshall and Jaggers 2007, 13). One is the presence of institutions and procedures through
which citizens can express effective preferences about alternative policies and leaders. The other

15


is the existence of institutionalized constraints on the exercise of power by the executive. 5 Thus,
conceptually, one half of the Polity index is based on elections and the correspondence between
leaders and the public, and the other half on constraints to the main executive leader. In terms of
conceptual relationship between this index and veto players, the first part is completely
independent, and the second, focuses only on the constraints to the main leader. The role of the
legislature in the process is ignored, the existence of a second chamber is not considered.
Let us study closely an example from the Polity data collection, in order to see the
relationship of this index with veto players. When France moved from having a president and
premier of the same party to its current “cohabitation” between a (popularly elected) president
of one party, and a (legislatively chosen) premier of another party, France’s democracy score
went up (Marshall and Jaggers 2007, 14). Yet, the French president does not have veto power,
and cannot influence the legislative process, as Mitterrand learned when he tried to block
decisions of the Chirac government. When Mitterrand refused to sign governmental decisions
which he disagreed with, Premier Chirac introduced the measures he wanted as legislation,
which Mitterrand had to sign into law (actually, according to the constitution he could send them
back to Parliament for a new decision under the same simple majority requirement, which is not
a veto as commonly understood). By contrast, a veto players score of the French V Republic
does not depend on the agreement or disagreement of the President and the Prime Minister, since
the President has no veto power.6
5

According to Diamond (1999, 9), Polity “acknowledges civil liberties as a major component of

democracy but, because of the paucity of data, does not incorporate them.”
6

Actually, the measure of veto players we are using makes the same mistake: it attributes veto

power to the president on the grounds that he can dissolve Parliament and proclaim an election
but this is not equivalent to veto unless the people replace the parliamentary majority in the
16


Concerning the “constraints on chief executive,” while usually legislatures in western
democracies, they may also be imposed by any “accountability groups.” Such groups might
include “the ruling party in a one-party state; councils of nobles or powerful advisors in
monarchies; the military in coup-prone polities” (Marshall and Jagger 2007, 23). (For an
extensive discussion of this index and surrounding criticisms, see Keech forthcoming). So, the
conceptual relationship between the polity index and veto players is weak (only part of the index
covers issues related to veto players, but only partially (chief executive only) and non-law
making bodies are included in the mix. The correlation between the two indexes is 0.83.
Finally, all the game theoretic literature on democratic peace is based on the first part of
the index, that is, the relationship between the public and its elected officials (audience costs etc);
this is why the polity index was introduced in the empirical tests. Veto players measures have no
conceptual relationship with elections (competitive or not). In these ways, the Polity index, and
therefore the variable of democracy itself, cannot be said to have any conceptual relationship to
the veto players variable.
Empirical Results
Based on logit splines, Table 1 shows the replicated results of Oneal and Russett’s (2005) study
in Models 1, 3 and 5. As expected, democracy is statistically significant at the 0.001 level.
Democracy produces a dampening effect on any kinds of MIDs, fatal MIDs, and interstate wars.
The effects of the control variables are consistent with those reported in Oneal and Russett’s
study. Economic interdependence decreases international conflict across those three models.

When power asymmetric relations between two states in a dyad exist, international conflict is
less likely to occur. A long geographic distance hinders two states from engaging a conflict.
When at least a major power is involved, the likelihood of international conflict increases. As
upcoming election.
17


Oneal and Russett’s study reports, the significance of allies and contiguity is not consistent
across those three models.
(Table 1 about here)
Models 2, 4 and 6 are built to see the competing effect between veto players and
democracy. When the dependent variable is the onset of all MIDs, the veto players variable
makes the pacifying effect of the democracy variable disappear. While the former is significant
at the 0.001 level, the latter is not significant. Despite the fact that fatal MIDs are instead used
for the dependent variable, the same phenomenon is observed, indicating that veto players
outweigh democracy. These findings are consistent with our theoretical expectations: in
explaining peaceful coexistence between two states in a dyad, the function of veto players rather
than the type of regime should be highlighted. When the dependent variable is war occurrence,
both veto players and democracy show a dampening effect. The significance of the control
variables concur with that reported in the replicated models.
Table 2 employs the two improved estimation methods over the logit splines: GLLAMM
and heteroskedastic probit models. While Model 1, 3 and 5 use the former estimation, Model 2,
4 and 6 report the results from the heteroskedastic probit models. The effect of democracy fades
away in competition with veto players in Models 1 and 2, while both democracy and veto players
are not significant in Model 3. When an independent variable enters the mean and the variance
equation in the heteroskedastic probit estimation, the coefficient sign can switch depending on
the regressor values at which the marginal effect is evaluated (Corneliβen, 2005, 9). This makes
the interpretation of the coefficient difficult. Thus, we instead report the marginal effects and
their standard errors in Models 2, 4 and 6. The veto players variable matters across those three

18


models. Interestingly, none of the control variables is significant in a consistent manner across
the table.
(Table 2 about here)
How can we interpret these results? Democratic peace advocates could argue that the veto
players variable is highly correlated with democracy, and therefore what it captures is exactly the
effects of the democracy variable. The reason that it outperforms democracy is purely technical:
the variance of the veto payers variable (with 100 modalities from 0.00 to 1.00 ) is figher than
the variance of the democracy variable (with around 20 modalities). This is why the veto players
variable appears to outperform democracy. In order to test this argument we use a truncated
variable of veto players with 10 modalities only (we use the first decimal of the veto players
variable). Even this reduced variance veto players variable outperforms democracy.
Another argument could be that what we have found here is not an alternative
explanation of democratic peace, but which characteristic of democracy matters the most: the
electoral connection or the institutional structure. According to this argument, the veto players
variable is capturing essentially the same things as constraints to the executive in the Polity
dataset, and therefore the democratic peace argument remains intect, except for its interpretation
that in this case, it is not the electoral connection that defines democracy but it is the institutional
structure of the government. We considered democracies defined by only the electoral
connectionelements of the index, and we then introduced the veto players variable. In this case
also veto players replaces the electoral connection variables. This is a similar argument with the
one that a series of studies have made in the past: They have found out that democracy can be
replaced by freedom of the press or other variables that are conceptually correlated with
democracy but for some reason are better correlated with the occurrence of war.

19

Up to this point, we have not made a strong case as to which one of the two
interpretations is more likely. We will address this point in the conclusions.
Conclusions
The democratic peace literature is vast and spans over hundreds of years. While it has been
challenged in terms of its data (exclusion of older democratic wars), its models, and its statistical
coherence, its explanations and conclusions have remained quite compelling. We have provided
an alternative explanation of international conflict, based on veto players theory. In this
perspective wars are seen as departures from a peaceful situation (the status quo), and as such,
will be less likely to occur when the more veto players are involved in the government of the
conflicting countries. Furthermore, there can not be said to be any link between the common
operationalization of democracy and the number of veto players. This constitutes a large
departure from the democratic peace literature, so, we want to come back now to the big picture.
One other major theme in the international relations literature is the negative relationship
between economic interdependence and war. Our model is finding negative coefficients for
economic interdependence: in this respect it replicates the finding of a series of previous models.
Countries that have economic connections are less likely to go to warMost recently, another
research program was developed by Mansfield, Milner, and Pevehouse (2008), who investigate
further the conditions of deepening of economic interdependence (and therefore, indirectly
conditions that solidify peace). They divide economic agreements from free trade agreements to
economic and political unions. They find the agreements that represent significant departures
from the status quo and deepen significantly economic relations can be signed by countries with
few veto players, but are unlikely to be signed by countries with many veto players. The
empirical results of this research program are schematically presented below.
(Figure 3 about here)
20


As the number of veto players increases, the probability of signing agreements deepening
economic relationships decreases. Let us put together the two findings to understand the big
picture: Countries are interacting in a peaceful status quo mode. Whether they want to deviate

from this status quo to one direction (war) or the other (economic interdependence and
deepening of peace) the role of veto players is negative. They prevent significant departures
from the status quo.
(Figure 4 about here)
Figure 4, combines Figures 2 and 3, in order to show how these findings fit together.
In addition, Mansfield, Milner, and Pevehouse (2008)(unlike us) find a significant
independent role of regime type on the signature of deep economic agreements: democracies are
more likely to sign such agreements. The finding that democracy has an independent effect along
with veto players in terms of signing agreements deepening peaceful ties among nations, but not
in terms of war engagement can be the subject of further investigations. For the time being we
will highlight the implication that the variables “democracy” and “veto players” (used by both
Mansfield et al and us and with the same variables to capture their impact) are sufficiently
independent to produce statistically significant effects for both variables in other studies but not
in ours.
In conclusion, we use the variable veto players which is correlated with democracy and
outperforms it in explaining the (non) occurrence of war. The correlation is empirically high, but
conceptually only weak and only on the average: while authoritarian regimes have less veto
players than democratic ones, high number of vps in neither a necessary (Baath regimes) nor a
sufficient condition (UK) for democracy. While we have not excluded the possibility of using
democratic peace arguments with focus on the institutions of a country as opposed to the

21


electoral connection, we have provided a much simpler explanation which has to do with the
inability of multiple veto players to depart from the status quo whether it is towards war or
towards deepening peace.

22

FIGURE 1

Veto players A1-A3 produce more policy stability than B1-B5
(no matter where the status quo is)

A1

B1
SQ

B5

W(B)

W(A)

B2
B4
A2

B3

A3

23


FIGURE 2

Departure to war as a function of VPs

PEACEFUL COEXISTENCE

WAR
MANY VPS

FEW VPs

24


25