Professional Documents
Culture Documents
Gledistch & Hegre 1997 PDF
Gledistch & Hegre 1997 PDF
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
Sage Publications, Inc. is collaborating with JSTOR to digitize, preserve and extend access to The Journal of
Conflict Resolution.
http://www.jstor.org
Peace and Democracy
THREE LEVELS OF ANALYSIS
NILS PETTERGLEDITSCH
HAVARDHEGRE
InternationalPeace ResearchInstitute,Oslo (PRIO)
Departmentof Sociology and Political Science,
Norwegian Universityof Science and Technology
Peace and regime type can be examinedat the dyadic, nation,and system levels. At the dyadic level, it
is well establishedthatdemocraciesrarelyif ever fight each other.At the nationallevel, the broadconsensus
is that there is no significantrelationshipbetween democracyand war participation,but this conclusion
remainscontroversial.At the systemlevel, therehasbeenlittleresearch;most scholarshave takenforgranted
thatthe answer can be inferredfrom the findingsat the dyadic or nationallevels. The authorsshow that,if
the conventionalwisdom holds at the dyadicandnationallevels, the probabilityof warin a politicallymixed
dyad mustbe higherthanthe probabilityof warbetweentwo nondemocracies,andthe relationshipbetween
democracyandwaratthe systemlevel mustbe parabolic.Thusincreasingdemocratizationinitiallyproduces
more war, and the reductionof war startsonly at a higherlevel of democratization.
THREE QUESTIONS
AUTHORS' NOTE: Some of the Correlatesof Wardata and the Polity data used in this articlewere
madeavailablefromthe InteruniversityConsortiumof PoliticalandSocial Research(ICPSR),eitherdirectly
or throughthe Norwegian Social Science Data Archives (NSD). Neitherthese institutionsnor those who
originallygeneratedthe dataare responsiblefor our use of them. For commentson earlierversions of the
article, we are gratefulto ClarkAbt, StuartBremer,Scott Gates, WarwickMcKibbin,Sara McLaughlin,
Arvid Raknerud,RudolphRummel,Bruce Russett,Roslyn Simowitz, and ErichWeede, as well as to Dan
Smith and other colleagues at PRIO. John Wilken Aschehoug provided technical assistance with the
manuscript.Finally,we would like to express our gratitudeto the FridtjofNansen Foundationfor Science
and the Humanities,the NorwegianMinistryof Defense, and the NorwegianResearchCouncil (NFR) for
financialsupport.
JOURNALOF CONFLICTRESOLUTION,Vol. 41 No. 2, April 1997 283-310
? 1997 Sage Publications,Inc.
283
284 JOURNALOF CONFLICTRESOLUTION
The prevailingopinion appearsto answerthese three questions yes, no, and yes.
The first question has been extensively researched,with clear results. The second
question has also been analyzed a great deal, with conflicting results and with the
prevailingopinionleaning in the directionof a no, but now shiftingin the directionof
a perhaps.The thirdquestionhas rarelybeen subjectedto empiricalinvestigation,but
it is commonlyassumedthatit can be answeredby a simple deductionfrom one of the
two otherlevels. The most commonconclusionis thatif democraciesdo not fight each
other,an increasingnumberof democraciesin the systemwill producea morepeaceful
system. Othershave arguedthatif democraciesare as war prone as nondemocracies,
it makesno differenceat the systemlevel if the numberof democraciesincreases.Both
of these system-level statementscannot be true at the same time, so there must be
somethingwrong eitherwith the deductionsor with the empiricalregularities.
In this articlewe firstconfirmthe democraticpeace at the dyadic level andthe lack
of a clear relationshipat the nationallevel, using several measuresof conflict. Next
we show that, given the conventionalwisdom-that democracieshardly ever fight
each other but overall participatein war as much as other countries-it follows
logically that the probabilityof war in a politically mixed dyad must be higher than
the probabilityof war between two nondemocracies,and the relationshipbetween
peace anddemocracyat the system level mustbe bell-shaped(i.e., parabolic).Finally,
we look briefly at the empiricalevidence at the system level. Althoughdemocracyis
clearly relevantfor subnationalconflict, we do not examinethose effects here.1
RESEARCH DESIGN
SPATIALAND TEMPORALDOMAIN
deaths(Small and Singer 1982; Singerand Small 1994). Ourtwo otherdatasets have
a lower thresholdon violence. The militarizedinterstatedisputesdatafor 1816 to 1892
(Singer and Small 1994) includeall interstateconflicts with the use or threatof force.
The Uppsaladata(Wallensteenand Sollenberg1996) includeall armedconflicts with
more than 25 dead in a given year.3
DEMOCRACY
We use the most recentlycorrectedversion of the Polity III data set generatedby
Ted Gurrandassociates(Gurr,Jaggers,andMoore 1990; Jaggersand Gurr1995), the
only such dataset to cover the full spatialandtemporaldomainof the COWdata.4The
Polity data set includes 172 currentand historical countries. In terms of units of
analysis,the overlapbetweenthe Polity andCOWdatais very high (Gleditsch 1995a,
306). However, the Polity set of countriesis somewhat smaller than COW's. Thus
some country years and dyad years included here have no Polity data at all. For
simplicity, we have merged this category with the Polity codes of interruption,
interregnum,transition,and missing data(Gurr,Jaggers,and Moore 1989, 6-8).
For the measurementof democracy,we first compute for each country year the
difference between the Polity III indices of institutionalizeddemocracyand institu-
tionalized autocracy.If DEMOC-AUTOCis 3 or higher, we define the country as
democratic.By using the differencebetween the two scales, we avoid categorizing
ambiguousregimes as democracies.Primeexamplesof countrieswith a high score on
the two indices areJapanfrom 1868 to 1944 (DEMOC= 5, AUTOC= 4) andGermany
from 1908 to 1917 (DEMOC= 5, AUTOC= 3).
The DEMOC and AUTOC indexes in Polity are additiveindexes rangingfrom 0
to 10. Combiningthem(DEMOC-AUTOC)yields an additiveindex rangingfrom-10
to 10, combining assessmentsof the competitivenessof political participation(-2 to
3), regulationof politicalparticipation(-2 to 0), competitivenessof executive recruit-
ment (-2 to 2), openness of executive recruitment(-1 to 1), and constraintson the
chief executive (-3 to 4). The cutoff at 3 is fairly arbitrarybut is set to give roughly
the same proportionof democraciesas in previous studies with a cutoff of 6 on the
democracyscale alone. Ourdemocracyindex may be validatedintuitivelyby consid-
ering values for a few selected countries:The United States exceeds 3 for the entire
time span, Germanyachieves it duringthe WeimarRepublic (1919-1932) and again
from 1949, Russia is coded with a 4 in 1917 but did not reach this level again until
1991, andFranceis classified as a democracyfrom 1848 to 1950, from 1877 to 1939,
and from 1946 onward.
3. The Uppsala conflicts have been updateduntil 1995, but we can go only as far as the Polity data
have been updated(i.e., to 1994). The dataarepublishedby conflict.We have coded all countriesat opposite
sides in a war as being opponents.The Uppsaladatarequirethat at least one governmentto be among the
contendingparties.Whereveranothergovernmentis listed on the opposingside, we have coded the conflict
as international.
4. The correctedversion of the data set is availableby anonymousftp from <isere.colorado.edu.>.
Relative to the version publicized in Jaggers and Gurr (1995), the new version has corrected a small
summationerrorin the scores for democracyand the autocracyscores and a few other errorsthat barely
affect the statisticalresults.
286 JOURNALOF CONFLICTRESOLUTION
DO DEMOCRACIES USUALLY
MAINTAIN PEACE WITH EACH OTHER?
The evidence from previous studies is straightforward: there are few if any wars
between democracies.Some disagreementremainsas to whetherthe relationshipis
merely very strong (e.g., Weede 1992, 382) or virtuallywithout exception (Rummel
1983, 1995; Ray 1993).
Table 1 summarizesthe evidence for the entire 179-yearperiod. The dataindicate
thatthe relativefrequencyof war between two democraciesis abouttwo fifths of the
relative frequency of war between two nondemocracies.Politically mixed dyads
(democracy/nondemocracy)have an even higher relative frequency of war than
nondemocraticdyads.
Although this is a strong and highly significant relationship,it is by no means
perfect.Table2 lists the exceptions.The anomalouscases of warbetweendemocracies
comprise30 dyadyears.No less than24 of these aremadeup by Finlandversusvarious
Westerndemocraciesin WorldWarII. These cases could be interpretedas "derived
war," resulting from the change of sides of a major actor (the Soviet Union) in a
three-waycontest.It would notbe unreasonableto modify the theoryof the democratic
peace to incorporatesuch majorshifts in a multipolarwaras one of the circumstances
underwhich small democraciesmight unwittinglyfind themselves at war with other
democracies.However,it seems moreappropriate to dismissthese shiftsas a weakness
in our measurementof dyadic war data because there was no war action at all, even
where therewas a formaldeclarationof war.5
Two anomalousdyadyears occurbecausethe time variablein the Polity dataset is
too coarse. The 1971 BangladeshWarbetweenIndiaandPakistanwas precededby a
stateof emergencyin Pakistan.Similarly,theTurko-CypriotWarin 1974 was preceded
by a GreekCypriotmilitarycoup, instigatedby a militaryregimein Greece.Five days
later, Turkey responded by an invasion, which divided the island and, ironically,
broughtdown the Greekcolonels' regime. Thus these dyadyears should be classified
as wars between a democracyand a nondemocracy.These anomalies occur because
regime changes in Polity are coded by year ratherthanby date.
Of the remainingfour anomalousdyad years, the Lithuanian-PolishWarof 1919
was consideredtoo smallto be includedin earlierversionsof the COWset of interstate
wars6and thereforehas not turnedup in earlierlists (e.g., Gleditsch 1993, 313). It is
includedin the newest versionof the COWdataset butas a marginalwar (1,000 battle
deaths) between two brand-newdemocracies.Because Lithuania'sconstitutionwas
not adopteduntil 1922 (The Baltic States 1991, 180), one might question the Polity
coding of Lithuaniaas a democracyfrom 1918 (cf. also Weart1994). Coding Spain
5. The United States resisted Soviet pressureto declarewar on Finland,so the United States-Finland
dyads are particularlyinappropriateas war dyads. The senior authorhas dealt more extensively with the
case of Finlandelsewhere (Gleditsch 1993), as has Ray (1993, 271). The discussion in Spiro (1994, 61-2)
is somewhatmisleading.He chargesFinlandwith having"pursuedan alliancewith fascists and ... declared
war on democracies"(it was, of course,Englandthatdeclaredwaron Finland).Spiroalso countsEngland's
attackon Germanshippingin a Finnishharboras an attackon Finlandmore than4 monthsbefore England
declaredwar on Finland.
6. In Small and Singer(1982, 338) the LithuanianPolish Waris listed as a warbut with differentdates
(1920-1927) and is excluded because it did not meet the battlecasualtiesthreshold.For 1919 they also list
(p. 337) a Polish-UkrainianWar,excluded for the same reason.
TABLE 1
DemocracyandDyadic Relationshipsin War,1816-1994 (perc
MissingRegime
Two One No Data or
Typeof Relationship Democracies Democracy Democracies RegimeTransitio
0oo
*sj
288 JOURNALOF CONFLICTRESOLUTION
TABLE2
AnomalousCases: WarbetweenDemocracies, 1816-1994
as a democracyin 1898 has been questionedby Ray (1993). Coding Syriain 1948 and
Pakistanin 1965 as democraciesis also debatable.None of these four deviant cases
concernsstable or establisheddemocracies.
Obviously, one should be careful aboutreclassifyingdeviant cases without reex-
amining other cases. Temporalmismatchesand fictitious dyadic opposition in mul-
tipolarwars may occur among nondemocraciesand politically mixed dyads as well.
Moreover,correctingfor temporalmismatchmay conceivablyyield new warsbetween
democracies. But even in the absence of such systematic reconsideration,Table 1
confirmsthe very strongdyadicrelationshipbetweendemocracyandpeace, andgiven
our caveats aboutthe data,it is consistentwith the idea of a near-perfectrelationship.
Although democraciesrarely,if ever, fight each other,they are more frequently
allied in war.An averagepairof democraciesis allied in war more thanfour times as
frequentlyas the averagepair of nondemocracies.
How far can we lower the violence thresholdwhile retainingthe strong dyadic
relationshipbetween democracyand peace? Weede (1992, 380), using data with a
thresholdof 100 dead,foundno militaryconflictsbetweendemocraciesbetween 1962
and 1974. Likewise, no major military interventions(i.e., claiming more than 100
lives) betweendemocraciesarementionedin the dataset generatedby Tillema(1991).
If we look at all the interventionsin this dataset, the relationshipis no longerperfect.
But there are few such interventionsbetween democracies, and "almost all were
symbolic and short-termuses of armed force" with little loss of life (Kegley and
Hermann1996, 319). Studiesusing militarizedinterstatedisputes(MIDs) (Gochman
and Maoz 1984) find some conflict between democracies(Maoz and Russett 1992,
Table2, 254) but not for the highestcategoryof MIDs ("wardisputes")andfewer than
for othercombinationsof regime types.7
7. Manyof the MIDs betweendemocraciesarefisheriesdisputes(e.g., the Cod Warsbetween Iceland
and its neighbors).In such conflicts, the threator use of force is usually acted out betweenthe government
on one side and a privatefishing vessel on the other.The intergovernmental interactionis generallylimited
to diplomaticexchanges, and it is questionablewhether such conflicts have any place in a data set on
interstatedisputes.
Gleditsch,Hegre /PEACE AND DEMOCRACY 289
Apartfrom the higher risk of coding errorin the MID data, it is unreasonableto
expectjoint democracyto eliminateall militarizedconflict down to the level of force
found, for instance, in the Cod Wars.The theory of the democraticpeace does not
assume thatjoint democracywill eliminate all conflict, and we should expect some
conflicts to develop militaryovertones.But we shouldalso assumethatthe nonviolent
normsof democracieswill interveneto preventfurtherescalation.
In the secondline of Table3, we test the dyadicrelationshipfor the Uppsalaconflict
datafor the post-ColdWarperiod (1989-1994). For this dataset, more inclusive than
the COW data but not as inclusive as the MID data, we find a single case of armed
conflict between two democracies(Indiaand Pakistanin 1989) and little difference
between nondemocraticpairs and the mixed dyads. This is, in a sense, a stronger
finding thanthe one in Table 1 because the violence thresholdis lower. On the other
hand,the time span for the Uppsaladatais much shorterthanfor the COW data.The
thirdline of the tabletests the dyadicrelationshipon the disputesdata,where we also
find the least conflict for double democraticdyads, althoughthe relationshipis not
nearlyas strongas for interstatewar.Thusboth datasets providesupportingevidence
for the dyadic democraticpeace.
Just as we have asked whetherthe dyadic democraticpeace holds at lower levels
of violence, we may ask if the relationshipvaries with differinglevels of democracy.
If we lower the threshold of democracy minus autocracyto zero, the ratio of war
incidence among democraciesto that among nondemocraciesincreases from 2:5 to
more than 2:3. Conversely,if we raise the level of democracyminus autocracyfrom
3 to 8, we eliminateall wars betweendemocraciesexcept Finlandversus the Western
democraciesin 1944. This is anothercase of temporalmismatchbetween Polity and
COW because Finland'schange to a high level of democracyoccurredafter the end
of the war.Thus, at this level of democracy,the correlationmay be perfect.
Empiricalfindings such as those in Tables 1 and3 have frequentlybeen questioned
on the basis that the dyad years do not representindependentobservations.If two
countriesare at war in year t, the chancesaremuchbetterthatthey will remainat war
in year (t + 1) than that two new countrieswill go to war. Similarly,once a conflict
has brokenout betweencountriesa andb, thereis a higherprobabilitythatthe conflict
will spreadto countryc (particularlyif this is a neighboringor allied country)thanfor
an entirely new conflict to startbetween c and d. We may call these two forms of
dependencybetween the units dependenceon the past and simultaneousdependence.
To eliminate this problem, Bremer (1992, 320) limits his investigationto the dyad
years thatoccuron the first day of a war,arguingthat"thequestionof how warsbegin
is fundamentallydifferentfrom the questionsof why wars grow in size, duration,or
severity."But this is not a questionthatcan be settled a priori.Duringa war,decision
makers are constantlyforced to reexamineits costs, and a decision to stay in a war
ratherthangive up or withdrawfromconquestmay be a resultof the same forces that
made war break out in the first place. Bremer's approachhas the disadvantageof
reducinglong wars with many participantsto a single dyadic observationor, at most,
a few, if severalcountriesenterthe war on the first day.ThusWorldWarII is reduced
to one dyad year,Poland-Germanyin 1939. Anotherproblemwith Bremer'swork is
that his censoring is inconsistent: he eliminates dependent cases of war but not
t)
\C
TABLE3
DemocracyandDyadic ArmedConflict (percentageof dy
Missing Regime
Two One No Data or
Typeof Anned Conflict Democracies Democracy Democracies Regime Transition Tota
dependentcases of peace. His analysis lumps dyads continuingat war and dyads in
the process of joining an ongoing war with dyads at peace.
In a companionarticle from our project,Raknerudand Hegre (1997) tackle this
problemin a radicallydifferentway by modellingthe interstatedyad as a continuous
process (cf. also Beck and Tucker 1996). This leads to results that also confirm the
democraticpeace while seeing it in connection with war diffusion and recurrence.
Here, we choose a simpler approach,by comparingthe results obtainedin Tables 1
and 3 with the resultsof an analysisin which we have eliminatedthe unitdependency
for war (butnot for peace). The assumptionis thatif we find the same relationshipfor
the incidence of war and the onset of war,our confidencein the results will increase.
In Table4, we have cross-tabulateddemocracywith onsets of dyadic war and onsets
of war. The second half of the table correspondsto Bremer's radical reduction of
conflict dyads to new conflicts only, but the first half of the table is a less drastic
solution in which all new conflict dyads are countedin their first year.Although the
frequencyof dyadic war onsets is naturallylower than the dyadic incidence of war,
and the frequencyof war onsets even lower, Table4 confirmsthat war occurs much
more rarelyin jointly democraticdyads. The numberof anomalouscases is reduced,
mainly because the Finnishwar dyads from 1941 to 1944 are eliminated.
The dyadic relationshipbetween democracy and peace has been subjected to
various tests of thirdvariables(Maoz and Russett 1993; Oneal et al. 1996; Bremer
1992, 1993; Gleditsch 1995a); no evidence has been found for considering the
relationshipspurious.Indeed, if the relationshipbetween democracyand peace was
perfect, tests for spuriousnesswould be superfluous-unless a control variablewas
proposedthatin itself had a perfectrelationshipto the dependentandthe independent
variable.Even if the relationshipis just very strongratherthanperfect,the searchfor
single thirdvariablesseems unpromising.
MissingRegime
Two One No Data or
Typeof ArmedConflict Democracies Democracy Democracies RegimeTransition Tota
TABLE5
Democracyand ArmedConflict:The National-Level
Relationship(percentageof countryyears with conflict)
Missing Numberof
RegimeData Conflict All
No or Regime Dyad Dyad
Typeof ArmedConflict Democracy Democracy Transition Years Years X2 p
Incidenceof conflict
Interstatewar, 1816-1994 5.4 5.3 6.6 639 5.5 .04 .85
Armed conflict, 1989-1994 4.8 4.5 1.8 43 4.2 .03 .86
Militarizeddisputes, 1816-1992 33.6 32.9 22.8 3,632 32.1 .61 .43
Onset of new dyadic conflict
Interstatewar, 1816-1994 2.1 2.3 2.6 266 2.3 .78 .38
Armed conflict, 1989-1994 3.9 2.8 1.2 31 3.0 .85 .36
Militarizeddisputes, 1816-1992 28.0 26.3 16.9 2,928 25.9 3.45 .06
Onset of new conflict
Interstatewar, 1816-1994 1.1 1.8 1.6 182 1.6 5.92 .02
Armed conflict, 1989-1994 .7 1.0 .6 8 .8 .33 .56
Militarizeddisputes, 1816-1992 24.6 26.0 14.5 2,718 24.0 2.20 .14
NOTE: Conflict dataand democracydataas in Tables 1 and 3.
Although Rousseau et al. (1996, 526) found the previous evidence in supportof the
conventionalwisdom at the nationallevel "actuallyquite thin,"they also concluded
that the evidence for the dyadic thesis was much stronger.The bulk of the large-n
studies agrees with Chan (1984), who found that "relativelyfree" countriespartici-
pated in warjust as much as the "less free"-6.7% versus 6.1%, respectively,of all
country years between 1816 and 1980.8
Table 5 gives our national-levelresults for the correctedPolity III data using the
threeindicatorsof conflict. The same problemof dependencybetweenunits occurs at
the nationallevel. Therefore,we have computeddatafor the incidence of conflict as
well as for the two forms of onset. The frequencyof participationin waror militarized
interstatedisputes (whether measuredby incidence or onset) is not very different
between democraciesand nondemocracies.However,in an analysis dividing the war
data by time periods (not reproducedhere), we found that during the cold war,
democraciesparticipatedsignificantly less frequentlyin war than nondemocracies.
This patternappearsto continueinto the post-ColdWarperiod.
It mightbe suspectedthatthe "norelationship"findingwas dependenton the cutoff
for the level of democracy.Perhapsa positive relationshipbetween democracyand
peace would emergeif the requirementfor democracywere more stringent.This idea
was testedwith a negativeoutcome.At least for the periodas a whole, thereis no clear
trend in the war participationof democraciesrelative to nondemocracieswhen the
level of democracyis variedsystematically.
8. For extrasystemic wars (i.e., colonial and imperial wars), democratic countries were at war
significantlymorefrequently.However,this figureexaggeratesthe relativewarparticipationof democracies
because nondemocraticopponentsin colonial wars are not countedas separateactors.
294 JOURNALOF CONFLICTRESOLUTION
war. Rummel (1979) did not posit a correlationbetween freedom and the frequency
of involvement in war because free states by their very example represented a
subversive challenge to authoritarianand totalitariansystems. Therefore,libertarian
states would have to engage in defensive andreactiveviolence againstattemptsfrom
nonlibertarianstates to change the statusquo.
This idea was testedby Small and Singer(1976, 66), who foundthatin the 19 wars
in which democraciesparticipated,they initiated(or were on the side of the initiator)
in 58%. This result should have been adjustedfor the numberof democraciesin the
system, but this involved more work collecting democracy data than they were
preparedto undertakeatthetime.Becauseat all timesthereweremorenondemocracies
than democracies,nondemocraciesshould be expected to initiate more wars. When
the oppositeis found,it appearsthatdemocraciesareeven moreproneto initiatewars
than Small and Singer's resultsindicate.
Table6 lists all warsbetween 1816 and 1994 involving democraciesfrom the start
of the war.The tableincludes30 of the 75 interstatewarsin the Small-Singerdataset.
Twenty-two of these, or 73%, were initiated by a democracy.To determine war
initiationis a difficult coding task because it dependson identifyingthe countrythat
crosses the decisive borderlinein a processof escalation.In some cases, the threshold
is obvious, as when GermanyattackedPolandin 1939. In othercases, with a protracted
process of escalation in many small steps, identifying the initiatormay requirean
arbitrarycutoff. Moreover,the possibility of preventivewar muddiesthe waters.If A
intendsto attackB, andB strikesfirst to preventit, B will be the initiatorin Small and
Singer's (1976) terms, but in discussing nations' peacefulness, it may be just as
reasonable(or unreasonable)to identifythe otherpartyas the aggressor.
When the Small and Singer(1976) list of initiatorsis examinedmore closely, such
problemsemergequite clearly.Looking at some of the wars initiatedby democracies,
we find several protractedhigh-tension disputes (India vs. Pakistanin the Second
KashmirWarin 1965, Israelvs. its neighborsin the 1956 and 1967 wars,and Syriavs.
Israelin 1948) in which mutualthoughtsof preemptionmust have been so pervasive
that the coding of war initiationbecomes highly suspect. In four other wars (United
Statesvs. Vietnamin 1965, Indiavs. Pakistanover whatbecameBangladeshin 1971,
Turkeyvs. Cyprusin 1974, andArmeniavs. Azerbaijanin 1991) an interstatewarwas
initiatedonly afterviolence had alreadystartedin the form of civil war, government
massacre, or a coup d'etat. Thus all eight wars initiated by democracies in the
post-WorldWarII period seem ratherirrelevantto determiningthe peacefulness of
democracies.We have not analyzedin similardetail earlierwars initiatedby democ-
racies, but the data from the post-1945 period alone cast considerabledoubt on the
notion thatdemocraciesare as war prone as nondemocracies.At least in the modern
era, democracieswould appearto initiate violence very rarely,except in protracted
conflicts;but if violence has startedin some form,they arenot averseto interveningor
escalatingthe disputeto the point where it can be settledby superiorforce. The prior
existenceof violencemay serveas a justificationfor the interventionof democracies.
The six most violentinterstatewarsin theentireCOWperiod-World WarI (1914),
World War II (1939), the Sino-JapaneseWar (1937), the Korean War (1950), the
VietnamWar(1965), and the Iran-IraqWar(1980)-were all initiatedby nondemoc-
296 JOURNALOF CONFLICTRESOLUTION
TABLE6
Democracyand the Initiationof War,1816-1994
Initiator(s) Target(s)
Starting Democracy Democracy
Year War Level Country Level Country
TABLE6 Continued
NOTE: m.d. = missing data. In the 1969 Israeli-EgyptianWar,Israel (9) and Egypt (-7) participated,but
COW has no informationon initiation.This table includes all interstatewars in the Correlatesof Wardata
set involvingat least one democracyfromthe start.SmallandSinger(1982) have coded the initiatorvariable
only up to 1980. For the remainingyears,we have used Singer'supdateof the war datato 1992 (Singer and
Small 1994), the Wallensteenand Sollenberg(1996) data on wars between 1993 and 1994, and our own
coding of the initiator.Three wars had to be eliminatedfrom the COW list when the presenttable was
compiled. In the Israeli-EgyptianWar(1969-1970) and the Vietnamese-Cambodian War(1975-1980), no
initiatoris named.The war called RomanRepublic(1849), accordingto the COWdataset, was initiatedby
France,which was not one of the originalparties(Two Sicilies vs. Austria-Hungary). This makes no sense,
so this war has also been left out. Of these three, only the first involved a democracy.Note that the two
worldwars have also been excluded.Althoughthey involved severaldemocracies,the originalinitiatorand
victim (Austria-Hungaryvs. Serbia in WorldWar I and Germanyvs. Poland in WorldWarII) were not
democracieswhen the wars broke out. The same goes for the Crimeanwar, the FirstBalkanWar,and the
KoreanWar.Warsin bold printare those initiatedby democracies.Democracyscores are from Polity III.
a. Greece and Bulgarianot coded by COWas initiatorsbut as participatingon the initiator'sside from the
first day of the war.
racies, although mostly with other nondemocraciesas the initial victims. When
democraciesbecame involved, as they did in four of these wars,theyjoined the target
ratherthanthe initiator.This addsforceto the contentionthatif democraciesparticipate
at all, they tend to be on the reactiveside, at least in majorwars.
This is not tantamountto saying that democracies are more peaceful. Such a
statementwould require a more detailed analysis of the patternsof escalation and
considerationof a widerset of wars.For instance,manyextrasystemicwarshave been
initiatedby democraciesengagedin colonialconquest.Second, in the post-WorldWar
II periodthereappearto havebeen manymoremilitaryinterventionsabroadconducted
by democratic (Western)countries than by the Soviet Union and its allies. Some
interventionshave been justified with reference to stopping domestic violence or
promotingdemocracy,but others are more commonly interpretedas power politics.
Third,majorpowers, includinglargedemocracies,may fight war throughproxies.
Finally, what aboutcontrolsfor thirdvariables?Thattask is much more urgentat
the nationallevel thanat the dyadic level becausewe arenot dealing with a perfector
near-perfectrelationship.Many of the third variables controlled for at the dyadic
level-for instance,those testedby Bremer(1992)-could be translatedto the national
level. But few if any studies controlfor thirdvariablesin a convincing manner.10
7dnation
dyad= N 1 (1)
itnation tnation
- to
dyad
48-1 111 - 1
DYADICVERSUSSYSTEMLEVEL
At the system level, we are most interestedin the proportionof the system's
countriesin war outbreakin a year.For a start,we assume thatthe probabilityof war
between two democracies tDD is 0. Then the frequency of war is a function of JND, 7eNN,
d, and N, the number of countries in the system. Clearly, if tND < 1NN, the replacement
of a dyad of two nondemocraciesby a mixed dyad of one democracy and one
nondemocracymust involve a decreasein the frequencyof war.Therefore,as long as
lND < tNN (as we have found empirically to be the case for the onset of new interstate
war and for all measures of armed conflict at a lower threshold of violence), the
frequencyof war declines monotonicallywith increasingd. In otherwords, the more
democraciesin the system, the less war.
On the other hand, if ND > tNN (as we have found empirically to be the case for the
incidenceof war,the onset of new dyadicinterstatewar,andall measuresof militarized
Hegre/PEACEANDDEMOCRACY301
Gleditsch,
0% 10% 20% 30% 40/o 50% 60% 70% 80% 90% 100%
d
-..... Democracies --Non-Democracies
We see that if cDD = 0 and d = 1, n,ation has to be 0. The formula can also be
expressed as
lnation = Nd2(7tDD- 2lND + 7CNN)+ d(-TDD + 2NtND- 2N7NN + 7LNN)+ NtNN - CNN. (6)
tion and the frequencyof war at the system level. Thus, with increasingdemocratiza-
tion over time, we should expect the frequencyof war to increase initially and only
decline when reachinga breakpoint. The derivativeof (6) with respectto d is the rate
of change of nationwhen d increases:
dnation(d)
= 2dN(DD - 2 ND + 7NN)+ (-tDD + 2NnND- 2NnNN+ 7NN). (7)
dd
The maximizerof nat,on(d) is the value of d for which (7) equals0 (given values for
In the specialandnotveryrealisticcase of 7NN= 0 andnDD = 0 (i.e.,
the otherparameters).
no war among similarregimes), the frequencyof warpeaks for d = .5, at which point
half the dyads are mixed; hence any furtherdemocratizationmust replace a mixed
dyad with a pure dyad of one sort or the other. For our numerical example, the
maximizer is d = .36.
These points are illustratedin Figure2, where we have plotted7nationas a function
of d in a our subset of countries,using expression 6. If the observedfrequenciesare
representative,initial democratizationin the internationalsystem is followed by a
slightly increasingfrequencyof war until 36% are democratized,and then it startsto
decline.
It can be derivedfrom (7) thatthe maximizerof 7nation(d) is independentof N when
we assume that the dyadic probabilities CND, 7tNN, and 7DD are constants. In other words,
the expected frequency of war at the system level peaks for the same share of
democraciesregardlessof the size of the system.But, as arguedin theprevioussection,
the assumptionof constantdyadic probabilitiesis untenable:both cnationand ldyad are
dependenton the size of the system.Therelationshipbetweenthe levels is inextricably
tied to the size of the system, makingstrictempiricaltesting of our argumentdifficult.
The fact thatN has been constantlyexpandingin the time framecovered probably
does not alter the relationshipbetween the levels shown here, althoughwe have not
worked out the formalrelationship.In the absence of such an extendedformula,it is
thus necessaryto limit the analysisin some way to keepN roughlyconstantif we want
to test ourpropositions.We mightdo this by applyingsome variationof the politically
relevantdyads.Elsewhere,we (RaknerudandHegre 1997) proposea flexible version
of this procedurein which the irrelevantdyads are weighted down instead of being
deleted. Anotherway to bypass the problemis to confine the empiricalanalysis to a
regionthatis expandingonly moderately(suchas Europe)or to a fixed set of countries
as in our numericalexample.1 In Table7, the 140-yeartime span has been divided
into fourperiodsof 35 yearseach for the 30-countrysample.The shareof democracies
d and the observedshareof nationyears with the onset of a new militarizedinterstate
dispute has been computedfor each period. In the final column, we reportthe 7nation
predictedby expression(5), given the observedproportionsat the dyadic level. These
predictedfigures correspondwith Figure2.
Although double democraticdyads have a considerablylower proportionof dyad
years with new disputes(.0063 vs. .0105 and .0158), the predictedprobabilitiesat the
11. We conductedan analysis similar to the one reportedin Table 7 for all Europeancountrieswith
similarresults.
Gleditsch,Hegre/PEACE AND DEMOCRACY 303
25 %
20%
15%
10%
5%
0%
0% 10%0 20% 30% 40%0 500/ 60% 70% 80% 90% 100%
d
TABLE7
Observedand PredictedProbabilitiesof MIDs
at the NationalLevel in a Set of 30 Countries
70% -
60 % -
50% -
40 % -
30% -
20 % -
10%
0%
o so ' '0
so %o o0 ' ' 0 '0 0 '0 '0 '0 > 'I
so '0 '
0 r- 00 oo0as \c0 Yr-a o 0
o oo o' o o
Year
Figure 3: Relative Number of Democracies in the World and Incidence of War, 1816 to
1994 (%)
NOTE: Data as in Table 1. Wardata accumulatedby countryand by decades. Countriesin transitionor
withoutcodes in Polity have been excluded.
in the level of democracyin the 1920s and 1930s. Earlierperiodsof war accompanied
periods of democratization.All this is broadly consistent with the hypothesized
curvilinearpatternif we fix the turningpoint between 1910 and 1920.
A similar approachis taken in Figure 4, where the two variables are plotted on
separateaxes. Here the frequencyof war in the system (measuredby the percentage
of countryyears at war) is plottedagainstthe degree of democracy(measuredby the
percentageof countryyears accountedfor by democraticcountries)for each of six
time periods.The firstfive timeperiodseachcover 35 years,butthe final periodcovers
only the postcoldwarera.Figure4 shows thatfor the first 100 years,the worldbecame
more democraticand also more war prone. After the world wars, it became more
democraticand peaceful. However, in examining the changes between neighboring
periods, we find too little war in the thirdperiodand too little democracyin the fifth
for the patternto be completely consistentwith our theoreticalexpectations.Never-
theless, the curve as a whole may (with some imagination)be characterizedas
parabolic,whereincreasingdemocratization,as expected,is firstassociatedwith more
war,then with less.
Although it is difficult to predicttheoreticallywhere the breakpoint might occur,
we can simulate it. If we assume nND and nNNtobe constant over time and equal to the
frequencyof war in mixed andpurenondemocracydyads over the entiretime period,
using the actualfigures for d (the fractionof democracies),we get a predictedpattern
of war thatpeaks in the late 1920s and generallydeclines from then on. Because war
peakedonly 20 years laterandthe late 1920s were in fact quite a peacefulperiod,this
might seem to be wide of the mark.But if we regardthe two world wars as essentially
one conflict (not an unreasonableassumptionin view of the fact that the actors and
conflict lines were largely the same, at least in Europe),the midpointof that conflict
lies somewherein the late 1920s and early 1930s.
306 JOURNALOF CONFLICTRESOLUTION
12% -
10%-
) 8%-
6%-
4%-
2% I I
0% 1 . I I I I I I- 1
0% 5% 10% 15% 20% 25% 30% 35% 40% 45%
Percentof Country-YearsDemocratic
It is tempting to suggest that some kind of system shift has occurred, perhaps at the
end of the long European conflict known as the two world wars or perhaps at the end
of the cold war. This shift could be related to the fact that democratization has reached
such a high level-at least in certain regions-that there is a lack of opportunities for
war. The level of democracy in Europe is now higher than ever before. Because Europe
has accounted for so much of the world's war in the previous periods (Gleditsch
1995b), this could explain the recent decline in interstate war.
In one of the few studies at the system level,14 Maoz and Abdolali (1989) tested
regime type (democratic, anocratic, autocratic) against the occurrence of militarized
interstate disputes between 1816 and 1976. This study is not so relevant here because
it posits a positive relationship between democracy and peace at the national and
dyadic levels. Obviously, then, a simple monotonic relationship must follow at the
system level, although it would appear that these authors also think that the system-
level hypothesis is a logical extension of the dyadic-level hypothesis alone. The tests
reported in the empirical part of the article generally reflect the same lack of attention
to possible curvilinear relationships. They first report that the proportion of double
democracies is positively related to system conflict (Maoz and Abdolali 1989, 26) and
that this held even when corrected for autocorrelation (p. 27), but the proportion of
double democracies had a negative effect on the number of wars, although only a small
proportion of the variance was accounted for (p. 27). When broken down into two
subperiods, the relationship between the degree of democracy and conflict was found
14. Anotherimportantstudyat the system level (Maoz 1996) deals morewith systemic changesand its
causes thanwith consequencesof regime type measuredat the system level.
Gleditsch,Hegre /PEACE AND DEMOCRACY 307
to be differentin the 19th and 20th centuries.This is reminiscentof the famous early
findings by Singer and associates of differences between the 19th and the 20th
centurieswith regardto the influence of alliances and capabilitydistribution.15 Such
findings are unsatisfactorybecause "century" is not a theoreticalcategory. When the
difference between centuriesis interpretedas a question of crossing a thresholdof
democratizationin the internationalsystem,the shiftin therelationshipto warbecomes
theoreticallymeaningful,butthe shiftis unlikelyto follow the calendarquiteso neatly.
What about controlsfor thirdvariablesat the system level? The issue has hardly
been touchedin the literature.Clearly,the empiricalpatternfound in Figures 3 and 4
might be very differentif we had incorporatedthe influence of other variables.The
"shrinkingworld"might be one such variable,measuredfor instance by the time it
takes to travel between two randomly picked members of the interstate system.
Because there is more war between neighboringand proximatestates (Bremer1992;
Gleditsch 1995a),we mightexpect a higherfrequencyof waras countriescome closer
to each otherin termsof the time andcost expendedin interaction.This mightoutweigh
the effect of democratizationor influencethe relativesize of the probabilities7ND and
nNN, moving the breakpoint at which democratizationstartsto producepeace at the
system level. Once again, we might take all of Bremer's third-variabletests and
translatethemto the system level. To date,therehas been little theoreticalor empirical
researchof this kind.16
SUMMARY
15. Singer and Small (1968) found thata high numberof alliancestendedto be associatedwith peace
at the system level in the 19thcenturybut with warin the 20th century.Singer,Bremer,and Stuckey (1972)
found a balance-of-powermodel of peace to fit the 19thcentury,but a power preponderancemodel seemed
more suitablefor the 20th century.
16. For a first attempt,see McLaughlin(1996), who finds more supportfor a linearthan a curvilinear
relationshipbetweendemocracyandpeace atthe systemiclevel butcautionsthatatthis stage of herresearch,
inferenceis difficult.
308 JOURNALOF CONFLICTRESOLUTION
regions, may now have passed througha system shift in crossinga thresholdvalue for
democratization.
If this (admittedlysomewhatspeculative)conclusionis correct,furtherdemocrati-
zation should continue to lower the probabilityof war, at least in regions where
democracyis at a reasonablyhigh level. As noted, democraciestend to win the wars
in which they participate(Stam 1996), andunsuccessfulconductof a waris frequently
punished with a violent regime change (Bueno de Mesquita, Siverson, and Woller
1992). Losers seem morelikely to imitatethe winner,so the net outcomewill probably
be furtherdemocratization.The experienceof the two world wars and the end of the
cold war confirmsthis expectation.Although,in the shortterm,participationin war
is likely to underminedemocracy(e.g., throughrestrictionson freedom of speech or
thepostponingof elections),warwouldseem to promotemoredemocracyin the longer
run,and-at least above a certainlevel-more democracyin turnleads to a reduction
of war. Thus the optimism of the democraticpeace literaturewould seem warranted
in the long run,but on the basis of a somewhatmorecomplexreasoningthanordinarily
assumed.Of course,if the idea gains currencythatwarmay be pursuedas a deliberate
strategyof democratization,the world could be in for a transitoryunpeacefulperiod.
REFERENCES
The Baltic states: A reference book. 1991. Tallinn:Estonian Encyclopaedia Publishers;Riga: Latvian
EncyclopaediaPublishers;Vilnius:LithuanianEncyclopaediaPublishers.
Banks, A. S., and R. B. Textor.1963. A cross-politysurvey.Cambridge:MIT Press.
Beck, N., andR. Tucker.1996. Conflictin spaceandtime:Time-series-Cross-section analysiswith a binary
dependentvariable.Paperpresentedatthe annualmeetingof theAmericanPoliticalScienceAssociation,
29 August-1 September,San Francisco.
Benoit, K. 1996. Democracies really are more pacific (in general). Journal of Conflict Resolution 40
(4):636-57.
Brecher,Michael, JonathanWilkenfeld,and Sheila Moser. 1988. Crises in the twentiethcentury: Vol. 1.
Handbookof internationalcrises. Oxford,UK: Pergamon.
Bremer,S. A. 1992. Dangerousdyads: Conditionsaffecting the likelihood of interstatewar, 1816-1965.
Journalof ConflictResolution36:309-41.
.1993. Democracyandmilitarizedinterstateconflict,1816-1965.InternationalInteractions18:231-49.
Bueno de Mesquita,B., R. M. Siverson,and G. Woller.1992. Warand the fate of regimes:A comparative
analysis.AmericanPolitical Science Review 86:638-46.
Chan, S. 1984. Mirror,mirroron the wall.... Are the freer countriesmore pacific? Journal of Conflict
Resolution28:617-48.
Doyle, M. W. 1983a. Kant, liberal legacies, and foreign affairs, part 1. Philosophy and Public Affairs
12:205-35.
. 1983b. Kant,liberallegacies, and foreignaffairs,part2. Philosophyand PublicAffairs 12:323-53.
EllingsenT.,andN. P.Gleditsch.Forthcoming.Democracyandarmedconflictin the ThirdWorld.In Causes
of conflictin the ThirdWorld,editedby D. Smithand K. Volden.Oslo, Norway:North-SouthCoalition
and InternationalPeace ResearchInstitute.
Enterline,A. J. 1996. Driving while democratizing.InternationalSecurity20:183-96.
Galtung,J. 1996. Democracy:dictatorship= peace: war?In Peace by peaceful means: Peace and conflict,
development,and civilization,49-59. London:Sage.
Gleditsch,Hegre /PEACE AND DEMOCRACY 309
Gleditsch, N. P. 1993. Democracy and peace: Good news for humanrights advocates.In Broadeningthe
frontiers of human rights: Essays in honour of AsbjOrnEide, edited by D. Gomien, 283-306. Oslo,
Norway: ScandinavianUniversityPress.
.1995a. Geography,democracy,and peace. InternationalInteractions20:297-323.
. 1995b. Democracy and the future of the Europeanpeace. EuropeanJournal of International
Relations 1:539-71.
Gochman, C. S., and Z. Maoz. 1984. Militarized interstatedisputes 1816-1976. Journal of Conflict
Resolution28:585-616.
Gurr,T. R., K. Jaggers,and W. H. Moore. 1989. Polity II codebook.Boulder,CO: Universityof Colorado,
Departmentof Political Science.
.1990. The transformation of theWesternstate:Thegrowthof democracy,autocracy,andstatepower
since 1800. Studies in ComparativeInternationalDevelopment25:73-108. [Also in On measuring
democracy: Its consequences and concomitants,edited by A. Inkeles, 69-104. New Brunswick, NJ:
TransactionBooks, 1991.]
Haas, M. 1965. Societal approachesto the study of war.Journalof Peace Research2:307-23.
Jaggers,K., andT. R. Gurr.1995. Trackingdemocracy'sthirdwave with the PolityIIIdata.Journalof Peace
Research32:469-82.
Keesing'scontemporaryarchives. 1989. Cambridge,UK: Longman.
Kegley, C., and M. Hermann.1996. How democraciesuse intervention:A neglected dimensionin studies
of the democraticpeace. Journalof Peace Research33:309-22.
Mansfield, E. D., and J. Snyder. 1995. Democratizationand the danger of war. InternationalSecurity
20:5-38.
- 1996. The effects of democratizationon war.InternationalSecurity20:196-207.
Maoz, Z. 1996. Domestic sources of global change. Ann Arbor:Universityof MichiganPress.
Maoz, Z., and N. Abdolali. 1989. Regime types and internationalconflict, 1816-1976. Journal of Conflict
Resolution33:3-35.
Maoz, Z., and B. Russett. 1992. Alliance, contiguity,wealth, and political stability:Is the lack of conflict
between democraciesa statisticalartifact?InternationalInteractions17:245-67.
. 1993. Normativeand structuralcauses of democraticpeace. AmericanPolitical Science Review
87:624-38.
McLaughlin,S. 1996. Endogeneityandthe democraticpeace. Paperpresentedat the annualmeeting of the
Peace Science Society (International),25-27 October,Houston,TX.
Muller,EdwardN., and ErichWeede. 1990. Crossnationalvariationin politicalviolence: A rationalaction
approach.Journalof ConflictResolution34 (4):624-51.
Oneal, J. R., F H. Oneal, Z. Maoz, and B. Russett. 1996. The liberalpeace: Interdependence,democracy,
and internationalconflict, 1950-86. Journalof Peace Research33:11-28.
Raknerud,A., and H. Hegre. 1997. The hazardof war:Reassessing the evidence of the democraticpeace.
Journal of Peace Research34:forthcoming.
Ray, J. L. 1993. "Wars"between democracies:Rareor non-existent?InternationalInteractions18:251-76.
- 1995. Democracyand internationalconflict.Columbus:Universityof South CarolinaPress.
Rousseau, D. L., C. Gelpi, D. Reiter,and P. K. Huth. 1996. Assessing the dyadic natureof the democratic
peace, 1918-88. AmericanPolitical Science Review90:512-33.
Rummel,R. J. 1979. War,power,peace. Beverly Hills, CA: Sage.
.1983. Libertarianismand internationalviolence. Journalof ConflictResolution27:27-71.
.1994. Focus on: Power, genocide and mass murder.Journalof Peace Research31:1-10.
. 1995. Democracies ARE less warlike than other regimes. EuropeanJournal of International
Relations 1:457-79.
Schj01lset,A. 1996. Are some democracies more peaceful than others? A suggestion for extending the
democraticpeace perspective.Cand. polit. thesis, NorwegianUniversity of Science and Technology,
Trondheim.
Sen, A. 1994. Libertyand poverty:Politicalrightsand economics. The New Republic210:31-37.
310 JOURNALOF CONFLICTRESOLUTION