HomeDocuments on SitePersonalDemocratic PeaceDemocide20th C. DemocideMegamurderersLesser MurderersWhy DemocideDimensionsConflictMethodsTheoryPolicyLinks PHOTOS OF DEMOCIDEGalleries

Other Democratic Peace Documents On This Site


What is the "democratic peace"?

"Waging denuclearization and social justice through democracy"

"The rule of law: towards eliminating war"

"Freedom of the press--A Way to Global Peace"

"Convocation Speech"

Freeman Interview

City Times Interview


Bibliography on Democracy and War

Q & A On Democracies Not Making War on Each Other

But What About...?

"The democratic peace: a new idea?"


"Libertarianism and International Violence"

"Libertarianism, Violence Within States, and the Polarity Principle"

"Libertarian Propositions on Violence Within and Between Nations: A Test Against Published Research Results"


Vol. 2: The Conflict Helix (see Chapter 35)

Vol. 4: War, Power, Peace (see e.g., Propositions 16.11 and 16.27

Statistics of Democide

The Miracle That Is Freedom

Power Kills


By R.J. Rummel


The clear consensus in the literature is that democracies are no less or more warlike than other types of regimes. But when the core studies upon which this consensus is based are looked at closely, they imply in fact that democracies are less warlike. Regardless of this, these studies, as well as most others on this question, are based on a frequency count of wars, which gives the same count of one for a nation that lost a few dozen killed and for another that lost several million. I argue that a better theoretical interpretation of a regime's warlikeness is in terms of the severity of war, and that in these terms the degree to which a regime is democratic is inversely correlated with the severity of its wars, 1900-87. A survey of the published research further substantiates that democracies tend to fight less severe wars than other regimes.

While a consensus has grown that democracies don't make war on each other,1 a second consensus has developed in parallel that democracies are neither more nor less likely to make war or commit violence than other types of regimes. Just note the following quotes, which exemplify this agreement.

"Statistics can hardly be invoked to show that democracies have been less often involved in war than autocracies." (Wright, 1942: 841)
"Liberal states are as aggressive and war prone as any other form of government or society in their relations with nonliberal states." (Doyle, 1983: 225)
"And Kant was clearly wrong in his presumption that democracies are inherently peaceful." (Dixon, 1994: 1)
"[R]esearch on this question has been near unanimous in finding that democracies are as war prone as nondemocratic states. Only Rummel has dissented from this result...." (Starr, 1992: 43)
"[T]he results of most studies indicate that democracies are no less war-prone than other forms of government...." (Morgan and Campbell, 1991: 188)
"As the empirical evidence grew that democracies are just as warlike as other states, even if in a somewhat perplexing way, the idea arose that, nevertheless, there may be characteristics associated with democracies that tend to discourage war. An early attempt to identify such a characteristic was Rummel's (1983 ["Libertarianism and International Violence"]) argument that the political freedom of individuals promotes peace. But Rummel's empirical results were subsequently shown to lack generality...." (Kilgour, 1991: 267)
"... democracies fight about as much as nondemocracies...." (Geva, et al., 1993: 217)
"... how are we to explain the fact that [democracies] wage war with other types of states with as much frequency and vigor as do nondemocratic regimes...." (Bueno de Mesquita and Lalman, 1992: 146)
"In my reading of the evidence ... there is little difference in the war involvement of democracies and other regimes." (Weede, 1992: 377)

The thesis that the democratization of systems of rule eliminates the most important cause of war collides with many empirical findings.... [T]here is a long chain of articles denying a relationship between democracy and nonviolence. Rummel's is the significant exception." (Czempiel, 1992: 264) In spite of this consensus, it does not well reflect the evidence. As I try to show here, a careful reading of the studies underlying this consensus and of my own 1983 ("Libertarianism and International Violence") that is assumed to be the only exception (actually there are many more, which I cite below) shows that democracies are in fact the most pacific of regimes. Moreover, an analysis of the methodology of the core research studies that underlie this consensus further supports this conclusion.

To begin with my "exceptional findings", in Vol. 4: War, Power, Peace offered the Freedom Proposition that "[T]he more libertarian a state, the less it tends to be involved in violence."

A prior question has to do with the kind of violence limited by libertarian [democratic] systems. Libertarian systems are the natural enemies of authoritarian and totalitarian states. By their example and the products of freedom they are naturally subversive of authoritarian or totalitarian systems; and these freedoms seem to make libertarian states defenseless against unilateral changes in the status quo. Thus, libertarian states are often involved in reactive and defensive violence against the initiatives of nonlibertarian states. Therefore in general, I do not expect that there will be a correlation (on which, see Understanding Correlation) between libertarianism and the frequency [note: the frequency] of involvement in war or violence. Nor should there be for the conflict behavior variables. The predicted correlations for these variables are therefore random....

However, once a libertarian state is involved, domestic forces will usually begin to coalesce against increased violence and for a settlement of some sort. The growth in anti-Vietnam war sentiment and its impact on the American leadership's war policies and decisions are a paradigm case of [this proposition]. It follows that the intensity of violence variable (which measures the scope, occurrence, and degree of violence) and the conflict scale (which has intense violence at the extreme) should be negatively correlated with libertarianism [democracy]
----(Vol. 4: War, Power, Peace: 292-3, italics added)

There are two things to note about this quote. One is that it emphasizes the severity of violence as the crucial variable; and, second, it throws out the frequency of war involvements or other violence as a relevant variable, predicting that the correlation between democracy and the frequency of foreign violence should be random. Ironically, this zero or near zero correlation that I predicted is in fact what allegedly has been found by the subsequent studies underlying the consensus that democracies are no less or more violent than other types of regimes, to which my positive findings on severity are supposed to be an exception. Indeed, I also had found through factor analyses and correlational studies that there was little correlation between a dimension of foreign conflict and violence and a dimension of democratic versus authoritarian and totalitarian regimes (Rummel, 1968, 1972, 1979b).

In Vol. 4: War, Power, Peace I then looked at all the quantitative studies I could find, published and unpublished, to see if in fact the Freedom Proposition held up against findings in the literature on the severity of foreign violence and democracy. I found 23 directly relevant analyses and after considering their methods, scope and assumptions, I concluded that on balance they supported the Proposition.

But I was not entirely satisfied with so relying on the literature. Moreover, there were a number of negative studies that, although in the minority or relatively of minor scope, concerned me. Also, none of these studies had really focused on the relationship between democracy and the severity of its wars, although they had arrived at relevant findings in the process of researching other questions.

So I decided to do an intensive research study, the now well-cited "exception" to the consensus, the 1983 "Libertarianism and International Violence" study. The data I used in this study were those I had collected on all foreign conflict for all nations, 1976 to 1980. Those for a democracy scale were from the Freedom House ratings of all nations on civil and political rights and liberties. With these data and an appropriate scaling of conflict by its level of violence, I found that the highest levels of violence significantly increased as the level of freedom (for free, partially free and non-free regimes) decreased.2 This was strong confirmation of the Proposition.

After completion of this work and another on democracy and domestic violence ("Libertarianism, Violence Within States, and the Polarity Principle"), I tried to re-evaluate all the empirical studies relevant to this Proposition that I had surveyed in Vol. 4: War, Power, Peace, and those done since or which had subsequently come to my attention ("Libertarian Propositions on Violence Within and Between Nations: A Test Against Published Research Results"). This included those by Small and Singer (1976), Erich Weede (1984) and Steve Chan (1984) that had specifically directed their research to this Proposition. Again I evaluated the importance, methodology and relevance of the studies, quantitatively weighed them in terms of these considerations, and concluded that on balance the results supported the Proposition, although not robustly.3

I will deal in detail with the Small and Singer, Chan, and Weede studies and show that their findings do not support the consensus mainly based upon them. But first I present the results of again empirically testing this Proposition, using the popular Small and Singer (1982) data. The Proposition, which I call the Freedom/Foreign Violence Proposition to distinguish it from others,4 is:

The More Democratic a Regime, the Less its Foreign Violence

To test this and show that my results are not due to the peculiarities of my own data set, I first used the most commonly employed data on war in this area--those of Small and Singer (1982)--in spite of their problems, to be subsequently described. They define war as any military action in which there are 1,000 or more battle dead and provide figures on battle dead for each participant in a war. It is these battle-dead data that I employed to operationalize foreign violence, since, as should be clear from the above, it is severity and not frequency of war that the theory predicts (the more democratic a regime and the more deadly a potential war, the more domestic and psychological restraint a leader will have to go to war5).

The sample consisted of all state regimes participating in the Small-Singer wars 1900-80. Each was classified as to whether democratic (= 0), authoritarian (= 1), or totalitarian (= 2). For this I largely relied on the Small and Singer (1976) classification of democracies and Gurr's Polity II classification (1990) to make these distinctions. I rated a regime as totalitarian based on how absolute was a regime's power and the degree to which it commanded virtually every aspect of society, as did the Soviet and communist Chinese regimes. All communist regimes were classified as totalitarian, as was Hitler's Germany. The underlying theoretical variable was the freedom of people to govern their own lives as reflected in their civil liberties and political rights. In democracies people ordinarily have the most freedom, under totalitarian regimes the least.

Table 1 gives the comparison of means, 1900-80, for the battle dead for each type of state regime.6 There is here a clear step function--the less democratic the regime, the higher its mean battle dead.

The Table also presents the comparison of means for battle dead as a percentage of the regime's population. This is a theoretically less important measure than that of battle dead itself. For democratic people and interest groups, as well as the governing elite, that a war may cost thousands of dead, or is in fact causing hundreds of deaths per week, is the more salient factor--not that a certain percentage of the population is being killed. Indeed--whether in the US pre-Pearl Harbor debate about coming actively to the aid of Great Britain (whose defeat appeared imminent), or in the great domestic debate about ending the Korean or Vietnam wars, or in the debate over launching military action against Iraq after its invasion of Kuwait--no one, not at least according to my resources, phrased the concern about casualties in terms of the number of US citizens as a percentage of the population that would be or were being killed. Nonetheless, this is a favorite indicator among researchers and is included in Table 1 for that reason. Figure 1 plots the means listed in the Table.

As can be seen from the Figure, there is a clear increase in means as we move from democratic through authoritarian to totalitarian regimes. While the means for the battle dead are too close for democratic and authoritarian regimes to be significant, those for both regimes differ very significantly from the totalitarian ones.7 Moreover, the differences among the battle dead means overall are significant as well.8 For the percentage of population means, however, none of the comparisons among the means is significant and the analysis of variance is barely significant, if at all.9 Still, the means increase as expected, and this, along with the highly significant and more pertinent results for battle dead, gives clear empirical support to the Freedom/Foreign Violence Proposition.

I also tested the Proposition using a second set of data more to my liking. These I collected in the process of doing research on democide (genocide and mass murder) (Death By Government, Statistics of Democide). For all 141 regimes committing any kind of democide in this century I surveyed a wide variety of estimates of their war dead in international violence and determined a total that best accorded with the quality and central tendency of the estimates. Among these 141 demociders, 77 were involved in some sort of violent international conflict.

I then selected 73 additional regimes that did not commit democide, that reflected major regional and cultural patterns, and that involved large differences in type of regime from previous or succeeding ones. For the foreign violence dead of these non-demociders, I again relied on the Singer and Small data, supplemented by my own data for the years 1981-87. This therefore gives us a sample of 214 regimes, 129 of which have been involved in some sort of foreign violence (the sample of 214 in this case equaling that tested above for the Small and Singer war data is entirely coincidental).10 The sample is about half of all regimes existing during the period 1900-87.11

Table 2 and Figure 2 give the results. Those for the most relevant means between the types of regimes show the proper direction and are about as significant as was the case for the Small and Singer war data.12 Those for the means of percent population killed, however, are less favorable, showing no difference between the means for democratic and authoritarian regimes, although both differ in the proper direction from totalitarian ones. This notwithstanding, the means of those killed in battle are as they should be, and provide more support for the Proposition.

It may be that there is some underlying factor responsible for this relationship between democracy and violence. To check on this, I divided states into three different levels of wealth (economic development) in accordance with their energy consumption per capita: those below the global mean, those between the mean and one standard deviation above, and those above one standard deviation. Although the 214 regimes in my sample exist for different periods throughout this century, the data on wealth for all regimes were for 1981.13 I chose to do it this way so that the regimes were comparable. If I had compared the different regimes on their mid-period development, this would have meant that a country like Kaiser's Germany, among the wealthiest in 1914, would be among the bottom third compared to present regimes. Indeed, all the major powers that fought wars in the early part of the century would, taking their contemporary data, be considered among the poor and developing states today. To compare regimes in the context of their relative wealth at the time I have therefore fixed the measure of wealth at one point in time. I should also point out that energy consumption per capita has been found by factor analysis to be among the best indicators of the wealth of nations (Rummel, 1972, 1979b--on factor analysis, see "Understanding Factor Analysis").

With this understood, for the selected sample of 214 regimes with or without foreign violence, Table 3 subclassifies regime types by level of wealth (economic development), mean battle dead, and the mean of battle dead as a percentage of the population. Figure 3 shows the plot of these regime means at each level of wealth. One can see immediately that the relationship between regime type and the severity of its foreign violence--the central theoretical indicator here--is independent of its wealth, as it should be. Moreover, also in the proper direction are six of nine differences within each level of wealth for the mean of battle dead as a percentage of the population, with wealthy nations most departing from the Proposition. This does not question the Proposition, however, since the theoretical expectation is that it is the absolute number, not the proportion, of one's potential battle dead that most restrains democracies from war. And this is shown in the perfect consistency with this theory of the table's mean differences for battle dead across levels of wealth.

Now, more important, national attributes are more highly correlated with a nation's wealth than with any other dimension among nations. As shown by factor analyses (a method for determining the independent patterns of intercorrelation. among, in this case, national attributes--see "Understanding Factor Analysis"), it is a nation's economy (measured by national income per capita or energy consumption per capita) that accounts for its differences in educational levels, economic modernization, transportation web, health statistics, urbanization, science and technology, international transactions, diplomatic involvement abroad, membership in international organizations, and, of course, trade.14 The wealth available to its regime being therefore independent of the relationship of its democraticness to battle dead is a critical test. However, there is another dimension that many would consider even more important, that of physical power or, what I call here, capability.

For describing a nation per se, because of the number and quality of attributes correlated with the dimension, its capability to exercise power is the second most important dimension. This dimension reflects such attributes as a nation's population, gross national product, defense expenditures, military personnel, energy resources and consumption, number of treaties, density of international communications, and sheer exports.15 A nation's geographic size is also very highly correlated with, and thus indexed by, this dimension. Even in considering the role of capability in the behavior of nations, it is primarily their level of wealth that accounts for the greatest variety of behavior, while that of capability is most related to their conflict.16

Based on the findings of factor analysis, I measured capability by the amount of energy consumption in a nation, and for this used 1980 data.17 The argument for using current data here is the same as that given for wealth, perhaps even with greater force. I divided nations into low, mid and high capability just as I did on the wealth indicator: by below the mean, between the mean and the mean plus one standard deviation and above this. The resulting classification of regime type by its level of capability is listed in Table 4, and Figure 4 shows the associated plots. The results are all as they should be for mean battle dead: regardless of their capability (or power, GNP, size), democracies have the least foreign violence.

As with wealth, in Table 4 six of the nine differences in mean dead as percentages of the population are in the right direction, with the most capable nations departing from the Proposition. Considering also the results for wealth in Table 3, it is among the Big Powers that the correlation between democracy and the proportion of their population killed breaks down. This simply shows that, for the Big Powers, realpolitik dominates over proportional human costs. However, when it comes to the absolute costs of violence, as shown by the mean battle dead differences in Tables 3 and 4, even for Big Powers, realpolitik for democracies is secondary.

Up to this point, when measured properly and without exception, the Proposition has been well supported, contrary to the consensus that democracies are no more or not less warlike that other types of regime. But all this is empirical. There are also theoretical/methodological and logical reasons for questioning the consensus. Aside from my own tests, virtually all other tests of this Proposition have been based on frequency counts of foreign violence or war, resulting in the low correlations I had predicted, as previously quoted. Let us examine in more detail what the use of frequencies has entailed.

As mentioned, the most often used data in testing the correlation between regime type and war have been those on war published by Small and Singer (1982) and the related Gochman and Maoz militarized dispute data (1984). In both cases there must be at least 1,000 battle dead for a war to be defined as such. But a state is classified as a participant if first,

... the nation must be a qualified member of the interstate system. Second, it must have had regular, uniformed, national military personnel in sustained combat. Third, no matter how brief or lengthy that combat, these forces must have either numbered at least 1,000 or sustained at least 100 battle-connected deaths. Fourth, a nation need not have been, either formally or physically, at war with all of the nations on the opposing side in order to be classed as an active participant.
----Small and Singer, 1982: 67.18

This means that a regime coded as having a war can have virtually no one killed in the war as long as it had over 1,000 troops involved; or that a state suffering 800 killed in a violent confrontation in which the overall toll is 950 would not be counted as having fought a war. For example, in the Boxer Rebellion, classified as a war by Small and Singer since there were over 3,000 battle dead, Great Britain lost 34 killed, the United States 21, and France 24 (Small and Singer, 1982: 87). Yet this was classified as a war for each of these nations. Then consider the Falklands War of 1982 between Great Britain and Argentina. Figures vary on the number killed, but somewhat less than 1,000 seems a good number, with about 650 to 700 of those being Argentinians. But by virtue of the Small-Singer criterion that to be a war a 1,000 battle-dcad threshold has to be reached, the Falklands War would not be a war they would include, and thus, in spite of her high number killed compared to Great Britain, the USA and France, in the Boxer rebellion this would not be counted as a war for Argentina.19

The problem with this simplistic count of wars for a regime can be seen in another way. Counting wars or military actions equates conflicts that are vastly different. For example, according to Small and Singer the Philippines lost 90 killed in the Korean War (1982: 92), and this is counted as a war for the Philippines because there were more than 1,000 troops involved. But in the Small and Singer tables (1982: 91), the Soviet Union lost 7,500,000 battle dead in World War II, and this also is counted as one war. Thus, in comparing the democraticness of regimes and their use of force, if we measure force by a frequency count of wars, then Great Britain in the Boxer Rebellion, the Philippines in the Korean War, and the USSR in World War II are treated as equally using force, since each gets a count of one for war, even although Great Britain lost only 34 in combat, the Philippines 90, and the Soviet Union over 7,000,000. Yet, such frequency counts of wars or the use of force have been the main way the Propositions on democracy and violence have been tested by others.

Consider also that whatever we theorize to be the underlying conditions inhibiting or preventing democracies and near democracies from violence, to my knowledge no one argues that democracies are equally inhibited from using force in a conflict in which the expectation is of losing a dozen or so soldiers versus engaging in a total war in which the loss of millions may be suffered. But this is the theoretical implication of the use of a simple count of wars.

With all this as background, we can look at the studies most often cited as showing that democracies are as warlike as other regimes. Among these is certainly that by Melvin Small and J. David Singer (1976), which followed up the Dean Babst (1964, 1972) finding of no war between democracies. Although Babst's findings were dyadic and Small and Singer do present a list of wars, participants, and whether they were democratic or not, virtually all the statistical analyses are directed at the wars by, not between, democracies. In one table they show that wars involving democracies are not significantly shorter or longer than those fought by non-democracies. But I know of no theory arguing that wars involving democracies should be shorter.

Small and Singer do, however, have a table of battle deaths for democratic and non-democratic participants (1816-1965). This tests the Proposition properly in terms of severity. They find the mean battle deaths for all interstate wars to be 91,937 for democracies and 167,270 for nondemocracies (1976: Table 6), or almost 82% higher. This difference of means is comparable to what I found above for severity. However, they then apply a significance test (t-test) to this, which is incorrectly calculated--the corrected test gives odds of near 7 to 1 or greater against this result being by chance. They also give the mean battle dead excluding the two World Wars, which is 5,059 killed for democracies versus 38,436 for non-democracies. Here also their significance test is wrong; the correct one would make this highly significant--odds of 500 to 1 or more against this being a chance result.20 On balance, these results support the Proposition, but have been almost universally cited in line with Small and Singer's mistaken conclusion that democracies are no more nor no less warlike than nondemocracies.21

A second highly cited study on this Proposition is that by Steve Chan (1984). Using the Small and Singer war data, Chan did three analyses to determine the relationship between democracy and war. The first was on all wars (interstate, colonial and others), 1816-1980, measured by the number of war-years. This is a count of the years during which democratic or nondemocratic (he coded all regimes himself as free or unfree) regimes had war. For these measures he found significant positive relationships between democracy and their number of years of war for the period 1816-1980 and the subperiods 1816-1945 and 1946-72; however, for interstate war years there was only a significant relationship for 1946-76.22 But for all war-years, or only interstate ones, there was a significant negative relationship between democracy and war-years for 1973-80 (Chan, 1984: Table 2). But the results for all wars should be ignored, since they include mainly colonial wars. But count only the independent states fighting. Presumably those uncounted non-state groups fighting on the other side were generally unfree and by their omission Chan skews the results against democracies. Thus I consider only interstate wars where all sides were counted. And for the interstate war-years there is one significant positive (for 1973-80) and one negative result (for 1946-72), both equally significant. Thus, if one accepts these war-year counts, the result is mixed.

However, his analysis suffers from the severe problem of war frequencies mentioned above, since he is in fact counting the number of wars a regime is engaged in for each year. He justifies this, in part, by quoting from my 1983 study, where I apparently did the same ("Libertarianism and International Violence": 53). But I was then measuring the severity of violence reached for each year, not counting the frequency of violence.23

Finally, he used the Small and Singer (1982) war data, defined by at least 1,000 battle dead for a war. As discussed previously, this includes some nations with a few killed while possibly excluding those with several hundred, and in any case equates those with only dozens of battle dead with those with millions.

Chan's second analysis determines whether democracies initiated fewer wars, and is limited this time to the frequency of interstate wars (not war years) for which democracies or non-democracies were on the initiator or defender's side. While the direction of the frequencies favored the Proposition, they were non-significant (Chan, 1993: Table 3). But these were frequencies again, and, moreover, although many including Chan favor the idea that democracies initiate wars less often than non-democracies, this is irrelevant to the Proposition: nothing is stated in it about whether or not democracies tend to be or to side with the violent aggressor. After all, a democracy may find a hostile state clearly preparing to attack it, and the only recourse for its own survival is a preemptive strike (as was so in the case of Israel when it attacked Egypt in 1967, launching the Six-Day War).

Chan's third analysis was a cross-tabulation of a regime's change in internal freedom with its wars (1993: Table 4). He found that for 32 states, 8 had more wars than expected during their freer years; 24 had fewer. Moreover, leaving out two cases significantly opposing the Proposition that Chan discounts, there were seven cases "strongly" and significantly favoring the Proposition (only 1.6 significant cases would be expected by chance). That is, as a regime changes from democratic to non-democratic, or vice versa, its frequency of wars tends to change, as one would expect by the Proposition. Still, here too frequencies were used and thus this favorable result must also be ignored.

In sum, of Chan's three analyses, the first is ambiguous (though strongly positive for the post-World War II years), the second is irrelevant, and the third is supportive. This means that on balance Chan should be cited as tending to favor the Proposition, if one accepts the use of frequencies as relevant.24 Uniformly, however, he has been cited without qualification as showing that democracies are as warlike as non-clemocracics. In any case, because of his use of frequencies, and for the reasons given previously, his results are not really relevant to the Proposition.

The final study often cited as negative on the Proposition is that by Erich Weede (1984), who collected data on all interstate wars (1960-80) using the Small and Singer (1982) war data and those of Butterworth (1976) and Kende (1979). He measured democracy according to Bollen's scales (1980), the Freedom House ratings,25 and other sources. For each collection of data he then calculated the (product moment) correlation of the frequency of wars with democracy for the years 1960-80, 1960-74 and 1975-80. For Butterworth wars, 1960-74, he found a significant negative correlation (one-tailed); for Small and Singer he found interstate wars an almost or barely significant negative correlation for 1960-80, a highly significant negative correlation for 1975-80, and a non-significant positive correlation for 1960-74;26 and for Kende's wars he found a non-significant negative correlation for 1960-80, a non-significant positive correlation for 1960-74, and a significant negative correlation for 1975-80 (Weede, 1984: Table 3 ).27 That is, of seven correlations out of ten which were in the proper direction, five were significant and one was almost or barely (at p <0.09) significant, depending on how much random error one assumes in the data. Three were in the opposite direction, none was significant and all for one subperiod 1960-74. On balance, therefore, these results tend to favor the Proposition, by Weede's way of evaluating significance. Actually, they very strongly favor the Proposition, even though Weede concludes that:

... democratic states were about as frequently involved in war as other states were in the sixties and seventies. This is the most important finding of this analysis ... [which] contradicts Rummel's ... [1983] findings.
----Weede, 1984: 637

Weede carried out one more study. To compensate for difficulties in his democracy scale, he dichotomized the scale into democracies and nondemocracies, and war frequencies into the occurrence of at least one war or no war for the regime. The result for the different data sets on war and the three different periods is that all the cross-tabulations are in the proper direction, although only one specific cross-tabulation is significant by itself (Weede, 1984: Table 3).28 But here too the results are questionable. Although frequencies are dichotomized into the presence or absence of war, they still do not avoid the problem of regimes with a few killed in war being equated with those in which millions are lost.

Over all, then, were we to accept frequencies as an adequate test of the Proposition, we would conclude that Weede's results are on balance favorable for the Proposition. Nonetheless, he has also been cited as finding that democracies are neither more nor less warlike than other types of regimes.29

Thus, for the three most-cited studies,--those of Small and Singer, Chan, and Weede--at the core of the consensus in the field that the Proposition is false, Small and Singer's results tend to support the Proposition, and if one could accept their measurement of violence Chan and Weede would also support the Proposition. Small and Singer used an appropriate measure of severity (battle dead) and their results are relevant to the Proposition. But the Chan and Weede studies are inappropriate, since they measure violence by the number of years at war, by the frequency of wars, or by the existence of war, none of which measures the severity of violence central to the Proposition.

There are five other studies following on these that do analyses bearing on the Proposition, but they all use the Small and Singer (1982) war or the Gochman and Maoz (1984) militarized dispute data and cross-tabulate or correlate violence or war frequencies with some measure of democracy. If their use of frequencies was relevant to the Proposition, one study would be positive (Morgan and Schwebach, 1992),30 two would tend to be ambiguous (Domke, 1988; Maoz and Abdolali, 1989), and two studies would be negative (Cole, 1990; Morgan and Campbell, 1991), neither one strongly so.

In Crises in the Twentieth Century, Jonathan Wilkenfeld, Michael Brecher and Sheila Moser cross-tabulate 627 actors involved in 278 crises (1929-79) according to whether they were democracies, authoritarian or military dictatorships. They found that:

The more authoritarian a regime the greater was the probability of violent crisis triggers [provoking crisis by the use of violence ] (democratic--37%, civil authoritarian--49%, military--56%).... As expected, there was a higher frequency of violence responses by military regimes (50%) than by democratic (30%) and civil authoritarian regimes (33%). Non-violent military responses were most often employed by democratic regimes (32%), compared to 20% for the other two types.... in short, the effect of type of regime on an actor's responsive behavior was evident for violent response--the more authoritarian a regime the more likely its response to a crisis would be violent.

The data on crisis management technique reveal an even sharper escalation of violence--(democratic (37%), civil authoritarian (49%) and military regime (63%), with a considerable higher tendency toward full-scale war as well (18% and 21% for civil authoritarian and democratic, 39% for military regimes). Conversely, democratic regimes which were most likely to perceive non-violent acts as triggers to their crises tended to choose pacific [crisis management techniques], with negotiation the most frequent among them: it was highest among democratic regimes (32%), and dropped to 11% for military regimes.
----Wilkenfeld et al., 1988: 197

Obviously, this is a study largely based on frequencies, but those results relating to the initiation of the violence which triggers a crisis and the tendency to escalate violence, relate to the severity of violence, are relevant to the Proposition, and positive.

Overall, then, we find that when the Freedom/Foreign Violence Proposition is properly tested in terms of the severity of violence, all correlations or cross-tabulations of democracy and violence are in the proper direction. That is, democracy is less warlike (severity) than other regimes. This is contrary to the prevailing wisdom among students of war, but upon careful inspection the results underlying their consensus have not only been shown to equate for a nation wars involving a few dozen killed with wars killing millions, but also, were frequencies relevant, to support the Proposition, not negate it.

Regardless, one does not need to rely on the nuances of these studies or their methodology. As empirically shown in my "Libertarianism and International Violence" study, by other work in the field in which relevant and appropriate variables have been used, and by the analyses I did here, when properly measured, democracies are less violent than non-democracies.


* Scanned from "Democracies ARE Less Warlike Than Other Regimes," European Journal of International Relations 1 (December 1995): 457-479. Typographical errors have been corrected, clarifications added, and style updated.

1. See Ray (1993, 1995), Rummel ("Libertarianism and International Violence", "Libertarian Propositions on Violence Within and Between Nations: A Test Against Published Research Results", Power Kills), Russett (1993) and Weart (1994, 1998), and the literature cited therein.

2. This significance was for two different ways of calculating the highest violence in the sample, significant at chi-squares, p <0.006 and p <0.5 x 10-10, one-tailed. Also the correlation between regime type and severity was significant at p <0.05, one-tailed.

3. Although I am often cited as the exception to the alleged findings that democracies are no less warlike than other regimes, in my Vol. 4: War, Power, Peace (Freedom Proposition) and "Libertarian Propositions on Violence Within and Between Nations: A Test Against Published Research Results" article, I list over a dozen studies the empirical results of which also support the Proposition.

4. These are the Joint Freedom (or Interdemocratic Peace) Proposition (democracies do not make war on and rarely commit lesser violence against each other), the Freedom/Dyadic Foreign Violence Proposition (the more democratic two regimes, the less severe their violence against each other), the Freedom/Domestic (Collective) Violence Proposition (the more democratic a regime, the less severe its internal collective violence), the Freedom/Democide Proposition (the more democratic a regime, the less its democide), and the summary the Democratic Peace Proposition (democracy is a method of non-violence). Fore a chapter length discussion and presentation of the evidence on each of these propositions, see Power Kills. Note that as far as these propositions and their evidence are concerned, the term "freedom" can be replaced by "democracy".

5. See Vol. 4: War, Power, Peace, "Libertarianism and International Violence", and Power Kills.

6. The average battle dead for the given type of regime is the sum of all those killed in battle of all the regimes of that type (such as US citizens killed in battle for the United States, plus Canadians for Canada, plus Australians for Australia, and so on for democracies) divided by the number of regimes of that type. Since only state regimes are involved, all battle dead are for interstate wars.

7. For both Bonferroni adjusted and Tukey pairwise comparisons, p <0.015 (one-tailed). The Bonferroni adjustment takes into account the number of means on which the significance test is being conducted (the more means the more likely a particular test will be significant).

8. F-ratio of 4.03, p <0.008 (one-tailed).

9. The significance for the analysis of variance is p <0.1 (one-tailed), and those for the paired comparisons among the averages were non-significant.

10. All the data on violence and type of regime used here are given in Statistics of Democide.

11. How many regimes have existed depends on how different regimes have been defined. For the Gurr (1990) Polity II data there are 427 regimes, 1900-87. And I largely have relied on Polity II characterizations of regimes, but adjusting the count where I thought Polity II was mistaken. For example, Polity II gives three Chinese regimes from the years 1949-87; I count only one. Polity II gives three regimes for the Soviet Union, 1918-87; 1 count only one here as well.

12. The Bonferroni adjustment and Tukey pairwise comparisons were significant at p <0.01 (one-tailed).

13. Data were from Global Data Manager (1990).

14. Wealth is a distinct dimension from one with which political characteristics are correlated. That is, political characteristics, and thus type of regime, have little correlation with wealth.

15. As with wealth, the political characteristics of nations also have little correlation with their physical power, that is, their capability.

16. This statement is based on a simultaneous factor analysis (called super-P factor analysis) of behavior and attributes for 1950, 1955, 1960 and 1963; and a canonical analysis of dyadic behavior regressed on the distances between nations, such as along wealth and capability dimensions, among others. See Rummel (1979b).

17. Data are from Global Data Manager (1990). Not only is energy consumption among the best indicators of capability, but it also avoids the currency convertibility problem of its close rival, GNP.

18. See also Small and Singer (1982: 55).

19. In previously using the Small-Singer data I avoided the problem of the vastly different number killed in a war for each country by using battle dead as my measure of war. But my results are still affected by the 1,000 battle-dead cutoff for a war. Thus, a state that had a few killed in a war in which it had many troops was included in my analysis while a state that lost near 1,000 in a military action in which 1,000 were not killed overall was not included.

20. The t-test should be 2.84, p <0.0023 (one-tailed), not the 1.45 Small and Singer give. In a personal communication of 10 September 1980, I informed Singer of the errors in this and the previously mentioned t-test. Scott G. Gates, an assistant on Singer's Correlates of War Project (COW), subsequently responded "on behalf of the COW project" that they had verified my calculations. In a note written on Gate's response Singer confirmed his acceptance of this.

21. They do have another table giving the number of democracies and nondemocracies participating in wars (Table 3). The number of democracies is always a small percentage of the total number of nations involved, but nothing can be made of this without knowing the total number of democratic regimes in the international system.

22. The significance tests were by chi-square.

23. Admittedly, my phrasing of this led Chan to conclude I was using frequencies. However, my previous discussion of the methodology and Proposition should have warned him away from this interpretation. I should also say that when I wrote the 1983 article I did not realize that virtually without exception subsequent research by others would use frequencies.

In any case, Chan (1993: 207) now also recognizes the problem in frequency counts equating vastly different kinds of wars. He writes that "research on the democracy/war proneness question has focused on war involvement as a dichotomous dependent variable. At the very least, this approach assumes that the analyst has a homogeneous set of war involvements. Thus, World War II and the Western power's participation in China's Boxer Rebellion are given the same weight." Moreover, he sees the value of using a severity measure like "casualties", and points out that "To focus on an undifferentiated categorical variable of war involvement [frequencies] when more sensitive scalar data on the destructiveness of war efforts arc available does not seem to be helpful to our analytic interests. I suspect that depending on the particular measure of war or conflict used, one can arrive at very different conclusions about the relative pacifism of democracies".

24. Chan is now persuaded that democracies may be more peaceful (but still in terms of frequencies). In his latest piece (1993: 206) he wrote: "To the extent that due to their limited capabilities many nondemocracies have had less opportunities to get into wars with one another (such as between Nepal and Senegal, or between Ethiopia and El Salvador), and to the extent that due to their greater capabilities democracies have had more such opportunities among themselves and vis-a-vis many more nondemocracies, the historical data of war involvement would appear to lend greater support to the hypothesis that democracies are more pacific than nondemocracies"

25. These are given in the January-February issues of Freedom at Issue for each year.

26. I will ignore the results for extrasystemic wars, since only independent states are counted.

27. All significance levels were one-tailcd.

28. The direction of the relationship is given by Yule's Q; significance by chi-square (one-tailed). If we assume that there is no relationship between democracy and having fought wars, then the probability of getting 10 cross-tabulations out of 10 in the hypothesized direction by chance is 0.001. But these 10 cross-tabulations were not all independent. The different data sets overlap, as do the subperiods with the full periods, and this is enough to vitiate the probability. Nonetheless, it is possible that these 10 cross-tabulations can be either positive or negative and therefore that they were all in the proper direction should be considered substantively important, if one accepts the underlying measurements. Note also that it is incorrect to test separately the significance of coefficients arrayed in a table as Weede does here. For 100 such coefficients, 5 should be significant at p 0.05 entirely by chance.

29. Weede ignored the Proposition that democracies do not make war on each other, and, as he later wrote, did not accept it at this time. He now does. In his latest article (1992: 378) he wrote "I did not accept the "peace among democracies" proposition then [in his previous three articles].... Now, the weight of the evidence makes me accept this proposition. . . ."

30. Two readers, one strongly, argued that Bremer (1992) should also be included as positive, since he found that democratic vs. non-democratic dyads are significantly less likely to become engaged in war than non-democratic vs. non-democratic dyads. But these results are dyadic, not monadic, and thus irrelevant to whether democracies are more or less warlike. However, these results do bear positively on the different Proposition that the more democratic two regimes, the less likely there will be violence between them. Russett (1993) does not present results relevant to the monadic Proposition, but in a personal communication he reported reanalyzing his data on democracy, wars and disputes. He found that the data, 1946-86, clearly supported the Proposition.


Babst, Dean V. (1964) "Elective Government--A Force for Peace", Wisconsin Sociologist 3 (1): 9-14.

Babst, Dean V. (1972) "A Force for Peace", Industrial Research (April): 55-8.

Bollen, Kenneth A. (1980) "Issues in the Comparative Measurement of Political Democracy", American Sociological Review 45 (June): 370-90.

Bremer, Stuart A. (1992) "Dangerous Dyads: Conditions Affecting the Likelihood of Interstate War, 1816-1965", Journal of Conflict Resolution 36 (2): 309-41.

Bueno de Mesquita, Bruce and David Lalman (1992) War and Reason: Domestic and International Imperatives. New Haven, CT: Yale University Press.

Butterworth, Robert Lyle (1976) Managing Interstate Conflict, 1945-1974.--Data with Synopsis. Pittsburgh: University Center for International Studies, University of Pittsburgh.

Chan, Steve (1984) "Mirror, Mirror on the War ... Are Democratic States More Pacific?", Journal of Conflict Resolution 28 (4): 617-48.

Chan, Steve (1993) "Democracy and War: Some Thoughts on Future War Agenda", International Interactions 18 (3): 205-13.

Cole, Timothy Michael (1990) "Politics and Meaning: Explaining the Democratic Peace", paper delivered at the Annual Meeting of the American Political Science Association, August 30-2 September.

Czempiel, Ernst-Otto (1992) "Governance and Democratization", in Ernst-Otto Czempiel and James N. Rosenau (eds) Government without Government: Order and Change in World Politics, pp. 250-71. Cambridge: Cambridge University Press.

Dixon, William J. (1994) "Democracy and the Peaceful Settlement of International Conflict", American Political Science Review 88 (March): 1-17.

Domke, William (1988) War and Changing Global System. New Haven, CT: Yale University Press.

Doyle, Michael (1983) "Kant, Liberal Legacies, and Foreign Affairs, Part I", Philosophy and Public Affairs 12 (Summer): 205-35. Part 11, ibid.: 323-53.

Geva, Nehemia, Karl DeRouen, and Alex Mintz (1993) "The Political Incentive Explanation of the 'Democratic Peace': Evidence from Experimental Research", International Interactions 18 (3): 215-29.

Global Data Manager. Macintosh Version 2.3 Computer Application (1990). Philadelphia, PA: World Game Institute.

Gochman, Charles and Zeev Maoz (1984) "Militarized Disputes 1816-1975", Journal of Conflict Resolution 29 (4): 585-615.

Gurr, Ted Robert (1990) Polity II--Political Structures and Regime Change, 1800-1986. Ann Arbor, MI: Inter-University Consortium for Political and Social Research (Polity II data code book).

Kende, Istvén (1979) "Dynamics of Wars, or Arms Trade and of Military Expenditure in the "Third World", 1945-1976", Instant Research on Peace and War 2 (2): 59-67.

Kilgour, D. Marc (1991) "Domestic Political Structure and War Behavior: A Game Theoretic Approach", Journal of Conflict Resolution 35 (June): 266-84.

Maoz, Zeev and Nasrin Abdolah (1989) "Regime Types and International Conflict, 1816-1976", Journal of Conflict Resolution 33 (March): 3-35.

Morgan, T. Clifton and Sally Howard Campbell (1991) "Domestic Structure, Decisional Constraints, and War--So Why Kant Democracies Fight?", Journal of Conflict Resolution 35 (June): 187-211.

Morgan, T. Clifton and Valerie L. Schwebach (1992) "Take Two Democracies and Call Me in the Morning: A Prescription for Peace?", International Interactions 17 (4): 305-20.

Ray, James Lee (1993) "Wars between Democracies: Rare, or Nonexistent?", International Interactions 18 (3): 251-76.

Ray, James Lee (1995) Democracy and International Politics: An Evaluation of the Democratic Peace Proposition. Columbia, SC: University of South Carolina Press.

Rummel, R. J. (1968) "Domestic Attributes and Foreign Conflict" in J. David Singer (ed.) Quantitative International Politics, pp. 187-214. New York: Free Press.

Rummel, R. J. (1972) The Dimensions of Nations. Thousand Oaks, CA: Sage.

Rummel, R. J. (1979a) Vol. 4: War, Power, Peace. Beverly Hills, CA: Sage.

Rummel, R. J. (1979b) National Attributes and Behavior. Thousand Oaks, CA: Sage.

Rummel, R_ J. (1983) "Libertarianism and International Violence", Journal of Conflict Resolution 27 (March): 27-71.

Rummel, R_ J. (1984) "Libertarianism, Violence Within States, and the Polarity Principle", Comparative Politics 16 (July): 443-62.

Rummel, R. J. (1985) "Libertarian Propositions on Violence Within and Between Nations: A Test Against Published Research Results", Journal of Conflict Resolution 29 (September): 419-55.

Rummel, R. J. (1994) Death By Government. New Brunswick, NJ: Transaction Publishers.

Rummel, R. J. (1997) Power Kills: Democracy as a Method of Nonviolence. New Brunswick, NJ: Transaction Publishers.

Rummel, R. J. (1997) Statistics of Democide. Charlottesville, VA: Center for National Security Law, University of Virginia..

Russett, Bruce (1993) Grasping the Democratic Peace: Principles for a Post-Cold War World. Princeton, NJ: Princeton University Press.

Small, M. and J. D. Singer (1976) "The War Proneness of Democratic Regimes, 1816-1965", Jerusalem Journal of International Relations 1 (Summer): 50-69.

Small, Melvin and J. David Singer (1982) Resort to Arms: International and Civil Wars 1816-1980. Thousand Oaks, CA: Sage.

Starr, Harvey (1992) "Why Don't Democracies Fight Each Other? Evaluating the Theory--Findings Feedback Loop", Jerusalem Journal of International Relations 14 (4): 41-59.

Weart, Spencer (1994) "Peace among Democratic and Oligarchic Republics", Journal of Peace Research 32 (3): 299-316.

Weart, Spencer (1998) Never at War: Why Democracies Will Not Fight One Another, New Haven: Yale University Press.

Weede, Erich (1984) "Democracy and War Involvement", Journal of Conflict Resolution 28 (4): 649-64.

Weede, Erich (1992) "Some Simple Calculations on Democracy and War Involvement", Journal of Peace Research 29 (4): 377-83.

Wilkenfeld, Jonathan, Michael Brecher and Sheila Moser (1988) Crises in the Twentieth Century. Volume 11: Handbook of International Crisis. New York: Pergamon.

Wright, Quincy (1942) A Study of War, Vols. I-II. Chicago, IL: University of Chicago Press.

You are the visitor since 11/25/02

Go to top of document