Other Democratic Peace Documents On This Site
[A] country which develops strong forces of repression to keep down its own citizens is likely to be a danger beyond its frontiers. No tyranny, therefore, is an island. Democratic leaders should recall that wars between democratic states have not been seen in the twentieth century. Indeed, wearied and demoralized though modern man's democracies have sometimes been, there is no instance at all of any of them being bellicose against each other
This article subjects the following propositions to systematic tests against the quantitative literature: that violence does not occur between libertarian states; that the more libertarian two states are the less violence there is between them; and that the more libertarian a state is the less it will be involved in foreign and domestic violence. A comprehensive survey of the literature uncovered 67 relevant analyses. An evaluation, scoring, and weighting of these analyses showed that they significantly support the propositions.
This is a third follow-up report, but of a different kind. Rather than apply new empirical tests to these answers, they are now tested against the accumulated quantitative literature-essentially a more systematic and comprehensive retest of the effort described in Vol. 2: The Conflict Helix and Vol. 4: War, Power, Peace of the above-mentioned work. In detailing this, I will elaborate first the relevant propositions, and then present the philosophy and mechanics of the testing procedures. This will be followed by the results and their discussion.
At a less abstract level, there are the explanations common to liberal scholars: the aggregating and compromising, and therefore conflict reducing, effects of competitive party systems; the institutionalization of societywide conflict resolution through competitive politics and the ballot ("the ballot replaces the bullet"); the formalization and regulation of conflict and violence (e.g., labor-management collective bargaining laws); the democratic emphasis on exchange instead of authority and coercion; the unwillingness of democratic majorities to pay in blood and taxes for the foreign adventures of a political elite.
Keeping in mind then that a libertarian state emphasizes individual freedom and civil liberties and the rights associated with the competitive and open election of leaders, the following propositions (with their operational statements in parentheses) should be true.
These propositions, which answer the questions raised at the beginning of this report, are to be tested against the quantitative literature.
But this lack of popularity is no index of the strength and value of the test. For example, a test against the literature is stronger (presuming relevant findings do exist) than any particular data-based test,
This report will do a literature test of the libertarian propositions (for some of them, a retest) against quantitative international studies. This involves a number of considerations to which I will now turn.
Defining the relevant literature. First, the relevant literature is limited to publications (unlike Vol. 4: War, Power, Peace). In this way I could hope for a virtually complete survey. To include unpublished papers and reports would considerably increase the sample by those I happen to be familiar with--a good number of which would be my own--and, considering that these would necessarily constitute but a small number of unpublished works, the sample might thereby be biased. Moreover, if others wish to check my ratings, the published studies are easily available while unpublished material usually is not. By publication is meant any commercially available book, report, or journal.
Second, a relevant publication is defined as any quantitative study involving both violence and political variables or statistics. A quantitative study is any analysis presenting numerical data, ratings, evaluations, and the like, in some systematic form (even if only in a table, ordered by variable) or, but not necessarily, subjecting such to some kind of systematic analysis. Thus Tilly et al. (1975) is relevant even though only tables and plots of conflict frequencies are presented.
Locating relevant literature. To achieve a complete survey, several approaches were used. I wrote to 116 relevant senior or active researchers asking for their own reprints and references to publications I might not ordinarily come across; checked pertinent references in quantitative international studies; screened the appropriate university library stacks; and skimmed all the possible germane articles in the relevant journals published since the mid-1960s.
Upon accepting this report for publication, the editor of this journal suggested including the Weede (1984) and Chan (1984) articles soon to be published. I agreed, and also resurveyed the most relevant journal numbers to appear since the survey was completed. Aside from my own ("Libertarianism, Violence Within States, and the Polarity Principle"), I could find no other relevant publications. However, this resurvey could not be comprehensive (many journal numbers were at the binder, for example) and therefore violates the comprehensiveness of the original sample.
Defining relevant quantitative results. Once a relevant publication was found, I determined whether its quantitative results actually bore on the libertarian propositions (as operationally worded) and whether the study had prima facie relevance. The result is a sample of 54 studies involving 67 separate analyses to be evaluated, and rated as to whether they support or negate the propositions (a study is a published work; an analysis is a specific research design, of which there may be several in a study). All these studies are listed in the bibliography and will be subsequently classified by the propositions to which their analyses are relevant.
Unfortunately, limited space precludes comparing and contrasting these many studies here and describing their methods and data sets. However, there is one potential problem in these studies that must be considered, and that concerns the independence of their data and analyses. In most cases the studies use originally collected data; in some cases a data set, like the Singer and Small (1972) war sample, is used across several analyses; and some studies are virtual duplicates in data and methods of a previous one. As will be subsequently made clear, duplicate studies will be excluded from the test, and different studies analyzing the same data will be included only if they use techniques or approaches different enough to yield potentially contrary results (as for cross-sectional versus longitudinal analysis).
The sample. Quantitative studies employ many different kinds of data sets (or samples), several aspects of which are critical to testing the propositions. These aspects can be made clear by classifying the data sets into four types.
Not all four data sets will be relevant to each proposition; and if relevant, the nature of the set may limit what evidence is appropriate and whether it bears directly on the propositions. Consider a war data set, for example, that includes all states that have had war from 1945 to 1980. Let the set also contain such variables as censorship, competitive party politics, and freedom of group opposition in each state, and let a scale of democracy be formed from these variables in such a way that it can also be interpreted as a scale of libertarianism. Now, would the number of states that are democratic on this scale, versus those that are not, be a test of the Freedom/Foreign Violence Proposition that libertarian states are less inclined to foreign violence? No, and for two reasons.
First, to assess whether democratic states tend to have more or less foreign violence, one must know not only how much violence democratic states have engaged in relative to that of nondemocratic states, but also how many democratic states relative to nondemocratic states have not had violence. Another reason why the war set would not be appropriate is because it limits foreign violence to war. The Freedom/Foreign Violence Proposition refers to foreign violence, not specifically war. True, war is a subset of violence and what is true for violence should be true for war. Because of this, studies employing war data are considered indirectly relevant. However, what is true of wars may not be true for other kinds of violence or violence overall. For example, data for all states on the number and intensity of wars may show that democratic states are about as prone to war, regardless of intensity, as nondemocratic states. But democratic states may have relatively very little foreign violence short of war (democracies may avoid violence until nothing but war is left) while other states may be far more willing to employ violence at all levels.
The upshot of these considerations is not only that a war set is not directly appropriate to test the Freedom/Foreign Violence Proposition, but that the nature of any data set must be carefully considered for each proposition. For this reason the four data sets are distinguished and the appropriateness of each to the different propositions will be given below.
The dependent variable. For all four propositions, violence is the dependent variable. In foreign affairs this is the official use of violent means (those causing injury, death, or destruction of property) against another state; domestically, it is social or political collective violence, including the violence of a regime (e.g., mass or terroristic executions).
As the theory focuses on the inhibitions of the governing elite in a social field, the limitation to official or regime foreign violence should be clear. Attacks on a foreign embassy or a national representative--for example, the assassination of an ambassador--may have no official sanction and be the work of some small terrorist group. There is nothing in the theory to link such extremist violence to the degree of libertarianism.
Violence can be measured in many different ways, not all relevant to the propositions. Table 1 lists a number of such measurements and their appropriateness or usefulness for the propositions. Note especially that foreign violence and war are not the same thing; nor is domestic violence to be equated with revolution. Moreover, violence and conflict are not the same. The theoretical inverse relationship between libertarianism and violence stated in the propositions does not necessarily hold for conflict. Indeed, libertarian societies are theoretically expected to have more nonviolent conflict than other societies due to their diversity, lack of overriding custom, and limited governmental authority.
Of greatest importance is how violence is measured. By theory the social field will increasingly inhibit the outbreak of domestic or foreign violence the more intense the violence is likely to be. These inhibitions should be most effective at the upper ranges of violence intensity, least effective at the lower ones. Thus there should be a gradient of violence intensity: The less libertarian a state, the more likely intense violence. This means, then, that violence, as far as the propositions are concerned, cannot be measured adequately by a count of violent events or, as an indirect measure, a count of revolutions or wars. Most relevant would be some measure of severity overall, such as the number of people killed in official foreign violence.
The independent variable. Broadly conceived, libertarianism refers to social systems in which its members (individuals, groups, or states) have the maximum freedom of choice consistent with like freedom for others. As a political variable it incorporates a spectrum of libertarian types, including anarchy as the most libertarian. I will ignore anarchy here, however, because virtually no research of which I am aware allows for a comparison between violence in an anarchy and in governed societies.
Focusing then on governed societies, libertarianism--a libertarian state, regime, government, or system--is a structural variable. It refers to a particular cluster of political characteristics that should always be found together: free and open elections for top leaders, competitive party systems, freedom of speech and the press, freedom of groups to oppose government, individual rights and limited government, and so forth.
Aside from anarcho-libertarianism, there are two contending ideas of libertarianism. In its broad sense, libertarianism is equated with civil liberties and political rights--what we usually mean by democracy. Sweden, Japan, and the United States would thus be almost equally libertarian. In its narrow sense, libertarianism is equivalent to classical liberal democracy, adding to democracy the requirement of a free market. India would then be less libertarian than Japan; Israel less libertarian than West Germany. Theoretically, liberal democracies should have significantly less violence than socialist democracies, but both should have significantly less than nondemocracies. Unfortunately, few studies have included variables or scales that distinguish between the two types of libertarianism, although many do include measures that scale from libertarianism, in its narrow sense, to totalitarianism. Therefore for this test against the literature, libertarianism will be understood in terms of civil liberties and political rights, without the requirement of a free market. In effect, what is being tested here is that political democracy (the narrower version of libertarianism) is inversely related to violence, as should be clear from the choice of indicators of libertarianism in Table 2.
In sum, then, to determine whether a study has an analysis appropriate to the propositions, I would look for a measurement of violence like those listed in Table 1; and some such taxonomy or indicator of libertarianism as given in Table 2. But this only defines the propositional variables. What about their relationship?
Predicted correlations. Quantitative international studies involve different approaches, methods, and techniques. Correlational techniques (such as simple correlations, regression analysis, and factor analysis), however, are the most common, and other relevant techniques (such as the chi-square or analysis of variance) can be interpreted correlationally. Moreover, all but the first proposition are explicitly correlational. The first proposition (Joint Freedom) gives a point prediction: Libertarian states will have zero violence. If a study reports such violence, except for marginal cases or under very unusual circumstances, it constitutes a negative result.
Table 3 shows the direct and indirect correlational results or findings predicted for the four propositions, were they true. These are divided by type of data set (sample) and conflict variable (keeping in mind which of the measurements shown in Table 1 are relevant). This table provides a map, then, for locating the results that an analysis should or could have, directly or indirectly, for the data set it uses if any one of the four propositions is true.
For example, if an analysis calculates the intensity of violence for a conflict data set and also includes measures of libertarianism, then if the libertarian propositions are correct one should find zero intensity for libertarian dyads (Joint Freedom Proposition); a moderate or high negative correlation between the intensity of violence and libertarianism (Freedom/Dyadic Violence Proposition); and a moderate or low but still statistically significant negative correlation between the intensity of violence and libertarianism (Freedom/Foreign Violence and Freedom/Domestic Violence Propositions).
Methodological-analytic considerations. In rating whether a study's results or findings conform to those predicted by the propositions, its method of analysis had to be more carefully considered than it had been in selecting the sample of relevant studies. As a result, some analyses (not necessarily a whole study) were found unusable because of incorrect application of a method, numerically incorrect or contradictory tests or correlations, or inadequate technical details about an otherwise relevant result. Wherever possible, I redid the calculations or analysis myself, and in all cases ignored statements in the text that were inconsistent with its tables or figures.
Qualitative factors. Some studies may give more or less weight to their results than I do. For example, for a given data set and sample size, I may be looking for a significant correlation that might be .25 or more, although the study analyzing these data is concerned only with variance and thus with correlations of, say, .50 or more. Some studies give insufficient methodological information to evaluate or rate their findings precisely (for example, as to whether a given correlation is a multiple R, simple r, or partial r; whether a factor matrix is rotated or unrotated--on these terms see Understanding Correlation of "Understanding Factor Analysis"). Where this is a critical ambiguity the study is rated unusable.
Moreover, my focus is always at the operational level, where sometimes a variable's measurement or description may not agree with the label the variable may have; or the quantitative results or findings may not agree with an author's interpretation of the results. For me, the specific data set, indicators, measurements, and numerical results are the basis for rating a study, not an author's labels, evaluations, or conclusions.
Finally, there is a sense for the propositions and underlying theory that plays a role in rating the literature that is difficult to communicate. After all the "objective" factors have been considered, a judgment must be made. And this judgment is inevitably influenced by one's work and that of others. In any case, I suspect that at least the authors of the studies rated and their students will be interested in double-checking my evaluations; or those who believe their studies should have been included will wonder why they were omitted. In all cases, before blaming this on error, a mind set, or bias, check the following:
Importance. The importance of an analysis depends on the strength of its methodology, the time frame, the amount of data, and the number of variables. For example, a study cross-tabulating war data for 30 states in the period 1961-1963 is hardly as important as a correlational analysis of all wars for all states over two centuries. The former would be of low importance; the latter highly important. Not all analyses are so clear or extreme, and so a moderately important middle evaluation is also used.
Relevance. The results of most analyses covered here bear directly on a proposition. Sometimes, however, the results are only inferentially or indirectly related to a proposition. In this case, the indirect relationship should be indicated and lesser weight given them. Determining the relevance of an analysis is helped by the codings in Table 1 and Table 3, which show the measurements of conflict and the correlations for those data sets that should be considered indirect and those that should be considered direct.
Usableness. Most analyses are usable--their results can be rated. But the for which relevant results of some analyses are unratable because of methodological problems or lack of clarity. When such is the case, the reason for this should be indicated.
Positive versus negative support. Finally, the results and findings of a usable analysis are rated for a proposition as follows:
- SP = strongly supportive
- P = supportive
- A = ambiguous
- N = negative (not supportive)
- SN = strongly negative (supports an opposite proposition)
Any number of aspects of an analysis could make the difference between a P and an SP rating. A much greater level of significance (say, from .05 to .001), a very high correlation, a set of consistently positive results, a strong test, and the like. The difference between an N and an SN can be made clearer. Generally, negative results are those inconsistent with the proposition. But strongly negative results are those that are opposite to what is predicted by the proposition. For example, if a proposition predicts a significant negative correlation between libertarian institutions and domestic violence and an analysis finds no significant correlation between them, it would be coded negative (N) for the proposition. If instead this analysis found, quite in opposition to the proposition, a significant positive correlation, it would be coded SN (strongly negative). Note that this is a conservative approach: Only those results conforming to the proposition are considered positive; any other relevant and nonambiguous results are treated as at least negative. An ambiguous analysis is one whose results conflict--some negative, some positive. In some cases my reanalysis of a study's data may conflict with the original analysis, but both analyses may be valid in terms of the methodology used.
Bias compensating ratings. I suspect that my ratings will not be completely unprejudiced. Possibly I might be too wary of bias and the charge of stacking the ratings, and thus be overcautious. Prudently, however, I should assume that whatever my bias, it is toward a higher positive rating. To compensate for the maximum possible bias, therefore, all the analyses are double-rated: First, as I judge the proper rating; second, a one level decrement in this rating (SP to P, P to A, etc.) in order to maximally compensate for possible bias. Given the results rated and the criteria presented, I need to be persuaded that I would be off by as much as one full rating per analysis; so this compensation should provide a most conservative test of a proposition's support.
Examples. Space precludes discussing each analysis and its ratings, but some examples can be presented in order to make the rating system clearer. One example will be selected for each proposition tabulated in Tables 6, 7, and 8.
The first example is the Small and Singer (1976) analysis of all interstate wars with at least 1,000 battle deaths, 1816-1965. In their publication they tabulate the dyads involved in a war and whether a participant was democratic or not. Their definition of democracy meets (barely) the criteria of Table 2; and although they are using a war data set, it is relevant to the Joint Freedom Proposition. According to Table 3, if the proposition is correct there should be no cases of war between democracies in the Small-Singer results. In fact they report that except for two marginal cases there is no war between democracies. By my calculations 317 dyads made war from 1816 to 1965; 135 were democracies versus others; and of these there were 11 marginal cases of democracies versus democracies. But all except one of these involved Finland formally at war against the allies in World War II (due to Finland joining Germany in fighting the USSR), but no actual military action between them took place. The other case is of an "ephemeral republican France attacking an ephemeral republican Rome in 1849" (Small and Singer, 1976: 67). There are a number of other results having to do with whether democracies have more or less fatalities in war or shorter or longer wars. But these are not given in dyadic form and are thus irrelevant to the Joint Freedom and Freedom/Dyadic Propositions; and although they might at first seem relevant to the Freedom/Foreign Violence Proposition, they are not. For one thing, number of fatalities is irrelevant by Table 3 as this is a war data set; for another, duration is an irrelevant indicator by Table 1.
How then rate this analysis for the Joint Freedom Proposition? First, the analysis is obviously important. It tabulates wars and whether a state involved in war is a democracy or not over the whole period 1816-1965. Second, the analysis is direct. The measurement of democracy and the tabulation of whether or not they had wars bears directly on the proposition--any nonmarginal cases or war between democracies would violate the proposition. Remember, this proposition makes a point prediction, unlike the other correlational propositions. Third, the actual finding is rated strongly supportive (SP) because for no interstate war (as classified by Small and Singer) occurring during one and one-half centuries did states that were democracies for more than a few years actually fight each other. Finally, the application of a one-level decrement to compensate for possible bias in the above rating yields an alternative rating of supportive (P).
The second example is the analysis by Schwerin (1977) that is related to the Freedom/Dyadic Foreign Violence Proposition. Schwerin did a canonical analysis of the foreign behavior of the United States and its diverse distances (political, economic, etc.) from other states in order to test social field theory. Included among the behavioral variables was a dimension (factor scores) of military violence; and among the distances were dimensions of authoritarianism and totalitarianism. The third canonical variate emerging from the canonical analysis had a canonical correlation of .72, with a Z-transformation of 8.48 (df > 30) and a significance of p < .1x10-16. Military violence was the main dependent variable (canonical coefficient = .75; since all the variables were near orthogonal, the canonical coefficients can be interpreted roughly as correlation coefficients) and among the independent variables distance in totalitarianism was one of the main predictors (canonical coefficient = .51, taking into account the direction of scaling). This canonical equation is saying in effect that the more democratic the other state (democracy being at the opposite pole of the totalitarian dimension), among its other attributes, the less military violence the United States directs toward that state.
Clearly this finding has some relationship to the Freedom/Dyadic Foreign Violence Proposition, but how should it be rated? In importance this analysis is low. Although it involved several hundred variables (including those involved in the separate dimensions that were input to the canonical analysis), the analysis is only for 1963 and comprises only dyads with the United States as actor. However, the results are directly relevant: Democracy is opposed to totalitarianism, and military violence factor scores define the relevant results. In spite of the great significance of the canonical correlation for the equation relating these variables, the results are rated as only supporting (P) the proposition. This is because totalitarianism has only a moderate canonical coefficient on the equation, while three other distances have higher coefficients. Moreover, authoritarianism is not at all loaded on the equation. While a definite inverse relationship between democracy and military violence is shown, therefore, it appears a partial and minor one more in accord with simply a supportive rating than anything stronger. Applying a one-level decrement gives an adjusted rating of ambiguous (A).
A third example is the study of Chan (1984) that relates to the Freedom/Foreign Violence Proposition. Using an updated list of wars (collected by J. David Singer's Correlates of War Project), Chan did three analyses to determine the relationship between democracy and war. The first was on a mixed data set, 1816-1980, and found that for all wars (interstate, colonial, and others) there was a significant (chi-square) positive relationship between democracy (states were coded as relatively free or unfree) and war that held regardless of subperiod; however, for interstate wars there was no significant relationship for the whole period, a significant positive relationship for the period 1946-1972, and a significant negative one (in line with the proposition) for the period from 1973-1980. 1 do not consider the noninterstate war results as legitimate, for only the state participants (one side of each war) were coded. Considering that most of these wars involved nonstate countries or groups that were unfree, in Chan's terms, the coding could be biased against democracies. If these results are set aside, then those for interstate wars do not generally support the hypothesis and should be rated negative (N). They are not strongly negative because they are not significantly opposed to the proposition overall and for one period they do support the proposition.
Chan's second analysis found that democracies did not significantly tend to be on the initiator's side of a war, or on the opponents (Chan, 1984: Table 3). That is, democracies were not inclined to initiate or side with either the aggressor or the victim in war. However, this is irrelevant to the proposition: Nothing is stated about whether libertarian states tend to be or to side with the violent aggressor or not; nor can such be inferred from the theory. After all, a libertarian state may find a hostile state clearly preparing to attack it, and the only recourse for its own survival is a preemptive strike (as was the case for Israel in its attack on Egypt in 1967, launching the Six Day War). Chan's third analysis is a cross-tabulation of a state's change in internal freedom with its wars (Chan, 1984: Table 4). He found that for 32 states, 8 had more wars than expected during their freer years; 24 had less. Moreover, leaving out the 2 cases significantly opposing the proposition that Chan discounts, there were 7 cases "strongly" and significantly favoring the proposition (only 1.6 significant cases would be expected by chance). Accordingly, I would rate this analysis as supporting (P) the proposition. It is not rated strongly supportive because out of the 24 that tended to favor the proposition, only 7 did so significantly.
In sum, out of Chan's three analysis, one is negative, one is irrelevant and one is supportive. Overall, therefore, I gave the Chan analyses an ambiguous (A) rating for the proposition; and when the one-level decrement is applied, a negative support (N). In importance it is clear that the analyses should be rated highly--they cover all wars for over one and a half centuries. In relevance, however, these analyses bear only indirectly on the proposition: War, not violence, is the dependent variable and war is also measured indirectly in terms of frequency, presence/absence, or war-years.
The third example comprises Gurr's two studies (1979a, 1979b) related to the Freedom/Domestic Violence Proposition. They are too similar in data and analyses to be considered independent studies and are coded as one. The analysis published in Violence in America (1979a) is the more extensive and is the one focused on here. Gurr codes over 2,200 episodes of conflict in 87 states for the years 1961 to 1970. He distinguished 37 democratic (including 18 Western democratic), 19 autocratic, and 31 elitist states and cross-tabulated their conflict. By calculating the proportions from information given in his Table 2.1 and the text (Gurr, 1979a) I determined that Western democracies have proportionally more riots than other states, but much fewer guerrilla wars, civil wars, local rebellions, and revolts and private wars. Moreover, his Table 2.3 shows that the median conflict deaths for 18 Western democracies is 0 (10 reported no deaths) compared to a median of 21 per 10 million for all states; the United States, considered by some a most violent state during this period, was below the median with 18 deaths per 10 million. A problem with these findings, however, is that 19 non-Western democracies are included among the other states. Fortunately, Gurr also includes a cross-tabulation for all 37 democracies (Gurr, 1979a: Table 2.5) compared to autocratic and elitist states. He finds that the median conflict deaths per 10 million population is 8 for the democracies, 14 for the autocratic, and 550 for the elitist (a classification including totalitarian states as well).
How should Gurr's analysis be coded? First, it covers only 10 years, but it includes almost all states and the data collection is thorough, across a variety of conflict variables. The limitation in time precludes a rating of high importance, but the comprehensiveness of the data and analysis also precludes a rating of low importance. Accordingly, it is rated medium (M). In relevance, some of the tables are at best indirectly related to the proposition. His Table 2.5, however, is to the point: it clearly discriminates a scale of democracy and measures violence in terms of intensity. Gurr's analysis is therefore rated directly relevant. As to whether they support the proposition or not, all results are consistently in favor, are in the appropriate direction, and seem large in magnitude. But Gurr's analysis cannot be coded strongly supportive without more information to do significance tests. Surely, however, these results are substantial enough to code them as supportive (P) of the proposition. And applying the one-level decrement for possible bias yields an alternative coding of ambiguous (A).
As shown, three kinds of scores will be used finally to judge whether a proposition tests well against the literature. First, there is a total score (sum of the ratings) for a proposition, which will be positive if the literature tends to be supportive, zero if balanced, and negative if opposed. As this total will be influenced by the number of analyses rated, an average of the ratings is also calculated. The total still is useful, however, as a measure of the weight of evidence (obviously 20 supportive studies are more important than 2).
Neither the total N, or the average takes into account the importance or directness of an analysis. It hardly seems reasonable to accord the same weight to one study's correlations between indirectly relevant variables for a sample of 20 nations on 1960 data and that of another study's correlations between democracy and violence for all nations since 1900. Accordingly, for each proposition I also calculated a weighted average that takes into account the importance and relevance of an analysis. Note from Table 4 that the measurement of importance and relevance is such as to subtract from a rating unless the analysis is important and direct. This rating and scoring system will be applied to the ratings and to their bias compensating one-level decrement.
Finally, a t-test will be made of the average and weighted average to determine their significance (with regard to the null hypothesis that average = weighted average = 0);
Aside from the individual evaluations, the totality of the ratings should be weighed. The propositions are independent in the sense that the empirical truth or falsity of one does not necessitate the truth or falsity of the others. Therefore, that all four should be positively supported (as well as are the many theoretically related propositions in Vol. 2: The Conflict Helix and Vol. 4: War, Power, Peace) lends additional credibility to each and to the principle that unites them: Libertarianism tends to minimize collective violence.
In conclusion, given my own direct tests of the four propositions ("Libertarianism and International Violence" and "Libertarianism, Violence Within States, and the Polarity Principle") and the results in the literature evidenced here, there is a reasonable basis for accepting the four propositions as probably true. And if they are indeed true, they have obvious implications for a more peaceful world order.
(1) Often the same data bases are used in the separate studies and therefore many of the ratings are not independent. First, 31 of the 47 usable studies tabulated in Tables 5 to 8 are based on either originally collected data or secondary data not used by any other of the studies. Of the remaining 16, all employ different segments of a larger data set (as does Rummel, 1972b, and Schwerin, 1977); independent approaches to the same data set (as in Chan's longitudinal and cross-sectional analyses of Singer and Small's war data), or independent methods applied to the same data base (as for the Zinnes and Wilkenfeld, 1971, Marchov Chain analysis of the Rummel, 1972a, domestic conflict data that they factor analyzed). One of the criteria for determining whether an analysis was usable here was whether its results were independent of the other analyses. Where such independence did not exist or was questionable, the study was not included as usable, or was combined with the analysis it duplicated and both were rated as one. Such was the case, for example, with the 1. K. Feierabend and R. L. Feierabend (1971) and R. L. Feierabend and I. K. Feierabend (1972) studies that are rated together in Table 8.
(2) Considering that I have a self-interest in the results, I may have unconsciously raised the ratings in anticipation of applying the one-level decrement. There is no denying this problem; it is one that permeates much of our research on conflict and violence, because whether we are collecting event data (that we may unconsciously bias), testing hypotheses, or exploring the data, we have mind-sets, implicit theories, and ideological positions that can distort the data collection and results. Although the one-level decrement is at least an attempt to bracket such bias, the only real protection is replication by different researchers with different interests and different biases. If then the results are consistent, we may legitimately discount self-interests. Here I have tried to be as explicit as space allows to facilitate such replication.
(3) Too many diverse things are being reduced to their lowest common denominator--the methods employed here are too gross to be of real use. As far as the tabulation of the different analyses and their rating is concerned, nothing is being reduced. Each analysis is taken as is, in all its diversity. For each analysis I went through all the relevant findings, ignoring or reducing none, and determined whether a correlation coefficient, chi-square, factor loading, or regression coefficient supported or did not support one of the four propositions. Once the different analyses were tabulated, I then calculated the summary statistics shown in Tables 5, 6, 7, and 8. But these only make all the ratings easier to evaluate; the original ratings are still given for those who wish to apply a different way of counting them. Nor is the approach here a gross one. Each analysis was looked at in all its rich detail, including in many cases redoing or adding to the analysis done by a study.
(4) The propositions refer to libertarianism but only the relationship of democracy to violence is being tested. Democracy is a form of libertarianism. As the term is used among contemporary libertarians it incorporates anarchy, liberal democracy (minimum government), and democratic socialism. Theoretically, among these different libertarian types, anarchy should have the least violence, democratic socialism the most; but all three should have less violence than authoritarian and totalitarian systems. There are no comparative empirical studies of anarchies; and few empirical studies include variables that distinguish between liberal and socialist democracies (two that do are my "Libertarianism and International Violence" and "Libertarianism, Violence Within States, and the Polarity Principle"). Accordingly, libertarianism was treated here as a general category including both types of democracy. This neither contradicts the propositions or weakens their tests.
(5) Spurious correlations may account for some of the results; also there may be explanations for the relationships found other than democracy. Of course, we can never be sure that there are not hidden or third variables accounting for our results and unquestionable explanations can only be left to religion. But there are degrees of confidence in one's results; and were this the usual kind of survey trying to accumulate similar results across the literature, the criticism would be particularly potent. Here, however, what was found was predicted and theoretically explained before the survey was done. Although the results may still be spurious, until an alternative theory is tested that explains the results equally or even better, such a possibility should be discounted.
(6) One cannot really say whether democracy will create a more peaceful world; once the whole world is democratic new forces may be unleashed creating even more war than we have now. This simply is saying that we cannot be certain about any route to peace. All we can do is move incrementally and carefully along any prescribed route, always ready to change direction if new theory and evidence show that we are mistaken. I can say here only that theory and empirical results agree, on balance, that libertarianism offers a way of reducing collective violence. This may be wrong and it might be wiser to wait until much more study has been done before drawing any policy implications. But I must ask: Do any of the current policy proposals for creating a more peaceful world have such accumulated empirical research and theory supporting them?
* Scanned from "Libertarian Propositions on Violence Within and Between Nations: A Test Against Published Research Results," The Journal of Conflict Resolution. 29 (September 1985): 419-455. Typographical errors have been corrected, clarifications added, and style updated.
1. Note omitted.
2. This and the previous proposition were combined elsewhere into a Freedom Proposition: freedom inhibits violence (the more libertarian a state, the less it tends to be involved in violence). For greater clarity in surveying the literature, this has been divided into the Dyadic Foreign Violence and Foreign Violence versions.
3. Of course, there are exceptions, as possibly for a carefully constructed first-time critical test of a theory. But in the social sciences our theories and data are not that precise to allow for such tests, even were one to accept philosophically the possibility of such a critical test in science.
4. In alphabetical order, these journals are (with the year in which the screening began shown in parentheses; all were screened up to their most recent number): American Academy of Political and Social Science Annals (1970), American Behavioral Scientist (1970), American Journal of Political Science (1966, formerly Midwest Journal of Political Science), American Political Science Review (1961), American Journal of Sociology (1966), American Sociological Review (1976), Behavioral Science (1964), Behavioral Science Research (1981), British Journal of Political Science (1971), Comparative Political Studies (1968), Comparative Politics (1968), Comparative Social Research (1978), Cooperation and Conflict (1970), Current Research on Peace and War(1978), Democracy (1981), Instant Research on Peace and War (1976), international Interaction (1975), International Journal of Comparative Sociology (1978), International Organization (1970), International Peace Research Newsletter (1982), International Studies Quarterly (1966, successor to Background), Journal of Conflict Resolution (1964), Journal of Peace Research (1964), Journal of Peace Science (1975), The Journal of Politics (1972), Journalism Quarterly (1965), Peace and Change (1976), Peace Research (1969), Peace Research Society (International) Papers (1964), Political Quarterly (1981), Political Science Quarterly (1978), Political Studies (1981), Polity (1979), Review of Politics (1965), Social Forces (1968), 7he Sociological Review (1967), Sociology and Social Research (1969), The Western Political Quarterly (1965), and World Politics (1964).
5. For a more thorough discussion of libertarianism and a scaling of nations, 1975-1980, see "Libertarianism and International Violence."
6. In his review of my Understanding Conflict and War Hakan Wiberg (1982) compared his ratings of 25 analyses to mine. Across the propositions rated, these 25 analyses were distributed by ratings as follows:
Considering that one can say from Wiberg's review that he was not sympathetic to the philosophy or theory of my work, he well may have a bias toward underrating the positiveness of an analysis, as I might have a bias toward overrating it. Therefore, it is interesting to note that his ratings are a little less than midway between mine and the one-level decrement.
7. This is a conservative procedure: It is treating ambiguous results as essentially negative. The logic of the test is this. Assume that there is really no relationship between libertarianism and violence. Then the universe of relevant studies will have a total and average of zero. Assume also that the sample of relevant studies coded for a proposition constitute a random sample from this universe, a fair assumption as we can argue that the characteristics of the sample are independent of libertarianism, violence, and their correlation. Then the probability that the sample average would deviate from the total = average = 0 by as much as they do or more can be determined from the t-distribution.
8. Since Wright (1942) published the first such, there have been relatively few quantitative dyadic analyses of nations; and among these hardly any include relevant variables, not to mention appropriate operationalizations.
9. One use of the probability values for gauging the propositions is in giving us a precise measure of how positive the ratings must be to overcome a small sample of studies.
10. These have been spelled out, especially in terms of a just world order, in my Vol. 5: The Just Peace.
ADELMAN, 1. and C. T. MORRIS (1973) Economic Growth and Social Equity in Developing Countries. Stanford, CA: Stanford University Press.
- - - (1967) Society, Politics and Economic Development: A Quantitative Approach. Baltimore: Johns Hopkins Univ. Press.
ALCOCK, N (1976) "Towards a causal model of violence." Peace Research 8 (July): 91 103.
- - -. (1975) "The political dimensions of compassion." Peace Research 7 (January): I 23.
- - -. (1974) "Freedom and equality among nations." Peace Research 5 (July): 53-67.
- - -. and W. ECKHARDT (1974) "Comparisons between the attitudes and behaviors of individuals and nations." Peace Research 5 (April): 33-44.
ANDERSON, C. W. and B. A. NESVOLD (1972) "A Skinnerian analysis of conflict behavior." Amer. Behavioral Scientist 15: 883-910.
BABST, D. V. (1972) "A force for peace." Industrial Research (April): 55-58.
BANKS, A. (1972) "Correlates of democratic performance." Comparative Politics 4 (January): 217-230.
BARRINGER, R. E. (1972) War: Patterns of Conflict. Cambridge: MIT Press.
BROCH, T. and J. GALTUNG (1966) "Belligerence among the primitives." J. of Peace Research 1: 33-45.
BWY, D. P. (1972) "Political instability in Latin America: the cross-cultural test of a causal model," pp. 223-241 in 1. K. Feierabend and T. R. Gurr (eds.) Anger, Violence, and Politics: Theories and Research. Englewood Cliffs, NJ: Prentice-Hall.
CATTELL, R. B. (1953) "A quantitative analysis of the changes in the culture patterns of Great Britain, 1837-1937, by P-technique." Acta Psychologica 9: 99-12 1.
- - -. and R. GORSUCH (1965) "The definition and measurement of national morale and morality." J. of Social Psychology 67: 77-96.
CHAN, S. (1984) "Mirror, mirror on the wall ... : are the freer countries more pacific? " J. of Conflict Resolution 28 (December): 617-648.
COOPER, M. N. (1974) "A reinterpretation of the causes of turmoil: the effects of culture and modernity." Comparative Pol. Studies 7 (October): 267-291.
CUTRIGHT, P. (1963) "National political development: measurement and analysis." Amer. Soc. Rev. 28: 253-263.
EAST, M. A. and P. M. GREGG (1967) "Factors influencing cooperation and conflict in the international system." Int. Studies Q. 2 (September): 244-269.
- - - and C. F HERMANN (1974) "Do nation-types account for foreign policy behavior?," pp. 269-303 in J. N. Rosenau (ed.) Comparing Foreign Policies: Theories, Findings, and Methods. New York: Halsted Press.
ECKHARDT, W. (1975) "Primitive militarism." J. of Peace Research 12, 1: 55-62.
- - -. (1969) "The factor of militarism." J. of Peace Research 2: 123-132.
- - - and C. YOUNG (1977) Governments Under Fire: Civil Conflict and Imperialism. New Haven, CT: Human Relations Area Files.
- - - (1974) "A preliminary report on civil conflict and inequality." Peace Research 5 (April): 45-50.
FABBRO, D. (1978) "Peaceful societies: an introduction." J. of Peace Research 15 (No. 1): 67-83.
FEIERABEND, 1. K. and R. L. FEIERABEND (1973) "Violent consequences of violence," pp. 187-219 in H. Hirsch and D. C. Perry (eds.) Violence as Politics: A Series of Original Essays. New York: Harper & Row.
- - -. (1971) "The relationship of systematic frustration, political coercion, and political instability: a cross-national analysis," pp. 417-440 in J. V. Gillespie and B. A. Nesvold (eds.) Macro-Quantitative Analysis: Conflict, Development, and Democratization. Beverly Hills, CA: Sage.
- - - and B. A. NESVOLD (1969) "Social change and political violence: cross-national patterns," pp. 632-687 in H. D. Graham and T. R. Gurr (eds.) Violence in America: Historical and Comparative Perspective. New York: Bantam Books.
FEIERABEND, 1. K. with R. L. FEIERABEND and B. NESVOLD (1973) "The comparative study of revolution and violence." Comparative Politics 5 (April): 393-424.
FEIERABEND, R. L. and 1. K. FEIERABEND (1972) "Appendix: invitation to further research-designs, data, and methods," pp. 369-416 in 1. K. Feierabend et al., (eds.) Anger, Violence, and Politics: Theories and Research. Englewood Cliffs, NJ: Prentice-Hall.
FLANIGAN, W. H. and E. FOGELMAN (1970) "Patterns of political violence in comparative perspective." Comparative Politics 3 (October): 1-20.
GOLDMAN, K. (197 1) International Norms and War between States: Three Studies in International Politics. Stockholm, Sweden: Laromedelsforlagen.
GREGG, P. and A. S. BANKS (1965) "Dimensions of political systems: factor analysis of a cross-polity survey." The Amer. Pol. Sci. Rev. 59 (September): 602-613.
GURR, T. R. (1979a) "Political protest and rebellion in the 1960s: the United States in world perspective," pp. 49-76 in H. D. Graham and T. R. Gurr (eds.) Violence in America: Historical and Comparative Perspectives. Beverly Hills, CA: Sage.
- - -. (1979b) "Some characteristics of political terrorism in the 1960s," pp. 23-49 in M. Stohl (ed.) The Politics of Terror: A Reader in Theory and Practice. New York: Dekker.
- - -. (197 1) "The conditions of civil violence: first tests of a causal model," pp. 187-249 in J. V. Gillespie and B. A. NESVOLD (eds.) Macro-Quantitative Analysis: Conflict, Development, and Democratization. Beverly Hills, CA: Sage.
- - -. (1969) "A comparative study of civil strife," pp. 572-632 in H. D. Graham and T. R. Gurr (eds.) Violence in America: Historical and Comparative Perspective. New York: Bantam Books.
- - - and M. 1. LICHBACH (1979) "Forecasting domestic political conflict," pp. 153193 in J. D. Singer and M. Wallace (eds.) To Augur Well: Early Warning Indicators in World Politics. Beverly Hills, CA: Sage.
HAAS, M. (1974) International Conflict. Indianapolis: Bobbs-Merrill.
HAZLEWOOD, L. A. (1973) "Externalizing systematic stresses: international conflict as adaptive behavior," pp. 148-190 in J. Wilkenfeld (ed.) Conflict Behavior and Linkage Politics. New York: David McKay.
HIBBS, D. A. JR. (1973) Mass Political Violence. New York: John Wiley.
MARKUS, G. B. and B. A. NESVOLD (1972) "Governmental coerciveness and political instability: an exploratory study of cross-national patterns." Comparative Pol. Studies 8 (July): 231-244.
MELKO, M. (1973) 52 Peaceful Societies. Oakville, Ontario: Canadian Peace Research Institute Press.
MIDLARSKY, M. 1. (1975) On War: Political Violence in the International System. New York: Free Press.
- - -. and S. T. THOMAS (1975) "Domestic social structure and international warfare," pp. 530-548 in M. A. Nettleship, R. Dalegivens, and A. Nettleship (eds.) War, Its Causes and Correlates. The Hague: Mouton Publishers.
NESVOLD, B. A. (1971) "Scalogram analysis of political violence," pp. 167-186 in J. V. Gillespie and B, A. Nesvold (eds.) Macro-Quantitative Analysis: Conflict, Development, and Democratization. Beverly Hills, CA: Sage.
ORBELL, J. M. and B. M. RUTHERFORD (1973) "Can Leviathan make the life of man less solitary, poor, nasty, brutish and short'?" British J. of Pol. Sci. 3: 383-40T
OTTERBEIN, K. F (1970) The Evolution of War: A Cross-Cultural Study. New Haven, CT: Human Relations Area Files.
- - -. (1968) "Internal war: a cross-cultural study." Amer. Anthropologist 70: 277-289.
PHILLIPS, W. and D. R. HALL (1970) "The importance of governmental structure as a taxonomic scheme for nations." Comparative Pol. Studies 3 (April): 63-89.
ROSENAU, J. N. and G. D. HOGGARD (1974) "Foreign policy behavior in dyadic relationships: testing a pre-theoretical extension," pp. 117-149 in J. N. Rosenau (ed.) Comparing Foreign Policies: Theories, Findings, and Methods. New York: Halsted Press.
RUMMEL, R. J. (1984) "Libertarianism, Violence Within States, and the Polarity Principle," Comparative Politics 16 (July): 443-462.
- - -. (1983a) "Libertarianism and International Violence," J. of Conflict Resolution 27 (March): 27-71.
- - -. (1983b) Social Field Theory, Libertarianism and Violence. Honolulu: Department of Political Science, University of Hawaii.
(1981) Vol. 5: The Just Peace. Beverly Hills, CA: Sage.
- - -. (1979a) National Attributes and Behavior. Beverly Hills, CA: Sage.
- - -. (1979b) Vol. 4: War, Power, Peace. Beverly Hills, CA: Sage.
- - -. (1976) Vol. 2: The Conflict Helix. Beverly Hills, CA: Sage.
- - -. (1976a) Understanding Correlation. Honolulu: Department of Political Science, University of Hawaii.
- - -. (1975-1981) Understanding Conflict and War (Vols. 1-5). Beverly Hills, CA: Sage.
- - -. (1972a) Dimensions of Nations. Beverly Hills, CA: Sage.
- - -. (1972b) "U.S. foreign relations: conflict, cooperation, and attribute distances," pp. 71-113 in B. Russett (ed.) Peace, War, and Numbers. Beverly Hills, CA: Sage.
- - -. (1970) Applied Factor Analysis. Evanston, IL: Northwestern Univ. Press.
- - - (1967) "Understanding Factor Analysis," J. Of Conflict Resolution 11 (December 1967): 444-480.
- - -. (1963) "Dimensions of conflict behavior within and between nations." General Systems Yearbook of the Society for General Systems Research 8: 1-50.
RUSSETT, B. M. and R. J. MONSEN (1975) "Bureaucracy and polyarchy as predictors of performance: a cross-national examination." Comparative Politics 8 (April): 5-3 1.
SCHWERIN, E. W. (1977) "U.S. foreign relations for 1963," pp. 301-314 in R. J. Rummel (ed.) Field Theory Evolving. Beverly Hills, CA: Sage.
SINGER, J. D. and M. SMALL (1972) The Wages of War 1816-1965: A Statistical Handbook. New York: John Wiley.
SMALL, M. and J. D. SINGER (1976) "The war-proneness of democratic regimes, 1816-1965." The Jerusalem J. of Int. Relations I (Summer): 50-69.
SOROKIN, P. (1957) Social and Cultural Dynamics. Boston: Porter Sargent.
- - -. (1937-194 1) Social and Cultural Dynamics (Vols. 1-4). Boston: Porter Sargent.
TEXTOR, R. B. (1967) A Cross-Cultural Summary. New Haven, CT: Human Relations Area Files.
THOMAS, H. (1979) A History of the World. New York: Harper & Row.
TILLY, C., L. TILLY, and R. TILLY (1975) The Rebellious Century 1830-1930. Cambridge: Harvard Univ. Press.
VINCENT, J. E. (1979) Project Theory: Interpretations and Policy Relevance. Washington, D C: Univ. Press of America.
WEEDE, E. (1984) "Democracy and war involvement." J. of Conflict Resolution 28 (December): 649-664.
- - - (1970) "Research communication: conflict behavior of nation-states." J. of Peace Research 3: 229-235.
WIBERG, H. (1982) "Review essay: Rudolph J. Rummel: Understanding Conflict and War Vols 1-5." J. of Peace Research 19, 4: 369-386.
WRIGHT, Q. (1942) A Study of War (Vols. I and 2). Chicago: Univ. of Chicago Press.
YOUGH, S. N. (1981) "Modernization, institutionalization, and political violence: a cross-national study." J. of East Asian Affairs I (January): 1-48.
ZINNES, D. A. and J. WILKENFELD (197 1) "An analysis of foreign conflict behavior of nations;' pp. 167-213 in W. F Hanrieder (ed.) Comparative Foreign Policy: Theoretical Essays. New York: David McKay.