Statistics of Democide
Contents | Figures | Tables | Preface
Other Democide Related Documents On This Site
All murder'd: for within the hollow crown|
That rounds the mortal temples of a king
Keeps Death his court.
----Shakespeare King Richard III, ii, 152
I had already found this hypothesis consistent with the case studies of all megamurderers that I reported in my Death By Government and with the limited statistical analyses I reported there. As a result of that work I restated the hypothesis as a social principle: Power kills, Power kills absolutely. The diverse analyses I give here consistently and solidly further confirm this.
In sum, among a variety of socio-economic, cultural, social diversity, geographic, and other indicators, the best way of accounting for and predicting democide is by the degree to which a regime is totalitarian. That is, the extent to which a regime controls absolutely all social, economic, and cultural groups and institutions, the degree to which its elite can rule arbitrarily, largely accounts for the magnitude and intensity of genocide and mass murder. The best assurance against this democide is the democratic openness, political competition, regularly scheduled elections, and limited government of a free people.
That Power kills is the primary and for domestic democide singular general explanation of democide. This is true even when we consider how regimes differ in their underlying ethnic, religious, and racial diversity. It is true when we consider whether they are Christian or Moslem, or the cultural region they are from. It is true when taking into account their different levels of education or economic development. It is true for their differences in sheer size. And it also is true even for the trend of overall democide through time.
However, the tendency of regimes to fight severe domestic rebellions or foreign wars also predicts to democide. But for both rebellions and wars Power is also a causal agent. The more totalitarian a regime's power, the more total its wars or rebellions are likely to be; and the more totalitarian power and bloody its wars and rebellions, the more it probably will commit democide.
I can best sum up the conclusions of Statistics of Democide by three charts reproduced from later chapters. The first, Figure 1.1, shows the mean domestic democide (logged) at different levels of totalitarian power.
Figure 1.2 displays how the democratic and totalitarian scales making up totalitarian power act together to sharply increase democide the more closely a regime is to absolute Power. Finally, Figure 1.3 shows the yearly trend over this century of democide, democide by totalitarian regimes, and that by democratic ones. These figures speak for themselves.
For the rest of this summary I will briefly sketch the questions and considerations, techniques and methods used, and results of the statistical chapters. My purpose is to communicate an understanding of the enterprise that may get lost among the diverse statistical considerations involved in each chapter.
I applied a variety of analyses, including analysis of variance, multiple linear and curvilinear regression, component (factor) analysis, canonical analysis, and discriminant analysis. And I used plots extensively to uncover or display relationships. The paramount question throughout was whether the principle that Power kills holds up under a variety of methods, under diverse ways of partialling the data on democide and politics, and different conditions.
The theoretical framework within which I carried out these analyses was social field theory.
With all this as background, the first empirical problem was to define the empirical patterns of democide and their best indicators. To this end I subjected fourteen different kinds of democide, including deaths from genocide, concentration camps, forced labor, terror, massacres, and the like, to various component analyses. This is a powerful and robust method for defining independent empirical patterns in data and partialling out the influences of third, fourth, and other variables. The democide patterns thus uncovered were five, one centrally involving domestic democide and another foreign democide. A third pattern had at its center the domestic democide annual rate; a fourth the democidal bombing dead. Genocide stood out by itself as a singular pattern. The five patterns are statistically independent, which means that in general domestic and foreign democide and the other patterns have quite different specific causes and conditions, although Power may still be a general cause.
With the empirical patterns of democide in the social field thus defined, the next problem was to test for the stated relationship between a democracy versus totalitarianism dimension--Power--and these empirical patterns. The theoretical expectation was that Power would be most related to domestic democide, genocide, and the annual rate of democide, while the islands of Power created even within democratic regimes in time of war would obfuscate the relationship between Power and foreign democide.
To test this demanded quantifying Power beyond the simple three-point scale of democracy, authoritarianism, and totalitarianism used Death By Government. The logic of the Power principle also demanded including in the analysis state regimes that had not committed any democide. Accordingly, I added an additional seventy-three such regimes with no democide to the 141 with. I selected these to reflect the major variation among political characteristics, as well as the major cultures and geographic regions. The resulting sample of 214 state regimes was the basic sample in all subsequent analyses and comprised about half of all state regimes that existed sometime during 1900 to 1987.
To then best quantify Power, I created or selected from the literature seventeen measures of a regime's political characteristics, such as whether it had an elected legislature, was communist, the extent of its political power, the power of the traditional elite, the degree of political competition, and so on. I then applied component analysis to these measures and delineated five statistically independent political patterns. By far, the most dominant pattern was that of totalitarian power (or Power for short--this now empirically defines what I meant by Power in Death By Government and previous pages), best indicated by an inverse combination of democratic and totalitarian scales (those used in Figure 1.2) that I labeled TotalPower. This became the basic indicator of Power for all subsequent analyses. The other political patterns comprised political power, traditional elite power, monarchy, and authoritarianism versus totalitarianism.
With these patterns so defined, I could carry out the tests of Power's relationship to democide. I did this first by determining how the patterns of democide and politics were interrelated. I should find that TotalPower would be most related to the democide, genocide, and annual rate patterns, and this is what resulted from component analysis. Indeed, the three democide patterns combined into one interrelated cluster, at the center of which was TotalPower. Secondly, I did an interactive regression analysis of the different democide patterns on all the political characteristics and selected interaction terms (e.g., TotalPower squared). I found, as I should have, that for the three democide patterns, TotalPower was the best and only significant predictor. For domestic democide and the annual rate, this was TotalPower squared, which means that as regimes approach absolute Power the effect of Power on democide multiplies.
This effect was verified through the three-dimensional surface plot reproduced above as Figure 1.2. Moreover, a triangular plot of democratic, totalitarian, and authoritarian scales further showed the sheer dominance of the totalitarian end, that is of absolute Power, in accounting for the domestic democide pattern. A plot of the mean domestic democide for different levels of TotalPower, reproduced above as Figure 1.1, and the contingency analysis of different domestic democide magnitudes against different levels of TotalPower, further confirmed and displayed this fundamental connection.
This should not end the tests, however. The observed relationships were within a social field in which many underlying forces and conditions could produce misleading empirical results, even creating high correlations (on the nature of correlation, see Understanding Correlation) that disappear when other measures are taken into account. So I did several additional analyses to test what I had so far found.
One popular explanation for the linkage between Power and democide is the existence of significant racial, ethnic, religious, national, and other such minorities. Indeed, this social pluralism may be the underlying cause of democide and the Power correlation only epiphenomena. To determine if this were so, I first had to quantify social diversity. I did this by selecting from the literature eight measures of social diversity, including those of ethnic divisions, religious divisions, overall diversity, and national disunity. As for the previous analyses, I reduced these measures through component analysis to their basic patterns, which are two. One is a very strong general pattern defined by a general diversity index, and the second a single variable pattern comprising a percent measure of the minorities at risk under a regime.
I then did a component analysis of the two indicators of social diversity together with the indicators of the democide patterns. In contradiction to what the literature would predict, there was absolutely no relationship between diversity and democide patterns or indicators. This was further confirmed through a scatter plot. These results alone say much, for they show that in spite of the obvious relationship between race or ethnicity in such cases as the Holocaust and Turkey's genocide of its Armenians, in general diversity and democide, including genocide, are statistically uncorrelated.
But this finding is a preliminary to further tests of the relationship between democide and Power. To do this I next included all democide, political, and diversity indicators together in a component analysis. The relationship between Power and democide was unaffected, showing that diversity is not a situational explanation or condition for this relationship. This also is true even when I did the analysis within high and low diversity groups alone.
However, it may be instead that culture is responsible for the Power-democide linkage. Following the same procedures as above, I quantified the culture of a nation through fourteen measures, including the percent Christians a regime governed, the percent Moslems, whether anti-women or not in terms of pro-women legislation, clan basis, and six regional dichotomous measures locating regimes in one of the major cultural areas, such as Europe, Central and South American, and South Asia. I did the usual component analyses to isolate the cultural patterns and select their indicator. The major cultural patterns among all 214 regimes are African, Moslem, Latin American, Asian, and those with an English influence. When I did a component analysis of their indicators along with the five for the democide patterns, as I had done for diversity, I could uncover no relationship between culture and democide.
The more significant component analysis came next, for then I included with the cultural and democide indicators, those for politics and social diversity. Again, there was no effect on the positive relationship between Power and domestic democide. Nor did the inclusion of culture alter the lack of relationship between diversity and democide.
This is the general finding for all 214 regimes. It may be, however, that when the analyses are carried out within a cultural pattern or region, interrelationships may change. And they do for some regions. For non-Moslem, European, or Asian regimes there is no significant change in the Power-democide connection. For Christian regimes the relationship is less strong. For Central/South American regimes a relationship between Power and democide is still there but made more complex; only for African and Moslem regimes is the connection eliminated altogether. These results warn against assuming without further analysis that within all cultural regions Power and democide go together in a straight forward way.
Still, for most cultural regions and in general the principle that Power kills holds simply and solidly. But if social diversity and culture generally do not effect the relationship, perhaps the education of a regime's people, or their socio-economic development and modernization may inhibit democide or change the Power-democide equation. Perhaps the critical context is whether a nation is large or small, or has many or few people. Relying particularly on the accumulated results of published cross-national component and factor analyses (on factor analysis, see "Understanding Factor Analysis"), I selected twenty-one indicators of the major independent, empirical socio-economic and geographic cross-national patterns. Among these the dominant one I tried to index is wealth, a comprehensive pattern among nations that includes measures of economic development, the quality of health, the transportation system, educational attainment, and the like. The second pattern involves political variables, essentially reflecting the dimension of Power already measured. The third is national power, the natural and demographic resources available to a regime, and for obvious reasons I was particularly interested is seeing it well represented in the analysis. Measures of other patterns, such as that of population density, were also included.
As I did for the political, social diversity, and cultural measures, I first component analyzed these twenty-one socio-economic and geographic measures to find their empirical patterns for the 214 regimes. There was no surprise. Among the 214 regimes the same patterns found in cross-national data emerged, primarily wealth, national power, and density. I then included the indicators of these and the other socio-economic patterns in an overall component analysis of all the democide, political, diversity, and cultural indicators, twenty-four in all. This was now a near comprehensive analysis of the social field and the context within which democide occurs (only leaving out measures of war and rebellion soon to be discussed). What happened? No change. Power remained tied in with domestic democide, and no other measure besides political power had any even moderate relationship to domestic democide.
It may be, as with cultural regions, that there is a difference here between rich and poor regimes. I thus redid the analysis within each of these groups. Although there was some shifts among patterns and correlations, the relationship between domestic democide and Power remained.
Finally, I filled out the context of democide by including in the analysis the number of war and rebellion-dead for each regime. These are especially important. Unlike the other contextual measures that I included (either because they were favored in the literature as causes or conditions of democide or they filled out the social field) I had a theoretical reason to expect that these measures of non-democidal violence to be highly correlated with democide. They manifest or themselves bring about a breakdown in the structure of expectations and supporting balance of powers within a society and its regime. Thus war and rebellion catalyze democide, promote it (as in democidal urban bombing), or provide an excuse and cover for it to be committed. Moreover, one would expect that the more warlike a regime the more likely it would commit democide.
And this comes out quite clearly when both war-dead and rebellion-dead are component analyzed along with the democide patterns. The number killed in rebellions during the life of a regime is highly related to its domestic democide; its war-dead to foreign democide. These then are tough tests for Power. Will Power remain related to democide when I include these measures of rebellion and war with them in a component analysis? The answer is a straight forward yes. There is no change in the Power-domestic democide nexus.
Even when I put all the indicators of democide, politics, diversity, culture, etc., together and component analyze them, the relationship between domestic democide and Power remains largely the same. And the characteristic severity of rebellion is correlated with the domestic democide pattern; the characteristic severity of war with foreign democide.
The causal linkages for the Power-democide-war-rebellion connections are theoretically clear. Power not only causes democide, but also the blood shed in a regime's wars and the rebellions against it. And a regime's characteristic involvement in such violence is also related to its democide. Power thus directly causes democide, while also indirectly causing it through its influence on the occurrence and characteristic severity of rebellion and war. Several plots were made to test for this relationship of Power to war and rebellion, which with the exception of the war-dead of the democracies in World War II--a war unleashed by totalitarian power--were consistent with the theory.
Up to this point I have shown that for all 214 regimes, including all 141 with democide of some sort, an indicator of an empirical pattern of Power among a variety of political characteristics of regimes is most highly correlated with a pattern of domestic democide (which also involves genocide and the annual democide rate), as expected by theory. Second, I have shown that a variety of contextual measures spanning the social field of regimes have virtually no effect on this relationship in general. Indeed, the correlation of Power with domestic democide is second to none and almost unique except for political power and the characteristic intensity of rebellions against a regime.
But this has been a social field type of analysis, relying on component analysis as the main vehicle for uncovering interrelationships and partialling out contextual and situational influences. What would happen with straight forward regression analysis? Now regression analysis is a useful way of assessing how well the variation in a variable can be accounted for (the favored term is "predicted") by some other variables. I have already used regression analysis to verify the dominant role of Power among political patterns alone in predicting domestic democide. Now I will use it to successively regress the overall democide itself and then the different democide patterns on all the political, diversity, cultural, etc., indicators together, and some theoretically specified interaction terms, such as TotalPower squared, war-dead squared, and an indicator of national power times war-dead; twenty-four independent variables in total.
As a result, the best predicted (accounted for, explained) was overall democide (which includes both domestic and foreign democide, of course), with six indicators predicting over 70 percent of the variation in domestic democide across 214 regimes. And the best predictor was TotalPower squared. As in the previous regression with Power, its causal influence is magnified the nearer to absolute Power (see Figure 1.2, for example). The lesser but significant predictors involved rebellion-dead, war-dead, and national power or their interaction terms. I then did a contingency table analysis of the actual versus predicted domestic democide resulting from this regression and found that Power and the other five significant indicators or interaction terms were able to well predict regimes committing no or little democide, as well as the megamurderers.
However, a difficulty with regression is that only one dependent variable can be analyzed at a time. Consequently, I also used canonical analyses, which enabled me to take the democide space of the fourteen types of democide--the raw democide data I began with--and fit it to the space of all twenty-three independent indicators, including interaction terms. Canonical analysis is like component analysis, except that rather than delineating independent patterns among all the variables, one is finding the independent patterns among one set of variables that best fit another set. The result of applying this method was a pattern of social indicators that accounted for 85 percent of the variation in a pattern of democide across all 214 regimes. The best indicators of this were Power, national power, and war and rebellion-dead or their interaction terms.
I also did a discriminant analysis. This is a form of canonical analysis, except the dependent variables comprise categorical groups. In my case, the groups were those with no or little democide, with democide between 1,000 to 9,999 killed, between 10,000 and 99,999, and so on for up to the group of deka-megamurderers. I found essentially the same small set of predictors that came out of the regression and canonical analysis. As clear from a contingency analysis of the results, these indicators and interaction terms, centrally involving Power, war, and rebellion, were well able to predict whether regimes had democide and at what level.
Finally, I looked at total democide by year for all the regimes. I found, as Figure 1.3 shows, that democide peaked during world War II, that it is related through time to the severity of war and rebellion, and especially that by far the larger part of the overall democide trend is due to that committed by totalitarian power--that is, Power.
* From the pre-publisher edited manuscript of Chapter 1 in R.J. Rummel, Statistics of Democide, 1997. For full reference to Statistics of Democide, the list of its contents, figures, and tables, and the text of its preface, click book.
1. In Rummel (1997, Part 2) I develop the theoretical argument for the inverse relationship of democracy to collective violence and democide.
Go to top of document