Statistics of Democide
Chapter 1: Summary and Conclusions [Why Democide?...]
Other Democide Related Documents On This Site
As we know from previous chapters, 141 state regimes committed some form of democide. The determination of how much democide these regimes committed overall and for what years involves several data aggregation decisions. For some regimes their democide was more or less spread over their lifetime or the democide estimates in the literature were such that no specific years for their democide could be determined. In both cases I calculated their democide for each year as: (regime's total democide)/(regime's duration in years). For some other regimes, such as Pakistan in 1971, Burundi in the same year, or Indonesia in 1965-66, there were specific years in which most if not virtually all their democide was committed, and I allocated their total democide accordingly. For the remaining regimes their democide could be divided among time periods in their lives. In this regard some regimes bear special mention.
For the Soviet Union, with a huge democide of 62,000,000 that would markedly effect the overall democide time series, I divided its total democide into the political periods given in my Lethal Politics. Thus I proportioned the Soviet total civil war democide of near 3,300,000 for each year 1917 to 1922; the NEP period democide of 2,200,000 for each year 1923-1928, and so on. I also similarly divided and proportioned Nationalist and communist Chinese democide by political period. There were clearly two periods in Nazi German's democide, pre-invasion of Poland domestic democide and post-invasion occupation of Poland in September, 1939, and most of the rest of continental Europe in subsequent years. Germany's democide was accordingly differently proportioned for each year during these two different periods.
For the United States, United Kingdom, and France, among other democracies, their domestic and foreign democide was tabulated for those years in which it occurred. For the United States, for example, this largely would be in the Philippine Independence War at the beginning of the century, World War II indiscriminate bombing, and Vietnam. The United Kingdom committed almost all its democide in and immediately after World War I and through the years 1941 to 1945 in World War II. The new post World War II French republic committed its major democide during the Algerian War of Independence.
Similar periodization was done for the communist regimes of Eastern Europe. The greater majority of their killing was done in the years immediately after communist governments were established by the Soviet Union or seized power. Not only were executions then wide-spread, but in each forced labor camps were established on the Soviet model. But with the death of Stalin in 1953 and subsequent reform within the totalitarian model, reform similarly occurred in Eastern Europe. I took account of these changes in proportioning the democide of each Eastern European regime to its earlier years.
And so on. Space does not allow for the inclusion of the full 88-years by 141 regime table, but the periodization can be reconstructed from the information given elsewhere in this book, such as that of Table 4.1A for Cambodia.
Once I determined the democide for each year of each regime, 1900 to 1987, I then summed the democide across the regimes for each year to get a year-by-year total. I did this also for all non-state regimes. The area plot of Figure 23.1 compares both year-by-year totals. The filled in area is that of state regimes and the white beneath the topmost line plot is that democide added by the non-state regimes. As can be seen, virtually all democide is due to state regimes. The most democide non-state regimes add is in the early 1920s and in the post-War II years. The former is largely due to the White armies that established independent regimes during the Russian civil war, and the latter to the Chinese communists armies that took over large parts of China during the concluding phase of the Chinese civil war.
Before evaluating state regime democide, it will be helpful to also show the war-dead by year. I summed for each year up through 1980 the battle-dead for all international and civil wars from Small and Singer's Resort to Arms (1982). Figure 23.2 shows the result along with the state-regime democide. I have labeled some of the period's episodes to provide an explanation for the plots highs or blips.
World War II clearly appears the greatest context, catalyst, or excuse for democide in this century. To fully appreciate this democide I must again point out that I have tried to clearly separate it from battle-dead or those noncombatants killed as a by-product of military action. This World War II democide counts the genocide against the Jews and mass murder of the Poles, Ukrainians, Chinese, and others. The degree to which this democide exceeded the war's battle dead can be seen from the plot.
As also evident, democide tends sharply upward from World War I to World War II and then with some bumps and jumps tends to slope downward after that toward the pre-World War I level. I argue that this mirrors the growth and decline of absolute totalitarian power. With the Bolshevik coup in 1917 we saw for our century the birth of the first true totalitarian power over a major country. The world was then made up of democracies, monarchies, feudal political systems, and warlords (as in China). The consolidation of totalitarian power in what became the Soviet Union was soon followed by the birth of fascism in Italy, Nazism in Germany, military-fascism in China, Japan and then Spain, and fascism in Eastern Europe. Fascism was not necessarily totalitarian, but it was more absolutist than preceding authoritarian or monarchical regimes and in some places, as in Japan and especially in Germany, the fascist political system took over the control of virtually all aspects of society.
With the end of world War II fascism was defeated, but replaced in many parts of Europe and Asia with totalitarian communism. The democide associated with the communization of Eastern Europe and China can be seen in Figure 23.2. But since the early years of communist consolidation and revolutionary social reconstruction, and especially in the Soviet Union after the death of Stalin and in China with Mao Tse-tung's death, the most totalitarian and absolutist type of communism was in decline, its subsequent victories in South East Asia and Africa notwithstanding.
The cross-regime results of previous chapters have shown that a regime's totalitarian power, and its characteristic war and rebellion-dead, are closely associated with democide. The plots in Figure 23.2 clearly show this association of democide with war and rebellion-dead across years. The serial product moment correlation is .66, but this is misleading. (On interpreting this coefficient, see Understanding Correlation.) From the plot we can see that each major war causes a jump in democide, but that is within an overall increasing or decreasing trend in democide centered around World War II. The trends upward and downward themselves, I argue, are due to the increase or decrease in totalitarian power.
To also picture this, I will use the 0-18 point TotalPower scale from previous chapters and divide it into three parts. The first part will consist of those among the 214 regimes of my sample coded 0-4, the most democratic regimes. I will call these the democracies. The second part will be those regimes coded 14-18, those exercising the greatest totalitarian power. These I will label the totalitarians. Those regimes coded in between I will discard.
I then counted for each year the number of democratic versus totalitarian regimes and overlaid the democide plot with the resulting count per year. But this is misleading, because it would equate regimes with very small populations with those having very large ones. For regimes with a small population they could massacre virtually every citizen and not make a blip on the plot. I will instead total the yearly populations for all the democracies in our sample of 214 state regimes and those separately for the totalitarians. The population total I will use for each regime is its population at the mid-year of its life span, counted from 1900 for those regimes in existence then and ending with 1987. So for the United States, its population for mid-year 1944 was allocated to each year 1900 to 1987. The exception to this procedure was for the population controlled by Germany during its occupation of Europe and that by Japan at the greatest extent of its Asian empire. I used the population controlled by each for their world War II years. Once I did such allocation for all democracies and totalitarians, I separately summed the populations for each group and plotted it against total democide. The result is shown in Figure 23.3.
As can be seen from the figure, the increase in democide is correlated with that in totalitarianism, understood as the population controlled by these regimes, up through World War II. But democide gradually decreased after this in spite of the huge increase in totalitarian populations after the war. The huge postwar totalitarian population lasted until the late 1970s and early 1980s, when with reforms in the Soviet Union, China, and other totalitarian regimes and the disappearance of absolutist near totalitarian regimes in Uganda and Somalia, there was a sharp drop in this population. Meanwhile, democracy has tended to increase and, except for a brief setback in India, as shown by the population chasm in the 1970s, has maintained its level achieved in the 1950s for our sample of regimes.
Does not the plot of Figure 23.3 negate the presumed relationship between democide and totalitarianism?
Totalitarian power is, theoretically, a necessary and not sufficient cause of democide. The greatest democide is associated with totalitarian power, but totalitarian power does not necessarily mean that tremendous democide will be committed. The underlying history of the post-World War II period displays this. The most democide occurred when totalitarians seized power and began to destroy the old socioeconomic and cultural structures and construct new ones. This also included eliminating the old elite and potential and actual opponents. But within several years or a decade or so, the original revolutionaries were purged or died, new and less fanatical revolutionaries seized or got power, and unworkable policies and impossible plans were seen for what they were and corrected. In other words, there was a pulling back from idealistic revolutionary aims, gradual reform, and an increasing reluctance to kill. This process occurred, for example, in the USSR (over a much more extended time period), China (over a less extended period), Poland, Hungary, Czechoslovakia, and to a lesser extent in East Germany. The result was that after the early inception of such regimes democide has declined in spite of the continued existence of these totalitarian regimes, and that decline is displayed in the figure. Eventually, even the regimes themselves have collapsed, as the figure also shows. Thus as totalitarianism grows, so does democide, but while totalitarianism is maintained at near its peak for some thirty years, democide declines. For this reason and that the stimulus of war and rebellion is also involved, there is a low positive serial correlation between totalitarian populations and democide of .25.
However, I can control for all this by overlaying the democide curves for totalitarian, democratic, and other regimes. This is done in Figure 23.4.
Now it can be seen that the vast majority of democide was perpetuated by totalitarian regimes, even though they comprise only twenty-nine out of our sample of 214 state-regimes, or 14 percent. Some was done by the thirty democracies in the sample, but in the context of that done by the others it was relatively little, as shown in the figure. Moreover, although World War II caused the huge spike in democide during the war-years, the overwhelming democidal killing was done by totalitarians. The product moment correlation between totalitarian democide and overall democide is .988. The probabilistic necessity of totalitarianism for democide could not be better shown.
I can now make more precise what is seen in all these figures. Table 23.1 gives the result of regressing democide by year onto the four major series we have looked at, democide by totalitarian regimes, the total population of these regimes, the total population for democratic regimes, and the yearly war and rebellion-dead. The annual democide committed by the democracies was relatively small, as Figure 23.4 shows, and thus not included ignored in the regressions.
The first regression shown includes the annual totalitarian democide and consistent with the high correlation evident in Figure 23.4, the multiple correlation is .99. The question is then, given the obviously very close association between the two, does war and democracy add anything significant to the regression, which they should do by theory? The answer can be seen from the t-values. That for the democratic population is significantly negative, as it should be; that for war and rebellion is highly significant in the positive direction (two-tailed probabilities are shown, but since the theoretical direction is stipulated and empirically correct, these should be divided in half). Totalitarian population is not significant at all, since its covariance with total democide is taken up by the totalitarian democide variable.
Would this and the others measures be good predictors if the very highly correlated totalitarian democide variable is removed? The answer is given be the second regression, which has a good multiple correlation of .77. The former predictors are even more significant in the same direction, and totalitarian population is now also highly significant. In words, the relative size of the democratic and totalitarian populations and the number of war-dead in war and rebellions significantly predicts the amount of democide in the world. An almost perfect prediction is given if one adds the amount of totalitarian democide.
All this further confirms our fundamental findings in previous chapters. Whether looking at this across regimes or years, Power kills, either directly in causing democide, or indirectly in causing war and rebellion which themselves then cause democide.
* From the pre-publisher edited manuscript of Chapter 23 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. Rummel (1990, Table 1.1).
2. From Rummel (1991, Table 1.1).
3. The product moment correlation coefficient used here only measures how close the shape of two curves are, and not the similarity in their magnitudes. But that the magnitudes of the total democide and totalitarian democide curves are also close can be seen from Figure 23.4.
4. These are serial regressions with the possibility of serial dependency. Even if I conceive of the p-values as combinatorial probabilities, and not for a sample, if the data are serially dependent, no significance can be assessed. However, I assume that the amount of democide, or war-dead, or the totalitarian population for a regime can be zero in one year and non-zero in the next (that is, the totalitarian regime can be replaced by a nontotalitarian one). This is not true of national population or GNP, for example, where for a nation one can hardly assume that the series will be zero for any year or even change by any large amount. But for regimes, there can be no war, then intense war for one, two, or more years, then no war. Similarly with democide. How totalitarian populations can decrease sharply from one year to the next can be seen for the early 1980s in Figure 23.4. And the total, totalitarian regime population among the 214 regimes for the first decade of the century was zero
5. Consistent with the analyses of previous chapters it may seem useful to do component analysis (so called T-factor analysis) of the years across the regimes to identify time series patterns in annual democide, or a component analysis (S-factor analysis) of regimes across the years to identify groups of regimes with similar democide patterns across the years. The reason this is not done here is because, with few exceptions such as the United States or Canada, the regimes usually exist for no more than a decade or so. There is no choice but to treat the years for which they do not exist as inapplicable (missing data). But to then estimate these missing data by inserting zeros would be nonsense. However if one analyzes around the missing data by omitting pairs with one or both years missing, given the large amount of missing data the resulting non-Gramian correlational matrix would cause serious distortions in the resulting component analysis, such as large negative eigenvalues and variable correlations with the orthogonal factors exceeding an absolute 1.00.