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

Other Democide Related Documents On This Site


"Democide vs genocide. Which is what?"

"War isn't this century's biggest killer"

"How many did communist regimes murder?"


"Democide in totalitarian states: mortacracies and megamurderers"

"The Holocaust in comparative and historical perspective"

Graduate Syllabus on Repression and Democide


"Power kills: genocide and mass murder"

"Power predicts democide"


Lethal Politics

China's Bloody Century

Democide: Nazi Genocide and Mass Murder

Death By Government

Statistics of Democide (entire)

Other Method Related/Applied Documents On This Site

On Methods:

Understanding Correlation

event data collection

Application of Correlation/Factor Analysis

"Libertarianism and International Violence"

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

Statistics of Democide

Vol. 2: The Conflict Helix

Vol. 4: War, Power, Peace

Power Kills

"The Conflict Helix and the Probability of a Korean War,"

"Power kills: genocide and mass murder,"

"Power predicts democide,"

"Is collective violence correlated with social pluralism?"

Application of Event Data Collection

"Libertarianism and International Violence"

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

Vol. 2: The Conflict Helix

Vol. 4: War, Power, Peace

"Is collective violence correlated with social pluralism?"

Application of Democide Collection Methods and Resulting Estimates

Lethal Politics

China's Bloody Century


Statistics of Democide

"Power kills: genocide and mass murder,"

"Power predicts democide,"


By R.J. Rummel

The statistical problem of estimating democide is clear, if not easily resolvable: how to determine within some range of error the most likely estimate of those murdered by the various governments of China (the de facto governments), given all kinds of estimates in the literature of all kinds of killing during all kinds of events in all kinds of time periods; with no definitive, authoritative estimate available. There is no entirely satisfactory solution to this. Available estimates are neither comprehensive or independent enough to apply formal probability models (sometimes there is only one or two estimates of a particular massacre), nor are their any census results that can be relied on for formal demographic analysis.

The approach has to be one of reasonable approximation, therefore, which involves successively narrowing the range of estimates to what an hypothetical, reasonable analyst would arrive at in viewing all the available information. And then defining within this range a prudent figure that more or less reflects the central thrust of the statistics.

In doing this, it must be clear at the outset that we cannot expect to determine the actual number of deaths. Even were all Chinese archives opened and every record of those murdered by the police, secret police, or the military available, we might then have a figure closer to the truth, although still probably off by millions. This is because many people undoubtedly were killed or died without a record being made, especially in time of rebellion, civil or international war, or mass movements like collectivization. Moreover, the numbers killed in some instances may have been highly exaggerated by the sources, especially when they are of Nationalist estimating communist democide, or vice versa.

In regard to difficulty of arriving at a definitive total, the Jewish holocaust is instructive. Although the Germans are noted for their rule following and record keeping, and the relevant German archives were available after the war, and participants could be systematically interviewed, there is still considerable disagreement among experts over the number of Jews murdered. Estimates vary from 4,200,000 to 6,000,000 dead, or by near 43 percent.

The job here can only be to find a most likely and prudent total of Chinese democide, where most likely means what a reasonable person would agree to on the basis of the evidence and analysis. Several statistical principles will be applied in doing this.

The variation principle: dependence on only a few or one-sided estimates will obviously bias the results. While one cannot include every estimate appearing in every book or article by whomever, estimates should be selected that are at the extremes, as well as being reasonably authoritative and credible; such as from Chinese sources, former Chinese officials, experts on China, refugees, and Western journalists formerly stationed there.

The comprehensiveness principle: as many estimates as possible should be included, even if redundant against other estimates, as long as they are independent in their sources. For example, if two estimates are based on the same source, only one should be included.

The disaggregation principle: this a basic principle of estimation. Overall estimates should be subdivided into the smallest sub-estimates available. For example, one should not simply rely on an estimate of those killed by the PRC, but should subdivide this into the various campaigns, movements, deportations, camp deaths, massacres, and the like for which the communists are responsible. The sum of these sub-estimates tends to cancel out errors of exaggeration and underestimation and to achieve a more reliable overall estimate than just an estimate of the PRC's toll by itself. Of course, such overall estimates should also be included and where great disparities between the norm of these and the sum of the sub-estimates occurs, they should be reconciled (consolidated). This disaggregation and reconciliation is done here by subdividing estimates into seven pre-PRC and five PRC historical periods, and then within each period by further subdividing into killing groups, events, institutions, processes, and years.

The error range principle: Overall low and high sub-estimates should be made for each case of democide, as well as a most probable estimate. The idea is that not all the low (high) estimates of all the different cases of democide can be correct, that some must be underestimating (overestimating) the true number, and that when all the lows (highs) are summed into a total for a period, the resulting low (high) is probably under (over) the true figure. Therefore, these total lows and highs will provide a probable range of error for the central estimate.The consolidation principle: If two different sets of low, mid, and high estimates are to be reconciled into one set, then the lowest and highest of the two sets should be taken, and the average of the two mid-estimates should be calculated (unless there are qualitative or contextual reasons to select one or the other). For example, based on available estimates of the communist labor camp population, calculate the number of dead in these camps for a given year as a low of 1,000,000, a mid-estimate of 2,000,000, and a high of 4,600,000. Let also the dead for this year be estimated from a variety of sources as 850,000, 2,500,000, and 3,500,000. These two sets of figures follow, with their resolution (consolidation) below them.

		 1,000,000  2,000,000  4,600,000
		   850,000  2,500,000  3,500,000
		   850,000  2,250,000  4,600,000 

As can be seen, the low is taken from that set with the lowest low; the high is taken from that set with the highest high, and the mid-estimate is the average.

If there are more than two sets, and unless there are qualitative or contextual factors to the contrary (such as a lowest low by over 50 percent for deaths during the land reform period coming from a communist official), they should be resolved into one set of estimates by also taking the lowest low and highest high. The mid-estimate may or may not be based on the average depending on what is being estimated and the nature of the different estimates. For example, there may be a large number of estimates near the high, which would cause the average also to be near the high, but this may seem undesirable if the historical context points to an estimate closer to the low. Moreover, in cases of ambiguity or the lack of much historical material on an event, to be conservative a mid-estimate closer to the low is more desirable than one closer to the high.

This principle assures me that final, prudential total for Chinese killing will be bracketed by the lowest reasonable lows and highs. And since these are the sum of lows or highs throughout, they in effect give the lower and upper bounds on the prudential estimate.

The disclosure principle: all estimates, their sources and relevant comments and qualifications should be disclosed, as well as all calculations, formulas, and what ever else is done to reach the final totals. This is to enable an informed and detailed evaluation of the final figures by experts, and to allow them to make additions and adjustments from their own perspective.


A number of procedures were used to translate estimates into tables to which calculations and consolidation could be applied, and in carrying out calculations on these estimates.

1. Translation of qualitative figures to quantitative: Estimates of those killed are not always numerically precise. The following protocols have been used to translate from such qualitative assertions to usable numbers.
1.1. If a source says that the number killed is "close to", or "just over", or "about", or "nearly X", then X is made the estimate. If this is not also said to be a low or high, or conservative, or otherwise similarly qualified, it is treated as a mid-estimate.
1.2. If a source says that X is "conservative" or "prudent", or qualifies X by "at least" or "more than", or otherwise indicates X is a low figure, then X is treated as a low. To say "no more than", "less than", "at most", or "approaches X", implies that X is a high. "As many as" is treated as suggesting a mid-estimate.
1.3. Qualitative approximations like "several thousand", or "thousands" is interpreted as a low of 2,000; a "couple of thousand" is interpreted as a mid-estimate of 2,000.

2. Filling in the range: In some cases when results are consolidated or there is only a single figure for a massacre or killing available, the range of killed may have to be approximated. An analysis of available estimates suggests that the following procedure is very conservative.

2.1. If a low and mid-value is missing, but there is high, then the high can be halved to estimate the low. If there is a low, but no mid-value and high, then the low can be doubled to get the high.
2.2. If a mid-value is available but the low is missing, then the low can be estimated as 2/3rds the mid-value; if a high is missing, it can be estimated as 1/3rd higher then the mid-value. If the contextual information suggests an average is more desirable, that is calculated for the mid-value.


Battle dead:
Those combatants killed in a violent conflict, such as a war or revolution. War-dead comprise battle-dead and those civilians killed in military action without themselves being the direct target of military action (as in civilians being caught in an house-to-house urban battle).


Any actions by government:
(1) designed to kill or cause the death of people
(1.1) because of their religion, race, language, ethnicity, national origin, class, politics, speech, actions construed as opposing the government or wrecking social policy, or by virtue of their relationship to such people;
(1.2) in order to fulfill a quota or requisition system; (1.3) in furtherance of a system of forced labor or enslavement;
(1.4) by massacre;
(1.5) through imposition of lethal living conditions of life;
(1.6) by directly targeting noncombatants during a war or violent conflict.

(2) that cause death by virtue of an intentionally or knowingly reckless and depraved disregard for life, as in

(2.1) deadly prison, or concentration, forced labor, prisoner of war, or recruit camp conditions;
(2.2) killing medical or scientific experiments on humans;
(2.3) torture or beatings;
(2.4) encouraged or condoned murder, or rape, looting, and pillage during which people are killed;
(2.5) a famine or epidemics during which government authorities withhold aid or knowingly act in a way to make it more deadly;
(2.6) forced deportations and expulsions in which many die.

Qualifications and clarifications:

(a) "government" includes de facto governance, as by the Communist Party of the People's Republic of China; or by a rebel or warlord army over a region and population it has conquered, as by the brief rule of Moslem Turks (East Turkistan Republic) over part of Sinkiang Province (1944-1946).

(b) "actions by governments" comprise official or authoritative actions by government officials, including the police, military, or secret service; or such non-governmental actions (e.g., by brigands, press-gangs, or secret societies) receiving government approval, aid, or acceptance.

(c) clause 1.1 includes, for example, directly targeting noncombatants during a war or violent conflict out of hatred or revenge, to depopulate an enemy region, or to terrorize or force the population into urging surrender; this would involve, among other actions, indiscriminate urban bombing or shelling, or blockades that cause mass starvation.

(d) "relationship to such people" (clause 1.1) includes their relatives, colleagues, co-workers, teachers, or students.

(e) "massacre" (clause 1.4) includes the mass killing of prisoners of war or of rebellions.

(f) "quota" system (clause 1.3) includes randomly selecting people for execution in order to meet a quota; or arresting people according to a quota, some of whom are then executed.

(g) "requisition" system (clause 1.3) includes taking from peasants or farmers all their food and produce, leaving them to starve to death.

(h) excluded from the definition is or are:

(h.1) execution for what are internationally considered capital crimes, such as murder, rape, spying, treason, and the like, so long as evidence does not exist that such allegations were invented by the government in order to execute the accused;
(h.2) actions taken against civilians during violent mob action or a riot (e.g., killing people with guns in their hands is not democide);
(h.3) the death of noncombatants killed during attacks on military targets so long as the primary target is military (e.g., during bombing enemy logistics)

Genocide :

Any actions by government designed to kill or cause the death of people because of their religion, race, language, ethnicity, national origin, or class (bourgeoise, landlords, capitalists, rich). Genocide is defined as a type of democide, and is included within clause (1.1) of its definition.


Democide, as defined above. This is an international crime against humanity.
Any actions by government designed to kill or cause the death of people because of their politics, speech, or actions construed as opposing the government or wrecking social policy. Politicide is defined as a type of democide, and is included within clause (1.1) of its definition.


* From the pre-publisher edited manuscript of "Appendix A" in R.J. Rummel's Lethal Politics and the "Methodological Appendix" in his China's Bloody Century.

You are the visitor since 11/28/02

Go to top of document