STATISTICS OF DEMOCIDE
Estimates, Calculations, And Sources*
By R.J. Rummel
In proportion to its population, Cambodia underwent a human catastrophe unparalleled in this century. Out of a 1970 population of probably near 7,100,0001 Cambodia probably lost slightly less than 4,000,000 people to war, rebellion, man-made famine, genocide, politicide, and mass murder. The vast majority, almost 3,300,000 men, women, and children (including 35,000 foreigners), were murdered within the years 1970 to 1980 by successive governments and guerrilla groups. Most of these, a likely near 2,400,000, were murdered by the communist Khmer Rouge.
The Khmer Rouge were fanatical communists who wanted to establish the most advanced and purist form of communism in the world. With military victory over the Lon Nol government in 1976 and absolute power thus in their hands, they hastily proceeded to construct their utopia. No actual or potential opponent was allowed to stand in their way; no violation of their draconian rules could go unpunished; no independent thoughts or groups could be allowed. No independent movement or property or enterprise was permitted. All Cambodians were as bricks in the hands of these supreme social engineers and human lives counted for little.
In Table 4.1A I present democide and war estimates from the sources and my calculations on these. Determining the number killed for 1970 to 1987 is especially difficult for Cambodia. Not only are central estimates often unavailable, but the parties themselves have made estimates for propaganda purposes that might well be too high or low. Moreover, scholars themselves have had strong points of view about the Lon Nol, Khmer Rouge, or Samrin regimes that could influence their estimates. Fortunately, for this one nation the population is small enough and the war-dead and democide large enough--over 50 percent of the 1970 population--that demographic analysis can discipline our assessment of the estimates and calculations on them.
For this reason I did a variety of regression analyses on available population estimates in order to calculate a range of population deficits. At the end of the Table 4.1B (lines 604 to 771) I list a sample of 167 population estimates collected for this purpose. First, I eliminated those estimates for the same period for which another source used the same reference.2 I also eliminated all lows or highs standing by themselves.3 Where there was both a low and high (as on line 712), I averaged them to get a mid-population figure and then erased the low and high. Next, I discarded all extreme high or low mid-estimates, such as that of 4,000,000 people in January, 1979 (from the Samrin regime--line 711).4 Finally, in the process of analysis I found the estimate for 1950 (line 605) created a counter-intuitive J-curve in population for this early period and had to be eliminated.
For the remaining estimates I assumed that any one given for a year without a month being indicated is for mid-year (July 1). Then for each population estimate I converted its year and month to year + (month/12). Thus July 1970 equals 1970 + (7/12) equals 1970.58. Since I was going to calculate polynomial regressions, I further had to reduce the size of the variable "year" (very large X-values will cause too near singular matrices for higher degree equations), such as "1970". I thus subtracted 1950 from each year, and in the above case got "20". I then carried out several polynomial regressions of the remaining population estimates on the transformed year and found that the extreme variation of estimates for the 1970s and early 1980s made it difficult to determine a reasonable population curve. This led me to prepare two alternative sets of transformed estimates.
Both are made up of a low, mid-value, and high for each year. For the first set, called the "average set", I made the low and high the lowest and highest estimates for the year and the mid-value the average of all estimates for that year. For the second set, called the "rolling average set", I calculated from the average set the low, mid-estimate, and the high for each year as the average of that and the previous two-years. Obviously both sets considerably lessen the year to year variability, the second set much more than the first. My aim here is to best bring out the trend curve of population. The cost is that the curves will be shallower and less steep than otherwise; and the final population deficits based on them will be more conservative in their range and size.
I applied polynomial regression to each low, high, and mid-estimates of both sets5 and compared them. I did this for the overall period 1953-1990, and then for each of the regimes.6 I then selected the best substantive curve7 and calculated the best fits. These are given in Table 4.2. The footnotes to the Table indicate which are calculated based upon the rolling averages and from what degree equation. Figure 4.1 plots the overall low, mid, and high polynomial regression curves for the overall period, and the untransformed population estimates.
To calculate a Cambodian population deficit, however, there must be some prediction about what the population would have been had not war and democide occurred. This I did by first determining the best polynomial regressions for the years 1953 to 1970 and the lows, highs, and mid-estimates of both sets mentioned above. From these I then calculated population predictions for the years 1971 to 1988. The results are given as the last three columns in Table 4.2. Figure 4.2 shows the actual mid-population fits for the three regimes and the mid-population prediction, and illustrates what is meant by deficit.
Another way of getting at population predictions is from the literature. Table 4.1B lists a number of population predictions (lines 775 to 791). I did a polynomial regression on those fourteen projections from 1962 and 1970 (lines 775 to 788). Since these projections were all to the years 1975 and after, I added the mid-population fit for 1970 (from row series #18, Table 4.2) to the sample to anchor it to the beginning of the killing years. The best fit to these projections was a straight line, the values for which are given in the "Forecast" column of Table 4.2. The curve falls between the low and highs of the predictions previously determined from the 1953 -1970 estimates (given in the last three columns of Table 4.2). Consequently, I will use only these predictions to determine the deficits.
Turning to the calculation of these deficits in Table 4.1B (lines 798 to 828), the first is for the Lon Nol regime. The low, mid-value, and high for the population fits at the end of the Lon Nol Regime (line 799) were subtracted from the predictions for that time (line 800) to get a gross deficit (line 801). Since this deficit may well include those that the regime expelled from Cambodia, and thus should not be counted as dead, I subtracted an estimate of this number (line 802) determined elsewhere (line 516) to get a net deficit (line 803).
My calculations of the deficit for the Khmer Rouge and Samrin regimes follow a similar pattern. However, to calculate this deficit also requires that the prediction curves be lowered such that they begin at the same level that the population was at when the regimes came to power. To do this for the Khmer Rouge regime, for example, the population fit for the month they came into power (line 806) is subtracted from the predictions for that month (line 808) to get a prediction difference (line 810). Then when I calculate the deficit as for the Lon Nol regime, I also subtract this difference to adjust the deficit downward (line 811). Of course, the resulting gross deficit may also include those who fled the country during the regime, and thus these refugees should be subtracted as well (line 812).
After following a similar procedure for the Samrin regime, I calculate the overall net population deficit (line 828). This is a straight sum of the deficits for the three regimes. This is permissible, since for the Khmer rouge and Samrin regimes the deficit has been adjusted to take account of the population level at the beginning of each regime.8
With these deficits to discipline our democide calculations, I can now work between the various consolidations, calculations, and totals of the dead or killed for each regime and the net population deficits. The first of our four periods to consider is that of Sihanouk, particularly the late 1960s (lines 2a to 16a in Table 4.1A). However only two estimates are given in the sources for democide (lines 6a to 7a), and I had to determine my own estimates for the Samlaut rebellion and the Khmer Rouge democide from events and their context during these years (lines 3a and 13a). These all give me a minimum democide of 13,000 killed.
Fortunately, many more estimates are available for the following Lon Nol period (lines 2 to 124 in Table 4.1A). Those for the war and rebellion dead are presented and consolidated first (lines 3 to 74). Note that for the consolidation of combatant dead (line 7) and the totaling of battle-dead (line 15) the highs (500,000 and 800,000, respectively) that I would have ordinarily selected have had to be reduced. This is because my acceptance of them and that of other high estimates would have led to an overall high for the Lon Nol regime that would be many millions in excess of the net population deficit; and when the other regimes are included, an overall high that would nearly have wiped out the total initial population.
While there are estimates of those killed in the American bombing campaign in Cambodia, there are none of how many of these were democidal. That is, what proportion of the deaths were due to intentional attacks on civilian targets or the result of an utterly reckless disregard of civilian casualties. Many of the sources bear on this, in particular Shawcross's revised Sideshow and the critiques of his arguments included therein.9 Taking these materials into account and the reports of survivors that are scattered throughout the literature, I believe that from 10 to 25 percent of the deaths, perhaps most likely 15 percent, were probably democidal (line 37).
I could find no overall or even partial estimates of the Khmer Rouge democide during the Lon Nol period. That it was high was clearly suggested in the sources, particularly by Becker.10 The Khmer Rouge imposed evacuations on the townspeople and a regime on the peasants under their control similar to that when they came into power--but far less deadly, apparently. Accordingly, I calculated the toll for thirty-six months and the population they controlled (roughly half) as one-fourth as lethal during their subsequent forty-five month government (line 107). The result, from 64,000 to 304,000 dead, is reasonable and consistent with the other figures for this period.
All told, 745,000 Cambodians were killed or died unnatural deaths during these years (line 122). This and the associated low and high are compared to the net population deficit (line 123). As can be seen, my calculated low is much lower, the mid-value is under, and the high slightly higher than the net deficit. Understanding that we want to keep the low as conservative as the estimates allow, the deficit gives us no reason to change this conclusion (line 124).
In their diversity and divergence, the estimates for the number killed or died during the Khmer Rouge period (lines 127 to 345) are particularly difficult to assess. This is especially true in light of the partisan nature of so many of them. To best handle this, I divided and subdivided the estimates into the most discrete conceptual or event defined groups as possible, and then tried to use the various sums of these diverse groups to check on the overall totals.
Relatively few of the dead are due to war (line 135) or rebellion (138); virtually all the deaths during these years are democidal. These were caused by the evacuation of the cities and towns, or displacement from one region to another (lines 141 to 168), pure genocide of minorities (lines 171 to 209), Khmer Rouge caused famine and disease (lines 219 to 223--virtually nothing was done to alleviate the conditions causing this or to provide the massive aid to save those dying), or executions (line 225 to 252). Of course these sources of death overlap, and assuming a very conservative 90 percent overlap in some cases, they add to up to 1,815,000 dead (line 352b).
There are also estimates of the overall dead for part of the years the Khmer rouge were in power. These and other partial estimates are shown together in the Table (lines 256 to 271). Of these I took those overall estimates and proportionally (months of estimate to months in power) projected them to cover the whole period. The validity of this depends on whether one assumes that conditions were equally deadly throughout the whole period. Once the initial urban evacuations were over conditions briefly improved for the former urbanites but then gradually got worse each year for all. If this be so, then the projected estimates (line 272) are conservative.
Many estimates are available for the overall democide (lines 275 to 329) for the whole period. I consolidated these (line 330) and pegged the mid-value at 2,000,000. This is higher than 1,828,000, the average of all the estimates between the low and the highest estimate of 4,000,000 (this particular high was not selected in order to keep the final high within reasonable range of the population deficit for this period). Whether this mid-value is excessive will be seen from the totals soon to be considered.
The final set of estimates is of population deficits and losses during this period.11 In the Table (lines 335 to 345) I ordered these by size and consolidated them (line 346). Vickery gives a lowest low of 400,000 population decline (line 335), but in a later work he ignores this (line 336) and we will do so here.
Now all this can be put together and compared (line 350 to 357). First I show the four partial results (lines 350 to 352a) and their consolidation (line 352b). Then from this consolidation and four other ways of calculating the Khmer Rouge regime's toll (lines 353a to 357), I end up with a proposed range of 600,000 to 3,000,000 domestic killed, and a mid-value of 2,000,000, which is close to the average of the various ways of calculating the democide. To these I add the foreign democide (line 358a) to get the overall total range (line 359). It is tempting also to round the mid-total democide to 2,000,000. But to do so would create arithmetical problems down the line--things would not add up. To this total I can now add war and rebellion dead (lines 361 to 362) to get a total mortality (line 364) for the Khmer Rouge.
In contrast to the wealth of estimates for the numbers who died under the Khmer Rouge, there are relatively few for the Samrin regime. Those I could find I organized and consolidated in the Table for war-dead (lines 367 to 381), famine-dead (lines 382 to 383), and democide by this regime (lines 386 to 391), Khmer Rouge guerrillas (lines 394 to 395), and Vietnam (392). For the latter, because of the clear dominance of Vietnam over the regime and their carrying the major burden of the guerrilla war, I estimate that they are responsible for two-thirds of the democide.
Gathering the different sources of death (lines 399 to 405), I get a mid-sum of 1,160,000 Cambodian dead (line 405). This is reasonably in line with the net deficit (line 404), and because of the particular nature of the population curve for this regime (early decline followed initially by rapid growth--I will discuss this further below) the mid-total estimate is probably more conservative than it appears by comparison to the deficit. The deficit low of zero is due to a possible population growth that exceeds the predicted, and thus is negative. I replaced the negative with a zero.
Finally we can turn to the totals for all Cambodian dead beginning with the Sihanouk period. I organized these in the Table (lines 407a to 417), and summed them (lines 417a and 418). The overall domestic democide is 3,151,000. This can be compared to the one estimate in the literature for those killed from 1970 to 1986 (line 419). Then I added the war and rebellion dead to the democide total to get an overall mortality range (line 420).
There are various demographic based estimates of the population loss or deficit over most of these years. These I collected together (lines 437 to 441) and beneath them show the deficit from my polynomial regression (line 442). My mid-value tends to be conservative by comparison (the average of all the loss and deficit estimates on lines 437 to 441 is 2,800,000).
Finally, I can compare the grand total of domestic dead (line 445) to the independently calculated overall net deficit (line 446) for the years 1970 to 1987. The two-mid-values and highs are surprisingly close, and the low of the total is lower then the deficit, which it should be. In other words, the calculated deficit gives us no reason to question the final total.
In sum, then, 3,944,000 Cambodians were killed in war, or from famine, disease, execution, torture, massacres, and the like (line 448); 3,151,000 of them were murdered (line 417a). Although I consider this overall total prudent, it is much beyond the consensus of scholars, journalists, and officials. However, as mentioned it is virtually the same as the mid-population net deficit calculated from the diverse population estimates. And this I consider conservative because of the way the population estimates were transformed and calculated.
At the start of the mass killing in 1970 the highest population estimate was 8,000,000 (lines 656 and 657 in Table 4.1B). One estimate of average population growth in the previous five years was 2.8 percent.12 Let us assume the high was a not unreasonable 3 percent. Then the population in 1987 would be 13,619,000, or 5,619,000 higher. How then do I get the net deficit high shown (line 446 of Table 4.1A); and how can I accept the domestic death toll high (line 445) which is practically the same? First, based on the trend in population estimates for each year 1953 to 1970, the predicted population high for 1987 was 14,615,000 (row series #150, Table 4.2), equivalent to a growth rate of slightly over 3.4 percent from a population high fit of 7,406,000 in 1970. The predicted population adds 7,244,000 people. This net population increase is much greater than the net deficit and total deaths high shown in the Table. Moreover, note that the actual population fits for the population in 1987 are not much different from the population fits in 1970. It is as though the population was almost flat for eighteen years.
There is, however, another way of looking at this. In the case of Cambodia, calculating the overall population deficit as the difference between the 1987 fitted and predicted population grossly underestimates the actual population loss. This is because the population declined shapely during the Khmer Rouge period and continued its decline during the first months of the Samrin regime and then thereafter grew rapidly for several years, possibly even greater than 5 percent (women made up almost two-thirds of the surviving population--in 1982 Samrin's Minister of Health claimed a growth rate of about 7 percent for 198113). This would be far in excess of normal growth rates estimated for Cambodia prior to 1970. And it would mean that a portion of the population deficit created during previous regimes would be erased. For this reason, I have determined the net deficit as the sum of the deficits for this and previous regimes, rather simply the difference between fitted and predicted populations for 1987. The result is that while the deficit and total domestic death highs given in the Table (lines 445 and 446 of Table 4.1A) are, indeed, high, they are not inconceivable.
So much for the deficit and total dead. Next in the Table I classify various estimates of the refugees and expellees from Cambodia (lines 450 to 534 of Table 4.1B). The consolidated sub-totals, as for those under the Khmer Rouge (line 519) have been used to determine the net population deficit (e.g., lines 811 to 813).
Then there are the population estimates. I have divided out those for various subgroups, as of the urbanites (lines 538 to 543) and Chinese (lines 558 to 563) from those for the population as a whole (lines 605 to 771). The later, as mentioned above, were used to generate population polynomial fits and predictions. Those population projections given in the Table (lines 775 to 791) were similarly used. The last of the population estimates comprises the gross and net population deficits I have calculated and discussed above (lines 798 to 828).
At the end of the Table (lines 831 to 845) I list the democide rates for the three regimes, as well as for the Khmer Rouge when they fought the Lon Nol and Samrin regimes as guerrillas (line 840 to 841). The population figures used in these calculations are the population fits given in Table 4.2.
* From the pre-publisher edited manuscript of Chapter 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. This is a most probable mid-estimate from Table 9A.2, row series #20 mid-population.
2. Removed were line 655 (duplicated reference of line 654); line 680 (duplicates line 679); line 738 (duplicates line 740-figures differ due to line 680 round off).
3.These are on lines 663, 667, 696, 708, and 718.
4. Additional estimates omitted were lines 694, 702, 703, 725, 746, and 769.
5. I could not apply the usual growth curve to these estimates, since population growth was interrupted in the early 1970s and the population took a dive from 1975 to 1980. This thus produced a curve with several inflection points, the fit to which can be determined by polynomial regression of the form Y = a + bX + cX2 + dX3 + eX4. . . . The degree of the equation (for example, if x is taken to the fourth power the equation would be of degree 4) depends on the number of inflection points that can be assumed in the data. Here a substantive knowledge of the period being thus fitted is essential.
6. In order to best bring out the curves involved, two additional years had to be included at each end. Thus, estimates for the Khmer Rouge were analyzed for 1973 to 1980.
7. "Best" fit is not meant in significance test or R-squared terms. A best fit here is that which best accords with the population estimates and the generally agreed loss of life during each of the periods. For example, a curve that shows little loss of population during the Lon Nol years or continues unchanged upward through the first deadly year or two of the Khmer Rouge period is unsatisfactory.
8. The deficits also include the unborn. But this has to be balanced against a deficit's assumption about natural deaths. Many who would have died naturally, however, were killed beforehand or, especially among the infirm, sick, and aged (just the group contributing most to the normal death rate), died prematurely due to conditions for which these regimes were responsible. Given the widely diverse estimates of birth and natural death rates during these regimes, trying to balance them would require the same kind of regressions as done on the population estimates, and introduce a second level of complex assumptions. Here I settle for the simpler assumption that the unborn and decrease in natural deaths (due to a regime causing people to die prematurely) balance out.
9. Shawcross (1986, pp. 422ff).
10. See Becker (1986, pp. 149, 151, 164-68).
11. A "population loss" is the difference between the current population and what the population was on some previous date, assuming it was higher then. A population deficit is the difference between the current population and what the current population is predicted to be.
12. Kampuchea in the Seventies: Report of a Finnish Inquiry Commission (1982, p. 32).
13. Gough (1986, p. 66). I am aware of, but not convinced by Vickery's (1988, p. 71) criticism of a rate over 5 percent .