Death Tolls for the Man-made Megadeaths of the 20th Century


Site Index

Alphabetical Index

My collection of atrocity statistics seems to be the most popular (?) part of my website, so I get a ton of questions on the subject. Here are some answers:

Q: Why?

A: I have a simple theory of history: Things that happen to a lot of people are usually more important than than things that happen to only a few people.

For example, you hear news and debates every day about the smallest events in Israel but rarely about events in Indonesia, yet by every objective standard, events in Indonesia are a lot more important. Indonesia has more people, more resources, more struggles and more exports. Still, I'm sure that the number of Anglophones who can name three leaders of Israel far outweighs the number who can name three leaders of Indonesia. In fact, any attempt to defend the attention we pay to Israel seems to be either 1) circular reasoning (Israel is important because we consider it important.), or 2) incredibly self-centered (Israel is important because it effects me personally.)

I was rather naively hoping to correct this in my atlas. For example, my original plan was to map the wars of the 20th Century in descending order of size. That meant that I would be drawing many maps of the World Wars and a few maps of Vietnam and Korea before I would even think of starting on Granada and Panama. Unfortunately, when I tried to find a list of wars and mass killings arranged by size, there wasn't one, so I had to compile a list from scratch, one atrocity at a time.

After many months of this unwholesome research, I discovered that there is an actual, legitimate field of scholarship (euphemistically called "peace research") whose practitioners have already compiled the lists I was seeking, so all my work was in vain.

Then, as I tried to reconcile all the discrepancies in my research, I discovered that most atrocity collectors have a political agenda to push. Each list seemed to be slanted in favor of a preconceived result, so I had to plunge in again and continue to create my own list from scratch.

Now that you've come this far in the explanation, forget everything I've just said. Theories of historiography are just fine, but as a practical matter, I just haven't had the time to dig out all the obscure details of Indonesian history, while the history of Israel is just sitting there, readily available.

Even so, once I published the results of my "peace" research, I began to get questions and suggestions that pushed me to fill gaps and analyze my data, which meant that I could never put it away for very long.

Q: Do you have a list or map of all currently ongoing wars?

A: I'm mostly interested in history, so I don't fanatically follow current events. If you want up-to-the-minute information, try one of these sites:

Q: How about a list of current dictators?

A: Try Freedom House or Parade. If those exact pages are gone, search those servers for your answer. You'll find it there somewhere.

Q: Doesn't [random war or atrocity] belong on your list?

A: In a word: NO. Most likely, you just aren't seeing it (Look again) or it wasn't very bad (Check your sources).

One of the reasons these pages have grown wildly out of control is that I would start by posting the Top Ten Something, and then someone would write in, "What about This...?" Well, This was actually way down around number 23, so I'd e-mail back an explanation. Then someone else would ask about This. Finally, in order to stop people from asking me about This yet again, I'd increase my list to the Top Twenty-five. Then I'd get, "Why isn't This Other Thing on your Top Twenty-five?" and I'd have to explain that This Other Thing is actually number 34.

Sure there are gaps in my coverage, but please, please, if you want to disagree with my numbers, please, provide an alternate estimate and a published source. "IIRC" is not a source.

Q: What about [random famine]?

Q: What do you mean by "atrocity"? What does it take to get on your list?

A: By atrocity, I mean anything that people just go on and on about.

Sometimes I get the feeling that some researchers start with the conclusion ("Communists committed the worst genocide in history") and then fiddle with the definition of genocide so that it fits their conclusion. They'll include as genocide all the things Communists excel at -- such as creating famines and working political prisoners to death -- while simultaneously excluding things that the Communists are mere apprentices at -- like unprovoked blitzkrieg and plantation slavery. Or they'll start with the conclusion that the USA is the worst terrorist nation of the modern era. To prove their point, they'll define the modern era as beginning, oh, say 1945, once the Germans and Japanese are out of the running. And then they'll define terrorism to include things Americans do well -- like aerial bombing and financing foreign tyrants -- while excluding things Americans don't do, like female infanticide, ethnic cleansing and labor camps.

Those who want to prove that war is the century's biggest killer will count the Armenian genocide under World War I and the Holocaust under World War II, while those who want to prove that war is not the biggest killer will separate the Armenian and Jewish genocides from their respective World Wars. The war-is-bad party will count China's Cultural Revolution as a civil war, while the democide-is-even-worse party will count the Cultural Revolution as an internal, peacetime oppression. Meanwhile, collectors of both these categories tend to ignore, for example, the Congo Free State where the means of death weren't quite war and weren't quite government.

Rather than get bogged down in quibbling over definitions, I include everything. See, for example, my list of the 20th Century's worst massacres.

Q: The number of dead is unknowable.

A: No more so than any other social statistic. Every ten years, we go through the ritual of a nationwide census, followed by the ritual of explaining why the census is wrong. That doesn't stop us from trying.

No one expects to tally every single death down to the last individual, but we can at least decide whether we're talking about thousands, hundreds, or millions. And if it's thousands, is it tens of thousands or hundreds of thousands? And is it the upper tens of thousands, or the lower tens of thousands?

Once we get the number narrowed down into the right neighborhood, then we can start assessing it's meaning.

Q: How can we possibly know how many people died in an atrocity?

A: The best answer would vary on a case-by-case basis, but the short answer is money.

Even if a general is reluctant to tell the newspapers how many men he lost in a bungled offensive, he still has to tell the accountants to drop 4,000 men from the payroll. Even if a dictator doesn't really care how many civilians are killed in the crossfire, his finance minister will still note the disappearance of 100,000 taxpayers. Even if no one cares how many slaves are kidnapped from the jungle, a customs official at the harbor will be collecting duties on each cargo.

Head counts (and by extension, body counts) are not just an academic exercise; they have been an important part of government financing for centuries. It's no coincidence that the US Census Dept. is part of the Treasury.

Q: How reliable are ancient and medieval atrocity statistics?

Q: Do numbers even matter?

A: Maybe they do; maybe they don't. There are certainly many valid reasons to decide that they really don't.

On the other hand, I frequently see statements that could stand or fall on the basis of numbers:

I've noticed that people sometimes avoid using numbers because that would create a tangible hypothesis that can be scientifically tested. They know it's a trap.

I once read an article in, I forget, the Weekly Standard or something that dealt with the question of civilian casualties of American bombing in Afghanistan. The main thrust of the article was that the commonly quoted estimates of Professor Marc Herold were wrong, and we should all write letters to the University of Wherever complaining that they were subsidizing this man's treasonous research.

Do you think this article offered a counter-estimate of civilian casualties that we could use to reasonably drive Herold's numbers out of the marketplace of ideas? No, of course not. If they gave their own number, then they'd be cornered. Someone could come along behind them and prove that they don't know what they're talking about. Even worse, their enemies could use these numbers for their own purposes, and declare, "Why, even the Weekly Standard or Whoever admits that this many people were killed."

Of course, if there's a number vacuum, then someone will step in to fill it. That's how Herold's numbers became so common -- not that he was right, only that he was first. More to the point, that's why I link to Herold's website, but not to the article that refutes him. I'm only concerned with numbers, so no numbers, no cite. The lesson here is that once you've decided that numbers don't matter, you've abandoned the field to people who like numbers, and the only way to reclaim it is to fight numbers with numbers.

Q: Abbreviations: Five hundred thousand people were killed in the Chicago gang wars under Prohibition!? Really? That's incredible. Five hundred thousand!?

A: Huh? Oh, I see... I said "500 k", but k means killed, not thousand. In these pages, I never use K for thousand. K means killed; T means thousand. I'll admit, I'm inconsistent about whether I use M for million or missing, but generally, I'm hoping that all my abbreviations are apparent by context as a shortening of common English words.

Blame and Responsibility

There's one question I see quite often, but it's phrased in different ways that need entirely different answers:

Q: Is government responsible for more deaths in the 20th Century than any other cause? In Dunnigan's Dirty Little Secrets of the Twentieth Century he asks, "What do you think killed the most people this century? Epidemics, war or democide? If you took a chance and chose democide, you were right."

A: No. You'd be wrong. Rummel's estimate of democide, the highest estimate there is, puts the number killed at 170M. Almost twice that many, 300M, died of smallpox. Even the most anti-government statistics would probably agree that some 19 out of every 20 deaths of 20th Century were by natural causes.

Q: Is government responsible for most of the deaths by violence and oppression in the 20th Century?

A: Well, yeah. Of course. Organized thugs kill more efficiently than disorganized thugs.

Wars and oppressions are collective endeavors, and whenever humans work collectively, they work through government. Whether wars are fought using citizen militia, standing armies, hired mercenaries, tribal warriors, corporate security teams, street gangs, paramilitaries or feudal levies, there's always a ruler or governing body to determine who does what to whom. Call it a cabinet, junta, council of elders, general staff, board of directors, politburo or capo de tutti capi -- it's a government.

Unfortunately, blaming war and tyranny on government is like blaming house fires on oxygen. Strictly speaking, it's true, but no fire marshal would ever get a good performance review if that's his answer to everything. To understand house fires, you'll have to look at wiring, storage, smoking, materials and cooking, and not just insist, "It's oxygen, I tell you! Oxygen is evil! Why won't anyone listen to me?"

Most attempts to blame government for democides depend on a kind of circular reasoning. They define "government" as any organization capable of mass violence, and then are amazed when all mass violence is produced by government. If they were to instead define "government" as, say, any organization that delivers mail or builds roads, then at least they would be moving away from obvious tautologies.

Q: But isn't excess government responsible for more avoidable deaths in the 20th Century than any other cause?

A: Debatable. Consider, for example, those 300 million smallpox deaths. By 1900 smallpox was entirely preventable; vaccination was a common procedure, and most of the world had been incorporated into advanced industrial states and colonial empires, so a concentrated government effort might have eradicated the disease seventy years early and prevented those 300 million deaths. That single example suggests at least 300 million deaths by too little government versus 170 million by too much government. [n.1]

In fact, if we add up annual deaths from unclean water (12M/yr), air pollution (2.7M/yr), smoking (5M/yr) and traffic accidents (1.2M/yr) -- you know, all those things that statists want to regulate -- we could find an entire century's worth of democide in a single decade of government inaction. This doesn't mean that I want to ban smoking or confiscate cars; it only means that problems and solutions are more complicated than some people want to admit.

[n.1: Actually, my estimate is 83M democides, but for this example, we'll go with Rummel's more popular estimate of 170M.]

Q: Don't governments kill more of their own people than foreigners?

A: Yes, but it's so much easier that way. I mean, they're right there. You'd have to go elsewhere to kill foreigners.

Actually, before I concede without a fight, I probably should point out that what you and I consider "their own people" is not always the same as what various tyrants considers "their own people". Hitler didn't consider the Jews his own people; Stalin probably didn't consider the Ukrainians his own people; Leopold certainly didn't consider the Africans his own people, nor did Saddam consider the Kurds his people. This is why your various whiny minorities get so nervous about being treated as second-class citizens. It makes them fair game.

Q: Did democide kill four times as many people as wars in the 20th Century?

A: Maybe yes, maybe no. This statement is usually supported by Rummel's calculation that 169 million people were killed by democide, while only 34 million died in wars. This assertion can be disputed on two points -- the numbers and the definitions.

On the first point, estimates vary, and Rummel's democide total is generally higher than anyone else's. Even the Black Book of Communism is generally lower than Rummel.

On the second point, the estimate of 34 million war dead only includes soldiers killed in combat. Of the 55 million people who died in World War 2, only some 20 million were soldiers. Of the remaining 35 million civilians killed, most were deliberate, which means that Rummel tallies these under democide, not war. On the other hand, we could easily argue that these properly belong under war dead; after all, it's unlikely that the Allies would have bombed Dresden had there not been a war, nor would the Nazis have had access to Poland's 3 million Jews without a war. Thus, by merely reclassifying the civilian deaths of World War Two, we could boost the estimate of 34 million war dead by as much as 35 million, while decreasing the estimate of democide by a comparable number. That by itself reduces Rummel's ratio of 4 democides for every war death to more like 2 democides for every war death. Applying the same analysis to other wars narrows the gap even further.

Q: Is religion responsible for more more violent deaths than any other cause?

A: No, of course not -- unless you define religion so broadly as to be meaningless. Just take the four deadliest events of the 20th Century -- Two World Wars, Red China and the Soviet Union -- no religious motivation there, unless you consider every belief system to be a religion.

Q: So, what you're saying is that religion has never killed anyone.

A: Arrgh... You all-or-nothing people drive me crazy. There are many documented examples where members of one religion try to exterminate the members of another religion. Causation is always complex, but if the only difference between two warring groups is religion, then that certainly sounds like a religious conflict to me. Is it the number one cause of mass homicide in human history? No. Of the 22 worst episodes of mass killing, maybe four were primarily religious. Is that a lot? Well, it's more than the number of wars fought over soccer, or sex (The Trojan and Sabine Wars don't even make the list.), but less than the number fought over land, money, glory or prestige.

In my Index, I list 41 religious conflicts compared with 27 oppressions under "Communism", 24 under Colonialism, 2 under "Railroads" and 2 under "Scapegoats". Make of that what you will.

Q: How much oppression is caused by racism?

A: Well, if you define racism as targeting a group for oppression primarily because they look different (as opposed to speaking, worshipping or voting differently), then probably not as much as you'd think.

Because so much conflict within the US has been along racial lines, we (Americans) tend to be hypersensitive to it; however, if you look across the globe and across history, most conflicts and oppressions have been between peoples who look pretty much the same as one another. And even when an oppression is directed at a group which slightly differs in appearance from the oppressors, we often find groups that look as much or even more different surviving the same oppressors relatively unscathed.

(An odd and trivial example of this: I once visited a big old Southern graveyard, loaded up with tons of Confederate dead, and about twenty feet from the grave of Jefferson Davis was a large tombstone covered in Japanese, and presumably covering a Japanese as well. I'm not saying that this proves that the old South was a model of racial harmony; I'm just saying that we (humans) (also Southerners) don't just uncontrollably and instinctively lash out at anyone who looks different. Jim Crow Virginians apparently considered the Japanese to be white enough to bury them in white cemeteries. Racial animosity was focused instead very specifically at African-Americans.)

Weirdly, it sometimes seems that we (Americans) see racism even where it isn't. I've heard opponents label the US invasions of Afghanistan and Iraq as racist, even though the Afghans and Iraqis are white. I mean, aren't they? If you were to dress them from Sears, teach them the speak American with a Texas drawl and transport them back in time to segregated Alabama, they'd be able to use the white bathroom without even attracting a sideways scowl, right?

Q: How many wars have been fought over spelling?

A: Possibly the scariest thing in the whole world is that this is not a totally stupid question.

For example, there's a language known to linguists as Serbo-Croatian, which sounds the same when spoken, but becomes two variants when written -- Serbian, which uses the Cyrillic alphabet, and Croatian, which uses the Roman alphabet. These people have gone to war with one another several times.

There's another language that used to be called Hindustani, which also sounds the same when people speak it, but becomes two languages when written -- Urdu, which uses the Nastaliq alphabet, and Hindi, which uses the Devanagari alphabet. These people have also gone to war with one another several times.

I'm sure there are serious, tenured scholars who will say that these groups are at odds because of "cultural differences", but when I actually dissect what these "cultural differences" are, it worries me that maybe we've actually seen people killing each other over which alphabet to use.

Q: Is it fair to judge the past by the moral standards of today?

A: Sure. Why not?

Sometime the curses of posterity are the only punishments that a monster from history will ever get. Stalin died peacefully in his sleep, beloved and mourned by a nation. Are we going to let him get away with that?

Also, if the purpose of studying history is to avoid past mistakes, then we have to decide what those mistakes were. That means passing judgement. Did Augustus bring tyranny or stability to the Roman world? Do states' rights allow oppressive local governments to flourish, or prevent the rise of federal tyranny?

And if we want to promote heroic role models, we first have to find someone we admire. That too requires that we judge the past. I for one am not going to promote posthumous veneration for someone I'm glad we're rid of.

One other thing I've noticed is that "the moral standards of today" are not always a recent invention. To say that slaveholders of the 1800's can't be held morally accountable ignores the fact that huge segments of the population were already opposed to slavery at that time. If we argue that the massacre of civilians was an accepted aspect of medieval warfare ("Everybody did it"), then we slander the various saints and holy men that spoke against it, and the occasional general who showed mercy to his captives. After all, aren't you including them in the "everybody" who did it?

Q: But -- regardless of which standards you use -- why judge at all?

A: To quote Alan T. Nolan from Lee Considered: "[T]he historian must -- as a practitioner of a discipline regarded as one of the humanities -- take into account the human and social consequences of [historic events]."

Q: So who killed more -- Hitler or Stalin? How about gun control? Christianity? Communism? Imperialism? Islam? The Masons? Left-handers?

Q: OK, then who is responsible for all these deaths?

A: Beats the heck out of me. You know, I could probably score some good think-tank funding if I could only prove statistically that human cruelty is getting steadily worse, and it's all someone's fault. Unfortunately, it just looks like the numbers fluctuate randomly over time, and we fight wars and oppress the weak because that's what we're good at.

Q: Why should I believe anything you say? You're just some guy on Internet.

A: If that's your best counter-argument, then you're in trouble.

Q: The Turks did no such things! You must be in the pay of Armenian nationalists!

A: I wish. Those Armenian nationalists pay top dollar, plus they have a fantastic dental plan.

Methods and Sources

Q: What are some of the common mistakes, tricks and distractions you've encountered when counting bodies?

A: Excellent question. I'm glad I asked that. Here are some problems I've run into.

  1. Don't stop with the first answer you find. If you want to know how many people died in the Civil War, don't just look in the encyclopedia and believe what they tell you. Get a second opinion. Even if the first source is generally trustworthy, they might have a typographical error for all you know. Or you might have stumbled across the one topic where they have a blind spot.
  2. Go back and read your source again. Are you sure you understand what they're saying? For example, did the Vietnam War kill two million civilians total in the North and South, or two million civilians in the North and two million civilians in the South.
  3. Even though the exact number may be unknown and possibly unknowable, you can still be absolutely, inarguably wrong if you misquote or misinterpret your source. If, for example, you claim that 20,000 Catholics were executed by a dictator, when in fact, the source you cite specifically says "20,000 political prisoners, many of them Catholic, were executed", then you are wrong, simple as that. (Well, unless you explain yourself...)
  4. Apples and Oranges: Do you know exactly what is being counted in an estimate?
  5. It is common for debaters to ascribe all the deaths in a particular event to whatever trend, nation, policy, individual or practice they're trying to discredit. For example, the 600,000 people who died in the Cambodian Civil War (both government and Khmer Rouge, in combat, by bombing, by hunger and disease) are often described as 600,000 killed by American bombing. The 72,000 common criminals executed during the reign of Henry VIII sometimes become 72,000 Catholic martyrs.
  6. Definitions: When you start to blur the lines defining what words mean, then they end up meaning nothing at all. For example, genocide is usually defined as the attempted killing of every member of an ethnic group, but when you start calling the suppression of a native language "cultural genocide", or the eradication of the bourgeoisie "class genocide", you end up treating "genocide" as just another synonym for "repression". Is that what you want to do?
    1. Casualties are not the same as deaths. Casualties include wounded and prisoners as well. For example, notice the confusion concerning the Pentagon estimates of Communist casualties in Korea, the CNN tally for D-Day, and the wide discrepancies over the number of dead at Verdun, Passchendaele and the Somme. Someone, somewhere has seen the word "casualties" and translated it as "killed", or vice versa.
    2. Losses are usually the same as casualties, but sometimes they mean whatever category we were just talking about. ("5,000 Germans were killed, while the British lost only 2,000")
    3. Missing means they weren't there when the sergeant took roll after the battle and no one saw what happened to them. Maybe even the entire unit is unaccounted for. The missing might be prisoners, dead, wounded, AWOL or straggling. If a unit comes through the battle intact, there will be only a few missing, and those are probably dead or wounded somewhere. A shattered unit will have many missing who will later turn up as POWs, although a few might evade capture and return to their lines. In the long run, if soldiers stay missing very long, the odds are, they're dead. Whether to count missing as POWs or dead depends on case-by-case context.
    4. Victims are not the same as deaths. Victims of a dictator might include those who were merely imprisoned, exiled or beaten. For example, notice how some estimates for deaths at the hands of the Spanish Inquisition are close to the total number of (non-fatal) victims. Did someone interpret "victims" as "executions"?
    5. Killed vs. Died: "Killed in the war" usually implies directly, by violence, while "died in the war" ("deaths in the war") implies all causes, including disease, famine and accident. It's best not to use these words interchangeably.
  7. When you start combining the effects of problems 4, 5, and 6, you might end up counting the same deaths twice or more.
  8. Innocent victims: Irrelevant. Hundreds of thousands of German POWs were worked to death in Stalin's Gulag. Was this a war crime? Yes. Were they innocent? No, they were invaders. Calling victims innocent is a distraction. If a crime boss eliminates his rivals, it's still murder.
  9. Double standards
    1. Standard double standard: If our guys do it, it's an accident, but if their guys do it, it's a deliberate policy.
    2. Backward double standard: You know the type -- If our government does it, it's an evil policy, but if the enemy does it, that's beside the point.
    3. Valid backward double standard: The reason I get more worked up over Americans torturing POWs than I do over Ugandans torturing POWs, is that I'm responsible for what the USA does, but not for what Uganda does.
    4. Triple standard: If our guys do it to the enemy, it's OK; if the enemy does it to us, it's evil, but if someone who's not even part of this debate does it or has it done to them, we don't care. For example, when debating the morality of bombing cites, you always hear about the Americans bombing Germany and Japan, and the Germans bombing England, and you get the impression that that's all there was. I had been collecting statistics for six years before I learned that some 500,000 Soviet civilians died in German air raids. The upshot of these double and triple standards is that when we draw up lists of horrible happenings, we only include the ones we do or the ones that are done to us, which is a bit self-centered if you ask me.
  10. Newer isn't necessarily better: "The latest studies show..." Presumably, newer studies have access to more data than earlier studies, but that doesn't necessarily free them from biases or carelessness. Also, remember that no academic career ever gets a boost by publishing studies that show that someone else was right all along. No, you get more promotions and funding if your studies show something new.
  11. Official isn't necessarily more accurate: "The official death toll is put at ..." Maybe the officials are padding the numbers to get more money, or maybe they're underestimating to hide the full extent of their incompetence.
  12. Bias: While you should always question someone's motives for putting forth a particular estimate, that by itself doesn't count as a valid counter-argument. Almost everyone has a political motivation when they publicize a death toll, so merely pointing that out wins you no points.
  13. If all else fails, simply deny that anyone disagrees with you. From Stannard's American Holocaust: "[W]hile debate continues as to the actual population of the Americas prior to the arrival of Europeans at the end of the fifteenth century, few informed scholars any longer contend that it was not at least within the general range of 75 to 100,000,000 persons..." This statement is demonstrably false. There are many informed scholars who would contend this. Whether these scholars are correct in their contention is beside the point because Stannard is denying their existence, not their accuracy.

Q: What do all those modifiers like "at least", "as many as", "more than" actually mean?

A: They don't mean anything. Not mathematically. When somebody says "at least 10,000 were killed", they are really saying "10,000 is the higher number I can provide evidence for (but I was hoping it would be higher)". The phrase "possibly as many as 10,000" means pretty much the same thing.

In fact, I've come to regard "more than" and "no more than" to be synonymous. Technically speaking, "more than 10,000" could mean 15,000, but, come on, if they could prove 15,000, they'd say "15,000". As for "no more than 10,000", I suppose it could mean 5,000 or even 200, but the strongest line of proof points to 10,000, so they'll grudgingly admit to it.

Q: What do they mean when they call something a "conservative estimate"?

A: They're lying.

Q: Are raw numbers the best measure of an atrocity's impact? Wouldn't it be better to calculate death tolls as a percentage of the overall population per year, instead of just the absolute total?

A: Unfortunately, numbers like that are often slanted by how you chose your populations and times. They open up too many possibilities to manipulate the numbers.

Basically, at their epicenter, all major atrocities kill at pretty much the same rate: a lot, very quickly.

Q: How accurately do you summarize your sources?

The longer I've worked on this project, the more fanatical I've gotten about keeping the exact words used by my sources; however, with some of my earlier sources, I paraphrased and summarized, perhaps too much. In any case, I'd feel more comfortable if you would go back to my original sources before you take my word as gospel.

Q: Among your sources, you cite Chomsky [or the Cuban American National Foundation or the New York Times or the Washington Times or the Black Book of Communism or whoever], but that guy is so full of bull[oney] that now I can't trust any of your numbers.

A: Regardless, they're experts, and we can't ignore what they say. My definition of expert isn't "someone I agree with", nor is it "someone who admits right up front that Communism sucks", nor is it "someone who has never been on the payroll of the US Government". To me, an expert is someone who knows enough about the subject to write at length about it. In a field as controversial and as prone to hyperbole as atrocitology, just citing actual facts puts a person a cut above the usual commentator. It adds credibility if this "expert" has managed to convince a large publishing house, NGO or newspaper that he knows what he's talking about. Peer review would be an even nicer little bonus, but it's rarely available in such things. Obviously, not every expert is right, but if we gather enough of them, we may see a consensus emerge.

Q: So basically, you're putting the numbers to a vote... But you can't determine truth by majority rule.

A: No, but when two experts disagree, who decides the truth?

Q: I do -- by looking at the facts.

A: But then, you're setting yourself up as an expert as well, and not meaning to be rude or anything, but just who are you? You're not omniscient. At best, you're just another expert who is adding to the noise of all the different estimates.

Q: You're assuming the the numbers will arrange themselves neatly as a bell curve around a central number that yields the True Answer, but there's no support for that assumption.

A: The way I figure it, either some of the estimates are close to being correct, or else none are. If none are, then we might as well just give up right now.

On the other hand, if some of the estimates are correct, how do we decide which ones? If we take the highest estimate available, then we're assuming that everybody who has studied the problem (except for one guy) has been consistently, predictably wrong. Every single one of them (except that one guy) has fallen short. Not only is it a bit arrogant to ignore the bulk of the estimates in favor of this one guy, it's also harder to defend. If someone challenges this estimate as being to high, they have all those other studies supporting their assertion, while you only have this one guy.

The same goes for picking the lowest available estimate.

But if we grab the middle estimate, we're assuming that the other experts have been wrong in a variety of unpredictable ways. Some are too high, and some too low, but even the wrongest among them are only about half as far off as they would have been if we had picked the highest estimate. If someone challenges this estimate as being too high, you can always point to another expert who would say that this estimate is actually too low -- and vice versa.

That's why I like the median. First you strike out the highest and lowest estimates. Then the second highest and second lowest cancel each other out. Then the third highest cancels out the third lowest, etc. Finally, you end up with a number that an equal number of experts would condemn as too high and too low. This has to be right.

to Table of Contents

Last updated June 2007

Copyright © 1999-2007 Matthew White