November 15, 2002



(See previous post.) Okay, Marc Herold's stats for Afghan deaths January to March, 2002. As we go farther back in time, the numbers start going up, obviously , with 182 civilian fatalities listed for this three month period, in 11 separate incidents. Seven of those incidents, involving 122 deaths, occurred in Paktia province, site of the U.S. Operation Anaconda that March. The others were single incidents in Herat, Uruzgan, Lowgar, and Paktika.

Looking at the 4 out-of-Paktia ones first, we have:
1) 13-18 Jan, 12 killed in Herat: based solely on this quote, from a Guardian story by Ian Traynor:

An internal document from the medical charity Medécins Sans Frontières, written last week and obtained by the Guardian, says that "a large number of civilian deaths and casualties" have been caused by recent US cluster bomb attacks on the Herat region.

Note there is no reference to a number (Herold adds that), no reference to any date, and no confirmation. Discountable.

2) 24 Jan, 15 killed in Uruzgan: it has been conceded by Donald Rumsfeld himself that this was a mistake by Special Forces raiders, who killed friendly Afghans when they believed they were being fired on. Believable.

3) 29 Jan: 10 killed in Lowgar: based solely on a Reuters report (reprinted here):

Residents of Alwazak village, around 50 km south of the capital, said a US bomb or missile had obliterated a residential compound late on Wednesday night, killing 10 members of the Khaderkhal family. They said the attack had caused an old munitions dump to explode, but local commanders loyal to the Afghan interim administration said the munitions dump probably belonged to the Taliban and may have exploded as they attempted to move it.

Reuters (whose reporter later in the story does confirm seeing 10 fresh graves) cannot say conclusively that this was U.S. action, or even that those killed weren't partly or entirely Taliban soldiers. Yet Herold still says it was definitely a U.S. "missile attack." Discountable.

4) 7 March: 23 people, including 16 in the same pick-up truck (!) killed in a US airstrike in Paktika: based on two sources, a Pakistani reporter, Sailab Mahsud, posting from across the border in Wana, based solely on interviews with witnesses, and a BBC Pashto interview with one Haji Hazrat. The official U.S. investigation eventually put the number at 14... Herold, however, doesn't mention that..

So out of 60 fatalities outside of Paktia January to March, we can find sufficient confirmation for inclusion in Herold's list for 29 of them. That's a grand total of 64 successful hits out of the 186 fatalities Herold has listed that we've examined so far, or a rate of 34 per cent. His total of Afghan fatalities has fallen further, down to 3,019.

Next: those in Paktia in this period.

Posted by BruceR at 06:38 PM



Okay, if Herold's logical fallacy isn't quite clear yet, maybe this will help. Imagine a city, with 100 blocks. In each block there's a number of buildings, and one of those buildings in one block is a military target of some kind. The rest of the buildings in that block, and in all those other blocks are civilian housing. The civilian density per block varies from block to block, depending on whether the block has offices, schools, churches, parks, etc. or just houses.

Now imagine you have a bomb that can destroy 1 block, and your mission is to take out that military target. You have two choices of guidance system, an accurate one which hits the block you aim at 90 per cent of the time, and another, random block the other 10 per cent of the time, and a guidance system that's more or less completely random. Herold's point, boiled down, is that since the number of civilian fatalities you're getting over repeated tests, destroying 1 block each time, seems constant regardless of the guidance system, you must be cruelly aiming your accurate weapon randomly, instead. But that implies an unacknowledged assumption, however: that the civilian population density in the block with the legit military target is drastically lower than the average density per block for the whole city. If it's not, if the number of civilians in the military block does not approach zero in fact, then his whole argument falls apart (ie, you just can't in fact easily distinguish random inhumane bombing from targeted humane bombing using that form of measurement.)

My counter-argument is that, even if both systems were found to on a per-use basis to kill a comparable number of civilians (which does appear credible: Herold's more or less right about that) it's still more humane to use the accurate one. For the goal isn't to minimize the number of civilians killed per use, because given the conditions above you can have no effect on that. The goal is to minimize the number of civilians killed before the target is destroyed, and EVEN IF the per-use numbers were comparable, on those terms the more accurate system will still inevitably reduce the amount of damage to civilians, because after the first success that target is destroyed and no further bombing is required at all. So precision bombing must logically still be more humane, even if Herold's data is 100 per cent correct, because Herold's data is in fact showing something entirely different from what he thinks it shows.

Posted by BruceR at 04:19 PM



While we continue to look at the Herold database, the professor is keeping up his campaign to get his own numbers out. Once again, he claims that the exact same numbers we're looking at are unrefuted.

Herold estimates that war in Afghanistan has resulted in between 3,100 and 3,600 deaths from the impact of bombs alone, he said. Throughout his extensive research, he has compiled a 200-page report about every single civilian death during the first months of the war on terrorism. "This has not been refuted," he said.

We'll resume the refutation in a little while. More interesting as an indication of Herold's intellectual rigour is his evidence for his argument that on a per-ton basis, the Afghan war is a particularly bloody aerial war.

He said about 2 million tons of bombs were dropped during a major campaign in Laos during the Vietnam War in the 1960s. Between 1,750 and 2,500 civilians were killed for every tonnage of bomb. In Cambodia, about 540,000 tons of bombs were dropped, which he said resulted in between 926-2,700 deaths per tonnage.

This is contrasted with the 14,000 tons of bombs dropped in Afghansitan when he compiled his report several months ago. In terms of the ratio of people killed to tonnage of bombs dropped, the Afghanistan bombing is more deadly.

"The carnage that took place in Afghanistan is on par with the carnage that took place in Laos and Cambodia," he said.

This paragraph is confusing (the reporter means "per 10 kilotons" when he says "per tonnage" in most cases), so here's Herold's actual math in tabular form:

Laos: bombs dropped: 2000kT; civilan fatalities due to bombing, 350,000; deaths per 10kT: 1,750
Cambodia: 540kT; 150,000; per 10kT: 2,700
Afghanistan: 14kT; 3,600; per 10kT: 2,570.

The fatality figures for Laos and Cambodia are, predictably, fuzzy, but not off the scale of believability. (By comparison, Christopher Hitchens, in The Trial of Henry Kissinger, offers 350,000 and 600,000 respectively). Nearly every other commentator has pegged the Afghanistan fatalities at one-third or less of what Herold says they are (I'd say c.900 fatalities per 10kT myself). Herold's numbers for the other conflicts are certainly within a factor of 3 of Afghanistan's, one way or the other, which at first seems surprisingly close. Here are some other figures, also fuzzy, which are also within that same factor-of-3 range, though:

Kosovo: 17kT; 500; per 10kT: 290;
Iraq (1991): 85kT; c. 3,000; per 10kT: c. 380;
World War 2 (USAAF only): 2150kT; c.400,000; per 10kT: c. 1800.

So, if you ranked U.S. air efforts on Herold's humaneness scale, in terms of civilian deaths per 10kT (with percent of precision weapons in use in brackets, for comparison), you'd get:
1. Kosovo (30 per cent precision)
2. Iraq (9)
3. Afghanistan--other people's fatality estimates (90)
4. Laos (0)
5. World War 2 (0)
6. Afghanistan--Herold's estimate (90)
7. Cambodia (0)

Now, let's take as an assumption that Herold's estimate of the fatalities is correct. (In fact, this list itself would be strong evidence that it's not, but never mind that for now.) One can then draw one of two inferences from this. Precision weapons, we can agree, generally go where they are aimed. So either a) the Americans must have been deliberately targeting large groups of civilians in Afghanistan, because they were certainly killing them much more efficiently apparently than either waves of B-17s in 1945 or waves of B-52s in 1972 over Laos could do; or you can conclude that deaths per 10kT really isn't a particularly useful statistic, and Herold's fixated on a canard here.

I mean, think about it. There's no logical reason there needs to be a correlation. 10,000 tons of high explosive, evenly applied over an area, will cause x number of casualties. Even if one assumes that the individual weapons are better targeted at military targets (and that kill rate will go up correspondingly), those military targets aren't discrete geographically. On a national or regional or even city-wide level, it's still the same number of tons over the same area. By my own estimate, the USAF day bombing offensives over Germany and Japan killed about 0.06 civilians per weapon dropped. B-1s dropping GPS-guided bombs over Afghanistan killed about 0.07. Does that mean a B-1 is as indiscriminate a killing weapon as a B-17? By Herold's analysis it would have to be.

No, the difference is that in 1945, the vast majority of bombs dropped failed to affect the target being aimed at. In 2001, the vast majority did. That means the total amount of tonnage dropped on that target, or on the country or city as a whole by extension, to achieve the same military effect could drop dramatically. That, in turn, has a dramatic downward effect on total civilian deaths. You don't need to bomb the same target over and over again, plastering all the civilian dwellings around it, because it was killed the first night you tried.

Here's an example. In the Gulf War, where precision weapons were still relatively rare, the air campaign had to drop around 2kT a day to achieve its aims, about the same as it had to in Vietnam and WW2, as a matter of fact, and no doubt would have dropped more if it had the planes and bombs to do it. In Afghanistan the air force met all its aims with 0.1 kT a day, a significant reduction in the tons of bombs being dropped, which even if the civilian-loss-per-10kT figure remains difficult to reduce, meant that the aggregate tonnage (and hence civilian casualties) was kept down.

Carl Conetta has already explained much of the remaining discrepancies in Afghan casualty stats to most others' satisfaction (it has a lot to do with the relative circular-error stats on laser-guided vs. GPS-guided weaponry.)

In short, Herold's right, in an irrelevant kind of way. Yes, precision weapons, per ton of explosive, kill about as many civilians as unaimed ones do. The difference he pays no attention to, and the reason that recent American wars have set a new standard for humaneness, is that in the 21st century, you need fewer tons of explosive to do the same job.

Posted by BruceR at 03:06 PM