The numbers are too big
One of my favourite blogging moments is when the now-retired left-wing scienceblogger Chris Murtaugh wrote to his co-ideologists regarding race and IQ: "The universe is not here to please you." Facts are not to be denied simply because you find them inconvenient.
The same is true for those of us who supported the Iraq war regarding the Lancet study. The universe is not here to please us; the fact that we would like there not to have been massive civilian casualties does not mean that there were, in fact, no such casualties.
Indeed, I would be surprised if there weren't. Wars disrupt things, like food shipments and water. THat makes people die.
This does not mean that we should never go to war. If you take a group of cancer patients and split them into two groups, one of which gets surgery and the other of which does not, you will see sharply increased deaths among the surgery patients during and after the surgery. But over ten years, it may be a very different story.
However, when assessing the benefits of surgery, you do have to consider the cost. Sometimes--perhaps even often--the benefit is not worth the added risk.
Kieran Healy takes on the right-wing reaction:
The Lancet paper by Burnham et al. study estimates about 655,000 excess deaths in Iraq for the period of March 2003 to July of 2006, of which about 600,000 are directly attributable to violence—an appalling number. Right-wing reaction has been, understandably, that the 600,000 estimate is unbelievably high. (Tim Lambert gives a roundup.) Convincing those critics who see this number and declare “that can’t possibly be right,” or “my gut says no” or “this doesn’t even pass the smell test” is difficult. This is partly because some will just think that any estimate that sounds bad must be false, and take refuge in old saws about lies, damned lies, and what have you. But it’s also partly because six hundred thousand violent deaths since the war began seems huge—and, frankly, it is. As this typical guy says, that’s equivalent to 3 to 10 Hiroshima atomic blasts, 6 to 20 Nagasaki atomic blasts or 10 Dresden bombing campaigns. Yes, that’s right. Those events happened in a single day or over a very short period. The present estimate is for a large country of twenty six million people over three and a half years. Sadly, this means it’s quite achievable. As Juan Cole points out, you just have to believe that for our five people a day are being shot or otherwise killed in each of Iraq’s major towns outside of Baghdad.
I agree that there are a lot of deaths. But I don't think that there are as many as the Lancet study makes out. Not because I am comparing them to one-time events like Nagasaki, which I agree is silly. But even comparing them to other long term wartime figures makes it look to big.
Germany, with a prewar population of just about 80 million, suffered 1.8 million civilian deaths during six years of invasion, concentrated aerial bombing of civilian targets, and occupying forces that in the case of the English and Americans, frankly didn't give a [expletive deleted] what happened to the Krauts, and in the case of the Russians, took great pleasure in terrorising, raping and killing the local populace in revenge for their own dead. How likely is it that Iraq has lost a higher percentage of its civilian population in three years--especially given the vast advances in medical care, field treatment for water supply issues and famine, and GDP? With my admittedly limited knowlege of World War II, I find it very difficult to believe that the insurgents are worse than the Russians were, not to mention the Allied Air Command.
The Netherlands lost 30,000 people out of a population of roughly 9 million during six months of famine, during which the average calorie consumption dropped well below 1,000 per day. There was also a total famine of medical and other supplies, which could not pass through the battle lines. How likely is it that there is a larger humanitarian crisis in Iraq than there was in a country getting no food or supplies whatsoever?
Or to compare it to another civil war, this is more deaths than America's civilian and military deaths combined (union and confederate) during 4 years of brutal civil war with no medical care worth having, Sherman's march to the sea, and the tragic mistake of using massed formations against repeating rifles and modern artillery.
The average report is of about 30 civilian casualties a day--a horrifying number that should sting the consciences of those who advocated war. I'm sure that there are more whose deaths go unreported. But assuming that violence is the major cause of death, how likely is it that the newspapers are all consistently underestimating the number of violent civilian deaths by a factor of five or more? Okay, maybe they're all happening outside of Baghdad. Except outside of Baghdad includes the Kurdish north, where 10% of the population is mostly not getting shot by insurgents. And a lot of Iraq's other towns outside of Baghdad aren't that big. Bayji, a major oil centre, has 60,000 inhabitants. As anyone from a town that size can tell you, it wouldn't go unnoticed if it was losing 1,600 people a year to murder.
The Lancet study is arguing that in the space of 3 years, Iraq has lost 2.5% of its population in extra deaths. That doesn't sound like much, but it's an enormous figure, as these things go. It's even more enormous if you exclude the Kurdish areas, which are pretty stable, and have about 2.5 million people in them; the various stable places in the Shiite south, and the roughly 5% of the population that, according to the Atlantic Monthly, has fled the country. That suggests that Iraq is losing more than 1% of its remaining population a year to violence--as if 3 million Americans a year were getting shot in the streets.
I find it unlikely--not impossible, but unlikely. Certainly, in a country like Iraq, which as war zones go is pretty well organized and supplied (don't look at me like that, anti-war people: read some history, for heaven's sake. Or go check out the Congo) it's much bigger than we should expect, even with horrible sectarian violence.
Update One of Mr Healy's commenters points out something that makes the figure even more unlikely:
One point on the 470 per day explanation proferred by Cole.
The original study published in October 2004 went with 100,000 excess deaths as the confidence interval mid-point. This was for the period Mar 2003 – Sep 2004. The new study has revised upwards the figure for this period to 112,000. This means that for the 22 month period Oct 2004 to Jul 2006, the study finds 543,000 excess deaths (655,000 – 112,000). This breaks down to an average of 822 excess deaths every day for the 670 or so days between Oct 2004 and July 2006. Given we’re talking an average of 822, this means that >1,000 daily excess deaths must be commonplace.
Update II Anyone want to calculate the odds that the Lancet's habit of publishing these studies right before American elections is a product of chance? As I say, I think that conservatives are protesting too much . . . but the other side also seems to be way too emotionally attached to the idea of heavy casualties. Neither makes me willing to trust the results, and the fairly obviously political timing of these studies makes me suspicious. As Medpundit says:
I admit, this headline caught my eye. 655,000 dead in Iraq is an impressive number. Then I read the first sentence and saw that the number was gathered by public health researchers and it lost some credibility. The American public health community has a decidedly left leaning cast to it.1 It is more politically homogenous than any other medical specialty. How homogenous are they? Well, you won't find statements like this on the website of any other medical speciality. One is obliged to assume that the researchers started with a bias.
Then I read that it was published in The Lancet and I lost all interest. This is the journal that gave us the infamous MMR-causes-autism study and that published a similarly discredited tally of Iraqi casualities before the last American election. In the ranks of medical journals, I place them on a par with The Guardian.
Obviously, the fact that the researchers are likely to be left-wing doesn't invalidate the study. But observer bias does matter, especially in survey studies. My personal feeling is that given the difficulties of doing research in a war zone, any study, whether it bolsters or refutes my opinions, is likely to be crap. We'll know how many excess deaths there were when Iraq calms down, the refugees return, and they get a decent census; not before.
As Medpundit points out, this stuff may be bog-standard for public health work, but public health work is not exactly known for its outstanding statistical methodology:
And sorry, but the defense that it's as soundly designed as can be expected for these kinds of public health surveys is a weak one. Retrospective, interview-based studies like this are poor designs. It may be the standard way of gathering data in the public health field, but that doesn't make it the best methodology, and it certainly doesn't make its statistics sound. For too long the field of public health has relied on these types of shotty shoddy numbers to influence public policy, whether it's the number of people who die from second hand smoke or the number who die from eating the wrong kinds of cooking oils.
Hello, cancer clusters!
Posted by Jane Galt at October 12, 2006 10:15 AM
janegalt.net
Selected comments
Another possible problem with "excess deaths" is that you shouldn't trust the pre-war Iraqi death rates. As it stands, the reported death rate was dramatically below the US, UK, Germany, France, Brazil, and Mexico. (8.1,10.3,10.3,9.0,9.3,10.1 per 1000) {I used the easiest to find rates over the past four years for each country, which isn't strictly sound but considering the lack of huge disasters in any of the reference countries I suspect isn't the cause of an error}. The reported Iraq number appears to be around 6.4.
When I was looking at the numbers originally I suspected it might be a function of age, which is why I added two relatively younger countries--Mexico and Brazil.
So under the sanctions regime (which was allegedly killing thousands) Saddam had apparently a drastically better death rate than the US, UK, Germany, France, Brazil or Mexico. That is drastically as in 20% better than the US and 35% better than the UK and Germany. (Am I doing that calculation right? Normally I'm math-confident but that seems ridiculous.)
Posted by: Sebastian Holsclaw on October 12, 2006 1:02 PM
janegalt.net
In the first survey, the one that estimated roughly 100,000 excess deaths
in the first twenty-one or so months of the war, a very small group of
pollesters asked a relatively small group of households to list the number
of people that had died in their households in the preceeding three years.
Something like a normalized forty or so more people died after the war than
before. If the same was assumed to be true of the rest of the population
then we'd have 100,000 people dead that wouldn't have been if it weren't for
the war.
It seems like a straightforward calculation, but there are some interesting
assumptions behind it. For one there is the whole question of what 'household'
means. The people making the study had a narrow definition: the people living
in a dwelling plus those that used to live in that dwelling but had died. Did
the people answering the pollesters have the same definition? Suppose you
have a son that died. But your son hadn't been living in the same building
for some time. When asked if someone in your household had died, are you
really going to leave out your son?
The authors of the Lancelot survey assume you would leave out your son.
I don't believe that this is what people would really do. So in fact the
survey wasn't a 'household' survey; it was somewhere between a household and
extended family survey. Since extended families are much larger than households
the practical impact would be to significantly lower the number of excess deaths
when extrapolated to the population as a whole. All this is obvious or
would be obvious if one was actually doing such a survey. An ethical
researcher would have attempted to estimate the magnitude of such an effect
and included a second estimate for the number of deaths making the assumption
that extended families were being surveyed instead of households.
Another interesting assumption is the belief that everyone will answer honestly.
Obviously some people are and were deeply committed to one side or the
other. If a person is willing to aid and abet those striving to kill
americans can there really be any doubt they would lie if there seemed a benefit to
lying? Given the small number of actual reported deaths it would have only
taken a handful of people answering disingenuously to completely throw the survey.
Posted by: Mark Amerman on October 12, 2006 1:37 PM
janegalt.net
Looking at the CIA 2003 Factbook entry, updated 8/1/2003:
Death rate: 5.84 deaths/1,000 population (2003 est.)
Looking at the CIA 2006 Factbook entry:
5.37 deaths/1,000 population (2006 est.)
So why would the authors of the Lancet paper rely on the 2003 CIA Factbook to boster their case aas to the 2003 death rate without admitting that the 2006 Factbook completely undermines their theory?
If we rely on the CIA factbook, there have been NEGATIVE "excess deaths" in Iraq.
Posted by: Al on October 12, 2006 2:00 PM
janegalt.net |
