Daniel Davies comments on the current outbreak of Lancet denialism, including Shannon Love's latest effort which Love describes as a "Fisking". ("Fisking" is a term bloggers use for especially lame posts.) Love gets taken to bits in the comments to his post; I don't need to add anything.
From Glenn Reynolds
Reader Peter Malloy emails: "The inability to sense of irony among the anti-war left never ceases to amaze me. The way this group clings to the 100,000 deaths figure makes clear that they WANT it, desperately, to be true. This group of people, ostensibly against the killing involved in war, actually desires that Iraqi civilians have suffered the worst case outcome in order to support their political views. Real humanitarians, ain't they?"
I don't have the ability to read the minds of the "anti-war left" and neither does Malloy. I can, however, read my own mind and Malloy can read his own mind. When he makes a statement about their desires it is based, not on actual knowledge of their desires, but whatever he can extrapolate from his own desires in the matter. So it well may be that Malloy wants to deny the results of the Lancet study, to pretend that Iraqi civilians have not suffered, in order to support his political views. Real humanitarian, isn't he?
Those who are humanitarians will want to Count the Casualties in Iraq.
- Log in to post comments
It's called projection, Tim, and in this case it's kinda creepy, rather moronic, and to use your description, innumerate. But that's what thinking people are up against and no sense in denying it.D
LOL Lancet denialism? How about your Lancet fantasism?
I think it's beyond obvious that a study that proves only that there were at least 8,000 excess deaths and no more than 194,000 is totally worthless, even leaving aside the many methodology problems that have been brought up.
I'd like to see a real study done, with solid methodology agreed on by representatives of both the left and right beforehand, to settle this issue fairly. We could all agree to accept its result, whether it finds 250,000 excess deaths or 250,000 lives saved.
TallDave, you are not motivated by problems with the methodology, and you don't understand enough about the math to know whether there are any real problems with it*. You are clearly motivated by a need to minimize the carnage and destruction the Bush administration is causing in Iraq.
* You still don't get that the study does not "proves only that there were at least 8,000 excess deaths and no more than 194,000", do you?
You still don't get that the study does not "prove only that there were at least 8,000 excess deaths and no more than 194,000"
Sorry, that's what it proves (within a 95% confidence level.) If we were to tighten the band a bit (say to 20% around the midpoint) we'd be looking at a mere 16% confidence level. And if we were to include the discarded Falluja cluster altogether (rather than admitting it capriciously and unevenly) the confidence interval might be wide enough to extend into positive territory.
You are clearly motivated by a need to minimize the carnage and destruction the Bush administration is causing in Iraq.
Considering that only 9 out of 21 ex Falluja violent deaths are attributed to the coalition I'd say this statement is somewhat inaccurate.
pop quiz: who said there is no very good way to make general statements about percentile confidence levels of this sort of distribution...
If your answer was dsquared, you win a kewpie doll. I actually agree with him.
No, telluride, that is NOT what it proves at all. And the Falluja data was not "admitted capriciously and unevenly", it was excluded altogether from the calculations. It was certainly included in the discussion, and it would have been very inappropriate not to include it.
"the confidence interval might be wide enough to extend into positive territory.
What on Earth is that supposed to mean?
The Bush administration is not merely responsible for the death and massive destruction caused by its attacks. As the aggressor, the Bush administration is responsible for all the consequences,directe and indirect, of its aggression. Furthermore, as the occupying power, the Bush administration bears full legal and moral responsibility for the lives, welfare, and maintenance of the people and property under its occupation. The Bush administration bears full legal and moral responsibility to abide by both the letter and the spirit of the Genevan Conventions, and all other applicable instruments to which it is signatory.
To say that the Bush administration has virtually completely abrogated its legal and moral responsibilities in virtually every aspect of its conduct of the invasion and occupation of Iraq is far too kind.
Talldave: "I think it's beyond obvious that a study that proves only that there were at least 8,000 excess deaths and no more than 194,000 is totally worthless"
As Tim has pointed out, the range of 8,000-194,000 is informative in that it doesn't include "0".
In other words, to date, more Iraqis have died as a result of the invasion than would have died had the invasion not occurred.
I am, in general, part of the PRO-WAR left, having supported military intervention in Kososo, Somalia; Afghanistan and East Timor amongst other places.
In specific, I opposed the invasion of Iraq because I believed from the outset that it was going to kill more Iraqis than would have died had Saddam reamined in power and that the predictions of the ultimate benefits of the invasion were wildly unrealisitic.
. It was certainly included in the discussion, and it would have been very inappropriate not to include it.
actually it's quite inappropriate to include it. and it is included in "the calculations" represented outside of the study as when garfield & roberts say the total is "probably more" than 100k. Admitting falluja to any quantitative assessment should yield a vastly larger confidence interval they dare not express. They are stealing bases.
"we concluded that the civilian death toll was at least around 100,000 and probably higher"
-Garfield & Roberts, in The Independent.
"Probably higher" from the mouth of a statistician, sounds to me like a positive scientific, quantitative assessment of probability. Either their assumption is a normal distribution without the falluja data, or a normal distribution WITH the falluja data, and the concomitant increase in variability/standard deviation. it is altogether conceivable that "zero excess deaths" is encompassed by this vastly larger confidence interval.
telluride,
That quote is probably accurate as presented here, but I notice that we haven't been provided with a link to the original article in order to confirm it.
""Probably higher" from the mouth of a statistician, sounds to me like a positive scientific, quantitative assessment of probability."
It depends on the context and the intent behind the statement. It could just be an opinion. Believe it or not scientists and statisticians have them. The main difference is that they are more likely to have formed their opinion based on real evidence than on wishful thinking.
And no, it would have been as inappropriate not to have included Falluja in the discussion as it would have been to include it in the calculations.
It could just be an opinion. Believe it or not scientists and statisticians have them.
The problem lies in the defnition of "real evidence." "At least 100,000" is factually accurate (as the lower bound of the 95% confidence interval is actually 8,000.) & "probably more" admits the data from falluja without admitting its "bad qualities" -- ie the effect on overall variability of results. "Real evidence" is clearly in the eye of the beholder here.
accurate s/b "factually inaccurate..."
Actually, telluride, the best evidence we have so far indicates exactly that the number of excess deaths is at least 100,000.
Sorry - I did not say that very well. I should have said that the best evidence we have so far supports the view that the number of excess deaths is at least 100,000.
The authors also have stated that there should be further studies, and I don't think there is anyone, including those of us who find the study credible, who does not agree whole heartedly with that.
Actually, telluride, the best evidence we have so far indicates exactly that the number of excess deaths is at least 100,000.
Oh, ok. That explains everything. [/sarcasm.] Whatever your "best evidence" is, it aint in the Lancet study.
By the way, telluride, maybe you simply have not had the chance yet, but I am a bit disappointed that you have not seen fit to acknowledge the fact that your source directly contradicts your claim that malnutrition rates increased from 1996-2003. It shows that there was a consistent decrease in the rates of acute malnutrition and underweight, with a very dramatic net decrease, and a very significant net decrease in chronic malnutrition. Your source shows very clearly that by 2003 the rates for all three were very close to pre-sanction rates.
"Whatever your "best evidence" is, it aint in the Lancet study."
Really, telluride? What better evidence is there, then?
contradicts your claim that malnutrition rates increased from 1996-2003.
I ignored your post, since it was more irrelevant afactual argument from assertion and it didnt address anything I said. The 1996-2003 time frame came from you, not me. As I stated, the chronic malnutrition rate was higher in 2002 than in 1991 (as was the acute rate), and yet you would have me believe the infant mortality rate was 25% lower in 2002 than 1991, after 10 years of sanctions, simply on your say-so. That seems counterintuitive to me, if not to you. Keep pecking away at off-topic afactual minutiae if it makes ya feel better.
"what better evidence is there" is simply an ad ignorantium fallacy. There isnt better evidence. That doesn't make this evidence good.
Another post on the confidence interval. For what it's worth, I did the calculations (summarised in the comments section of Tim's last post - comment number 46, I think) on the probability that the number of excess deaths is greater than zero, if we take authors' 8000-194,000 confidence interval at face value. It's 98%. Alternatively, we could say that there's a 90% chance that the number is above 38,000. The probability that it's above 98,000 is 50%. Of course, there's some implicit statistical assumptions here, as always. But without going into a tonne of statistical theory, it'll have to do.
"you would have me believe the infant mortality rate was 25% lower in 2002 than 1991"
should read
"you would have me believe the infant mortality rate was 39% lower in 2002 than 1991"
Dan
If there was a 50% probability that deaths exceeded 100,000 as of September 2004, subsequent fatalities would puch that considerably higher.
Ian: Agreed, for the most part. But I don't have those data, so I'm not going to hazard a guess as to what they might look like. In theory, it's possible (not very likely, I would think, but certaintly possible) that Iraq got much more peaceful since Sept 2004, and the death rates dropped below pre-war rates, so the number of excess deaths would actually be decreasing. However, while I don't think that's a very plausible scenario, I don't have any actual data to work with, so I don't want to try to say anything about it.
"There isnt better evidence. That doesn't make this evidence good."
No, but it most certainly DOES make it the best evidence we have to date.
And of course, the fact that the people who are working so hard to pick it to pieces are, almost without exception, people who know virtually or absolutely nothing about epidemiology or the principles or mathematics involved in using samples to predict population statistics, and who virtually invariably believe the estimate is too high, does not make the evidence good, either, but it certainly discredits its detractors. Add to that the fact that professionals who are well versed in exactly the methodology used in the study, or in using samples to predict population values have not found serious flaws in either the methodology or the results, and the balance certainly tips in favour of the study and its results.
"it's possible (not very likely, I would think, but certaintly possible) that Iraq got much more peaceful since Sept 2004"
It did not become more peaceful, believe me. Both U.S. perpetrated violence and the violence from the so-called "insurgency" actually increased.
"people who know virtually or absolutely nothing about epidemiology [insert overlong pompous, ostentatious appeal to authority here]..."
Richard Garfield knows something about epidemiology. So does Richard Horton. Yet each one has flagrantly misrepresented the results of the Lancet survey. You claim to know something about statistics. Yet for some reason you deny the existence of a lower bound of 8,000 in the study's confidence interval. You also seem happy to let a discarded outlier inform your view of the survey's core finding (but not another, equally important finding.)
I do wonder why these lavishly credentialed people are being ignored!
Shirin: Yeah. This is what all the newspapers tell me. I guess that it's probably true, sadly. All I really wanted to say is a rise in excess deaths from Sept 2004 to the present isn't mathematically guaranteed. But unless every news source besides Fox (sorry, that was snide) is lying to me, it's pretty bloody likely to be true. However, I'm not the expert in this area. I find the papers depressing and infuriating, so I avoid them a lot of the time for mental health reasons.
Either their assumption is a normal distribution without the falluja data, or a normal distribution WITH the falluja data
No? It's normal without the Fallujah data, but not normal with the Fallujah data; this would be what you'd expect if the non-Fallujah data was drawn from a different distribution from the Fallujah data.
You're in danger of missing the wood for the trees here; if the death rate had gone down, how likely would it be that you would get a sample like the ex-Fallujah data? Not very likely at all. When you then add the Fallujah data to your information set, does that make you think that a falling death rate is more or less likely? Surely not more!
I suspect that bringing sampling and distribution theory into the discussion may make the flypaper even more attractive. But what the hell.
As I understand it the only reason why need to assume normality of any distribution is to construct a CI, and for that purpose what matters is that the estimator has a normal distribution. To justify that assumption we invoke the Central Limit Theorem. Within reason you can assume anything you like about the underlying population distribution.
Do the critics have a problem with that? Or rather, can they say plainly what their problem is?
Surely not more!
It needn't have any impact. As with the example in the other thread: you have 32 numbers between 30 and 50. Adding 200 to this set tweaks the mean quite a bit, but not the mode or median at all. You and I seem to be working off of different notions of 'likelihood' which explains some part of our dispute. I contend that "likelihood" as defined by the confidence interval in a radically skewed distribution does not describe "expected value" or anything related to the mean but rather properties of median and most distinctly the mode. I think you and I agree that applying normal-distribution-based confidence intervals to this study is somewhat meaningless (where it is applied consistently, which is nowhere.) In any case, citing the outlier reintroduces discarded data, and renders the cited CI [even more] meaningless. Allude to Falluja, you must allude to the volatility and/or distributional issues connoted by that data.
[OT?: This is why I brought up the derivs example in the other thread. When the "expected value" of a given low-delta skewed-distribution-based instrument (call it a weather call since there is no hedgeable underlier) increases, the probability of its likely exercise necessarily - and counterintuitively - decreases. For similar reasons any lognormally distributed variable has a greater chance of going down than up.]
Kevin, my problem is that the CI for the study ex-Falluja is plainly expressed [8,000-194,000] but the CI including Fallujah is unknown. Yet in pops Falluja now and again to "qualify" the midpoint of the ex-Fallujah numbers ("at least 98,000 and probably more"), unencumbered by any revised CI. This conceptual amalgam is statistically meaningless.
telluride,
When you say âthis conceptual amalgam is statistically meaninglessâ you lose me. I learned statistics without philosophical jargon, using a mixture of methods. The most illuminating involved numerical examples. So, consider the following case:
10,000 towns are neatly arranged by latitude and longitude in a 100x100 grid. Each town has 1,000 inhabitants hence total population is 10 million. The true mortality situation, known only to God, is as follows. In 100 towns 10 people die and in the other 9,900 just 4 people die. So the death toll is 40,600 and the mortality rate is 4.06 per thousand.
Fallible statisticians survey 30 towns, selected at random from the grid. There is a 74% probability that they will select none of the high-mortality towns. If that happens, they will estimate the mortality rate as 4.00 per thousand and since there is zero variance in their sample they will have high confidence in that figure â which as it happens is not far out.
They more interesting case is where, by chance, they select just one of the high-mortality towns. They get an estimate of 4.20 per thousand. This is slightly worse than the first case, but since they havenât experienced the first case they might reasonably go with it. However, the single high-mortality case worries them. Maybe it is a freak observation. Maybe the locals lied to them.
They report their findings as: conservatively-estimated mortality rate 4.00 per thousand, omitting one town whose experience they discuss separately.
If this is wrong, why so?
Is Telluride making a sensible mathematical point? I'm asking, not taking a shot. His claim seems unlikely.
If you included the Fallujah outlier, the naive extrapolated death toll would be close to 300,000. I assume the confidence interval would be much larger , but it'd be surprising to us naive nonstatisticians if it actually made it more likely that the death excess from the invasion was negative. Suppose the Fallujah cluster had shown 250 deaths rather than 52--would that make it even more likely that the invasion had lowered death rates?
My wild guess is that non-Fallujah clusters really do represent what's happening in most of Iraq. The Fallujah cluster shows what sort of casualty rate you can expect to see in neighborhoods that have been suffered repeated aerial attacks and what we can't tell from this study is how many such neighborhoods there are in Iraq. But if the non-Fallujah areas probably suffered between 8000 and 194,000 excess deaths, it doesn't seem likely that including the Fallujah cluster is going to give us an estimate that veers into the negative range. But maybe statistics is more counterintuitive than I realize.
The study does give the 95% CI when Falluja is included -- it is 1.6-4.2 for the risk of death. subtract 1 and multiply by 200k to get the CI for excess deaths: 120k-640k.
Actually, it's reasonably intuitive. What does the confidence interval depend on?
In layman's language, on the sample size, and on a measure of how much each sample point is likely to vary from the mean. The more often we sample the tighter the confidence interval, the greater the tendency to vary, the wider the confidence interval.
So, think of us attempting to measure something, say a volume, and we know the likelihood of deviations from the true value for each measurement, say 90% within +/- 1, 99% to within +/- 2 and 99.99% to within +/-3 (which is a description of the variability of the results).
If we do enough measurements, we can get the likelihood of being +/-1 for the mean up to 99.99, as well. And that depends on the number of samples and the variability of our measurements (the example shows that I am an engineer I suppose).
Now the key here is what do we do when we don't know the variability of the data? Well, we've got to estimate it. And we could use the data themselves to do that. If we've got death rate estimates per cluster before and after varying a little (from 0 to 15 say) and then see it varying a lot in one case (to 200 say), our estimate of the variability of the death rate goes up, and therefore the confidence interval becomes wider.
The variability goes both ways you see. It kind of says that there might be a cluster or two pre-war (with mortality of 200) we may not have sampled, and to account for that the confidence interval would then widen to negative values.
To put this another way, it says that rare clusters might have an undue influence on the average, and to get the average right, we therefore need to sample more clusters.
Tim, I don't see how you are making that calculation. Ex-Falluja the rate of risk is given as (1.1-2.3) which the report translates numerically to (8,000-194,000.) By your logic that range should be ([.1 x 98k - 1.3 x 98k] = 9800 - 127,400 And why would you multiply by 200k and not 298k for the falluja inclusive numbers: ie 180-1MM ish?
Kevin the difference in variability connoted by your example I don't think comes near the one suggested by the Lancet findings. The fallujah cluster result alone is 58 times larger than the average of the rest. your a priori distribution is far closer to ideal than Iraq could possibly have given the results of the survey. which is why the results were thrown out (and then wiped clean and hauled out again and again).
Donald, I am not saying it would lower the excess death number. But I know that holding to any conventional distibutional assumptions it would blow the CI out to beyond the point of ridiculousness (see guesswork above, 800,000 deaths wide.) Which is why I'm guessing the including-falluja-cluster confidence interval isn't cited as a firm figure on the summary.
Of course, there are several ways to calculate this variability. In simple terms, if you make the assumption that the pre-war death rate is unlikely to fluctuate as much as the post-war one for each cluster, you won't get a negative value for your 95% confidence interval.
Donald Johnson,
Your intuition matches mine. The main purpose of studying the probability distribution of an estimator is to test some hypothesis - in this case whether excess mortality is greater than zero. A large positive observation like Fallujah widens the confidence interval but I can see no sensible way it could alter the verdict of the test.
But if a Lancet critic can teach me something about statistics so much the better. I'm not proud.
Telluride,
Thanks for your reply. Obviously it doesn't matter in the least whether the top end of the CI goes skywards. The rejection of the null hypothesis (zero excess deaths) is all that matters.
"The fallujah cluster result alone is 58 times larger than the average of the rest." Could you expand on that?
Could you expand on that?
Sorry, Heiko, a sloppy cuff sloppily phrased. I meant that 98k excess deaths divided by 32 clusters yields about 3k excess deaths per cluster, whereas Fallujah ostensibly yielded 200k, which I suppose makes it even higher than 58. You are more familiar than I with the specific breakdown by cluster, so I'm sure you know the exact fig.
Ok, I see.
One of my postings seems to have been lost. The gist of that was that we could also regard Fallujah and the rest of Iraq as two separate populations that need to be sampled separately. In that case the sample size for Fallujah would be too small, and we'd have to estimate Fallujah and the rest of Iraq separately and could then add up the two estimates.
I did a bit of playing with this "effect of outliers" thing (telluride, Donald, Kevin), and I think I can explain why it's such a mess. When you add a big positive outlier to data that come from a normal distribution, it increases both the mean and the variability, and because the normal distribution is symmetric, this sometimes pushes the lower bound of the confidence interval downwards. For example, if we have data X=[5 7 2 -2 10], we get a mean of 4.4, standard deviation of 4.6, and thus a standard error of the mean of 2.1. So the lower bound of the 95% CI is set at 0.35. Zero isn't inside the CI. Now add an outlier, so our data set is Y=[5 7 2 -2 10 20]. The mean is now 7, the standard deviation is 7.6, and the standard error is 3.1. That puts the lower bound of the 95% CI at 0.93. Zero is still outside the CI. However, if the outlier is much larger, things are different. If the data are X=[5 7 2 -2 10 40], the mean is 10.3, the standard deviation is 15.1, and the standard error is 6.2. This puts the lower bound of the 95% CI at -1.8. So zero is now inside the 95% CI. I drew some pretty pictures of the distributions, but I don't know how I'd post them.
As far as the Lancet data goes, my guess is that it would be wrong to assume that the data all come from the same distribution. An different analysis could assume that different regions have different underlying distributions. So the overall estimate of excess deaths would be found by averaging across the various regions/distributions. This sort of hierarchical modelling is fairly common in some disciplines, but I've no idea about whether its been used much in epidemiology.
Oops. The third data set should've been labelled Z for the sake of clarity.
Thanks for the various replies. To Dan--
Wouldn't the Fallujah cluster be a rather obvious case of sampling from a population with a very different death rate from most of Iraq? The small areas which have been heavily bombed will have much higher death rates. Heiko suggested there might be some prewar places with exceptionally high death rates, so finding a high postwar cluster might mean the Lancet study missed a prewar cluster of similar magnitude, but that doesn't follow, because we know why there was an outlier in Fallujah--the place was heavily bombed and there are other places in Iraq which might have suffered similar levels of American violence. There presumably would have been Fallujah-type outliers in Iraq in in 1991, when Saddam was crushing the Shiite and Kurdish rebellions or in 1988 during the Anfal campaign, but is there any reason to think that in 2002?
Not that I'm aware of.
I think in one of the interviews I read with one of the Lancet authors he expressed some regret that they didn't get to sample another neighborhood in Fallujah--I assume the idea here was that to get a better handle on the death rate in the heavily bombed areas.
Yes, to get a tighter bound it would be necessary to oversample Falluja. The dpressing thing is that you would need about as many clusters there as in the rest of Iraq. And of course, if the authors had done this, you can imagine what the Shannon Loves of the world would be saying ("The Baathist-loving Lancet oversampled Falluja to make the numbers higher").
Donald: I reckon you're right, and it would be nice to be able to do that kind of analysis with the Lancet data. Part of the problem is that most of the standard methods used in science came out of work by Sir Ronald Fisher, Karl Pearson and Jerzy Neyman back in the 1920s through 1950s. Back then all the calculations had to be done by hand, so it was just ridiculously hard to do the kinds of analyses that we'd want to try today. However, machine learning people have gotten very good at this kind of thing, so it might be possible to learn a more plausible model from the raw data. In fact, it might be a nice thing for serious stats people to do (some stats folks would even say ethically necessary). After all, a bunch of people put their lives at risk trying to collect the data, so the least we can do is see how effectively we can model the data we've got (of course, doing science in a political arena is a surefire way not to get tenure). Off the top of my head, I can imagine that Fallujah would show marked differences from other regions, and would end up with it's own distribution, which would be much noisier since the estimate would be based on less data.
Tim: How very depressing. But I wouldn't bet against you on it.
Donald,
I am essentially saying exactly the same thing as Dan, but I've been trying to do it in simpler language.
I've described three possible choices you could make in the calculation. If, as you do, you assume that the Fallujah cluster gives a good idea of possible cluster death rate variability post-war only, but does so for all of Iraq, then the mean will zoom, and the CI will stay above zero. If you don't assume that and try to estimate the variability of the data just from the data themselves (essentially what Dan's exercise is about), the CI will in all likelihood widen to include 0 (just that Dan's exercise doesn't tell you why that would be so in simple language).
And, if you make the choice to regard Fallujah as an entirely separate population, which I personally think is the most accurate, as it was the only area actually under tight insurgent control that required an outright invasion with heavy casualties last year, you need to sample it better, because at the moment there is only one cluster of 30 households in it. These 30 households aren't independent from each other, and with a heterogeneous phenomenon like bombing, chances are that their effective sample size is then just a handful of interviews.
Tim: Some people with an agenda might get confused, as you suggest, but I don't think Shannon Love would.
And let me add, the third choice would automatically give a tighter range than the second or first.
Why? Because the worst that can happen is that everybody in Fallujah is dead, so we can't get a million.
From the data as is, we'd probably get something like a 95% confidence interval between 1,000 and 300,000 for Fallujah alone.
Heiko (and anyone else reading my last couple of posts): Apologies for the unnecesarily complex language. In an effort to say the same thing clearly, I've put the "pretty picture" up on my blog No de Qur'tuba, now a grand total of 7 hours old. I suspect that, in this case, the picture is worth the thousand words. Permalink for the post here. Incidentally, any comments on layout, content, or structure would be greatly appreciated, since I'm new at this (Except for comments about the small fontsize, of which I'm already aware. Currently trying to fix this).
Donald: Just realised I didn't answer a big part of your comment. I'll excuse myself on the grounds that (a) it's subtle, and (b) I'm not going to be very helpful. In short, I suspect you're right. Part of what you're talking about is what things might look like in other parts of Iraq. After all, the "West" has invaded twice, the Anfal campaign devastated huge parts of the country, and before that was the Iran-Iraq war. All those things are likely to have caused death rates to vary across the country, and not just around Falluja (viewed on a long enough time scale, we ought to model Halabja, for instance). The "ideal statistician" (i.e., someone with infinite time, patience, money, and data) would try to build a model of what the impact of each of those events did, which would also allow us to take a guess about unobserved parts of the country. Of course, the ideal statistician ain't been born, so the rest of us just have to cope. As it relates to the current debate, my guess is that the Anfal campaign left a large enough mark on the country that it would still be observable in the death rates in different regions. On the other hand, it was long enough ago that the difference between the Anfal-effect at time-pre-Gulf-War-II and time-post-Gulf-War-II must be pretty small, so I'd expect that they would cancel out in the Lancet study. Just a guess, though.
As far as I can tell, this is (more or less) how the study arrived at the 100,000 "additional" death figure (please feel free to correct me if I'm wrong or point out any flaws with this conclusion). I am almost positive that I am missing out on a few minor details or steps in the equation, but I think the general method I am presenting here to be accurate:
If someone provide me with the exact mathematical equation the study used to come up with the range 8,000 to 194,000 it would be much appreciated. Thanks.
---Ray D.
I would still like to know why they settled on or how they mathematically arrived at exactly "98,000" for the 32 clusters. If anyone can provide a detailed explanation, it would be much appreciated.
---Ray D.
Ray D., I don't think any such calculation was performed or could be performed without access to the raw data. The increase in risk is I am pretty sure not enough to know what the expanded CI [in terms of excess death] would be including Fallujah. I still think it is possible that the expanded interval includes zero, since the excluded outlier is 66 times the average size of the other samples. Note Dan's example, but try 33 data points where one is 66 times the average of the rest.
FYI fiddling around in excel it is easy to get a confidence interval 6-12 times wider with a 33-number data set matching those criteria.
Note that the calculation of a confidence interval in this manner implicitly assumes that all the observations are being drawn from a single underlying normal distribution, whereas the very existence of the Fallujah outlier is good evidence that they aren't being drawn from a single normal distribution.
whereas the very existence of the Fallujah outlier is good evidence that they aren't being drawn from a single normal distribution...
any use of the expression "confidence interval" or indeed any probabilistic measure needs to state these non-standard distributional assumptions to be meaningful (as they almost surely do not.) The authors had no problems referring us to a confidence interval for "risk of death" including falluja- why again is there no equivalent cited for excess deaths? This same confidence interval is being hauled out again and again to point to the "most likely" real number, and the percent chance that number is >40k, >100k etc etc etc when we all agree that normal assumptions are probably wrong. I am objecting to the inconsistency. It is not trivial precisely because the Falluja outlier is being used to color the notion of "likelihood" of the unaugmented central estimate ["probably more" etc. etc]
[btw I am not sure what makes 'excess deaths' more likely to be bimodal vs. skewed, sinc only 1 outlier was observed here with no 2nd cntral tendency, but I am happy to agree with you that 'normal' isn't correct.]
FYI I also asked Dr. Garfield via email what the confidence interval net of Falluja would look like, and he wrote back that it was never calculated "because it would be very wide."
dsquared says: "Note that the calculation of a confidence interval in this manner implicitly assumes that all the observations are being drawn from a single underlying normal distribution...."
Here you lose me. In my way of looking at things, since the deaths are simply non-random (because past) events, the randomness arises entirely in the sampling. Is it not true to say the distribution of the estimator is normal or something very like it?
This is pure curiosity on my part - if it turns on the great struggle between Bayesians and frequentists then never mind. I don't doubt that the Lancet referees know their stats, and in any case I regard the whole CI thing as a convention, a kind of homage to the Central Limit Theorem. The bottom line is, it beggars belief to suppose you would get a sample like that in a population whose mortality rate had not risen at least a little and probably a lot.
I have to love commenters who directly or obliquely claim I don't understand the math, but fail to even offer their own competing explanation. That's the difference between the 4th-grade schoolyard and Socratic debate. So, ironic ad hominems notwithstanding:
I tend to disagree that with the idea that the fact that the 50% mark may be around 100,000 means anything. There's a reason statisticians do the 95% confidence interval: to PROVE something. 50% is not proof. 95% generally is.
This study, despite methodology problems including not sampling the heaviest Shiite and Kurdish provinces that had the fewest fatalities and benefitted the most from Saddam's removal, could only prove 8,000 excess deaths.
And no, a longer time period does NOT necessarily mean more excess deaths, in fact more likely the opposite. Sanctions have ended, and there are huge ongoing infrastructure improvements in Iraq. Sadr City, for instance, no longer has raw sewage in its streets. Tell me that's not a health improvement that will save lives vs. pre-war.
When you look at the 2 million deaths DIRECTLY attributable to Saddam's regime, anti-war arguments for leaving Saddam in power based on the cost in human lives look pretty weak.
For the most complete list of infrastructure improvements in Iraq since the war, I recommend Chrenkoff's 21-part series. You may want to set aside a weekend if you're going to read the whole thing.
TallDave, you might want to go to Iraq and travel throughout the country spreading the word about how much better the infastructure is*. You see, Iraqis need to hear this from you since what they are experiencing is the exact opposite. After all, until you tell them, all they have to go by is their own lying eyes.
While you are there you might also tell them that what what may seem to them to be a significant increase in the death rate is really just an illusion. Again, right now all they have to go by is their own lying eyes, and it would be really reassuring for them to know that they are not suffering and dying nearly as much as they think they are.
*This claim reminds me of one of the best comic moments of the early days of the occupation, when Donald Rumsfeld made a televised speech to the Iraqi people while ensconced in one of the grand seats in one of the grand rooms of the Republican Palace (the message THAT sent was not lost on the Iraqi people, you can be sure). He spoke with a strange kind of folksiness, which played very strangely to Iraqis, introducing himself as "Don", and proceeded to inform the Iraqi people that they now had more electricity and more water than they did before they were "liberated". The problem was that almost no Iraqis saw the speech because there was no electricity to power TV sets. Services of all kinds have steadily deteriorated since then, and are worse now throughout the country than they have been since the first weeks after the 1990 war against Iraq.
Again, who are you going to believe, "Don", or your lying eyes?
TallDave,
Your charge of failure to provide a "competing explanation" is completely without foundation. Any number of people here, including myself, have addressed your misconceptions, often repeatedly, and often at great length. It is obvious from the fact that you continue to repeat the same fallacies you came here with that you are impervious to explanations that compete with what you choose to believe. There seems little point in repeating oneself ad nauseum to someone who appears deaf to what one has to say.
"There's a reason statisticians do the 95% confidence interval: to PROVE something."
More evidence of your lack of knowledge and understanding of statistics. Statisticians do not use 95% CI's to PROVE things. They use them to make predictions about populations based on samples from those populations.
The charges of selective undersampling are also false, as has been discussed here at some length.
And now we have the always predictable, "Iraqis are better off without Saddam" argument:
"Sanctions have ended"
And with them the oil for food program on which the majority of Iraqis were completely dependent for food, medicine, and other necessities. On top of that unemployment and underemployment are now in the region of 60-70%, and many Iraqi businesses have been destroyed or are suffering enormous losses, which means that more Iraqis have no income, and less ability to feed, house, and clothe their families than at any time in Iraq's history. We also have a massive number of homeless refugee families - around 200,000 people from Falluja alone - whose homes have been destroyed or made uninhabitable by American attacks, or bulldozed to the ground as a punishment for not cooperating with occupation forces.
And the sanctions also have been replaced by bombs, missiles, and bullets, skies filled with attack helicopters, and streets filled with heavily-armed, trigger-happy Americans who have no knowledge or respect for Iraqis, do not speak their language, and most of whom are scared, hate-filled, and/or drunk with absolute power, and who know they can act with near complete impunity. The country is filled with tanks and humvees whose crews have been trained to run over anyone - man, woman, child, elderly - who does not get out of their way fast enough, and to blow away any vehicle that gets too close. And of course, there are the checkpoints at which one false move can result in being sprayed with heavy machine gun fire.
And the sanctions have been replaced by a so-called "insurgency", which includes groups who target innocent Iraqis simply because of their religious affiliation. And they have been replaced by religious fanatics who harass and even kill women who do not dress in a certain way, or murder barbers for offering shaves, or giving the "wrong" kind of haircuts, and who operate freely throughout the country. In place of the sanctions we now have hordes of thieves, robbers, car jackers, kidnappers, and murderers-for-hire freely roaming the cities and the countryside.
"and there are huge ongoing infrastructure improvements in Iraq."
As I suggested above, why don't you go to Iraq and inform the Iraqi people of that. They will be very relieved to hear how much better things are than their lying eyes tell them.
"Sadr City, for instance, no longer has raw sewage in its streets."
The lying eyes of people living in Sadr City tell them a different story.
The living hell the U.S. has turned Iraq into makes life under sanctions look pretty attractive by comparison.
"When you look at the 2 million deaths DIRECTLY attributable to Saddam's regime..."
What is your source for this assertion, TallDave?
"anti-war arguments for leaving Saddam in power based on the cost in human lives look pretty weak."
On the contrary. War is considered by human rights and humanitarian professionals to be such an extreme measure that it can only be justified on humanitarian grounds if it is the only way to halt something that is 1) currently ongoing, and 2) rises to the level of genocide. It is debatable whether even Saddam's worst atrocities, such as the Anfal campaign, constituted genocide. If they did rise to the level of genocide the time to intercede was when they were going on, not decades after they were over and done with.
TallDave, what was the U.S. government doing about Saddam while the Anfal was ongoing? What did the U.S. government do about the torture and execution chambers, and the rape rooms in Abu Ghraib while they were being most actively used? What did the U.S. government do about the mass murder committed at Halabja in the months and years right after it happened?
What did the U.S. government do while the Shi`a and Kurdish rebels of 1991 were being slaughtered by the tens of thousands for eight straight days right under its nose? And what did it do while the regime followed up the eight days of slaughter of Shi`a rebels by committing mass summary executions of Shi`as, again right under its nose in the southern no-fly zone?
If the U.S. gives a damn about the Iraqi people, why did the U.S. wait to protect the people of Iraq from Saddam's atrocities until years after they needed protection? Why did they wait until years after Saddam had stopped committing such atrocities, and after a period of years during which his regime was killing fewer Iraqis than at any other time in its history?
Heiko, according to the Lancet authors there were several places in Iraq that were heavily bombed--the Fallujah cluster was the only one of their 33 which fell inside one of those towns. So Fallujah might not be unique--it might be one of several hard-hit areas.
To Dan and Heiko--I mentioned 1988 and 1991 because those are two years when the death toll was comparable to or possibly greater than what the Lancet study found for the post-invasion period. I'm not sure what Iraq was like in 1987 vs. 1988, but I think a study comparing 1990 to 1991 would probably have found roughly the kind of numbers the Lancet study showed (though probably somewhat worse). You'd see an enhanced violent death rate in many clusters, an enhanced nonviolent death rate (the US had destroyed the sewage system in the war) and maybe a cluster or two like Fallujah, where either Saddam had conducted a massacre putting down the Shiite or Kurdish revolts or (less likely) where the US had heavily bombed a neighborhood. In the actual study as opposed to my imaginary 1991 study, Heiko is suggesting that it was just a fluke that no outlier was found with 53 deaths in the pre-invasion period. I don't think that's correct. You might expect to see such an outlier in Iraq's most violent years, but from what I understand, 2002 was not one of those years.
I suppose I'm reasoning like an amateur nonquantitative Bayesian here, since I'm applying what I think I know about Iraq's recent history to argue against how Heiko suggests we interpret the Fallujah cluster. Saddam wasn't slaughtering massive numbers of his own people in 2002, so you shouldn't expect to see a Fallujah-sized outlier from the pre-invasion period and the existence of the Fallujah outlier therefore gives one no reason to think the invasion actually lowered death rates in Iraq compared to the previous year.
Donald. Seems like a good method to me. You'd have death-rate data for lots of clusters over lots of years, and could use them to compare your "informed model" to several different "uninformed" models. One could assume death-rates in all regions and all years follow the same distribution. Another would allow each region to have a different fixed (i.e., constant across years) distribution. Some models could allow for some years to be worse than others (across all regions), and others could allow all kinds of arbitrary variation. You could then throw in the theoretically-motivated "informed model" by making specific a priori predictions about some regions in some years (e.g., Halabja, 1998; Falluja, 2002), though you'd have to be careful to avoid looking at the new data before formulating the model. I think that a process like this would provide a pretty good test.
Donald,
you misunderstand me. I don't think we'd be likely to find many clusters in 2002 with a death rate of 200. I also don't think, though, that we are likely to find many places like Fallujah in the rest of Iraq.
So as far as I am concerned, it's option 3 we'd have to go for. If we do that, we haven't got enough data for Fallujah, and outside of Fallujah we've only got enough data to make (barely) a statement about overall death rates that holds some statistical significance, with the confidence intervals for subgroups (such as bombing deaths) far too wide to be meaningful.
What I was trying to explain was what options 1 and 2 would mean in simple language (ie why the confidence interval for option 1 would be something like -100,000 to + 500,000 and for option 2 something like 100,000 to a million), not that I agree that either was the best choice.
So, when you write:
"Heiko is suggesting that it was just a fluke that no outlier was found with 53 deaths in the pre-invasion period."
You are completely misinterpreting what I wanted to say.
I've said a lot already about how reliable the death rate estimates are in my mind and how they are composed, and how that matters. But I know it's a lot of work to follow all these threads, so let me summarise.
I think we've got insufficient evidence for non-violent death rate movements, and good evidence for a number of subcategories of violence, such as crime (10,000 extra deaths, I mostly base this on morgue data), bombing deaths and shootings to be blamed on the coalition (about 3000, with shootings edging out bombings, basis Human Rights Watch, Iraqi Body Count, media reports, blog accounts and a few other considerations), insurgents killed (of the order of 25,000, basis, military analyst estimates, kill ratio estimates, some information provided by coalition press releases) and innocent civilians and Iraqi forces killed by terrorists and insurgents (about 10,000, basis Iraqi Body Count database, and information released by the Iraqi Health Ministry and the Iraqi government).
I also agree that the numbers killed by Saddam in 2002 were most likely in the thousands, and relatively low compared to other periods.
So, that gives an estimate of 40,000-50,000 violent excess deaths, and some much less well known excess death (or lives saved) from non-violent causes, such as road traffic accidents and infant mortality.
But it's not appropriate to compare (as in morally equate, or partially equate) that with Saddam's killings. He targeted civilian Shiites and Kurds to the point of genocide. The victims of Anfal weren't "colateral damage", or the work of frightened soldiers at checkpoints, they were carefully planned by the regime with the express motive of killing those civilians.
What we need to compare the "death toll" with, and we need to keep the composition of that death toll in mind for that, is what would have happened, over the past two years and the next few decades, if Saddam had been dealt with in another way.
Heiko,
The statement "outside of Fallujah we've only got enough data to make (barely) a statement about overall death rates that holds some statistical significance" is simply false. Since that has been pointed out often enough I won't elaborate.
In general, you ask for extraordinary standards of rigour from others while you yourself play fast and loose with the evidence. For example, I don't quarrel with the description of Anfal as genocide, but just imagine how a Baathist Heiko would spin it. How many died? How do you arrive at that number? What's the confidence interval? How many were rebels?
At the risk of being rude, I think it would do you good to be locked in a room with Riverbend so that you could see how hard it is to hold a rational discussion with someone who is so incapable of acknowledging the opposing view. She has a pretty good excuse: her country invaded, friends tortured or killed, her career prospects ruined by the fundamentalists who are now set to control her life. I can understand why she is unable to tke a more balanced view. But how you got this way is beyond me.
The ChigagoBoyz/Shannon Love posts got me interested in the Roberts paper again; I've made some relevant comments over there but won't repeat them here, especially with this intermittent dialup connection. I only just discovered this active thread (duh). The technical/statistical points and counterpoints being made here are interesting; thanks to the statistically-capable for the pickup lessons.
A couple of things:
--How much more straightforward the statistical debate would be if Roberts had been required to post finer-grained summaries of their collated data as Supplemental Online Material, as is now routine at many journals.
-- In discussing the Fallujah outlier, Kevin Donoghue gave useful "imagine 10,000 towns on a grid" example earlier, at 26/3 08:49. An extension of this analogy explains why I think Roberts' handling of that 33rd cluster is embarrassing, and thus likely honest. Unlike the truly "wild data" outliers discussed so elegantly in the William Kruskal essay cited at CB by dsquared (sorry, I can't link from here), Fallujah is an "outlier" only in that it is not part of some quasi-normal mortality distribution. But since Roberts did this research for the very purpose of studying postwar mortality, and since they, like us, read the papers, they had every reason to anticipate a priori that Fallujah's death-by-violence mortality would be very, very high. They didn't allow for that in their study design (e.g. by geographical stratification). Perhaps they didn't think about it, perhaps they were operating under KISS rules, perhaps they figured that the chance of one of only 33 chosen clusters landing in Fallujah was negligibly small. In any case, they got "unlucky," from a statistical-analysis point of view. I'm claiming that this is embarrassing because it's a result of an inadequate study design. I take it as a sign of honesty because there's no evidence of fudging or massaging of the primary data. (Their description of the results of the analysis of that data is another story--we disagree on that).
--A descriptor was mentioned over at CB that allows an estimate to be made of what the size a non-clustered study would be that has equivalent power to the cluster study being analyzed. That number was, from memory, given by Roberts as about 2 for the 32-cluster data set, and about 26 for the 33-cluster set. Meaning that the heterogeneity introduced by Fallujah mortality caused about a 13-fold decline in the power of the data set, for this reason. Would one of the statistically-knowledgable people here be able to provide some background or comment on this point? Thanks.
Heiko, you said:
Sorry if I missed the thread or post in question, but with whom do you agree on the number being "in the thousands"? Certainly not with Amnesty International, or (though they don't discuss numbers per se) with Human Rights Watch. If there are more authoritative sources, I'd very much like to know who they are.
AMac, probably all you'd need to know on the subject of your last question can be found in this excellent online discussion from MORI: Cluster Sampling: A False Economy?
To see how it applies to the Lancet survey you'd also need a copy of that, of course; and to search within it for the term design effect.
AMac,
Whenever I see the phrase "inadequate study design" the question that springs to mind is, what would you have done?
Of course it is true that in a situation like Iraq there are going to be small pockets of exceptionally high mortality while the rest of the country mostly suffers only secondary effects from the conflict. But what is a researcher supposed to do about that within the obvious constraints of cost and physical security? That's the question that I was trying to get Telluride to address with my numerical example, but he either overlooked it or ducked it.
Sometimes there are no good solutions. I have thought quite a bit about other ways the survey might have been done and I really can't think of any method which would be an improvement. Oversampling the hot-spots has been suggested but that obviously increases the physical risks.
This is not aimed at you, since you evidently respect Roberts et al more than S. Love does, but ironically the "design" critique mostly comes from people who resist the notion that the Lancet estimates are more likely to be too low than too high - yet that is the clear implication of that argument.
Thanks for the link, JoT. I'll read it with my copy of Roberts to hand.
K. Donoghue,
> "inadequate study design" / what would you have done?
Fair question. I've thought about that too; it's a question a peer-reviewer asks (ought to ask) when looking at a manuscript with significant limitations or problems. "Roberts should have put their Iraqi colleagues/employees at ten times the risk by sampling ten clusters in Fallujah" doesn't strike me as a good response. My short answer is that the study is what it is, and should be presented as such, and hopefully interpreted as such by the wider scientific community. I will try to be more specific on this point at a later time, when my 'net access isn't so iffy.
Amac: I think that there's actually much more to be said for the study design than you do. Outliers are outliers; they're not the spawn of Satan, and IMO the decision to use 33 clusters is basically an indication that they wanted to use 30 clusters and planned to have to throw three away.
As a defender of the study, I think my job would have been much, much harder if they'd used geographically stratified sampling; it would have been very hard to come up with an objective and defensible way of defining a "high-violence" area. In general, all the problems with this survey appear to be to do with the use of extrapolated deaths numbers; it's true that these extrapolated numbers rely on very strong distributional assumptions not really supported by the data, but if you're gonna hang people for that, you'll need a lot of rope because it is absolutely universal practice in the epidemology literature (score one point for econometrics versus epidemiology in my book).
Hi JoT,
What Donald said was:
"Saddam wasn't slaughtering massive numbers of his own people in 2002"
which I agree with.
The question of whether killings by state associated security forces in 2002 were in the range 1000-10,000 or in the range 100-1,000 was not addressed by Donald, and I was clearly sloppy in my wording.
It has been discussed before, though, on one of these Lancet threads (and I can't currently locate the discussion).
And the only study dealing with the matter we could find was:
http://www.phrusa.org/research/iraq/release_032304c.pdf
And while the report says that nearly half of the surveyed households reported at least 1 serious human rights abuse (which included killings and disappearances) between 91 and 2003 and 30% of those abuses occurred between 2000 and 2003, it doesn't split the numbers down sufficiently to be sure of the rate of killing in 2002, though the numbers to me appear consistent with continued levels of killings in the thousands.
Human Rights Watch doesn't take a position on how low the number of killings were in 2002.
The link to AI doesn't work, I've looked at their press releases in the past though, and I don't think they've done field work to measure the level. What they've got is individual reports that make it clear it was at a minimum in the hundreds. At least that's my understanding.
At any rate, the level of violent death (however inflicted) was clearly a fraction in 2002 of what it was between March 2003 and now, and small enough to be unlikely to substantially affect the excess death estimate.
The individual groupings of human rights abuses in the study were (out of about a 1000):
Torture 212
Killings 210
Separation and disappearance 166
Forced conscription 136
Beating 131
Gunshot wounds 64
Kidnapped 36
Held hostage 18
And a few others with lower numbers.
And 95% of abuse where the perpetrator was mentioned was done by Baath regime associated groups.
Thanks for the reply, Heiko. Sorry about the AI link - another bit of bum html code from yours truly. Here it is again: http://web.amnesty.org/report2003/Irq-summary-eng
Hi Kevin,
without "elaboration" I've got no idea what explanation you are referring to. Don't you accept that the data for the subgroups will have confidence intervals like 1,000 - 26,000 for say civilians killed by coalition forces?
Where do I ask for "extraordinary" standards of rigeur or play "fast and lose" with the evidence?
If a Baathist Kevin asked me to show that Anfal was genocide I'd point to captured documents by the bucketload that unequivocably prove that the regime intended and executed genocide, and used, among other means, "special" weapons to do so, and that the euphemism "special" can very clearly be proven, in numerous ways, to refer to chemical weapons. I'd refer to thousands of witness accounts, and mass graves that have been undug. There may be uncertainty about the exact numbers killed, and some of those were indeed "rebels", but there is no scope for claiming that this wasn't a premeditated campaign of mass killing of civilians for the purpose of ethnic cleansing.
(to provide a link, I've read about the captured documents in great detail elsewhere)
http://www.indict.org.uk/crimedetails.php?crime=Anfal
Riverbend is quite exceptional among Iraqi bloggers, Sunni, moderately pro-Baathist and pro-insurgency, secular and the content of the blog is tailored towards a radical leftwing audience. She's probably also a frightened young woman in a war zone, and clearly has many genuine grievances.
I am following many bloggers in Iraq, opinions there are very diverse, and I am certainly keen on trying to see your point of view, as well.
AMac,
thinking this through, they could have done individual rather than cluster sampling for Fallujah, ask 30 non clustered individuals, or if that's too much of a safety concern, 15, it would still give a better estimate than a single cluster (I suspect their one cluster of 30 people might be about the same as sampling 3 or 4 random individuals).
Hi JoT,
ok, here they say "scores of people", and because my English is too weak I had to actually look that up to be sure what it means (occasionally I still get tripped up as a non-native speaker, even after ten years in England). My dictionary says "scores of people" means "large numbers".
They also give a few examples. I don't think they take a position here on whether the actual number is in the range 0-100, 100-1,000 or 1,000 to 10,000?
Heiko,
A "score" in numerical terms is 20, therefore, I would interpret "scores" as at least 40, and less than 200. I say less than 200, because when it reaches 200, it should become hundreds, not scores. In any case, it sounds like AI is suggesting that the number was, at most, hundreds, not thousands or tens of thousands.
The Human Rights Watch report JoT linked to is, of course, not a report on human rights abuses of Saddam Hussein. It lays out the reasons that the invasion and occupation of Iraq was not justified on humanitarian grounds (which was an afterthought anyway, to cover for the fact that the WMD and terrorism excuses were becoming thinner by the day). It does not, to the best of my recollection, include any kind of information about violent deaths at the hands of the regime.
I don't know where the notion comes from that Riverbend is pro-Ba`thist, except of course for the fact that this is the almost universal knee-jerk response by war cheerleaders to any Iraqi who refuses to pretend being attacked, shocked and awed by tens of thousands of bombs, brutally and repressively occupied, and forcibly "reconstructed" (read deconstructed and transformed) by a foreign power is any kind of improvement over what they had to deal with before. I happen to know Riverbend personally, and pro-Ba`thist is not a term that comes to mind. I also don't know what being Sunni (or Shi`a or Kurd, or Christian, or Yezidi, or Turkmen, or any other such classification) has to do with it, except in the minds of those who have bought into the popular myth that Iraqis are incapable of thinking or forming viewpoints about their situation independently of their religion/ethnicity. (How you fit the many, many Iraqis of ethnically and religiously mixed background into that model I don't know.)
Heiko
A score is an almost-obsolete English expression meaning "twenty". "Scores" typically means something like "more than twenty but less than one hundred."
In the context of the HRW report on Saddam-era Iraq, the term is used to refer to the confirmed killings of specific named individuals. It seems reasonable to assume that the actual number of killings by the Baathists was substantially higher.
<body>Heiko: a "score" means "a set of 20" ("20 Stück"). It's a term used in much the same way as "dozens". While "scores" (like "dozens") can in colloquial usage mean "many", it's not generally used where "hundreds" is meant, let alone "thousands". Compare AI's report for 2000, which opens with the line:
to its 2002 equivalent which reads:
AI's 2001 report also uses "scores":
These AI figures may be inexact, but when they mean to say "hundreds" they say it.
Ian: in the UK, at least, "scores" is far from obsolete. "A brace" - now, that is almost obsolete!
All right, after all that help I think I've learnt another word in the English language.
I still think though, as apparently does Ian Gould, that AI likely refers to named, or at least known cases, but doesn't attempt to estimate the true figure.
I am not sure what JoT thinks about this?
Heiko you'd have to check with AI to be certain about this, but my assumption is that it's a mix of detailed case histories which included names and some less well-documented cases that didn't. It seems unlikely that AI had the names and details for "hundreds" of deaths for the years in which they used that terminology, so I think for both "hundreds" and "scores" they were making (quite likely lower-bound) estimates for executions they felt fairly certain had taken place. Of course Amnesty is opposed to all forms of capital punishment so it's not only political executions they would have had in mind.
Ok, I agree with that, and I was aware that AI is opposed to capital punishment in all cases.
(Sorry--long comment)
Here is a more complete answer to Kevin Donoghueâs question to me of 28/3 02:19, "what would you have done?" Iâve addressed it in the form of "If I had been asked to peer-review the manuscript." (In the form that it ultimately appeared, obviously.) Actually, thereâs a subtle distinction: "If I had been a reviewer, and had known what I know now as a result of online discussions." A review along the lines that follow would be more wide-ranging than a typical peer-review, where a scientist knowledgeable in the area of the manuscript takes a few hours, reads carefully, and comments. Everybody involved in the peer-review process knows that peer-review misses a lot. The hope is that it picks up the most important flaws, and the reality is that itâs better than what came before. Here I outline the main issues, my actual review would be about three times as long, and cite specifics.
In conclusion, if I had been a reviewer whose advice had been followed, the Roberts study would have been published in the Lancet, perhaps on the same schedule. Working from the same data set, the message imparted by such an article to non-statisticians would have been very different from that delivered by the actual paper.
For those who argue that the Roberts article, as published, was not easily misinterpreted, here is the opening of the Invited Comment that was published alongside it (Bushra Ibrahim Al-Rubeyi, "Mortality before and after the invasion of Iraq in 2003," Lancet v. 364, # 9448, p. 1834, 20 Nov 2004):
If you believe that Dr. Al-Rubeyiâs conclusions are justifed by Robertsâ paper, I would be grateful if you would add a comment explaining why.
Hi Amac,
I think it is entirely reasonable to treat Fallujah separately. I don't accept dsquared's point regarding the difficulty of defining "high violence areas".
Fallujah was treated differently, in fact, it wasn't sampled the same way as the rest, because it wasn't under government control. Eg there was the issue of using GPS.
Therefore, and as it was known a priori that Fallujah was the single most violent town in all of Iraq, I think they should have treated it as an entirely different population representative only of Fallujah, and not of "high violence areas" in Iraq, as dsquared implicitly does, when he makes comments about how high the rate in Najaf or Sadr city might be, none of which were fully under insurgent control for months, requiring a conventional military invasion in an urban environment, as was the case with Fallujah.
Treating Fallujah as an "outlier", which gets excluded from explicit statistical calculations, but then brought back in to make (statistically not supported) qualitative statements about the data is unacceptable.
If they did treat Fallujah separately and presented the calculations, I fully expect the confidence interval to be something like 1,000 to 300,000 (which in itself assumes that they weren't lied to, or that the underlying population figures can be applied, abandoned houses may be an indication of people killed, or of people having fled).
By not calculating this figure, they can sneak Fallujah back in to make qualitative statements that are in fact not supported by any hard data.
In summary, I would not delete your point 3 and make a suggestion, such as:
The authors could have sampled 30 randomly chosen households in Fallujah, and also drawn upon other estimates for this particular city, such as those claimed by the insurgents themselves or of Iraqi Body Count. Using a single cluster and assuming that this cluster represented a 33rd of Iraq's population was clearly not appropriate. Falluja's population at the time could be estimated to be between 100,000 and 200,000, and a single cluster with a heterogeneous phenomenon such as bombing could be expected to yield much less information than random sampling of individuals.
Sorry, your point 2, not your point 3.
Anytime someone says "Excel" you know that they don't know or care to learn about statistics.
Amac,
Thanks for the response to my question. My main interest is in the criticisms of the methodology and I donât really want to get into the presentation of results, but since that is obviously important to you letâs explore it.
To begin with a point on which we agree: you would like more of the cluster-level data so that we can see how the aggregate numbers are arrived at. I would go further: I would like that done with every worthwhile piece of research I have ever seen! It would be an enormous help to students to see how the numbers are crunched. When I was learning econometrics I could have done with more of that and less waffle about distributed lags. But thatâs a gripe with the world at large, not just the Lancet. And given that standards of disclosure are what they are, there is no justification for raising the bar for Roberts et al.
As to the layout of the report, I would have preferred to see first the cold facts about deaths in the reported clusters, then the statistical analysis concluding with the extrapolations from mortality rates to excess deaths. But editors shouldnât tell authors how to do these things. You say that if the authors insisted on the version we see, publication should have been refused. I strongly disagree. In essence your objection is that the authorsâ politics are showing. I grant you that, arguendo, and ask: so what? It is not unusual. Did a Milton Friedman paper ever fail to express his distaste for socialism?
You say the text is âmisleading on key points.â Certainly your quotation from Dr Al-Rubeyi suggests he read what he wanted between the lines. However, the Economist newspaper managed to produce a pretty good summary. That may reflect the fact that the Economist supported the invasion so there was an incentive to read the report critically. There is nothing to prevent intelligent readers with similar politics from doing the same thing. Sadly, all too many of them prefer to denounce it first and read it later, or not at all.
Your criticism (5) of the way pre- and post-war mortality estimates are reported is unfounded, unless recall bias pushes down the pre-war figures only. I havenât seen a convincing argument that it should. People are not bad at recalling deaths in the family as a rule. Since this is a longitudinal study there is no real need to reconcile the pre-war data with figures from other sources. It would be nice of course but there was no need for the authors to put out warning flags.
Shirin,
Where can I start on how wrong you are? First off, Iraqis don't need me to explain things are getting better: they say they are by 62% to 23%. Lying eyes, indeed. Are you claiming all these infrastructure improvements are made up? Or are you arguing they like dying more?
http://story.news.yahoo.com/news?tmpl=story&cid=676&e=1&u=/usatoday/200…
The rest of your commentary is similarly uninformed.
"Sanctions have ended"
And with them the oil for food program on which the majority of Iraqis were completely dependent for food, medicine, and other necessities.
Uhhhhh... no. The rations were not ended, and Iraq is still selling oil. I'm sorry, but if you don't even know the basics of the Iraq situation, I can't really be bothered to take your comments seriously.
If the U.S. gives a damn about the Iraqi people, why did the U.S. wait to protect the people of Iraq from Saddam's atrocities until years after they needed protection? -- and so forth in that vein
Probably mostly because the international community was so against it. Was that wrong? I've thought so for 14 years, and I think I've been thoroughly vindicated in that view. But you would have opposed it then too, so again what's your point?
"There's a reason statisticians do the 95% confidence interval: to PROVE something."
More evidence of your lack of knowledge and understanding of statistics. Statisticians do not use 95% CI's to PROVE things. They use them to make predictions about populations based on samples from those populations.
ROTFLMAO!!! More evidence you're not worth taking seriously. I have to love it when people say the same thing I said in different words, then try to claim I'm wrong on that basis. Or are you arguing there's no difference in reliability of prediction between a 95% interval and a 50% interval?
So, once again we are left with a study that only proves 8,000 people were killed, and even that left out sampling the most heavily Shiite and Kurdish areas, as the authors themselves admit.
Thanks for the feedback, Kevin...we disagree, obviously. Any comment on Point 4, requiring statistical treatment of data disaggregated by cause-of-death?
AMac,
AFAIC the more disclosure the better, but again I don't fault the Lancet for agreeing to publish without it.
Talldave, people are trying to have an argument here that rises a little above the ROTFLMAO level.
To Heiko--Sorry I misunderstood your point. If I understand you correctly, you agree that the number of violent deaths is probably tens of thousands higher in the year or so post-invasion compared to the pre-invasion period. Where we disagree is that you are certain that most of these violent deaths are caused by the insurgents, whereas I think there's a good chance (I'm not certain, of course) that a large fraction (quite possibly the majority) are caused the US.
I have a couple of subjective reasons for thinking this. First, governments engaged in counterinsurgency wars generally kill at least as many civilians as insurgents. Israel, for instance, has probably killed more civilians than actual terrorists in the last few years. Governments usually lie about this if they can. The US seems to be acting like a typically brutal government in this situation, as the widespread torture scandal shows. If the US has killed tens of thousands of insurgents, there's a good chance it's killed comparable numbers of civilians and it probably identifies all the dead as insurgents, as sometimes occurred in Vietnam.
My other point is that Fallujah may not be unique. I don't have it in front of me, but I think the Lancet paper mentions that the neighborhood they sampled in Fallujah didn't look any more devastated than many other neighborhoods they saw not only in Fallujah, but in the other high violence places. You can't tell much from the fact that one extremely devastated neighborhood showed up in a sample of 33, but I'm not sure how far one can go with this. If the samples were randomly taken from all parts of Iraq, I think one could argue that it's likely such neighborhoods show up with a frequency of one in several dozen, but since this one had to come from Anbar Province (I think), I'm not sure what one can argue. But there were other towns with devastated areas, I think.
As for the nonviolent death toll, I was surprised it wasn't higher. In 1991 the US deliberately destroyed some of Iraq's civilian infrastructure to put pressure on Saddam, figuring that increased civilian misery might even lead to Saddam's toppling. This was admitted by Pentagon targeting planners to Barton Gellman in the Washington Post in an article dated (I think) June 23, 1991. This time it's the insurgents doing the same thing for the same reason, and the US has been less successful putting things back together. So one would expect the sewage in the street problem to be causing increased death rates--there was also a story last year that said malnutrition had gone up since the war.
Donald Johnson,
Talldave, people are trying to have an argument here that rises a little above the ROTFLMAO level.
Tell me about it, I'm one of them. Did you have a point?
Donald,
Also, some of us are trying to have an argument that rises above the The US seems to be acting like a typically brutal government in this situation, as the widespread torture scandal shows. level.
Anytime someone says "Excel" you know that they don't know or care to learn about statistics.
This is a needless slur; I know a lot about statistics and Excel is a fine program. You can't do everything in it, but you can't do everything with R either.
Amac: just one point. When you say
For the total figure and for subsets A, B, and C suggested immediately above (not an issue for subset D), excess deaths are the subtraction of a pre-war estimate from a post-war estimate, where each estimate has a large uncertainty. Figures so calculated can have particularly large uncertainties associated with them
then that's not quite 100%. The calculation of the excess deaths is based on the pre- and post-war death rates, but it's not a simple subtraction. You have to take into account the fact that this is a cohort study; it's not two independently calculated death rates from different samples but the death rates in the same sample at two different times. (These clearly aren't independent; people who died in the prewar period can't die again postwar). This means that the uncertainty in the relative risk rate calculation is a little bit better than a simple subtraction would imply. Small beer I know, but I bring it up to illustrate the fact that this is the sort of thing I'd expect a Lancet reader to have off pat.
TallDave: as I've said a few times while discussing this study, it really really is a good idea to read things before citing them. The Yahoo story you linked to in support of the contention that "62% of Iraqis think that their country is heading in the right direction" referred to a survey carried out in February of this year. The article also contains the following passage:
In September, 45% of Iraqis thought the country was headed in the wrong direction and 42% thought it was headed in the right direction
Can you guess which month of 2004 the Lancet team were carrying out their fieldwork? I'll give you a clue; it begins with "S" and marks the start of the oyster season.
dsquared,
It really really is a good idea to read my posts before criticizing them. I was arguing Shirin's point of whether Iraqis are better off now.
dsquared,
Shirin said TallDave, you might want to go to Iraq and travel throughout the country spreading the word about how much better the infastructure is*
Note his use of the present tense.
As I've been saying for awhile, lets agree on specifications and methodology for a new study which we can all agree is fair before it is done. Then all claims of bias and bad methodology and flawed sampling can be thrown out the window, and we can get some real factual perspective about what the impact of the war has been. Naturally such a study would include the most current data.
Talldave, if you don't know that counterinsurgency campaigns are typically brutal you need to read more. Even democratic governments are usually fight these things in a criminally brutal manner. Britain was bad in Kenya in the 50's (there are a couple of recent books about this), France was bad in Algeria, and the US was bad in the Philippines and Vietnam. And democratic governments usually lie about their human rights violations too--if you relied on French or the US governments to tell about the civilian casualties they inflicted you'd wait a pretty long time, but of course both were more than happy to publicize the atrocities committed by their enemies.. The US torture scandal is not really out of the ordinary compared to the past history of this kind of war and the burden of proof should be on the person who thinks we're being extraordinarily careful about human rights given both past history and what we know has been happening with respect to torture.
One other comment to Heiko--if you stick to the non-Fallujah data there were 21 violent deaths, which extrapolates to around 60,000. I think the breakdown was 12 due to Iraqi criminals and insurgents and 9 from Americans, and if I continue to do the naive extrapolating, that's 27,000 dead from Americans and 36,000 from Iraqis. I realize the error bars are huge, but the number you'd get for Iraqi murders and terrorist attacks seems in the right ballpark. So why would the number for deaths caused by Americans be far too high? It seems odd that the survey would give a believable figure for the criminal and insurgent killings and a wildly inflated one for the American-inflicted deaths. And again, we're leaving out the heavily bombed areas.
1) No: âFiskingâ is" NOT "a term bloggers use for especially lame posts."
"âFISKING: Three people asked what âgroup-Fiskingâ means in this post, which borrows the term from an InstaPundit post.
The term refers to Robert Fisk, a journalist who wrote some rather foolish anti-war stuff, and who in particular wrote a story in which he (1) recounted how he was beaten by some anti-American Afghan refugees, and (2) thought they were morally right for doing so. Hence many pro-war blogsâmost famously, InstaPunditâoften use the term âFiskingâ figuratively to mean a thorough and forceful verbal beating..."
2) What's the point ofd jointly designing a good study? The Americans don't want a good study and wouldn't allow one. If the occupying forces wanted to know how many people had died, it could go and examine all the actual death certificates in all the towns - wouldn't take long (andif the American Army can't handle security, that says something too). It won't because it doesn't want to know, and the fact that it hasn't proves that it doesn't want to know; which strongly suggests that its own estimates are worse than the Lancet figures.
It seems to me that Chris Borthwick and others are falling into the "you can accuratly count all the people in a country" fallacy. Not unless everyone stands still for a couple of weeks you can't and even then you probably can't. Same for deaths, including the factor that there is not always a death certificate.
"Robert Fisk, a journalist who wrote some rather foolish anti-war stuff,"
You can call what Robert Fisk writes foolish if you like, but he is one of the few western journalists covering the Middle East who actually knows something about the region he covers. In particular, he is one of the few journalists covering Iraq who 1) speaks, reads, and writes Arabic fluently, 2) knows and understands Middle Eastern history, culture, and politics, 3) doesn't spend most or all of his time holed up in the green zone getting all his information from U.S. military news briefings, talking to people working for the occupation. He has also turned out to be right far, far, far more often than he has been wrong.
"and who in particular wrote a story in which he (1) recounted how he was beaten by some anti-American Afghan refugees, and (2) thought they were morally right for doing so"
Utter rubbish. He did not say they were morally right, he said he was able to understand why they did what they did, and that in their place he would most likely have done the same thing. You might disagree with Robert Fisk, but at least don't misrepresent what he says, please.
âIraqis don't need me to explain things are getting better: they say they are by 62% to 23%.â
No, TallDave. You have, no doubt unintentionally, misrepresented what the article says. The question was not about current conditions on the ground. It was also not about whether there has been any improvement in conditions on the ground. The question was about something altogether different - whether or not the country is headed in the right direction. It does not, of course, elaborate on why or how the respondents see it as headed in the right direction, but you can bet it has everything to do with hope about the political situation, and nothing whatsoever to do with improvements in security or living conditions, both of which remain ghastly.
âAre you claiming all these infrastructure improvements are made up?â
TallDave, we have been hearing ad nauseum from propagandists and war cheerleaders about massive rebuilding, and infrastructure improvements since April, 2003. None of them has held up under any kind of examination. With a very few minor very localized exceptions Iraqis have seen no improvement in their living conditions, and continued deterioration overall. Pardon me if I do not get to excited about further such claims, particularly coming from the likes of Chrenkov.
âOr are you arguing they like dying more? â
Complete non sequitur. I thought you were arguing that massive infrastructure improvements have made great improvements in daily life.
Iraqis are still dying and being maimed from violence and other war-related causes at a very high rate, although the rate does seem significantly lower at the moment. Based on daily reports so-called "insurgents" and common criminals are killing about the same number of Iraqis as usual, but for the present U.S. forces do not appear to be bombing cities at the same rate, which lowers the death rate considerably.
" The rations were not endedâ¦â
You are technically right, and I misspoke. Even so, ending the sanctions is only a benefit if you do not replace them with something as bad or worse. As a direct result of the Bush administrationâs invasion and occupation of the country more Iraqis have less ability to feed, house, and clothe their families than at any time in Iraqâs history. Child malnutrition rates, which had been on a downward trend since 1996, and by 2002 were very close to pre-sanctions rates, doubled in the first year and half or so of occupation, and appear to have continued to rise. Rates of diseases caused by lack of clean water and adequate sanitation continue to increase. Medical workers report that patients often leave hospitals sicker than they arrived because they contract new diseases there.
âand Iraq is still selling oil.â
How much oil is Iraq selling now? Where is the money from that oil going? Whom is it benefiting?
âare you arguing there's no difference in reliability of prediction between a 95% interval and a 50% interval? â
What I am suggesting, TallDave, is that after studying elementary statistics for only a few weeks, anyone should understand that statistics are not used to PROVE things. They should also know that 50% intervals do not provide more certainty than 95% intervals, and that they are virtually never used. (Perhaps there is some application for a 50% CI, and if so I hope one of the scientists here will enlighten me, but I cannot recall ever seeing one outside of an elementary statistics textbook.)
âSo, once again we are left with a study that only proves 8,000 people were killedâ
Once again, you repeat your mistake, thus showing your lack of understanding of the subject you are discussion. Statistical studies based on sampling do not PROVE things. What they do is make predictions about populations based on studying a random sample. They deal in probabilities, not in proof.
One more thing, TallDave. The poll you cited comes from the International Republican Institute, which deceptively represents itself to the press (which shamefully does not fact-check the claim) as "a non-partisan, U.S. taxpayer-funded group". A look at the board of directors roster indicates, and further research confirms that it is, in fact, a distinctly right wing republican organization with strong neocon ties, and a neocon-like agenda. Most of the board members have been affiliated with one or more of the Reagan, Bush I, or Bush II administrations, and a some are very close indeed to George W.
All of this suggests, of course, that their activities are strongly agenda-driven, and they cannot be trusted to be objective with respect to Iraq. The fact that they misrepresent themselves as non-partisan suggests than they cannot be trusted to be honest, either. For this reason, before taking any of their polls very seriously I would want to examine their methodology very closely (including the questions they asked and the way they were worded, and the characteristics of their samples), and compare their results to other polls taken at about the same time. My guess is that while the results might not be contradictory, there is a likelihood they would be significantly different.
Hi Donald,
with some engagements one hears one story from the US military (we killed 45 insurgents) and another from insurgent friendly stringers (no insurgents killed, but 9 innocent civilians).
I think it's virtually impossible to clear that sort of thing up without very detailed and lengthy investigations (ie I think it'll take years for the new Iraqi government to estimate good figures based on investigating all individual cases, assuming they aren't too busy fighting a civil war, or are led by another dictator who couldn't care less and would just publish an estimate without any underlying work), which is a major reason why the US military doesn't estimate insurgent or civilian casualties. Because, when challenged that their estimate for insurgents is too high and their estimate for civilians too low, they'd have to admit that their method of counting is too crude to be sure, and that would be spun into an admission of lying.
I suppose it's a damned if you do, and damned if you don't thing, just like trading with Saddam. If you do, you "support" him and his killing, if you don't, you are killing innocent Iraqi children. And in both cases, the blame for any and all killing will get squarely laid at the feet of the US government.
You wonder why I distrust the figure for civilians killed by Americans produced by the Lancet study more than the figure for violent death overall. The answer for that is that all women and children claimed to have been killed by coalition forces are due to just 2 accounts, while I think the 12 non-coalition deaths are based on 1 report each. That gives less scope for error, be it because somebody confused an insurgent mortar with a coalition bomb, or because the Lancet team picked up a real, but rare case, which if it involves 3 women and children has a disproportionate impact.
If one did try to do a confidence interval, with 12 different deaths, largely in different governorates, one would get something much more narrow than with 2 bombing incidents, with both of those in one governorate (Thiqar I think).
The main reason I believe that most killed are either insurgents, innocents killed by insurgents, criminals or victims of crime is that Iraq has a free press with stringers (or bloggers, Najma has done a good job for Mosul there) bound to report major massacres, and that together with the evidence from morgues and government figures from hospitals, tens of thousands of civilians just don't add up for me.
That's not to say that I exclude the possibility. Najma reports that obtaining a death certificate in Mosul is more difficult than it used to be. There may be some high violence areas where people shot would be buried without death certificates, and the deaths may be so dispersed that stringers wouldn't have a good enough story to sell ... (though in the deaths that are reported coalition forces are not the most frequent target of blame, and recently it's running at more than 90% of reported deaths from military and terrorist activity being due to suicide bombings, beheadings and the like, just have a look at the Iraqi Body Count database)
I think it's possible, though on present evidence, unlikely. But I also think the Lancet study adds little additional evidence to suggest tens of thousands, because it's got one cluster in Fallujah with high claimed and likely entirely unverified deaths, and two accounts of women and children dying from coalition air strikes in Thi Qar and 4 accounts of individual men being killed by coalition forces, where the status is either entirely unclear, or where the family claims that they were not combatants (I think in two cases the story was told in more detail and sounded believable).
dsquared (29/3 09:16):
Thanks for sticking up for me re. Excel (âneedless slurâ), but I was, alas, unclear. What I actually meant was merely that Excelâs â.xlsâ format is cleaner in its presentation of tabular data than â.csvâ or â.txt,â and is widely read, by WordPerfect, SPSS, and beyond.
As far as the cohort-study nature affecting how excess deaths are calculated (âit's not a simple subtraction⦠the relative risk rate calculation is a little bit better than a simple subtraction would imply.â): Sounds like a solid point to me. Iâll persist in my doubts that the Lancet reader has such distinctions down pat. FWIW. And I still miss the calculations of excess-deaths-by-cause that arenât in the paper.
"one hears one story from the US military (we killed 45 insurgents) and another from insurgent friendly stringers (no insurgents killed, but 9 innocent civilians)."
So, any "stringer" whose reports do not coincide with those of the US military is, by definition, "insurgent (sic) friendly".
Enough said.
Chris Borthwick, I am familiar with the origin of the term "fisking". My point was that "fiskings" tend to be lame posts. Occasionally someone will do a good one, but that's like being a tall dwarf -- they are only good in comparison with the generaly low quality of fiskings.
Shirin,
Well, I have to give some credit for bowing to reality on the fact oil is being sold and rations given out.
But this is a howler:
"What I am suggesting, TallDave, is that after studying elementary statistics for only a few weeks, anyone should understand that statistics are not used to PROVE things"
Oh, so statistical analysis is done strictly for fun and is never meant to mean or prove anything. OK, so factoring in this somewhat insane premise, we're left with either a study that proves 8,000 people were killed, or a study that proves nothing because statistics never prove anything.
Regarding your IRI comment: If there was sampling bias, it would still represent a big change from polls done earlier by the same organzation. Also, does the word "irony" mean anything to you? Or do you think the Lancet authors (who insisted on publishing the study before the election) were big Bush supporters?
Donald,
"Talldave, if you don't know that counterinsurgency campaigns are typically brutal you need to read more."
Of course a counterinsurgency is brutal; they're not going to give up arms because you offer them lollipops. Please don't attribute your silly strawman argument to me.
Donald, if you don't think most prisons in America have comparable rates of abuse to Abu Ghraib, then you need to read more. If you think America has a policy approving torture of POWs, you need a swift kick in the butt from a U.S. Marine sergeant. The prison abuse scandal doesn't prove that the counterinsurgency is typically brutal; in fact just the opposite. By investigating and punishing those few miscreants who indulged their personal predilection for mistreating a defenseless human being against official policy, it proves the US is conducting an untypically humane counterinsurgency operation.
"Typical" brutality would be, say, prospective EU member Turkey marching thousands of civilians out into the desert to die horrible deaths of starvation in the 1960s, or Saddam gassing Halabja. Nothing remotely like that is happening in Iraq today.
Heiko,
Excellent points. I think it can be argued the West in general owes some responsibility for removing Saddam because of buying billions in oil from him that he had no real right to claim as his (beyond being the strongest thug, that is), which wealth he then used to oppress his people and invade his neighbors. His own people might have had some chance to overthrow him decades ago if not for foreign wealth and arms. But this does tend to all get laid at the feet of the U.S., whether we support him or depose him.
TallDave,
I said you are technically correct in saying that the ration system was not terminated. I also said that as a direct result of the Bush administration's war and occupation many more Iraqis are starving, and going without shelter than before. Did you miss that part?
Yes, I know that "oil is being sold", and I also asked you how much oil is being sold, where the money is going and who is benefitting from it. Did you miss that part?
"Oh, so statistical analysis is done strictly for fun and is never meant to mean or prove anything. OK, so factoring in this somewhat insane premise, we're left with either a study that proves 8,000 people were killed, or a study that proves nothing because statistics never prove anything."
TallDave, I really don't know how to respond to this. You appear to be playing some kind of silly, childish game here - at least I hope so.
Regarding the IRI poll, I will wait to see the full poll results. However, I have looked at some of their past polls, and in addition to clear sampling bias, there are also a lot of very big problems with their interpretation, and presentation. There is, of course, also a serious problem with your interpretation of just one question we have seen from the current poll - you seem to think it refers to the current situation on the ground when it clearly does not.
As for your other remarks, your charge of sampling bias in the Lancet study is without foundation, and that issue has been addressed very effectively here again and again and again and again and again by several people who are better qualified than I am, so I will not belabour it now, particularly since it is more likely than not to fall on deaf ears.
TallDave,
The other day I asked you some questions, which you must have missed, since you did not respond to them. As a courtesy, I will ask them again:
What was the U.S. government doing with respect to Saddam while the Anfal - which arguably rises to the level of genocide - was ongoing? What was the U.S. government doing with respect to Saddam during the period that the torture and execution chambers, and the rape rooms in Abu Ghraib were being most actively used? What was the U.S. government doing with respect to Saddam when his military committed mass murder at Halabja and what was their reaction in the days, weeks, months and years right after it happened?
What was the U.S. government doing during the eight days that Saddam's military was slaughtering by the tens of thousands the Shi`a and Kurdish rebels of 1991 right under its nose in territory the U.S. military controlled? And what was it doing while the regime followed up the eight days of slaughter of Shi`a rebels by committing mass summary executions of Shi`as, right under the nose of the U.S. military in the southern no-fly zone, which was supposedly set up to protect the Shi`a from Saddam?
Please answer the questions, TallDave, or if you don't know the answers just say so.
Talldave, your view of the torture scandal reveals yet again that the internet allows people in parallel universes to make contact with each other.
Heiko, some decent arguments there,though I disagree. But I'm too sleepy to elaborate.
Shirin,
I didn't answer your questions because as I said before you would have opposed removing Saddam then anyway, so what's the point of the question? In fact, your questions tend to underline the need for regime change.
As for the statistics stuff, I see you've run smack into the blindlingly obvious logical brick wall you built around yourself. So which interpretation are you choosing? Your insane suggestion all statistics including this one are meaningless and prove nothing or that the only statistically meaningful conclusion is that between 8,000 and 194,000 excess deaths occurred?
And no, many methodology problems have not been addressed, esp. not the undersampling of Kurd/Shia region the authors admit to.
Donald,
I hope your visit to the real universe has been helpful, perhaps even educational.
"Shirin, I didn't answer your questions because as I said before you would have opposed removing Saddam then anyway, so what's the point of the question?"
With all due respect, you have no way on earth of knowing what I would have opposed 2-3 decades ago.
"act, your questions tend to underline the need for regime change."
"Regime change" being, of course, a euphemism for a violent U.S. invasion and occupation of the country, in which case no, they do not.
"Wich interpretation are you choosing? Your insane suggestion all statistics including this one are meaningless and prove nothing or that the only statistically meaningful conclusion is that between 8,000 and 194,000 excess deaths occurred?"
Is exaggerating your opponent's statements beyond all recognition and reason and then presenting two absurd extremes as the only alternatives one of your socratic debate techniques?
"And no, many methodology problems have not been addressed, esp. not the undersampling of Kurd/Shia region the authors admit to."
Indeed, this has been addressed, both here and in the paper published in the Lancet.
TallDave, here are the answers to the questions I asked you:
What was the U.S. government doing with respect to Saddam while the Anfal - which arguably rises to the level of genocide - was ongoing?
It was treating him as a friend and ally, heaping him with praise, sending envoys on regular visits to Baghdad, inviting high level members of his regime to visit Washington DC, providing him with financial, logistical and diplomatic support, and selling him attack helicopters, weapons, and some of the raw material and equipment he used in his unconventional weapons programs. Some of these helicopters and weapons were used against Kurds.
What was the U.S. government doing with respect to Saddam during the period that the torture and execution chambers, and the rape rooms in Abu Ghraib were being most actively used?
See above.
What was the U.S. government doing with respect to Saddam when his military committed mass murder at Halabja?
See above.
What was their reaction in the days, weeks, months and years right after it happened?
1) To carry on business as usual (see above).
2) The CIA attempted a cover up by claiming the gas came from the Iranians, and that it had drifted into the village because of a mistake by the Iranian military.
3) When Congress learned that the helicopters used in the attack were American-made and had been sold to the regime by approval of the U.S. government, they tried to impose sanctions that would prevent Iraq from obtaining any further American technology. The White House promptly killed this bill.
And speaking of poison gas, somewhere around that time the White House dispatched Donald Rumsfeld to Baghdad with the specific task of reassuring Saddam not to worry, that the administration's recent public criticisms of chemical weapons use was not directed at him, and their relations remained as strong and as warm as ever.
What was the U.S. government doing during the eight days that Saddam's military was slaughtering by the tens of thousands the Shi`a and Kurdish rebels of 1991 right under its nose in territory the U.S. military controlled?
For eight straight days the U.S. military gave permission to the Iraqi regime to fly armed helicopters to the rebel areas where for eight straight days U.S. military aircraft circled over the areas while Saddam's military slaughtered the rebels by the tens of thousands. When asked later why he had allowed this, General Schwartzkopf replied that Saddam had tricked him into it by claiming they were using the helicopters to transport officials. Tricked him. For eight days in a row.
Additionally, the U.S. military intercepted rebels on the roads, confiscated their weapons and vehicles, and blocked their movement into rebel areas.
And what was it doing while the regime followed up the eight days of slaughter of Shi`a rebels by committing mass summary executions of Shi`as, right under the nose of the U.S. military in the southern no-fly zone, which was supposedly set up to protect the Shi`a from Saddam?
Flying patrols, bombing here and there. Other than that, absolutely nothing.
TallDave, you asked what was the point of my questions.
The point is very simply that throughout the years during which there was the closest thing to a justification for "taking him out" on humanitarian grounds, the U.S. government did the exact opposite - it supported Saddam. It supported him, in fact, in his aggression against Iran. The point is that by the time the G.W. Bush administration made the decision to "take him out" (a decision taken no later than September, 2001), there was no justification whatsoever to so on humanitarian grounds. By that time there was an abundance of regimes that had far worse human rights records - and were far more dangerous - than his had been for years.
Well for the most part I'm through wasting my time with someone who doesn't understand logic and only wants to engage in personal attacks or endless "yeah but the US did this or that so they're bad" statements. You seem more than happy to blame the U.S. for eveything Saddam did but ardently oppose holding Saddam responsible. The obvious illogic of this position apparently escapes you, just as you don't understand the logical consequences of your insane statements that statistics don't prove anything.
TallDave, I am afraid I did not make myself clear. My point was not that the U.S. is bad. My point was that if at any time there was justification for invading and occupying Iraq on humanitarian grounds, it was while the regime was committing major crimes and atrocities, such as the Anfal, or while he was actually using chemical weapons in Iran and against Iraqis. During that time the U.S. not only did not intervene on humanitarian grounds, it supported the regime. The appropriate time to attack, invade, and occupy a country on humanitarian grounds is not after all these things have been over and done with for years.
"You seem more than happy to blame the U.S. for eveything Saddam did but ardently oppose holding Saddam responsible."
I'm sorry, TallDave. I have always been told that my English is excellent, but I do seem to be communicating very poorly these days. I am sure I have never suggested anything like that.
"your insane statements that statistics don't prove anything."
Again, I seem to have failed to make myself clear. I have not said that statistics do not prove anything. What I have said is that CI's are used not to prove things, but to make predictions about populations by estimating population values based on studies of samples.
You are, in fact, wasting your time, Tall Dave, but it's not because of any failings of those you're having discussions with.
Apologies. The Qur'tuba comment is me figuring out trackback pings.
A Brookings institute scholar is reported as saying that currently 500-100 Iraqis per motnh are dying as a result of the insurgency. http://dailytelegraph.news.com.au/story.jsp?sectionid=1268&storyid=3004…
It's now been pretty close to exactly 24 months since the end of the initial military campaign.
If. as the US government and its partisans tell us, everything in Iraq is constantly improving. it would seem reasonable to assume that fewer Iraqis are dying now than the average for that period.
That implies a minimum of 12,000-24,000 casualties - excluding casualties in the initial invasion and deaths indirectly caused by the violence and the lack of security (e.g. miscarriages and maternal deaths due to unassisted home deliveries as a result of the curfew).
Interestingly if one visits the Brooking institute's website http://www.brookings.edu/views/op-ed/ohanlon/20021009.htm, Michael O'Hanlon predicted before the war that likely Iraqi losses would be 10-50,000 military casualties and "tens of thousands" of civilian deaths.
O'Hanlon's article was hardly an anti-war diatribe. For example, he estimated the likely financial cost of the war at around $50 billion plus $10 billion per year for the occupation. In fact, the cost, includign two years occupation costs is now around US$300 billion.
So those who wish to argue for very low Iraqi casualties figures need to believe that O'Hanlon simultaneously serioudly overestimated likely casualties AND seriously underestimated likely budgetary costs.
Maybe O'Hanlon just isn't very good at his job - or maybe sometimes, despite what Winston Smith concluded, two plus two really does add up to four even if The Party says different.
Looking at the Brookings Institute's "Iraq Index" http://www.brookings.edu/dybdocroot/fp/saban/iraq/index.pdf (page 8), we find that the Institute estimates 26-45,000 civilian fatalities including crime-related fatalities and notes that this is probably an underestimate. The report also cites 1,500+ Iraqi security force casualties; 700+ deaths of security force recruits and 50+ deaths of Iraqis employed as translators by the Occupation forces.
So we're getting up to 50,000+ deaths without including Iraqi military casualties in the initial campaign, insurgents and, once again, indirect deaths from causes such as disruptions to medical services.
Did the Occupation Forces deliberately kill all these people?
Of course not.
But most of them would probably still be alive if the invasion had not occured.
Looking at the Brookings Institute's "Iraq Index" http://www.brookings.edu/dybdocroot/fp/saban/iraq/index.pdf (page 8), we find that the Institute estimates 26-45,000 civilian fatalities including crime-related fatalities and notes that this is probably an underestimate. The report also cites 1,500+ Iraqi security force casualties; 700+ deaths of security force recruits and 50+ deaths of Iraqis employed as translators by the Occupation forces.
So we're getting up to 50,000+ deaths without including Iraqi military casualties in the initial campaign, insurgents and, once again, indirect deaths from causes such as disruptions to medical services.
Did the Occupation Forces deliberately kill all these people?
Of course not.
But most of them would probably still be alive if the invasion had not occured.