Unless, of course, you'd like to see what a doctor who is ACTUALLY FAMILIAR with the metholodology, unlike any of us here apparently, you can read this: http://www.truthdig.com/report/item/200601017_655000_iraq_war_deaths/

Cheers!

Seriously, if you actually have a professional critique, you ought to comment specifically on why the phenomenological modeling in the report, parametric or otherwise, is so clueless here (i.e. differentiate it from hypothesis testing, as you have failed to do so thus far). If you're really interested in honest debate, you'd share your oh-so-high-and-lofty view.

And John, yes 92% had death certificates. But I find it unlikely that one could find some alleged central repository of certificates.

Imagine how a developing country functions on a day-to-day basis, and then imagine how chaotic it would be after decades of an exceedingly barbaric dictatorship, followed by an invasion and three years of foreign occupation and civil war.

Most of the certificates were probably issued by hospitals or other local authorities, in which case it would be unlikely that someone will risk their life to transfer records of people who are already dead to central authorities in Baghdad.

This is probably one of the rationales for making the approach a classic epidemiological method.

Hence also my skepticism that the estimate of the Iraqi government is too low (in addition to the facts that their political situation dictates that they must placate the Bush administration, and they are hopelessly overwhelmed on all fronts).

Cheers!
Jason

perhaps a better point of comparison would be to either the total civillian death rate in germany, or (more relevantly) the combined german military/civillian death rate. wikipedia has more than 10% of the german population dying in ww2.

well, yes, i must have missed it. perhaps you can post it again, because the only relevant link i see is to a post at your own site (where i can see no evidence of davies having admitting what you claimed).

Why bother, I think the study is not worth the time. Consider this, according to this study more civilians have died in Iraq than died during the Allied Strategic Bombing of Germany (about 400,000). That was when we really were bombing cities pretty much indiscriminantly.

IIRC there was no civil war in germany.

In case you didn’t notice, no one is impressed with your proclivity to use vulgarity, force, and insults to convey your viewpoint.

Please, your are just as insulting, spare me your faux indignation.

Yes I did read the study, fine sir, the 500 figure was in response to your posts, not the article. Are you familiar with your own posts?

Yes, and it is quite clear from the second update, that upon reading the study my 500 number was too low.

You still have no true critique of their methodology that I can detect.

Your inability to detect my criticism isn't my problem.

Go read up on log-linear models and then let’s talk.

This is so childish.

I find it hard to believe your claims that essentially all epidemiologists are losers.

Sorry, those are your words not mine.

If the methods of this study are so wrong, why haven’t at least a few stood up and said so?

Why bother, I think the study is not worth the time. Consider this, according to this study more civilians have died in Iraq than died during the Allied Strategic Bombing of Germany (about 400,000). That was when we really were bombing cities pretty much indiscriminantly.

In case you didn't notice, no one is impressed with your proclivity to use vulgarity, force, and insults to convey your viewpoint.

In fact, using less of these would make you sound less like a cornered cat hissing.

Yes I did read the study, fine sir, the 500 figure was in response to your posts, not the article. Are you familiar with your own posts?

You still have no true critique of their methodology that I can detect. Go read up on log-linear models and then let's talk. Fair enough, they should have gone Bayesian, at least for a comparison if nothing more. Why don't you give them a call if you're so right and so concerned?

I find it hard to believe your claims that essentially all epidemiologists are losers. If the methods of this study are so wrong, why haven't at least a few stood up and said so?

Cheers!

Do you think you do yourself any credit by framing all your responses around insults?

Why not, that is you tactic. If you don't like it, don't do it.

That calling me stupid constitutes scoring a point for your perspective?

I didn't call you stupid, I called you ignorant. Considering your track record around here calling others cretins and other names I fail to see why this bothers you. You crapped in your mess kit, now live with it, most think your a complete jerk.

I understand the distinction between frequentist and subjectivist approaches in statistics.

I know.

You object to the study in question because it makes inferences about people dying, which is a one-time event. I think it a silly objection.

This is because of you lack of understanding that you admitted above. What can I do beyond this other than point out that fequentists concepts don't hold for one time events. This is obvious from the frequentist definition of probability.

That people die once is indisuputable. That you can amass data on that event multiple times, is also true.

Again, the problem isn't that you have multiple observations, but that it doesn't fit the frequentist definition of probability. Now you can say, "Well, okay, but it is close enough for me," but then you are starting to get perilously close to the realm of subjective probability and one should at least admit that this makes the results some what less reliable than say a lab experiment that is more highly controlled.

I don’t want to put words in your mouth, but it seems to me that your approach would be to say that you had a prior sense (perhaps pulled from thin air, or wishful thinking) that the number was markedly lower, and that this study is so dramatically out of line with those beliefs that it cannot be considered of high probability. Which would mean, of course, that it is all about you and your subjective beliefs, and not about what the actual numbers are.

Not bad. Yes, this captures a large part of my view of this study. However, the number of deaths isn't something pulled out of thin air. For example, there are several other numbers out there. There is the 30K that Pres. Bush mentioned a while ago, probably way too low. There is Iraq Body Count, but also low. There are other accounts that put the numbers higher as well, but not this high. These numbers seem to be several times larger than other estimates. Using these other numbers, and one's subjective beliefs about them, to inform one's on beliefs about the number of deaths is not unreasonable in that any and all scientists do it everywhere on anything they are studying.

Further, using the "machinery" of Bayesian statistics one can take their subjective beliefs about the the number of deaths and give them mathematical form, called a prior. This prior is actually a probability distribution that could be say, normal with a given mean and variance: say 200,000 for the mean and 25,000 for the variance. Then using Bayes theorem you can update that prior probability distribution using the data you observe. Hence one's subjective beliefs become tied to the data and even wildly bizarre initial beliefs will be swamped by the data.

So your view of this is not bad, but your final take on it is wrong. There are lots of Bayesian statisticians out there and to imply that they are all dishonest is inaccurate. In fact, I'd say they are perhaps more honest in that they put their subjective beliefs out there for everyone to see, and at the same time agree to let those beliefs be changed by the data. Frequentists don't do any of this save in a ad hoc manner.

Your questions regarding the entire approach taken, rather than the specific methodology, is a critique of epidemiology as a whole, and not this specific study.

No, not necessarily. While I think that Bayesian methods are superior, Frequentist methods can be used in situations that conform to the Frequentist concept of probability. One can argue that there is a problem with the Frequentist approach in that its measure of probability is a limit,

Prob(A) = limN -> infinity m(N)/N.

Since that is a limit one can change a finite number of outcomes without affecting the limit, but since in reality this isn't the case (i.e. we only have a finite number of observations) the above isn't an operational concept since it can't form the basis of empirical measurement of probability. I don't tend to go that far since we can often get very large numbers of observations that probably do the trick nicely.

Sorry if all this is a mischaracterization, but perhpas you could explain to us how you would go about deriving an accurate estimate of the number of excess deaths?

Well, I'd look at the other numbers on this topic, use them to form my prior, then I'd gather the data (their method there looks good) then update my prior. I'd probably select a number of priors to see how sensitive my results are to the selection of my prior including what is called an non-informative prior (think of a very flat distribution). This way, people would see my "biases" and could see how my biases impacted the results (by comparing them to the other priors).

A methodologicial problem with the new study, or perhaps with the prior estimates. To the extent that it may refer to the new study, then an appropriate response might be “this seems weird, I suspect there might be some methodological problem that needs to be investigated.” Not ” this study presents carefully and selectively packaged truths, half-truths and falsehoods presented to sway opinion towards a political viewpoint”.

I didn't say that, my view is that this study should be taken with a great deal of caution. For example, every Frequentist result can be obtained using Bayesian methods by carefully selecting your prior. However, the Frequentist method doesn't make this prior explicit. The prior, as I've noted, is a way of putting your biases out there for everyone to see. So, in a sense one could make the claims Charles Austin has made.

Charles,

First, if you’re a statistician you surely know that if you fit a normal distribution to a mean of 500 deaths, anything over 1,000 is just as likely to occur in Iraq as 0 (depending on the variance you choose – though I’m sure you’d choose a large one, since you are clearly biased), and would be well outside 3 standard deviations (i.e., an outlier). That would mean that only 10 or so days during the entire war would have had body counts exceeding 1,000. 2,000 would mean a Fallujah debacle, and they excluded such outliers in the recent analysis.

Did you even read the article? My guess is no, since for the last 13 months of their study the average number of deaths per day is about 1,000 on a yearly basis. Further, we need to know the variance of the distribution since you don't know it either you are talking out of your ass.

Further, the point wasn't that 1,000 deaths/day or even 2,000 deaths/day were likely or even frequent, but that those kinds of things would be somewhat news worthy. As has been pointed out that would be like mutliple 9/11s here in this country. I noted that these deaths aren't as highly concentrated nor did they happen in such a grandiose manner as 9/11, but it still seems like a huge number of deaths to go completely under the radar.

Second, of course there are theoretical quibbles with measuring one-time events with statistical methods. In biology at least, we refer to it as pseudoreplication. But what the authors have employed is a largely descriptive statistical technique that does not require true replication. All parametric stats (and non-parametrics) were simply in the form of modeling or estimation, not hypothesis testing, so calm down. Thank god we don’t have the replication you seem to suggest.

Uhhmmm, I'm not saying we must have replication, just that if we are going to use Frequentist methods, that replication is a requirement.

Actually, why don’t we turn all of central Asia into a theoretical analysis Steve (i.e. by invading each country independently, all at the same time)? THEN we’d know, right? You’re probably one of the “statisticians” hired by Exxon to argue that there was no way to establish that the Valdez killed anything, because it was only one spill.

This is just stupid. Nowhere have I indicated that anything of the sort be done.

Finally, neither of you are scientists (if Steve is a statistician as he has claimed). According to Charles, “this ain’t rocket science.” I’m not convinced you know your ass from a hole in the ground” here, in your words, Steve.

I think I've demonstrated that I know far more about this than you do. Not that I expect you to admit it.

Neither of you are epidemiologists.

Thank God.

Another commenter was correct to note that a fellow epidemiologist claimed that the methods were impeccable.

One can follow Frequentist methods to the "T" and still be "wrong" in that it is the wrong methods to apply to the situation. Further, I've seen some evidence suggesting that epidemiology has a problem with over estimating the efficacy of things like drug efficacy, and that part of the problem might be the dependence of Frequentist methods. You can see more in my comment above to Tano.

Besides, newsworthy scientific papers are often rushed through, and not to change election results. Your spittling is tantamount to an attack on science and scientists, and I for one take severe exception.

Oh this is just simply bullshit. Because I disagree with a methodology, I'm attacking scientists. What errant nonsense.

One more thing — as far as repeatability of the method goes in terms of post-war mortality, their figures line up pretty well. Of course this won’t satisfy your mis-placed statistical yearnings for replication and conventional statistics, Steve.

Which raises an intriguing question...how do those pre-war death rates match up with pre-war claims of infant deaths? I seem to recall much chest thumping about the sanctions killing a million babies.

Oh, and you notion of repeatability seems to need some work. The fact that their results match up with other estimates isn't repeatability in terms of an individual death. And also, I'm not saying we must have repeatability (because, IMO, we can't), but that we need to shift to a method that does not rely on repeatability. For somebody who thinks they understand statistics, you sure are advertising your ignorance.

Snuh,

still waiting for a link to where you had Davies “agree[] that he misrepresented what a CI is.”

Well, I just read in one of the thinks to Crooked Timber. I don't know what more to tell you, other than you missed it.

Stephen Daedalus,

And I note that your pose of condescension is… transparent and laughable.

Hey, just giving my pal Tano there a taste of his own medicine. What can I say, turn about is fair play, IMO.

As I noted your mathematical and statistical sophistication is…lacking.

And I note that your pose of condescension is... transparent and laughable.

still waiting for a link to where you had Davies "agree[] that he misrepresented what a CI is."

also, it does not strike me as inherently unreasonable to suppose that only 1 in 13 [50,000 (the IBC number) in 650,000(the lancet)] of the "excess deaths" in iraq were corroborated by two accounts in the foreign press [the criteria for inclusion in the IBC].

But it does indicate that as we scientists know is of the essence, lack of repeatability of measurements does not explain their figures.

I'd downgrade your shovel of salt back to the pinch. These are not "real" numbers, it is a ballpark. You ought to read up on the sort of modeling we apply in the biological sciences.

Cheers!

As you’ve both indicated, this is a debate about science. So, as a scientist, I’ll address your points.

Let’s look at your oh-so-sophisticated statistical and scientific claims (contrary to your tone Steve, you are not a stat demi-god worthy of knee-scraping).

First, if you’re a statistician you surely know that if you fit a normal distribution to a mean of 500 deaths, anything over 1,000 is just as likely to occur in Iraq as 0 (depending on the variance you choose – though I’m sure you’d choose a large one, since you are clearly biased), and would be well outside 3 standard deviations (i.e., an outlier). That would mean that only 10 or so days during the entire war would have had body counts exceeding 1,000. 2,000 would mean a Fallujah debacle, and they excluded such outliers in the recent analysis.

Second, of course there are theoretical quibbles with measuring one-time events with statistical methods. In biology at least, we refer to it as pseudoreplication. But what the authors have employed is a largely descriptive statistical technique that does not require true replication. All parametric stats (and non-parametrics) were simply in the form of modeling or estimation, not hypothesis testing, so calm down. Thank god we don’t have the replication you seem to suggest. Actually, why don’t we turn all of central Asia into a theoretical analysis Steve (i.e. by invading each country independently, all at the same time)? THEN we’d know, right? You’re probably one of the “statisticians” hired by Exxon to argue that there was no way to establish that the Valdez killed anything, because it was only one spill.

Finally, neither of you are scientists (if Steve is a statistician as he has claimed). According to Charles, “this ain’t rocket science.” I’m not convinced you know your ass from a hole in the ground” here, in your words, Steve. Neither of you are epidemiologists. Another commenter was correct to note that a fellow epidemiologist claimed that the methods were impeccable. What more can you ask for? As an ecologist, I’d tell you to go to hell if you thought I didn’t know how to measure nitrogen mineralization or plant community composition. Stay within the realm of your expertise and stop over-extending yourself, committing precisely the mistakes you criticize others for.

Besides, newsworthy scientific papers are often rushed through, and not to change election results. Your spittling is tantamount to an attack on science and scientists, and I for one take severe exception.