Where he gets this is from is the fact that within that nation-wide survey, one data set (that refering to population living near Fallujah) was a statistical 'outlier', that is, not coherent with the rest of the survey's findings.  Because of that, during the statistical analysis portion of the survey, the Jhons Hopkins medical school team decided to exclude that data set.  Not only is that good statistics (the exclusion of outliers is a bog-standard aspect of statistical analysis), it's actually a conservative approach that tends to minimize (not increase) measured casualties. 
Furthermore, the Jhons Hopkins team published not only their top line results but also the methodology followed and even the data sets that were *excluded* from the final results.  Far from being "a record of doing their work improperly", that's the hallmark of a carefully done and thoroughly documented scientific research.

To claim that excluding the Fallujah data was done in a professional and scientific manner is laughable.  To have included it would have totally tanked their findings, as the Fallujah outlier was so large as to be completely indefensible.  Yet, you could see how much it pained them to exclude it, as they continued to cite results using the Fallujah cluster despite the fact that it wasn't plausible (keep in mind that the study was completed prior to the November 2004 Operation "Al Fajr").  The Fallujah results are discussed in 5 of the 7 pages of discussion from the study article.  If the results were such an outlier, why discuss them in such depth across nearly the entire paper while making statements that maybe it might not be such an outlier (as if other neighborhoods would yield an extrapolation of 56K KIA!).

 

Here's what I wrote last year on this:

 

 Fallujah. “In Falluja, the team noted that vast areas of the city had been devastated to an equal or worse degree than the area they had randomly chosen to survey.”

 Here are some calculations that make the above statement appear very biased. Now, since I don’t have access to their data, I use the average persons per cluster calculation below to extrapolate the findings’ predicted number of deaths in Fallujah. This means that the calculations will most likely be slightly off, but that doesn’t detract from the solid conclusion that Fallujah is an outlier and that the study in the same breath calls it an outlier while implying that it’s the norm as well.

 7868 people surveyed / 33 clusters  = 238 people per cluster

 238 people per cluster / 256K people living in Fallujah = sample is 1/1075^th of the population

 1075 x 52 violent deaths = 55,900 Fallujans dead

 If there’s that many dead, what is the number wounded? 

 So, if you look at the calculations above coupled with their qualitative statement, the conclusion that is drawn from the study is that surveying a different neighborhood means that they still would have arrived at sample statistics resulting in at least 55,900 Fallujans dead. WTF?

Where he gets this is from is the fact that within that nation-wide survey, one data set (that refering to population living near Fallujah) was a statistical 'outlier', that is, not coherent with the rest of the survey's findings.  Because of that, during the statistical analysis portion of the survey, the Jhons Hopkins medical school team decided to exclude that data set.  Not only is that good statistics (the exclusion of outliers is a bog-standard aspect of statistical analysis), it's actually a conservative approach that tends to minimize (not increase) measured casualties. 
Furthermore, the Jhons Hopkins team published not only their top line results but also the methodology followed and even the data sets that were *excluded* from the final results.  Far from being "a record of doing their work improperly", that's the hallmark of a carefully done and thoroughly documented scientific research.

To claim that excluding the Fallujah data was done in a professional and scientific manner is laughable.  To have included it would have totally tanked their findings, as the Fallujah outlier was so large as to be completely indefensible.  Yet, you could see how much it pained them to exclude it, as they continued to cite results using the Fallujah cluster despite the fact that it wasn't plausible (keep in mind that the study was completed prior to the November 2004 Operation "Al Fajr").  The Fallujah results are discussed in 5 of the 7 pages of discussion from the study article.  If the results were such an outlier, why discuss them in such depth across nearly the entire paper while making statements that maybe it might not be such an outlier (as if other neighborhoods would yield an extrapolation of 56K KIA!).

 

Here's what I wrote last year on this:

 

 Fallujah. ̶

Swhitebull,

 

They don't state as much (that other neighborhoods would have yielded the exact same result), but the strong implication is there IMO. 

 

To me, it is simple, if the outlier status is questionable, then it is fair to discuss the with and without.  When the outlier status is firmly locked, such as a neighborhood that equates to 56K killed when the most that had ever been operating in the area had been a brigade plus, to drag on the discussion is outlandish.  Talk about how the data points from the cluster don't extrapolate into a result that checks with common sense, and then leave it at that.  Don't belabor the point by continuing to discuss with/without results as if that will somehow make 56K killed begin to make sense and seem plausible. 

"Coming back to a more 'serious' level.... Given all that, I find it a little hard to comprehend how and why you'd think that their critique of the Lancet's statistical methodology would be any more credible than, say, Tickle-me Elmo (arf!.. arf-arf!... ooops, sorry)."
If their observation is correct, then why should it be disqualified.

========
Right you are.  I just find it funny that a group whose notorious characteristic is that they use comically poor statistical methodologies would be cited as source of insightfull questions regarding statistics.  However, you are right; address the question, not the questioner.
==============

"Regarding you quotes - the one you posted was not from the Lancet, but from Iraqbodycount.org's own text in their "rebutal".  Which is great, but not particularly informative of what the Lancet published."

You see, I quoted what IBC said and provided reference. Then you I said where they got the number (in The Lancet). That was obviously not specific enough for you.

"Regarding your reference to the Lancet's study, well, hum...it's not in page 12 of the Lancet Survey, as you claim. Correct me if I'm wrong, but I think we're both reading the same document, and Page 12 contains the "Limitations" chapter and half of the "Conclusions" chapter.  Neither one of those even mentions the words "male" or "female", let alone make the claims that you state they do."

Take a look at the graph on p.12. Here is the quote, also p.12:

"In the next graph is shown the deaths from violent causes by age and sex. As can be seen, violent deaths account for most of the deaths, and violent deaths are almost entirely in males. Among the males, there were no practical survey methods to determine which of the deaths were among active combatants. It is interesting to note that the largest single age group of female deaths was among the under age 15 years."

"It's possible that it's elsewhere on the Lancet's survey - if you find it please let me know."

Done.


"In the meantime, I dont follow how you've concluded that a population that has suffered 2.5% mortality could *possibly* have a 7% mortality amongt males, even if every single casualty was male (which they were not).  Unless that happened *and* the population was overwhlmingly female (which it isnt)."

The other way round. Adults constitute less than 66% of pop. Of this group 50% are males. So this group comprisese less than 33% of entire pop. However, the survey found almost all violent deaths were suffered in this group. So

33%*7.5% = 2.5% = Lancet claim of nat'l average.

The actual numbers are

"Of the 287 violent post-invasion deaths recorded by the Lancet authors where the age and sex was known, 235 (82%) were adult males between 15 and 59 years old. Extrapolating to the population as a whole would mean that around 470,000 men in this age group have been killed violently, i.e. one in 15 (7%) of adult males aged 15 to 59"

link... even without the 60+ segment (as I read it).

So the numbers are consistent either way.

7.5% amongst adult males, or more than 7% according to IBC.


"The only way I can see around that is if you only count males in demographic groups where the impact of the war is particularly severe (of a particular age group, of a particular ethnic group, living in a particular location, etc..)  In which case, that claim has no meaning.  If you cherry-pick like that *of course* you're goign to find populations that have a much higher death rate.  Gee whiz, no wonder.

And in no way is that a disqualification of the Lancet's results - of course armed conflict afects certain demographic groups more heavily. For instance: military age males in places where the fighting is hardest tend to get the short end of the stick more often than 50-year olds living in peacefull provinces. That is, quite simply, the nature of warfare - dunno about the guitarists and retired librarians at Iraqbodycount.org, but I'd be truly amazed if anyone here would find *that* to be a surprising result...

Dont want to put words in your mouth, so I'd love to learn more about what this surprising result is and how you came to it."

Of course you wouldn't. I was just responding to this:

"- The Lancet survey does *not* indicate that 7% of the male population of Iraq has died over 3+ years of armed conflict.   The good folk at Iraqbodycount just made that one up.
Let's "reality-check" the math: since the Lancet survey found that 2.5% of the overall population of Iraq died over the course of the war, even if every single casualty was male that still does not account for 7% of t

Regardin the sample size.  On this I think you're plain wrong; a randomized sample of 12,000 people in 1850 households most certainly *is* large enough to provide adecuate sample data for a population of 22 million.  I'm amazed you think otherwise if you have worked in a professional capacity in statistics - but let me turn that question around to you; what woul *you* think is an adecuate sample size for measuring mortality rates in a country of 22 million or so?

eu4ea,

 

I have never heard a sample described as randomized before.  This was not a random sample.  This was a multistage cluster sample that used random sampling within the clusters.  There is a difference, and that difference is huge. 

 

However, your underlying point that the sample size was large enough to produce results is fine; however, because of its cluster design, the poor precision of the results make it much more sensitive when conducting your hypothesis testing (i.e. because your confidence interval at standard levels of significance is huge, you are more likely to fail to reject your null hypothesis), and also much more sensitive when you have bias in your sample. 

"Coming back to a more 'serious' level.... Given all that, I find it a little hard to comprehend how and why you'd think that their critique of the Lancet's statistical methodology would be any more credible than, say, Tickle-me Elmo (arf!.. arf-arf!... ooops, sorry)."
If their observation is correct, then why should it be disqualified.

========
Right you are. 

I just find it funny that a group so notorious for their comically inadecuate statistical methodologies would be cited as source of insightful questions regarding statistics.  However, you are right: address the question, not the questioner. Even if it *is* pretty comical.
==============

"Regarding you quotes - the one you posted was not from the Lancet, but from Iraqbodycount.org's own text in their "rebutal".  Which is great, but not particularly informative of what the Lancet published."

You see, I quoted what IBC said and provided reference. Then you I said where they got the number (in The Lancet). That was obviously not specific enough for you.

"Regarding your reference to the Lancet's study, well, hum...it's not in page 12 of the Lancet Survey, as you claim. Correct me if I'm wrong, but I think we're both reading the same document, and Page 12 contains the "Limitations" chapter and half of the "Conclusions" chapter.  Neither one of those even mentions the words "male" or "female", let alone make the claims that you state they do."

Take a look at the graph on p.12. Here is the quote, also p.12:

"In the next graph is shown the deaths from violent causes by age and sex. As can be seen, violent deaths account for most of the deaths, and violent deaths are almost entirely in males. Among the males, there were no practical survey methods to determine which of the deaths were among active combatants. It is interesting to note that the largest single age group of female deaths was among the under age 15 years."

"It's possible that it's elsewhere on the Lancet's survey - if you find it please let me know."

Done.


Ah, ok!   You're refering to Page 11 which contains a couple of graphs, not Page 12.  I suspect the difference is that you were counting the cover, while I was goign by the Study's page numbering. No big deal. 

Regading the figures, I did not see them before, but looking at your reference, it seems to me that the most plausible explanation is that some enterprising soul at Iraqbodycount.org took a ruler to the Page 11 graphs, and came out with the numbers you're refering to.


"In the meantime, I dont follow how you've concluded that a population that has suffered 2.5% mortality could *possibly* have a 7% mortality amongt males, even if every single casualty was male (which they were not).  Unless that happened *and* the population was overwhlmingly female (which it isnt)."

The other way round. Adults constitute less than 66% of pop. Of this group 50% are males. So this group comprisese less than 33% of entire pop. However, the survey found almost all violent deaths were suffered in this group. So

33%*7.5% = 2.5% = Lancet claim of nat'l average.

The actual numbers are

"Of the 287 violent post-invasion deaths recorded by the Lancet authors where the age and sex was known, 235 (82%) were adult males between 15 and 59 years old. Extrapolating to the population as a whole would mean that around 470,000 men in this age group have been killed violently, i.e. one in 15 (7%) of adult males aged 15 to 59"

link... even without the 60+ segment (as I read it).

So the numbers are consistent either way.

7.5% amongst adult males, or more than 7% according to IBC.

"The only way I can see around that is if you only count males in demographic groups where the impact of the war is particularly severe (of a particular age group, of a particular ethnic group, living in a particular location, etc..)  In which case, that claim has no meaning.  If you cherry-pick like that *of course* you're goign to find populations that have a much higher death rate.  Gee whiz, no wonder.

And in no way is that a disqualification of the Lancet's results - of course armed conflict afects certain demographic groups more heavily. For instance: military age males in places where the fighting is hardest tend to get the short end of the stick more often than 50-year olds living in peacefull provinces. That is, quite simply, the nature of warfare - dunno about the guitarists and retired librarians at Iraqbodycount.org, but I'd be truly amazed

Ok - you *are* cherry picking. 

We're not talking about "males"; we're talking about "males between 15 and 59 who died a violent death".  In that case, I have no problem with the figure as such - if you cherry pick the segments with the highest mortality, you can easily get to that.  Above it, too.

However,  it's worth pointing out that you're buying into IBC's false assumption that violent deaths are the only conflict-related deaths, and are accepting numbers extrapolated from that flawed basis.

Naturally, they're not - and excess deaths from causes other than violence (such as privations, lack of medicine, heatstroke etc.) tend to strike the very old and the very young disproportionatelly. That reality was cheerfully ignored by the IBC team - they just saw that big "violent deaths" chart, took a ruler to it, and extrapolated to their heart's content.

When you think of it, it's not surprising; IBCs own methodology relies exclusively on newspaper reports of violent deaths. Further, that sort of extreme mathematical sloppiness is not unheard-of coming from a guitarist, a psychologist and "freelance researcher" for Greenpeace (the 3 guys who were the material authors of that particular pearl of wisdom). 

The upside is that it makes it a lot easier to catch - you dont have to worry about any sort of elaborate deception; just think "how could you screw this up at the most sophomoric level?" and run with that.

Heart,

eu4ea