6/20 Curiosity got the best of me, and I read the Lancet study, too. It's
quite thorough, undeceptive, and straightforward. Responding to
emarkp's post (http://csua.com/?entry=38170
RE: Confidence Intervals. (1)Assuming a normal(it's not, it's
skewed), there's a 90% certainty there were over 40k dead, 85%
certainty over 51k dead, 75% certainty over 68k dead. (2) this CI
DOES NOT INCLUDE Falluja. The author's purposely excluded a huge
outlier, and the most violent region in all of Iraq in order to form
a conservative estimate. (3) the author's are plainly honest about
the difficulties inherent in this kind of war-time study. (4)
Furthermore, this study estimated the number dead at the time of the
study. Extrapolating to today yields: 90%: 65, 85%: 82k, 75% 108k.
Think about that: lowest quadrile = 108k dead. Not including Falluja.
RE: death certificates: Out of 142 deaths, 78 certificates were
asked for, and 63 provided(81%). Additionally, "When households
could not produce the death certificate, interviewers felt in all
cases that the explanation offered was reasonable"
RE: non-war violent crime. That's part of the point of the study.
war-related conditions cause an increase in overall violent crime,
including murder, faction in-fighting, etc.
RE: the cluster analysis: "Because the probability that clusters
would be assigned to any given Governorate was proportional to the
population size in both phases of the assignment, the sample remained
a random national sample. This clumping of clusters was likely to
increase the sum of the variance between mortality estimates
of clusters and thus reduce the precision of the national
mortality estimate." So yes, confidence intervals get bigger. That
was a call they had to make, to reduce researcher risk. But it's
still a random sample. -nivra
\_ Thsi is ordinarily true, but I would be curious to see some
\_ This is ordinarily true, but I would be curious to see some
pre-Saddam data. Naturally, I doubt we could find something like
that. -- ilyas
\_ The study actually recorded pre-Saddam data. Pre-Saddam death
rates were 5.0(3.7,6.3)/1000/yr, Post-Saddam data were
12.3(1.4-23.2)/1000/yr. -nivra
\_ You are telling me that during Saddam's tenure Iraq had a lower
death rate than the United States? -- ilyas
\_ So is the CIA:
http://www.cia.gov/cia/publications/factbook/geos/iz.html
\_ I don't think it's true, myself. -- ilyas
\_ Why not? Look at the population age breakdowns.
\_ Why not? Because it simply doesn't make sense.
Almost every cause of death is more sereve the
less developed you are. I also think certain
types of deaths are simply not reported. I would
also like to note that your typical 'hellhole
middle eastern arab states' all have extremely
low death rates for some reason. It's very
suspicious. -- ilyas
\_ It's even better than that. Post-invasion,
excluding Falluja, the Iraqi mortality rate is
7.9 per 1000 people. According to the CIA
World Factbook, the estimated 2005 mortality
rate in the US is 8.25 per 1000.
\_ The slate article cites UN data as saying:
"Iraq's mortality rate from 1980-85 was 8.1 per
1,000. From 1985-90, the years leading up to
the 1991 Gulf War, the rate declined to 6.8 per
1,000. After '91, the numbers are murkier, but
clearly they went up. Whatever they were in
2002, they were almost certainly higher than 5
per 1,000. In other words, the wartime
mortality rate--if it is 7.9 per
1,000--probably does not exceed the peacetime
rate by as much as the Johns Hopkins team
assumes." -emarkp
\_ I think it's incredibly shocking if true.
Saudi Arabia death rates are apparently
2.62/1k. Anybody want to comment on why
this might be? Has anybody plotted if
the population growth matches birth/death
figures in the Middle East? -- ilyas
\_ Better diet and more exercise?
\_ I have no idea but I have come to view
the US as a weird place... in many areas
we're not much better off than the 3rd
world places. Iraq was nowhere near as
fucked up as some 3rd world places...
people are relatively educated and so
forth, they have infrastructure. What is
the US death rate among young men in
ghettos? I'm playing GTA: San Andreas
right now and based on this research,
large crowds of people regularly get
shot or run over in downtown areas.
\_ It's not just the US, the Middle East
rates are much lower than those of the
entire industrialized West. I think
it smells of your good ol'fashioned
Soviet-era underreporting. -- ilyas
\_ Look at the age structure of the
populations. I'm pretty sure
80-90 year old Americans are going
to have a higher death rate than
young people in third world
countries. Middle eastern
countries mostly have very young
populations.
\_ I love how this quote exposes Kaplan's
hackery. Well, gee pre-Invasion is
"certainly higher" than what the methods
in the study indicate. The _logical_
conclusion is that the study's methodology
is conservative and under-estimates actual
death rate. Of course, Kaplan doesn't
understand the definition of "unclear,"
so I'm just expecting too much of him, I
guess. -nivra
\_ His assertion that the numbers "clearly
went up" is weak, yes. But weren't we
getting complaints from the world about
how the sanctions were killing Iraqis?
Isn't it reasonable to guess that the
pre-2001 mortality rate was at least 6.8?
\_ sure. -nivra
\_ That data is pre-invasion, not pre-Saddam, and it is from the
same study (they asked people about deaths in the period
before and after invasion). -emarkp
\_ Yes. I had assumed ilyas meant "When Saddam Was Ousted,"
by "Saddam".
\_ Yeah, that's what I meant. -- ilyas
=======
>>>>>>> Your Changes Above
RE: the cluster analysis: "Because the probability that clusters
would be assigned to any given Governorate was proportional to the
population size in both phases of the assignment, the sample remained
a random national sample." -nivra
\_ Could you post a link to the actual study, please? Thanks.
\_ It's changed since the slate article. Here's the URL
(registration required, but bugmenot has a login):
http://www.thelancet.com/journals/lancet/article/PIIS0140673604174412/fulltext
\_ /csua/tmp/~nivra/lancet.pdf
\- does your asessment of iraq wildly change whether the casualty count
is 50k or 100k? i mean it is ok to invade a country on false
premises, if it only leads to "1 vietnam" [usa deaths = 50k]
worth of deaths ... but at 2V, they really ought to have done
better?
\_ my assessment? No. Iraq is massively fucked up, regardless. -nivra
\_ my assessment? No. Iraq's massively fucked up, regardless. -nivra
\- i meant to sort of just throw it out there. i think the
these kinds of reactions shore up the claim if there were
no pix out of abu graib, it would have been a total non-
story. instead of a lot of people being outraged, probably
at least half the country would have had the "you have to
break some eggs to make an omlette" type attitude. what was
amazing and depressing is the rush limbaugh types not taking
a hard line about what goes on in war but these flip comments
about frat hazing and such. anyway, now i am getting
depressed again.
\_ I brought this up before,
\_ I brought this up before, what went on in AG is the same
kind of shit that goes on in American prisons. Why is
so much attention paid to this one incident? Don't
American citizens incarcerated in our prisons deserve
more humane treatment? -- ilyas
\- that is not what i am saying. i am more contrasting
the comments about gitmo [no pix, only stories and
the "what do you epxect, it is a club med?] type
reaction. and certainly there is evidence of beatings
and such in american prisons, but i dont think guards
wipe their asses with the mexican flag, or tell
or dress up black inmates in hooded klan outfits.
\_ So you think psych torture is uniquely worse than
physical torture? Btw, it's not just 'beatings'
that go on in US prisons. -- ilyas
\- are you deliberately being difficult or
is this your natural mode of thought? this
subthread is no longer worth my time.
\_ Note that this subthread is about as long as my
initial post which was deleted because it was
too long. -emarkp
\_ Look dumbass, a *single* post is easily
moved to another location. There is not
a simple way to move an enitre thread
composed of short responses. In the
Iliad thrad, nobody cut and pasted
the wikipedia article on "aspect".
If they had, that would have been deleted
with a request to leave a pointer.
A pointer to a sloda file or a URL are
functionally the same. Do you have some
persecution complex or is it a full on
messiah complex?
\_ "So I'm glad we could agree that the Lancet study was a pile of
crap." -emarkp
Why the big disparity in perceptions?
\_ The Lancet study is a pile of crap the same way I feel the Bible
is a pile of crap. Did you even fucking graduate from Berkeley?
\_ "It's quite thorough, undeceptive, and straightforward."
-nivra
\_ But wrong. Their methodology punches a lot of holes in a
random sampling, and the results aren't valid. -emarkp
\_ So I guess there is a genuine disparity in perceptions
then (as opposed to a political/fake disparity).
\_ What part of "... the sample remained a random
national sample" do you not understand? You read the
study... tell me what these "holes in random sampling"
are. Don't regurgitate Kaplan, b/c he's full of shit.
The authors are clear on both the methodology and the
impacts of the methodology: "This clumping of clusters
was likely to increase the sum of the variance
between mortality estimates of clusters and thus
reduce the precision of the national mortality
estimate. We deemed this acceptable since it reduced
travel by a third." Certainly, it affected the
precision of the estimate, but not the accuracy.
That's one of the main reasons for the large CI. -nivra
\_ They reassigned random locations to other
governorates that were chosen non-randomly. That
means they are no longer random. Yes, they tried to
"match" governorates but you /can't do that/ or you
bias the data--not just the variance. -emarkp
\_ Did you understand the methodology? The
reassignment _did_ occur randomly. Yes, it
decreases statistical power, but No, it did not
compromise randomness. cf. wolfram link below.
\_ No, they did /not/ reassign them randomly. The
chose a paired governorate and randomly chose
between the two. The choice of pairs was /not/
random. -emarkp
\_ Neither are the initial governates. Duh.
It wouldn't make a difference if they
paired all the fucking country up. Again,
look into the wolfram link if you don't
understand. The pairing screws with
precision, not accuracy. -nivra
\_ ??? The paper says: "We obtained January,
2003, population estimates for each of
Iraq.s 18 Governorates from the Ministry
of Health. No attempt was made to adjust
these numbers for recent displacement or
immigration. We assigned 33 clusters to
Governorates via systematic equal-step
sampling from a randomly selected start."
How is that not random for the initial
selection? -emarkp
\_ The clusters are random, the
governates are not. 18 governates
already assigned by M. of Health.
The authors simply collapsed 18 into
12, then proceeded randomly. -nivra
\_ Sure, but the clusters were
assigned to the governorates at
random. I don't see a problem with
that. It's the non-random
substitution of governorates that
is a problem. -emarkp
You're ok with: _/
P1 = P(cluster_init|population_governates)
They did:
P2 = P(cluster_sampled|population_governates,P1)
Why are you ok with P1 and not ok with P2? (reformatted) -nivra
\_ Ah. I'll have to think about that more carefully. That
does suggest a problem in my reasoning. -emarkp
precision, not accuracy.
you're ok with: P1 = P(cluster_init|population_governates) _/
They did: P2 = P(cluster_sampled|population_governates,P1)
Why are you ok with P1 and not ok with P2?
\_ It's widely known that conditioning on an extra thing can
reverse any inequality among probabilities. This
phenomenon even has a name in statistics. This may be
what is giving you pause. -- ilyas
\_ It's not surprising that people who agree with the motivations of
the war doubt the study while people who disagree trust it. I'm
distrustful of statistical correlative studies in general because
they're so hard to get right and so easy to make bad assumptions.
The Lancet study was a bit cavalier with adjustments to the
sampling. To the authors' credit, they did make effort to
decrease possible effects of adjusting the sampling, but I don't
think it was sufficient. -emarkp
\_ Shrug. Random individuals on the net don't make a big diff to
me. nivra and emarkp saying two opposite things means
something.
\_ Why were death certificates not asked for in 64 out of 142 deaths?
Instead of 63 death certificates out of 78 requested, shouldn't it
be 63 certificates out of 142 deaths (44%)?
\_ no. They only asked for death certificates 78 times. emarkp's op
has the relevant quote. It was a methodolical decision, but still
connsistent with random sampling. -nivra
\_ RE: the clustering. Asserting that changing the sampling doesn't
lose randomness doesn't make it so. They can't use the clustering
as a representative sample just as if the pairs of governorates are
exactly equal.
\_ they don't just assert. They show how they changed the sampling,
and if you understood basic probability, you would understand
that it doesn't change the randomness of the sample. Here, a
stats refresher: http://csua.org/u/cfp (mathworld) -nivra
stats refresher: http://csua.org/u/cfp (mathworld)
\_ No, they did /not/ reassign them randomly. The chose a paired
governorate and randomly chose between the two. The choice of
pairs was /not/ random. -emarkp
\_ see above. precision not accuracy. -nivra
\_ see above. precision not accuracy.
RE: non-war violent crime. Why is that part of the study? Did they
check what crime was like here in the US before and after the
invasion? Was there any correlation?
\_ this has to do with what? The estimate of the number dead
is an estimate of total dead, regardless of prior death-rate.
Furthermore, you don't think the stoppage of electricity, water,
food, etc. could have affected non-violent death rates?
food, etc. could have affected non-violent death rates? -nivra
RE: Confidence Intervals. Without seeing their inputs, I don't
trust the results. Especially when they're just dumping the data
into software developed by "Save the Children". -emarkp
\_ Firstly, they used two different data packages, and results
concurred. Secondly, this is exactly what it comes down to:
you don't "trust" the results. You have no background in
epidemiological studies and are not familiar with software or
terminology, yet you don't trust the results from methodologies
that are clearly described, checked by two presumably standard
statistical packages published in The Lancet. That shows the
underlying reason for the "disparity in opinion".
underlying reason for the "disparity in opinion". -nivra
\_ Yes, the study claims that they compared the "Save the
Children" results with EpiInfo. They don't provide their
source data however so it's pretty much impossible to check
their results. -emarkp
\_ Firstly, the software was not called "Save the Children."
The software was designed to measure death-rates on what
I presume was a project paid for by "Save the Children."
Secondly, they do show source data. They provided all
raw numbers. If you distrust the results, go find other
epidemiological software and run it yourself. I happen
to trust editorial boards of major scientific journals
to trust epidemiological results that the authors ran
on two different statistical packages.-nivra
on two different statistical packages.
\_ I used to trust them more. I was convinced by the
"hockey stick" that global warming was correlated with
human industrial activity. Then someone put random
data into the same analysis engine and the hockeystick
appeared with that data as well. So much for that
trust. -emarkp
\_ Fine. That means the heart of your opinion resides
blatant mistrust of the editorial board of first-
rate medical journal. Finally, something we agree
on. -nivra
\_ No, the reason to investigate is my distrust of
what appeared to be a politically motivated study
and a distrust of /all/ statistical correlation
studies. The heart of my opinion is based on what
I read in the study itself. -emarkp
\_ mistrust of "correlation equals causation" is
reasonable. The authors make no such claim.
In fact, the 98k has nothing to do with either
correlation or causation. It's a simple
extrapolation of the death count based on a
statistical sample -nivra
\_ Neither did the hockey-stick claim it. It
was an extrapolation of global temperatures
based on a sample. And it has been entirely
busted. -emarkp
\_ So your argument is: Hockey stick
extrapolation incorrect. Therefore,
Iraq casualty extrapolation incorrect.
Sounds reasonable to me. -nivra
\_ Take emarkp's post: Of the 6 points he made, 3 were blatantly
invalid: (3) undermines his point, (4) is a non-point, (5)
reflects a failure to understand the study, 2 were directly
addressed in the study: (2) & (6) as quoted above. That leaves
the CI criticism (1), which is a valid criticism of the
"repeated talking point," and not of the "study," since both
times the study used the 98k number, it was immediately followed
by the 95% CI. Furthermore, the CI still reflects a 75%
certainty that there were over 68k dead in Iraq at the time of
the study. -nivra
\_ The very wide CI indicates a severe weakness of the study. Your
assertion that my points are invalid doesn't make it so. -emarkp
\_ At least Kaplan wasn't dense enough to make this argument.
The CI shows the inherent difficulties in war-related
epidemiological studies. It's not a "weakness of the study."
He, at least, understand the difference between the study
and the security situation in Iraq. The CI, if anything,
shows the integrity of the study - it accurately reflects
the difficulty of obtaining a precise estimate. -nivra
\_ Amend my comment to "weakness of selecting any one value"
as the mortality. -emarkp
\_ The study quotes "98k" twice; both times immediately
followed by the CI. Well, gee, what value do
researchers normally choose? How about a mean? -nivra
the difficulty of obtaining a precise estimate. |