Lancet-1 study vindicated by NEJM/WHO study?

[FireGarden]

This could also go in Science, since it would be nice to have a stats comment. But most of the discussion has been in the politics section.


Using the comments from:
http://scienceblogs.com/deltoid/2008/01/ifhs_study_on_violent_deaths_i.php

Regarding the new study published in the NEJM:

According to Table 4 from the Supplementary Materials:

Mortality Rate All causes Pre-Invasion = 3.17
Mortality Rate All causes for Mar03 to Dec04 = 5.92

The excess death for Mar03 to Sep04 period ( = Lancet 1):

(5.92-3.17)/1000 * (17.8/12) * 24400000 = 99531

The Lancet came up with 100,000 excess deaths for that period. But labelled most as being violent.


Also, see the comments regarding the 400,000 figure for total excess deaths implied by the new study. In this case, Lancet-2 again counted most of the excess deaths as being violent.

 

[casebro]

I just don't think that a poll is the way to count something physical, vs attitudes. And it is only good for attitudes because we have no alternative.

Good science is repeatable.

Now, if there was some kind of back-up statistics for the death toll poll, it would be more believable. Like mens shoe sales statistics dropping by 10%- cigar sales? fishing pole sales? Something to show that there are 10% fewer men in Iraq? Birth rate perhaps (fewer 'fathers' around)? Enrollments in seminary schools? Some ratio change in male vs female apparel? Just anydamnthing that can be physically counted, to supply a statistic that would verify the poll.

How about in addition to asking "how many men died in this household" the pollsters also ask "how many live here now- men and women?". That ought to show a drop in the ratio of men/women.

We get nuthin. From nobody. Just one unverified figure. from interviews. not from physical count.

 

[FireGarden]

Taking a poll is repeatable.
And the method is not just for opinions. You can't ask everybody what their shoe size is. So you aim for an estimate of that real average.

I don't feel the numbers above are cherry-picked. Maybe I'm being misled. But they seem to estimate the same things, and seem to come up with the same value.

The only thing that's inconsistent is the proportion of violent to non-violent deaths. The estimate of total excess deaths is very consistent between Lancet-1 and NEJM/WHO. I can't call that coincidence.

 

[casebro]

But if I wanted shoe size data, I'd go to the shoe manufacturers. not the known-to-fib consumers. At least for some back up verification. Of which there ain't none from Iraq.

 

[Darth Rotor]

The Lancet came up with 100,000 excess deaths for that period. But labelled most as being violent.

What is an excess death?

DR

 

[Garrette]

What is an excess death?

DR

If I read it correctly, it is deaths beyond the number expected given the baseline deaths in the year prior to the war. e.g., (using numbers I'm plucking out of the air), if there were 5 deaths per thousand in the year prior to the war and 7 deaths per thousand each year since the war, excess deaths amount to 2 per thousand each year.

Of those 2 per thousand, they are subcategorized as either violent or nonviolent.

It's also done in reverse so that violent deaths per thousand per year since the war are compared to violent deaths per thousand per year prior to the war.

 

[Darth Rotor]

If I read it correctly, it is deaths beyond the number expected given the baseline deaths in the year prior to the war. e.g., (using numbers I'm plucking out of the air), if there were 5 deaths per thousand in the year prior to the war and 7 deaths per thousand each year since the war, excess deaths amount to 2 per thousand each year.

Of those 2 per thousand, they are subcategorized as either violent or nonviolent.

It's also done in reverse so that violent deaths per thousand per year since the war are compared to violent deaths per thousand per year prior to the war.

Thanks for the clarification.

The use of the term "excess" troubles me.

The last time I checked, war tends to create death, and in particular violent death, so calling this excess is an absurdity, from where I sit.

It's rather expected, when that nasty old war thing shows up, this violent death thing. When it's a civil war, things tend to be even bloodier.

Let me show you how I view excess death.

If Firegarden dies, it's a death.

If you die, it's an excess death.

DR

 

[geni]

But if I wanted shoe size data, I'd go to the shoe manufacturers.

Would work if there were a fairly small number of shoe manufacturers in iraq minimal imports and few people doing shoe repair. Otherwise no.

 

[casebro]

Would work if there were a fairly small number of shoe manufacturers in iraq minimal imports and few people doing shoe repair. Otherwise no.

Okay, fine. You, or Lancet, come up with some corroborating evidence.

Ya know, some verification using a different method. Marriage license numbers have dropped, due to a dearth of eligible males? Anything?

 

[geni]

Okay, fine. You, or Lancet, come up with some corroborating evidence.

Ya know, some verification using a different method. Marriage license numbers have dropped, due to a dearth of eligible males? Anything?

Marriage licenses? In Iraq? Anyway you appear to be forgetting certian elements of the shia belife system that would probably keep marriage rates steady.

 

[FireGarden]

Let me show you how I view excess death.

If Firegarden dies, it's a death.

If you die, it's an excess death.

DR

Well, cheers.
Now that you've widened your vocabulary (not necessarily in line with everyone else) maybe we can get back to the question of whether or not Lancet-1 is vindicated by the new study.

The excess violent deaths from Lancet-1 were 60,000 according to a poster on scienceblogs. (I can't find the necessary figures in the pdf I've downloaded). That's excluding Fallujah, the data for which was regarded as an outlier by the Lancet-1 study.

From the new NEJM report: "...the IFHS data indicate that every day 128 persons died from violence from March 2003 through April 2004".

Lancet-1 covered 17.8 months, that's slightly outside the range given. The average rate (according to NEJM) drops to 115 violent deaths per day for the next given period.

128*(365)*(17.8/12)=66,186

So as a rough and ready estimate, we have two very independent surveys saying pretty much the same thing.

Is Lancet-1 vindicated or not?
It was trashed when it first came out: Too small, main-street bias and even accusations of fraud.

 

[casebro]

Marriage licenses? In Iraq? Anyway you appear to be forgetting certian elements of the shia belife system that would probably keep marriage rates steady.

Like what- polygamy?

OH, I assume that in a country without a death registry at all that there would be only poor marriage counting too. But if somewhere between 100,000 and 800,000 'mostly males' are missing, there ought to be some kind of measurable-for-the-sake-of-verification impact on the economy. Something that does not depend on opinion polls. I've given enough ideas that I can come up with. There have been plenty of doubters about the counts. Why has NOBODY come up with any verifictaion? Even the polling group has not verified their numbers in any way.

 

[geni]

Like what- polygamy?

amoung other things.

OH, I assume that in a country without a death registry at all that there would be only poor marriage counting too. But if somewhere between 100,000 and 800,000 'mostly males' are missing, there ought to be some kind of measurable-for-the-sake-of-verification impact on the economy.

With the unemployement rate that seems doubtful.

Something that does not depend on opinion polls. I've given enough ideas that I can come up with.

And every one has been trivial to shoot down.

There have been plenty of doubters about the counts.

Appeal to popularity logical fallacy there.

Why has NOBODY come up with any verifictaion?

Because there is no worthwhile record keeping in raq that yo can build on.

Even the polling group has not verified their numbers in any way.

There have now been two seperate polls. You are free to conduct a third.

 

[Praktik]

So you can see by my post count that I'm a relative newby here. I think I'm starting to appreciate the culture here - so I can see why things like the Lancet study are good fodder for a discussion of methologies and statistics, and the political aspects given to numbers and so on.

Call me a bleeding heart pinko if you will - but can we all agree that there's been a terrible toll in Iraq as a direct (and indirect!) result of the decision to invade?

This may seem a diversion from a discussion primarily focused on the numbers side - and I guess the emotional part of me wonders sometimes what the answer to that question might be.

I'm going to go get some kleenex and get ready for my 28th viewing of The Bridges of Madison County, carry on...

 

[FireGarden]

But if somewhere between 100,000 and 800,000 'mostly males' are missing, there ought to be some kind of measurable-for-the-sake-of-verification impact on the economy.

What proportion of refugees are men?

Do you have an explanation as to how two independent studies came up with such similar numbers? Or do you just call it coincidence?

 

[FireGarden]

So you can see by my post count that I'm a relative newby here. I think I'm starting to appreciate the culture here - so I can see why things like the Lancet study are good fodder for a discussion of methologies and statistics, and the political aspects given to numbers and so on.

I'm interested in knowing how good the studies are. It helps to post ideas on a forum like JREF so that others can give their ideas.

The debate can seem heartless and self-serving, but I think it's better than talking to myself! (most of the time, anyway)

Call me a bleeding heart pinko if you will - but can we all agree that there's been a terrible toll in Iraq as a direct (and indirect!) result of the decision to invade?

That has been agreed. A few times. Even by some of those who support the war.

 

[casebro]

Do you have an explanation as to how two independent studies came up with such similar numbers? Or do you just call it coincidence?

No coincidence. Both studies used the same methodology. Polls of a very small percentage of the population. If their method is flawed, no matter how many times they 'count', the answer is still wrong.

 

[UserGoogol]

No coincidence. Both studies used the same methodology. Polls of a very small percentage of the population. If their method is flawed, no matter how many times they 'count', the answer is still wrong.

What? Polling a very small percentage of the population is how all polls work. Lancet may have methodological flaws, but if you're saying that polling a very small percentage of the population is methodologically unsound, then you're saying that the entire field of statistics is flawed.

 

[casebro]

What? Polling a very small percentage of the population is how all polls work. Lancet may have methodological flaws, but if you're saying that polling a very small percentage of the population is methodologically unsound, then you're saying that the entire field of statistics is flawed.

We don't use the polls for elections, we use actual ballots. And how many polls were wrong in the last couple elections?

 

[FireGarden]

No coincidence. Both studies used the same methodology. Polls of a very small percentage of the population. If their method is flawed, no matter how many times they 'count', the answer is still wrong.

So you're saying that all their errors added up to the nearly the same wrong answer? And that there is nothing surprising about that?

We don't use the polls for elections, we use actual ballots. And how many polls were wrong in the last couple elections?

Attitudes can change quite abruptly (due to crying women, or whatever). Tears don't change who's dead.