Exchange with John Rentoul re Death Toll from Iraq War
By Joe Emersberger at Oct 20, 2013
It is worth summing up the best available studies of the death toll from the war (i.e. studies based on scientific sampling of the Iraqi population and published in journals after being subjected to peer review)
2006 Lancet study -------------------650,000 deaths by June 2006
2008 Iraqi government / WHO study--400,000 deaths by June 2006
(estimate made by Mohamed Ali, lead author of study, a few months after it was published)
2013 PLOS Medicine study------------500,000 deaths by June, 2011
You wrote in your blog that the new Iraq moralilty study is "not peer-reviewed".
However in the link to the journal article that you provide - right up near the top left corner of the opening page (just above the title), it says that the article is both "open access" and "peer-reviewed"
Why did you say that the study was not peer-reviewed?
Rentoul immediately acknowledged the error and corrected it. I then sent him this follow up note.
I’m glad you corrected that promptly.
One more thing - your post strongly suggests that ALL estimates within
the confidence interval are equally probable.
In fact the central estimate is the most probable and is several times
more probable than upper and lower bounds of the confidence interval.
The following is a crude illustration of the point, the cluster data
distribution in the study is more complex, but the basic logic holds up
as I’ve confirmed with a epidemiologist.
Consider this graph of a probability distribution:
If I take a very narrow slice of area under the curve at the mean and
compare it to an equally narrow slice of area at -2sigma, then the
ratio of the two areas will basically be the ratio of the two heights:
0.4/.05 = 8
In other words the central estimate is 8 times more probable than the
estimate at plus or minus 2 sigma.
Dear Mr Emersberger
Thank you for pointing out my error in the original post.
As for the confidence interval, I hope I suggested no such thing. You are of course quite right about the shape of a normal distribution, and your points would be valid were there no other independent sources of evidence. Fortunately there is other evidence, and the Iraq Body Count and Iraqi Family Health Survey both suggest a figure below the middle of the distribution.
Please feel free to reproduce this email on Media Lens message boards.
Hi again Mr. Rentoul
My point is off to the side of whether or not you believe other studies
or IBC's count show that the TRUE value is higher or lower than the
Readers should not be led to believe that this study says that any
value within the upper and lower bound of the CI is equally probable.
They shoud be told that the central estimate is the most probable
ACCORDING TO THIS STUDY. You clearly did NOT do that when you wrote
"Probability surveys of this kind produce a range of estimates. In this
case, the authors say that the 95% confidence interval for excess
deaths is 48,000–751,000 (it is 95% likely that the true number lies
between these numbers). "
and when you wrote in your opening sentence that the study found the
"death toll from violence might have been around 250,000, to the
nearest 200,000 or so."
How can readers possibly learn from what you wrote that the central
estimate is the most probable according to this study?
It is one to thing to make arguments in favor of a lower or higher
number than this study estimates based on other information.
It is something else entirely to mislead people as to what the study
Bear in mind, according to a Comres poll done in May, 59% of the UK
public believes that fewer than 10,000 Iraqis (civilains and combatans
combined) died as a result of the war.
Only 16% estimated more than 100,000.
To put it kindly, the UK media has been less than diligent about
informing peolpe about the death toll from the war.