WikiLeaks Analysis Suggests Hundreds of Thousands of Unrecorded Iraqi Deaths
Dan Clore
http://groups.yahoo.com/group/smygo
http://www.informationclearinghouse.info/article27625.htm
WikiLeaks Analysis Suggests Hundreds of Thousands of Unrecorded Iraqi Deaths
By Les Roberts
March 07, 2011
"ICH"
Imagine that the New York Times revealed that five Senators were known
to be taking bribes from a particular corporation. Some days later the
Washington Post runs a story saying they had independent sources
suggesting that four Senators were taking bribes from that same
corporation but goes on to state that this was nothing new as the story
was already covered, neglecting to mention that three of the four names
were different than those previously reported by the Times. This is hard
to imagine because eight named Senators in a scandal is not the same as
five named Senators, and because healthy competition between papers
would tend to point out the information missed by a rival. Yet, this is,
at least numerically, what happened following the October 22nd, 2010
release of the Iraq War Logs by WikiLeaks.
The release which supposedly included over 391,000 classified DoD
reports described violent events after 2003 including 109,000 deaths,
the majority (66,000) being Iraqi civilians. At the time of the release,
the most commonly cited figure for civilian casualties came from
Iraqbodycount.org (IBC), a group based in England that compiles press
and other descriptions of killings in Iraq. In late October, IBC
estimated the civilian war death tally to be about 104,000. Virtually
all authorities, including IBC themselves, acknowledge that this count
must be incomplete, although the fraction missed is debated. The press
coverage of the Iraq War Logs release tended to focus on the crude
consistency between the number recorded by WikiLeaks, 66,000 since the
start of 2004, and the roughly 104,000 recorded deaths from
Iraqbodycount since March of 2003. The Washington Post even ran an
editorial entitled, “WikiLeaks’s leaks mostly confirm earlier Iraq
reporting” concluding that the Iraq War Log reports revealed nothing new.
A research team from the Columbia University Mailman School of Public
Health released a report this week analyzing the amount of overlap
between the 66,000 WikiLeaks reports and the previously known listing of
IBC. The team developed a system for grading the likelihood that the
WikiLeaks War Log record matched an entry in IBC, scoring the match
between 0 (not a match) to 3 (very likely a match). The matching records
were graded by at least two reviewers and then a third reviewer
arbitrated any discrepancies. The conclusion? Only 19% of the WikiLeaks
reports of civilian deaths had been previously recorded by IBC. With so
little overlap between the two lists, it is almost certain that both
tallies combined are missing the majority of civilian deaths, suggesting
many hundreds of thousands have died.
On some level, not noticing that the WikiLeaks list of 66,000 deaths
were different events than those previously recorded by IBC is somewhat
understandable. Reporters have precious few hours to read, assess, reach
out to experts, and then produce copy on the topic of the day. It takes
several minutes to review a particular War Log and then go to the public
database on Iraqbodycount.org and see if on that specific day there was
an event that seems to match the War Log description. In fact, many
papers ran an AP wire article on the WikiLeaks release so it is likely
very few reporters actually looked at the Iraq War Logs.
On the other hand, WikiLeaks gave these records in advance to five
papers including the New York Times and it took the Columbia University
team just minutes to realize that for most events reported outside of
Baghdad (where matching takes more work) there were no reported killings
in a particular city or province on that day within IBC’s database.
This is not the first time this topic has been inadequately covered by
the US press. A study I coauthored in The Lancet in estimating 100,000
excess deaths by September of 2004 (an estimate confirmed three times
since then) received extraordinary press coverage almost everywhere in
the world, but almost none within the US. Project Censored cited the
topic of Iraqi civilian deaths as the second most under-reported topic
of 2004. A survey by researchers from Johns Hopkins University suggested
there had been 600,000 deaths due to the invasion by mid-2006. A poll by
the Opinion Research Business in late 2007 put the tally over 1 million.
Both estimates were viciously attacked by critics, largely supported by
experts in their respective disciplines, but consistently labeled as
“controversial” by the press.
The implications of the WikiLeaks Iraq War Logs for the US standing in
the Middle-East are profound. The only public estimate of the Iraqi
death toll ever provided by the US was President Bush’s response at a
public forum in December of 2005 in which he said, "I would say 30,000
more or less have died as a result of the initial incursion and the
ongoing violence against Iraqis," with the Whitehouse spokesmen later
attributing this estimate to media accounts. This number matched the IBC
estimate at that time. WikiLeaks’ War Logs suggest the US had
information to know that this estimate was only a small fraction of the
reality.
See also Do WikiLeaks and Iraq Body Count tell the same story? No! A
Comparison of the Reports of Iraqi Civilian Deaths - Les Roberts - Full
report [http://www.brussellstribunal.org/pdf/wikileaks_report_march_5.pdf ]
Les Roberts is a Clinical Associate Professor in the Program on Forced
Migration and Health at Columbia University.
http://www.brussellstribunal.org/
--
Dan Clore
New book: _Weird Words: A Lovecraftian Lexicon_:
http://tinyurl.com/yd3bxkw
My collected fiction: _The Unspeakable and Others_
http://www.lulu.com/product/paperback/the-unspeakable-and-others/6124911
Lord Weÿrdgliffe & Necronomicon Page:
http://tinyurl.com/292yz9
News & Views for Anarchists & Activists:
http://groups.yahoo.com/group/smygo
Skipper: Professor, will you tell these people who is
in charge on this island?
Professor: Why, no one.
Skipper: No one?
Thurston Howell III: No one? Good heavens, this is anarchy!
-- _Gilligan's Island_, episode #6, "President Gilligan"
ishikawa
Body Count (IBC) and Wikileaks worked together on Wikileaks' Iraq
logs, the result being an analysis which found there was an 81%
overlap between the Wikileaks logs and IBC's data.
If the overlap is indeed that high, then the conclusions about the
number of dead would be more in line with the WHO estimate of 151,000
violent deaths - a point made by 'Science' journal (18 January 2008,
p. 273).
The 'Columbia University' study cited by Les Roberts, above (and on
which Roberts bases all his conclusions), seems to have been conducted
by Roberts himself, with his students (although the authors aren't
clearly noted on the paper). What would explain the large discrepancy
between the IBC/Wikileaks estimate of an 81% overlap and Roberts's
claim of a 19% overlap? Well, it looks as if Les Roberts and his
students have decided to pretty much automatically classify all the
data which IBC got from morgues (rather than press reports) as "no
match" or "unlikely match" (at best). That's obviously a problem.
Dan Clore
Very interesting information. Could you provide a link to the
IBC/WikiLeaks study?
--
Dan Clore
New book: _Weird Words: A Lovecraftian Lexicon_:
http://tinyurl.com/yd3bxkw
My collected fiction, _The Unspeakable and Others_:
http://www.lulu.com/product/paperback/the-unspeakable-and-others/6124911
Lord Weÿrdgliffe & Necronomicon Page:
http://tinyurl.com/292yz9
News & Views for Anarchists & Activists:
http://groups.yahoo.com/group/smygo
Strange pleasures are known to him who flaunts the
immarcescible purple of poetry before the color-blind.
-- Clark Ashton Smith, "Epigrams and Apothegms"
ishikawa
>
> Very interesting information. Could you provide a link to the
> IBC/WikiLeaks study?
IBC published the figures estimating the overlap:
http://www.iraqbodycount.org/analysis/numbers/warlogs/
It was well covered by the media, because Julian Assange mentioned it
in press conferences, eg:
"Working with Iraq Body Count, we have seen there are approximately
15,000 never previously documented cases of civilians who have been
killed by violence in Iraq." (Julian Assange) http://tinyurl.com/5szpvoh
I think Les Roberts must be aware of this, so it's odd that he doesn't
mention it, especially when he makes bizarre comments such as: "On
some level, not noticing that the WikiLeaks list of 66,000 deaths were
different events than those previously recorded by IBC is somewhat
understandable."
But then Roberts has a history of misrepresenting and/or
misinterpreting IBC's data (he even acknowledged that he was in error
once, when called on it*). See, for example a previous "study" that he
conducted with his students, and misleading statements he made about
its results: http://tinyurl.com/les-roberts
Further work needs to be done to verify who is closest to the truth
about the overlap, but based on previous history (and on the obvious
fact that IBC know their own data better than Roberts's students, and
that IBC worked in collaboration with Wikileaks, and were present at
the main televised Wikileaks press conference), I would put my money
on IBC.
Roberts, it should be noted, is an epidemiologist, so this type of
comparison of datasets isn't really within his field of expertise (or
even experience).
* Les Roberts's error (just under half way down article): http://tinyurl.com/yjpjnfd
Dan Clore
> On Mar 8, 7:07 pm, Dan Clore <cl...@columbia-center.org> wrote:
>> Very interesting information. Could you provide a link to the
>> IBC/WikiLeaks study?
>
> IBC published the figures estimating the overlap:
> http://www.iraqbodycount.org/analysis/numbers/warlogs/
>
> It was well covered by the media, because Julian Assange mentioned it
> in press conferences, eg:
>
> "Working with Iraq Body Count, we have seen there are approximately
> 15,000 never previously documented cases of civilians who have been
> killed by violence in Iraq." (Julian Assange)
> http://tinyurl.com/5szpvoh
Thanks. I think I've read some of this before. Very interesting to
compare with Roberts' piece.
ishikawa
>
> Thanks. I think I've read some of this before. Very interesting to
> compare with Roberts' piece.
>
Indeed. I've just had a look at some of the bigger Wikileaks incidents
(eg over 50 deaths) which Les Roberts and his students claim have "no
match" with IBC's data. It seems that Roberts hasn't been very
thorough (I'm being charitable here). For example, there's a Wikileaks
war log which documents 50 dead, involving a truck, and damage to a
mosque, dated 19 June 2007: http://213.251.145.96/id/43C86B72-0546-B08C-CF0ED9AC57F6AA84
No match with IBC? Well, check that date (19 June 2007) in IBC's
database, and you find this entry: http://www.iraqbodycount.org/database/incidents/k6733
IBC record a greater number of deaths for the incident, but it's a
truck bomb targeting a mosque - on the same date. And some of the
other details suggest a definite match (eg the IBC "gas cylinders", or
"TRUCK FILLED WITH %%% TANKS" in the Wikileaks log).
On their spreadsheet, Roberts and his students put this down as
"0" ("no match") - indeed it's one of the *biggest* incidents that
they classify as "no match". And what about examples that aren't as
clear - and that require more of a judgment call from Roberts's
students (who perhaps aren't particularly adept at interpreting IBC's
data?). Bear this in mind, and then read the third paragraph in Les
Roberts's article (above) again. The "grading system" and the
"reviewing" that Roberts talks about basically means Roberts's
students making judgment calls in many cases, given the type of data
involved. And from the rather shaky premise of the study, Roberts
concludes that: "it is almost certain that both tallies combined are
missing the majority of civilian deaths, suggesting many hundreds of
thousands have died".
I don't think this is the type of study on which to base claims of
"near certainty" on such issues.
Dan Clore
I don't think that that's actually a match. There's the same or very
similar MO, and both reports give the target of the attack as a Shiite
mosque in Baghdad. But not only is the number of casualties different
(50 killed versus 86-87) the name of the mosque given is different --
Khillani versus Ghai Lani. Using the Antiwar.com archives, I've found a
Yahoo report that is definitely the same incident as IBC reports (IBC
doesn't list it as a source), and it gives the name of the mosque as
Khulani. Khillani and Khulani are obviously just different
transliterations, but Ghai Lani looks different enough that it seems
unlikely to be another transliteration of the same name, though it's
possible. The Yahoo report also gives the number wounded as over 200,
whereas the WikiLeaks report gives the number wounded as 125.
But here's the real kicker: The WikiLeaks report gives the time of issue
as 07:50:00. The IBC report gives the time of the incident as 2:00 p.m.
Now, these WikiLeaks reports use the military 24 hour system rather than
a.m. and p.m. So if the WikiLeaks report were for the same incident as
the IBC report, it would have to be issued some time after 14:00:00.
So, I conclude that these are clearly two separate incidents, if the
information is at all accurate. It's far from unknown for insurgents to
coordinate attacks using the same MO at around the same time, so I
wonder if that isn't the case here.
--
Dan Clore
New book: _Weird Words: A Lovecraftian Lexicon_:
http://tinyurl.com/yd3bxkw
My collected fiction, _The Unspeakable and Others_:
http://www.lulu.com/product/paperback/the-unspeakable-and-others/6124911
Lord We˙rdgliffe & Necronomicon Page:
ishikawa
>
> So, I conclude that these are clearly two separate incidents, if the
> information is at all accurate. It's far from unknown for insurgents to
> coordinate attacks using the same MO at around the same time, so I
> wonder if that isn't the case here.
I'm not convinced by this, at all.
It would require that there were two such incidents on the same day,
with the same details (50+ civilian deaths, mosque, truck filled with
gas cyclinders) - one of which somehow escaped the notice of anyone
(except the US military), and the other recorded by multiple media
sources (IBC have made the point that large incidents of this type
generate a lot of media coverage). Smaller incidents may be totally
overlooked, but not these large ones - that's stretching credulity too
much, I think.
To simply dismiss it as "no match" seems seriously problematic for a
study of this type.
Joshua Dougherty
> I don't think that that's actually a match. There's the same or very
> similar MO, and both reports give the target of the attack as a Shiite
> mosque in Baghdad. But not only is the number of casualties different
> (50 killed versus 86-87) the name of the mosque given is different --
> Khillani versus Ghai Lani. Using the Antiwar.com archives, I've found a
> Yahoo report that is definitely the same incident as IBC reports (IBC
> doesn't list it as a source), and it gives the name of the mosque as
> Khulani. Khillani and Khulani are obviously just different
> transliterations, but Ghai Lani looks different enough that it seems
> unlikely to be another transliteration of the same name, though it's
> possible. The Yahoo report also gives the number wounded as over 200,
> whereas the WikiLeaks report gives the number wounded as 125.
>
> But here's the real kicker: The WikiLeaks report gives the time of issue
> as 07:50:00. The IBC report gives the time of the incident as 2:00 p.m.
> Now, these WikiLeaks reports use the military 24 hour system rather than
> a.m. and p.m. So if the WikiLeaks report were for the same incident as
> the IBC report, it would have to be issued some time after 14:00:00.
>
> So, I conclude that these are clearly two separate incidents, if the
> information is at all accurate. It's far from unknown for insurgents to
> coordinate attacks using the same MO at around the same time, so I
> wonder if that isn't the case here.
Um Dan, those are the same incident. Khulani, Khillani, Khulani and
Ghai Lani are all just transliterations. I've worked with iraqi place
names and transliterations like this for years now so am familiar with
it. There is nothing unusual about differences like those above. They
all are the same word, just with a difference in whether the sound of
the Arabic word should start with an English 'Kh...' or 'Gh...' sound.
In this case, the press mostly decided to write it up in English with
a 'Kh', while the US military decided to use 'Gh' in this log.
Also, your "real kicker" about the time of day is just some kind of
mistake on the website ishikawa cited. WikiLeaks provided IBC with the
unredacted logs to work with. The date and time given in the log is:
"Date: 2007-06-19 13:50:00"
And the log gives a detailed timeline in the text. These numbers get
redacted on the website, but you can confirm 1:50PM from the
unredacted text:
"TIMELINE:
1350: ROMEO ELEMENT CALLS STRIKE TO REPORT A LARGE EXPLOSION ON RTE
WILD THEY BELIEVE IT TO BE A VBIED
1432: SHIA MOSQUE GHAI LHANI WAS TARGETED BUT THERE IS A REPORT THAT
IT POSSIBLY TARGETED AN MP PATROL ALSO
1458: UPDATE FROM THE UNIT ON THE GROUND IS THAT THE DRIVER WAS INSIDE
OF A TRUCK FILLED WITH PROPANE TANKS WHEN HE TRIED TO CROSS THE CURB
THE TRUCK GOT STUCK AND WAS DETONATED THERE.
1800: BJCC REPORTS 50X KILLED AND 125 X WOUNDED"
The time of 7:50 AM never appears anywhere in the actual log, so i
don't know why it says that on that website. Perhaps there is some
kind of bug with the dateline on that site. You can however see again
that 1:50PM is correct in the 'tracking number' column on the left on
the website:
"20070619135038SMB"
That means 2007, June 19 at 1350 (or 1:50 PM)
But even if there had been a disparity with the time of day between
the two sources that wouldn't necessarily mean it didn't match, it
could just mean that a media report got the time wrong.
And really Dan, think of how ludicrous your conclusion is. There is no
other WL log entry with a large number killed like this in on that
day. This is the only one. So if these are two separate incidents
(bomb 1 and bomb 2), then the US military notices huge truck bomb 1
but fails to notice the huge truck bomb 2. At the same time, the press
widely reports huge truck bomb 2 (to the tune of 54 separate reports),
while not one of these reports, many quite detailed, notice or mention
the other huge truck bomb 1, that is just like it at a different
mosque that day. This is just lunacy.
I've also done some further checking with the Roberts study. The cases
where they say there is no match are wrong most of the time. But for
example, look at the four biggest incidents they claim have no IBC
match. The big incidents are typically the easiest to find matches, as
there usually are not multiple huge things that could be confused. One
of these is the incident ishikawa brought up. So how about the next
three, using the same website as ishikawa for the WL logs?
1) http://213.251.145.96/id/400F2C60-E4FC-16ED-630FB2A7A107D5A8/
66 by car bomb in Karbala
This is the match:
http://www.iraqbodycount.org/database/incidents/k6218
IBC gives a sowewhat higher number but same area, same day. Was there
another car bomb in the same area that day killing over 60 people that
nobody noticed? Right! Apparently so.
66 unrecorded deaths in IBC (to then be multiplied exponentially with
'capture-recapture' techniques later).
2) http://213.251.145.96/id/11812
65 by suicide car bomb at KDP and PUK Headquarters near Arbil.
That's this:
http://www.iraqbodycount.org/database/incidents/k021
Again, IBC gives a higher number, but again was there a second,
different suicide bomb that killed over 60 people in the same area on
the same day with the same target? Or is it perhaps just that the war
log is an early field report and the IBC entry is based on later
updated casualty totals?
No, that's not plausible. 65 more unrecorded deaths.
3) http://213.251.145.96/id/390B3593-053C-B45E-345FE49337F39672
55 by car bomb near Karbala. This is just another report about the
first bombing (the logs often have multiple entries for the same
thing, this is not unusual). Or... were there now two additional car
bombings in the same area that day that nobody noticed, while noticing
the one that IBC has. Must be. That's another 55 unrecorded deaths.
The matching is just incompetent.
Joshua Dougherty
> The time of 7:50 AM never appears anywhere in the actual log, so i
> don't know why it says that on that website. Perhaps there is some
> kind of bug with the dateline on that site. You can however see again
> that 1:50PM is correct in the 'tracking number' column on the left on
> the website:
I did a little more checking to figure out what is going on with the
times on that website, and there is definitely some kind of systematic
bug going on with the time they give under the headline.
Looking at the Karbala log here:
http://213.251.145.96/id/400F2C60-E4FC-16ED-630FB2A7A107D5A8/
It has the same exact problem with the time. The time given under the
headline reads "2007-04-28 13:15:00", or 1:15 PM. But the time given
in the tracking number is "20070429191538SLA" or 7:15 PM. The time it
gives under the headline is 6 hours too early, the same exact error as
in the previous Baghdad log. So it seems like there's some kind of bug
there shifting the times down by 6 hours somehow.
The unredacted log says, "Date: 2007-04-28 19:15:00", or 7:15 PM. The
log never says 1315 or 1:15 PM. IBC places the time as 7:10 PM.
Further, the unredacted summary says the bomb "DETONATED AT THE AL
ABBAS INTERSECTION", and IBC says "checkpoint near Imam Abbas shrine",
which further reinforces that this is an unambiguous match.
Joshua Dougherty
> The 'Columbia University' study cited by Les Roberts, above (and on
> which Roberts bases all his conclusions), seems to have been conducted
> by Roberts himself, with his students (although the authors aren't
> clearly noted on the paper). What would explain the large discrepancy
> between the IBC/Wikileaks estimate of an 81% overlap and Roberts's
> claim of a 19% overlap? Well, it looks as if Les Roberts and his
> students have decided to pretty much automatically classify all the
> data which IBC got from morgues (rather than press reports) as "no
> match" or "unlikely match" (at best). That's obviously a problem.
Indeed. We've already seen that their matching is a big part of the
problem, but their assumptions are also a big part. They use extremely
liberal assumptions so that, effectively, morgue entries in IBC
tallying thousands of reported deaths don't exist. So if a log says
police find a body in Baghdad, Roberts assumes this never (0) or
almost never (1) goes to the morgue even though that is the main place
where they take them. So he is inflating the numbers by using extreme
assumptions that push the numbers skyward.
Moreover, they aren't even reading the full text of the logs, and so
don't notice that in many cases their assumption is just demonstrably
false. They say they downloaded the logs from the Guardian, but the
Guardian does not include the text summaries. So they are basically
just reading the headline, columns and coordinates. But if you
actually read the summaries you'd see that quite often the logs state
explicitly that a particular body found was taken to the morgue. They
code all of these as 0 ("no match") or 1 ("unlikely match") even
though the correct coding here is 3, since those cases are just
unambiguous matches to the IBC morgue entries.
They seem to apply this assumption not just to morgue either, but to
any IBC entry that is in any way aggregated. So if a log says the
police find 1 body in Sadr City, Baghdad, and IBC has an entry saying
the police found 30 bodies around Baghdad that day, this too is coded
as "no match" (or "unlikely match" if they are feeling generous).
So these crazy assumptions would be a big part of the problem.
I like Roberts's spin with the definitions of these number codes too.
They have a "no match" category (0), but there isn't any category of
just "match". The best it can ever be given is only a "high
probability that it is a match" (3). So nothing is ever just a match.
Cute.
Then also look at the double-think going on when it comes time to do
the 'capture recapture' estimate at the end. All of the above
assumptions they use in the matching to drive the 'no match' numbers
skyward effectively assume that IBC and the logs are 'fishing in
different ponds'. This is required if the log deaths almost never
match the morgue deaths or aggregate bodies found entries. But when it
comes time to do the capture recapture, and the goal here is argue
that the final estimate will probably be too low, you need to invert
the previous assumption and assert that IBC and the logs are 'fishing
in the same pond':
"The WikiLeaks Iraq War Logs often attribute reports to Iraqi
officials (estimated at 20% of all reports), the same sources of many
reports cited in the newspaper articles cited by IBC. Many other IBC
reports cite US Government or military sources. Both of these
processes would result in an artificially large overlap between the
two listings."
Double-plus good Winston! The assumptions he used for the matching are
the opposite of this. So what is going on here? Clearly, there is an
agenda here to drive the numbers skyward, to create an illusion that
the WL data somehow supports Roberts's 2006 survey. So assumptions are
chosen to serve that end. In the case of the matching it drives up the
numbers if you assume 'different ponds', but when it comes time for
the capture-recapture estimate, it drives up the numbers if you assume
'same pond'. So you just double-think as needed.
Joshua Dougherty
> The 'Columbia University' study cited by Les Roberts, above (and on
> which Roberts bases all his conclusions), seems to have been conducted
> by Roberts himself, with his students (although the authors aren't
> clearly noted on the paper). What would explain the large discrepancy
> between the IBC/Wikileaks estimate of an 81% overlap and Roberts's
> claim of a 19% overlap? Well, it looks as if Les Roberts and his
> students have decided to pretty much automatically classify all the
> data which IBC got from morgues (rather than press reports) as "no
> match" or "unlikely match" (at best). That's obviously a problem.
Indeed. We've already seen that their matching is a big part of the
Les Roberts
I am not sure that this is a very honest discussion. Given that the
vast majority of 50 death events are in both, and that we had only 1
reviewer on those coded as 0, looking at the biggest "0"s would be the
most efficient way to find a false negative. We say in the report we
are sure this happened, probably more than 100 times.... When you
find 200 such cases, then clearly we need to go back and reanalyze the
data. We expect you to find a few if you look.
A) Please take a random sample of 10 or 20 of the records (not some
selected group likely to be right or wrong) and look for yourself.
B) in Anne Sieglers article a couple years ago the 170+ deaths in
Baghdad showed the vast majority of violent deaths in Baghdad do not
end up in a morgue. Thus, Josh is right, some of the discrepancy is
about the underlying belief that single Baghdad deaths are ending up
in the morgue by the IBC crowd and belief they are not by the Columbia
crowd. But if one removed the 20% of deaths in IBC that are morgue
tallies, and match, we would still come up with hundreds of thousands
and the conclusion that both lists miss the vast majority would remain
unchanged.
Please sample, look, and tells us the percent that are wrong! Outside
of Baghdad it is abundantly clear after just minutes that most Iraq
War Logs are not in IBC.
cheers, Les
Joshua Dougherty
I will be looking more at the data, but I've already checked the
matching in your spreadsheet for the all incidents of 4 or more
deaths. There are 94 such incidents which total 835 deaths. That
amounts to almost a third of the deaths in your sample already. And
there is a reason why I looked at these. They are the easiest to
match. The correct match for the big things tend to jump off the
screen at you, so if you can't get these right there is a big problem.
59 of the 94 are definite direct matches. That is, these are
unambiguos incident level matches, or in just a few cases, morgue
matches where the DoD summary says explicitly that the bodies were
taken to 'Justice Medical', i.e., the Baghdad morgue and are therefore
definite matches to the morgue entries.
So before going any further, you are wrong at least 63% of the time.
This is a big problem.
For the remaining 34 logs:
4 have an issue with civilian-combatant status. Two of these were
deaths of foreign contractors which IBC just doesn't include, and
shouldn't really be counted against IBC. The other two were debatable
in terms of how they should be classified in terms of civ-com status.
5 cases were Baghdad deaths where the log doesn't explicitly say they
went to 'Justice Medical', so they may conceivably not have gone to
morgue. IBC's analysis would have coded these as morgue overlaps, that
is, IBC would add these incidents to the database, but would subtract
the equivalent number from the morgue entry to avoid double counting,
so they would not add 'new deaths' to IBC, but would add the incident-
level details and any victim details present. Your analysis assumes
they are new deaths to add on top of the morgue entry.
19 cases are possible matches where these would be debatable to some
degree what would be the correct conclusion on the matching.
7 are definitely no match.
That is, your codings on these things are wrong _at least_ 63% of the
time, or at worst 93% of the time. And it seems more likely to be
closer to the latter to me. As I said above, these bigger incidents
should be the easiest ones to match correctly, and therefore the least
likely place you should find false negatives. It's hard to see why the
matching wouldn't be just as bad or worse for the smaller things that
often require more care and can often have several possible matches.
But I'll be looking at it more in the future.
"in Anne Sieglers article a couple years ago the 170+ deaths in
Baghdad showed the vast majority of violent deaths in Baghdad do not
end up in a morgue."
I believe you co-authored this paper. The paper does not say what you
claim it does. The only thing it says about morgue is this:
"61 (38%) were absolutely not in the database, meaning the interviewee
was
certain that the body was not brought to the morgue. "
38% is not "the vast majority", Les. Also, you should recall that this
Siegler sample was not really a random sample in the first place, and
it's also a very small sample, so it involves some wishful thinking to
treat various percentages from it as representative of all deaths in
Baghdad. I would also have some reservations about how reliable
answers to such a question would be in this instance. It's not even
clear in the study in what way the interviewee's are related to the
dead people they are describing, how they came to know of these deaths
in the first place, or why they would even be in a position to know
what eventually happened to the bodies of all 10 people killed
"closest to their home" since 2003, that you're asking them to report.
Moreover, the origin of this data on morgue is something of a mystery
as the Siegler paper first explains that you asked the respondents
five questions, and you list the five:
"Interviewees were asked to provide five items of information
regarding each
death: (1) gender; (2) approximate age; (3) date; (4) place; and (5)
circumstance."
But then somehow later in the paper you have answers to a sixth
question about morgue. I also recall that when you corresponded with
IBC some time ago about this very Siegler paper that, at that time,
you had already collected the data and had already promoted some of
the results (though incorrectly) in interviews, but at that time when
IBC brought up the morgue problem, you stated that you did not ask
about morgue and didn't know how many went to morgue. A couple months
later the paper is published and there's a percentage for morgue.
Strange.
One last point on morgue, I looked up the 100 incidents with 1
civilian death each, that were in our sample for our analysis, and I
checked the ones in Baghdad where the only possible match was to
morgue or to an aggregated entry like bodies found. I checked to see
how many of these stated explicitly in the log that they went to the
morgue. I found that 36% of them said this explicitly. This true
percentage must be considerably higher than this since the logs won't
necessarily always state this even when they know it, and often they
just wouldn't have that information to report in the first place.
"Outside of Baghdad it is abundantly clear after just minutes that
most Iraq
War Logs are not in IBC."
This might be true. I'd have to check our sample again to see what we
came to for in vs. out of Baghdad, but be careful not to confuse "most
Iraq War Logs" with "most deaths". After all, it can take 50 or more
Iraq War Logs recording 1 death missed, to equal one big bombing that
isn't missed. So a percentage of logs/incidents and a percentage of
deaths are two very different things.
"But if one removed the 20% of deaths in IBC that are morgue tallies,
and match, we would still come up with hundreds of thousands and the
conclusion that both lists miss the vast majority would remain
unchanged."
I'm not sure off the top of my head if the 20% figure is correct. I'll
take your word for it, but as i noted previously, it appears that
morgue tallies are not the only IBC deaths that are effectively
ignored by your analysis. It appears that anything that has any level
of aggregation is likewise ignored, and not just in Baghdad. This
would include bodies found tallies that cover a whole day or some
period of time. The numbers we have like this are quite large, e.g.:
http://www.iraqbodycount.org/database/incidents/k4435
These kind of entries seem to get treated the same way morgue is
treated. So if you look at a log saying 1 body found shot dead in Sadr
City on that day, and IBC does not have an entry for 1 body found in
Sadr City, but only has the aggregate entry above, it gets coded as 0
or 1. The number of deaths recorded in such entries is quite large.
There are also aggregate entries for many areas and time periods
outside Baghdad too, e.g.:
http://www.iraqbodycount.org/database/incidents/x490
http://www.iraqbodycount.org/database/incidents/d3464
http://www.iraqbodycount.org/database/incidents/d3734a
It seems to me that all of these type of entries get treated the same
as morgue does. As such, the number of IBC-recorded deaths your
approach effectively ignores would be far higher than 20%.
But as I've also shown, the problem here is not just the different
treatment of aggregates. Though that is a big part of the problem
there is also clearly a big systematic problem in the matching
itself.
As a last point, I'd also ask what exactly is it that your students
are supposed to be learning from these little exercises?
Dan Clore
Thanks, Josh. You've provided more than adequate explanations for two of
the three major discrepancies here. (The third is the number of
casualties.) I think that's good enough to conclude that these reports
are about the same incident. There was a reason I qualified my conclusions.
--
Dan Clore
New book: _Weird Words: A Lovecraftian Lexicon_:
http://tinyurl.com/yd3bxkw
My collected fiction, _The Unspeakable and Others_:
http://www.lulu.com/product/paperback/the-unspeakable-and-others/6124911
Lord Weÿrdgliffe & Necronomicon Page:
jmh
Sounds like some time zone issue.
jmh
Joshua Dougherty
> I will be looking more at the data, but I've already checked the
> matching in your spreadsheet for the all incidents of 4 or more
> deaths. There are 94 such incidents which total 835 deaths. That
> amounts to almost a third of the deaths in your sample already.
I did not put this quite right, so to clarify, I looked at all the
incidents of 4 or more deaths which were coded as 'no match' or 0.
Joshua Dougherty
"Moreover, IBC reports could not be any days before the WikiLeaks
event time. (This is because the press reporting process might have
delayed the arrival of reports in IBC. However, since IBC data are
based on press reports, it would be impossible for them to have dates
prior to those recorded in the WikiLeaks data)."
This is just wrong. This seems to confuse the date of death with the
date of the press report. The date IBC uses is meant to refer to the
date of the incident or death, which is often reported as being
earlier than the date of publication of the report.
For example, a report of 10 March may say that someone was killed
'yesterday' or 'last night', then IBC would put the date as 9 March.
But then imagine if it actually happened in the early morning hours of
10 March, and the press report was a little bit off on this, then the
log will say 10 March and IBC would say 9 March.
Also, the DoD logs seem to occasionally be a little off in terms of
date or time themselves, and in some cases may assert that something
happened later than it actually did, in which case the IBC match could
be dated earlier.
I believe there were a couple cases like this in our sample, where the
IBC match was dated earlier than the DoD report. Not very many at all,
so this would not be any kind of huge bias, but it does seem like
another flawed assumption that goes in the same direction of inflating
the non-matches.
On Mar 9, 12:25 pm, Les Roberts <lfroberts...@gmail.com> wrote:
ishikawa
>
> I am not sure that this is a very honest discussion. Given that the
> vast majority of 50 death events are in both, and that we had only 1
> reviewer on those coded as 0, looking at the biggest "0"s would be the
> most efficient way to find a false negative. We say in the report we
> are sure this happened, probably more than 100 times.... When you
> find 200 such cases, then clearly we need to go back and reanalyze the
> data. We expect you to find a few if you look.
It took me a while to decipher this, partly because it's badly worded
("50 death events"), partly because it's logical nonsense (the point
about "the most efficient way" doesn't follow from the premise), and
partly because it doesn't address the point about *incompetence* in
the matching process. A "false negative" (which, in this context,
presumably denotes a failure to find a match where one actually
exists) doesn't necessarily imply sloppiness, but sloppiness would
certainly be required to fail to find the matches that Josh identified
above for those large-scale incidents. This worrying level of
incompetence in the matching process is what Les Roberts should be
addressing.
Joshua Dougherty
> "But if one removed the 20% of deaths in IBC that are morgue tallies,
> and match, we would still come up with hundreds of thousands and the
> conclusion that both lists miss the vast majority would remain
> unchanged."
>
> I'm not sure off the top of my head if the 20% figure is correct. I'll
> take your word for it, but as i noted previously, it appears that
> morgue tallies are not the only IBC deaths that are effectively
> ignored by your analysis. It appears that anything that has any level
> of aggregation is likewise ignored, and not just in Baghdad. This
> would include bodies found tallies that cover a whole day or some
> period of time. The numbers we have like this are quite large, e.g.:
>
> http://www.iraqbodycount.org/database/incidents/k4435
>
> These kind of entries seem to get treated the same way morgue is
> treated. So if you look at a log saying 1 body found shot dead in Sadr
> City on that day, and IBC does not have an entry for 1 body found in
> Sadr City, but only has the aggregate entry above, it gets coded as 0
> or 1. The number of deaths recorded in such entries is quite large.
>
> There are also aggregate entries for many areas and time periods
> outside Baghdad too, e.g.:
>
>
> It seems to me that all of these type of entries get treated the same
> as morgue does. As such, the number of IBC-recorded deaths your
> approach effectively ignores would be far higher than 20%.
To expand a bit on this. The approach the Roberts analysis is usually
in terms of the assumptions is basically equivalent to the one IBC
used for determining proportions of new _incidents_, as opposed to new
_deaths_.
That is, for 'new incidents' IBC assumes that aggregate matches of any
kind are never matches, because the purpose here was to see how much
new incident-level details would be added, not how many new deaths
would be added to the numbers. These findings are discussed here:
http://www.iraqbodycount.org/analysis/beyond/warlogs/
That is the 2nd part of our 3-part analysis, plus accompanying
appendix. The other parts are here:
http://www.iraqbodycount.org/analysis/numbers/warlogs/
http://www.iraqbodycount.org/analysis/qa/warlogs/
http://www.iraqbodycount.org/analysis/numbers/warlogs-appendix
The main finding for the 'Details' piece is:
"IBC estimates that the Iraq War Logs describe hitherto unreported
details of 23,000 violent incidents in which Iraqi civilians were
killed or their bodies were found."
That is on the basis of assuming that Baghdad bodies never match to
morgue, or never match to daily bodies found tallies, or that
assassinations in Basra never match to the big Basra assassination
entry above (x490), etc. etc. etc. It assumes that any and all
aggregate matches are not matches. So this is assuming there are never
any matches but direct and specific incident-level matches.
Even under those assumptions, we come to a total of 23,000 incidents
(or logs), so about half the logs, that will provide new incident
details. This is also almost the total number of incidents already
present in the IBC database (24,212). So our finding is that the
number of incidents in IBC will basically double after all the work is
complete. That is, IBC already had about 50% matching even under the
most extreme of assumptions possible with matching (which are
appropriate if estimating incident percentages, but absurd assumptions
if estimating death percentages.)
Moreover, these new incidents are almost never the big incidents. They
are mostly the logs with 1, 2 or 3 deaths. So even if 23,000 amounts
to about 50% of the logs, it does not amount to 50% of the deaths in
the logs. It will be much less than 50% of the deaths in the logs,
because these 50% of logs are weighted heavily toward the logs with
the fewest deaths each.
Thus there is simply no way for us to even come close to Roberts'
conclusion of only "19%" matching deaths (or even 19% of matching
logs). There are at least around 50% of matching logs, and this
necessarily means much more than 50% matching deaths, even when you
are assuming away all possible aggregate matches. This shows still
further that the problem with Roberts' analysis can not only be the
assumptions about morgue or aggregates, but must also involve a very
serious problem with their matching, entailing a large systematic bias
toward coding things as '0' when the correct coding is '3'.
Les Roberts
> On Mar 9, 2:27 pm, Joshua Dougherty <jdoc135...@gmail.com> wrote:
>
>
>
>
>
> > "But if one removed the 20% of deaths in IBC that are morgue tallies,
> > and match, we would still come up with hundreds of thousands and the
> > conclusion that both lists miss the vast majority would remain
> > unchanged."
>
> > I'm not sure off the top of my head if the 20% figure is correct. I'll
> > take your word for it, but as i noted previously, it appears that
> > morgue tallies are not the only IBC deaths that are effectively
> > ignored by your analysis. It appears that anything that has any level
> > of aggregation is likewise ignored, and not just in Baghdad. This
> > would include bodies found tallies that cover a whole day or some
> > period of time. The numbers we have like this are quite large, e.g.:
>
> >http://www.iraqbodycount.org/database/incidents/k4435
>
> > These kind of entries seem to get treated the same way morgue is
> > treated. So if you look at a log saying 1 body found shot dead in Sadr
> > City on that day, and IBC does not have an entry for 1 body found in
> > Sadr City, but only has the aggregate entry above, it gets coded as 0
> > or 1. The number of deaths recorded in such entries is quite large.
>
> > There are also aggregate entries for many areas and time periods
> > outside Baghdad too, e.g.:
>
>
> > It seems to me that all of these type of entries get treated the same
> > as morgue does. As such, the number of IBC-recorded deaths your
> > approach effectively ignores would be far higher than 20%.
>
> To expand a bit on this. The approach the Roberts analysis is usually
> in terms of the assumptions is basically equivalent to the one IBC
> used for determining proportions of new _incidents_, as opposed to new
> _deaths_.
>
> That is, for 'new incidents' IBC assumes that aggregate matches of any
> kind are never matches, because the purpose here was to see how much
> new incident-level details would be added, not how many new deaths
> would be added to the numbers. These findings are discussed here:http://www.iraqbodycount.org/analysis/beyond/warlogs/
>
> That is the 2nd part of our 3-part analysis, plus accompanying
>
> - Show quoted text -
Hi Josh,
I read your thoughts and last evening we looked at all the events >20
killings and ....depending on how you handle the double report, we
think they are all in IBC. I am not sure what my students are
learning, but I am learning humility.
I can envision three ways why a systematic sample of 200 files taken
with match=0 from the datafile could be showing such a low false
negative rate and why virtually all the largest matches called “0”s
are false. One would change the overall results little, the other two
would change figure 3 and the deaths estimate substantially, clearly
we have some investigation and reanalysis to undertake before I can
make a fully informed comment.
Your reading of the Siegler article is reasonable and wrong. We had a
questionnaire, it was short. After the initial interviews a subset of
these could not be assessed because of the morgue tally issue. So the
students with such reports were asked to skype / call / e-mail back to
their interviewees and ask if the body had gone to the morgue. I
believe (as I read this again after a couple years) 61 said, no the
body did not go to the morgue. An additional 48 could not be
contacted, or said they did not know, or said the body did go to the
morgue. Thus the phrase vast majority was of those that we called
back and that knew.
I was not aware of your analysis until I shared our report last
weekend. I had looked at your website around 10.24 when the 16,000
number was being bantered around but I think the analysis was not yet
up. I suspect you have a lot of false positives. I am starting to
think, at least with the fraction of deaths reported, we
underestimated our false negatives. That would drive our analyses in
opposite directions. We will reanalyze what we did. In the meantime,
I think you can pretty easily measure your false positive rate
associated with the inability to match to morgue data.
If you assume that all events, large and small, have an equal chance
to end up with the morgue (and I suspect this is not true, I suspect
the big events that involve police and body collection vehicles more
likely take bodies to the morgue….but stay with me for a moment) if
21% of your deaths (on Oct 24th anyway) are from morgue tallies, when
you look at the single killings for example, only ~21% of those you
match should be from morgue tallies. That is to say, if for example
24% of single incidents matched, you would expect no more than an
additional 5% of singleton reports to be morgue tally matches.(...of
the overlap, we would expect the incident matches to be 5 times more
than the morgue tally matches)
I will look into our data and get back to you with what we find. If
we have made major errors, we will reanalyze and share the report
through the same people we sent it out with this time and correct the
record. Next week everyone is away, I suspect it will take a couple
weeks to properly reanalyze.
We had two dozen people take just a few War logs from an arbitrary
starting point. Zero of them found more matches than definite not
matches. Only a couple found more maybe + matches than non matches.
I remain very confident that most war logs are not in IBC. Let me
explore what we have done wrong on our quality control efforts….and I
hope you will do the same….and we can have a constructive chat in a
couple weeks.
Cheers, Les
ishikawa
>
> I can envision three ways why a systematic sample of 200 files taken
> with match=0 from the datafile could be showing such a low false
> negative rate and why virtually all the largest matches called “0”s
> are false. One would change the overall results little, the other two
Les, there's a problem of credibility and trust here. You are
promoting fairly extreme claims, such as: "the WikiLeaks list of
66,000 deaths were different events than those previously recorded by
IBC". But the examples we've seen above reveal a basic incompetence in
your attempts to match to IBC records.
Josh makes the highly relevant point that those large incidents should
be the easiest to match. Before you start talking about false negative
rates again, perhaps you could take a step back and explain in
straightforward language why we should take seriously a study which
seems to lack the basic level of proficiency needed to carry out its
main task?
Joshua Dougherty
Thanks Les,
Certainly i think more double-checking is needed.
"I remain very confident that most war logs are not in IBC."
Yes, you are correct about that. In fact the IBC analysis i cited says
that is so. But again we need to avoid confusing log percentages with
death percentages.
And here i must correct my previous posting saying that 50% was the
incident-level (no aggregates) match percentage for logs. I was using
the wrong denominator there to get 50%. The Civilian category has
34,000 logs, and i was thinking of the total log count. So we came to
32.4% logs matched, which led to the conclusion of 23,000 new detailed
incidents for civilians. But again keep in mind that you are referring
here to a percentage of incidents or logs, and not to a percentage of
deaths. These are very different things because the logs that match to
IBC tend to contain more deaths than the logs that do not match.
So the percentage of deaths matched has to be considerably higher than
32.4% matching rate for logs. Working this out quickly, i come to
about 35,000 deaths contained in that 32.4% of matched logs, so around
53% of the deaths in the logs matched even when we were ruling out any
kind of aggregate matches.
And one last point on this would have to do with the capture-recapture
assumptions, and whether one should actually include all of IBC's
aggregate deaths in the totals at the end if you've ruled out matching
to them with the assumptions at the start. If you are effectively
matching only specific incidents to specific incidents, then it would
seem appropriate to remove all the aggregates from the IBC total
before doing the estimate. You suggested that 20% in IBC are morgue,
and there is much more than morgue. I'm not sure exactly how much off
the top of my head, but i'd guess perhaps this would go up to 40% if
including non-morgue aggregates. So then it seems like perhaps the
more appropriate capture-recapture would be 66,000 in WL, vs. maybe
60,000 in IBC, and not 66,000 vs. 90,000 or 100,000. What choice is
made about this would make a pretty big difference in the estimate,
and of course who's matching rate you believe would also make a big
difference.
ishikawa
>
> [...] But again we need to avoid confusing log percentages with
> death percentages.
Yes, it's an important point. Les Roberts's claim of a "19%" overlap
(of Wikileaks and IBC) is probably the figure that will stick in
people's minds after reading Roberts's article. But that "19%" figure
doesn't refer to a percentage overlap of recorded *deaths*. It refers
to logs (ie incident reports).
This is problematic because most public discourse (eg media,
scientific, etc) is about death counts/estimates. Roberts's own Lancet-
published studies, for instance, are about estimating the number of
deaths - not about the number of incidents leading to deaths. IBC have
rich data on both deaths and incidents and, to their credit, are very
clear about distinguishing the two in these matters. I think Les
Roberts should make an effort to be a little clearer when he's talking
about percentages of incidents/logs, as opposed to percentages of
deaths. His study does provide a figure for the claimed percentage
overlap of deaths (31.6%), but he doesn't mention this in his article.
Joshua Dougherty
this I thought I'd post an update. I previously looked at all the
incidents in Roberts' spreadsheet of 4 or more killed that were coded
as '0' or no match. I've since looked at those with 3 killed that were
coded as '0' or no match.
There were 75 logs of 3 killed in the sample for which there is
supposedly no IBC match. I found there were only 21 where there is
clearly no match.
19 were direct incident-level matches
18 were aggregate matches - 4 of which would require morgue alone, the
rest matching to daily bodies found entries or other types of
aggregate entries.
14 are partial or possible matches, which under Roberts' system should
probably be coded as 1 or 2 rather than 0
3 have a civilian-combatant issue which likely explains the non-match,
1 of these being deaths of foreign security contractors
And then the 21 where we'd agree there is no match and the correct
code is 0.
So the same problem is carrying through with these. Lots of 'false
negatives' at the incident level and with regard to aggregate matches
if estimating 'new deaths', rather than 'new incident details'.
In Roberts' first posting of Mar 9, he wrote: "Please take a random
sample of 10 or 20 of the records (not some selected group likely to
be right or wrong) and look for yourself."
So I've also tried this. I sorted the full spreadsheet by report key
(Column A) and just looked at the first 20 in the sorted list. This
gave a pretty diverse selection of dates and event types, and is easy
for anyone to replicate. Another advantage here is that unlike the
previous checks i did on the 4+ and the 3's, this one would also
include matched incidents, not just the non-matched 0's, so I can
check for "false positives" too.
Of these 20 logs, the original spreadsheet codes only 4 as matched, or
at least partially/possibly matched. The other 16 are all coded as no
match (0). I don't quite understand the match codes for these four,
however. Log #4 gives a code of "32", which i guess means that one
person thought it should be a '3' and another thought it should be a
'2', but k023 is definitely the match, so it should be just a 3. Log
#5 gives a code of "113" which I guess means two people thought it was
unlikely a match (1), while another thought it was matched (3), but
k022 is the correct match and this should be a 3. Then Log #7 gets
only a 2, but k047 is clearly the match, so this should just be a 3.
Log #8 gets a '4' but that isn't even one of their codes, so i guess
this is supposed to be equivalent to 3, which is correct.
So there are no "false positives" at all, and even the few 'positives'
they do find are being (mis)interpreted to still drive the results
more negative.
When I reanalyze the whole 20, I come to about the opposite
conclusions, that there are only 4 that clearly have no match, and 16
which match either fully or partially/possibly:
8 are direct incident-level matches
1 is an aggregate match to morgue, unambiguously so, as the log says
"His body was taken to Justice Medical" (i.e., the Baghdad morgue)
4 are aggregate matches mainly to daily bodies found tallies
3 are possible matches
4 are no match
So again here the same problems keep coming up: a systematic failure
to find matching incidents when they are there, a failure to
adequately handle any kind of aggregate IBC entries (when trying to
estimate 'new deaths', as opposed to 'new incident details'), and the
sample suggests a new failing to properly code the full value even of
those cases where matches are found, building in even more negative
bias. All of these problems appear to be systematic and, taken
together, clearly drive the results hugely in the direction of
exaggerating the number of new deaths.
To sum up, I simply so no reason here to doubt or alter any of the
conclusions given in IBC's original report on this issue, published at
the time of the WikiLeaks release of the logs. The difference between
those conclusions and the newer ones offered by Roberts a couple weeks
ago are adequately explained by all the failings discussed above. I'd
hope that Roberts will follow through on his promise of Mar 10 to
correct the record, as it is quite clear that their published
conclusions are very wrong and a correction is in order.
Below is a more detailed listing of the 20 incidents discussed above,
and the appropriate IBC match/coding as I see it:
1. x271
2. k021
3. No match
4. k023
5. k022
6. No match
7. k047
8. k052
9. k6103g - aggregate match to bodies found - Suleikh is Adhamiya
district
10. k6098 - Log says 2 killed-4 wounded, IBC says 1 killed-4 wounded -
otherwise identical, so this is just a match with a disparity over the
number killed
11. Possible - k6248
12. x493g - aggregate match to morgue - log says "His body was taken
to Justice Medical."
13. k6013a - aggregate match to bodies found - log says body found and
retreived by IP in Doura area - IBC reports say "Police patrols found
in Baghdad on Tuesday, April 17, twenty-five unidentified bodies... 5
in ( Doura)"
14. Possible - k6097
15. No match
16. k6099
17. d3459 - aggregate match - log gives no coordinates or location,
but says the reporting unit was 'BASRAH OPS, NCC', which suggests this
was Basrah and so would match d3459
18. k3062 - aggregate match, or k3058c, or x493e morgue
19. Possible - k6100
20. No match
Joshua Dougherty
> 17. d3459 - aggregate match - log gives no coordinates or location,
> but says the reporting unit was 'BASRAH OPS, NCC', which suggests this
> was Basrah and so would match d3459
I just looked at this one again and it actually does give coordinates,
but these would place it in Kuwait, over the border from Basrah. I've
noticed other logs where the coordinates are clearly wrong, so I'd
guess that these are wrong and it is Basrah.
ishikawa
he's presented appears to be very damning (I looked at several of the
matches, and they certainly check out.). It wouldn't be so serious if
Les Roberts hadn't already promoted his conclusions across the
internet -- conclusions based on a study irretrievably flawed with
systematic failures.
I hope Dr Roberts puts as much effort into publishing corrections as
he put into promoting his mistaken claims...
TDNY
ishikawa
Hundreds of Thousands of Unrecorded Iraqi Deaths' for reference
purposes.