ArticlePDF Available

Conflict Deaths in Iraq: A Methodological Critique of the ORB Survey Estimate


Abstract and Figures

In September of 2007 ORB, a British opinion polling firm, released an estimate that 1.2 million Iraqis had been killed in the conflict, subsequently lowering its estimate to 1 million. We com-pare three ORB polls and find important irregularities in ORB's mortality data in four central governorates of Iraq that account for more than 80% of the estimated deaths. These internal validity checks indicate that the ORB mortality data are not credible and would suggest a much lower estimate than ORB has published. We also analyze a number of specific error sources in the poll. Systematic errors, which include non-coverage and measurement errors, mostly point toward overestimation. Variable errors are also substantial but they are difficult to quantify in part due to incomplete disclosure of methodological details by ORB. External validity checks, including comparisons with two much larger and higher quality surveys, reinforce the conclu-sion that ORB has overestimated the number killed in Iraq by a wide margin. Thus, our paper answers a challenge facing the field of survey methodology, to explain how different surveys have produced such divergent mortality estimates for Iraq.
Content may be subject to copyright.
Survey Research Methods (2010)
Vol.4, No.1, pp. 3-15
ISSN 1864-3361
European Survey Research Association
Conflict Deaths in Iraq: A Methodological Critique of the ORB Survey
Michael Spagat
Royal Holloway, University of London
Josh Dougherty
Iraq Body Count
In September of 2007 ORB, a British opinion polling firm, released an estimate that 1.2 million
Iraqis had been killed in the conflict, subsequently lowering its estimate to 1 million. We com-
pare three ORB polls and find important irregularities in ORB’s mortality data in four central
governorates of Iraq that account for more than 80% of the estimated deaths. These internal
validity checks indicate that the ORB mortality data are not credible and would suggest a much
lower estimate than ORB has published. We also analyze a number of specific error sources in
the poll. Systematic errors, which include non-coverage and measurement errors, mostly point
toward overestimation. Variable errors are also substantial but they are dicult to quantify in
part due to incomplete disclosure of methodological details by ORB. External validity checks,
including comparisons with two much larger and higher quality surveys, reinforce the conclu-
sion that ORB has overestimated the number killed in Iraq by a wide margin. Thus, our paper
answers a challenge facing the field of survey methodology, to explain how dierent surveys
have produced such divergent mortality estimates for Iraq.
Keywords: Iraq, conflict mortality, survey quality, ORB poll, survey error
1 Introduction
In September of 2007 a British polling firm called Opin-
ion Research Business (ORB) published an estimate of “1.2
million murders” in Iraq since the U.S. invasion of 2003
(ORB, 2007b). ORB subsequently revised this estimate
down to just over 1 million in January of 2008 (ORB, 2008a).
The polling data which formed the basis of this estimate were
compiled by an Iraqi polling firm called IIACSS on behalf of
ORB and represents the first ever attempt by ORB to use its
polling to create a national mortality estimate, either for Iraq
or for any other country.
All credible sources on conflict mortality since the 2003
invasion have shown a staggering level of human losses suf-
fered by the people of Iraq, yet the ORB estimate is excep-
tionally high even within this universe of figures. ORB’s
estimate is comparable only to the Burnham et al. (2006)
one published in the Lancet which was, in turn, much higher
than any other estimate until the ORB estimate appeared al-
most a year later. ORB (2007b) leans on this Lancet estimate
as support for its conclusions. Likewise, two authors of the
Lancet study have regularly pointed to ORB as corrobora-
tion for their high 2006 estimate, with one going as far as to
claim that ORB represents the “best estimate” of Iraqi deaths
currently available (KUOW radio, 2008). Moreover, the
Bloomberg School of Public Health of Johns Hopkins Uni-
versity, the parent organization behind the Lancet estimate,
issued two separate ocial statements claiming the ORB poll
Contact information: Michael Spagat, Department of
Economics, Royal Holloway, University of London, e-mail:
as important corroboration for the Lancet estimate.1How-
ever, the Lancet study itself has been widely questioned in
articles such as Bohannon (2006), Dardagan et al. (2006),
Daponte (2007), Guha-Sapir and Degomme (2007), Johnson
et al. (2008), Laaksonen (2008), Munro and Canon (2008),
Rosenblum and van der Laan (2009) and Spagat (2008 &
2010). Recently, the principal researcher of the Lancet study
has been censured by the American Association for Public
Opinion Research because he “repeatedly refused to make
public essential facts about his research” and suspended by
Johns Hopkins University from human subject research for
protocol violations that compromised the safety of survey re-
spondents (AAPOR, 2009 and Bloomberg School of Public
Health, 2009).
However, aside from the authors of the Lancet estimate
and Bloomberg ocials, the ORB estimate has not been
taken very seriously in the scientific community. For exam-
ple, the Iraq Family Health Survey (IFHS) ignores the ORB
estimate in its article on violent conflict mortality in Iraq
that was published in the New England Journal of Medicine
(IFHS, 2008a). Daponte (2007) also does not mention the
ORB poll in its overview of conflict mortality estimates in
Iraq. Tapp et al. (2008) mentions but does not discuss the
ORB poll beyond repeating some information that appeared
in ORB’s first press release accompanying its poll (ORB,
2007b).2Thus, implicitly, the scientific literature seems to
have dismissed the ORB poll without much discussion.
We argue, however, that for two main reasons the ORB
1One such Bloomberg statement is quoted in Munro and Canon
(2008). Another is Bloomberg School of Public Health (2007). See
also Burnham and Roberts (2007).
2The Tapp et al. (2008) article did not notice that ORB reduced
its estimate by about 200,000 in January of 2008.
poll should not simply be dismissed a priori. First, ORB is a
professional polling organization and, at least on the surface,
appears to have applied some standard survey methods to ar-
rive at its estimate. If ORB has overestimated violent deaths
in the Iraq conflict by a wide margin, as we argue in this
paper, then it is a challenge to the field of survey method-
ology to explain what went wrong. Conversely, if ORB’s
estimate is more or less accurate then two other very large
and well-funded surveys of Iraq must have underestimated
violent deaths by a wide margin (see section 6, table 2), in
which case it would be a challenge to survey methodologists
to explain what went wrong with these surveys. More gen-
erally, if survey methodologists are unable to separate bad
conflict mortality surveys from good ones then the sample
survey approach to determining conflict mortality must be
considered unviable. Second, as noted above, the authors of
the Lancet estimate and Bloomberg ocials lean heavily on
the ORB poll. Outside of academic circles, the ORB poll
has been cited as important evidence in a range of public and
media discussion, including Steele and Goldenberg (2008),
National Public Radio (2007) and many blogs and websites
such as (McElwee, 2008). Thus, the ORB estimate
has had an impact in much of the public discourse on mor-
tality in the Iraq conflict and, therefore, should not simply be
In this paper we use the perspective of the survey qual-
ity literature (Biemer and Lyberg, 2003) to examine ORB’s
estimate for killings in the Iraq war. In section 2 we briefly
examine the sample survey methodology for estimating war
deaths and then present what has been disclosed of ORB’s
particular implementation of this methodology. In section
3 we examine sampling error, concluding that ORB’s pub-
lished “ranges” substantially underestimate the true sampling
error of its estimate, but that ORB has not released enough
information about its methodology to allow for a proper as-
sessment of sampling error.
In section 4 we present an internal validity check of
ORB’s data that includes two additional ORB polls con-
ducted just before and just after the one that we focus on
in this paper. We find internal contradictions indicative of
compromised data collection practices which greatly exag-
gerate the resulting estimate. In particular, in a contiguous
region of four governorates accounting for more than 80%
of ORB’s estimate, a higher percentage of respondents re-
port deaths of household members than the percentage of re-
spondents reporting deaths of extended family members in an
ORB poll conducted in February, 2007. ORB also conducted
a third poll in March, 2008 in which it reports switching
back to its extended-family question but with only a slight
increase in the percent of respondents reporting deaths. Yet
in southern Iraq ORB reports that the percent of households
reporting deaths did, as expected given Iraq’s large extended
family networks, drop sharply when ORB switched from an
extended family question to a household question. In light of
these problems it is clear that ORB’s data cannot support an
estimate of 1 million deaths as claimed, but might support a
very rough estimate of 300,000.
In sections 5 and 6 we identify further error sources, fo-
cusing mainly on systematic errors in the form of errors of
measurement, non-coverage and non-response. These sug-
gest that even a revised estimate of 300,000 is likely to be bi-
ased upward. In section 7 we compare ORB’s estimate with
other sources, including two larger and much higher qual-
ity mortality surveys. These external validity checks suggest
that the ORB estimates greatly exceed those that derive from
more credible methodologies and stand in contradiction to a
wide range of evidence.
2 Methodology and Disclosure
In recent years there has been growing use of sample
survey methodology to estimate deaths linked to violent con-
flicts. The most prominent examples of surveys that have
produced national estimates have been Spiegel and Salama
(2000) for Kosovo, a series of surveys done in the Demo-
cratic Republic of Congo (DRC) by the International Rescue
Committee (IRC) (e.g., Coghlan et al., 2006) and five surveys
in Iraq that we will discuss in the present paper.
The basic ideas of the survey approach are straightfor-
ward for survey professionals. A random sample of house-
holds is drawn, normally a cluster sample. Questions are
asked aimed at discovering, among other things, the num-
ber of people living and dying in each household. There are
questions about causes of death that distinguish, at a mini-
mum, between violent deaths and non-violent deaths. Re-
searchers estimate the number of violent deaths during the
survey’s recall period. In some cases, researchers also make
estimates for what are known as “excess deaths”. These are
deaths above and beyond a baseline rate thought to be the
one that would have prevailed if there had not been a war.
In practice, the baseline mortality rate is generally defined as
a pre-war rate and this is often measured within the survey
itself by extending the recall period back to before the war
ORB has not disclosed many details about its methodol-
ogy but it is clear that this methodology was roughly along
the lines described above, although ORB does not give an
excess death estimate. ORB’s estimate for violent deaths in
Iraq first appeared in a September 2007 press release pub-
lished on the ORB website entitled “More than 1,000,000
Iraqis murdered”:
“. .. this data suggests a total of 1,220,580 deaths
since the invasion in 2003. Calculating the af-
fect from the margin of error we believe that the
range is a minimum of 733,158 to a maximum
of 1,446,063.” ORB, (2007b)
ORB’s ocial description of its methodology is con-
tained within six bullet points in its press release:
“Results are based on face-to-face inter-
views amongst a nationally representative
sample of 1,720 adults aged 18+through-
out Iraq (1,499 agreed to answer the ques-
tion on household deaths)
The standard margin of error on the sample
who answered (1,499) is +2.5%
The methodology uses multi-stage random
probability sampling and covers fifteen of
the eighteen governorates within Iraq. For
security reasons Karbala and Al Anbar
were not included. Irbil was excluded as
the authorities refused our field team a per-
Interviews conducted August 12th-19th
Full results and data tabulations are avail-
able at
ORB is a full member of the British
Polling Council and abides by its rules.”
ORB, (2007b)
The only other methodological information provided is
an English language version of the questions. ORB declines
to disclose the field versions of its questionnaire in Arabic
and Kurdish, on the ground that such disclosures would lead
to what ORB believes would be unproductive discussions of
the quality of its translations.3
After its initial release ORB did further rural sampling4
and in January of 2008 lowered its estimate by almost
200,000 deaths:
“. . .we now estimate that the death toll between
March 2003 and August 2007 is likely to have
been of the order of 1,033,000. If one takes
into account the margin of error associated with
survey data of this nature then the estimated
range is between 946,000 and 1,120,000.” ORB
ORB (2008a, press release)5provides the additional
methodological information that there were “112 unique
sampling points” and that in the end it conducted 2,163 suc-
cessful interviews out of 2,414 attempts.
ORB does not disclose what its sampling frame is but
this would appear to be a list of dwellings in Iraq since it is
clear that the measurement units for the mortality question
are intended to be households. Within each such unit a sin-
gle respondent was chosen using the next birthday method.
Thus, it is assumed that each potential respondent within a
household would provide identical information about mor-
tality so that the selection of one over another does not mat-
ter. It is also assumed that respondents provide information
on deaths only for persons residing within their households
although we argue in section 6 that this assumption is not
well founded.
ORB (2008a, New Casualty Tabs) tabulates responses to
questions and includes some information on the weighting
scheme which should adjust for diering selection probabil-
ities. From Table 1 of ORB (2008a, New Casualty Tabs)
we infer that answers are weighted by governorate and by
whether respondents are classified as urban or rural dwellers.
For example, 474 Baghdad interviews are scaled up to 681
while 190 Salahadin interviews are scaled down to 161. 590
rural interviews are scaled up to 756 and 1,824 urban inter-
views are scaled down to 1,658. ORB does not explain how
it arrives at its weighting or how it classifies urban and rural
ORB estimates the number of deaths as follows. It
considers only the 2,163 individuals who responded to the
mortality question. Among these, ORB finds that 20.2%
report deaths with these respondents reporting 1.26 deaths
on average, where these figures incorporate the weighting
scheme. Taking these figures to be representative of the
household unit, ORB derives a mortality estimate by apply-
ing them to an estimated number of households nationwide.
ORB states that the 1997 census in Iraq “indicated a total
of 4,050,597 households”, and therefore calculates 0.202 x
1.26 x 4,050,597 =1,030,958, with the tiny dierence from
ORB’s reported number presumably being due to rounding.
ORB’s “estimated range” of 946-000-1,120,000 comes from
subtracting and adding 1.7% from and to its 20.2% figure
where the 1.7% is approximately the margin of error on a
simple random sample of size 2,160 for a yes-no question to
which approximately 20% of the responses are “yes”.
ORB’s parsimony with information about its methodol-
ogy is an indicator of low survey quality and weakens confi-
dence in its estimate. However, nondisclosure of information
cannot, per se, explain estimation error. We now turn, there-
fore, to potential sources of error.
3 Sampling error
ORB does not provide enough information about its
sampling methodology to allow a proper assessment of sam-
pling error but, for a variety of reasons, it is clear that its
range of plus or minus 8% is inappropriately narrow.
First, the fact that ORB (2008a) used 112 unique sam-
pling points suggests that the poll was a cluster survey with
112 clusters. Any proper error calculation must account for
the survey being a cluster survey, this being especially impor-
tant since households located near to each other may have
quite similar violence experiences. Yet ORB does not use
cluster-survey sampling error methods and, instead, calcu-
lates its error margins as if it had drawn a simple random
3This refusal was made, and this reason given, in a meeting
Michael Spagat had with two ORB representatives at the ORB oce
in London on February 28, 2008.
4This development raises some concerns about the quality of the
sampling. ORB (2007b) states that they drew a “nationally repre-
sentative” sample but then ORB (2007c) notes that this sample had
“limited coverage in rural districts” so ORB, therefore, performed
a “rural booster”. ORB (2008a, press release) then declared this
sample to also be “nationally representative.” Precise details about
the sample design would be welcome.
5ORB posts all its releases online. To contain the size of our
bibliography we cite only the main pages for each poll and provide
information when appropriate guiding readers to further documents
accessible from these main pages.
sample. The impact of ignoring clustering eects in ORB’s
error calculation is likely to be large. Two other much larger
cluster surveys of mortality in Iraq had much wider confi-
dence intervals. The Iraq Living Conditions Survey (ILCS),
with 2,200 clusters and 21,668 households, reported a con-
fidence interval of -25% to +21% (ILCS, 2005a). The Iraq
Family Health Survey (IFHS), with 979 clusters and 9,345
households, reported a confidence interval of -31% to +48%.
In comparison, ORB’s claimed plus or minus 8% does not
seem credible.
Second, ORB accounts for sampling error only in its es-
timate of the percent of households experiencing deaths but
not in the average number of deaths reported per household.
Third, the ORB (2008a) sample is weighted by governorate
and also has rural-urban weightings. However, the error cal-
culation does not account for the weighting scheme.
4 Internal Validity Check
ORB conducted three polls within a year of each other
in Iraq which asked questions on mortality. These three polls
are very hard to reconcile with each other and, as a group, un-
dermine the ORB mortality estimate derived from the second
of these polls.
1. There was first an ORB poll (ORB, 2007a) conducted
in February of 2007 (hereafter ORB1). In ORB1 26%
of respondents reported “the murder of a member of my
family/relative” [emphasis added] in the three previous
years (table 3 of ORB1).
2. Six months later, in August of 2007, there was the
ORB poll that is the main object of this paper, and
which was used by ORB to create its mortality es-
timate (hereafter ORB2). In ORB2 20% of respon-
dents reported deaths “as a result” of conflict/violence
of a household member where household membership
was defined by living under the same roof (table 1 of
3. Lastly, there was a third ORB poll (ORB, 2008b) con-
ducted in February/March of 2008 (hereafter ORB3).
In ORB3 24% of respondents reported “the murder of
a member of my family/relative” since the invasion (ta-
ble 20 of ORB3).6
ORB itself, and virtually all commentary on ORB2, has
interpreted its estimate to mean direct violent deaths, i.e.,
the same type of deaths measured for extended families in
ORB1 and ORB3. However, it is further claimed that ORB2
measured these deaths only for formal household members
rather than for any extended-family members as in ORB1 and
ORB3. The claim that ORB2 was limited to formal house-
hold members is the main basis for ORBs extrapolation of its
figures to a national mortality estimate.
The number of respondents with an extended family
member who has been killed has to be far higher than the
number of respondents with a household member killed as
long as all three polls are indeed measuring the same kinds
of deaths and respondents are following the “family/relative”
or “household” instructions of each poll correctly. Yet ORB
gives three percentages in separate polls conducted within
one year of each other that are very close to one another:
one for respondents reporting deaths of household members,
sandwiched in between two others reporting deaths of ex-
tended family members.7
To quantify the discrepancy between the three polls we
use the network scale-up equation given in Moody (2006),
which estimates the proportion of a population connected to
someone with direct experience of a particular event type as
a function of the total population size, the number of peo-
ple with direct experience of the event type and the average
number of connections per person.8We first assume that the
population of Iraq is 29 million, that average extended family
size in Iraq is only 26 and that 250,000 people were killed in
Iraq during the ORB2 coverage period, i.e., that the ORB2
estimate is too high by a factor of 4. These figures imply
that about 20.2% of the population would have had a family
member or relative killed. Thus, we can generate percentages
much like the ORB1 and ORB3 ones based on very small
extended family sizes for a Middle Eastern country (White
and Houseman, 2002) and far fewer deaths than estimated by
ORB2. If we raise the extended family size to 49, still a low
number, then we can lower the number of deaths to 150,000
and still 22.4% of the population would have experienced the
killing of an extended family member. If we use the ORB2
estimate of 1 million people killed and an extended family
size of 49 then more than 82.1% of the population would
have experienced the killing of an extended family member,
far more than the percentages measured in ORB1 or ORB3.
Thus, the ORB2 estimate of 1 million people killed is di-
cult to reconcile with the ORB1 and ORB3 polls.
Taken together, the three ORB polls present quite a puz-
zle; why is it that ORB finds a pattern of percentages of 26%-
20%-24% when the middle number should be much lower
than the first and third numbers? This pattern makes little
sense if ORB2 is truly limited to formal household members
while ORB1 and ORB3 include extended family members.
We argue that the ORB data themselves provide the an-
swer to this puzzle.
6The number of interviews attempted for the three polls was
5,019, 2,414 and 4,000 for ORB1, ORB2 and ORB3 respectively.
Thus, the ORB2 poll, the basis for ORB’s national estimate of
deaths due to the war, is the smallest of the three.
7The confidence intervals for ORB2’s 20% and ORB3’s 24%
overlap even if we use the dubiously narrow margins of error
claimed by ORB in its methodological bullet points. It is clear
that on any proper calculation, all three confidence intervals overlap
substantially with one another.
8The formula, is pr=1(1 e
t)cwhere pris the proportion con-
nected to those with direct experience, eis the number of events, t
is the total population and cis the average number of connections
per person. This formula can be derived analytically assuming inde-
pendence in the assignment of events to people and Moody (2006)
reports on simulations showing that the independence assumption
can be relaxed.
Table 1 compares results of the three polls, conducted
between March of 2007 and March of 2008, reported by gov-
ernorate.9The figures clearly show one pattern in what we
refer to as the geographic ‘Center’ and another very dierent
pattern in the ‘South’. Note that these regions are geograph-
ically distinct from each other but internally contiguous (see
First, in the South the percent of respondents reporting
deaths of household members in ORB2 is consistently and
substantially lower than the percent of respondents report-
ing deaths of family members/relatives in ORB1 [(7, 7, 4,
15, 2, 8) versus (27, 52, 23, 43, 46, 29)]. This large drop,
by a factor of five on average (to 7 from 35), is reasonable
and expected since the ORB1 question allows the inclusion
of any extended family members, and extended family net-
works are much larger than households. A network scale-
up calculation (Moody, 2006) also generates similar factors
using plausible parameter values.11 This pattern appears not
only at the regional level, but also in every single governorate
in the region, i.e., the pattern is robust.
Second, in the Center, these four contiguous gover-
norates (see map), which account for more than 80% of the
deaths in ORB2, show a pattern that is completely dierent
and incompatible with the pattern found in the South. The
percent of respondents reporting deaths of household mem-
bers in ORB2 rises higher than the percent that had reported
deaths of extended-family members in the same region in
ORB1 [(43, 40, 18, 27) versus (31, 35, 18, 14)]. It is not just
that the percentages fail to display the substantial and pre-
dictable drops that are seen throughout the entire South; they
actually move in the opposite direction. Again, this is true
not just at the regional level but also in every governorate in
the region. The household percentages in these governorates
in ORB2 are never below the extended-family percentages in
Third, the ORB3 poll adds in a further problem in the
South. The ORB3 extended-family numbers in this region
are uniformly far below the ORB1 extended-family numbers
despite the fact that another rather violent year had elapsed
between the two polls [(2, 12, 1, 8, 9, 8) versus (27, 53, 23,
43, 46, 29)]. Moreover, the ORB3 extended-family numbers
are strikingly similar to the ORB2 household numbers in the
South [(2, 12, 1, 8, 9, 8) versus (7, 7, 4, 15, 2, 8)] despite the
fact that the former should have been much higher than the
The empirical patterns in the data displayed in table 1 are
not credible. It appears that the boundaries of the declared
units of analysis, the household in ORB2 and the extended
family in ORB3, were not properly or uniformly enforced
in either poll. Given this, the ORB mortality data cannot be
considered valid, whatever the precise explanation for these
One explanation that might resolve the strange pattern
across the three polls in the four central governorates is that
the field workers in this region did not change to the new
household question for the ORB2 poll, but instead asked a
family member/relative question like those asked in ORB1
and ORB3. The introduction of a dierent mortality ques-
tion into the ORB2 questionnaire than was used in ORB1
could have generated confusion or miscommunication in the
field between dierent field teams working in dierent re-
gions. Again, note that these regions are geographically con-
tiguous. Thus, it seems logical that one field team or teams
might cover the contiguous ‘Center’ region (Baghdad up to
Kurdistan), while a separate team or teams might handle the
contiguous ‘South’ (everything below Baghdad).
For the South, a similar explanation could resolve the
pattern evident in the ORB3 poll. Namely, the field work-
ers in this region correctly asked the household question in
the ORB2 poll, but then incorrectly asked a household ques-
tion again in ORB3. This would explain why the ORB2
and ORB3 numbers are so similar in the South, and why
the ORB3 numbers are so much lower than the ORB1 num-
bers in the South. More broadly, this would explain why
ORB3’s overall figure of 24% is lower than the 26% figure
from ORB1, despite the fact that it would have been expected
to be higher given the additional year of violence covered by
March 2008.12
A second possible explanation for the strange pattern in
the Center is that some people involved in the ORB2 field
work purposefully manipulated the survey in this region with
the goal of producing a high estimate. This is always a risk
with politically charged research questions, and the degree to
which this question would be both politically and emotion-
ally charged in the case of Iraq in mid-2007 may have few
parallels. Some known facts are consistent with this manip-
ulation scenario. The release of the ORB2 poll was timed to
coincide with the September, 2007 Congressional testimony
of General David Petraeus on the impact of the U.S. “surge”
policy, suggesting political motivation. Moreover, Munqeth
Daghir, the director of IIACSS, the firm which implemented
the field work for these polls has stated that he became a poll-
ster in 2003 as a non-violent means of opposing the US inva-
sion of Iraq (Research Talk, 2006, Munro, 2007) so for him
political motivation seems fairly explicit. However, political
biases on the part of researchers do not necessarily translate
into biased data. The strange pattern across the three polls
in the South displayed in table 1, where ORB3 is inappro-
priately low, would yield no similar political advantage and
9For completeness at the bottom of table 1 we give figures for
the five governorates that were not polled all three times. The in-
complete patterns in these governorates are consistent with the pat-
terns for the governorates in the respective regions for which data
are complete.
10 We thank Gabriela Guerrero Serd´
an for providing us with this
11 For example, if 280,000 households experienced the murder of
a member, the average extended family size is 45 and the population
is 29 million then about 7% of households would have experienced
deaths whereas about 35% of the population would have experi-
enced the death of an extended family member.
12 Looking only at the overall figures one might imagine that sam-
pling error might explain why the ORB3 figure is below the ORB1
figure. However, the numbers in the South in table 1 virtually elim-
inate the possibility that sampling error is the right explanation for
this anomaly.
Table 1: Three ORB Polls by Governorate
% with death of % with death of % with death of
family household member household member
member/relative August, 2007 member/relative
Region March, 2007 (ORB1) (ORB2) March, 2008 (ORB3)
Center 26 36 41
Baghdad 31 43 45
Diyala 35 40 42
Salahudin 18 18 22
Ninewa 14 27 39
South 35 7 7
Babyl 27 7 2
Najaf 53 7 12
Muthana 23 4 1
Dhi Qar 43 15 8
Maysan 46 2 9
Basrah 29 8 8
North 8 7 6
Kirkuk/Tameem (Kirkuk) 20 (Kirkuk) 17 (Tameem) 9
Sulaymania 2 2 6
Dohuk 0 3 1
Center Ramadi/Al-Anbar (Ramadi) 29 NA 49
South Karbala 30 NA 13
Wassit NA 12 8
Qaddisiya NA 11 16
North Irbil NA NA 7
Sources: All figures for governorates in this table come directly from three tables posted on the ORB web site. They are ORB (2007a, TablesFINALMarch07Irq) for ORB1, ORB
(2008a, New Casualty Tabs) for ORB2 and ORB (2008b, FINALTABLESMarch08) for ORB3. For regional percentages we take averages across the governorates of each region
using the population weightings for the governorates that are implicit in ORB’s posted tables.
Table 1 compares results of the three polls, conducted between March of 2007 and March of
2008, reported by governorate.9 The figures clearly show one pattern in what we refer to as the
geographic ‘Center’ and another very different pattern in the ‘South’. Note that these regions are
geographically distinct from each other but internally contiguous (see map10).
First, in the South the percent of respondents reporting deaths of household members in ORB2 is
consistently and substantially lower than the percent of respondents reporting deaths of family
members/relatives in ORB1 [(7, 7, 4, 15, 2, 8) versus (27, 52, 23, 43, 46, 29)]. This large drop,
9 For completeness at the bottom of table 1 we give figures for the five governorates that were not polled all three
times. The incomplete patterns in these governorates are consistent with the patterns for the governorates in the
respective regions for which data are complete.
10 We thank Gabriela Guerrero Serdán for providing us with this picture.
Figure 1. Governorates Linked by Reporting Pattern in Table 1 are also linked Geographically
seems more consistent with the mistaken field work scenario.
Whatever the precise explanation for these inconsisten-
cies, the data from the three polls suggest that there were
significant problems in the implementation of the mortal-
ity questions in the Center and the South during the ORB2
and ORB3 polls respectively. Specifically, the responses to
ORB2’s household question in the Center are much too high
compared to the responses to the family/relative questions of
ORB1 and ORB3, and conflict sharply with the more rea-
sonable pattern seen in the South between ORB1 and ORB2.
In addition, ORB3’s family/relative figures in the South are
much too low compared to ORB2’s household figures.
Given the very high numbers of deaths provided to the
ORB2 poll by the four central governorates, the implications
of having inflated data in this region are large. The consistent
and sensible pattern seen across the South between ORB1
and ORB2 suggests that the percentages in the Center should
have dropped by a roughly similar factor between these two
polls. Making this adjustment, while otherwise following the
same estimation approach as ORB, the estimate would come
to roughly 300,000, rather than 1,000,000 as reported.13
5 Measurement Error
In this section we argue that ORB’s questionnaire is in-
adequate to fully exclude the reporting of non-violent deaths,
focusing on the systematic error resulting from this problem
although this will probably cause some variable error as well.
The original ORB (2007b) press release presents the sur-
vey as measuring the number of Iraqis who have been “mur-
dered” (1.2 million). A few months later, a follow-up release
revised the estimate down to 1 million (ORB, 2008a), but in
this case the survey is described as measuring “Iraqi citizens
[who] have died as a result of the conflict”. The first for-
mulation could only include direct, violent killings whereas
the second is vaguer and could potentially include a wide
range of non-violent deaths that respondents view as indi-
rectly linked to the war. This inconsistency between press
releases already hints at fuzziness on the boundary line be-
tween violent and non-violent deaths in the poll.
The ORB2 mortality question posted on the ORB web-
site asks respondents about deaths “resulting from the con-
flict”. Rather than defining such deaths clearly or checking
these designations, ORB delegates to respondents the author-
ity to decide whether or not a death is “a result of the conflict”
and therefore appropriate to include:
“How many members of your household, if any,
have died as a result of the conflict in Iraq since
2003 (ie as a result of violence rather than a nat-
ural death such as old age). Please note that I
mean those who were actually living under your
roof?”14 Question 1 of the ORB2 poll.
While the parenthetical follow-up in the question does
attempt to link deaths “as a result of the conflict” with vio-
lence, and may have helped to reduce the reporting of non-
violent deaths, it is not clear how each respondent would
interpret this part of the question. The parenthetical could
easily be interpreted as giving non-exclusive examples of a
kind of death they want reported and a kind of death they do
not want reported, leaving plenty of gray area in between.
If ORB intended to measure violent deaths exclusively then
it would have been better to just ask respondents specifi-
cally and directly about violent deaths, rather than basing
its question around the often broader concept of conflict-
related deaths and then attempting to correct this ambiguity
by adding on another, still ambiguous, parenthetical remark.
The wording of ORB’s question 1 could encourage some
respondents to report deaths they perceive to have been in-
directly caused by the conflict rather than direct violent
deaths as normally understood in conflict research. Exam-
ples of non-violent yet conflict-related deaths are varied.
Such deaths may include those directly caused by diseases
but carried through impure, conflict-aected water supplies,
or deaths suered in hospitals experiencing conflict-related
shortages of medical personnel, equipment or electricity. On-
going violence in an area can also prevent access to treatment
for conditions such as complicated births or heart attacks.
Such deaths are easily interpreted as proper to include in an
answer to ORB2’s question, although doing so would con-
vert the ORB estimate into an estimate of “excess deaths”, or
deaths that would not have occurred without the conflict, ac-
cording to the subjective judgment of each respondent, rather
than an estimate of direct violent deaths only.15 A loose in-
terpretation also becomes more likely when the closest these
conflict/violence-related deaths come to being defined in the
ORB question is that they are placed into a dichotomy against
“natural deaths such as old age”. Many respondents will feel,
sometimes strongly, that the kinds of deaths described above
are very much not “natural deaths such as old age”, but are
instead unnecessary (un-natural) deaths brought on by the
environment of conflict and violence, and are appropriate to
13 The following is a rough calculation which illustrates that the
stakes are very high, hundreds of thousands of deaths, on the
anomalies of Table 1. More than 80% of ORB’s estimate of more
than 1 million deaths come from the Center, i.e., more than 800,000
deaths. In the South, the percent reporting household deaths in
ORB2 is 1/5 of the percent reporting family member/relative deaths
in ORB1. If the ORB2 percent in the Center had also dropped by
a factor of 5 relative to the ORB1 percentage then the ORB2 per-
centage in the Center would have been 5.2 rather than 36. Since
800,000x(1-(5.2/36) =680,000 we conclude that ORB’s estimate
would be reduced by more than 680,000 deaths if the percentage
dierence in the Center for ORB2 compared to ORB1 had been
similar to the more credible dierence measured in the South.
14 We have copied ORB’s question exactly, including the errors
such as placing a full stop rather than a question mark at the end of
the first sentence, and a question mark rather than a full stop at the
end of the second sentence.
15 To the extent that ORB’s data do capture non-violent deaths be-
lieved by respondents to be conflict related, and therefore measures
“excess deaths” rather than just violent deaths, this would create an
additional estimation problem in that the ORB poll does not provide
any way to measure negative excess deaths. Including positive ex-
cess deaths but providing no counterbalancing measurement of neg-
ative excess deaths in regions where death rates may have declined
would be a source of upward bias.
include. One related example evident in the ORB2 data is
that 6% of respondents reporting deaths give “accident” as
the cause of death, but deaths from accidents are not nor-
mally classified as direct violent deaths in conflict mortality
Moreover, some respondents may perceive certain non-
violent deaths as results of conflict/violence when they are
not. There is often no clear way to judge whether a partic-
ular non-violent death would have been avoided if a conflict
had never started in the first place. Consequently, all sorts of
deaths might be seen by respondents as conflict-related even
when many are not really conflict-related.
Finally, ORB assumes that all respondents followed its
restriction on allowable types of deaths to report: “(ie as
a result of violence rather than a natural death such as old
age)”. Yet, ORB2’s parenthetical clarification on allowable
types of injuries is very similar to this one, “(ie as a result of
violence rather than accidental injury)”, and 10% of respon-
dents reporting injuries nevertheless gave “accident” as the
cause of injury. If many respondents did not actually exclude
accidental injuries when instructed to do so, then it is almost
certainly wrong for ORB to assume that all of these same
respondents did exclude natural deaths such as old age when
instructed to do so with virtually identical wording.
Therefore, it seems inevitable that the ORB2 poll would
have captured a number of non-violent deaths, including
some that are likely to have been conflict-related and some
that are less likely to have been conflict-related. Thus, the
ORB2 mortality estimate would overstate the true number of
direct violent deaths in Iraq because it can not really claim to
measure exclusively violent deaths.
Non-coverage and non-response
The ORB2 poll has a number of non-coverage errors and
non-response errors. Again we focus on systematic error al-
though each factor will probably add some variable error to
the survey. We try to quantify upward and downward biases
to the extent possible.
First, the attempt to define household boundaries in this
poll, essentially with an afterthought attached to the same
question, is inadequate to fully exclude the reporting of
deaths of extended family members. How dierent respon-
dents might interpret “actually living under your roof” is not
known. The question does not pin down when, how long,
or in what context either the respondent or the deceased had
to live under this roof to be included in that household. With
the high level of displacement in Iraq since the invasion, with
individuals or whole families often migrating and moving in
with other families or relatives for various lengths of time, or
moving in and out of various dwellings over time, the com-
position of many Iraqi households will have been in a state
of flux across the reporting period. A respondent could inter-
pret this part of the question in a variety of ways depending
on his or her experiences. For example, if some respondents
lived in more than one dwelling at dierent times over the
course of the long recall period used in this question, they
could consider all of the other inhabitants of each dwelling
to qualify for inclusion, whereby their answers could reflect
deaths across two or more households. Moreover, the above
analysis still assumes, as ORB does, that each respondent
really did attempt to rigorously restrict reports of deaths to
only people who lived under the same roof that the respon-
dent did. Yet, as pointed out in section 4, 10% of respon-
dents reporting injuries ignored the instruction not to report
accidental injuries. This suggests that ORB is also wrong to
assume that 100% of respondents followed the instruction on
household boundaries.
Even if the exact wordings of ORB2’s questions were
clear on household boundaries, it is still possible that re-
spondents nevertheless reported deaths of extended-family
members. Few respondents to a household poll will under-
stand the statistical imperative to set clear limits on house-
hold boundaries, or place such concerns above any other
concerns of their own. To the contrary, many respondents
might feel that, since interviewers are evidently trying to dis-
cover violent or conflict-related deaths, it is right and proper
for respondents to report deaths of, e.g., cousins or uncles,
even though they were not formally household members. It
is likely that respondents would have a natural tendency to-
ward what might be called “household inflation” so that they
can bear witness to deaths of relatives who were near and
dear to them. Taking household rosters, with demographic
information on each household, is one defense mechanism
against household inflation but ORB did not implement this
Second, the treatment of non-response to the ORB2
mortality question is unsound and its assumptions unac-
knowledged. In making its estimate, ORB ignores the 251
(weighted) responses out of 2,414 attempted interviews in
which the respondent either did not answer the main mortal-
ity question or responded “don’t know”. ORB then extrapo-
lated the remaining 2,163 responses as if they were the com-
plete sample. The eect of this treatment of non-response is
to introduce an unstated assumption that these non-response
households had the same proportion of conflict deaths as the
average of the other households which did respond. Eec-
tively, it is assumed that there were also conflict deaths in
20% of these households that did not actually report any
deaths. This baseless assumption has a large impact on the
ORB2 estimate, possibly generating upward bias of as large
as 10% (i.e., about two percentage points).
Third, ORB (2007b) notes that, for security reasons, it
was unable to sample in the governorates of Al Anbar and
Karbala, implying that this is a source of downward bias. Al
Anbar in particular has been one of the most violent gover-
norates in Iraq so it is plausible that these security problems
did cause some downward bias in the ORB2 estimate. On
the other hand, for reasons related to Kurdish sovereignty,
ORB2 was also unable to sample in the peaceful governorate
of Irbil. ORB includes the populations of these three un-
16 On the importance of household rosters see (SMART, 2006, p.
75) and (London School of Hygiene and Tropical Medicine (un-
dated) p. 109).
sampled governorates in its national population estimate and
extrapolates a national estimate out of data collected from
only 15 out or Iraq’s 18 governorates.17 This estimation pro-
cedure eectively assumes that these three un-sampled gov-
ernorates have the same death rate as the average death rate
across the 15 sampled governorates. According to the Iraq
Body Count database (IBC), these three governorates contain
roughly 8.6% of the violent deaths of civilians during the
ORB2 coverage period while, according to the IFHS, they
contain about 12.3% of the population. Thus, it is not at all
clear that these coverage problems would have been a source
of downward bias as suggested by ORB, and in fact could
have been another source of upward bias.
Fourth, ORB (2007c) also states that they were unable to
sample in “more volatile areas”, which suggests that even in
governorates where the field teams did operate they failed to
conduct interviews at some selected “sampling points” due
to security problems. If so, these failures could be a source
of downward bias, although they would also point to further
problems with the way this poll has been presented to the
public. ORB should have at least give a report on these prob-
Fifth, survivor bias, i.e., the general problem in conflict
surveys that one cannot interview members of households
which were wiped out completely, would be a source of
downward bias in ORB2. However, this is thought to be a
fairly small source of bias in most conflict surveys (Gakidou
and King, 2006).
We can find no indication that ORB did anything to
account for the non-coverage and non-response errors dis-
cussed in this section.
In sections 5 and 6 we focused on systematic errors but it
is clear that the issues we discuss will contribute to variable
errors as well, although these would be dicult to quantify.
Moreover, in section 3 we argued that, properly calculated,
the sampling error in the ORB2 survey should be much larger
than claimed by ORB. Together these considerations suggest
that a reasonable confidence interval for an estimate based on
ORB’s data could be very wide indeed.
7 External Validity Checks
In this section we place the ORB2 estimate within the
context of other evidence on violent mortality during the Iraq
conflict. This analysis further reinforces the conclusion that
ORB has overestimated violent deaths in Iraq by a wide mar-
Table 2 compares ORB2 with four other surveys: the
ILCS, the IFHS, Roberts et al. (2004), hereafter designated
as “L1”, and Burnham et al. (2006), to be designated “L2”.18
The table also builds in the data of IBC (continuously up-
dated) which tallies violent civilian deaths based on media
reports and data from hospitals, morgues and NGOs. Since
IBC uses a uniform methodology and includes the coverage
periods of all the surveys, it provides a useful means for com-
paring surveys with dierent coverage periods.
The ORB national mortality estimate is much higher
than those of the ILCS, IFHS and L1. While the latter 3
surveys exceed IBC figures for their respective coverage pe-
riods by factors between 1.7 and 3.1, ORB and L2 both ex-
ceed IBC figures for their respective time periods by a factor
of 12.2. Although the national estimates of ORB and L2
appear to be well in line with each other, they diverge sub-
stantially on the geographical distribution of violent deaths,
with ORB placing a much higher fraction of its deaths in
Baghdad than L2 does. The IFHS and ILCS are much larger
and higher quality surveys than ORB and L2. For exam-
ple, the IFHS and ILCS took full household rosters for each
of their respondents. This procedure is considered a crucial
step in household conflict mortality surveys to establish clear
household boundaries and to reduce inaccurate or fabricated
reports of deaths (footnote 16). L2 and the ORB poll did not
take household rosters, which is perhaps instructive, given
that these two surveys produced far higher estimates than
those which did follow this procedure. Another quality in-
dicator is that the Arabic and Kurdish questionnaires for the
ILCS and IFHS are freely available online whereas neither
ORB nor the L2 researches have disclosed the field versions
of their questionnaires. ILCS (2005b) and IFHS (2008) and
Iraq Family Health Survey Study Group (2008) both give de-
tailed accounts of how they designed and field tested their
questionnaires, developed their sample frames, drew their
samples, trained their sta, conducted and supervised their
fieldwork and coded and quality-checked their data. ORB
provides virtually no information on any of these issues.19
To close this section we present three further anomalies
generated by the ORB2 data.20 First, the ORB2 estimate im-
plies roughly 220,000 deaths in car bombings, about 130,000
of which are in Baghdad.21 In contrast, the IBC database has
roughly 11,500 car-bomb deaths during the ORB coverage
period, roughly 5,500 of which are in Baghdad. Thus, the
ORB2 estimate implies that the international media have no-
ticed only about 5% of car-bomb deaths both nationwide and
17 It is inappropriate in the first place for ORB to simply extrapo-
late its data gathered in just 15 governorates to all 18 governorates.
Accepted practice would have been to publish an estimate covering
only the governorates actually sampled.
18 The shorthand “L1” is in reference to this being the first war
mortality survey of Iraq published in the Lancet whereas “L2” was
the second such survey.
19 Here we give just two other indicators of poor quality control in
the ORB2 poll. ORB states that its interviews were “face-to-face”
yet it did not establish the genders of 4% of its respondents. The
first table of results released by ORB implied that the population
of Baghdad is 60% Christian but then, without explanation, ORB
replaced this table with another one that reduced this percentage to
3% (Mclean, 2007).
20 Blogger Will Mclean has posted (Mclean, 2007/2008) a num-
ber of good critiques of this kind on the ORB poll.
21 ORB (2008a, New Casualty Tabs, Table 2) gives a breakdown
of causes of death. Nationally, 22.3% are attributed to car bombs
which would work out to 220,000 deaths if this percentage is ap-
plied to 1,033,000. In Baghdad 21.5% are attributed to car bombs
which would work out to 130,000 deaths if this percentage is ap-
plied to 600,000 (notes to Table 2 above). Alternate calculations
could produce slightly dierent figures but these would not aect
our analysis.
Table 2: ORB and Surveys of Violent Deaths in Iraq since March 2003
Coverage Period Ends May 1, 2004 September 20, 2004 June 30, 2006 July 10, 2006 August 31, 2007
Survey Estimates of Violent Deaths 26,000 56,700 98,000 or 151,000 601,000 1,033,000
Ratio to IBC 1.7 3.0 2.0 or 3.1 12.2 12.2
Violent Deaths in Baghdad 8,063 18,900 - 27,000 52,920 or 81,540 150,000 600,000
Ratio to IBC in Baghdad 1.0 1.9 - 2.7 1.9 or 2.9 5.2 12.5
Number of clusters 2,200 32 971 47 112
Field Questionnaire Available Yes No Yes No No
Household Roster Taken Yes Yes Yes No No
Sources: Pedersen (2007) for the ILCS; EPIC (2004) discloses that 57,600 of the L1 estimate are violent and Deltoid (2004) that between 7 and 10 of the 21 violent deaths in the
L1 sample are in Baghdad; Iraq Family Health Survey Study Group (2008) for the IFHS; Spagat (2008) for L2. The calculation for the Baghdad ORB figures is 0.61 deaths per
household in Baghdad times 988,346 households in Baghdad (4,050,597 households in Iraq of which about 24.4% are in Baghdad according to IFHS (2008).
*The IFHS argued that there is a general tendency for the underreporting of deaths in household surveys. On these grounds, the IFHS adjusted its estimate upwards by more than
50% compared to a conventional estimate as produced by all the other surveys in this table. It is, therefore, best to remove this adjustment when comparing across surveys so in the
IFHS column we always provide two estimates; the first is a conventional estimate and the second is an adjusted one as given in the IFHS.
inside Baghdad.22 Yet, the IBC database shows that, on av-
erage, a lethal car-bombing generates six independent media
reports. The notion that 95% of all car-bomb deaths go un-
reported is exceptionally implausible. Indeed, car-bombings
are highly visible and newsworthy events that generate loud
explosions that are heard over wide areas followed by noisy
and obvious emergency responses. Many Iraqis carry cell
phones that can be used to report car bombings. The per-
petrators of car-bombings have a strong interest in making
sure that these events are noticed by the international me-
dia, an interest shared by the Iraqi and US governments for
dierent reasons. Moreover, the violent-death figures of the
Iraqi Ministry of Health and the Baghdad Morgue use dier-
ent counting systems and are very similar to the IBC figures
both inside and outside of Baghdad, so ORB’s figures would
imply that hospitals and morgues have also failed to detect
the vast majority of car bomb deaths.23 In fact, if the ORB
figures are right then it would appear that the number of car-
bomb deaths alone is already higher than the total number of
violent deaths of all causes recorded ocially, even in Bagh-
Second, the ORB2 findings on injuries seem to be well
out of line with existing evidence for two main reasons. First,
if injuries are estimated the same way that ORB estimated
deaths, the injury data suggest that there should have been
roughly 1 million injuries between the beginning of the war
and the end of August, 2007, a number that seems to be too
high by a wide margin.24 Civic Worldwide (2003) identified
4,000 injuries during the first 50 days of the war. IBC (2003)
found “at least 20,000” civilian injuries as of August, 2003.
IBC (2005) found “at least 42,500” civilian injuries between
the beginning of the war and March 19, 2005. The Iraqi
Ministry of Health (MoH) recorded 53,634 injuries due to
“terrorist incidents and military clashes” outside of Kurdis-
tan between April 5, 2004 and June 1, 2006 [Roug and Smith,
2006].25 The UN Assistance Mission for Iraq (UNAMI) re-
ported 36,685 injuries in 2006 (UNAMI [2007]). The MoH
reported 38,609 injuries in 2007 (NINA [2008]) It is dicult
to extract a single clear figure for the whole ORB2 coverage
period from these overlapping data, but it is clear that the
number of injuries consistent with these sources during that
period would be much closer to 200,000 than to 1 million.
Thus, the ORB2 data would suggest that 800,000 or more
injured Iraqis were either never treated for these injuries or
have never been recorded as receiving any treatment.26
The other problem with ORB’s injury findings is that the
ratio of injuries to killings, as measured by ORB is implau-
sibly low, roughly just 1 to 1, and inconsistent with other
sources on this question as well. Coupland and Meddings
(1999) report ratios of wounded to killed for 9 conflicts rang-
ing from 1.9 to 13.0 with a median of 4.0. Civic Worldwide
(2003) recorded a ratio of 2 injuries per death during the first
50 days of the war. IBC (2003) and IBC (2005) found ratios
of 2.5 to 1 and 1.7 to 1 respectively. The MoH recorded a
ratio of 2.8 to 1 between April 5, 2004 and June 1, 2006.
UNAMI (2007), with a ratio of 1.1 to 1, is the only source
with a ratio nearly as low as ORB’s but it applies only to 2006
when a large portion of killings were by execution, leaving
few injured behind. For 2007 the MoH reported a ratio of 2.3
22 IBC continuously monitors dozens of international media
23 Roug and Smith (2006) obtained data from the Iraqi Ministry
of Health and Baghdad Morgue and concluded that “at least 50,000
Iraqis had died violently since the 2003 U.S. led invasion”. Their
figures closely match the IBC ones both at the national and gover-
norate levels. Gamel (2009) obtained annual national data on death
certificates issued for violent deaths from 2005 through February of
2009 that are also close to IBC figures.
24 According to ORB (2008a, New Casualty Tabs, Table 3) there
were 0.25 conflict injuries per household, which is precisely equal
to the number of deaths per household shown in ORB (2008a, New
Casualty Tabs, Table 1).
25 Roug and Smith (2006) published data on violent deaths but
not injuries. However, the authors of the article collected MoH data
on both deaths and injuries which they shared with us: personal
communication with Doug Smith.
26 ORB2 data suggest roughly 575,000 injuries in Baghdad, a fig-
ure that exceeds the MoH Baghdad figure of 28,343 injuries for the
shorter period April 5, 2004 to June 1, 2006 by a factor of about 19.
injuries per killing (NINA [2008]). Even ORB’s ratios in car
bombings, aerial bombardments and other blasts are respec-
tively just 1.3, 1.2 and 1.4. Such ratios are lower than those
measured in explosions in a variety of dierent contexts.27
Thus it would appear that if ORB has not overestimated the
number of deaths in Iraq then it has underestimated the num-
ber of injuries. Yet, as argued above, ORB’s injury numbers
already appear to be implausibly high to begin with, so it is
dicult to square this circle.
Third, the ORB2 numbers imply that approximately
600,000 Baghdad residents have been killed (notes to Table
2 above). If so, then there should be a noticeable gender im-
balance in Baghdad. Several independent sources show that
more than 80% of violent deaths during the ORB timeframe
were adult males. The roughly 600,000 people killed in
Baghdad implied by ORB’s estimate would include roughly
500,000 adult males and 55,000 adult females.28 Roughly
60.6% of the population of Iraq is adult (ILCS, 2005a) and
the estimated population of Baghdad in 2006 was 6,962,650
(IFHS, 2008), implying an adult population in Baghdad of
about 4,200,000. The ORB figures, together with these pop-
ulation figures, suggest that nearly a quarter of the adult male
population of Baghdad had been killed by August of 2007.29
If ORB’s mortality figures for Baghdad are reasonably ac-
curate the remaining adult females should outnumber the re-
maining adult males by a ratio of about 1.3 to 1.30 Yet ORB2
itself found a ratio of adult females to adult males in Bagh-
dad of just slightly under 1 and ORB1 and ORB3 found, re-
spectively, ratios of females to males in Baghdad of roughly
1.04 and 0.92 to 1.31 Note that ORB uses the “next birth-
day” method of selecting respondents within households, a
method that is not biased toward selecting either males or
females.32 Thus, the gender balance within Baghdad mea-
sured in ORB’s own polls, with possibly even more males
than females, appears to be inconsistent with ORB’s mortal-
ity estimate.33
Without a doubt the war in Iraq has inflicted immense
human costs on the Iraqi people. Yet the conflict mortal-
ity estimate of 1 million deaths published by ORB does not
withstand scrutiny. It is inconsistent with the survey data of
the ILCS and the IFHS which are much larger and higher-
quality surveys. The ORB2 estimate also generates a num-
ber of implications which are not plausible: about 200,000
car-bomb deaths unnoticed by all international media, hun-
dreds of thousands of unnoticed injured people who have ap-
parently either not sought treatment or not been recorded as
receiving any, a surprisingly low ratio of injuries to killings,
and an implied gender imbalance in Baghdad that seems in-
consistent with ORB’s own polling data.
ORB gives little information about its methodology, for
example, not disclosing its sample design or translations of
its questions, while information that has been released sug-
gests some significant weaknesses. ORB’s presentation of
margins of error and “ranges” for its estimates suggests that
ORB has not calculated proper 95% confidence intervals.
ORB2 did not record household rosters, asked an ambigu-
ously worded mortality question and did not account for non-
response in its estimates or “ranges”. The claim of ORB
(2007b) to have drawn a “nationally representative sample”
followed immediately by the ORB (2007c) admission of the
need for substantial further rural sampling also casts doubt
on ORB’s sampling procedures.
Most importantly, our analysis of figures across the three
ORB polls finds fundamental flaws in the data underpinning
ORB’s estimate. The ORB2 data are not suitable for deriv-
ing any credible estimate but, given proper scrutiny, it is clear
that ORB has overestimated by a wide margin.
We are especially grateful to two anonymous referees and
to Peter Lynn for substantially improving our paper. We also
thank Beth Daponte, Hamit Dardagan, David Kane, Will
McLean and Robert Shone for excellent comments on an
early draft. Finally, we thank Mohammed Ali for an early
observation that led us eventually to the section of the paper
on internal validity.
27 Coupland and Meddings (1999) give wounded-to-killed ratios
ranging from 3.3 to 20.0 with a median of 7.0 in a list of incidents
involving explosives between 1996 and 1998.
28 Gender proportions are derived from IBC, IFHS and L2. IBC
covers the whole ORB time-frame, while IFHS and L2 are the clos-
est of the existing surveys, covering to mid-2006. These sources
give 83%, 89% and 85% respectively for adult male deaths, and
9%, 3% and 5% for adult females. In order to give ORB the best
chance of consistency on gender balance, our calculations use the
lowest of the three for males (83%) and the highest of the three for
females (9%). ORB itself did not collect gender information on
29 Note that, even using ORB’s low measured ratio of injuries to
killings of about 1 to 1, if one quarter of adult males in Baghdad
have been killed then one third of the remaining adult males should
be injured. Using a standard rule of thumb of 3 injuries per killing in
conflict we would expect that all remaining adult males in Baghdad
should be injured.
30 The number of adult males in Baghdad should be something
like ((4,200,000)/2)-500,000 =1,600,000. The number of adult
females in Baghdad should be roughly ((4,200,000)/2)-55,000 =
2,045,000. The ratio of the two numbers is 1.28. This calculation
sets aside migration (see footnote 36).
31 The sources are ORB (2007a, TablesFINALMarch07Irq, Table
11), ORB (2008b, New Casualty Tabs, Table 7) and ORB (2008b,
FINALTABLESMarch08, Table 51).
32 “When knocking on the door of the household the interviewer
initially collects the birthdays of each member of the household. If a
respondent does not know the date of their birthday (something that
is common amongst older Iraqis) then a random birthday is gener-
ated from a table each interviewer has. The individual with the next
birthday is then selected for interview.” (ORB, 2008b) Senior ORB
staconfirmed to Spagat that they used this method in the ORB2
33 Conceivably migration might resolve this discrepancy if,
among the adult population, there has been large-scale, predomi-
nantly female outflows from Baghdad, large-scale predominantly
male inflows into Baghdad or some combination of the two. We are
not aware of data that would enable a convincing argument along
these lines to be made.
American Association for Public Opinion Research. (2009). AA-
POR Finds Gilbert Burnham in Violation of Ethics Code. Finds Gilbert Burnham in Vio-
lation of Ethics Code/1383.htm.
Biemer, P., & Lyberg, L. (2003). Introduction to survey quality.
Hoboken, New Jersey: Wiley.
Bloomberg School of Public Health. (2007). Johns Hop-
kins Researchers Respond to ORB Poll on Iraq Mortality. tools/iraq/burnham
Bloomberg School of Public Health. (2009). Review Com-
pleted of 2006 Mortality Survey. http:// www. jhsph.
edu/publichealthnews/press releases/2009/iraq review.html.
Bohannon, J. (2006). Iraqi death estimates called too high: methods
faulted. Science,314, 396-397.
Burnham, G., Lafta, R., Doocy, S., & Roberts, L. (2006). Mortality
after the 2003 invasion of Iraq: a cross-sectional cluster sample
survey. The Lancet,368, 1421-1428.
Burnham, G., & Roberts, L. (2007). U.S. must face
huge death toll of Iraqi civilians. Baltimore Sun, October
Civic Worldwide. (2003). Iraqi Civilian War Casualties.
Coghlan, B., Brennan, R. J., Ngoy, P., Dofara, D., Otto, B.,
Clements, M., et al. (2006). Mortality in the Democratic Re-
public of Congo: a nationwide survey. The Lancet,367, 44-51.
Coupland, R., & Meddings, D. (1999). Mortality associated with
use of weapons in armed conflicts, wartime atrocities, and civil-
ian mass shootings: literature review. British Medical Journal,
319, 407-410.
Daponte, B. (2007). Wartime estimates of Iraqi civilian casualties.
International Review of the Red Cross,89, 943-957.
Dardagan, H., Sloboda, J., & Dougherty, J. (2006). Real-
ity checks: some responses to the latest Lancet estimates.
Deltoid. (2004). Breakdown of violent deaths in Lancet study.
EPIC. (2004). An interview with EPIC advisor Richard Garfield. Interview with EPIC A.html.
Gakidou, E., & King, G. (2006). Death by Survey: Estimating
Adult Mortality Without Selection Bias from Sibling Survival
Data. Demography,43, 569-585.
Gamel, K. (2009). AP Impact: Secret Tally has 87,215 Iraqis Dead.
Associated Press, April 23.
Guha-Sapir, D., & Degomme, O. (2007, June). Estimating Mortal-
ity in Civil Conflicts: Lessons from Iraq: Triangulating dierent
types of mortality data in Iraq. CRED Working Paper.
Iraq Body Count. (2003). Adding indierence to injury.
Iraq Body Count. (2005). A Dossier of Civilian Casualties
in Iraq 2003-2005.
Iraq Body Count (IBC). (n.d.).
Iraq Family Health Survey. (2008). IFHS web site. ifhs.htm.
Iraq Family Health Survey Study Group. (2008). Violence-related
mortality in Iraq from 2002 to 2006. New England Journal of
Medicine,358, 484-493.
Iraq Living Conditions Survey 2004. (2005a). Content of IMIRA.
Iraq Living Conditions Survey 2004. (2005b). Volume II:
Analytical report.
FilesByRWDocUNID FileName/KHII-6CC44A-undp-irq-
Johnson, N., Spagat, M., Gourley, S., Onnela, J., & Reinert, G.
(2008). Bias in epidemiological studies of conflict mortality.
Journal of Peace Research,45, 653-664.
KUOW Radio. (2008). How many Iraqi Civilians Died because of
the War.
Laaksonen, S. (2008). Retrospective Two-Stage Cluster Sampling
for Mortality in Iraq. International Journal of Market Research,
50, 403-417.
London School of Hygiene and Tropical Medicine (LSHTM) . (un-
dated). The use of epidemiological tools in conflict-aected pop-
ulations: open-access educational resources for policy-makers.
McElwee, P. (2007). A Million Iraqis Dead? The US
Press Buries the Evidence., http://
Mclean, W. (2007). Baghdad now 60% Christian.
Mclean, W. (2007/2008). Iraq Posts. http://willscommonplacebook.
Moody, J. (2006). Fighting a Hydra: A Note on the Network
Embeddedness of the War on Terror. Structure and Dynamics:
eJournal of Anthropological and Related Sciences,1, 1-5.
Munro, N. (2007, December 1). Iraq’s Slippery Polls. National
Munro, N., & Canon, C. (2008, January 4). Data bomb. National
National Public Radio. (2007). Survey puts Iraq War
Dead above One Million.
NINA. (2008, June 21). MOH Reveals Figures of 2007
Violence Victims.
Details.asp?ar95 VQ =EDGKJM (login required).
ORB. (2007a). Despite Violence only
26% Preferred Life under Sadaam. details.aspx?NewsId=
ORB. (2007b). Iraq Casualties Poll Update. details.aspx?NewsId=
ORB. (2007c). More than 1,000,000 Iraqis Murdered.
ORB. (2008a). Iraqis Confident in Se-
curity by Concerned about Economy. details.aspx?NewsId=
ORB. (2008b). Update on Iraqi Casualty Data. details.aspx?NewsId=
Pedersen, J. (2007). Mortality and Missing Family Members in Iraq
Following the 2003 Invasion. unpublished manuscript, Fafo-
AIS, Norway.
Research Talk. (2006). ESOMAR: Life and Death MR.
Roberts, L., Lafta, R., Garfield, R., Khudhairi, J., & Burnham, G.
(2004). Mortality before and after the 2003 invasion of Iraq:
cluster sample survey. The Lancet,364, 1857-1864.
Rosenblum, M., & van der Laan, M. J. (2009). Confidence Inter-
vals for the Population Mean Tailored to Small Sample Sizes,
with Applications to Survey Sampling. International Journal of
Biostatistics,5, 118.
Roug, L., & Smith, D. (2006, June 25). War’s iraqi death toll tops
50,000. Los Angeles Times.
SMART. (2006). Measuring mortality, nutritional status, and
food security in crisis situations: SMART METHODOLOGY. Methodology 08-07-
Spagat, M. (2008). Mainstreaming and Outlier: The
Quest to Corroborate the Second Lancet Survey of Mor-
tality in Iraq. unpublished manuscript. http://personal.
Spagat, M. (2010). Ethical and Data-Integrity Problems in the Sec-
ond Lancet Survey of Mortality in Iraq. Defense and Peace Eco-
nomics,21(1), 1-41.
Spagat, M. (forthcoming). Ethical and Data-Integrity
Problems in the Second Lancet Survey of Mortality in
Iraq. Defense and Peace Economics. http://personal.rhul
Spiegel, P., & Salama, P. (2000). War and mortality in Kosovo,
1998-99: an epidemiological testimony. Lancet,355, 2204-
Steele, J., & Goldenberg, S. (2008, March 19). What is the Real
Death Toll in Iraq? The Guardian.
Tapp, C., Burkle, F. M., Wilson, K., Takaro, T., Guyatt, G. H.,
Amad, H., et al. (2008). Iraq War Mortality Estimates: a Sys-
tematic Review. Conflict and Health,2(1).
UN Assistance Mission for Iraq (UNAMI). (2007). Hu-
man Rights Report: 1 November - 31 December 2006.
White, D., & Housman, M. (2002). The Navigability of Strong Ties:
Small Worlds, Tie Strength and Network Topology. Complexity,
8, 72-81.
... Second, there are strong discrepancies between the ORB poll and L2 in the geographical pattern of violent deaths, rendering corroboration from the ORB poll double-edged at best for L2. ORB data imply an estimate of about 600,000 violent deaths in Baghdad compared to an estimate of about 150,000 implied by L2 data (Spagat and Dougherty, 2009). Thus, there would need to be roughly 450,000 violent deaths in ...
... Baghdad very close to the national average. In L1, 7-10 out of 21 post-invasion violent deaths outside of the outlying Fallujah cluster were in Baghdad (Spagat and Dougherty, 2009), a governorate which contained 7 out of the 32 clusters outside Fallujah. This suggests that Baghdad had at least 1.5 times the violent death rate of the average outside Fallujah during the L1 coverage period. ...
... Bloomberg (2008a) claims of corroboration for its mortality surveys -from the ILCS, the BBC poll, the IFHS (apparently), Pentagon data and the Baghdad morgue -are all incorrect. The Bloomberg (2008a) claim of corroboration from one ORB poll has some surface plausibility but is ultimately weak because the data and methodology underpinning the ORB estimate are fatally flawed (Spagat and Dougherty, 2009) and, in any case, the ORB poll finds a very different geographical distribution of violent deaths than L2 does. Bloomberg (2008a) makes further incorrect claims about the ILCS, on death certificates and many other matters documented in the present paper. ...
I survey much evidence on mortality in Iraq, including data from the first and second Lancet surveys. The second Lancet survey is inconsistent with all credible and relevant information on levels and trends in violent deaths and on the geographical distribution of violence. I discuss weaknesses in attempts made by The Bloomberg School of Public Health and authors of the second Lancet survey to claim corroboration for the second Lancet survey from other sources. These attempts notwithstanding, the second Lancet survey is a clear outlier within a wide body of evidence on mortality in Iraq.
... IBC data are readily comparable to those from any of the five surveys because IBC data is daily, covers the entire conflict and is compiled using a uniform methodology. Table 1, taken from Spagat and Dougherty (2010), compares the five surveys along various dimensions, with IBC units serving as a measuring rod in some of the rows. ...
... AAPOR (2009b) lists the methodological details that the principal researcher on L2 has refused to disclose.30 Spagat (2010) includes evaluations of the methodological quality of the ILCS, IHFS and L2 andSpagat and Dougherty (2010) evaluates the quality of the ORB survey. ...
... Special national surveys have been used to estimate excess mortality in several conflicts, such as those in the Democratic Republic of the Congo, Afghanistan, and Iraq, often resulting in much debate about the methods and results. [18][19][20][21] Large numbers of small-scale surveys, generally conducted by non-government organisations involved in the response, are frequently conducted in crisis situations but might give a biased picture of overall mortality trends. ...
Women and children bear substantial morbidity and mortality as a result of armed conflicts. This Series paper focuses on the direct (due to violence) and indirect health effects of armed conflict on women and children (including adolescents) worldwide. We estimate that nearly 36 million children and 16 million women were displaced in 2017, on the basis of international databases of refugees and internally displaced populations. From geospatial analyses we estimate that the number of non-displaced women and children living dangerously close to armed conflict (within 50 km) increased from 185 million women and 250 million children in 2000, to 265 million women and 368 million children in 2017. Women's and children's mortality risk from non-violent causes increases substantially in response to nearby conflict, with more intense and more chronic conflicts leading to greater mortality increases. More than 10 million deaths in children younger than 5 years can be attributed to conflict between 1995 and 2015 globally. Women of reproductive ages living near high intensity conflicts have three times higher mortality than do women in peaceful settings. Current research provides fragmentary evidence about how armed conflict indirectly affects the survival chances of women and children through malnutrition, physical injuries, infectious diseases, poor mental health, and poor sexual and reproductive health, but major systematic evidence is sparse, hampering the design and implementation of essential interventions for mitigating the harms of armed conflicts.
... Data collected were coded and analyzed using statistical package for social sciences (SPSS), version 23. under windows 7. Simple descriptive analysis in the form of percentage distribution, Means, Simple Linear Regression, R Square, were done. However, suitable inferential statistics was done to test the level of significance with a confidence level of 0.05 (Thomas andHeck, 2001: Spagat andDougherty, 2010). ...
Full-text available
Stock Markets are an important part of the economics of the Arabian countries. Securities traded on a stock exchange include stock issued by listed firms and bonds. The current study aims to investigate the effect of Stock Split on Market Capitalization and Market Value in Hospitality and Tourism in three Bourses. For the purpose of the study, determinations of whether Stock Split made by Hospitality and Tourism Sector Index’s firms on the Egyptian Exchange, Saudi Arabian Stock Exchange and Bahrain Bourse have an effect on the splitting firms. The study uses historical quantitative data during the period (2010:2018) collected from (ASMs). Based on linear regression analysis, the findings in Travel and Leisure Index in The Egyptian Exchange indicated that there is a significant positive relationship effect of Stock Split on Market Capitalization, Market Value, and Market Value on Market Capitalization. Furthermore, the results showed that in Tourism and Hotels index in Saudi Arabian Stock Exchange there is a significant positive relationship effect of Stock Split on Market Capitalization, Market Value, and Market Value on Market Capitalization. In addition, the results showed that in Hotels and Tourism in Bahrain Bourse there is a significant positive relationship between of Stock Split on \Market Capitalization, Market Value, and Market Value on Market Capitalization. The results also revealed that Stock Split on Market Capitalization, Market Value, and Market Value on Market Capitalization was considered the most criterion validity in the three Bourses. Keywords: Arabian Stock Markets (ASMs), Travel & Leisure Sector Index, Finance, Investment, Hospitality & Tourism Projects, Stock Split, Market Value, Market Capitalization.
... Estimates for the total war-dead from the Iraq War instigated in 2003 vary widely; a credible assessment suggests the number of deaths in that country from war and related violence and lawlessness is more than 450,000 people (Hagopian et al. 2013). Higher estimates have been published, but have come under criticism for methodological shortcomings (e.g., Spagat and Dougherty 2010). Even a tenth of the number reported by Hagopian et al. (2013) though is a horrific body-count emblematic of profound individual, familial and societal loss and suffering. ...
Full-text available
Markets and marketing are integral to human welfare and survival. When used however for the purposes of war and other systemically violent conflict, they can be devastating and pose an existential threat to humanity. Drawing on experience in war-ravaged and recovering economies, the author examines a stream of research on marketing systems disrupted or destroyed by war. Some underlying conditions and predictors of war and its peaceful resolution are introduced, including social traps and their mitigation or elimination. An argument is revisited for marketing as a form of constructive engagement, which must be implemented to affect and to develop equitable and sustainable marketing systems, flourishing communities, societal well-being and sustainable peace. The article concludes with some considerations for further research.
... The only reasonable explanation is that serious bias affects the 'representativeness' of the ORB surveys. ORB was previously criticised by an academic paper for its opaque and 'incomplete disclosure' of method and 'important irregularities' in their estimates of deaths from the war in Iraq (Spagat and Dougherty 2010). That unreliability is present in their Syrian data. ...
... The most reasonable explanation is that serious bias affects the 'representativeness' of selection for the ORB surveys. ORB was previously criticised by an academic paper for its opaque and 'incomplete disclosure' of method and 'important irregularities' in their estimates of deaths from the war in Iraq (Spagat and Dougherty 2010). That unreliability is also present in their Syrian data. ...
Full-text available
Normal ethical notions of avoiding conflicts of interest, searching for independent evidence and disqualifying self-serving claims from belligerent parties have been ignored in much of the western debate. This toxic atmosphere invites further fabrications, repeated to credulous audiences, even when the lies used to justify previous invasions (e.g. of Iraq in 2003) and dirty wars (e.g. in Libya, 2011)are still relatively fresh in our minds. As in previous wars, the aim is to demonise the enemy, by use of repeated atrocity claims, and so mobilise popular support behind the war (Knightley 2001)
This article reviews sample survey methods for estimating the number of war-related deaths. That different surveys using seemingly similar methodologies have produced widely differing estimates of Iraq war deaths has made this method somewhat controversial. The existing literature focuses strongly on death, perhaps because this is the most dramatic human cost of war. The article identifies a number of factors to consider in evaluating conflict surveys, including sampling procedures, mechanisms for ensuring the integrity of the data-collection process, the appropriateness of extrapolations, and the setting of baselines in the case of excess-death calculations.
It has become a commonplace to say that war has changed radically since the early twentieth century to the point where civilians now comprise some 80 or 90% of war victims. This proposition has been supported by many writers and academics, some United Nations agencies, and the European Union in its European Security Strategy. Yet it rests on shaky foundations. It is possible that in some particular conflicts nine out of ten deaths are of civilians, but the proposition does not hold up as a generalisation about all war in the past two decades. There is persuasive evidence that certain wars have had civilian death tolls far lower than 90%. The proposition, intended to alert the world to the importance of protecting civilians, has probably had the unintended effect of reinforcing cynicism about efforts to limit the human costs of war.
Full-text available
Two-stage sampling has commonly been used in surveys of households and individuals. The standard strategy is first to stratify the frame population, then determine a reasonable number of primary sampling units (PSUs) within each stratum, to choose some of these with probability proportional to size (first stage) and, finally, to draw sampled units randomly within each cluster (second stage). Good determination of PSUs is the key point in this strategy. It is advantageous if the areas are fairly small. For each stage, the selection should be based on probability principles so that correct inclusion probabilities can be calculated for each individual of the target population. This requirement is not easy to satisfy well in standard surveys in developed countries. It is expected that the problems met will be more complex in the developing world, and even harder in countries experiencing conflict. A really challenging example is the Iraq Mortality Survey (IMS), which was conducted in the summer of 2006. This survey is exceptional also in the sense that the main study variables are deaths due to both violent and non-violent causes. Such variables are not used in surveys in developed countries since reasonably good data are available from records of death registers or lists. Such records have not been considered as reliable in Iraq, hence survey methodology was attempted. The results on estimated deaths due to violence were surprisingly high. This aroused lively debate around the world. The paper comments on this debate, while trying to reconstruct country-level estimates using the initial micro data received from the IMS team. A survey methodologist such as this author cannot be happy with these data, hence many doubts are expressed here about the published estimates.
Full-text available
While we typically count the number of people directly affected by violence, war and state action with casualty reports, we often ignore how these people are embedded in larger networks. In this paper, I estimate the number of people who know somebody who has been killed, injured or detained in the US war on terrorism in Iraq and Afghanistan, and thus likely at higher risk to joining a resistance movement. Since such numbers are constantly changing, an on-line web-calculator for the estimates is included so that readers can make their own estimates as well. Better understanding how friends and families are affected by such actions will help us understand the wider implications of military action.
Preface. Chapter 1. The Evolution of Survey Process Quality. 1.1 The Concept of a Survey. 1.2 Types of Surveys. 1.3 Brief History of Survey Methodology. 1.4 The Quality Revolution. 1.5 Definitions of Quality and Quality in Statistical Organizations. 1.6 Measuring Quality. 1.7 Improving Quality. 1.8 Quality in a Nutshell. Chapter 2. The Survey Process and Data Quality. 2.1 Overview of the Survey Process. 2.2 Data Quality and Total Survey Error. 2.3 Decomposing Nonsampling Error into Its Component Parts. 2.4 Gauging the Magnitude of Total Survey Error. 2.5 Mean Squared Error. 2.6 An Illustration of the Concepts. Chapter 3. Coverage and Nonresponse Error. 3.1 Coverage Error. 3.2 Measures of Coverage Bias. 3.3 Reducing Coverage Bias. 3.4 Unit Nonresponse Error. 3.5 Calculating Response Rates. 3.6 Reducing Nonresponse Bias. Chapter 4. The Measurement Process and Its Implications for Questionnaire Design. 4.1Components of Measurement Error. 4.2 Errors Arising from the Questionnaire Design. 4.3 Understanding the Response Process. Chapter 5. Errors Due to Interviewers and Interviewing. 5.1 Role of the Interviewer. 5.2 Interviewer Variability. 5.3 Design Factors that Influence Interviewer Effects. 5.4 Evaluation of Interviewer Performance. Chapter 6. Data Collection Modes and Associated Errors. 6.1 Modes of Data Collection. 6.2 Decision Regarding Mode. 6.3 Some Examples of Mode Effects. Chapter 7. Data Processing: Errors and Their Control. 7.1 Overview of Data Processing Steps. 7.2 Nature of Data Processing Error. 7.3 Data Capture Errors. 7.4 Post-Data Capture Editing. 7.5 Coding. 7.6 File Preparation. 7.7 Applications of Continuous Quality Improvement: The Case of Coding. 7.8 Integration Activities. Chapter 8. Overview of Survey Error Evaluation Methods. 8.1 Purposes of Survey Error Evaluation. 8.2 Evaluation Methods for Designing and Pretesting Surveys. 8.3 Methods for Monitoring and Controlling Data Quality. 8.4 Postsurvey Evaluations. 8.5 Summary of Evaluation Methods. Chapter 9. Sampling Error. 9.1 Brief History of Sampling. 9.2 Nonrandom Sampling Methods. 9.3 Simple Random Sampling. 9.4 Statistical Inference in the Presence of Nonsampling Errors. 9.5 Other Methods of Random Sampling. 9.6 Concluding Remarks. Chapter 10.1 Practical Survey Design for Minimizing Total Survey Error. 10.1 Balance Between Cost, Survey Error, and Other Quality Features. 10.2 Planning a Survey for Optimal Quality. 10.3 Documenting Survey Quality. 10.4 Organizational Issues Related to Survey Quality. References. Index.
Challenges exist when making reliable and valid estimates of civilian mortality due to war. This article first discusses a framework used to examine war's impact on civilians and then considers challenges common to each statistical approach taken to estimate civilian casualties. It examines the different approaches that have been used to estimate civilian casualties associated with the recent fighting in Iraq to date and compares the results of different approaches. The author concludes by proposing that after fighting has ceased, other approaches to estimating Iraqi civilian mortality, such as post-war retrospective surveys and demographic analysis, should be employed.
We examine data on and models of small world properties and parameters of social networks. Our focus, on tie-strength, multilevel networks and searchability in strong-tie social networks, allows us to extend some of the questions and findings of recent research and the fit of small world models to sociological and anthropological data on human communities. We offer a ***navigability of strong ties*** hypothesis about network topologies tested with data from kinship systems, and potentially applicable to corporate cultures and business networks. A small world (SW) is a (large) graph with both lo-cal clustering and, on aver-age, short distances between nodes [1,2]. Short distances pro-mote accessibility, whereas local clustering and redundancy of edges, as some research suggests [3,4], promotes robustness to disconnection and, through multiple independent pathways, reliable accessibility as well. For paths to transmit materials and information via network traversal, a small world also requires navigability. This was the property investigated in the first small world experiment by Travers and Milgram [5]: Could people randomly selected in Omaha, Nebraska, successfully send letters to a predeter-mined target in Boston, when asked to direct their letters to single acquaintances who are asked in turn to forward the letters through what becomes a chain of personal acquaintan-ces? In many cases this task was accomplished in fewer than six steps, but success required letters sent to acquaintances who were successively closer, geographically or occupation-ally, to the target. The problem of navigability is whether the next step in such chains will be any closer to the target than the last. This cannot occur in a network of edges generated with uniform probabilities, as Kleinberg showed [6]. SW net-works with random rewiring, like random networks gener-ally, lack the ability to find the target person quickly via successive links in the network. Kleinberg also showed a far stronger result: the ability of decentralized algorithms to find short paths by sending messages along their incident edges using only local information about them depends, in regular lattices in which edge probability is an inverse power of lattice distance, on a unique value of that exactly matches the dimensionality of the lattice. The short paths that are relevant in this context are those whose lengths are bounded by a polynomial in logN, where N is the number of nodes, because this is what defines algorithmic efficiency for a random graph [7]. The right power-law decay of link frequency—in relation to geometric distance— creates fewer long jumps in the right direction that act as shortcuts Douglas R. White is a mathematical anthropologist and a professor at the University of California, Irvine, where he is also Graduate Director of Social Networks and member of the Institute for Mathematical Behavioral Sciences. As part of a Traveling Scholar program of the University of Minne-sota, where he received his Ph.D. in 1969, he was an NSF Cooperative Fellow at the University of Michigan in math-ematical psychology and anthropology and an NIMH Pre-doctoral Fellow at Columbia University in anthropology and mathematical sociology. He advanced to Associate Pro-fessor at the University of Pittsburgh (1967–1979) where he co-directed the Cross-Cultural Cumulative Coding Center and took leave (1971–1973) to co-direct the National Lan-guage Study project for the Irish Republic. He is a Directeur des Etudes Invité at the Ecole des Hautes Etudes in Paris and recipient of the A. von Humboldt distinguished senior sci-entist prize in Germany.
I survey much evidence on mortality in Iraq, including data from the first and second Lancet surveys. The second Lancet survey is inconsistent with all credible and relevant information on levels and trends in violent deaths and on the geographical distribution of violence. I discuss weaknesses in attempts made by The Bloomberg School of Public Health and authors of the second Lancet survey to claim corroboration for the second Lancet survey from other sources. These attempts notwithstanding, the second Lancet survey is a clear outlier within a wide body of evidence on mortality in Iraq.
This paper considers the second Lancet survey of mortality in Iraq published in October 2006. It presents some evidence suggesting ethical violations to the survey's respondents including endangerment, privacy breaches and violations in obtaining informed consent. Breaches of minimal disclosure standards examined include non-disclosure of the survey's questionnaire, data-entry form, data matching anonymised interviewer identifications with households and sample design. The paper also presents some evidence relating to data fabrication and falsification, which falls into nine broad categories. This evidence suggests that this survey cannot be considered a reliable or valid contribution towards knowledge about the extent of mortality in Iraq since 2003. Editor's Note: The authors of the Lancet II Study were given the opportunity to reply to this article. No reply has been forthcoming.
The validity of standard confidence intervals constructed in survey sampling is based on the central limit theorem. For small sample sizes, the central limit theorem may give a poor approximation, resulting in confidence intervals that are misleading. We discuss this issue and propose methods for constructing confidence intervals for the population mean tailored to small sample sizes.We present a simple approach for constructing confidence intervals for the population mean based on tail bounds for the sample mean that are correct for all sample sizes. Bernstein's inequality provides one such tail bound. The resulting confidence intervals have guaranteed coverage probability under much weaker assumptions than are required for standard methods. A drawback of this approach, as we show, is that these confidence intervals are often quite wide. In response to this, we present a method for constructing much narrower confidence intervals, which are better suited for practical applications, and that are still more robust than confidence intervals based on standard methods, when dealing with small sample sizes. We show how to extend our approaches to much more general estimation problems than estimating the sample mean. We describe how these methods can be used to obtain more reliable confidence intervals in survey sampling. As a concrete example, we construct confidence intervals using our methods for the number of violent deaths between March 2003 and July 2006 in Iraq, based on data from the study ``Mortality after the 2003 invasion of Iraq: A cross sectional cluster sample survey,'' by Burnham et al. (2006).