Civilian Victims in an Asymmetrical Conflict Operation Enduring Freedom, Afghanistan Aldo A Benini, Lawrence H Moulton

commentators on, recent Western military
interventions, notably in Iraq in 2003, and
are likely to become part of the ideological
arsenal of future wars. Here, we briefly
present these claims, then translate them
into specific hypotheses that can be tested
against our data on the communities
affected by Operation Enduring Freedom.
Readers interested in its political and
military aspects are referred to Cottey
(2003).

The first is that compared to the victims

of the former misrule and oppression, losses
caused by regime-changing military inter-
ventions are few. While Shaw himself is very
careful about such claims, Western poli-
ticians and media publicized many such pro-
jections in the run-up to those wars, as well
as to the recent one in Iraq. In this latter case,
for example, expected civilian losses were
contrasted, in almost routinized argument,
with totals from several past categories of
victims, including civilian victims of internal
repression and soldiers lost to the war with
Iran, estimated to exceed 1 million. The
topic endures; for a postwar example, see
Falconer (2003), who places mass murder in
Iraq prominently in the tradition of
Rummel’s research in murder by states
(Rummel, 1997).

The second claim is that the number of

civilians killed in these recent wars is very
small compared to the major wars that the
USA fought in the 20th century, namely, the
World Wars and the wars in Vietnam and
Korea (Shaw, 2002: 346). Much of this trend
is credited to the so-called ‘Revolution in
Military Affairs’. Modern precision
weaponry and improved intelligence resulted
in better targeting decisions by Western
military, reducing, if not eliminating, indis-
criminate fire against civilians. Civilian mas-
sacres, therefore, tend to be, in Shaw’s words,
‘small and accidental’, albeit at the same time
accepted and ‘programmed into the risk
analysis of the war’ (Shaw, 2002: 349). In

ideological terms, they are, if you will, the
price of liberation.

Using detailed data on 600 local com-

munities exposed to hostilities during Oper-
ation Enduring Freedom, we argue that the
Afghan case lends qualified support to the
first thesis: that civilian losses can be reduced
as a result of the intervention. It does not
support the ‘small massacre’ claim once the
motivation (‘accidental’) is taken out of the
equation and the actions of the opponents
are factored in. Some of the factors that
affected the distribution of civilian victims
during Operation Enduring Freedom may
operate in other military interventions by the
West. If so, we may expect local communi-
ties in countries other than Afghanistan to
suffer losses at such elevated levels that few
observers will want to qualify them as ‘small
massacres’.

Violence Against Civilians

We make three specific points, which we
then submit to testing against the Afghanis-
tan data. First, attention to direct violence
from the victorious Western forces under-
estimates the total number of civilian
victims. Some of the estimates offered on
civilian losses in Afghanistan, for example,
relied mainly on media reports as their
source; the media, however, privileged inci-
dents related to high-tech weaponry, to the
detriment of other causes of civilian deaths
and injuries such as landmine and unex-
ploded ordnance strikes, as well as non-
Western ground forces. When civilian-loss
information is collected directly from the
affected local communities, the tallies will be
substantially higher than those made by
observers restricted to media accounts. We
show this difference for our survey data vs.
various non-survey-based earlier estimates.

Our second point is that asymmetrical

conflict, as epitomized by the recent wars in
Afghanistan and Iraq, creates a highly

j o u r n a l o f P

E A C E

E S E A RC H

volume 41 / number 4 / july 2004

404

01 044474 (ds) 25/5/04 2:29 pm Page 404

unequal distribution of civilian victims
across affected local communities. On one
side, a higher than expected portion of all
communities that ever experience combat
come away with no civilian victims. On the
other hand, a small but significant minority
of communities takes heavy losses. Some of
these high-impact communities are places
where the defendants dig in and are met with
massive fire, and where civilians become
entrapped in battles. In other places, civilian
massacres may be intentional, with passing
local forces settling scores with opponent
ethnic groups or political adversaries. The
Afghan data document both extremes:
numerous zero-victim communities as well
as many outliers on the high end.

Our third point can be called the ‘back-

ground noise of a violent universe’. The
structure of the pre-intervention violence
shapes the pattern of civilian victims during
the war and, in fact, does so massively. One
implication is that communities with high
numbers of civilian victims in prewar
periods tend to be those again smitten hard
during the war. Some of this is expected,
particularly in countries with massive
prewar landmine contamination (where
residents, subject to numerous mine strikes
earlier, may have to flee across hazardous
areas). Other parts of the correlation may be
contingent on entrenched rifts in society
that run parallel with both prewar and
current military frontlines. An example,
known from war footage and borne out by
our data, is the frontline between the
Taliban and the Northern Alliance in
northern Afghanistan, which remained
static for quite some time after Operation
Enduring Freedom was begun.

Our dragnet in Afghanistan, therefore, is

larger than Shaw’s. It is larger in temporal,
social, and technical aspects. Our sources
detailed local victims not only for the time
after 9/11, but also for a one-year period
before the war. They did not discriminate by

any particular party to the war; all victims of
violence were counted, regardless of what
party was the source of the violence. They
include victims both from direct violence
and from landmine and unexploded
ordnance (UXO) strikes.

Our plan is to place our broad victim

picture within some of the recent conflict
literature that speaks to persistent wars
and/or Western military interventions. We
then describe the Afghanistan data and
the method by which they were collected,
the claims regarding civilian victims, and the
limits to their reliability. We will then
compare the overall tally from the 600
affected communities to estimates offered
by other researchers, without a detailed
review of their methodologies. We will sum-
marily describe demographic and war-
related features of those communities and
focus on describing their numbers of
victims. In the central part of our argument,
we develop a conceptual model of the dif-
ferential numbers of local victims and
estimate the significance of various factors,
notably the influence of violence suffered in
the year prior to 9/11. We discuss various
findings and place them in the perspective
of future wars that the West may fight and
of the research on civilian victims that these
wars, or the prospects for them, may
motivate.

Shaw in no way endorses the ‘relegitima-

tion of war’ that he observes as a recent (post-
Cold War) political development in the
West. In fact, he concludes that the legiti-
macy of war, ‘even its newly refined justifi-
cations’ in response to terrorism (2002: 357),
is fragile and liable to be reversed. Also, with
regard to Afghanistan, he discusses indirectly
caused civilian deaths as a result of Western
action (mainly reviewing Conetta, 2002).
Our article is concerned with two specific
implications, not with questions of legiti-
macy at large. Our data do not elucidate the
indirect-victims question.

Aldo A. Benini & Lawrence H. Moulton

I C T I M S I N

S Y M M E T R I C A L

O N F L I C T

405

01 044474 (ds) 25/5/04 2:29 pm Page 405

Operation Enduring Freedom in
Perspective

The wars that have devastated Afghanistan
since 1978 form a series of civil wars that
were internationalized to variable degrees,
most openly so during the long Soviet
military intervention and in the fulminant
Operation Enduring Freedom in 2001 and
the presence of Western forces since then.
These wars created an internal dynamic with
two important outcomes. In the long run,
from 1978 to 2001, large numbers of
Afghans were killed. The deaths came in
waves that moved with the different stages
of war. Goodson (2001: 87–88), whose
detailed analysis offers an eight-stage war
history prior to 2001, collects various esti-
mates assuming total war deaths in the range
of 1.5 to more than 2 million (p. 93). These
deaths include combatants. To those he adds
682,000 wounded persons, which he con-
siders a low estimate (p. 94). Other esti-
mates are consistent with Goodson’s; for
example, the Correlates of War Project
(Sarkees, 2000) uses 1.3 million as its total
battle-death estimate for the period
1978–92.

Second, the wars were persistent. Collier

et al. (2003: 82) established an average of
approximately 40 months for the duration of
civil wars started in the 1960s and 1970s;
this value jumped to 125 months for wars
started in the 1980s. The series of wars in
Afghanistan have lasted for more than
double that time. Although by 2001 the
Taliban were in control of most of
the national territory, they were still at war
in the northern and eastern regions. As a
result, when Operation Enduring Freedom
was launched, there were basically two wars
going on in parallel. The ground war was
fought between the Taliban and a Northern
Alliance boosted with Special Forces and
fresh supplies. The air support, epitomizing
the ‘Revolution in Military Affairs’, marked

the short-run addition to the conduct of this
long-drawn conflict.

Some may find a parallel here with the

two simultaneous wars that Kaldor (1999)
observed in Yugoslavia in 1999. On the
ground, Milosevic’s forces waged war against
the Kosovar Albanians. From the air, NATO
fought a ‘spectacle war’ (Kaldor, 1999: 154)
scripted by the Revolution in Military
Affairs. This pattern obtained in the 2003
Iraq war in very limited degree. True, the
Kurdish Peshmerga were there as an armed
opposition group, but hostilities between
them and Iraqi government forces had been
rare for the past ten years, and the Peshmerga
played a minor role in the war compared to
coalition ground forces.

The varying importance that local armed

opposition forces had for the Western
victories – greatest in Afghanistan 2001,
modest in Kosovo 1999, slight in Iraq 2003
– may ultimately make it difficult to charac-
terize these recent interventions as a
common type of armed conflict. For
example, Mueller (2003: 511, n. 7), who
postulates the contemporary obsolescence of
war and the need to police its remnants,
expresses doubts about the neat limits of
Afghanistan and Iraq engagements; their
‘messy aftermaths’ may ultimately make such
interventions unappealing.

With this cautious note, Mueller joins

Shaw’s conclusion that the justifications of
Western interventions remain fragile.
Because some of these wars are so recent,
none of these authors can say much defini-
tive about their nature and prospects for rep-
etition. There is greater agreement on some
of the long-run and short-term effects in the
affected countries. A growing literature looks
into the excess mortality from the effects of
war other than direct violence. Deaths
among Iraqi children, attributed to poor
nutrition, sanitation, and medical care, were
the subject of intensive study after the first
Gulf War; the estimates that Mueller (1995:

j o u r n a l o f P

E A C E

E S E A RC H

volume 41 / number 4 / july 2004

406

01 044474 (ds) 25/5/04 2:29 pm Page 406

105–106) reports all exceed, by a multiple,
those of civilian lives lost during the war. For
Afghanistan, Collier et al. (2003: 24) show a
65% increase in infant mortality due to civil
war. These are effects that continue ‘long
after the shooting stops’ (Ghobarah, Huth &
Russett, 2003: 189).

Agreement on short-run effects seems

firm as far as civilian victims are concerned.
The estimates do not exceed 10,000 killed
during any of the Western military inter-
ventions, by both sides to the conflict. Table
I brings typical estimates together, with the
Iraq 2003 estimate as a running tally. It could
be argued that these death tolls are signifi-
cantly smaller than the loss of civilian lives
during the precursory developments. These
include the violent dissolution of the
Yugoslav federation, the atrocities commit-
ted by the Iraqi government (in line with
Rummel’s thesis that undemocratic regimes
tend to kill their citizens in numbers exceed-
ing the battle deaths; see Rummel, 1995,
1997), and again, the immense death toll of
the Afghan wars since 1978. However, our
subject is the civilian victims in Afghanistan
during the limited period of Operation
Enduring Freedom, for which relatively low
estimates were offered. To comment on
these, we turn to our data.

600 Communities

Sample and Data Collection
This dataset on Afghan towns and villages
exposed to hostilities after 11 September
2001 is the by-product of a landmine and
UXO contamination assessment. The assess-
ment, with a view to creating an inventory
of freshly contaminated sites for rapid clear-
ance purposes, was done by the Afghan
NGO Mine Clearance Planning Agency
(MCPA) with the help of the Vietnam
Veterans of America Foundation (VVAF), an
advocacy and victim assistance organization
in humanitarian mine action (Benini &
Donahue, 2003).

MCPA had maintained a staff of several

thousand active in minefield surveys,
marking, and area reduction over the past ten
years and enjoyed acceptance throughout the
country. In spring and early summer 2002,
MCPA interviewer teams visited all com-
munities suspected to have been subject to
airstrikes or ground operations during Oper-
ation Enduring Freedom. These communi-
ties – villages or urban neighborhoods – had
been nominated by provincial administra-
tions and by neighboring communities;
moreover, MCPA had access to coalition air-
strike imprints. The teams visited 747
suspect communities, among which exactly
600 were determined to have had at least one
airstrike or ground operation. These affected
communities were scattered in 102 districts
in 25 of the 32 provinces. This expert- and

Aldo A. Benini & Lawrence H. Moulton

I C T I M S I N

S Y M M E T R I C A L

O N F L I C T

407

Table I.

Victim Estimates for Western Military Interventions, 1991–2003

Conflict

Estimates of civilians killed

Sources

Iraq 1991

2,500–3,500

Mueller (1995: 103)

Kosovo 1999

Ethnic cleansing: 10–12,000 [incl.

Shaw (2002: 347); Kaldor (1999: 157,

those before NATO intervened]

162)

Air campaign: 500–1,400

Afghanistan 2001

150–3,600

Various. See discussion, page 417.

Iraq 2003

7,300–9,200 [as of 26 September 2003]

Dardagan (2003)

01 044474 (ds) 25/5/04 2:29 pm Page 407

respondent-driven sampling was exhausted
for all leads and, because MCPA teams faced
virtually no security restrictions, is con-
sidered to be close to a full census of the
affected communities. Because these cluster
strongly in many areas and thus are generally
known to some neighbors, sampling among
non-suspected communities was not done.

In each community confirmed exposed to

post-9/11 hostilities, a team would conduct
an interview, using a modular questionnaire,
with a small group of local key informants.
These groups, variable in size and composi-
tion, would share information on dates and
types of hostilities, prewar and current popu-
lation, old and new contaminated areas and
broad types of munitions, types and numbers
of property damaged or destroyed, and
finally, victims. Victim numbers were elicited,
broken down in several dimensions – by age
and sex, cause (direct violence vs. landmine
and unexploded ordnance strikes), outcome
(deaths and injuries) – as well as two periods
of time. Counts were requested of all who
had come to harm between 11 September
2001 and the date of survey – a 9-month
period on average. Retrospective counts were
requested for the period of 12 months prior
to 9/11. No attempt was made to attribute
the violence that caused these victims to any
specific parties to the conflict. Before leaving
the community, teams took GPS (Global
Positioning System) measurements of the
coordinates of a central location such as its
mosque.

The affected community, and not, say, the

distinct danger area, violent incident, or indi-
vidual victim, is the unit of analysis. We use
the term ‘victim’ to designate both fatalities
and survivors from injuries and use specific
terms when we mean the one but not the
other. Obviously, the reliability of the claims
to victims that the surveyed communities put
forward is critical. In complex emergencies
with long traditions of relief, such as
Afghanistan’s, communities perceive an

incentive to overstate levels of suffering or
populations in need. For example, the 2,997
landmine and unexploded ordnance victims
reported by the 600 communities for an
average nine-month period after 9/11 stand
in marked contrast to the 658 for the whole
country on whom the International Com-
mittee of the Red Cross (ICRC) collected
data for the six months between January and
June 2002 (ICBL, 2002: 603). While the
ICRC says that its lists are not comprehensive
(Desvignes, 2003), some of the MCPA-inter-
viewed informants may have exaggerated
their local numbers. Similarly, for counts or
estimates proffered for pre-9/11 victim
numbers, retrospective bias (Tourangeau,
Rips & Rasinski, 2000: 125), due to attenu-
ation of memory or backwards extrapolation
from post-9/11 figures, cannot be excluded.

Two partial reliability tests using external

data are available. In addition, the significant
coefficients on the victim levels of neighboring
communities are meaningful in this context.
Claims regarding victims of direct violence
were compared to those reported in the
Human Rights Watch study on the use of
cluster bombs (HRW, 2002). Three of the
HRW case study villages are documented in
sufficient detail in order to match claims. For
two villages with direct correspondence, figures
for victims from the attacks and from bomblets
exploding later match closely; the third HRW
case study village is subsumed in the statistics
of an MCPA village 200 meters away.

For a second test, point coordinates of

over 6,000 contaminated areas recorded
prior to the war were used. This external
database (from the United Nations Mine
Action Programme for Afghanistan) was
considered reasonably complete for the
central region of the country, where techni-
cal survey teams used to enjoy access, but not
for the other four regions. It seems reason-
able to assume that communities claiming
some landmine or UXO victims from the
pre-9/11 period should be in close contact

j o u r n a l o f P

E A C E

E S E A RC H

volume 41 / number 4 / july 2004

408

01 044474 (ds) 25/5/04 2:29 pm Page 408

with some of those contaminated areas.
Although ‘close contact’ is hard to define,
and the average diameter of those areas is not
known, a distance of not more than 3 km
seems plausible if frequent exposure to the
hazard is to be expected. Two-thirds of the
86 central-region communities with such
claims in fact are within this distance from
the nearest recorded area.

The strongest support for the reliability of

the victim numbers, however, comes from the
high correlation of victim counts among neigh-
boring communities. This is seen in the regres-
sion model coefficients of the variables that
contain the numbers of victims in communi-
ties within 3 km. We will detail this for the
model of victims from direct violence further
below, but it also holds for victims from mine
and UXO strikes. If the victim data were not
very reliable, we would expect these coefficients
to be much closer to zero.

Community Demographics
Of the 600 communities exposed after 11
September 2001, all offered current popu-
lation estimates. Between them, the com-
munities totaled slightly less than 2 million
residents.

This corresponds to less than 7% of one

of the currently available estimates for the
national population (27,756,000, mid-year
2002

). Communities were of extremely

unequal size extending from totally deserted
villages to the 60,000 residents estimated
for one of the urban neighborhoods in
Kabul.

The hostilities took place in 26 of the 32

provinces. Much of the fighting was
concentrated around the centers of power of
the Taliban, in the provinces of Kabul (166
affected communities) and Kandahar (88),

with a secondary concentration in two
northern provinces. One of them – Takhar
(82) – had been cut by longstanding front-
lines between the Taliban and the Northern
Alliance. The other – Kunduz (39) – was the
scene of concentrated fighting against one of
the Taliban strongholds.

The pattern of population displacement

corresponded to the fighting concentration,
which was regionally very unequal. About a
third of the 600 communities (211) reported
that some or all of their residents had fled to
the outside; of 216,000 such displaced
persons, 71,000 were from Takhar and
55,000 from Kabul provinces.

That pattern extends to the community

level: among 497 communities with prewar
population estimates, 298 (60%) experienced
no population flight; 172 communities (35%)
saw some but not all people move away; and
a small group (27 communities, 5%) saw its
entire population flee from the war.

On the other hand, the returnees between

9/11 and the survey dates outnumbered the
displaced persons. An estimated 289,000
persons returned to the 600 communities.
Most of the gain, however, was in the 166
communities of Kabul province, which
received 164,000 returnees.

In sum, the communities exposed to hos-

tilities form a minority phenomenon, both
by their small number among the over
30,000 settlements registered by the
Afghanistan Information Management
Center, a UN humanitarian information
unit, and by the size of their combined popu-
lations. However, they form stark regional
clusters, around pre-9/11 frontlines and the
centers of the defeated regime.

Hostilities and Victims

Exposure
Significant numbers of communities started
becoming the object of ground attacks and

Aldo A. Benini & Lawrence H. Moulton

I C T I M S I N

S Y M M E T R I C A L

O N F L I C T

409

To be precise, 1,919,752, as added up from local key

informant estimates. These people were living in an esti-
mated 300,689 households.

US Bureau of the Census (2003). Population estimates

for Afghanistan vary widely. One of our reviewers had seen
estimates ranging from 15 to 30 million people.

01 044474 (ds) 25/5/04 2:29 pm Page 409

airstrikes three weeks after 11 September
2001. As the war intensified, larger and larger
numbers were engulfed, and stayed exposed
for longer periods of time. Figure 1 details the
flow by weekly cohort. Clearly, two peaks can
be made out in terms of entry cohorts and
periods of exposure, although the dynamics
are slightly different between those two
criteria. Starting in the first week of October,
a significant number of communities became
exposed to hostilities, primarily as a result of
coalition airstrikes on Kabul, Kandahar, and
Kunduz. This first wave peaked in the third
week of October. The first of its weekly
cohorts saw hostilities go on for longer than
any other cohort, perhaps because the Taliban
still believed they could resist.

By the end of the first wave of entrants,

ground attacks by the Northern Alliance in
Takhar province had finally gained mo-
mentum. As it progressed, 153 new entrants

created another sharp peak in the week
starting 5 November. The median exposure
of this cohort (9 days) was shorter than that
of the two preceding cohorts, indicating that
the Taliban were clearing from their
positions more readily. The fall, on 9
November, of the city of Mazar-i-Sharif to
the Northern Alliance was a watershed event.
Communities entering the war thereafter
had only brief median exposures. For most,
the war was over.

Of the 582 communities with defined

start and end points, 75% had less than 17
days of exposure to hostilities, and 146
(25%) of the communities were exposed for
only 1 day. The median exposure was 6
days. However, this varied greatly among
regions, from 1 day for communities in the
South, to 9 in the center, and 10 in the
North. As to the mode of attacks, 500 of the
affected communities reported at least one

j o u r n a l o f P

E A C E

E S E A RC H

volume 41 / number 4 / july 2004

410

Bars represent cohorts of communities by start of exposure to hostilities in given weeks. The length of the bars is pro-
portional to the median duration for first to last episode. Figures beside bars stand for the number of communities in
each cohort. Cohorts shown include 570 communities; 30 communities are missing duration data, or start dates are in
2002.

Figure 1.

Communities by Date of First Hostilities

01 044474 (ds) 25/5/04 2:29 pm Page 410

instance of airstrikes; only 262 saw ground
operations on their territories.

Victims – The Global Claim
For the period between 11 September 2001
and their respective survey dates, the 600
affected communities claimed a total of 10,770
victims among them. With 20 June 2002 as the
median survey date, the period in question was
roughly nine months. Key informants attrib-
uted 2,997 victims to landmine and UXO inci-
dents, and 7,773 to aerial or indirect fire
bombardment, shooting,

and other forms of

violence. These victims resided in 400 of the

affected communities; 200 communities did
not claim any human victims during this
period. The claims vary a great deal from com-
munity to community, from 0 to a maximum
of 399.

The victim claims can be broken down

along several dimensions. By outcome, 1,582
of the landmine and UXO victims died,
while 1,415 survived their injuries. Among
the victims of bombardments, shooting, and
other violence, 3,994 persons died and 3,779
survived. By age and sex, 1,937 of the
landmine and UXO victims were men, 388
were women, and 672 were children under
the age of 15 years. Among those killed or
injured by bombardments, shooting, or
other violence, 4,586 were men, 1,455
women, and 1,732 children. It is apparent
that the women-to-men ratio is higher
among the latter group. This may be so
because direct violence may have affected

Aldo A. Benini & Lawrence H. Moulton

I C T I M S I N

S Y M M E T R I C A L

O N F L I C T

411

Figure 2.

Communities and Post-9/11 Victims from All War-Related Causes (N = 600)

100

150

200

250

300

1–20

21–40

41–60

61–80

81–100

100+

Victims

Comm

unities

In military parlance, indirect fire refers to fire delivered

by such weapons systems as artillery, mortars, and multiple
rocket launchers. Direct-fire weapons include assault rifles
and other small arms, as well as anti-tank weapons such as
rocket-propelled grenades (RPG-7). However, in this
article we use the term ‘direct violence’ to embrace violence
from direct and indirect fire as well as from other actively
used weapons, as different from violence from victim-
actuated devices.

01 044474 (ds) 25/5/04 2:29 pm Page 411

residential areas (where women spend much
of their time) relatively more than landmines
and UXO did.

An Extreme Case
We illustrate those lifeless statistics with an
extreme case. Hazar Bagh, in the Khwaja
Ghar district of Takhar province, claimed the
highest number of recent victims among any
of the 600 surveyed communities – no fewer
than 399.

This tragedy did not simply coalesce out

of random circumstance. Hazar Bagh, a
town with an estimated current population
of 21,000, has been the object of hostilities
since as far back as the Soviet war. It changed
hands an unknown number of times
between Taliban and Northern Alliance
forces prior to 11 September 2001. The
people there estimated that the community
saw 475 of its members killed or injured in
the roughly one-year period prior to 11 Sep-
tember. A large part of the population was
displaced and returned only in 2002.

The most recent direct conflict occurred in

Hazar Bagh during 26–30 October 2001.
Coalition aircraft attacked the town, and
Taliban forces fired into buildings. The
MCPA interviewers were told that 210 resi-
dential buildings, three mosques, two schools,
and half a dozen shops were damaged to
varying degrees. The landmine problem was
serious – estimates were made of 16 sq. km of
farmland already contaminated before 9/11
plus 1 sq. km of newly contaminated land.
Mines blocked access also to some good water
supplies, and the road to the district center
was damaged and likely mined. The com-
munity lost an estimated 760 animals to the
direct post-9/11 violence, as well as another
300 to landmine and UXO incidents.

Correlation Between Types of Violence
As the account of Hazar Bagh suggests, the
numbers of victims claimed from the two
groups of violent causes are correlated. In

Table II, communities with victims of each
type were placed in three categories, labeled
low, medium, and high. Their ranges are set
so that each group comprises approximately
one-third

of the communities that claimed

some victims. The 66 communities that
reported a high number of victims from
landmines and UXO tended to have high
numbers of victims from bombardments,
shooting, and other kinds of violence as well.
Most of these communities were in the pre-
9/11 friction zone between the Taliban and
various factions, chiefly the Northern
Alliance. No fewer than 30 of them were in
Takhar province (with a median count of
34.5 in direct-violence victims), the scene of
the heaviest fighting before 9/11.

The reverse holds to a much smaller

degree. In fact, there were a considerable
number of communities (107) with medium
and high levels of victims from direct
violence, and none from landmines and
UXO. This reflects the nature of the mobile
war in the later stages of Operation Enduring
Freedom, as compared to the static frontlines
pre-9/11 and in the first three weeks of the
operation.

The main finding, however, seems to be

that a significantly higher proportion (57%)
of communities claimed some victims from
bombardment, shooting, or other violence
than those that reported some landmine and
UXO victims (34%).

j o u r n a l o f P

E A C E

E S E A RC H

volume 41 / number 4 / july 2004

412

This is a convenient categorization for count variables,

particularly for later estimation purposes. In these two and
several more variables used in regression models, ranges
were defined for low, medium, and high so as to have
roughly equal numbers in these non-zero categories
(‘roughly’ because of ties). The downside is that this entails
ranges that are different from variable to variable, which
some may find disturbing. The point here is that the
categories were formed for formal reasons (cross-tabulation
and estimation), not in any substantive relation to Shaw’s
‘small massacres’ concept.

01 044474 (ds) 25/5/04 2:29 pm Page 412

Analysis

Extreme Values
The distribution of victims from the post-
9/11 violence over the 600 affected com-
munities is of key importance to our
investigation. On the lowest end of the
distribution, large proportions of affected
communities had no victims: in Table II, we
see that 256/600 (43%) of the communities
had no direct-violence victims, and 397/600
(66%) had no landmine/UXO victims. On
the high end, looking just among those
communities with at least one victim, the
one-third of the communities with the
highest numbers of direct-violence victims
accounted for 5,832/7,773 = 75% of the
victims, and the corresponding percentage
for the landmine/UXO communities is
2,375/2,997 = 79%. Our goal is to assess
factors that explain the great variability
exhibited by these distributions. We begin by
placing them in a conceptual framework and
then describe the regression model used to
quantitate the relationships between the
factors and the victim counts.

Conceptual Model
Our model explaining the differences in
victim counts looks at four domains of
potentially significant factors:

• Community characteristics
• Recent war experience
• Neighboring communities’ recent war

experience

• Pre-9/11 violence

The variables by which we fathom out

each of those domains are limited to those
for which the contamination assessment col-
lected data, plus the calculation of distances
from the community coordinates to the
known main road network.

We assume that different community

characteristics attracted hostilities and
created vulnerability to different degrees. We
measure the location of communities in the
region with the longest fighting in recent
pre-9/11 history, the magnitude of the popu-
lation (at survey dates, not prewar, in order
to minimize loss of cases), and distance to the
nearest primary or secondary roads.

Unfortunately, we do not have accounts

of individual episodes through which each of
the communities went, and thus we do not
have a direct measure of exposure to hostili-
ties. Our measurements are restricted to the
time period from the first to the last episode
of exposure to any airstrikes or to any ground
operations ever, and the presence or not in
the community of munitions depots. Such
depots reportedly were frequent targets of
airstrikes, with expected collateral damage.

Aldo A. Benini & Lawrence H. Moulton

I C T I M S I N

S Y M M E T R I C A L

O N F L I C T

413

Table II.

Afghan Communities, by Levels of Direct and Indirect Violence Victims, Post-9/11

Direct: Victims from bombardments, shooting, other violence

Low

Medium

High

None

(1–6)

(7–22)

(23–280)

Total

Indirect: Victims

None

200

397

from landmines

Low (1–3)

and UXO

Medium (4–12)

High (13–185)

Total

256

120

116

108

600

Figures in cells are numbers of affected communities with particular combinations of victim levels. Gamma = 0.46,
p < .001.

01 044474 (ds) 25/5/04 2:29 pm Page 413

The combination of no airstrikes and no
ground operations does not occur because
only such communities were surveyed as
having suffered at least one type of attack.

The recent war experience of neighboring

communities was considered as a likely cor-
relate with the intensity of violence to which
the communities in point were subjected.
We reason that mine laying and UXO
littering do not respect community bound-
aries, and defensive positions and advancing
fronts tend to entrap the civilians of neigh-
boring communities simultaneously or in
short succession. We measure neighbors’
recent experience as the sums of post-9/11
direct violence, respectively landmine and
UXO victims in all neighboring communi-
ties within a 3-km radius. For each variable,
we create categorical variables as described in
footnote 4. We use medium and high levels
as two dichotomous variables for each of
them. In a formal perspective, these eco-
logical variables take care of the spatial auto-
correlation of the communities’ victim
counts.

The pre-9/11 violence is relevant for a

number of reasons that our model cannot
separate with the information extant. Local
regions with persistent pre-9/11 fighting
probably saw high levels of mine laying and
UXO littering, in addition to civilian
victims from direct collateral damage. A
number of communities may have suffered
significant oppression at the hands of
armed forces from other parts of the
country, producing violent behaviors after
9/11. Anecdotal accounts of setting up
defensive positions inside residential areas
of selected villages and towns by retreating
troops and score settling by advancing ones
fall in this category. We measure pre-9/11
violence levels by their outcomes during
the preceding approximate 12-month
period. As with the ecological variables
described above, we create dichotomous
variables for the medium and high levels of

direct violence, and for landmine and UXO
victims.

Descriptive statistics for the 15 model

covariates from four domains are given in
Table III.

Regression Models
We present the results of a zero-inflated
Poisson regression (ZIP) (Lambert, 1992) of
the number of post-9/11 victims from direct
violence on the above covariates. The choice
of this model was motivated by two con-
siderations. First, because the response
variable was a count with many small values,
some kind of Poisson regression model was
required. Second, there were an observed dis-
proportionately large number of zero counts.
This indicated the presence of two different
distributions mixed together: a group of
communities that were inherently at very low
risk in spite of being affected by hostilities,
and a group of communities that were at
higher risk. The ZIP model allows us to dis-
entangle two separate effects: (1) what
factors are related to being in the near-zero
risk distribution, as opposed to the higher-
risk group?; and (2) for the communities in
the higher-risk group, what factors explain
the variability of the rates among them? The
model is called ‘zero-inflated’ because it
allows for the large number of zero-count
communities. Two sets of coefficients are
estimated, one for a logistic model com-
ponent that distinguishes the two distri-
butions and one for a standard Poisson
model for those in the higher-risk group.

Our conceptual framework did not extend

to differentiating between which covariates
would be most likely to affect one or other
component of the ZIP model. Therefore, we
fit a full model in which all covariates were
entered in both parts of the model. In this

j o u r n a l o f P

E A C E

E S E A RC H

volume 41 / number 4 / july 2004

414

In addition to this model, a zero-inflated negative

binomial regression was run to test for even greater vari-
ability than assumed. The test was not statistically signifi-
cant (p = 0.15), and the ZIP model was retained.

01 044474 (ds) 25/5/04 2:29 pm Page 414

way, we could examine the relative contri-
bution of a variable to both parts, adjusted for
the presence of the other covariates.

We have estimated models for both types

of victims, those of direct violence and those
of landmines/UXO; for space reasons, we
will present only the model for direct-
violence victims.

We will interpret these results substan-

tively in the next section, but here give some

technical explanation. The coefficients in the
two parts of the ZIP model have opposite
meanings. A positive coefficient in the
Poisson part signifies that an increase in the
covariate increases the expected number of
victims. For a positive coefficient in the infla-
tion part, however, an increase in the covari-
ate indicates a lower probability that the
community had some victims (equivalently,
a higher probability of zero victims). Some
covariates are significant in both parts. For
example, in the northern region, not only
were there relatively fewer communities
without recent victims, but also the number
among those who did have some tended to
be higher than for the other regions. Note
the effects of ground operations: this variable
produces positive coefficients in both model
parts. The apparent paradox between two
tendencies – such attacks produced more
victims; they also improved the chances for
communities to come away without any
victims – will be commented on below. A

Aldo A. Benini & Lawrence H. Moulton

I C T I M S I N

S Y M M E T R I C A L

O N F L I C T

415

Table III.

Descriptive Statistics

Minimum Maximum

Mean

Std. dev.

Community characteristics
Is in northern region

600

0.28

Population (log

)

584

0.00

4.85

3.11

0.66

Distance to main road (meters, log

)

599

0.14

5.02

3.39

0.89

Recent war experience
Days start-to-end local hostilities

582

11.01

11.48

Local airstrike

595

0.84

Local ground operations

595

0.44

Had munitions depot

600

0.07

Neighboring communities
Direct violence 3 km radius post-9/11 medium

599

0.18

Direct violence 3 km radius post-9/11 high

599

0.18

Mines UXO 3 km radius post-9/11 medium

599

0.12

Mines UXO 3 km radius post-9/11 high

599

0.13

Pre-9/11 violence
Direct violence community’s own pre-9/11 medium

600

0.12

Direct violence community’s own pre-9/11 high

600

0.12

Mines UXO community’s own pre-9/11 medium

600

0.10

Mines UXO community’s own pre-9/11 high

600

0.11

Valid N (listwise)

561

In addition to this full model, we fit several other models.

In one set of models, we fit one covariate at a time, in both
components of the ZIP model, to look at the univariate
relationship between the covariate and the victim count. In
general, the coefficients in these models were somewhat
larger in magnitude than in the full model, as is usually the
case, yet there were no large differences in the patterns. We
also conducted an influence analysis, excluding the 5% of
the observations of the full model with the largest residu-
als on both tails. Finally, we produced a reduced model, in
which all variables of the full model that did not have at
least one statistically significant (p < 0.1) coefficient in
either the Poisson or in the inflation parts were eliminated.
For these latter two models, there were only minimal
differences from the full model, and so we report only the
latter.

01 044474 (ds) 25/5/04 2:29 pm Page 415

similar paradox is apparent among the four
coefficients of the levels of landmine and
UXO victims in neighboring communities.

Despite its many technicalities, the

significance of regression modeling must be

seen in the larger picture. Our models permit
us to discern simultaneously the influence of
historic and contemporary factors on the
distribution of victims. Since the data are
geo-referenced, neighborhood effects can be

j o u r n a l o f P

E A C E

E S E A RC H

volume 41 / number 4 / july 2004

416

Table IV.

Zero-inflated Poisson Regression, Victims from Direct Violence Post-9/11

Coefficient

P>|z|

Poisson part
Is in northern region

0.28

.051

Population (log

)

0.26

.079

Distance to main road (meters, log

)

0.07

.330

Days start-to-end local hostilities

–0.01

.115

Local airstrike

0.19

.184

Local ground operations

0.34

.048

Had munitions depot

0.26

.246

Direct violence 3 km radius post-9/11 medium

0.03

.875

Direct violence 3 km radius post-9/11 high

–0.07

.698

Mines UXO 3 km radius post-9/11 medium

–0.25

.116

Mines UXO 3 km radius post-9/11 high

–0.48

.044

Direct violence community’s own pre-9/11 medium

0.06

.769

Direct violence community’s own pre-9/11 high

0.91

<.001

Mines UXO community’s own pre-9/11 medium

0.43

.013

Mines UXO community’s own pre-9/11 high

0.56

<.001

Constant

1.33

.049

Inflation part
Is in northern region

–0.66

.017

Population (log

)

–0.25

.121

Distance to main road (meters, log

)

–0.23

.046

Days start-to-end local hostilities

–0.01

.545

Local airstrike

0.48

.141

Local ground operations

0.88

<.001

Had munitions depot

–0.16

.672

Direct violence 3 km radius post-9/11 medium

–0.56

.043

Direct violence 3 km radius post-9/11 high

–1.03

.002

Mines UXO 3 km radius post-9/11 medium

–0.89

.015

Mines UXO 3 km radius post-9/11 high

–0.56

.203

Direct violence community’s own pre-9/11 medium

–0.37

.267

Direct violence community’s own pre-9/11 high

–1.89

<.001

Mines UXO community’s own pre-9/11 medium

0.02

.956

Mines UXO community’s own pre-9/11 high

0.13

.759

Constant

1.30

.090

Correlation observed vs. predicted; fit
Spearman’s rho

.43

McFadden’s pseudo-R

adj.

.37

561

01 044474 (ds) 25/5/04 2:29 pm Page 416

controlled for as well. This provides a basic
(by econometric standards), but neverthe-
less attractive, spatio-temporally integrated
framework for estimation purposes.

Discussion

We restrict discussion of statistical findings
to those relevant for our three hypotheses.

Victim Estimates
Several researchers have publicized estimates
of civilian victims in this war. All of those
mentioned here restricted their focus to
victims of the US intervention. The scope of
these estimates varied enormously. At its
most narrow, the Human Rights Watch,
concerned with the use of cluster bombs,
tallied a minimum of 25 deaths from direct
impacts plus another 127 from later dud
explosions (HRW, 2002: 1, 25). HRW col-
lected this data through visits to 250
bombing sites in March and April 2002.
Herold, in a piece amplified by an article in
The Guardian (Herold, 2002), estimated
that just under 3,000 Afghan civilians had
been killed in US air attacks between
October and December 2001. Including
deaths from Special Forces attacks, he
increased his claim to 3,620 civilians killed
up to 31 July 2002, a period just slightly
longer than our survey period. Herold used
media reports and Internet searches. The
most sweeping claim has been advanced by
the Project on Defense Alternatives
(Conetta, 2002), adding to an estimated
1,000–1,300 civilian deaths from airstrikes
8,000–18,000 deaths from indirect war
effects. This latter estimate hinges on mor-
tality figures collected in IDP camps.
Conetta attributed 40% of it to the war.

Shaw (2002: 345) points out the

methodological difficulties of Conetta’s
indirect-effects claim, while Lemieux (2002)
takes Herold to task for the inclusion of
material from the Afghan Islamic Press, a

pro-Taliban agency. In the end, Shaw settles
for 1,000–1,300 ‘civilians killed by the West’
(2002: 347). Regardless of those problems,
we take issue with the idea that violent
deaths of civilians can be neatly attributed to
one or the other side of the conflict. Civilians
caught in crossfire, moving in contaminated
areas, or held hostage by troops hiding in
residential areas are harmed by the un-
distinguishable cause-and-effect mix from all
sides. In addition, counting only the dead,
and not also the injured, is tempting for the
benefit of easy comparison with death statis-
tics from other conflicts, but overlooks the
loss of life years among the injured who die
soon after the compilation period, and the
human suffering caused by permanent dis-
abilities.

Our figures, therefore, paint a different

picture from those taken with an exclusive
view to one party of the conflict. The 600
surveyed communities claimed 5,576 resi-
dents killed violently between 11 September
2001 and June 2002; another 5,194 were
injured. The combined figure is nearly a
magnitude higher than the signature range
that Shaw’s comparative table displays for the
Afghanistan war.

Exposed Communities
As noted, almost half of the communities
exposed to hostilities suffered no victims
from direct violence, and two-thirds claimed
no landmine/UXO victims. Given the means
over all exposed communities, these zero-
victim communities are unexpectedly
numerous (this is indicated, for example, by
the many statistically significant variables in
the inflation part of the ZIP model, Table
IV). They contrast with the 13 communities
that each reported more than 100 victims of
direct violence and the three that reported
more than 100 landmine/UXO victims.

The regression model for direct violence

identifies some of the factors of this polarized
distribution, and some of them can be

Aldo A. Benini & Lawrence H. Moulton

I C T I M S I N

S Y M M E T R I C A L

O N F L I C T

417

01 044474 (ds) 25/5/04 2:29 pm Page 417

speculatively connected with expected
outcomes of a highly asymmetrical conflict.
The speed of the operation’s advance and the
use of airpower to breach resistant front lines
are significant. Many of the 168 communi-
ties in the north were close to front lines
from before 9/11, which took relatively long
to soften up in the initial weeks of the war.
Their median number of victims from direct
violence was 12, as opposed to only 1 for the
communities in all the other regions through
which Operation Enduring Freedom parleyed
its advance with increasing speed. The coeffi-
cients for the northern region confirm this
logic even after controlling for all other regres-
sion covariates.

Refining the regional effects, local clusters

of violence at smaller spatial scales operate as
well. Communities in close neighborhood to
others with medium or high levels of victims
from direct violence were about half to a
third as likely to avoid victims in their own
ranks.

Another characteristic of asymmetrical

warfare seems to be that once static front
lines crumble, the militarily weaker party
quickly retreats from many of its defensive
positions and is subsequently met with
strong fire in those places where its forces try
to regroup and to offer coherent resistance.
Around the former, most of the exposed
communities should get away without many
victims from direct violence. Near the latter,
they should suffer high numbers. This is
borne out by the coefficients of the ground
operations variable. The fact that a com-
munity witnessed such operations on its
territory at first greatly improves its chances
to avoid victims – the Taliban ran away
without much of a fight. However, where
they did fight, communities would suffer
significantly. This can be made more graphic
with figures from the region where the war

progressed more rapidly. Of the 432 com-
munities outside the northern region, 218
reported victims from direct violence.
Among these, 58 communities exposed to
ground operations reported a median of 19.5
victims; the 160 exposed ‘only’ to airstrikes
scored a median of 6.

Background Noise of a Violent Universe
If there is any true surprise in the victim data,
it is the discovery of how strongly levels of
post-9/11 victims were determined by the
levels that existed before. These highly
significant effects persist when the structural
factors (region, population, etc.) and the
concurrent violence levels are taken into
account. In our model of direct violence, all
four pertinent variables in the Poisson part
are strongly significant. In addition, high
levels of direct violence in the year prior to
11 September 2001 demolished chances to
escape without victims after 9/11 – this
coefficient in the inflation part is the
strongest for all dichotomous variables.
Similar, although slightly weaker, effects of
this kind hold for the landmine and UXO
model that we developed outside this article.

All this points to the continuing effects of

the earlier violence into the period of Oper-
ation Enduring Freedom. While so much
was, of course, known for the large-scale situ-
ation of the Afghan nation, the background
noise from old conflicts is clearly measurable
at the fine grain of the individual communi-
ties. We can be more precise: if we eliminate
from the regression model the variables char-
acterizing the old conflicts – pre-9/11 victim
levels and whether a community was part of
the northern region or not – the measure for
regression fit drops by half.

In other words,

legacy and concurrent effects seem to be on
par in terms of explaining variability in
victim counts.

j o u r n a l o f P

E A C E

E S E A RC H

volume 41 / number 4 / july 2004

418

The odds of being in the near-zero risk group are calcu-

lated by base e exponentiation of the inflation part coeffi-
cients.

McFadden’s adj. pseudo-R

: from .37 to .19.

01 044474 (ds) 25/5/04 2:29 pm Page 418

Victim Counts and the Difficulty of
Baselines
Returning from the discussion of our three
specific hypotheses to Shaw’s first broad
claim – compared to earlier periods,
Western interventions reduced civilian
losses – the Afghanistan data support it.
This is true of short-run as well as long-run
contrasts. A caveat should nevertheless be
made regarding long-run baselines for such
comparisons.

During the 12 months preceding 9/11, an

estimated 12,421 residents of the 600 com-
munities became victims to violence – 8,935
to direct violence, 3,486 to landmine and

UXO strikes. Although this figure is lower
than an annualized post-9/11 figure from the
nine-month period of our survey (10,770

12 months/9 months = 14,360), by all
accounts the incidence of violence had
dropped very sharply by January 2002 and
had not as of May 2003 climbed back to
anywhere near war levels.

A comparison with Afghanistan’s long

past of war leads to a similar finding. Before
11 September 2001, Afghanistan had gone
through 22 years of internal conflict. These
wars together killed, if we take a middling
estimate (see page 406), 2 million people.
Annualized, the 5,576 deaths during our

Aldo A. Benini & Lawrence H. Moulton

I C T I M S I N

S Y M M E T R I C A L

O N F L I C T

419

–

Figure 3.

Districts with Victims from Direct Violence During Operation Enduring Freedom

The regional concentration of victims is illustrated by district-wise counts of post-9/11 victims from
direct violence. The two districts with over 800 such victims each were on the front lines between the
Taliban and the Northern Alliance both before and after 11 September 2001.

01 044474 (ds) 25/5/04 2:29 pm Page 419

post-9/11 survey period approximate one-
twelfth of the historic violence level.

A problem with such interpretations is

that they are volatile to the ideological
climates of the day. Until 1997, US and
Pakistani propaganda depicted the Taliban as
a source of stability for the region, particu-
larly in the context of a Central Asia-to-
Pakistan gas pipeline project led by a US
energy corporation (Shadid & Donnelly,
2001; Cottey, 2003: 170). Their law-and-
order regime was billed as deliverance from
the anarchy of local warlords. This implied
that violence against civilians had abated
from earlier periods. If so, then historic yard-
sticks of the long run, such as an average
calculated on 22 years, are not helpful as
baselines. We prefer, as both more authentic
and more reliable, the victim numbers
that our key informants recalled for the
12-month period preceding 9/11.

Conclusion

Operation Enduring Freedom, as we know
by now, was not the last intervention by the
West that pitted military forces of very
unequal strength against each other.
Although Iraq’s armed forces were in a
different league from the Taliban ragtag
troops, the spring 2003 Iraq war, too, was an
instance of highly asymmetrical conflict. It
may be too early for a numerical assessment,
and if the Pentagon has its way, none may
ever happen (Graham & Morgan, 2003;
however, see Dardagan, 2003). What tran-
spired from the media coverage of the war
nevertheless indicates a pattern of engage-
ments that likely produced a polarized local
distribution of victims similar to that in
Afghanistan. Western forces applied concen-
trated airstrikes to regime infrastructure and
enemy troop concentrations. They strove to
limit ground engagements to strategic
points, simply sidestepping the majority of
poorly defended localities. Judging from the

overflow of major hospitals, a small number
of urban neighborhoods suffered numerous
victims while a host of other communities
may have remained outside the theater and
without human loss. All this is speculative
and subject to vast observational and report-
ing bias.

In the end, the dense figurework from

Afghanistan and the impressionistic picture
of the Iraq war both urge the same
question: Does the new Western way of war
go hand in hand with considerable under-
reporting of civilian losses? There are
several factors that make this a systemic
likelihood: media management, the
omission of injuries and of landmine and
UXO strikes from victim counts, and the
efforts of researchers to attribute victims to
the actions of one or the other party to the
conflict. One of our reviewers stressed,
against the media bias charge, that quanti-
fying victims had become vastly more
accurate since the Armenian genocide. But
this is a ‘longue durée’-argument; it does
not refute the biases operating on publi-
cized counts of victims during recent wars.
Counting is never value-free: on one
extreme, ‘redefining war on our terms’, as
President Bush put it, will encourage a
focus on what went wrong in very narrow
terms of collateral damage. On the other
extreme, exemplified by Conetta’s research,
compilations of violent incidents are sup-
plemented by estimates of victims from
indirect effects that depend on arbitrary
attributions. We have chosen a third path,
that of surveying the communities that
suffered the loss.

Whether community surveys offer a

genuine methodological improvement, and
their findings will be heard in assessments of
the cost of war, only further research can
prove. This study of civilian victims in
Afghanistan emerged as a sideline to a
landmine contamination assessment; com-
munity surveys directly focused on victim

j o u r n a l o f P

E A C E

E S E A RC H

volume 41 / number 4 / july 2004

420

01 044474 (ds) 25/5/04 2:29 pm Page 420

questions will need to adapt their instru-
ments, such as by incorporating sampling for
false negatives and a record of individual
incidents that took place during a com-
munity’s period of exposure. Victimization
studies concerned also with indirect victims
may yet want to evolve other designs, such as
repeated measurements post-intervention or
household-level surveys, in order to
overcome the limitations of ecological
(macro-level) prevalence studies.

Despite these limitations, our data were

good enough to confirm two enduring facts
of war: the civilian population suffers more
than a few small massacres, and forces that
predate short-term military intervention
significantly shape the pattern of violence.

References

Benini, Aldo & Joseph Donahue, 2003. ‘Land-

mines, War and Victim Dynamics: Contami-
nation Assessment in Afghanistan’.
Washington, DC: Vietnam Veterans of
America Foundation (http://vvaf.policy.net/
humanitarian/immap_report_2/; accessed 7
November 2003).

Collier, Paul; Lani Elliott, Håvard Hegre, Anke

Hoeffler, Marta Reynal-Querol & Nicholas
Sambanis, 2003. Breaking the Conflict Trap:
Civil War and Development Policy. Washing-
ton, DC: World Bank; Oxford: Oxford
University Press (

http://econ.worldbank.org/

prr/CivilWarPRR/text-26671/

; accessed 29

August 2003).

Conetta, Carl, 2002. ‘Strange Victory: A

Critical Appraisal of Operation Enduring
Freedom and the Afghanistan War’. Project
on Defense Alternatives Research Monograph
6, 30 January. Cambridge, MA: Common-
wealth Institute (

http://www.comw.org/

pda/0201strangevic. pdf

; accessed December

2002).

Cottey, Andrew, 2003. ‘Afghanistan and the New

Dynamics of Intervention: Counter-
Terrorism and Nation Building’, in SIPRI
Yearbook 2003: Armaments, Disarmament and

International Security. Oxford: Oxford Uni-
versity Press (167–194).

Dardagan, Hamit, 2003. ‘The Iraq Body

Count Database’ (

http://www.iraqbody-

count.net/bodycount.htm

; accessed 26 Sep-

tember 2003).

Desvignes, Laurence-Marie, 2003. ‘Comments of

Report – Afghanistan’, e-mail message from
ICRC Kabul to Aldo Benini, 17 January.
Kabul: ICRC.

Falconer, Bruce, 2003. ‘The World in Numbers:

Murder by the State’, Atlantic Monthly 292
(November): 56–57.

Ghobarah, Hazem Adam; Paul Huth & Bruce

Russett, 2003. ‘Civil Wars Kill and Maim
People Long After the Shooting Stops’,
American Political Science Review 97(2):
189–202.

Goodson, Larry P., 2001. Afghanistan’s Endless

War: State Failure, Regional Politics and the Rise
of the Taliban. Seattle, WA: University of
Washington Press.

Graham, Bradley & Dan Morgan, 2003. ‘U.S.

Has No Plan to Count Civilian Casualties:
Congress Calls for “Assistance” to Iraqis for
War Losses’, Washington Post, 15 April: A13,
A16.

Herold, Marc W., 2002. ‘Counting the Dead:

Attempts To Hide the Number of Afghan
Civilians Killed by US bombs Are an Affront
to Justice’, The Guardian, 8 August
(http://www.guardian.co.uk/afghanistan/com
ment/stor y/0,11447,770999,00.html;
accessed 5 January 2003).

HRW, 2002. ‘Fatally Flawed: Cluster Bombs and

Their Use by the United States in Afghanis-
tan’, Report 14(7) (G), December. Washing-
ton, DC: Human Rights Watch.

ICBL, ed., 2002. Landmine Monitor Report 2002.

Washington, DC: International Campaign to
Ban Landmines.

Kaldor, Mary, 1999. New and Old Wars: Organ-

ised Violence in a Global Era. Stanford, CA:
Stanford University Press. [Reprinted with an
afterword, 2001.]

Lambert, Diane, 1992. ‘Zero-Inflated Poisson

Regression, with an Application to Defects in
Manufacturing’, Technometrics 34(1): 1–14.

Lemieux, Marc A., 2002. ‘From Baghdad to

Aldo A. Benini & Lawrence H. Moulton

I C T I M S I N

S Y M M E T R I C A L

O N F L I C T

421

01 044474 (ds) 25/5/04 2:29 pm Page 421

Kabul: The Impact of International Humani-
tarian Law on Coalition Airpower and Their
Consequences for Humanitarian Action’,
Journal of Humanitarian Assistance (July)
(http://www.jha.ac/articles/a110.htm;
accessed March 2003).

Mueller, John, 1995. ‘The Perfect Enemy: Assess-

ing the Gulf War’, Security Studies 5(1):
77–117.

Mueller, John, 2003. ‘Policing the Remnants of

War’, Journal of Peace Research 40(5):
507–518.

Rummel, Rudolph J., 1995. ‘Democracy, Power,

Genocide, and Mass Murder’, Journal of
Conflict Resolution 39(1): 3–26.

Rummel, Rudolph J., 1997. Statistics of

Democide: Genocide and Mass Murder Since
1900. Charlottesville, VA: Center for National
Security Law, School of Law, University of
Virginia; Somerset, NJ: Transaction.

Sarkees, Meredith Reid, 2000. ‘The Correlates of

War Data on War: An Update to 1997’,
Conflict Management and Peace Science 18(1):
123–144.

Shadid, Anthony & John Donnelly, 2001. ‘US

Tried To Woo Taliban In ’90s’, Boston Globe,
20 September (http://www.cooperativeresearch.
org/timeline/2001/bostonglobe092001.html;
accessed 30 October 2003).

Shaw, Martin, 2002. ‘Risk-Transfer Militarism,

Small Massacres and the Historic Legitimacy
of War’, International Relations 16(3):
343–359.

Tourangeau, Roger; Lance J. Rips & Kenneth

Rasinski, 2000. The Psychology of Survey
Response. Cambridge: Cambridge University
Press.

US Bureau of the Census, 2003. ‘International

Data Base: Summary Demographic Data for
Afghanistan’ (

http://blue.census.gov/cgi-bin/

ipc/idbsum?cty = AF

; accessed 9 January).

ALDO A. BENINI, b. 1950, PhD in Soci-
ology (University of Bielefeld, 1979); dual
career in rural development and humanitarian
action; Lutheran World Federation (1983–86),
International Committee of the Red
Cross (1987–94), Global Landmine Survey
(1999– ). Regions with significant work
experience: Bangladesh, Afghanistan, Africa.

LAWRENCE H. MOULTON, b. 1956, PhD
in Biostatistics (Johns Hopkins University,
1987); Assistant Professor, University of
Michigan (1986–91); Professor, Departments
of International Health and Biostatistics,
Bloomberg School of Public Health, Johns
Hopkins University (1991– ).

j o u r n a l o f P

E A C E

E S E A RC H

volume 41 / number 4 / july 2004

422

01 044474 (ds) 25/5/04 2:29 pm Page 422