DRONE STRIKES IN PAKISTAN:
Impact of Militant Retaliation against Alleged Spies
Michael Reese, University of ChicagoKeven Ruby, CPOSTVincent Bauer, CPOST
Program on International Security Policy (PISP)
February 24, 2015
Understanding Drone Warfare
§ Conventional approach to drones is to treat under rubric of targeted killings–Counterinsurgent à Insurgents–Organizational survivability (e.g., Jordan)–Militant effectiveness/violence (e.g., Johnston and others)–Propaganda effectiveness
§ What’s missing in the literature–Insurgent à Counterinsurgent?–Are there insurgent strategies for countering the US drone campaign?
§ Collaborator killings in tribal areas of Pakistan unique opportunity–Previous research found multiple reports of killings of individuals
specifically identified as collaborating with the United States
1
Taliban campaign against “US Spies” in the tribal areas of Pakistan
§ Wide-spread belief that spying central to drone program: "chips" and "magic pens" used to by spies to guide drones to target
§ Specialized counterintelligence unit, Lashkar-e-Khorasan, created by main militant groups in Waziristan in 2009
§ Notes on victims identify them as US spies and ward others that “all spies will share the same fate”
§ Confessions of spying videotaped and distributed on DVD
§ The Pakistani government reports losing 70 spies in the tribal areas between 2004 and 2011
2
“Taliban Use Swords to Slit the Throats of Afghan ‘Traitors’ in Public Executions before Thousands.” Daily Mail, June 27, 2008.
Al Qaeda warns of the danger of “Muslim Spies”
“We would not be exaggerating if we said that the frontline of the Crusader’s campaign…is the spying networks in all its types, shapes, and forms. …
The spying networks are their eyes to see the hidden things that they cannot see and are their hands that are still extending inside the houses, in the forests, up the mountains, into the valleys, and inside the dark caves in order to catch a target that their developed technology was not able to reach.”
Abu-Yahya al-LibiAl QaedaGuidance on the Ruling of the Muslim SpyJune, 2009
3
Research questions and Hypotheses
§ Questions– Is there a strategic logic to collaborator killings? Is there evidence that
these collaborator killings are linked to deterring drone strikes or are allegations of spying driven by local score settling instead?
– If so, does it work? Does killing alleged spies deter or reduce the effectiveness of drone strikes that rely on many sources of intelligence not limited to spies?
§ Hypotheses–H1: If collaborator killings are a response by militants to drone strikes,
then collaborator killings will be correlated in space and time with drone strikes.
–H2: If collaborators provide intelligence necessary for drone targeting, then killing collaborators will negatively effect the quality of drone strikes.
4
H1: Causal logic by which drone strikes drive insurgents to kill collaborators§ Militants believe collaborators are central to the effectiveness of
drone strikes.–Believe eliminating/deterring collaborators will improve security.–Anecdotal evidence suggests this is the case.
§ Alternative explanations for collaborator killings–Cover for eliminating political rivals (e.g., Cole 2014)– Intimidating/undermining local adversaries (e.g., Kydd & Walter 2006)–False denunciations for settling scores and private gain (e.g., Kalyvas)
§ How can we know?–If collaborator killings systematically follow drone strikes in space and
time, increased confidence that killings are strategic.
5
H2: Causal logic by which killing of alleged US spies degrades drone strike quality
HUMINT Needed
Actual Selectivityvs Spies
Perceived Selectivity
Elimination Mechanism
DeterrenceMechanism
Predict Degradation
Yes Selective Selective ü ü Yes
Yes Selective Indiscriminate ü û Yes
Yes Indiscriminate Selective û ü Yes
Yes Indiscriminate Indiscriminate û û No
No NA NA NA NA No
6
1. If effect, unclear whether because of elimination or deterrence.
2. If no effect, unclear whether spies unnecessary or militants ineffective at counterespionage.
1
2
Building the Collaborator Killing Database
§ How was the data collected?–Events were collected by RAs at CPOST from Newswires (LexisNexis)
and Dawn (Google) using keyword searches.
§ What were the inclusion criteria?–Civilian victimization (killed, shot, bodies…)–Indications of spying (spy, spying, spied…) –Suspected responsibility by militants (militants, fighters, Taliban…)–Presumed connection with the United States (United States, America…)–Story reported in Pakistan.
§ What makes spying so visible?–Notes are left with the vast majority of victims, marking them as
collaborators and accusing them of working with the United States.
7
What do these news sources look like?
AP: “Militants shot dead a man execution-style suspected of spying for Pakistani and U.S. authorities and providing intelligence used in this week's air raid on an Islamic school that killed 80 people, witnesses and an official said Thursday.…The bullet-riddled, mutilated body of tribesman Mohammed Jan Khan was found in a dry stream bed in a mountainous area near Chingai village. There was no immediate claim of responsibility.Tribespeople told the AP that after the school attack, Khan ran from the scene when armed militants spotted him talking on a satellite phone."People were removing debris when they saw him speaking on the phone. He got suspicious and ran away," said Wazir Khan, a Chingai resident and restaurant cook aged in his 30s. "Two or three masked gunmen chased him, shot him in the back and took his body to the river bed and left a message next to it under a stone."Three tribespeople said the note, written in the locally spoken Pashto language, accused Khan of spying for American and Pakistan troops in the area for a long time and being under observation…”
8
Habibullah Khan, Associated Press International, “Militants execute suspected spy after deadly Pakistan army attack on school”, November 2, 2006
9
A. Total killed B. Mode of killing C. Occupation
D. Nationality
0
10
20
30
40
50
60
70
2005 2006 2007 2008 2009 2010 2011
Elder Other Unknown
Afghan Pakistani Unknown
The data on killing of alleged US spies in Pakistan, 2005-2011 (N=299)
<1%
2%
3%
5%
15%
22%
53%
0 50 100 150
Burning
Bombing
Hanging
Stabbing
Unspecified
Beheading
Shooting
The BIJ data on drone strikes in Pakistan
10
Total Strikes
Source: Bureau of Investigative Journalism
Fatalities Fatalities per Strike
0
20
40
60
80
100
120
140
0
10
20
30
40
50
60Civilians
Militants
0
200
400
600
800
1000
1200Civilians
Militants
Drone strikes and collaborator killings follow similar trajectory over time
11
Sources: Drones (BIJ); Collaborator Killings (CPOST)
0
10
20
30
40
50
60
70
0
20
40
60
80
100
120
140
2004 2005 2006 2007 2008 2009 2010 2011
Col
labo
rato
rs K
illed
Dro
ne S
trik
es
12100km
Afghanistan
Pakistan
India
Afghanistan
Miran Shah
Mir Ali
13100km
Afghanistan
Pakistan
India
Afghanistan
14100km
Afghanistan
Pakistan
India
Afghanistan
15100km
Afghanistan
Pakistan
India
25km x 25km
Afghanistan
Controls
§ Population density: – Controls for the possibility that more
people=more collaborators to kill§ Rough terrain:
– Controls for the possibility that collaborators will be more difficult to identify in rough terrain
§ Pakistani military operations– Controls for the possibility that collaborator
killings are driven by militant response to Pakistani military operations
§ Government territorial control– Controls for the possibility that sources of intel
on and access to collaborators lower in areas of government control
§ Previous collaborator killing– Controls for the possibility that killing
collaborators is effective in deterring collaboration
16
H1: DronesàCollaborator Killings
§ Population density: – Controls for the possibility that more people=more
militants to strike§ Rough terrain:
– Controls for the possibility that drone strikes will be less effective in killing militants in rough terrain
§ Pakistani military operations– Controls for the possibility Pakistani military
intelligence substitutes for local collaborators during operations
§ Government territorial control– Controls for the possibility that sources of intel on
militant targets greater in areas of greater government control
§ Previous drone strike – Controls for the possibility that defensive
measures other than killings affect drone strike quality
H2: Collaborator KillingsàDrones
Probability of collaborator killings increases after drone strikes
17
A) 30 Days B) 90 Days
0
0.0001
0.0002
0.0003
0.0004
0.0005
0.0006
0.0007
0.0008
0.0009
Min Median Max
Prob
abili
ty o
f Col
labo
rato
r Kill
ing
Drone Strikes (Prior 30 days)
0
0.0002
0.0004
0.0006
0.0008
0.001
0.0012
0.0014
0.0016
Min Median Max
Prob
abili
ty o
f Col
labo
rato
r Kill
ing
Drone Strikes (Prior 90 Days)
Logit, DV= Collaborator Killing; IV=Drone Strikes; Controls=Territorial Control, PakMil Operations, Pop Density, Rough Terrain
Probability of collaborator killings increases after drone strikes
18
C) 365 Days D) All Days
Logit, DV= Collaborator Killing; IV=Drone Strikes; Controls=Territorial Control, PakMil Operations, Pop Density, Rough Terrain
0
0.00002
0.00004
0.00006
0.00008
0.0001
0.00012
Min Median Max
Prob
abili
ty o
f Col
labo
rato
r Kill
ing
Drone Strikes (Prior 365 Days)
0
0.000005
0.00001
0.000015
0.00002
0.000025
0.00003
0.000035
0.00004
Min Median Max
Prob
abili
ty o
f Col
labo
rato
r Kill
ing
Drone Strikes (All Prior Days)
Drone strikes kill fewer militants in areas after collaborator killings
19
A) 30 Days B) 90 Days
0
1
2
3
4
5
6
7
8
9
10
Min Median Max
Mili
tant
s K
illed
by
Dro
ne S
trik
es
Collaborators Killed, Within 25km of Strike(Prior 30 Days)
0
1
2
3
4
5
6
7
8
9
10
Min Median Max
Mili
tant
s K
illed
by
Dro
ne S
trik
es
Collaborators Killed, Within 25km of Strike(Prior 90 Days)
nbreg, DV= Militants killed by drone strike; IV=Collaborators killed within 25km and n days prior;Controls=Territorial Control, PakMil Operations, Pop Density, Rough Terrain
Drone strikes kill fewer militants in areas after collaborator killings
20
C) 365 Days D) All Days
nbreg, DV= Militants killed by drone strike; IV=Collaborators killed within 25km and n days prior;Controls=Territorial Control, PakMil Operations, Pop Density, Rough Terrain
0
1
2
3
4
5
6
7
8
9
10
Min Median Max
Mili
tant
s K
illed
by
Dro
ne S
trik
es
Collaborators Killed, Within 25km of Strike(Prior 365 Days)
0
1
2
3
4
5
6
7
8
9
10
Min Median Max
Mili
tant
s K
illed
by
Dro
ne S
trik
es
Collaborators Killed, Within 25km of Strike(All Prior Days)
Implications
§ Effectiveness of drone strikes not solely measured over violence
§ Drone strikes will decline in effectiveness over time as militants’ countermeasures take effect
§ In the long-run, effectiveness of countermeasures depends on perception that militant targeting of collaborators is selective.–If local public perceives targeting as selective, it will undermine active
support for the state.– If indiscriminate, it will undermine toleration of militant groups and even
encourage collaboration with the state against the militant groups.
21
Questions?
22
Additional Slides
23
Mechanisms by which killing of alleged US spies degrades drone strike quality
§ For there to be any effect of collaborator killings on drone strike quality, drone strike quality must rely on spies.
There are two mechanisms:
§ Mechanism A: Elimination–By killing actual spies required by drone program, militant groups are
directly degrading drone capability.
§ Mechanism B: Deterrence –By being perceived as killing spies, militants inhibit current and potential
spies, indirectly degrading drone capability.
24
25
“Taliban Use Swords to Slit the Throats of Afghan ‘Traitors’ in Public Executions before Thousands.” Daily Mail, June 27, 2008.
Probability of collaborator killings increases after drone strikes
26
A) 30 Days B) 90 Days C) 365 Days D) All Days
0
0.0001
0.0002
0.0003
0.0004
0.0005
0.0006
0.0007
0.0008
0.0009
Prob
abili
ty o
f Col
labo
rato
r Kill
ing
Drone Strikes (Prior 30 days)
0
0.0002
0.0004
0.0006
0.0008
0.001
0.0012
0.0014
0.0016Pr
obab
ility
of C
olla
bora
tor K
illin
g
Drone Strikes (Prior 90 Days)
0
0.00002
0.00004
0.00006
0.00008
0.0001
0.00012
Prob
abili
ty o
f Col
labo
rato
r Kill
ing
Drone Strikes (Prior 365 Days)
0
0.000005
0.00001
0.000015
0.00002
0.000025
0.00003
0.000035
0.00004
Prob
abili
ty o
f Col
labo
rato
r Kill
ing
Drone Strikes (All Prior Days)
65 CI 95 CI Mean Predicted Probability
Logit, DV= Collaborator Killing; IV=Drone Strikes; Controls=Territorial Control, PakMil Operations, Pop Density, Rough Terrain
Drone strikes kill fewer militants in areas after collaborator killings
27
A) 30 Days B) 90 Days C) 365 Days D) All Days
0
1
2
3
4
5
6
7
8
9
10
Min Median Max
Mili
tant
s K
illed
by
Dro
ne S
trik
es
Collaborators Killed, Within 25km of Strike(Prior 30 Days)
0
1
2
3
4
5
6
7
8
9
10
Min Median Max
Mili
tant
s K
illed
by
Dro
ne S
trik
es
Collaborators Killed, Within 25km of Strike(Prior 90 Days)
0
1
2
3
4
5
6
7
8
9
10
Min Median Max
Mili
tant
s K
illed
by
Dro
ne S
trik
es
Collaborators Killed, Within 25km of Strike
(Prior 365 Days)
0
1
2
3
4
5
6
7
8
9
10
Min Median Max
Mili
tant
s K
illed
by
Dro
ne S
trik
es
Collaborators Killed, Within 25km of Strike(All Prior Days)
65 CI 95 CI Mean Predicted Probability
nbreg, DV= Militants killed by drone strike; IV=Collaborators killed within 25km and n days prior;Controls=Territorial Control, PakMil Operations, Pop Density, Rough Terrain