Modeling conflict, with examples from terrorism Kristian Skrede - - PowerPoint PPT Presentation

Modeling conflict, with examples from terrorism

Kristian Skrede Gleditsch

Based on joint work with Aaron Clauset (SFI), Lindsay Heger (UCSD), and Maxwell Young (Waterloo)

Department of Government, University of Essex & Centre for the Study of Civil War, PRIO http://privatewww.essex.ac.uk/∼ksg/

20 November 2007 CABDyN Seminar, Said Business School, Oxford University

K.S. Gleditsch (Essex) Modeling conflict Oxford CABDyN 1 / 18

Introduction

Scale invariance in conflict data Frequency and severity of terrorism (CYG JCR 2007) Implications for study of terrorism Frequency and severity in Israel-Palestine conflict (CHYG 2007)

K.S. Gleditsch (Essex) Modeling conflict Oxford CABDyN 2 / 18

Scale invariance in conflict data

L.F. Richardson 1948 demonstrated the scale invariant distribution

• f war magnitude/severity

However, almost all subsequent research considers conflict as incidence or binary events Some debate on general vs. separate theories for larger or smaller conflicts (International Interactions 1990) Specific work on war size: Cioffi-Revilla 1991 JCR forecast of Gulf War magnitude; Lacina 2006 JCR on civil wars Cederman 2003 APSR: computer simulation of geopolitical system that reproduces a scale invariant war distribution Johnson et al. 2006: scale invariance for a large range of conflicts, including events within conflict; Spirling ND for democide

K.S. Gleditsch (Essex) Modeling conflict Oxford CABDyN 3 / 18

Scale invariance in terrorism

Most studies of terrorism focus on incidence, or accounting for location where and when attacks occur CYG in J. Conflict Resolution 51 (2007): frequency-severity in MIPT data on terrorist events since 1968

1 10 100 1000 10000 10

−4

10

−3

10

−2

10

−1

10 severity of event, x P(X ≥ x) Deaths Injuries Total K.S. Gleditsch (Essex) Modeling conflict Oxford CABDyN 4 / 18

Summary of distributions

Distribution N x σstd xmax Ntail α xmin pKS ≥ Injuries 7456 12.77 94.45 5000 259 2.46(9) 55 0.41 Deaths 9101 4.35 31.58 2749 547 2.38(6) 12 0.94 Total 10878 11.80 93.46 5213 478 2.48(7) 47 0.99 A summary of the distributions with power-law fits from the maximum likelihood method. N (Ntail) depicts the number of events in the full (tail) distribution. The parenthetical value depicts the standard error of the last digit of the estimated scaling exponent. K.S. Gleditsch (Essex) Modeling conflict Oxford CABDyN 5 / 18

Why is this interesting?

Terrorist events vary dramatically in their severity Terrorist seek media attention and spectacular attacks More severe attacks can provide signals of resolve to governments Political and economic impact of terrorism a function of severity

11 Sept attack on WTC/Pentagon vs. previous 1993 WTC bombings 7 July London bombings vs. 21 July copy-cat attack

Suggested predictors in work on terrorist incidence (e.g., Li JCR 2005) unable to account for variation in severity Severity offers a complimentary perspective to incidence

K.S. Gleditsch (Essex) Modeling conflict Oxford CABDyN 6 / 18

Trends in average log-severity

1968 1972 1976 1980 1984 1988 1992 1996 2000 2004 1 2 3 4 mean log2 ( severity ) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 −1 −0.5 0.5 1 separation time, τ (years) auto−correlation mean + σ (upper) average log-severity (deaths), 24 months sliding window (lower) ACF of average log-severity K.S. Gleditsch (Essex) Modeling conflict Oxford CABDyN 7 / 18

Trends in scaling parameter

1968 1972 1976 1980 1984 1988 1992 1996 2000 2004 1.50 1.75 2.00 2.25 2.50 2.75 3.00 scaling exponent α year mean ± σ

1968 1972 1976 1980 1984 1988 1992 1996 2000 2004 0.25 0.5 0.75 1

pKS

1968 1972 1976 1980 1984 1988 1992 1996 2000 2004 1 5 10

xmin

1968 1972 1976 1980 1984 1988 1992 1996 2000 2004 1 7 14 21

interval (days) year

(L) Average scaling exponent α for two-year periods (R top) significance, one-sided KS test (R middle) estimated xmin (R bottom) average inter-event interval for events in tail K.S. Gleditsch (Essex) Modeling conflict Oxford CABDyN 8 / 18

Disaggregating by locus and weapon type

1 10 100 1000 10000 10

−4

10

−3

10

−2

10

−1

10 severity of event, x (total) P( X ≥ x ) Events inside OECD Events outside OECD

10

−4

10

−2

10 Chem/Bio

Explosives

10

−4

10

−2

10 P(X ≥ x) Fire

Firearms

10 10

2

10

4

10

−4

10

−2

10 severity, x (total) Knives 10 10

2

10

4

severity, x (total) Other

(L) Frequency-severity distributions for OECD and non-OECD nations (R) Frequency-severity distributions by weapon types K.S. Gleditsch (Essex) Modeling conflict Oxford CABDyN 9 / 18

Generative models for scale invariance

Many generative models can generate scale invariant distributions Self-organized criticality model applied to forms of conflict such as interstate wars, strikes Limitations

Terrorism is not inherently spatial phenomenon Severity not only function of size of explosion Substitution between targets/weapons

Johnson et al. fragmentation and coalescence model of insurgency CYG JCR: Toy model for scale invariance through competitive forces

K.S. Gleditsch (Essex) Modeling conflict Oxford CABDyN 10 / 18

Toy model for scale invariance

Competition non-state actor (terrorist) and government Severity function of planning and time invested Selection mechanism where probability event executed inversely related to planning required Payoff of additional planning proportional to time already invested Potential severity: p(t) ∝ eκt Severity of real event to planning time of a potential event: x ∝ eλt After selection of realized events:

• p(x) dx =
• p(t) dt → p(x) ∝ x−α where α = 1 − κ/λ

If slight advantage to state |κ| |λ|, then we get a power law with exponent α 2

K.S. Gleditsch (Essex) Modeling conflict Oxford CABDyN 11 / 18

Counterterrorism and beyond

Standard approach considers substitution of targets (Enders and Sandler 1993, 2002): Countermeasures make alternative targets relatively more attractive But, calculus of terrorism much more complicated, e.g.:

inter-group competition, political support violence vs. non-violence, severe vs. non-severe

Data allow evaluating these influences in Israel-Palestine conflict Focus on main players: Fatah, Hamas, PFLP , PIJ Plus, data on Israeli countermeasures and Palestinian support

K.S. Gleditsch (Essex) Modeling conflict Oxford CABDyN 12 / 18

Israeli countermeasures

2000 2001 2002 2003 2004 2005 2006 1 3 5 7 9 11 13 15

Year Number of incidents

Suicide attacks Israeli counter attacks 2000 2001 2002 2003 2004 2005 2006 20 40 60 80 100 120 140

Year Number of incidents

Non−suicide attacks Israeli counter attacks

(L) Counts for suicide attacks and Israeli counter-terrorism events (R) Counts for non-suicide and Israeli counter-terrorism events

K.S. Gleditsch (Essex) Modeling conflict Oxford CABDyN 13 / 18

Competition, imitation, and public opinion

1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 25 50 75 100 125 150 175 200 225 Severity Year 10 20 30 40 50 60 Public Opinion Percentage Severity Data Fatah Percentage 99% Confidence Interval 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 25 50 75 100 125 150 175 200 225 Severity Year 10 20 30 40 50 60 Public Opinion Percentage Severity Data Hamas Percentage 99% Confidence Interval

(L) Fatah suicide attack severity (left axis) and public approval (right) (L) Hamas suicide attack severity (left axis) and public approval (right)

K.S. Gleditsch (Essex) Modeling conflict Oxford CABDyN 14 / 18

No evidence of coordination

−40 −30 −20 −10 10 Fatah−PLO to Hamas Conflict−Cooperation 2001 2002 2003 2004 2005 −30 −25 −20 −15 −10 −5 5 Hamas to Fatah−PLO Conflict−Cooperation 2001 2002 2003 2004 2005

(L) Fatah to Hamas conflict cooperation score, by week (R) Hamas to Fatah conflict cooperation score, by week

K.S. Gleditsch (Essex) Modeling conflict Oxford CABDyN 15 / 18

Change in share of claimed attacks

2000 2001 2002 2003 2004 2005 0.2 0.4 0.6 0.8 1 fraction of events per 6 months claimed / attributed unknown

K.S. Gleditsch (Essex) Modeling conflict Oxford CABDyN 16 / 18

Attack modes and elections

1 2 3 4

event count

Hamas suicide attacks 2000 2001 2002 2003 2004 2005 2006 50 100 150 200

• ave. casualties per event

50 100 150

event count

2000 2001 2002 2003 2004 2005 2006 50 100 150 200

• ave. casualties per event

Hamas non−suicide attacks Palestinian elections (L) Incident frequency (upper pane) and average casualties per attack (lower), suicides (R) Incident frequency (upper pane) and average casualties per attack (lower), non-suicides K.S. Gleditsch (Essex) Modeling conflict Oxford CABDyN 17 / 18

Summary

Frequency-severity distributions in conflict data Thinking about event severity offers important new insights Calculus of terrorism is highly complex

Many possible strategies Many possible targets and modes Many possible interpretations of data

In Israel-Palestine conflict evidence of

inter-group competition: innovation, imitation interaction with political processes: public support, elections

K.S. Gleditsch (Essex) Modeling conflict Oxford CABDyN 18 / 18