is the rate of events of terrorism increasing?
DESCRIPTION
Terrorism is endemic to the modern world. It is impossible to board an airplane, attend a sporting event, or walk into a public building without experiencing its symptoms. However, is the incidence rate of such horrific events actually increasing? This paper draws data from The Global Terrorism Database, which collects information on terrorist events around the world (1970 through 2011), and attempts to answer this very question. This research applies G- Control Charts, most commonly used for monitoring of workplace accidents and various health care application, to determine if the time between incidences of terrorism has in fact decreased. Though not intended as a basis for policy decisions, the paper demonstrates a novel use of control charts and provides a basis for a better informed debate.TRANSCRIPT
Is the Rate of Events of Terrorism Increasing?
Brandon R. Theiss, [email protected]
About the Data Set
• The Global Terrorism Database (GTD) is an open-source database including information on terrorist events around the world from 1970 through 2012 (with additional annual updates planned for the future). Unlike many other event databases, the GTD includes systematic data on domestic as well as transnational and international terrorist incidents that have occurred during this time period and now includes more than 113,000 cases.
http://www.start.umd.edu/gtd/
How Terrorist Event Defined
• The GTD defines a terrorist attack as the threatened or actual use of illegal force and violence by a non state actor to attain a political, ‐economic, religious, or social goal through fear, coercion, or intimidation. In practice this means in order to consider an incident for inclusion in the GTD, all three of the following attributes must be present:• The incident must be intentional – the result of a conscious calculation on the
part of a perpetrator.• The incident must entail some level of violence or threat of violence including ‐
property violence, as well as violence against people.• The perpetrators of the incidents must be sub national actors. ‐ This database
does not include acts of state terrorism.
A note about 1993
• The original Pinkerton Global Intelligence Services (PGIS) data, upon which the 1970-1997 GTD data are based, consisted of hard-copy index cards, which were subsequently coded electronically by START researchers. Unfortunately, the set of cards for 1993 was lost prior to PGIS handing the data over to START
Limiting the Data
• Only Events that happened in the USA• Only Serious Events• At least one injury or• At least one fatality or• >$500,000 in Damage (inflation adjusted)
• Reduced to 160 observations
How to make sense of this data?
Industrial Engineering Approach to Analyzing Data• Control Charts- Shewhart charts are used to determine the extent of
common cause variation and identify possible points that have an assignable tool. It is a tool that allows for the critical few to be separated from the trivially many• CUSUM Chart - typically used for monitoring change detection• EWMA Chart - tracks the exponentially-weighted moving average of all prior
sample means• C- Chart - used to monitor the total number of events occurring in a given unit
of time
Dec-09Dec-05Dec-01Dec-97Dec-93Dec-89Dec-85Dec-81Dec-77Dec-73Jan-70
7
6
5
4
3
2
1
0
Date
Events
Time Series Plot of Events
Events 1970- 2012
2000199019801970
90
80
70
60
50
40
30
20
10
0
Decade
Count
172526
92Count of Events by Decade
Number of Events by Decade
J an-10Jan-06Jan-02Jan-98Jan-94Jan-90Jan-86Jan-82Jan-78Jan-74Jan-70
80
70
60
50
40
30
20
10
0
Date
Cum
ula
tive S
um
0UCL=1.60LCL=-1.60
CUSUM Chart of Events
Does a CUSUM chart Indicate a Change?
Not very useful as the chart signals on every point!
J an-10Jan-06Jan-02Jan-98Jan-94Jan-90Jan-86Jan-82Jan-78Jan-74Jan-70
3.5
3.0
2.5
2.0
1.5
1.0
0.5
0.0
Date
EWM
A
__X=0.333
UCL=0.732
LCL=-0.066
EWMA Chart of Events
What about the EWMA?
Does present a meaningful signal. However for greater than expected number of events
λ=0.2
J an-10Jan-06Jan-02Jan-98Jan-94Jan-90Jan-86Jan-82Jan-78Jan-74Jan-70
7
6
5
4
3
2
1
0
Date
Sam
ple
Count
_C=0.333
UCL=2.063
LCL=0
11111
1
1
1
1
C Chart of Events
What about the C Chart?
Signals on fewer points than EWMA. Still only signals for greater number of events
5/30
/200
9
11/2
7/20
00
4/12
/199
6
7/19
/198
4
9/12
/198
0
2/18
/197
8
1/24
/197
5
5/27
/197
2
11/15/
1970
5/26
/197
0
1/14
/197
0
800
700
600
500
400
300
200
100
0
Date
Elapse
d
Time Series Plot of Elapsed
Looking at the time between terrorism events (inter arrival rate)
7506004503001500
60
50
40
30
20
10
0
Elapsed
Frequency
Histogram of Elapsed
Inter Arrival Rate
Has the general appearance of a Poisson or Geometric Distribution with outliers
1/2/
2001
7/27
/199
6
2/22
/198
5
10/12/
1980
6/24
/197
8
3/2/19
75
8/17
/197
2
11/21/
1970
6/12
/197
0
1/14
/197
0
800
700
600
500
400
300
200
100
0
Date
Sam
ple
Count
_C=86.2UCL=114.0LCL=58.3
1
1
1
1
1
1
1
1
1111
1
1
111
1111
11
1
1
1
1
11
1
1
1
1
11
1
11
1
11
1
1
1
1
1
1
1111
1
1
111
1
1111
1
1
1
111
1
1
1
111
1
1
1
111
11
1
11111111111
11
1111
11111111111111111111111
111111
C Chart of Elapsed
C Chart of Inter Arrival Rate
Signals on the majority of points! Not very useful
G Control Charts
• Developed and evaluated for the monitoring the number of cases between hospital acquired infections and other adverse health events• Also can be used to monitor the number of days between events • Used to analyze rare events• Assumes that the data follows a geometric-type distribution• Assumes that the opportunities are reasonably constant.
1/2/
2001
7/27
/199
6
2/22
/198
5
10/12/
1980
6/24
/197
8
3/2/19
75
8/17
/197
2
11/21/
1970
6/12
/197
0
1/14
/197
0
800
700
600
500
400
300
200
100
0
Date
Days
Betw
een E
vents
CL=57.8
UCL=559.3
LCL=0
1
11
2
1
2222222222222222222222222
G Chart of DateEvent Probability = 0.012
G Chart of Inter Arrival Rate
Signals when there is a large number of days between events
Category81
078
075
072
069
066
063
060
057
054
051
048
045
042
039
036
033
030
027
024
021
018
015
012
0906030
70
60
50
40
30
20
10
0
Valu
e
ExpectedObserved
Chart of Observed and Expected Values
360
300
240
180
720
120
810
690
660
570
540
510
270
450
420
210
390
150
3306048
09030600
630
750
780
140
120
100
80
60
40
20
0
Category
Contr
ibute
d V
alu
e
Chart of Contribution to the Chi-Square Value by Category
Chi-Square Goodness-of-Fit Test for Observed Counts in Variable: Count N DF Chi-Sq P-Value158 26 259.591 0.000
Compared to a Geometric Distribution with a probability of 0.0117232
Does the inter arrival rate follow a Geometric Distribution?
2000199019801970
800
700
600
500
400
300
200
100
0
Decade
Elapse
d
202.588149.958
113.76939.7582
Boxplot of Elapsed
Is the number of days between events constant across decades?
Kruskal-Wallis Test on Elapsed
Decade N Median Ave Rank Z1970 91 20.00 59.8 -6.301980 26 104.00 102.6 2.811990 24 110.00 108.1 3.332000 17 91.00 109.2 2.84Overall 158 79.5
H = 40.01 DF = 3 P = 0.000H = 40.02 DF = 3 P = 0.000 (adjusted for ties)
Is the difference statistically significant?
1/2/
2001
7/27
/199
6
2/22
/198
5
10/1
2/19
80
6/24
/197
8
3/2/
1975
8/17
/197
2
11/21/
1970
6/12
/197
0
1/14
/197
0
1400
1200
1000
800
600
400
200
0
Date
Days
Betw
een E
vents
CL=27 CL=81 CL=78CL=143
UCL=268
UCL=781 UCL=753
UCL=1371
LCL=0 LCL=0 LCL=0 LCL=0
1970 1980 1990 2000
222222222222222222
G Chart of Date by DecadeEvent Probability = 0.024, 0.008, 0.009, 0.005
Dividing the arrival rate into stages by decade
ObamaBush IIClintonBushReaganCarterFordNixon
70
60
50
40
30
20
10
0
President
Count
1
14
19
8
19
25
11
63
Chart of President
The Number of Events by President in Office
ObamaBush IIClintonBushReaganCarterFordNixon
800
700
600
500
400
300
200
100
0
President
Elapse
d
211204.214
121.944
222.25
138.526
57.2482.181825.9516
Boxplot of Elapsed
The Inter Arrival Rate by President in Office
1/2/
2001
7/27
/199
6
2/22
/198
5
10/1
2/19
80
6/24
/197
8
3/2/19
75
8/17
/197
2
11/2
1/19
70
6/12
/197
0
1/14
/197
0
1600
1400
1200
1000
800
600
400
200
0
Date
Days
Betw
een E
vents
CL=18CL=52 CL=40 CL=99CL=119 CL=74CL=159UCL=177
UCL=500UCL=386
UCL=950
UCL=1141
UCL=716
UCL=1526
LCL=0LCL=0 LCL=0 LCL=0LCL=0 LCL=0LCL=0
Ford Carter Reagan BushClinton Bush IINixon
222222222
G Chart of Date by PresidentEvent Probability = 0.037, 0.013, 0.017, ..., 0.004
Dividing the arrival rate into stages by President in Office
Restricting To Only Events With Fatalities
Dec-09Dec-05Dec-01Dec-97Dec-93Dec-89Dec-85Dec-81Dec-77Dec-73Jan-70
3.0
2.5
2.0
1.5
1.0
0.5
0.0
Mon-Year
Events
Time Series Plot of Events
59 events where at least one individual was killed
Fatal Events 1970- 2012
J an-10J an-06J an-02J an-98J an-94J an-90J an-86J an-82J an-78J an-74J an-70
3.0
2.5
2.0
1.5
1.0
0.5
0.0
Mon-Year
Sam
ple
Count
_C=0.123
UCL=1.173
LCL=0
11
1
11
11
1
C Chart of Events
C Chart of Fatal Events by Month
Again signal that the 1970s had unusually high occurrences of terrorism
12/30/
1994
6/18
/198
4
1/28
/198
2
11/2
2/19
78
12/29/
1975
7/1/19
73
10/26/
1972
4/2/19
71
6/30
/197
0
4/24
/197
0
3000
2500
2000
1500
1000
500
0
Date
Elapse
d
Time Series Plot of Elapsed
Looking at the time between fatal events (inter arrival rate)
2400180012006000
35
30
25
20
15
10
5
0
Elapsed
Frequency
Histogram of Elapsed With Fatalities
Inter Arrival Rate of Fatal Events
7/27
/199
6
8/7/19
87
5/4/19
82
2/15
/197
9
9/10
/197
6
10/20/
1973
12/3
1/19
72
4/29
/197
1
7/17
/197
0
4/24
/197
0
3000
2500
2000
1500
1000
500
0
Date
Sam
ple
Count
_C=233UCL=278LCL=187
1
1
111
1
1
1
1
1
11
1111
1
1111
1
1111
11111111
111111111111
C Chart of Elapsed
C Chart of Fatal Events
Signals on almost every point!
7/27
/199
6
8/7/
1987
5/4/19
82
2/15
/197
9
9/10
/197
6
10/20/
1973
12/3
1/19
72
4/29
/197
1
7/17
/197
0
4/24
/197
0
3000
2500
2000
1500
1000
500
0
Date
Days
Betw
een E
vents
CL=173
UCL=1658
LCL=0
1
222
G Chart of DateEvent Probability = 0.004
G Chart of Fatal Events
Signals again that the 1970s had a high rate of events. Also signals that the ’00 had a lower rate
2000199019801970
40
30
20
10
0
Decade
Count
2
8
12
37
Chart of Decade
Fatal Events by Decade
7/27
/199
6
8/7/19
87
5/4/19
82
2/15
/197
9
9/10
/197
6
10/20/
1973
12/31/
1972
4/29
/197
1
7/17
/197
0
4/24
/197
0
4000
3000
2000
1000
0
Date
Days
Betw
een E
vents
CL=63 CL=177CL=397
UCL=610
UCL=1697
UCL=3791
LCL=0 LCL=0 LCL=0
1970 1980 1990
22
G Chart of Date by DecadeEvent Probability = 0.011, 0.004, 0.002
G Chart of Fatal Events by Decade
ObamaBush IIClintonBushReaganCarterFordNixon
3000
2500
2000
1500
1000
500
0
President
Elapse
d2818
803
548419.5
274.444168.111
274
55.8889
Boxplot of Elapsed
Fatal Events by President in Office
7/27
/199
6
8/7/19
87
5/4/19
82
2/15
/197
9
9/10
/197
6
10/20/
1973
12/31/
1972
4/29
/197
1
7/17
/197
0
4/24
/197
0
6000
5000
4000
3000
2000
1000
0
Date
Days
Betw
een E
vents
CL=40CL=412
CL=131 CL=225CL=267CL=569UCL=386
UCL=3940
UCL=1262
UCL=2150UCL=2553
UCL=5437
LCL=0LCL=0 LCL=0 LCL=0LCL=0LCL=0
Nixon FordCarter Reagan Bush Clinton
22
G Chart of Date by PresidentEvent Probability = 0.017, 0.002, 0.005, ..., 0.001
G Chart of Fatal Events by Decade
Conclusion
• Social and Political Science Researchers are presented with challenges of analyzing large data sets• Industrial Engineering Tools Techniques and methods can be
effectively used to analyze these data in novel ways• Like in a traditional industrial engineering context, using and analyzing
data with the proper tool can result in better outcomes for the customer
7506004503001500
48
36
24
12
0
7506004503001500
48
36
24
12
0
1970
Elapsed
Frequency
1980
1990 2000
Histogram of Elapsed
Panel variable: Decade
7506004503001500
50
40
30
20
10
0
Elapsed
Frequency
Histogram of ElapsedDecade = 1970
12/7/1
979
9/28
/197
8
9/10
/197
6
1/24
/197
5
2/27
/197
3
5/9/19
72
4/29
/197
1
8/24
/197
0
6/17
/197
0
5/1/19
70
1/14
/197
0
300
250
200
150
100
50
0
Date
Days
Betw
een E
vents
CL=27.2
UCL=268.0
LCL=022
222
222
22222222
22
G Chart of DateEvent Probability = 0.024
Category25
024
023
022
021
020
019
018
017
016
015
014
013
012
011
010
0908070605040302010
30
25
20
15
10
5
0
Valu
e
ExpectedObserved
Chart of Observed and Expected Values
240802012
022
021
06030200
100
250
1607015
013
014
017
040180
110
19090501023
0
9
8
7
6
5
4
3
2
1
0
Category
Contr
ibute
d V
alu
e
Chart of Contribution to the Chi-Square Value by Category
Chi-Square Goodness-of-Fit Test for Observed Counts in Variable: C-Obs
N DF Chi-Sq P-Value91 24 37.7535 0.037