Introduction

As a result of the events of 9/11, as well as the ongoing accounts of acts of terrorism in the media, we have seen rising national concerns about terrorists using sophisticated weapons (LaFree, Dugan, & Miller., 2015, p. 99). In response to these concerns, the U.S. government has invested more resources in counterterrorism measures, shifting the focus of the FBI from traditional crimes to counterterrorism after 9/11 (LaFree, Dugan, & Miller, 2015, p. 5). Beyond internal measures, Saddam Hussein’s association with terrorists and the threat of weapons of mass destruction (WMDs) were the main reasons given in the rationale for the invasion of Iraq in 2003 (Otterman, 2003). An ongoing international concern is the damage terrorists could cause if they obtain weapons of mass destruction (WMD) (NATO, 2015). However, the definition of a WMD is unclear (Kaszeta, 2014), so we will look at a more specific group of weapons with the capacity to cause significant damage: chemical, biological, radiological, and nuclear weapons, denoted, CBRN.

In this activity, we will study the extent to which CBRN weapons have been used, and analyze whether or not their past use fits with our perceptions. We will attempt to answer questions such as have CBRN weapons been used successfully in the past?; which weapons are more historically dangerous (measured by the number of fatalities and injuries) in the hands of terrorists, common weapons like firearms and explosives, or advanced weapons like chemical, biological, and nuclear weapons?; and what are the implications of past usage of CBRN weapons compared to other weapons in determining our priorities in counter-terrorism policies?

In this handout, we will use the Global Terrorism Database (GTD). The GTD contains information about more than 140,000 terrorist incidents occurring between 1970 and 2018.² The data in the GTD are gathered only from incidences that are reported in the media. The team managing the database verifies the information they gather through multiple news sources (LaFree, Dugan, & Miller, 2015).

Using Summary Statistics and Simple Plots to Identify Key Events

We start by reading in a simplified version of the full Global Terrorism Database downloaded April 26, 2019:

GTDdata <- read.csv("/home/share/Stat324/Data/GT2019.csv")

# View the first 5 rows of each column
datatable(head(GTDdata))

# View the structure of each column
str(GTDdata, max.level = 2)

## 'data.frame':    191464 obs. of  15 variables:
##  $ Year      : int  1986 1988 1988 1988 1991 1980 1984 1988 1988 1988 ...
##  $ Month     : int  3 5 7 12 9 12 12 2 4 4 ...
##  $ Day       : int  27 3 18 21 16 17 5 22 22 27 ...
##  $ Country   : chr  "Australia" "New Caledonia" "Australia" "Australia" ...
##  $ Region    : chr  "East Asia & Pacific" "East Asia & Pacific" "East Asia & Pacific" "East Asia & Pacific" ...
##  $ provstate : chr  "Victoria" "Sud" "Australian Capital Territory" "South Australia" ...
##  $ latitude  : num  -37.8 -22.3 -35.3 -34.9 -43.5 ...
##  $ longitude : num  145 166 149 139 173 ...
##  $ AttackType: chr  "Bombing" "Assassination" "Assassination" "Bombing" ...
##  $ TargetType: chr  "Armed Forces" "Armed Forces" "Armed Forces" "Armed Forces" ...
##  $ claimed   : int  NA NA NA NA NA NA NA NA NA NA ...
##  $ WeaponType: chr  "Explosives" "Explosives" "Explosives" "Explosives" ...
##  $ Fatalities: int  0 0 0 0 0 2 10 0 4 0 ...
##  $ Wounded   : num  21 0 0 0 0 0 0 0 5 0 ...
##  $ Success   : int  1 1 1 1 1 1 1 1 1 1 ...

EXERCISE 1. How many (total) incidents are there in this dataframe?
There are a total of 191,464 incidents in this dataframe.

EXERCISE 2. How many variables (columns)?
We have 15 columns.

Details for each variable are given in the GTdescription.csv file, which is in the Stat324/Data folder.

We will create plots and summary statistics to get a sense of how frequently attacks involving CBRN weapons have occurred, the danger they present, and the patterns of CBRN attacks around the world.

First we will restrict the data to the CBRN incidents. Then we’ll use the dim() command to determine how many CBRN incidents are in the restricted data. Finally, we’ll create a very basic histogram of the number of CBRN attacks each year.

# Create a subset of the original data by restricting it to only CBRN weapons:
CBRNIncidents <- GTDdata[GTDdata$WeaponType == "CBRN", ]

# Determine the number of terrorist incidents in the restricted data:
nrow(CBRNIncidents)

## [1] 384

# Determine the percentage of terrorist incidents that are CBRN attacks:
100 * nrow(CBRNIncidents)/nrow(GTDdata)

## [1] 0.2005599

# Create a histogram of the number of CBRN incidents each year:
ggplot(CBRNIncidents, aes(x = Year)) + geom_histogram(binwidth = 1, fill = "red",
    color = "black")

EXERCISE 3. How many CBRN attacks are listed in this dataset? What percentage of the total terrorist attacks were CBRN attacks?
We have 384 CBRN attacks in this dataset. This is .20% of the total terrorist attacks.

EXERCISE 4. Based on the histogram, have there been CBRN attacks during every decade from 1970-2018?
Yes there have been CBRN attacks in every decade.

EXERCISE 5. Have there been CBRN attacks during every year from 1970-2018?
No, there have not been CBRN attacks every year, such as 1972 and 1983-1984.

EXERCISE 6. Have the number of CBRN attacks become more frequent over time? How can you tell?
Yes, we can tell because of the spike in CBRN attacks after 2010.

Now let’s ask R to count the number of incidents in each Region, and color the histogram based on these regional counts. Recall that we need to use the summarize function to count, and we’ll need a named argument. (What name is used in the chunk below?) Because we have installed both the plyr and dplyr packages, R gets confused and doesn’t know which summarize function to use (it tries to use the plyr summarize function, which won’t work). So we will tell R what it should do by preceding the function name with the package name: dplyr::.

# Calculate and view the number of CBRN terrorist incidents in each region:
INC_per_Region <- dplyr::summarize(group_by(CBRNIncidents, Region), count = n())

# Look at the first few rows of the result:
datatable(INC_per_Region)

# Color the CBRN incidents histogram by regional counts, with a bin width of 10
# years:
ggplot(CBRNIncidents, aes(x = Year, fill = Region)) + geom_histogram(binwidth = 10,
    color = "black", linewidth = 0.1) + scale_fill_manual(values = customColors) +
    theme(legend.title = element_blank(), legend.key.size = unit(0.5, "cm"), legend.text = element_text(size = 8))

Consider questions 7 - 9 below. Can you answer all of them using this histogram with bin-widths of 10 years?

EXERCISE 7. Do CBRN attacks occur across the globe or only in some regions of the world?
They occur across the globe.

EXERCISE 8. Did all 7 regions experience at least 15 (total) CRBN attacks during 1970-2018?
Yes, all regions experienced minimum 15 total CRBN attacks.

EXERCISE 9. In what region(s) did most CBRN attacks occur during the 1970s? In what regions were there no CBRN attacks in the 70s?
There were more attacks in Europe & Central Asia and North America. There were no CBRN attacks in Sub Saharan Africa, East Asia & Pacific, Middle East & North Africa and South Asia regions.

The bin-widths of 10 years hide some of the information. Let’s change the bin-width to 3. Observe that when our bars have a width of 3 years we can see that this distribution is actually bimodal - there are two “peaks”. When the years are grouped in 10-year periods we can only see one peak.

# Make a histogram of CBRN incidents colored by regional counts, with a bin
# (rectangle) width of 10 years:
ggplot(CBRNIncidents, aes(x = Year, fill = Region)) + geom_histogram(binwidth = 3,
    color = "black", linewidth = 0.1) + scale_fill_manual(values = customColors) +
    theme(legend.title = element_blank(), legend.key.size = unit(0.5, "cm"), legend.text = element_text(size = 8))

EXERCISE 10. There is one more important feature that is lost by grouping the years in either 3 or 10-year periods. Compare these last two histograms to the initial histogram from the RESTRICT DATA CHUNK that used 1-year bin widths. What can you see that is not visible in the two later histograms?
We cannot see the corresponding count-y value for each region.

EXERCISE 11. Where did most CBRN attacks occur between 2010-2018?
The most CBRN attacks occured in South Asia and Middle East & North Africa.

The following code can be used to count CBRN attacks by the intended target (TargetType), as well as the number and percentage of “successful” attacks for each target. Please look carefully at how we calculate and report these values. We display the information using our datatable function.

# Create a table using counts of total incidents, total successful incidents by
# TargetType, and percent of successful incidents.
TOTAL_INCIDENT <- dplyr::summarise(group_by(CBRNIncidents, TargetType), TotalCount = n(),
    SuccessfulCount = sum(Success), PercentSuccessful = 100 * SuccessfulCount/TotalCount) %>%
    arrange(desc(SuccessfulCount))

datatable(TOTAL_INCIDENT, rownames = FALSE)

In this data, a successful attack is defined as “an act of terror was committed”; it does not necessarily mean that the terrorists succeeded in their goal.

EXERCISE 12. What target (TargetType) has been the site of the most CBRN attacks? How many attacks? Did this target also have the largest success rate /percentage?
Private Citizens & Property are the targets of most CBRN attacks. They have 92 total attacks with a smaller percent success rate than other TargetTypes.

EXERCISE 13. Which type of target had the second largest number of CBRN attacks? How many attacks? What was the success rate for these targets?
The second target was Government with 76 and a 80.26% success rate.

EXERCISE 14. Which target has the lowest percentage of successful attacks? Were there many attempts at these targets?
The target with the lowest percentage of successful attacks is Business. There were over 29 attempts.

EXERCISE 15. In general, do the targets with the smallest number of successes also have the smallest success percentages? Explain.
Yes, the targets with the smallest number of successes had the smallest success percentages.

EXERCISE 16. Modify the code above to find the count, number of successful, and percentage of successful AttackTypes. Which attack type was attempted most often in CBRN attacks? Do the most frequent types of attacks tend to correspond to the attacks with the largest percentage of successful attacks? Explain.

# Create a table using counts of total incidents, total successful incidents by
# TargetType, and percent of successful incidents.
ATTACK_Total <- dplyr::summarise(group_by(CBRNIncidents, AttackType), TotalCount = n(),
    SuccessfulCount = sum(Success), PercentSuccessful = 100 * SuccessfulCount/TotalCount) %>%
    arrange(desc(SuccessfulCount))

datatable(ATTACK_Total, rownames = FALSE)

An unarmed assault was the most attempted attack type. However, this was only ranked third in terms of most successful percent wise. Yes, there is a correlation between most attempted attack type and attacks with the largest percentage of successful attempts.

After looking at how many attacks occur, we may also be interested in knowing how dangerous these attacks have been - that is, how many fatalities or injuries did they cause? We will first investigate fatalities (deaths), using sum() to count how many people died from CBRN-related attacks, and the percentage of attacks with CBRN weapons that caused fatalities.

# Count the number of total number of deaths in a CBRN attack.  Note that we
# must tell R to ignore any missing/NA entries.
sum(CBRNIncidents$Fatalities, na.rm = TRUE)

## [1] 640

# Count the number of total number of deaths in the terrorist dataframe.
sum(GTDdata$Fatalities, na.rm = TRUE)

## [1] 434981

# Find the percentage of all terrorist attack deaths due to CBRN attacks:
100 * sum(CBRNIncidents$Fatalities, na.rm = TRUE)/sum(GTDdata$Fatalities, na.rm = TRUE)

## [1] 0.1471329

# Count the number of CBRN attacks that resulted in at least one fatality -i.e.
# Fatalities > 0:
sum(CBRNIncidents$Fatalities > 0, na.rm = TRUE)

## [1] 71

# Find the percentage of all terrorist attacks that were CBRN attacks with at
# least one fatality.
100 * sum(CBRNIncidents$Fatalities > 0, na.rm = TRUE)/nrow(GTDdata)

## [1] 0.03708269

# Find the percentage of the CBRN attacks that had at least one fatality.
100 * sum(CBRNIncidents$Fatalities > 0, na.rm = TRUE)/nrow(CBRNIncidents)

## [1] 18.48958

EXERCISE 17. (a) How many fatalities were the result of CBRN terrorist attacks in the last 45 years?
We had 640.
(b) What percentage of all deaths due to terrorists acts is this?
This is 0.1471 percent of all deaths.

EXERCISE 18. (a) How many CBRN attacks resulted in at least one fatality?
There were 71 CBRN attacks with at least one fatality.
(b) What percentage of all terrorist attacks is this?
This is roughly 0.03708% of all terrorist attacks.
(c) What percentage of the CBRN attacks resulted in at least one fatality?
About 18.49% of CBRN attacks resulted in at least one fatality.

Now we will display the ten most deadly attacks with information about their country and year, and make a scatterplot of fatalities with year on the x-axis.

# Create a table with three columns, Year, Country, and number of Fatalities,
# for each incident.  Sort the data from largest to smallest number of
# fatalities (decreasing, -).  Display the 10 cases that resulted in the most
# deaths from terrorist attacks involving CBRN weapons.
CBNTable <- CBRNIncidents %>%
    select(Year, Country, Fatalities) %>%
    arrange(-Fatalities)
datatable(CBNTable, rownames = FALSE)

# Create a scatterplot of fatalities vs. year
jitter <- position_jitter(width = 0.1, height = 0.1)
ggplot(data = CBRNIncidents, aes(x = Year, y = Fatalities)) + geom_point(size = 2,
    color = "purple3", position = jitter) + theme_bw()

Note that I added some “jittering” to the points in order to distinguish points that would otherwise lie on top of each other.

EXERCISE 19. (a) When, and where was the deadliest incident?
Uganda in 2000 had the deadliest incident.
(b) How many deaths occurred?
There were 200 deaths.
(c) Is this incident easy to spot on the scatterplot? Yes, it is easy to spot Uganda on the scatter plot.

• To find out more about the most deadly incidents, go to the GTD website. Click Access the GTD, then Search & Browse. Find the Advanced Search box on the top right. Expand the results and show more incidents per page to find the details of the event that caused the 200 fatalities in Uganda in 2000. (What day, what was the perpetrator group, what was the attack type, and what was the weapon type?)

EXERCISE 20. (a) When, and where was the CBRN attack that caused the second largest number of fatalities?
1999 in Columbia has the second largest number of fatalities.
(b) How many deaths in this incident?
There were 65 fatalities.

EXERCISE 21. Use the GTD website to provide more details about the event in Columbia in Exercise 20.
This was an Unarmed Assault that was considered successful. The target were Police, specifically the Colombian government. The weapon was chemical. The perpetrator group name was the FARC, or the Revolutionary Armed Forces of Colombia.

EXERCISE 22. Which type of CBRN weapon was the most deadly? Use the GTD website to determine which of chemical, biological, radiological or nuclear weapon types most often result in events with fatalities.
The CBRN weapon that was most deadly was Explosives followed by Firearms.

EXERCISE 23. Modify the code used in the COUNTING DEATHS and MOST DEADLY ATTACKS chunks to find the number of injuries (Wounded) caused by CBRN weapons.
(a) Find the total number of people wounded in a CBRN attack (taking into account missing/NA entries).
We have 15,056 total number of people wounded in a CBRN attack.

(b) Find the total number of wounded in the terrorist dataframe. We have a total of 549,754 wounded in the terrorist dataframe.

(c) Find the percentage of all those wounded in a terrorist attack who were wounded in a CBRN attack.
2.739% of all those wounded in a terrorist attack were wounded in a CBRN attack.

(d) Count the number of CBRN attacks that resulted in at least one injury.
184 attacks with at least one injury.
(e) Find the percentage of all terrorist attacks that were CBRN attacks with at least one injury.
0.0961% of all terrorist attacks were CBRN with at least one injury.
(f) Find the percentage of just CBRN attacks that had at least one injury.
47.92% of CBRN attacks had at least one injury.

# Count the number of total number of deaths in a CBRN attack Ignore null
sum(CBRNIncidents$Wounded, na.rm = TRUE)

## [1] 15056

# Count the number of total number of deaths in the terrorist dataframe.
sum(GTDdata$Wounded, na.rm = TRUE)

## [1] 549754

# Find the percentage of all terrorist attack deaths due to CBRN attacks:
100 * sum(CBRNIncidents$Wounded, na.rm = TRUE)/sum(GTDdata$Wounded, na.rm = TRUE)

## [1] 2.738679

# Count the number of CBRN attacks that resulted in at least one fatality -i.e.
# Fatalities > 0:
sum(CBRNIncidents$Wounded > 0, na.rm = TRUE)

## [1] 184

# Find the percentage of all terrorist attacks that were CBRN attacks with at
# least one fatality.
100 * sum(CBRNIncidents$Wounded > 0, na.rm = TRUE)/nrow(GTDdata)

## [1] 0.09610162

# Find the percentage of the CBRN attacks that had at least one wounded
100 * sum(CBRNIncidents$Wounded > 0, na.rm = TRUE)/nrow(CBRNIncidents)

## [1] 47.91667

(g) Provide a table of the 10 CBRN attacks that caused the most injuries, with information about their country and year.

CBRN10 <- CBRNIncidents %>%
    select(Year, Country, Wounded) %>%
    arrange(desc(Wounded)) %>%
    head(10)
datatable(CBRN10, rownames = FALSE)

EXERCISE 24. Create a scatterplot of Wounded by Year.

# Create a scatterplot of wounded by year
jitter <- position_jitter(width = 0.1, height = 0.1)
ggplot(data = CBRNIncidents, aes(x = Year, y = Wounded)) + geom_point(size = 2, color = "purple3",
    position = jitter) + theme_bw()

## Warning: Removed 26 rows containing missing values or values outside the scale range
## (`geom_point()`).

EXERCISE 25. From the statistics and graph, are CBRN attacks more likely to injure people or cause fatalities ?
They are most likely to injure than to cause fatalities.

EXERCISE 26. Use the advanced search at the GTD website to find the event which caused 5500 injuries in the attack listed in the first row of your table in Exercise 23. What type of event, where was it, when was it, and who initiated the event?
This event occured in Tokyo on 3/20/1995. This was an unarmed assault with a chemical weapon by the Aum Shinri Kyo.

EXERCISE 27. Describe the 1984 US attack that caused 751 injuries.
The 1984 US attack was an Unarmed Assault with a Biological weapon from the Rajneeshees.

These attacks reinforce what we found with fatalities: most successful attacks with CBRN weapons use chemical weapons², even though biological, radiological, and nuclear weapons threaten to do damage on a massive scale.

We’ve learned that attacks with sophisticated weapons do happen, and on occasion they can cause a considerable number of deaths and injuries. However, we have yet to see any attacks using CBRN weapons such as plagues, dirty bombs, and nuclear strikes that have caused mass injuries or fatalities. In the next section we compare CBRN weapons to other types of weapons.

Using Stacked-Line Graphs to Compare Weapon Types

The stacked-line graph allows us to compare various categorical variables over time. To prepare data to be displayed in a stacked-line graph, we organize it by year and a category of interest,e.g., region, religion, weapon, attack type, target, or success. We then count the number of incidents/fatalities/wounded in the given year and category. We will start by comparing the number of CBRN attacks to attacks using other types of weapons each year.

With some additional modifications, we can change the y-axis to display percentages, rather than absolute counts. We can also facet by the categories, separating the categories into different graphs. The stacked-line graph allows us to gain a better sense of how CBRN weapons compare to other weapons.

First we will count the number of Incidents for each Year and WeaponType. Then make graphs of incidents vs. year. Our first plots will be line graphs - they won’t be stacked. (Each weapon type will appear in it’s own graph.) In other words, we will facet by each weapon type in order to make comparisons easier.

# Sum the total number incidents for each year and type of weapon; store as a
# new dataframe
GTDweapon = dplyr::summarize(group_by(GTDdata, Year, WeaponType), Incidents = n())

# Create line graphs of number of incidents by weapon type
ggplot(data = GTDweapon, aes(x = Year, y = Incidents)) + geom_area() + aes(fill = WeaponType) +
    ylab("Incidents") + facet_wrap(~WeaponType, ncol = 3) + theme(legend.title = element_blank(),
    legend.key.size = unit(0.5, "cm"), legend.position = "bottom", strip.text = element_text(size = 10),
    legend.text = element_text(size = 8), axis.text = element_text(size = 7), axis.text.x = element_text(angle = 90))

EXERCISE 28. Which type of weapon saw the greatest number of incidents? During what years were the greatest number of this type of incident?
Explosives saw the greatest numbers of incidents during the 2010s-2020s.

EXERCISE 29. Can you see (from the graphs) the rarity of CBRN weapons? Can you tell in what years there were the greatest number of CBRN attacks?
Yes, from the graphs we can see CBRN weapons are rare. It seems there were CBRN attacks in 2015-2020, but it is hard to see.

Let’s modify the y-axis to extend to a maximum of 100 incidents, so that we can better see the CBRN weapons.

# Show a max of 100 incidents per weapon type each year
ggplot(data = GTDweapon, aes(x = Year, y = Incidents)) + geom_area() + aes(fill = WeaponType) +
    ylab("Incidents") + facet_wrap(~WeaponType, ncol = 3) + coord_cartesian(ylim = c(0,
    100)) + scale_fill_manual(values = customColors) + theme(legend.title = element_blank(),
    legend.key.size = unit(0.5, "cm"), legend.position = "bottom", strip.text = element_text(size = 10),
    legend.text = element_text(size = 8), axis.text = element_text(size = 7), axis.text.x = element_text(angle = 90))

EXERCISE 30. How was the code modified?
We added the coord_cartesian(ylim = c(0, 100)) and scale_fill_manual(values=customColors) to the code.

EXERCISE 31. What do these modified graphs show about CBRN versus other types of incidents?
These modified graphs show the frequency in which CBRN takes place, which is less than the other graphs, but it lets us make an estimate to the year.

We can also look at the percent of CBRN incidences each year. We will sum the total number of incidents for each year and then find the percentage of incidents each year corresponding to each weapon type. Notice our use of a left_merge, and the mutate verb. We’ll display using datatable so that we can search the entire table and sort by any column.

# Sum the total number incidents for each year
GTDbyYear = dplyr::summarize(group_by(GTDdata, Year), TotalIncidents = n())

# Merge the two data sets and find the percentage of incidents by weapon type.
GTDweapon = left_join(GTDweapon, GTDbyYear, by = "Year")
GTDweapon = mutate(GTDweapon, PercentIncident = 100 * Incidents/TotalIncidents)

# View the result to check that it is correct:
datatable(GTDweapon, rownames = FALSE)

EXERCISE 32. Sort the table and determine the year in which the greatest percentage of incidents occurred. What type of incident and what percentage of the incidents was this? ?
2012 had the greatest percentage incident, with Explosives as the Weapon Type and 63.45% percent incident.

EXERCISE 33. In what year was the smallest number of incidents? What was the weapon type and the percentage of incidents for that situation?
1971 with incidents of CBRN nature. This has a .21% incident percent.

Now we will create our first stacked line graph. Observe that we no longer facet - there is only one graph. But the percentages of incidents for each weapon type are stacked - one on top of the other in the graph.

# Create a stacked-line graph
ggplot(data = GTDweapon, aes(x = Year, y = PercentIncident)) + geom_area() + aes(fill = WeaponType) +
    ylab("Percentage of all Incidents") + scale_fill_manual(values = customColors) +
    scale_y_continuous(labels = percent) + theme(legend.title = element_blank(),
    legend.key.size = unit(0.5, "cm"), legend.position = "bottom", legend.text = element_text(size = 8))

To determine the percentage of firearm incidents 1970, you need to find the difference in the percentages indicated by the top and bottom of the blue (firearm) region. In 1970, the blue region seems to extend from roughly 35% to 48%. That makes the percentage of incidents that used firearms in 1970 about 48 - 35 = 13%. (If we look this up in our datatable, we find the actual percentage was 12.44%. We did a pretty good job estimating from the graph.) Observe that the only weapon type whose percentage of incidents you can determine solely from the height of its region is Unknown/Other beccause this category is at the bottom of the graph All of the other regions are “stacked” on top of this region. Also note that the sum of all the percentages is 100% - indicated by the fact that the top of all the stacked regions is at a height of 100%.
Use this graph to answer Exercises 34-39. (If you have to, you could use the datatable, but you should be able to answer each of these just from the graph.) `

EXERCISE 34. In 2018, was the percentage of incidents involving incendiary devices just under 25%?
Yes, they were under 6% roughly.

EXERCISE 35. In 1970, approximately what percentage of the incidents involved explosives? Firearms?
In 1970, approximately 51% incidents involved explosives and 12% involved firearms.

EXERCISE 36. In 1980 was the percentage of unknown/other incidents greater or less than the percentage of incendiary incidents?
The unknown was 13%, whereas incendiary was only 6%, so the percentage of unknown incidents was greater than incendiary.

EXERCISE 37. Which two weapon types are used most frequently in terrorist attacks?
Explosives and firearms were frequestly used.

EXERCISE 38. Was the use of CBRN weapons usually less than 1% of all incidents?
Yes, most of the time. The most CBRN was used was in 1998 with 1.49%.

EXERCISE 39. Does this support the media image that terrorists regularly rely on sophisticated weapons?
No, this shows the media tries to make terrorism more sophisticated. Most attacks relied on Explosives and Firearms.

Next we will look at how many deaths were caused by CBRN weapons compared to other weapons. The steps are the same, but now use counts of Fatalities instead of Incidents.

# Sum the total number of deaths (Fatalities) for each year
GTDdeath = dplyr::summarize(group_by(GTDdata, Year), TotalDeaths = sum(Fatalities,
    na.rm = TRUE))

# Sum the fatalities grouped by weapon type and year
GTDweaponDeath = dplyr::summarize(group_by(GTDdata, Year, WeaponType), Deaths = sum(Fatalities,
    na.rm = TRUE))

# Merge the two data sets and find percentage of deaths by weapon type.
GTDweaponDeath <- left_join(GTDweaponDeath, GTDdeath, by = "Year")
GTDweaponDeath = mutate(GTDweaponDeath, PercentDeath = Deaths/TotalDeaths)

# Create a stacked-line graph
ggplot(data = GTDweaponDeath, aes(x = Year, y = PercentDeath)) + geom_area() + aes(fill = WeaponType) +
    ylab("Percentage of Deaths") + scale_y_continuous(labels = percent) + scale_fill_manual(values = customColors) +
    facet_wrap(~WeaponType, ncol = 2) + theme(legend.title = element_blank(), legend.key.size = unit(0.5,
    "cm"), legend.position = "bottom", legend.text = element_text(size = 8))

EXERCISE 40. What new dataframe was created, and what information does it contain?
The new dataframe, GTDweaponDeath, contains information of the total number of deaths caused by different weapon types for each year.

EXERCISE 41. Which data table did you need to merge with the new table?
We merged GTDeath and GTData.

EXERCISE 42. In the stacked-line graph, can you see a small spike in the percentage of CBRN deaths? In what year? (If you can’t see the spike, or can’t tell what year, you may want to facet the graph by weapon. See the GTDWEAPON INCIDENT PERCENTAGES CHUNK for how to do this.)
Yes, there is a small spike in the year 2000.

EXERCISE 43. The CBRN “spike” accounts for approximately what percent of the terrorism-related fatalities in that year?
Roughly less than 5%.

EXERCISE 44. Identify the event we found earlier that might have caused this spike. (Hint: Look back at Exercise 19.)
Uganda in 2000.

While 200 deaths are hardly insignificant, it is still only a small percentage of all fatalities. Other weapons currently cause a much larger loss of life.

EXERCISE 45. Repeat the process (CHUNK COMPARE PECENTAGES OF DEATHS FOR EACH WEAPON) for non-fatal injuries, (Wounded), and describe any patterns that you find related to CBRN injuries.

# Sum the total number of non-fatal injuries (Wounded) for each year
GTDinjuries <- dplyr::summarize(group_by(GTDdata, Year), TotalInjuries = sum(Wounded,
    na.rm = TRUE))

# Sum the non-fatal injuries grouped by weapon type and year
GTDweaponInjuries <- dplyr::summarize(group_by(GTDdata, Year, WeaponType), Injuries = sum(Wounded,
    na.rm = TRUE))

# Merge the two data sets and find percentage of injuries by weapon type.
GTDweaponInjuries <- left_join(GTDweaponInjuries, GTDinjuries, by = "Year")
GTDweaponInjuries <- mutate(GTDweaponInjuries, PercentInjuries = Injuries/TotalInjuries)

# Create a stacked-line graph
ggplot(data = GTDweaponInjuries, aes(x = Year, y = PercentInjuries)) + geom_area() +
    aes(fill = WeaponType) + ylab("Percentage of Injuries") + scale_y_continuous(labels = scales::percent) +
    scale_fill_manual(values = customColors) + facet_wrap(~WeaponType, ncol = 2) +
    theme(legend.title = element_blank(), legend.key.size = unit(0.5, "cm"), legend.position = "bottom",
        legend.text = element_text(size = 8))

There was a spike in CBRN injuries in 1984 (which I believe was the attack caused by the Rajneeshee group) and once more in 1995. Aside from that, the CBRN followed a similar pattern to Melee and Incendiary, meaning there was not as many of CBRN involved.

Drawing Conclusions from our Graphs

These graphs and the GTD website allowed us to analyze CBRN attacks. We see that they typically resulted in a relatively small number of deaths and injuries, but occasionally produced a considerable number of deaths \((200)\) or injuries (~\(5500)\). However, these numbers are far from the millions threatened.

It is worth noting that the most harmful CBRN attacks used chemical weapons (with one exception),³ informing us that most attacks with biological, radiological, or nuclear weapons were either not attempted, or were not successful. In Putting Terrorism in Context, LaFree, Dugan, & Miller confirm that few organizations are willing to adopt biological or chemical weapons (2015, p. 189), and that there were only 13 out of 113,000 recorded cases of radiological weapons at that time. There are also no cases of nuclear weapons, in part due to the difficulty of obtaining and weaponizing the materials involved (2015, p. 191).

Using the stacked-graph plot, we were able to see that in terms of incidents, fatalities, and injuries, CBRN weapons are relatively rare compared to firearms and explosives. However, we saw that the attack by Aum Shinrikyo was damaging enough to result in over 1/3 of all the injuries in the year it took place, suggesting that CBRN weapons do have the potential to do significant damage. NATO’s webpage “Weapons of Mass Destruction” reinforces that the use of CBRN weapons could produce “incalculable consequences for global stability and prosperity” (2015).

It is important to correct the image that terrorists routinely use sophisticated weapons. Resources are primarily needed to stop terrorism using common weapons (explosives and firearms), which are currently causing the most deaths and injuries. However, given their high potential for damage, we can conclude that it is well worth taking preventative counterterrorism measures against future CBRN attacks.

Since there have been just a few CBRN attacks, it does mean that the first major attack will be difficult to predict. One potential avenue for indirectly anticipating major attacks is looking at what causes terrorist organizations to escalate the caliber of weapons they use. For example, terrorist organizations on the brink of defeat or with doomsday beliefs may be more inclined to use superweapons (Laqueur, 1996). Another avenue is performing case studies on the few organizations which made attempts, similar to the case studies in the Global Terrorism Index report (Institute for Economics and Peace, 2014) on the most dangerous terrorist organizations in 2014. It may be advisable to investigate each type of weapon separately, since patterns of attack with chemical weapons are unlikely to resemble biological weapons which are unlikely to resemble radiological or nuclear weapons. Another sophisticated and highly dangerous weapon raised by Laqueur (1996) is cyber-terrorism, which could cause as much damage as CBRN weapons without the difficulty in procuring and processing materials. Unfortunately, data on cyber-terrorism is not currently available in the GTD.

EXERCISE 46. Develop your own question of interest in relation to these data and create a graph to address your question. Do you see any interesting patterns if you use different colors, filters, or facets for particular variables of interest? You may also go to the Executive Summary of the Global Terrorism Index report (Institute for Economics and Peace, 2014) (http://www.visionofhumanity.org/sites/default/files/Global%20Terrorism%20Index%20Report%202014_0.pdf) and look for interesting claims that you can investigate. Submit at least one plot with a brief interpretation.

# Filter the data for the 1980s and the late 2010s
decade_80s <- GTDdata %>%
    filter(Year >= 1980 & Year <= 1989)
decade_2010s <- GTDdata %>%
    filter(Year >= 2010 & Year <= 2019)

# Sum the total number of incidents for each decade
total_incidents_80s <- decade_80s %>%
    summarize(TotalIncidents = n())
total_incidents_2010s <- decade_2010s %>%
    summarize(TotalIncidents = n())

# Sum the incidents grouped by weapon type for each decade
incidents_by_weapon_80s <- decade_80s %>%
    group_by(WeaponType) %>%
    summarize(Incidents = n())
incidents_by_weapon_2010s <- decade_2010s %>%
    group_by(WeaponType) %>%
    summarize(Incidents = n())

# Calculate the percentage of incidents by weapon type for each decade
incidents_by_weapon_80s <- mutate(incidents_by_weapon_80s, PercentIncidents = Incidents/total_incidents_80s$TotalIncidents)
incidents_by_weapon_2010s <- mutate(incidents_by_weapon_2010s, PercentIncidents = Incidents/total_incidents_2010s$TotalIncidents)

# Combine the data for both decades
combined_data <- bind_rows(mutate(incidents_by_weapon_80s, Decade = "1980s"), mutate(incidents_by_weapon_2010s,
    Decade = "2010s"))

# Create a bar plot to compare the distribution of incidents by weapon type
# between the decades
ggplot(data = combined_data, aes(x = WeaponType, y = PercentIncidents, fill = Decade)) +
    geom_bar(stat = "identity", position = "dodge") + ylab("Percentage of Incidents") +
    scale_y_continuous(labels = scales::percent) + theme(axis.text.x = element_text(angle = 45,
    hjust = 1)) + scale_fill_manual(values = c(`1980s` = "skyblue", `2010s` = "salmon")) +
    labs(fill = "Decade") + ggtitle("Comparison of Terrorist Incidents by Weapon Type: 1980s vs 2010s")

I wanted to compare the distribution of terrorist incidents between the decades 1980s and the 2010s. Interesting to note, Explosives and Firearms are still the most used weapon type, but Explosives were used more often in the 2010s. CBRN was used in the 2010s.

Homework

1. Answer the 46 questions in this document in the space provided below each exercise.

2. Change the name and date at the top of the file. Knit your file to an HTML document.

3. Open the HTML file in a browser, and save the file WITH YOUR NAME(s) in it.

4. Post the file in Moodle no later than midnight, Friday, May 3. Note: Late submissions will not receive a grade.