EDA Project_Mass Shooting_v4.Rmd

Mass Shootings in the U.S. 1966 - 2017 by Jesse Montgomery
========================================================

Here, I analyze the Stanford Mass Shootings in America (MSA) dataset found on\
Kaggle.com (https://www.kaggle.com/carlosparadis/stanford-msa). This includes \
data of mass shootings (defined as 3 or more victims) occuring in the United \
States between 1966 and 2016. This is described as a high-quality, but not\
comprehensive dataset. To capture recent shootings not included in this \
dataset, I added data from the US Mass Shootings dataset\
(https://www.kaggle.com/zusmani/us-mass-shootings-last-50-years). This an open\
dataset that is continually updated, but of lower quality. To explore \
additional variables, I combined this dataset with US census data\
(https://www2.census.gov/programs-surveys/popest/datasets/2010-2016/cities/\
totals/sub-est2016_all.csv) as well as climate data from the National Oceanic \
and Atmosphere Administration\
(https://www1.ncdc.noaa.gov/pub/data/ccd-data/nrmavg.txt).

```{r echo=FALSE, message=FALSE, warning=FALSE, packages}
# Load all of the packages that you end up using in your analysis in this code
# chunk.

# Notice that the parameter "echo" was set to FALSE for this code chunk. This
# prevents the code from displaying in the knitted HTML output. You should set
# echo=FALSE for all code chunks in your file, unless it makes sense for your
# report to show the code that generated a particular plot.

# The other parameters for "message" and "warning" should also be set to FALSE
# for other code chunks once you have verified that each plot comes out as you
# want it to. This will clean up the flow of your report.
options(scipen=999) #Suppress scientific notation to make plots more readable.
library(ggplot2)
library(gridExtra)
library(knitr)
library(tidyr)
library(dplyr)
library(reshape2)
library(ggmap)
library(colorspace)
```

```{r echo=FALSE, Load_the_Data}
# Load the Data
# Work
setwd(paste("C:\\Users\\jesse_montgomery\\Google Drive\\Scripts\\R\\",
"Udacity_DAND\\Exploratory Data Analysis\\Project_Mass Shootings", sep=""))
# # Home
# setwd(paste("C:\\Users\\jesse\\Google Drive\\Scripts\\R\\",
# "Udacity_DAND\\Exploratory Data Analysis\\Project_Mass Shootings", sep=""))
# Mass shooting data
shootings <- read.csv(
  "mass_shooting_events_stanford_msa_release_06142016_2.csv", 
  stringsAsFactors = F)
# Climate data
climate <- read.csv("MonthlyAverageTempByCity.csv", stringsAsFactors = F)
# Population data
populations <- read.csv("sub-est2016_all.csv", stringsAsFactors = F)
```


```{r echo=FALSE, Clean_the_Data}
# Remove rows with only NA values
shootings = na.omit(shootings)
# Create a column for numeric Julian date
shootings$Date.Numeric = shootings$Date
# Convert date strings to Julian date. Make formatting conditional based on 
# whether there is 2 or 4 digit year
for (i in 1:length(shootings$Date)) { 
  if (as.Date(shootings$Date[i], "%m/%e/%y") < as.Date("2018-01-01")) { 
    shootings$Date.Numeric[i] <- as.Date(shootings$Date[i], "%m/%e/%y")
  } else {shootings$Date.Numeric[i] <- as.Date(shootings$Date[i], "%m/%e/%Y")
  }
}
shootings$Date.Numeric = as.numeric(shootings$Date.Numeric)
shootings$Date = as.Date.character(shootings$Date, format = "%m/%d/%Y")
# Make a separate column for year, month, and day
shootings$Year <- lapply(shootings$Date.Numeric, function(x) {
                         format(as.Date(x, origin=as.Date('1970-01-01')), 
                                format = '%Y')})
shootings$Year <- as.numeric(shootings$Year)
shootings$Month <- lapply(shootings$Date.Numeric, function(x) {
                          format(as.Date(x, origin=as.Date('1970-01-01')), 
                                 format = '%b')})
shootings$Month <- factor(shootings$Month, levels = 
                            c("Jan", "Feb", "Mar", "Apr", "May", "Jun", "Jul", 
                              "Aug", "Sep", "Oct", "Nov", "Dec"))
shootings$Day <- lapply(shootings$Date.Numeric, function(x) {
                        format(as.Date(x,origin=as.Date('1970-01-01')), 
                               format = '%d')})
shootings$Day <- as.numeric(shootings$Day)

# Fix formatting on other columns of interest. 
shootings$Day.of.Week <- factor(shootings$Day.of.Week, levels = 
                                  c("Sunday", "Monday", "Tuesday", "Wednesday",
                               "Thursday", "Friday", "Saturday"))
shootings$Average.Shooter.Age <- suppressWarnings(as.numeric(
  shootings$Average.Shooter.Age))
shootings$Number.of.Shotguns <- suppressWarnings(as.numeric(
  shootings$Number.of.Shotguns))
shootings$Number.of.Rifles <- suppressWarnings(as.numeric(
  shootings$Number.of.Rifles))
shootings$Number.of.Handguns <- suppressWarnings(as.numeric(
  shootings$Number.of.Handguns))
shootings$Number.of.Automatic.Guns <- suppressWarnings(as.numeric(
  shootings$Number.of.Automatic.Guns))
shootings$Number.of.Semi.Automatic.Guns <- suppressWarnings(as.numeric(
  shootings$Number.of.Semi.Automatic.Guns))
shootings$Total.Number.of.Guns <- suppressWarnings(as.numeric(
  shootings$Total.Number.of.Guns))
shootings$Shooter.Sex <- factor(shootings$Shooter.Sex, levels = 
                                  c("Male", "Female"))
shootings$Fate.of.Shooter.at.the.scene <- factor(
  shootings$Fate.of.Shooter.at.the.scene)
shootings$Shooter.s.Cause.of.Death <- factor(shootings$Shooter.s.Cause.of.Death)
shootings$School.Related <- factor(shootings$School.Related)
shootings$History.of.Mental.Illness...General <- factor(
  shootings$History.of.Mental.Illness...General)
shootings$Military.Experience <- factor(shootings$Military.Experience)
shootings$State <- factor(shootings$State)
shootings$Shooter.Race <- factor(shootings$Shooter.Race)
shootings$Possible.Motive...General <- factor(
  shootings$Possible.Motive...General)
shootings$Targeted.Victim.s...General <- factor(
  shootings$Targeted.Victim.s...General, levels = 
    sort(unique(shootings$Targeted.Victim.s...General), decreasing = TRUE))
shootings$City <- factor(shootings$City)
shootings$Place.Type <- factor(shootings$Place.Type)
```

```{r echo=FALSE, warning=FALSE, message=FALSE}
# This function will create a vector of indices that will be used to map each 
# shooting city to the nearest city in the climate data. Proximity is 
# determined by latitude and longitude in both datasets. The distance of each
# shooting location to the cities in the climate dataset was calculated using
# the euclidean distance formula (a^2 + b^2 = c^2).
get.ref.indices <- function(reference.data) {
  # Create an empty vector for the indices
  reference.data.indices <- vector(mode = "numeric", length = nrow(shootings))
  # Loop through each row of the shootings dataset.
  for (i in 1:nrow(shootings)) {
  distance.vector <- 
    # Calculate the euclidean distance of the shooting location to each city
    # in the climate dataset.
    sqrt((shootings$Longitude[i] - reference.data$Longitude)^2 + 
           (shootings$Latitude[i] - reference.data$Latitude)^2)
  # Find the minimum distance and get the index. 
  reference.data.indices[i] <- which.min(distance.vector)
  }
  # Return the index for all shooting locations corresponding to the closest
  # city in the climate dataset.
  return(reference.data.indices)
}

# Run the above function on the climate data set.
climate.indices <- get.ref.indices(climate)

# Create a column of the average monthly temperature of the shooting location
# based on the climate dataset.
for (i in 1:nrow(shootings)) {
shootings$Location.Temp.Month.Avg[i] <- 
  climate[climate.indices[i], as.character(shootings$Month[i])]
}
shootings$Location.Temp.Month.Avg <- 
  as.numeric(shootings$Location.Temp.Month.Avg)

# Create a column of population based on the census data. The for loop below
# simply searches for the name of the city and state in the shooting dataset 
# matching the city and state in the census data.
shootings$Population <- numeric(nrow(shootings))
for (i in 1:nrow(shootings)) {
   city_population <- as.numeric(
    populations[(as.character(shootings$City[i]) == populations$NAME) & 
                   (as.character(shootings$State[i]) == populations$STNAME), 
                 "CENSUS2010POP"])
   if (length(city_population) > 0) {
     shootings$Population[i] = city_population
   } else {
     shootings$Population[i] = NA
   }
}
```

# Univariate Plots Section
I will break the univariate analysis up in five sections: 

1. The magnitude of shootings. What is the distribution of those killed/injured?
    + Total number of fatalities 
    + Total number of inuries
2. Shootings over time. Are more shootings happening? Is there a \
seasonality?
    + By day of the week
    + By year
    + By month 
3. Location. Where do mass shootings occur most often? Are there common \
characteristics about the places?
    + State
    + City
    + Population
    + Type of place
4. Weapons. Are there types of weapons that are used more than others? Would \
gun control be more effective for a type of gun?
    + Type of gun
    + Number of guns
5. Shooter demographics. Are there common characteristics among shooters?
    + Race
    + Age
    + History of mental illness
    + Possible motives
    + Targeted victims

## 1. Magnitude of Shootings

### Variable Summaries
Number of fatalities and number of victims injured.

```{r echo=FALSE}
# Isolate the more interesting columns for summary
Magnitude_Vars_of_Interest = 
  c("Total.Number.of.Fatalities", "Number.of.Victims.Injured")
kable(summary(shootings[Magnitude_Vars_of_Interest]))
```

There is a very large difference between Max and all other stats for each of \
these variables indicating some outliers or a very skewed distribution. Let's \
take a look.

### Plots of Fatalities and Victims

```{r echo=FALSE}
p1 <- ggplot(data = shootings) +
  geom_histogram(aes(x = Total.Number.of.Fatalities), fill = 'orange', 
                 binwidth = 1) 
p2 <- ggplot(data = shootings) +
  geom_histogram(aes(x = Number.of.Victims.Injured), fill = 'orange', 
                 binwidth = 1) 
grid.arrange(p1, p2)
```

There are several outliers. The most extreme is the shooting with over 500 \
injured victims. Let's see what shooting this is.

```{r echo=FALSE}
kable(shootings[shootings$Number.of.Victims.Injured > 500, 
          c("Title", "Date...Detailed", "Location", 
            "Total.Number.of.Fatalities", "Number.of.Victims.Injured")])
```

It is no surprise that this was the recent shooting in Las Vegas, the most \
deadly in history. It also accounts for the highest fatality outlier.

This is one analysis where outliers cannot be ignored. But to visualize the \
distributuion of other shootings better, I will plot the lower 90% of data.

```{r echo=FALSE}
# Data is subsetted using the quantile function, using a quantile of 0.9.
p1 <- ggplot(data = subset(shootings, Total.Number.of.Fatalities < 
                             quantile(shootings$Total.Number.of.Fatalities, 
                                      0.9))) +
  geom_histogram(aes(x = Total.Number.of.Fatalities), fill = 'orange', 
                 binwidth = 1) +
  scale_x_continuous(breaks = seq(0, 8, 1))
p2 <- ggplot(data = subset(shootings, Number.of.Victims.Injured < 
                             quantile(shootings$Number.of.Victims.Injured, 
                                      0.9))) +
  geom_histogram(aes(x = Number.of.Victims.Injured), fill = 'orange', 
                 binwidth = 1) +
  scale_x_continuous(breaks = seq(0, 10, 1))
grid.arrange(p1, p2)
```

The distribution for mass shootings is skewed to the right. Most shootings tend \
to have less victims.

Let's look at the magnitude of the top 5 mass shootings, including when and \
where they occurred. I will define the top 5 mass shootings as those with the \
greatest number of fatalities.

```{r echo=FALSE}
# Sort the data by number of fatalities
shootings.Sorted.by.Fatalities <- 
  shootings[order(-shootings$Total.Number.of.Fatalities), ]
kable(shootings.Sorted.by.Fatalities[1:5, 
      c("Title", "Date...Detailed", "Location", "Total.Number.of.Fatalities", 
            "Number.of.Victims.Injured")])
```


With the exception of the mass shooting at Virginia Tech, the most severe \
shootings have occured in the last 5 years. It is a grim trend to consider.

## 2. Shootings Over Time
### Variable Summaries
Year, day of the week, and the month.

```{r echo=FALSE}
# Isolate the more interesting columns for summary
Time_Vars_of_Interest = 
  c("Year", "Day.of.Week", "Month")
kable(summary(shootings[Time_Vars_of_Interest]))
```

The median year being very near the Max indicates a skewed distribution.There \
doesn't appear to be an obvious trend with day of the week, except that mass \
shootings are less common on Tuesday and Saturday.  Winter and Spring\
months appear to have more shootings.

### Shootings by Year

```{r echo=FALSE}
ggplot(data = shootings) +
  geom_histogram(aes(x = Year), fill = 'orange', binwidth = 1) +
  scale_x_continuous(breaks = seq(1965, 2015, 5))

```

The distribution is certainly skewed to the left. It looks like there has \
been an increase in the last five years with a major spike in 2015 and 2016. It\
would appear that things are getting worse. 

### Shootings by Day of the Week

```{r echo=FALSE, warning=FALSE}
ggplot(data = shootings) +
  geom_histogram(stat = 'count', aes(x = Day.of.Week), fill = 'orange')
```

Perhaps Tuesday and Saturday are safer days to be in public?

### Shootings by Month

How about month? Is there a seasonality to mass shootings?

```{r warning=FALSE, echo=FALSE}
ggplot(data = shootings) +
  geom_histogram(stat = 'count', aes(x = Month), fill = 'orange')
```

Indeed, it does appear that most shootings occur in winter and early spring \
months with the exception of January. This surprises me. I think of mass \
shootings being random attacks at large gatherings which I picture occuring \
outdoors more often during the summer. Perhaps depression influenced by darker \
and colder months is a factor. Let's see if temperature is a factor. I have \
created a column of the likely temperature that month in the area pulled from\
National Oceanic and Atmospheric Administration (NOAA) climate data.

```{r echo=FALSE, message=FALSE}
ggplot(data = shootings) +
  geom_histogram(aes(x = Location.Temp.Month.Avg), fill = 'orange')
```

There is a fairly normal distribution centered about 55 degrees, which is \
the average annual temperature in the United States (55.7 $^\circ$F, \
https://www.usclimatedata.com). If colder temperatures were a contributing\
factor, I would expect data to be skewed to the right or at least the mean\
shifted to the left of the national average temperature. 

## 3. Shootings by Location
### Variable Summaries
State, city, population, and type of place.

```{r echo=FALSE}
# Isolate the more interesting columns for summary
Location_Vars_of_Interest = 
  c("State", "City", "Population", "Place.Type")
kable(summary(shootings[Location_Vars_of_Interest]))
```

Let's at the locations of mass shootings. First by state.

### Shootings by State

```{r echo=FALSE, warning=FALSE, Univariate_Plots}
# First reorder the factors in order of decreasing frequency.
# Order states by frequency. Create a table and then sort it. Then refactor by 
# the sorted names of the table.
states_decreasing = sort(table(shootings$State), decreasing = T)
shootings.Sorted.by.State <- 
  within(shootings, State <- factor(State, levels=names(states_decreasing)))
## plot
ggplot(data = shootings.Sorted.by.State) +
  geom_histogram(stat = 'count', aes(x = State), fill = 'orange') +
  theme(axis.text.x = element_text(angle = 35, hjust = 1))

# Sum of incidents in the top 5 states
SumTopFiveStates <- sum(states_decreasing[1:5])
# Sum of incidents in all states
SumAllStates <- sum(states_decreasing)
# Fraction of all shootings that occured in the top 5 states
FractionTop5States <- SumTopFiveStates/SumAllStates
```

The top 3-5 states tower over the rest. The top 5 states account for 35% of \
all shootings. Is there something unique about these locations? 

### Shootings by City

Mass shootings are concentrated in a few states. Are they repeated in \
individual cities? Again, we will order the histogram by frequency.

```{r, warning=FALSE, echo=FALSE}
city_decreasing = sort(table(shootings$City), decreasing = T)
shootings.Sorted.by.City <- within(shootings, City <- 
                                    factor(City, levels=names(city_decreasing)))
ggplot(data = shootings.Sorted.by.City) +
  geom_histogram(stat = "count", aes(x = City), fill = 'orange') +
  theme(axis.text.x = element_blank())
```

Fewer than a quarter of the cities have more than one incident. Let's zoom in \
on those.

```{r, warning=FALSE, echo=FALSE}
shootings.Sorted.by.City.Top <- 
  subset(shootings.Sorted.by.City, City %in% 
           names(city_decreasing[city_decreasing > 1]))
ggplot(data = shootings.Sorted.by.City.Top) +
  geom_histogram(stat = "count", aes(x = City), fill = 'orange') +
  theme(axis.text.x = element_text(angle = 35, hjust = 1))
```

There are several highly populated cities in this list, but that isn't the \
case for all of them.

Now let's look at the population of cities with mass shootings.

### Shootings and Population

```{r, warning=FALSE, echo=FALSE}
ggplot(data = shootings) +
  geom_histogram(aes(x = Population), fill = 'orange', binwidth = 10000)
```

There are a handful of cities with very high populations. Let's look at the \
lower 90%

```{r, warning=FALSE, echo=FALSE}
ggplot(data = subset(shootings, Population < 
                       quantile(shootings$Population, 0.9, na.rm = T))) +
  geom_histogram(aes(x = Population), fill = 'orange', binwidth = 10000)
```

The distribution is skewed to the right. Mass shootings occur more frequently in\
cities with lower populations. Half of them occur in cities with less than \
100,000 people.

### Shootings by Type of Place

What types of locations or facilities are most frequently targeted by shooters?

```{r, warning=FALSE, echo=FALSE, PlotTypeHist}
ggplot(data = shootings) +
  geom_bar(aes(x = Place.Type), fill = 'orange') +
  theme(axis.text.x = element_text(angle = 35, hjust = 1))
```

Neighborhoods are the most frequently targeted. Places of business, education \
and government are also common.

## 4. Weapons
What weapons were involved in mass shootings? Should gun control efforts be \
placed on certain types of weapons?

Number of each type of gun.

```{r, warning=FALSE, echo=FALSE, message=FALSE}
# Melt data with reshape2 to put into long form that can be easily plotted
# with ggplot2.
Weapons_Vars_of_Interest = c("Number.of.Shotguns", "Number.of.Rifles", 
                             "Number.of.Handguns", "Number.of.Automatic.Guns",
                             "Number.of.Semi.Automatic.Guns")

Gun.type <- melt(shootings[Weapons_Vars_of_Interest], 
                 variable.name = "Gun.description",
                 value.name = "Count")

ggplot(data = Gun.type) +
  stat_summary(aes(x = Gun.description, y = Count), 
               fun.y = "sum", fill = "orange", geom = "bar") +
  theme(axis.text.x = element_text(angle = 35, hjust = 1))

```

Handguns guns are the most common weapons. Automatic and semi-automatic are \
likely a subcategories of the other three gun types. Gun control efforts \
would probably be best concentrated on handguns and especially those that are\
semi-automatic.

## 5. Shooter Deomographics
Do shooters have anything in common?


### Variable Summaries
Race, age, history of mental illness, possible motives

```{r echo=FALSE}
# Isolate the more interesting columns for summary
Shooter_Vars_of_Interest = 
  c("Shooter.Race", "Average.Shooter.Age", 
    "History.of.Mental.Illness...General", 
    "Possible.Motive...General")
kable(summary(shootings[Shooter_Vars_of_Interest]))
```

Most shooters are White Caucasian, followed by African American and about 30 \
yrs old. Mental illness is known to affect about a third of the shooters. The\
rest either had no history of mental illness or it was unknown. Most\
motives are not known.

### Race

```{r, warning=FALSE, echo=FALSE}
ggplot(data = shootings) +
  geom_histogram(stat = "count", aes(x = Shooter.Race), fill = 'orange') +
  theme(axis.text.x = element_text(angle = 35, hjust = 1))
```

Races besides White and African American are few.

### Age

```{r, warning=FALSE, echo=FALSE}
ggplot(data = shootings) +
  geom_histogram(aes(x = Average.Shooter.Age), fill = 'orange', binwidth = 1) + 
  scale_x_continuous(breaks = seq(0, 70, 10))

Number.Minor.Shooters <- nrow(subset(shootings, shootings$Average.Shooter.Age <
                                       18, select = Average.Shooter.Age))
```

There appears to be an almost bimodal distribution around ages 20 and 40. A\
large proportion of shooters are very young. At least 35 of them were minors.

### Possible Motive
It is apparent that most motives are unknown. Let's look at the rest of the\
motives.

```{r, warning=FALSE, echo=FALSE}
ggplot(data = subset(shootings, shootings$Possible.Motive...General
                     !='Unknown')) +
  geom_histogram(stat="Count", aes(x = Possible.Motive...General), 
                 fill = 'orange', binwidth = 1) +
  theme(axis.text.x = element_text(angle = 35, hjust = 1))
```

Four distinct categories are dominant, excluding multiple motives. The multiple\
motives category includes more than one of the categories. For example, one\
such shooting involved a young man that was bullied (social dispute) and had a \
history of mental illness. Most of those labeled multiple motives are a \
combination of these top four motives.

1. Mental illness
2. Domestic dispute
3. Social dispute
4. Terminated/Denied/Reprimanded.

### Targeted Victims

Who do shooters target?

```{r, warning=FALSE, echo=FALSE}
ggplot(data = shootings) +
  geom_histogram(stat="Count", aes(x = Targeted.Victim.s...General), 
                 fill = 'orange', binwidth = 1) +
  theme(axis.text.x = element_text(angle = 35, hjust = 1))
```

Four types of victims stand above the rest:

1. School associates (Students/Classmates/Teacher)
2. Family
3. Business associates (Colleague/Workmate/Business acquaintance)
4. General public

# Univariate Analysis

### What is the structure of your dataset?

The data is tidy, long-form data where each row is an observation (shooting \
event) and each column is a feature of the observation. There are not many \
missing values for each feature, making this a good dataset from which to \
draw conclusions. 

### What is/are the main feature(s) of interest in your dataset?

The main features of interest are the number of fatalities and the number of \
injuries.

### What other features in the dataset do you think will help support your \
investigation into your feature(s) of interest?

To further understand this dataset, I would like to also look at geographical \
distribution of the shootings and the type of place that shootings occurred.

### Did you create any new variables from existing variables in the dataset?

I did not create any new variables from the dataset, but I did pull in other \
variables I was interested in. I pulled population data from a U.S. Census\
and climate data from NOAA. I was able to match up location to these additional\
by the longitude and latitude (climate) and city and state names (population)\
in my primary dataset.

### Of the features you investigated, were there any unusual distributions? \
Did you perform any operations on the data to tidy, adjust, or change the form \
of the data? If so, why did you do this?

The most striking observation for me was the increase in shooting attacks over\
time. It hints at great public danger for the future. I also found the  \
population distribution interesting. Less populated areas are more prone to \
mass shootings. 

I had to rearrange the data to sum the types of weapons used in shootings. This\
was necessary to arrange the data into a single column that could be easily \
plotted in ggplot2. Other tidying included declaring numeric columns that were\
imported as strings and converting string columns with distinct levels into\
factors.

# Bivariate Plots Section

### Number of Fatalities with Time
The histogram of shootings with time showed that shootings are increasing in\
frequency. Are these shootings becoming more deadly? First, let's look at the\
number of fatalities with time. 

```{r echo=FALSE, message=FALSE}
# Convert date character string to timestamp for plotting
ggplot(data = shootings) +
  geom_point(aes(x = Date, y = Total.Number.of.Fatalities),
             size = 4, alpha = .5, color='red')
```

While there are some shootings within the past decade with higher fatalities\
than previous years, most shootings appear to have similar fatalities with the\
majority having less than 10. 

### Number of Victims with Time

Let's look at the same for number of victims\
(fatalities + injuries). The Las Vegas shooting in October is a serious\
outlier in this metric. To get a better look at other shootings, I will \
remove it. 

```{r echo=FALSE, message=FALSE, VictimsWithTime}
ggplot(data = subset(shootings, CaseID!=327)) +
  geom_point(aes(x = Date, y = Total.Number.of.Victims),
             size = 4, alpha = .5, color = 'red')
```

Plots of Total Victims and Total Fatalities with time look very similar. There\
are some shootings within the last 5 to 10 years with higher numbers. However,\
it is not clear if there is a definite trend of mass shootings towards \
increasing severity. It is possible that recent shootings are outliers that \
do not reflect future activity. We can only hope this is the case. 

### Geographic Distribution of Fatalities

Let's see how fatalities are distributed across the country. 

```{r echo=FALSE, message=FALSE, DistributionFatalities}
map<-get_map(location='united states', zoom=4, maptype = "terrain",
             source='google',color='color')
ggmap(map) + geom_point(
        aes(x=Longitude, y=Latitude, size=Total.Number.of.Fatalities 
            ), 
        data=shootings, alpha=.5, na.rm = T, color='red') + 
  theme(axis.text.x = element_blank(), axis.text.y = element_blank()) +
  xlab("") + ylab("") 
```

Shootings are concentrated mostly along the Eastern United States and the West\
Coast with a cluster in Southeast Texas. Fewer shootings have occured in the\
Midwest and Central United States and may be influenced by more sparse \
population density in these regions. Still, shootings have occured in remote\
areas. There are several regions with dense clusters, indicating a large number\
of fatalities. Some that stand out to me are  Southern California, Southern\
Florida, Southeast Texas, and Northwest Washington. 

### Place Type and Shooter Age

What type of places do these shootings occur at? Are different types of places\
targeted by different age demograpics? Let's look at this with a box plot.

```{r echo=FALSE, message=FALSE, PlaceTypeShooterAge}
ggplot(data = shootings) +
  geom_boxplot(aes(x = Place.Type, 
                   y = Average.Shooter.Age)) +
  theme(axis.text.x = element_text(angle = 35, hjust = 1))
```

Places of work (Company/Factory/Office) and government facilities are targeted\
by older shooters while primary and secondary schools are targeted by younger \
shooters. Retail facilities and residential neighborhoods have the broadest \
distributions and are targeted by shooters of all ages.

### Motive and age

Do shooters of different ages have different motives? 

```{r echo=FALSE, Motive_Age}
ggplot(data = shootings) +
  geom_boxplot(aes(x = Possible.Motive...General, y = Average.Shooter.Age)) +
  theme(axis.text.x = element_text(angle = 35, hjust = 1))
```

The youngest shooters were motivated by expulsion or suspension from school and\
harassment. Other motivations for younger shooters include drug use, financial\
difficulties, race, and social disputes. Older shooters were motivated most by\
financial and legal disputes. 

### Age Group and Total Victims

The previous two plots lead to the question what is the most deadly age group?

```{r echo=FALSE, message=FALSE, AgeGroupTotalVictims}
# Create discrete age ranges for easier viewing on stacked bar chart. A 
# continuous scale is too difficult to distinguish age groups.
shootings$Shooter.Age.Bucket <- cut(shootings$Average.Shooter.Age, 
                                    breaks = c(10, 20, 30, 40, 50, 60, 70))
ggplot(data=shootings) + 
  geom_col(aes(x = Shooter.Age.Bucket, y = Total.Number.of.Victims), 
           fill = "orange") +
  theme(axis.text.x = element_text(angle = 35, hjust = 1))
```

Shooters ages 20 to 30 are the most deadly. 

### Motive and Total Victims

Are some motives more deadly than others?

```{r echo=FALSE, MotivesTotalVictims}
ggplot(data = subset(shootings, CaseID!=327)) +
  geom_col(aes(x = Possible.Motive...General, y = Total.Number.of.Victims), 
           fill = "orange") +
  theme(axis.text.x = element_text(angle = 35, hjust = 1))
```

Previously, it was observed that domestic and social disputes were behind as \
many shootings as mental illness. However, we see that mental illness is \
responsible for as many victims as both of these combined. The top most deadly\
known individual motives are:

1. Mental illness
2. Social dispute
3. Political/Religious ideals
4. Terminated/Denied/Reprimanded
5. Domestic dispute
6. Race
7. Financial dispute

# Bivariate Analysis

### Talk about some of the relationships you observed in this part of the \
investigation. How did the feature(s) of interest vary with other features in \
the dataset?

The number of fatalities and victims for each shooting has remained mostly\
constant over time. There are some shootings in the past 5 to 10 years that \
have an unusually high number of victims. It is possible that recent shooters\
attempt to outdo previous shooters. More data would be required to confirm such\
a trend. This is data I hope we won't collect.

Geographic distribution is fairly spread out. No place in the United States\
appears to be immune to this trauma. However, some regions are more prone to \
mass shootings. There is greater conentration towards the East and West Coasts\
with various clusters scattered across the states. Indiana, New Hampshire, \
North Dakota, and Rhode Island are the only states without recorded mass\
shootings in this dataset.

### Did you observe any interesting relationships between the other features \
(not the main feature(s) of interest)?

Shooters have various motives and some of these tend to be dependent on age. In\
some cases, different age groups also tend to target different types of places.\
This is likely due to the type of facilities they often interact with and may \
be related to their motives. For example, young shooters targeted primary and\
secondary schools and one major motivation for this group was expulsion or\
suspension from these institutions. 

### What was the strongest relationship you found?

This is a complex societal issue with many contributing factors, many of which\
probably are not captured in this dataset. No strong relationships were found\
and this is probably not likely for any small set of variables given the \
complexity of the problem. However, some variables explain certain aspects. The\
relationship of age with motive as well as how deadly different motives are \
were the most impressive to me. Perhaps a greater understanding of the shooters\
themselves would help shed light on unfathomable phenomenon and maybe even lead\
to more effective prevention measures. 

# Multivariate Plots Section

In this section I will explore more about the shooters and their motives.

### Geographical Distribution of Motives

We've established the top 7 deadliest motives overall. Do these cluseter in \
certain regions? 

```{r echo=FALSE, message=FALSE, GeographyMotives}
# Select the top most deadly motives
Top.Motives <- c("Mental illness", "Social dispute", 
                 "Political/Religious ideals", "Terminated/Denied/Reprimanded",
                 "Domestic dispute", "Race", "Financial dispute")
# Set custom color scale
cbPalette <- c("#000000", "#E69F00", "#56B4E9", "#009E73", "#F0E442", "#0072B2",
               "#D55E00", "#CC79A7")
# Plot map
map<-get_map(location='united states', zoom=4, maptype = "terrain",
             source='google',color='color')
ggmap(map) + geom_point(
        aes(x=Longitude, y=Latitude, size=Total.Number.of.Fatalities,
            color = Possible.Motive...General),
        data=subset(shootings, shootings$Possible.Motive...General %in%
                    Top.Motives), na.rm = T) +
  theme(axis.text.x = element_blank(), axis.text.y = element_blank()) +
  xlab("") + ylab("") +
  scale_colour_manual(values=cbPalette)
```

Every motive is observed from coast to coast. No distinct clusters are apparent\
for any particular motive.

### Total Victims, Motives, and Time

How do the top motives trend with time? Are there motives that are more \
currently of concern? 

```{r echo=FALSE, VictimsMotivesTime}
# Perform a group by to get the total victims for each motive in a given year
VictimsByMotive <- shootings %>% 
  group_by(Possible.Motive...General, Year) %>%
  summarise(Total.Victims = sum(Total.Number.of.Victims))

ggplot(data=subset(VictimsByMotive, VictimsByMotive$Possible.Motive...General
                   %in% Top.Motives)) +
  geom_line(aes(x = Year, y = Total.Victims, 
                color = Possible.Motive...General), size = 2) 
```

Viewing all three variables on a single plot is far too busy. Let's break this\
up and look at each motive on a separate plot. 

```{r echo=FALSE, VictimsMotivesTimeFacet}
# Perform a group by to get the total victims for each motive in a given year
VictimsByMotive <- shootings %>% 
  group_by(Possible.Motive...General, Year) %>%
  summarise(Total.Victims = sum(Total.Number.of.Victims))

ggplot(data=subset(VictimsByMotive, VictimsByMotive$Possible.Motive...General
                   %in% Top.Motives)) +
  geom_line(aes(x = Year, y = Total.Victims, 
                color = Possible.Motive...General), size = 2) +
  facet_wrap(~Possible.Motive...General) +
  theme(axis.text.x = element_text(angle = 35, hjust = 1))
```

Mental illness has been a dominant factor throughout the years. Domestic \
and social disputes as well as political and religious ideals have seen an \
increase in the last 10 years. Financial disputes and employment issues such as\
termination have remained a constant influence. At least one encouraging aspect\
is that race has had a diminishing influence on mass shootings. 

### Total Victims, Motives, and Shooter Age

Previously we identified the top most deadly motives. We saw certain motives \
are associated with age. How do different age groups associate with these \
motives?  

```{r echo=FALSE, message=FALSE, VictimMotivesAge}
# Create discrete age ranges for easier viewing on stacked bar chart. A 
# continuous scale is too difficult to distinguish age groups.
shootings$Shooter.Age.Bucket <- cut(shootings$Average.Shooter.Age, 
                                    breaks = c(10, 20, 30, 40, 50, 60, 70))
# Only include those with known ages and only the top deadliest motives.
plot_data <- subset(shootings, (!is.na(shootings$Shooter.Age.Bucket)) &
                      (shootings$Possible.Motive...General %in% Top.Motives))

ggplot(data=plot_data) + 
  geom_col(aes(x = Possible.Motive...General,
               y = Total.Number.of.Victims,
               fill = Shooter.Age.Bucket)) + 
  theme(axis.text.x = element_text(angle = 35, hjust = 1))
```

It appears that shooters ages 10 to 30 dominate the motives from mental illness\
to social disputes. This should be considered high risk age groups for these \
particular challenges and require special attention. 

### Total Victims, Type of Victim and Shooter Age

Who do the different age groups of shooters target? Which are the most targeted\
types of victims?

```{r echo=FALSE, message=FALSE, VictimTypeAge}
# Only include those with known ages and only the top deadliest motives.
plot_data <- subset(shootings, (!is.na(shootings$Shooter.Age.Bucket)) &
                      (shootings$Possible.Motive...General %in% Top.Motives))

ggplot(data=plot_data) + 
  geom_col(aes(x = Targeted.Victim.s...General,
               y = Total.Number.of.Victims,
               fill = Shooter.Age.Bucket)) + 
  theme(axis.text.x = element_text(angle = 35, hjust = 1))
```

It appears that places of work and school are where shooters target most. \
Creating awareness of possible violence in these settings could help reduce \
future shootings.

# Multivariate Analysis

### Talk about some of the relationships you observed in this part of the \
investigation. Were there features that strengthened each other in terms of \
looking at your feature(s) of interest?

In this section, I explored the motives of shooters. Specifically I looked at\
whether motives were influenced by geographical location and whether they have\
changed with time. Looking at the time component for each motive revealed that\
some have changed over time and certain motives of special concern in recent\
history.  I also looked at what challenges motivate different age groups and\
what types of victims they target. Looking at all these variables together\
showed that certain age groups are at a higher risk for responding violently to\
certain challenges and that they tend to target particular relationships. 

### Were there any interesting or surprising interactions between features?

I was surprised to see that race is a declining factor in mass shootings. It \
was revealing to see that a younger demographic is more susceptible to violence\
with mental illness. This is a problem that requires speciall attention. Also,\
conflict at work and school can have serious consequences. Careful measures\
should be implemented to facilitate peaceful conflict resolution in these\
settings.

------

# Final Plots and Summary

### Plot One
```{r echo=FALSE, Plot_One}
ggplot(data = shootings) +
  geom_histogram(aes(x = Year), fill = 'orange', binwidth = 1) +
  scale_x_continuous(breaks = seq(1965, 2015, 5)) +
  ggtitle("Number of Mass Shootings by Year") +
  ylab("Number of Shootings")
```

### Description One

This was the first plot in the investigation that alerted me to the serious\
problem we have with mass shootings. These are increasing in frequency and pose\
a great threat to public safety. 

### Plot Two
```{r echo=FALSE, Plot_Two}
# Select the motives of greatest current concern.
Current.Motives.of.Concern <- c("Mental illness", "Social dispute", 
                 "Political/Religious ideals", "Domestic dispute")

ggplot(data=subset(VictimsByMotive, VictimsByMotive$Possible.Motive...General
                   %in% Current.Motives.of.Concern)) +
  geom_line(aes(x = Year, y = Total.Victims, 
                color = Possible.Motive...General), size = 2) +
  facet_wrap(~Possible.Motive...General) +
  ggtitle("Dominant or Increasing Motives in Last Decade") +
  ylab("Total Victims") +
  theme(legend.position = "none") # Legend redundant with subplot titles.
```

### Description Two

This plot illustrates the social challenges that are of greatest concern in \
recent history. Four motives were identified that are dominant or increasing \
within the past decade:

1. Domestic dispute. This motive was first recorded as a factor in a 2005 \
shooting. It continued to increase in frequency until it became the most deadly\
motive in 2015.
2. Mental illness. This has continued to be one of the dominant factors behind \
mass shootings. Mental illness has always been a concern in society and appears\
to have had a constant risk of violence.
3. Political and religious ideals. This is also increasing in frequency. It's \
not surprising given the political and religious polarization that is so \
evident in media.
4. Social dispute. This is a general description of conflicts that may occur \
with friends, schoolmates, associates, or strangers. These violent encounters \
are an increasing concern.  

Efforts should be implemented to provide support for those with in these \
challenges in these areas. For example, greater awareness and assistance should\
be provided for those with mental illness, realizing that there is a small \
possibility of violence if it remains unaddressed. It also occurs to me that we\
could do more as a society to reduce polarization. Each group of people sharing\
certain ideals represent a distribution. As harsh rhetoric builds and opponents\
are increasingly demonized, it only reasons that more and more of the tail ends\
of those within ideological groups will be pushed over social barriers and take\
to violence. The more we can unite and work to understand each other's\
differences, we may be able to discourage extreme violent reactions.

### Plot Three
```{r echo=FALSE, Plot_Three}
# Only include those with known ages and only the top deadliest motives.
plot_data <- subset(shootings, (!is.na(shootings$Shooter.Age.Bucket)) &
                      (shootings$Possible.Motive...General %in% Top.Motives))

ggplot(data=plot_data) + 
  geom_col(aes(x = Targeted.Victim.s...General,
               y = Total.Number.of.Victims,
               fill = Shooter.Age.Bucket)) + 
  theme(axis.text.x = element_text(angle = 35, hjust = 1)) +
  ggtitle("Types of Victims Targeted by Shooters") +
  ylab("Total Victims") +
  xlab("Type of Victim") +
  scale_fill_discrete(name="Shooter Age Range")
```

### Description Three

This plot highlights two types of victims that are most frequently targeted. \
Comparing this data to the bar plot of place type reveals that places of work \
and school are common targets. These are two areas where efforts should be \
applied to create awareness of violence and implement prevention measures.

------

# Reflection

Mass shootings are a tragic and highly complex societal issue in the United\
States. While admittedly incomplete, this dataset captures many important\
variables that provide valuable insight. My initial exploration confirmed my\
suspicion that mass shootings are a growing problem with a marked increase in\
recent years. As I dug deeper into the data, these events appeared random and \
it was difficult to find strong correlations in the data. However, I was able \
to find several associations that were intriguing and some that were surprising.

In the beginning of my exploration I was interested in victims, locations,\
weapons, and other other variables external to the shooters. As my work\
progressed, I noticed my focus shift to the shooters themselves. What were \
their motives, their ages, and the relationships they target? I began to see\
associations of motive and shooting location were related to age. I began to \
get a sense that there are certain intense challenges that may drive \
individuals to reacting in violent ways, especially when they have a\
predisposition to certain emotional difficulties. It was found that mental\
illness plays a large factor, but there may be other variables to consider as\
well. 

There are several other variables within this dataset that should be explored.\
For example, a more detailed analysis of gun type could yield valuable\
information that could guide gun control efforts. Are there certain brands that\
are common and do these have distribution networks that could be improved? I \
started to dig into the type of place shootings occured, but I did not explore \
the shooters relationship to location. In addition, collection of more \
variables could be insightful. For example, economic status of the shooter or \
the economics of the area in which the shooter lives. This could help answer if\
there are financial stratas that are more at risk. Also, what about education \
level? It is my belief that if we dig into the details of what drives shooters \
to these atrocities, we can come together to create awareness of risk factors \
and implement prevention measures that will help reduce this terrible burden on\
society.