crimes_and_police_violence.R

library(dplyr)
library(scales)
library(ggplot2)

shot <- read.csv("https://raw.githubusercontent.com/washingtonpost/data-police-shootings/master/fatal-police-shootings-data.csv",
                 stringsAsFactors = F)
# https://www.fbi.gov/about-us/cjis/ucr/crime-in-the-u.s/2014/crime-in-the-u.s.-2014/tables/table-8/Table_8_Offenses_Known_to_Law_Enforcement_by_State_by_City_2014.xls
crimes <- read.csv("us_crimes_2015.csv", stringsAsFactors = F)
head(shot)
head(crimes)

shot <- filter(shot, city != "Sand City") # one crazy outlier with population 324

shot.agg <- shot %>%
  group_by(state, city) %>%
  summarise(police_shootings = n()) %>%
  left_join(crimes) %>%
  mutate(police_shootings = police_shootings / population,
         violent = violent / population,
         property = property / population)

formatter <- comma_format(drop0trailing=T)

# Only cities larger than 10,000 (small population makes per-capita rates erratic)
ggplot(filter(shot.agg, population > 10000)) + 
  geom_point(aes(x=violent * 10000, y=police_shootings*10000, size=population),
             alpha=0.4) +
  scale_size(labels=formatter, range=c(0.5, 10), breaks=10^seq(1, 8), name="Poplulation") +
  geom_smooth(aes(x=violent * 10000, y=police_shootings*10000),
              method=lm) +
  xlab("Violent crimes") +
  ylab("People killed by police") +
  ggtitle("People killed by police vs. violent crime rate (per 10,000)")


# All cities
ggplot(shot.agg) + 
  geom_point(aes(x=violent * 10000, y=police_shootings*10000, size=population),
             alpha=0.4) +
  scale_size(labels=formatter, range=c(0.5, 10), breaks=10^seq(1, 8), name="Poplulation") +
  geom_smooth(aes(x=violent * 10000, y=police_shootings*10000),
              method=lm) +
  xlab("Violent crimes") +
  ylab("People killed by police") +
  ggtitle("People killed by police vs. violent crime rate (per 10,000)")

us_crimes_2015.csv