Racial Bias in Fatal Police Shootings
Presentation:
Hear me talk through the making of this visualization. All code featured in the presentation (and more!) can be found in the sections below.
The Data:
Despite the frequency with which incidents of police brutality occurs in the US, an official centralized record of police violence does not exist. In 2015 The Washington Post took matters into their own hands and created a database of every fatal shooting in the US by a police officer. The record is regularly updated (right now it is current as of June 8, 2020). Currently there are 5401 fatalities represented in the dataset. Here is just a glimpse at the data:
## # A tibble: 5,401 × 12
## name date armed age gender race city state threat_level flee
## <chr> <date> <fct> <dbl> <fct> <fct> <chr> <fct> <fct> <fct>
## 1 Tim Elliot 2015-01-02 gun 53 M A Shel… WA attack Not …
## 2 Lewis Lee… 2015-01-02 gun 47 M W Aloha OR attack Not …
## 3 John Paul… 2015-01-03 unar… 23 M H Wich… KS other Not …
## 4 Matthew H… 2015-01-04 toy … 32 M W San … CA attack Not …
## 5 Michael R… 2015-01-04 nail… 39 M H Evans CO attack Not …
## 6 Kenneth J… 2015-01-04 gun 18 M W Guth… OK attack Not …
## 7 Kenneth A… 2015-01-05 gun 22 M H Chan… AZ attack Car
## 8 Brock Nic… 2015-01-06 gun 35 M W Assa… KS attack Not …
## 9 Autumn St… 2015-01-06 unar… 34 F W Burl… IA other Not …
## 10 Leslie Sa… 2015-01-06 toy … 47 M B Knox… PA attack Not …
## # ℹ 5,391 more rows
## # ℹ 2 more variables: body_camera <lgl>, signs_of_mental_illness <lgl>
While the Washington Post data allows us to see raw number of shootings that occur, if we want to see how Black and White people are differentially targeted by police we are going to need to calculate the proportion of Black/White people that are shot. To do this I need data on the population size of Black and White people living in the US. I got this data from the US Census Bureau.
Explanation of Variables in Washington Post dataset
Qualitative variables: name, armed, gender, race, city, state, signs_of_mental_illness, threat_level, flee, body_camera
Quantitative variables: age
armed: Did the victim have a weapon? If yes, what kind? There are 89 weapons represented in the dataset, ranging from a gun to a chair.
race:
W= White, non-Hispanic (2468)B= Black, non-Hispanic (1291)A= Asian (93)N= Native American (78)H= Hispanic (900)O= Other (48)NA= Unknown (523)
signs_of_mental_illness: Did the victim exhibit signs of mental illness?
TRUE = 1216
FALSE = 4185
threat_level: Was there a direct and immediate threat to the life of the police officer? This includes incidents where officers or others were shot at, threatened with a gun, attacked with other weapons or physical force, etc.
attack = 3487
other = 1914
flee: Was the victim moving away from the officers?
Not fleeing = 3406
Car = 898
Foot = 689
Other = 162
NA = 246
body_camera: Reported as TRUE if news reports indicated an officer was wearing a body camera and it may have recorded at least a portion of the incident.
TRUE = 615
FALSE = 4796
Links to datasets
The Visualization:
## Warning: Using `size` aesthetic for lines was deprecated in ggplot2 3.4.0.
## ℹ Please use `linewidth` instead.
## This warning is displayed once every 8 hours.
## Call `lifecycle::last_lifecycle_warnings()` to see where this warning was
## generated.
Intended Audience
It’s never fun to talk about violence, especially violence present in a system that is supposed to hold honor, however I think that it is important that everyone be made aware of the racial bias that is reflected in statistics describing police brutality. This audience for this visualization is therefore broad: people of all ages, genders, education levels, socioeconomic status, etc. This plot is easily understood even without previous experience with dumbbell plots.
About Dumbbell Plots
Also called connected dot plots or dumbbell dot plots, dumbbell plots are a version of lollipop charts that features comparison between 2 (or 3) groups. Lollipop charts are closely related to bar charts but are only effective in conveying information about a single group. By using the dumbbell layout you can increase the depth of information conveyed. While a grouped bar chart would also convey information about the two groups, dumbbell plots take advantage of the Gestalt principle of continuity to aid the eye in following the directionality of the relationship.
How to Read it and What to Look For
Dumbbell plots consist of 2 (or 3) points connected by a line. Often there are multiple “dumbbells” to represent different groups, timepoints, etc. The points indicate the numerical (or categorical) value for a group. The line connecting two points exists to indicate the relationship between the two points, both in directionality and magnitude. It also functions to guide the eye in appropriate grouping of points. In my visualization I am trying to convey the relationship between Black and White victims of fatal shootings. The x-axis carries information about the number of victims. The y-axis represents time, in years. We can then follow the incidence of shootings across time for both races.
Representation Description/Intended Message
My goal for this visualization was to illustrate how Black and White populations are targetd differently by gun violence, specifically gun violence in the contect of fatal line-of-duty police shootings. To do this, I cannot present the raw number of shootings in each population becuase the population sizes are vastly different (Black people are a minority, after all). Instead, I’m showing the number of fatal shootings per 1 million people of a given race (ie. the proportion of each population that dies due to gun violence). Below is a table of those numbers:
| Year | White | Black | White | Black | White | Black |
|---|---|---|---|---|---|---|
| 2015 | 497.00 | 258.00 | 247.78 | 42.63 | 2.01 | 6.05 |
| 2016 | 468.00 | 234.00 | 248.50 | 43.00 | 1.88 | 5.44 |
| 2017 | 459.00 | 224.00 | 249.62 | 43.50 | 1.84 | 5.15 |
| 2018 | 454.00 | 229.00 | 250.14 | 43.80 | 1.81 | 5.23 |
| 2019 | 405.00 | 249.00 | 249.42 | 43.43 | 1.62 | 5.73 |
| 2020 | 424.69 | 222.67 | 249.73 | 43.58 | 1.70 | 5.11 |
| * 2020 data is projected based on current data. | ||||||
| † Data is current as of June 8, 2020. |
From both the table and the visualization it is clear that Black people are consistently disproportionately targeted by gun violence in this context. This holds up over all years for which there is data available.
Presentation
I chose to make this plot in portrait orientation (vs landscape) because I thought it made the relationships more readily interpretable. I removed major and minor grids on the y-axis since I found them distracting and the segment connecting the two points already provides the same information. While years are technically quantitative, the way I’m using them makes them almost categorical and thus it is less important to have grid lines. I did keep grid lines for the x-axis to help interpretation of the value of the points. The color choices were deliberate and were chosen to represent skin color. A text and arrow annotation was used to indicate that the 2020 data is projected based on current data. A footnote was used to indicate how current the data is. I positioned the legend under the title such that it is readily available but doesn’t occupy too much space. For the title I used a bold-face font for the title and regular font for the subtitle. I found that by using that bold font I was able to better visually separate the title and subtitle.
How I created it
**1. Import and wrangle the shooting data. **
shootings_raw <- read_csv("fatal-police-shootings-data.csv")
glimpse(shootings_raw)
## Rows: 5,401
## Columns: 14
## $ id <dbl> 3, 4, 5, 8, 9, 11, 13, 15, 16, 17, 19, 21, 22,…
## $ name <chr> "Tim Elliot", "Lewis Lee Lembke", "John Paul Q…
## $ date <date> 2015-01-02, 2015-01-02, 2015-01-03, 2015-01-0…
## $ manner_of_death <chr> "shot", "shot", "shot and Tasered", "shot", "s…
## $ armed <chr> "gun", "gun", "unarmed", "toy weapon", "nail g…
## $ age <dbl> 53, 47, 23, 32, 39, 18, 22, 35, 34, 47, 25, 31…
## $ gender <chr> "M", "M", "M", "M", "M", "M", "M", "M", "F", "…
## $ race <chr> "A", "W", "H", "W", "H", "W", "H", "W", "W", "…
## $ city <chr> "Shelton", "Aloha", "Wichita", "San Francisco"…
## $ state <chr> "WA", "OR", "KS", "CA", "CO", "OK", "AZ", "KS"…
## $ signs_of_mental_illness <lgl> TRUE, FALSE, FALSE, TRUE, FALSE, FALSE, FALSE,…
## $ threat_level <chr> "attack", "attack", "other", "attack", "attack…
## $ flee <chr> "Not fleeing", "Not fleeing", "Not fleeing", "…
## $ body_camera <lgl> FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALS…
If we take a quick look at how it was imported, you can see that a lot of the variables were imported as character class when they should operate as factor class. I’ll use forcats::as_factor() to turn those variables to factor class. I also need to create a new “Year” column since I am interested in the number of shootings per year.
shootings_raw <- shootings_raw %>%
mutate(manner_of_death = as_factor(manner_of_death),
armed = as_factor(armed),
gender = as_factor(gender),
race = as_factor(race),
state = as_factor(state),
threat_level = as_factor(threat_level),
flee = as_factor(flee))
# Extract "Year" from "date"
shootings_raw$Year <- as.numeric(format(shootings_raw$date, '%Y'))
Now I’m going to use summarise() to calculate the number of shootings per year for each race. Then I’m going to filter the data to only look at data for Black and White victims.
# Count W/B shootings per year
shootings <- shootings_raw %>%
group_by(Year, race) %>%
summarise(fatalshootings = n()) %>%
filter(race == "W" | race == "B")
## `summarise()` has grouped output by 'Year'. You can override using the
## `.groups` argument.
shootings %>% head(12)
## # A tibble: 12 × 3
## # Groups: Year [6]
## Year race fatalshootings
## <dbl> <fct> <int>
## 1 2015 W 497
## 2 2015 B 258
## 3 2016 W 468
## 4 2016 B 234
## 5 2017 W 459
## 6 2017 B 224
## 7 2018 W 454
## 8 2018 B 229
## 9 2019 W 405
## 10 2019 B 249
## 11 2020 W 185
## 12 2020 B 97
In order to combine this with my census data, I’m going to need to pivot the data wider such that each row represents a single year.
shootings <- shootings %>%
pivot_wider(names_from = race, values_from = fatalshootings)
shootings
## # A tibble: 6 × 3
## # Groups: Year [6]
## Year W B
## <dbl> <int> <int>
## 1 2015 497 258
## 2 2016 468 234
## 3 2017 459 224
## 4 2018 454 229
## 5 2019 405 249
## 6 2020 185 97
2. Import and wrangle the census data.
# How many B/W people are in America from 2015-2020?
census_raw <- read_xlsx("census-data.xlsx")
# Convert units from millions
census <- census_raw %>%
mutate(W_pop = White * 1000000,
B_pop = Black * 1000000)
census
## # A tibble: 21 × 5
## Year White Black W_pop B_pop
## <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 2020 250. 43.6 249726667. 43577778.
## 2 2019 249. 43.4 249420000 43433333.
## 3 2018 250. 43.8 250140000 43800000
## 4 2017 250. 43.5 249620000 43500000
## 5 2016 248. 43 248500000 43000000
## 6 2015 248. 42.6 247780000 42630000
## 7 2014 247. 42.2 246660000 42160000
## 8 2013 246. 41.7 245590000 41710000
## 9 2012 245. 41.3 244510000 41260000
## 10 2011 243. 40.8 243380000 40810000
## # ℹ 11 more rows
3. Combine the census data with the shooting data.
# Combine shooting & census data
shootings <- inner_join(shootings, census, by="Year") %>%
rename(W_mil = White,
B_mil = Black) %>%
# Calculate the number of shootings per 1 million people
mutate(W_per = W*1000000/W_pop,
B_per = B*1000000/B_pop)
4. For 2020: calculate the projected numbers. I’ll do this by (1) calculating the number of days represented in the data and (2) dividing my values by the number of days and multiply by 365.
# Calculate projected data for 2020
# How many days of data is represented for 2020 data? Latest date is 06/08/2020
days <- as.numeric(as.Date(as.character("2020/06/08"), format="%Y/%m/%d")-as.Date(as.character("2020/01/01"), format="%Y/%m/%d"))
# Calculate projected shootings for 2020
proj2020 <- shootings[6,]
proj2020 <- tibble(Year = "2020 proj",
W = pull(proj2020[1,2])*365/days,
B = pull(proj2020[1,3])*365/days,
W_mil = pull(proj2020[1,4]),
B_mil = pull(proj2020[1,5]),
W_pop = pull(proj2020[1,6]),
B_pop = pull(proj2020[1,7]),
W_per = pull(proj2020[1,8])*365/days,
B_per = pull(proj2020[1,9])*365/days)
proj2020
## # A tibble: 1 × 9
## Year W B W_mil B_mil W_pop B_pop W_per B_per
## <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 2020 proj 425. 223. 250. 43.6 249726667. 43577778. 1.70 5.11
# Add projected values to shootings data frame
shootings <- shootings %>% ungroup() %>% mutate(Year = as.character(Year))
shootings <- bind_rows(shootings, proj2020)
shootings
## # A tibble: 7 × 9
## Year W B W_mil B_mil W_pop B_pop W_per B_per
## <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 2015 497 258 248. 42.6 247780000 42630000 2.01 6.05
## 2 2016 468 234 248. 43 248500000 43000000 1.88 5.44
## 3 2017 459 224 250. 43.5 249620000 43500000 1.84 5.15
## 4 2018 454 229 250. 43.8 250140000 43800000 1.81 5.23
## 5 2019 405 249 249. 43.4 249420000 43433333. 1.62 5.73
## 6 2020 185 97 250. 43.6 249726667. 43577778. 0.741 2.23
## 7 2020 proj 425. 223. 250. 43.6 249726667. 43577778. 1.70 5.11
5. Time to get the data into the format we need for ggplot. First, I need to pivot the data longer. Then I can get rid of the 2020 data and replace it with the projected data.
# Pivot long
shootings_long <- shootings %>%
pivot_longer(W_per:B_per, names_to = "race", values_to = "fatal_per")
# Get rid of 2020, replace it with the projected numbers
shootings <- shootings %>% filter(Year != "2020")
shootings[6,1] = "2020"
shootings_long <- shootings_long %>% filter(Year != "2020")
shootings_long[11,1] = "2020"; shootings_long[12,1] = "2020"
shootings_long
## # A tibble: 12 × 9
## Year W B W_mil B_mil W_pop B_pop race fatal_per
## <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <chr> <dbl>
## 1 2015 497 258 248. 42.6 247780000 42630000 W_per 2.01
## 2 2015 497 258 248. 42.6 247780000 42630000 B_per 6.05
## 3 2016 468 234 248. 43 248500000 43000000 W_per 1.88
## 4 2016 468 234 248. 43 248500000 43000000 B_per 5.44
## 5 2017 459 224 250. 43.5 249620000 43500000 W_per 1.84
## 6 2017 459 224 250. 43.5 249620000 43500000 B_per 5.15
## 7 2018 454 229 250. 43.8 250140000 43800000 W_per 1.81
## 8 2018 454 229 250. 43.8 250140000 43800000 B_per 5.23
## 9 2019 405 249 249. 43.4 249420000 43433333. W_per 1.62
## 10 2019 405 249 249. 43.4 249420000 43433333. B_per 5.73
## 11 2020 425. 223. 250. 43.6 249726667. 43577778. W_per 1.70
## 12 2020 425. 223. 250. 43.6 249726667. 43577778. B_per 5.11
6. Build the plot in ggplot. A dumbbell plot can be made in ggplot using a combination of geom_segment() and geom_point(). Let’s look at the most basic version. Note that I’m using data from shootings for geom_segment() and data from shootings_long for geom_point(). Note that I’m already implementing some customizations. I’ve specified the size of line to use for the segment. I’ve also specified the size, shape, and outline color for the points.
ggplot() +
geom_segment(data = shootings,
mapping = aes(x=W_per, xend=B_per, y=Year, yend=Year),
size = 1) +
geom_point(data = shootings_long,
mapping = aes(x = fatal_per, y = Year, fill = race),
size=5, shape = 21, color = "black")
There’s a lot that I want to change:
- I want the vizualization to be in “portrait” orientation. I’ll specify the dimensions in my code chunk options within the chunk header using
fig.width=xandfig.height=y. (Where x and y are numbers, in inches.) - The x-axis should start at 0. I’m also going to expand on the right side so that there’s roughly equal padding on both sides of the dumbbells. For this I’ll use
coord_cartesian(). - I want to customize my colors. For this I’ll specify the names and hex codes manually and use
scale_fill_manual()to implement the names and values. - I don’t want to use the default ggplot theme. Instead I’m going to use
theme_minimal()to strip down everything to a lighter palette. Undertheme()I’m also going to add additional customizations, namely the legend size, location, and orientation as well as the title, subtitle, and footnote formatting. I’m also going to remove the y-axis gridlines. - To make it very obvious that the 2020 data is projected data, I’m adding a text annotation (
annotate(geom = "text")) and an arrow annotation (annotate(geom = "curve")). - Finally, I will specify my title, subtitle, footnote, legend title, and axis labels using
labs().
See the final code and visualization:
mycolors <- c("W_per" = "#fed2b7", "B_per" = "#55160d")
ggplot() +
geom_segment(data = shootings,
mapping = aes(x=W_per, xend=B_per, y=Year, yend=Year),
size = 1) +
geom_point(data = shootings_long,
mapping = aes(x = fatal_per, y = Year, group = race, fill = race),
size=5, shape = 21, color = "black") +
# Customize appearance
coord_cartesian(xlim = c(0,8)) +
scale_fill_manual(values = mycolors,
labels = c("Black","White")) +
scale_y_discrete(expand = c(0.1,0,0,1)) + # Expand margins on top and bottom of plot
theme_minimal() +
theme(legend.position = c(0.115,0.98),
legend.background = element_rect(fill = "white", color = "white"),
legend.text = element_text(size = 10),
legend.title = element_text(size = 10),
legend.direction = "horizontal",
plot.title = element_text(face = "bold", size = 15, hjust=0.2),
plot.title.position = "plot",
plot.subtitle = element_text(hjust = 0.1),
plot.caption = element_text(hjust = 0),
panel.grid.major.y = element_blank()) +
# Arrow annotation to projected data
annotate(geom = "curve", size = 1, color = "black",
x = 6.5, y = 5.8, xend = 5.35, yend = 6.1, curvature = 0.7,
arrow = arrow(length = unit(2.5, "mm"))) +
# Text annotation to projected data
annotate(geom = "text", x = 6.5, y = 5.63,
label = "projected based on\ncurrent data", color = "black",
size = 3.5, lineheight = 0.8, hjust = 0.5) +
# Customize labels
labs(title = "Police Shooting Fatalities by Victim Race",
subtitle = "Racial bias is evident in the police system. Black people \nare disproportionately victimized and murdered by police.",
fill = "Race",
y = "Year",
x = "Number of Fatal Shootings\n(per 1,000,000 people of indicated race)",
caption = "*Data current as of June 8, 2020")
Jennifer Jahncke
Postdoctoral Scholar
I am a neuroscientist studying how brain cells form and maintain connections.