On average, police in the United States shoot and kill more than 1,000 people every year, according to an ongoing analysis by The Washington Post.
-
-
The Washington Post Fatal Force Database
+
+
Proposal
+
We propose a case study to explore the relationship between police residence and fatal police shootings, employing advanced data science methodologies. Focusing on officers residing in the cities they serve, our project aims to uncover insights and patterns that contribute to a nuanced understanding of this complex issue.
+
+
Objectives:
+
+
Investigate the correlation between police residence and fatal police shootings.
+
Utilize a comprehensive dataset spanning 2015 to 2023, focusing on police agencies involved in at least one fatal shooting.
+
Apply advanced statistical methods and machine learning techniques to identify patterns and potential biases.
+
Examine disparities in incident rates based on officers’ residency status, considering demographic, socioeconomic, and policing variables.
+
+
Methodology:
+
+
Data Collection:
+
+
Compile a dataset comprising information on police agencies involved in fatal police shootings.
+
Compile a data set of census variables such as officer residency, race, community demographics, and departmental policies.
+
+
Analysis:
+
+
Employ advanced statistical methods and machine learning techniques to discern patterns and correlations.
+
Conduct a comprehensive exploration of variables influencing fatal police shootings.
+
+
+
Hypothesis and Expected Outcomes:
+
We will conduct two hypothesis tests to analyze both;
+
+
the nominal relationship between an increasing proportion of in-city officer residency and number of fatal police shooting deaths and
+
the categorical difference in fatal police shooting deaths between cities where a majority or or minority of police officers live in the city.
+
+
+
Inference for a Difference in Proportions
+
+
\(H_0\): The mean total number of fatal shootings per agencies does not differ based on if a majority of the officers live in the city or not.
+
\(H_A\): The mean total number of fatal shootings per agencies is fewer in cities where a majority of the officers live in the city then cities where they do not.
\(H_O\): There is no relationship between percentage of the total police force that lives in the city they serve and number of fatal shootings.
+
\(H_A\): There is a relationship between percentage of the total police force that lives in the city they serve and number of fatal shootings.
+
+
\(H_0 : \rho = 0\)
+
\(H_0 : \rho \neq 0\)
+
+
+
+
+
+
+
The Washington Post Fatal Force Database
In 2015, The Washington Post began tracking details about each police-involved killing in the United States — the race of the deceased, the circumstances of the shooting, whether the person was armed and whether the person was experiencing a mental-health crisis — by manually culling local news reports, collecting information from law enforcement websites and social media, and monitoring independent databases such as Fatal Encounters and the now-defunct Killed by Police project. In many cases, The Post conducts additional reporting.
In 2022, The Post updated its database to standardize and publish the names of the police agencies involved in each shooting to better measure accountability at the department level.
The 2014 killing of Michael Brown in Ferguson, Mo. began a protest movement culminating in the Black Lives Matter movement and an increased focus on police accountability nationwide. In this data set, The Post tracks only shootings with circumstances closely paralleling those like the killing of Brown — incidents in which a police officer, in the line of duty, shoots and kills a civilian. The Post is not tracking deaths of people in police custody, fatal shootings by off-duty officers or non-shooting deaths in this data set.
diff --git a/_site/about.html b/_site/codebook.html
similarity index 73%
rename from _site/about.html
rename to _site/codebook.html
index c1b9615..2897ab9 100644
--- a/_site/about.html
+++ b/_site/codebook.html
@@ -7,7 +7,7 @@
-Police Shootings - About
+Police Shootings - Codebook
@@ -117,6 +83,10 @@
Percentage of the total police force that lives in the city
+
+
+
white
+
Percentage of white (non-Hispanic) police officers who live in the city
+
+
+
non-white
+
Percentage of non-white police officers who live in the city
+
+
+
black
+
Percentage of black police officers who live in the city
+
+
+
hispanic
+
Percentage of Hispanic police officers who live in the city
+
+
+
asian
+
Percentage of Asian police officers who live in the city
+
+
+
+
+
Incident Information
+
+
+
+
+
+
+
+
Name
+
Description
+
+
+
+
+
id
+
A unique identifier for each fatal police shooting incident.
+
+
+
date
+
The date of the fatal shooting.
+
+
+
body_camera
+
Whether news reports have indicated an officer was wearing a body camera and it may have recorded some portion of the incident.
+
+
+
city
+
The municipality where the fatal shooting took place
+
+
+
county
+
County where the fatal shooting took place.
+
+
+
state
+
The two-letter postal code abbreviation for the state in which the fatal shooting took place.
+
+
+
latitude
+
The latitude location of the shooting expressed as WGS84 coordinates, geocoded from addresses. Please note that the precision and accuracy of incident coordinates varies depending on the precision of the input address which is often only available at the block level.
+
+
+
longitude
+
The longitude location of the shooting expressed as WGS84 coordinates, geocoded from addresses.
##Tidying Data
-
-#creating dfs from .csv files
-police_locals <-read_csv("data/police-locals.csv")
-
-
Rows: 75 Columns: 10
-── Column specification ────────────────────────────────────────────────────────
-Delimiter: ","
-chr (6): city_old, city, state, black, hispanic, asian
-dbl (4): police_force_size, all, white, non-white
-
-ℹ Use `spec()` to retrieve the full column specification for this data.
-ℹ Specify the column types or set `show_col_types = FALSE` to quiet this message.
Rows: 3422 Columns: 6
-── Column specification ────────────────────────────────────────────────────────
-Delimiter: ","
-chr (4): name, type, state, oricodes
-dbl (2): id, total_shootings
-
-ℹ Use `spec()` to retrieve the full column specification for this data.
-ℹ Specify the column types or set `show_col_types = FALSE` to quiet this message.
Rows: 9129 Columns: 19
-── Column specification ────────────────────────────────────────────────────────
-Delimiter: ","
-chr (12): threat_type, flee_status, armed_with, city, county, state, locati...
-dbl (4): id, latitude, longitude, age
-lgl (2): was_mental_illness_related, body_camera
-date (1): date
-
-ℹ Use `spec()` to retrieve the full column specification for this data.
-ℹ Specify the column types or set `show_col_types = FALSE` to quiet this message.
-
-
#removing old `city` tag from data set that we created when decatenated the city names
-police_locals <- police_locals |>
-select(-city_old)
-
-# creating `agencies` df with just police departments
-agencies <- agencies |>
-filter(grepl("department", tolower(name))) |>
-filter(!grepl("county", tolower(name)))
-
-#creating binned categorical account of if shooting victim was `armed`
-shootings <- shootings |>
-mutate(armed =case_when(is.na(armed_with) ~"NO",
- armed_with =="unarmed"~"NO",
- armed_with =="unknown"~"NO",
- armed_with =="undetermined"~"NO",
- armed_with =="gun"~"YES",
- armed_with =="knife"~"YES",
- armed_with =="blunt_object"~"YES",
- armed_with =="other"~"YES",
- armed_with =="replica"~"YES",
- armed_with =="vehicle"~"YES"))
-
-#creating df with only agency `names`, `id`, and `state`
-agencies_ids <- agencies |>
-select(name, id, state)
-agencies_ids
-
-
# A tibble: 2,057 × 3
- name id state
- <chr> <dbl> <chr>
- 1 Aberdeen Police Department 2576 WA
- 2 Abilene Police Department 2114 TX
- 3 Abington Township Police Department 2088 PA
- 4 Acworth Police Department 3375 GA
- 5 Ada Police Department 2579 OK
- 6 Adel Police Department 3107 GA
- 7 Akron Police Department 815 OH
- 8 Alamogordo Police Department 1434 NM
- 9 Alamosa Police Department 2354 CO
-10 Albany Police Department 1443 GA
-# ℹ 2,047 more rows
-
-
#creating df with `city`, `agency`, and `state` info for each shooting
-shooting_agencies <- shootings |>
-select(city, agency_ids, state)
-shooting_agencies
-
-
# A tibble: 9,129 × 3
- city agency_ids state
- <chr> <chr> <chr>
- 1 Shelton 73 WA
- 2 Aloha 70 OR
- 3 Wichita 238 KS
- 4 San Francisco 196 CA
- 5 Evans 473 CO
- 6 Guthrie 101 OK
- 7 Chandler 195 AZ
- 8 Assaria 490 KS
- 9 Burlington 287 IA
-10 Knoxville 26254 PA
-# ℹ 9,119 more rows
-
-
#changing `shooting` var in `shooting_agencies` df to numeric
-shooting_agencies$agency_ids <-as.numeric(shootings$agency_ids)
-
-
Warning: NAs introduced by coercion
-
-
#creating df with `city` and `state` info for each agency by joining `agencies_ids` and `shooting_agencies`
-agencies_w_cities <- agencies_ids |>
-left_join(shooting_agencies, by =c("id"="agency_ids", "state"="state")) |>
-drop_na(city) |>
-distinct(id, .keep_all =TRUE)
-agencies_w_cities
-
-
# A tibble: 1,781 × 4
- name id state city
- <chr> <dbl> <chr> <chr>
- 1 Aberdeen Police Department 2576 WA Aberdeen
- 2 Abilene Police Department 2114 TX Abilene
- 3 Abington Township Police Department 2088 PA Abington Township
- 4 Acworth Police Department 3375 GA Acworth
- 5 Ada Police Department 2579 OK Ada
- 6 Adel Police Department 3107 GA Adel
- 7 Akron Police Department 815 OH Akron
- 8 Alamogordo Police Department 1434 NM Alamogordo
- 9 Alamosa Police Department 2354 CO Alamosa
-10 Albany Police Department 1443 GA Albany
-# ℹ 1,771 more rows
-
-
#creating df with census data for each agency by joining `agencies_w_cities` and `police_locals`
-agencies_census <- agencies_w_cities |>
-full_join(police_locals, by =c("city"="city", "state"="state")) |>
-drop_na(police_force_size) |>
-distinct(id, .keep_all =TRUE) |>
-mutate(majority =if_else(all >=0.5, "TRUE", "FALSE"))
-agencies_census
-
-
# A tibble: 109 × 12
- name id state city police_force_size all white `non-white` black
- <chr> <dbl> <chr> <chr> <dbl> <dbl> <dbl> <dbl> <chr>
- 1 Albany P… 2237 NY Alba… 890 0.185 0.160 0.364 **
- 2 Albuquer… 508 NM Albu… 1340 0.616 0.630 0.602 **
- 3 Amtrak P… 1657 IL Chic… 12120 0.875 0.872 0.877 0.89…
- 4 Atlanta … 447 GA Atla… 2950 0.137 0.186 0.111 0.10…
- 5 Austin P… 141 TX Aust… 1985 0.295 0.195 0.427 0.25
- 6 Baltimor… 4784 MD Balt… 2800 0.257 0.133 0.362 0.39…
- 7 Baltimor… 149 MD Balt… 2800 0.257 0.133 0.362 0.39…
- 8 BART Pol… 2015 CA Oakl… 1530 0.0948 0.0267 0.160 0.06…
- 9 Baton Ro… 1098 LA Bato… 980 0.214 0.144 0.321 0.34…
-10 Boston P… 3 MA Bost… 2560 0.477 0.442 0.583 0.68…
-# ℹ 99 more rows
-# ℹ 3 more variables: hispanic <chr>, asian <chr>, majority <chr>
-
-
#creating df of only shootings involving agencies within `agencies` df
-shootings_case <- shootings |>
-right_join(agencies_census, by =c("city"="city", "state"="state")) |>
-select(-agency_ids) |>
-rename(agency_ids = id.y, id = id.x, agency = name.y, victim = name.x) |>
-select(-location_precision, -race_source)
-
-
Warning in right_join(shootings, agencies_census, by = c(city = "city", : Detected an unexpected many-to-many relationship between `x` and `y`.
-ℹ Row 4 of `x` matches multiple rows in `y`.
-ℹ Row 29 of `y` matches multiple rows in `x`.
-ℹ If a many-to-many relationship is expected, set `relationship =
- "many-to-many"` to silence this warning.
-
-
shootings_case
-
-
# A tibble: 3,677 × 27
- id date threat_type flee_status armed_with city county state
- <dbl> <date> <chr> <chr> <chr> <chr> <chr> <chr>
- 1 5 2015-01-03 move not unarmed Wichita Sedgw… KS
- 2 8 2015-01-04 point not replica San Francis… San F… CA
- 3 8 2015-01-04 point not replica San Francis… San F… CA
- 4 22 2015-01-07 threat not knife Columbus Frank… OH
- 5 22 2015-01-07 threat not knife Columbus Frank… OH
- 6 27 2015-01-07 shoot foot gun New Orleans Orlea… LA
- 7 325 2015-01-09 point not gun El Paso El Pa… TX
- 8 46 2015-01-13 shoot foot gun Albuquerque Berna… NM
- 9 46 2015-01-13 shoot foot gun Albuquerque Berna… NM
-10 56 2015-01-15 shoot foot gun Indianapolis Marion IN
-# ℹ 3,667 more rows
-# ℹ 19 more variables: latitude <dbl>, longitude <dbl>, victim <chr>,
-# age <dbl>, gender <chr>, race <chr>, was_mental_illness_related <lgl>,
-# body_camera <lgl>, armed <chr>, agency <chr>, agency_ids <dbl>,
-# police_force_size <dbl>, all <dbl>, white <dbl>, `non-white` <dbl>,
-# black <chr>, hispanic <chr>, asian <chr>, majority <chr>
-
+
##Tidying Data
+
+#creating dfs from .csv files
+police_locals <-read_csv("data/police-locals.csv")
+agencies <-read_csv("data/fatal-police-shootings-agencies.csv")
+shootings <-read_csv("data/fatal-police-shootings-data.csv")
+
+#removing old `city` tag from data set that we created when decatenated the city names
+police_locals <- police_locals |>
+select(-city_old)
+
+# creating `agencies` df with just police departments
+agencies <- agencies |>
+filter(grepl("department", tolower(name))) |>
+filter(!grepl("county", tolower(name)))
+
+#creating binned categorical account of if shooting victim was `armed`
+shootings <- shootings |>
+mutate(armed =case_when(is.na(armed_with) ~"NO",
+ armed_with =="unarmed"~"NO",
+ armed_with =="unknown"~"NO",
+ armed_with =="undetermined"~"NO",
+ armed_with =="gun"~"YES",
+ armed_with =="knife"~"YES",
+ armed_with =="blunt_object"~"YES",
+ armed_with =="other"~"YES",
+ armed_with =="replica"~"YES",
+ armed_with =="vehicle"~"YES"))
+
+#creating df with only agency `names`, `id`, and `state`
+agencies_ids <- agencies |>
+select(name, id, state)
+
+#creating df with `city`, `agency`, and `state` info for each shooting
+shooting_agencies <- shootings |>
+select(city, agency_ids, state)
+
+#changing `shooting` var in `shooting_agencies` df to numeric
+shooting_agencies$agency_ids <-as.numeric(shootings$agency_ids)
+
+#creating df with `city` and `state` info for each agency by joining `agencies_ids` and `shooting_agencies`
+agencies_w_cities <- agencies_ids |>
+left_join(shooting_agencies, by =c("id"="agency_ids", "state"="state")) |>
+drop_na(city) |>
+distinct(id, .keep_all =TRUE)
+
+#creating df with census data for each agency by joining `agencies_w_cities` and `police_locals`
+agencies_census <- agencies_w_cities |>
+full_join(police_locals, by =c("city"="city", "state"="state")) |>
+drop_na(police_force_size) |>
+distinct(id, .keep_all =TRUE) |>
+mutate(majority =if_else(all >=0.5, "TRUE", "FALSE"))
+
+#creating df of only shootings involving agencies within `agencies` df
+shootings_case <- shootings |>
+right_join(agencies_census, by =c("city"="city", "state"="state")) |>
+select(-agency_ids) |>
+rename(agency_ids = id.y, id = id.x, agency = name.y, victim = name.x) |>
+select(-location_precision, -race_source)
# A tibble: 108 × 2
- agency n
- <chr> <int>
- 1 Albany Police Department 1
- 2 Albuquerque Police Department 66
- 3 Amtrak Police Department 54
- 4 Atlanta Police Department 36
- 5 Austin Police Department 40
- 6 BART Police Department 13
- 7 Baltimore City Police Department 30
- 8 Baltimore Police Department 30
- 9 Baton Rouge Police Department 15
-10 Boston Police Department 10
-# ℹ 98 more rows
#count shootings by agency
+shootings_by_agency <- shootings_case |>
+count(agency)
+
+#find top 25 agencies with the most shootings
+top_25_agencies <- shootings_by_agency |>
+slice_max(n, n =25)
+
+# visulize top 25 agencies with the most shootings
+ggplot(data = top_25_agencies,
+mapping =aes(x = agency, y = n)) +
+geom_col() +
+theme(axis.text.x =element_text(angle =75,
+vjust =1,
+hjust =1,
+margin =margin(t =5, b =5)))
-
+
-
#mapping Locations of Police-Involved Shootings between 2015 and 2023
-
-library(ggmap)
-
-
Warning: package 'ggmap' was built under R version 4.3.1
-
-
-
ℹ Google's Terms of Service: <https://mapsplatform.google.com>
- Stadia Maps' Terms of Service: <https://stadiamaps.com/terms-of-service/>
- OpenStreetMap's Tile Usage Policy: <https://operations.osmfoundation.org/policies/tiles/>
-ℹ Please cite ggmap if you use it! Use `citation("ggmap")` for details.
-
-
library(maps)
-
-
Warning: package 'maps' was built under R version 4.3.1
-
-
-
-Attaching package: 'maps'
-
-The following object is masked from 'package:purrr':
-
- map
-
-
library(mapdata)
-
-usa <-map_data("usa")
-states <-map_data("state")
-
-ggplot(data = states) +
-geom_polygon(aes(x = long, y = lat, fill = group, group = group), color ="white") +
-coord_fixed(1.3) +
-guides(fill=FALSE) +# do this to leave off the color legend
-geom_point(data = shootings_case, aes(x = longitude, y = latitude), color ="black", size = .2) +
-geom_point(data = shootings_case, aes(x = longitude, y = latitude), color ="red", size = .1) +
-labs(title ="Locations of Police-Involved Shootings between 2015 and 2023",
-captions ="This is only includes cities where we have agency census data.",
-x ="Longitude",
-y ="Latitude")
-
-
Warning: The `<scale>` argument of `guides()` cannot be `FALSE`. Use "none" instead as
-of ggplot2 3.3.4.
#mapping Locations of Police-Involved Shootings between 2015 and 2023
+
+#load geo-viz libraries
+library(ggmap)
+library(maps)
+library(mapdata)
+
+#create blank map
+usa <-map_data("usa")
+states <-map_data("state")
+
+#add locations of shootings to maps
+shot_map <-ggplot(data = states) +
+geom_polygon(aes(x = long, y = lat, fill = group, group = group), color ="white") +
+coord_fixed(1.3) +
+guides(fill=FALSE) +# do this to leave off the color legend
+geom_point(data = shootings_case, aes(x = longitude, y = latitude), color ="black", size = .2) +
+geom_point(data = shootings_case, aes(x = longitude, y = latitude), color ="red", size = .1) +
+labs(title ="Locations of Police-Involved Shootings between 2015 and 2023",
+captions ="This is only includes cities where we have agency census data.",
+x ="Longitude",
+y ="Latitude")
-
agencies_census <- agencies_census |>
-left_join(shootings_by_agency, by =c("name"="agency"))
-agencies_census
-
-
# A tibble: 109 × 13
- name id state city police_force_size all white `non-white` black
- <chr> <dbl> <chr> <chr> <dbl> <dbl> <dbl> <dbl> <chr>
- 1 Albany P… 2237 NY Alba… 890 0.185 0.160 0.364 **
- 2 Albuquer… 508 NM Albu… 1340 0.616 0.630 0.602 **
- 3 Amtrak P… 1657 IL Chic… 12120 0.875 0.872 0.877 0.89…
- 4 Atlanta … 447 GA Atla… 2950 0.137 0.186 0.111 0.10…
- 5 Austin P… 141 TX Aust… 1985 0.295 0.195 0.427 0.25
- 6 Baltimor… 4784 MD Balt… 2800 0.257 0.133 0.362 0.39…
- 7 Baltimor… 149 MD Balt… 2800 0.257 0.133 0.362 0.39…
- 8 BART Pol… 2015 CA Oakl… 1530 0.0948 0.0267 0.160 0.06…
- 9 Baton Ro… 1098 LA Bato… 980 0.214 0.144 0.321 0.34…
-10 Boston P… 3 MA Bost… 2560 0.477 0.442 0.583 0.68…
-# ℹ 99 more rows
-# ℹ 4 more variables: hispanic <chr>, asian <chr>, majority <chr>, n <int>
-
-
agencies_census |>
-ggplot(mapping =aes(x = all, y = n, fill=)) +
-geom_point() +
-geom_smooth(method ="lm", formula = y ~poly(x, 2), se =FALSE)
+
#creating df with total shootings per agency and census data
+agencies_census <- agencies_census |>
+left_join(shootings_by_agency, by =c("name"="agency"))
+
+#prelim visualization of relationship between percentage of officer residency and number of fatal shootings per agency
+agencies_census |>
+ggplot(mapping =aes(x = all, y = n)) +
+geom_point() +
+geom_smooth(method ="lm", se =TRUE)
-
-
-
shootings_case |>
-ggplot(aes(x = majority, fill = armed)) +
-geom_bar() +
-labs(title ="Shootings in Cities where a Majority of Officers Reside",
-captions ="This is only includes shootings where we have agency census data.",
-x ="Does a majority a of the total police force live in the city?",
-y ="Number of fatal shootings",
-fill ="Victim Armed?")
+
+
+
#creating visualization of comparison Shootings in Cities where a Majority/Minority of Officers Reside
+p0 <- shootings_case |>
+ggplot(aes(x = majority, fill = armed)) +
+geom_bar() +
+labs(title ="Shootings in Cities where a Majority of Officers Reside",
+caption ="This is only includes shootings where we have agency census data.",
+x ="Does a majority a of the total police force live in the city?",
+y ="Number of fatal shootings",
+fill ="Victim Armed?")
+p0
#calculate mean number of shootings per agency in cities where a majority of officers reside in the city
+majority_mean <- shootings_case |>
+filter(majority ==TRUE) |>
+count(agency) |>
+summarize(maj_mean =mean(n))
+
+#calculate mean number of shootings per agency in cities where a minority of officers reside in the city
+minority_mean <- shootings_case |>
+filter(majority ==FALSE) |>
+count(agency) |>
+summarize(min_mean =mean(n))
+
+#calculate a difference in means between the `majority` and `minority`
+diff_in_means <- majority_mean - minority_mean
+diff_in_means
maj_mean
1 -2.575
-
knitr::kable(head(diff_in_means))
+
#tidy table
+knitr::kable(head(diff_in_means))
@@ -540,8 +342,9 @@
Data
-
fit <-lm(n ~ all, data = agencies_census)
-fit
+
#fit single linear regression model for correlation between percentage of officer residency and number of fatal shootings per agency
+fit <-lm(n ~ all, data = agencies_census)
+fit
Warning in right_join(., agencies_census, by = c(agency = "name")): Detected an unexpected many-to-many relationship between `x` and `y`.
-ℹ Row 63 of `x` matches multiple rows in `y`.
-ℹ Row 2 of `y` matches multiple rows in `x`.
-ℹ If a many-to-many relationship is expected, set `relationship =
- "many-to-many"` to silence this warning.
-
-
shootings_by_agency_census
-
-
# A tibble: 202 × 15
-# Groups: agency [108]
- agency armed n.x id state city police_force_size all white
- <chr> <chr> <int> <dbl> <chr> <chr> <dbl> <dbl> <dbl>
- 1 Albany Police … YES 1 2237 NY Alba… 890 0.185 0.160
- 2 Albuquerque Po… NO 10 508 NM Albu… 1340 0.616 0.630
- 3 Albuquerque Po… YES 54 508 NM Albu… 1340 0.616 0.630
- 4 Amtrak Police … NO 8 1657 IL Chic… 12120 0.875 0.872
- 5 Amtrak Police … YES 46 1657 IL Chic… 12120 0.875 0.872
- 6 Atlanta Police… NO 5 447 GA Atla… 2950 0.137 0.186
- 7 Atlanta Police… YES 31 447 GA Atla… 2950 0.137 0.186
- 8 Austin Police … NO 3 141 TX Aust… 1985 0.295 0.195
- 9 Austin Police … YES 37 141 TX Aust… 1985 0.295 0.195
-10 BART Police De… YES 12 2015 CA Oakl… 1530 0.0948 0.0267
-# ℹ 192 more rows
-# ℹ 6 more variables: `non-white` <dbl>, black <chr>, hispanic <chr>,
-# asian <chr>, majority <chr>, n.y <int>
fit_multi <-lm(n.x ~ all + armed, data = shootings_by_agency_census)
-fit_multi
+
#add `armed` and `majority` to `shootings_by_agency` df
+shootings_by_agency_census <- shootings_case |>
+group_by(agency) |>
+count(armed) |>
+drop_na(n, armed) |>
+right_join(agencies_census, by =c("agency"="name")) |>
+distinct(armed, .keep_all =TRUE)
+
+shootings_by_agency_census <- shootings_by_agency_census |>
+select(n.x, armed, all)
+
+#fit multiple linear regression model for correlation between percentage of officer residency and victim armament and number of fatal shootings per agency
+fit_multi <-lm(n.x ~ all + armed, data = shootings_by_agency_census)
+fit_multi
ggplot(data = shootings_by_agency_census, aes(x = all, y = n.x)) +
-geom_jitter(jitter =15, alpha =0.5) +
-geom_smooth(method ="lm", formula = y ~poly(x, 2), se =FALSE) +
-labs(title ="Number of Shootings on a Scale of Police Force Residency",
-x ="Percentage of the total police force that lives in the city",
-y ="Number of fatal shootings in that city")
#visualize polynomial relationship between percentage of officer residency and number of fatal shootings per agency
+ggplot(data = shootings_by_agency_census, aes(x = all, y = n.x)) +
+geom_jitter(width =0.10, height =0, alpha =0.45) +
+geom_smooth(method ="lm", formula = y ~poly(x, 2), se =TRUE) +
+labs(title ="Number of Shootings on a Scale of Police Force Residency",
+x ="Percentage of the total police force that lives in the city",
+y ="Number of fatal shootings in that city")
+
+
+
+
#visualize polynomial relationship between percentage of officer residency and victim armament and number of fatal shootings per agency
+ggplot(data = shootings_by_agency_census, aes(x = all, y = n.x, color = armed)) +
+geom_jitter(width =0.10, height =0, alpha =0.45) +
+geom_smooth(method ="lm", formula = y ~poly(x, 2), se =TRUE) +
+labs(title ="Number of Shootings on a Scale of Police Force Residency",
+x ="Percentage of the total police force that lives in the city",
+y ="Number of fatal shootings in that city",
+color ="Victim Armed?")
-
+
-
ggplot(data = shootings_by_agency_census, aes(x = all, y = n.x, color = armed)) +
-geom_jitter(jitter =15, alpha =0.5) +
-geom_smooth(method ="lm", formula = y ~poly(x, 2), se =FALSE) +
-labs(title ="Number of Shootings on a Scale of Police Force Residency",
-x ="Percentage of the total police force that lives in the city",
-y ="Number of fatal shootings in that city")
The intercept, \(35.7782\), is the estimated number of fatal shootings when the percentage of officer in-city residency (all) is \(0\). For each one-unit increase in the percentage of officer residency, the number of fatal shootings is expected to decrease by \(0.5874\) (\(-0.5874\)) units, assuming all other factors remain constant.
+
+
This model suggests that there is a negative association between the percentage of officer residency and the number of fatal shootings. However, it’s important to interpret the results in the context of your data and consider potential confounding factors, like whether or not the victim was armed.
+
The model equation for fit_multi considering victim armament (armed) is:
+
[ = 4.117 + 1.211 + 24.921 ]
+
+
The intercept, \(4.117\), is the estimated number of fatal shootings where the percentage of officer in-city residency (all) is \(0\) and the victim was un-armed. For each one-unit increase in the percentage of in-city officer residency compared to the total force (all), we expect an increase of \(1.211\) fatal shootings, assuming the victim’s armament status (armedYES) remains constant.
+
The coefficient for ‘armedYES’, \(24.921\), indicates that the victim is armed (armed is YES), we expect an increase of \(24.921\) fatal shootings compared to when the victim is not armed (armed is No), assuming the percentage of officer residency (all) remains constant.
+
+
In summary, the model suggests that the percentage of officer residency and whether the victim is armed are associated with the number of fatal shootings per agency even as we control for victim armament. However, as correlation does not imply causation, and other factors not included in the model may influence the outcomes.
+
+
#generate null distribution
+null_dist <- agencies_census |>
+specify(n ~ majority) |>
+hypothesize(null ="independence") |>
+generate(reps =1000, type ="permute") |>
+calculate(stat ="diff in means", order =c("TRUE", "FALSE"))
+
+#compute observed test statistic
+test_stat <- agencies_census |>
+specify(n ~ majority) |>
+calculate(stat ="diff in means", order =c("TRUE", "FALSE"))
+
+#visualize p-value
+null_dist |>
+visualize() +
+shade_p_value(obs_stat = test_stat, direction ="less")
-
+
+
+
#compute p-value
+ null_dist |>
+get_p_value(obs_stat = test_stat, direction ="less")
+
+
# A tibble: 1 × 1
+ p_value
+ <dbl>
+1 0.263
+
Inference for a Difference in Means
+
+
\(H_0\): The mean total number of fatal shootings per agencies does not differ based on if a majority of the officers live in the city or not.
+
\(H_A\): The mean total number of fatal shootings per agencies is fewer in cities where a majority of the officers live in the city then cities where they do not.
A Case Study on the Relationship between Police Residence and
-
-
Proposal
+
+
On average, police in the United States shoot and kill more than 1,000 people every year…and then they go home to their families
+
+
+
Abstract
This case study investigates the intricate relationship between police residence and fatal police shootings, employing a data science approach to uncover insights and patterns within the context of law enforcement agencies. Focused on police officers residing in the cities they serve, the study examines whether this residency factor correlates with the incidence of fatal police shootings. The data set, spanning the years 2015 to 2023, is composed of information on police agencies involved in at least one fatal shooting, and is subjected to rigorous analysis using advanced statistical methods and machine learning techniques.
This study aims to discern patterns, trends, and potential biases associated with the geographical proximity of police officers to the communities they police. A comprehensive exploration of demographic, socioeconomic, and policing variables contributes to a nuanced understanding of the factors influencing fatal police shootings. Furthermore, the study seeks to identify any disparities in incident rates based on officers’ residency status, considering variables such as race, community demographics, and departmental policies.
The insights derived from this case study bear substantial implications for informing public policy, refining police training protocols, and strengthening community relations. By unraveling the nuanced dynamics surrounding police residence and fatal police shootings, this case study aims to provide evidence-based recommendations to enhance transparency, accountability, and trust between law enforcement agencies and the communities they serve. In doing so, it contributes to the broader discourse on police reform, fostering a data-driven approach to address critical issues and promote safer, more resilient communities.
-
-
Explanatory Variables
-
-
-
-
-
+
+
+
Hypotheses
+
We will conduct two hypothesis tests to analyze both;
+
+
The nominal relationship between an increasing proportion of in-city officer residency and number of fatal police shooting deaths
+
+
\(H_0\): The mean total number of fatal shootings per agencies does not differ based on if a majority of the officers live in the city or not.
+
\(H_A\): The mean total number of fatal shootings per agencies is fewer in cities where a majority of the officers live in the city then cities where they do not.
The categorical difference in fatal police shooting deaths between cities where a majority or or minority of police officers live in the city.
+
+
\(H_0\): There is no relationship between percentage of the total police force that lives in the city they serve and number of fatal shootings.
+
\(H_A\): There is a relationship between percentage of the total police force that lives in the city they serve and number of fatal shootings.
+
+
\(H_0 : \rho = 0\)
+
\(H_0 : \rho \neq 0\)
+
+
+
+
+
+
Methods
+
+
Tidying Data
+
+
+Show the code
+
##Tidying Data
+
+#creating dfs from .csv files
+police_locals <-read_csv("data/police-locals.csv")
+agencies <-read_csv("data/fatal-police-shootings-agencies.csv")
+shootings <-read_csv("data/fatal-police-shootings-data.csv")
+
+#removing old `city` tag from data set that we created when decatenated the city names
+police_locals <- police_locals |>
+select(-city_old)
+
+# creating `agencies` df with just police departments
+agencies <- agencies |>
+filter(grepl("department", tolower(name))) |>
+filter(!grepl("county", tolower(name)))
+
+#creating binned categorical account of if shooting victim was `armed`
+shootings <- shootings |>
+mutate(armed =case_when(is.na(armed_with) ~"NO",
+ armed_with =="unarmed"~"NO",
+ armed_with =="unknown"~"NO",
+ armed_with =="undetermined"~"NO",
+ armed_with =="gun"~"YES",
+ armed_with =="knife"~"YES",
+ armed_with =="blunt_object"~"YES",
+ armed_with =="other"~"YES",
+ armed_with =="replica"~"YES",
+ armed_with =="vehicle"~"YES"))
+
+#creating df with only agency `names`, `id`, and `state`
+agencies_ids <- agencies |>
+select(name, id, state)
+
+#creating df with `city`, `agency`, and `state` info for each shooting
+shooting_agencies <- shootings |>
+select(city, agency_ids, state)
+
+#changing `shooting` var in `shooting_agencies` df to numeric
+shooting_agencies$agency_ids <-as.numeric(shootings$agency_ids)
+
+#creating df with `city` and `state` info for each agency by joining `agencies_ids` and `shooting_agencies`
+agencies_w_cities <- agencies_ids |>
+left_join(shooting_agencies, by =c("id"="agency_ids", "state"="state")) |>
+drop_na(city) |>
+distinct(id, .keep_all =TRUE)
+
+#creating df with census data for each agency by joining `agencies_w_cities` and `police_locals`
+agencies_census <- agencies_w_cities |>
+full_join(police_locals, by =c("city"="city", "state"="state")) |>
+drop_na(police_force_size) |>
+distinct(id, .keep_all =TRUE) |>
+mutate(majority =if_else(all >=0.5, "TRUE", "FALSE"))
+
+#creating df of only shootings involving agencies within `agencies` df
+shootings_case <- shootings |>
+right_join(agencies_census, by =c("city"="city", "state"="state")) |>
+select(-agency_ids) |>
+rename(agency_ids = id.y, id = id.x, agency = name.y, victim = name.x) |>
+select(-location_precision, -race_source)
+
+
+
+
+
Counting Shootings
+
+
+Show the code
+
#count shootings by agency
+shootings_by_agency <- shootings_case |>
+count(agency)
+
+#find top 25 agencies with the most shootings
+top_25_agencies <- shootings_by_agency |>
+slice_max(n, n =25)
+
+
+
+
+
Mapping Locations of Police-Involved Shootings between 2015 and 2023
+
+
+Show the code
+
shot_map
+
+
+
+
+
+
+
+Show the code
+
#creating df with total shootings per agency and census data
+agencies_census <- agencies_census |>
+left_join(shootings_by_agency, by =c("name"="agency"))
+
+#creating visualization of comparison Shootings in Cities where a Majority/Minority of Officers Reside
+p0 <- shootings_case |>
+ggplot(aes(x = majority, fill = armed)) +
+geom_bar() +
+labs(title ="Shootings in Cities where a Majority of Officers Reside",
+caption ="This is only includes shootings where we have agency census data.",
+x ="Does a majority a of the total police force live in the city?",
+y ="Number of fatal shootings",
+fill ="Victim Armed?")
+
+
+
+
+Show the code
+
p0
+
+
+
+
+
+
+
+Show the code
+
#calculate mean number of shootings per agency in cities where a majority of officers reside in the city
+majority_mean <- shootings_case |>
+filter(majority ==TRUE) |>
+count(agency) |>
+summarize(maj_mean =mean(n))
+
+#calculate mean number of shootings per agency in cities where a minority of officers reside in the city
+minority_mean <- shootings_case |>
+filter(majority ==FALSE) |>
+count(agency) |>
+summarize(min_mean =mean(n))
+
+#calculate a difference in means between the `majority` and `minority`
+diff_in_means <- majority_mean - minority_mean
+
+
+
+
+Show the code
+
#tidy table
+knitr::kable(head(diff_in_means))
+
+
+
-
Name
-
Description
+
maj_mean
-
city
-
U.S. city
-
-
-
police_force_size
-
Number of police officers serving that city
-
-
-
all
-
Percentage of the total police force that lives in the city
-
-
-
white
-
Percentage of white (non-Hispanic) police officers who live in the city
-
-
-
non-white
-
Percentage of non-white police officers who live in the city
-
-
-
black
-
Percentage of black police officers who live in the city
-
-
-
hispanic
-
Percentage of Hispanic police officers who live in the city
-
-
-
asian
-
Percentage of Asian police officers who live in the city
+
-2.575
-
Incident Information
-
+
+
+
+
+Show the code
+
#fit single linear regression model for correlation between percentage of officer residency and number of fatal shootings per agency
+fit <-lm(n ~ all, data = agencies_census)
+
+#add `armed` and `majority` to `shootings_by_agency` df
+shootings_by_agency_census <- shootings_case |>
+group_by(agency) |>
+count(armed) |>
+drop_na(n, armed) |>
+right_join(agencies_census, by =c("agency"="name")) |>
+distinct(armed, .keep_all =TRUE)
+
+shootings_by_agency_census <- shootings_by_agency_census |>
+select(n.x, armed, all)
+
+#fit multiple linear regression model for correlation between percentage of officer residency and victim armament and number of fatal shootings per agency
+fit_multi <-lm(n.x ~ all + armed, data = shootings_by_agency_census)
+
+
+
+
+
+
Results
+
+
Multiple Linear Regression of relationship between percentage of officer residency and number of fatal shootings per agency fit
+
The model equation for fit is:
+
\[
+\text{Number of Fatal Shootings (n)} = 35.7782 - 0.5874 \times \text{Percentage of Officer Residency (all)}
+\]
A unique identifier for each fatal police shooting incident.
-
-
-
date
-
The date of the fatal shooting.
-
-
-
body_camera
-
Whether news reports have indicated an officer was wearing a body camera and it may have recorded some portion of the incident.
-
-
-
city
-
The municipality where the fatal shooting took place
-
-
-
county
-
County where the fatal shooting took place.
+
intercept
+
35.778
+
6.887
+
5.195
+
0.000
+
22.126
+
49.430
-
state
-
The two-letter postal code abbreviation for the state in which the fatal shooting took place.
-
-
-
latitude
-
The latitude location of the shooting expressed as WGS84 coordinates, geocoded from addresses. Please note that the precision and accuracy of incident coordinates varies depending on the precision of the input address which is often only available at the block level.
-
-
-
longitude
-
The longitude location of the shooting expressed as WGS84 coordinates, geocoded from addresses.
+
all
+
-0.587
+
14.902
+
-0.039
+
0.969
+
-30.130
+
28.955
-
Agency Information
-
+
+
+
Interpretation:
+
+
The intercept, \(35.7782\), is the estimated number of fatal shootings when the percentage of officer in-city residency (all) is \(0\). For each one-unit increase in the percentage of officer residency, the number of fatal shootings is expected to decrease by \(0.5874\) (\(-0.5874\)) units, assuming all other factors remain constant.
+
+
This model suggests that there is a negative association between the percentage of officer residency and the number of fatal shootings. However, it’s important to interpret the results in the context of your data and consider potential confounding factors, like whether or not the victim was armed.
+
+
+Show the code
+
#visualize polynomial relationship between percentage of officer residency and number of fatal shootings per agency
+ggplot(data = shootings_by_agency_census, aes(x = all, y = n.x)) +
+geom_jitter(width =0.10, height =0, alpha =0.45) +
+geom_smooth(method ="lm", formula = y ~poly(x, 2), se =TRUE) +
+labs(title ="Number of Shootings on a Scale of Police Force Residency",
+x ="Percentage of the total police force that lives in the city",
+y ="Number of fatal shootings in that city")
+
+
+
+
+
+
+
+
Multiple Linear Regression of relationship between percentage of officer residency/victim armament and number of fatal shootings per agency fit_multi
+
The model equation for fit_multi considering victim armament (armed) is:
The intercept, \(4.117\), is the estimated number of fatal shootings where the percentage of officer in-city residency (all) is \(0\) and the victim was un-armed. For each one-unit increase in the percentage of in-city officer residency compared to the total force (all), we expect an increase of\(1.211\) fatal shootings, assuming the victim’s armament status (armedYES) remains constant.
+
The coefficient for ‘armedYES’, \(24.921\), indicates that the victim is armed (armed is YES), we expect an increase of\(24.921\) fatal shootings compared to when the victim is not armed (armed is No), assuming the percentage of officer residency (all) remains constant.
+
In summary, the model suggests that the percentage of officer residency and whether the victim is armed are associated with the number of fatal shootings per agency even as we control for victim armament. However, as correlation does not imply causation, and other factors not included in the model may influence the outcomes.
+
+
+Show the code
+
#visualize polynomial relationship between percentage of officer residency and victim armament and number of fatal shootings per agency
+ggplot(data = shootings_by_agency_census, aes(x = all, y = n.x, color = armed)) +
+geom_jitter(width =0.10, height =0, alpha =0.45) +
+geom_smooth(method ="lm", formula = y ~poly(x, 2), se =TRUE) +
+labs(title ="Number of Shootings on a Scale of Police Force Residency",
+x ="Percentage of the total police force that lives in the city",
+y ="Number of fatal shootings in that city",
+color ="Victim Armed?")
+
+
+
+
+
+
+
+Show the code
+
#generate null distribution
+null_dist <- agencies_census |>
+specify(n ~ majority) |>
+hypothesize(null ="independence") |>
+generate(reps =1000, type ="permute") |>
+calculate(stat ="diff in means", order =c("TRUE", "FALSE"))
+
+#compute observed test statistic
+test_stat <- agencies_census |>
+specify(n ~ majority) |>
+calculate(stat ="diff in means", order =c("TRUE", "FALSE"))
+
+#visualize p-value
+null_dist |>
+visualize() +
+shade_p_value(obs_stat = test_stat, direction ="less")
+
+
+
+
+
+Show the code
+
#compute p-value
+ null_dist |>
+get_p_value(obs_stat = test_stat, direction ="less")
+
+
+
# A tibble: 1 × 1
+ p_value
+ <dbl>
+1 0.252
+
+
+
At a significance level of \(\alpha = 0.05\), the p-value of \(0.248\) suggests that, there is insufficient evidence to reject the null hypothesis. In this context, since our null hypothesis asserts that mean total number of fatal shootings per agencies does not differ based on if a majority of the officers live in the city or not, our p-value indicates that, assuming our null is true, the probability of observing our given test statistic (difference in means; \(\mu_{maj} − \mu_{min}\)) is \(-4.92\) is around \(25\%\) (\(0.248\)). Meaning our observed difference in means between the groups is likely to have occurred by random chance.
#compute p-value
+null_dist_cor |>
+get_p_value(obs_stat = test_stat, direction ="two.sided")
+
+
+
# A tibble: 1 × 1
+ p_value
+ <dbl>
+1 0
+
+
+
At a significance level of \(\alpha = 0.05\), the p-value of \(0.248\) suggests that, there is sufficient evidence to reject the null hypothesis. In this context, since our null hypothesis asserts that there is no relationship between percentage of the total police force that lives in the city they serve and number of fatal shootings, our p-value indicates that, assuming our null is true, the probability of observing our given test statistic (correlation coefficient; \(\rho = 0\)) is \(-0.0470\) is around \(0\%\) (\(0\)). Meaning our observed correlation coefficient likely would not happen if there was no relationship between percentage of officer residency and number of fatal shootings for a given agency.
+
+
+
+
Conclusion
+
+
General Conclusions
+
+
+
Study Limitations
+
+
+
Improvements for Future Study
+
+
+
+
Citations
diff --git a/_site/index_files/figure-html/Hypothesis Testing for Correlation between Total Fatal Shootings and Residency Prop-1.png b/_site/index_files/figure-html/Hypothesis Testing for Correlation between Total Fatal Shootings and Residency Prop-1.png
new file mode 100644
index 0000000..fe27455
Binary files /dev/null and b/_site/index_files/figure-html/Hypothesis Testing for Correlation between Total Fatal Shootings and Residency Prop-1.png differ
diff --git a/_site/index_files/figure-html/Hypothesis Testing for Diff in Mean Total Fatal Shootings between Residency Prop-1.png b/_site/index_files/figure-html/Hypothesis Testing for Diff in Mean Total Fatal Shootings between Residency Prop-1.png
new file mode 100644
index 0000000..5fe78ce
Binary files /dev/null and b/_site/index_files/figure-html/Hypothesis Testing for Diff in Mean Total Fatal Shootings between Residency Prop-1.png differ
diff --git a/_site/data_files/figure-html/unnamed-chunk-4-1.png b/_site/index_files/figure-html/unnamed-chunk-1-1.png
similarity index 100%
rename from _site/data_files/figure-html/unnamed-chunk-4-1.png
rename to _site/index_files/figure-html/unnamed-chunk-1-1.png
diff --git a/_site/index_files/figure-html/unnamed-chunk-2-1.png b/_site/index_files/figure-html/unnamed-chunk-2-1.png
new file mode 100644
index 0000000..a1e7dba
Binary files /dev/null and b/_site/index_files/figure-html/unnamed-chunk-2-1.png differ
diff --git a/_site/index_files/figure-html/unnamed-chunk-7-1.png b/_site/index_files/figure-html/unnamed-chunk-7-1.png
new file mode 100644
index 0000000..fa72337
Binary files /dev/null and b/_site/index_files/figure-html/unnamed-chunk-7-1.png differ
diff --git a/_site/index_files/figure-html/unnamed-chunk-9-1.png b/_site/index_files/figure-html/unnamed-chunk-9-1.png
new file mode 100644
index 0000000..46fc51f
Binary files /dev/null and b/_site/index_files/figure-html/unnamed-chunk-9-1.png differ
diff --git a/_site/search.json b/_site/search.json
index 67f3d6e..652322a 100644
--- a/_site/search.json
+++ b/_site/search.json
@@ -4,42 +4,70 @@
"href": "background.html",
"title": "Background",
"section": "",
- "text": "#On average, police in the United States shoot and kill more than 1,000 people every year, according to an ongoing analysis by The Washington Post."
+ "text": "On average, police in the United States shoot and kill more than 1,000 people every year, according to an ongoing analysis by The Washington Post.\n\n\nProposal\nWe propose a case study to explore the relationship between police residence and fatal police shootings, employing advanced data science methodologies. Focusing on officers residing in the cities they serve, our project aims to uncover insights and patterns that contribute to a nuanced understanding of this complex issue.\n\nObjectives:\n\nInvestigate the correlation between police residence and fatal police shootings.\nUtilize a comprehensive dataset spanning 2015 to 2023, focusing on police agencies involved in at least one fatal shooting.\nApply advanced statistical methods and machine learning techniques to identify patterns and potential biases.\nExamine disparities in incident rates based on officers’ residency status, considering demographic, socioeconomic, and policing variables.\n\nMethodology:\n\nData Collection:\n\nCompile a dataset comprising information on police agencies involved in fatal police shootings.\nCompile a data set of census variables such as officer residency, race, community demographics, and departmental policies.\n\nAnalysis:\n\nEmploy advanced statistical methods and machine learning techniques to discern patterns and correlations.\nConduct a comprehensive exploration of variables influencing fatal police shootings.\n\n\nHypothesis and Expected Outcomes:\nWe will conduct two hypothesis tests to analyze both;\n\nthe nominal relationship between an increasing proportion of in-city officer residency and number of fatal police shooting deaths and\nthe categorical difference in fatal police shooting deaths between cities where a majority or or minority of police officers live in the city.\n\n\nInference for a Difference in Proportions\n\n\\(H_0\\): The mean total number of fatal shootings per agencies does not differ based on if a majority of the officers live in the city or not.\n\\(H_A\\): The mean total number of fatal shootings per agencies is fewer in cities where a majority of the officers live in the city then cities where they do not.\n\n\\(H_0 : p\\_{maj} − p\\_{min} = 0\\), or equivalently \\(H_0 : p\\_{maj} = p\\_{min}\\)\n\\(H_A : p\\_{maj} − p\\_{min} < 0\\), or equivalently \\(H_A : p\\_{maj} < p\\_{min}\\)\n\n\nInference for a Correlation\n\n\\(H_O\\): There is no relationship between percentage of the total police force that lives in the city they serve and number of fatal shootings.\n\\(H_A\\): There is a relationship between percentage of the total police force that lives in the city they serve and number of fatal shootings.\n\n\\(H_0 : \\rho = 0\\)\n\\(H_0 : \\rho \\neq 0\\)\n\n\n\n\n\n\nThe Washington Post Fatal Force Database\nIn 2015, The Washington Post began tracking details about each police-involved killing in the United States — the race of the deceased, the circumstances of the shooting, whether the person was armed and whether the person was experiencing a mental-health crisis — by manually culling local news reports, collecting information from law enforcement websites and social media, and monitoring independent databases such as Fatal Encounters and the now-defunct Killed by Police project. In many cases, The Post conducts additional reporting.\nIn 2022, The Post updated its database to standardize and publish the names of the police agencies involved in each shooting to better measure accountability at the department level.\nThe 2014 killing of Michael Brown in Ferguson, Mo. began a protest movement culminating in the Black Lives Matter movement and an increased focus on police accountability nationwide. In this data set, The Post tracks only shootings with circumstances closely paralleling those like the killing of Brown — incidents in which a police officer, in the line of duty, shoots and kills a civilian. The Post is not tracking deaths of people in police custody, fatal shootings by off-duty officers or non-shooting deaths in this data set.\nThe FBI and the Centers for Disease Control and Prevention log fatal shootings by police, but officials acknowledge that their data is incomplete. Since 2015, The Post has documented more than twice as many fatal shootings by police as recorded by federal officials on average annually. That gap has widened in recent years, as the FBI in 2021 tracked only a third of departments’ fatal shootings.\n\n\nMost Police Don’t Live In The Cities They Serve\nIn Ferguson, Missouri, where protests lamented for months following the shooting of a teenager by a police officer this month, more than two-thirds of the civilian population is black. Only 11 percent of the police force is. The racial disparity is troubling enough on its own, but it’s also suggestive of another type of misrepresentation. Given Ferguson’s racial gap, it’s likely that many of its police officers live outside city limits.\nIf so, Ferguson would have something in common with most major American cities. In about two-thirds of the U.S. cities with the largest police forces, the majority of police officers commute to work from another town.\nOn average among the 75 American cities with the largest police forces, 49 percent of black police officers and 47 percent of Hispanic officers live within the city limits. But just 35 percent of white police officers do. The disparity is starkest in cities with largely black populations. In Detroit, for example, 57 percent of black police officers live in the city but just 8 percent of white ones do. Memphis, Tennessee; Baltimore; Birmingham, Alabama; and Jackson, Mississippi — also majority black — likewise have large racial gaps in where their police officers live."
},
{
- "objectID": "background.html#wapo-fatal-force-database",
- "href": "background.html#wapo-fatal-force-database",
- "title": "Background",
- "section": "WaPo Fatal Force Database",
- "text": "WaPo Fatal Force Database\nIn 2015, The Washington Post began tracking details about each police-involved killing in the United States — the race of the deceased, the circumstances of the shooting, whether the person was armed and whether the person was experiencing a mental-health crisis — by manually culling local news reports, collecting information from law enforcement websites and social media, and monitoring independent databases such as Fatal Encounters and the now-defunct Killed by Police project. In many cases, The Post conducts additional reporting.\nIn 2022, The Post updated its database to standardize and publish the names of the police agencies involved in each shooting to better measure accountability at the department level.\nThe 2014 killing of Michael Brown in Ferguson, Mo. began a protest movement culminating in the Black Lives Matter movement and an increased focus on police accountability nationwide. In this data set, The Post tracks only shootings with circumstances closely paralleling those like the killing of Brown — incidents in which a police officer, in the line of duty, shoots and kills a civilian. The Post is not tracking deaths of people in police custody, fatal shootings by off-duty officers or non-shooting deaths in this data set.\nThe FBI and the Centers for Disease Control and Prevention log fatal shootings by police, but officials acknowledge that their data is incomplete. Since 2015, The Post has documented more than twice as many fatal shootings by police as recorded by federal officials on average annually. That gap has widened in recent years, as the FBI in 2021 tracked only a third of departments’ fatal shootings.\n#Most Police Don’t Live In The Cities They Serve\nIn Ferguson, Missouri, where protests lamented for months following the shooting of a teenager by a police officer this month, more than two-thirds of the civilian population is black. Only 11 percent of the police force is. The racial disparity is troubling enough on its own, but it’s also suggestive of another type of misrepresentation. Given Ferguson’s racial gap, it’s likely that many of its police officers live outside city limits.\nIf so, Ferguson would have something in common with most major American cities. In about two-thirds of the U.S. cities with the largest police forces, the majority of police officers commute to work from another town.\nOn average among the 75 American cities with the largest police forces, 49 percent of black police officers and 47 percent of Hispanic officers live within the city limits. But just 35 percent of white police officers do. The disparity is starkest in cities with largely black populations. In Detroit, for example, 57 percent of black police officers live in the city but just 8 percent of white ones do. Memphis, Tennessee; Baltimore; Birmingham, Alabama; and Jackson, Mississippi — also majority black — likewise have large racial gaps in where their police officers live."
+ "objectID": "codebook.html",
+ "href": "codebook.html",
+ "title": "Codebook",
+ "section": "",
+ "text": "Name\nDescription\n\n\n\n\ncity\nU.S. city\n\n\npolice_force_size\nNumber of police officers serving that city\n\n\nall\nPercentage of the total police force that lives in the city\n\n\nwhite\nPercentage of white (non-Hispanic) police officers who live in the city\n\n\nnon-white\nPercentage of non-white police officers who live in the city\n\n\nblack\nPercentage of black police officers who live in the city\n\n\nhispanic\nPercentage of Hispanic police officers who live in the city\n\n\nasian\nPercentage of Asian police officers who live in the city\n\n\n\n\n\n\n\n\n\n\n\n\nName\nDescription\n\n\n\n\nid\nA unique identifier for each fatal police shooting incident.\n\n\ndate\nThe date of the fatal shooting.\n\n\nbody_camera\nWhether news reports have indicated an officer was wearing a body camera and it may have recorded some portion of the incident.\n\n\ncity\nThe municipality where the fatal shooting took place\n\n\ncounty\nCounty where the fatal shooting took place.\n\n\nstate\nThe two-letter postal code abbreviation for the state in which the fatal shooting took place.\n\n\nlatitude\nThe latitude location of the shooting expressed as WGS84 coordinates, geocoded from addresses. Please note that the precision and accuracy of incident coordinates varies depending on the precision of the input address which is often only available at the block level.\n\n\nlongitude\nThe longitude location of the shooting expressed as WGS84 coordinates, geocoded from addresses.\n\n\n\n\n\n\n\n\n\n\nDescription\n\n\n\n\nid\nDepartment Database Id\n\n\nname\nDepartment Name\n\n\nstate\nState in which the agency is located."
},
{
- "objectID": "index.html",
- "href": "index.html",
- "title": "A Case Study on the Relationship between Police Residence and Fatal Police Shootings",
+ "objectID": "codebook.html#explanatory-variables",
+ "href": "codebook.html#explanatory-variables",
+ "title": "Codebook",
"section": "",
- "text": "This case study investigates the intricate relationship between police residence and fatal police shootings, employing a data science approach to uncover insights and patterns within the context of law enforcement agencies. Focused on police officers residing in the cities they serve, the study examines whether this residency factor correlates with the incidence of fatal police shootings. The data set, spanning the years 2015 to 2023, is composed of information on police agencies involved in at least one fatal shooting, and is subjected to rigorous analysis using advanced statistical methods and machine learning techniques.\nThis study aims to discern patterns, trends, and potential biases associated with the geographical proximity of police officers to the communities they police. A comprehensive exploration of demographic, socioeconomic, and policing variables contributes to a nuanced understanding of the factors influencing fatal police shootings. Furthermore, the study seeks to identify any disparities in incident rates based on officers’ residency status, considering variables such as race, community demographics, and departmental policies.\nThe insights derived from this case study bear substantial implications for informing public policy, refining police training protocols, and strengthening community relations. By unraveling the nuanced dynamics surrounding police residence and fatal police shootings, this case study aims to provide evidence-based recommendations to enhance transparency, accountability, and trust between law enforcement agencies and the communities they serve. In doing so, it contributes to the broader discourse on police reform, fostering a data-driven approach to address critical issues and promote safer, more resilient communities.\n\n\n\n\n\n\n\n\n\nName\nDescription\n\n\n\n\ncity\nU.S. city\n\n\npolice_force_size\nNumber of police officers serving that city\n\n\nall\nPercentage of the total police force that lives in the city\n\n\nwhite\nPercentage of white (non-Hispanic) police officers who live in the city\n\n\nnon-white\nPercentage of non-white police officers who live in the city\n\n\nblack\nPercentage of black police officers who live in the city\n\n\nhispanic\nPercentage of Hispanic police officers who live in the city\n\n\nasian\nPercentage of Asian police officers who live in the city\n\n\n\nIncident Information\n\n\n\n\n\n\n\nName\nDescription\n\n\n\n\nid\nA unique identifier for each fatal police shooting incident.\n\n\ndate\nThe date of the fatal shooting.\n\n\nbody_camera\nWhether news reports have indicated an officer was wearing a body camera and it may have recorded some portion of the incident.\n\n\ncity\nThe municipality where the fatal shooting took place\n\n\ncounty\nCounty where the fatal shooting took place.\n\n\nstate\nThe two-letter postal code abbreviation for the state in which the fatal shooting took place.\n\n\nlatitude\nThe latitude location of the shooting expressed as WGS84 coordinates, geocoded from addresses. Please note that the precision and accuracy of incident coordinates varies depending on the precision of the input address which is often only available at the block level.\n\n\nlongitude\nThe longitude location of the shooting expressed as WGS84 coordinates, geocoded from addresses.\n\n\n\nAgency Information\n\n\n\n\nDescription\n\n\n\n\nid\nDepartment Database Id\n\n\nname\nDepartment Name\n\n\nstate\nState in which the agency is located."
+ "text": "Name\nDescription\n\n\n\n\ncity\nU.S. city\n\n\npolice_force_size\nNumber of police officers serving that city\n\n\nall\nPercentage of the total police force that lives in the city\n\n\nwhite\nPercentage of white (non-Hispanic) police officers who live in the city\n\n\nnon-white\nPercentage of non-white police officers who live in the city\n\n\nblack\nPercentage of black police officers who live in the city\n\n\nhispanic\nPercentage of Hispanic police officers who live in the city\n\n\nasian\nPercentage of Asian police officers who live in the city\n\n\n\n\n\n\n\n\n\n\n\n\nName\nDescription\n\n\n\n\nid\nA unique identifier for each fatal police shooting incident.\n\n\ndate\nThe date of the fatal shooting.\n\n\nbody_camera\nWhether news reports have indicated an officer was wearing a body camera and it may have recorded some portion of the incident.\n\n\ncity\nThe municipality where the fatal shooting took place\n\n\ncounty\nCounty where the fatal shooting took place.\n\n\nstate\nThe two-letter postal code abbreviation for the state in which the fatal shooting took place.\n\n\nlatitude\nThe latitude location of the shooting expressed as WGS84 coordinates, geocoded from addresses. Please note that the precision and accuracy of incident coordinates varies depending on the precision of the input address which is often only available at the block level.\n\n\nlongitude\nThe longitude location of the shooting expressed as WGS84 coordinates, geocoded from addresses.\n\n\n\n\n\n\n\n\n\n\nDescription\n\n\n\n\nid\nDepartment Database Id\n\n\nname\nDepartment Name\n\n\nstate\nState in which the agency is located."
+ },
+ {
+ "objectID": "codebook.html#project-thoughts",
+ "href": "codebook.html#project-thoughts",
+ "title": "Codebook",
+ "section": "Project thoughts",
+ "text": "Project thoughts\nI am interested in exploring data related to…\n\nPolitical Extremism\nBlack American Opinion"
},
{
- "objectID": "index.html#proposal",
- "href": "index.html#proposal",
+ "objectID": "index.html#sec-abstract",
+ "href": "index.html#sec-abstract",
"title": "A Case Study on the Relationship between Police Residence and Fatal Police Shootings",
- "section": "",
- "text": "This case study investigates the intricate relationship between police residence and fatal police shootings, employing a data science approach to uncover insights and patterns within the context of law enforcement agencies. Focused on police officers residing in the cities they serve, the study examines whether this residency factor correlates with the incidence of fatal police shootings. The data set, spanning the years 2015 to 2023, is composed of information on police agencies involved in at least one fatal shooting, and is subjected to rigorous analysis using advanced statistical methods and machine learning techniques.\nThis study aims to discern patterns, trends, and potential biases associated with the geographical proximity of police officers to the communities they police. A comprehensive exploration of demographic, socioeconomic, and policing variables contributes to a nuanced understanding of the factors influencing fatal police shootings. Furthermore, the study seeks to identify any disparities in incident rates based on officers’ residency status, considering variables such as race, community demographics, and departmental policies.\nThe insights derived from this case study bear substantial implications for informing public policy, refining police training protocols, and strengthening community relations. By unraveling the nuanced dynamics surrounding police residence and fatal police shootings, this case study aims to provide evidence-based recommendations to enhance transparency, accountability, and trust between law enforcement agencies and the communities they serve. In doing so, it contributes to the broader discourse on police reform, fostering a data-driven approach to address critical issues and promote safer, more resilient communities.\n\n\n\n\n\n\n\n\n\nName\nDescription\n\n\n\n\ncity\nU.S. city\n\n\npolice_force_size\nNumber of police officers serving that city\n\n\nall\nPercentage of the total police force that lives in the city\n\n\nwhite\nPercentage of white (non-Hispanic) police officers who live in the city\n\n\nnon-white\nPercentage of non-white police officers who live in the city\n\n\nblack\nPercentage of black police officers who live in the city\n\n\nhispanic\nPercentage of Hispanic police officers who live in the city\n\n\nasian\nPercentage of Asian police officers who live in the city\n\n\n\nIncident Information\n\n\n\n\n\n\n\nName\nDescription\n\n\n\n\nid\nA unique identifier for each fatal police shooting incident.\n\n\ndate\nThe date of the fatal shooting.\n\n\nbody_camera\nWhether news reports have indicated an officer was wearing a body camera and it may have recorded some portion of the incident.\n\n\ncity\nThe municipality where the fatal shooting took place\n\n\ncounty\nCounty where the fatal shooting took place.\n\n\nstate\nThe two-letter postal code abbreviation for the state in which the fatal shooting took place.\n\n\nlatitude\nThe latitude location of the shooting expressed as WGS84 coordinates, geocoded from addresses. Please note that the precision and accuracy of incident coordinates varies depending on the precision of the input address which is often only available at the block level.\n\n\nlongitude\nThe longitude location of the shooting expressed as WGS84 coordinates, geocoded from addresses.\n\n\n\nAgency Information\n\n\n\n\nDescription\n\n\n\n\nid\nDepartment Database Id\n\n\nname\nDepartment Name\n\n\nstate\nState in which the agency is located."
+ "section": "Abstract",
+ "text": "Abstract\nThis case study investigates the intricate relationship between police residence and fatal police shootings, employing a data science approach to uncover insights and patterns within the context of law enforcement agencies. Focused on police officers residing in the cities they serve, the study examines whether this residency factor correlates with the incidence of fatal police shootings. The data set, spanning the years 2015 to 2023, is composed of information on police agencies involved in at least one fatal shooting, and is subjected to rigorous analysis using advanced statistical methods and machine learning techniques.\nThis study aims to discern patterns, trends, and potential biases associated with the geographical proximity of police officers to the communities they police. A comprehensive exploration of demographic, socioeconomic, and policing variables contributes to a nuanced understanding of the factors influencing fatal police shootings. Furthermore, the study seeks to identify any disparities in incident rates based on officers’ residency status, considering variables such as race, community demographics, and departmental policies.\nThe insights derived from this case study bear substantial implications for informing public policy, refining police training protocols, and strengthening community relations. By unraveling the nuanced dynamics surrounding police residence and fatal police shootings, this case study aims to provide evidence-based recommendations to enhance transparency, accountability, and trust between law enforcement agencies and the communities they serve. In doing so, it contributes to the broader discourse on police reform, fostering a data-driven approach to address critical issues and promote safer, more resilient communities."
},
{
- "objectID": "index.html#project-thoughts",
- "href": "index.html#project-thoughts",
+ "objectID": "index.html#hypotheses",
+ "href": "index.html#hypotheses",
"title": "A Case Study on the Relationship between Police Residence and Fatal Police Shootings",
- "section": "Project thoughts",
- "text": "Project thoughts\nI am interested in exploring data related to…\n\nPolitical Extremism\nBlack American Opinion"
+ "section": "Hypotheses",
+ "text": "Hypotheses\nWe will conduct two hypothesis tests to analyze both;\n\nThe nominal relationship between an increasing proportion of in-city officer residency and number of fatal police shooting deaths\n\n\\(H_0\\): The mean total number of fatal shootings per agencies does not differ based on if a majority of the officers live in the city or not.\n\\(H_A\\): The mean total number of fatal shootings per agencies is fewer in cities where a majority of the officers live in the city then cities where they do not.\n\n\\(H_0 : p\\_{maj} − p\\_{min} = 0\\), or equivalently \\(H_0 : p\\_{maj} = p\\_{min}\\)\n\\(H_A : p\\_{maj} − p\\_{min} < 0\\), or equivalently \\(H_A : p\\_{maj} < p\\_{min}\\)\n\n\nThe categorical difference in fatal police shooting deaths between cities where a majority or or minority of police officers live in the city.\n\n\\(H_0\\): There is no relationship between percentage of the total police force that lives in the city they serve and number of fatal shootings.\n\\(H_A\\): There is a relationship between percentage of the total police force that lives in the city they serve and number of fatal shootings.\n\n\\(H_0 : \\rho = 0\\)\n\\(H_0 : \\rho \\neq 0\\)"
},
{
- "objectID": "about.html",
- "href": "about.html",
- "title": "About",
- "section": "",
- "text": "About this site\n\n1 + 1\n\n[1] 2"
+ "objectID": "index.html#methods",
+ "href": "index.html#methods",
+ "title": "A Case Study on the Relationship between Police Residence and Fatal Police Shootings",
+ "section": "Methods",
+ "text": "Methods\n\nTidying Data\n\n\nShow the code\n##Tidying Data\n\n#creating dfs from .csv files\npolice_locals <- read_csv(\"data/police-locals.csv\")\nagencies <- read_csv(\"data/fatal-police-shootings-agencies.csv\")\nshootings <- read_csv(\"data/fatal-police-shootings-data.csv\")\n\n#removing old `city` tag from data set that we created when decatenated the city names\npolice_locals <- police_locals |>\n select(-city_old)\n\n# creating `agencies` df with just police departments\nagencies <- agencies |>\n filter(grepl(\"department\", tolower(name))) |>\n filter(!grepl(\"county\", tolower(name)))\n\n#creating binned categorical account of if shooting victim was `armed`\nshootings <- shootings |>\n mutate(armed = case_when(is.na(armed_with) ~ \"NO\",\n armed_with == \"unarmed\" ~ \"NO\",\n armed_with == \"unknown\" ~ \"NO\",\n armed_with == \"undetermined\" ~ \"NO\",\n armed_with == \"gun\" ~ \"YES\",\n armed_with == \"knife\" ~ \"YES\",\n armed_with == \"blunt_object\" ~ \"YES\",\n armed_with == \"other\" ~ \"YES\",\n armed_with == \"replica\" ~ \"YES\",\n armed_with == \"vehicle\" ~ \"YES\"))\n\n#creating df with only agency `names`, `id`, and `state`\nagencies_ids <- agencies |>\n select(name, id, state)\n\n#creating df with `city`, `agency`, and `state` info for each shooting\nshooting_agencies <- shootings |>\n select(city, agency_ids, state)\n\n#changing `shooting` var in `shooting_agencies` df to numeric\nshooting_agencies$agency_ids <- as.numeric(shootings$agency_ids)\n\n#creating df with `city` and `state` info for each agency by joining `agencies_ids` and `shooting_agencies`\nagencies_w_cities <- agencies_ids |>\n left_join(shooting_agencies, by = c(\"id\" = \"agency_ids\", \"state\" = \"state\")) |>\n drop_na(city) |>\n distinct(id, .keep_all = TRUE)\n\n#creating df with census data for each agency by joining `agencies_w_cities` and `police_locals`\nagencies_census <- agencies_w_cities |>\n full_join(police_locals, by = c(\"city\" = \"city\", \"state\" = \"state\")) |>\n drop_na(police_force_size) |>\n distinct(id, .keep_all = TRUE) |>\n mutate(majority = if_else(all >= 0.5, \"TRUE\", \"FALSE\"))\n\n#creating df of only shootings involving agencies within `agencies` df\nshootings_case <- shootings |>\n right_join(agencies_census, by = c(\"city\" = \"city\", \"state\" = \"state\")) |>\n select(-agency_ids) |>\n rename(agency_ids = id.y, id = id.x, agency = name.y, victim = name.x) |>\n select(-location_precision, -race_source)\n\n\n\n\nCounting Shootings\n\n\nShow the code\n#count shootings by agency\nshootings_by_agency <- shootings_case |>\n count(agency)\n\n#find top 25 agencies with the most shootings\ntop_25_agencies <- shootings_by_agency |>\n slice_max(n, n = 25)\n\n\n\n\nMapping Locations of Police-Involved Shootings between 2015 and 2023\n\n\nShow the code\nshot_map\n\n\n\n\n\n\n\nShow the code\n#creating df with total shootings per agency and census data\nagencies_census <- agencies_census |>\n left_join(shootings_by_agency, by = c(\"name\" = \"agency\"))\n\n#creating visualization of comparison Shootings in Cities where a Majority/Minority of Officers Reside\np0 <- shootings_case |>\n ggplot(aes(x = majority, fill = armed)) +\n geom_bar() + \n labs(title = \"Shootings in Cities where a Majority of Officers Reside\",\n caption = \"This is only includes shootings where we have agency census data.\",\n x = \"Does a majority a of the total police force live in the city?\",\n y = \"Number of fatal shootings\",\n fill = \"Victim Armed?\")\n\n\n\n\nShow the code\np0\n\n\n\n\n\n\n\nShow the code\n#calculate mean number of shootings per agency in cities where a majority of officers reside in the city\nmajority_mean <- shootings_case |>\n filter(majority == TRUE) |>\n count(agency) |>\n summarize(maj_mean = mean(n))\n\n#calculate mean number of shootings per agency in cities where a minority of officers reside in the city\nminority_mean <- shootings_case |>\n filter(majority == FALSE) |>\n count(agency) |>\n summarize(min_mean = mean(n))\n\n#calculate a difference in means between the `majority` and `minority`\ndiff_in_means <- majority_mean - minority_mean\n\n\n\n\nShow the code\n#tidy table\nknitr::kable(head(diff_in_means))\n\n\n\n\n\nmaj_mean\n\n\n\n\n-2.575\n\n\n\n\n\n\n\nShow the code\n#fit single linear regression model for correlation between percentage of officer residency and number of fatal shootings per agency\nfit <- lm(n ~ all, data = agencies_census)\n\n#add `armed` and `majority` to `shootings_by_agency` df\nshootings_by_agency_census <- shootings_case |>\n group_by(agency) |>\n count(armed) |>\n drop_na(n, armed) |>\n right_join(agencies_census, by = c(\"agency\" = \"name\")) |>\n distinct(armed, .keep_all = TRUE)\n\nshootings_by_agency_census <- shootings_by_agency_census |>\n select(n.x, armed, all) \n\n#fit multiple linear regression model for correlation between percentage of officer residency and victim armament and number of fatal shootings per agency\nfit_multi <- lm(n.x ~ all + armed, data = shootings_by_agency_census)"
+ },
+ {
+ "objectID": "index.html#results",
+ "href": "index.html#results",
+ "title": "A Case Study on the Relationship between Police Residence and Fatal Police Shootings",
+ "section": "Results",
+ "text": "Results\n\nMultiple Linear Regression of relationship between percentage of officer residency and number of fatal shootings per agency fit\nThe model equation for fit is:\n\\[\n\\text{Number of Fatal Shootings (n)} = 35.7782 - 0.5874 \\times \\text{Percentage of Officer Residency (all)}\n\\]\n\n\nShow the code\n#tidy `fit`\np1 <- get_regression_table(fit)\nknitr::kable(head(p1))\n\n\n\n\n\n\n\n\n\n\n\n\n\n\nterm\nestimate\nstd_error\nstatistic\np_value\nlower_ci\nupper_ci\n\n\n\n\nintercept\n35.778\n6.887\n5.195\n0.000\n22.126\n49.430\n\n\nall\n-0.587\n14.902\n-0.039\n0.969\n-30.130\n28.955\n\n\n\n\n\nInterpretation:\n\nThe intercept, \\(35.7782\\), is the estimated number of fatal shootings when the percentage of officer in-city residency (all) is \\(0\\). For each one-unit increase in the percentage of officer residency, the number of fatal shootings is expected to decrease by \\(0.5874\\) (\\(-0.5874\\)) units, assuming all other factors remain constant.\n\nThis model suggests that there is a negative association between the percentage of officer residency and the number of fatal shootings. However, it’s important to interpret the results in the context of your data and consider potential confounding factors, like whether or not the victim was armed.\n\n\nShow the code\n#visualize polynomial relationship between percentage of officer residency and number of fatal shootings per agency\nggplot(data = shootings_by_agency_census, aes(x = all, y = n.x)) +\n geom_jitter(width = 0.10, height = 0, alpha = 0.45) +\n geom_smooth(method = \"lm\", formula = y ~ poly(x, 2), se = TRUE) +\n labs(title = \"Number of Shootings on a Scale of Police Force Residency\",\n x = \"Percentage of the total police force that lives in the city\",\n y = \"Number of fatal shootings in that city\")\n\n\n\n\n\n\n\nMultiple Linear Regression of relationship between percentage of officer residency/victim armament and number of fatal shootings per agency fit_multi\nThe model equation for fit_multi considering victim armament (armed) is:\n\\[\n\\text{\\ of Fatal Shootings (n.x)} = 4.117 + 1.211 \\times \\text{Percentage of Officer Residency (all)} + 24.921 \\times \\text{Armed (YES)}\n\\]\n\n\nShow the code\n#tidy `fit_multi`\np2 <- get_regression_table(fit_multi)\nknitr::kable(head(p2))\n\n\n\n\n\n\n\n\n\n\n\n\n\n\nterm\nestimate\nstd_error\nstatistic\np_value\nlower_ci\nupper_ci\n\n\n\n\nintercept\n4.117\n3.362\n1.225\n0.222\n-2.512\n10.747\n\n\nall\n1.211\n6.335\n0.191\n0.849\n-11.281\n13.702\n\n\narmed: YES\n24.921\n2.891\n8.619\n0.000\n19.219\n30.622\n\n\n\n\n\n\nThe intercept, \\(4.117\\), is the estimated number of fatal shootings where the percentage of officer in-city residency (all) is \\(0\\) and the victim was un-armed. For each one-unit increase in the percentage of in-city officer residency compared to the total force (all), we expect an increase of \\(1.211\\) fatal shootings, assuming the victim’s armament status (armedYES) remains constant.\nThe coefficient for ‘armedYES’, \\(24.921\\), indicates that the victim is armed (armed is YES), we expect an increase of \\(24.921\\) fatal shootings compared to when the victim is not armed (armed is No), assuming the percentage of officer residency (all) remains constant.\n\nIn summary, the model suggests that the percentage of officer residency and whether the victim is armed are associated with the number of fatal shootings per agency even as we control for victim armament. However, as correlation does not imply causation, and other factors not included in the model may influence the outcomes.\n\n\nShow the code\n#visualize polynomial relationship between percentage of officer residency and victim armament and number of fatal shootings per agency\nggplot(data = shootings_by_agency_census, aes(x = all, y = n.x, color = armed)) +\n geom_jitter(width = 0.10, height = 0, alpha = 0.45) +\n geom_smooth(method = \"lm\", formula = y ~ poly(x, 2), se = TRUE) +\n labs(title = \"Number of Shootings on a Scale of Police Force Residency\",\n x = \"Percentage of the total police force that lives in the city\",\n y = \"Number of fatal shootings in that city\",\n color = \"Victim Armed?\")\n\n\n\n\n\n\n\nShow the code\n#generate null distribution\nnull_dist <- agencies_census |>\n specify(n ~ majority) |>\n hypothesize(null = \"independence\") |>\n generate(reps = 1000, type = \"permute\") |>\n calculate(stat = \"diff in means\", order = c(\"TRUE\", \"FALSE\"))\n\n#compute observed test statistic\ntest_stat <- agencies_census |>\n specify(n ~ majority) |>\n calculate(stat = \"diff in means\", order = c(\"TRUE\", \"FALSE\"))\n\n#visualize p-value\nnull_dist |>\n visualize() +\n shade_p_value(obs_stat = test_stat, direction = \"less\")\n\n\n\n\n\nShow the code\n#compute p-value\n null_dist |>\n get_p_value(obs_stat = test_stat, direction = \"less\")\n\n\n# A tibble: 1 × 1\n p_value\n <dbl>\n1 0.252\n\n\nAt a significance level of \\(\\alpha = 0.05\\), the p-value of \\(0.248\\) suggests that, there is insufficient evidence to reject the null hypothesis. In this context, since our null hypothesis asserts that mean total number of fatal shootings per agencies does not differ based on if a majority of the officers live in the city or not, our p-value indicates that, assuming our null is true, the probability of observing our given test statistic (difference in means; \\(\\mu_{maj} − \\mu_{min}\\)) is \\(-4.92\\) is around \\(25\\%\\) (\\(0.248\\)). Meaning our observed difference in means between the groups is likely to have occurred by random chance.\n\n\nShow the code\n#generate null distribution\nnull_dist_cor <- agencies_census |>\n specify(n ~ white) |>\n hypothesize(null = \"independence\") |>\n generate(reps = 1000, type = \"permute\") |>\n calculate(stat = \"correlation\")\n\n#compute observed test statistic\ntest_stat_cor <- agencies_census |>\n specify(n ~ white) |>\n calculate(stat = \"correlation\")\n\n\n#visualize p-value\nnull_dist_cor |>\n visualize() +\n shade_p_value(obs_stat = test_stat, direction = \"two.sided\")\n\n\n\n\n\nShow the code\n#compute p-value\nnull_dist_cor |>\n get_p_value(obs_stat = test_stat, direction = \"two.sided\")\n\n\n# A tibble: 1 × 1\n p_value\n <dbl>\n1 0\n\n\nAt a significance level of \\(\\alpha = 0.05\\), the p-value of \\(0.248\\) suggests that, there is sufficient evidence to reject the null hypothesis. In this context, since our null hypothesis asserts that there is no relationship between percentage of the total police force that lives in the city they serve and number of fatal shootings, our p-value indicates that, assuming our null is true, the probability of observing our given test statistic (correlation coefficient; \\(\\rho = 0\\)) is \\(-0.0470\\) is around \\(0\\%\\) (\\(0\\)). Meaning our observed correlation coefficient likely would not happen if there was no relationship between percentage of officer residency and number of fatal shootings for a given agency."
+ },
+ {
+ "objectID": "index.html#conclusion",
+ "href": "index.html#conclusion",
+ "title": "A Case Study on the Relationship between Police Residence and Fatal Police Shootings",
+ "section": "Conclusion",
+ "text": "Conclusion\n\nGeneral Conclusions\n\n\nStudy Limitations\n\n\nImprovements for Future Study"
+ },
+ {
+ "objectID": "index.html#citations",
+ "href": "index.html#citations",
+ "title": "A Case Study on the Relationship between Police Residence and Fatal Police Shootings",
+ "section": "Citations",
+ "text": "Citations"
},
{
"objectID": "data/README-fatal-police-shoot.html",
@@ -67,13 +95,6 @@
"href": "data.html",
"title": "Data",
"section": "",
- "text": "library(tidyverse)\n\nWarning: package 'lubridate' was built under R version 4.3.1\n\n\n── Attaching core tidyverse packages ──────────────────────── tidyverse 2.0.0 ──\n✔ dplyr 1.1.3 ✔ readr 2.1.4\n✔ forcats 1.0.0 ✔ stringr 1.5.0\n✔ ggplot2 3.4.2 ✔ tibble 3.2.1\n✔ lubridate 1.9.3 ✔ tidyr 1.3.0\n✔ purrr 1.0.2 \n── Conflicts ────────────────────────────────────────── tidyverse_conflicts() ──\n✖ dplyr::filter() masks stats::filter()\n✖ dplyr::lag() masks stats::lag()\nℹ Use the conflicted package (<http://conflicted.r-lib.org/>) to force all conflicts to become errors\n\nlibrary(usmap)\nlibrary(sf)\n\nLinking to GEOS 3.11.0, GDAL 3.5.3, PROJ 9.1.0; sf_use_s2() is TRUE\n\nlibrary(infer)\nlibrary(moderndive)\n\n\n##Tidying Data\n\n#creating dfs from .csv files\npolice_locals <- read_csv(\"data/police-locals.csv\")\n\nRows: 75 Columns: 10\n── Column specification ────────────────────────────────────────────────────────\nDelimiter: \",\"\nchr (6): city_old, city, state, black, hispanic, asian\ndbl (4): police_force_size, all, white, non-white\n\nℹ Use `spec()` to retrieve the full column specification for this data.\nℹ Specify the column types or set `show_col_types = FALSE` to quiet this message.\n\nagencies <- read_csv(\"data/fatal-police-shootings-agencies.csv\")\n\nRows: 3422 Columns: 6\n── Column specification ────────────────────────────────────────────────────────\nDelimiter: \",\"\nchr (4): name, type, state, oricodes\ndbl (2): id, total_shootings\n\nℹ Use `spec()` to retrieve the full column specification for this data.\nℹ Specify the column types or set `show_col_types = FALSE` to quiet this message.\n\nshootings <- read_csv(\"data/fatal-police-shootings-data.csv\")\n\nRows: 9129 Columns: 19\n── Column specification ────────────────────────────────────────────────────────\nDelimiter: \",\"\nchr (12): threat_type, flee_status, armed_with, city, county, state, locati...\ndbl (4): id, latitude, longitude, age\nlgl (2): was_mental_illness_related, body_camera\ndate (1): date\n\nℹ Use `spec()` to retrieve the full column specification for this data.\nℹ Specify the column types or set `show_col_types = FALSE` to quiet this message.\n\n#removing old `city` tag from data set that we created when decatenated the city names\npolice_locals <- police_locals |>\n select(-city_old)\n\n# creating `agencies` df with just police departments\nagencies <- agencies |>\n filter(grepl(\"department\", tolower(name))) |>\n filter(!grepl(\"county\", tolower(name)))\n\n#creating binned categorical account of if shooting victim was `armed`\nshootings <- shootings |>\n mutate(armed = case_when(is.na(armed_with) ~ \"NO\",\n armed_with == \"unarmed\" ~ \"NO\",\n armed_with == \"unknown\" ~ \"NO\",\n armed_with == \"undetermined\" ~ \"NO\",\n armed_with == \"gun\" ~ \"YES\",\n armed_with == \"knife\" ~ \"YES\",\n armed_with == \"blunt_object\" ~ \"YES\",\n armed_with == \"other\" ~ \"YES\",\n armed_with == \"replica\" ~ \"YES\",\n armed_with == \"vehicle\" ~ \"YES\"))\n\n#creating df with only agency `names`, `id`, and `state`\nagencies_ids <- agencies |>\n select(name, id, state)\nagencies_ids\n\n# A tibble: 2,057 × 3\n name id state\n <chr> <dbl> <chr>\n 1 Aberdeen Police Department 2576 WA \n 2 Abilene Police Department 2114 TX \n 3 Abington Township Police Department 2088 PA \n 4 Acworth Police Department 3375 GA \n 5 Ada Police Department 2579 OK \n 6 Adel Police Department 3107 GA \n 7 Akron Police Department 815 OH \n 8 Alamogordo Police Department 1434 NM \n 9 Alamosa Police Department 2354 CO \n10 Albany Police Department 1443 GA \n# ℹ 2,047 more rows\n\n#creating df with `city`, `agency`, and `state` info for each shooting\nshooting_agencies <- shootings |>\n select(city, agency_ids, state)\nshooting_agencies\n\n# A tibble: 9,129 × 3\n city agency_ids state\n <chr> <chr> <chr>\n 1 Shelton 73 WA \n 2 Aloha 70 OR \n 3 Wichita 238 KS \n 4 San Francisco 196 CA \n 5 Evans 473 CO \n 6 Guthrie 101 OK \n 7 Chandler 195 AZ \n 8 Assaria 490 KS \n 9 Burlington 287 IA \n10 Knoxville 26254 PA \n# ℹ 9,119 more rows\n\n#changing `shooting` var in `shooting_agencies` df to numeric\nshooting_agencies$agency_ids <- as.numeric(shootings$agency_ids)\n\nWarning: NAs introduced by coercion\n\n#creating df with `city` and `state` info for each agency by joining `agencies_ids` and `shooting_agencies`\nagencies_w_cities <- agencies_ids |>\n left_join(shooting_agencies, by = c(\"id\" = \"agency_ids\", \"state\" = \"state\")) |>\n drop_na(city) |>\n distinct(id, .keep_all = TRUE)\nagencies_w_cities\n\n# A tibble: 1,781 × 4\n name id state city \n <chr> <dbl> <chr> <chr> \n 1 Aberdeen Police Department 2576 WA Aberdeen \n 2 Abilene Police Department 2114 TX Abilene \n 3 Abington Township Police Department 2088 PA Abington Township\n 4 Acworth Police Department 3375 GA Acworth \n 5 Ada Police Department 2579 OK Ada \n 6 Adel Police Department 3107 GA Adel \n 7 Akron Police Department 815 OH Akron \n 8 Alamogordo Police Department 1434 NM Alamogordo \n 9 Alamosa Police Department 2354 CO Alamosa \n10 Albany Police Department 1443 GA Albany \n# ℹ 1,771 more rows\n\n#creating df with census data for each agency by joining `agencies_w_cities` and `police_locals`\nagencies_census <- agencies_w_cities |>\n full_join(police_locals, by = c(\"city\" = \"city\", \"state\" = \"state\")) |>\n drop_na(police_force_size) |>\n distinct(id, .keep_all = TRUE) |>\n mutate(majority = if_else(all >= 0.5, \"TRUE\", \"FALSE\"))\nagencies_census\n\n# A tibble: 109 × 12\n name id state city police_force_size all white `non-white` black\n <chr> <dbl> <chr> <chr> <dbl> <dbl> <dbl> <dbl> <chr>\n 1 Albany P… 2237 NY Alba… 890 0.185 0.160 0.364 ** \n 2 Albuquer… 508 NM Albu… 1340 0.616 0.630 0.602 ** \n 3 Amtrak P… 1657 IL Chic… 12120 0.875 0.872 0.877 0.89…\n 4 Atlanta … 447 GA Atla… 2950 0.137 0.186 0.111 0.10…\n 5 Austin P… 141 TX Aust… 1985 0.295 0.195 0.427 0.25 \n 6 Baltimor… 4784 MD Balt… 2800 0.257 0.133 0.362 0.39…\n 7 Baltimor… 149 MD Balt… 2800 0.257 0.133 0.362 0.39…\n 8 BART Pol… 2015 CA Oakl… 1530 0.0948 0.0267 0.160 0.06…\n 9 Baton Ro… 1098 LA Bato… 980 0.214 0.144 0.321 0.34…\n10 Boston P… 3 MA Bost… 2560 0.477 0.442 0.583 0.68…\n# ℹ 99 more rows\n# ℹ 3 more variables: hispanic <chr>, asian <chr>, majority <chr>\n\n#creating df of only shootings involving agencies within `agencies` df\nshootings_case <- shootings |>\n right_join(agencies_census, by = c(\"city\" = \"city\", \"state\" = \"state\")) |>\n select(-agency_ids) |>\n rename(agency_ids = id.y, id = id.x, agency = name.y, victim = name.x) |>\n select(-location_precision, -race_source)\n\nWarning in right_join(shootings, agencies_census, by = c(city = \"city\", : Detected an unexpected many-to-many relationship between `x` and `y`.\nℹ Row 4 of `x` matches multiple rows in `y`.\nℹ Row 29 of `y` matches multiple rows in `x`.\nℹ If a many-to-many relationship is expected, set `relationship =\n \"many-to-many\"` to silence this warning.\n\nshootings_case\n\n# A tibble: 3,677 × 27\n id date threat_type flee_status armed_with city county state\n <dbl> <date> <chr> <chr> <chr> <chr> <chr> <chr>\n 1 5 2015-01-03 move not unarmed Wichita Sedgw… KS \n 2 8 2015-01-04 point not replica San Francis… San F… CA \n 3 8 2015-01-04 point not replica San Francis… San F… CA \n 4 22 2015-01-07 threat not knife Columbus Frank… OH \n 5 22 2015-01-07 threat not knife Columbus Frank… OH \n 6 27 2015-01-07 shoot foot gun New Orleans Orlea… LA \n 7 325 2015-01-09 point not gun El Paso El Pa… TX \n 8 46 2015-01-13 shoot foot gun Albuquerque Berna… NM \n 9 46 2015-01-13 shoot foot gun Albuquerque Berna… NM \n10 56 2015-01-15 shoot foot gun Indianapolis Marion IN \n# ℹ 3,667 more rows\n# ℹ 19 more variables: latitude <dbl>, longitude <dbl>, victim <chr>,\n# age <dbl>, gender <chr>, race <chr>, was_mental_illness_related <lgl>,\n# body_camera <lgl>, armed <chr>, agency <chr>, agency_ids <dbl>,\n# police_force_size <dbl>, all <dbl>, white <dbl>, `non-white` <dbl>,\n# black <chr>, hispanic <chr>, asian <chr>, majority <chr>\n\n\n\nshootings_by_agency <- shootings_case |>\n count(agency)\nshootings_by_agency\n\n# A tibble: 108 × 2\n agency n\n <chr> <int>\n 1 Albany Police Department 1\n 2 Albuquerque Police Department 66\n 3 Amtrak Police Department 54\n 4 Atlanta Police Department 36\n 5 Austin Police Department 40\n 6 BART Police Department 13\n 7 Baltimore City Police Department 30\n 8 Baltimore Police Department 30\n 9 Baton Rouge Police Department 15\n10 Boston Police Department 10\n# ℹ 98 more rows\n\nggplot(data = shootings_case,\n mapping = aes(x = agency)) +\n geom_bar() +\n theme(axis.text.x = element_text(angle = 90,\n vjust = 1,\n hjust = 1,\n margin = margin(t = 5, b = 5)))\n\n\n\n\n\n#mapping Locations of Police-Involved Shootings between 2015 and 2023\n\nlibrary(ggmap)\n\nWarning: package 'ggmap' was built under R version 4.3.1\n\n\nℹ Google's Terms of Service: <https://mapsplatform.google.com>\n Stadia Maps' Terms of Service: <https://stadiamaps.com/terms-of-service/>\n OpenStreetMap's Tile Usage Policy: <https://operations.osmfoundation.org/policies/tiles/>\nℹ Please cite ggmap if you use it! Use `citation(\"ggmap\")` for details.\n\nlibrary(maps)\n\nWarning: package 'maps' was built under R version 4.3.1\n\n\n\nAttaching package: 'maps'\n\nThe following object is masked from 'package:purrr':\n\n map\n\nlibrary(mapdata)\n\nusa <- map_data(\"usa\")\nstates <- map_data(\"state\")\n\nggplot(data = states) + \n geom_polygon(aes(x = long, y = lat, fill = group, group = group), color = \"white\") + \n coord_fixed(1.3) +\n guides(fill=FALSE) + # do this to leave off the color legend\n geom_point(data = shootings_case, aes(x = longitude, y = latitude), color = \"black\", size = .2) +\n geom_point(data = shootings_case, aes(x = longitude, y = latitude), color = \"red\", size = .1) +\n labs(title = \"Locations of Police-Involved Shootings between 2015 and 2023\",\n captions = \"This is only includes cities where we have agency census data.\",\n x = \"Longitude\",\n y = \"Latitude\")\n\nWarning: The `<scale>` argument of `guides()` cannot be `FALSE`. Use \"none\" instead as\nof ggplot2 3.3.4.\n\n\nWarning: Removed 309 rows containing missing values (`geom_point()`).\nRemoved 309 rows containing missing values (`geom_point()`).\n\n\n\n\n\n\nagencies_census <- agencies_census |>\n left_join(shootings_by_agency, by = c(\"name\" = \"agency\"))\nagencies_census\n\n# A tibble: 109 × 13\n name id state city police_force_size all white `non-white` black\n <chr> <dbl> <chr> <chr> <dbl> <dbl> <dbl> <dbl> <chr>\n 1 Albany P… 2237 NY Alba… 890 0.185 0.160 0.364 ** \n 2 Albuquer… 508 NM Albu… 1340 0.616 0.630 0.602 ** \n 3 Amtrak P… 1657 IL Chic… 12120 0.875 0.872 0.877 0.89…\n 4 Atlanta … 447 GA Atla… 2950 0.137 0.186 0.111 0.10…\n 5 Austin P… 141 TX Aust… 1985 0.295 0.195 0.427 0.25 \n 6 Baltimor… 4784 MD Balt… 2800 0.257 0.133 0.362 0.39…\n 7 Baltimor… 149 MD Balt… 2800 0.257 0.133 0.362 0.39…\n 8 BART Pol… 2015 CA Oakl… 1530 0.0948 0.0267 0.160 0.06…\n 9 Baton Ro… 1098 LA Bato… 980 0.214 0.144 0.321 0.34…\n10 Boston P… 3 MA Bost… 2560 0.477 0.442 0.583 0.68…\n# ℹ 99 more rows\n# ℹ 4 more variables: hispanic <chr>, asian <chr>, majority <chr>, n <int>\n\nagencies_census |>\n ggplot(mapping = aes(x = all, y = n, fill=)) +\n geom_point() +\n geom_smooth(method = \"lm\", formula = y ~ poly(x, 2), se = FALSE)\n\n\n\nshootings_case |>\n ggplot(aes(x = majority, fill = armed)) +\n geom_bar() + \n labs(title = \"Shootings in Cities where a Majority of Officers Reside\",\n captions = \"This is only includes shootings where we have agency census data.\",\n x = \"Does a majority a of the total police force live in the city?\",\n y = \"Number of fatal shootings\",\n fill = \"Victim Armed?\")\n\n\n\nmajority_mean <- shootings_case |>\n filter(majority == TRUE) |>\n count(agency) |>\n summarize(maj_mean = mean(n))\nmajority_mean\n\n# A tibble: 1 × 1\n maj_mean\n <dbl>\n1 32.4\n\nminority_mean <- shootings_case |>\n filter(majority == FALSE) |>\n count(agency) |>\n summarize(min_mean = mean(n))\nminority_mean\n\n# A tibble: 1 × 1\n min_mean\n <dbl>\n1 35\n\ndiff_in_means <- majority_mean - minority_mean\ndiff_in_means\n\n maj_mean\n1 -2.575\n\nknitr::kable(head(diff_in_means))\n\n\n\n\nmaj_mean\n\n\n\n\n-2.575\n\n\n\n\n\n\nfit <- lm(n ~ all, data = agencies_census)\nfit\n\n\nCall:\nlm(formula = n ~ all, data = agencies_census)\n\nCoefficients:\n(Intercept) all \n 35.7782 -0.5874 \n\np1 <- get_regression_table(fit)\nknitr::kable(head(p1))\n\n\n\n\n\n\n\n\n\n\n\n\n\nterm\nestimate\nstd_error\nstatistic\np_value\nlower_ci\nupper_ci\n\n\n\n\nintercept\n35.778\n6.887\n5.195\n0.000\n22.126\n49.430\n\n\nall\n-0.587\n14.902\n-0.039\n0.969\n-30.130\n28.955\n\n\n\n\nshootings_by_agency_census <- shootings_case %>%\n group_by(agency) %>%\n count(armed) %>%\n drop_na(n, armed) %>%\n right_join(agencies_census, by = c(\"agency\" = \"name\")) |>\n distinct(armed, .keep_all = TRUE)\n\nWarning in right_join(., agencies_census, by = c(agency = \"name\")): Detected an unexpected many-to-many relationship between `x` and `y`.\nℹ Row 63 of `x` matches multiple rows in `y`.\nℹ Row 2 of `y` matches multiple rows in `x`.\nℹ If a many-to-many relationship is expected, set `relationship =\n \"many-to-many\"` to silence this warning.\n\nshootings_by_agency_census\n\n# A tibble: 202 × 15\n# Groups: agency [108]\n agency armed n.x id state city police_force_size all white\n <chr> <chr> <int> <dbl> <chr> <chr> <dbl> <dbl> <dbl>\n 1 Albany Police … YES 1 2237 NY Alba… 890 0.185 0.160 \n 2 Albuquerque Po… NO 10 508 NM Albu… 1340 0.616 0.630 \n 3 Albuquerque Po… YES 54 508 NM Albu… 1340 0.616 0.630 \n 4 Amtrak Police … NO 8 1657 IL Chic… 12120 0.875 0.872 \n 5 Amtrak Police … YES 46 1657 IL Chic… 12120 0.875 0.872 \n 6 Atlanta Police… NO 5 447 GA Atla… 2950 0.137 0.186 \n 7 Atlanta Police… YES 31 447 GA Atla… 2950 0.137 0.186 \n 8 Austin Police … NO 3 141 TX Aust… 1985 0.295 0.195 \n 9 Austin Police … YES 37 141 TX Aust… 1985 0.295 0.195 \n10 BART Police De… YES 12 2015 CA Oakl… 1530 0.0948 0.0267\n# ℹ 192 more rows\n# ℹ 6 more variables: `non-white` <dbl>, black <chr>, hispanic <chr>,\n# asian <chr>, majority <chr>, n.y <int>\n\nshootings_by_agency_census <- shootings_by_agency_census |>\n select(n.x, armed, all) \n\nAdding missing grouping variables: `agency`\n\nfit_multi <- lm(n.x ~ all + armed, data = shootings_by_agency_census)\nfit_multi\n\n\nCall:\nlm(formula = n.x ~ all + armed, data = shootings_by_agency_census)\n\nCoefficients:\n(Intercept) all armedYES \n 4.117 1.211 24.921 \n\np2 <- get_regression_table(fit_multi)\nknitr::kable(head(p2))\n\n\n\n\n\n\n\n\n\n\n\n\n\nterm\nestimate\nstd_error\nstatistic\np_value\nlower_ci\nupper_ci\n\n\n\n\nintercept\n4.117\n3.362\n1.225\n0.222\n-2.512\n10.747\n\n\nall\n1.211\n6.335\n0.191\n0.849\n-11.281\n13.702\n\n\narmed: YES\n24.921\n2.891\n8.619\n0.000\n19.219\n30.622\n\n\n\n\nggplot(data = shootings_by_agency_census, aes(x = all, y = n.x)) +\n geom_jitter(jitter = 15, alpha = 0.5) +\n geom_smooth(method = \"lm\", formula = y ~ poly(x, 2), se = FALSE) +\n labs(title = \"Number of Shootings on a Scale of Police Force Residency\",\n x = \"Percentage of the total police force that lives in the city\",\n y = \"Number of fatal shootings in that city\")\n\nWarning in geom_jitter(jitter = 15, alpha = 0.5): Ignoring unknown parameters:\n`jitter`\n\n\n\n\nggplot(data = shootings_by_agency_census, aes(x = all, y = n.x, color = armed)) +\n geom_jitter(jitter = 15, alpha = 0.5) +\n geom_smooth(method = \"lm\", formula = y ~ poly(x, 2), se = FALSE) +\n labs(title = \"Number of Shootings on a Scale of Police Force Residency\",\n x = \"Percentage of the total police force that lives in the city\",\n y = \"Number of fatal shootings in that city\")\n\nWarning in geom_jitter(jitter = 15, alpha = 0.5): Ignoring unknown parameters:\n`jitter`"
- },
- {
- "objectID": "background.html#the-washington-post-fatal-force-database",
- "href": "background.html#the-washington-post-fatal-force-database",
- "title": "Background",
- "section": "The Washington Post Fatal Force Database",
- "text": "The Washington Post Fatal Force Database\nIn 2015, The Washington Post began tracking details about each police-involved killing in the United States — the race of the deceased, the circumstances of the shooting, whether the person was armed and whether the person was experiencing a mental-health crisis — by manually culling local news reports, collecting information from law enforcement websites and social media, and monitoring independent databases such as Fatal Encounters and the now-defunct Killed by Police project. In many cases, The Post conducts additional reporting.\nIn 2022, The Post updated its database to standardize and publish the names of the police agencies involved in each shooting to better measure accountability at the department level.\nThe 2014 killing of Michael Brown in Ferguson, Mo. began a protest movement culminating in the Black Lives Matter movement and an increased focus on police accountability nationwide. In this data set, The Post tracks only shootings with circumstances closely paralleling those like the killing of Brown — incidents in which a police officer, in the line of duty, shoots and kills a civilian. The Post is not tracking deaths of people in police custody, fatal shootings by off-duty officers or non-shooting deaths in this data set.\nThe FBI and the Centers for Disease Control and Prevention log fatal shootings by police, but officials acknowledge that their data is incomplete. Since 2015, The Post has documented more than twice as many fatal shootings by police as recorded by federal officials on average annually. That gap has widened in recent years, as the FBI in 2021 tracked only a third of departments’ fatal shootings."
+ "text": "library(tidyverse)\nlibrary(usmap)\nlibrary(sf)\nlibrary(infer)\nlibrary(moderndive)\n\n\n##Tidying Data\n\n#creating dfs from .csv files\npolice_locals <- read_csv(\"data/police-locals.csv\")\nagencies <- read_csv(\"data/fatal-police-shootings-agencies.csv\")\nshootings <- read_csv(\"data/fatal-police-shootings-data.csv\")\n\n#removing old `city` tag from data set that we created when decatenated the city names\npolice_locals <- police_locals |>\n select(-city_old)\n\n# creating `agencies` df with just police departments\nagencies <- agencies |>\n filter(grepl(\"department\", tolower(name))) |>\n filter(!grepl(\"county\", tolower(name)))\n\n#creating binned categorical account of if shooting victim was `armed`\nshootings <- shootings |>\n mutate(armed = case_when(is.na(armed_with) ~ \"NO\",\n armed_with == \"unarmed\" ~ \"NO\",\n armed_with == \"unknown\" ~ \"NO\",\n armed_with == \"undetermined\" ~ \"NO\",\n armed_with == \"gun\" ~ \"YES\",\n armed_with == \"knife\" ~ \"YES\",\n armed_with == \"blunt_object\" ~ \"YES\",\n armed_with == \"other\" ~ \"YES\",\n armed_with == \"replica\" ~ \"YES\",\n armed_with == \"vehicle\" ~ \"YES\"))\n\n#creating df with only agency `names`, `id`, and `state`\nagencies_ids <- agencies |>\n select(name, id, state)\n\n#creating df with `city`, `agency`, and `state` info for each shooting\nshooting_agencies <- shootings |>\n select(city, agency_ids, state)\n\n#changing `shooting` var in `shooting_agencies` df to numeric\nshooting_agencies$agency_ids <- as.numeric(shootings$agency_ids)\n\n#creating df with `city` and `state` info for each agency by joining `agencies_ids` and `shooting_agencies`\nagencies_w_cities <- agencies_ids |>\n left_join(shooting_agencies, by = c(\"id\" = \"agency_ids\", \"state\" = \"state\")) |>\n drop_na(city) |>\n distinct(id, .keep_all = TRUE)\n\n#creating df with census data for each agency by joining `agencies_w_cities` and `police_locals`\nagencies_census <- agencies_w_cities |>\n full_join(police_locals, by = c(\"city\" = \"city\", \"state\" = \"state\")) |>\n drop_na(police_force_size) |>\n distinct(id, .keep_all = TRUE) |>\n mutate(majority = if_else(all >= 0.5, \"TRUE\", \"FALSE\"))\n\n#creating df of only shootings involving agencies within `agencies` df\nshootings_case <- shootings |>\n right_join(agencies_census, by = c(\"city\" = \"city\", \"state\" = \"state\")) |>\n select(-agency_ids) |>\n rename(agency_ids = id.y, id = id.x, agency = name.y, victim = name.x) |>\n select(-location_precision, -race_source)\n\n\n#count shootings by agency\nshootings_by_agency <- shootings_case |>\n count(agency)\n\n#find top 25 agencies with the most shootings\ntop_25_agencies <- shootings_by_agency |>\n slice_max(n, n = 25)\n\n# visulize top 25 agencies with the most shootings\nggplot(data = top_25_agencies,\n mapping = aes(x = agency, y = n)) +\n geom_col() +\n theme(axis.text.x = element_text(angle = 75,\n vjust = 1,\n hjust = 1,\n margin = margin(t = 5, b = 5)))\n\n\n\n\n\n#mapping Locations of Police-Involved Shootings between 2015 and 2023\n\n#load geo-viz libraries\nlibrary(ggmap)\nlibrary(maps)\nlibrary(mapdata)\n\n#create blank map\nusa <- map_data(\"usa\")\nstates <- map_data(\"state\")\n\n#add locations of shootings to maps\nshot_map <- ggplot(data = states) + \n geom_polygon(aes(x = long, y = lat, fill = group, group = group), color = \"white\") + \n coord_fixed(1.3) +\n guides(fill=FALSE) + # do this to leave off the color legend\n geom_point(data = shootings_case, aes(x = longitude, y = latitude), color = \"black\", size = .2) +\n geom_point(data = shootings_case, aes(x = longitude, y = latitude), color = \"red\", size = .1) +\n labs(title = \"Locations of Police-Involved Shootings between 2015 and 2023\",\n captions = \"This is only includes cities where we have agency census data.\",\n x = \"Longitude\",\n y = \"Latitude\")\n\n\n#creating df with total shootings per agency and census data\nagencies_census <- agencies_census |>\n left_join(shootings_by_agency, by = c(\"name\" = \"agency\"))\n\n#prelim visualization of relationship between percentage of officer residency and number of fatal shootings per agency\nagencies_census |>\n ggplot(mapping = aes(x = all, y = n)) +\n geom_point() +\n geom_smooth(method = \"lm\", se = TRUE)\n\n\n\n#creating visualization of comparison Shootings in Cities where a Majority/Minority of Officers Reside\np0 <- shootings_case |>\n ggplot(aes(x = majority, fill = armed)) +\n geom_bar() + \n labs(title = \"Shootings in Cities where a Majority of Officers Reside\",\n caption = \"This is only includes shootings where we have agency census data.\",\n x = \"Does a majority a of the total police force live in the city?\",\n y = \"Number of fatal shootings\",\n fill = \"Victim Armed?\")\np0\n\n\n\n#calculate mean number of shootings per agency in cities where a majority of officers reside in the city\nmajority_mean <- shootings_case |>\n filter(majority == TRUE) |>\n count(agency) |>\n summarize(maj_mean = mean(n))\n\n#calculate mean number of shootings per agency in cities where a minority of officers reside in the city\nminority_mean <- shootings_case |>\n filter(majority == FALSE) |>\n count(agency) |>\n summarize(min_mean = mean(n))\n\n#calculate a difference in means between the `majority` and `minority`\ndiff_in_means <- majority_mean - minority_mean\ndiff_in_means\n\n maj_mean\n1 -2.575\n\n#tidy table\nknitr::kable(head(diff_in_means))\n\n\n\n\nmaj_mean\n\n\n\n\n-2.575\n\n\n\n\n\n\n#fit single linear regression model for correlation between percentage of officer residency and number of fatal shootings per agency\nfit <- lm(n ~ all, data = agencies_census)\nfit\n\n\nCall:\nlm(formula = n ~ all, data = agencies_census)\n\nCoefficients:\n(Intercept) all \n 35.7782 -0.5874 \n\n#tidy `fit`\np1 <- get_regression_table(fit)\nknitr::kable(head(p1))\n\n\n\n\n\n\n\n\n\n\n\n\n\nterm\nestimate\nstd_error\nstatistic\np_value\nlower_ci\nupper_ci\n\n\n\n\nintercept\n35.778\n6.887\n5.195\n0.000\n22.126\n49.430\n\n\nall\n-0.587\n14.902\n-0.039\n0.969\n-30.130\n28.955\n\n\n\n\n#add `armed` and `majority` to `shootings_by_agency` df\nshootings_by_agency_census <- shootings_case |>\n group_by(agency) |>\n count(armed) |>\n drop_na(n, armed) |>\n right_join(agencies_census, by = c(\"agency\" = \"name\")) |>\n distinct(armed, .keep_all = TRUE)\n\nshootings_by_agency_census <- shootings_by_agency_census |>\n select(n.x, armed, all) \n\n#fit multiple linear regression model for correlation between percentage of officer residency and victim armament and number of fatal shootings per agency\nfit_multi <- lm(n.x ~ all + armed, data = shootings_by_agency_census)\nfit_multi\n\n\nCall:\nlm(formula = n.x ~ all + armed, data = shootings_by_agency_census)\n\nCoefficients:\n(Intercept) all armedYES \n 4.117 1.211 24.921 \n\n#tidy `fit_multi`\np2 <- get_regression_table(fit_multi)\nknitr::kable(head(p2))\n\n\n\n\n\n\n\n\n\n\n\n\n\nterm\nestimate\nstd_error\nstatistic\np_value\nlower_ci\nupper_ci\n\n\n\n\nintercept\n4.117\n3.362\n1.225\n0.222\n-2.512\n10.747\n\n\nall\n1.211\n6.335\n0.191\n0.849\n-11.281\n13.702\n\n\narmed: YES\n24.921\n2.891\n8.619\n0.000\n19.219\n30.622\n\n\n\n\n#visualize polynomial relationship between percentage of officer residency and number of fatal shootings per agency\nggplot(data = shootings_by_agency_census, aes(x = all, y = n.x)) +\n geom_jitter(width = 0.10, height = 0, alpha = 0.45) +\n geom_smooth(method = \"lm\", formula = y ~ poly(x, 2), se = TRUE) +\n labs(title = \"Number of Shootings on a Scale of Police Force Residency\",\n x = \"Percentage of the total police force that lives in the city\",\n y = \"Number of fatal shootings in that city\")\n\n\n\n#visualize polynomial relationship between percentage of officer residency and victim armament and number of fatal shootings per agency\nggplot(data = shootings_by_agency_census, aes(x = all, y = n.x, color = armed)) +\n geom_jitter(width = 0.10, height = 0, alpha = 0.45) +\n geom_smooth(method = \"lm\", formula = y ~ poly(x, 2), se = TRUE) +\n labs(title = \"Number of Shootings on a Scale of Police Force Residency\",\n x = \"Percentage of the total police force that lives in the city\",\n y = \"Number of fatal shootings in that city\",\n color = \"Victim Armed?\")\n\n\n\n\nThe model equation for fit is:\n[ = 35.7782 - 0.5874 ]\nInterpretation:\n\nThe intercept, \\(35.7782\\), is the estimated number of fatal shootings when the percentage of officer in-city residency (all) is \\(0\\). For each one-unit increase in the percentage of officer residency, the number of fatal shootings is expected to decrease by \\(0.5874\\) (\\(-0.5874\\)) units, assuming all other factors remain constant.\n\nThis model suggests that there is a negative association between the percentage of officer residency and the number of fatal shootings. However, it’s important to interpret the results in the context of your data and consider potential confounding factors, like whether or not the victim was armed.\nThe model equation for fit_multi considering victim armament (armed) is:\n[ = 4.117 + 1.211 + 24.921 ]\n\nThe intercept, \\(4.117\\), is the estimated number of fatal shootings where the percentage of officer in-city residency (all) is \\(0\\) and the victim was un-armed. For each one-unit increase in the percentage of in-city officer residency compared to the total force (all), we expect an increase of \\(1.211\\) fatal shootings, assuming the victim’s armament status (armedYES) remains constant.\nThe coefficient for ‘armedYES’, \\(24.921\\), indicates that the victim is armed (armed is YES), we expect an increase of \\(24.921\\) fatal shootings compared to when the victim is not armed (armed is No), assuming the percentage of officer residency (all) remains constant.\n\nIn summary, the model suggests that the percentage of officer residency and whether the victim is armed are associated with the number of fatal shootings per agency even as we control for victim armament. However, as correlation does not imply causation, and other factors not included in the model may influence the outcomes.\n\n#generate null distribution\nnull_dist <- agencies_census |>\n specify(n ~ majority) |>\n hypothesize(null = \"independence\") |>\n generate(reps = 1000, type = \"permute\") |>\n calculate(stat = \"diff in means\", order = c(\"TRUE\", \"FALSE\"))\n\n#compute observed test statistic\ntest_stat <- agencies_census |>\n specify(n ~ majority) |>\n calculate(stat = \"diff in means\", order = c(\"TRUE\", \"FALSE\"))\n\n#visualize p-value\nnull_dist |>\n visualize() +\n shade_p_value(obs_stat = test_stat, direction = \"less\")\n\n\n\n#compute p-value\n null_dist |>\n get_p_value(obs_stat = test_stat, direction = \"less\")\n\n# A tibble: 1 × 1\n p_value\n <dbl>\n1 0.263\n\n\nInference for a Difference in Means\n\n\\(H_0\\): The mean total number of fatal shootings per agencies does not differ based on if a majority of the officers live in the city or not.\n\\(H_A\\): The mean total number of fatal shootings per agencies is fewer in cities where a majority of the officers live in the city then cities where they do not.\n\n– \\(H_0 : \\mu_{maj} − \\mu_{min} = 0\\), or equivalently \\(H_0 : \\mu_{maj} = \\mu_{min}\\) – \\(H_A : \\mu_{maj} − \\mu_{min} < 0\\), or equivalently \\(H_A : \\mu_{maj} < \\mu_{min}\\)\n\n#generate null distribution\nnull_dist_cor <- agencies_census |>\n specify(n ~ white) |>\n hypothesize(null = \"independence\") |>\n generate(reps = 1000, type = \"permute\") |>\n calculate(stat = \"correlation\")\n\n#compute observed test statistic\ntest_stat_cor <- agencies_census |>\n specify(n ~ white) |>\n calculate(stat = \"correlation\")\ntest_stat_cor\n\nResponse: n (numeric)\nExplanatory: white (numeric)\n# A tibble: 1 × 1\n stat\n <dbl>\n1 -0.0470\n\n#visualize p-value\nnull_dist_cor |>\n visualize() +\n shade_p_value(obs_stat = test_stat, direction = \"two.sided\")\n\n\n\n#compute p-value\nnull_dist_cor |>\n get_p_value(obs_stat = test_stat, direction = \"two.sided\")\n\n# A tibble: 1 × 1\n p_value\n <dbl>\n1 0\n\n\nInference for a Correlation\n\n\\(H_O\\): There is no relationship between percentage of the total police force that lives in the city they serve and number of fatal shootings.\n\\(H_A\\): There is a relationship between percentage of the total police force that lives in the city they serve and number of fatal shootings.\n\n\\(H_0 : \\rho = 0\\)\n\\(H_0 : \\rho \\neq 0\\)"
}
]
\ No newline at end of file
diff --git a/about.qmd b/about.qmd
deleted file mode 100644
index 0c6f4a2..0000000
--- a/about.qmd
+++ /dev/null
@@ -1,9 +0,0 @@
----
-title: "About"
----
-
-About this site
-
-```{r}
-1 + 1
-```
diff --git a/background.qmd b/background.qmd
index 1b759fb..3a977ec 100644
--- a/background.qmd
+++ b/background.qmd
@@ -4,7 +4,54 @@ title: "Background"
> # On average, police in the United States shoot and kill more than 1,000 people every year, according to an ongoing analysis by The Washington Post. {style="red"}
-## The Washington Post Fatal Force Database
+# Proposal
+
+We propose a case study to explore the relationship between police residence and fatal police shootings, employing advanced data science methodologies. Focusing on officers residing in the cities they serve, our project aims to uncover insights and patterns that contribute to a nuanced understanding of this complex issue.
+
+### **Objectives:**
+
+1. Investigate the correlation between police residence and fatal police shootings.
+2. Utilize a comprehensive dataset spanning 2015 to 2023, focusing on police agencies involved in at least one fatal shooting.
+3. Apply advanced statistical methods and machine learning techniques to identify patterns and potential biases.
+4. Examine disparities in incident rates based on officers' residency status, considering demographic, socioeconomic, and policing variables.
+
+**Methodology:**
+
+a. Data Collection:
+ - Compile a dataset comprising information on police agencies involved in fatal police shootings.
+ - Compile a data set of census variables such as officer residency, race, community demographics, and departmental policies.
+b. Analysis:
+ - Employ advanced statistical methods and machine learning techniques to discern patterns and correlations.
+ - Conduct a comprehensive exploration of variables influencing fatal police shootings.
+
+**Hypothesis and Expected Outcomes:**
+
+We will conduct two hypothesis tests to analyze both;
+
+- the nominal relationship between an increasing proportion of in-city officer residency and number of fatal police shooting deaths and
+
+- the categorical difference in fatal police shooting deaths between cities where a majority or or minority of police officers live in the city.
+
+1. Inference for a Difference in Proportions
+
+ - $H_0$: The mean total number of fatal shootings per agencies does not differ based on if a majority of the officers live in the city or not.
+
+ - $H_A$: The mean total number of fatal shootings per agencies is fewer in cities where a majority of the officers live in the city then cities where they do not.
+
+ - $H_0 : p\_{maj} − p\_{min} = 0$, or equivalently $H_0 : p\_{maj} = p\_{min}$
+ - $H_A : p\_{maj} − p\_{min} < 0$, or equivalently $H_A : p\_{maj} < p\_{min}$
+
+2. Inference for a Correlation
+
+ - $H_O$: There is no relationship between percentage of the total police force that lives in the city they serve and number of fatal shootings.
+
+ - $H_A$: There is a relationship between percentage of the total police force that lives in the city they serve and number of fatal shootings.
+
+ - $H_0 : \rho = 0$
+
+ - $H_0 : \rho \neq 0$
+
+# The Washington Post Fatal Force Database
In 2015, The Washington Post [began tracking](https://www.washingtonpost.com/graphics/investigations/police-shootings-database/) details about each police-involved killing in the United States --- the race of the deceased, the circumstances of the shooting, whether the person was armed and whether the person was experiencing a mental-health crisis --- by manually culling local news reports, collecting information from law enforcement websites and social media, and monitoring independent databases such as [Fatal Encounters](https://fatalencounters.org/) and the now-defunct Killed by Police project. In many cases, The Post conducts additional reporting.
diff --git a/codebook.qmd b/codebook.qmd
new file mode 100644
index 0000000..370941e
--- /dev/null
+++ b/codebook.qmd
@@ -0,0 +1,44 @@
+---
+title: "Codebook"
+---
+
+## Explanatory Variables
+
+| Name | Description |
+|-------------------------|-----------------------------------------------|
+| `city` | U.S. city |
+| `police_force_size` | Number of police officers serving that city |
+| `all` | Percentage of the total police force that lives in the city |
+| `white` | Percentage of white (non-Hispanic) police officers who live in the city |
+| `non-white` | Percentage of non-white police officers who live in the city |
+| `black` | Percentage of black police officers who live in the city |
+| `hispanic` | Percentage of Hispanic police officers who live in the city |
+| `asian` | Percentage of Asian police officers who live in the city |
+
+### Incident Information
+
+| Name | Description |
+|-------------------------|------------------------------------------------|
+| `id` | A unique identifier for each fatal police shooting incident. |
+| `date` | The date of the fatal shooting. |
+| `body_camera` | Whether news reports have indicated an officer was wearing a body camera and it may have recorded some portion of the incident. |
+| `city` | The municipality where the fatal shooting took place |
+| `county` | County where the fatal shooting took place. |
+| `state` | The two-letter postal code abbreviation for the state in which the fatal shooting took place. |
+| `latitude` | The latitude location of the shooting expressed as WGS84 coordinates, geocoded from addresses. Please note that the precision and accuracy of incident coordinates varies depending on the precision of the input address which is often only available at the block level. |
+| `longitude` | The longitude location of the shooting expressed as WGS84 coordinates, geocoded from addresses. |
+
+### Agency Information
+
+| | Description |
+|---------|---------------------------------------|
+| `id` | Department Database Id |
+| `name` | Department Name |
+| `state` | State in which the agency is located. |
+
+## Project thoughts
+
+I am interested in exploring data related to...
+
+- Political Extremism
+- Black American Opinion
diff --git a/data.qmd b/data.qmd
index 2323c87..a4a5619 100644
--- a/data.qmd
+++ b/data.qmd
@@ -2,7 +2,11 @@
title: "Data"
---
-```{r}
+```{r setup, include=FALSE}
+knitr::opts_chunk$set(warning = FALSE, message = FALSE)
+```
+
+```{r pacakges}
library(tidyverse)
library(usmap)
library(sf)
@@ -10,8 +14,7 @@ library(infer)
library(moderndive)
```
-
-```{r Tidyin}
+```{r Tidying and Wrangling Data}
##Tidying Data
@@ -45,12 +48,10 @@ shootings <- shootings |>
#creating df with only agency `names`, `id`, and `state`
agencies_ids <- agencies |>
select(name, id, state)
-agencies_ids
#creating df with `city`, `agency`, and `state` info for each shooting
shooting_agencies <- shootings |>
select(city, agency_ids, state)
-shooting_agencies
#changing `shooting` var in `shooting_agencies` df to numeric
shooting_agencies$agency_ids <- as.numeric(shootings$agency_ids)
@@ -60,7 +61,6 @@ agencies_w_cities <- agencies_ids |>
left_join(shooting_agencies, by = c("id" = "agency_ids", "state" = "state")) |>
drop_na(city) |>
distinct(id, .keep_all = TRUE)
-agencies_w_cities
#creating df with census data for each agency by joining `agencies_w_cities` and `police_locals`
agencies_census <- agencies_w_cities |>
@@ -68,7 +68,6 @@ agencies_census <- agencies_w_cities |>
drop_na(police_force_size) |>
distinct(id, .keep_all = TRUE) |>
mutate(majority = if_else(all >= 0.5, "TRUE", "FALSE"))
-agencies_census
#creating df of only shootings involving agencies within `agencies` df
shootings_case <- shootings |>
@@ -76,21 +75,18 @@ shootings_case <- shootings |>
select(-agency_ids) |>
rename(agency_ids = id.y, id = id.x, agency = name.y, victim = name.x) |>
select(-location_precision, -race_source)
-shootings_case
```
-```{r}
+```{r Counting Shootings}
#count shootings by agency
shootings_by_agency <- shootings_case |>
count(agency)
-shootings_by_agency
#find top 25 agencies with the most shootings
top_25_agencies <- shootings_by_agency |>
slice_max(n, n = 25)
-top_25_agencies
# visulize top 25 agencies with the most shootings
ggplot(data = top_25_agencies,
@@ -102,18 +98,22 @@ ggplot(data = top_25_agencies,
margin = margin(t = 5, b = 5)))
```
-```{r}
+
+```{r shot_map}
#mapping Locations of Police-Involved Shootings between 2015 and 2023
+#load geo-viz libraries
library(ggmap)
library(maps)
library(mapdata)
+#create blank map
usa <- map_data("usa")
states <- map_data("state")
-ggplot(data = states) +
+#add locations of shootings to maps
+shot_map <- ggplot(data = states) +
geom_polygon(aes(x = long, y = lat, fill = group, group = group), color = "white") +
coord_fixed(1.3) +
guides(fill=FALSE) + # do this to leave off the color legend
@@ -131,7 +131,6 @@ ggplot(data = states) +
#creating df with total shootings per agency and census data
agencies_census <- agencies_census |>
left_join(shootings_by_agency, by = c("name" = "agency"))
-agencies_census
#prelim visualization of relationship between percentage of officer residency and number of fatal shootings per agency
agencies_census |>
@@ -155,14 +154,12 @@ majority_mean <- shootings_case |>
filter(majority == TRUE) |>
count(agency) |>
summarize(maj_mean = mean(n))
-majority_mean
#calculate mean number of shootings per agency in cities where a minority of officers reside in the city
minority_mean <- shootings_case |>
filter(majority == FALSE) |>
count(agency) |>
summarize(min_mean = mean(n))
-minority_mean
#calculate a difference in means between the `majority` and `minority`
diff_in_means <- majority_mean - minority_mean
@@ -174,7 +171,6 @@ knitr::kable(head(diff_in_means))
```
-
```{r}
#fit single linear regression model for correlation between percentage of officer residency and number of fatal shootings per agency
@@ -186,13 +182,12 @@ p1 <- get_regression_table(fit)
knitr::kable(head(p1))
#add `armed` and `majority` to `shootings_by_agency` df
-shootings_by_agency_census <- shootings_case %>%
- group_by(agency) %>%
- count(armed) %>%
- drop_na(n, armed) %>%
+shootings_by_agency_census <- shootings_case |>
+ group_by(agency) |>
+ count(armed) |>
+ drop_na(n, armed) |>
right_join(agencies_census, by = c("agency" = "name")) |>
distinct(armed, .keep_all = TRUE)
-shootings_by_agency_census
shootings_by_agency_census <- shootings_by_agency_census |>
select(n.x, armed, all)
@@ -225,69 +220,90 @@ ggplot(data = shootings_by_agency_census, aes(x = all, y = n.x, color = armed))
```
+The model equation for `fit` is:
+
+\[ \text{Number of Fatal Shootings (n)} = 35.7782 - 0.5874 \times \text{Percentage of Officer Residency (all)} \]
+
+Interpretation:
+
+- The intercept, $35.7782$, is the estimated number of fatal shootings when the percentage of officer in-city residency (`all`) is $0$. For each one-unit increase in the percentage of officer residency, the number of fatal shootings is expected to decrease by $0.5874$ ($-0.5874$) units, assuming all other factors remain constant.
+
+This model suggests that there is a negative association between the percentage of officer residency and the number of fatal shootings. However, it's important to interpret the results in the context of your data and consider potential confounding factors, like whether or not the victim was armed.
+
+The model equation for `fit_multi` considering victim armament (`armed`) is:
+
+\[ \text{# of Fatal Shootings (n.x)} = 4.117 + 1.211 \times \text{Percentage of Officer Residency (all)} + 24.921 \times \text{Armed (YES)} \]
+
+- The intercept, $4.117$, is the estimated number of fatal shootings where the percentage of officer in-city residency (`all`) is $0$ and the victim was un-armed. For each one-unit increase in the percentage of in-city officer residency compared to the total force (`all`), we expect an increase of $1.211$ fatal shootings, assuming the victim's armament status (`armedYES`) remains constant.
+
+- The coefficient for 'armedYES', $24.921$, indicates that the victim is armed (`armed` is `YES`), we expect an increase of $24.921$ fatal shootings compared to when the victim is not armed (`armed` is `No`), assuming the percentage of officer residency (`all`) remains constant.
+
+In summary, the model suggests that the percentage of officer residency and whether the victim is armed are associated with the number of fatal shootings per agency even as we control for victim armament. However, as correlation does not imply causation, and other factors not included in the model may influence the outcomes.
+
```{r Hypothesis Testing for Diff in Mean Total Fatal Shootings between Residency Prop}
#generate null distribution
-null_dist <- agencies_census %>%
- specify(n ~ majority) %>%
- hypothesize(null = "independence") %>%
- generate(reps = 1000, type = "permute") %>%
+null_dist <- agencies_census |>
+ specify(n ~ majority) |>
+ hypothesize(null = "independence") |>
+ generate(reps = 1000, type = "permute") |>
calculate(stat = "diff in means", order = c("TRUE", "FALSE"))
#compute observed test statistic
-test_stat <- agencies_census %>%
- specify(n ~ majority) %>%
+test_stat <- agencies_census |>
+ specify(n ~ majority) |>
calculate(stat = "diff in means", order = c("TRUE", "FALSE"))
#visualize p-value
-null_dist %>%
+null_dist |>
visualize() +
shade_p_value(obs_stat = test_stat, direction = "less")
#compute p-value
- null_dist %>%
+ null_dist |>
get_p_value(obs_stat = test_stat, direction = "less")
```
-Inference for a Difference in Proportions
-- $H_0$: The mean total number of fatal shootings per agencies does not differ based on if a majority of the officers live in the city or not.
-- $H_A$: The mean total number of fatal shootings per agencies is fewer in cities where a majority of the officers live in the city then cities where they do not.
+Inference for a Difference in Means
-– $H_0 : p_{maj} − p_{min} = 0$, or equivalently $H_0 : p_{maj} = p_{min}$
-– $H_A : p_{maj} − p_{min} < 0$, or equivalently $H_A : p_{maj} < p_{min}$
+- $H_0$: The mean total number of fatal shootings per agencies does not differ based on if a majority of the officers live in the city or not.
+- $H_A$: The mean total number of fatal shootings per agencies is fewer in cities where a majority of the officers live in the city then cities where they do not.
+-- $H_0 : \mu_{maj} − \mu_{min} = 0$, or equivalently $H_0 : \mu_{maj} = \mu_{min}$ -- $H_A : \mu_{maj} − \mu_{min} < 0$, or equivalently $H_A : \mu_{maj} < \mu_{min}$
```{r Hypothesis Testing for Correlation between Total Fatal Shootings and Residency Prop}
#generate null distribution
-null_dist_cor <- agencies_census %>%
- specify(n ~ white) %>%
- hypothesize(null = "independence") %>%
- generate(reps = 1000, type = "permute") %>%
+null_dist_cor <- agencies_census |>
+ specify(n ~ white) |>
+ hypothesize(null = "independence") |>
+ generate(reps = 1000, type = "permute") |>
calculate(stat = "correlation")
#compute observed test statistic
-test_stat_cor <- agencies_census %>%
- specify(n ~ white) %>%
+test_stat_cor <- agencies_census |>
+ specify(n ~ white) |>
calculate(stat = "correlation")
test_stat_cor
#visualize p-value
-null_dist_cor %>%
+null_dist_cor |>
visualize() +
- shade_p_value(obs_stat = test_stat, direction = "greater")
+ shade_p_value(obs_stat = test_stat, direction = "two.sided")
#compute p-value
-null_dist_cor %>%
- get_p_value(obs_stat = test_stat, direction = "greater")
+null_dist_cor |>
+ get_p_value(obs_stat = test_stat, direction = "two.sided")
```
Inference for a Correlation
-- $H_O$: There is no relationship between percentage of the total police force that lives in the city they serve and number of fatal shootings.
-- $H_O$: There is a relationship between percentage of the total police force that lives in the city they serve and number of fatal shootings.
+- $H_O$: There is no relationship between percentage of the total police force that lives in the city they serve and number of fatal shootings.
+
+- $H_A$: There is a relationship between percentage of the total police force that lives in the city they serve and number of fatal shootings.
+
+ - $H_0 : \rho = 0$
-– $H_0 : \rho = 0$
-– $H_0 : \rho \neq 0$
+ - $H_0 : \rho \neq 0$
diff --git a/index.qmd b/index.qmd
index 27acd93..1698490 100644
--- a/index.qmd
+++ b/index.qmd
@@ -14,9 +14,26 @@ output:
toc: true
toc_float: true
code_folding: true
+
+format:
+ html:
+ code-fold: true
+ code-summary: "Show the code"
---
-## Proposal
+```{r setup, include=FALSE}
+knitr::opts_chunk$set(warning = FALSE, message = FALSE)
+
+library(tidyverse)
+library(usmap)
+library(sf)
+library(infer)
+library(moderndive)
+```
+
+> # On average, police in the United States shoot and kill more than 1,000 people every year...and then they go home to their families
+
+## Abstract {#sec-abstract}
This case study investigates the intricate relationship between police residence and fatal police shootings, employing a data science approach to uncover insights and patterns within the context of law enforcement agencies. Focused on police officers residing in the cities they serve, the study examines whether this residency factor correlates with the incidence of fatal police shootings. The data set, spanning the years 2015 to 2023, is composed of information on police agencies involved in at least one fatal shooting, and is subjected to rigorous analysis using advanced statistical methods and machine learning techniques.
@@ -24,43 +41,324 @@ This study aims to discern patterns, trends, and potential biases associated wit
The insights derived from this case study bear substantial implications for informing public policy, refining police training protocols, and strengthening community relations. By unraveling the nuanced dynamics surrounding police residence and fatal police shootings, this case study aims to provide evidence-based recommendations to enhance transparency, accountability, and trust between law enforcement agencies and the communities they serve. In doing so, it contributes to the broader discourse on police reform, fostering a data-driven approach to address critical issues and promote safer, more resilient communities.
-### Explanatory Variables
-
-| Name | Description |
-|------------------------|------------------------------------------------|
-| `city` | U.S. city |
-| `police_force_size` | Number of police officers serving that city |
-| `all` | Percentage of the total police force that lives in the city |
-| `white` | Percentage of white (non-Hispanic) police officers who live in the city |
-| `non-white` | Percentage of non-white police officers who live in the city |
-| `black` | Percentage of black police officers who live in the city |
-| `hispanic` | Percentage of Hispanic police officers who live in the city |
-| `asian` | Percentage of Asian police officers who live in the city |
-
-**Incident Information**
-
-| Name | Description |
-|-----------------------|------------------------------------------------|
-| `id` | A unique identifier for each fatal police shooting incident. |
-| `date` | The date of the fatal shooting. |
-| `body_camera` | Whether news reports have indicated an officer was wearing a body camera and it may have recorded some portion of the incident. |
-| `city` | The municipality where the fatal shooting took place |
-| `county` | County where the fatal shooting took place. |
-| `state` | The two-letter postal code abbreviation for the state in which the fatal shooting took place. |
-| `latitude` | The latitude location of the shooting expressed as WGS84 coordinates, geocoded from addresses. Please note that the precision and accuracy of incident coordinates varies depending on the precision of the input address which is often only available at the block level. |
-| `longitude` | The longitude location of the shooting expressed as WGS84 coordinates, geocoded from addresses. |
-
-**Agency Information**
-
-| | Description |
-|---------|---------------------------------------|
-| `id` | Department Database Id |
-| `name` | Department Name |
-| `state` | State in which the agency is located. |
-
-## Project thoughts
-
-I am interested in exploring data related to...
-
-- Political Extremism
-- Black American Opinion
+## Hypotheses
+
+We will conduct two hypothesis tests to analyze both;
+
+1. The nominal relationship between an increasing proportion of in-city officer residency and number of fatal police shooting deaths
+
+ - $H_0$: The mean total number of fatal shootings per agencies does not differ based on if a majority of the officers live in the city or not.
+
+ - $H_A$: The mean total number of fatal shootings per agencies is fewer in cities where a majority of the officers live in the city then cities where they do not.
+
+ - $H_0 : p\_{maj} − p\_{min} = 0$, or equivalently $H_0 : p\_{maj} = p\_{min}$
+ - $H_A : p\_{maj} − p\_{min} < 0$, or equivalently $H_A : p\_{maj} < p\_{min}$
+
+2. The categorical difference in fatal police shooting deaths between cities where a majority or or minority of police officers live in the city.
+
+ - $H_0$: There is no relationship between percentage of the total police force that lives in the city they serve and number of fatal shootings.
+
+ - $H_A$: There is a relationship between percentage of the total police force that lives in the city they serve and number of fatal shootings.
+
+ - $H_0 : \rho = 0$
+
+ - $H_0 : \rho \neq 0$
+
+## Methods
+
+#### Tidying Data
+
+```{r Tidying and Wrangling Data}
+##Tidying Data
+
+#creating dfs from .csv files
+police_locals <- read_csv("data/police-locals.csv")
+agencies <- read_csv("data/fatal-police-shootings-agencies.csv")
+shootings <- read_csv("data/fatal-police-shootings-data.csv")
+
+#removing old `city` tag from data set that we created when decatenated the city names
+police_locals <- police_locals |>
+ select(-city_old)
+
+# creating `agencies` df with just police departments
+agencies <- agencies |>
+ filter(grepl("department", tolower(name))) |>
+ filter(!grepl("county", tolower(name)))
+
+#creating binned categorical account of if shooting victim was `armed`
+shootings <- shootings |>
+ mutate(armed = case_when(is.na(armed_with) ~ "NO",
+ armed_with == "unarmed" ~ "NO",
+ armed_with == "unknown" ~ "NO",
+ armed_with == "undetermined" ~ "NO",
+ armed_with == "gun" ~ "YES",
+ armed_with == "knife" ~ "YES",
+ armed_with == "blunt_object" ~ "YES",
+ armed_with == "other" ~ "YES",
+ armed_with == "replica" ~ "YES",
+ armed_with == "vehicle" ~ "YES"))
+
+#creating df with only agency `names`, `id`, and `state`
+agencies_ids <- agencies |>
+ select(name, id, state)
+
+#creating df with `city`, `agency`, and `state` info for each shooting
+shooting_agencies <- shootings |>
+ select(city, agency_ids, state)
+
+#changing `shooting` var in `shooting_agencies` df to numeric
+shooting_agencies$agency_ids <- as.numeric(shootings$agency_ids)
+
+#creating df with `city` and `state` info for each agency by joining `agencies_ids` and `shooting_agencies`
+agencies_w_cities <- agencies_ids |>
+ left_join(shooting_agencies, by = c("id" = "agency_ids", "state" = "state")) |>
+ drop_na(city) |>
+ distinct(id, .keep_all = TRUE)
+
+#creating df with census data for each agency by joining `agencies_w_cities` and `police_locals`
+agencies_census <- agencies_w_cities |>
+ full_join(police_locals, by = c("city" = "city", "state" = "state")) |>
+ drop_na(police_force_size) |>
+ distinct(id, .keep_all = TRUE) |>
+ mutate(majority = if_else(all >= 0.5, "TRUE", "FALSE"))
+
+#creating df of only shootings involving agencies within `agencies` df
+shootings_case <- shootings |>
+ right_join(agencies_census, by = c("city" = "city", "state" = "state")) |>
+ select(-agency_ids) |>
+ rename(agency_ids = id.y, id = id.x, agency = name.y, victim = name.x) |>
+ select(-location_precision, -race_source)
+```
+
+#### Counting Shootings
+
+```{r Counting Shootings}
+#count shootings by agency
+shootings_by_agency <- shootings_case |>
+ count(agency)
+
+#find top 25 agencies with the most shootings
+top_25_agencies <- shootings_by_agency |>
+ slice_max(n, n = 25)
+```
+
+#### Mapping Locations of Police-Involved Shootings between 2015 and 2023
+
+```{r shot_map, include=FALSE}
+#mapping Locations of Police-Involved Shootings between 2015 and 2023
+
+#load geo-viz libraries
+library(ggmap)
+library(maps)
+library(mapdata)
+
+#create blank map
+usa <- map_data("usa")
+states <- map_data("state")
+
+#add locations of shootings to maps
+shot_map <- ggplot(data = states) +
+ geom_polygon(aes(x = long, y = lat, fill = group, group = group), color = "white") +
+ coord_fixed(1.3) +
+ guides(fill=FALSE) + # do this to leave off the color legend
+ geom_point(data = shootings_case, aes(x = longitude, y = latitude), color = "black", size = .2) +
+ geom_point(data = shootings_case, aes(x = longitude, y = latitude), color = "red", size = .1) +
+ labs(title = "Locations of Police-Involved Shootings between 2015 and 2023",
+ captions = "This is only includes cities where we have agency census data.",
+ x = "Longitude",
+ y = "Latitude")
+
+```
+
+```{r}
+shot_map
+```
+
+```{r bolstering dfs}
+#creating df with total shootings per agency and census data
+agencies_census <- agencies_census |>
+ left_join(shootings_by_agency, by = c("name" = "agency"))
+
+#creating visualization of comparison Shootings in Cities where a Majority/Minority of Officers Reside
+p0 <- shootings_case |>
+ ggplot(aes(x = majority, fill = armed)) +
+ geom_bar() +
+ labs(title = "Shootings in Cities where a Majority of Officers Reside",
+ caption = "This is only includes shootings where we have agency census data.",
+ x = "Does a majority a of the total police force live in the city?",
+ y = "Number of fatal shootings",
+ fill = "Victim Armed?")
+```
+
+```{r}
+p0
+```
+
+```{r}
+#calculate mean number of shootings per agency in cities where a majority of officers reside in the city
+majority_mean <- shootings_case |>
+ filter(majority == TRUE) |>
+ count(agency) |>
+ summarize(maj_mean = mean(n))
+
+#calculate mean number of shootings per agency in cities where a minority of officers reside in the city
+minority_mean <- shootings_case |>
+ filter(majority == FALSE) |>
+ count(agency) |>
+ summarize(min_mean = mean(n))
+
+#calculate a difference in means between the `majority` and `minority`
+diff_in_means <- majority_mean - minority_mean
+```
+
+```{r}
+#tidy table
+knitr::kable(head(diff_in_means))
+```
+
+```{r}
+#fit single linear regression model for correlation between percentage of officer residency and number of fatal shootings per agency
+fit <- lm(n ~ all, data = agencies_census)
+
+#add `armed` and `majority` to `shootings_by_agency` df
+shootings_by_agency_census <- shootings_case |>
+ group_by(agency) |>
+ count(armed) |>
+ drop_na(n, armed) |>
+ right_join(agencies_census, by = c("agency" = "name")) |>
+ distinct(armed, .keep_all = TRUE)
+
+shootings_by_agency_census <- shootings_by_agency_census |>
+ select(n.x, armed, all)
+
+#fit multiple linear regression model for correlation between percentage of officer residency and victim armament and number of fatal shootings per agency
+fit_multi <- lm(n.x ~ all + armed, data = shootings_by_agency_census)
+```
+
+## Results
+
+### Multiple Linear Regression of relationship between percentage of officer residency and number of fatal shootings per agency `fit`
+
+The model equation for `fit` is:
+
+$$
+\text{Number of Fatal Shootings (n)} = 35.7782 - 0.5874 \times \text{Percentage of Officer Residency (all)}
+$$
+
+```{r}
+#tidy `fit`
+p1 <- get_regression_table(fit)
+knitr::kable(head(p1))
+```
+
+Interpretation:
+
+- The intercept, $35.7782$, is the estimated number of fatal shootings when the percentage of officer in-city residency (`all`) is $0$. For each one-unit increase in the percentage of officer residency, the number of fatal shootings is expected to decrease by $0.5874$ ($-0.5874$) units, assuming all other factors remain constant.
+
+This model suggests that there is a negative association between the percentage of officer residency and the number of fatal shootings. However, it's important to interpret the results in the context of your data and consider potential confounding factors, like whether or not the victim was armed.
+
+```{r}
+#visualize polynomial relationship between percentage of officer residency and number of fatal shootings per agency
+ggplot(data = shootings_by_agency_census, aes(x = all, y = n.x)) +
+ geom_jitter(width = 0.10, height = 0, alpha = 0.45) +
+ geom_smooth(method = "lm", formula = y ~ poly(x, 2), se = TRUE) +
+ labs(title = "Number of Shootings on a Scale of Police Force Residency",
+ x = "Percentage of the total police force that lives in the city",
+ y = "Number of fatal shootings in that city")
+```
+
+### Multiple Linear Regression of relationship between percentage of officer residency/victim armament and number of fatal shootings per agency `fit_multi`
+
+The model equation for `fit_multi` considering victim armament (`armed`) is:
+
+$$
+\text{\ of Fatal Shootings (n.x)} = 4.117 + 1.211 \times \text{Percentage of Officer Residency (all)} + 24.921 \times \text{Armed (YES)}
+$$
+
+```{r}
+#tidy `fit_multi`
+p2 <- get_regression_table(fit_multi)
+knitr::kable(head(p2))
+```
+
+- The intercept, $4.117$, is the estimated number of fatal shootings where the percentage of officer in-city residency (`all`) is $0$ and the victim was un-armed. **For each one-unit increase in the percentage of in-city officer residency compared to the total force (`all`), we expect an increase of** $1.211$ fatal shootings, assuming the victim's armament status (`armedYES`) remains constant.
+
+- The coefficient for 'armedYES', $24.921$, indicates that the victim is armed (`armed` is `YES`), **we expect an increase of** $24.921$ fatal shootings compared to when the victim is not armed (`armed` is `No`), assuming the percentage of officer residency (`all`) remains constant.
+
+In summary, the model suggests that the percentage of officer residency and whether the victim is armed are associated with the number of fatal shootings per agency even as we control for victim armament. However, as correlation does not imply causation, and other factors not included in the model may influence the outcomes.
+
+```{r}
+#visualize polynomial relationship between percentage of officer residency and victim armament and number of fatal shootings per agency
+ggplot(data = shootings_by_agency_census, aes(x = all, y = n.x, color = armed)) +
+ geom_jitter(width = 0.10, height = 0, alpha = 0.45) +
+ geom_smooth(method = "lm", formula = y ~ poly(x, 2), se = TRUE) +
+ labs(title = "Number of Shootings on a Scale of Police Force Residency",
+ x = "Percentage of the total police force that lives in the city",
+ y = "Number of fatal shootings in that city",
+ color = "Victim Armed?")
+```
+
+```{r Hypothesis Testing for Diff in Mean Total Fatal Shootings between Residency Prop}
+
+#generate null distribution
+null_dist <- agencies_census |>
+ specify(n ~ majority) |>
+ hypothesize(null = "independence") |>
+ generate(reps = 1000, type = "permute") |>
+ calculate(stat = "diff in means", order = c("TRUE", "FALSE"))
+
+#compute observed test statistic
+test_stat <- agencies_census |>
+ specify(n ~ majority) |>
+ calculate(stat = "diff in means", order = c("TRUE", "FALSE"))
+
+#visualize p-value
+null_dist |>
+ visualize() +
+ shade_p_value(obs_stat = test_stat, direction = "less")
+
+#compute p-value
+ null_dist |>
+ get_p_value(obs_stat = test_stat, direction = "less")
+
+```
+
+At a significance level of $\alpha = 0.05$, the p-value of $0.248$ suggests that, **there is insufficient evidence to reject the null hypothesis**. In this context, since our null hypothesis asserts that mean total number of fatal shootings per agencies does not differ based on if a majority of the officers live in the city or not, our p-value indicates that, assuming our null is true, the probability of observing our given test statistic (difference in means; $\mu_{maj} − \mu_{min}$) is $-4.92$ is around $25\%$ ($0.248$). Meaning our observed difference in means between the groups is likely to have occurred by random chance.
+
+```{r Hypothesis Testing for Correlation between Total Fatal Shootings and Residency Prop}
+
+#generate null distribution
+null_dist_cor <- agencies_census |>
+ specify(n ~ white) |>
+ hypothesize(null = "independence") |>
+ generate(reps = 1000, type = "permute") |>
+ calculate(stat = "correlation")
+
+#compute observed test statistic
+test_stat_cor <- agencies_census |>
+ specify(n ~ white) |>
+ calculate(stat = "correlation")
+
+
+#visualize p-value
+null_dist_cor |>
+ visualize() +
+ shade_p_value(obs_stat = test_stat, direction = "two.sided")
+
+#compute p-value
+null_dist_cor |>
+ get_p_value(obs_stat = test_stat, direction = "two.sided")
+
+```
+
+At a significance level of $\alpha = 0.05$, the p-value of $0.248$ suggests that, there is sufficient evidence to reject the null hypothesis. In this context, since our null hypothesis asserts that there is no relationship between percentage of the total police force that lives in the city they serve and number of fatal shootings, our p-value indicates that, assuming our null is true, the probability of observing our given test statistic (correlation coefficient; $\rho = 0$) is $-0.0470$ is around $0\%$ ($0$). Meaning our observed correlation coefficient likely would not happen if there was no relationship between percentage of officer residency and number of fatal shootings for a given agency.
+
+## Conclusion
+
+### General Conclusions
+
+### Study Limitations
+
+### Improvements for Future Study
+
+## Citations
