Police Killings
Elisha McMillen, Julia Morelli, Lauryn Sall, & Namoos Haider
PSYC 500 Final Project
*Sensitive Content Warning*
Today our group will explore trends in police killings such as racial disparity in police killings and overall killing rates over the course of time. Before we discuss our data, we want to start off by mentioning that we chose to work with data surrounding a socially and racially sensitive subject matter. We understand and acknowledge that people may have different sentiments and views on the topic at hand. Our goal is not to discuss opinions or speculate at this time, but rather to present the facts as listed in the data. Regardless, it is important that we address this topic with care and caution.
Section I: Data Curation and Ethics
Source of Data
Our team’s data comes from https://mappingpoliceviolence.org/. We chose this dataset because we would like to learn about statistics of police killings throughout the country, that is, killings committed by police officers in our country. The data is constantly being updated through the excel sheet on the website.
Background
These particular datasets provide information about geographic locations of police killings, the race of the victims, and other specific information about each incident. We are seeking to form conclusions about trends in police killings and the variables that come into play in these particular crimes. Our data contains categorical variables and continuous variables.
Categorical Variables:
Race
Geographical location
Continuous Variables:
The number of people killed
Black/White dissimilarity index
The flow of this presentation will begin with national police killing statistics, then we will move to state police killings with special regards to Kansas, then we will take a look at racial correlations to killings, and finally we will explore racial dissimilarity with regards to police killings. For the sake of operationalizing our variables, a police killing in this particular set of data is defined by https://mappingpoliceviolence.org/ as “A case where a person dies as a result of being shot, beaten, restrained, intentionally hit by a police vehicle, pepper sprayed, tasered, or otherwise harmed by police officers, whether on-duty or off-duty.”
Ethical Concerns
Ethical Concern #1: The names of victims are included in the original data which could put family members in danger due to lack of privacy
Ethical Concern #2: The names of the officers involved are listed rather than all of the officers involved which could lead to a lack of justice & accountability being served
Ethical Concern #3: The street address of the killing is listed, which could potentially identify the victim if it’s a home address and again can put the home/whoever resides there in danger
We decided to leave the names out as they didn’t aid in our data exploration process, but we want to acknowledge the importance of every individual victim on this list.
Research Questions
Research Question #1: Do police kill Black citizens at a rate higher than White citizens?
Research Question #2: Do certain states have a higher police killing rate than others and how does Kansas compare?
Research Question #3: In areas where there is a high rate of racial dissimilarity, is there a higher rate of police killings?
Hypotheses
Hypothesis #1: Police officers kill Black citizens at a higher rate than they kill White citizens.
Hypothesis #2: Certain states are definitely going to have higher police killing rates than others.
Hypothesis #3: In areas with high rates of racial dissimilarity, there are higher rates of police killings.
Section II: Data Preparation
Dataset #1: Victims Killed by Police Across the Nation
Dataset #2: Killings by Police That Were Reported by the Police Department in That City & State
*NOTE: For each of the states, a discrepancy occurs between the number of deaths reported by police departments included in the dataset and the number of deaths known due to news reports. In other words, there are fewer deaths recorded by police than those that actually occur because only a small proportion of departments from each state reported their data.
Dataset #3: Killings by Police Recorded in Each State
Descriptive Statistics
.info() function to determine datatypes
.value_counts() function to determine the total number of victims per race
.describe() function to describe statistics such as mean, standard deviation, and frequency in order to better understand the distribution of each dataset
Furthermore, each of the original datasets were condensed to display only the rows/columns of interest to our analysis.
Section III: Exploratory Data Analysis
National Data
The Rate of Police Killings by Victims' Race
When controlling for population by using the rate at which a race is killed by police, we found that White citizens are not being killed at a higher rate despite having the highest number of deaths by police. Black citizens are killed at a rate nearly 3 times higher than White citizens. Hispanic citizens also have a higher rate than White citizens at the national level.
State Data
The Rate of Police Killings by State
Population was standardized by using the overall rate of police killings per state population. Unexpectedly, New Mexico is the state with the highest rate of police killings (11.197). Rhode Island has the lowest police killing rate (0.811). California (15th highest, rate=4.923), Texas (25th highest, rate=4.124), and Florida (23rd, rate=4.425 highest) are not in the top ten states (+ Washington D.C.) with the highest police killings. Kansas ranked 28th overall with a rate of 3.978.
The Rate of Police Killings by Race by State
The above bar plots show the rates of all people killed by police, White people killed by police, Black people killed by police, and Hispanic people killed by police in the top 10 highest overall (e.g. all people) police killing rate states and Kansas (recall Kansas is ranked 28th overall). The Y axis has been shared between the various subplots to accurately show the disproportionate killing of Black citizens in most of the states. With the exception of Montana and Wyoming (both of which reported no police killings of Black citizens), all states have higher rates of Black deaths by police compared to both White and Hispanic citizens. Further, most of the top ten states (exception of Colorado, Mississippi, Montana, and Wyoming) the highest rate of police killings was still Black citizens even when previously dropped races were included (e.g. Asian, Native American, etc.). Although not visualized, all four exceptions had Native Americans as their highest rate.
Kansas Data
The Rate of Police Killings by Race
As with the national graph shown earlier, White citizens are more often killed by police than Black or Hispanic citizens. However, White citizens make up a large majority of the population of Kansas (~76.1%). While Black citizens make up only ~5.6% and Hispanic citizens account for ~11.7% of Kansas population.
Yet again, Kansas follows the same trend as was seen nationally. Black citizens are killed at a disproportionate rate compared to White citizens. Just like the national rates, police kill Black citizens in Kansas at a rate 2.75 times higher than White citizens. Hispanic citizens are also killed at a rate higher than White citizens.
Black-White Dissimilarity Index (BWDI)
The Black-White Dissimilarity Index is essentially a measurement of segregation of Black and White neighborhoods within a city/county (in our case, the index is being measured at the city level). Values range from 0 to 100 and represent the percentage of either race which would need to move out of their current neighborhood in order to fully integrate the two races. The higher the index value, the more segregated the neighborhoods are; in other words, a city with a high Black-White Dissimilarity Index is split into predominantly Black versus predominantly White neighborhoods. For a more in-depth explanation as well as the mathematical calculation visit: http://www1.udel.edu/uapp800/Lecture%20Material/Index%20of%20Dissimilarity%20Example.htm.
Highest Black-White Dissimilarity Index
This graph is looking at the top 10 cities that have the highest Black/White dissimilarity Index of the cities which reported police killing data (and is included in our dataset). Recall for example that Witchita was the only city within Kansas included in the dataset. So it is possible that other cities in the United States have higher Black-White Dissimilarity Index than the cities listed here. However, all ten cities listed here (based upon the 2010 census) show high levels of segregation between Black and White neighborhoods. For each city at least 69% of either Black or White citizens would have to move out of their current neighborhood and move across all neighborhoods in order to achieve full integration. In Chicago, 82.5% would need to relocate.
Heat Maps:
The two heatmaps below go hand-in-hand and were used to see if there was possibly a correlation between the Black-White Dissimilarity Index and the number of Black or White citizens killed.
Quick note: "Black" & "White" in the heatmaps refers to the population of the respective race
We examined if the Average Annual Police Rate for Black/White citizens was correlated with the Black-White Dissimilarity Index. While the rate of Black/White citizens killed by police showed a weak negative correlation (-0.35 for Black, -0.32 for White), the number of Black citizens killed by police showed a positive moderate correlation (0.59) with Black-White Dissimilarity Index. At the same time, White citizens killed had a very weak negative/no correlation (-0.0063) with the Black-White Dissimilarity Index.
Scatterplots:
These scatterplots are yet another way of visualizing the relationship between Black/White/all people killed by police and the Black-White Dissimilarity Index.
The Black people killed by police chart has a clear positive correlation. As the Black-White Dissimilarity Index increases, the number of Black citizens killed by police increases. Analyzing the White people killed by police chart is not as clear. The relationship between the number of White people killed by police and the Black-White Dissimilarity Index is not as strong. Correlational hypothesis testing will be used in later sections to determine
The scatterplots below illustrate the comparison between the average annual police homicide rate for both Black and White people and the Black-White Dissimiliarity Index.
Note that the Avg. Annual Police Homicide Rate for Black People, Avg. Annual Police Rate for White People, and Avg. Annual Police Homicide Rate have different scales on the x-axis. This is important to note because the rate for Black people goes up by 20 whereas the other two rates goes up by 2.5. Our original hypothesis was that Avg. Annual Police Homicide Rate for Black People would increase as the Black-White Dissimilarity Index increased. The scatter plot created does not appear to support this hypothesis. Most of the data falls between 0-20 for the rate irregardless of the Black-White Dissimilarity Index. This is similar to both of the other graphs which also have rates which fall between 0-20. However, unlike the other two graphs, there are extreme outliers for the Avg. Annual Police Homicide Rate for Black People. The outliers additionally tend to have lower Black-White Dissimilarity Index scores.
Section IV: Linear Regression & Hypothesis Testing
Needed Functions
Permutation Replicates
Correlation Coefficient
BS Pairs Linear Regression
Linear Regression
Note: The same process that is shown below was carried out for White people killed by police and rates for both races.
The distribution of all the variables analyzed were non-normal.
Intercept: Average BWDI where 0 black people were killed was ~42.
Slope: Number of black people killed by police increases by 0.83 for each BWDI increase. So for every 10 points increase, 8.3 more black people are killed by police (of course people can not be 0.3 killed so 8 whole people are killed by each 10 point increase).
Minimum RSS value is not pictured (when messing with the range, the minimum is between -0.5 & -1) however the regression line produced had a slope of 0.83. This means that linear regression is not the best model for this data.
Pairs Bootstrap
Note: The same process that is shown below was carried out for White people killed by police and rates for both races.
Estimated slope was 0.73; 95% CI[0.68, 1.13]. As BWDI increases by 1, the rate of police killings of white people increases by ~0.73. This relationship is statistically significant as 0 is not included in the interval; null hypothesis of no linear relationship between Number of Black People Killed by Police and BWDI can be rejected.
Note: The results were similar for both rate analyses (Black Rate vs BWDI, White Rate vs BWDI). Analysis resulted was statistically significant and thus the null hypothesis of the relationship being not linear can be rejected. White killings was found to have a non statistically significant result and thus the null hypothesis failed to be rejected.
Correlation Hypothesis Test (Permutation)
We can clearly reject the null hypothesis (p-value 0.0). Based upon this analysis, there is high certainty that the observed correlation coefficient of 0.59 is statistically significant and not caused by chance. Very likely, there is a positive linear relationship between number of Black citizens killed and the Black-White Dissimilarity Index.
The null hypothesis (H0) fails to be rejected (p-value 0.5). Based upon this correlational hypothesis testing, there is evidence that there may be no correlation between number of White citizens killed and the Black-White Index. The relationship between these two variable may be non-linear. Further research is necessary to determine if the relationship is nonlinear, if null hypothesis is "true", or if an error perhaps occurred somewhere in the data.
Section V: Discussion/Implication
Research Questions
Research Question #1: Do police kill Black citizens at a rate higher than White citizens?
Research Question #2: Do certain states have a higher police killing rate than others and how does Kansas compare?
Research Question #3: In areas where there is a high rate of racial dissimilarity, are more Black citizens killed?
Hypotheses
Hypothesis #1: Police officers kill Black citizens at a higher rate than they kill White citizens.
Hypothesis #2: Certain states are definitely going to have higher police killing rates than others.
Hypothesis #3: In areas with high rates of racial dissimilarity, more black citizens are killed by police.
Discussion/Implications
Our first hypothesis was confirmed by data. Although overall killings show a higher total of White people killed, this is misleading as there is a higher population of White citizens than Black. Nationally, Black people are killed at a rate, standardized by population, of 6.5, and White people are killed at a rate, standardized by population, of 2.415.
Our second hypothesis was also confirmed by the data, as certain states have higher police killing rates than others. We found that New Mexico is the state with the highest rate of police killings, at a standardized rate of 11.197. Rhode Island has the lowest police killing rate, at 0.811.
Finally, we analyzed the relationship between Black-White Dissimilarity Index and police killings. As the index increased, meaning the area is more segregated into Black neighborhoods vs. White neighborhoods, more Black individuals were killed by police. Yet fewer/no change in White people killings occurred as the index increased. This perhaps suggests that in predominantly Black neighborhoods, more police killings occur, but in predominantly White neighborhoods fewer police killings occur. These correlations however cannot definitively say determine causation. There may be other factors which cause these correlations.
Limitations
Limitation #1: The data is obtained through police records. So, if a death is not recorded by the police then it will not show up in the data. Further, not all deaths contained all the relevant information (for example, many deaths did not include the victim’s name, age, or race) because they were not included in the police report. We briefly looked at the 83 police killings which occurred in Kansas from 2013-2021 (March) and 6 victim’s races were not known. Although not likely, if all 6 victims were Black this would drastically change the data (the number of Hispanic and Black deaths would not be the same and the rates would also change) but we have no way of knowing these victims.
Limitation #2: Much of the data is based upon the 2010 census as the results for the 2020 census have not yet been made public. So analysis is based upon numbers that are no longer accurate for the current population and as such will need to be updated once 2020 census results are published.
Limitation #3: This is all correlational data so we can’t formulate any actual solutions to police killings because we can’t prove causation.
Limitation #4: We ended up finding that the third research question we had was distributed non-linearly; our analysis of this question therefore is not likely to be completely accurate in answer the question of how the Black-White Dissimilarity Index is related to the number of (and rate of) Black and White citizens killed by police.
Recommendation
Our suggestion is to look into why predominantly White neighborhoods potentially have lower police killings. What factors are affecting both variables? Comparison of policies and training is a starting point, but the reasoning may be more complex. Research into why police may be more likely to kill individuals in predominantly Black neighborhoods is needed. Is it because of implicit bias? Gun regulation/laws? The nation needs to find which factors are leading to this increase and then implement legislation and police training/reforms to better protect individuals in these neighborhoods. We loosely make this recommendations as again their are a multitude of factors which could cause the disproportionate rates of Black killings by police officers.