Updated: Sep 28, 2020
Part One: Fatal Police Encounters
Driven by high-profile lethal use of force cases in the United States, race relations have become a major topic of conversation in 2020. Racism is a topic that divides our nation, our families, and our relationships. It's a emotionally charged and often leads to violence, protests, and hate crimes. If there was ever a topic that needed clear-thinking, objective analysis, it's racism.
A significant portion of the discussion focuses on systemic racism. Systemic racism refers to the structure, organization, and rules of society. It indicates that society favors or gives advantage to one race over another. Systemic racism manifests itself in justice, employment, wealth, housing, health care, and education. To investigate the existence of systemic racism, we'll look at a variety of indicators within these areas, make determinations as to their statistical relevance, and decide if evidence of systemic racism exists on a national level.
Before we get started, let's clarify some issues. In the event that we discover evidence of systemic racism exists, it does not mean that
you are racist
your neighbor/police officer/banker/etc. is racist
your local police department is racist
The American people are racist or that America is a bad nation
Systemic racism means that the system favors one group over another. Also, we are looking at the data from a national level, so we are determining if the nation as a whole favors one group or another. You may have an entirely different experience at a local level.
Likewise, in the event that we discover there is no evidence of systemic racism, that doesn't mean
your personal experience with racism isn't valid.
you don't have systemic racism at a local level
We are making broad determinations to inform discussion at a national level so that there can be constructive conversations around what- needs to be done. For example, the Justice Department determined that there was systemic racism in the Maricopa County Sherriff's Office. That doesn't mean the entire state of Arizona or the United States have systemic racism. The findings only apply to that office. In the interest of fairness and clear thinking, we need to be clear about what we discover, and do our best to eliminate bias from the findings. Then we can have an honest discussion that minimizes the emotional fireworks that occur when discussing race relations.
To identify evidence of system racism, we'll focus on five broad categories: wealth, employment, education, housing, and justice. We'll do a series of articles that analyze publicly available data and determine if the data provides statistically significant evidence of systemic racism. Throughout the analysis, we'll apply the scientific method. This means that we will identify the null hypothesis (that there is not evidence of systemic racism), and attempt to prove the null hypothesis. If we succeed in proving the null hypothesis, we have to accept that there is not evidence of systemic racism for that category. If we fail to prove the null hypothesis, we must accept that there is evidence of systemic racism for that category. When we're finished, we'll look at all the data together and make a determination based on the complete picture.
Fatal Police Encounters and Systemic Racism
Protests and violence over fatal police encounters have risen to historic levels. We'll examine for systemic racism in fatal police encounters by looking at a variety of indicators. For this study, we'll focus on Blacks. Are fatal police encounters for Blacks in proportion to the population? What other criteria can we examine to understand the relationship?
For each of these indicators, we'll focus on proving the null hypothesis: there is no difference in how the system treats Blacks and Whites. If we succeed, we accept that there is no difference between how Blacks and Whites are treated and there is not evidence of systemic racism. If we fail to prove the null hypothesis, we accept that there is a difference between how Blacks and Whites are treated for these indicators and that represents evidence of systemic racism.
When people think about 'looking at the data,' a vision usually pops into their head of a huge database where everything is stored neatly for the taking. That database doesn't exist. The United States is a de-centralized government. Consequently, data on crime is also de-centralized. Every police department and agency has their own standard for how they report data on arrests, incidents, and convictions. The FBI and the Bureau of Justice Statistics attempt to gather this data into a central location, but it is spread out among hundreds of tables with helpful, meaningful names like '2018-Table 42A'. Fortunately, the University of Michigan has aggregated much of the data.
Some data simply isn't collected. There isn't a government database or table on fatal police encounters. We rely on organizations like Mapping Police Violence, Fatal Encounters, and the Washington Post to scour official records and media reports and gather the information together for the rest of us. Before you start screaming about media bias, blah, blah, blah. This isn't some book of numbers thrown together. Every incident is documented with links and supporting evidence. The Bureau of Justice Statistics recognizes the Fatal Encounters data as a sound source of evidence. We'll be using both the Mapping Police Violence and the Fatal Police Encounters data for our analysis. We aren't paying any attention to their findings. We're going to do our own work. We just need the data.
As much as possible, we'll be viewing data at rates per million population. This allows us to compare findings proportionately.
Fatal Encounters by Police Department
The Mapping Police Violence database provides a breakdown of fatal encounters by almost a hundred major police departments around the country. The dataset currently covers 2013 through 2019 and includes supplementary data about violent crime and murder rates. We can extract the data and average the fatal encounters by race for each police department. Keep in mind that we are looking at rates per million. Because of the way ratios work, police departments with very low black populations will show unusually high rates. Departments like Anchorage, Reno, and Scottsdale will look particularly bad, because they have such low Black populations.
The graph shows a stark difference in the rates of Blacks versus Whites. We can test for statistical significance by creating 10,000 simulations and checking to see how often this distribution occurs. The value we receive is referred to as a 'p-value'. A p-value of less than 0.05 (5%) is considered statistically significant. In this case, the p-value is zero. This ratio of Black deaths to White deaths never occurs in our sample, so we reject the null hypothesis that there isn't a difference between Black encounters and White encounters, and accept the hypothesis that there is a difference. For this dataset, Blacks are victims of fatal encounters at a higher rate than whites. We can use the Fatal Encounters database to compare outcomes going back to 2000. Looking at the data by year takes out the skew from low populations in exchange for fewer samples. The annual data from Fatal Encounters also provides a p-value of zero, so the differences are statistically significant. The average rate of incidents among Blacks is 3.9 times higher than the average number of incidents among Whites. In this case, we also need to accept that Blacks are victims of fatal encounters at a proportionately higher rate than Whites.
Making Fair Comparisons
Our first instinct is to ask if Blacks commit proportionately more crimes than Whites. However, this is not a valid comparison. We are testing for disparities in police encounters, so we can't use police encounters as a testing criteria. It's circular logic. If disparities exist, those disparities would distort the analysis. We need to find neutral comparisons that remove police encounters from the testing criteria. As an example, let's look at marijuana usage vs. marijuana arrests.
Marijuana usage is determined through surveys conducted by mental health organizations. Usage is not determined by police encounters, so it functions as a control in our experiment. The Substance Abuse and Mental Health Services Administration (SAMHSA) provides data on drug usage, and I found a set of summary data from Infoplease. Since Infoplease is not a research level organization, I validated the results against a 2013 report from SAMHSA and a report from the NCBI. The chart below displays marijuana usage by race.
For this dataset, usage of marijuana among Whites is statistically higher than among Blacks and Hispanics. The SAMHSA report found no statistical difference in illicit drug use (all drugs, not just marijuana), and the NCBI report found higher use among Whites than Blacks. We can accept that marijuana usage is the same or greater among Whites than Blacks.
We expect arrests to be proportionate to usage. Data on arrests for marijuana usage is provided by the University of Michigan's aggregation of the FBI Uniform Crime Reporting statistics. The chart below aggregates arrests for marijuana possession by state from 2009 through 2016. (No data is available for Florida.)
On average, arrests for marijuana possession of Blacks is almost four times higher than arrests for Whites, even though Whites use marijuana at the same or higher rate than Blacks. The arrest rate is not indicative of the actual usage rate. This is why we can't use encounters as a criteria in our testing. The encounters can be distorted by the conditions that we're testing for. We need to look for criteria that will neutralize the effect of police encounters on the outcome.
What criteria can we use to examine the relationship between crime and fatal police encounters? One consideration is the rate of violent crime, which is conveniently available to us in the MPV database. The rationale is that high incidents of violent crime would predict an increase in fatal police encounters. We can take the rate of violent crime, which doesn't consider race or police encounters, and use it to 'predict' fatal police encounters. A statistical model called linear regression analysis allows us to look at the relationship between violent crime and fatal police encounters. The model will tell us if the relationship is statistically significant and how much violent crime explains fatal police encounters. To clarify, we aren't saying that violent crime was or wasn't occurring during the incident. We're examining whether an overall environment of violent crime results in more frequent fatal police encounters. This has nothing to do with the justification of the events themselves.
In the chart below, a relationship exists if the dots trend in the same direction as the line. The closer the dots are to the line, the better the relationship. In our example, there is a statistically significant relationship between violent crime and fatal encounters, but it is weak. The model provides a value that tells us what percentage of deaths are explained by the violent crime rate. In this case, the violent crime rate only explains 5% of the outcome.
What if we looked at the relationship between the violent crime rate and fatal encounters among Blacks? Are higher levels of violent crime related to more fatal encounters among police and the Black population? Running the analysis and looking at the chart below shows no relationship between the level of violent crime and the number of deaths among Blacks. The relationship is not statistically significant, so we can't say that higher levels of violent crime result in more deaths among Blacks. Comparing the overall murder rate produces similar results. There is a weak (1%) relationship between the murder rate and all fatal police encounters but no relationship between the murder rate and fatal police encounters among Blacks. So, for this dataset, we have to accept the null hypothesis. There is no relationship between the deaths of Blacks in police encounters and the overall murder rate or overall violent crime rate.
What If the Victim Is Armed?
Should it matter if the victim is armed? Ideally, the threat to the police officer by an unarmed assailant is the same whether the assailant is Black or White. We want to know if there's a difference in fatal encounters by race when the subject is unarmed. Again, using the scientific method, we have to test for no difference in fatal encounters by race when the subject is unarmed. The MPV database provides information on whether the subject is armed vs unarmed or not having an actual weapon (such as a toy gun). We'll look at the outcomes by year.
As the graph shows, unarmed Blacks are proportionately more likely to be killed in a police encounter than their White counterparts. Testing shows that these differences are statistically significant. We have to reject the null hypothesis and accept that there is a difference in outcomes for unarmed Black encounters than for unarmed White encounters.
For each of our tests, we determined that there is a difference in outcomes for Blacks vs. Whites. Given these outcomes and the data, we must accept that there is evidence of systemic racism in regards to fatal police encounters among the Black population. Let's review what that does and doesn't mean:
It does not mean that
you are racist
your neighbor/police officer is racist
your local police department is racist
The American people are racist or that America is a bad nation
The analysis shows us that there is evidence of a problem, and that problem should be investigated and addressed. It doesn't tell us the cause of the problem or the nature of any specific incident. It doesn't tell us anything about how your community handles systemic racism. It does tell us that--as a nation--this is a problem worthy of our attention and further study. Regardless of the cause, it's certainly something that we want to improve upon.
To further understand if there is evidence of systemic racism within our criminal justice system, the next article will dig into traffic stops. Does race play a role in incidences and outcomes of traffic stops? Let's find out.