Description of Data

Our data comes from “The Guardian,” an American news hub that covers domestic and international affairs. They created a dataset depicting all the police killings that occurred in the United States during 2015 (however, only from January to June 2nd, it appears). It’s incredibly comprehensive and depicts a lot of information regarding the individual killed, the nature of the killing, and county-level information about the area the killing occurred in. There is additionally geospatial information as we have data down to the individual’s address which is very rich for analysis. All the variables are entered as “character” class, but some are functional as quantitative variables depending on their nature. The variables listed here are the ones we used in our analysis:

Research Questions

In particular, we were curious about these particular topics of discussion:

Relevant Visualizations

The median household income of individuals’ killed by police in the first five months of 2015 is more than $10,000 less than the median household income of US household in 2015. This suggests that individuals killed by police may come from more economically disadvantaged circumstances than the average American.

Although there are more armed indivudals killed by police than unarmed, it appears that there is no distinct relationship between an individual’s household income and whether they were armed.

This graph represents the relationship between average county level income per observation and average county level unemployment rate per observation. Additionally, observations are colored by Census region and identified by state. Generally, it appears there is no relationship, but it is interesting to observes trends by region. For example, midwestern states appear to have high unemployment and low-income, but northeastern states appear to have more observations of high income/low unemployment.

This is a bar plot depicting the average unemployment rate of the counties each observation belongs to, grouped by racial/ethnic groups. Generally, we see the unemployment rate is somewhere between 0.10 and 0.15 for all groups. Additionally, there doesn’t appear to be a significant level of variation when distinguishing between armed/unarmed observations.

Interestingly the only cause of death to have an equal proportion of Armed/Unarmed individuals across the facetted races/ethnicities is deaths from gunshots. Otherwise we can see Black people have a disproportionate number of Unarmed deaths resulting from being in Custody as well as from being struck by vehicles, something the Other category shares with it. Overall most races/ethnicities seem to have relatively low proportions of Unarmed deaths in most categories of Cause of Death except for 1, with Black being the only race to have more than 1 high Unarmed proportion category.

Here we can see that the largest, and really only, cluster can be found in a low college graduatate rate and statistically high unemployment rate area. While there are some points in areas with much higher unemployment rates, the cluster we can see here lies over double the national average for the year the data is from (2015).

States with the most number of police killings (i.e., California, Texas, Florida) are also states with the three largest populations in the US. Therefore, state population size may be a confounder for determing the relative frequency of police shootings across states.

However, on a per capita basis, the states with a higher concentration of police killings noticeably shift. States like North Dakota, Vermont, and Arkansas have the largest frequency of police killings per capita. It is important to note that North Dakota and Vermont rank in the bottom five for state populations.

CONCLUSION

From viewing the aggregate of the plots and graphs we can draw several conclusions. One is that of those killed by police during the time period represented by this data, people of lower socioeconomic status from areas of high unemployment/low higher educational attainment (with the ones from poorer areas tending to be from the south and wealthier from the northeast or west) were the most effected. The deceased also tended to be armed regardless of the unemployment rate from their area or their household income. While the highest body counts can be found in the most populous states, per captia shows us that its the midwest and other least populated states that are the deadliest per capita (specifically North Dakota and Vermont). In terms of how they died, we see that surprisingly gunshot death proportions for armed/unarmed individuals was consistent between various races (with the vast majority of those deaths being armed individuals) while other categories are more heavily skewed towards specific armed status. For example, the majority of black people killed in custody were unarmed, but they’re the only race to have unarmed people killed in this category at all.

Overall this paints a picture that there does seem to be a racial prejudice present for some causes of death (but not the one you’d expect, e.g., gunshots). But more than that the issue revealed by this dataset is that killings by police disproportionately effect the poor and uneducated. Surprisingly, we can see that the chance a victim was armed or not was not effected noticably by either their household income nor their race, but unsuprisingly being armed definitely was more common among most casues of death, especially when it comes to uses of force that require the police to be face to face with the victim by significant margins in almost all cases (bar hispanics for tasers).