Using Filters to Drill Down into Your Data

Published on January 16, 2017

If you watch or listen to the news, you can’t help but hear the debate about law enforcement use of deadly force, in particular as it relates to minorities. Regardless of where you sit within the political spectrum, the truth of the matter is that mistakes happen and people, good people, are on both sides of the debate.

Without question, each death is a tragedy; controversy erupts around questions of whether anyone death was avoidable. For instance, Bettie Jones, a 55-year old African-American woman was mistakenly shot to death by responding officers – all while she was trying to help a neighbor during a family disturbance.

The question is, can data serve to pinpoint patterns: are there similarities that highlight systemic bias, or a fixable problem? Are there situations in which law enforcement erred on the side of racial bias? So, with data from The Guardian and Census, I decided to see what I could learn about the situation. I created an app in Arcadia Instant using the dataset Police Killings to visualize all the information I need in one place, with easy drill downs to get into the specifics of the situation.

A bird’s eye view of the app shows five visuals; I’ll talk briefly about each one and then show how neatly Arcadia Instant arranges data across all of them with filters for drilling down to the needed granularity.

  • Police Incidents: This table shows a summary containing the number of fatalities, law enforcement agencies and cities present in the data.

  • Incidents/Million by State: The choropleth country map shows states with the number of encounters per million people, colored by the density of incidents in each state. It shows that Oklahoma has the highest incidents with almost 5.62 probable incidents per million people.

Number of Incidents/ethnic group by City: I want to see if a particular ethnic group is more prevalent in such incidents. The horizontal stacked bar chart Number of Incidents/ethnic group by City shows the number of people who died and their ethnicities for each city.

  • Trend Comparison: I also want to analyze how factors like Unemployment Rate, Personal Income (median) and College Degree (25+ Population with B.A. or higher) are associated with such incidents. The Trends Comparison line chart shows a comparison of these factors with the incidents over a period of five months. We can see that the number of incidents increases when unemployment rate increases for some months but there is no definitive trend, as is the case with the portion of the population with a B.A. or higher degrees and median personal income

  • Cause of Death: The pie chart below shows the different ways people died in encounters with the police. It shows that gunshots are the leading cause of death in violent encounters, followed by taser. Looking at this visual may make us believe the rumors are true about police being needlessly violent unless we dig deeper with the next visual.

  • Carrying Weapon: This is an important factor to consider while deciding if police pull the trigger easily or only under pressing circumstances. The pie chart shows whether the person was carrying a weapon at the time of the encounter or not and it’s clear that a good portion of those shot carried firearms or knives. The pie chart clearly shows that the majority carried weapons; in fact, almost 77% of them were armed, which shows that though there are accidents where the wrong person is engaged, by and large, the police do not often kill innocent people.

I stitched the visuals into an app, applied on-click filters to zoom in on the data I want to explore. A user can now hover over any component in a visual to view more information about it.

Clicking on the state of California in the Incidents/Million by State visual, and looking at the Number of Incidents/ethnic group by City visual shows that the largest number of people who died in police encounters were Hispanic in Los Angeles; only black in Anaheim; one Hispanic and two whites in Bakersfield.

Similarly, clicking on the state of Texas, we see that the police encounter deaths involved the highest number of Hispanics in Houston and El Paso, blacks in Fort Worth, and whites in Austin. Before jumping to any conclusion about whether there is a racial bias, we have to bear in mind that this particular dataset does not compare the exact number of firearms or other violent deaths with each ethnic group as a share of the population, so it does not indicate racial bias.

The app makes one thing clear, that the police often face life-threatening situations where the average citizen response would be to run away. Instead, they take a stand and do their job. Although I cannot really comment on the racial bias argument because of the lack of sufficient information, seeing a large number of armed criminals, I can observe that notwithstanding rare accidents, these data show a consistent pattern of police primarily shooting armed suspects.