Blog Post

Technical Issues with Network Analysis on Police Brutality

Technical Issues with Network Analysis on Police Brutality

Using data on racialized police violence from “The Counted,” I attempted to create a network analysis map using Cytoscape, an open-source program that creates network visualizations. I wanted to visually represent the distribution of race and ethnicity in fatal law enforcement brutality using data from 2015 and 2016. Unfortunately, I ran into some technical problems that I could not resolve.

Attached below are my two network visualizations (2015 is first, 2016 is second):

I think the easiest way to reflect on this practicum is to just discuss each technical problem separately.

1] I was unable to configure the race/ethnicity labels such that they would display when I zoomed out.

To rectify this issue, I played around with the tools on the sidebar extensively, but I was ultimately unsuccessful. As a result, it was difficult to analyze the network visualization because I was not able to view the entire network at once with all of the relevant labels. Eventually, I memorized the color labels for each race/ethnicity that I placed, so I was able to make sense of the network as a whole, when zoomed out. Unfortunately, I realized that my label size constraint would make it difficult to convey my network visualization to others. The key takeaways from this problem were twofold. First, label clarity is critical to collaboration/presentation. Second, individual solutions to a technical problem do not always translate into universal fixes.

2] I could not come up with a practical means of displaying which law enforcement agencies targeted which demographics of victims.

One of the key issues we sought to address in this practicum was: which law enforcement agencies targeted with demographics of victims? I was not able to orient the visualization such that each law enforcement agency had its own central node from which categories of race/ethnicity expanded from. Yet, I do not think a network visualization would have been the best way to display that data anyway. The more I thought about it, the sillier it seemed to display hundreds of law enforcement agencies in a visual map, filled with many labels. “The Counted” project visually displays the geographic and temporal distribution of police killings in the United States. While their map may lose some critical racial/ethnic organization, I think if one works from the assumption that certain demographics are disproportionately targeted in the United States (which is statistically supported), then their means of data visualization is quite effective for activist watchdog efforts. Thus, this issue led me to the conclusion that there is no one method for data visualization that can best represent all information. Strategies for data visualization need to be tailored to both the end goal and the audience. While that conclusion seems intuitive, I think working through a concrete example in this situation made me critically reflect on my approach to programs for data visualization. For instance, I now think of Cytoscape as a useful program for displaying data with few central nodes, but I also no longer think that the program is great for data with many central nodes.

To briefly return to the original analysis, while there were more white people killed by law enforcement agencies than any other one demographic in 2015 and 2016, people of color (especially black people) were (and are) disproportionately harmed by police brutality. Abby does an excellent overview of this in her blogpost on this practicum by viewing the data in people per million:

“In people per million 10.3% are Native American, 6.66% are African American, 3.23% are Hispanic/ Latino, and 2.9% are white.” -Abby Darch (

This data illustrates that even though formal legal equality exists with regards to police brutality, structural racism still exists in how police violence is carried out and judicially rendered (law enforcement officers that unjustly target people of color often receive minimal consequences). As digital humanists, we ought to continue to tackle issues like this by innovating and refining means of data visualization. A key component of activism is public awareness, and we can play an important role in that sect through persuasive data visualization. 


1 comment

The technical problems you encountered are intimately connected with the dataset -- they're not tool-based alone, as you point out. So: if you were going to redo this assignment and create a Practicum for Cytoscape with a different dataset or a different task with the same data set, what would it be? Or did I simply need to spend more time teaching the tool? :)