The COVID-19 pandemic and 2020 police brutality protests highlighted the disparities racial minorities face when interacting with social institutions, particularly police. New York City (NYC) operates the nation's most extensive public transport system and acts as an intersection between workers, home, and work. When NYC initiated the PAUSE Program that mandated all non-essential workers work from home, many people of color were classified as essential. With the heavy presence of subsequent police brutality protests in NYC, this study models geospatial and demographic policing pattern shifts relating to the Manhattan subway in response to significant events of 2020. This analysis examines if/how police use infrastructure systems and institutional power, such as arrests, in retaliation beyond excessive use of force. The project suggests policing patterns did have geospatial arrest shifts around subway stations that diverged based on race, and that there is a correlation between overall arrest patterns and Manhattan subways.
11 thoughts on “Response to the COVID-19 Pandemic and 2020 Police Brutality Protests: Changes in Manhattan Policing Patterns”
Comments are closed.
Hello!
Welcome to the poster and this digital poster fair! I am very interested in public transportation, particularly in its use in creating denser more travelable cities. I am not intimately familiar with New York, but the subway system is phenomenal. With the advent of the COVID-19 pandemic, however, I became aware of the ways in which public transportation can make a population vulnerable to natural scenarios. I also kept up to date on the Hong Kong protests and noticed how the police there weaponized the trains and train stations to combat protesters. The intersection of the pandemic and police brutality protests in Manhattan intrigued me, and I wanted to evaluate the ways institutions become aggressors on systems people have little choice but to use and how these aggressions are racially motivated. I think these questions are important in evaluating how vulnerable populations move and interact with the urban fabric and the institutions that control the city. The results were not as clear-cut as I expected, but interesting nonetheless. If you have any questions/feedback, please let me know!
Hi Spencer, thank you for such an interesting poster! I particularly liked how your first three figures displayed arrest data as well as the census track information. If you proceed with the project and look at the relationship between protest location and arrest, I encourage you to also consider whether these protests are occurring in predominantly white or black areas of the city. Also, many of the media narrative around this summer’s Black Lives Matter protests focused on the idea that more white and non-black individuals had joined in on the protests than in years past. Does this show up in the arrest records in the areas where the summer and fall protests occurred? Did the racial makeup of arrests change in the census tracks that contained protests and was there a rise in white and non-black arrests?
Good job!
Hello Samantha,
Thank you very much! I appreciate the kind words.
Comparing arrest data to protest locations was something I was very interested in working on. Unfortunately, I was unable to find a comprehensive ordering of protest locations and necessary information. ACLED/Princeton have the US Crisis Monitor, which includes a good listing of protests. The location data for the protests are based only on the city level, though, so all the Manhattan protests were labeled as the middle of Central Park. For fun, I read some local newspapers to get an overview of the protests’ general areas, and a lot of reporting was in the Times Square area and lower Manhattan. From my understanding, however, more sustained protests were present in other boroughs like Brooklyn. Newspapers tended to report specific locations for property violence (IE police car was set on fire at this intersection). In contrast, for peaceful protests and protestor arrests the geographic information tended to be vaguer.
As to the notion of more white and non-black individuals joining the protests, I do not know how to properly evaluate that, as it seems a project in itself. Gallup and Reuters polls show white support for BLM decreasing throughout the summer, which potentially influenced white actors’ involvement in protesting. Similarly, accurate tallying of the races of protestors would be difficult and then classifying that based on violent involvement or being out past curfew. Within the BLM movement, there were several calls throughout the protests for people joining from the suburbs to leave before curfew, as they did not live in the area and would not have to live with any long-term police reprisals. Not having researched this particular question, I would expect the results to show longer-term higher arrest rates for people of color in areas of protests, regardless of the protest actors’ race. Largely, I think it comes down also to a question of data. NYPD is required to report all arrest data, including protest arrests. However, the publically accessible data does not address if charges stuck. There is also a massive amount of arrest data to sift through. 2020 was a relatively slow year for arrests, and there were still roughly 20,000 on Manhattan Island alone. It is a fascinating question, though, and one I would love to explore more. These are just my off-the-cuff thoughts on the topic.
Hi Spencer! I recently did a project for another class on the disproportionate placement of hazardous waste facilities near census tracts with high minority populations, and am also interested in the ways that infrastructure can be targeted and/or racialized, and how to remedy these injustices. I was wondering if you had any predictions for this data if you also factored in income level, whether that be by census tract or by the average subway user, and if you think that this would impact your results? Thank you for sharing!
Hello Alyssa!
That sounds like a super interesting project. My initial thoughts are that there would be a significant overlap between the placement of hazardous material and income level, likely leading to a strong relationship with the number of subway riders and race. As you mentioned and already knew, the long history of placing hazardous waste facilities in majority-minority areas is closely tied to segregation and its continued legacy of disproportionate harm. I am unsure how that relates specifically to Manhattan. Perhaps, due to the high density of Manhattan, there are likely fewer facilities; however, as I say that, I am looking at a map with Rikers Island right next to East Harlem, one of the most black areas of Manhattan. There is almost certainly a strong correlation in less-dense boroughs like the Bronx.
I did not address income level in my project, as the primary objective was to address the impacts of race and its relationship to public transportation and policing. I would expect to find some correlation between lower arrest rates and higher income. I have a map I did not include in my poster due to space constraints, breaking down general arrest trends. One of the most interesting parts of the unpublished map is that there is a decrease in arrests in the high-income area around Fifth Avenue and Central Park for all races. However, if one divides the same map by race, black arrests in the same area decreased less than white arrests.
Specific to my project and subway data, my guess is that the income of subway riders dropped after the initiation of the PAUSE Program. Essential workers are likely to be paid less, so the ones going to work are likely to be of lower-income. I do think factoring income would be a good idea, but I was concerned about adding too many variables to the already convoluted project. I also felt the primary purpose was to evaluate race, which, due to a long history of oppression, has links to income, is also its own distinct variable concerning police interactions.
Thank you for the questions!
Hi Spencer!
Thank you for working to tackle such important questions. There’s a whole lot in 2020 that could have an effect on arrest data and that makes it more challenging to tease out the effects of deeper societal patterns and structural inequalities in a dataset like this. It seems like an extremely difficult proposition to try to determine what are the effects of Covid vs protests vs changes in policing policy vs a myriad of other elements that may have had an effect. So don’t be too discouraged that the results were not as clear as you were hoping!
One thing that occurs to me is that it may be worth looking at the cyclical annual variation in arrests prior to 2020, as there could be more fundamental trends that need to be accounted for prior to doing direct comparisons within the year. For example, people may spend less time outside (or change primary transportation methods) in NYC in January than they do in August simply due to the temperature (True in Chicago at least) which may affect arrest rates unevenly across the population. Or police precincts may have incentives to make more arrests at some points in the year than others. Or there may be other factors I haven’t thought of.
My biggest question for you is, if you could imagine any 3 (or more if you want) data sources that you think would help clarify exactly what is going on in your model, what would they be?
Hello James!
Definitely, social and systemic patterns spanning a city as complex as New York are difficult to tease out, especially for someone like me who doesn’t have intimate knowledge of Manhattan. Similarly, dividing the arrest data into three distinct periods fails to fully account for the cumulative nature of particularly the COVID-19 pandemic.
As you mentioned, seasonal variances impact crime and arrest rates in Chicago, and I expect a similar phenomenon in New York. I would assume that lower-income areas would have less seasonal variance when comparing arrest rates to public transportation than in higher-income areas. Also, regarding public transportation, the subway system is a commodity of sorts. Particularly after the PAUSE Program’s initiation, subway ridership dropped dramatically, but bus ridership did not decline by the same extent. This is likely in part because the subway system does not service lower-income areas as fully as buses do, where I would expect less flexibility was given to work from home. I wanted to include both prior years as a standard for arrest data and arrest data around bus stops. However, I was frankly unsure of my ability to properly analyze the data I was devoting my project to and did not want to misrepresent data from earlier years. Regarding the bus stops, there are just too many of them. One of the biggest problems with geolocating arrests in New York is the city is super vertical, so a radius around subways might include arrests made in the building above or even nearby buildings that are fully unrelated to public transportation. I figured I could limit that inclusion to roughly the building the subway station was situated under, but even small radii around bus stations would necessitate a large margin of error.
Three data sources I would like to utilize are the protest locations, locations of police calls, and increased access to travel data. Various outlets documented the protest locations, but creating a comprehensive compilation would have been a herculean task, I think. Similarly, arrest data does not really show where crime is happening but where it is being responded to, and if high police presence is “justified.” To go back to Chicago, CPD now claims they have a roughly 50% homicide solvency rate, but up until recently, that number was in the 20% range. NYPD claims roughly a 70% solve rate, but even that incorporates a huge margin of error. If the police aren’t arresting 30% (I know this is a flawed assumption as a person can kill more than one person and still be arrested for just one of those murders), that can dramatically shift the location of crimes. So, I wanted to evaluate police reports (which would include its own host of problems largely concerning community trust), but unfortunately, there is no associated location data reported. I felt like the association with crime data might dilute the purpose of the graph, and as you mentioned, a comparison to previous years’ arrest trends would more accurately answer the question of shifting police response. Finally, the subway data reported by NY MTA is scarce and non-specific. It would have been nice to see daily traffic fluctuation by station and what percentage of people in an area rely on public transit. The only source I had on subway ridership data was from the NYT; they wrote a handful of articles talking about maybe five or six stations and the general ridership trends there post-PAUSE Program.
Hi Spencer, excellent project! I wonder what led to you differentiate between white and non-white instead of some other division of race?
Hello Monica,
This is the methodology question I probably struggled with the most. The straightforward answer is that the largest difference in arrest patterns were between white and any other race. Similarly, many of the arrest patterns of many non-white races complimented each other.
In terms of confidence that police interaction will be positive, there is a slider of sorts. Many reports that I read since this year had focused largely on the black-white differences in treatment. Still, I was also curious to get a potential perspective into the world of policing and how perspectives were potentially enforced. In this case, my reasoning is more personal curiosity than tied to larger social questions.
Spencer–good initial sounding in what is a very complex topic. What you’re attempting is essentially a difference-in-difference research design, comparing before and after “treatment” effects. The effects here are arrests by race/ethnicity (dependent variable). The trick would be to try to build a model (regression) that controls for cyclical patterns in arrests (James’ point above), covid-related downturns in ridership, to better isolate protests as an independent variable. It’s tough also because of the vagaries inherent in the data–where were all the protests in Manhattan? though not impossible. Then the question is whether there is spatial nonstationarity in the correlation, such that a geographically weighted regression is warranted. This is all doable in ArcGIS. If you are interested in pursuing this further we can suss out the next steps. My question for now is what you think the subtle shifts in the hotspots might mean? If I’m not mistaken, the main hotspot is down around Chelsea, Hell’s Kitchen, and the Meat Packing boroughs, then extend up into midtown during protest times. What might be driving that hotspot? Why those places in particular? The core of it doesn’t change through the sequence, but it expands northward slightly. What would be your hypothesis for this ? On a related note, how did you define your neighborhoods in the Getis-Ord Gi*? k nearest neighbors? Fixed distance band? IDW? That parameter can of course have a big effect on the result, as we discussed in class. Secondly, as a suggestion it seems Local Moran’s I might provide more insights here because then you would also see feature-level relationships with their neighborhood. What you’re seeing here is clustering of high values, not individual features in relation to neighborhoods of features. Really excellent layout and aesthetics in the poster–good going on that front for sure. Great initial study.
Hello Dr. Wernke,
Concerning the shift in hotspot to Times Square, I believe it is likely twofold. First, several major protests, confronted by sometimes violent counterprotests, occurred in the Times Square/42nd St. Second, potentially shifting work patterns as COVID-19 restrictions loosened. Tourism to NYC is still remarkably low, but as my hotspot comparison starts at the beginning of the PAUSE Program, my potential guess is that due to the density of shops and restaurants, public life returned to the Times Square area perhaps slightly faster than in residential areas. My argument toward East Harlem and Hell’s Kitchen hotspots is at best that the decreased amount of people allowed to work from home pushed arrest rates to decline less substantially than other areas. I also hypothesize a racial connection. In my optimized hot spot analysis, I bounded incidents in aggregated polygons along census tract lines. The shift to Local Moran’s I does seem logical; I will continue poking into that.