NYC Shootings

Background

Much has been written about the increase in shootings throughout New York City since the latter half of 2020. I pulled the New York Police Department’s (NYPD) publicly available data to see the actual shooting numbers.

Method

I collected data from two different links on NYC Open Data and combined the information into one pandas DataFrame. Working at the NYPD taught me just how much seasonality impacts crime rates (more crime in the summer, less in the winter). To control for that, I limited my analysis to the 365 days before the first NYC Covid-19 lockdown, and the 365 days after the lockdown began.

Determining how to calculate shooting increase

As with many of my analyses at the NYPD, I wanted to look at data by precinct to get a better understanding of what was going on. While looking at citywide numbers is useful, it doesn’t paint as detailed a picture as local predict data does.

Historical shooting activity varies drastically across precincts. For example, a year-over-year increase of 3 shootings in the 1st precinct (downtown Manhattan) creates a very large percentage increase, which might overstate the situation. Whereas an increase of 3 shootings in the 75 precinct (Brooklyn) creates a small percentage increase and might understate the situation.

*Note that the below maps are interactive and additional information, including the precinct number, can be seen when hovering over the map.

Analysis: Raw change in shootings by precinct

I first wanted to look at data using maps. Choropleth maps make it easy to spot outliers and clusters of geographic areas with similar trends.

The plot above shows the raw change in shooting count by precinct for the year prior to NYC’s lockdown vs the year starting with the lockdown. By hovering over the darker precincts, I saw that the 67 and 75 precincts had the largest shooting increases with +66 and +56, respectively. The map also reveals that these two precincts are adjacent to each other and perhaps suffer from violence along the shared precinct border.

But what about some of the other precincts, like those in lower Manhattan or southern Brooklyn? Many had an increase of 1-4 shootings, so they were colored light blue above. But is that the case when looking at percentage increases?

Analysis: Percent change in shootings by precinct

Looking at the same map using percentage increases instead of raw counts, new possible “hotspots” appear. For example, the 6th precinct in lower Manhattan had an increase of +1 shooting, but the former 365 period had zero. This led to an infinite % increase. Meanwhile, the 76 precinct in Brooklyn South had an increase of 1,600% due to shootings increasing from 1 to 17 year-over-year.

Takeaways

Both of these plots should be used to make deployment decisions. Using the 76 precinct as an example, the first map does not indicate the 76 is a “hotspot” because other precincts have had much larger raw increases in shootings. But when you look at the 2nd map, it is clear there has been a significant shift that likely requires some sort of further inquiry.

Conversely, the 73 precinct is emphasized in the first map and de-emphasized in the 2nd. Despite the percentage increase being “only” +189%, the raw increase of +66 is the highest in the city and is likely a major priority for the NYPD.

These decisions are easier to make with the help of simple visuals. But it’s important to look at the data in multiple ways, as any one chart might bias the reader.

Analysis: Raw change in shootings by precinct #2

The below slope chart depicts the same data as in the first map above, but simplifies the takeaways by featuring the five precincts with the highest raw increase in year-over-year shootings. This plot would be more appropriate if someone was less interested in the geographic layout of “hotspots” and just wanted to prioritize where shooting count changes are highest.

*Note that the below slope chart is interactive and additional information, including counts for every non-labeled precinct, can be seen when hovering over the data.