How we analyzed a decade of shootings near US schools

On June 18, we published an article about the frequency of school shootings in the United States. The story was based on our analysis of 10 years of data from the Gun Violence Archive as well as records from other sources. Here is how we did it:

Shooting Data

Trace Analysis accessed the Gun Violence Archive via API to create a database of all shootings from 2014 to 2023 – 522,964 incidents in total. GVA relies on media reports supplemented by other sources, including urban violence dashboards when available, to identify shootings. For this reason, some shootings are missing from the database and some location data is imprecise.

School data

Past and present K-12 school location data was collected from the National Center for Education Statistics' EDGE geographic datasets. We removed from our analysis schools those that the NCES had coded as closed or virtual only.

Public school location data is updated annually and private school location data is only updated twice a year. We therefore associated the filmings with the most recent location data available for each school year. It is therefore possible that in rare cases, schools – particularly private establishments – were already closed when shootings occurred nearby. Likewise, it is possible that a small number of shootings near schools were missed because a school was open before appearing in the NCES data.

Private school data is also known to be incomplete and The Trace has not extrapolated from it. As a result, some private schools are missing from our analysis, making our shooting totals underestimated.

Student Data

We extracted public school demographic data from the Common Core of Data 2023-2023 and private school data from the Private School Universe Survey. The PSS is only updated every two years, so we used the 2021-2022 values ​​to estimate the 2022-2023 student populations for our analysis.

Locale data

To determine whether a shooting occurred in an urban, suburban, or rural school, we used the NCES school classification system. For precise definitions, see here.

School names

Districts enter school names and there is no standard format. For example, Abraham Lincoln Elementary School could be entered as “Lincoln, Abraham Elementary” or simply as “Abraham Lincoln.” Where possible we have standardized the names, but some names are probably still confusing. We apologize.

Segregation data

Census tract-level demographic data were derived from the 2022 American Community Survey five-year estimates.

Software

Our analysis was carried out using R version 4.1.2 (2021-11-01), and in particular the 'sf' package (Pebesma, 2021) version 1.0.16.

Population density and school shootings

Our 500 meter radius is more likely to capture shootings in cities where the population is densely concentrated around schools. In places where schools are on campuses bounded by parking lots, fewer shootings are likely to be captured. However, population density is not enough to explain the particularly high findings of nearby shootings in places like Philadelphia and Chicago, since New York, Boston and other dense cities did not have such high rates.

Shootings during and outside school hours

GVA's data on the time the shootings occurred is quite sparse and therefore cannot distinguish between shootings during and after school hours. Our analysis therefore includes both. However, our reporting reveals that after-school shootings still have a profound impact on students, who are exposed to crime scene tapes, police cars, blood, shell casings and other signs of violence. Even the presence of impromptu memorials on sidewalks can affect learning outcomes.