Law enforcement agencies and the government fail to provide accurate numbers on the number of lives taken at the hands of police. While there have been efforts by the government to document the number of police killings such as the USA National Vital Statistics System (NVSS) or a recent FBI program which has failed to deliver results (as of 2021), there is still no way to understand the scope of police violence.
We believe the data represented in Mapping Police Violence is the most comprehensive accounting of people killed by police since 2013. A report from the Bureau of Justice Statistics estimated approximately 1,200 people were killed by police between June, 2015 and May, 2016. Our database identified 1,104 people killed by police over this time period. While there are undoubtedly police killings that are not included in our database (namely, those that go unreported by the media), these estimates suggest that our database captures 92% of the total number of police killings that have occurred since 2013. We hope these data will be used to provide greater transparency and accountability for police departments as part of the ongoing work to end police violence in America.
In addition to CZ’s Mapping Police violence, there are two additional non-governmental datasets which attempt to fill this void:
Fatal Encounters (FE)
Attempts to “document all deaths that happen when police are present or that are caused by police: on-duty, off-duty, criminal, line-of-duty, local, federal, intentional, accidental–all of them. We enable people to filter out the deaths that they aren’t interested in examining.”
Data Collection Mechanism
Exclusively documents incidents of police violence where people are fatally shot by on-duty police officers.1
Data Collection Mechanism
Using these three data sources, researchers have shown that the NVSS program has been severely undercounting police killings. Specifically, the researchers estimate the following:
"From 1980 to 2018, the NVSS did not report 17,100 deaths (95% UI2 16,600–17,600) out of 30,800 deaths (30,300–31,300) that we estimated, accounting for 55.5% (54.8–56.2) of all police violence deaths from 1980 to 2018." — The Lancet
Moreover, recent research studying differences across the three data sources has found that while the three data sources are similar, they have become more dissimilar throughout time (Comer and Ingram, 2022).
Both findings indicate the importance of continuing the three non-governmental datasets to not only understand the extent of police violence but as an added layer to identify gaps across data sources and continue to hold law enforcement agencies accountable and transparent.
Understanding “Police Violence”
Most studies employ the term “Police Violence” to indicate a civilian death at the hands of a law enforcement officer (e.g., Desmond, Kirk, and Papachristos, 2016). Given data constraints around collecting incidents of police violence that do not result in a death, most datasets only include fatal incidents of police violence, Mapping Police Violence included.
Admittedly, this restricts and constrains our understanding of the true breadth of police violence and the other harms and traumas that are missing in the data. While we find that on average there are over 1,000 police killings every year in the US, we need to find ways in which we can start collecting non-fatal incidents of police violence. This is our next step and we’re committed to partnering with different research entities to document these incidents of police violence.
For example, it is important to note that the Mapping Police Violence database is more comprehensive than the Washington Post police shootings database: while WaPo only tracks cases where people are fatally shot by on-duty police officers, our database includes additional incidents such as cases where police kill someone through use of a chokehold, baton, taser or other means as well as cases such as killings by off-duty police.
UI stands for “Uncertainty Interval.” Because these numbers are estimated, there’s a 95% chance that the true number is in the range given by the UI (based on the researchers’ statistical model).