Finding More Evidence Than Just The Word of the Cops

Exploring the StreetCred Police Killings in Context Data

Nick Selby

This week, the StreetCred Police Killings in Context (PKIC) dataset launches. You can read about the dataset, its methodology and peer review, and get, work with and create your own analysis products or derivative works from the data, free. This article is one of several deep-dives into the PKIC data.

In studying the killings by police of unarmed people in the United States, especially in light of media coverage that is overwhelmingly skeptical of the American cops’ take on things, one of the most important things we needed to consider was where we would get information, and how we can know that it is truthful.

Some excellent journalists raised an important issue: in many cases, the cops simply can not be trusted - that absent some other evidence, based on history, it’s just too difficult to study data that’s primarily based on the word of police. As a cop myself I get upset at that, but I understand their position. And in any case, I would like to get that data they ask for.

Sometimes, we just can’t. We can sure try. We can institute a strong methodology and rules about how we cover, we can get peer reviewers from across the political spectrum (I even invited some serious data people involved with data in the excellent Mapping Police Violence site, which supports the Black Lives Matter movement, but didn’t get a reply — the invitation still stands), and we can make it easy to correct, update and add to the database.

But at the end of the day, the very nature of policing — anywhere in the world, not just in the United States — means that a lot of the time, the only people who see an incident are the officer and the person involved.

One thing we can do is look to the data and see what it tells us about non-police input. We start with the basic count of incidents. As we set this out, remember: none of the following suggests that these crimes on their face mean that police have the right to use deadly force.

This year, of 125 incidents in which police killed an unarmed civilian, 25% (31) began on traffic stops, and 65% (81) began as a response to a 911-call about a violent crime (robbery, carjacking, domestic violence or assault) or property crime (burglary, car theft or vandalism) in progress. There were 9 people (7%) whom 911 callers described as being, “crazy,” or, “on drugs”, “covered with blood”, and “yelling”, or threatening people. Three (2%) were wanted fugitives in the act of escape — and one was unarmed when he died but was acting as part of a gang of three who were wanted in a recent homicide and were at the time of the incident in the progress of a kidnapping a woman. One was a training incident in which one cop shot another accidentally. In nearly three of four of cases, then, the police were called by someone else for help. These were not self-dispatched, and the police didn’t select the person who ultimately died, the police were called (we will dig into the incidents that began on traffic stops soon). It is certainly not a case of police “targeting” any race or group other than, “people committing crimes.”

Again: none of this is to remotely suggest that these crimes on their face mean that police have the right to use deadly force. Instead we are making a different and important point that there were other people involved in bringing the police to the scene in three out of four of the cases.

Witness Corroboration

There were civilian, non-police witnesses in 64 (about 51%) cases in the PKIC database. There are no known human witnesses in 61 (about 49%) cases. In another 11 cases (8.79%), when there were not witnesses, there was video, either from the police, surveillance or a witness. And in one more case (0.8%) when there were neither witnesses or video, there was audio. This means that there was some independent, non-police version of events in just under 61% of cases.

What The Witnesses Are Saying

In general, in the 51% of cases in which there were witnesses, they generally sided with the police’s account. In 60 (48%) cases within the StreetCred PKIC dataset, some or all witnesses supported the police account of events.

Witnesses exclusively supported the police account in 40 (32%) cases, and some witnesses supported, and some disputed, the police account in 20 (16%) of cases.

Witnesses disputed the police in a total of 24 (19%) cases. Witnesses exclusively disputed the police account in 4 (3%) of cases, and some witnesses disputed, while others supported the police account in 20 (16%) of cases.


There is video available in 32 (26%) of incidents in the StreetCred PKIC data. And in two of these cases, it was witness video — not police video — that brought attention to apparently unjustified use of deadly force by the officers (see below). This is a definitive sign that the national push towards body-worn and dashcam video for officers is sorely needed, and cannot move fast enough. Especially in this climate, police video is a crucial form of evidence that clearly promotes, and is in the true interest of, justice. (Read a separate piece about the challenges to departments in launching an officer video program here).

What The Video Shows

It is important to understand that in many cases, the video of an incident is known to be available but has yet to be released. There have been two very notorious cases this year in which witness video demonstrated that officers appeared to have fabricated their accounts in the killing of unarmed men. As important, there have been two cases this year in which the video, once released by police, demonstrated conclusively that the police had told the truth in the killing of unarmed men. Video, we cannot stress enough, is absolutely essential to policing in the 21st century.

How You Can Help

Disagree with what you’ve read here? Have another take? Download the data here and perform your own analysis. Tell us what you would fix, or how you would do it better.