StreetCred® Police Killings in Context Methodology
The StreetCred® Police Killings in Context (PKIC) dataset was populated through a range of sources. As others have rightly noted, there is no compulsory national accounting by the United States government of killings by police officers of civilians. StreetCred supports the creation of such a system and stands ready to advise on its creation.
To compile this dataset, we began by using datasets put forth by reputable private entities. These include the Fatal Encounters database (which maintains a superb Silk visualization site, and which launched a crowd-funded effort to track killings of immigrants who are illegally in the US). Many other databases (including this one) are based on the excellent and important work of Fatal Encounters. This includes the excellent Mapping Police Violence dataset and visualization tool; The Guardian’s The Counted: People killed by police in the US database; and The Washington Post’s Distraught People, Deadly Results database. We also view, and sometimes quote, other resources, including Photography is Not a Crime, Killed By The Police (which also runs a Silk-based visualization site at https://killed-by-police.silk.co/) and, with caution about its sources (see below), The Daily Kos.
Using our subject matter expertise in law enforcement and use of force (some of us are current or former police officers, detectives, and agency administrators), and data gathering, cleaning and analysis (some of us are data researchers, crime analysts, intelligence analysts, fraud analysts or scientists; StreetCred Software, Inc., produces data analysis tools, and one of our number runs an open source intelligence service provider) we added columns that would provide contextually relevant, factual data-points, without drawing conclusions.
Our goal is to give stakeholders — community members, activists, police departments, city administrations, scholars, policy-makers and researchers — another tool to better understand whether law enforcement is treating fairly and justly those it is sworn to protect. We do not seek to stanch criticism or suppress controversy; on the contrary, our goal is to produce clean, clear and actionable data that both sides can use to debate intelligently and meaningfully the way forward.
StreetCred PKIC does not draw conclusions as to culpability or blame for any incident, and has no opinion on any incident or its outcome. The dataset is maintained entirely separate from any analysis by any party (including StreetCred Software, Inc.) of any datum or data within. The data within this dataset are intended to be factual, open and peer-reviewed. A mechanism for publishing corrections and errata is provided.
We do all this with the hope that our work will inform police procedures, policies and the training of both new and veteran officers, as well as to inform the criteria by which future incidents of deadly force are examined.
Corrections & Updates
If you see problems, inaccuracies, data-quality issues, or to submit new incidents, please use our PKIC Contact Form. See below to become a peer reviewer.
The StreetCred PKIC dataset considers any death that occurs within the United States, in which the decedent is unarmed, that occurs during an encounter with American law enforcement, from first contact through booking, except vehicle accidents.
- We do not consider vehicle fatalities. These are incredibly difficult to track, and data on them is very dirty and unreliable;
- We do not consider deaths in county jail, or in Municipal jail after booking. A jail is a confinement facility designed to hold people awaiting trial, or serving short sentences; typically of less than one year. Jails are typically managed either by the office of the county Sheriff, or by a municipality, or a county government. Jail officers are not typically police officers, even if they are working as a corrections officer or jailer run by a county Sheriff, though they ultimately are responsible to, and the Sheriff is ultimately responsible for, the jail. These are outside the scope of this data set;
- We do not consider deaths in prison. A Prison is a confinement facility, typically run by the state or federal government, that generally houses prisoners convicted of crimes punishable by one year or more. Corrections officers in prisons are typically not police officers.
- We do not consider unofficial acts committed off-duty. A personal dispute between family members or friends, who happen to be police officers, that turns deadly, is not in the scope of this database. Certainly domestic disputes involving police officers are a problem – they are just not this problem. This database seeks to highlight killing of unarmed civilians by police, whether on- or off-duty, who are acting in an official capacity.
This project seeks relevant, factual, non-biased news sources. The source of each data entry should be documented, in the dataset (Links columns) and/or bibliography (“or” the bibliography in cases when all three links columns are already occupied and we have more links to provide). The order in which sources are checked, trusted and then used are as follows:
- Official Coroner reports, releases, or statements;
- Official district or county or state attorney reports, releases, or statements;
- Grand jury reports, records, transcripts, indictments, or evidence;
- Official police reports, statements, or releases;
- News reports (print, web or television/video) by reputable news sources that quote more than one source and do not have a demonstrable bias towards any given version of the story.
By focusing on these sources, in the order stated, StreetCred PKIC makes every effort to remain unbiased and factual.
In general, local television news, local newspaper and news sites and other traditional local, county, state, regional and national news sources provide acceptable depth and acceptably unbiased reporting to be included as primary or secondary sources. In certain cases, articles in various Patch outlets, New York Newsday, The Daily Kos, The Washington Times, The Guardian (UK), PhotographyIsNotACrime, etc., are acceptable only as secondary sources. Articles on the websites of groups with an overt agenda – CopBlock, We Support Our Police, CopWatch and similar groups – are legitimate outlets to discuss issues, but in our view are not sufficiently objective for our purposes except in cases in which there are no other sources and there is evidence that an event occurred. We believe that the watchdog aspects that call attention to potentially illegal activity by police these groups provide are an important First Amendment-protected activity. However the veracity of claims they make is often difficult to verify.
Known Limitations of Data
The StreetCred PKIC methodologies are designed to reduce to the lowest number possible those incidents in which we only have the word of the police as to what happened in an incident. Specifically, records that contain official autopsy data, prosecution or grand jury testimony, information that comes from witnesses (who either support or dispute the account of the police), video and audio (taken by police, surveillance cameras or bystanders), timestamps on 9-1-1 calls, interviews with 9-1-1 callers, computer-aided dispatch records and other objective records, are more reliable than those that do not contain these entries. There are incidents in the PKIC database in which it is impossible to verify claims and statements made by police, and we acknowledge this. Such is the nature of policing. This means that researchers are unable, in some cases, to verify claims by police as to specifics of an incident. It also means that researchers are unable to reach conclusions as to whether the police behave differently in cases when they know that they are being observed by others. There are incidents in the PKIC database for which the only sources are media reports. While sub-optimal, using our methods and techniques to assure multiple references and quotes from officials, we feel these records are better than no record.
Cause of Death
News reports try to get cause of death correct, yet it is, more often than one would think, difficult to understand the cause of death until an autopsy is released. PKIC prefers publicly released autopsy results, or news reports that quote autopsy results, over news reports that state “probable” cause of death.
A word about, “Homicide”
Homicide is a word that is somewhat confusing to those outside the criminal justice system. As a medical finding, “homicide” means, “Death at the hands of another.” The medical finding of homicide is not a statement of culpability, or an indictment for a crime. Not all homicides are crimes. For example, a killing by one person of another self-defense, or in the defense of another, is generally not a crime. Therefore, an autopsy finding of manner of death of, “Homicide” does not necessarily mean that a crime has been committed. It is included to distinguish the manner of death from other manners of death, such as “natural causes,” “suicide,” “drug overdose,” “accidental death,” and other findings. The criminal term, “Homicide” refers to the illegal taking of the life of a human being by another. This refers, generally, to the crimes of capital murder, or murder, or manslaughter.
A word about, “Armed”
We consider a person to be armed if they are holding an object that may appear to a reasonable person to be a gun, knife, or any object that may be used as a deadly weapon, or if they are behaving in a manner that a reasonable person would confuse for that of a person drawing, displaying or otherwise brandishing a deadly weapon. While civilians and those without police training sometimes have a difficult time understanding how a police officer could mistake an air gun or a replica for a real gun, this is a matter of law and precedent. There is one record in the PKIC database that may need to be cleared, however we do not yet have enough information. Ernesto Canepa was apparently unarmed when killed by police, however police have said that they recovered a BB gun from his vehicle. If later updates to the story reveal that Canepa was holding or pointing the BB gun, we would consider him to have been armed at the time of the incident and therefore out of the scope of this dataset. There is one other case in the PKIC data in which a man was holding a cell phone that police later stated appeared to be a gun; this is more complex, and we have left that case in the dataset.
Almost all news reports report on the race of the decedent. The race of the officer(s) involved are less frequently stated. In order to determine the race of the officer involved, we read primary and secondary sources seeking specific articulation of the race of the officer. This includes sentences such as, “Officer X, who is white…” or the like. Other constructs, such as, “Mr Y, a black man, was killed by a white police officer,” are only acceptable if the source provides further information about the provenance of this determination. We have read several news articles that state constructs like that in cases that turned out to be false – the officer was in fact not the stated race. Further information can be the name of the officer and his rank, or, if the name is not released, a reference to an official statement by police that contains the racial characteristics of the officer involved. A photo of the officer in an article that is not accompanied by a specific statement of the race of the officer is not acceptable evidence of the race of the officer.
Almost all news reports report on the gender of the decedent. The gender of the officer(s) involved are less frequently stated. In order to determine the gender of the officer involved, we read primary and secondary sources seeking specific articulation of the gender of the officer, however this is not as specific as race. Many articles will specifically state “female officer,” while others will indirectly state it, for example, “Police have not released the name of the officer involved in the shooting, but state she is a six-year veteran of the force,” or “Police say the officer deployed his TASER…” It is also marginally acceptable to infer the gender of the officer through the officer’s obviously male or female name, such as “Michael” or “Elizabeth,” however great care must be taken when dealing with names that are dual-gender in the American culture, such as “Ashley,” “Blair”, “Sydney,” and the like. A photo of the officer in an article that is not accompanied by a specific statement that the officer pictured is the officer involved in the incident.
StreetCred Software, Inc., sponsored the creation and maintenance of the StreetCred PKIC dataset. The dataset is non-commercial. The technical director for the StreetCred PKIC Project, and the StreetCred VisionTM data-viewing and manipulation tool is James Jelinek.
StreetCred PKIC conducted, and continues, a peer-review process. Peer review in an open data-set means that we seek input on issues of:
- Data Quality – Is the data accurate and as complete as possible?
- Data Structure – Does our methodology make sense, and how can we improve it?
- Data Normalization – Are we providing data that can be used by everyone?
- Errors & Omissions – What do we have wrong, what can we add?
- Bias – How do we succeed in our project goal to gather un-biased and objective data, and how do we uncover bias in our figures?
If you would like to become a peer reviewer, please contact us.
The initial release of the StreetCred PKIC dataset was peer-reviewed. Reviewers include:
- Nick Selby, Chief Executive Officer, StreetCred Software, Inc; & Police Detective, Dallas-area agency;
- D. Brian Burghart, Fatal Encounters & Reno News & Review
- Andrew Hay, Director of Research, OpenDNS;
- Eric Olson, Senior Vice President, Cyveillance;
- Roger Lee, F1 Consulting
- James Jelinek, Senior Ruby On Rails Developer/DevOps Engineer StreetCred Software, Inc;
- Ed Flosi, President and CEO, PROELIA Defense and Arrest Tactics, LLC
- Ryan George, Lyft
- Corinna Selby;
- Wim Remes;
- boB Rudis;
- Dean Cornelison;
- Elisabeth Nybo, Santa Barbara County Sheriff’s Office;
- Tiffany Andrews, Code for America community;
- Adam Becker, Code for America community