Data last updated 05/01/2022

The full methodology can be found here.

Our previous methodology primarily sourced data from Fatal Encounters, The Washington Post, publicly-accessible media sources, and official data sources from local and state agencies required to report this data (e.g., California Department of Justice).

We posit that the data collected, coded, and made available belongs to the public and is not and should not be owned by any platform. We encourage communities, organizations, and other individuals to develop their own databases - this only strengthens the integrity and validity of the data by allowing us to study the differences across datasets and improve data collection efforts when there are attempts to harmonize and study the data. For example, a recent study has found that while the three major datasets are similar, they have started to become more dissimilar over recent years (Comer and Ingram, 2022).

Thus, we are always open to partnerships and collaborations to improve the methodology or share best practices. We center co-creation over ego, which can be a departure from orthodox academic norms and culture.

We want to acknowledge and thank Samuel Sinyangwe, a co-founder of Campaign Zero, for leading this work at Campaign Zero and the development of the  initial methodology. We also want to acknowledge and thank D. Brian Burghart, founder of Fatal Encounters, who advised the initial development of the project and continues to advise and collaborate on the methodology.

We are working to make our methodology more rigorous and efficient by partnering with different organizations and automating how we collect and validate data. We’re constantly striving to improve our methods and systems. Please contact us if you have any suggestions or would like to partner in any way.

Our methodology below was last updated on Mar 1, 2022.

Data Collection Mechanism

Mapping Police Violence integrates data from a number of sources, most notably Fatal Encounters. In order to ensure a smooth transition through different methodologies, we were grateful for the partnership with Fatal Encounters and the expertise from its Principal Investigator, D. Brian Burghart, in both ensuring a natural transition and for recommending a methodology that allows the public to recreate the data. Fatal Encounters was the first attempt to collect data on police killings and is the most comprehensive dataset. We could not have improved the methodology without their support and expertise.

While we strive to employ official data sources from local and state government agencies, we believe it is important to continue collecting data from publicly-accessible media sources. This allows us to identify gaps in government data, and further triangulate and validate the data.

Together with our partners at Meltwater, we have made data collection more rigorous and comprehensive than the previous approach which relied primarily on Google News Alerts. We are continuing to develop automated systems to reduce potential human coding errors.

The steps we now follow to collect data are:

  • 1
    Incident Identification is done primarily through the following systems:
    • A logo for Meltwater
    • A logo for Google News Alerts
  • 2
    Incident Information Extraction

    Researchers extract out demographic and contextual data from the source

  • 3
    Incident Coding & Review

    Each Incident is reviewed for unique variables and coded by researchers

  • 4
    Second-Review System

    Second-reviewer corrects and updates any incorrect or missing information which triggers incident to be published onto MPV database.

  • 5
    Incident Validation

    System conducts periodic review of incidents to validate and match with following data sources:

    • Fatal Encounters
    • The Washington Post
    • Henry A. Wallace Police Crime Database
  • 6
    Missing Data

    System conducts periodic review of incidents to input missing information from initial incident input

Cadence of Update

It is important to note that incidents are reviewed at different times despite when the incident occurs. This can be for several reasons including alerts and news incidents being published later or second reviewers flagging incidents for another review given data and incident reliability concerns. Thus, the data presented on the live tracker does not necessarily mean that all incidents until that date are included. Our team will do their best to keep it as comprehensive and updated as possible.

  1. We have automated the ingestion of Google and Meltwater alerts into our AirTable.
  2. Reviewers are assigned automatically every day and they will indicate after an incident is reviewed.
  3. Second reviewers will indicate after they have completed review which will automatically trigger the incident to be published onto the database.

Second-Review System

The previous system employed Google Sheets to collect and code data. While Google Sheets allows for many advantages, it makes it difficult to deploy a systematic approach to allow for second-review in the way AirTable permits.

We believe having a two-tiered review system is critical to both detecting human errors in addition to machine errors such as news alert concerns. This will provide further validity and improve data integrity. Nevertheless, if you find any errors or want to flag any concerns, please reach out to Campaign Zero. Your feedback and concerns only co-creates a better platform.

Coding of Incidents

Armed/Unarmed Status

A person is coded as Unarmed/Did Not Have a Weapon in the database if they were one or more of the following:

  1. not holding any objects or weapons when killed
  2. holding household/personal items that were not used to attack others (cellphone, video game controller, cane, etc.)
  3. holding a toy weapon (BB gun, pellet gun, air rifle, toy sword)
  4. an innocent bystander or hostage killed by a police shooting or other police use of force
  5. a person or motorist killed after being intentionally hit by a police car or as a result of hitting police stop sticks during a pursuit

A person was coded as having a Vehicle as a weapon if they were one or more of the following:

  1. a driver who was killed while hitting, dragging or driving towards officers or civilians
  2. a driver who was driving and/or being pursued by police at high speeds, presenting a danger to the public
  3. People who were killed by a civilian driver or crashed without being hit directly by police during a police pursuit are not included in the database. Note that an estimated 300 people are killed in police pursuits each year and only a small proportion of these cases are included in the database (most deadly pursuits end after the driver crashes themselves into something or hits a civilian vehicle without being directly rammed/hit by police).

A person was coded as Allegedly Armed in the database if they:

  1. were alleged to have possessed objects or weapons in circumstances other than those stated above

Alleged Threat Level

Extracted from Washington Post Methodology which can be accessed here.

MPV researchers are only responsible for coding incidents where a gun is not the highest level of force applied in the incident since WaPo data only includes observations with a gun present.

“The threat_level column was used to flag incidents for the story by Amy Brittain in October 2015.

As described in the story, the general criteria for the attack label was that there was the most direct and immediate threat to life. That would include incidents where officers or others were shot at, threatened with a gun, attacked with other weapons or physical force, etc. The attack category is meant to flag the highest level of threat. The other and undetermined categories represent all remaining cases. Other includes many incidents where officers or others faced significant threats.”

Fleeing

Extracted from Washington Post Methodology which can be accessed here.

“News reports have indicated the victim was moving away from officers

  1. Foot
  2. Car
  3. Not fleeing

The threat column and the fleeing column are not necessarily related. For example, there is an incident in which the suspect is fleeing and at the same time turns to fire at gun at the officer. Also, attacks represent a status immediately before fatal shots by police while fleeing could begin slightly earlier and involve a chase.”

Encounter Types and Initial Reported Reason for Encounter

The methodology for Encounter Type and Initial Reported Reason was created by Samuel Sinyangwe. We believe it is important to continue the coding of this variable. The coding for these unique variables started in 2017. MPV has expanded the scope of data collection to include information on the initial reported reason(s) for police to be on the scene prior to using deadly force. This information is obtained from a review of existing media reports on each case as well as statements from police, prosecutors, and other officials. These initial reported reasons are grouped into broader Encounter Types that are standardized within the following taxonomy, ranked in order of severity whereby cases are coded according to the most severe encounter type:

  1. Violent Crime includes cases where officers engaged the person because of reported or suspected murder, rape, robbery, aggravated assault, assault with a deadly weapon, or shooting/shots fired including reported attempts to commit these offenses and executing warrants related to these offenses.
  2. Other Crimes Against People includes offenses not classified as Part 1 Violent Crimes but nevertheless involve causing harm to other people including assault (without a weapon), battery, sex offenses not including rape, domestic assault (without a weapon), child abuse, kidnapping, and any warrants related to these offenses.
  3. Person with a Weapon is when a person is reported for being seen with a gun or other weapon, though no violent crimes or crimes against people were reported.
  4. Other Non-Violent Offenses are cases involving alleged offenses that are not classified as violent crimes or crimes against people (for example, cases involving alleged drug offenses, trespassing, theft, fraud, etc.), including warrants involving non-violent offenses, and cases where a person is pulled over by police for a specific non-traffic related offense or warrant.
  5. Mental Health/Welfare Check is when police initially encounter the person in response to reports or observations that they were acting "erratically," "suicidal," and/or "disorderly," reports of a person in need of medical/mental health care or otherwise in need of someone to check on their welfare with no other specific offenses alleged.
  6. Domestic Disturbance is when responding to an alleged domestic disturbance. However,  if officers responded to reports that indicated the person was committing a violent crime (i.e. aggravated assault/assault with a deadly weapon) or crime against a person (domestic assault or battery) in the context of a domestic disturbance, the incident is coded as "Violent Crimes or Other Crimes Against People", consistent with the most severe offense alleged.
  7. Traffic Stop is when police initially encounter the person because of a traffic violation or traffic-related offense, regardless of whether any other offenses are reportedly discovered after the person has already been stopped by police.
  8. None/Unknown corresponds to non-criminal situations such as car crashes as well as situations where generalized "suspicion" was cited as the reason for officers to engage with someone despite no specific offense being alleged.