Justin Nix and Jason Lozada are wrong about Mapping Police Violence. Here’s why.
In the absence of reliable police violence data from the federal government, we launched Mapping Police Violence in 2015 to find, analyze and evaluate every case where a person is killed by police in the United States. Today, this database includes 6,838 cases from 2013–2018. This week, two academics — Justin Nix and James Lozada — wrote a paper claiming they found “93 Misclassified Incidents” in our database that they believe should not have been coded as unarmed — which they then use to attempt to discredit recent research that used our data to examine the relationship between police violence and mental health. While I welcome efforts to review the data in the database we built (this is why all the source data for every project we’ve done is posted online), the paper these men produced is deeply flawed and appears to be an attempt to discredit the ongoing work to compile and analyze data to better understand the impact police violence has in America.
As their paper asserts, cases involving police violence — and the information made publicly available about the circumstances behind them — are not always clear-cut to code or evaluate. This is true for us and for anyone doing this work — including for other groups that compile similar data, such as the Washington Post’s Fatal Force project. The goal of Mapping Police Violence database is to understand killings by police in every form that they take. While 9 in every 10 incidents in our database are fatal police shootings, the remainder of cases reflect situations that involve other types of police use of force and police pursuits that are often more complicated to evaluate. Moreover, there is no universal definition of what constitutes “unarmed” in the context of an incident of police violence. For example, if someone is holding a spoon are they unarmed? What about a fork or screwdriver? How do you factor in the way they’re using the object at the time? This is why we have always articulated clearly our methodology on our site, reflecting how we’ve managed the complexity of coding these incidents. After re-examining the 93 cases they cite as “misclassified” in their paper, we believe the vast majority of these cases to be properly coded as unarmed while the remaining cases reflect ambiguities inherent to performing analyses of often incomplete and conflicting police data. Here’s the breakdown of these incidents and why they are included in our database:
1. Police Killings that were Domestic Violence Incidents
Of the 93 incidents Nix and Lozada claim are “misclassified” as unarmed, 14 of these incidents were people who were killed by law enforcement officers in domestic violence incidents. We believe these killings by police officers should be accounted for in any database and analysis that attempts to evaluate the impact of police violence in America, especially given research showing that domestic violence incidents are far more likely to occur at the hands of police and that police often use their own department-issued weapons to carry out this violence. Our database determines how many people were killed by police — and that includes intimate partners killed by police officers.
2. Killings of People who were alleged to have Toy Guns
Another 27 incidents involved people who were alleged to have a toy gun or toy sword at the time they were shot by police. Nix and Lozada claim these incidents should be coded as armed. This would mean that children like Tamir Rice who were killed by police while playing with a toy gun would be systematically defined as armed for the purposes of analyzing police behavior. I disagree. A toy is not a weapon. A person holding a toy is not an armed assailant. Moreover, communities know the difference between police shooting someone with a toy and someone who’s using a real gun and likely will react differently to these cases. Precisely because we recognize there is disagreement among researchers on this issue and researchers may want to apply a different methodology to understanding these cases — our database notes which incidents involve someone who was holding such a toy at the time they were killed. As such, our database enables researchers to analyze these incidents separately, exclude them entirely, or code them as unarmed.
3. Pedestrian and Pursuit Related Incidents
Another 21 incidents are unarmed people killed as the result of police pursuits; drivers, pedestrians and bystanders struck and killed by police vehicles; and unarmed passengers who were killed in vehicles that were shot at by the police. I believe these incidents should be included in analyses of killings by police, as they represent a major contributor to civilian deaths from police operations. For example, analyses have found as many as 300 people are killed in police pursuits each year. Collecting data on these incidents can be used to inform police pursuit policies that can limit these police actions and save lives. In contrast to other databases like Fatal Encounters, we employ a more conservative methodology for deciding which of these incidents to include. We do not include deaths of civilians hit by other civilians who are fleeing police (even though this would probably be a better indicator of the overall human “cost” of police pursuits) — but do include civilians who are killed as the result of being directly chased by police and civilians killed after being hit directly by police vehicles. While people who are pedestrians and passengers who are not possessing or using a weapon are coded as unarmed (as they should be), those who are using vehicles to allegedly threaten officers or civilians are appropriately coded in our database as weapon = “vehicle.”
4. Deaths in Custody
Another 17 incidents noted by Nix and Lozada were deaths in custody — 14 of which occurred soon after law enforcement used force against the person and 1 (Kyam Livingston) which occurred after police locked her in jail and then repeatedly refused to provide requested medical assistance. Again, these cases are difficult to categorize because information about them is often severely limited. In many of these cases, deaths are also attributed to pre-existing medical conditions that can be aggravated by police use of force and physical restraints. Moreover, given the lack of transparency around in-custody deaths, the actual number of people killed under these circumstances is likely far higher than the information that is publicly available for us to evaluate. For example, an estimated 1,000 people die in jails in America each year — only a fraction of these deaths are included in our database since we are coding the data according to these specifications. Additionally, two cases involved people who died in custody under suspicious circumstances that were called suicides by police investigators — Tyree Woodson who Baltimore police claimed somehow shot himself in jail after being arrested, searched and placed in restraints and Sandra Bland who was arrested, threatened with a taser and then, police claim, hung herself in Waller County jail. The police narrative in these cases has differed substantially from what subsequent reporting (and video evidence in Bland’s case) has found. We note the inclusion of cases with these special circumstances in the methodology for this reason and understand that these cases may be coded differently in other databases — nevertheless these cases are outliers compared to the total number of cases in the data and academics can choose how to account for these specific incidents in their methodological framework.
5. Vehicles and Screwdrivers as “Weapons”
Nix and Lozada claim there were 3 incidents where the person was coded as unarmed despite having a screwdriver and another 3 incidents where the person did not have a weapon but was allegedly driving towards officers when they were shot. I agree that the 3 incidents involving vehicles should be reclassified to “vehicle,” consistent with how we’ve been coding these incidents generally. And we have and will continue to code incidents involving screwdrivers as “unclear” reflecting the gray area between armed and unarmed that are associated with these situations. These changes are now reflected on our site.
6. Unarmed but allegedly “Reaching for a Weapon”
Nix and Lozada also note that 5 cases involved people who did not have a weapon in their possession but who police claim were “reaching for the officer’s weapon” when the police killed them. A straight-forward interpretation of unarmed indicates the person did not have a weapon on them when the police killed them. We believe these cases were appropriately coded as unarmed. Had the person actually taken the weapon into their possession, then they would have been coded as armed. It is seriously problematic for Nix and Lozada to recommend that all cases where police merely claim that an unarmed person “reached for an officer’s gun” be removed from analyses, a choice that would elevate a police narrative over the simple fact that the person did not actually have any weapons in their possession when they were killed by police. Finally, there were 4 cases where Nix and Lozada claim the person was armed with a gun but was coded as unarmed in the database — Freddie Blue, Antoine Dominique Hunter, Dontre Bennett and Leroy Browning. Nix and Lozada apparently misclassified at least two of these cases — Freddie Blue reportedly had a toy gun and Leroy Browning was alleged to have reached for an officer’s gun but was not reported to have removed it from its holster/to be in possession of it. In their paper, they justify coding these cases as armed with a gun because it says so in the Washington Post database. However, both they and the Post appear to have gotten these wrong, according to local reporting and court documents. And while there were conflicting reports about the shooting of Antoine Dominique Hunter (for example, the LA Times referred to him as unarmed) and Donte Bennett (some witnesses said he was holding a cell phone), since most articles have indicated that there was a gun recovered from the scene we’ve reclassified those two cases to “armed” pending further information being made available. Again, we continually update our data in response to new information reported in the news, the release of new video footage, findings of ongoing investigations, and in response to the feedback we get from researchers and data scientists all across the country who regularly review and utilize our data in their work.
Altogether, of the 93 cases Nix and Lozada claimed were “misclassified” in our database — our review finds 27 cases that could reasonably have been coded differently and even these cases tend to reflect ambiguities that are by no means easy to resolve. Their paper, and their analysis of the 2018 study from Bor et al., should be revised, at minimum, to reflect these numbers. I look forward to seeing the revised results of that analysis.
More broadly, however, Nix and Lozada could have simply reached out with their concerns about data coding so we could work together to address them. Instead, they chose to author a paper designed to discredit a project that has meticulously built the most comprehensive database of police violence in America — as well as the research that has been produced with this data. We welcome serious critique and analysis — that is how we better understand the causes and solutions to police violence in America. The effort by Nix and Lozada, however, appears motivated by something else entirely.
