Kate Wheeling is a staff writer at Pacific Standard, where she specializes in criminal justice and the environment. She was previously an associate editor and editorial fellow.
A new study uses a statistical technique from wildlife ecology to estimate the number of people killed by cops. It turns out every database is undercounting them.
How many people are killed by police every year in the United States? It depends on whom you ask.
The federal government tries to track this subset of the population with databases like the National Vital Statistics System, which is based on death certificates. As public attention on police violence has increased in recent years, media organizations began making databases of their own—like the Guardian's The Counted or the Washington Post's Fatal Force—to track law enforcement-related deaths. Comparisons between the data sets suggested that the official government data was severely undercounting police-related deaths. However, no one really knew how accurate those media databases were either.
Now, in a new study published today in PLoS Medicine, researchers borrowed techniques from wildlife ecology to estimate how many people are really killed by police officers in the U.S. They found that, while the media database The Counted documented roughly twice as many cases of police-related deaths than the NVSS, it still missed up to 7 percent of cases.
The researchers matched cases of police-related deaths from NVSS mortality records and The Counted, and used a statistical tool called capture-recapture analysis to estimate the number of cases missing from both data sets. Wildlife ecologists often use this technique to estimate the size of a wild population. They'll trap animals, tag and release them, and then try to trap them again. "With this method, if you have two ways of collecting data, you look at to what degree do they overlap," says Justin Feldman, a doctoral candidate at the Harvard T.H. Chan School of Public Health and lead author on the new study. If there's not a lot of overlap, the estimate of uncounted animals—or, in this study, cases of police-related deaths—would be large, Feldman explains. Conversely, a large amount of overlap would lead to a small estimate of uncounted cases.
The NVSS classifies police-related incidents based on death certificates; it only captures those whose death certificates explicitly state that the deceased passed away due to injuries from an altercation with law enforcement. The Counted, meanwhile, collects relevant cases from news reports.
In 2015, the NVSS recorded 523 law enforcement-related deaths, while The Counted identified 1,086 such cases. There was significant overlap between the two sources, according to the new study; 487 cases appeared in both lists. From this data, the authors estimated that at least 1,166 people were killed by police in the U.S. in 2015. The news-based system counted over 93 percent of the deaths, while NVSS captured less than 45 percent.
"The two main messages, I think, are pretty simple," Feldman says. "One is that these media based sources do a pretty good job of capturing the number of deaths; and two, that the death certificate data, which is supposed to be the gold standard for causes of death in the United States, is bad."
The researchers also found that deaths in poorer counties were more likely to be misclassified in NVSS records, as were deaths from injuries other than gunshot wounds. Improving federal databases could be as simple as expanding training for medical examiners and coroners, to ensure that they include police involvement on death certificates. Legislation to change how these deaths are reported might also improve data sets, according to Feldman. Tennessee, for example, passed a law last year requiring state police to report all fatal police shootings to the state health department, though it remains to be seen if the new law will help close the data gap.
But incorporating news reports into official counts will almost certainly make a difference. "This approach that the Guardian and the Washington Post have taken of capturing these deaths based on local media reports is very good, it captures nearly all—93 percent—of the deaths," Feldman says. "The Department of Justice actually adopted this methodology too." In 2016, the DOJ unveiled a new system, modeled after The Counted, which requires police departments to report fatalities involving officers to the federal agency quarterly, which agency officials would use to confirm police-related deaths reported by the media. "But it's not really clear under the new administration what's happening with that," Feldman says.
Next Feldman and his colleagues plan to look at non-fatal injuries as well, to find out if emergency rooms are better at classifying police-related injuries than mortality records. "Having better data will better inform the conversation about what to do about policing in the United States," he says. In other words, researchers can't evaluate policies aimed at reducing police violence when the underlying data on police-related injuries and deaths is flawed.
One thing we know for sure: Sound policies to reduce police violence are desperately needed. The Counted database currently counts 1,093 police-related deaths in 2016—and now we know even that database is missing some cases.