The Center for Data Innovation spoke with Thomas Hargrove, founder of the Murder Accountability Project, a nonprofit working to improve accuracy of accounting for homicides in the United States. Hargrove discussed how he developed an algorithm that can help detect serial killers, as well as the challenges to obtaining comprehensive data about murder from law enforcement agencies.
This interview has been lightly edited.
Joshua New: Can you describe why you founded the Murder Accountability Project?
Thomas Hargrove: I was a newspaper reporter for 37 wonderful years, including a two-year stint as a White House correspondent for the now-defunct Scripps Howard News Service. In 2004, I conducted an inquiry into the optional crime called prostitution. It’s an “optional crime” because, in many cities, anti-solicitation laws are hardly ever enforced. In many cities, anti-solicitation laws are hardly ever enforced. Others have vice squads who aggressively look for prostitutes, while in other cities only the “johns” are arrested. To sort out how prostitution is regulated, I contacted the database library at the University of Missouri’s Journalism School and purchased a copy of the FBI’s Uniform Crime Report for 2002, the most recent year for which complete data were available.
At no extra cost, the data librarian included a copy of a file I’d never heard of before, the Supplemental Homicide Report (SHR). When I opened the database, I looked up row after row of the case-level information of every murder reported to the FBI. This included the victim’s sex, race, age, ethnicity, the year and month of the murder as well as the method of killing and other details about the circumstances of the crime. Demographics about the killer was also included in those cases in which an arrest had been made.
There’s no way of telling where these ideas come from. But the first time I looked at the SHR, I immediately wondered if it would be possible to teach a computer how to spot serial killings. I worked at the Birmingham Post-Herald newspaper in Alabama at the time the entire South was transfixed by the Atlanta child murders in the late 1970s and early 1980s. The Atlanta Police Department was criticized for not recognizing sooner that too many killings of young black males were going unsolved. It’s a phenomenon criminologists call “linkage blindness.” If someone is murdered, a detective is assigned to the case. When someone else is killed, a different investigator is assigned. If both victims had a common killer, it’s unlikely detectives will compare notes to discover commonalities in the cases. It is such a common problem that many criminologists fear most homicidal series remain unconnected.
For six years, I urged my editors to let me try to develop an algorithm to spot serial killers. They agreed in 2010 and we began a national reporting project called “Murder Mysteries.” The goal was to study the broader problem of unsolved homicide as well as to look into the problem of serial murder. We found that every year, about 5,000 or 6,000 homicides are not “cleared,” meaning no one was arrested in connection with the killing. The rate at which we solve murder has declined over time. In the mid 1960s, America was solving about 90 percent of our killings. In recent years, that has dropped to only 62 percent or so. We estimate that from 1980 through 2015, more than 220,000 murders were unsolved. In a growing number of major cities, less than half of homicides are cleared through arrest.
When I retired in 2015, I founded a nonprofit group called the “Murder Accountability Project” (MAP) to track the many disturbing trends in homicides. We now have a Board of Directors that includes retired FBI investigators, police officials, criminologists and retired investigative journalists. Our purpose is to assist homicide investigators by using our database to test theories when they have suspects who may have committed crimes across multiple jurisdictions. We also want to educate the public about the growing problem of unsolved murders.
New: Just how severe is the lack of accountability for homicide, and how does data come into play?
Hargrove: When we started running the serial-killer algorithm for the first time in 2010, we noticed a cluster of 15 unsolved strangulations of women in the area in and around Gary, Indiana. None of the murders were reported as solved. We contacted the Gary Police Department and were assured there were no unsolved serial killings on their books. They refused our attempts to consider the problem further, even after we’d assembled the names and narratives of most of these murdered women who mostly were found in abandoned properties.
Police in nearby Hammond, Indiana, were summoned to a disturbance in 2014 at a Motel 6 and eventually arrested Darren Deon Vann who confessed to killing a young woman found in the hotel room’s bathtub and to killing many more women going back to the 1990s. Vann led police to the bodies of six more victims found in abandoned properties inside Gary city limits. I contacted Gary Police to inform them that Vann should be considered for many other murders in Gary prior to 2014. Corporal Gabrielle King, a spokeswoman for the Gary Police Department, said she didn’t “know that we have unsolved strangulations.”
In no way do we think Corporal King was deceitful. Police often have difficulty tracking their own records, especially patterns of unsolved homicides. Police were often slow—because of financial constraints—to move to computer records. Old cases are often paper records stored in warehouses. Those records that are in electronic form are not always searchable in ways that would produce a list of, for example, all unsolved strangulations in Gary, Indiana.
Veteran homicide detectives in New York City complain it’s difficult to call up records of their own cases because of coding problems and slow citywide computer systems. If police had difficulty accounting for their own cases, it is virtually impossible to give them access to records in neighboring jurisdictions.
The Murder Accountability Project was established to give detectives easy access to unsolved homicide records within their own jurisdictions and anywhere else in America. Our online database is a powerful tool.
New: Can you describe the algorithm you developed to spot signs of a potential serial killer? What kind of factors do you look at?
Hargrove: The ultimate purpose of the “Murder Mysteries” project was to develop a computer program that might spot serial victims. For this, we had an invaluable teacher in the form of Gary Leon Ridgway, the so-called “Green River Killer,” who was convicted of murdering 48 teenage girls and women in the greater Seattle area during the 1980s and 1990s. This is the most prolific killer in U.S. history, according to court records. If we could find a statistical procedure that could identify Ridgway’s victims, we might have a useful tool for law enforcement. To do this, we used the Supplemental Homicide Report and tried dozens of statistical procedures to see if Seattle stood out as unusual in any way.
First, we looked at whether the presence of a serial killer caused an elevated murder rate for women. Experts at the FBI’s Violent Criminal Apprehension Program told us that 70 percent of the known serial victims in their files were women. But this did not work. We looked at elevated murder rates for women within specific age ranges, also with no luck. We looked at whether the presence of a serial killer could cause elevated numbers of murders using unusual methods of killing such as strangulation, asphyxiation, knives and other cutting instruments, and blunt instruments. Again, no luck. Then we tried looking at whether the presence of a serial killer caused an elevated number of unsolved murders among women.
This seemed to work a little.
The presence of an active serial killer did somewhat reduce the rate at which Seattle police solved female homicides, although this difference was slight. But it suggested that the rate at which failure-to-solve homicides occur could be a factor in locating serial killings.
After months of trial and error, what did work was a technique called “cluster analysis” in which we succeeded in getting America’s hundreds of thousands of usually unrelated murders to form meaningful groups. We give each murder a group number, called a MURDGRP, based upon its county or metro area, the gender of the victim, the age group of the victim and the weapon used. We then use a technique called “aggregation” in which we count how many victims had the same nine-digit MURDGRP, how many of those killings were solved, and what percentage were solved out of total murders. We then look at this aggregated file looking for groups with extremely low solution rates. Anyone can see this algorithm by going to the “Data & Docs” page at our website.
The first time we ran the algorithm, most of Ridgeway’s victims popped up prominently in two clusters of teenagers and young women who’d been murdered in similar ways and for whom almost none of the cases were reported to have been solved. The algorithm also reported dozens and dozens of other highly suspicious clusters of victims. Many of these were known serial killings. But most appeared to be potential series completely unknown to police.
New: You spent several years scraping data from official sources to create a database of hundreds of thousands of unsolved murders, which you make freely available on your website. How challenging was this? Are law enforcement agencies generally pretty good at making this data available?
Hargrove: The volunteers at the Murder Accountability Project have spend thousands of hours gathering and assembling the most complete accounting of homicide available anywhere in the United States. We have gathered tens of thousands of records of homicides that were not reported to the U.S. Department of Justice.
Most of the records we have were provided by the FBI’s Criminal Justice Information Services (CJIS) division in Clarksburg, West Virginia. But CJIS oversees the entirely voluntary reporting programs called the Uniform Crime Report and Supplemental Homicide Report. That means police are not legally required to report crime information to the FBI.
We’ve found that entire states have failed to report tens of thousands of murders to the FBI. In recent years, these states include Florida, Illinois, Alabama, and the District of Columbia.
We were successful in convincing the DC Metropolitan Police Department to resume reporting case level details to the FBI. We also were successful in convincing the Florida Division of Law Enforcement to report information to us, although they still do not report to the FBI.
The Illinois State Police stopped reporting clearance data and case-level details to the FBI in 1994. They refused a Freedom of Information Act request from MAP, prompting us to file a lawsuit in Cook County Chancellery Court in 2015. We reached a settlement in 2016 in which Illinois provided all of the records they possessed, which represented only about 20 percent of the estimated 2,100 unreported murders. Illinois has also pledged to resume reporting Supplemental Homicide Report data starting with 2016 and to be in full compliance with FBI reporting standards by 2021.
We are currently in negotiations with Alabama law enforcement groups. Alabama ceased reporting any data to the Supplemental Homicide Report in 2010. The Alabama Law Enforcement Agency (ALEA) agreed to give us all records reported to them. But many major agencies, including police in Birmingham, Montgomery, and Mobile, have declined to report data to ALEA. We are in continuing negotiations with those agencies.
New: Predictive policing has become a pretty popular buzzword in recent years, but from what I can tell about your experiences with how police manage and use data, this almost seems unrealistic. What can law enforcement agencies do to improve? Or is this something policymakers need to address?
Hargrove: We believe police departments across America would benefit from improved crime accounting, especially given the recent advances in artificial intelligence. Although the early attempts have produced mixed results, predictive policing can utilize crime data to more effectively allocate limited personnel to specific neighborhoods at specific times when crimes are more likely to occur.
The board of directors of the Murder Accountability Project believes the worsening declines in homicide clearances in recent years are a failure by local politicians to provide adequate resources to combat major crimes. We don’t have enough trained detectives with full resources —overtime pay to chase evidence in the critical early hours of an investigation, rapidly responding forensic evidence technical support, and adequate laboratory response.
It should also be said that the directors of the Murder Accountability Project believe what we do should be a government function rather than an effort by a small nonprofit group of volunteers. Reporting unsolved homicides to a central authority should be mandatory. And there should be a process of multi-layered review at the state and federal level for failed local homicide investigations.
In short, we’d be grateful if Congress were to act to put the Murder Accountability Project out of business by mandating reforms that would render our mission moot. Given the political realities, however, we expect MAP’s mission to continue into the foreseeable future.
We urge anyone who supports our mission to consider making a tax-deductible donation, no matter how small, to our group and its ongoing mission to improve the accountability of murder.