The Serial-Killer Detector
By Alec Wilkinson The New Yorker
A former journalist, equipped with an algorithm and the largest collection of murder records in the country, finds patterns in crime.
Hargrove estimates that two thousand serial killers are at large in the U.S.Illustration by Harry Campbell
Thomas Hargrove is a homicide archivist. For the past seven years, he has been collecting municipal records of murders, and he now has the largest catalogue of killings in the country—751,785 murders carried out since 1976, which is roughly twenty-seven thousand more than appear in F.B.I. files.
States are supposed to report murders to the Department of Justice, but some report inaccurately, or fail to report altogether, and Hargrove has sued some of these states to obtain their records. Using computer code he wrote, he searches his archive for statistical anomalies among the more ordinary murders resulting from lovers’ triangles, gang fights, robberies, or brawls. Each year, about five thousand people kill someone and don’t get caught, and a percentage of these men and women have undoubtedly killed more than once. Hargrove intends to find them with his code, which he sometimes calls a serial-killer detector.
Hargrove created the code, which operates as a simple algorithm, in 2010, when he was a reporter for the now defunct Scripps Howard news service. The algorithm forms the basis of the Murder Accountability Project (map), a nonprofit that consists of Hargrove—who is retired—a database, a Web site, and a board of nine members, who include former detectives, homicide scholars, and a forensic psychiatrist. By a process of data aggregating, the algorithm gathers killings that are related by method, place, and time, and by the victim’s sex. It also considers whether the rate of unsolved murders in a city is notable, since an uncaught serial killer upends a police department’s percentages. Statistically, a town with a serial killer in its midst looks lawless.
In August of 2010, Hargrove noticed a pattern of murders in Lake County, Indiana, which includes the city of Gary. Between 1980 and 2008, fifteen women had been strangled. Many of the bodies had been found in vacant houses. Hargrove wrote to the Gary police, describing the murders and including a spreadsheet of their circumstances. “Could these cases reflect the activity of one or more serial killers in your area?” he asked.
The police department rebuffed him; a lieutenant replied that there were no unsolved serial killings in Gary. (The Department of Justice advises police departments to tell citizens when a serial killer is at large, but some places keep the information secret.) Hargrove was indignant. “I left messages for months,” he said. “I sent registered letters to the chief of police and the mayor.” Eventually, he heard from a deputy coroner, who had also started to suspect that there was a serial killer in Gary. She had tried to speak with the police, but they had refused her. After reviewing Hargrove’s cases, she added three more victims to his list.
Four years later, the police in Hammond, a town next to Gary, got a call about a disturbance at a Motel 6, where they found a dead woman in a bathtub. Her name was Afrikka Hardy, and she was nineteen years old. “They make an arrest of a guy named Darren Vann, and, as so often happens in these cases, he says, ‘You got me,’ ” Hargrove said. “Over several days, he takes police to abandoned buildings where they recover the bodies of six women, all of them strangled, just like the pattern we were seeing in the algorithm.” Vann had killed his first woman in the early nineties. In 2009, he went to jail for rape, and the killings stopped. When he got out, in 2013, Hargrove said, “he picked up where he’d left off.”
Researchers study serial killers as if they were specimens of natural history. One of the most comprehensive catalogues is the Radford Serial Killer Data Base, which has nearly five thousand entries from around the world—the bulk of them from the United States—and was started twenty-five years ago by Michael Aamodt, a professor emeritus at Radford University, in Virginia. According to the database, American serial killers are ten times more likely to be male than female. Ray Copeland, who was seventy-five when he was arrested, killed at least five drifters on his farm in Missouri late in the last century, and is the oldest serial killer in the database. The youngest is Robert Dale Segee, who grew up in Portland, Maine, and, in 1938, at the age of eight, is thought to have killed a girl with a rock. Segee’s father often punished him by holding his fingers over a candle flame, and Segee became an arsonist. After starting a fire, he sometimes saw visions of a crimson man with fangs and claws, and flames coming out of his head. In June of 1944, when Segee was fourteen, he got a job with the Ringling Brothers circus. The next month, the circus tent caught fire, and a hundred and sixty-eight people were killed. In 1950, after being arrested for a different fire, Segee confessed to setting the tent ablaze, but years later he withdrew his confession, saying that he had been mad when he made it.
Serial killers are not usually particularly bright, having an average I.Q. of 94.5, according to the database. They divide into types. Those who feel bound to rid the world of people they regard as immoral or undesirable—such as drug addicts, immigrants, or promiscuous women—are called missionaries. Black widows kill men, usually to inherit money or to claim insurance; bluebeards kill women, either for money or as an assertion of power. A nurse who kills patients is called an angel of death. A troller meets a victim by chance, and a trapper either observes his victims or works at a place, such as a hospital, where his victims come to him.
The F.B.I. believes that less than one per cent of the killings each year are carried out by serial killers, but Hargrove thinks that the percentage is higher, and that there are probably around two thousand serial killers at large in the U.S. “How do I know?” he said. “A few years ago, I got some people at the F.B.I. to run the question of how many murders in their records are unsolved but have been linked through DNA.” The answer was about fourteen hundred, slightly more than two per cent of the murders in the files they consulted. “Those are just the cases they were able to lock down with DNA,” Hargrove said. “And killers don’t always leave DNA—it’s a gift when you get it. So two per cent is a floor, not a ceiling.”
Hargrove is sixty-one. He is tall and slender, with a white beard and a skeptical regard. He lives with his wife and son in Alexandria, Virginia, and walks eight miles a day, to Mount Vernon or along the Potomac, while listening to recordings of books—usually mystery novels. He was born in Manhattan, but his parents moved to Yorktown, in Westchester County, when he was a boy. “I lived near Riverside Drive until I was four,” he said. “Then one day I showed my mom what I learned on the playground, which is that you can make a switchblade out of Popsicle sticks, and next thing I knew I was living in Yorktown.”
Hargrove’s father wrote technical manuals on how to use mechanical calculators, and when Hargrove went to college, at the University of Missouri, he studied computational journalism and public opinion. He learned practices such as random-digit-dialling theory, which is used to conduct polls, and he was influenced by “Precision Journalism,” a book by Philip Meyer that encourages journalists to learn survey methods from social science. After graduating, in 1977, he was hired by the Birmingham Post-Herald, in Alabama, with the understanding that he would conduct polls and do whatever else the paper needed. As it turned out, the paper needed a crime reporter. In 1978, Hargrove saw his first man die, the owner of a convenience store who had been shot during a robbery. He reported on a riot that began after police officers shot a sixteen-year-old African-American girl. Once, arriving at a standoff, he was shot at with a rifle by a drunk on a water tower. The bullet hit the gravel near his feet and made a sound that “was not quite a plink.” He also covered the execution of a man named John Lewis Evans, the first inmate put to death in Alabama after a Supreme Court abrogation of capital punishment in the nineteen-sixties and seventies. “They electrocuted people in Alabama in an electric chair called the Yellow Mama, because it was painted bright yellow,” Hargrove said. “Enough time had passed since the last execution that no one remembered how to do it. The first time, too much current went through too small a conduit, so everything caught fire. Everyone was crying, and I had trouble sleeping for days after.”
In 1990, Hargrove moved to Washington, D.C., to work for Scripps Howard, where, he said, “my primary purpose was to use numbers to shock people.” Studying the Social Security Administration’s Death Master File—“where we will all end up one day,” Hargrove said—he noticed that some people were included for a given year and dropped a few years later: people who had mistakenly been declared dead. From interviews, he learned that these people often have their bank accounts suddenly frozen, can’t get credit cards or mortgages, and are refused jobs because they fail background checks. Comparing a list of federal grants for at-risk kids in inner-city schools against Census Bureau Zip Codes, he found that two-thirds of the grants were actually going to schools in the suburbs. “He did all this through really clever logic and programming,” Isaac Wolf, a former journalist who had a desk near Hargrove’s, told me. “A combination of resourceful thinking and an innovative approach to collecting and analyzing data through shoe-leather work.”
In 2004, Hargrove was assigned a story about prostitution. To learn which cities enforced laws against the practice and which didn’t, he requested a copy of the Uniform Crime Report, an annual compilation published by the F.B.I., and received a CD containing the most recent report, from 2002. “Along with it, at no extra cost, was something that said ‘S.H.R. 2002,’ ” he said. It was the F.B.I.’s Supplementary Homicide Report, which includes all the murders reported to the Bureau, listing the age, race, sex, and ethnicity of the victim, along with the method and circumstances of the killing. As Hargrove looked through it, “the first thing I thought was, I wonder if it’s possible to teach a computer to spot serial victims.” Hargrove said that for six years he told each of his editors at Scripps Howard that he wanted to find serial killers using a computer, and the response was always, “You’re kidding, right?”
In 2007, Hargrove did an investigation into SIDS, Sudden Infant Death Syndrome, after wondering why, according to the Centers for Disease Control’s infant-mortality records, so many more babies in Florida died from accidental suffocation than did babies in California, even though California had many more babies. During the following year, Hargrove interviewed coroners and pathologists around the country. “A growing number of them began saying, ‘To be honest, I might get in trouble for saying this, but sids doesn’t exist as such,’ ” he said. Hargrove concluded that sids wasn’t a diagnosis or a mysterious disease but the result of people putting babies in their cribs in such a way that they suffocated during sleep. Florida tended to attribute these deaths to accidental suffocation, California to sids. In the aftermath of his story, the C.D.C. created the Sudden Unexpected Infant Death Case Registry to evaluate each death. Frank Lautenberg, a New Jersey senator, met with Hargrove and then introduced the Sudden Death Data Enhancement and Awareness Act, which President Obama signed in 2014. After the sids story, Hargrove’s stock rose “insanely high in the newsroom,” he said. He told his boss that he still wanted to try to teach a computer to detect serial killers, and this time his boss said, “You’ve got a year.”
Hargrove began by requesting homicide reports from 1980 to 2008; they included more than five hundred thousand murders. At the start, he knew “what the computer didn’t know,” he said. “I could see the victims in the data.” He began trying to write an algorithm that could return the victims of a convicted killer. As a test case, he chose Gary Ridgway, the Green River Killer, who, starting in the early eighties, murdered at least forty-eight women in Seattle, and left them beside the Green River. Above his desk, Hargrove taped a mugshot of Ridgway in which he looks tired and sullen. Underneath it, he wrote, “What do serial victims look like statistically?”
Creating the algorithm was laborious work. “He would write some code, and it would run through what seemed like an endless collection of records,” Isaac Wolf told me. “And we did not have expensive computer equipment, so it would run for days. It was sort of jerry-rigged, Scotch-Taped. He was always tinkering.”
“As soon as it clusters into a cultural bubble, it’s done.”
Ridgway was eventually identified by DNA and was arrested in 2001, as he was leaving his job at a Kenworth truck plant, where he had worked as a painter for thirty-two years. He told the police that strangling women was his actual career. “Choking is what I did, and I was pretty good at it,” he said. Ridgway’s wife—his third—was astonished to find out what he had done. They had met at a Parents Without Partners gathering and had been married for seventeen years. She said that he had always treated her like a newlywed. Ridgway had considered killing his first two wives but decided that he was too likely to get caught. Mostly, he killed prostitutes, and, if he killed one who had money on her, he regarded it as his payment for killing her.
Hargrove began each day with a review of what had failed the day before. He sorted the homicides by type, since he had been told that serial killers often strangle or bludgeon their victims, apparently because they prefer to prolong the encounter. He selected for women, because the F.B.I. said that seventy per cent of serial-killer victims are female. Each test took a day. He had no idea if anything would work. For a while, the only promising variable seemed to be “failure to solve.” “After a hundred things that didn’t work, that worked a little bit,” Hargrove said, holding his right thumb and forefinger close together. “I started making the terms more specific, looking at a group of factors—women, weapon, age, and location.”
With those terms, the algorithm organized the killings into approximately ten thousand groups. One might be: Boston, women, fifteen to nineteen years old, and handguns. Another might be: New Orleans, women, twenty to fifty, and strangulation. Since “failure to solve” had produced results, even if feeble, Hargrove told the computer to notify him of places where solution rates were unusually low. Seattle came in third, with most of the victims women whose cause of death was unknown—unknown because the bodies had been left outside and sufficient time had passed that the coroners could no longer determine how they had died. The computer, Hargrove knew, had finally seen Ridgway’s victims.
By reading meaning into the geography of victims and their killers, Hargrove is unwittingly invoking a discipline called geographic profiling, which is exemplified in the work of Kim Rossmo, a former policeman who is now a professor in the School of Criminal Justice at Texas State University. In 1991, Rossmo was on a train in Japan when he came up with an equation that can be used to predict where a serial killer lives, based on factors such as where the crimes were committed and where the bodies were found. As a New York City homicide detective told me, “Serial killers tend to stick to a killing field. They’re hunting for prey in a concentrated area, which can be defined and examined.” Usually, the hunting ground will be far enough from their homes to conceal where they live, but not so far that the landscape is unfamiliar. The farther criminals travel, the less likely they are to act, a phenomenon that criminologists call distance decay.
Rossmo has used geographic profiling to track terrorists—he studied where they lived, where they stored weapons, and the locations of the phone booths they used to make calls—and to identify places where epidemics began. He also worked with zoologists, to examine the hunting patterns of white sharks. Recently, Rossmo studied where the street artist Banksy left his early work, and found evidence to support the British Daily Mail’s assertion, made in 2008 but never corroborated, that Banksy is a middle-aged man from Bristol, England, named Robin Gunningham.
“In a murder investigation, when you step away from the Hollywood mystique, it’s about information,” Rossmo told me. “In any serial-murder case, the police are going to have thousands and thousands, even tens of thousands, of suspects.” In the Green River case, the police had eighteen thousand names. “So where do you start? We know quite a lot about the journey to a crime. By noting where killings took place or the bodies were discovered, you can actually create probability distributions.” In his book “Geographic Profiling,” Rossmo notes research that found, among other things, that right-handed criminals tend to turn left when fleeing but throw away evidence to the right, and that most criminals, when hiding in buildings, stay near the outside walls.
Using computers to find killers has historical precedent. Eric Witzig, a retired homicide detective and a former F.B.I. intelligence analyst who is on map’s board, worked on the F.B.I.’s Violent Criminal Apprehension Program, vicap, which was started by a Los Angeles homicide detective named Pierce Brooks. Witzig told me how, in the fifties, Brooks worked on the case of Harvey Glatman, who became known as the Lonely Hearts Killer. Glatman was a radio-and-TV repairman and an amateur photographer who would invite young women to model for him, saying that the photographs were for detective magazines. He would tie his victim up for the shoot, and then never remove the bonds. “The victim, a young woman, was not just tied up, but the turns of the bindings were sharp and precise, indicating that the offender took a lot of pleasure in it,” Witzig said.
Brooks started to research the way that some killers seemed to commit the same crime repeatedly. He began putting all his murder records onto three-by-five cards, and after becoming interested in computers in the late nineteen-fifties he asked the L.A. police department to buy him one. He was told that it was too expensive. In 1983, he presented the idea of a homicide-tracking computer database to Congress, after which the F.B.I. offered him a job at Quantico and bought him the equipment to start vicap. The program was meant to be an accessory to investigations, but detectives didn’t take to it. “The first and maybe main problem was the original vicap reporting form,” Witzig said. Brooks wanted to record every element of a homicide and, as a result, there were more than a hundred and fifty questions. “Of course, there was user resistance,” Witzig said. “No one wanted to do more paperwork.” He added that the program had “some of the brightest law-enforcement deep thinkers in the world involved, but we exist at map because they failed.”
Map has its own limitations. Since the algorithm relies on place as a search term, it is blind to killers who are nomadic over any range greater than adjacent counties. There is also a species of false positive that Hargrove calls the Flint effect: some cities, such as Flint, Michigan, are so delinquent in solving murders that they look as if they were beset by serial killers.
Someone versed in statistics can run the algorithm, which appears on map’s Web site. The rest of us, who might, for example, wish to know how many killings are unsolved where we live, can use the site’s “search cases” function. Deborah Smith, who lives in New Orleans, is a hobby map searcher and a forum moderator on Websleuths, an online watering hole for amateur detectives. “I keep spreadsheets of murdered and missing women around the country, with statistics, and I highlight murders that I think might be related,” she told me. “I have them for nearly every state, and that comes from map. If I have a killer, like, say, Israel Keys, who was in Seattle about fifteen years ago, I’ll look up murders in Seattle and parts of Alaska, because he lived there, too, and see if there were any the police might have overlooked.” She added, “map is just extremely, extremely useful for that. There isn’t really anything else like it.”
map’s board hasn’t determined what to do with the algorithm’s findings, however, and the question presents moral and practical difficulties for Hargrove. “We have to figure out our rules of engagement,” he told me. “Under what scenario do we start calling police?” A few months ago, Hargrove informed the Cleveland police that there appeared to be about sixty murders, all of women, that might involve a serial killer, or, from the range of methods, perhaps even three serial killers. Twelve of the women had convictions for prostitution, and their bodies were found in two distinct geographic clusters. Hargrove can’t say anything about his exchange with the Cleveland police, because map’s rules dictate that such communications are privileged. The police wrote me that, as a result of Hargrove’s analysis, “a small taskforce is being considered to look at several unsolved homicides.” The head of the department’s special-investigations bureau, James McPike, told the Cleveland Plain-Dealer, of map, “We’re going to be working with the group to help us identify what we might be able to do.”
Hargrove is pleased about the investigation but he also worries that something may go awry. “What if they arrest the wrong guy, and he sues?” he asked. “I contacted a bunch of police departments in 2010, when I was a reporter, because I wanted to see if the algorithm worked. Now I know it works—there’s no question in my mind. In certain places, we can say, ‘These victims have an elevated probability of having a common killer.’ In 2010, though, I had a big media company behind me, with lawyers and media insurance. Now I’m a guy with a nonprofit that has fourteen hundred dollars in the bank and a board of nine directors and no insurance.”
One of map’s most public benefits has been making people aware of how few murders in America are solved. In 1965, a killing led to an arrest more than ninety-two per cent of the time. In 2016, the number was slightly less than sixty per cent, which was the lowest rate since records started being kept. Los Angeles had the best rate of solution, seventy-three per cent, and Detroit the worst, fourteen per cent. As Enzo Yaksic, a map board member and the director of the Northeastern University Atypical Homicide Research Group, told me, the project “demonstrates that there’s a whole population of unapprehended killers that are clearly out there.”
Another of map’s board members, Michael Arntfield, is a professor at the University of Western Ontario, where he runs a cold-case society. It is focussing on the largest finding of the algorithm, a collection of a hundred unsolved murders of women in the Atlanta area over forty years. Most of the victims were African-American, and all were strangled. From the Atlanta police, Arntfield got names for forty-four of the women, and has been learning more about them. (Studying the backgrounds of murder victims in the hope of discovering how they met their killers is a discipline called victimology.) Arntfield and his colleagues separated the victims into two groups: a small group of older women, who were killed in their homes, and a larger group of young women, many of whom may have been prostitutes. From newspaper accounts, Arntfield has found two men who have committed crimes with strikingly similar attributes, both of whom are already in prison. Adam Lee, the head of Atlanta’s Major Crime Section, which includes homicide, told me that the police haven’t yet linked these murders to a particular killer, but he said that he considers map a useful tool and is “very interested in sitting down with Arntfield.”
Hargrove told me he hopes that eventually detectives will begin to use the algorithm to connect cases themselves, and that map will help solve a murder. Meanwhile, he is considering creating a companion site that tracks arson, and has begun compiling data on fires, though he hasn’t had time to post it yet. “There’s a link between serial arson and serial killings,” he said. “A lot of guys start out burning things.”
We were walking in Alexandria by the river, along Hargrove’s usual route, and he said, “Our primary purpose is to gather as many records as possible.” He paused. “It’s seductive how powerful these records are, though. Just through looking, you can spot serial killers. In various places over various years, you can see that something god-awful has happened.” ♦