Technology Could Help DoD Better Address Workplace Violence
By Annette M. Boyle
CAMBRIDGE, MA — As the DoD struggles with implementing a uniform workplace violence program that includes prevention and response protocols, a key tool has been lacking: How to predict who has the highest risk of committing violent acts — and a comprehensive way to intervene to reduce that risk.
Using that data and developing a comprehensive system to intervene could help the services respond to a recent report by the inspector general of the DoD which slammed the department for failing to provide a comprehensive approach to address workplace violence.
“Military personnel, DoD civilian employees and contractors were not equally prepared to prevent and respond to an act of workplace violence, which could jeopardize their safety during a workplace violence threat or incident,” according to the IG.
In the workplace or otherwise, 5,771 soldiers committed murder, manslaughter, kidnapping, robbery or other violent felonies during a five-year period ending in 2009, according to a study published recently in Psychological Medicine. The more than 15,000 cases of domestic violence as well as 718 familial and 6,198 non-familial sex crimes were not included in the study because researchers suggested that those offenders have risk patterns that are distinct from other types of criminals. 1
For years, the military services have relied on psychological interviews and reporting to identify individuals at high risk of violent behavior, but turning toward big data and the DoD’s existing administrative database might be even more reliable, according to the recent report.
Part of the problem arises from the implementation of different sets of recommendations by different organizations. The DoD followed the Secretary of Defense’s August 2010 final direction, written in response to a 2009 attack at Fort Hood, TX, which left 13 dead. The Defense Threat Reduction Agency adopted the Fort Hood review board’s recommendations. Neither of them, according to the IG, ensured the recommendations were properly implemented.
To date, only the Marine Corps has developed and implemented a servicewide violence prevention and response program. The other services have updated or developed guidance that addresses only certain aspects of workplace violence.
The IG report referenced DoD Directive 5205.16, which called for DoD policies to effectively address insider threats, including workplace violence, and to use information derived from previous violent acts to identify, minimize and counter those threats. The IG noted that the “Directive does not specify how this will be accomplished, especially when a comprehensive DoD-wide policy or program on workplace violence does not exist.”
The Psychological Medicine study provides crucial insight into the problem of insider threats. Conducted in collaboration with the Army Study to Assess Risk and Resilience in Servicemembers (Army STARRS), the study found that 5% of the soldiers identified by a model as having the highest predicted risk accounted for 36.2% of all the major physical violent crimes committed by men and 33.1% of those committed by women from 2004 to 2009. When applied to the 2011 to 2013 cohort, the top 5% identified by the as highest risk accounted for more than 50% of all major physical violent crimes. >>Next Page
A handful of factors served as key predictors of future violence, according to the model. Among them were disadvantaged socioeconomic and minority status, early-career stage, prior crime and mental-disorder treatment. The model was developed using a machine learning process that analyzed the administrative data available for all 975,057 U.S. Army soldiers on active duty from 2004 to 2009.
Identifying the soldiers at highest risk for violence allows for targeted intervention strategies. To head off violence, the DoD needs to know “where are these people who are high risk,” said co-author Ronald Kessler, PhD, a professor at Harvard Medical School, Cambridge, MA, and principal investigator of the Behavioral-Based Predictors of Workplace Violence in the Army STARRS research project. With the model, “I can find them. I can show psychiatrists who they are. Now the question is what to do with that knowledge.”
Being able to identify those most at risk for violence raises some challenging issues, Kessler explained, “We don’t want to use the information to punish somebody for something they haven’t done. They may be murderers, but they may be heroes. If you do something for this person, you could keep them from ruining their lives. But you don’t want to stigmatize them.”
The model would allow the Army to sort out individuals and then provide those at highest risk with skills for dealing with impulses, which would make them a stronger, more-effective person, in the military or outside, he noted.
“The fact that the model identifies such a high proportion of violent crimes is especially exciting, because the variables used in the model are routinely collected administrative data the Army can use to identify high-risk soldiers without carrying out expensive one-on-one clinical assessments,” added lead author Anthony Rosellini, PhD, a postdoctoral fellow at Harvard Medical School.
The Army model developed using machine learning had better predictive value than other tools identified in a comprehensive review requiring clinicians to make in-depth assessments of the risk of violence, noted the authors.
That’s a finding that “makes quite a few clinicians feel threatened,” Kessler conceded. “They want to talk to the person, evaluate whether they will commit suicide.”
Getting it right requires a balance, however, he said, noting, “Computers never forget a variable, and they can calculate very well. But they aren’t the only thing. They don’t know that your girlfriend broke up with you. Figuring out how to combine skills will take time. We’re not the enemy.”
The research results also challenged some common assumptions about the risk of violence. “Our finding that never-deployed and previously- deployed soldiers had comparably elevated violent crime risk is striking, given that recent research has suggested that combat exposure leads to increased violence among soldiers returning from deployment,” the authors wrote. They suggested that previous results could perhaps be explained by variables in the current model.
The machine learning process underlying the model enabled the researchers to try thousands of combinations of variables to find the factors that created the most parsimonious, stable model across the entirely of the database and specific subsets, Kessler said.
While the existing database worked for this purpose, it was no easy feat. “It took five people three years going back and forth to get the data,” he emphasized. “As secondary uses become more common, people will become more aware of how they’re entering and maintaining the data.”
- Rosellini AJ, Monahan J, Street AE, Heeringa SG, Hill ED, Petukhova M, Reis BY, Sampson NA, Bliese P, Schoenbaum M, Stein MB, Ursano RJ, Kessler RC. Predicting non-familial major physical violent crime perpetration in the US Army from administrative data. Psychol Med. 2015 Oct 6:1-14. [Epub ahead of print] PubMed PMID: 26436603.