Spotting soldiers at risk for crime

Analyzing the records of nearly 1 million U.S. soldiers, researchers have developed a computer program that they believe can identify soldiers most likely to commit severe, violent crimes.

The research, led by Harvard University psychologist Anthony Rosellini, drew on 38 databases containing information on 446 variables for each soldier who served between 2004 and 2009.

During that period, a total of 5,771 soldiers committed murder, manslaughter, kidnapping, robbery or aggravated arson or assault, according to the study, which was funded by the Army. Domestic violence and sex crimes were not included in the study, because they are thought to follow risk patterns distinct from other offenses.

Researchers created a risk model based on the soldiers’ demographic characteristics, health histories, career details and other factors predating their crimes.

For men, who accounted for the vast majority of soldiers and offenders, 24 factors were found to be at play. Those most at risk were young, poor, ethnic minorities with low ranks, disciplinary trouble, a suicide attempt and a recent demotion, according to the study published last week in the journal Psychological Medicine.

The highest-risk group — just 5% of the total population of male soldiers — accounted for 36% of the crimes perpetrated by men, the researchers found. Each year on average, 15 of every 1,000 of those men committed a violent offense, more than seven times the overall rate for male soldiers.

Using an altered algorithm, the researchers found the highest-risk female soldiers were responsible for 33% of crimes perpetrated by women, who overall were about half as likely as men to commit violent offenses.

To test their model, the researchers applied it to a sample of 43,248 soldiers who served between 2011 and 2013. They found that the 5% identified as most at-risk were responsible for 51% of the violent crimes committed by the group.

The tool gives the Army the ability “to identify highrisk soldiers without carrying out expensive one-onone clinical assessments,” Rosellini said.

While violence-prevention training is standard for all soldiers, the algorithm could allow the Army to determine who should receive more intensive risk evaluations or interventions, the researchers wrote.

If the new analytical tool can be validated in more tests, the Army will still have to figure out how to use it.

There is still much debate about how to redirect people prone to violence. Even in the highest-risk group, most people do not become offenders.

An intensive violence-prevention program “would make sense only if the interventions are shown to be highly efficient — something that has not yet been demonstrated,” said study coauthor John Monahan, a law professor at the University of Virginia.

The study, part of a massive research effort funded by the Army, builds on a paper published last year that used the same method to create a tool for identifying soldiers at risk of suicide.

Ronald Kessler, a Harvard sociologist and coauthor, said the team is working on other algorithms.