A Harvard Study Says It May Be Possible to Predict Soldiers Likeliest to Commit Violent Crimes

The researchers used a machine learning model to identify the 5 percent of Army soldiers at highest risk for criminal behavior.

Violent crime in the military is a much-studied and discussed issue, linked to everything from PTSD to predilections toward violence prior to service. Now, a new study from Harvard Medical School (HMS) says there may be a way to predict which soldiers are likeliest to commit violent crimes, and thus try to prevent that activity before it happens.

The report was led by HMS healthcare policy professor Ronald Kessler and based on a study funded by the Department of Defense (DoD). The results are based on a massive analysis of data collected from the 975,057 U.S. Army soldiers who were on active duty between 2004 and 2009. Using a machine learning model, the researchers were able to look at a large number of possible predictors of violent crime—everything from health to career information—for each soldier, and narrow the pool down to the 5 percent selected as most likely to commit a transgression in the future.

That 5 percent, the researchers found, was responsible for about 36 percent of major violent crimes committed by men during that time period, and roughly 33 percent by women. After re-doing the task for soldiers on active duty between 2011 and 2013, the likeliest 5 percent was found to be responsible for about half of violent crimes during that span.

In a statement, Kessler said the numbers could lead to a strategy for curtailing violence in the military:

“These numbers are striking. They show us that predictive analytic models can pinpoint the soldiers at highest violence risk for preventive interventions. Targeting such interventions might be the best way to bring down the violent crime rate in the Army.”

Of course, it’s important to note that these numbers do not prove that the soldiers with the highest indication of potential violence will, in fact, commit crimes—or that others outside that group won’t commit crimes. As the researchers say in the statement, further study will be necessary to see just how much of an impact these findings have moving forward.