Image credit: Grand Theft Auto V

It’s hard to go a day without hearing about another violent incident: a local robbery, a public brawl, a school shooting. Disturbing events hit media platforms and spread like wildfire. As a result, there has been an incredible pressure on scientists and legal authorities to figure out, “How can we catch the bad guys before they do bad things? I mean, there has to be something we can do, right?”


It’s understandable. We have been raised to believe that most everything can be predicted, from the weather and planetary movements to the stock market and dating compatibility. Statistics (a field of mathematics focused on probabilities) has given us an incredible ability to make decisions based on what is likely to happen. However, violence is tricky. After decades of research, most psychologists have agreed that an acceptably reliable predictor of violence is fundamentally just not possible.

This begs the questions: Why is violence so hard to predict, and what can we do about it?

Problem 1: What Violence Is, and What Violence Isn’t.
If we go by the dictionary definition, violence is “the use of physical force so as to injure, abuse, damage, or destroy.” Violence is a behavior. It’s not a personality trait. It happens on a case-by-case basis. There are many motivations for violent behavior. It can be used to express emotion, to convey an idea, or even to manipulate others. As a result, violence is largely circumstantial. Most violence occurs when specific, personally meaningful factors come together: the exact feelings, situation, and consequences for a given person. Even when personality plays a role in violence, it’s not the only thing involved, and sometimes it’s better explained by something else .

Think back on news reports covering violent crimes. How often were drugs and alcohol involved? How about easy access to a deadly weapon or a significant life event like getting fired or a recent argument with a spouse? If the person didn’t experience or have access to these things, chances are that the person wouldn’t have committed the violent act. Violence itself is a situational occurrence — which is one of the things that makes it so hard to predict.

Problem 2: What Prediction Is, and What Prediction Isn’t.

When we talk about “predicting” behaviors, we’re looking at statistics. In essence, statistics allows us to examine what has happened previously in order to give us a clearer idea of what may happen in the future. But statistics, while driven by reliable mathematics, is NOT a crystal ball. There is always, ALWAYS room for error. Why? Let’s break down a few examples:

Correlation is not Causation. Have you ever heard the expression “correlation does not equal causation?” Basically, it means that just because two events are related (correlation), does not mean that one made the other happen (causation).

Here’s a famous example: in many US cities, there is a positive correlation between ice cream sales and murder rates. Meaning that, when ice cream sales increase, so do homicide rates. (It’s true. Seriously).

Take a moment to think about that: can you explain why that relationship might exist?

I imagine your first thought was, “I don’t think ice cream makes people kill each other or that killing makes people want to eat ice cream.” You’d be right. Those things are definitely not true. But you know what does encourage people to kill each other? Hot weather. And hot weather, incidentally, also makes people want to eat ice cream.

Hot weather contributes to both those things happening, making it look to some people like one caused the other. Without looking at the data, that’s an easy trap to fall into! It’s also important to point out that hot weather may contribute to buying ice cream, but it doesn’t always make people buy ice cream. Maybe someone doesn’t like ice cream. Maybe they can’t afford it. Maybe they’re on a diet and resisting the temptation of it.

When it comes to violence, even our best predictors (like a history of violence or poor impulse control) aren’t especially telling of whether someone will definitely commit violence.

We can’t do “good” science with violence.  Okay, now that we’ve got correlation versus causation down, we want to figure out what causes violence.

Modern science has found a LOT of causal relationships, right? Cigarettes cause lung cancer. Stress causes high blood pressure. Why not apply those same principles to causes of violence?

The only way we can truly determine cause-and-effect relationships is with a controlled experiment. In controlled experiments, we control all the factors and keep all of them the same except for the one we’re measuring. (These types of studies are common when testing new medications.) In the previous example of ice cream and homicide, hot weather is what we call a confounding variable, which is something that happens when we can’t control all the factors.

Confounding variables confuse us about the relationships between things, and they make statistics messy. Worse yet, confounding variables can be just about anything. And the real world — well, it’s full of confounding variables. FULL of them. On top of that, we can’t run controlled experiments directly testing violence for a variety of reasons. When looking for cause and effect with violence, we can’t ethically allow people to just hurt each other for the sake of scientific data. As my boss put it, “We can’t just give a kid a knife and see what they do.” Instead, we can only take measurements of what’s already happened in the real world or create alternate tests which may or may not have real-world validity (e.g. the “hot sauce test”).  This means we generally can’t eliminate confounding variables in violence research because we can’t use our cleanest scientific methods to test our ideas, so we can’t confidently determine causes of violence.

Problem 3: Even if we could predict violence — then what?

A huge amount of research has already been done trying to identify our best predictors of violence. We have found quite a few culprits: a history of violence, having low self-worth, experiencing abuse or neglect in childhood, witnessing violence in childhood, access to lethal weapons, and several others. (Despite common perception, most people with mental illnesses are not violent, and most people who commit violence do not have a diagnosable mental illness.)

So…now what? When we have the information, what can we do with it?

Every year, more than 3.6 million referrals are made to child protection agencies, suggesting abuse or neglect. Forty million adult Americans grew up with domestic violence in their homes, and a huge chunk of us have access to lethal weapons. These people are all categorized as “at-risk” for committing violence. Some of them have, some of them may, but many of them never have and never will. The truth is, if we want to identify people who are “at-risk offenders,” we’re looking at…well, just about everyone.

In short, predicting violence is incredibly difficult. There are a lot of variables that get in the way, there is no single factor that can predict violence, and there is no combination of factors that can tell us for certain who will definitely be violent.

The best thing we can do for violence prevention is the best thing we can do for people: help people get help. As soon as you suspect abuse or endangerment, take steps to keep yourself and others safe. Because it’s dangerous to go alone. If you suspect a violent crime is about to occur or someone is in danger, dial 911 or your local emergency number immediately.


Violence and Abuse Hotlines, if you want to talk to someone who can help you decide what to do:

Chelsea Hughes, PhD is a veteran volunteer and Content Manager of She is a therapist, professor, and researcher with a knack for combining her professional work with her love of video games and geek culture. In her spare time, you can usually find her strumming her guitar, building her latest cosplay, or accidentally stealing kills on League of Legends.

Help us give hope at events around the world. Support Take This on Patreon!