Can we predict domestic homicide? New research suggests we can’t
- Written by: Troy McEwan, Professor of Clinical and Forensic Psychology, Swinburne University of Technology
In 2024, 38 Australian women were murdered by a partner or ex-partner. Thankfully, new data show the number of women killed by intimate partners has reduced to 32 over the most recent reporting period. The annual rate to June 2025 was among the lowest on record.
Nonetheless, more needs to be done to get the number of intimate partner homicides to zero. Thanks to tireless advocacy by many, the federal government has a target to reduce female victims of homicide by 25% per year. It’s part of the National Plan to End Violence against Women and Children.
One common way of trying to prevent intimate partner and family homicide is through risk assessment and management. Risk assessments are used by police, family violence sector agencies and others to prioritise cases where harm seems most likely.
Sadly, our new research, published in the Journal of Family Violence, suggests it’s almost impossible to use these risk assessments to accurately predict who will attempt to kill their partner.
How do we assess risk?
Completing a risk assessment involves identifying the presence of “risk factors” in a potential perpetrator: characteristics that are thought to be associated with increased risk of homicide.
Past research has found women who are killed by a partner are more likely than abused women generally to have experienced things like strangulation, stalking, controlling behaviour and threats to kill.
Because they are more common in homicide cases, many people believe these kinds of characteristics are risk factors that can help them predict future homicide. Unfortunately, that’s not true.
Homicide is (thankfully) extremely rare. In Victoria, there were about 18 family violence-related homicides for every 100,000 family violence reports made to police in 2024–25.
Because it’s so rare, it’s virtually impossible to predict who might be a victim of homicide, or who will commit a homicide, even when lots of risk factors are present.
Our research
Our recent research showed this in a population of nearly 40,000 family violence reports recorded by Victoria Police.
We followed every person for 12 months in police databases to see who was involved in a subsequent family violence homicide or an incident that could have resulted in death (such as a very serious assault).
We tested the most commonly identified risk factors for intimate partner homicide to see if any of them alone, or combined, could predict a fatal or near fatal outcome.
We found none of them could. More than 99% of people with these risk factors were not involved in a fatal or near fatal attack in the 12 months we followed them.
For example, police recorded that 7,337 people had, in the past, threatened their partner or family member with serious harm or death. Among those who did go on to very seriously harm or kill, about 22% had previously made such threats.
However, 99.84% of those who had made a threat did not kill or attempt to kill a partner or family member in the 12 months after the police risk assessment.
The same is true of those who used jealous and controlling behaviour towards a partner or family member. Police recorded 12,123 people as having done this, and of those who very seriously harmed or killed, 29% had a history of jealous and controlling behaviour.
But again, 99.87% of people with jealous and controlling behaviour at the time of the original police report did not go on to kill or seriously harm.
The same was true of strangulation, stalking, and other risk factors for homicide.
We then tested whether combinations of these risk factors could predict homicide, with similar outcomes. Whichever way we looked at the data, the result was the same – every previously assumed risk factor or combination of risk factors got it wrong more than 99% of the time.
It’s possible we could have made more accurate predictions if we followed people for a longer period, say ten years.
But, while more homicides would have occurred, the overall rate of lethal and near lethal violence would have still been extremely low. Even more importantly, the results would not mean much in practice because risk assessments tend to guide responses in the short to medium term, not over many years.
What’s behind this?
We got these results because fatal or near fatal violence is very rare. It was only present in 55 cases in our sample of nearly 40,000 police reports.
While 55 deaths are of course 55 too many, all the risk factors we examined were unfortunately much more common.
Access to firearms was the least common factor, recorded in about 1,300 family violence cases. The most common factor, the perpetrator having identified mental health problems, was present in around 13,500 family violence cases.
Together, this means that even when a risk factor is present, most of those with it do not go on to use fatal or near-fatal violence.
These results don’t mean we should stop paying attention to strangulation, stalking or threats to kill. They do mean we should stop thinking these behaviours can help us predict homicide.
Of course, we must respond when these awful behaviours are identified and prevent people who’ve done such things from causing further harm.
But risk management should not be based on the idea that all people who act in this way are likely to kill, when the reality is that the overwhelming majority will not.
In some circumstances, very intrusive risk management is warranted to ensure immediate safety (such as remanding the person who has been violent). But in most cases, the presence of these risk factors doesn’t indicate that a homicide is imminent.
Rather, they indicate the need for interventions that can reduce family violence and its harms. Examples include psychological treatments that can change emotions and thoughts linked to violence, mental health and substance use treatment where it is required, effective multi-service collaborations, and wrap-around victim support services.
The evidence suggests while prediction is not possible, prevention is. The best way forward is not to create a false expectation that we can ever know who will kill.
Instead, we must adopt evidence-based preventative strategies and fund them fully, so they are available to everyone who needs them. Perhaps then the goal of zero intimate partner or family homicides will be closer to a reality.
The National Sexual Assault, Family and Domestic Violence Counselling Line – 1800 RESPECT (1800 737 732) – is available 24 hours a day, seven days a week for any Australian who has experienced, or is at risk of, family and domestic violence and/or sexual assault.
Authors: Troy McEwan, Professor of Clinical and Forensic Psychology, Swinburne University of Technology
Read more https://theconversation.com/can-we-predict-domestic-homicide-new-research-suggests-we-cant-268290
Daily Bulletin
