How accurately can people assess their own likelihood of future self-harm or violence?
We are studying the value of psychiatric patients’ self-predictions of violence, compared to predictions made by clinicians or risk assessment instruments.
Statement of the opportunity
Most people with serious mental illness are not violent. When violence does occur, it is often intertwined with factors such as substance use, adverse experiences, and situational triggers. Clinicians play a critical role in assessing and managing violence risk, especially during emergencies.
Validated tools improve risk assessment accuracy, but their complexity limits their use in crisis settings. Clinicians must often rely on intuitive judgment, which can result in overestimating risks and overly restrictive interventions when individuals are most vulnerable. There is an urgent need for practical, efficient approaches that improve accuracy, de-escalate crises, and involve individuals in decision-making.
Testing the promise of self-perceived risk
This research adapts the Conditional Model of Prediction (CMP) to enhance risk assessment by focusing on patients’ self-predictions rather than using clinical evaluations alone. With a lifetime of experience in a wide variety of situations, patients often forecast their own behavior more accurately than external evaluators. Despite potential incentives to deceive, patients often provide honest self-reports.
According to our CMP, individuals have experience-based schemas that specify the kind of violence they might become involved in, given particular conditions (e.g., drinking). Patients’ knowledge of their own “if…then” patterns (Mischel & Shoda, 1995) equips them to assess their risk state effectively.
Study design and objectives
We aimed to:
- Compare the accuracy of patients’ self-perceptions of violence risk with clinical judgment and actuarial tools.
- Explore whether self-prediction accuracy stems from patients’ understanding of their own “if…then” patterns.
- Assess whether patients disclose different levels of self-perceived risk to researchers versus clinicians, exploring generalizability to real-world contexts.
Our study included over 574 psychiatric patients with co-occurring mental health and substance use disorders. Risk assessments were conducted during hospitalization (baseline), followed by interviews with patients and collateral informants at 8 and 20 weeks post-discharge to evaluate self-harm and violence. Hospital and arrest records supplemented these data. Self-predicted risk levels in the research setting were also compared to those elicited by hospital clinicians.
Key findings
- Accuracy of self-perceptions. Patients’ responses to a single question about their self-perceived risk were more accurate than clinicians’ judgments and as accurate as formal risk assessment tools in predicting self-harm and violence within two months post-discharge. This highlights the value of incorporating self-perceptions into risk assessments, which outperform the current standard of practice (clinical judgment) and rival the gold standard (risk assessment).
- Disclosure consistency. Patients disclosed similar levels of risk to researchers (under confidentiality) and clinicians (during discharge planning). This suggests that self-reports remain accurate even when there may be incentives to misrepresent risk.
Next steps & implications
Our findings support the potential of self-perceptions of risk as a practical, patient-centered approach to improve violence prevention and health outcomes. We are developing policy briefs and publications to disseminate these results. We are also collaborating with a psychiatric emergency triage center to implement self-prediction methods and evaluate whether they can reduce restrictive placements without increasing rates of self-harm or violence.
Partners & funding
Jennifer Skeem conceptualized and led this project, with Sarah Manchak overseeing data collection. Co-Investigators included Edward Mulvey, Charles Lidz, and John Monahan. We are grateful for key partnerships with College Hospital, Royale Therapeutic Residential Center, UC Irvine Medical Center, and Western Medical Center. Funding was provided by the National Institute of Mental Health.
