Actuarial vs. Structured Professional Judgment in Violence Risk Assessment
Author: Ukrainian Psychological Hub · Published: October 11, 2026 · Editorial Policy
Actuarial and structured professional judgment (SPJ) are two major evidence-informed approaches to violence risk assessment. Both organize information about factors associated with future violence, but they differ in how information is combined, how the final judgment is reached, and what an assessor is expected to do with the result. Their most important practical distinction concerns decision processes and risk management, rather than a simple contest over which method predicts violence more accurately.
Across major systematic reviews, well-studied actuarial and SPJ instruments often show broadly comparable discrimination of later violent outcomes. Performance varies by population, outcome, setting, follow-up period, instrument, and implementation. Neither method can establish that a particular person will commit violence. Neither can turn a statistical association into a legal finding. Singh, Grann, and Fazel (2011); Fazel et al. (2012).
What is the difference between actuarial and structured professional judgment?
An actuarial violence risk assessment applies a prespecified scoring or statistical rule to defined predictors and produces a score, category, or estimated probability according to the tool’s model. The role of professional expertise is to gather accurate information, apply the model correctly, understand its validation limits, and interpret its output for the referral question. The final score or classification follows the specified algorithm rather than the examiner’s discretionary weighting.
Structured professional judgment uses a defined set of empirically or theoretically informed risk factors, structured evidence gathering, and explicit professional reasoning. The evaluator considers whether each factor is present and relevant in this case, formulates plausible scenarios of future violence, identifies risk-management needs, and reaches an overall judgment under the instrument’s guidance. The evaluator retains discretion, but should explain and document it.
These are families of methods, not uniform technologies. Some actuarial tools include dynamic variables; some SPJ research analyzes summed item scores rather than the final professional judgment. Different versions and implementations of the same instrument can therefore answer different questions. Skeem and Monahan (2011).
The essential comparison
Actuarial assessment emphasizes fixed combination rules, reproducible scoring, and explicit reference to the populations and outcomes on which a model was developed or validated. Its output may be a total score, risk category, or model-derived probability, depending on the instrument. A percentile, score, category, and absolute probability are different quantities and should be identified correctly.
SPJ emphasizes structured case formulation, the relevance of specific risk factors, potential future scenarios, the anticipated circumstances of violence, and interventions or safeguards that could modify those circumstances. The endpoint may include structured summary judgments and a management plan. SPJ does not automatically produce a calibrated individual percentage.
Unstructured professional judgment is a third approach: a clinician or other professional reaches a conclusion without consistently applying a validated structure. Comparisons of actuarial methods with SPJ should not be confused with comparisons of either approach with unstructured judgment. The NICE guideline on violence and aggression explicitly advises structured assessment rather than unaided judgment alone in specified inpatient circumstances.
What exactly does violence risk assessment predict?
Before selecting either method, the assessor must specify the outcome. “Violence” can refer to threats, physical aggression in a hospital, assault in the community, injury-producing violence, violent arrest, violent conviction, or violent reconviction. These outcomes have different detection mechanisms, frequencies, and consequences.
The time horizon also matters. A short-term ward instrument used during a shift is addressing a different question from a tool estimating violence over months or years after discharge. A measure validated for adult forensic psychiatric patients should not be assumed to work for adolescents, people in community supervision, or the general population. Tools constructed for intimate partner violence or sexual recidivism answer other, more specific questions.
An assessment concerns future behavior under particular circumstances. It is distinct from clinical diagnosis, competency to stand trial, criminal responsibility, capacity, a finding that past allegations are true, or an adjudication of guilt. Diagnosis can inform clinical care and sometimes a risk formulation, but diagnosis itself does not establish dangerousness.
A conditional judgment, not a personal destiny
A risk estimate is conditional on the available data, the population on which a tool was evaluated, how future behavior is measured, the follow-up period, and the environments or interventions a person encounters. Detention, treatment, housing stability, supervision, interpersonal conflict, and access to services can change the opportunities or conditions in which violence might occur.
Accordingly, an apparently accurate forecast can become difficult to interpret when interventions occur between assessment and follow-up. Preventing an event may be the desired outcome; treating every non-event as evidence that the earlier judgment was incorrect would oversimplify the evaluation. Prospective research needs to document interventions and the conditions under which outcomes were observed.
How actuarial violence risk assessment works
An actuarial tool begins with specified predictors and scoring instructions. These may be derived from prior research, statistical modeling, or both. Relevant information is collected from records, interviews, and corroborated sources as the instrument requires. The evaluator follows scoring instructions rather than deciding which predictors seem most compelling based on intuition.
The resulting output is compared with validation evidence. For some tools the output is a score associated with a rank ordering of observed outcomes; other tools offer predefined categories or estimated event rates. A score should never be silently converted into a percentage unless the tool provides a validated, applicable mapping from score to probability.
Standard rules can improve consistency across assessors and support audits. They can also make systematic problems repeatable: if the source data are incomplete, the outcome definition is unsuitable, or the original sample differs substantially from the assessed population, faithful scoring will not fix those weaknesses.
Strengths of actuarial methods
A prespecified model makes the combination rule visible and constrains arbitrary adjustments. It can be tested for discrimination, calibration, and error rates in independent populations. It can also be monitored across sites and groups when the necessary outcome data exist.
These are meaningful advantages when institutions need consistent triage or when a referral question requires a validated quantitative estimate. Yet consistency is not the same as accuracy. The usefulness of the score depends on outcome relevance, the original research design, transportability, recency, reliable measurement, and the costs of classification errors.
Limits of actuarial methods
Many actuarial tools rely heavily on historically stable information. Such predictors can help rank risk but may offer limited guidance on what to change through treatment or environmental intervention. They may be sensitive to how historical records were generated, including inequalities in surveillance, arrest, or institutional documentation.
A model built in one setting can lose calibration after migration to another. Local violence rates, population composition, interventions, and definitions of violent outcomes may differ. An old validation result cannot substitute for local performance monitoring.
How structured professional judgment works
SPJ assessment also follows a structured guide. The assessor identifies relevant factors through interviews, documentary evidence, collateral information where appropriate, and contextual observations. The evaluator then considers the significance of those factors in the person’s current and anticipated circumstances.
A full SPJ formulation asks how violence might occur, under what conditions, toward whom, with what warning indicators, and what measures could reduce the relevant opportunities or triggers. The purpose is to organize testable propositions about potential situations and to connect them to proportionate risk management. Those propositions are professional hypotheses, not factual findings about future acts.
SPJ guides are sometimes confused with checklists. Recording the presence of 20 factors is not equivalent to demonstrating why several of them matter to a specific case or how a management plan follows. A professional judgment must be traceable from the evidence to the formulation and recommendations.
Strengths of SPJ
SPJ can distinguish the mere presence of a variable from its case-specific relevance. It can incorporate changes in context, treatment engagement, environmental stressors, protective circumstances, and realistic management resources. It also creates a framework for communicating which interventions have a plausible relationship to the hypothesized risk scenarios.
These advantages matter in settings where assessment and risk reduction are connected, such as forensic mental health treatment, discharge planning, and coordinated community care. Their value depends on trained evaluators, accurate records, disciplined formulation, and the actual implementation of a management plan.
Limits of SPJ
Professional discretion introduces opportunities for confirmation bias, inconsistent weighting, hindsight effects, and differences between evaluators. Using a structured instrument does not make those problems disappear. Some SPJ tools require extensive training and substantial information; a missing collateral record or poorly documented formulation can undermine their usefulness.
Importantly, research showing that an SPJ instrument’s numerical total score predicts outcomes does not necessarily validate the complete SPJ process. A review of 206 HCR-20 papers identified a strong reliance on summed scores and comparatively little direct examination of scenario planning, management, or whether using the guide reduces violence. Review of HCR-20 evidence (2021).
Examples of actuarial and SPJ instruments
The Violence Risk Appraisal Guide (VRAG) is commonly discussed in the actuarial tradition. Its evidentiary relevance depends on version, population, outcome, and follow-up. The HCR-20 family is commonly associated with SPJ and is organized around historical, clinical, and risk-management considerations. SARA is an SPJ approach developed for intimate partner violence risk. SAVRY is a structured professional judgment framework used in youth violence assessments.
These are examples of differing intended uses, not interchangeable instruments. A HCR-20 score derived for research may be analyzed statistically even when the guide’s intended clinical process is structured judgment. An adolescent instrument should not be imported into adult assessments simply because a published study reports strong accuracy. A partner-violence instrument should not become a general homicide prediction tool.
Instrument-specific manuals, qualifications, administration protocols, and validation literature control proper use. Detailed evidence about HCR-20 belongs in its dedicated article; the present comparison concentrates on what distinguishes the two assessment approaches.
Which approach predicts violence more accurately?
The strongest answer is conditional. The systematic review and metaregression by Singh, Grann, and Fazel (2011) included 68 studies and 25,980 participants and reported broadly comparable predictive validity for actuarial and structured clinical judgment approaches. Accuracy differed according to instrument characteristics and study samples. This supports comparison of well-specified instruments in specified settings, not a universal claim that all tools are equally good.
A broader BMJ systematic review by Fazel and colleagues (2012) analyzed 73 samples involving 24,847 people. For tools evaluating violent offending, positive predictive value was often modest, while negative predictive value was substantially higher. The authors cautioned against using instruments as the sole determinant of detention, sentencing, or release.
A more recent systematic review and meta-analysis in The Lancet Psychiatry (2023) studied 50 publications, 36 tools, and 10,460 people discharged from forensic mental health hospitals. Most studies were judged at high risk of bias. Pooled discrimination estimates for certain instruments were moderate; for example, HCR-20 version 2 and VRAG each had pooled AUC estimates around 0.69 in the reported external validations. Critical calibration and classification metrics were often absent.
Together, these reviews establish that structured tools can help rank outcomes in some populations, while demonstrating substantial evidence gaps. They do not show that one assessment family consistently dominates across forensic psychiatry, policing, prisons, community supervision, and courts.
AUC: What a reported 0.70 does and does not mean
Area under the receiver operating characteristic curve (AUC) measures discrimination. An AUC of 0.70 means that, for a randomly selected observed event case and non-event case from the evaluated setting, the model assigns the higher score to the event case approximately 70% of the time, with standard handling of ties.
It does not mean that the evaluator is 70% certain a given person will become violent. It is not the probability of violence for a person scoring “high.” Nor does AUC tell a court how many people will be incorrectly labeled high risk at a particular threshold.
A second-order systematic review found that AUC was frequently reported or interpreted inconsistently and that categorical estimates were much less often evaluated with appropriate classification metrics. Singh et al. (2013).
Calibration, base rates, and positive predictive value
Discrimination asks whether an instrument ranks observed event cases above non-event cases. Calibration asks whether the predicted frequencies match observed frequencies. If a validated model assigns 20% risk to a large group over a defined time period, good calibration would mean that roughly 20% of that group experience the defined outcome in the observed setting and period.
Calibration is essential when a tool communicates absolute probability. A model may retain a reasonable AUC after being transported into a new jurisdiction yet overestimate or underestimate the absolute event rate. A qualitative SPJ label such as “high risk” should not be translated into a precise probability unless an appropriately validated quantitative interpretation exists.
The outcome base rate strongly affects positive predictive value. When the target event is infrequent, even a classification rule with apparently respectable sensitivity and specificity may generate many false positives. Claims about success must identify the event rate and the decision threshold.
Worked example: Why high-risk labels produce false positives
Consider a purely hypothetical cohort of 1,000 people with a 10% event rate over one stated follow-up period. Suppose a classification rule has 70% sensitivity and 80% specificity. Of 100 people who experience the defined outcome, 70 would be classified positive and 30 negative. Of 900 people without the outcome, 180 would be classified positive and 720 negative.
There would be 250 positive classifications, but only 70 would be true positives. Positive predictive value would be 70 divided by 250, or 28%. The other 180 positive results would be false positives. Negative predictive value would be 720 divided by 750, or 96%, while 30 outcomes would still be missed.
This arithmetic is an illustration, not an estimate for VRAG, HCR-20, any institution, or any population. Changing the base rate, threshold, follow-up, or intervention environment changes the numbers. It shows why sensitivity, specificity, positive predictive value, negative predictive value, calibration, and the costs of errors belong beside AUC.
False positives and false negatives have different consequences
A false positive identifies an individual as above a chosen risk threshold when the specified event does not occur during follow-up. Depending on the use, that may contribute to unnecessarily restrictive decisions, stigma, lost opportunities, and unequal treatment. A false negative occurs when an event follows a negative classification, potentially leaving important protection needs unidentified.
Whether an error is acceptable is a practical and ethical question involving the seriousness of the anticipated harm, the consequences of restrictions, the alternatives available, and the governing legal framework. No scientific performance statistic determines the proper legal or moral balance by itself.
A high-risk label should therefore be accompanied by an account of what it means, the available performance evidence, uncertainty, the decision threshold if one was applied, and the less restrictive measures considered. The APA Specialty Guidelines for Forensic Psychology emphasize appropriate methods, impartiality, evidence, and communication of limitations.
Static and dynamic factors: Why the distinction matters
Static factors are historical characteristics that cannot be changed through current intervention, such as some aspects of past conduct. Dynamic factors may change over time, such as current substance use, acute instability, treatment engagement, or accessible support. Their value depends on reliable measurement and evidence relating them to the particular outcome.
SPJ often draws attention to dynamic factors because risk management needs potential points of intervention. Actuarial approaches can also incorporate dynamic information. Therefore, “static equals actuarial” and “dynamic equals SPJ” are misleading shortcuts.
A dynamic score decreasing over time is not automatically evidence that violence will decline. A 2026 systematic review of longitudinal HCR-20 studies identified typically small declines in some scores but little evidence that changes in those scores predicted subsequent violent recidivism. This is a crucial distinction between sensitivity to measured change and evidence of improved real-world outcomes.
The separate English Hub article reserved for static and dynamic forensic risk factors will address measurement and mechanisms in depth. That page is not linked here until it is published.
Does structured professional judgment prevent violence?
Prediction and prevention must be evaluated separately. A tool may distinguish groups with different event frequencies without improving outcomes after its introduction. To establish preventive benefit, studies need to examine what changes in practice occurred, whether suggested interventions were implemented, and whether violence or related harms declined compared with appropriate alternatives.
The 2021 HCR-20 evidence review found very limited direct research on whether deploying the full SPJ process changed violence rates. That finding does not mean risk management has no value. It means a predictive validation study should not be treated as proof of intervention effectiveness.
A small study can also find useful incremental predictive information in professional formulation, but the effect should be evaluated in its own population and design. Douglas and colleagues’ structured judgment study reported added predictive validity in a particular forensic psychiatric sample. Its result contributes to the evidence base without establishing a universal advantage.
Management quality should be assessed through feasible plans, accountable implementation, review intervals, coordination among responsible services, and outcome monitoring. These are operational questions that extend beyond whether a summary score predicts a future record.
Choosing the right approach in different settings
Forensic psychiatric hospitals
Assessment may inform clinical planning, safety measures, leave decisions, and discharge arrangements. A structured approach should specify inpatient versus post-discharge outcomes and the expected horizon. Staff should revisit current circumstances and address treatment, environmental conditions, and patient participation.
For immediate ward aggression, a short-term instrument may be more relevant than a long-term violence prediction instrument. NICE NG10 recommends considering instruments such as the Brøset Violence Checklist or the Dynamic Appraisal of Situational Aggression for specified inpatient monitoring and risk management, rather than relying on unstructured judgment alone. That is a UK clinical guideline for its defined setting, not a worldwide legal mandate.
Courts, sentencing, and release decisions
Courts ask legal questions under particular statutes and rules. Assessment may provide probabilistic or contextual information when relevant and admissible, but the legal decision-maker must apply the governing law. The instrument does not decide guilt, criminal responsibility, statutory dangerousness, or whether deprivation of liberty is justified.
In United States federal proceedings, Federal Rule of Evidence 702 governs relevant aspects of expert evidence and emphasizes sufficient facts, reliable principles and methods, and reliable application. State rules and other countries’ laws may differ. A high AUC from a distant population does not itself establish that the opinion meets the applicable evidentiary standard.
Correctional supervision and community care
The choice should follow the decision: allocation of resources, targeted care, release planning, case management, or a documented comparison of outcomes. Predictive data may support consistent triage; structured formulation may support identification of modifiable conditions and responsibilities. Mixing these purposes without explaining them invites misuse.
Threat assessment and imminent danger
Threat assessment typically addresses a specific concern or emerging situation and requires verification of behavior, context, access, protective circumstances, and possible interventions. It cannot be reduced to the output of a long-term actuarial reoffending tool. Emergency responses follow their own legal and professional procedures and need timely reassessment.
Cultural fairness, sex, age, and transfer across populations
Fairness has multiple measurable dimensions. Groups may differ in base rates, how predictors are recorded, calibration, sensitivity, specificity, or false-positive rates. It is possible for a model to have similar overall AUCs across groups while producing substantively different error burdens or calibration.
Published evidence does not resolve these issues for every instrument and community. The 2011 comparative review identified variation across samples and demographic characteristics; the 2023 Lancet Psychiatry review found a marked underrepresentation of women and widespread high risk of bias in the examined forensic hospital literature. Results from heavily male research samples require careful qualification when applied elsewhere.
Cultural fairness also concerns interpretation. The NICE recommendations call on professionals to avoid negative assumptions based on culture, religion, or ethnicity and to consider whether unfamiliar practices may be misread. Past criminal justice contact can reflect patterns of policing and reporting as well as individual conduct; its inclusion demands transparent justification and careful assessment of local data.
Relevant questions include whether predictive values and calibration have been externally evaluated in the group of interest, whether disability and communication differences affect assessment, and whether the instrument permits valid use for the age and setting. Absence of validation should be stated explicitly, not compensated for by impressionistic confidence.
Data quality, scoring reliability, and implementation
Every method depends on the underlying evidence. Records may contain unresolved allegations, missing periods, biased descriptions, inconsistent offense definitions, and information gathered for purposes unrelated to prediction. A structured format cannot transform unsupported claims into verified historical facts.
Actuarial users must observe item definitions, scoring rules, missing-data procedures, and whether the tool was validated for their population. SPJ users must document both item-level evidence and the reasoning connecting factors to relevant risk scenarios. In either method, limitations should follow the assessment into the report.
It also matters who performs the assessment. Training, supervision, adherence to instrument requirements, independent review, and ongoing performance audits are material safeguards. Interrater consistency of a score or summary judgment should be measured rather than presumed.
A study of violence risk communication in practice found that reports did not consistently or clearly cover the topics expected within an SPJ framework, including risk scenarios and summary judgments. Good guidance must be implemented before its proposed benefits can be evaluated.
How an evidence-based forensic report should communicate results
A clear report states the precise referral question, decision context, outcome, time horizon, relevant jurisdiction, assessment sources, method and version, limitations, and reasons for selecting the approach. It identifies verified facts, disputed information, interview statements, clinical observations, scores, and professional inferences separately.
For an actuarial result, the report explains what the score means, what comparator sample or validation evidence supports its use, whether a threshold or probability is provided, and which performance metrics apply. A “high” classification without a reference population and outcome period is incomplete.
For SPJ, the report identifies relevant factors, why they matter in this case, potential scenarios, foreseeable changes, protective circumstances, and proportionate management recommendations. A judgment should disclose the strength of evidence supporting the proposed scenarios and the limitations of discretionary weighting.
The final synthesis makes clear which statements are research findings, which are clinical or behavioral observations, which are risk opinions, and which questions remain for the court or legally authorized decision-maker. The recommended action is not established solely by a risk category.
Common myths and misconceptions
“Actuarial tools are always more scientific”
Prespecified scoring makes a method reproducible and testable, but scientific adequacy also requires appropriate validation, calibration where probabilities are given, representative data, reliable administration, and a suitable use. An invalid model does not become valid because it uses an equation.
“SPJ is just personal intuition”
Properly conducted SPJ requires structured assessment and explicit reasoning. Its professional-discretion component can be useful for case formulation and management, but it creates reliability and bias questions that must be evaluated empirically.
“A high score means this person will be violent”
A rank, category, or score describes an assessment output within specific evidence and time constraints. It neither proves future behavior nor establishes personal culpability. Base rates and false-positive rates remain crucial.
“AUC tells us the probability of violence”
AUC summarizes rank discrimination across thresholds. Absolute probability needs a separately justified and calibrated interpretation.
“If a tool predicts well, using it prevents violence”
Predictive validity and prevention are different outcomes. A study must evaluate implemented decisions or interventions to demonstrate risk reduction.
“Mental illness makes a person criminal or dangerous”
Clinical diagnosis and criminal offending are distinct constructs. Violence risk must be considered through relevant behavior, circumstances, empirically supported factors, and the assessment question, without equating psychiatric illness with crime.
Frequently asked questions
Is actuarial risk assessment the same as an AI algorithm?
No. Actuarial models may be simple scoring tables or statistical equations and need not use artificial intelligence. Machine-learning systems create further questions about transparency, training data, external validation, calibration, and governance. The method must be evaluated as actually deployed, not by a marketing label.
Is HCR-20 actuarial or structured professional judgment?
The HCR-20 is designed as an SPJ framework. Research sometimes sums its item scores and evaluates those numerical totals, which is analytically different from testing the intended final judgment and management process. A study’s operationalization must be inspected before interpreting it.
Can SPJ produce a numerical risk percentage?
A professional may use external quantitative research as context, but an SPJ label does not automatically translate into a validated individualized percentage. Any numeric estimate needs an explicit evidentiary and methodological basis.
Which approach is best for courts?
There is no universal answer. An appropriate choice depends on the legally relevant question, population, validated outcome, horizon, available evidence, expert competence, and applicable rules. Neither approach replaces a legal finding.
Can these instruments predict homicide by a specific person?
Their performance for broad violent outcomes cannot automatically be extended to rare outcomes such as homicide. Low base rates create particular problems for positive predictive value. Offense-specific prediction requires offense-specific validation, and even then remains probabilistic.
Do risk assessments have to be repeated?
Review intervals should follow the setting, instrument, change in circumstances, and purpose. Some dynamic considerations require timely review; a score from a previous context may become stale. Reassessment schedules supported by one forensic inpatient study are not universal requirements.
Do actuarial and SPJ scores identify causes of violence?
No. A predictor can be associated with an outcome without being a direct cause. Causal claims and intervention selection require additional evidence about mechanisms and the effect of changing relevant conditions.
Can a risk assessment determine criminal responsibility or competency?
No. Competency, criminal responsibility, and risk are separate psycholegal questions. A court applies the relevant jurisdiction’s legal standards using admissible evidence and authorized procedures.
Conclusion
Actuarial and structured professional judgment methods are best understood as distinct structured ways of organizing evidence for future-oriented decisions. Actuarial assessment applies predefined combination rules and may provide comparatively transparent numerical outputs. SPJ structures professional formulation and may help connect risks to scenarios and management planning. Available systematic reviews do not establish a consistent overall winner in predictive discrimination.
The decisive requirements are a well-defined outcome and time horizon, valid evidence for the target population, careful treatment of calibration and error rates, competent administration, transparent reasoning, cultural fairness, and an explicit separation between prediction, intervention, forensic opinion, and legal judgment. Decisions affecting liberty and safety deserve that complete chain of accountability.
For broader context, see Violence Risk Assessment: What It Can Predict and Where It Fails.
Related Articles
References
American Psychological Association. (2013). Specialty guidelines for forensic psychology. American Psychologist. APA official guideline.
Campbell, D., Cohen, M., Macpherson, G., & O’Rourke, S. (2026). The longitudinal properties of the HCR-20: A systematic review. Aggression and Violent Behavior, 86, 102111. https://doi.org/10.1016/j.avb.2025.102111.
Douglas, K. S., et al. (2003). Evaluation of a model of violence risk assessment among forensic psychiatric patients. PubMed record.
Fazel, S., Singh, J. P., Doll, H., & Grann, M. (2012). Use of risk assessment instruments to predict violence and antisocial behaviour in 73 samples involving 24,847 people: Systematic review and meta-analysis. BMJ, 345, e4692. https://doi.org/10.1136/bmj.e4692.
Federal Rules of Evidence. (2023 amendment). Rule 702: Testimony by expert witnesses. Official rule text reproduced by Cornell LII.
National Institute for Health and Care Excellence. (2015). Violence and aggression: Short-term management in mental health, health and community settings (NG10). NICE recommendations.
Review of the empirical evidence base for use of the HCR-20. (2021). PubMed indexed article.
Singh, J. P., et al. (2013). Measurement of predictive validity in violence risk assessment studies: A second-order systematic review. Behavioral Sciences & the Law. https://doi.org/10.1002/bsl.2053.
Singh, J. P., Grann, M., & Fazel, S. (2011). A comparative study of violence risk assessment tools: A systematic review and metaregression analysis of 68 studies involving 25,980 participants. Clinical Psychology Review. https://doi.org/10.1016/j.cpr.2010.11.009.
Skeem, J. L., & Monahan, J. (2011). Current directions in violence risk assessment. Current Directions in Psychological Science, 20(1). https://doi.org/10.1177/0963721410397271.
Violence risk assessment instruments in forensic psychiatric populations: A systematic review and meta-analysis. (2023). The Lancet Psychiatry. PubMed record.