Psychology by BogdanovaThe Encyclopedia of Psychology for the Artificial Era

RESEARCH DOSSIER

Targeted Violence and Threat Assessment

Abstract. Targeted violence and threat assessment investigate how recognizable conflicts, behavior, communications, institutional interactions, and protective opportunities may precede harm directed toward identifiable people or places. This fundamental research monograph reconstructs the protective-intelligence tradition and integrates contemporary studies of school, university, community, workplace, stalking, lone-actor, and public mass-violence contexts. Its method is a critical narrative synthesis of official retrospective case research, randomized school intervention evidence, prospective and observational case records, structured professional judgment studies, and methodological research on predictive validity. Particular attention is given to selection bias, the base-rate problem, assessment reliability, disability equity, reporting, victim safety, and the distinction between program effectiveness and favorable case outcomes. Research through 2026 supports structured contextual inquiry, multidisciplinary coordination, and measured protective intervention while exposing limits of claims drawn from perpetrator-only data. The synthesis establishes that the scientific future of threat assessment depends on prospective denominators, transparent decisions, independent field evaluation, and defensible human–artificial evidence handling.

Keywords: targeted violence; behavioral threat assessment; threat management; protective intelligence; school threat assessment; warning behaviors; stalking; lone-actor violence; structured professional judgment; violence prevention; risk calibration; referral bias; public health prevention; victim safety; forensic psychology; Artificial Era

The Research Object and the Prevention Question

Targeted violence refers to intentional harm directed toward an identified person, group, institution, or place through a process in which grievance, decisions, opportunities, and behavioral preparation can become linked. The category includes selected forms of public-figure attacks, school attacks, workplace attacks, stalking-related harm, and ideologically motivated violence, although those events differ in prevalence, context, and legal definition. Threat assessment is the disciplined investigation of a concerning situation and the formulation of proportionate protective action. A threat management program must answer what has happened, what is credible, what information remains missing, which people require protection, and what changes in the environment might interrupt an escalation. This object concerns specific situations and decisions, not the assignment of permanent dangerousness to a demographic or diagnostic group. The protective-intelligence research of [Fein and Vossekuil (1998)] Fein & Vossekuil, 1998 and the contemporary National Threat Assessment Center guidance National Threat Assessment Center, 2024 establish its foundational behavioral orientation.

The principal scientific questions are whether observable conduct and context support an interpretable model of emerging intent, how assessors should integrate uncertain and sometimes contradictory information, what kinds of intervention can reduce harm, and how prevention programs can be evaluated without relying solely on attacks that happened. The present work uses a critical narrative synthesis of original case-series research, field evaluations, experimental school research, instrument validation, official investigations, and methodological scholarship. It distinguishes the empirical observation that a behavior appeared in some completed attacks from evidence that the same behavior predicts an attack among all persons displaying it. This distinction governs every subsequent chapter. In a low-base-rate outcome, a seemingly informative indicator can produce many erroneous accusations when applied without context, comparison cases, and review Fazel et al., 2012.

Conceptual Boundaries: Threat, Risk, Intent, Capability, and Harm

A communicated threat is an expression that may convey a wish, warning, demand, fantasy, intimidation, or intent to harm; its wording alone cannot determine whether violence will occur. Intent refers to an individual's current orientation toward a contemplated action, while capability concerns practical ability within an environment. Target selection is a relational process involving perceived grievance and a potential victim or institution. Risk is the conditional possibility of harm over a specified period and under specified circumstances. Protective urgency concerns the speed with which proportionate safeguarding decisions must be made, even when a probability cannot be estimated. These concepts overlap in real cases, but each is measured through different evidence. Researchers and practitioners have repeatedly emphasized the limitations of equating a direct verbal threat with a completed violent outcome Fein & Vossekuil, 1998 Meloy et al., 2012.

The dossier's unit of analysis is the evolving person–situation–environment configuration. A single incident may involve several people, several reporting channels, and rapidly changing access to help or resources. Risk estimates drawn from correctional follow-up studies address different populations and time horizons than threat management in an active school, workplace, or community case. Thus the [Fazel et al. (2012)] Fazel et al., 2012 meta-analysis is used here to clarify statistical prediction limits rather than to replace contextual assessment with a criminal-recidivism score. Similarly, stalking-specific evidence describes repeated pursuit and victim safety, while lone-actor terrorist assessment studies a particular violent-outcome domain. Disciplined assessment specifies the outcome and intervention decision before selecting observations or tools.

Historical Foundations of Protective Intelligence

The modern United States research tradition grew from protective investigations of attacks and near attacks against public figures. The Secret Service's Exceptional Case Study Project was analyzed in the National Institute of Justice guide by [Fein and Vossekuil (1998)] Fein & Vossekuil, 1998. Its historical contribution was to replace reliance on offender profiles with examination of behavior, situational development, motivation, and interactions with institutions. The study reviewed 83 individuals who attacked or approached an attack on public officials or figures, a highly selected population. It could reveal features of these known cases, while offering limited evidence about how frequently similar features occur among nonattackers. The original guide explicitly approached its findings as principles for investigations rather than a formula for predicting an individual's future behavior.

Subsequent research extended the behavioral perspective to schools after the late-1990s school-violence crisis, to public mass attacks, workplace conflicts, stalking, and ideologically driven violence. The National Threat Assessment Center's [2018 school guide] National Threat Assessment Center, 2018, the FBI's [2018 study] Federal Bureau of Investigation, 2018, and later NTAC community publications National Threat Assessment Center, 2023 National Threat Assessment Center, 2024 formalized the relationship between reporting, multidisciplinary assessment, intervention, and follow-up. This history also shows the evolution of the scientific question. Earlier retrospective case reconstruction chiefly asked how an attack developed; contemporary program evaluation asks whether institutions detect actionable concerns, distinguish urgent cases from benign ones, and implement beneficial responses without producing avoidable harms. These are distinct evidentiary problems requiring distinct designs.

A Process Model of Escalation and Interruption

A process model interprets targeted violence as a contingent sequence of changes in motivation, perceived options, behavior, and opportunity. Grievances may become salient; social isolation or perceived humiliation may narrow deliberation; communication with others may reveal conflicts; and institutions may or may not recognize a need for support or protection. The sequence does not occur identically in every case, and no single stage is necessary or sufficient for serious harm. A process model becomes scientifically useful when its variables can be defined and its alternative trajectories observed. The Secret Service and FBI retrospective studies offer evidence about selected completed attacks, including changes in observable behavior Federal Bureau of Investigation, 2018 National Threat Assessment Center, 2023. Their case selection means they cannot establish universal progression rates.

Interruption is the central outcome of a prevention-oriented model. Protective relationships, timely reporting, access to assistance, conflict mediation, coordinated care, and lawful environmental safeguards can alter the course of a concerning situation. These pathways should be treated as hypotheses to evaluate with time-stamped case records and outcome information. A favorable case closure is informative only if it is assessed against comparable cases and a clear definition of success. A person's eventual nonviolence after an intervention is consistent with prevention but does not, on its own, prove that violence would otherwise have occurred. Strong evaluation documents intermediate changes in threat communication, welfare, access, participation, and protective conditions National Threat Assessment Center, 2024.

Grievance, Identity, and Social Context

Perceived injustice and grievance are recurring themes in retrospective research on targeted attacks. A grievance may concern a relationship, workplace, perceived exclusion, ideological cause, or conflict with an institution. The 2023 NTAC review of mass attacks in public spaces describes grievances among the motivations observed in its selected attack cases National Threat Assessment Center, 2023. This evidence establishes the importance of understanding the actor's interpretation of events, but grievance itself is common in everyday life and has limited diagnostic specificity. Assessment therefore considers the form, persistence, context, behavioral consequences, and changing function of an expressed grievance rather than treating anger or complaint as an independent sign of violent intent.

Identity-related processes can intensify the personal meaning of a conflict. A person may link a perceived injury to social status, belonging, or a larger narrative of humiliation and restoration. Research on lone-actor terrorism explores identity, fixation, and other variables through structured judgment instruments Meloy & Gill, 2016 Guldimann & Meloy, 2020. Most ideological expression and most identity-based grievances are peaceful. The scientific task is to examine behavior and circumstances, not to infer violence from political, religious, or cultural membership. Appropriate assessment also asks whether institutional conduct, prejudice, discrimination, or escalating interpersonal responses contribute to the problem. An institution that treats an allegation of grievance as proof of danger may magnify the very conflict it intends to resolve.

Communication, Leakage, and Reporting

Leakage describes communication to a third party of a person's interest in or intent toward serious harm, whereas direct threats are addressed to a potential target. Meloy and colleagues included both among a broader proposed warning-behavior typology Meloy et al., 2012. Their conceptual importance lies in showing that people may disclose meaningful information through different channels before an act. Neither the mere existence of provocative expression nor the absence of direct threats establishes a reliable future-outcome judgment. Communications must be assessed for context, changes in meaning, corroboration, access to support, and the receiver's capacity to report safely.

A 2026 study by [Silva and Lankford] Silva & Lankford, 2026 examined 483 U.S. mass shootings from 2000 through 2023, differentiating public, family, and felony-related events. In these known completed cases, leakage and threats varied across event types; the study examined public reporting and institutional response as separate stages. Because its sample consists of completed mass shootings, the findings support analysis of missed information and reporting channels, but do not supply positive predictive values for everyone making concerning statements. The study also underscores the difference between an observed warning and an actionable report. Programs must make reporting accessible, document how reports are assessed, and protect reporters and subjects against retaliation or unsupported escalation.

The Epistemology of Warning Behaviors

Warning behaviors are short-term changes in conduct that may indicate increasing concern within a particular case. The typology proposed by [Meloy, Hoffmann, Guldimann, and James (2012)] Meloy et al., 2012 groups warning patterns into eight conceptual families. These categories organize observations for professional discussion. They are not a validated checklist that can be converted directly into a universal percentage of attack probability. A pattern reported in a completed attack remains a descriptive association until its frequency in comparison cases, interrater reliability, temporal relation to outcomes, and effect on decisions have been evaluated. The analytic value of a framework depends on transparent operational definitions and evidence that raters can apply them consistently.

Behavioral observations must be interpreted within a person's baseline, developmental characteristics, social relationships, and immediate context. A sudden change from prior functioning may demand supportive outreach in one case and an urgent protective response in another. Reports may be incomplete, exaggerated, misunderstood, or shaped by institutional incentives. Independent corroboration, explicit recording of uncertainty, and deliberate search for disconfirming evidence reduce the danger of converting ambiguous signs into fixed labels. Assessors should also document how an interpretation would change if new evidence emerged. This approach treats warning behaviors as questions guiding inquiry rather than conclusions that end it American Psychological Association, 2013.

Schools as a Distinct Evidence Environment

School threat assessment must account for developmental variation, common impulsive expression, peer conflict, social media communication, disability accommodation, and the obligation to preserve educational opportunity. The comprehensive school threat assessment tradition developed at the University of Virginia offers a procedure for distinguishing relatively transient conflicts from situations requiring further protection and supportive intervention Cornell et al., 2012. The aim is to understand the context of a threat and intervene proportionately. Students should be evaluated in the setting of their developmental and linguistic abilities rather than through adult-centered assumptions about expression, humor, or emotional regulation.

The best-known experimental evidence concerns school discipline and support rather than the prevention of rare shootings. [Cornell, Allen, and Fan (2012)] Cornell et al., 2012 randomized 40 schools and studied outcomes for 201 students making threats. The intervention schools' students were more likely to receive counseling and parent conferences and less likely to receive severe exclusionary discipline. Those outcomes are directly informative for institutional practice; the sample was not sized to establish effects on school homicide. The scientific implication is to measure tangible administrative and student outcomes while acknowledging that rare catastrophic events require other designs and longer surveillance to study reliably.

Community-Based Programs and Institutional Networks

Targeted-violence prevention extends beyond a single office because important information is distributed across families, education, workplaces, community services, health care, and public-safety organizations. The Secret Service's [2024 guide to behavioral threat assessment units] National Threat Assessment Center, 2024 describes structured intake, assessment, management, and coordination for local law enforcement agencies. The guide is an official practice framework, not a randomized demonstration that every proposed component independently reduces violence. A credible program defines decision authority, referral pathways, thresholds for emergency assistance, privacy responsibilities, and procedures for periodic reevaluation.

Information-sharing mechanisms must distinguish relevant, corroborated observations from rumor and protected personal attributes. The network benefits from a designated coordinating function and a record of who acts on each recommendation. It also requires boundaries: educational personnel should not be used as covert intelligence collectors, clinical records should not circulate without lawful grounds, and referrals should not automatically become punitive events. The performance of a network should be evaluated through time to triage, completeness of follow-up, resolution of immediate safety needs, disproportionate referrals, and stakeholder trust. This turns a general appeal to collaboration into a testable institutional intervention.

Workplace, Public-Figure, and Community Settings

Workplace or public-figure threats can involve repeated grievances, disputes with institutions, interpersonal boundaries, stalking, and communications distributed across several agencies. The original protective-intelligence research examined threats toward notable public figures, but its process-centered lessons can inform other settings when ecological differences are made explicit Fein & Vossekuil, 1998. Workplaces have distinctive duties of care, employment law, confidential human-resource records, and incentives to interpret conflict as misconduct. Public institutions must also distinguish unpleasant criticism from threats of harm. A risk process that ignores these structural contexts can misread organizational conflict as an individual pathology.

Case management is most defensible when protective measures are directly tied to observed conditions and are adjusted as those conditions change. The operational emphasis is safe access to reporting, preservation of relevant evidence, designated support for affected people, and coordination with qualified personnel where imminent danger is suspected. A workplace program cannot be validated simply by counting the number of threats referred; increased referrals may reflect greater trust and earlier detection. Outcome monitoring requires independent definitions of harm, near misses, disruptions, inappropriate sanctions, employee well-being, and lawful privacy protections. Such measures permit comparisons among programs without implying that threat assessment confers omniscience.

Stalking and Repeated Pursuit as a Distinct Threat Process

Stalking comprises repeated unwanted pursuit, contact, monitoring, or intrusion that can threaten a person's safety and autonomy. Its course differs from that of a single public attack because victim exposure, repeated interactions, relationship histories, and episodic changes in conduct are central. Structured professional judgment approaches such as the Guidelines for Stalking Assessment and Management aim to organize information about the individual, the behavior, the victim, and potential interventions. These tools require evidence about their reliability and applicability rather than reliance on a title or formal training credential.

In a retrospective study of 146 referred adult stalkers, [Shea, McEwan, Strand, and Ogloff (2018)] Shea et al., 2018 reported initially poor interrater reliability that improved with clearer definitions and rescoring; case prioritization and continued-stalking judgments discriminated outcomes with reported AUC values of .69 and .76. These are sample-specific discrimination statistics, not individual probabilities. [Penney, Ulrich, and Maheandiran (2023)] Penney et al., 2023 subsequently studied changing risk factors across repeated assessments and found temporal information relevant to stalking outcomes. The evidence favors reassessment and victim-centered management, while selected forensic samples and heterogeneous outcomes limit generalization to all community complaints.

Lone-Actor Violence and Instrument Validation

Studies of lone-actor terrorism have attempted to formalize risk formulation through the Terrorist Radicalization Assessment Protocol, commonly called TRAP-18. The protocol distinguishes proximal behavioral concerns from more distal contextual characteristics. Its development offers an example of structured professional judgment that combines theoretical organization with attempts at empirical validation Meloy & Gill, 2016 Guldimann & Meloy, 2020. The scientific issue is whether the instrument supplies reproducible information that improves real decisions under conditions in which many individuals express alarming ideas and very few commit attacks. Research based only on perpetrators cannot determine the rate of false positives in the wider referral population.

[Meloy and Gill (2016)] Meloy & Gill, 2016 examined 111 known lone-actor terrorists and reported associations between several protocol variables and subgroup outcomes. [Goodwill and Meloy (2019)] Goodwill & Meloy, 2019 compared 33 attackers with 23 nonattackers using multidimensional scaling. The case-control structure moves beyond perpetrator-only description, but sample selection and small sizes constrain calibration for general screening. [Guldimann and Meloy (2020)] Guldimann & Meloy, 2020 explicitly recognized the need for more independent validation and disclosed the developers' intellectual and financial interests. Responsible synthesis treats encouraging discriminative results as provisional and calls for independent, prospectively followed cohorts with well-defined interventions, selection processes, and adverse consequences.

Diagnosis, Distress, and the Problem of Stigma

Assessment must separate symptoms, diagnoses, functional difficulties, and specific risk-relevant conduct. Mental health problems may require clinical attention and can complicate a case through distress, fear, intoxication, or impaired judgment, but diagnostic labels have limited specificity for rare targeted attacks. The Secret Service's retrospective mass-attack research describes mental health symptoms in a proportion of known perpetrators National Threat Assessment Center, 2023. Because such research begins with attackers, it cannot estimate the likelihood that a person with a particular diagnosis will commit violence. Clinical concern and public safety concern sometimes overlap, but their thresholds, legal authorities, and appropriate interventions differ.

A Columbia database study by [Brucato and colleagues (2022)] Brucato et al., 2022 analyzed mass murders across a long historical period and reported that psychotic symptoms were documented in only a minority of perpetrators, with differing rates according to the type of incident. The dataset depended on available historical reports and selected event definitions, so its estimates are not diagnostic prevalence rates for all people or all violence. Nevertheless, it demonstrates the scientific weakness of treating psychosis as a general explanation for mass violence. A humane threat assessment protocol uses trained clinical expertise when relevant, prioritizes support and welfare, and avoids making treatment a substitute for other protective measures.

Risk Prediction, Base Rates, and Decision Thresholds

Threat assessment decisions occur under uncertainty, often with very low base rates for the most serious outcomes. Suppose an indicator were applied indiscriminately to a population where the event is exceptionally rare: even a high sensitivity and specificity could produce many false positive classifications, because the population of noncases is vastly larger. Positive predictive value depends on prevalence as well as test properties; this is a mathematical constraint, not an ethical preference. AUC measures rank discrimination, whereas calibration asks whether estimated probabilities match observed frequencies. Neither answers whether a particular intervention improves safety. Programs should publish the population to which a model applies, the observation period, the outcome definition, and the costs of errors.

The [Fazel et al. (2012)] Fazel et al., 2012 systematic review evaluated violence and offending risk tools across 73 samples and 24,847 individuals. Its violence instruments had a median positive predictive value of approximately 41 percent in the studied populations, which were not representative of general community targeted-violence referrals. The result cannot be transferred as a predictive rate for school or workplace violence. Its methodological lesson is stronger: even established instruments have limits as sole determinants of coercive decisions. Case-specific threat assessment therefore uses evidence to guide proportionate safeguards, reopens conclusions when circumstances change, and refuses spurious precision in individual attack forecasts.

The Public Health Model of Targeted-Violence Prevention

DeGue, Liu, Frederique, Le, and Simon (2026) DeGue et al., 2026 articulate a public health research approach in which targeted violence is defined and monitored, risk and protective factors are investigated, prevention strategies are developed and evaluated, and beneficial practices are implemented and studied in real settings. This framework widens the unit of analysis from the potential attacker to the social and institutional conditions of prevention. It asks whether communities can strengthen supportive relationships and access to assistance while improving the response to credible threats. The article is a narrative methodological contribution; its proposed framework should not be counted as experimental proof that any particular national program has prevented mass attacks.

The conceptual reach of the public health approach matters because safety outcomes are generated by an ecological system. Exposure to interpersonal conflict, access to assistance, organizational trust, institutional reporting practices, and the effects of exclusionary responses may all influence the evolving case. By studying protective conditions as well as potential warning behaviors, investigators can formulate prevention hypotheses that do not require labeling large groups as dangerous. The four-stage public health cycle also exposes two often neglected steps: a prevention practice needs causal testing and subsequent implementation evaluation. A compelling mechanism or favorable expert consensus cannot replace measured outcomes DeGue et al., 2026.

Multistate School Threat Assessment Outcomes in 2026

Kerere, Cornell, Konold, and Maeng (2026) Kerere et al., 2026 studied 2,349 threat assessment cases from 166 schools across five U.S. states, analyzing whether the classification assigned to a threat and the presence of common school safety practices were associated with a subsequent student attack. Their paper reported 234 cases (10 percent) involving an attack after assessment and seven cases (0.3 percent) involving serious injury. Serious threat classifications were associated with substantially greater odds of a subsequent attack than nonserious classifications. In their observational analysis, the particular school safety measures studied were not significantly associated with attacks or injury. The study's outcome is a subsequent student attack, a much broader class than the exceptionally rare mass-casualty attack.

This study substantially improves the available empirical denominator because it follows assessed cases, rather than beginning solely with offenders who completed major attacks. Nevertheless, associations between classifications and events do not demonstrate that the classification itself causes violence or that a given action prevented an outcome. School circumstances, program fidelity, reporting differences, and case-mix may shape observed rates. A predictive association must also be separated from discriminative accuracy, calibration, and decision utility. The null associations for selected school safety resources are findings about this sample and design rather than proof that every such resource is useless in all contexts Kerere et al., 2026.

University Case Series and the Meaning of Persistent Contact

Cotton, Grydehøj, Ramzah, and Scalora (2026) Cotton et al., 2026 coded 494 university threat assessment cases to study problematic approach behaviors and patterns of communication. Their article examined intensity of effort, the pattern through which a person makes repeated and varied communications about a conflict, concern, or target. The abstract reports that intensity-of-effort patterns were associated with problematic approach toward persons with whom the individual had prior engagement, and that digital modalities should be assessed in relational context. The empirical outcome was problematic approach behavior; it is not equivalent to attempted homicide, and the analysis cannot be used as a direct estimate of lethal violence.

The result illustrates the importance of aligning a construct with its outcome. Frequency of communication may be salient in stalking or harassment investigation while having poor specificity for severe physical violence. College referral samples also differ in age, social relationships, institutional authority, and service availability from school or general community populations. Replication should use independent assessors, distinguish baseline contact from escalating behavior, and report whether the assessments improved victim safety or prevented harmful events. The evidence is best used to refine contextual interviewing and protect affected persons, rather than as a mechanical rule assigning dangerousness to someone who communicates persistently Cotton et al., 2026.

Contemporary Pathway Research and the Limits of Retrospective Selection

Schildkraut (2026) Schildkraut, 2026 analyzed documented pathway-related behaviors in 171 U.S. mass public shooting incidents involving 175 perpetrators between 1999 and 2024. The study restricted its analytical sample to attacks for which preceding concerning behaviors or communications could be confirmed, then coded several observable categories and their combinations. It found substantial diversity in how these categories accumulated and co-occurred. The study is a meaningful improvement over purely conceptual accounts of an invariant progression toward violence, because it operationalizes multiple patterns for empirical comparison.

The restriction to attacks with documented warning information is a decisive limitation for interpretation. Neither the observed frequencies of pathway categories nor their co-occurrence establish how often similar conduct appears among nonattackers, and sample ascertainment may be influenced by the availability of records. This research can test descriptive theories about already-completed violence, yet prospective assessments require different cohorts and independent false-alarm measurement. Read alongside the 2026 leakage comparison by Silva and Lankford Silva & Lankford, 2026, the main lesson is methodological: identify the sampled event type, the information available before an incident, and the selection rules before making an inference about prevention Schildkraut, 2026.

Case Ascertainment, Referral, and Selection Bias

A threat assessment system begins with referrals, but referrals are products of human attention, institutional rules, and reporting opportunities. Two institutions can have the same underlying level of danger and very different numbers of cases because their students, employees, or community members differ in willingness to report. Formal intake records should distinguish source, allegation, observed conduct, dates, corroboration status, and reason for referral. These records create the denominator needed to evaluate whether a program is selecting meaningful cases or reproducing institutional stereotypes. They also allow auditors to measure whether cases close without follow-up and whether people receive appropriate services. Counts of concerning statements alone provide a limited picture of safety.

The empirical relevance of selection becomes visible in [Cornell and colleagues' (2025)] Cornell et al., 2025 statewide school study. It reported 15,301 assessments across 1,221 schools and found that students with disabilities accounted for 41 percent of referred cases. This proportion does not establish that disability causes dangerousness; it identifies a major assessment and equity question about who enters the process and why. Any evaluation should compare referral rates with the relevant student population, referral triggers, communication supports, disciplinary practices, and eventual outcomes. Referral is a decision made by an institution, not a biological property of a student.

Evidence Collection and Corroboration

Threat assessment requires an evidence architecture designed to handle inconsistent and sometimes emotionally charged accounts. Records may include witness statements, direct communications, documented events, prior referrals, contextual changes, and information from lawful institutional sources. Evidence quality depends on provenance, timing, completeness, independence, and relevance to the particular concern. An assessor should distinguish firsthand observation from hearsay, a verbatim communication from paraphrase, and a repeated copied claim from several independent observations. Establishing a chain of provenance reduces the risk that one rumor becomes multiple apparently corroborating facts. The same standards apply to digital submissions, which may be cropped, reposted, impersonated, or separated from their original context.

Evidence integration is a scientific process of updating competing explanations. An apparently alarming communication may reflect intimidation, distressed expression, artistic performance, interpersonal joking, or genuine mobilization toward harm. Further inquiry should identify what each explanation predicts and what observations would count against it. Missing information must be recorded as missing, rather than coded automatically as reassuring or incriminating. The [APA Specialty Guidelines for Forensic Psychology (2013)] American Psychological Association, 2013 provide a professional basis for documenting relevant methods, limitations, and sources when psychological judgments are used in legal contexts. A case record is strongest when another qualified evaluator can reconstruct the evidential route from observation to recommendation.

Assessment Interviews and Relational Context

Interviews may clarify the meaning of concerning behavior, the person's current stressors, access to supportive relationships, views of a conflict, and willingness to participate in a safe resolution. Their validity depends on rapport, clarity, voluntary cooperation where applicable, and sensitivity to age, culture, language, and disability. A confrontational interview can change subsequent behavior and damage the relationships on which future reporting depends. Interviewers should accurately explain their role and confidentiality limitations and separate supportive conversation from coercive investigative functions. The purpose is to understand a situation and identify protective opportunities rather than extract a confession or validate a preselected theory.

Multisource information is particularly important when an interviewee minimizes a concern, describes it differently, or has difficulty expressing themselves. Inconsistency alone has multiple interpretations; deception cannot be inferred from body language or discomfort. A full assessment documents the person's account alongside independent observations and explicitly records unresolved contradictions. For young people or persons requiring communication accommodations, adapted interviewing safeguards are part of scientific measurement quality. The Secret Service school guide National Threat Assessment Center, 2018 and behavioral threat assessment unit guide National Threat Assessment Center, 2024 emphasize information gathering across relevant systems and calibrated intervention. Whether each component improves accuracy remains a subject for empirical field testing.

Dynamic Formulation and Structured Professional Judgment

Structured professional judgment uses an explicit set of domains to organize a case formulation, while retaining a contextual explanation of how an outcome could become more or less likely. The formulation connects observed behavior with environmental pressures, protective conditions, uncertainty, and a planned response. It should state the time horizon and reasons for review rather than assign an immutable trait label. Effective structure makes assumptions visible, permits audit, and clarifies when different assessors disagree. The scientific value depends on reliability and predictive or management validity within the population for which the approach is used, not merely on the professional recognition of an instrument.

The evidence on stalking instruments illustrates both potential and limits. [Shea et al. (2018)] Shea et al., 2018 showed that definitional revisions affected interrater reliability, demonstrating that operational clarity matters even among trained raters. [Penney et al. (2023)] Penney et al., 2023 linked changes in dynamic factors to subsequent stalking outcomes in a forensic sample. In another domain, [Guldimann and Meloy (2020)] Guldimann & Meloy, 2020 described promising TRAP-18 findings while recognizing validation gaps. These studies support disciplined documentation and repeated evaluation. They do not license transferring one tool's score to another outcome, population, jurisdiction, or decision threshold without validation.

Temporal Measurement and Case Reassessment

A threat is a changing situation, and assessment quality therefore depends on when observations were recorded. Studies that summarize events after an attack often compress months of ambiguous experience into a coherent narrative. Prospective tracking instead records what was known at each decision point, what interventions were feasible, and how the individual's behavior and surrounding conditions changed. Event-time analysis can examine time to stabilization, recurrence of concerning behavior, re-referral, or adverse outcome. Repeated measurements also help distinguish a short-lived episode from a persistent pattern. Treating every record as static sacrifices precisely the temporal information that threat management seeks to use.

An adequate case chronology includes the timing of reports, evidence acquisition, case conferences, changes in supports or restrictions, contacts with affected persons, and reasoned decisions for continued review or closure. Independent evaluators can then determine whether escalation was detected promptly and whether management altered a relevant exposure. [Penney et al. (2023)] Penney et al., 2023 provide an empirical example of how changing risk factors can improve understanding of stalking outcomes. The more general hypothesis that time-sensitive assessment improves prevention across all forms of targeted violence still requires prospective multicenter tests. Case closure should be a reviewable outcome with the possibility of reopening on new evidence.

Threat Management and the Choice of Proportionate Responses

Intervention is a set of protective and supportive decisions matched to the circumstances, assessed urgency, and available legal authority. Some cases are resolved through clarification and restorative communication; others require mental health services, workplace adjustments, victim support, lawful restrictions, or emergency response by appropriately qualified professionals. The objective is to reduce credible opportunities for harm and improve the conditions that allow safe conduct. An assessment that stops at a severity label omits the scientifically important management stage. Every management decision should identify its intended protective effect, its potential unintended consequences, and the information that will trigger review.

Threat management also needs to consider harms created by its own actions. Unwarranted exclusion, public labeling, excessive monitoring, or abrupt social disconnection can damage well-being and future trust in reporting systems. Therefore the least intrusive adequate response is an empirical as well as ethical principle: it keeps interventions matched to evidence and makes outcomes measurable. The [2024 NTAC guide] National Threat Assessment Center, 2024 is a contemporary operational framework for coordination and management, while the [Cornell et al. (2012)] Cornell et al., 2012 school trial provides evidence that a structured approach can shift disciplinary and support outcomes. Further trials are needed to establish how different response components work for particular kinds of cases.

Victim Safety, Reporting Relationships, and Support

People who are targeted by stalking, intimidation, repeated threats, or credible violent conduct require a distinct place in assessment design. Their accounts may reveal patterns unavailable from an institutional incident record, and their preferences, needs, and safety constraints are directly relevant to management. At the same time, responsibility for preventing violence belongs to institutions and the person making harmful decisions; victims should not be burdened with predicting or controlling another person's behavior. A victim-centered assessment documents changes in exposure, access to protective services, communications of concern, and the safety implications of institutional decisions.

Reporting systems should offer multiple accessible routes and clear information about what happens after a report. A program that collects disclosures but fails to act or communicate can reduce future willingness to report. [Silva and Lankford (2026)] Silva & Lankford, 2026 distinguish leakage, public reporting, and responses as different elements of a possible prevention chain. That conceptual separation is especially valuable for evaluation: successful detection requires both recognition and the appropriate institutional action. Outcomes should include victim-reported safety and service access while protecting confidentiality. Causal effects on victim safety remain difficult to establish when cases differ greatly and interventions are combined.

Development, Disability, Culture, and Disparate Impact

Threat assessment is vulnerable to errors when institutions interpret behavior without regard to development, disability, language, and social norms. A student's literal language, restricted interests, repeated communication, emotional dysregulation, or difficulties reading social expectations may be misunderstood if assessors treat adult normative behavior as the universal baseline. Cultural stereotypes and structural inequalities can alter whose behavior is reported and how an ambiguous event is interpreted. These problems affect both the fairness and accuracy of evidence. Schools and other institutions must implement accessible communication, relevant accommodations, and genuine opportunity for clarification.

The statewide study by [Cornell et al. (2025)] Cornell et al., 2025 is an important warning about institutional selection: students receiving disability services represented a substantial fraction of referrals. Its descriptive design does not determine why referrals differed or whether each decision was justified. A rigorous program audit follows the full pipeline from initial concern to investigation, interventions, exclusion, re-referral, and outcomes. It should compare appropriately defined groups, document disability accommodations, and separate referral disparity from outcome disparity. Appropriate ethical safeguards arise through general principles of proportionality, privacy, educational access, and procedural fairness rather than through a single statistical threshold.

Measurement Error, False Alarms, and the Denominator Problem

Threat assessment research has frequently relied on samples of completed attacks because their histories are accessible and memorable. Such designs are useful for identifying possible prevention opportunities, but they omit the far larger population of people whose concerning behavior never culminated in serious violence. Without this denominator, investigators cannot estimate false-positive rates, precision, calibration, or the burden imposed by protective interventions. A study of attack cases may identify an observed characteristic as common while the same characteristic is equally common among peaceful comparison subjects. The distinction between sensitivity among known attacks and positive predictive value among all referrals is fundamental.

The [Fazel et al. (2012)] Fazel et al., 2012 review is important precisely because it illustrates the dependence of predictive usefulness on the outcome and sampled population. AUC, odds ratios, and retrospective fit do not by themselves show that a tool prevents violence. Programs should define prospective cohorts at initial referral and follow all cases under transparent outcome protocols. Separate estimates are required for serious harm, repeated threats, service use, sanctions, educational exclusion, and case closure. Researchers must report missing data, censoring, loss to follow-up, and how intervention itself may alter the observed outcome. This is the empirical foundation of honest safety claims.

Scientific Designs for Evaluating Prevention

The strongest feasible evaluation design depends on the intervention and ethical context. Cluster randomized school trials can compare adoption of a structured protocol with usual practice, as demonstrated by [Cornell, Allen, and Fan (2012)] Cornell et al., 2012. Stepped-wedge designs may be useful when multiple institutions adopt a program in phases, although secular changes and implementation variability must be modeled. Interrupted time-series designs require stable measurement across enough periods to distinguish policy effects from natural variation. Matched cohort and difference-in-differences analyses can help when randomization is infeasible, but their conclusions rely on assumptions about selection and comparable trends.

Because completed attacks are rare, evaluation should not depend on counts of catastrophic outcomes alone. Meaningful intermediate endpoints include timeliness and fidelity of triage, harmful recurrence, support engagement, victim safety, unnecessary exclusion, and quality of follow-up. These are process and welfare outcomes, and improvement in them does not prove that a prevented attack occurred. Preregistration, independent evaluation, multi-institution data harmonization, and publication of null and adverse findings are essential. Program developers should disclose their roles and financial interests, and studies should report both benefits and harms, including whether reporting creates surveillance or chilling effects.

Interpretation of Retrospective Attack Data

The National Threat Assessment Center's [2023 mass-attacks report] National Threat Assessment Center, 2023 examined 173 attacks in public spaces during 2016–2020. It documented observed behavioral and contextual patterns among perpetrators and recommended early reporting and coordinated intervention. This is a descriptive sample of attacks that occurred, not a prospectively followed population of people referred for threat assessment. Its findings help identify where institutions may have had opportunities to respond, but the report cannot show the probability of an attack given a listed characteristic. Data collected after severe incidents may also be more complete and more intensely scrutinized than records of noncases.

The [FBI's 2018 analysis] Federal Bureau of Investigation, 2018 similarly reconstructs pre-attack behavior in an identified active-shooter case set. Different institutional reports use different eligibility rules, definitions of mass attack and active shooter, time windows, and information sources. These differences cannot be ignored when comparing numerical findings. A comprehensive review states who was included, what evidence was available, whether cases were solved, and how missing observations were treated. Aggregating incompatible case series into a single risk rate would create a misleading appearance of precision. Their proper function is hypothesis generation and identification of potentially modifiable features of prevention systems.

The Emerging 2026 Evidence on Public Reporting

The 2026 publication by [Silva and Lankford] Silva & Lankford, 2026 broadened the empirical comparison of mass shooting types with a study of 483 U.S. events from 2000 to 2023, all involving at least four victims killed. Its analysis separated leakage from threats, and reporting from criminal justice response. It reported that public mass shootings showed higher levels of these pre-event communications than family or felony-related events under its definitions. The work supplies important descriptive comparisons and proposes explanations for underreporting, including uncertainty about whether behavior signals serious danger. Its event selection, dependence on available records, and retrospective reconstruction limit generalization.

A 2026 research report by [Schildkraut, Martaindale, and Greene-Colozzi] Schildkraut et al., 2026 examined 171 mass public shooting incidents from 1999 through 2024 involving 175 perpetrators with identifiable pre-attack communications or warning behaviors. The report helps characterize information-sharing opportunities and institutional intersections in a different selected case corpus. The two 2026 projects should be read together for their complementary definitions, while respecting that both analyze completed attacks and neither establishes a calibrated referral-population predictor. Their strongest contribution is to make public reporting, institutional response, and documentation separate objects of empirical investigation.

Artificial Systems, Digital Evidence, and Decision Support

Digital communications create new evidence channels and new uncertainties for threat assessment. An institution may receive text screenshots, messages, multimedia, or algorithmic alerts, each requiring provenance checks and careful interpretation. Automated classification can help organize large quantities of material, but language models and machine learning systems can misread sarcasm, quoted speech, context, dialect, mental distress, and cultural reference. Human–artificial decision support therefore belongs within the evidence-handling and audit framework rather than outside it. The central empirical question is whether such systems improve relevant outcomes beyond trained human teams while limiting false accusations and unnecessary surveillance.

A defensible evaluation requires independently labeled datasets, disjoint training and validation populations, performance by subgroup, prospective calibration, transparent error analysis, and safeguards for appeal and correction. Historical institutional reports are a biased training source if they reflect earlier discriminatory practices. Machine-produced explanations should be checked against actual evidence rather than treated as evidence of inner motivation. Automated outputs cannot turn an uncorroborated allegation into a verified fact. The broader Artificial Era makes these questions central to psychology because judgment is increasingly distributed across people, interfaces, institutions, and artificial systems. The relevant outcome remains safe, accountable human decision-making.

Professional Competence, Ethics, and Governance

Targeted violence assessment can affect civil liberties, education, employment, clinical confidentiality, and personal reputation. Accordingly, institutions require professional role definitions, a documented basis for information access, safeguards against conflicts of interest, and mechanisms for review and redress. The [APA Specialty Guidelines for Forensic Psychology (2013)] American Psychological Association, 2013 frame relevant questions of competence, appropriate methods, communication, and limits of professional opinions when psychological expertise enters a legal setting. The guidelines are professional norms and should not be mistaken for outcome-validation studies. The applicable legal duties vary by jurisdiction and setting.

Governance should separate reporting, assessment, intervention authorization, and independent audit where possible. Cases with serious imminent danger require appropriate emergency procedures, while ordinary ambiguous complaints require contextual investigation and fair handling. Teams should record reasons for decisions and their expiration or review dates, preserve confidentiality as law permits, and provide an avenue to correct errors. A scientifically mature organization reports both successes and erroneous interventions. Organizational accountability is therefore part of predictive validity: a method's value depends not only on the accuracy of a judgment but also on the consequences of the decision made using it.

Interdisciplinary Integration Across Forensic Psychology

Targeted violence assessment draws on social cognition, developmental psychology, clinical assessment, criminology, public health, victimology, organizational behavior, and decision science. Each discipline contributes distinct evidence. Social psychology studies grievance and group interpretation; developmental science explains age-specific expression and social regulation; clinical research addresses distress and treatment needs; forensic methods evaluate information quality and professional inference; public health studies systems that distribute preventive resources. No single disciplinary construct explains all targeted attacks. Multidisciplinary investigation has scientific value when evidence is integrated through explicitly defined questions and decision rules.

The relationship to neighboring research dossiers is therefore directional rather than duplicative. Violence and Recidivism Risk Prediction examines population instruments and follow-up offending; Stalking assessment emphasizes repeated unwanted pursuit; Algorithmic Risk Assessment focuses on scoring and automation; Forensic Expert Ethics concerns expert accountability and admissibility. This dossier integrates those contributions around a distinct outcome: the investigation and management of an identifiable potential harm in a changing situation. The parent Science, Forensic and Legal Psychology, supplies the overarching psycholegal framework. Police and Public Safety Psychology and Community Psychology offer additional institutional contexts, subject to the canonical architecture actually published.

Critical Synthesis: Established Knowledge and Open Questions

Several scientific findings are well grounded within their respective designs. Targeted attacks may be preceded by observable changes or communications, and retrospective case analysis can locate possible points for intervention Fein & Vossekuil, 1998 National Threat Assessment Center, 2023 Silva & Lankford, 2026. Structured school threat assessment has experimental evidence for improvements in service referral and reduced exclusionary discipline among students making threats Cornell et al., 2012. Stalking instruments show context-specific discrimination and illustrate how operational definitions alter reliability Shea et al., 2018 Penney et al., 2023. Literature on violence prediction places strong limits on claims about individual forecasting, particularly outside studied populations Fazel et al., 2012. These are complementary, not interchangeable findings.

The central unresolved question is whether behaviorally guided interventions causally reduce severe violence across different settings while preserving individual rights. Other open problems include false alarms, underreporting, assessor disagreement, instrument transportability, differences in institutional referral practices, unintended harms of monitoring, and how protective decisions change future data. Developers and independent evaluators should collect all referred cases rather than only dramatic outcomes, preregister intervention hypotheses, provide uncertainty intervals, and permit reproducibility assessments. The criterion of progress is a credible account of safer outcomes and fewer avoidable harms, measured prospectively and reviewed independently.

Conclusion

Targeted violence and threat assessment form a prevention-centered psychological research field concerned with emerging risk in specific, evolving situations. Its most durable intellectual contribution is the shift toward examining behavior, context, communication, institutional response, and possibilities for interruption. Case reconstruction documents the ways some completed attacks developed. Structured assessment offers a disciplined language for inquiry. Program evaluation determines whether these practices provide benefit. Each level requires its own evidence, and none justifies claims of certain prediction or an invariant offender profile. The central scientific object is a dynamic person–situation–institution relationship in which protective decisions can alter outcomes.

The synthesis supports a research agenda that combines prospective referral cohorts, transparent measurement, rigorous intervention evaluation, victim-centered protection, careful accommodation for disability and cultural context, and accountability for institutional error. Emerging digital and artificial systems should be evaluated by the same principles, with additional attention to provenance, calibration, subgroup disparities, and contestability. The field matures when it can report not only the characteristics of attackers, but also the outcomes of interventions, the experiences of people assessed, and the safety of those whom institutions are obligated to protect. That standard makes prevention a testable scientific responsibility rather than a retrospective narrative of missed warning signs.

References