Some mental disorders increase the risk for violence (1–6), with higher rates of violence now firmly established most prominently for individuals with diagnoses of substance abuse, followed by cluster B personality disorders, and to a lesser extent, schizophrenia spectrum disorders (7–9). The means by which mental disorders increase the risk for violence have, however, remained controversial. Two complementary approaches—actuarial and clinical—have until recently produced divergent findings. On one hand, actuarial models have demonstrated the predictive power of age and history (e.g., references 1, 2); on the other hand, clinical and personal distress variables, most notably symptoms and diagnoses, have traditionally failed to predict violence (10). This has led some to argue that the same actuarial risk factors operate for both mentally disordered and non-mentally-disordered offenders, suggesting that clinical factors contribute little to the risk for violence (10, 11).
Several especially well-designed studies have now established specific clinical risk factors for violence (e.g., references 7–9, 12–14). These data have suggested that these risk factors are formed early in development as dynamic individual difference variables related to temperament (15), subsequently crystallized into quantifiable personality traits (16, 17), and differentially associated with primary and comorbid mental disorders (18) that are known to have a higher risk for violence (e.g., references 4, 7, 9, 13). These diagnostic effects may very well be mediated by means of different clinical risk factors related to individual differences in personality. The risk for violence may therefore be dynamic, varying as a function of the extent to which certain personality dimensions are present and the degree to which environmental events moderate or exacerbate their expression. Thus, only some persons with mental disorders will show a greater risk for violence, and these at-risk subjects are predicted to have certain personality configurations that are vulnerable to pathological expression.
The aim of this review was to examine the relationship of a greater risk for violence among persons with certain mental disorders in terms of four fundamental personality dimensions: 1) impulse control, 2) affect regulation, 3) threatened egotism or narcissism, and 4) paranoid cognitive personality style. Two of these dimensions—impulse control and affect regulation—are likely substantially compromised by virtually all mental disorders linked to violence. Both dimensions parallel those that have helped to define a general personality and social psychology of criminal conduct (10). Narcissism or threatened egotism and paranoid cognitive personality style have also been empirically linked to violence and mental disorder. These dimensions may play a more specific role in violence and may be particularly important as additional critical features in explaining acts of aggression in individuals with cluster B personality disorders, psychopathy, and schizophrenia.
Substance abuse disorders represent by far the strongest correlates of violence among all mental disorders. Even when one compares substance abuse and comorbid major mental illnesses, substance abuse by itself produces a remarkably strong effect, with rates of violence that are 12 to 16 times higher for individuals with substance abuse disorders in comparison to five times higher for individuals with major mental illness, such as schizophrenia and affective illness (4). More common, however, is that substance abuse disorders are accompanied by several distinct comorbid conditions, most prominently cluster B personality disorders, particularly antisocial personality disorder. In fact, antisocial personality disorder and its often childhood precursor—conduct disorder—have been most strongly associated with substance abuse disorder, especially alcohol dependence (19), and both conditions may share a genetic liability (20). In addition, anxiety and depression often accompany substance abuse and antisocial personality disorder, and both antisocial personality and depressive disorders may also share a genetic diathesis (21).
Substance abuse, antisocial personality disorder, and depressive disorder may thus share a common genetic origin, with the association among these disorders evident relatively early in development and persisting into adulthood. Each of these disorders invariably compromises both impulse control and emotional regulation but in varying degrees. Antisocial personality disorder, for example, may have a more pronounced impact on impulse control than on affect regulation, whereas the opposite pattern may hold for depressive disorders. The important point, however, is that of comorbidity: each condition may exponentially increase the risk for violence, by means of either additive or interactive effects. Risk will increase as the connections between these personality dimensions and mental disorders increase.
Among the most extreme acts of violence, various subtypes of murder-suicide incidents have been linked to the clinical configuration of depression, antisocial personality disorder, and substance abuse. In one subtype of murder-suicide, a depressed, often substance-abusing offender with antisocial personality disorder murders a significant other and then immediately or soon thereafter commits or attempts suicide (22, 23). In another subtype, more common in aging persons in declining health, anxiety and depression are linked to substance abuse directly and, when combined with suicidal intention, lead to acts of lethal violence, even in the absence of antisocial personality disorder (22). Similarly, for instances commonly referred to as "suicide by cop," a suicidal, depressed, and substance-abusing offender intentionally engages in life-threatening and criminal behavior to provoke a police officer to shoot the offender either in self-defense or to protect others (24). And finally, more common but no less troubling, is that clinical configurations of depression, antisocial personality disorder, and substance abuse are often overrepresented in domestic abusers (25).
Other comorbid configurations with substance abuse have also been consistently associated with a greater risk for violence. For relatively young patients (ages 18 through 40) discharged from state mental hospitals with major mental disorders (e.g., schizophrenia, schizoaffective disorder, delusional disorder), prevalence rates nearly twice as high (31.1%) occurred for patients with major mental disorder and co-occurring substance abuse than for patients with major mental disorder without co-occurring substance abuse (17.9%) (8). Moreover, the highest rates of violence (43%) occurred for patients with co-occurring substance abuse and personality disorders (8).
To summarize, substance abuse may owe its strikingly high risk for violence to the several different pathways through which it is linked to violence. These include a direct pathway through which drugs, such as CNS depressants, impair affective regulation and impulse control leading to violence, as well as indirect pathways mediated by any or all of several families of comorbid disorders and conditions—antisocial personality disorder, anxiety and depression, and chronic psychosis.
Longitudinal studies have provided strong evidence of personality disorders representing a significant clinical risk for violence (7, 26). For example, following over 700 community-based adolescents into early adulthood, Johnson and colleagues (7) found an elevated base rate for violence of 14.4 % among subjects with a DSM-IV personality disorder diagnosis. Perhaps more important, personality disorder symptoms proved to be even stronger predictors of violence than did an overall diagnosis. In fact, increased symptoms of DSM-IV cluster A or cluster B personality disorder corresponded to a greater likelihood of violence in the community during adolescence and early adulthood. Paranoid, narcissistic, and passive-aggressive personality disorder symptoms correlated significantly with violence.
These findings are especially important in that they are derived from a community-based sample with subjects unselected for treatment and followed over a course of 10 years during a developmental epoch (adolescence to early adulthood) known to be a peak period for violence. The longitudinal nature of the study provided evidence that personality disorder symptoms of adolescence preceded subsequent acts of violence in early adulthood. Moreover, a greater risk for violence among both subjects with a personality disorder diagnosis and subjects with particular symptoms remained significant, even when there was statistical control for a number of potential confounds, including comorbid axis I conditions, parental level of psychopathology, age, sex, and socioeconomic status (7).
These findings conform well to studies of clinical and forensic samples, with one notable exception: the absence of a specific diagnosis effect for borderline personality disorder and an elevated risk for violence. Another especially well done study of a community-based sample of adult subjects with personality disorder also failed to demonstrate a diagnostic effect for borderline personality disorder (26). However, as Johnson and colleagues (7) suggested, clinical and forensic studies likely sampled a more severe form of borderline personality disorder than those studied in community samples, and an elevated risk for violence might be evident only in these typically hospitalized subjects with severe forms of borderline personality disorders, which are invariably characterized by significant difficulties with both emotional regulation and impulse control.
Personality disorders and violence may also be examined in reference to a third dimension or clinical risk factor: narcissism or threatened egotism. Social psychological studies have identified threatened egotism as an important normally distributed personality dimension predictive of aggressiveness in nonclinical samples (27–30). In these studies (29), subjects completed self-report measures of both narcissism and self-esteem that, although correlated, assess different personality characteristics. Subjects then wrote brief essays expressing their opinions on abortion that they thought would be critically evaluated by another subject. The experimenters in reality randomly marked the essays as good ("No suggestions. Great essay!") or bad ("This is one of the worst essays I have ever read!"). Subjects were then paired with the subjects whom they believed had graded their essays.
As a laboratory analogue to aggression, the subject pairs then competed in a reaction-time test, and the fastest responder earned the right to blast his or her opponent with a loud noise with the decibel level set by the winner. Subjects who scored highest on measures of narcissism responded with the highest levels of aggression, in the form of noise blasts in the negative—but not positive—evaluation condition. Moreover, aggression did not occur in a follow-up experiment in which subjects with high narcissism played the same reaction-time game with a new person, different from the person who had (ostensibly) praised or insulted them. These findings indicated that anger in response to a narcissistic injury is tied to a particular person and is not a matter of displacement or lashing out at an unrelated party (29).
Narcissistic injury may also play an especially important role in the prodigious rates of violence in persons defined by structured diagnostic instruments as psychopaths. Although not diagnosable within the nomenclature of the DSM, psychopathy represents a virulent form of antisocial personality disorder that is almost eerily free of emotional distress, such as anxiety, psychosis, and depression, as well as other comorbid conditions of antisocial personality disorder (with the exception of substance abuse) (31). Hare (31), arguably the founding figure of the contemporary study of psychopathy, clearly described narcissism as central to an interpersonal factor that, together with an antisocial behavior factor, formed psychopathy. He wrote of the egotism of psychopaths, who have a "narcissistic and grossly inflated sense of self-worth and importance [and] a truly astounding egocentrism and entitlement, and see themselves as the center of the universe, as superior beings" (p. 38). Not surprising, Baumeister and colleagues (28) quoted this same passage in arguing that the narcissism of psychopathy explained the subjects’ high rates of violence.
There is growing evidence for the importance of variables related to a paranoid cognitive personality style in the risk of violence in schizophrenia spectrum disorders. Cognitive personality style reflects a characteristic tendency to behave, feel, think, and perceive—in other words, an attributional bias to view the world in certain consistent ways. It may be expressed on a continuum of psychopathology ranging from an attributional bias of the world as hostile and threatening to relatively encapsulated symptoms of Schneiderian first-rank paranoid delusional symptoms of alien control.
Arsenault et al. (9) examined subclinical symptoms of paranoia, described as a psychological predisposition or cognitive personality style of perceiving excessive threat in the environment, in a birth cohort of 961 young adult community residents in New Zealand who were unselected on the basis of mental health and/or criminal justice considerations. Subjects completed a questionnaire consisting of 17 true/false items, such as "When people act friendly, they usually want something from me" and "Some people go against me for no good reason." The groups consisting of subjects with schizophrenia spectrum disorder, alcohol dependence, and marijuana dependence all demonstrated elevated rates of violence. For subjects with schizophrenia spectrum disorders but not for the two substance dependence groups, elevated rates of violence could have been statistically attributable only to a paranoid cognitive personality style, as objectively defined by the 17 true/false items. Arsenault et al. (9) noted, "Our finding suggests that this cognitive personality style may tip schizophrenic patients toward violence during episodes of psychosis, but also that distorted information processing may promote the violence of a broader group of individuals in whom subclinical syndromes in the schizophrenia spectrum develop" (pp. 984–985).
Link et al. (12) compared various groups of mentally ill former inpatients living in the community with a randomly selected sample of comparison subjects without any mental disorder who were recruited from the same community. Active psychotic symptoms were assessed with such questions as "How often have you felt that thoughts put into your head were not your own?" "How often have you thought you were possessed by a spirit or devil?" and, "How often have you felt that your mind was dominated by forces beyond your control?" Responses to these rather simple but pointed questions proved to be the strongest correlates of recent violent behavior. In fact, violence could only be attributed to affirmative endorsements of these questions, which were indicative of current paranoid psychotic symptoms. Group differences became statistically insignificant when there was control for active psychotic symptoms. Actuarial data and social desirability did not influence the relationship of psychotic symptoms and recent violent behavior, which also remained significant, even when alcohol and drug use were taken into account.
Link and Stueve (13) proposed that these paranoid psychotic symptoms only, which they described as delusions of "threat/control override," represented significant clinical risk factors for violence. In a large epidemiological sample of more than 10,000 adults living in the community, Swanson et al. (6) compared delusions of threat control with other prominent positive symptoms, such as hallucinations and delusions of grandeur. The findings again demonstrated the strongest support for relationship of delusions of threat/control override and violence for mentally disordered patients living in the community. For example, mentally disordered subjects with threat/control override had probabilities of self-reported violence of 0.63 in comparison to 0.39 for mentally disordered subjects without these paranoid symptoms. Of interest, these symptoms were also associated with greater rates of violence in comparison subjects, perhaps suggestive of a paranoid personality style that is expressed much more prominently and intensely in the form of clear-cut paranoid threat/control override delusions in mentally disordered subjects.
Findings from a specifically selected group of psychotic patients committed to a maximum-security facility compared favorably to those from unselected epidemiological samples of community residents. For example, Nestor et al. (32) examined the phenomenology of psychotic symptoms experienced around the time of arrest for subjects committed to a maximum-security psychiatric hospital after being charged with serious felonious crimes. Paranoid delusions described as organized, incorporating personal targets, and marked by alien control and impostor beliefs occurred significantly more frequently in severely violent psychotic patients. A similar pattern of findings emerged when acquittees who were considered not guilty of murder by reason of insanity were compared with mentally disordered murderers who had been committed to the same maximum-security hospital (33). Acquittees who had been deemed not guilty by reason of insanity were also more likely to have killed blood relatives, especially parents, in contrast to the convicted murderers, who were more likely to have killed significant others.
For these paranoid psychotic offenders, acts of violence often reached lethal proportions, driven by apparent motives of psychotic self-defense in the face of intense paranoid feelings directed toward personal, often blood-related targets. Of interest, these subjects often showed evidence of relatively preserved executive and self-organizational abilities, as assessed by formal neuropsychological testing, a profile similar to that associated with some forms of paranoid schizophrenia (34). The relative preservation of neuropsychological function might provide a mentally disordered, violent subject with the organizational abilities needed to engage in a sequence of behaviors directed toward targeted victims.
In summary, the evidence thus pointed to a fairly specific relationship between violence on one hand and paranoid ideas and perceptions on the other, with the latter expressed on a continuum ranging from cognitive personality style and referential thinking to malignant, encapsulated delusions. For violent subjects seen as having psychotic conditions but not showing symptoms related to the kind of directed paranoia of those described, other clinical mechanisms were likely operative. But the extent to which these are characteristic of other particular symptoms of psychosis, such as hallucinations, formal thought disorder, or other types of delusions, is unclear. In fact, there is evidence to suggest that psychotic symptoms, exclusive of paranoia, might be associated with lower levels of violence over time, as assessed in recidivism rates, especially for subjects who are engaged in and compliant with active treatment (1, 8, 35).
Psychosis, especially chronic schizophrenia, is another schizophrenia spectrum disorder. It more commonly exerts a disorganizing effect on behavior, thought, and perception, in contrast to the sparing of executive function associated with paranoid schizophrenia. For chronic schizophrenia, a breakdown in the regulation of various negative emotions—most notably anger, hostility, and irritability, along with a loss of impulse control—may lead to reactive states of aggression and violence (e.g., references 35–39). Considerably less important might be the paranoid ideation and delusional symptoms that mediate violence with other schizophrenia spectrum disorders. Also less important, if not absent, is the type of targeted, ideationally driven violence seen in the subjects most commonly diagnosed with paranoid schizophrenia committed to maximum-security facilities.
Not surprising then, violence committed within the context of chronic schizophrenia might be more frequent, yet less severe, than that seen in subjects with other schizophrenia spectrum disorders. Victims are often members of the subject’s social and treatment network, such as caretakers and family members (e.g., see reference 37). Violence may be related more to general factors of impulse control and affect regulation than to any specific, if not unique, feature of schizophrenia spectrum disorders. Comorbid substance abuse intensifies general problems with impulse control and affective regulation. Treatment and prevention might target these general problems and secondary conditions. In fact, Steadman et al. (8) reported that the elevated rates of violence for their diagnostic groups around the time of hospitalization declined over time, especially for the comorbid substance abuse groups. This might very well had been attributable to compliance with outpatient treatment and abstention from problem substances (e.g., reference 35).
The present review mapped clinical risk factors for violence onto four fundamental personality dimensions—two related to regulatory functions of impulse control and affect regulation and two to the personality surface traits of narcissism and paranoid cognitive personality style. These dimensions are argued to operate jointly and in varying degrees within the context of particular families of disorders and their comorbid conditions that either singly or together are known to be associated with elevated rates of violence. Taken together, the findings of these studies have provided the strongest empirical evidence that the relationship between particular mental disorders and violence may be mediated by different configurations across these personality dimensions (9, 16–18).
The risk for violence may be transmitted by distinct—if not independent—pathways, as, for example, narcissistic injury in subjects with the antisocial conditions and paranoia in the subjects with schizophrenia spectrum disorders. Objective measures of psychopathy and paranoia are each correlated with violence but are uncorrelated or negatively correlated with each other (40). Such a pattern of correlation among measures is often considered to be ideal for statistical modeling (see reference 41), such as taxonomic classification and multiple regression procedures. Objective personality measures may thus be used in these statistical procedures to test the extent to which particular personality dimensions mediate the elevated rates of violence for particular classes of mental disorders.
Among the most powerful of these personality measures is the Multidimensional Personality Questionnaire (42), a pencil-and-paper true/false test that has been used to identify personality correlates of violence and crime across cultures, countries, races, and genders (e.g., reference 17). In fact, Arsenault et al. (9) used the alienation scale of the Multidimensional Personality Questionnaire to assess paranoid cognitive style in a large cohort of subjects. They found the elevated rates of violence for the schizophrenia spectrum disorder group to be mediated by high Multidimensional Personality Questionnaire alienation scores, indicative of a paranoid cognitive style. Likewise, the Multidimensional Personality Questionnaire provides a robust measure of impulsivity, which correlates significantly with serotonin agonist increases in serum prolactin that have been linked to violence (43, 44). The Multidimensional Personality Questionnaire also includes a scale to measure affect regulation, with an assessment of self-perceptions regarding the management and modulation of both internalizing (anxiety, depression) and externalizing (hostility, anger) negative affective states (17). Failed affect regulation may lead to lower thresholds for frustration, agitation, and irritability, as well as cognitive and behavioral disorganization (45).
Finally, the review also served to return the person (qualities, attributes, traits, etc.) to the scientific debate regarding the important question of violence and mental disorders. By identifying features and dimensions that distinguish those who act violently from the majority who do not, the very real and pernicious stigma of all mentally ill subjects as violence prone may eventually be eliminated. Similarly, clinical risk factors for violence might be identified as targets of potential treatment and prevention, especially important considering that most—if not all—violent mentally disordered subjects are seen in mainstream mental health settings (46), often long before becoming violent (47). A violence risk assessment might be especially important for those without any prior history of violence, one that is not only actuarial but clinical, attentive to diagnoses, symptoms, and personality features—all of which can be evaluated economically, efficiently, and objectively (e.g., references 7, 9, 17, 18). Indeed, the clinical risk assessment measures used in the foregoing studies have excellent demonstrable psychometric properties of reliability and validity evident across different formats, such as structured interviews or objective self-report inventories. Incorporating these measures of risk for violence poses minimal cost but has potentially incalculable benefit in terms of violence prevented and, for extreme cases, lives saved.
Received Aug. 8, 2001; revisions received Dec. 27, 2001, and April 22, 2002; accepted May 13, 2002. From the Department of Psychology, University of Massachusetts, Boston; the Laboratory of Neuroscience, Department of Psychiatry, Brockton VA Medical Center, Brockton, Mass.; and Harvard Medical School, Boston. Address reprint requests to Dr. Nestor, Psychiatry 116A, Brockton VAMC, 940 Belmont St., Brockton, MA 02301; firstname.lastname@example.org (e-mail). Supported by a Young Investigator Award from the National Alliance for Research on Schizophrenia and Depression, a Joseph P. Healey Award, a public service grant from the University of Massachusetts, Boston, to Dr. Nestor, and the Department of Veterans Affairs Medical Research Service.