Douglas, Guy, & Hart, 2009, Psychosis as a Risk Factor for Violence to Others - A Meta-Analysis

Psychosis as a Risk Factor for Violence to Others: A Meta-Analysis

Kevin S. DouglasSimon Fraser University

Laura S. GuyUniversity of Massachusetts Medical School

Stephen D. HartSimon Fraser University

The potential association between psychosis and violence to others has long been debated. Past researchfindings are mixed and appear to depend on numerous potential moderators. As such, the authorsconducted a quantitative review (meta-analysis) of research on the association between psychosis andviolence. A total of 885 effect sizes (odds ratios) were calculated or estimated from 204 studies on thebasis of 166 independent data sets. The central tendency (median) of the effect sizes indicated thatpsychosis was significantly associated with a 49%–68% increase in the odds of violence. However, therewas substantial dispersion among effect sizes. Moderation analyses indicated that the dispersion wasattributable in part to methodological factors, such as study design (e.g., community vs. institutionalsamples), definition and measurement of psychosis (e.g., diagnostic vs. symptom-level measurement,type of symptom), and comparison group (e.g., psychosis compared with externalizing vs. internalizingvs. no mental disorder). The authors discuss these findings in light of potential causal models of theassociation between psychosis and violence, the role of psychosis in violence risk assessment andmanagement, and recommendations for future research.

Keywords: psychosis, violence, risk assessment, mental illness, schizophrenia

People have long assumed that there is a link between majormental illness (MMI) and violence. This belief is prevalent acrosscultures and has been quite stable over at least the past fivemillennia (Monahan, 1992a). Surveys of community residentsindicate that this idea is still widespread in contemporary Ameri-can society (Link & Cullen, 1986; Link, Cullen, Frank, &Wozniak, 1987; Pescosolido, Monahan, Link, Stueve, &Kikuzawa, 1999; Riskind & Wahl, 1992; Socall & Holtgraves,1992). Starting in the 1960s and continuing into the mid-1980s,critics began to question this assumption. The basis for the criti-cisms was twofold: First, there was a general lack of empiricalevidence supporting the association and, second, the results offollow-up studies of psychiatric patients released into the commu-nity were interpreted as evidence against an association (Bonta,Law, & Hanson, 1998; Monahan, 1981/1995; Rabkin, 1979;Steadman & Halfon, 1971; Steadman & Keveles, 1972; Teplin,

1985). Some critics were cautious, concluding that the extantresearch was an inadequate basis from which to draw any firmconclusions (the “absence of proof” position; e.g., Monahan, 1981/1995). Others held more extreme views, arguing that, despite itsweaknesses, the evidence was sufficient to disprove a link betweenmental disorder and violence (the “proof of absence” position; e.g.,Teplin, 1985).

Whether there is a link between MMI and violence, and, moreimportantly, what the nuances of such a link may be, is importantfor several reasons. First, persons with MMI are among the moststigmatized groups in society (Corrigan & O’Shaughnessy, 2007;Corrigan & Penn, 1999). Moreover, persons who carry a psychi-atric diagnosis may internalize such stigma and experience dimin-ished self-esteem and self-efficacy (Watson, Corrigan, Larson, &Sells, 2007). A perception that persons with MMI are violent un-doubtedly contributes to this well-documented stigma (Phelan &Link, 1998). Understanding the relationship between these constructscould inform efforts to dispel myths (if there is not an association) or,conversely, could educate the public about the realities of MMI,including the limits of its association with violence.

Second, various symptoms and syndromes of MMI, such asdelusions and hallucinations, are often used in violence risk as-sessments to determine whether persons’ constitutionally protectedliberties should be violated in the name of public safety. Perhapsthe clearest example of this is involuntary civil commitment,where (depending on jurisdiction) laws dictate that a person maybe detained against his or her will if, inter alia, a risk to others isposed that is connected to a MMI.

Third, research has demonstrated that violence perpetration iscorrelated with other adverse outcomes for persons with MMI,such as victimization and suicide/self-harm (Hillbrand, 2001;

Kevin S. Douglas and Stephen D. Hart, Department of Psychology,Simon Fraser University; Laura S. Guy, Department of Psychiatry, Uni-versity of Massachusetts Medical School.

Kevin S. Douglas gratefully acknowledges the support of the MichaelSmith Foundation for Health Research, Career Scholar Program. Duringthe study, Kevin S. Douglas served as a Guest Professor of AppliedCriminology at Mid-Sweden University. Laura S. Guy gratefully acknowl-edges the support of the Social Science and Humanities Research Councilof Canada, Doctoral Fellowship. We thank Ray Koopman for his statisticalconsultation, and Sarah Spain for her assistance in conducting the study.

Correspondence concerning this article should be addressed to Kevin S.Douglas, Department of Psychology, Simon Fraser University, 8888 Uni-versity Drive, Burnaby, British Columbia, Canada, V5A 1S6. E-mail:[email protected]

Psychological Bulletin © 2009 American Psychological Association2009, Vol. 135, No. 5, 679–706 0033-2909/09/$12.00 DOI: 10.1037/a0016311

679

Nicholls, Brink, Desmarais, Webster, & Martin, 2006). Althoughcausal statements cannot be made about this association, it high-lights the need to understand the connection between MMI andviolence in that the perpetration of violence could inform under-standing of risk for the experience of victimization and suicide/self-harm. Similarly, violence to others ostensibly could disrupttreatment efforts, as well as lead to psychiatric deterioration andoverall worse mental health and quality of life. In turn, theseoutcomes could elevate the risk for self-harm, self-neglect, victim-ization, and alienation from familial and other supports.

Fourth, violence to others poses a salient public health concern.To the extent that MMI raises the risk for violence, persons in thecommunity will be victimized. Often, when persons with MMI areviolent, the victims of violence are family members (Estroff,Swanson, Lachicotte, Swartz, & Bolduc, 1998), who thereforemay bear a disproportionate risk of victimization and personalsuffering.

Finally, there are enormous potential costs associated with apotential link between MMI and violence. Financial costs associ-ated both with MMI/mental health treatment and with violence areestimated to be in the hundreds of billions, or trillions, of dollarsper year (Anderson, 1999; D. P. Rice, Kelman, & Miller, 1992). Tothe extent that violence is associated with MMI, there may beconcomitant economic and personal costs due to increased violentbehavior (costs to members of the public, to the patient) andsociety at large (costs of hospitalization, incarceration).

Numerous narrative reviews of the literature on mental disorderand violence or crime have been written. Popular academic posi-tion has shifted over the years. Prior to the 1990s, the conventionalview held that there is no association (e.g., Teplin, 1985) or at leastno demonstrable association (Monahan, 1981/1995; Rabkin, 1979)between mental disorder and violence. Starting in the 1990s,however, the conventional view evolved to one acknowledgingthat the association between mental disorder and violence wassmall but statistically robust and practically important (Monahan,1992b; Mulvey, 1994). This position is aptly summarized by E.Silver (2006):

The vast body of research conducted . . ., including studies of violencecommitted by psychiatric inpatients before, during and after hospital-ization, studies of rates of mental disorder among inmates incarceratedfor violent offenses, epidemiological and birth cohort studies measur-ing the onset and prevalence of mental disorder and violence cross-sectionally and over the life course, and several comprehensive liter-ature reviews summarizing these studies, suggests that: Althoughmost people with major mental disorder do not engage in violence, thelikelihood of committing violence is greater for people with a majormental disorder than for those without. (p. 685)

Similarly, in a review of a cluster of family, twin, adoption, andepidemiological studies, Tehrani, Brennan, Hodgins, and Mednick(1998) stated the following:

Violent offending by the mentally ill and the offspring of the mentallyill tends to be recidivistic. These groups tend to commit a dispropor-tionately high percentage of the total number of violent crimes com-mitted by the total cohort. The absolute number of the violent crimescommitted by these individuals has some societal significance.(p. S85)

Of the various forms of MMI, most reviewers identified psy-chotic disorders as having the clearest association with violence

(Bjørkly, 2002a, 2002b; Bloom, 1989; Bradford, 1983; De Pauw &Szulecka; Junginger, 1996; Krakowski, Volavka, & Brizer, 1986;Monahan, 1992b; Tardiff, 1984; Wessely, 1993). For example, ina Danish birth cohort of more than 358,000 people, Brennan,Mednick, and Hodgins (2000) reported that the risk for violentoffending for men with schizophrenia was 4.6 times higher thanthat of the general population, even after controlling for the po-tentially confounding factor of low socioeconomic status (SES).For women, the risk was 23.2 times higher. Similarly, a recent U.S.study of 1,410 patients with schizophrenia drawn from 57 mentalhealth sites across 24 states (Swanson et al., 2006) found thatpositive symptoms of schizophrenia were associated with bothminor and serious violence, even after controlling for numerouspossible confounds and covariates.

However, others have rejected the view that violence has anysubstantial association with MMI more generally, or with psycho-sis more specifically. A large, multisite follow-up of patientsdischarged from psychiatric emergency units in three states foundthat MMI was typically unrelated, and sometimes slightly nega-tively related, to violence (Appelbaum, Robbins, & Monahan,2000; Monahan et al., 2001; Steadman et al., 1998). Some veryspecific symptoms, such as command hallucinations to perpetrateviolence, predicted violence (Monahan et al., 2001), as did comor-bid MMI and substance use disorders (Steadman et al., 1998).However, a fair summary of this study is that MMI, includingpsychotic disorders, played a very small role in the violence ofpatients.

Elbogen and Johnson (2009) analyzed a two-wave epidemio-logical data set of 34,653 persons residing in the community in theUnited States. The authors reported that MMI, measured at Time1, did not predict violence, measured between Times 1 and 2.However, comorbid MMI and substance use disorders did predictlater violence, more so than substance use disorders alone, sug-gesting an interaction between MMI and substance use disorders.Although the odds of future violence among those with schizo-phrenia alone were about double the odds of future violenceamong persons without schizophrenia, this association was notstatistically significant, despite the large sample size, perhapsattributable to extreme skew in cell frequencies.

A lack of association has been reported in studies of mentallydisordered offenders. For example, on the basis of studies of alarge cohort of forensic psychiatric patients, Quinsey, Harris, Rice,and Cormier (2006) concluded the following:

The presence of schizophrenia and psychotic symptoms exhibitedaround the time of the index offense or admission to hospital werenegatively related to risk. Psychosis, psychotic symptoms, and exac-erbation of those symptoms have little value as indicators of the riskof violence in offender populations. (p. 113)

The same conclusion was reached in a meta-analysis of variablesassociated with recidivism among mentally disordered offenders.Bonta et al. (1998) found that the average association betweenpsychosis and violence was small and negative (r � �.04) acrossthe 11 studies in their meta-analysis reporting on psychosis.

Why have findings and opinions been so discrepant? One likelyexplanation is methodological diversity. Researchers have exam-ined the association between MMI and violence in many differentsamples, using many different research designs, defining and mea-suring MMI and violence in many different ways. As such, one

680 DOUGLAS, GUY, AND HART

cannot easily determine from a cursory reading of the literaturewhy some studies find support for an association and others do not.One can find studies from all types of settings that report positiveassociations, including samples of criminal offenders (Etherington,1993), forensic patients (Erb, Hodgins, Freese, Muller-Isberner, &Jockel, 2001), civil patients (Aarsland, Cummings, Yenner, &Miller, 1996), and the general population (Brennan et al., 2000).Similarly, one can locate studies from these same settings that failto find a positive association between mental disorder and vio-lence: criminal offenders (Arboleda-Florez et al., 1995), forensicpatients (Harris, Rice, & Quinsey, 1993), civil patients (Appel-baum et al., 2000), and the general population (Elbogen & John-son, 2009; Swartz & Lurigio, 2004). Given such discrepant find-ings across different populations, it is simply understandable whydifferent conclusions are reached by different commentators.

Another explanation is confounding factors. Understanding thelink between MMI and violence is complicated by the fact thatmental illness is associated with a host of other variables, many ofwhich, in turn, increase the odds for violence (Elbogen & Johnson,2009). Some researchers have raised the sensible possibility thatthe association between MMI and violence, should it be observed,could be attributable to confounding factors, such as young age,low SES, comorbid substance use, or personality disorder (Walsh,Buchanan, & Fahy, 2002). Controlling or adjusting for potentialconfounding factors may result in a dramatic change in the inter-pretation of findings. This is a legitimate possibility. Some studieshave reported that, once other putative risk factors are controlled oradjusted for, the relationship between MMI and violence becomesnonsignificant (Mojtabai, 2006). Yet, many studies—often ofgood quality—have found the converse: MMI remains predictiveof violence in the presence of other risk factors that have beenentered into multivariate analyses (Swanson et al., 2006) or afteradjusting for population parameters, such as age (Brennan et al.,2000; Eronen, Hakola, & Tiihonen, 1996b), marital status, andSES (Brennan et al., 2000). Resolving this confounder issue willrequire as much theory as data in that some possible covariatescould be consequences, rather than causes, of MMI. For instance,some research shows that MMI causes low SES, rather than theconverse (e.g., the geographic drift hypothesis; Dembling,Rovnyak, Mackey, & Blank, 2002; see also Hudson, 2005). Hence,it is not clear whether such variables should be controlled for inanalyses of the association between mental illness and violence.

Despite these methodological and conceptual complicationswithin the literature, one theme that has emerged from past nar-rative reviews and research is that the key to understanding therelationship between MMI and violence may lie with a morefocused approach on its symptoms. In particular, commentatorshave suggested that psychotic disorders (e.g., schizophrenia) andsymptoms (e.g., delusions) might form the basis for the bridgebetween MMI and violence (Monahan, 1992b). We turn our at-tention to this issue.

Possible Conceptual Links Between Psychosisand Violence

Psychosis is a syndrome found in mental disorders such asschizophrenia, delusional disorders, bipolar mood disorder, andsome forms of severe depression. The syndrome comprises symp-toms reflecting profound disturbances in thought, perception, and

behavior. The thought disturbances include delusions (false be-liefs, often bizarre, held with conviction even in the face ofdisconfirmatory evidence) and impaired communication (disorga-nized, illogical, or incoherent speech). Perceptual disturbancesinclude hallucinations (perception in the absence of an externalstimulus, such as hearing voices or seeing things that are notthere), derealization (a sense that the external world is not real),and depersonalization (a sense that one is no longer a real, auton-omous human being). The behavioral disturbances include disrup-tions in activity level (extreme agitation or lethargy) and disorga-nized or purposeless behavior (odd gestures, gait, and posturing;social withdrawal and poor hygiene; and strange habits, such aseating cigarette butts or playing with feces). People with psychosisalso frequently experience disturbances of mood (flat or bluntedemotions, silliness) and motivation (extreme apathy).

Clearly, psychosis is a heterogeneous syndrome. But it is em-pirically robust: Certain symptoms co-occur much more frequentlythan would be expected by chance and wax or wane together overtime. Moreover, events or circumstances that tend to exacerbate oralleviate one symptom often influence several others at the sametime. In fact, factor analytic techniques suggest that the broadsyndrome of psychosis appears to subsume at least three distinctdomains of symptomatology: positive, disorganized, and negative(Ratakonda, Gorman, Yale, & Amador, 1998; see also Baxter,1997, who provided a conceptual discussion of the three-domainmodel as it relates to violence). The positive domain includessymptoms that are pathological by their presence, such as delu-sions and hallucinations. The negative domain includes symptomsthat are pathological by their absence, such as flat or bluntedaffect, poverty of speech, apathy, and social withdrawal. Finally,the disorganization domain includes symptoms that reflect impair-ment in basic cognitive functioning, such as in odd or purposelessbehavior, impaired communication, and inappropriate affect.

Why might psychosis be associated with violence? There are atleast three possible explanations. It is possible that psychosis is acause of violence. Psychotic symptoms may provide a clear, ifdelusional, motivation for violence or interfere with the ability tomanage interpersonal conflict. An important point here is thatproof of causation requires demonstration that psychotic symp-toms occur before (certain) acts of violence. Psychosis also may bea consequence of violence. Perhaps the stress of perpetratingviolence triggers the onset of psychotic symptoms in people whoare so predisposed. Finally, psychosis may be a simple correlate ofviolence. The association between the two may be statistical ratherthan causal, the result of links with some third variable, such asstressful life events, lack of social support, personality traits,substance use, victimization, and so forth. If this explanation weretrue, we would expect to find no clear temporal or statisticalassociation between psychosis and violence, at least after control-ling for potential confounding factors. At the present time, thereexists no clear, unambiguous evidence to rule out any of thesepotential mechanisms.

If psychosis is a cause of violence, there are three major roles itmay play. First, psychosis may play a role in focusing (organizing)decision and behavior, giving individuals a clear motivation forviolence. Many people with psychosis report clear explanations fortheir behavior and commit acts that are complex, organized, andgoal-oriented, even if appearing illogical to outside observers. Thishas been described aptly as the “principle of rationality within

681PSYCHOSIS AND VIOLENCE

irrationality” (Link & Stueve, 1994, p. 143). If psychotic symp-toms play a focusing role in violence, they are almost certain to besymptoms from the positive domain (i.e., delusions or hallucina-tions).

Second, certain types of psychosis may play a role in destabi-lizing (disorganizing) decisions and behavior, interfering with theability of individuals to manage interpersonal conflicts. Here,disturbances of thought, behavior, and affect (illogical speech,agitation, labile mood) may frustrate psychotic individuals or thepeople with whom they interact, increasing the likelihood thatsomeone will become angry or make an impulsive decision to actviolently. Hiday (1995; see also Hiday, 1997) has posited a modelin which violence may occur through “tense situations” elicited byactive psychotic symptoms. Baxter (1997) has hypothesized thatthis type of psychosis can lead to “disorganized/impulsive” vio-lence and crime.

Third, psychosis may play a disinhibiting role in violence.Whereas positive symptoms motivate or excite behavior and dis-organization symptoms destabilize behavior, negative symptomsinterfere with goal-directed behavior. Because of their retardinginfluence on behavior, negative symptoms probably play a ratherlimited role in decisions to act violently. However, negative symp-toms may increase violence risk in specific circumstances. Forexample, if individuals have comorbid substance use or personalitydisorders, symptoms or consequences of the comorbid disordermay give rise to motivations for violence, and negative psychoticsymptoms may result in a lack of inhibitions to act violently by(further) impairing one’s ability to experience empathy, remorse,or anxiety. Moreover, negative symptoms that result in depressionor suicidality may increase violence risk, as morbid thoughts ofself-harm may change or expand in focus to include others. Thesetheoretical routes to violence underscore the need for an analysis atthe level of symptom or at least that of symptom domain.

Methodological Challenges in the Study of Psychosisand Violence

The interest sparked by the psychosis–violence debate has led toincreased awareness of methodological weaknesses in the research(see discussions by Junginger & McGuire, 2004; Krakowski et al.,1986; Monahan, 1981/1995, 1988, 1992b; Monahan & Steadman,1994; Mulvey, 1994; Mulvey, Blumstein, & Cohen, 1986;Walsh etal., 2002; Wessely, 1993). At least four major areas of concernhave been identified: study design, conceptualization and measure-ment of MMI, conceptualization and measurement of violence,and the role of potential confounding factors.

Study design. In the past, many studies were based on rela-tively small samples of convenience comprising cohorts of civil orforensic psychiatric patients who had undergone assessment, weretreated for months or years, and then were followed up some timeafter their release into the community. Problems with such a designinclude the lack of an appropriate comparison group (e.g., Was therate of violence among patients higher or lower than that amongnonpatients?) and an inability to control for interventions (e.g., Didtreatment or community support alter the likelihood of violence?).

Further, studies on the connection between mental illness andviolence have been conducted in a variety of samples, includingcivil psychiatric patients (Monahan et al., 2001), forensic psychi-atric patients or “insanity acquittees” (Rosenfeld & Harmon,

2002), criminal offenders (Porporino & Motiuk, 1995), communityresidents (Brennan et al., 2000; Elbogen & Johnson, 2009), orsome mixture of the foregoing (Ryan, Hart, Messick, Aaron, &Burnette, 2004). Comparisons can be made among persons withdifferent disorders (Tardiff, Marzuk, Leon, & Portera, 1997) or toindividuals without a diagnosis of mental illness (Beaudoin, Hod-gins, & Lavoie, 1993). Some studies have used a form of randomsampling, such as consecutive admissions (Klassen & O’Connor,1988a), whereas others have not (Asnis, Kaplan, van Praag, &Sanderson, 1994). Some have used matching designs (Hodgins,1992). Both prospective (Monahan et al., 2001) and postdictivedesigns (Modestin & Wuermle, 2005) commonly have been used,and violence has sometimes counted only if it occurred in ahospital or prison (Krakowski & Czobor, 2004b) and in other caseshas counted only if it occurred in the community (Swanson,Borum, Swartz, & Hiday, 1999).

MMI. In past research, mental illness often was considered aunitary, static construct and was inferred from an individual’s(former) status as a psychiatric patient. The state of clinical prac-tice with respect to diagnosis and record keeping often did notpermit comprehensive and reliable assessments of mental disorder.This made it impossible to determine whether violence was dif-ferentially associated with various forms of mental disorder (e.g.,Which is more strongly related to violence: schizophrenia orsubstance use disorder?) and whether the severity or course of thedisorder had an impact (e.g., Is remission of symptoms associatedwith decreased likelihood of violence?). Monahan (1988; Mona-han & Steadman, 1994) has described the problem of “impover-ished” predictor variables in the field of violence risk assessment,which is closely allied with the issue of the relationship betweenpsychosis and violence. The problem is rooted in gross categori-zations of complex phenomena, such as MMI. Finer gradations ofconceptualization, to the symptom level if possible, are likely toprovide more meaningful data on the relationship between psy-chosis and violence. Very broadband definitions, such as “anymental disorder,” are apt to blur important distinctions betweenspecific psychotic syndromes and violence. Junginger (1996) madea similar point in his narrative review, in which he described theconcept of “psychotic action,” which refers to violence that stemsfrom and is consistent with the contents and themes of symptoms,such as delusions and hallucinations.

In addition to the level of conceptualization, studies differ interms of how they measure psychosis: Some use only chart-baseddiagnoses (e.g., Quinsey et al., 2006), others use well-validatedresearch-based interview measures (e.g., Crocker et al., 2005), andyet others simply rely on the diagnoses arrived at by clinicians inthe course of their everyday duties. All such differences can impactthe observed association between violence and mental illness, ifonly because of the systematically higher reliability of research-based, standardized interview procedures relative to unstructuredclinical diagnostic procedures (e.g., First, 2003).

Violence. There is no simple way to define or measure vio-lence (e.g., Douglas & Ogloff, 2003; Monahan & Steadman, 1994;Monahan et al., 2001; Mulvey, Shaw, & Lidz, 1994). In the past,there was a tendency to conceptualize violence in dichotomousterms—that is, present versus absent—on the basis of reviews ofofficial criminal or psychiatric records. This practice may bedescribed as providing an impoverished criterion variable (Mona-han, 1988; Monahan & Steadman, 1994) and possesses the same


problems as do impoverished predictor variables. Recently, re-searchers have paid more attention to the nature of the violencecommitted. Numerous studies on the topic, however, still declineto define violence (e.g., Alexander, Crouch, Halstead, & Piachaud,2006). Others include within their definitions ostensibly nonvio-lent behavior, such as damaging property (e.g., Cooper, Browne,McClean, & King, 1983; Grassi, Peron, Marangoni, Zanchi, &Vanni, 2001) or self-harming behavior (e.g., Barlow, Grenyer, &Ilkiw-Lavallle, 2000; Barnard, Robbins, Newman, & Carrera,1984).

Research has varied greatly in interpretations of exactly whatbehaviors constitute violence. Some have limited the definition tophysical acts that cause demonstrable harm to victims (Monahan etal., 2001); others have studied only homicide (Schanda et al.,2004); and yet others have included less serious forms of behavior,such as minor physical acts (i.e., pushing) or verbal behavior (i.e.,threats to harm someone; Troisi, Kustermann, Di Genio, & Sir-acusano, 2003). Moreover, the method of detecting violence hasranged from relatively simple procedures, such as sole reliance onofficial state or federal recidivism records (Kaliski & Zabow,1995), to more comprehensive procedures involving officialrecords, self-report, and reports from collateral informants (Mona-han et al., 2001).

Confounding factors. As reported above, some studies havereported on the relationship between MMI and violence aftercontrolling for possible competing risk factors. A related issuestems from the particular diagnostic mix in a given sample. Forinstance, as Quinsey et al. (2006) have argued, it could be that,when compared with the risk posed by persons with primarydiagnoses of personality disorder or substance abuse/dependence,MMI poses an inverse relative risk for violence. Evidence hasaccumulated to support both substance use (especially alcohol) andcertain personality disorders (Cluster B; psychopathic personality)as robust risk factors for violence (for substance use/alcohol, seeMulvey et al., 2006; Steadman et al., 1998; Swanson et al., 2002;for psychopathic personality, see Grann, Långstrom, Tengstrom, &Kullgren, 1999; Monahan et al., 2001; Skeem & Mulvey, 2001).Some studies have indeed reported that once such diagnoses arecontrolled, or when major mental illness is compared with theseother diagnoses, its relative association with violence is decreased(Tengstrom, Hodgins, Grann, Långstrom, & Kullgren, 2004). Yet,as with other areas of this literature, one can locate studies that failto demonstrate a reduced association (Stueve & Link, 1997).

Purpose of the Meta-Analysis

Given the divergent findings on the association betweenpsychosis and violence reported within the body of research (sum-marized above), a quantitative synthesis of the literature seemsnecessary to move the field forward. We decided to conduct ameta-analysis of the association between MMI and violence, fo-cusing more specifically on psychosis. We had two goals for thismeta-analysis: first, to characterize, in general terms, the magni-tude of the association between psychosis and violence observed inresearch to date and, second, to identify moderating factors—suchas study design and issues related to the conceptualization andmeasurement of psychosis and violence (as discussed above)—thatmight moderate this association.

Method

Studies Included in the Meta-Analysis

Search procedure. Relevant studies were located in four ways.First, we conducted searches of the PsycINFO, MEDLINE, Crim-inal Justice Abstracts, Google Scholar, National Criminal JusticeReference Service (NCJRS), Sociological Abstracts, Humanitiesand Social Sciences Index Database, and Dissertation Abstractscomputerized literature databases. The stems of the followingidentifier and subject words were used in separate and combinedsearches: mental illness/disorder, psychopathology, psychosis,schizophrenia, hallucination, delusion, affective disorders, mooddisorders, violent/aggressive behavior, aggression, homicide, as-sault, crime, and criminal. Second, we searched the reference listsof major narrative review articles. Third, we examined the tablesof contents of 65 journals. The journals selected comprised allsources in which studies included in the present meta-analysiswere published, as well as several additional journals known topublish high-quality research. Finally, we searched the referencelists of articles identified in the first three steps. We limited oursearch to articles published in the scientific literature before orduring August 2006. Unpublished research was limited to disser-tations.

Inclusion and exclusion criteria. Studies were coded for themeta-analysis if they met two inclusion criteria. First, they neededto present data on the association between psychosis and violentbehavior. Any disorder characterized by psychosis was included(e.g., any schizophrenia-spectrum disorder, bipolar disorder, de-pression with psychotic features) so long as the assessment was notbased solely on a self-report measure of personality or psychopa-thology (e.g., Brief Symptom Inventory, Derogatis & Melisaratos,1983; Minnesota Multiphasic Personality Inventory-2, Butcher,Dahlstrom, Graham, Tellegen, & Kaemmer, 1989; PersonalityAssessment Inventory, Morey, 1991), as self-reports do not di-rectly assess the presence of psychotic symptoms (e.g., the studiesof Parker, Morton, Lingefelt, & Johnson, 2005, and Valliant,Gristey, Pottier, & Kosmyna, 1999, were deemed not to meet thisinclusion criterion because the only assessment of psychosis com-prised MMPI scores). Studies that did not state explicitly that thementally ill participants in their sample had some form of psycho-sis were deemed not to meet this inclusion criterion (e.g., Lagos,Perlmutter, & Saexinger, 1977, p. 1134, was not coded becauseparticipants were described only as having a “mental illness”). Forsome studies, despite there being a focus on psychosis, we couldnot parse out psychotic from nonpsychotic participants.

In an effort to strike a balance between our competing interestsof wanting to compare “uncontaminated” groups of people withand without psychosis and wanting to include as many studies aspossible for the sake of generalizability, we created the followingrule: Studies were deemed to have met this inclusion criterion if nomore than 5% of the persons in the “psychosis group” had or mighthave had disorders without psychosis and/or if no more than 5% ofthe persons in the “non–psychosis group” had or might have hadsome form of psychosis. Several high-quality studies (e.g., McNiel& Binder, 1989, 1994) were not included in the meta-analysis onthis basis (i.e., more than 5% of the study’s “psychosis group” didnot have psychosis and/or more than 5% of the study’s non–psychosis group” had psychosis). The number of studies that were


included in the meta-analysis according to these criteria wassmall.1 Further, in an effort to ensure that persons without psy-chosis were not included within the “psychosis group,” we did notcode effect sizes on the basis of total scores from clinician- orresearcher-rated measures, such as the Brief Psychiatric RatingScale (BPRS; Overall, & Gorham, 1962) or the Positive andNegative Syndrome Scale (PANSS; Kay, Fiszbein, & Opler,1987). When psychosis was assessed with these types of measures,only individual scales and items that assessed psychosis specifi-cally (e.g., “hallucinatory behavior”) were coded (e.g., scales thatassessed “general psychopathology” were not coded).

Violence was defined as any actual, attempted, or threatenedharm to another person or persons (Webster, Douglas, Eaves, &Hart, 1997). Thus, studies in which episodes of self-harm orviolence against objects formed part of the definition of violencewere not included in the meta-analysis. As such, many studies thatused the total score on structured measures of violence that recordseveral types of “violence” (such as the Overt Aggression Scale;Yudofsky, Silver, Jackson, Endicott, & Williams, 1986) were notincluded because harm to property or self is part of the definitionof violence or aggression in these measures. We also followed the“5% rule” for violence such that only studies in which less than 5%of the “nonviolent” group was violent and/or less than 5% of the“violent” group was not violent were included. Very few studieswere included on the basis of this rule.2

The second inclusion criterion was that the data needed to bepresented in a manner that allowed us to calculate odds ratios or toestimate them from other statistics, such as Pearson’s r or Cohen’sd. As such, we did not include studies or effect sizes in which theassociation between psychosis and violence was described asbeing “not significant” or otherwise lacked sufficient informationto compute an effect size (i.e., that psychosis was significant butonly a p value was reported). Although a few studies were deemednot to meet this inclusion criterion on this basis (e.g., Walsh et al.,2004), this strategy more typically led to the exclusion of a subsetof effect sizes from a study rather than exclusion of the entirestudy. In addition, some studies did not meet this inclusion crite-rion because, although they reported the percentage of violentpersons who were psychotic or vice versa (i.e., the percentage ofnonviolent individuals who were not psychotic), there was nocomparison group (e.g., Fazel & Grann, 2004) and, hence, esti-mating an effect size for the effect of psychosis on violenceliterally was impossible. A detailed record pertaining to selectionof studies and effect sizes (as well as all transformational proce-dures used; see below) is available from the authors.

Overlapping studies and data sets. When two or more dissem-inations examined the same sample and no unique odds ratios forpsychosis and violence could be calculated, the more methodolog-ically sound dissemination was included (e.g., peer-reviewed pub-lications were selected rather than the dissertations on which theywere based; studies that provided more descriptive informationabout relevant moderator variables were selected). Studies thatclearly presented data on the same sample were coded as beingfrom the same data set. Studies also were coded as representing thesame data set if, in the opinions of the authors, the majority of theparticipants in the samples were judged to overlap. Such judgmentwas exercised for two groups of studies.3 In one case, two dis-seminations (Wallace, Mullen, & Burgess, 2004; Wallace et al.,

1998) were judged to overlap only minimally and were coded ascomprising separate data sets.4

Choice and Calculation of Effect Size

Odds ratios were chosen as the effect size because they areappropriate for characterizing the association between two natu-rally dichotomous variables (e.g., psychotic or not; violent or not)and are easily calculated from data presented in the form of 2 � 2contingency tables. Of special relevance to the present investiga-tion, odds ratios are directly comparable across studies with dif-ferent experimental designs (i.e., retrospective, cross-sectional,and prospective), which typifies this literature. As Fleiss (1994)noted,

The odds ratio is not prone to the artifactual appearance of interactionacross studies due to the influence on other measures of association oreffect of varying marginal frequencies or to constraints on one or theother sample proportion. On the basis of this and all of its otherpositive features, the odds ratio is recommended as the measure ofchoice for measuring effect or association when the studies contrib-uting to the research synthesis are summarized by fourfold tables.(p. 259, emphasis added)

As the majority of studies presented frequency data reducible tofourfold tables, we opted for the odds ratio as the effect size ofchoice. The use of odds ratios in meta-analyses also has beendiscussed and recommended by Haddock, Rindskopf, and Shadish(1998). Odds ratios are relatively easy to interpret, with the ob-served values indicating the increased relative risk of one variable(in this case, violence) associated with the presence of the othervariable (in this case, psychosis). Odds ratios of 1 indicate noassociation between the two variables; those greater than 1 indicatea positive association (i.e., increased risk), and those less than one

1 Two studies were coded in which the “psychotic” group contained(fewer than 5%) individuals without psychosis (Fido, Razik, Mizra, &El-Islam, 1992; Fulwiler, Grossman, Forbes, & Ruthazer, 1997), and onestudy was coded in which there were (fewer than 5%) persons withpsychosis in the “nonpsychotic” group (Oulis, Lykouras, Dascalopoulou, &Psarros, 1996).

2 In one study (Nestor, Haycock, Doiron, Kelly, & Kelly, 1995), 2participants in the nonviolent group had committed acts of violence; in onestudy (Raja & Azzoni, 2005), self-harm was included in the definition ofviolence; and in one study (Buckley et al., 2004), individuals who com-mitted minor acts of violence against property might have been included inthe violent group.

3 Three disseminations by researchers at the Penetanguishene mentalhealth center (Harris et al., 1993; M. E. Rice & Harris, 1992, 1995), andfour disseminations by Krakowski and colleagues (Krakowski, & Czobor,1997, 2004a, 2004b; Krakowski, Czobor, & Chou, 1999) were judged tohave samples in which the majority of participants overlapped betweenstudies.

4 Participants in these studies were selected via case linkage from acrime database and a psychiatric registrar in the Australian state of Vic-toria. Wallace et al. (1998; N � 1,044) linked persons convicted between1993 and 1995 to the psychiatric register, whereas Wallace et al. (2004;N � 2,861) studied individuals with criminal records who were diagnosedwith schizophrenia during a first admission to a psychiatric hospital in1975, 1980, 1985, 1990, and 1995 and compared them with communityresidents. Thus, some overlap in the two samples was possible.


indicate a negative association (i.e., decreased risk). Epidemiolo-gists usually consider odds ratios of 2.0–2.5 and higher to repre-sent clinically or practically meaningful associations (Fleiss, Wil-liams, & Dubro, 1986); odds ratios of 3.0 or greater are consideredto be large (Haddock et al., 1998).

Whenever possible, we calculated or estimated multiple oddsratios from a single study so that we could determine whethervarious aspects of study design were related systematically to thestrength of the association between psychosis and violence (seebelow). The majority of studies either presented odds ratios orprovided the necessary information to calculate them directly.When a study presented both adjusted and unadjusted odds ratios,we coded both but used the unadjusted value in the main analyses.5

We opted for unadjusted effect sizes because the majority ofstudies reported only such effects. We report the differences be-tween adjusted and unadjusted effects in the Results section. Whena study only provided other types of effect size—such as r, d, or�2—odds ratios were estimated by converting the effect sizeprovided with formulas presented in Lipsey and Wilson (2001) andRosenthal (1991). We converted rate ratios to odds ratios using theformula detailed in Zocchetti, Consonni, and Bertazzi (1997).Unstandardized beta coefficients from logistic regression analyseswere converted to odds ratios with eb and were exponentiated ifnecessary. The Appendix presents all transformational formulasused.

Most odds ratios (k � 446; 50.4%) were calculated directly fromfrequency counts provided in the studies of the number of peoplewho were violent or not violent, as a function of having or nothaving psychosis. The remaining odds ratios were reported in thestudy (k � 158; 17.9%) or estimated from formulas given in theAppendix (k � 281; 31.7%). Seven of these latter studies reporteda mean rating or the mean number of violent acts committed bypeople with and without psychosis. For those studies, a Cohen’s dvalue was calculated and then converted to an odds ratio. Simi-larly, when a study presented mean scores for violent and nonvi-olent participants on scales or items of measures of psychosis, aCohen’s d value was calculated and then converted to an oddsratio.

In total, then, 885 effect sizes were calculated or estimated from204 studies of 166 independent data sets. (Studies included in themeta-analysis are marked with an asterisk in the Reference sec-tion.) Coding of the odds ratios and moderator variables (describedbelow) were based on consensus ratings of the first two authors.

Weighting

Because an effect size estimate obtained from a large sample isassumed to be a more precise estimate of the population of effectsizes in comparison with one based on a relatively small sample,effect sizes typically are weighted such that those supplied bylarger samples make a relatively more substantial contribution.Optimally, effect sizes should be weighted by their inverse vari-ances (Lipsey & Wilson, 2001). Because of the number of effectsizes that needed to be estimated, it was impossible to use theinverse variance to weight each effect size. Therefore, weweighted all 885 odds ratios using weights that were proportionalto the inverse of the sum of the variance of the true log odds ratiosin the universe of studies and the sampling variance of the log oddsratio for each study.6 This procedure—a random effects model—is

an approximation to the standard weighting practices presented byHedges and Vevea’s (1998) Equation 13 or DerSimonian andLaird’s (1986) Equation 3 but one that avoids using n_i wherepossible and gives more weight to studies with larger samples butdown-weights them appropriately to the extent that there is evi-dence of true between-studies variability (DerSimonian & Laird,1986; Hedges & Vevea, 1998). We considered down-weightingextreme sample sizes important in the current analyses becauseseveral studies had population-sized Ns (i.e., 100,000–400,000),and we were concerned about giving them undue influence inanalyses.7

Choice of Statistical Tests and Analyses

Our first goal was to characterize the association between psy-chosis and violence in overall terms across all studies, so theprimary data analytic strategy was descriptive in nature. Becauseodds ratios are not normally distributed, we used nonparametricmeasures of central tendency and dispersion—the median (Mdn)and interquartile range (IQR), respectively—to describe the distri-

5 Forty-seven studies presented only adjusted odds ratios, which there-fore were used in the main analyses. An adjusted odds ratio takes intoaccount the influence of potential covariates (e.g., age) specified by theresearcher on the association between the two variables of interest that areused to form the odds ratio (i.e., psychosis and violence).

6 By using the natural logarithm of the odds ratios, ln(ORs), weightswere applied that were proportional to 1/(U � Vi), where U � the varianceof the true ln(ORs) in the universe of studies (i.e., tau squared, whichrepresents the random effects variance component) and Vi � the samplingvariance of the ln(ORs) for study i. The lower bound, 16/ni, was substitutedfor Vi, which is the smallest value that Vi can have. This lower boundoccurs when the four cells are the same, n/4: v_min � 1/(n/4) � 1/(n/4) �1/(n/4) � 1/(n/4) � 16/n. U was estimated as U � {[�(Li – L)2]/(k � 1)} �{[16 � �(1/ni)]/k}, where ni � the sample size for study i; Li � theln(ORs) for study i; L � the unweighted average ln(ORs); and k � thenumber of studies. The weight for study i was proportional to wi � 1/(u �16/ni) � ni /[(ni � (u � 16)].

7 Because the inverse variance weighting method often is preferred, wealso calculated inverse variance weights for as many effect sizes as pos-sible so that we could compare weighting methods. We were able tocalculate inverse variance weights for 79.5% of effect sizes. We comparedthe mean weighted logged odds ratios for this 79.5% with the meanweighted logged odds ratios using the method we ultimately used foranalyses, described in the text and Footnote 6. The inverse varianceweighting method produced somewhat larger logged odds ratios (M �.628, 95% CI � .468–.788) compared with the method we adopted (M �.378, 95% CI � .294–.463). The back-transformed raw odds ratios were1.87 and 1.46, respectively. We also compared the two methods using theentire sample of 885 effect sizes (though weights for 181 effect sizesnecessarily were computed with our original approach, as the inversevariance was not calculable). Using the inverse variance weighting method,we obtained a mean log odds ratio of .637, with a 95% CI of .510–.765.Using the method we adopted, we obtained a mean log odds ratio of .441with a 95% CI of .363–.520. In back-transforming, we obtained odds ratiosof 1.89 and 1.56, respectively. The findings that the inverse varianceweighting method produced larger logged odds ratios than did our pre-ferred method are consistent with our concern that the inverse weightingmethod might give undue influence to extreme sample sizes (e.g.,100,000�), some of which were present amongst studies that also hadlarge odds ratios (i.e., epidemiological studies; studies of communityresidents rather than persons in institutions).


bution of odds ratios, although we report the mean and theweighted mean as well. Possible values of odds ratios range from0 to �, with 1 representing chance levels of association (negativerelationships are indicated by values between 0 and 1, and positiverelationships are represented by values between 1 and �). Wethen performed parametric analyses to examine the distributionof the natural logarithms of the odds ratios (log odds),ESLOR � ln�OR. Log odds are distributed more normally thantheir corresponding raw odds ratios. The theoretical range of logodds is �� to �, with 0 indicating chance association (negativerelationship 0 and positive relationship �0). Results were thenconverted back into odds ratios via the inverse natural log function,OR � eESLOR. To ensure that the overall findings were represen-tative, we repeated the descriptive analyses after controlling formultiple effect sizes per study and per data set. The latter controlswere accomplished by replacing the multiple effect sizes per studyor data set with a single value, namely, the median of effect sizeswithin that study or data set.8,9 For studies in which odds ratioswere coded both for scales on a measure (e.g., BPRS positivesymptoms scale) and for the individual items that comprised thescale, only the odds ratio for the scale was included among thevalues considered in computing the median for the overall oddsratio estimate (i.e., the “one per study or data set” odds ratio).

We also aggregated weighted effect sizes using a meta-analysismacro for SPSS written by David B. Wilson (see Lipsey & Wilson,2001; available online at http://mason.gmu.edu/�dwilsonb/ma.html). The macro calculates upper and lower 95% confidenceintervals (CIs) and performs a homogeneity test with the Q statis-tic. The Q statistic (Hedges & Olkin, 1985) is used to identify datasets that produce significant heterogeneity and thus could beconstrued as outliers. It is distributed as a chi-square with k � 1degrees of freedom, where k is the number of data sets.

Moderators

Our goal in terms of moderators was to identify variablesrelating to method, disorder, violence, and demographics thatmight moderate the association between psychosis and violence, asdescribed in the introduction. We examined three categories ofmethodological factors: study design, definition and measurementof psychosis, and definition and measurement of violence.

Study design. Study design comprised eight variables relatedto the basic design of the studies. Sample setting was codedaccording to whether studies included civil psychiatric patients,forensic psychiatric patients, correctional offenders, communityresidents, or a combination of participants from two or more of thefour settings. The method used to select participants from within agiven setting was coded as sampling procedure. This variable hadfour levels, including random sampling, matching, nonrandomsampling, and unspecified sampling procedure. The distinctionbetween retrospective or cross-sectional, archival (or pseudo-)prospective, and true prospective studies was coded as method.Sample size was coded continuously. The gender of participants inthe sample was coded as men only, women only, or both men andwomen. Country of data collection was coded dichotomously asUnited States or any other country. The nature of the comparisongroup was coded as no mental disorder or other mental disorder.Finally, for those studies in which the comparison group wasanother mental disorder, mental disorder comparison group was

coded for whether the disorder was an internalizing, externalizing,organic/cognitive, or other disorder.

Definition and measurement of psychosis. We coded four vari-ables related to the definition and measurement of psychosis. Thenature of the psychotic group was coded as type of psychosis. Wedifferentiated among studies that examined schizophrenia, affec-tive psychoses, mixed or unspecified psychotic disorders, andpresence versus absence of psychotic symptoms. We parsed type ofpsychotic symptoms into the following categories: positive, nega-tive, disorganized, and other or unspecified. We also codedwhether positive symptoms comprised hallucinations and delu-sions versus other positive symptoms, such as grandiosity andsuspiciousness. Symptoms that were of the threat/control-overridetype also were coded. A threat/control-override psychotic symp-tom is one that involves either a threatening component, such asdelusions of persecution in which a person believes that otherswish to harm him or her, or the delusional belief that outside forcescontrol one’s mind, such as thought insertion or withdrawal (Link& Stueve, 1994). Purpose of assessment reflected whether diag-noses were made for clinical, research, or other/unspecified pur-poses. Although studies typically had not examined whether therewere active symptoms of psychosis at the time of violence perpe-tration, we were able to code two variables that approximated thistemporal relation. Timing of assessment was coded according towhether psychosis was assessed before, at the time of, or after theoccurrence of violence; in some studies, psychosis was assessedbefore and after the violence, and in other studies the timing wasunspecified.

Definition and measurement of violence. The final four vari-ables pertained to the manner in which studies defined and mea-

8 Of course, several analytic choices are available in meta-analysis. Thesingle odds ratio for each data set could have been computed in at least twoways, both of which take into account the non-normal distribution of oddsratios. In one approach, the median value of a data set’s raw odds ratioscould be taken and subsequently transformed into a log odds ratio. In theother approach, each raw odds ratio in a data set first could be transformedinto a log odds ratio, and then the mean of those values could be computed.Calculations using both approaches were completed in determining thesingle independent effect size for each data set. Both approaches yieldedsimilar findings (r � .97, p .01, across the 166 pairs of effect sizes). Forall analyses reported, we used the first of the two strategies described here.

9 As described in more detail below, one of the moderator variablescoded was whether people with psychosis were compared with individualswithout a mental illness or with individuals with another mental illness. Forstudies in which odds ratios could be calculated for both comparisongroups, it typically was the case that more effect sizes could be calculatedfor comparisons with other mental illness (e.g., separate effects could becomputed for psychosis versus depression, psychosis versus substanceabuse). To determine whether the computation of the overall associationbetween psychosis and violence might have been biased in either directionby not considering this variable, we took an alternative approach tocomputing a single effect size for each data set. Specifically, we alsocomputed the median of effect sizes within a study or data set by taking theaverage of two values: (a) the median of effect sizes based on comparisonsbetween persons with psychosis and with persons without a mental illnessand (b) the median of effect sizes based on comparisons between personswith psychosis and with another mental illness. Results obtained withthis alternative approach yielded different values for 9 of 166 data sets(mean difference among the 9 pairs of odds ratios was 1.29).


sured violence. Setting of violence was coded according to whetherstudies examined violence that occurred in an institution, in thecommunity, or both. Method of assessing violence had three levels:official records (including direct observation), self-report (includ-ing interviews), or multiple methods (i.e., both official records andself-reports). Severity of violence contrasted studies that examinedsevere violence (homicide, attempted homicide, sexual violence,or assaults resulting in serious physical injury), less severe vio-lence (assaults not resulting in serious physical injury and threatsof violence), and any or unspecified acts of violence. Finally, wealso coded violence comparison group, which detailed whetherparticipants who committed violence were compared with partic-ipants who engaged in criminal/antisocial but nonviolent acts (e.g.,theft or fraud) or in no detected criminal/aggressive behavior at all.

Using a nonparametric test of the differences between medians(�2), we determined whether the magnitude of observed (un-weighted) odds ratios varied significantly as a function of themethodological factors described above. We also aggregatedweighted effect sizes using the SPSS macro written by David B.Wilson, a meta-analytic analogue to the one-way analysis of vari-ance (ANOVA), to investigate each moderator variable.

Results

Association Between Psychosis and Violence

Central tendency. Table 1 summarizes the central tendency(mean, median) and distribution (standard deviation, 95% confi-dence interval, interquartile range, standard error) of unweightedraw and log odds ratios across (a) all 885 effect sizes, (b) one effectsize per study, and (c) one effect size per data set. The raw andlogged odds ratios for the analyses evaluating one effect size perstudy were statistically significant, indicating a very small proba-bility that we would have observed these effect sizes if the popu-lation effect sizes had been zero ( p .0000001).

The mean raw odds ratios were highly negatively skewed in partbecause of the natural distribution of odds ratios. That is, very highodds ratios can influence the mean more so than can very low oddsratios. As such, although we present them for descriptive purposes,we consider the median raw odds and both the mean and medianlog odds ratios to be more accurate depictions of central tendency.

Analyses of the (natural) log-transformed odds ratios indicatedthat their distribution was quasi-normal, as the stem-and-leaf plotin Table 2 indicates. Back-transformation of the mean and medianlog odds ratio in Table 1 yielded values ranging from 1.49 to 1.68,which is highly consistent with the reported median raw oddsratios. The consistency of findings across these analyses indicates

that psychosis was reliably and significantly associated with anapproximately 49% to 68% increase in the odds of violencerelative to the odds of violence in the absence of psychosis. Thismay be interpreted as a small, though reliable effect, correspond-ing to a standardized mean difference score (d) of .24 to .32 (seeHaddock et al., 1998, p. 349).

Dispersion. As Tables 1 and 2 also indicate, there was con-siderable dispersion of odds ratios around the central tendency.The IQR for all 885 odds ratios revealed that at least 25% of all ofthe effect sizes were negative in magnitude (�0.73), indicating anegative association between psychosis and violence. On the otherhand, about 25% of the odds ratios were large in magnitude(�3.30), including about 14% that were very large (�5.00). Asimilar pattern emerged when controlling for multiple effects perstudy and per data set.

The distribution of log odds closely approximated a normalcurve. We evaluated the heterogeneity of the 166 independenteffect sizes (one effect size per sample) by calculating Q, follow-ing Shadish and Haddock (1994, p. 266). Q (169.17) was notsignificant. Although these findings suggest that the distribution ofeffect sizes is homogeneous, the observed variability did, in fact,exceed expectation but merely not to such a degree as to bestatistically significant at p .05. More specifically, because theexpected value of a chi-square is equal to its degrees of freedom,and the observed Q value of 169.17 is larger than its 166 associateddegrees of freedom, the observed variability exceeds what wouldbe expected by chance (see Lipsey & Wilson, p. 135). Theseanalyses indicate that, notwithstanding the nonsignificance of Q,there was considerable variability with respect to the magnitude ofeffect sizes. This supports the search for factors that potentiallymoderate the psychosis–violence association. This is also consis-tent with our observation that 25% of the effect sizes were nega-tive, indicating an inverse association between psychosis and vi-olence, and 25% were positive and large (�3), indicating a strongassociation between psychosis and violence.

Potentially Moderating Variables Between Psychosisand Violence

Publication bias. Meta-analysts have acknowledged that theremay exist a publication bias that favors positive findings. Thismeans that the effect sizes reported in the published scientificliterature may systematically overestimate the true magnitude ofthe effect. This is often referred to as the “file drawer” problem, asnumerous studies with null results—especially studies based on

Table 1Association Between Psychosis and Violence: Distribution of Nonweighted Effect Sizes (Odds Ratios and Natural Log Odds)

Odds ratio Natural log odds

Effect size k M 95% CI SD Mdn Q1 to Q3 M ( SE) 95% CI SD Mdn Q1 to Q3

All effect sizes 885 3.50 2.53 to 4.46 14.59 1.53 0.73 to 3.33 .42 (.04) .34 to .50 1.15 0.43 �0.32 to 1.20One effect size per study 204 2.62 2.20 to 3.03 2.99 1.69 0.81 to 3.36 .51 (.07) .37 to .64 0.97 0.52 �0.21 to 1.21One effect size per data set 166 2.17 1.85 to 2.50 2.11 1.66 0.79 to 3.09 .40 (.07) .26 to .54 0.91 0.51 �0.24 to 1.13

Note. k � number of effect sizes; Q1 � first quartile; Q3 � third quartile.


small samples—may go unpublished and be relegated to the filedrawer.

Logically, bias in favor of positive findings is unlikely to plaguethis area of research as much as it might affect other areas ofresearch because many, if not most, studies have investigatednumerous possible predictors of violence. Hence, if psychosis wasnot significant, but other predictors were, the paper, in theory,would not be more likely to be rejected for publication than studiesfinding a positive effect for psychosis. In addition, on the basis ofour narrative review at the outset of this article, we were much lessinterested in the overall effect size than we were in moderatedeffect sizes, which promise to be much more meaningful.

To help minimize any potential publication bias, we searchedand included unpublished research in the form of dissertations. Welocated only six dissertations that were not subsequently publishedas journal articles. The mean log odds for effect sizes estimatedfrom unpublished dissertations was 0.90, as compared with 1.68for published articles. Although the odds ratio for dissertations wassmaller, the difference was not statistically significant, �2(1,166) � 0.69, p � .41.

We also investigated possible publication bias by conductingtrim-and-fill analyses (Duval & Tweedie, 2000). This procedure isa quantitative form of the more traditional funnel graph, a scatter-plot of the association between sample size and effect size. The“filled” trim-and-fill funnel graph is presented in Figure 1. Thefigure includes a reference line for the overall mean log odds effectsize as well as 95% confidence intervals. Visual inspection doesnot reveal evidence of publication bias.

We used the metatrim procedure in Stata 10 (StataCorp LP,2007) with the linear estimator (the most stable estimator). Thisprocedure indicated that there were 0 effect sizes to trim and thatthe adjusted effect size was identical to the raw effect size. Usingthe run estimator, we obtained a “number to trim” result of 2,meaning that there may be two missing studies with large, negativeeffect sizes (see funnel plot, which shows the imputed effectssurrounded by large squares). The adjusted effect size obtainedwith a random effects model and the run estimator was ln(OR) �0.427, z � 10.35, p .001, 95% CI � 0.347–0.508, where OR

represents the odds ratio. This is not meaningfully different thanthe unadjusted ln(OR) of .43. On the basis of these results, we donot believe that publication bias was operative in the current study.

Moderating Effects: Study Design

For all moderator analyses, we tested the difference amonglevels of moderators using weighted log odds and the Q statistic,which provides an ANOVA analog. However, the odds ratiospresented in Tables 3–17 are raw odds (as opposed to log odds), tofacilitate interpretation. We note that nonparametric significancetests (median tests) of the raw odds ratios produced the samefindings, in terms of which moderators were significant.

Setting. As Table 3 indicates, most studies were of civil psy-chiatric patients. The difference among the effect sizes across thefour settings was statistically significant, with the highest oddsratios obtained in community (i.e., general population) samples.Odds ratios were significantly greater than chance among civilpsychiatric, correctional, and community samples, indicating thatpersons with psychosis in those samples were more likely to beviolent than persons without psychosis in those samples.

Figure 1. Trim-and-fill funnel plot. s.e. � standard error.

Table 2Association Between Psychosis and Violence: Stem-and-LeafPlot of Distribution of Effect Sizes (Natural Log Odds)

Frequency Stem Leaf

5 Extremes �2.89 �2.0 4&

28 �1.0 55678954 �1.0 00011222333487 �0.0 555556666777778888899

130 �0.0 000000111111222222223333334444151 0.0 0000001111111122222222223333334444444150 0.0 5555555566666666677777777788888999999126 1.0 000011111111222222233333334444480 1.0 555566666778888999938 2.0 00111223412 2.0 57&11 3.0 03&4 Extremes �3.7

Note. N � 885 effect sizes; stem width � 1.00; each leaf � 4 cases;& � fractional leaf.

Table 3Moderating Effect of Setting

Setting kMedian odds

ratio IQR (Q1 to Q3)

Civil psychiatric�� 86 1.69b 0.88 to 2.80Forensic psychiatric 38 0.91c 0.43 to 3.07Correctional� 21 1.27b,c 0.89 to 2.87Community�� 15 3.46a 2.21 to 5.50Mixed 9 1.30b,c 0.58 to 2.30

Note. k � number of effect sizes; IQR � interquartile range; Q1 � firstquartile (25th percentile); Q3 � third quartile (75th percentile). OverallQB(4) � 16.70, p � .0022. Median odds ratios that do not share subscriptsdiffer significantly from one another at p � .05. Moderators in the left-hand column marked with designated symbols have weighted mean logodds ratios that differed significantly from chance, as indicated by a z-test.Tau squared∧ for each level is as follows: u � .41 for civil psychiatric; u �1.01 for forensic psychiatric; u � .46 for correctional; u � .67 forcommunity; and u � 1.12 for mixed.� p .05. �� p .001.


Sampling procedure. Most studies used some form of randomsampling (see Table 4), although a fair number did not specifytheir sampling procedure. Odds ratios differed significantly as afunction of sampling procedure, ranging from a low of 1.49 forsimple random sampling to a high of 5.50 for matching (mostlyepidemiological studies), a procedure designed to control for pos-sible confounding factors. Odds ratios were significant at eachlevel of the moderator except for nonrandom sampling, althoughthis effect was based on only seven studies.

Design. Design was not a statistically significant moderatingfactor, and odds ratios were homogeneous across types of design(see Table 5). Odds ratios were significant only for retrospectiveand true prospective designs, although the number of pseudopro-spective studies (k � 18) was small and, hence, power was limitedwithin this level of the variable.

Sample size. We tested whether sample size influenced ob-served effect sizes by treating it as a continuous variable in ameta-analytic analog to regression analysis (MetaReg macro inSPSS; Lipsey & Wilson, 2001). The model was not significant(Q � 0.88, ns). The unstandardized beta for sample size was 0.00.Thus, sample size appears not to have influenced effect size, afinding that is consistent with our trim-and-fill analyses reportedabove.

Gender. Gender did not moderate effect size. Studies that usedonly men, only women, or both men and women as participants didnot produce effect sizes that differed significantly from one an-other. Effect sizes were significantly different from chance at eachlevel of this moderator (see Table 6).

Country of data collection. Effect sizes did not differ betweenU.S. and non-U.S. samples and were significantly different fromchance at both levels of this moderator (see Table 7).

Comparison group. Effect size varied significantly as a func-tion of the nature of the comparison group (see Table 8). As mightbe expected, odds ratios were higher when individuals with psy-chosis were compared with persons without mental disorder(Mdn � 3.68) than when compared with individuals with othermental disorders (Mdn � 1.51). However, odds ratios were sig-nificantly greater than chance at both levels of the moderator.

We further broke down the nature of the comparison group, asshown in Table 9. Psychosis had a stronger effect when compared

with internalizing disorders (nonpsychotic mood disorders, anxietydisorders) than when compared with externalizing disorders (Clus-ter B personality disorders, substance-related disorders). In fact,the effect size for psychosis was not significantly different fromchance when the comparison group comprised externalizing dis-orders.

We also tested the relationship between psychosis and violencewhen the comparison group was limited to personality disorder(k � 34). For approximately half of the 34 data sets for which wecould parse out this subgroup, the nature of the personality disor-der was not specified; for those in which it was specified, the groupconsisted of antisocial or borderline personality disorders or psy-chopathic personality disorder. The median odds ratio for thissubset of analyses was 0.59, and the mean log odds was �0.38(SD � 1.00), which was a statistically significant inverse effect(Z � �2.06, p .05). That is, psychosis was associated withsignificantly lower odds for violence than were these personalitydisorders.

In addition, we tested the impact of substance-related disorderson the psychosis–violence association. The cleanest comparisonswe could make were from studies that divided their samples intoparticipants with psychosis who either did or did not have acomorbid substance-related disorder. We located 12 independentdata sets that permitted this comparison. The mean log odds for

Table 4Moderating Effect of Sampling Procedure

Sampling procedure kMedian odds


Random�� 125 1.49b 0.65 to 2.64Nonrandom 7 1.84b 1.40 to 3.10Matched to population�� 17 5.50a 3.33 to 7.85Not specified�� 20 1.69b 0.91 to 2.94

Note. k � number of effect sizes; IQR � interquartile range; Q1 � firstquartile (25th percentile); Q3 � third quartile (75th percentile). OverallQB(3) � 64.99, p � .0001. Median odds ratios that do not share super-scripts differ significantly from one another at p � .05. Moderators in theleft-hand column marked with the designated symbol have weighted meanlog odds ratios that differed significantly from chance, as indicated by az-test. Tau squared∧ for each level is as follows: u � .62 for random; u �.71 for nonrandom; u � .40 for matched to population; and u � .14 for notspecified.�� p .001.

Table 5Moderating Effect of Study Design

Design kMedian odds


Retrospective�� 123 1.88a 0.81 to 3.10Archival (pseudo-)prospective 18 1.55a 0.62 to 3.89True prospective�� 37 1.49a 0.83 to 2.96

Note. k � number of effect sizes; IQR � interquartile range; Q1 � firstquartile (25th percentile); Q3 � third quartile (75th percentile). OverallQB(2) � 1.06, ns. Median odds ratios that do not share subscripts differsignificantly from one another at p � .05. Moderators in the left-handcolumn marked with designated symbols have weighted mean log oddsratios that differed significantly from chance, as indicated by a z-test.Tau-square∧ for each level is as follows: u � .71 for retrospective; u �1.36 for archival (pseudo-)prospective; and u � .54 for true prospective.�� p .01. �� p .001.

Table 6Moderating Effect of Gender

Gender group kMedian odds


Men�� 56 1.37a 0.76 to 3.22Women�� 21 1.73a 0.92 to 4.98Men and women�� 109 1.76a 0.79 to 2.83

Note. k � number of effect sizes; IQR � interquartile range; Q1 � firstquartile (25th percentile); Q3 � third quartile (75th percentile). OverallQB(2) � 2.00, ns. Median odds ratios that do not share superscripts differsignificantly from one another at p � .05. Moderators in the left-handcolumn marked with designated symbols have weighted mean log oddsratios that differed significantly from chance, as indicated by a z-test.Tau-square∧ for each level is as follows: u � .87 for men; u � 1.43 forwomen; and u � .71 for men and women.�� p .01. �� p .001.


noncomorbid psychosis was 1.54 (SD � 0.55), and that for co-morbid psychosis was 2.37 (SD � 1.16). The difference betweenthese effect sizes was large (Cohen’s d � 0.97). The raw oddsratios associated with the log odds are 4.67 and 10.70, indicatingthat these 12 studies had odds ratios that were larger than theaverage odds across all studies.

Finally, we tested the difference between adjusted and unad-justed effect sizes reported in studies, with the rationale thatadjusted effect sizes control for various other potential risk factorsfor violence. We did so with some caution, as it is not clear whichvariables, in fact, should be controlled statistically (Dembling etal., 2002; Hudson, 2005), in that some covariates, such as SES,might actually themselves be consequences of mental illness ratherthan truly competing causes of violence. Further, different studiescontrolled for different factors. Nonetheless, we located 59 pairs ofadjusted and unadjusted odds ratios. The mean (SD) unadjustedodds ratio was 5.11 (4.95), and the mean adjusted odds ratio was3.22 (3.06), which represented a significant difference, t(58) �3.98, p .001. Median odds were 3.50 and 2.21, respectively. Thestandardized difference between means was moderate (d � 0.47).

Moderating Effects: Definition and Measurementof Psychosis

Type of psychosis. The manner in which psychosis was con-ceptualized in studies significantly moderated the psychosis–

violence association (see Table 10). That is, studies that coded atthe level of the symptom had significantly higher effect sizes thanthose that either failed to specify how they defined psychosis orcombined different types of psychotic disorders into one category.The effect sizes for both the symptom and the schizophrenia levelof analysis were significantly greater than chance.

We looked in more detail at the potential impact of type ofsymptom on the strength of the association between psychosis andviolence. As shown in Table 11, the overall moderating effect oftype of symptom (positive, negative, disorganized, other/unspecified) showed a trend toward significance. Pairwise com-parisons revealed that positive symptoms were associated with alarger odds ratio (2.32) than were negative symptoms (1.32).However, effect sizes at all levels of this moderator were signifi-cantly greater than chance.

We evaluated whether more specific categories of positivesymptoms were associated with violence. Of the data sets, 37permitted a test of whether hallucinations and/or delusions wereassociated with violence, 25 permitted a test of other positivesymptoms (e.g., bizarre behavior, excitement, suspiciousness, non-

Table 7Moderating Effect of Country of Data Collection

Country kMedian odds


United States�� 82 1.72a 0.81 to 2.75Not United States�� 83 1.56a 0.73 to 3.47

Note. k � number of effect sizes; IQR � interquartile range; Q1 � firstquartile (25th percentile); Q3 � third quartile (75th percentile). OverallQB(1) � 0.35, ns. Median odds ratios that do not share subscripts differsignificantly from one another at p � .05. Moderators in the left-handcolumn marked with the designated symbol have weighted mean log oddsratios that differed significantly from chance, as indicated by a z-test.Tau-square∧ for each level is as follows: u � .57 for United States, and u �.76 for not United States.�� p .001.

Table 8Moderating Effect of Type of Comparison Group

Comparison group kMedian odds


Other mental disorder�� 149 1.51a 0.73 to 2.59No mental disorder�� 29 3.68b 0.98 to 6.20

Note. k � number of effect sizes; IQR � interquartile range; Q1 � firstquartile (25th percentile); Q3 � third quartile (75th percentile). OverallQB(1) � 13.24, p � .0003. Median odds ratios that do not share subscriptsdiffer significantly from one another at p � .05. Moderators in the left-hand column marked with the designated symbol have weighted mean logodds ratios that differed significantly from chance, as indicated by a z-test.Tau-square∧ for each level is as follows: u � .55 for other mental disorder,and u � 1.13 for no mental disorder.�� p .001.

Table 9Moderating Effect of Type of Mental Disorder ComparisonGroup

Comparisongroup k

Median oddsratio IQR (Q1 to Q3)

Internalizing� 27 2.15a 0.76 to 4.17Externalizing 47 0.85b 0.41 to 2.00Organic 16 1.41a,b 0.96 to 2.62Other�� 123 1.69a 0.96 to 2.80

Note. k � number of effect sizes; IQR � interquartile range; Q1 � firstquartile (25th percentile); Q3 � third quartile (75th percentile). OverallQB (3) � 14.90, p � .0019. Median odds ratios that do not share super-scripts differ significantly from one another at p � .05. Moderators in theleft column marked with designated symbols have weighted mean log oddsratios that differed significantly from chance, as indicated by a z-test:Tau-square∧ for each level is as follows: u � 1.29 for internalizing; u � .79for externalizing; u � .46 for organic; and u � .47 for other.� p .05. �� p .001.

Table 10Moderating Effect of Type of Psychosis

Type of psychosis kMedian odds


Schizophrenia�� 68 1.71a,b 0.61 to 3.83Affective psychoses 22 1.36a,b 0.66 to 3.27Mixed/unspecified psychoses 62 1.25a 0.62 to 2.85Psychotic symptoms�� 77 2.07b 1.29 to 3.17

Note. k � number of effect sizes; IQR � interquartile range; Q1 � firstquartile (25th percentile); Q3 � third quartile (75th percentile). OverallQB(3) � 9.28, p � .026. Median odds ratios that do not share subscriptsdiffer significantly from one another at p � .05. Moderators in the left-hand column marked with the designated symbol have weighted mean logodds ratios that differed significantly from chance, as indicated by a z-test.Tau-square∧ for each level is as follows: u � 1.10 for schizophrenia; u �.64 for affective psychoses; u � .98 for mixed/unspecified psychoses; andu � .26 for psychotic symptoms.�� p .001.


delusional or nonhallucinatory paranoia) in relation to violence, 16allowed us to assess the relationship between some type of para-noid symptom and violence, and 18 enabled us to examine threat/control-override symptoms in relation to violence.

The median odds ratio for hallucinations/delusions was 2.31,with a mean log odds of .66 (SD � .91), which was significantlydifferent from chance (Z � 4.32, p .0001). For other positivesymptoms, the median was 2.37, and the mean log odds was 0.87(SD � .61), Z � 6.75, p .0001. These types of symptoms did notdiffer from one another. Paranoid symptoms showed a trend to-ward significance: median odds � 1.11; mean log odds � 0.30(SD � 0.73), Z � 1.81, p � .07. Threat/control-override symptomswere significantly different than chance: median odds � 1.92;mean log odds � 0.68 (SD � 0.98), Z � 3.09, p � .002.

Method of assessment. As shown in Table 12, whether eval-uations of psychosis occurred in the context of clinical practice,took place in the context of research, or were unspecified did notmoderate the association between psychosis and violence. In thevast majority of studies, either the purpose of evaluations wasspecified or it could be inferred. Odds ratios were significant forboth of these types of evaluation.

Timing of assessment. Overall, timing of assessment ap-proached significance as a moderating factor (Table 13). The onlysignificant pairwise difference occurred between assessments ofpsychosis made before the measurement of violence and studiesthat did not specify the relative timing of assessment. It is notablethat the largest effect size was that for studies in which theassessment of psychosis occurred at approximately the same timeas the violent incident. It likely did not differ from the othercategories because there were few studies (k � 10) that fell intothis category. When all effect sizes were used, this odds ratio wassignificantly greater than those in the “before” or “after” catego-ries.

Moderating Effects: Definition and Measurementof Violence

Setting of violence. Setting was not a significant moderator.As Table 14 indicates, the strength of association between psy-chosis and violence did not differ as a function of whether violencewas measured within an institution (i.e., psychiatric hospital) or inthe community, and effect sizes were significantly greater thanchance at both levels of this moderator.

Method of violence detection. The method of violence detec-tion did not significantly moderate effect sizes (see Table 15), andthere were no pairwise differences between levels. Effects weresignificant when either official records or self-report was used tomeasure violence, but not when a combination of these methodswas used.

Severity of violence. Severity of violence was not a significantmoderator (Table 16). Psychosis was related to violence whether itwas severe, moderate, or not specified.

Violence comparison group. As shown in Table 17, thepsychosis–violence relationship was stronger when the compari-son of violent individuals was made to individuals without anti-social behaviors, as opposed to nonviolent but still antisocial/criminal incidents. When the comparison was to nonviolent,though antisocial conduct, the odds ratio was not different fromchance.

Table 13Moderating Effect of Timing of Assessment

Timing of diagnosis kMedian odds


Assessment made beforeviolence 51 1.30a 0.62 to 2.40

Assessment made at violence� 10 2.51a,b 0.78 to 4.43Assessment made after

violence�� 84 1.57a,b 0.76 to 2.30Unspecified�� 36 2.22b 1.01 to 3.91

Note. k � number of effect sizes; IQR � interquartile range; Q1 � firstquartile (25th percentile); Q3 � third quartile (75th percentile). OverallQB(3) � 6.89, p � .076. Median odds ratios that do not share subscriptsdiffer significantly from one another at p � .05. Moderators in the leftcolumn marked with designated symbols have weighted mean log oddsratios that differed significantly from chance, as indicated by a z-test.Tau-square∧ for each level is as follows: u � .72 for assessment madebefore violence; u � .64 for assessment made at violence; u � .59 forassessment made after violence; and u � .77 for unspecified.� p .05. �� p .001.

Table 11Moderating Effect of Type of Psychotic Symptom

Type of psychoticsymptom k


Positive�� 62 2.32a 1.23 to 3.46Negative� 22 1.32b 0.87 to 2.05Disorganized� 20 1.85a,b 1.20 to 3.06Other/unspecified�� 22 1.78a,b 1.23 to 3.42

Note. k � number of effect sizes; IQR � interquartile range; Q1 � firstquartile (25th percentile); Q3 � third quartile (75th percentile). OverallQB(3) � 6.88, p � .076. Median odds ratios that do not share subscriptsdiffer significantly from one another at p � .05. Moderators in the leftcolumn marked with designated symbols have weighted mean log oddsratios that differed significantly from chance, as indicated by a z-test.Tau-square∧ for each level is as follows: u � .43 for positive psychoticsymptoms; u � .25 for negative psychotic symptoms; u � .46 for disor-ganized psychotic symptoms; and u � .51 for other/unspecified psychoticsymptoms.� p .05. �� p .01. �� p .001.

Table 12Moderating Effect of Purpose of Assessment

Diagnosticmethod k


Clinical�� 79 1.51a 0.62 to 3.09Research�� 87 1.88a 0.94 to 3.15Unspecified/other 6 1.27a 0.91 to 2.27

Note. k � number of effect sizes; IQR � interquartile range; Q1 � firstquartile (25th percentile); Q3 � third quartile (75th percentile). OverallQB(2) � 1.87, ns. Median odds ratios that do not share subscripts differsignificantly from one another at p � .05. Moderators in the left-handcolumn marked with designated symbols have weighted mean log oddsratios that differed significantly from chance, as indicated by a z-test.Tau-square∧ for each level is as follows: u � 1.08 for clinical; u � .43 forresearch; and u � .36 for unspecified/other.�� p .01. �� p .001.


To gain a better sense of how the strength of association be-tween psychosis and violence varied as a function of the number ofsignificant moderators present, we plotted the mean log odds ratiosas a function of the number of moderators present in a given study.The result is shown in Figure 2 and illustrates that with nomoderators present, log odds ratios were below chance levels.With one moderator present, log odds ratios were small and lowerthan the sample mean. With two to four moderators present, thelog odds ratios were larger than the sample mean. Translating fromthe log odds shown in the data labels in Figure 2, we obtained rawodds ratios corresponding to the number of moderators as follows:no moderators, OR � 0.81; one moderator, OR � 1.09; twomoderators, OR � 1.83; three moderators, OR � 3.66; four mod-erators, OR � 4.56. These data illustrate quite clearly why suchdivergent conclusions have been drawn about the association be-tween psychosis and violence.

Discussion

The findings of our meta-analysis provided strong support forthe view that psychosis and violence are associated with oneanother, albeit with a small overall effect size that varies consid-erably and in important ways across study features and aspects of

psychosis and violence themselves. Across 885 effect sizes from204 studies of 166 independent data sets, psychosis was reliablyassociated with a 49%–68% increased likelihood of violence. Weobserved substantial variability as a function of the moderators wetested. In fact, perhaps of more importance than the overall effectsize estimates were the moderator analyses, which shed light onthe discrepant findings and professional opinions discussed in theintroduction. That is, depending on the number of moderatorspresent in a given study, odds ratios ranged from a meaningfulinverse association to a large positive association. Given this rangein effect sizes, the range in conclusions about psychosis andviolence makes complete sense.

The Status of Psychosis as a Risk Factor for Violence

Putting aside the issue of moderation for the moment, one mightask: How does psychosis compare with other risk factors forviolence in terms of strength of association? It is difficult tointerpret the effect size for the psychosis–violence association inrelation to other risk factors because of a lack of appropriatecomparison data. Only some of the studies included in our meta-analysis permitted us to examine systematically the effect sizeassociated with other diagnoses of mental disorder, demographiccharacteristics, or criminal history variables. Furthermore, thereare no other meta-analyses of which we are aware that have

Table 14Moderating Effect of Setting in Which Violence Occurred

Setting in which violenceoccurred k


Institutional�� 47 1.84a 1.00 to 3.09Community�� 117 1.59a 0.73 to 3.08

Note. k � number of effect sizes; IQR � interquartile range; Q1 � firstquartile (25th percentile); Q3 � third quartile (75th percentile). OverallQB(1) � 0.36, ns. Median odds ratios that do not share subscripts differsignificantly from one another at p � .05. Moderators in the left-handcolumn marked with the designated symbol have weighted mean log oddsratios that differed significantly from chance, as indicated by a z-test.Tau-square∧ for each level is as follows: u � .57 for institutional, and u �.69 for community.�� p .001.

Table 16Moderating Effect of Severity of Violence

Severity of violence kMedian odds


Severe�� 52 1.79a 0.64 to 3.54Moderate�� 72 1.86a 0.88 to 2.97Unspecified/other�� 67 1.45a 0.73 to 2.83

Note. k � number of effect sizes in analysis; IQR � interquartile range;Q1 � first quartile (25th percentile); Q3 � third quartile (75th percentile).Overall QB(2) � 0.79, ns. Median odds ratios that do not share subscriptsdiffer significantly from one another at p � .05. Moderators in the left-hand column marked with the designated symbol have weighted mean logodds ratios that differed significantly from chance, as indicated by a z-test.Tau-square∧ for each level is as follows: u � .96 for severe; u � .54 formoderate; and u � .79 for unspecified/other.�� p .001.

Table 17Moderating Effect of Violence Comparison Group

Comparison group kMedian odds


Nonviolent antisocial incident 42 1.07a 0.64 to 1.88No incident�� 119 1.92b 0.80 to 3.15

Note. k � number of effect sizes; IQR � interquartile range; Q1 � firstquartile (25th percentile); Q3 � third quartile (75th percentile). OverallQB(1) � 8.13, p � .0043. Median odds ratios that do not share subscriptsdiffer significantly from one another at p � .05. Moderators in the left-hand column marked with the designated symbol have weighted mean logodds ratios that differed significantly from chance, as indicated by a z-test.Tau-square∧ for each level is as follows: u � .55 for nonviolent antisocialincident, and u � .72 for no incident.�� p .001.

Table 15Moderating Effect of Method of Detecting Violence

Method of detecting violence k

Medianoddsratio

IQR(Q1 to Q3)

Official records (including observation)�� 114 1.66a 0.79 to 3.09Self-reports (including interview)�� 38 2.02a 1.25 to 3.02Mixed/not specified 18 0.80a 0.60 to 1.78

Note. k � number of effect sizes; IQR � interquartile range; Q1 � firstquartile (25th percentile); Q3 � third quartile (75th percentile). OverallQB(2) � 4.18, ns. Median odds ratios that do not share subscripts differsignificantly from one another at p � .05. Moderators in the left-handcolumn marked with the designated symbol have weighted mean log oddsratios that differed significantly from chance, as indicated by a z-test.Tau-square∧ for each level is as follows: u � .66 for official records(including observation); u � .72 for self-reports (including interview); andu � .50 for mixed/not specified.�� p .0001.


summarized the effect size of a violence risk factor across multiplesettings. Some meta-analyses have examined a range of risk fac-tors in a specific setting (e.g., mentally disordered offenders, as perBonta et al., 1998) or a specific risk factor in a few settings (e.g.,psychopathy in correctional and forensic psychiatric settings, asper Guy, Edens, Anthony, & Douglas, 2005; Hemphill, Hare, &Wong, 1998; Salekin, Rogers, & Sewell, 1996).

With this caveat in mind, it appears that the average effect sizefor psychosis in the present meta-analysis (r � .12–.16, as derivedfrom transformation from OR to r) is comparable to numerousindividual risk factors studied in the Bonta et al. (1998) meta-analysis (most of which were estimated from fewer than 10 stud-ies). For example, Bonta et al. obtained mean effect sizes (Zr) forthe following predictors: history of violence (.16), marital status(.13), weapon use (.12), nonviolent criminal history (.13), institu-tional adjustment (.14), and hospital admissions (.17). Antisocialpersonality (.18) and juvenile delinquency (.20) were only slightlymore strongly associated with violence in Bonta et al. than waspsychosis in the current study. Several risk factors had smallereffect sizes in Bonta et al. than psychosis had in the present study:substance use (.08), offense seriousness (.06), mood disorder (.01),violent index offense (�.04), race (.09), intelligence (�.02), andeducation (�.02).

The effect size for psychosis appears meaningfully smaller thanthat for psychopathy according to several meta-analyses that re-ported mean correlations typically in the .25�.30 range (Hemphillet al., 1998; Salekin et al., 1996; Walters, 2003). Some meta-analyses (Guy et al., 2005) have reported smaller effect sizes (rs �.11–.16) between psychopathy and violence in particular settings,such as within prisons.

In general, then, psychosis appears to be intermediate in com-parison with other putative risk factors. It is comparable to, or

larger than, numerous other risk factors, though tends to be smallerthan externalizing disorders, such as antisocial personality or psy-chopathy, or early-onset criminal behavior. Our finding that psy-chosis is associated with violence is consistent with that of previ-ous narrative reviews (e.g., Bradford, 1983; De Pauw & Szulecka,1988; Junginger, 1996; McNiel, 1994; Monahan, 1992b; Mulvey,1994; Taylor, 1995; Taylor et al., 1994; Wessely, 1993). However,some readers may be surprised by our finding in light of themeta-analysis of factors associated with recidivism among men-tally disordered offenders described above (Bonta et al., 1998).Bonta et al. reported that the association between psychosis andviolence was heterogeneous across studies, with the overall asso-ciation being small and negative in direction (Zr � �.04). How-ever, it is important to recognize that Bonta et al.’s review waslimited to a small number of effect sizes dealing with psychosisand violence (k � 11) from a specific type of study (longitudinalstudies of mentally disordered offenders released from correctionalinstitutions).

The Importance of Moderators

Perhaps the most important finding to emerge from the presentmeta-analysis is the extent to which the strength of associationbetween psychosis and violence differed as a function of moder-ator variables. Indeed, if one generalized from a certain group ofstudies, the conclusion would be that psychosis is inversely relatedto violence. With another set of studies, the conclusion would bethat psychosis is a very strong correlate of violence. To someextent, this systematic variation in strength of association, depictedclearly in Figure 2, should provide guidance for researchers interms of framing hypotheses about the anticipated relationshipbetween psychosis and violence in any given study, depending on

Figure 2. Effect of number of significant moderators on size of log odds ratio.


the characteristics of the study (i.e., nature of comparison groups,level of measurement).

First, several potential moderators were not significant. As such,a reasonable inference is that the psychosis–violence relationshipis relatively robust across levels of these moderators, includinggender, country, and severity of violence.

Design-related moderators. Of the study design or samplingvariables, both sample type and sampling procedure moderated thepsychosis–violence relationship. In particular, studies with com-munity samples and studies using epidemiological designs pro-duced larger effect sizes than did studies conducted in othersettings and those using other designs. However, psychosis wasstill a significant predictor of violence within correctional and civilpsychiatric settings and in studies where random sampling wasused. Study design was not a significant moderator, with bothretrospective and prospective designs producing significant andcomparable effects.

Another one of the stronger moderators came from the studydesign category of potential moderators: the nature of the compar-ison group. That is, to whom are people with psychosis beingcompared? If the comparison was to individuals with other—nonpsychotic—mental illnesses, the odds of violence were slightlyhigher for persons with psychosis (OR � 1.51). However, whenthe comparison group consisted of individuals who had not beenidentified as having mental disorders, the odds ratio was substan-tially higher, and large (OR � 3.68). Splitting up “other mentalillness” further, we found that comparing psychosis with internal-izing disorders produced a meaningful effect size (OR � 2.15),whereas comparing it with externalizing disorders did not (OR �0.85).

Although we were not able to conduct exactly the same type ofanalyses with substance use disorders as we could for externalizingdisorders more broadly, our findings indicated that comorbid psy-chosis and substance-related diagnoses produced substantiallylarger effect sizes than did psychosis alone (d � .97). This obser-vation is consistent with leading studies (Elbogen & Johnson,2009; Monahan et al., 2001) pointing to the crucial role of sub-stance use problems in understanding risk for violence.

There are three important points to consider with respect to thecomparison group issue. First, these analyses shed some light onthe relative ranking of categories of mental illness vis-a-vis theirrelationship to violence. Notably, externalizing disorders are likelymore strongly related to violence than are psychotic disorders. Thisassertion is consistent with several meta-analyses showing meancorrelational effect sizes between psychopathy and violence in the.25 to .30 range (i.e., Hemphill et al., 1998; Salekin et al., 1996;Walters, 2003), although there are exceptions in which smallereffect sizes have been reported (Guy et al., 2005).

Second, the findings suggest that, compared with persons with-out mental illness (or, presumably, any externalizing disorders),psychosis is, in fact, a strong risk factor for violence, increasing theodds of violence three- to fourfold. This issue of comparison grouphas important implications for the role that psychosis might play inviolence risk assessment. It is tempting to interpret these findingssuch that psychosis should be considered a noncorrelate in samplesconsisting of many highly externalizing people. This is true, how-ever, only under two conditions: (a) that only group-level behavioris of interest and (b) that there is no possibility that anotherreference group would be of relevance for decisions.

Concerning the first point, we note that at an individual assess-ment level, decisions must be made about a specific person, notabout groups of people. The presence of some comparison groupis of only peripheral relevance to a given individual’s risk. Theevidence from this meta-analysis suggests that psychosis be con-sidered a possible risk factor at an individual level whenever it ispresent. For instance, upon release from a correctional institution,mutually exclusive groups of persons with psychosis may be atlower risk than groups of persons with primary externalizingdiagnoses. However, when released to the general population(largely comprising people without any externalizing disorders),individuals with psychosis may be considered to be at higher riskrelative to the majority of people in the population. Viewed in thisway, individuals with either psychosis or externalizing disorderswould be at elevated risk compared with their nondisorderedcommunity counterparts. Those with externalizing disorderswould just be at higher risk than those with psychotic disorders.

Second, if analysis is at the level of the sample or the popula-tion, then the composition of the reference group clearly matters interms of hypothesis generation. Researchers should expect effectsizes that range from negative (comparison with personality dis-order) to small (comparison with nonpsychotic mental disorderbroadly) to large (comparison with internalizing disorders or nodisorder at all), depending on the nature of the comparison group.This could be an important issue in the context of a particularstudy. For instance, if a researcher were interested in predictingviolence within a prison therapeutic community setting in whichmost participants had antisocial personality disorder, it would bereasonable to predict that psychosis would have a minimal predic-tive effect. If, however, a researcher were predicting violencewithin a community residential setting consisting of numerousindividuals with internalizing disorders, then psychosis might beexpected to play a more important predictive role.

Psychosis-related moderators. Turning to the measurement ofpsychosis as a potential moderator, we observe several noteworthyfindings. The psychosis–violence association was significant whenpsychosis was measured as a diagnosis of schizophrenia or mea-sured at the level of the symptom. It was not significant foraffective psychoses or when researchers either failed to specifyhow they measured it or aggregated across all types of psychosis.These findings argue against the practice in research of lumpingindividuals with psychoses into a single, gross category, a pointthat has been made cogently in the past (e.g., Monahan & Stead-man, 1994). At the level of the symptom, positive symptoms weresignificantly more strongly related to violence than were negativesymptoms, providing some support for causal models that empha-size these types of symptoms (i.e., “psychotic action”; Junginger,1996). However, we also note that symptoms of disorganization(and other/unspecified) symptoms predicted violence. It is impos-sible to comment on the other/unspecified category in that it ispossible that it included positive symptoms. The positive findingfor symptoms of disorganization would suggest that researchersshould investigate whether such symptoms might play a role inviolence, perhaps by disrupting executive functioning, by compro-mising the ability to remain composed, or by leading to tense orstressful interpersonal encounters (see Hiday, 1997, 2006). Wealso note that most studies did not report the effects of differenttypes of symptoms while controlling for the others, which wouldbe an important step in future research.


Much has been made in the psychosis–violence literature aboutthe role of different types of positive symptoms: hallucinations,delusions, paranoid symptoms, threat/control-override symptoms,and so forth. Our findings revealed few differences among these interms of their relation to violence: Hallucinations and delusions,threat/control-override symptoms, and other positive symptomswere all associated with violence, and symptoms with some sort ofparanoid element approached significance. Whereas some paststudies (e.g., Link, Andrews, & Cullen, 1992; Link & Stueve,1994) have reported that threat/control-override symptoms predictviolence more strongly than other symptoms, other studies havefailed to replicate this. For example, Appelbaum et al. (2000)provided a detailed analysis of the association between threat/control-override symptoms and violence and reported that theassociation did not hold up when patients’ reports of putativelydelusional experiences were confirmed for true delusional status,although subthreshold (non-delusional) perceptions similar in na-ture to threat/control-override delusions were associated with pastviolence. Few studies in this meta-analysis provided such a de-tailed level of analysis. As such, our findings cannot address suchfine, though important, distinctions. However, we argue that it isunlikely that a median odds ratio of 2.32 for positive symptoms isgenerally artifactual across five dozen samples. The precise con-tours and boundaries of the symptom–violence association willneed to be addressed in further primary studies, drawing on inves-tigations such as Appelbaum et al. (2000) and models ofpsychosis–violence such as those proposed by Hiday (1997); Jung-inger (1996), or Taylor and colleagues (Taylor, 1998, 2006; Taylor& Hodgins, 1994).

Finally, with respect to the timing of the psychosis determina-tion vis-a-vis the occurrence of violence, the largest effect was forthe small number of studies (k � 10) where diagnosis and violenceoccurred at approximately the same time. This potential moderatoris muddied by the fact that 36 studies failed to specify the timingof diagnosis vis-a-vis violence. In theory, if psychosis is a causalrisk factor, then its presence must precede the outcome of violence.However, when psychosis was assessed prior to the measurementof violence, it produced a small and nonsignificant odds ratio of1.30. Before concluding that this means that psychosis cannot,therefore, be considered a causal risk factor for violence, we pointout two very important limits to this general timing variable.

First, in addition to information about whether the diagnosis ofpsychosis occurred before or after violence, it is important to knowhow close in time the two occurred, particularly if it is activesymptoms that are most crucial in the psychosis–violence associ-ation. Although for most studies we were able to code whetherdiagnosis preceded or followed the occurrence of violence, wewere not able to systematically determine how far apart diagnosisand violence were. That is, an assessment of psychosis in, say,1990, may have little relevance for a person’s behavior in 1997,being separated by 7 years. This idea is consistent with calls in theliterature to focus on the dynamic, changing nature of risk factorsand on the consideration that certain risk factors that are prone tochange, such as psychotic symptoms, may only be of relevance toviolence when they are in some active or elevated state (seeDouglas & Skeem, 2005; Skeem & Mulvey, 2002). As such, theincreased odds ratio associated with assessments of psychosis andviolence that occurred close in time is meaningful. However, we

encourage researchers to always report the time between diagnosisand violence.

Second, it is artificial to presume that studies capture, in anysystematic way, all of the violence that a person has perpetratedand either the first break of psychosis or the complex waxing andwaning of psychotic symptoms across time. As such, most studiesrepresent some snapshot of the timing of psychosis and violencevis-a-vis one another, a view that could change if the snapshotwere taken at a different point in a person’s life or included a largertime frame. As such, one cannot presume that if psychosis werediagnosed after violence had occurred (say, for instance, thatpsychosis were diagnosed in an individual in a psychiatric hospitaland that it was clear from arrest records or the person’s self-reportthat she or he had been violent in the past), psychosis did not existprior to the violence. Rather, the only conclusions that could bemade under such circumstances would be either that (a) psychosiswas not diagnosed or measured at that time or (b) studies failed toreport or measure age of first onset of psychosis. Similarly, ifpsychosis were diagnosed today and a person was not detected ashaving been violent in a 6-month follow-up period, it would notmean that psychosis and violence had not co-occurred for thatindividual at some point or that these variables would not co-occurat some point in the future. In such a case, it could only be said thatviolence was not observed to occur in the 6 months after psychosiswas measured.

Studies on mental illness and violence rarely document eitheronset or course of illness (waxing and waning of symptoms) orviolence across a person’s life, or whether these are systematicallyrelated to one another. Such a design is exactly what is needed,however, to make firm conclusions about exactly how psychosisand violence may or may not be related. On the basis of thepreceding issues, we argue that because (a) symptom-level mea-surement produced the largest odds and (b) close timing betweenassessment of psychosis and measurement of violence producedthe largest odds, there is some support for the view that risk forviolence is particularly elevated when the course of psychosis isactive.

It is also important to note that the correlation between psycho-sis and violence could, in part, stem from other forms of possiblerelationships between them. That is, a person may become psy-chotic in the stressful context of being violent or in a transactionthat involves both violence and victimization. In such a scenario,it would not be appropriate to describe psychosis as a cause of thatparticular violent act. A complicating factor to such a scenario,however, is that it is unlikely that psychosis would develop inresponse to violence-related stress without there being an under-lying psychotic disorder present, such as schizophrenia. In such acase, then, a psychotic disorder precedes the violence, which inturn leads to decompensation involving onset of florid symptoms.Moreover, it is unlikely that acting violently would cause a disor-der itself, such as schizophrenia, but more likely that it couldexacerbate some of the symptoms of the disorder. We were notable to meta-analyze the literature at such a fine level, although weagain call for researchers to clearly indicate specific details aboutthe timing and onset not only of active symptoms but also of thecourse of the disorder more generally, as well as how thesevariables relate to the occurrence of violent behavior. It is certainlypossible that for some individuals, or for the same person atdifferent times, violence may cause psychotic decompensation. At


other times or for other people, psychosis may cause violence. Thetwo possibilities are not mutually exclusive.

Violence-related moderators. In terms of violence-relatedmoderators, strongest support was found for the outcome compar-ison group. That is, psychosis was not predictive of violence whenthe outcome comparison was other antisocial (though nonviolent)behavior, as opposed to no known antisocial behavior. Given therelatively high correlation between violent and nonviolent antiso-cial behavior, however, we expect that this effect would be ob-served for other risk factors as well. Results also indicated that thepsychosis–violence relationship was not moderated by the severityof violence, with severe and moderate violence producing compa-rable odds ratios. This finding should indicate that, to the extentthat psychosis is a risk factor for violence, it is not limited to trivialacts of violence.

The method for assessing violence (i.e., reliance on single vs.multiple methods) was not a significant moderator. Studies thatrelied on official records, observation, or self/other reports pro-duced significant effects. Studies that incorporated multiple meth-ods, or failed to specify the measurement approach for violence,did not. There were relatively few (k � 18) of the latter type ofstudy. This finding was not expected, given that multiple methodsof measurement typically capture a greater amount of violence(Steadman et al., 1998). However, it may be the case that theofficial records and observation methods in the current studieswere confounded by other variables (such as retrospective design)or were based on cases with known outcomes (i.e., all violentoffenders in a prison sample were “detected” to be violent accord-ing to official records). As such, the value of multiple methods ofmeasurement is likely greater in prospective studies in whichparticipants are tracked in the community after release from aninstitution, and there is a very real risk of failing to detect violenceby sole reliance on single measurement procedures, such as formalrecidivism records.

Implications for Future Research

Posing the question, “Are individuals with psychosis morelikely to be violent than individuals without psychosis?” is sort oflike asking whether 10-year olds are tall. Compared with toddlers,they certainly are. Compared with adults, they are decidedly short.And so it is with psychosis: Compared with individuals with nomental disorders, people with psychosis seem to be at a substan-tially elevated risk for violence. Compared with individuals withexternalizing psychopathology, psychosis does not appear to fur-ther elevate the risk for violence. Whereas we did find a statisti-cally significant, though small, overall relationship between psy-chosis and violence, we argue that the general question should beabandoned in favor of the more complex and sophisticated ques-tion, “What particular symptoms of psychosis, under which situ-ational circumstances, and in combination with which personal orsituational factors, are associated with increased or decreased riskof various kinds of violence?”

With respect to psychosis, future research should focus onspecific diagnostic groups (formed according to standardized di-agnostic criteria) and on specific clusters of psychotic symptoms(assessed with standardized measures). It would be helpful if, inthe future, researchers provided more detail concerning their as-sessments of psychosis, including analyses of the prevalence and

interrater reliability of, as well as the risk associated with, indi-vidual symptoms of psychosis. In addition, future research shouldcontrol for the effects of comorbid mental disorders within bothpsychotic and nonpsychotic samples statistically, by selection, orby matching. Such control would enhance the precision of theestimate of risk for violence associated with psychosis and alsomay identify important cofactors, such as substance use or person-ality disorder. Finally, future research should examine which psy-chotic disorders or symptoms are active versus in remission at thetime of violence, and, if possible, track both the onset and courseof psychosis and violence. It may be that the association betweenpsychosis and violence is strongest when the symptoms are active(Krakowski et al., 1986; Monahan, 1988, 1992b; Monahan &Steadman, 1994), a view partially supported by the present anal-ysis. The general point here is that better measurement of psycho-sis should allow strong tests of competing causal models of itsassociation with violence.

Researchers also need to pay more attention to the measurement,conceptualization, and context of violence. Although relyingsolely on records or self-reports is sometimes necessary, it oftenprecludes measurement of other features of violence, such asmotivation, targets, or location. It is probable that some types ofviolence (predatory, instrumental violence against strangers) arenot frequently committed by persons with psychosis but that othertypes may be (reactive violence against family members or cohab-itants following stressful or tense encounters). Indeed, research hasidentified that family members are commonly at risk of acts ofviolence by persons with psychosis (Estroff et al., 1998).

Longitudinal research would be well-suited to identifying thepossible transactions between elements of psychosis and violenceacross time, an area that little research has been able to address. Itwould be particularly interesting to track the first onset of psycho-sis and violence. In addition, additional studies of the possiblemediating role of third variables are necessary. Psychosis might beassociated with violence because it is related to other risk factors(Elbogen & Johnson, 2009), such as unemployment or housingproblems. From one perspective, this would reduce psychosis to anuisance variable if these other variables were considered caus-ative of both psychosis and violence. However, if mental illnessitself plays a causal role in housing problems or unemployment(Dembling et al., 2002; Hudson, 2005), then it is not clear, theo-retically, that these variables should be controlled for. Rather, itcould be the case that violence is but one of a host of adverseoutcomes associated with mental illness, many of which are asso-ciated with basic problems in living (employment, housing, rela-tionships, education, health). We expect that many of these vari-ables could influence onset of illness as well as transact withcourse of illness and violence over time. A simple A3B concep-tualization of the issue is unlikely to capture it adequately.

Implications for Clinical Practice: Evaluating Riskof Violence

The findings reported herein have a number of implications forviolence risk assessment. First, psychosis should be evaluated inall violence risk assessments. At first, this seems inconsistent withfindings that psychosis was not associated with violence undercertain conditions (e.g., comparison with externalizing psychopa-thology, within forensic psychiatric samples). However, we base


this argument on two premises: (a) There was an overall associa-tion between psychosis and violence, and this was particularlystrong when the comparison group consisted of individuals with-out mental disorder, and (b) as we outlined above, all of ourfindings, and those of the original studies, are based on groups.Further, some studies (Elbogen & Johnson, 2009) have reportedthat close to half of persons with MMI have a co-occurringsubstance-related disorder, and comorbidity elevates the risk ofviolence substantially (see also Monahan et al., 2001). As such, itis important to know whether a person with either an MMI or asubstance-related disorder has a co-occurring disorder.

Clinical assessment of risk for violence inherently is aboutmaking individual decisions. In our view, the evidence from thecurrent meta-analysis is sufficient to conclude that psychosis mayelevate a person’s risk compared with the case in which psychosisis absent in that individual. At the very least, the current meta-analysis supports treating the violence relevance of psychosis in anindividual as a hypothesis to be disconfirmed through an assess-ment of its relevance to a person’s history of violence and likeli-hood for future violence. In essence, evaluators should be askingthemselves what potential role psychosis might play vis-a-visviolence for the given individual whom they are evaluating.

Further, the issue of “compared with what or whom?” can befactored into an individual risk assessment. That is, an evaluatormight conclude that risk is lower for his or her patient withpsychosis compared with persons with externalizing disorders butthat it is still higher than it would be if psychosis were absent. Theevaluator might also conclude, depending on the context in whichthe patient might be residing in the future, that the patient likely isat lower risk for violence than is his or her externalizing counter-parts but that he or she is at higher risk than individuals withoutany mental disorders. Note that this does not mean that cliniciansneed to perform exhaustive evaluations of psychosis in every case;rather, a brief screen for psychosis could be conducted, with amore comprehensive assessment limited to cases in which theresults of the screen are positive.

The second implication is that the presence of psychosis shouldbe considered neither necessary nor sufficient for a determinationof high risk. This conclusion is based on the finding that the overallmagnitude of the psychosis–violence association was found to besmall; most violent individuals are not psychotic, and most psy-chotic individuals are not violent. This means that clinicians con-ducting a risk assessment should consider the role that psychosismay have played in past violence (according to the individual’shistory) and could play in future violence (e.g., as an organizing ora disorganizing factor). Only in cases in which the psychosis seemslogically linked to the occurrence of future violence should clini-cians consider it an important risk factor. Of course, other cofac-tors may also be present that increase (e.g., substance abuse,psychopathy) or decrease (e.g., social support, stable housing) theindividual’s risk for violence. As such, psychosis should never bethe sole factor that is considered in determining a person’s risk– orlack thereof—for violence.

Limitations

First, although we were able to evaluate some interesting andimportant issues, such as the composition of the comparison group,types of symptoms, and the impact of comorbidity, many questions

were out of reach. For example, it was not possible to test partic-ular theories about why psychosis might elevate the odds forviolence. Such theories come in various forms and are crucial formoving the field forward. Relatively focused theoretical modelswere developed specifically to explain the relationship betweenpsychosis (or MMI) and violence. Hiday (1997, 2006), for exam-ple, posited that the features of major mental illness (odd behavior,difficulty handling stimulation) can lead to tense interpersonalsituations in which the probability of violence is increased. Jung-inger (1996) developed a theoretical framework to explain whenpsychosis might lead to violence. In particular, violence in theform of “psychotic action” is more likely when the themes andcontent of the delusions and hallucinations are aggressively ladenand, hence, consistent with acting violently.

Other theoretical explanations of the link between psychosis andviolence draw from classic psychological theories, such as psy-chodynamic theory (Mizen, 2003). For instance, some scholarshave tested the role that defense mechanisms such as projectionmight play in violence (Porcerelli, Cogan, Kamoo, & Leitman,2003). Such theoretical approaches—narrow or broad—have greatpotential merit in furthering understanding of any link betweenpsychosis and violence. For example, in principle, it should bepossible to measure whether a particular psychotic symptom, suchas paranoid delusions, was acting as a defense mechanism for anindividual. However, the existing empirical literature on psychosisand violence did not permit testing such theoretical models in thismeta-analysis, because most studies do not actually measurewhether such psychotic symptoms are indeed manifestations ofthese (or other) theoretical postulates. Similarly, very few of theincluded studies measured the themes of symptoms and whethersuch themes were consistent or inconsistent with violence. Hadthey done so, we could have tested Junginger’s (1996) model ofpsychotic action. There are simply too few empirical studies thatincorporate measurement of psychosis and violence that also in-clude tests of these theories. Again, we encourage such steps infuture research.

Moreover, it was not possible to determine the appropriate roleof covariates, such as SES or employment (i.e., are such variablescauses or correlates of mental illness?), or to evaluate the impor-tance of timing between psychosis and violence across the lifespan. These specific and complicated aspects of the psychosis–violence association are best suited for future in-depth, individualstudies.

Finally, any meta-analysis is limited by the nature and quality ofthe research on which it is based. In our view, the researchliterature on psychosis and violence is sufficiently large to permitgeneral conclusions about the overall association, as well as aboutsome of the specific aspects of the association. However, we alsonote—echoing the comments of others (e.g., Krakowski et al.,1986)—that researchers need to include more detail in their pub-lished findings that would facilitate reviews, whether narrative orquantitative. That is, for some moderators, there were a largenumber of studies in which important methodological featureswere not specified. Despite this, we were able to code enoughstudies in sufficient detail to permit a meaningful evaluation of theoverall psychosis–violence relationship and the impact of putativemoderators.


References

References marked with an asterisk were included in the meta-analysis.

*Aarsland, D., Cummings, J. L., Yenner, G., & Miller, B. (1996). Rela-tionship of aggressive behavior to other neuropsychiatric symptoms inpatients with Alzheimer’s disease. American Journal of Psychiatry, 153,243–247.

*Abu-Akel, A., & Abushua’leh, K. (2004). ‘Theory of mind’ in violent andnonviolent patients with paranoid schizophrenia. Schizophrenia Re-search, 69, 45–53.

*Abushua’leh, K., & Abu-Akel, A. (2006). Association of psychopathictraits and symptomatology with violence in patients with schizophrenia.Psychiatry Research, 143, 205–211.

Alexander, R. T., Crouch, K., Halstead, S., & Piachaud, J. (2006). Long-term outcome from a medium secure service for people with intellectualdisability. Journal of Intellectual Disability Research, 50, 305–315.

Anderson, D. A. (1999). The aggregate burden of crime. Journal of Lawand Economics, 42, 611–642.

*Appelbaum, P. S., Robbins, P. C., & Monahan, J. (2000). Violence anddelusions: Data from the MacArthur Violence Risk Assessment Study.American Journal of Psychiatry, 157, 566–572.

*Arango, C., Bombın, I., Gonzalez-Salvador, T., Garcıa-Cabeza, I., &Bobes, J. (2005). Randomised clinical trial comparing oral versus depotformulations of zuclopenthixol in patients with schizophrenia and pre-vious violence. European Psychiatry, 21, 34–40.

*Arango, C., Calcedo Barba, A., Gonzalez-Salvador, T., & CalcedoOrdonez, A. (1999). Violence in inpatients with schizophrenia: A pro-spective study. Schizophrenia Bulletin, 25, 493–503.

*Arboleda-Florez, J., Love, E. J., Fick, G., O’Brien, K., Hashman, K., &Aderibigbe, Y. (1995). An epidemiological study of mental illness in aremanded population. International Medical Journal, 2, 113–126.

*Arseneault, L., Cannon, M., Murray, R., Poulton, R., Caspi, A., & Moffitt,T. E. (2003). Childhood origins of violent behaviour in adults withschizophreniform disorder. British Journal of Psychiatry, 183, 520–525.

*Arseneault, L., Moffitt, T. E., Caspi, A., & Taylor, A. (2002). The targetsof violence committed by young offenders with alcohol dependence,marijuana dependence and schizophrenia-spectrum disorders: Findingsfrom a birth cohort. Criminal Behaviour and Mental Health, 12, 155–168.

*Arseneault, L., Moffitt, T. E., Taylor, P. J., & Silva, P. A. (2000). Mentaldisorders and violence in a total birth cohort: Results from the Dunedinstudy. Archives of General Psychiatry, 57, 979–986.

*Asnis, G. M., Kaplan, M. L., van Praag, H. M., & Sanderson, W. C.(1994). Homicidal behaviors among psychiatric outpatients. Hospitaland Community Psychiatry, 45, 127–132.

*Barkataki, I., Kumari, V., Das, M., Hill, M., Morris, R., O’Connell, P., etal. (2005). A neuropsychological investigation into violence and mentalillness. Schizophrenia Research, 74, 1–13.

Barlow, K., Grenyer, B., & Ilkiw-Lavallle, O. (2000). Prevalence andprecipitants of aggression in psychiatric inpatients. Australian and NewZealand Journal of Psychiatry, 34, 967–974.

Barnard, G. W., Robbins, L., Newman, G., & Carrera, F. (1984). A studyof violence within a forensic treatment facility. Bulletin of the AmericanAcademy of Psychiatry and Law, 12, 339–348.

*Bassiony, M. M., Steinberg, M. S., Warren, A., Rosenblatt, A., Baker,A. S., & Lyketsos, C. G. (2000). Delusions and hallucinations in Alz-heimer’s disease: Prevalence and clinical correlates. International Jour-nal of Geriatric Psychiatry, 15, 99–107.

Baxter, R. (1997). Violence in schizophrenia and the syndrome of disor-ganisation. Criminal Behaviour and Mental Health, 7, 131–139.

*Beaudoin, M. N., Hodgins, S., & Lavoie, F. (1993). Homicide, schizo-phrenia and substance abuse or dependency. Canadian Journal of Psy-chiatry, 38, 541–546.

*Bihm, E. M., Poindexter, A. R., & Warren, E. R. (1998). Aggression and

psychopathology in persons with severe or profound mental retardation.Research in Developmental Disabilities, 19, 423–438.

*Binder, R. L., & McNiel, D. E. (1986). Victims and families of violentpsychiatric patients. Bulletin of the American Academy of Psychiatry andthe Law, 14, 131–139.

*Bjorkdahl, A., Olsson, D., & Palmstierna, T. (2006). Nurses’ short-termprediction of violence in acute psychiatric intensive care. Acta Psychi-atrica Scandinavica, 113, 224–229.

Bjørkly, S. (2002a). Psychotic symptoms and violence toward others—Aliterature review of some preliminary findings: Part 1. Delusions. Ag-gression and Violent Behavior, 7, 617–631.

Bjørkly, S. (2002b). Psychotic symptoms and violence toward others—Aliterature review of some preliminary findings: Part 2. Hallucinations.Aggression and Violent Behavior, 7, 605–615.

Bloom, J. D. (1989). The character of danger in psychiatric practice: Arethe mentally ill dangerous? Bulletin of the American Academy of Psy-chiatry and the Law, 17, 241–255.

Bonta, J., Law, M., & Hanson, R. K. (1998). The prediction of criminal andviolent recidivism among mentally disordered offenders: A meta-analysis. Psychological Bulletin, 123, 123–142.

Bradford, J. M. W. (1983). The forensic psychiatric aspects of schizophre-nia. Psychiatric Journal of the University of Ottawa, 8, 96–103.

*Brennan, P. A., Mednick, S. A., & Hodgins, S. (2000). Major mentaldisorders and criminal violence in a Danish birth cohort. Archives ofGeneral Psychiatry, 57, 494–500.

*Brownstone, D. Y., & Swaminath, R. S. (1989). Violent behaviour andpsychiatric diagnosis in female offenders. Canadian Journal of Psychi-atry, 34, 190–194.

*Buckley, P. F., Hrouda, D. R., Friedman, L., Noffsinger, S. G., Resnick,P. J., & Camlin-Shingler, K. (2004). Insight and its relationship toviolent behavior in patients with schizophrenia. American Journal ofPsychiatry, 161, 1712–1714.

Butcher, J. N., Dahlstrom, W. G., Graham, J. R., Tellegen, A., & Kaem-mer, B. (1989). Manual for administration and scoring MMPI-2. Min-neapolis: University of Minnesota Press.

*Calcedo-Barba, A. L., & Calcedo Ordonez, A. (1994). Violence andparanoid schizophrenia. International Journal of Law and Psychiatry,17, 253–263.

*Carrithers, V. C. (1994). A regression/correctional analysis study ofviolence, six demographics, and three mental illness symptoms. Unpub-lished doctoral dissertation, University of Louisville.

*Cheung, P., Schweitzer, I., Crowley, K., & Tuckwell, V. (1997). Aggres-sive behavior in schizophrenia: The role of psychopathology. Australianand New Zealand Journal of Psychiatry, 31, 62–67.

*Chuang, H. T., Williams, R., & Dalby, J. T. (1987). Criminal behaviouramong schizophrenics. Canadian Journal of Psychiatry, 32, 255–258.

*Cirincione, C., Steadman, H. J., Robbins, P. C., & Monahan, J. (1992).Schizophrenia as a contingent risk factor for criminal violence. Interna-tional Journal of Law and Psychiatry, 15, 347–358.

*Clare, P., Bailey, S., & Clark, A. (2000). Relationship between psychoticdisorders in adolescence and criminally violent behaviour: A retrospec-tive examination. British Journal of Psychiatry, 177, 275–279.

*Coid, J., Yang, M., Roberts, A., Ullrich, S., Moran, P., Bebbington, P., etal. (2006). Violence and psychiatric morbidity in the national householdpopulation of Britain: Public health implications. British Journal ofPsychiatry, 189, 12–19.

*Convit, A., Isay, D., Otis, D., & Volavka, J. (1990). Characteristics ofrepeatedly assaultive psychiatric inpatients. Hospital and CommunityPsychiatry, 41, 1112–1115.

Cooper, S. J., Browne, F. W., McClean, K. J., & King, D. J. (1983).Aggressive behaviour in a psychiatric observation ward. Acta Psychiat-rica Scandinavica, 68, 386–393.

Corrigan, P. W., & O’Shaughnessy, J. R. (2007). Changing mental illnessstigma as it exists in the real world. Australian Psychologist, 42, 90–97.


Corrigan, P. W., & Penn, D. L. (1999). Lessons from social psychology ondiscrediting psychiatric stigma. American Psychologist, 54, 765–776.

*Cote, G., Lesage, A., Chawky, N., & Loyer, M. (1997). Clinical speci-ficity of prison inmates with severe mental disorders: A case-controlstudy. British Journal of Psychiatry, 170, 571–577.

*Craig, T. J. (1982). An epidemiologic study of problems associated withviolence among psychiatric inpatients. American Journal of Psychiatry,139, 1262–1266.

*Crocker, A. G., Mueser, K. T., Drake, R. E., Clark, R. E., McHugo, G. J.,Ackerson, T. H., et al. (2005). Antisocial personality, psychopathy, andviolence in persons with dual disorders: A longitudinal analysis. Crim-inal Justice and Behavior, 32, 452–476.

*Daffern, M., Howells, K., Ogloff, J. R. P., & Lee, J. (2005). Individualcharacteristics predisposing patients to aggression in a forensic psychi-atric hospital. Journal of Forensic Psychiatry & Psychology, 16, 729–746.

*Dalal, B., Larkin, E., Leese, M., & Taylor, P. J. (1999). Clozapinetreatment of long-standing schizophrenia and serious violence: A two-year follow-up study of the first 50 patients treated with clozapine inRampton high security hospital. Criminal Behaviour and Mental Health,9, 168–178.

*Danielson, K. K., Moffitt, T. E., Caspi, A., & Silva, P. A. (1998).Comorbidity between abuse of an adult and DSM-III-R mental disor-ders: Evidence from an epidemiological study. American Journal ofPsychiatry, 155, 131–133.

De Pauw, K. W., & Szulecka, K. (1988). Dangerous delusions: Violenceand the misidentification syndromes. British Journal of Psychiatry, 152,91–96.

*Dean, K., Walsh, E., Moran, P., Tyrer, P., Creed, F., Byford, S., et al.(2006). Violence in women with psychosis in the community: A pro-spective study. British Journal of Psychiatry, 188, 264–270.

*Dell, S., Robertson, G., James, K., & Grounds, A. (1993). Remands andpsychiatric assessments in Holloway Prison. I: The psychotic popula-tion. British Journal of Psychiatry, 163, 634–640.

Dembling, B. P., Rovnyak, V., Mackey, S., & Blank, M. (2002). Effects ofgeographic migration on SMI prevalence estimates. Mental Health Ser-vices Research, 4, 7–12.

*Denman, N. (1996). Relationships among diagnoses and features ofviolence in adult males. Unpublished doctoral dissertation, University ofKansas.

Derogatis, L. R., & Melisaratos, N. (1983). The Brief Symptom Inventory:An introductory report. Psychological Medicine, 13, 595–605.

DerSimonian, R., & Laird, N. (1986). Meta-analysis in clinical trials.Controlled Clinical Trials, 7, 177–188.

*Desai, R. A., Lam, J., & Rosenheck, R. A. (2000). Childhood risk factorsfor criminal justice involvement in a sample of homeless people withserious mental illness. Journal of Nervous and Mental Disease, 188,324–332.

*Dhossche, D. M. (1999). Aggression and recent substance abuse: Absenceof association in psychiatric emergency room patients. ComprehensivePsychiatry, 40, 343–346.

*Dickerson, F., Ringel, N., Parente, F., & Boronow, J. (1994). Seclusionand restraint, assaultiveness, and patient performance in a token econ-omy. Hospital and Community Psychiatry, 45, 168–170.

Douglas, K. S., & Ogloff, J. R. P. (2003). Violence by psychiatric patients:The impact of archival measurement source on violence prevalence andrisk assessment accuracy. Canadian Journal of Psychiatry, 48, 734–740.

Douglas, K. S., & Skeem, J. L. (2005). Violence risk assessment: Gettingspecific about being dynamic. Psychology, Public Policy, and Law, 11,347–383.

*Doyal, M., Dolan, M., & McGovern, J. (2002). The validity of NorthAmerican risk assessment tools in predicting in-patient violent behav-iour in England. Legal and Criminological Psychology, 7, 141–154.

*Durbin, J. R., Pasewark, R. A., & Albers, D. (1977). Criminality and

mental illness: A study of arrest rates in a rural state. American Journalof Psychiatry, 134, 80–83.

Duval, S. D., & Tweedie, R. T. (2000). A nonparametric “trim and fill”method of accounting for publication bias in meta-analysis. Journal ofthe American Statistical Association, 95, 89–98.

*Edwards, J. G., Jones, D., Reid, W. H., & Chu, C. C. (1988). Physicalassaults in a psychiatric unit of a general hospital. American Journal ofPsychiatry, 145, 1568–1571.

*Ehmann, T. S., Smith, G. N., Yamamoto, A., McCarthy, N., Ross, D., Au,T., et al. (2001). Violence in treatment resistant psychotic inpatients. TheJournal of Nervous and Mental Disease, 189, 716–721.

Elbogen, E. B., & Johnson, S. C. (2009). The intricate link betweenviolence and mental disorder: Results from the national epidemiologicsurvey on alcohol and related condition. Archives of General Psychiatry,66, 152–161.

*Elbogen, E. B., Swanson, J. W., Swartz, M. S., & Van Dorn, R. (2005).Family representative payeeship and violence risk in severe mentalillness. Law and Human Behavior, 29, 563–574.

*Erb, M., Hodgins, S., Freese, R., Muller-Isberner, R., & Jockel, D. (2001).Homicide and schizophrenia: Maybe treatment does have a preventiveeffect. Criminal Behavior and Mental Health, 11, 6–26.

*Eronen, M. (1995). Mental disorders and homicidal behavior in femalesubjects. American Journal of Psychiatry, 152, 1216–1218.

*Eronen, M., Hakola, P., & Tiihonen, J. (1996a). Factors associated withhomicide recidivism in a 13-year sample of homicide offenders inFinland. Psychiatric Services, 47, 403–406.

*Eronen, M., Hakola, P., & Tiihonen, J. (1996b). Mental disorders andhomicidal behavior in Finland. Archives of General Psychiatry, 53,497–501.

*Eronen, M., Tiihonen, J., & Hakola, P. (1996c). Schizophrenia andhomicidal behavior. Schizophrenia Bulletin, 22, 83–89.

Estroff, S. E., Swanson, J. W., Lachicotte, W. S., Swartz, M., & Bolduc, M.(1998). Risk reconsidered: Targets of violence in the social networks ofpeople with serious psychiatric disorders. Social Psychiatry and Psychi-atric Epidemiology, 33, S95–S101.

*Estroff, S. E., & Zimmer, C. (1994). Social networks, social support, andviolence among persons with severe, persistent mental illness. In J.Monahan & H. J. Steadman (Eds.), Violence and mental disorder:Developments in risk assessment (pp. 259–295). Chicago: University ofChicago Press.

*Etherington, R. (1993). Diagnostic and personality differences of juvenilesex offenders, non–sex offenders and non-offenders. Unpublished doc-toral dissertation, California School of Professional Psychology, LosAngeles, Alliant University.

Fazel, S., & Grann, M. (2004). Psychiatric morbidity among homicideoffenders: A Swedish population study. American Journal of Psychiatry,161, 2129–2131.

*Fazel, S., & Grann, M. (2006). The population impact of severe mentalillness on violent crime. American Journal of Psychiatry, 163, 1397–1403.

*Fido, A. A., Razik, M. A., Mizra, I., & El-Islam, M. F. (1992). Psychiatricdisorders in prisoners referred for assessment: A preliminary study.Canadian Journal of Psychiatry, 37, 100–103.

First, M. B. (Ed.). (2003). Standardized evaluation in clinical practice.Washington, DC: American Psychiatric Association.

*Flannery, R. B., Fisher, W., Walker, A. P. M., Littlewood, K. B., &Spillane, M. J. (2001). Nonviolent psychiatric inpatients and subsequentassaults on community patients and staff. Psychiatric Quarterly, 72,19–27.

Fleiss, J. L. (1994). Measures of effect size for categorical data. In H.Cooper & L. V. Hedges (Eds.), The handbook of research synthesis (pp.245–260). New York: Russell Sage Foundation.

Fleiss, J. L., Williams, J. B. W., & Dubro. A. F. (1986). The logistic


regression analysis of psychiatric data. Journal of Psychiatric Research,20, 145–209.

*Fresan, A., Apiquian, R., de la Fuente-Sandoval, C., Loyzaga, C., Garcıa-Anaya, M., Meyenberg, N., et al. (2005). Violent behavior in schizo-phrenic patients: Relationship with clinical symptoms. Aggressive Be-havior, 31, 511–520.

*Fulwiler, C., Grossman, H., Forbes, C., & Ruthazer, R. (1997). Early-onset substance abuse and community violence by outpatients withchronic mental illness. Psychiatric Services, 48, 1181–1185.

*Gardner, W., Lidz, C. W., Mulvey, E. P., & Shaw, E. C. (1996). Acomparison of actuarial methods for identifying repetitively violentpatients with mental illnesses. Law and Human Behavior, 20, 35–48.

*Gosden, N. P., Kramp, P., Gabrielsen, G., Andersen, T. F., & Sestoft, D.(2005). Violence of young criminals predicts schizophrenia: A 9-yearregister-based followup of 15- to 19-year-old criminals. SchizophreniaBulletin, 31, 759–768.

Grann, M., Långstrom, N., Tengstrom, A., & Kullgren, G. (1999). Psy-chopathy (PCL-R) predicts violent recidivism among criminal offenderswith personality disorders in Sweden. Law and Human Behavior, 23,205–217.

Grassi, L., Peron, L., Marangoni, C., Zanchi, P., & Vanni, A. (2001).Characteristics of violent behavior in acute psychiatric in-patients: A5-year Italian study. Acta Psychiatrica Scandinavica, 104, 273–279.

*Greenberg, W. M., Shah, P. J., & Seide, M. (1993). Recidivism on anacute psychiatric forensic service. Hospital and Community Psychiatry,44, 583–585.

Guy, L. S., Edens, J. F., Anthony, C., & Douglas, K. S. (2005). Doespsychopathy predict institutional misconduct among adults? A meta-analytic investigation. Journal of Consulting and Clinical Psychology,73, 1056–1064.

Haddock, C. K., Rindskopf, D., & Shadish, W. R. (1998). Using odds ratiosas effect sizes for meta-analysis of dichotomous data: A primer onmethods and issues. Psychological Methods, 3, 339–353.

*Hafner, H., & Boker, W. (1982). Crimes of violence by mentally disor-dered offenders (H. Marshall, Trans.). Cambridge, United Kingdom:Cambridge University Press. (Original work published 1973)

*Harris, G. T., Rice, M. E., & Cormier, C. A. (2002). Prospective repli-cation of the Violence Risk Appraisal Guide in predicting violent recid-ivism among forensic patients. Law and Human Behavior, 26, 377–394.

*Harris, G. T., Rice, M. E., & Quinsey, V. L. (1993). Violent recidivism ofmentally disordered offenders: The development of a statistical predic-tion instrument. Criminal Justice and Behavior, 20, 315–335.

*Harris, G. T., & Varney, G. W. (1986). A ten-year study of assaults andassaulters on a maximum security psychiatric unit. Journal of Interper-sonal Violence, 1, 173–191.

*Harwood, D. G., Ownby, R. L., Barker, W. W., & Duara, R. (1998). Thebehavioral pathology in Alzheimer’s disease scale (BEHAVE-AD): Fac-tor structure among community-dwelling Alzheimer’s disease patients.International Journal of Geriatric Psychiatry, 13, 793–800.

Hasselblad, V., & Hedges, L. V. (1995). Meta-analysis of screening anddiagnostic tests. Psychological Bulletin, 117, 167–178.

Hedges, L. V., & Olkin, I. (1985). Statistical methods for meta-analysis.Orlando, FL: Academic Press.

Hedges, L. V., & Vevea, J. L. (1998). Fixed- and random-effects models inmeta-analysis. Psychological Methods, 3, 486–504.

Hemphill, J. F., Hare, R. D., & Wong, S. (1998). Psychopathy andrecidivism: A review. Legal and Criminological Psychology, 3, 139–170.

Hiday, V. A. (1995). The social context of mental illness and violence.Journal of Health and Social Behavior, 36, 122–137.

Hiday, V. A. (1997). Understanding the connection between mental illnessand violence. International Journal of Law and Psychiatry, 20, 399–417.

Hiday, V. A. (2006). Putting community risk in perspective: A look at

correlations, causes and controls. International Journal of Law andPsychiatry, 29, 316–331.

*Hill, C. D., Neumann, C. S., & Rogers, R. (2004). Confirmatory factoranalysis of the Psychopathy Checklist: Screening Version in offenderswith Axis I disorders. Psychological Assessment, 16, 90–95.

Hillbrand, M. (2001). Homicide-suicide and other forms of co-occurringaggression against self and against others. Professional Psychology:Research and Practice, 32, 626–635.

*Hodelet, N. (2001). Psychosis and offending in British Columbia: Char-acteristics of a secure hospital population. Criminal Behaviour andMental Health, 11, 163–172.

*Hodgins, S. (1992). Mental disorder, intellectual deficiency, and crime:Evidence from a birth cohort. Archives of General Psychiatry, 49,476–483.

*Hodgins, S., Hiscoke, U. L., & Freese, R. (2003). The antecedents ofaggressive behavior among men with schizophrenia: A prospectiveinvestigation of patients in community treatment. Behavioral Sciences &the Law, 21, 523–546.

*Holcomb, W. R., & Ahr, P. R. (1988). Arrest rates among young adultpsychiatric patients treated in inpatient and outpatient settings. Hospitaland Community Psychiatry, 39, 52–57.

*Hoptman, M. J., Yates, K. F., Patalinjug, M. B., Wack, R. C., & Convit,A. (1999). Clinical prediction of assaultive behavior among male psy-chiatric patients at a maximum-security forensic facility. PsychiatricServices, 50, 1461–1466.

Hudson, C. G. (2005). Socioeconomic status and mental illness: Tests ofthe social causation and selection hypotheses. American Journal ofOrthopsychiatry, 75, 3–18.

*Inada, T., Minagawa, F., Iwashita, S., & Tokui, T. (1995). Mentallydisordered criminal offenders: Five years’ data from the Tokyo districtpublic prosecutor’s office. International Journal of Law and Psychiatry,18, 221–230.

*Jamieson, L., & Taylor, P. J. (2004). A re-conviction study of special(high security) hospital patients. British Journal of Criminology, 44,783–802.

*Janofsky, J. S., Spears, S., & Neubauer, D. N. (1988). Psychiatrists’accuracy in predicting violent behavior on an inpatient unit. Hospitaland Community Psychiatry, 39, 1090–1094.

Junginger, J. (1996). Psychosis and violence: The case for a contentanalysis of psychotic experience. Schizophrenia Bulletin, 22, 91–103.

Junginger, J., & McGuire, L. (2004). Psychotic motivation and the paradoxof current research on serious mental illness and rates of violence.Schizophrenia Bulletin, 30, 21–30.

*Kaliski, S. Z. (2002). A comparison of risk factors for habitual violencein pre-trial subjects. Acta Psychiatrica Scandinavica, 106(Suppl. 412),58–61.

*Kaliski, S. Z., & Zabow, T. (1995). Violence, sensation seeking, andimpulsivity in schizophrenics found unfit to stand trial. Bulletin of theAmerican Academy of Psychiatry and Law, 23, 147–155.

Kay, S. R., Fiszbein, A., & Opler, L. A. (1987). The Positive and NegativeSyndrome Scale (PANSS) for schizophrenia. Schizophrenia Bulletin, 13,261–276.

*Kay, S. R., Wolkenfeld, F., & Murrill, L. M. (1988). Profiles of aggres-sion among psychiatric patients II: Covariates and predictors. Journal ofNervous and Mental Disease, 176, 547–557.

*Kessler, R. C., Molnar, B. E., Feurer, I. D., & Appelbaum, M. (2001).Patterns and mental health predictors of domestic violence in the UnitedStates: Results from the National Comorbidity Survey. InternationalJournal of Law and Psychiatry, 24, 487–508.

*Kienlen, K. K., Birmingham, D. L., Solberg, K. B., O’Regan, J. T., &Meloy, J. R. (1997). A comparative study of psychotic and nonpsychoticstalking. Journal of the American Academy of Psychiatry and Law, 25,317–334.

*Kjelsberg, E. (2004). Gender and disorder specific criminal career profiles


in former adolescent psychiatric in-patients. Journal of Youth and Ad-olescence, 33, 261–269.

*Klassen, D., & O’Connor, W. A. (1988a). Crime, inpatient admissions,and violence among male mental patients. International Journal of Lawand Psychiatry, 11, 305–312.

*Klassen, D., & O’Connor, W. A. (1988b). Predicting violence in schizo-phrenic and non-schizophrenic patients: A prospective study. Journal ofCommunity Psychology, 16, 217–227.

*Klassen, D., & O’Connor, W. A. (1989). Assessing the risk of violence inreleased mental patients: A cross-validation study. Psychological As-sessment: A Journal of Consulting and Clinical Psychology, 1, 75–81.

*Krakowski, M. I., & Czobor, P. (1997). Violence in psychiatric patients:The role of psychosis, frontal lobe impairment, and ward turmoil.Comprehensive Psychiatry, 38, 230–236.

*Krakowski, M. I., & Czobor, P. (2004a). Gender differences in violentbehaviors: Relationship to clinical symptoms and psychosocial factors.American Journal of Psychiatry, 161, 459–465.

*Krakowski, M. I., & Czobor, P. (2004b). Suicide and violence in patientswith major psychiatric disorders. Journal of Psychiatric Practice, 10,233–238.

*Krakowski, M. I., Czobor, P., & Chou, J. C. (1999). Course of violencein patients with schizophrenia: Relationship to clinical symptoms.Schizophrenia Bulletin, 25, 505–517.

*Krakowski, M. I., Jaeger, J., & Volavka, J. (1988). Violence and psycho-pathology: A longitudinal study. Comprehensive Psychiatry, 29, 174–181.

Krakowski, M. I., Volavka, J., & Brizer, D. (1986). Psychopathology andviolence: A review of the literature. Comprehensive Psychiatry, 27,131–148.

*Kravitz, H. M., Cavanaugh, J. L., & Rigsbee, S. S. (2002). A cross-sectional study of psychosocial and criminal factors associated witharrest in mentally ill female detainees. Journal of the America Academyof Psychiatry and the Law, 30, 380–390.

*Kumari, V., Das, M., Hodgins, S., Zachariah, E., Barkataki, I., Howlett,M., et al. (2005). Association between violent behaviour and impairedprepulse inhibition of the startle response in antisocial personality dis-order and schizophrenia. Behavioural Brain Research, 158, 159–166.

*Kunik, M. E., Snow-Turek, A. L., Iqbal, N., Molinari, V. A., Orengo,C. A., Workman, R. H., et al. (1999). Contribution of psychosis anddepression to behavioral disturbances in geropsychiatric inpatients withdementia. Journals of Gerontology: Series A, Biological Sciences andMedical Sciences, 54, 157–161.

*Lafayette, J. M., Frankle, W. G., Pollock, A., Dyer, K., & Goff, D. C.(2003). Clinical characteristics, cognitive functioning, and criminal his-tories of outpatients with schizophrenia. Psychiatric Services, 54, 1635–1640.

Lagos, J. M., Perlmutter, K., & Saexinger, H. (1977). Fear of the mentallyill: Empirical support for the common man’s response. American Jour-nal of Psychiatry, 134, 1134–1137.

*Lamb, H. R., Shaner, R., Elliott, D. M., DeCuir, W. J., & Foltz, J. T.(1995). Outcome for psychiatric emergency patients seen by an outreachpolice-mental health team. Psychiatric Services, 46, 1267–1271.

*Lamb, R., & Grant, R. W. (1982). The mentally ill in an urban county jail.Archives of General Psychiatry, 39, 17–22.

*Landau, S. F. (1996). Possession of firearms, psychiatric hospitalizationand violent criminal behaviour: The Israeli experience. Studies on Crime& Crime Prevention, 5, 166–181.

*Leonard, R., Tinetti, M. E., Allore, H. G., & Drickamer, M. A. (2006).Potentially modifiable resident characteristics that are associated withphysical or verbal aggression among nursing home residents with de-mentia. Archives of Internal Medicine, 166, 1295–1300.

*Lidz, C. W., Mulvey, E. P., & Gardner, W. (1993). The accuracy ofpredictions of violence to others. Journal of the American MedicalAssociation, 269, 1007–1011.

*Lindqvist, P., & Allebeck, P. (1990). Schizophrenia and crime: A longi-tudinal follow-up of 644 schizophrenics in Stockholm. British Journal ofPsychiatry, 157, 345–350.

*Link, B. G., Andrews, H., & Cullen, F. T. (1992). The violent and illegalbehavior of mental patients reconsidered. American Sociological Re-view, 57, 275–292.

Link, B. G., & Cullen, F. T. (1986). Contact with the mentally ill andperceptions of how dangerous they are. Journal of Health and SocialBehavior, 27, 289–303.

Link, B. G., Cullen, F. T., Frank, J., & Wozniak, J. F. (1987). The socialrejection of former mental patients: Understanding why labels matter.American Journal of Sociology, 96, 1461–1500.

*Link, B. G., Monahan, J., Stueve, A., & Cullen, F. T. (1999). Real in theirconsequences: A sociological approach to understanding the associationbetween psychotic symptoms and violence. American Sociological Re-view, 64, 316–322.

*Link, B. G., & Stueve, A. (1994). Psychotic symptoms and the violent/illegal behavior of mental patients compared to community controls. InJ. Monahan & H. J. Steadman (Eds.), Violence and mental disorder:Developments in risk assessment (pp. 137–159). Chicago: University ofChicago Press.

*Link, B. G., Stueve, A., & Phelan, J. (1998). Psychotic symptoms andviolent behaviors: Probing the components of “threat/control-override”symptoms. Social Psychiatry and Psychiatric Epidemiology, 33, S55–S60.

Lipsey, M. W., & Wilson, D. B. (2001). Practical meta-analysis. ThousandOaks, CA: Sage.

*Lowenstein, M., Binder, R. L., & McNiel, D. E. (1990). The relationshipbetween admission symptoms and hospital assaults. Hospital and Com-munity Psychiatry, 41, 311–313.

*Mamak, M. (1997). Violence and major mental illness. Unpublisheddoctoral dissertation, University of Toronto, Toronto, Ontario, Canada.

*Margari, F., Matarazzo, R., Casacchia, M., Roncone, R., Dieci, M.,Safran, S., et al. (2005). Italian validation of MOAS and NOSIE: Auseful package for psychiatric assessment and monitoring of aggressivebehaviours. International Journal of Methods in Psychiatric Research,14, 109–118.

McNiel, D. E. (1994). Hallucinations and violence. In J. Monahan & H. J.Steadman (Eds.), Violence and mental disorder: Developments in riskassessment (pp. 183–202). Chicago: University of Chicago Press.

McNiel, D. E., & Binder, R. L. (1989). Relationship between preadmissionthreats and later violent behavior by acute psychiatric inpatients. Hos-pital and Community Psychiatry, 40, 605–608.

McNiel, D. E., & Binder, R. L. (1994). The relationship between acutepsychiatric symptoms, diagnosis, and short-term risk of violence. Hos-pital and Community Psychiatry, 45, 133–137.

*McNiel, D. E., Eisner, J. P., & Binder, R. L. (2000). The relationshipbetween command hallucinations and violence. Psychiatric Services, 51,1288–1292.

*Milton, J., Amin, S., Singh, S. P., Harrison, G., Jones, P., Croudace, T.,et al. (2001). Aggressive incidents in first-episode psychosis. BritishJournal of Psychiatry, 178, 433–440.

Mizen, R. (2003). A contribution towards an analytic theory of violence.Journal of Analytic Psychology, 48, 285–305.

*Modestin, J., & Ammann, R. (1995). Mental disorders and criminalbehaviour. British Journal of Psychiatry, 166, 667–675.

*Modestin, J., & Ammann, R. (1996). Mental disorder and criminality:Male schizophrenia. Schizophrenia Bulletin, 22, 69–82.

*Modestin, J., & Wuermle, O. (2005). Criminality in men with majormental disorder with and without comorbid substance abuse. Psychiatryand Clinical Neurosciences, 59, 25–29.

*Moffit, T. E., Brammer, G. L., Caspi, A., Fawcett, J. P., Raleigh, M.,Yuwiler, A., et al. (1998). Whole blood serotonin relates to violence inan epidemiological study. Biological Psychiatry, 15, 446–457.


*Mojtabai, R. (2006). Psychotic-like experiences and interpersonal vio-lence in the general population. Social Psychiatry and Psychiatric Epi-demiology, 41, 183–190.

Monahan, J. (1988). Risk assessment of violence among the mentallydisordered: Generating useful knowledge. International Journal of Lawand Psychiatry, 11, 249–257.

Monahan, J. (1992a). “A terror to their neighbors”: Beliefs about mentaldisorder and violence in historical and cultural perspective. Bulletin ofthe American Academy of Psychiatry and the Law, 20, 191–195.

Monahan, J. (1992b). Mental disorder and violent behavior. AmericanPsychologist, 47, 511–521.

Monahan, J. (1995). Predicting violent behavior: An assessment of clinicaltechniques. Beverly Hills, CA: Sage. (Original work published 1981)

Monahan, J., & Steadman, H. J. (1994). Toward a rejuvenation of riskassessment research. In J. Monahan & H. J. Steadman (Eds.), Violenceand mental disorder: Developments in risk assessment (pp. 1–17).Chicago: University of Chicago Press.

Monahan, J., Steadman, H. J., Silver, E., Appelbaum, P. S., Robbins, P. C.,Mulvey, E. P., et al. (2001). Rethinking risk assessment: The MacArthurstudy of mental disorder and violence. New York: Oxford UniversityPress.

Morey, L. C. (1991). Personality Assessment Inventory: Professional man-ual. Tampa, FL: Psychological Assessment Resources.

*Mullen, P. E., Burgess, P., Wallace, C., Palmer, S., & Ruschena, D.(2000). Community care and criminal offending in schizophrenia. TheLancet, 355, 614–617.

Mulvey, E. P. (1994). Assessing the evidence of a link between mentalillness and violence. Hospital and Community Psychiatry, 45, 663–668.

Mulvey, E. P., Blumstein, A., & Cohen, J. (1986). Reframing the researchquestion of mental patient criminality. International Journal of Law andPsychiatry, 9, 57–65.

Mulvey, E. P., Odgers, C. O., Skeem, J., Gardner, W., Schubert, C., &Lidz, C. (2006). Substance use and community violence: A test of therelation at the daily level. Journal of Consulting and Clinical Psychol-ogy, 74, 743–754.

Mulvey, E. P., Shaw, E., & Lidz, C. W. (1994). Why use multiple sourcesin research on patient violence in the community? Criminal Behaviourand Mental Health, 4, 253–258.

*Myers, K. M., & Dunner, D. L. (1984). Self and other directed violenceon a closed acute-care ward. Psychiatric Quarterly, 56, 178–188.

*Naples, M., & Steadman, H. J. (2003). Can persons with co-occurringdisorders and violent charges be successfully diverted? InternationalJournal of Forensic Mental Health, 2, 137–143.

*Nestor, P. G., Haycock, J., Doiron, S., Kelly, J., & Kelly, D. (1995).Lethal violence and psychosis: A clinical profile. Bulletin of the Amer-ican Academy of Psychiatry and the Law, 23, 331–341.

Nicholls, T. L., Brink, J., Desmarais, S. L., Webster, C. D., & Martin, M.(2006). The Short-Term Assessment of Risk and Treatability (START):A Prospective Validation Study in a Forensic Psychiatric Sample. As-sessment, 13, 313–327.

*Nijman, H., Merckelbach, H., Evers, C., Palmstierna, T., & a Campo, J.(2002). Prediction of aggression on a locked psychiatric admissionsward. Acta Psychiatrica Scandinavica, 105, 390–395.

*Noble, P., & Rodger, S. (1989). Violence by psychiatric in-patients.British Journal of Psychiatry, 155, 384–390.

*Oulis, P., Lykouras, L., Dascalopoulou, E., & Psarros, C. (1996). Ag-gression among psychiatric inpatients in Greece. Psychopathology, 29,174–180.

Overall, J. E., & Gorham, D. R. (1962). The Brief Psychiatric Rating Scale.Psychological Reports, 10, 799–812.

*Paradis, C. M., Broner, N., Maher, L. M., & O’Rourke, T. (2000).Mentally ill elderly jail detainees: Psychiatric, psychosocial and legalfactors. Journal of Offender Rehabilitation, 31, 77–86.

Parker, J. S., Morton, T. L., Lingefelt, M. E., & Johnson, K. S. (2005).

Predictors of serious and violent offending by adjudicated male adoles-cents. North American Journal of Psychology, 7, 407–418.

*Pearson, M., Wilmot, E., & Padi, M. (1986). A study of violent behaviouramong in-patients in a psychiatric hospital. British Journal of Psychia-try, 149, 232–235.

Pescosolido, B. A., Monahan, J., Link, B. G., Stueve, A., & Kikuzawa, S.(1999). The public’s view of the competence, dangerousness, and needfor legal coercion of persons with mental health problems. AmericanJournal of Public Health, 89, 1339–1345.

Phelan, J., & Link, B. (1998). The growing belief that people with mentalillness are violent: The role of the dangerousness criterion for civilcommitment. Social Psychiatry and Psychiatric Epidemiology, 33, 7–12.

*Philipse, M. W. G., Koeter, M. W. J., van der Staak, C. P. F., & van denBrink, W. (2006). Static and dynamic patient characteristics: A prospec-tive study in a Dutch forensic psychiatric sample. Law and HumanBehavior, 30, 309–327.

*Phillips, M. R., Wolf, A. S., & Coons, D. J. (1988). Psychiatry and thecriminal justice system: Testing the myths. American Journal of Psy-chiatry, 145, 605–610.

*Phillips, P., & Nasr, S. J. (1983). Seclusion and restraint and prediction ofviolence. American Journal of Psychiatry, 140, 229–232.

Porcerelli, J. H., Cogan, R., Kamoo, R., & Leitman, S. (2003). Defensemechanisms and self-reported violence toward partners and strangers.Journal of Personality Assessment, 82, 317–320.

*Porporino, F. J., & Motiuk, L. L. (1995). The prison careers of mentallydisordered offenders. International Journal of Law and Psychiatry, 18,29–44.

*Quanbeck, C., Stone, D. C., Scott, C. L., McDermott, B. E., Altshuler,L. L., & Frye, M. A. (2004). Clinical and legal correlates of inmates withbipolar disorder at time of criminal arrest. Journal of Clinical Psychia-try, 65, 198–203.

Quinsey, V. L., Harris, G. T., Rice, M. E., & Cormier, C. A. (2006). Violentoffenders: Appraising and managing risk (2nd ed.). Washington, DC:American Psychological Association.

*Rabinowitz, J., & Mark, M. (1999). Risk factors for violence amonglong-stay psychiatric patients: National study. Acta Psychiatrica Scan-dinavica, 99, 341–347.

Rabkin, J. G. (1979). Criminal behavior of discharged mental patients: Acritical appraisal of the research. Psychological Bulletin, 86, 1–27.

*Raja, M., & Azzoni, A. (2005). Hostility and violence of acute psychiatricinpatients. Clinical Practice and Epidemiology in Mental Health, 1:11.(doi:10.1186/1745-0179-1-11)

*Rasanen, P., Tiihonen, J., Isohanni, M., Rantakallio, P., Lehtonen, J., &Moring, J. (1998). Schizophrenia, alcohol abuse, and violent behavior: A26-year followup study of an unselected birth cohort. SchizophreniaBulletin, 24, 437–441.

Ratakonda, S., Gorman, J. M., Yale, S. A., & Amador, X. F. (1998).Characterization of psychotic conditions: Use of the domains of psycho-pathology model. Archives of General Psychiatry, 55, 75–81.

*Repo, E., Virkkunen, M., Rawlings, R., & Linnoila, M. (1997). Criminaland psychiatric histories of Finnish arsonists. Acta Psychiatrica Scan-dinavica, 95, 318–323.

Rice, D. P., Kelman, S., & Miller, L. S. (1992). The economic burden ofmental illness. Hospital & Community Psychiatry, 43, 1227–1232.

*Rice, M. E., & Harris, G. T. (1992). A comparison of criminal recidivismamong schizophrenic and nonschizophrenic offenders. InternationalJournal of Law and Psychiatry, 15, 397–408.

*Rice, M. E., & Harris, G. T. (1995). Psychopathy, schizophrenia, alcoholabuse, and violent recidivism. International Journal of Law and Psy-chiatry, 18, 333–342.

Riskind, J. H., & Wahl, O. (1992). Moving makes it worse: The role ofrapid movement in fear of psychiatric patients. Journal of Social andClinical Psychology, 11, 349–364.

*Ritsner, M., Modai, I., & Gibel, A. (2003). Decreased platelet peripheral-


type benzodiazepine receptors in persistently violent schizophrenia pa-tients. Journal of Psychiatric Research, 37, 549–556.

*Robbins, Monahan, J., & Silver, E. (2003). Mental disorder, violence, andgender. Law and Human Behavior, 27, 561–571.

*Robertson, G. (1988). Arrest patterns among mentally disordered offend-ers. British Journal of Psychiatry, 153, 313–316.

*Rogers, P., Watt, A., Gray, N. S., MacCulloch, M., & Gournay, K. (2002).Content of command hallucinations predicts self-harm but not violencein a medium secure unit. The Journal of Forensic Psychiatry & Psy-chology, 13, 251–262.

*Rosenfeld, B., & Harmon, R. (2002). Factors associated with violence instalking and obsessional harassment cases. Criminal Justice and Behav-ior, 29, 671–691.

Rosenthal, R. (1991). Meta-analytic procedures for social research (Rev.ed.). Newbury Park, CA: Sage.

*Rossberg, J. I., & Friis, S. (2003). Staff members’ emotional reactions toaggressive and suicidal behavior of inpatients. Psychiatric Serves, 54,1388–1394.

*Rossi, A. M., Jacobs, M., Monteleone, M., Olsen, R., Surber, R. W.,Winkler, E. L., & Wommack, A. (1986). Characteristics of psychiatricpatients who engage in assaultive or other fear-inducing behaviors.Journal of Nervous and Mental Disease, 174, 154–160.

*Russo. G., Salomone, L., & Della Villa, L. (2003). The characteristics ofcriminal and noncriminal mentally disordered patients. InternationalJournal of Law and Psychiatry, 26, 417–435.

*Ryan, E. P., Hart, V. S., Messick, D. L., Aaron, J., & Burnette, M. (2004).A prospective study of assault against staff by youths in a state psychi-atric hospital. Psychiatric Services, 55, 665–670.

Salekin, R. T., Rogers, R., & Sewell, K. W. (1996). A review andmeta-analysis of the Psychopathy Checklist and Psychopathy-Checklist-Revised: Predictive validity of dangerousness. Clinical Psychology:Science and Practice, 3, 203–215.

*Schanda, H., Knecht, G., Schreinzer, D., Stompe, T., Ortwein-Swoboda,G., & Waldhoer, T. (2004). Homicide and major mental disorders: A25-year study. Acta Psychiatrica Scandinavica, 110, 98–107.

*Schneider, L. S., Katz, I. R., & Park, S., Napolitano, J., Martinez, R. A.,& Azen, S. P. (2003). Psychosis of Alzheimer disease: Validity of theconstruct and response to Risperidone. American Journal of GeriatricPsychiatry, 11, 414–425.

*Schwartz-Watts, D., & Morgan, D. W. (1998). Violent versus nonviolentstalkers. Journal of the American Academy of Psychiatry and the Law,26, 241–245.

*Sendi, I. B., & Blomgren, P. G. (1975). A comparative study of predictivecriteria in the predisposition of homicidal adolescents. American Journalof Psychiatry, 132, 423–427.

*Shader, R. I., Jackson, A. H., Harmatz, J. S., & Appelbaum, P. S. (1977).Patterns of violent behavior among schizophrenic inpatients. Diseases ofthe Nervous System, 38, 13–16.

Shadish, W. R., & Haddock, C. K. (1994). Combining estimates of effectsize. In H. Cooper & L. V. Hedges (Eds.), The handbook of researchsynthesis (pp. 261–281). New York: Russell Sage Foundation.

*Shore, D., Filson, C. R., Johnson, W. E., Rae, D. S., Muehrer, P., Kelley,D. J., et al. (1989). Murder and assault arrests of white house cases:Clinical and demographic correlates of violence subsequent to civilcommitment. American Journal of Psychiatry, 146, 645–651.

Silver, E. (2006). Understanding the relationship between mental disorderand violence: The need for a criminological perspective. Law andHuman Behavior, 30, 685–706.

*Silver, H., Goodman, C., Knoll, G., Isakov, V., & Modai, I. (2005).Schizophrenia patients with a history of severe violence differ fromnonviolent schizophrenia patients in perception of emotions but notcognitive function. Journal of Clinical Psychiatry, 66, 300–308.

Skeem, J. L., & Mulvey, E. P. (2001). Psychopathy and communityviolence among civil psychiatric patients: Results from the MacArthur

Violence Risk Assessment Study. Journal of Consulting and ClinicalPsychology, 69, 358–374.

Skeem, J. L., & Mulvey, E. P. (2002). Monitoring the violence potential ofmentally disordered offenders being treated in the community. In A.Buchanan (Ed.), Care of the mentally disordered offender in the com-munity (pp. 111–142). New York: Oxford Press.

Socall, D. W., & Holtgraves, T. (1992). Attitudes toward the mentally ill:The effects of labels and beliefs. Sociological Quarterly, 33, 435–445.

*Sosowsky, L. (1978). Crime and violence among mental patients recon-sidered in view of the new legal relationship between the state and thementally ill. American Journal of Psychiatry, 135, 33–42.

StataCorp. (2007). Stata 10 [Software program]. College Station, TX:Author.

Steadman, H. J., & Halfon, A. (1971). The Baxstrom patients: Back-grounds and outcomes. Seminars in Psychiatry, 3, 376–386.

Steadman, H. J., & Keveles, G. (1972). The community adjustment andcriminal activity of the Baxstrom patients: 1966–1970. American Jour-nal of Psychiatry, 129, 80–86.

Steadman, H. J., Mulvey, E. P., Monahan, J., Robbins, P. C., Appelbaum,P. S., Grisso, T., et al. (1998). Violence by people discharged from acutepsychiatric inpatient facilities and by others in the same neighborhoods.Archives of General Psychiatry, 55, 393–401.

*Steadman, H. J., Silver, E., Monahan, J., Appelbaum, P. S., Robbins,P. C., Mulvey, E. P., et al. (2000). A classification tree approach to thedevelopment of actuarial violence risk assessment tools. Law and Hu-man Behavior, 24, 83–100.

*Steels, M., Roney, G., Larkin, E., Jones, P., Coudace, T., & Duggan, C.(1998). Discharged from special hospital under restrictions: A compar-ison of the fates of psychopaths and the mentally ill. Criminal Behaviourand Mental Health, 8, 39–55.

*Straznickas, K. A., McNiel, D. E., & Binder, R. L. (1993). Violencetoward family caregivers by mentally ill relatives. Hospital and Com-munity Psychiatry, 44, 385–387.

*Stueve, A., & Link, B. G. (1997). Violence and psychiatric disorders:Results from an epidemiological study of young adults in Israel. Psy-chiatric Quarterly, 68, 327–342.

*Swanson, J. W. (1994). Mental disorder, substance abuse, and communityviolence: An epidemiological approach. In J. Monahan & H. J. Stead-man (Eds.), Violence and mental disorder: Developments in risk assess-ment (pp. 101–136). Chicago: University of Chicago Press.

*Swanson, J. W., Borum, R., Swartz, M. S., & Hiday, V. (1999). Violentbehavior proceeding hospitalization among persons with severe mentalillness. Law and Human Behavior, 23, 185–204.

*Swanson, J. W., Borum, R., Swartz, M. S., & Monahan, J. (1996).Psychotic symptoms and disorders and the risk of violent behavior in thecommunity. Criminal Behaviour and Mental Health, 6, 309–329.

*Swanson, J. W., Holzer, C. E., III, Ganju, V. K., & Jono, R. T. (1990).Violence and psychiatric disorder in the community: Evidence from theepidemiologic catchment area surveys. Hospital and Community Psy-chiatry, 41, 761–770.

*Swanson, J. W., Swartz, M. S., Borum, R., Hiday, V. A., Wagner, H. R.,& Burns, B. J. (2000). Involuntary out-patient commitment and reduc-tion of violent behaviour in persons with severe mental illness. BritishJournal of Psychiatry, 176, 324–331.

*Swanson, J. W., Swartz, M. S., & Elbogen, E. B. (2004). Effectiveness ofatypical antipsychotic medications in reducing violent behavior amongpersons with schizophrenia in community-based treatment. Schizophre-nia Bulletin, 30, 3–20.

Swanson, J. W., Swartz, M. S., Essock, S. M., Osher, F. C., Wagner, H. R.,Goodman, L. A., et al. (2002). The social–environmental context ofviolent behavior in persons treated for severe mental illness. AmericanJournal of Public Health, 92, 1523–1531.

*Swanson, J. W., Swartz, M. S., van Dorn, R. A., Elbogen, E. B., Wagner,H. R., Rosenheck, R. A., et al. (2006). A national study of violent


behavior in persons with schizophrenia. Archives of General Psychiatry,63, 490–499.

*Swartz, J. A., & Lurigio, A. A. (2004). Psychiatric diagnosis, substanceuse and dependence, and arrests among former recipients of supplemen-tal security income for drug abuse and alcoholism. Journal of OffenderRehabilitation, 39, 19–38.

*Swartz, M. S., Swanson, J. W., Hiday, V. A., Borum, R., Wagner, R., &Burns, B. J. (1998). Violence and severe mental illness: The effects ofsubstance abuse and nonadherence to medication. American Journal ofPsychiatry, 155, 226–231.

*Swett, C., & Mills, T. (1997). Use of the NOSIE to predict assaults amongacute psychiatric patients. Psychiatric Services, 48, 1177–1180.

*Tam, E., Engelsmann, F., & Fugere, R. (1996). Patterns of violentincidents by patients in a general hospital psychiatric facility. Psychiat-ric Services, 47, 86–88.

*Tanke, E. D., & Yesavage, J. A. (1985). Characteristics of assaultivepatients who do and do not provide visible cues of potential violence.American Journal of Psychiatry, 142, 1409–1413.

*Tardiff, K. (1981). Assault in hospitals and placement in the community.Bulletin of the American Academy of Psychiatry and the Law, 9, 33–39.

*Tardiff, K. (1984). Characteristics of assaultive patients in private hospi-tals. American Journal of Psychiatry, 141, 1232–1235.

*Tardiff, K., & Koenigsberg, H. W. (1985). Assaultive behavior amongpsychiatric outpatients. American Journal of Psychiatry, 142, 960–963.

*Tardiff, K., Marzuk, P. M., Leon, A. C., & Portera, L. (1997). A pro-spective study of violence by psychiatric patients after hospital dis-charge. Psychiatric Services, 48, 678–681.

*Tardiff, K., Marzuk, P. M., Leon, A. C., Portera, L., & Weiner, C. (1997).Violence by patients admitted to a private psychiatric hospital. AmericanJournal of Psychiatry, 154, 88–93.

*Tardiff, K., & Sweillam, A. (1980). Assault, suicide, and mental illness.Archives of General Psychiatry, 37, 164–169.

*Tardiff, K., & Sweillam, A. (1982). Assaultive behavior among chronicinpatients. American Journal of Psychiatry, 139, 212–215.

Taylor, P. J. (1995). Schizophrenia and the risk of violence. In S. R. Hirsch& D. R. Weinberger (Eds.), Schizophrenia (pp. 163–183). Oxford,United Kingdom: Blackwell.

Taylor, P. J. (1998). When symptoms of psychosis drive serious violence.Social Psychiatry and Psychiatric Epidemiology, 33, S47–S54.

Taylor, P. J. (2006). Delusional disorder and delusions: Is there a risk ofviolence in social interactions about the core symptom? BehavioralSciences and the Law, 24, 313–331.

Taylor, P. J., Garety, P., Buchanan, A., Reed, A., Wessely, S., Ray, K.,Dunn, G., & Grubin, D. (1994). Delusions and violence. In J. Monahan& H. J. Steadman (Eds.), Violence and mental disorder: Developmentsin risk assessment (pp. 161–182). Chicago: University of Chicago Press.

*Taylor, P. J., & Gunn, J. (1984). Violence and psychosis: I. Risk ofviolence among psychotic men. British Medical Journal, 288, 1945–1949.

Taylor, P. J., & Hodgins, S. (1994). Violence and psychosis: Criticaltimings. Criminal Behaviour and Mental Health, 4, 267–289.

*Taylor, P. J., Leese, M., Williams, D., Butwell, M., Daly, R., & Larkin,E. (1998). Mental disorder and violence: A special (high security)hospital study. British Journal of Psychiatry, 172, 218–226.

*Teasdale, B., Silver, E., & Monahan, J. (2006). Gender, threat/control-override delusions and violence. Law and Human Behavior, 30, 649–658.

Tehrani, J. A., Brennan., P. A., Hodgins, S., & Mednick, S. A. (1998).Mental illness and criminal violence. Social Psychiatry and PsychiatricEpidemiology, 33, S81–S85.

*Tengstrom, A., Hodgins, S., Grann, M., Långstrom, N., & Kullgren, G.(2004). Schizophrenia and criminal offending: The role of psychopathyand substance use disorders. Criminal Justice and Behavior, 31, 367–391.

*Tengstrom, A., Hodgins, S., Muller-Isberner, R., Jockel, D., Freese, R.,Ozokyay, K., et al. (2006). Predicting violent and antisocial behavior inhospital using HCR-20: The effect of diagnoses on predictive accuracy.International Journal of Forensic Mental Health, 5, 39–53.

*Tennent, G., Loucas, K., Fenton, G., & Fenwick, P. (1974). Male admis-sions to Broadmoor hospital. British Journal of Psychiatry, 125, 44–50.

*Tennent, G., & Way, C. (1984). The English Special Hospital—A 12–17year follow-up study: A comparison of violent and non-violent re-offenders and non-offenders. Medicine, Science, and the Law, 24,81–91.

*Teplin, L. A. (1985). The criminality of the mentally ill: A dangerousmisconception. American Journal of Psychiatry, 142, 593–599.

*Teplin, L. A. (1994). Psychiatric and substance abuse disorders amongmale urban jail detainees. American Journal of Public Health, 84,290–293.

*Teplin, L. A., Abram, K. M., & McClelland, G. M. (1994). Does psy-chiatric disorder predict violent crime among released jail detainees? Asix-year longitudinal study. American Psychologist, 49, 335–342.

*Teplin, L. A., Abram, K. M., & McClelland, G. M. (1996). Prevalence ofpsychiatric disorders among incarcerated women. I: Pretrial jail detain-ees. Archives of General Psychiatry, 53, 505–512.

*Thomas, S., Leese, M., Walsh, E., McCrone, P., Moran, P., Burns, T., etal. (2005). A comparison of statistical models in predicting violence inpsychotic illness. Comprehensive Psychiatry, 296–303.

*Tiihonen, J., Eronen, M., & Hakola, P. (1993). Criminality associatedwith mental disorders and intellectual deficiency. Archives of GeneralPsychiatry, 50, 917–918.

*Tiihonen, J., Isohanni, M., Rasanen, P., Koiranen, M., & Moring, J.(1997). Specific major mental disorders and criminality: A 26-yearprospective study of the 1966 Northern Finland birth cohort. AmericanJournal of Psychiatry, 154, 840–845.

*Troisi, A., Kustermann, S., Di Genio, M., & Siracusano, A. (2003).Hostility during admission interview as a short-term predictor of aggres-sion in acute psychiatric male inpatients. Journal of Clinical Psychiatry,64, 1460–1464.

*Valdiserri, E. V., Carroll, K. R., & Hartl, A. J. (1986). A study of offensescommitted by psychotic inmates in a county jail. Hospital and Commu-nity Psychiatry, 37, 163–166.

Valliant, P. M., Gristey, C., Pottier, D., & Kosmyna, R. (1999). Riskfactors in violent and nonviolent offenders. Psychological Reports, 85,675–680.

*Ventura, L., Cassel, C., Jacoby, J., & Huang, B. (1998). Case manage-ment and recidivism of mentally ill persons released from jail. Psychi-atric Services, 49, 1330–1337.

*Verma, S., Poon, L. Y., Subramaniam, M., & Chong, S. A. (2005).Aggression in Asian patients with first-episode psychosis. InternationalJournal of Social Psychiatry, 51, 365–371.

*Villeneuve, D. B. (1994). Predictors of general and violent recidivismamong mentally disordered federal incarcerates. Unpublished doctoraldissertation, Queen’s University, Kingston, Ontario, Canada.

*Volavka, J., Laska, E., Baker, S., Meisner, M., Czobor, P., & Krivelevich,I. (1997). History of violent behaviour and schizophrenia in differentcultures: Analyses based on the WHO study on determinants of outcomeof severe mental disorders. British Journal of Psychiatry, 171, 9–14.

*Vreugdenhil, C., Vermeiren, R., Wouters, L., Doreleijers, T. A. H., vanden Brink, W., & Vreugdenhil, C. (2004). Psychotic symptoms amongmale adolescent detainees in the Netherlands. Schizophrenia Bulletin,30, 73–86.

*Wallace, C., Mullen, P. E., & Burgess, P. (2004). Criminal offending inschizophrenia over a 25-year period marked by deinstitutionalizationand increasing prevalence of comorbid substance use disorders. Ameri-can Journal of Psychiatry, 161, 716–727.

*Wallace, C., Mullen, P. E., Burgess, P., Palmer, S., Ruschena, D., &


Browne, C. (1998). Serious criminal offending and mental disorder:Case linkage study. British Journal of Psychiatry, 172, 477–484.

Walsh, E., Buchanan, A., & Fahy, T. (2002). Violence and schizophrenia:Examining the evidence. British Journal of Psychiatry, 180, 490–495.

Walsh, E., Gilvarry, C., Samele, C., Harvey, K., Manley, C., Tattan, T., etal. (2004). Predicting violence in schizophrenia: A prospective study.Schizophrenia Research, 67, 247–252.

Walters, G. D. (2003). Predicting institutional adjustment and recidivismwith the psychopathy checklist factor scores: A meta-analysis. Law andHuman Behavior, 27, 541–558.

Watson, A. C., Corrigan, P., Larson, J. E., & Sells, M. (2007). Self-stigmain people with mental illness. Schizophrenia Bulletin, 33, 1312–1318.

Webster, C. D., Douglas, K. S., Eaves, D., & Hart, S. D. (1997). HCR-20:Assessing risk for violence, Version 2. Burnaby, British Columbia, Canada:Mental Health, Law, & Policy Institute, Simon Fraser University.

*Weibe, D. J. (1998). Mental health factors in the adjustment of youthdetention center residents. Simon Fraser University, Burnaby, BritishColumbia, Canada.

*Werner, P. D., Rose, T. L., & Yesavage, J. A. (1983). Reliability,accuracy, and decision-making strategy in clinical predictions of immi-nent dangerousness. Journal of Consulting and Clinical Psychology, 51,815–825.

Wessely, S. (1993). Violence and psychosis. In C. Thompson & P. Cowen(Eds.), Violence: Basic and clinical science (pp. 120–134). Oxford,England: Butterworth-Heinemann.

*Wessely, S. C., Castle, D., Douglas, A. J., & Taylor, P. J. (1994). Thecriminal careers of incident cases of schizophrenia. Psychological Med-icine, 24, 483–502.

*Yesavage, J. A. (1983a). Bipolar illness: Correlates of dangerous inpatientbehaviour. British Journal of Psychiatry, 143, 554–557.

*Yesavage, J. A. (1983b). Inpatient violence and the schizophrenic patient.Acta Psychiatrica Scandinavica, 67, 353–357.

*Yesavage, J. A. (1984). Correlates of dangerous behavior by schizophren-ics in hospital. Journal of Psychiatric Research, 18, 225–231.

*Yesavage, J. A., Benezech, M., Larrieu-Arguille, R., Bourgeois, M.,Tanke, E., Rager, P., et al. (1986). Recidivism of the criminally insanein France: A 22-year follow-up. Journal of Clinical Psychiatry, 47,465–466.

*Yesavage, J. A., Werner, P. D., Becker, J., Holman, C., & Mills, M.(1981). Inpatient evaluation of aggression in psychiatric patients. Jour-nal of Nervous and Mental Disease, 169, 299–302.

Yudofsky, S. C., Silver, J. M., Jackson, W., Endicott, J., & Williams, D.(1986). The Overt Aggression Scale for the objective rating of verbaland physical aggression American Journal of Psychiatry, 143, 35–39.

*Zitrin, A., Hardesty, A. S., Burdock, E. I., & Drossman, A. K. (1976).Crime and violence among mental patients. American Journal of Psy-chiatry, 133, 142–149.

Zocchetti, C., Consonni, D., & Bertazzi, P. A. (1997). Relationship be-tween prevalence rate ratios and odds ratios in cross-sectional studies.International Journal of Epidemiology, 26, 220–223.

(Appendix follows)


Appendix

Transformational Formulas

Estimating an Odds Ratio From Frequencies Provided inan Article (Hasselblad & Hedges, 1995)

Status Psychosis No psychosis

Violent a bNot violent c d

Note. OR � ad/bc. When a twofold table contained one or more emptycells, 0.5 was added to each cell of the table to allow the calculation of anOR (Fleiss, 1994).

Estimating an Odds Ratio From a Standardized MeanDifference (d) Effect Size (Hasselblad & Hedges, 1995;see also Lipsey & Wilson, 2001, p. 198)

OR � e�d/�3

1. If d needed to be estimated, formulas presented in Table B10by Lipsey and Wilson (2001, pp. 198–200) were used dependingon the information presented in the article. Most commonly, stan-dardized mean difference effect sizes were estimated with each ofthe two means (X), the standard deviation (s), and the sample size(n) for each group (Lipsey & Wilson, 2001, p. 198), as follows:

d � �X1 � X2/�spooled

spooled � ��n1 � 1s12 � �n2 � 1s2

2/��n1 � n2 � 2�

2. d was estimated from r (Lipsey & Wilson, 2001, p. 199), asfollows:

d � 2r/��1 � r2�

3. r, which subsequently was converted to d, was estimated froma chi-square value (�2) with df � 1 and total sample size (N)(Lipsey & Wilson, 2001, p. 201), as follows:

|r| � ��2/N

4. d was estimated from an independent t test (t) and totalsample size (N) (Lipsey & Wilson, 2001, p. 198), as follows:

d � �2t/N �assumes n1 � n2)

5. d was estimated from an independent t test (t) and the samplesize for each group (n) (Lipsey & Wilson, 2001, p. 198), asfollows:

d � �t �n1 � n2/�n1n2

6. d was estimated with an F ratio (F) from a one-way analysisof variance and total sample size (N) as follows (Lipsey & Wilson,2001, p. 199):

|d| � �z F/N �assumes n1 � n2

Estimating an Odds Ratio From a Rate Ratio (Zocchetti,Consonni, & Bertazzi, 1997)

POR � PRR��1 – PrE � �PRR � PrE – PrD/�1 � PrE

� �PRR � PrE – �PRR � PrD�,

where PrD � prevalence of psychosis, PrE � prevalence of theexposure, POR � prevalence odds ratio, and PRR � prevalencerate ratio.

Received February 25, 2008Revision received April 9, 2009

Accepted April 20, 2009 �


Douglas, Guy, & Hart, 2009, Psychosis as a Risk Factor for Violence to Others - A Meta-Analysis

Documents

odds of violence

violence risk assessments

measurement of psychosis

role of psychosis

1037a0016311 psychosis

association bonta

university drive

health research