Media Analysis

Real in Their Consequences: A Sociological Approach to Understanding the Association between Psychotic Symptoms and Violence

Author(s): Bruce G. Link, John Monahan, Ann Stueve and Francis T. Cullen

Source: American Sociological Review , Apr., 1999, Vol. 64, No. 2 (Apr., 1999), pp. 316- 332

Published by: American Sociological Association

Stable URL: https://www.jstor.org/stable/2657535

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org. Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at https://about.jstor.org/terms

American Sociological Association is collaborating with JSTOR to digitize, preserve and extend access to American Sociological Review

This content downloaded from �������������76.183.236.69 on Fri, 04 Dec 2020 22:04:21 UTC�������������

All use subject to https://about.jstor.org/terms

REAL IN THEIR CONSEQUENCES:

A SOCIOLOGICAL APPROACH TO UNDERSTANDING THE

ASSOCIATION BETWEEN PSYCHOTIC SYMPTOMS AND VIOLENCE

Bruce G. Link Columbia University and

New York State Psychiatric Institute

John Monahan Ann Stueve Francis T. Cullen The University of Virginia Columbia University University of Cincinnati

Studies conducted over the past three decades have consistently reported an

association between mental illness and violence. We propose a sociologi-

cally inspired explanation for this association by referring to the Thomas

Theorem-if situations are defined as real, they are real in their conse-

quences. We identify a small subset of psychotic symptoms, termed "threat!

control-override" symptoms, that tend to induce violence because they in-

fluence the definitions of situations. Our data come from an epidemiological

study conducted in Israel that includes a psychiatrist-administered diagnos-

tic interview. We find an association between violent behaviors and psychi-

atric diagnosis that cannot be accountedfor by sociodemographic variables.

Threat/control-override symptoms also are strongly related to violent behav-

iors and explain a substantial part of the association between violence and

psychiatric diagnoses. Other equally severe psychotic symptoms are not re-

lated to indicators of violence when threat/control-override symptoms are

controlled. These findings support our explanation for the association be-

tween mental illness and violence, and challenge the stereotype that most

people with mental illnesses are dangerous.

n developing his concept of the "self-ful-

filling prophecy," Merton ([1948]1968) drew attention to what he called the Thomas Theorem-"If men define situations as real they are real in their consequences." Accord-

ing to Merton, the theorem, attributed to W. I. Thomas,1 "lacks the sweep and preci-

sion of a Newtonian theorem [but] possesses the same gift of relevance, being instruc-

tively applicable to many, if indeed not most, social processes" (p. 475). Within sociology the Thomas Theorem has been most closely associated with a symbolic interactionist ap-

proach that directs attention to the impor- tance of people's interpretations of situations for understanding their behavior. We propose a specific application of the idea that people's definitions of situations shape their behavior: We propose that this insight can

elucidate the association between psychotic symptoms and violence.

Violence, or "dangerousness," has been a strong component of the stereotype of men- tal illness throughout recorded history (Monahan 1992) and is a principal reason why people with mental illnesses are rejected (Link et al. 1987). Given the ramifications of the stereotype, many have questioned its va- lidity by asking whether people with mental illnesses are, in fact, more likely to engage in violent acts. Until the mid-1960s, research findings seemed to disconfirm the stereo-

Direct all correspondence to Bruce G. Link,

Epidemiology of Mental Disorders, 100 Haven

Avenue, Apt. 31D, New York, NY 10032 (BGL1 @Columbia.edu). The authors thank the anony- mous ASR reviewers and Sharon Schwartz for

helpful comments. This research was supported in part by NIMH Grants MH30710 and MH38773.

1 The quotation that became the theorem is con- tained in a book by W. I. Thomas and Dorothy Swaine Thomas (1928:572), but it was attributed to W. I. Thomas by Merton and others. For docu- mentation of the appropriateness of this attribu- tion to only one of the authors, see Merton

(1995).

316 American Sociological Review, 1999, Vol. 64 (April:316-332)

This content downloaded from �������������76.183.236.69 on Fri, 04 Dec 2020 22:04:21 UTC�������������

All use subject to https://about.jstor.org/terms

MENTAL DISORDERS AND VIOLENT BEHAVIOR 317

type. But subsequent studies have shown an elevated rate of arrest among discharged pa- tients in both the United States (Rabkin 1979) and Europe (Eronen, Tiihonen, and Haola 1996; Hodgins et al. 1996; Wessely 1997), leading some investigators to con- clude with Steadman (1981) that research on ex-mental patients supports public stereo- types "to an extent rarely acknowledged by mental health professionals" (p. 310).

Others thought this conclusion premature, however, noting that the arrest-rate studies are subject to important biases (Link, Andrews, and Cullen 1992). For example, these studies usually compared arrest rates of patients from public mental hospitals and clinics serving relatively poor people from violence-prone neighborhoods with arrest rates of the general population without con-

trolling for relevant confounding factors. Moreover, there is evidence that people with mental illnesses are more likely to be ar- rested than people without mental illnesses, even when the offending behaviors are the

same (Robertson 1988; Teplin 1984; but also see Klassen and O'Connor 1989). These bi- ases pertaining to who, among the population of people with mental disorders, are selected for mental hospitalization and who, among the people who commit violent acts, are se- lected for arrest indicated the need for fur- ther research.

Three epidemiological studies of commu- nity-based samples, however, provide fur- ther evidence for an association between mental illness and violence. While these studies can also be criticized, they are not subject to the same biases as the arrest-rate studies. Using data from three sites of the Epidemiological Catchment Area study, Swanson and colleagues (Swanson 1994; Swanson et al. 1990) showed that violence (self-reported hitting, fighting, or weapon use) was five times higher among commu- nity residents who met criteria for a DSM- III Axis I diagnosis2 of mental disorder than

2 The American Psychological Association' s Diagnostic and Statistical Manual employs a multi-axial system: The first axis addresses men- tal disorders that can be defined as having onsets and offsets; other axes address personality disor- ders, physical health, stressful circumstances, and social functioning.

for community residents who did not. This connection between mental disorder and violent behavior held when age, gender, and socioeconomic status were controlled and for persons with mental disorders who did not meet diagnostic criteria for substance abuse.

A second study (Link et al. 1992) com- pared arrest rates and self-reported violence in samples of psychiatric patients (current and former) with arrests and self-reported violence among community residents drawn from the same New York City neighbor- hoods. Patient groups tended to be more violent than never-treated community resi- dents across several indicators of violence, even controlling for individual-level socio- demographic factors, census-based homi- cide rates, and social desirability. Finally, Stueve and Link (1997) examined the asso- ciation between psychiatric disorders and violent behaviors using data from the epide- miological study in Israel analyzed below. Individuals diagnosed with psychotic or bi- polar disorders were more likely than re- spondents without these diagnoses to report recent fighting or weapon use, controlling for sociodemographic characteristics, life- time substance abuse, and antisocial person- ality disorder.

Every study mentioned above can be chal- lenged on the basis of some methodological weakness-such as the exclusive use of ar- rest data or self-report data, reverse causa- tion, inappropriate control groups, or the failure to control for an important variable. But because many studies with different po- tential biases find the same thing, it is diffi- cult to dismiss the observed connection be- tween violence and mental illness as simply artifactual.

If there is an association between mental illness and violence, what is it about mental illness that elevates rates of violence? Re- sults from the Link et al. (1992) study pro- vide a clue: Psychotic symptoms played a significant role in accounting for the associa- tion between patient status and violence. The authors did not explain, however, why psy- chotic symptoms lead to violence. We pro- pose a sociological understanding of the con- nection between psychotic symptoms and violence derived from the idea that people's definitions of situations-even when those

This content downloaded from �������������76.183.236.69 on Fri, 04 Dec 2020 22:04:21 UTC�������������

All use subject to https://about.jstor.org/terms

318 AMERICAN SOCIOLOGICAL REVIEW

definitions are based on delusions-have real consequences.3

DEFINITIONS AND CONSEQUENCES

If people define situations as real, they are real in their consequences. The Thomas

Theorem resonates with the more general symbolic interactionist principle that in or- der to understand a person's behavior, one must understand his or her definition of the situation. The idea that definitions are impor- tant determinants of violent and illegal be- havior can be found in several strands of in- quiry on crime, delinquency, and youth vio- lence (Heimer 1997). For example, our un- derstanding of violence has been enhanced by studies that investigated how perpetrators and victims defined situations preceding a violent act (Athens 1980; Luckenbill 1977). "Definitions" also are central to Sutherland's (1947) differential association theory and to Akers's (1973) expansion of that theory into social learning theory. In this work, defini- tions are "orientations, rationalizations, defi- nitions of the situation, and other moral atti- tudes that define the commission of an act as right or wrong, good or bad, desirable or un- desirable, justified or unjustified" (Akers 1997:64). They are important determinants of violence because they "operate as cues or signals to the individual as to what responses are appropriate or expected in a given situa- tion" (Akers 1997:65). Definitions, in short, are consequential because they influence the motivation to act. Indeed, Heimer (1997) identifies definitions as the common causal pathway for the effects of more distal influ- ences on youth violence such as association with aggressive peers, socioeconomic status, and parenting practices.

We use the interactionist principle with its

emphasis on definitions to elucidate the im- plications of psychotic symptoms for vio-

lence. In doing so, we turn from instances in

which definitions are more obviously shaped

by social factors, to psychosis, where the definitions of importance are associated with

delusional processes that are not so obviously shaped by social circumstances. We draw on the interactionist insight in claiming that,

whatever their source, definitions play a key role in violence. That is, we propose that once

one suspends concern about the irrationality of psychotic symptoms and accepts that they are experienced as real, violence will unfold in a "rational" (understandable) fashion ac- cording to the general principle derived from the symbolic interactionist approach.

But which psychotic symptoms are likely

to result in definitions of situations that are conducive to violence? We predict that psy- chotic symptoms are more likely to promote violence (1) if they cause the person to feel threatened by others (threat), or (2) if they override internal controls that might other- wise block the expression of violence (con- trol-override). In the first instance, threat, the afflicted person defines the situation as so gravely threatening that a violent re- sponse is viewed as justified. In the second instance, control-override, the symptom leads to a definition of the situation that overwhelms routine behavioral constraints.

For example, a person may believe that God is commanding him or her to kill someone. In such an instance, that person's definition of the situation includes a divine prescrip- tion for violence.

Thus, according to the idea that definitions produce real consequences, it is these "threat/control-override symptoms" that are likely to be associated with violence. Other types of psychotic symptoms-such as hear- ing voices, hearing one's thoughts spoken aloud, or having one's thoughts taken away-are not likely to result in violence in the absence of threat/control-override symp- toms. While these symptoms are severe, they are not as likely to affect a person's defini- tion of the situation in such a way as to lead to violence. For example, experiencing one's thoughts as having been taken away is an ex- tremely disturbing and disruptive experi- ence: However, this experience does not en-

3 Our confidence in applying the Thomas Theo- rem to psychotic symptoms and violence is bol- stered by the fact that the original example used by W. I. Thomas involved these same themes. The passage that precedes the Thomas Theorem

tells of a prisoner at Dannemora prison who had "killed several persons who had the unfortunate habit of talking to themselves on the street. From the movement of their lips he imagined that they were calling him vile names, and he behaved as if this were true. If men define situations as real, they are real in their consequences" (Thomas and Thomas 1928:572).

This content downloaded from �������������76.183.236.69 on Fri, 04 Dec 2020 22:04:21 UTC�������������

All use subject to https://about.jstor.org/terms

MENTAL DISORDERS AND VIOLENT BEHAVIOR 319

tail the insertion of other thoughts that might

override prohibitions against violence; nor does it necessarily involve believing that oth- ers are out to do harm.4 Consequently, this

symptom is unlikely to foster a definition of the situation that leads to violence.

Note that our theorizing about threat/con-

trol-override symptoms is designed to ex- plain differences in levels of violence be-

tween people who have developed mental ill- nesses and people who have not. As such, it

addresses a very different problem than "risk-assessment" research (Tardiff 1996), which aims to predict violence within popu- lations of people with mental illnesses. This latter approach employs a large battery of

questions that includes measures of alcohol and drug use, impulsivity, anger, stability of social relationships, history of previous vio- lence, and so on. While these models are in- formative in predicting violence, the vari- ables assessed would be expected to predict violence in almost any group and are not spe- cifically directed at explaining why people with mental illnesses have higher rates of violence than people without mental ill- nesses. Moreover, focused as they are on ex- plaining violent behavior within groups, studies using the risk-assessment approach have not incorporated the requisite design- samples of diagnosed cases of mental illness and appropriate comparison samples of those without mental illnesses-needed to examine differences between groups (see Steadman et al. 1998 for an exception).

Our predictions also differ from those one would make using a medical/diagnostic ap- proach to understanding mental disorders and their consequences. In the medical/diag- nostic approach, symptoms are used to char- acterize syndromes or disease entities (American Psychiatric Association 1994), and the disease construct becomes the major factor to be understood. The search for causes, consequences, and treatments fo- cuses on the identified disease entity rather than on the specific symptoms encompassed

4 It is, of course, possible that one might be- lieve that others were intending harm by stealing

one's thoughts. But if so, one presumably would also respond positively to a threat/control-over- ride symptom asking about other people's inten- tions to cause harm.

by it. Consequently, hypotheses concerning mental illness and violence are likely to fo-

cus on diagnosis and not specific symptoms.

While such a diagnostic approach has en-

joyed considerable success in medicine, par- ticularly with regard to discovering causes

and treatments for disease, there are reasons to doubt its usefulness for understanding so-

cial consequences of mental illness such as violence. This is because there is no one-to-

one correspondence between symptoms and

diagnosis, and because it is the social mean- ing of the symptoms that influences the like- lihood of violence. Thus, our approach dif- fers from the medical/diagnostic perspec-

tive-we cluster symptoms according to

their meaning for social interactions involv- ing violence. If our approach is correct,

threat/control-override symptoms will have primacy over diagnosis in predicting violent

behavior. Specifically, diagnostic categories will have associations with violence prima- rily because they include people who experi- ence threat/control-override symptoms.

PREVIOUS RESEARCH ON THREAT/ CONTROL-OVERRIDE SYMPTOMS

Although they did not explicitly connect their theorizing to the Thomas Theorem or the interactionist perspective, Link and Stueve (1994) used these principles implic- itly. In their reanalysis of the Link et al.

(1992) data, they identified three of the 13 items in the Psychiatric Epidemiology Re- search Interview's (PERI) psychotic symp- toms scale (Dohrenwend et al. 1980) as in- volving either threat or the override of self control. Ten other items indicative of severe psychosis but not necessarily involving threat or the override of personal control formed a scale of other psychotic symptoms (see Appendix A). Link and Stueve (1994) found that the three-item threat/control-over- ride scale was strongly related to recent self- reported hitting, fighting, and weapon use, even when controlling for a scale composed of the other 10 psychotic symptoms. More- over, the three threat/control-override (TCO) symptoms accounted for differences in rates of violence between patients and community controls. By contrast, the 10 other psychotic symptoms, although equally severe and in- dicative of psychosis, showed no association

This content downloaded from �������������76.183.236.69 on Fri, 04 Dec 2020 22:04:21 UTC�������������

All use subject to https://about.jstor.org/terms

320 AMERICAN SOCIOLOGICAL REVIEW

with the indicators of violence when the TCO symptoms were controlled.

Swanson et al. (1996) replicated Link and

Stueve's finding with data from the Epide- miological Catchment Area (ECA) study. They found that ECA respondents who re-

ported threat/control-override symptoms were twice as likely as those with only non- TCO psychotic symptoms to report violence, and about six times as likely as those with

no mental disorder. In sum, there is growing evidence from

studies of the epidemiology of mental disor- ders concerning the association between vio- lence and mental illness generally, and be-

tween threat/control-override symptoms and violence in particular. Nevertheless, this growing body of evidence can be challenged for three reasons.

First, the evidence to date is based largely

on samples from U.S. urban settings, where the prevalence of violence is much greater than in many other countries. While epide- miological studies of mental disorder and

violence have been conducted outside the

United States, only one investigation-our preliminary analyses of data from Israel

(Stueve and Link 1997)-has reported find- ings using assessments that are independent of official statistics gleaned from mental hos- pitals and criminal justice records. That in- vestigation, however, did not consider the role of psychiatric symptoms. Furthermore, no non-U.S. study has reported data relating

threat/control-override symptoms to vio- lence. The absence of such studies is particu- larly important in light of claims made by several investigators concerning the possible influence of contextual factors on the vio- lence/mental illness association (Estroff et al. 1994; Hiday 1995; Link and Stueve 1995).

Second, it has been argued that the asso- ciation between mental illness and violence

is spurious owing to comorbidity with anti- social personality and drug and alcohol abuse. According to this view, there is noth- ing about mental illness (save antisocial per-

sonality and substance abuse) that causes violence, and as a result there is no need to probe aspects of mental illness such as threat/control-override symptoms.

Third, it can be argued that diagnostic con- structs are crucial to explaining violence, and that once diagnosis is controlled, specific

symptoms-including threat/control-over-

ride symptoms-will matter little. In sharp

contrast, with this medical/diagnostic ap- proach, our theory predicts that threat/con-

trol-override symptoms themselves carry the

essential information about mental illness

that is relevant to violence, and that when these symptoms are taken into account, the association between specific diagnoses and outcomes will be explained.

In this paper, we report the results of a community-based epidemiological study of mental disorders conducted in Israel (Dohrenwend et al. 1992), a nation in which rates of citizen-to-citizen violence are much

lower than those in the United States and where rates of substance abuse and antiso- cial personality disorder are also low.5 As a result, our study can assess both the general-

ity of the association between mental illness and violence and the validity of comorbidity as a potential alternative explanation for the association. In addition, rich diagnostic data collected through structured clinical inter-

views allow us to assess whether threat/con-

trol-override symptoms account for the asso- ciation between diagnoses and violent behav- ior, or whether diagnoses contain the essen-

tial information regarding violence.

METHODS

Sample

Israeli-born Jews ages 24 to 33 were the tar- get population in a study designed to imple- ment Dohrenwend et al.'s (1992) quasi-ex-

perimental test of the social causation/social selection issue. Using the Israeli population

5 A relatively direct comparison of the preva- lence of violence is possible between the Israeli sample (Stueve and Link 1997) and the Washing- ton Heights sample (Link et al. 1992) because identical questions were used. The five-year prevalence of fighting and weapon use among never-treated community residents in Washington Heights was 15.1 percent and 2.7 percent respec- tively, whereas in Israel the corresponding figures were 5.2 percent and .6 percent. Rates of lifetime alcoholism are 6 to 10 times higher in Epidemio- logical Catchment Area sites of New Haven, Bal- timore, and St. Louis than in Israel, and lifetime drug-use disorder and antisocial personality dis- order are about three times higher in these cities than in Israel (Levav et al. 1993).

This content downloaded from �������������76.183.236.69 on Fri, 04 Dec 2020 22:04:21 UTC�������������

All use subject to https://about.jstor.org/terms

MENTAL DISORDERS AND VIOLENT BEHAVIOR 321

register, 19,000 people were screened for data on socioeconomic status and ethnicity with a 98 percent completion rate. This "screening" sample was used to generate the stratified "PERI" sample; 4,914 people were inter-

viewed with the Psychiatric Epidemiology Research Interview (PERI), which contains

screening scales and demographic informa- tion. Interviews were conducted in person, and the completion rate was 94.5 percent.

Following this, a subsample of respondents who completed the PERI was reinterviewed.

This subsample included all persons who screened positive for a mental disorder on the screening scales as well as a sample of 18 percent of those who screened negative. These respondents were administered a ver- sion of the Schedule for Affective Disorders and Schizophrenia (SADS) by psychiatrists. This "SADS" sample included 2,741 respon- dents and had a 90.7 percent completion rate. In the current study, we use the second stage sample minus 63 persons (2.3 percent) who were unable, mainly because of mental retar-

dation, to complete the lay-administered PERI. This left 2,678 cases for analysis.

Because of our sampling strategy, we weighted cases to account for stratification based on gender, ethnicity (North African versus European backgrounds), birth cohort, education, parental exposure to holocaust conditions, and the PERI screen (Dohrenwend et al. 1992). We used the soft-

ware program SUDAAN (Shah et al. 1992) to compute standard errors appropriate to our complex survey design.

Although the sample was not specifically drawn to answer questions about the associa- tion between mental disorder and violence, it has several advantageous characteristics. First, the sample is register-based rather than household-based, which means that all iden-

tified respondents are interviewed, including those in mental hospitals, prisons, jails, and so on. Clearly, such respondents are relevant to a consideration of the association between

mental disorder and violence, but many would be excluded from a household sample. Second, the completion rates were high by U.S. standards, thereby reducing the possi- bility of sample selection bias that might also affect the association between mental illness

and violence. Finally, data gathered by skilled clinical interviewers supplement in-

formation gathered using lay-administered

screening scales.

Measures

Violent behavior was assessed using self-re- port measures of fighting and weapon use in the five-year period preceding the PERI in-

terview. The measures are identical to the ones used by Link et al. (1992) and Link and Stueve (1994), and include questions similar to those employed by Swanson et al. (1990). Fighting was assessed by asking, "Have you

been in a physical fight in the past five years?"; weapon use was assessed by asking, "Have you ever had anything in your hands

like a knife, a stick or a gun during a fight?"

Weapon use was placed within the five-year period before the interview by asking, "How old were you the last time you fought with something in your hands?" While these ques-

tions do not fully cover the domain of vio-

lence, previous research has shown that the

associations between mental illness and vari- ous indicators of violence are quite consis- tent (Link et al. 1992).

Psychiatric diagnoses were made using the psychiatrist-administered Schedule for Af- fective Disorders and Schizophrenia-Israel Version (SADS-J). The SADS-J is a modified version of the SADS (see Levav et al. 1993) that yields psychiatric diagnoses according to Research Diagnostic Criteria (RDC) (Spitzer, Endicott, and Robins 1978). We fo- cused on five general types of diagnoses: (1) psychotic disorders, including schizophrenia, schizoaffective, unspecified functional psy- chosis, and major depression with psychosis; (2) bipolar disorders, including bipolar I dis- order, bipolar II disorder, hypomanic disor- der, cyclothymia, and mania;6 (3) major de- pression without psychosis; (4) generalized anxiety disorder; and (5) phobias. To deal with comorbidity among the five categories, we gave primacy to the diagnosis with the highest rank on the above list (1 is highest) in assigning a person to a category. We used

6 The bipolar category includes disorders that were previously called manic-depressive psycho- sis (bipolar I disorder and bipolar II disorder) as well as diagnoses that share features of bipolar disorder but that do not fully meet criteria for one of these diagnostic categories.

This content downloaded from �������������76.183.236.69 on Fri, 04 Dec 2020 22:04:21 UTC�������������

All use subject to https://about.jstor.org/terms

322 AMERICAN SOCIOLOGICAL REVIEW

data from the SADS-I about the onset and

offset of episodes of these disorders to deter-

mine whether a person had the disorder in

question during the five-year period covered

by the questions about violence.

We control for lifetime substance abuse (drug and alcohol) and antisocial personality disorders to determine whether comorbidity

accounts for the association between mental disorder and violent acts.7 While the low fre- quency of these diagnoses is desirable for this inquiry, it is important to recognize that findings regarding these variables should not be generalized to settings in which they are much more common, such as in the United

States.

Psychotic symptoms were assessed using two different instruments: the lay-adminis- tered PERI, and the psychiatrist-adminis-

tered SADS-J. In the PERI, all respondents were asked fixed-format questions about the frequency with which they had experienced each of 13 psychotic symptoms in the past year. This 13-item measure has been shown to be reliable in both community and patient samples (Dohrenwend et al. 1980). Evidence for the validity of the scale comes from a dis- criminant function analysis showing that this scale is effective in separating people with psychosis from people with other diagnoses (Shrout, Dohrenwend, and Levav 1986). The three symptoms taken from this larger scale to measure threat/control-override were mea-

sured by: "How often have you felt: (1) that your mind was dominated by forces beyond your control, (2) that thoughts were put into your head that were not your own, and (3) that there were people who wished to do you harm." The remaining 10 items were used to create a scale of other psychotic symptoms (see Appendix A).

In administering the SADS-J, the clinician begins by asking a series of questions de- signed to assess whether a person has any

psychotic symptoms. If the clinician con- cludes that evidence of psychosis exists, ad- ditional questions are asked to determine a

7 A history of violent behavior can contribute to a diagnosis of antisocial personality disorder. Our control for antisocial personality represents a conservative step implemented to address the claim that mental disorder and violence are linked only because of comorbidity with antisocial per-

sonality.

precise diagnosis, to assess when an episode

started and stopped (but not individual symp-

toms), and to categorize symptoms into spe-

cific types. The clinicians identified 64 per- sons as having symptoms that might indicate

a psychotic disorder. For these individuals

we were able to classify psychotic symptoms

into threat/control-override and other psy-

chotic symptoms. We implemented this cat- egorization for disorders occurring during the five-year period covered by the questions about violent behaviors. Persecutory delu-

sions and delusions of being controlled were

operationalized as threat/control-override symptoms. Other psychotic symptoms were operationalized by summing somatic, reli- gious, grandiose, or nihilistic delusions with-

out persecutory content along with visual

hallucinations, thought broadcasting, and

other symptoms that did not involve delu- sions of control or of persecution.

Each of these approaches assesses general

categories of symptom experience and does not seek to capture specific details of symp- tom content, such as exactly who is control- ling one's mind or who is out to do one harm. Thus, specifics such as believing that God is demanding that one kill one's next- door neighbor or that one's landlady is poi- soning the air in one's apartment are not in- cluded in the operationalization. Rather a person responding to "God's demands" would fit into the more general category of

symptom in the PERI of "having one's mind dominated by forces beyond your control"; a person who believed his apartment was being poisoned would fit the symptom in the PERI of feeling "there were people who wished to do you harm." As is typical in the assessment of psychiatric disorders, we combine specific symptom experiences into types of symptoms. Thus, our theorizing is about which types of symptoms are more or less likely to induce violence. Our predic- tion is that TCO-type (threat/control-over- ride) symptoms include specific symptom experiences that are likely to induce vio- lence, whereas the specific experiences un- derlying the other types of symptoms are not as likely to induce violence.

Our two approaches to symptom assess- ment (PERI and SADS-J) deal with measure- ment error in different ways (Dohrenwend 1990). The fixed-format, lay-administered

This content downloaded from �������������76.183.236.69 on Fri, 04 Dec 2020 22:04:21 UTC�������������

All use subject to https://about.jstor.org/terms

MENTAL DISORDERS AND VIOLENT BEHAVIOR 323

Table 1. Five-Year Prevalence of Fighting and Weapon Use among People with Five Selected Types

of Psychiatric Disorders Compared with Controls: Israel, 1983 to 1988

Involved Psychiatric Diagnosis Unweighteda Weighteda in a Fightb Used Weaponsb

Psychotic disorders, including major 29 19 28.8** 11.1* depression with psychosis

Bipolar disorders 123 81 23.7*** 6.7

Major depression without psychosis 519 388 11.0 1.7

Generalized anxiety disorders 319 328 10.9 1.3

Phobias 112 112 7.8 .8

None of the above disorders 1,576 1,778 8.1 1.1

Total number of cases 2,678 2,706

Source: Adapted from Stueve and Link (1997).

a Missing cases for weighted and unweighted data are 35 and 63, respectively.

b Weighted percentage.

**p < .01 **8 < .001 (two-tailed tests)

items in the PERI rely on psychometric prin-

ciples related to asking multiple questions from the same domain and summing across

like items to more closely approach true

scores. The SADS-J uses the skill of the cli- nician to "probe" and "cross-examine" re-

spondents to determine whether a symptom

is truly present. Neither method is perfect for

identifying the symptoms of interest, as each

method carries its own possibilities for bias and error. Because each method has its own rationale for dealing with measurement error

and also has different potential biases, we give equal weight to each measure in an ef-

fort to create the best composite measure possible.

We operationalized threat/control-over- ride symptoms and other psychotic symp-

toms by first creating scales for each symp- tom-type from the PERI and the SADS-I. Next, we placed each of the four scales in a metric that varied from a possible low of 0 to a possible high of 5 by dividing each scale by the appropriate constant. Finally, we added the PERI scale for each symptom- type (threat/control-override and other psy- chotic symptoms) to the SADS-I scale for each symptom-type to create two scales that could theoretically vary from 0 if none of the symptoms was present to 10 if all were. Scored in this way, the mean for threat/con- trol-override symptoms was .489 (range 0 to 9.17; S.D. = .746); the mean for other psy-

chotic symptoms was .348 (range 0 to 4.61;

S.D. = .515).

Control measures include gender (female = 1, male = 0), years of education, ethnicity

(North African parentage = 1, European par-

entage = 0), age, and a 15-item Crowne-

Marlowe scale (Crowne and Marlowe 1960). The Crowne-Marlowe scale is included to

control social desirability response bias re- lated to the reporting of socially undesirable

characteristics such as violent behavior and psychotic symptoms.

RESULTS

Is There an Association between Psychiatric Diagnosis and Violent Behaviors in Israel?

The results in Table 1, which were adapted from a previous study using these data

(Stueve and Link 1997), show that the five- year prevalence of fighting and weapon use is substantially and significantly elevated

among people diagnosed with psychotic and bipolar disorders. Depending on the com- parison, people with these diagnoses are from about three to over nine times more likely than people with none of the disorders assessed to have engaged in a violent behav- ior. In contrast, people diagnosed with major

depression (without psychosis), generalized anxiety disorders, and phobias are no more

This content downloaded from �������������76.183.236.69 on Fri, 04 Dec 2020 22:04:21 UTC�������������

All use subject to https://about.jstor.org/terms

324 AMERICAN SOCIOLOGICAL REVIEW

Table 2. Logistic Coefficients from the Regression of Violent Behaviors on Psychiatric Diagnoses and

Other Independent Variables: Israel, 1983 to 1988

Violent Behavior

Fighting Weapon Use

Independent Variables Model 1 Model 2 Model 1 Model 2

Psychotic/bipolar disorders 1.21*** 1.08*** 1.92** 1.79***

(.27) (.29) (.46) (.45) Sex (1 = female; 0 = male) -.98*** - -1.31**

(.21) (.40)

Ethnicity (1 = North African; 0 = European) .50** -.26 (.19) (.45)

Age (in years) -.10*** -. I I* (.03) (.06)

Education (in years) -.1o*** -.19** (.03) (.07)

Crowne-Marlowe (social desirability) scale -.72* -1.54** (.30) (.51)

Substance abuse disorder (1 = yes; 0 = no) 1.01 *** 1.68** (.26) (.54)

Antisocial personality disorder (1 = yes; 0 = no) .95** 1.27* (.31) (.53)

Chi-square 10.1 113.1 7.1 50.3

Degrees of freedom 1 8 1 8

Note: Numbers in paretheses are standard errors; unweighted N = 2,678.

*p < .05 **p < .01 ***p < .001 (two-tailed tests)

likely than controls to have engaged in these

types of violence.8 Because the number of cases in the psychotic disorder group is small and because the prevalence of violence in this group is similar to that in the bipolar dis- order group, we combine these groups in our subsequent statistical analyses into a psy- chotic/bipolar disorder group. We also com- bine respondents diagnosed with major de- pression (without psychosis), generalized anxiety disorders, and phobias with the con- trol group, as differences in reported vio- lence between these individuals and the con-

trols are not statistically significant.

Is the Association between Diagnosis and Violent Behavior Spurious?

Results in Table 1 suggest the possible inde- pendent influence of psychosis/bipolar disor-

8 Ten individuals were classified as having ma- jor depression with psychosis. Because this group was small, we decided not to form a separate group for this diagnosis. Three of the 10 reported having been in a fight, and none of the 10 re- ported having used a weapon.

der on violence. However, the prevalence of

violent behaviors is much greater among

people with substance abuse and antisocial diagnoses, and comorbidty between these di- agnoses and psychotic/bipolar disorders does

exist. The prevalences of antisocial person- ality and substance abuse among people with psychotic/bipolar disorders are 3.4 percent

and 10.3 percent respectively, whereas in the general population the corresponding figures

are 1.7 percent and 4.9 percent. This indi- cates the need to statistically adjust the asso- ciation in Table 1 for comorbid substance abuse and antisocial personality disorder. Model 2 in Table 2 shows the association be- tween psychiatric diagnosis and violent be- haviors while holding constant variables that

might render such an association spurious, including antisocial personality, substance abuse, sociodemographic variables, and the

Crowne-Marlowe (social desirability) scale. The control variables in Table 2 relate to

the indicators of violence in a predictable fashion. Gender, education, age, the Crowne- Marlowe scale, and diagnoses of substance abuse/antisocial personality all produce a

This content downloaded from �������������76.183.236.69 on Fri, 04 Dec 2020 22:04:21 UTC�������������

All use subject to https://about.jstor.org/terms

MENTAL DISORDERS AND VIOLENT BEHAVIOR 325

40

Threat/Control-Override

Symptoms

I Low (N = 1,922)

30 D Medium Low (N = 550)

Medium High (N = 161)

* High (N = 73)

-6

12.5

......... ..............,,,_

6.9_

0~~~~~~~~~~

. . . . . . . . . .,...,_ ............... .....

.........._ ......................... . ............ _ . . . .. .. .. . . ..........,.,_ O . ,..,. , ,, ,,.,I ............................. .. Fighting Weapon Use

Figure 1. Threat/Control-Override (TCO) Symptoms and Five-Year Prevalence of Violence: Israel,

1983 to 1988

predicted pattern of results with the violence

indicators. In addition, respondents from

North African backgrounds were signifi- cantly more likely than respondents from European backgrounds to report fighting, but

not weapon use.

Table 2 also shows that associations be-

tween psychiatric diagnosis and indicators of violent behavior remain significant even when controls for comorbidity and other po- tential confounding variables are entered. In- deed, the regression coefficients for psychi- atric diagnosis change very little from Model 1 to Model 2 of each analysis, and when

exponentiated, yield adjusted odds-ratios (fighting 2.94; weapon use 5.99) that are

only slightly lower than the unadjusted odds- ratios (fighting 3.35; weapon use 6.82). The fact that several potent predictors of violence failed to render the association between psy- chiatric diagnosis and violent behaviors spu- rious reinforces the possibility that there is something about psychotic bipolar disorders per se that lead to violent behavior.

Do Threat/Control-Override Symptoms Account for the Association between Psychiatric Diagnoses and Violent Behaviors?

Figure 1 shows the bivariate association be-

tween four levels of threat/control-override symptoms and the two indicators of violence (fighting and weapon use). The prevalence of violence increases in a monotonic fashion as scores on the TCO scale increase. The asso-

ciation is also strong, with those who score high on threat/control-override symptoms

being much more likely than those who score low to have engaged in fighting and weapon

use. Moreover, this monotonic "dose-re- sponse" association holds in groups defined by gender, education, ethnicity, substance- abuse diagnosis, antisocial personality diag- nosis, and patient status.

The scale measuring other psychotic symptoms also shows a significant and rela- tively strong monotonic association with both indicators of violent behavior (not

This content downloaded from �������������76.183.236.69 on Fri, 04 Dec 2020 22:04:21 UTC�������������

All use subject to https://about.jstor.org/terms

326 AMERICAN SOCIOLOGICAL REVIEW

Table 3. Logistic Coefficients from the Regression of Violent Behaviors on Psychiatric Diagnoses,

Threat/Control-Override Symptoms, and Other Independent Variables: Israel, 1983 to 1988

Violent Behavior

Fighting Weapon Use

Independent Variables Model 1 Model 2 Model 3 Model 1 Model 2 Model 3

Psychotic/bipolar disorders 1.21*** .48 .68* 1.92** .70 1.23* (.27) (.33) (.30) (.46) (.75) (.53)

Threat/control-override symptoms - .62*** .54*** - .71*** .58** (.12) (.12) (.22) (.18)

Other psychotic symptoms .02 -.17 .07 -.15

(.17) (.17) (.30) (.25) Sex (1 = female; 0 = male) -1.06*** - -1.48***

(.22) (.42) Ethnicity (1 = North African; 0 = European) .51 ** - -.13

(.19) (.45) Age (in years) -.09** - -.10

(.03) (.06)

Education (in years) -.08* - -.13 (.03) (.07)

Crowne-Marlowe (social desirability) scale - -.58 - -1.37** (.30) (.47)

Substance abuse disorder (1 = yes; 0 = no) .90*** - 1.57*** (.26) (.47)

Antisocial personality disorder - .94** - - 1.27*

(1 = yes; 0 = no) (.31) (.51)

Chi-square 10.1 42.6 129.4 7.1 19.4 56.2

Degrees of freedom 1 3 10 1 3 10

Note: Numbers in paretheses are standard errors; unweighted N = 2,678.

*p < .05 **p < .01 ***p < .001 (two-tailed tests)

shown). However, consistent with our defi-

nitions-based prediction, Table 3 shows that the association between other psychotic

symptoms and violence is attributable to the fact that they covary with TCO symptoms.

For both fighting and weapon use, the asso-

ciation between other psychotic symptoms

and violence diminishes to nonsignificance when threat/control-override symptoms are controlled. In contrast, threat/control-over-

ride symptoms remain strongly associated

with the violence indicators, even when the scale of other psychotic symptoms is held constant. Moreover, in analyses (not shown)

we determined that separate measures of threat and of control-override independently predicted fighting and weapon use. Finally, as Model 3 in each analysis shows, TCO symptoms remain a significant predictor of violent behaviors even when substance abuse, antisocial personality, and demo- graphic variables are held constant.

Next we consider the consequences of holding threat/control-override symptoms

constant on the association between psychi-

atric diagnosis and violent behaviors

(Model 1 of each analysis compared with

Model 2). In sharp contrast to Table 2, in

which many potential confounders were en-

tered with little resulting change in the re- gression coefficients for psychiatric diagno-

sis, Table 3 shows substantial declines in the coefficients for psychiatric diagnoses when TCO symptoms are controlled. The adjusted

odds-ratios for diagnosis decreases from 3.35 to 1.62 for fighting and from 6.82 to 2.01 for weapon use from Model 1 to Model 2.

Interestingly, in Model 3, when the same

control variables that had little effect on the diagnosis/violence association in Table 2 are

entered, the coefficient for diagnosis in- creases. This is due to a pattern of suppres- sion involving diagnosis, psychotic symp-

This content downloaded from �������������76.183.236.69 on Fri, 04 Dec 2020 22:04:21 UTC�������������

All use subject to https://about.jstor.org/terms

MENTAL DISORDERS AND VIOLENT BEHAVIOR 327

toms and the control variables.9 The signifi- cant effect of diagnosis in Model 3 indicates that variables other than TCO symptoms are

also involved in the diagnosis/violence asso- ciation and need to be assessed in future re- search. However, even with the suppression effect the coefficients for diagnosis are sub- stantially smaller in Model 3 than in Model

1, dropping by 44 percent for fighting and 36 percent for weapon use. These sharp declines in the adjusted odds-ratios indicate that threat/control-override symptoms play a sub- stantial role in explaining why diagnosis is related to violence.

Because we operationalized threat/control-

override symptoms in two ways, we checked whether the same pattern of results was ob- tained using each operationalization sepa- rately (not shown). We found that whether indicated by the PERI or the SADS-I, threat/ control-override symptoms are associated with violence when other psychotic symp- toms are held constant, but the opposite is not true. This means that our results are ro- bust regarding variations in the method of measuring our key independent variable. 10

Finally, holding threat/control-override symptoms constant affects the association between violence and diagnosis and not the association between violence and the "con-

trol" variables. When we compare the effects of gender, education, age, ethnicity, the Crowne-Marlowe scale, substance abuse, and antisocial personality on violence before (Table 2) and after (Table 3) threat/ control-

override symptoms are entered, we find com- paratively little change in the magnitude of the coefficients for these variables. For coef-

ficients that were initially significant (N = 13), and thus at least of modest size, the av-

9 Three variables-years of education (r = -.006), substance abuse (r = .046), and antisocial personality (r = .024)-all have relatively small bivariate associations with psychotic/bipolar di- agnosis. At the same time, each of these variables is significantly related to both TCO symptoms and violent behaviors. And, of course, TCO symptoms are strongly related to a psychotic/bi- polar diagnosis. As a consequence of this set of relationships, when TCO is held constant the ef- fect of the control variables is to make the asso- ciation between psychotic/bipolar disorders and violence larger.

10 Tables reporting these results are available from the first author.

erage percentage change in the associations is only 11.0 percent (range 0 to 31.5 percent).

Compare this to relatively sharp declines in coefficients for diagnosis of 60.3 percent

(fighting) and 63.5 percent (weapon use)

when TCO symptoms are added in Table 3 (Model 1 compared to Model 2). This com- parison indicates that threat/control-override

symptoms account for only the associations they were predicted to explain (that between

psychiatric diagnosis and violent behavior).

DISCUSSION

A growing body of research documents an association between indicators of mental ill- ness and indicators of violence. The scope and meaning of the association has been questioned, however, with respect to whether it generalizes to other contexts and whether it is spurious, due solely to comorbidity with substance abuse and antisocial personality disorder. More important, theories and tests

designed to explain why people with mental illnesses exhibit elevated levels of violence were lacking. We have addressed these issues using a large epidemiological study con- ducted in Israel that allowed us to test pre-

dictions derived from the idea that defini- tions of situations have real consequences even when the definitions are shaped by de- lusions or hallucinations.

We found an association between indica- tors of mental illness and indicators of vio- lence in the very different (when compared with the United States) social context of Is- rael. The association was little affected when adjusted for sociodemographic variables, a social desirability scale, and the presence of a lifetime diagnosis of substance abuse or an- tisocial personality disorder. We also found a strong dose-response association between threat/control-override symptoms and indica- tors of violence that held within groups de- fined by gender, education, ethnicity, age, substance-abuse diagnosis, antisocial person- ality diagnosis, and patient status. Consistent with our prediction, when entered either as a composite variable or as separate threat and control-override variables, threat/control- override symptoms remained strongly asso- ciated with violence even controlling for other psychotic symptoms. The threat/con- trol-override symptoms also played a large

This content downloaded from �������������76.183.236.69 on Fri, 04 Dec 2020 22:04:21 UTC�������������

All use subject to https://about.jstor.org/terms

328 AMERICAN SOCIOLOGICAL REVIEW

role in explaining the association between

selected psychiatric diagnoses and violent

behaviors, but they did not mediate associa- tions between other variables and violence.

That is, holding constant threat/control-over- ride symptoms did not appreciably affect co-

efficients for sociodemographic variables, social desirability, or substance abuse/antiso- cial personality disorder.

Potential Limitations

Possible limitations in our study are related

to the validity of our measures and to the cross-sectional, nonexperimental nature of our design. Our inquiry relies on self-report measures that may not validly assess symp-

toms or violence because of the under- reporting of socially undesirable informa-

tion. While this is possible, we doubt this se- riously biases our results for three reasons.

First, we controlled for social desirability (using the Crowne-Marlowe scale) and found that doing so did not appreciably affect the association between psychiatric diagnoses and indicators of violence. Second, studies checking self-reports of violence-related variables (e.g., arrest) suggest that mis- reporting is relatively rare and that official data may represent a greater problem of undercounting than do self-reports (Link et al. 1992). Third, both self-report data and of- ficial records were available in the Link et

al. (1992) study and showed consistent evi- dence regarding the association between mental illness and violence.

Another validity concern is whether the threat/control- override scale truly measures psychotic symptoms or whether it is "reality based." In a classic investigation related to this issue, Lemert (1962) points out that when the people around a paranoid person actually do conspire and plot in an effort to control the afflicted person's behavior, they simultaneously reinforce and provide a real- ity base for that person's paranoid beliefs. Rather than challenge the validity of our

measure of threat/control-override symp- toms, Lemert's insight concerning the man- ner in which social contexts provide a reality basis for some paranoid beliefs suggests the need for further study of the interplay be- tween psychotic symptoms and social con- texts (Estroff et al. 1994; Hiday 1995).

There is another form of the "reality- based" argument, however, that would deny

the importance of threat/control-override symptoms entirely. According to this view, our measure taps not symptoms but a gen-

eral distrust of others or a reality-based as- sessment of the actual level of threat in the social context that has nothing to do with

psychotic symptoms. Kendler et al. (1996)

indicate that in community surveys, many positive responses to scale items measuring

"psychotic symptoms" are in fact "realistic" and not symptoms at all. According to Kendler et al., the only way to avoid such false positive symptoms is to use clinicians who can skillfully separate true symptoms from impostors. Based on this reasoning, the distrust/true-threat interpretation is more plausible when applied to the fixed-format lay-administered PERI questions than to the measures derived from the psychiatrist-ad- ministered SADS-I. It is instructive that the SADS-I-generated symptoms showed the same pattern of association with violent be- havior when used alone. In addition, the threat/control-override measure explains (mediates) the association between indica- tors of mental illness and violence but not the association between other variables and violence. If the measure captured a more general distrust or embeddedness in danger- ous situations, one would expect it to ex- plain some part of these other associations.

The fact that it explains only the association our theory predicts supports our interpreta- tion of the measure as an indicator of a par- ticular type of psychotic symptom.

With regard to the cross-sectional, non- experimental design, two generic causal-in- ference issues can be raised-reverse causa- tion and unmeasured confounding. Our cross-sectional design does not allow us to

definitively place the occurrence of violence and symptoms in temporal order and there- fore to rule out the possibility of reverse causation. Further, the observational, non- experimental nature of our design means that we cannot rule out the possibility that some unmeasured third variable causes both psychosis and violence. On close inspec- tion, however, alternative explanations fall short in accounting for the full pattern of associations we have reported. It is difficult to explain why threat/control-override

This content downloaded from �������������76.183.236.69 on Fri, 04 Dec 2020 22:04:21 UTC�������������

All use subject to https://about.jstor.org/terms

MENTAL DISORDERS AND VIOLENT BEHAVIOR 329

symptoms are related to violent behavior when other psychotic symptoms are held

constant and why the reverse is not true. Reverse causation would predict a general

elevation across a variety of psychotic symptoms, including the symptoms in the

other psychotic symptoms scale. Moreover, the fact that separate measures of threat and control-override independently predict vio- lent behaviors means that a reverse causa- tion explanation needs to account for why both of these components are related to vio- lence and none of the other psychotic symp- toms is.

The observed pattern of results also leads

us to question the possibility that some un- measured confounding variable accounts for both psychosis and violent behaviors. We measured and controlled many possible

"confounders" that were strongly linked to violent behavior and found that they had little bearing on the association between mental illness and violence. Of course, it is possible that we missed the key confounder. However, the fact that several obvious po-

tential confounders were unable to account for even part of the association between psychiatric diagnoses and violent behaviors makes it unlikely that adding other con-

founding variables would strikingly alter our interpretations. Moreover, any unmea-

sured confounder would need to be quite powerful to account for the relatively strong dose-response association between threat/ control-override symptoms and violence. Fi- nally, why would such a confounder pro- duce a stronger association between threat/ control-override symptoms and violent be- haviors than between other psychotic symp- toms and violent behaviors? The specificity of our prediction along with the results sup- porting it demand the same kind of specific- ity from alternative explanations.

Implications

Our inquiry offers both theory and evidence pointing to the central role of threat/control- override symptoms for explaining the men- tal illness/violence association and suggests that these symptoms operate by influencing how situations are defined. As such, our re- sults are consistent with Heimer's (1997) claim that definitions are a common pathway

linking many circumstances, including spe-

cific psychotic symptoms, to violence.

We leave to future research, however, the task of assessing the specific processes and mechanisms involved. For example, detailed

measures of threat/control-override symp-

toms would allow one to assess whether, for example, a delusion of persecution involved

a specific person and whether that specific

person was the person targeted in a violent attack and whether mind control must in- volve explicit prescriptions for violence.

Such research would provide further evi- dence bearing on our interpretation of the as- sociation between mental illness and vio-

lence and also offer the possibility of identi- fying interventions that might reduce the im- pact of symptoms on violence.

Our inquiry also challenges the medical/ diagnostic approach to understanding the mental illness-violence connection. While

some diagnoses involve an elevated risk of violence, our results show that the threat/con-

trol-override symptoms have primacy over diagnostic distinctions in explaining vio- lence. When threat/control-override symp- toms are held constant, the effect of diagno- sis is substantially reduced while the threat/ control-override symptoms remain powerful predictors of violence when diagnosis is con- trolled. Moreover, whereas a focus on the meaning of symptoms for the definition of situations allows us to understand why some- one experiencing threat/control-override symptoms might be prone to violence, sole reliance on diagnosis does not inform our understanding. At least in some situations, social science theorizing furthers understand- ing of the implications of specific psychiatric symptoms regardless of their place in or rel- evance to a diagnostic system or construct. This is especially the case when the phenom- enon to be explained is social (like violence) and is a potential consequence of the disor- der. Diagnostic categorizations were formed to characterize and define disorders, not to account for the social consequences of the symptoms they encompass. Here, social sci- ence concepts can be usefully applied to psy- chiatric symptoms, as our application of ideas about the definition of situations to issues of mental illness and violence suggests.

Our analysis also has implications for how we think about the risk of violence posed by

This content downloaded from �������������76.183.236.69 on Fri, 04 Dec 2020 22:04:21 UTC�������������

All use subject to https://about.jstor.org/terms

330 AMERICAN SOCIOLOGICAL REVIEW

people with mental illnesses. Recall that a central aspect of the stereotype of mental ill- ness involves dangerousness, and that this stereotype is a major reason for the rejection and exclusion of people with mental ill- nesses. By quantifying the magnitude of the risk and by specifying the violence-inducing component of mental illness, our study calls for a sharp modification of such stereotyped views. Not all mental illnesses are associated with an elevated risk of violent behaviors. Our results suggest that major depression (without psychosis), generalized anxiety dis- order, and phobias are not related to indica-

tors of violence. Moreover, even among people with psychotic or bipolar disorders, it is primarily those individuals reporting threat/control-override symptoms who were at greater risk of violent behavior; once threat/control-override symptoms are held constant, other psychotic symptoms have no significant association with such behaviors.

Finally, the magnitude of the association is modest in comparison to the fear engendered by stereotypes of dangerousness. The odds- ratios relating psychotic/bipolar disorders to violent behaviors are within the range of odds-ratios for other predictors of violence such as age, gender, and education. If a com- munity were to use the risk of violence as the sole basis for the exclusion of people with mental illnesses, such a community might just as well exclude men in favor of women, teenagers in favor of people who are 50 years old, and grade school graduates in favor of college graduates. Even for people who ex- perience the highest levels of threat/control- override symptoms, our results suggest that only a minority engage in fighting (37.4 per- cent) or weapon use (13.3 percent). In sum, our results support neither the extreme fear associated with the mental illness stereotype nor its broad application to all persons with mental illnesses.

In light of sociological insights derived

from the Thomas Theorem, our analysis has

illuminated the nature of the association be-

tween mental illnesses and violent behaviors

and challenged the destructive stereotype

that haunts people who suffer from mental

illnesses.

Bruce G. Link is Professor of Public Health at Columbia University and a Research Scientist at New York State Psychiatric Institute. His re-

search interests include the impact of labeling and stigma on people with mental and physical illnesses, the role of social factors as fundamen- tal causes of disease, public conceptions of men- tal illnesses and the relationship between mental illness and violence.

John Monahan, a psychologist, holds the Doherty Chair in law at the University of Vir- ginia, where he is also Professor of Psychology and Legal Medicine. In 1998 he was elected to membership in the Institute of Medicine in the National Academy of Sciences. For the past 10 years, he has directed the Research Network on Mental Health and the Law of the John D. and

Catherine T. MacArthur Foundation.

Ann Stueve is on the faculty of Epidemiology at the Joseph L. Mailman School of Public Health at Columbia University. Trained in sociology and psychiatric epidemiology, her work focuses on the design and evaluation of HIV prevention pro- grams with high risk populations; the causes and consequences of violence, substance use, and early unprotected sex among urban adolescents; the role of psychiatrists in mental health settings; and the relationship between mental illness and violence.

Francis T. Cullen is Distinguished Research Professor of Criminal Justice and Sociology at the University of Cincinnati. He has recently co- authored Combating Corporate Crime: Local Prosecutors at Work; Criminological Theory: Past to Present; and Offender Rehabilitation: Ef- fective Correctional Intervention. His current re- search interests include the measurement of sexual victimization, the impact of social support on crime, and the principles of effective correc- tional rehabilitation programs.

Appendix A. Psychotic Symptoms Scale Separated into "Threat/Control-Override" and "Other Psy-

chotic Symptoms" Subscales

Threat /Control-Override Symptoms Subscale

During the past year ...

1. How often have you felt that your mind was dominated by forces beyond your control?

2. How often have you felt that thoughts were put into your head that were not your own?

3. How often have you felt that there were people who wished to do you harm?

(Continued on next page)

This content downloaded from �������������76.183.236.69 on Fri, 04 Dec 2020 22:04:21 UTC�������������

All use subject to https://about.jstor.org/terms

MENTAL DISORDERS AND VIOLENT BEHAVIOR 331

(Appendix A continued from previous page)

Other Psychotic Symptoms Subscale

During the past year ...

1. How often have you felt that you do not exist at all, that you are dead, dissolved?

2. How often have you heard things that other people say they can't hear?

3. How often have you felt that your unspoken thoughts were being broadcast or transmitted, so that everyone knows what you are thinking?

4. How often have you thought that you were possessed by a spirit or a devil?

5. How often have you had visions or seen things that other people say they cannot see?

6. How often have you felt you have special powers?

7. How often have you thought something odd was going on?

8. How often have you felt your thoughts were taken away from you by some external force?

9. How often have you had ideas or thoughts that nobody else would understand if you talked about them?

10. How often have you seemed to hear your thoughts spoken aloud-almost as if someone standing

nearby could hear them?

Source: The Psychotic Symptoms Scale is from the Psychiatric Epidemiology Research Interview (PERI) (Dohrenwend et al. 1980).

REFERENCES

Akers, Ronald. 1997. Criminological Theories:

Introduction and Evaluation. 2d ed. Los Ange-

les, CA: Roxbury.

. 1973. Deviant Behavior: A Social Learn- ing Approach Belmont, CA: Wadsworth.

American Psychiatric Association. 1994. Diag-

nostic and Statistical Manual of Mental Disor-

ders (DSM-IV). 4th ed. Washington, DC: American Psychological Association.

Athens, Lonnie. 1980. Violent Criminal Acts and

Actors: A Symbolic Interactionist Study. Bos-

ton, MA: Routledge and Kegan Paul. Crowne, Douglas P. and David Marlowe. 1960.

"A New Scale of Social Desirability Indepen- dent of Psychopathology." Journal of Consult- ing and Clinical Psychology 24:349-54.

Dohrenwend, Bruce. 1990. "The Problem of Va-

lidity in Field Studies of Psychological Disor- ders Revisited." Psychological Medicine 20: 195-208.

Dohrenwend, Bruce, Itzhak Levav, Patrick

Shrout, Sharon Schwartz, Guedalia Naveh, Bruce Link, Andrew Skodal, and Ann Stueve. 1992. "Socioeconomic Status and Psychiatric Disorders: The Causation-Selection Issue." Science 255:946-52.

Dohrenwend, Bruce P., Patrick E. Shrout, Gladys Egri, and Frederick Mendelsohn. 1980. "Mea- sures of Nonspecific Psychological Distress and Other Dimensions of Psychopathology in the General Population." Archives of General Psychiatry 37:1229-36.

Eronen, Markku, Jari Tiihonen, and Panu Haola. 1996. "Mental Disorders and Homicidal Be-

havior in Finland." Archives of General Psy-

chiatry 53:497-501. Estroff Sue, Catherine Zimmer, William

Lachicotte, and Julia Benoit. 1994. "The Influ- ence of Social Networks and Social Support on Violence by Persons with Serious Mental Ill- ness." Hospital and Community Psychiatry 45:

669-79. Heimer, Karen. 1997. "Socioeconomic Status,

Subcultural Definitions, and Violent Delin-

quency." Social Forces 75:799-833. Hiday Virginia. 1995. "The Social Context of

Mental Illness and Violence." Journal of Health and Social Behavior 36:122-37.

Hodgins, Sheilagh, Saranoff Mednick, Patricia

Brennan, Fini Schulsinger, and Marianne Engberg. 1996. "Mental Disorder and Crime: Evidence from a Danish Birth Cohort." Ar- chives of General Psychiatry 53:489-96.

Kendler, Kenneth, Timothy Gallagher, Jamie Abelson, and Ronald Kessler. 1996. "Lifetime Prevalence, Demographic Risk Factors, and Diagnostic Validity of Nonaffective Psychosis as Assessed in a US Community Sample: The National Comorbidity Study." Archives of Gen- eral Psychiatry 53:1022-31.

Klassen, Diedre and William O'Connor. 1989. "Assessing the Risk of Violence in Released Mental Patients: A Cross Validation Study." Psychological Assessment 1:75-81.

Lemert, Edwin. 1962. "Paranoia and the Dynam- ics of Exclusion." Sociometry 25:2-20.

Levav, Itzhak, Robert Kohn, Bruce P. Dohrenwend, Patrick E. Shrout, Andrew E. Skodol, Sharon Schwartz, Bruce Link, and Guedalia Naveh. 1993. "An Epidemiological Study of Mental Disorders in a 10-Year Cohort of Young Adults in Israel." Psychological

This content downloaded from �������������76.183.236.69 on Fri, 04 Dec 2020 22:04:21 UTC�������������

All use subject to https://about.jstor.org/terms

332 AMERICAN SOCIOLOGICAL REVIEW

Medicine 23:691-707.

Link, Bruce G., Howard A. Andrews, and Francis T. Cullen. 1992. "The Violent and Illegal Be- havior of Mental Patients Reconsidered." American Sociological Review 57:275-92.

Link, Bruce G., Francis T. Cullen, James Frank, and John F. Wozniak. 1987. "The Social Re- jection of Former Mental Patients: Understand- ing Why Labels Matter." American Journal of Sociology 92:1461-1500.

Link, Bruce and Ann Stueve. 1994. "Psychotic Symptoms and the Violent/Illegal Behavior of Mental Patients Compared to Community Con- trols." Pp. 137-60 in Violence and Mental Dis- order, edited by J. Monahan and H. Steadman.

Chicago, IL: University of Chicago Press. Link, Bruce and Ann Stueve. 1995. "Evidence

Bearing on Mental Illness as a Possible Cause

of Violent Behavior." Epidemiologic Reviews 17:172-80.

Luckenbill, David. 1977. "Criminal Homicide as Situated Transaction. Social Problems 25:176- 86.

Merton, Robert K. [1948] 1968. "The Self-Ful- filling Prophecy." Pp.475-92 in Social Theory

and Social Structure, edited by R. K. Merton. New York: Free Press.

. 1995. "The Thomas Theorem and the Matthew Effect." Social Forces 74:379-422.

Monahan, John. 1992. "Mental Disorder and Vio- lent Behavior: Attitudes and Evidence." Ameri- can Psychologist 47:5 1 1-2 1.

Rabkin, Judith. 1979. "Criminal Behavior of Dis- charged Mental Patients." Psychological Bul- letin 86:1-27.

Robertson, Graham. 1988. "Arrest Patterns among Mentally Disordered Offenders." Brit- ish Journal of Psychiatry 153:313-16.

Shah, Babubhai, Beth Barnwell, P. Nileen Hunt, and Lisa Lavange. 1992. SUDAAN User's Manual. Release 6.0. Research Triangle Park,

NC: Research Triangle Institute. Shrout, Patrick E., Bruce P. Dohrenwend, and

Itzhak Levav. 1986. "A Discriminant Rule for Screening Cases of Diverse Diagnostic Types: Preliminary Results." Journal of Consulting and Clinical Psychology 54:314-19.

Spitzer, Robert, Jeanne Endicott, and E. Robins. 1978. "Research Diagnostic Criteria: Rationale

and Reliability." Archives of General Psychia-

try 35:773-82.

Steadman, Henry J. 1981. "Critically Assessing the Accuracy of Public Perceptions of the Dan-

gerousness of the Mentally Ill." Journal of

Health and Social Behavior 22:310-16.

Steadman, Henry J., Edward P. Mulvey, John Monahan, Pamela C. Robbins, Paul S. Apple-

baum, Thomas Grisso, Loren Roth, and Eric Silver. 1998. "Violence by People Discharged from Acute Psychiatric Inpatient Facilities and

by Others in the Same Neighborhoods." Ar- chives of General Psychiatry 55: 393-401.

Stueve, Ann and Bruce Link. 1997. "Violence and Psychiatric Disorders: Results from an Epi- demiological Study in Israel." Psychiatric Quarterly 68:327-42.

Sutherland, Edwin. 1947. Principles of Criminol- ogy. 4th ed. Chicago, IL: Lippincott.

Swanson, Jeffrey. 1994. "Mental Disorder, Sub- stance Abuse and Community Violence: An Epidemiological Approach." Pp. 101-136 in Violence and Mental Disorder, edited by J. Monahan and H. Steadman. Chicago, IL: Uni- versity of Chicago Press.

Swanson, Jeffrey W., Charles E. Holzer III, Vijay K. Ganju, and Robert Tsutomu Jono. 1990.

"Violence and Psychiatric Disorder in the Community: Evidence from the Epidemiologic Catchment Area Surveys." Hospital and Com- munity Psychiatry 41:761-70.

Swanson, Jeffrey, Randy Borum, Marvin S. Swartz, and John Monahan. 1996. "Psychotic Symptoms and Disorders and the Risk of Vio- lent Behaviour in the Community." Criminal Behaviour and Mental Health 6:317-38.

Tardiff, Kenneth. 1996. Assessment and Manage- ment of Violent Patients. 2d ed. Washington, DC: American Psychiatric Press.

Teplin, Linda. 1984. "Criminalizing Mental Dis- order: The Comparative Arrest Rate of the Mentally Ill." American Psychologist 39:794- 803.

Thomas, W. I. and Dorothy Swain Thomas. 1928. The Child in America: Behavior Problems and

Programs. New York: Knopf. Wessely, Simon. 1997. "The Epidemiology of

Crime, Violence and Schizophrenia." British Journal of Psychiatry 170(32 supp.):8-1 1.

ERRATUM

VOLUME 64, NUMBER 1, FEBRUARY 1999

Quadagno, Jill. Creating a Capital Investment Welfare State: The New American Exceptionalism.

Page 5, Figure 1: The heading of the right-most column in the figure should read "Capital In-

vestment Welfare State."

This content downloaded from �������������76.183.236.69 on Fri, 04 Dec 2020 22:04:21 UTC�������������

All use subject to https://about.jstor.org/terms