Mental Illness and Violence: Proof or Stereotype?

Next | Previous | Table of Contents

3. Critical Review: Does Mental Illness Cause Violence?

In the previous chapter, the main statistical relationships reported in the literature between mental illness and violence were summarized. However, understanding these is only the first step in critically assessing the question: Does mental illness cause violence? A strict reasoning process which goes beyond demonstrated statistical associations is required to make judgements concerning causal mechanisms. Epidemiology provides such a framework (Lilienfield and Stoley, 1994) that will be applied to assess whether there is sufficient evidence to support the conclusion that mental illness can be said to cause violence.

The starting point for the discussion is the oft-quoted passage from Monahan and Steadman's now classic literature review published in 1983. Until recently, it has served as the touchstone by which mental health providers and scientists alike have understood the relationship between mental illness and violence.

“The conclusion to which our review is drawn is that the relations between ... crime and mental disorder can be accounted for largely by demographic and historical characteristics that the two groups share. When appropriate statistical controls are applied for factors such as age, gender, race, social class, and previous institutionalization, whatever relations between crime and mental disorder are reported tend to disappear” (cited in  Monahan, 1993, p. 287).

By 1993, Monahan had questioned and indeed reversed this conclusion.

“I now believe that this conclusion is at least premature and may well be wrong for two reasons. First, to control statistically for factors, such as social class and previous institutionalization, that are highly related to mental disorder is problematic. For example, in some cases mental disorder causes people to decline in social class (perhaps because they became psychotic at work) and also to become violent, then to control for low social class is, to some unknown extent, to attenuate the relationship that will be found between mental disorder and violence.... If, in other cases, mental disorder causes people to be repeatedly violent and therefore institutionalized, then to control for previous institutiona- lization also masks, to some unknown degree, the relationship that will be found between mental disorder and violence” (Monahan, 1993, p. 287-288).

Two things are clear from these passages. First, ideas about the relationship between mental illness and violence have changed. Secondly, much of what is found by way of a statistical association between mental disorder and violence will hinge on certain technicalities such as how age, sex, or socio-economic status should be considered in the statistical analysis. The epidemiological framework chosen for this review will clarify these issues as well as highlight the main sources of error in studies of this nature.

Control for Confounding Factors:

Confounding occurs when the effects of two or more factors become mixed in a dataset, making it difficult to see, or actually distorting, the effects of the main study relationship (Last, 1988). The extent to which any variable will be considered to be a confounding factor will depend on the investigator's understanding of the causal mechanism under investigation. For example, Monahan (1993) postulates a mechanism for the causal action of mental illness on violence that places socio-economic status as one intervening step in the hypothesized causal chain between these variables. In Monahan's view, persons who become seriously mentally ill drift downward in socio-economic status as their illness increasingly interferes with their ability to work. When socio-economic status is treated as a consequence of mental illness, Monahan is correct in arguing that statistically removing the effects of socio-economic status could seriously distort the results. Any factor that represents a plausible step in the causal chain under study cannot be considered to be extraneous to the analysis and its effects should not be statistically controlled (Rothman, 1986). Unfortunately, the issue is not so easily settled because, in some cases, downward drift may not occur, or may not be the consequence of a mental illness. Thus, the extent to which controlling for socio-economic status would distort  results would depend, in part, on the population studied. In the presence of uncertainty about the causal mechanism, it might be best to assess the relationship between mental illness and violence with and without statistical controls for socio-economic status, in order to compare the differences in the results.

Monahan incorrectly makes the same argument for other factors such as age, or sex. To be considered to be intervening variables, it would have to be argued that age and sex are caused by mental illness. This is clearly not plausible. But, because all of these factors are known to be statistically associated with violence, they could interfere with our ability to draw valid conclusions about the relationship between mental illness and violence, depending upon how they are represented in the study and comparison groups. For example, if the study group of persons with mental illness also contained a disproportionately high number of young males who were prone to violence, a straight comparison across study and control groups would make it appear that there was a statistical association between mental illness and violence. In fact, this apparent association could be explained as a result of the mixing of the effects of age and sex. To assess the true relationship between mental illness and violence, the confounding effects of age and sex would have to be controlled using one of several statistical techniques designed for this purpose. When critically reviewing the literature, then, it is appropriate to consider factors such as age, sex, or past violence as confounding factors that warrant careful control.

Confounding by Definition:

Another highly pertinent issue is that it may be impossible to define violence independent of mental disorder. For a number of psychiatric disorders described in DSM-III-R (and subsequently in DSM-IV) (American Psychiatric Association, 1987, 1994), violent behaviour is a key diagnostic feature. These include antisocial personality disorder, borderline personality disorder, intermittent explosive disorder, and sexual sadism. For a number of other diagnoses, such as schizophrenia, bipolar disorder, and substance abuse, DSM lists violent behaviour as an associated feature, although it is not a symptom. Thus, manifestations of violence may increase the likelihood that these disorders will be diagnosed (Swanson, Holzer, Ganju, and Jono, 1990). In essence, this is ‘confounding by definition'.

Harry (1985) conducted a content analysis of DSM-I, DSM-II, and DSM-III to assess to what degree diagnostic conceptualizations have changed vis-à-vis the relationship between mental illness and violence. The descriptive paragraphs and diagnostic criteria for each disorder were reviewed for words pertaining to violent behaviours. Unfortunately, no distinction was made between words reflecting violence toward others and violence toward the self. Nevertheless, the results are illuminating. In DSM-I, 6 of the 276 possible disorders (2.17%) were ‘violent'. In DSM-II, 9 of the 337 possible disorders were ‘violent' (2.67%). In DSM-III, this proportion jumped to 162 of 348 or 46.6%! Ninety-one of these (26.15%) included violent words as part of their diagnostic criteria. In DSM-III, the diagnostic groups with the largest number of ‘violent' disorders were substance use disorders, organic mental disorders, affective disorders (typically suicide and self-harm), and those disorders first manifesting before adulthood.

DSM-I appeared in 1952 and DSM-II appeared in 1968 and was used until approximately 1980, when DSM-III was introduced. A number of authors have remarked on the seeming inexplicable reversal in research findings. Early studies tended not to find a relationship between mental illness and violence but studies conducted during the past 15 years have reversed this trend (Link, Andrews, Cullen, 1992; Teplin, 1985). If we consider that this reversal coincides with the adoption of DSM-III, changing conceptualizations of mental disorder that incorporate notions of violence may be at the root of many current findings.

Possible Confounding due to Psychiatric Medications:

A final issue related to confounding is the effect of psychiatric medications on aggression. In a review of clinical aspects of dangerous behaviour, Menuck (1983) describes a number of iatrogenic or paradoxical drug reactions including:

• Tranquilizer-sedatives (e.g. benzodiazepine and barbiturates) may have a disinhibiting effect on affect and behaviour. Paradoxical drug reactions involving violent behaviour have been reported among incarcerated offenders, psychiatric patients, and control group volunteers receiving these drugs.
• Drugs that are capable of producing hypomania may elicit aggressive behaviour by arousal of the central nervous system. Aggressive behaviour has been reported among persons treated with imipramine and amitriptyline, phenylzine, prednisone, and bromocriptine.
• Neuroleptic drugs sometimes increase aggressivity. This phenomenon has been observed during both low and high dose pharmacotherapy.

Even if a statistical relationship could be demonstrated between mental illness and violence in a number of studies, it would still not be clear to what extent the violence would be due to the mental illness or to the psychopharmacological treatments. Within the context of community care and public perceptions of the mentally ill, however, this may be a subtlety of little practical consequence.

Selecting Subjects to Avoid Bias:

The third issue raised in the earlier passages (Monahan, 1993) that warrants careful thought is how to best deal with prior institutionalizations that may be related to violence.

Epidemiologists recognize this difficulty as a problem in selecting subjects for study and they refer to the resulting distortion that can occur as “selection bias” (Rothman, 1986). In the present context, as Monahan's example illustrates, selection bias would occur if study subjects represented sub-groups of the mentally ill that were more likely to be violent.

Selection bias is a universal problem for studies using persons who are undergoing treatment for a mental illness, particularly inpatient hospitalization, or those incarcerated for a violent crime. Because both of these groups may be more prone to violence, studies of treated populations could lead to an exaggerated estimate of the effect of mental illness. Neither sub-group is representative of their respective larger populations (i.e., the mentally ill population and the offender populations).

Selection bias is an insurmountable problem for these studies because there is good evidence that:

(a) Persons with a mental illness who come into contact with health services, particularly hospitals, may be more likely to be violent. For the last two decades, civil committal criteria governing involuntary psychiatric hospitalization have moved away from criteria that were based on a clinical judgement that the individual was in need of treatment toward a standard of dangerousness (Monahan, 1984). In Canada, Alberta was the first to adopt the dangerousness criterion in their Mental Health Act in 1972 (Davis, 1992). Studies that describe an increased incidence of violence among hospitalized psychiatric populations are consistent with this legislative shift toward a dangerousness standard for admission.

(b) Persons coming into contact with the police because of violent or disturbed behaviour may be more likely to have a mental illness. The more restrictive civil committal criteria are considered to have provided an important impetus for the deinstitutionalization of persons with mental illness from mental hospitals to community care settings. For example, in the United States between 1955 and 1975, the resident population of state mental hospitals declined by more than 365,000 persons (Morrissey and Goldman, 1981). In Canada, between 1961 and 1976, 34,000 patients were discharged from psychiatric facilities with similar trends occurring across Europe (Holley and Arboleda-Flórez, 1988), although perhaps at a more gradual pace (Morrissey and Goldman, 1981).

Based on a recent analysis of mental health legislation in Canada (Arboleda-Flórez and Copithorne, 1994, updates, 1995), it is clear that these laws give police officers considerable discretionary power in their handling of mentally ill persons in the community. All provincial mental health legislation provide the officer with two choices. If these individuals are apparently suffering from  a mental disorder and potentially a danger to themselves or others, the officer may transport them to a psychiatric facility for examination and, if appropriate, treatment. Secondly, the officer may proceed with a charge and an arrest. Psychiatric services may then be sought through the various provisions outlined in the Criminal Code for mentally disordered offenders (see Kunjukrishnan and Bradford, 1985 for a description). Since civil committal criteria have been tightened, some have argued that processing through the criminal justice system has become a more expedient means of removing persons with a mental illness from the community while at the same time gaining entrance to mental health services.

The notion that there is a flow between the mental health and criminal justice systems was first described in 1939 by a British researcher named Penrose. Penrose explained the apparent association between mental illness and crime by documenting an inverse relationship in the size of prison and mental hospital populations across 18 European countries. Where prison populations were extensive, mental hospital populations were small, and vice versa. Basic to Penrose's theory is the notion that the volume of persons requiring institutional care remains relatively stable and that these individuals are shunted from mental hospital to correctional facility, and back again, as standards and policies change (Holley and Arboleda-Flórez, 1988).

Weller and Weller (1988) have plotted separation data from psychiatric hospitals against admission data from prisons in England between 1950 and 1985. A correlation coefficient of -94 describes a strong inverse relationship consistent with Penrose's original formulations. Stated another way, knowledge of the psychiatric bed population over this period would have permitted predictions as to the scope of prison populations with only 11.6% of the prison population remaining unaccounted for in the prediction. These authors argue that it is difficult to put forward convincing explanations for such a strong relationship except by postulating that psychiatric hospitals have decanted their patients to prisons. Studies that (a) fail to show differences in violent behaviour of civilly committed psychiatric patients compared to forensic patients (e.g. Beran and Hotz, 1984), (b) show high arrest/conviction rates among persons with mental illness (e.g. Hodgins, 1992; Lindqvist and Allebeck, 1990), and (c) show a high prevalence of mental illness among incarcerated offenders (e.g. Arboleda-Flórez, 1994; Bland, Newman, Dyck, and Orn, 1990; Gingell, 1991) all support the thesis that a sub-population of persons are moving between mental health and criminal justice systems.

The high prevalence of mental illness among incarcerated populations in Canada (e.g. Arboleda-Flórez, 1994; Bland et al., 1990) have been explained in a number of specific ways. Mentally disordered offenders may be arrested at a disproportionately high rate  compared to non-mentally disordered offenders. Persons with mental illness may also be less skilful at crime or more easily caught. Or, once arrested, they may be more likely to plead guilty because of inability to pay, or inability to understand legal representation (Davis, 1992).

Given the wide range of plausible non-etiological explanations for the high prevalence of mental illness among incarcerated populations or the high incidence of violence among psychiatric patients, it is clear that only studies focussing on unselected samples of persons defined as mentally ill or violent can be used to derive etiological inferences. It is unfortunate that recent reviews (e.g. Monahan, 1993; Torrey, 1994) have failed to appreciate the limitations of these bodies of research for drawing etiological inferences.

Gunn (1977, p. 317) has noted:

“....Most discussions on the mentally abnormal offender concentrate on either those who are resident in hospitals or prisons, or on serious offenders, especially those who are violent or sexually deviant. In view of the complexities mentioned above, this selectivity is understandable, but it should always be remembered that it excludes the majority of mentally disordered and the majority of criminals.”

Classification of Mental Illness and Violence
(Information Bias):

Systematic errors in obtaining information that is used to classify subjects on either exposure or outcome factors can result in invalid conclusions (Rothman, 1986). For example, many of the studies examining the relationship between mental illness and violence have relied on institutional records to classify their subjects, such as an admission for a psychiatric illness to classify mental illness or a criminal arrest or conviction for a violent crime to classify violence. Clearly these do not capture all persons who suffer from a mental illness or all violent acts. Estimates of the relationship between mental illness and violence may be inflated or underestimated depending on how subjects are misclassified (Rothman, 1986). To illustrate the problem of underreporting, Dietz (1981) cited a study indicating that in cases of assaultive violence without theft or rape, 62% of attempts and 46% of completed attacks were not reported to police. Lion, Synder, and Merrill (1981) report similar difficulties in psychiatric patient populations.

Gunn (1977) highlights the difficulties associated with relying on secondary data (such as institutional records) for classifying subjects:

“We are all aware that the very existence of mental illness has been challenged and that definitions are extremely difficult to formulate. Yet most of us believe that somewhere in the confusion there is a biological reality of mental disorder, and that this reality is a complex mixture of diverse conditions, some organic, some  functional, some inherited, some learned, and some acquired, some curable, others unremitting. It would be surprising if such a mélange had a clear-cut relationship with any social parameter, specially one which is arbitrarily determined by legislation. Criminal behaviour is simply the breaking of the criminal laws in force at any particular time.”

Too often, official statistics reflect political biases and social trends that may affect these measures and their interpretation. As a result, many investigators have concluded that primary data collected from the general population through the use of self-report measures, rather than official samples, will provide a more accurate picture of the nature and scope of criminal activity and violence (Convit, O'Donnell, and Volavka, 1990).

While population surveys that involve primary data collection provide the best opportunities for overcoming the shortcomings of archival data, they are not entirely without difficulty. Underreporting of violence may be a general problem in self-report measures of violence, particularly if the violence is directed toward children (for which strict legal sanctions exist) or spouses (Swanson et al., 1990). Further, this reporting bias may differ for persons who suffer from mental illness, compared to those who do not. For example, Convit et al. (1990) examined the validity of self-report measures of arrests among psychiatric patients by comparing these to officially reported arrests and found them to be only slightly better than chance alone. Of the 41 patients studied, 66% gave accurate reports, 12% denied having arrests when their record showed arrests, and 22% reported arrests when their official records showed none. While this study was based on an extremely small and selected sample, it does raise the issue of misclassification bias, and highlights the importance of investigators taking steps to minimize this bias in their study designs. It also raises the importance of using appropriate caution to interpret findings, particularly when drawing etiological inferences.

What Comes First? Temporal Ordering of Factors:

In order for mental illness to cause violence, it must precede it. In order to infer causality from empirical evidence, therefore, a clear temporal ordering of events must be established.

In a cohort (i.e. follow-up) study, two or more groups of individuals who differ with respect to the purported causal factor under study (termed “exposure”) are followed through time and compared with respect to their outcomes. An essential element of a cohort study is that all groups are free from the outcome of interest at the outset of the research (Rothman 1986). This makes it possible to establish the temporal ordering of factors with absolute certainty. In the present context, this would require cohorts to be defined on the basis of the presence or absence of mental illness, excluding any individuals who, at the outset, report a past history of violence.

The importance of excluding individuals with a prior history of violence is highlighted by Steadman, Vanderwyst, and Ribner (1978). These authors compared the arrest rates of former mentally ill patients with criminal offenders released in the same jurisdiction in New York State (United States). Discharged mentally ill patients were found to have an overall arrest rate that was substantially higher than the general population. However this masked the fact that these patients differed dramatically with respect to their prior history of arrest. Approximately three-quarters had no previous arrest history and were arrested about as often or less often than the general population. Those with previous arrest rates were arrested more often than the general population. The authors conclude that as deinstitutionalization has changed the composition of state mental hospitals to include a higher proportion of individuals with prior criminal histories, so too have the rates of criminality among former mentally ill patients increased. Epidemiologists refer to this as identifying a cohort that is “at risk” for the outcome. Individuals who, at the outset, have experienced the outcome under study are excluded from study because they are no longer at risk (Rothman, 1986).

In a case-control study, subjects are chosen on the basis of the outcome of interest (in this case violence), then information is collected on the exposure of interest (i.e. previous mental illness). In the present context, cases would be defined on the basis of violence, controls would be defined on the lack of violence. Temporal ordering of factors may be difficult in case-control studies because investigators must rely on participants' memories to pinpoint the timing of crucial events. Because of the possibility of “recall bias”, epidemiologists require findings from case-control studies to be corroborated by other study designs (ideally cohort studies) prior to drawing etiological inferences.

In a cross-sectional survey, a representative sample of individuals are surveyed and exposure and outcome information are obtained at the same time. Because both exposure and outcome are measured simultaneously, it may not be clear which predated which. For this reason, cross-sectional surveys are typically considered to provide the weakest evidence of causality in epidemiological enquiry (Rothman, 1986).

Epidemiologists adhere to a hierarchy of evidence placing the most credence in statistical associations demonstrated in well-designed and executed cohort studies. Case-control study designs can provide persuasive evidence but are usually not deemed to be sufficiently strong to make a causal judgement. Descriptive cross-sectional studies are used to generate hypotheses for further testing. They are not used to infer causality.

Biological Plausibility:

Finally, the biological plausibility of a hypothesis is an  important epidemiological concern in assessing causality. Biological plausibility refers to whether or not an observed statistical relationship can be interpreted within the context of current biological theories. Biological plausibility is often difficult to assess because the state of the knowledge may be such that biological mechanisms are unknown. Therefore, the absence of a biologically plausible hypothesis does not invalidate a causal judgement. However, the presence of a biologically plausible hypothesis lends certain strength to a causal inference that is supported by strong empirical evidence (Rothman, 1986). Therefore, it is useful to assess whether there are biological mechanisms that have been postulated to link mental illness to violence in a causal framework.

Genetic and inherited diseases, pernicious influences in-utero with a potential effect on the developing brain, perinatal damage to the brain, specific conditions associated with central nervous system pathology, and some personality syndromes as substratum for episodic dyscontrol, have been implicated, both with mental symptomatology and the expression of violence. An association between psychopathy and violence has been proposed, but requires further elaboration.

In 1950, Sandberg discovered a male with an extra Y chromosome (described in Heilbrun and Heilbrun, 1985). This chromosomal abnormality became known as the XYY syndrome. As females have two XX chromosomes, and males have one X and one Y chromosome, it is the Y chromosome that conveys “maleness”. Males with two Y chromosomes, as in the case of Sandberg's male, were immediately described as “supermales” and endowed with special characteristics such as being extra-tall and extra-aggressive. Soon, reports surfaced from mental and criminal institutions about an abundance of tall men who had committed heinous crimes of violence and who were XYY. Tall men defendants who had committed serious crimes of violence began pleading incompetence on the basis that they had an extra Y chromosome and, therefore, a biological abnormality which caused them to commit a crime. As might be expected, this led to a scientific and legal controversy on the accuracy of the relationship of the XYY syndrome and criminality, specifically, violent crimes. A large community epidemiological study carried out in Denmark settled the controversy. A birth cohort was gathered consisting of 31,436 men. The tallest among them were checked for the extra Y chromosome. Only 12 were found to be XYY, and none of these had ever committed a violent act.

Mednick and Finello (1983), leading exponents on the biology of crime, have noted similarities in findings on the antecedents and correlates of antisocial conduct across nations and continents. Specifically, they call attention to several leads:

1) the “cultural robustness” of findings such as the unresponsive autonomic nervous system characteristic of delinquents, adult offenders and prison inmates;

2) cross-national research implicating neuropsychological disturbance and hyperactivity as characteristic of violent offenders; and

3) slow frequency of electric brain activity that has been used to predict criminal behaviour.

In their own research, Mednick, Gabrielli and Hutchings (1984) compared court convictions of 14,427 adoptees with those of their biological and adoptive parents. They found a statistically significant correlation between the adoptees and their biological parents for convictions of property crimes, but not for violent crime. More significantly, siblings adopted separately into different homes tended to be concordant for convictions, especially if they shared a biological father with a record of criminal behaviour.

The proposal, which appeared over 50 years ago, for a neuroanatomical central nervous system (CNS) mechanism for the expression of emotions and behaviours has led to findings of alterations in CNS serotonin, a neurotransmitter, in association with violent behaviour in animals. These findings have been confirmed by many researchers the world over. The hypothesis that affective disorders in humans are associated with suicidal and violent behaviour has led to numerous replications of experiments reporting an association between low concentrations in the cerebrospinal fluid of 5-hydroxyindoleacetic acid, a metabolite related to serotonin, and impulsive, destructive, and violent behaviour (Brown, Linnoila, 1990; Apter et al., 1990). Equally, researchers and clinicians in many countries have described the beneficial effects of a variety of drugs with CNS activity such as lithium, propranolol, chlorpromazine, clozapine, and other antipsychotics, which are used for the treatment of violent behaviour whether or not associated with mental illness (Greendyke, Schuster, and Wooton, 1984; Craft et al., 1987; Herrera et al., 1988). Paradoxically, some anxiolitics such as the benzodiazepines, have been implicated in triggering violent reactions (Lader and Petursson, 1981).

Despite the XYY fiasco, and the difficulties in conducting this kind of research, some of which are sociopolitical and ethical more than technical ones, a theory and a body of knowledge are solidifying around the hypothesis that some mental conditions, or basic emotional dimensions, are correlated with the expression of violent behaviour. As more and more research findings are pointing towards the presence of brain pathology in major mental conditions such as schizophrenia and affective disorders, a convergence has started to develop, at the biological level, for a similar CNS substratum underlying mental illness and violence.

Unfortunately, the biology of violence has not advanced as rapidly as the biology of mental illness. Thus, it is yet too early to claim that this point has been proven beyond  any controversy. Nevertheless, a link between mental illness and violence remains a biologically plausible hypothesis that warrants future careful study.

Proof or Stereotype?

Does mental illness cause violence or is this perception an unfortunate stereotype?

In discussing the “criminalization” hypothesis, Teplin (1984) has pointed to the desire of some scientists and policy makers to accept hypotheses based largely on intuition and unsystematic observation without subjecting them to adequate empirical testing. She suggests that public policy decisions should be firmly grounded in scientific evidence. In speaking about deinstitutionalization Arboleda-Flórez (1993) also notes that mental health policy decisions have been based on statements of philosophy and social values rather than on solid empirical data and theory.

Much in the way of understanding the relationship between mental illness and violence has been accepted on the basis of intuition and unsystematic observation and, as yet, there remains a lack of adequate empirical testing. Following Monahan's (1992) judgement, mental illness has become understood as a likely cause of violence and criminality. Recent reviews (Monahan, 1992, Torrey 1994) leave the reader with the impression that an ever progressing science has passed a threshold of evidence that has finally allowed us to conclude that a causal association exists. Tracing the socio-historical roots of the belief that mental illness causes violence from the very origins of Western civilization to present day perceptions (Monahan, 1992) may strengthen our belief in the validity of this conclusion.

However, earlier, more critical reviews (Davis 1991; Teplin, 1983) succeeded in highlighting a number of methodological difficulties that seemed to plague studies in this area. This critical review of the recent literature shows few improvements.

Studies of violence among treated mental patients have demonstrated that these individuals (especially those who abuse substances) may have higher levels of criminality and violent criminality than the general population. Virtually all of these studies have relied on official arrest data to measure criminality to facilitate population comparisons and have assumed that violent crime is an adequate measure of violence. It has been argued that arrests and convictions are even inadequate measures of crime, because seriously disturbed defendants and those with histories of psychiatric hospitalization may be diverted to the mental health system and rehospitalized, rather than arrested. Also, because most crimes do not result in arrests and most arrests do not result in convictions, arrest rates underestimate the total number of arrestable incidents and overestimate the number of guilty verdicts. Studies of police-citizen encounters overcome these problems. It is interesting to note, however, that even these studies (Holley and Arboleda-Flórez, 1988;  Teplin, 1984) have not demonstrated a difference between mentally ill offenders and non-mentally ill offenders with respect to the prevalence of violent offenses.

Studies of mental illness among incarcerated offenders have shown a high prevalence of serious mental illness and substance abuse disorders. These studies have been largely descriptive and have not used comparison groups to assess the extent to which the prevalence of mental illness among incarcerated offenders is higher than could be expected given general population rates. Nonetheless, Canadian prevalence figures of approximately 60% (Arboleda-Flórez, 1994; Bland et al., 1990) leave little doubt that incarcerated offenders are a high risk group. Studies of violence among psychiatric inpatients could be said to have yielded similar findings, despite the difficulties experienced in accurately measuring violence. The logic underlying these investigations is straightforward. If mental illness and violence are causally related, then we might expect to see a high prevalence of mental illness among incarcerated offenders and a high incidence of violence among psychiatric inpatients. While this is true, studies conducted exclusively on institutional populations cannot provide empirical evidence of a causal relationship between mental illness and violence because they have systematically excluded the majority of persons with a mental illness. Because these studies focus on sub-groups of mentally ill that are most likely to exhibit violence, findings from these studies can only be used to disconfirm a causal explanation (i.e. if they revealed lower than expected prevalences).

Link and Stueve (1995) argue that the consistency of findings across the various populations studied, despite their specific limitations, must be considered as evidence of causality. Methodological weaknesses in one area cancel the methodological weaknesses in another. This perspective overlooks the importance of selection bias as one explanation for the consistency of results reported. Only studies that examine the relationship of mental illness and violence in unselected and representative samples of mentally ill can address the issue of causality. To date, only two studies (supporting three analyses) have examined the relationship between mental illness and violence in unselected samples of adults living in the general population, one in Canada (Bland and Orn, 1986) and one in the United States (Swanson, 1993; Swanson et al., 1990). Both studies used a structured diagnostic interview schedule (DIS) to survey representative samples of the population to arrive at DSM-III diagnoses. Both studies represent the “state-of-the-art” with respect to current psychiatric epidemiological methods. Because both studies have overcome the problem of selection bias by focussing on non-institutionalized samples, results are potentially useful in drawing epidemiologic inferences of causality.

Both studies report a statistical relationship between mental disorder and violence, particularly among persons  with substance abuse disorders. However, what is not clear from these findings is the extent to which this statistical relationship could be explained by definitional confounding (described earlier) where almost half of DSM-III diagnoses, particularly substance abuse disorders, could be defined in part on the basis of violent behaviours. To compound this problem, both studies abstracted diagnostic items from the DIS interview that referred to violent behaviours and used these as the basis of their measure of violence. The DIS items referring to violent behaviour used by Swanson and colleagues (1990) were taken from the diagnostic sections for antisocial personality disorder and alcohol abuse and dependence. Thus, it is not surprising that a significant statistical relationship was reported in this study between substance abuse disorders (single or comorbid) and violence. Using a similar procedure, Bland and Orn (1986), restricted their analysis to three diagnostic categories, antisocial personality disorder, substance abuse disorders, and major depression.

In addition, the cross-sectional designs of the studies make it difficult to establish the temporal ordering of the factors with any degree of certainty. The American study (Swanson, 1993) employed a one-year period prevalence measure of both mental illness and violence such that individuals meeting the criteria for either of these during the year prior to the survey would have been counted positively. Thus, violence could have predated the mental illness or vice versa. The Edmonton study (Bland and Orn, 1986) used a lifetime prevalence measure such that individuals meeting the diagnostic criteria at any point in their lives would be considered to be positive for the study factor. Again, it is not clear whether the mental illness predated the violence.

The American study was designed to provide longitudinal data. Participants were re-contacted and re-interviewed in a second wave one year following the first interview. Longitudinal studies of rare events such as violence require very large sample sizes. However, this study overcame this problem by combining representative samples from five cities to provide a total sample size of over 20,000. This is sufficiently large to provide follow-up information on new cases of mental illness with no prior history of violence. If the issue of definitional confounding could be overcome and independent measures of mental illness and violence achieved, this study could provide information that could be used to explore an etiological relationship between mental illness and violence.

Finally, a word on the generalizability of findings from the United States, where much of the research has been done, to Canadian populations. Both Borzecki and Wormith (1985) and Davis (1992) address this issue. The trend toward deinstitutionalization in Canada was as marked as in the United States with massive reductions in the census of hospital populations occurring in both countries. Restrictive admission policies in both countries, based on a standard of dangerousness, ensure that competent patients who are not dangerous to themselves or to others can refuse hospitalization. A major difference between the countries occurs in the access of persons with a mental illness to appropriate community treatment resources. Universal access to medical care coupled with greater spending on community mental health programs in Canada suggests that there may be less pressure to use criminal justice alternatives for mentally disordered persons living in the community (Borzecki and Wormith, 1985). Nevertheless, the high prevalence of mental illness noted in incarcerated populations in Canada (Arboleda-Flórez, 1994; Bland et al., 1990) and the consistency of findings regarding police-citizen contacts across Canadian (Arboleda-Flórez and Holley, 1988) and American studies (Teplin, 1985) suggests that similar selection pressures are operating in both countries.

Conclusions:

In light of the foregoing annotated review and discussion, it is possible to identify that which we are relatively confident of, and that about which we remain uncertain.

Given the methodological problems that have plagued this field, a causal inference that mental illness causes violence cannot yet be made for the following reasons:

a) Studies examining the relationship between mental illness and violence have been largely restricted to selected offender or patient populations. Because of their selected nature, these studies are not sufficient to draw etiological inferences about the role of mental illness in causing violence in general. They only provide evidence of an association among selected sub-groups of persons who come into contact with services because they are more likely to be violent in the first place.

b) General population studies that have overcome this selection bias have shown a statistical association between some mental illnesses and violence. However, these studies have been cross-sectional in nature so have been unable to establish the temporal ordering of study factors. Therefore, at best, they could provide only weak evidence to support an etiological inference. In addition, however, it is possible that the statistical associations noted are an artifact of the manner in which mental illness and violence were defined. These studies have failed to derive measures of mental illness that are independent of violence. Given this problem, findings from these studies cannot be used to form the basis of an etiological argument.

Without further well-controlled epidemiological investigations, there is insufficient evidence to infer an etiological relationship between mental illness and violence.

Based on the foregoing review, however, we can be relatively  confident that:

a) the prevalence of mental illness (particularly substance abuse disorders) among incarcerated populations is high, reflecting a population that is in particularly great need of services;

b) former mental patients may be at high risk of arrest and violence when released into the community, particularly if they have a history of prior arrests or violence or if they experience psychotic symptoms;

c) family members (not the general public) are the most likely targets of violence from former mentally ill patients in the community; and

d) hospitalized mental patients are at high risk of committing violence, particularly if they have a history of prior violence or if they experience psychotic symptoms.

Potential Directions for Future Research:

• Further research using selected samples of patients or incarcerated offenders is unlikely to further our knowledge on this issue. Future research will have to proceed with attention to a number of critical design issues including the:

a) development of independent measures of mental illness and violence that can be used within the context of primary data collection efforts to avoid confounding by definition;

b) measurement of the relationship between mental illness and violence in large and unselected populations, with appropriate exclusions or controls for persons with a prior history of violence; and

c) clear temporal ordering of factors such that it is unequivocal that mental illness predates any expression of violence.

• The paucity of published literature incorporating the perspective of those who have a mental illness should also be addressed. The views and experiences of consumers and their family members would contribute further to the examination of violence as it is experienced by these individuals in community and hospital settings, and among peers. This is an area where future work is needed.
• While not the subject of the current review, the research reviewed in this study raises questions in regard to the extent to which the following issues are being addressed:
• Appropriate identification of mental illness and substance abuse problems among incarcerated offenders in correctional settings.
• Management of disturbed behaviour by the mentally ill that is often directed towards family members. Such approaches are being used in hospital settings.
• Appropriate access to community resources for former mental patients in the community.
• Appropriateness of treatment for the mentally ill in correctional and community settings.

Next | Previous | Table of Contents