Criminal Behavior and Victimization Among Homeless Individuals With Severe Mental Illness: A Systematic Review

Studies were eligible for inclusion if they reported prevalence rates of one of the outcome variables (victimization; contacts with the criminal justice system, including arrests, charges, convictions, and incarcerations; and self-reported criminal behavior), they were based on primary data collection, their sample consisted of adults (18 years and older), participants were homeless at the time of data collection, and participants had at least one severe mental illness (schizophrenia, other psychotic disorder, bipolar disorder, or major depression) with or without a co-occurring substance use disorder, in accordance with DSM or ICD diagnostic criteria (10,11). No publication date or publication status restrictions were imposed. However, studies had to be published in either English or French.

Because this review focused on base rates of criminal behavior, criminal justice involvement, and victimization of homeless individuals living in the community, studies reporting data for samples recruited exclusively through jails, prisons, jail diversion programs, or mental health courts were excluded for two reasons. First, including homeless detainees would lead to overestimating rates of criminal behavior, arrests, convictions, and incarceration. Second, previous studies have suggested that profiles of victimization of detainees with severe mental illness may differ from those of community-based individuals with severe mental illness—for example, in the prevalence of sexual victimization among men. Studies that included participants with personality disorders or substance use problems without concomitant severe mental illness were also excluded, as were studies in which participants had a history of homelessness but were not necessarily homeless at the time of data collection.

Studies were first identified by searching MEDLINE, Embase, PsycINFO, Cumulative Index to Nursing and Allied Health Literature, and Web of Science. The search was limited to peer-reviewed studies in French or English with an abstract. Manual scanning of reference lists, correspondence with authors, and consultation with experts in the field of homelessness and justice supplemented the electronic search. The final search was run on July 20, 2012.

Two of the authors (LR and ARA) used the following keywords in all databases: (homeless* OR shelter*) AND (mental illness OR mental health OR mental disorder OR schizophrenia OR bipolar OR depression OR “affective disorders” OR “mood disorders” OR “psychotic disorders* OR psychos*) AND (justice OR crime OR criminality OR jail OR prison* OR police OR victim* OR arrest OR incarceration OR violence OR detention OR “forensic psych*” OR offens* OR offend* OR convict*). The complete search strategy in MEDLINE is available from the first author upon request.

Two authors (LR and ARA) independently assessed eligibility of studies in a standardized manner, using the abstract provided in the database. In some cases, the full-text publication was consulted to clarify eligibility criteria. For multiple publications of the same study, the article with the most complete primary outcomes was selected. Disagreement between reviewers was resolved by consensus, and the advice of a third author was sought if a consensus was not reached.

The first author developed a data extraction sheet that was pilot-tested in a preliminary database search and refined accordingly with all other authors. The two reviewers (LR and ARA) then independently extracted the following data: study objectives, geographic location, and design; sample size, mean age, and gender composition; diagnostic tool used to assess severe mental illness; rates of comorbid alcohol and substance use disorders; time frame; source of outcome data; rates of criminal justice contact, victimization, and criminal behavior; and significant correlates of criminal activity or victimization, with associated odds ratios. Authors were contacted for missing information, and odds ratios were calculated manually if sufficient information was provided (either in the published report or via data provided by the authors). Disagreements were resolved by discussion between the two reviewers, and the advice of a third author was sought if disagreements persisted.

Quality assessment of observational data in systematic reviews is different from that of randomized controlled trials and not as well established. Risk of bias in the case of observational studies is intrinsically linked to underreporting or overreporting of outcome variables, in this case criminal activity, contacts with the justice system, and victimization. For each of these variables, assessment methods are crucial in determining risk of underreporting or overreporting. Another potential source of bias is the representativeness of the sample because homeless individuals, particularly those with severe mental illnesses, are known to be difficult to contact. In an attempt to establish methodological quality of the observational studies reviewed, relevant items from the Newcastle-Ottawa Scale were used (12): representativeness of the sample, ascertainment of severe mental illness, assessment of outcome, and, for prospective studies, adequacy of follow-up. According to these criteria, methodological quality was assessed as good, fair, or poor. The full definition of each item is available from the first author upon request.

The search of all electronic databases provided a total of 1,438 citations. After the abstracts were read, 1,377 studies were discarded because they did not meet inclusion criteria. The full text of the remaining 61 articles was obtained and read, resulting in the exclusion of a further 42 studies. Three additional studies were obtained through manual searches of the reference lists, of which two were selected, for a final total of 21 studies included in the review. [The phases of the systematic review are shown in a figure in an online supplement to this article.]

Although 21 studies were selected, some of our selected outcome variables were cited more often than others. Criminal behavior was reported as a criterion variable in four studies. Arrest rates were reported in seven studies, conviction rates in eight studies, and incarceration rates in eight studies. One study cited rates of summons in the target population. Victimization rates were cited in six studies. Only one study cited rates of criminal behavior and victimization for the same sample. No study reported rates of persons charged with a crime. Selected studies were published between 1986 and 2011. Twenty were conducted in the United States and one in England.

The studies included had a total of 10,704 participants, of whom 3,705 (34.6%) were female and 6,999 (65.4%) were male, with a mean age of 38.32±9.11 at the time of the interview (baseline interview for longitudinal studies). Three studies reported outcomes for 7,222 participants (67.5% of total sample) from the Access to Community Care and Effect Services and Supports (ACCESS) study, a multisite housing program for homeless individuals with severe mental illness (13–15). This is noteworthy because participants in the ACCESS study had to receive shelter services, which may have altered the representativeness of this sample. According to the Newcastle-Ottawa Scale, four (19%) studies were classified as good (16–19), 12 (57%) as fair (13,15,20–28), and five (24%) as poor (29–33).

Table 1 summarizes characteristics and findings from the 15 studies that measured contacts with the criminal justice system. The studies were heterogeneous in terms of sources of data, type of design, and studied time frame.

Lifetime arrest rates (four studies) ranged between 62.9% and 90.0% (20,21,29,30). Two studies reported 12-month arrest rates: 12.8% in a sample of 39 individuals recruited from a specialist psychiatric service for people with severe mental illness who were homeless in Birmingham, United Kingdom (20), and 52.0% in a sample of 144 individuals with severe mental illness recruited from a variety of locations (16). One study reported a 27-month arrest rate of 35.4% (22). It should be noted that the study reporting a 12.8% arrest rate over a one-year period (20) was the only study conducted outside the United States. A large study of 7,222 homeless individuals participating in the ACCESS program yielded two-month charge rates of 2.7% for major crimes, 10.9% for minor crimes, and 2.1% for substance-related crimes (15). One study reported a 12-month summons rate of 41% (16). Significant correlates of arrest included all types of previous contacts with the criminal justice system, high perceived need for medical services, high intensity of mental health service use, young age, a diagnosis of conduct disorder before the age of 15, and African-American race-ethnicity (15,16,22,29).

Lifetime conviction rates (seven studies) ranged between 28.1% and 80.0% (20,21,23,24,29–31). One study reported a 12-month conviction rate of 33.3% among homeless persons with schizophrenia and substance abuse (17). Significant correlates of conviction included high intensity of mental health service use, high severity of substance use, and protracted homelessness (24,29,31).

Lifetime incarceration rates (four studies) ranged between 48.0% and 67.0% (14,20,30,32). Three studies reported 12-month incarceration rates ranging from 26.0% to 33.0% (16–18). Significant correlates of incarceration included male gender, young age, African-American race-ethnicity, previous contact with the criminal justice system, greater severity of psychiatric symptoms, higher level of substance use, and co-occurring mental and substance use disorders (14,16,17).

Table 2 presents characteristics and findings from the four studies that measured self-reported criminal behavior. Available time frames were considerably shorter than for contacts with the criminal justice system and ranged between two weeks and 12 months. In one study, 17.0% of participants reported committing a violent crime during a six-month period, whereas 32.0% reported committing a nonviolent crime during the same time frame (25). Over a 12-month period, participants enrolled in a supported housing trial reported a mean number of 1.8 crimes (33). In a study by Calsyn and colleagues (16), 12-month prevalence rates of 40.0% for substance-related offenses, 22.0% for minor offenses, and 9.0% for major offenses were reported. Finally, in one study, 7.8% of participants reported committing a violent crime in the previous two weeks (26). Of the two studies that reported correlates of self-reported criminal behavior, significant correlates included young age, severity of psychiatric symptoms, type of homelessness (street versus shelter), protracted homelessness, and previous contact with the criminal justice system (16,25).

Table 3 summarizes characteristics and findings from the six studies that measured victimization outcomes. Lifetime victimization rates ranged between 73.7% and 87.0% (17,19,27,34). Two studies reported short-term victimization rates, with a one-month prevalence of 45.0% (28) and a two-month prevalence of 76.7% (13). Significant correlates of victimization included female gender, diagnosis of depression, and a history of child abuse.

Lifetime prevalence data considered in this review indicate that 62.9%−90.0% of homeless individuals with severe mental illness have been arrested at least once, 28.1%−80.0% have been convicted of a crime, and 48.0%−67.0% have been incarcerated. Of note, for conviction rates, the lowest figure (28.1%) represents felony convictions only (which include more severe categories of crime than misdemeanors) (29); if this figure is excluded, lifetime rates for any conviction fall into a much tighter range, between 52.0% and 80.0%. These rates, as well as shorter-term prevalence rates, are much higher than for the general U.S. population, in which lifetime arrest rates are estimated to be about 15.0% (35). Ten-year arrest rates of community-dwelling adults with mental illness ranged between 25.0% and 33.0% (36–38).

Lack of uniformity in time frames and sources of data render any comparison tentative, but homeless individuals with severe mental illness seem to be arrested more often than their stably housed counterparts. As discussed below in the context of rates of criminal behavior, this could reflect both a greater frequency of criminal behavior as well as the visible nature of the homeless life, which may make these individuals more likely to come into contact with the criminal justice system.

Comparing rates of criminal justice contacts from the studies reviewed with rates from studies of homeless individuals without a severe mental illness is more challenging because there is considerable overlap in their characteristics at contact with services and because most studies on homeless populations do not typically exclude individuals with mental illnesses. Two of the studies selected for this review did, however, report comparative data for homeless participants with and without severe mental illnesses (20,30). Results suggested that both groups had similar rates of involvement with the criminal justice system (arrests, convictions, and incarcerations). This indicates that in terms of criminal justice involvement, homeless individuals with severe mental illness tend to be more similar to the general homeless population than to the general population of individuals with severe mental illness. However, any conclusions must be viewed as tentative given the few studies available and the challenge of drawing comparisons across studies that used different definitions of mental illness and different measures of criminal justice involvement and that had various other methodological differences (such as use of self-report data versus data from administrative records).

Common correlates of criminal behavior (either through self-reports or administrative records reflecting contacts with the justice system) included demographic variables (young age, male gender, and African-American race-ethnicity), clinical variables (level of psychiatric symptoms, substance use, and diagnosis of conduct disorder), and contextual variables (type and duration of homelessness, perceived need for medical services, and previous contact with the justice system). One of the most important correlates of all types of contact with the criminal justice system in both this review and previous studies of housed individuals with severe mental illness or homeless individuals without severe mental illness is the extent of criminal history or previous contacts with the criminal justice system (39,40). Number of prior convictions and age at onset of criminal activity have also been found to be the most common predictor of recidivism among offenders both with and without severe mental illness (41).

For two reasons, the extent of criminal history may be even more significant for individuals who live both with severe mental illness and in a homeless situation compared with housed persons with severe mental illness. The first one speaks to the interaction between demographic or clinical factors and environmental variables (particularly living context). When individuals with severe mental illness are homeless and thus visible in the public space, they are more likely to draw attention from the public and from law enforcement personnel and thus to be rearrested more frequently (26).

The second reason reflects common behavioral patterns that are more likely to be found in the homeless population, particularly antisocial personality disorder and substance abuse. Some evidence suggests that antisocial personality disorder and homelessness are related. For example, homelessness is one of the most significant correlates of antisocial personality disorder in community samples with (42) or without (43) severe mental illness. Connolly and colleagues (44) reported that 57% of a sample of homeless adults met Structured Clinical Interview for DSM-IV criteria (45) for a diagnosis of antisocial personality disorder. In a recent study of homeless young women, 66.2% met criteria for a lifetime diagnosis of antisocial personality disorder (46). One retrospective study of the course of antisocial symptoms among homeless adults indicated that these symptoms preceded the onset of homelessness for more than 90% of individuals in the sample (47). The co-occurrence of antisocial personality disorder with a substance use disorder among persons with a mental illness is also well documented (48,49), as well as the cumulative effect that these conditions have on rates of criminal behavior and violence (42,50,51). Future research in this area should include follow-up studies of the course of antisocial personality disorder symptoms after provision of housing or support services, as well as the specific nature of relationships between homelessness, personality disorders, severe mental illness, and substance use.

Another behavioral pattern likely to be found in homeless populations is substance misuse. A large body of literature indicates that substance abuse accounts for a large proportion of violent behavior (52–54), incarceration (55), and arrest (56–59) among adults with severe mental illness. In fact, many studies have found that in the larger population of adults with severe mental illness, substance abuse, along with antisocial tendencies, is the main cause of offending, rather than mental illness itself (41,57,60,61). This is particularly relevant given that 20%−40% of homeless individuals have a severe mental illness as well as a diagnosis of a substance use disorder (62,63). As noted above, there is also substantial evidence for a close link between substance abuse and personality disorders (48,49), with more than half of treated substance abusers having at least one personality disorder (64). In our review, three studies reported an association between substance abuse and charges, convictions, and incarcerations. No study specifically examined the relationship between substance use and arrests or self-reported criminal behavior. This reflects the more limited number of studies focusing specifically on homeless adults with severe mental illness, but it hinders our understanding of the role played by substance use in specific criminal behaviors and criminal justice system trajectories, which may vary for some subgroups of individuals. Indeed, Constantine and colleagues (59) found that the risk of felony arrest among persons with severe mental illness was independently associated with a substance use disorder but not with homelessness, whereas the risk of misdemeanor arrest was independently associated with homelessness but not with a substance use disorder.

It is also noteworthy that two of the studies included in the review sampled only homeless individuals with co-occurring disorders and failed to find a relationship between indicators of severity of substance abuse and arrests, incarcerations, or summons (16,22). This absence of a significant association could, of course, reflect the more limited range of severity of substance abuse in a sample that included only persons with co-occurring disorders. It is possible that for various subgroups of homeless individuals with severe mental illness, various causal pathways underlie the relationship between substance abuse and contacts with the criminal justice system. For example, the disinhibiting effect of drug and alcohol abuse might make some homeless individuals more disorderly and likely to come to the attention of law enforcement personnel in the public space, thus increasing the risk for nuisance offenses. For other individuals, such as those with comorbid personality disorders, substance abuse might trigger violent behavior and be related to more serious offenses. Research designed to disentangle the longitudinal course of substance use and personality disorders in relationship to onset and duration of homelessness and criminal behavior among persons with severe mental illness is needed.

Contextual factors, such as type and duration of homelessness, also appear as significant risk factors for self-reported criminal behavior and convictions. Contextual factors tend to receive less attention than clinical risk factors in studies of the relationship between mental illness and criminal justice involvement (65). Our results suggest that this is also the case for homeless populations of adults with severe mental illness, even though the very nature of homelessness should bring researchers to better understand how and when disturbing or violent behaviors and contacts with law enforcement personnel occur and lead to criminal justice involvement. An alternative perspective on our findings is that even though homelessness itself can be considered a crime (loitering and vagrancy are considered nuisance offenses), 10%−37% of homeless people with severe mental illness have never been arrested, according to data in the reviewed studies. Qualitative studies of the daily lives of homeless individuals can provide preliminary explanations about why some individuals never formally come in contact with the criminal justice system: the creation of small, protective groups of homeless individuals (66), the use of discretionary power by law enforcement personnel (67), the existence of retribution without recourse to formal police or justice mechanisms (68), and frequent refusals from fellow homeless individuals, spouses, or relatives to press charges (68). Research with the subgroup of people who have no contact with the criminal justice system might shed light on how to reduce contacts in this population.

Studies examining rates of victimization among homeless individuals tended to have more design features in common, and their findings were slightly more homogeneous than the other two categories of studies in terms of operational definitions, time frame, and correlates. All six victimization studies found that homeless individuals with mental illnesses, especially women, experienced extreme levels of victimization—higher than those of housed individuals with severe mental illness. Indeed, a systematic review of criminal victimization among adults with major mental disorders reported one-year prevalence rates between 4.3% and 35.0% for violent victimization and between 7.7% and 28.0% for nonviolent victimization (69). Studies (not included in our review) that examined the impact of past homelessness on victimization among larger samples of individuals with severe mental illness also found that homelessness was one of the most significant risk factors (70–73). Like homeless individuals with severe mental illness, stably housed individuals with severe mental illness are more likely than the general population to be victims of crime (69,74). There is also evidence that housed individuals with severe mental illness are more likely to be victims than to be perpetrators of violence (39,75). Finally, these findings can be contrasted with those of previous studies of victimization in the general homeless population, which have found one-year prevalence rates ranging from 50% to 60% (76,77). Compared with homeless individuals with severe mental illness, homeless individuals without severe mental illness are at higher risk of being victims of crime (26.0% versus 45.0%) (28). Thus it can be concluded that homeless individuals with severe mental illness are at higher risk of victimization than other comparable subgroups.

Only two studies identified correlates of victimization other than gender (27,34), namely child abuse and depression or anxiety. These two studies focused only on homeless women with severe mental illness, and both had small samples. In a study with a larger sample of adults with severe mental illness, child abuse was also found to be a risk factor for victimization, but other correlates included illness severity and substance misuse (72). The gender difference in rates of victimization among homeless adults with severe mental illness is similar to that found in previous studies of victimization among housed adults with severe mental illness (70,73). Future research should focus on the identification of correlates and predictors of victimization among larger samples of homeless men and women with severe mental illness. Investigations should also examine how revictimization occurs (the physical and social contexts, relationship to perpetrator, and so forth) and whether this process is also moderated by gender.

One important aspect of quality assessment for this systematic review is the representativeness of the sample. Homeless individuals are known to be hard to reach and difficult to recruit and retain for participation in research; moreover, studies tend to oversample homeless individuals who use services, particularly shelters (78). In this systematic review, one-third of the studies (seven studies) were considered representative or somewhat representative of the population of homeless persons in a given area, whereas the remaining studies recruited participants from designated areas (for example, skid row districts) or services (for example, a psychiatric clinic in an emergency shelter). This lack of representativeness may have biased results in terms of rates of criminal behavior, contacts with the justice system, or victimization, but the direction of bias is unpredictable. For example, homeless persons living in situations where they tend to be underrepresented (such as squatters) could be more subject to victimization, but they could also have developed street survival skills that lessen their exposure to violence.

Another caveat is related to the quality of the assessments of the outcomes of interest—criminal behavior, contacts with the justice system, and victimization. All of the included studies assessing criminal behavior and victimization examined these variables through self-reports. For victimization, use of self-report seems appropriate, and previous studies have indicated that official records (such as police logs) greatly underestimate the prevalence of victimization (79). Two studies used Lehman’s Quality of Life Interview (80) to assess victimization, three used home interviews, and one used the Conflict Tactics Scales (81). Both the Quality of Life Interview and home interviews use abuse-related labels (for example, rape) rather than behaviorally specific questions (for example, forced to have sex), which is known to lead to underreporting. The use of measures with established psychometric properties may not only alleviate the risk of underreporting; it could also help clarify the nature of victimization experiences, expand the focus of research to include severity levels, and facilitate comparison of findings across studies. Also, in the studies reviewed, contacts with the criminal justice system were examined either through administrative databases (six studies) or via self-report (nine studies) but not through multiple sources. Future research should consider strengthening the assessment of violence, victimization, and contacts with the criminal justice system through the use of multiple sources of information (administrative databases, self-report, and proxy report) to triangulate results and provide more valid prevalence estimates. In addition, the field should move beyond dichotomized outcomes to consider additional features of the outcome criterion, including severity, frequency, and the relationship of the perpetrator to the victim.

Overall, the ability to infer causal pathways between personal, clinical, and contextual factors and criminal activity in the studies reviewed was limited because of their use of cross-sectional, between-person, and mostly retrospective designs. One notable exception is the study by Fischer and colleagues (25), which used a within-person, prospective, longitudinal design to examine the relation between criminal activity, level of psychiatric symptoms, and housing status. Results indicated that levels of psychiatric symptoms and type of homelessness (shelter or street) had additive rather than multiplicative effects, particularly among individuals with protracted homelessness. Additional research examining the convergent effects of contextual, clinical, and personal factors affecting criminal behavior and victimization are urgently needed.

This systematic review had limitations. First, because of the heterogeneity of the studies, meta-analytic techniques were precluded. Second, most studies used cross-sectional designs to identify variables associated with criminal behavior, victimization, and contacts with the justice system. It was thus difficult to determine for most studies the nature and mechanisms of the relationships between independent and dependent variables. These findings need to be replicated in studies with large samples that use longitudinal designs to identify predictors of victimization and criminal activity. Third, as highlighted above, 67.5% of all participants were from a single study (ACCESS program). Fourth, all but one of the studies were conducted in the United States. Results may not generalize beyond the United States because of differences between countries in the nature of the homeless population (for example, homeless individuals in the United States are much more likely to be veterans); rates of contacts with the criminal justice system in the general population, which is higher in the U.S. than in other Western countries; and availability, accessibility, and costs of mental health and social services, which can vary significantly from one jurisdiction to the other. Fifth, in all studies reviewed, a significant portion of homeless individuals with severe mental illness may not have been identified as such, introducing potential bias. Finally, this systematic review used narrowly defined inclusion criteria so that conclusions could be drawn about a fairly precisely defined population (adults with both current severe mental illness and current homelessness).