Social Networks and Arrest Among Persons With Severe Mental Illness: An Exploratory Analysis

Individuals with severe mental illness in U.S. jails and prisons account for 14% to 17% of the 2.3 million people incarcerated, a rate three to six times higher than in the general population (1). Growing evidence suggests that trend is not directly attributable to deinstitutionalization, particularly in light of the impact of recent criminal justice policies on incarceration rates for the entire U.S. population (2). Current social-psychological criminological frameworks describe offending behavior as a function of social marginalization, lack of access to resources, and a high degree of contact with those similarly situated in the social hierarchy (3). In support of this framework, research has shown that disadvantaged neighborhoods, unemployment, poverty, substance abuse, and stigma, which disproportionately affect people with severe mental illness, are associated with criminal justice involvement (2–4).

Social network theories describe mechanisms that influence offending behavior, including social structures that affect behavior, attitudes, and access to resources (5). Relevant research links large social networks to increases in social capital, whereas small and densely interconnected networks that involve frequent contact with similar others have been linked to the spread of high-risk behaviors (6). We are aware of only one study examining network structures and criminal justice involvement among individuals with severe mental illness. Skeem and colleagues (7) found that social network composition, specifically networks largely made up of friends and substance users, increased probation violations. Given the paucity of research in this area, the aim of the exploratory study presented here was to examine whether current friendship network structures, such as size and frequency of contact, are related to lifetime arrest among persons with severe mental illness.

Individuals with severe mental illness generally have small networks of similarly situated individuals affected by mental illness, addiction, joblessness, and homelessness (7,8). It follows that frequent contact may increase high-risk behaviors and compound marginalization, thereby increasing the likelihood of arrest. We also expected to find that larger networks, which may provide greater access to resources, would be negatively associated with arrest. Because substance abuse and demographic characteristics have been widely shown to predict arrest in this population (2,3,9), a secondary aim was to examine whether social network characteristics had an impact on arrest above and beyond these factors.

Methods

The parent study, which examined functional improvement among persons with severe mental illness, employed a prospective follow-along design of patients at four community-based psychosocial rehabilitation programs in Los Angeles County. Data were collected between 2004 and 2006. All participants provided written informed consent, and the study was approved by the University of Southern California Institutional Review Board and the Los Angeles Department of Mental Health Human Subjects Research Committee.

The sample for the study reported here consisted of 119 individuals. Diagnoses were as follows: schizophrenia, 52 (44%); schizoaffective disorder, 19 (16%); bipolar disorder, 28 (23%); and major depression with psychotic features, 20 (17%). Sixty-three (53%) were African American, 25 (21%) were white, 17 (14%) were Latino, and 5 (4%) were Asian. Nine (8%) reported “other” race-ethnicity. Seventy-four (62%) were male, 110 (92%) were unmarried, and the mean±SD age of respondents was 38.1±9.8 years.

The outcome variable was lifetime arrest. Information about arrest was gathered as part of a demographic interview. The dichotomous variable was based on the item: “From the age of 18 until now were you ever arrested?” Information on the size of the current social network was gathered in a semistructured clinical interview referred to as the Community Adjustment Form (CAF) (10). Respondents’ self-reported number of friends was dichotomized to reflect small or large social networks because of the level of skew (11), which was not significantly improved after a log transformation (cut-point based on a median number of two friends) was applied.

Current degree of contact with friends was also gathered from the CAF (10) and reflected self-reported number of contacts with friends in the past two months. A ratio of number of contacts to number of friends provided an average amount of contact per friend (those reporting no friends had a ratio value of 0). This variable was also dichotomized on the basis of the level of skew; respondents with a contact ratio above the median value of 6 were coded as having high contact, and all others were coded as having low contact.

Substance abuse (7), race-ethnicity, age, and gender have been shown to be associated with arrest (9) and were included as covariates. A dichotomous variable indicating presence or absence of a lifetime diagnosis of substance abuse or dependence (determined by a trained mental health professional) and a dichotomous variable representing self-reported alcohol, marijuana, or street drug use in the previous six months were included. Logistic regression was the appropriate method of statistical analysis (11). We used a hierarchical regression approach to test the association of the outcome with demographic characteristics, substance use and abuse, and social network characteristics, entered in that order.

Results

Among the 119 persons, 78 (66%) had been arrested at least once as an adult, and 82 (69%) had received a diagnosis of a substance use disorder at some point in their life. Most were not actively using substances currently. In the past six months, 80 (67%) had not used alcohol, 100 (84%) had not used marijuana, and 96 (81%) had not used street drugs.

Results from the hierarchical logistic regression analysis are summarized in Table 1. In step 1, being male was positively related to lifetime arrest (Wald=13.59, df=1, p<.001), and the model with demographic covariates was significant (χ²=16.76, df=3, p<.001). In step 2, a lifetime diagnosis of substance abuse or dependence was positively related to lifetime arrest (Wald=18.81, df=1, p<.001), and substance use in the past six months was not. The model was significant (χ²=42.93, df=5, p<.001), and the incremental change from step 1 to step 2 was significant (Δχ²=26.17, df=2, p<.001).

In the final step, the size of the current friendship network (Wald=4.32, df=1, p=.04) and the degree of contact with the current friendship network (Wald=5.51, df=1, p=.02) were significantly related to the likelihood of lifetime arrest. The final model was significant (χ²=50.85, df=7, p<.001), and the incremental change from step 2 to step 3 was also significant (Δχ²=7.92, df=2, p=.02). The addition of current network variables significantly improved the overall model of risk and protective factors related to lifetime arrest; the final model correctly predicted membership in the arrest group 85% of the time.

Discussion

The findings suggest that among people with severe mental illness, larger current friendship networks were negatively related to lifetime arrest and a high frequency of contact with current networks was positively related. The addition of variables related to friendship networks significantly improved a model of risk and protective factors above and beyond inclusion of demographic and substance use and abuse variables. Recent research has indicated that mental illness and offending behavior may be primarily linked through social-psychological factors that establish “general risk factors for crime” (3). The preliminary results of this study suggest that social network structures may be a risk factor warranting further investigation in this population.

A methodological limitation was the use of current data to predict arrest at any time during adulthood. Therefore, conclusions must be viewed with caution. For example, associations among the variables may reflect a dynamic in which previous experiences in the criminal justice system shaped current social networks, rather than a dynamic in which networks influenced criminogenic behavior. Although the findings are speculative, the strength of association and theoretical justification for this line of inquiry merit further examination of network constructs and criminal justice involvement. This area of investigation is supported by social-psychological frameworks, including the notion of “risk environment” (12), which point toward the interplay of social and environmental factors that shape risk behaviors. Future risk factor research might examine the size, density, and composition of social networks as an important aspect of a risk environment, with a potential impact on offending behavior in this population.

A high degree of contact with the current network was associated with a five times greater likelihood of lifetime arrest, and having a larger network was associated with a 76% reduction in the likelihood. These findings may be viewed in the context of social network theory. Granovetter (5) postulated that larger networks promote multiple pathways of access to social resources and that more intense and regular network contact leads to restricted social capital. In this study frequent network contact was positively associated with lifetime arrest, which may reflect a dynamic—as Granovetter suggests—in which regular contact with persons who share exposure to socioeconomic disadvantage and high-risk behaviors may compound social disadvantage and increase risk of arrest. Also in line with Granovetter’s theory, the finding that larger current networks were negatively associated with arrest may suggest that networks promote greater access to resources that serve to reduce the arrest risk. Findings may inform efforts to delineate network features, such as members’ levels of social capital and the way in which these features interact with frequency of contact to influence criminal justice involvement.

A lifetime substance use disorder diagnosis was associated with a likelihood of arrest 22 times greater than that of study participants with no such diagnosis. This may point to increased opportunities for offending, such as disturbing the peace, resorting to theft and burglary to pay for drugs, and possessing illegal substances (13). In addition, those who report frequent network contact may engage in interactions involving the purchase and use of substances, increasing arrest risk. Again, future studies in this population are needed to disentangle relationships between risk behavior, network characteristics, and criminal justice involvement. Being male was also related to lifetime arrest, a finding consistent with previous research (9).

This study had several limitations. The network variables examined lack important information, such as degree of social capital and risk behavior among network members. Also, the impact on arrest of network members other than friends, such as family members and treatment providers, could not be assessed.

Several recommendations for future research can be made on the basis of these limitations. First, a prospective design is necessary. In this study, current data were used to predict arrest after age 18. Second, information on network composition has important implications—specifically, whether the network tends toward prosocial activity and conformity or toward nonconformity—because ties with people who commit crimes are a robust predictor of criminal behavior (14); whether network members abuse substances, which may undermine sobriety and lead to justice involvement (8); and whether the network consists of treatment providers and others who provide support (7). Third, information on the quality of network ties is highly relevant; social support has been shown to be related to reduced criminal justice involvement (15), and social undermining (dislike, criticism, and blocking of goals) has been shown to have even greater effects than social support on health outcomes (7). Skeem and colleagues (7) speculated that social support and undermining may have an impact on criminal justice involvement by way of their effects on stress and coping, including dealing with “criminogenic strains” that may lead to antisocial behavior. Finally, among persons with severe mental illness, Skeem and colleagues (7) found that the perception of coercion and control was associated with increased probation violations and that substance abuse was related to poor relationships with treatment providers.

Conclusions

Findings represent an incremental step in setting the stage for investigations of social network structures and arrest in a highly marginalized population. Attention to social structural conditions among people with severe mental illness may help reduce their risk of criminal justice involvement (12).