Individuals with serious and often disabling mental illnesses like schizophrenia, bipolar disorder, and major depression are grossly overrepresented in the criminal justice system. Compared to the general population, the current prevalence rate of these specific mental illnesses among jail detainees is higher for men by more than three times (1.8 vs. 6.4%; Teplin, 1990) and almost twice as high as for women (M w = 10.6 vs. 20.4%; Teplin, Abram, & McClelland, 1996). Moreover, regardless of gender, nearly three out of every four jail detainees with a serious mental illness have a co-occurring substance abuse disorder (Abram & Teplin, 1991; Abram, Teplin, & McClelland, 2003). These figures take on new meaning when considered in context. The number of people under correctional supervision in the USA recently reached an all-time high of 7.3 million (Bureau of Justice Statistics, 2009). Although prevalence estimates vary, a meta-analysis of 62 studies suggests that 14% of offenders suffer from a major mental illness (Fazel & Danesh, 2002; see also Steadman, Osher, Robbins, Case, & Samuels, 2009). If so, then there are over one million individuals with mental illness in the USA in jail, in prison, on probation, or on parole.

Individuals with mental illness are not only disproportionately represented in the criminal justice system; they also are disproportionately likely to fail under correctional supervision. The vast majority of individuals in the correctional system—70%—are supervised in the community on probation or parole (Glaze & Bonczar, 2007). Compared to their relatively healthy counterparts, probationers and parolees with mental illness are significantly more likely to have their community term suspended or revoked (Messina, Burdon, Hagopian, & Prendergast, 2004; Porporino & Motiuk, 1995; Skeem, Nicholson, & Kregg, 2008). Based on a sample of 44,987 offenders, Eno Louden and Skeem (in press) found that parolees with mental illness (52–62%) were about two times more likely than parolees without illness to return to prison within 1 year of release (30%).

Together, these figures are sobering. They indicate that a large number of individuals with mental illness enter the criminal justice system each year, and many penetrate deeply into the correctional system over time. As observed by the Council of State Governments (CSG), “the current situation not only exacts a significant toll on the lives of people with mental illness, their families, and the community in general, it also threatens to overwhelm the criminal justice system” (2002, p. 6).

This situation has attracted remarkable attention from policymakers and practitioners; especially those involved in the correctional system (see American Probation and Parole Association [APPA], 2003; Bureau of Justice Assistance, 2009; National Institute of Corrections, 2009). Since 1995, the National GAINS Center (Substance Abuse and Mental Health Services Administration, 2010) has focused intently on addressing the situation. More recently, the CSG Justice Center (2009) has been leading “an unprecedented national effort to help local, state, and federal policymakers and criminal justice and mental health professionals improve the response to people with mental illness who come into contact with the criminal justice system.” This laudable effort has brought together professionals in law enforcement, the courts, corrections, and mental health; identified and described programs across the USA for offenders with mental illness; and distilled the basic, perceived root of the problem.

With respect to the last point, “people on the front lines every day believe too many people with mental illness become involved in the criminal justice system because the mental health system has somehow failed. They believe that if many of the people with mental illness received the services they needed, they would not end up under arrest, in jail, or facing charges in court” (Council of State Governments [CSG], 2002, p. 26). In other words, the perceived root of the problem is “criminalization of mental illness” (Abramson, 1972; see also Human Rights Watch, 2003; Soros Foundation, 1996; Torrey, 1995; Torrey et al., 2002). Deviant behavior that was once (appropriately) defined and managed within a psychiatric framework has been (inappropriately) redefined and managed within a criminal framework. It is believed that because of deinstitutionalization (Lamb & Weinberger, 1998), increasingly restrictive laws for involuntary psychiatric hospitalization (Abramson, 1972), and the inadequacy of community-based psychiatric services (Torrey et al., 2002), “the nation’s jails and prisons have become, de facto, the nation’s largest psychiatric hospitals” (Treatment Advocacy Center, 2007, p. 1). When an individual with mental illness engages in deviant behavior that should have been prevented or managed through treatment, it is thought that the behavior is redefined in criminal terms by other agents of social control, including the police (cf. Engel & Silver, 2001; Fisher et al., 2006). Typically, an individual would be arrested for minor deviant behavior, perhaps in an effort to secure treatment in jail (Torrey et al., 2002; see also Lamb & Weinberger, 1998). Occasionally, an individual would be arrested for violence that is the direct “product of the person’s untreated mental illness” (Torrey et al., 2002, p. 48).

For this population, the chief policy goal is reduced recidivism or exit from the criminal justice system. Because (untreated) mental illness is perceived as the reason for criminal justice involvement, providing psychiatric treatment seems the logical way to end such involvement. At the federal level, this is implied by the very name of the “Mentally Ill Offender Treatment and Crime Reduction Act” (U.S. Congress, 208th, 2nd session, 2004), which authorizes funding for programs that target this population. Historically, access to effective mental health services has been cast as the lynchpin to successful response (e.g., CSG, 2002, Policy Statement #1 & Chapter 7). Indeed, virtually all contemporary programs are designed to link offenders with mental illness to community treatment services; for this population, there has been a “proliferation of case management services as the policy response” (Draine, Wilson, & Pogorzelski, 2007, p. 161). In general, the response focuses on one dimension: mental illness. Criminal justice involvement is used to mandate or link the individual to psychiatric treatment (e.g., a probationer is required to abide by a special condition to participate in treatment), and treatment is thought to reduce the risk of recidivism.

Given that much needed advocacy for this population has promoted the wide dissemination of such programs as mental health courts, the time is ripe to assess the extent to which these programs are reaching the chief policy goal of protecting public safety. In this article, we distill evaluations of contemporary programs, place their results in the context of relevant theory and research on the link between mental illness and criminal justice involvement, and propose a conceptual framework that may help advance policy and interventions for offenders with mental illness. Our framework suggests that the effect of mental illness on criminal behavior reflects “moderated mediation”; the link is direct in one subgroup, but mediated by other factors in another subgroup. Our framework further suggests that system bias and stigma—not criminal behavior per se—plays a role in community supervision failure. We conclude by proposing three specific research and practice priorities that expand the focus beyond mental health to explicitly embrace other dimensions and thereby better reduce recidivism for this population.

Our focus is on the large class of adults with mental illness who have been convicted of crimes or arrested and diverted from jail (rather than those deemed not guilty by reason of insanity), and on general recidivism (although we also address violent recidivism). We emphasize the context of community corrections (probation and parole) rather than institutions (jail and prison) because most offenders are supervised in the community and the bulk of work on evidence-based corrections focuses on that context.

To What Extent is the Current Policy Model “Working”?

Describing Program Types

The most common types of contemporary programs for offenders with mental illness are shown in Table 1, which describes each program type, summarizes its underlying premise and proposed solution (derived in part from Draine et al., 2007), and provides a program exemplar or prototype. As shown in Table 1, four program types are derived from general criminal justice models, i.e., jail diversion programs, problem-solving courts, specialty probation or parole caseloads, and jail transition or prison re-entry programs. These programs target a particular stage of case processing (e.g., arrest, re-entry) and/or a special population (e.g., mental health courts were derived from drug courts). Although ongoing judicial or correctional supervision is an integral component of some programs (e.g., mental health courts), others rely more exclusively on service brokerage (e.g., other jail diversion programs). Even though mental health courts are a specific form of jail diversion (i.e., specialty court-based, post-booking), we disaggregate mental health courts from the larger class because they (a) involve ongoing judicial supervision and (b) have spread prolifically over recent years (Bureau of Justice Assistance [BJA], 2009). As shown in the second column of Table 1, criminal justice-derived programs for this population are united by their emphasis on linkage with mental health services in the community as an essential component of their mission.

Table 1 Contemporary programs for offenders with mental illness
Full size table

Also shown in Table 1, are the other two program types that are derivatives of mental health models: Forensic Assertive Community Treatment (FACT) and Forensic Intensive Case Management (FICM). FACT and FICM are relatively intensive treatment models that may be used either independently or in conjunction with criminal-justice derived programs (e.g., a mental health court). FACT and FICM were adapted from the most extensively studied mental health service, Assertive Community Treatment (ACT) (Morrissey et al., 2007). As suggested by the composition of Table 1, treatment development efforts for this population have involved adapting existing evidence-based mental health services like ACT (see Osher & Steadman, 2007), that is, services that have been shown to improve patients’ clinical outcomes (e.g., reduced hospitalization). In contrast, evidence-based correctional practices that have been shown to reduce offenders’ recidivism have had little effect on practice for this population.

Distilling Evidence on Program Effectiveness

Our approach to distilling evidence on these programs’ effectiveness involved three steps. First, we focused on isolating studies of the programs defined in Table 1. Given the wide diversity among programs that adopt a particular label (e.g., “specialty probation,” Skeem et al., 2006; “jail diversion,” Hartford, Carey, & Mendonca, 2007), and the overlap among program types, we focused on (a) multi-site studies that simultaneously assessed multiple program exemplars, and (b) studies of programs with features that were prototypic of the target type and minimized overlap with other types. For example, we focused on studies of jail diversion programs that did not overlap with mental health courts.

Second, we conducted a comprehensive search in both PsychInfo and Medline for empirical evaluations of the effectiveness of each type of program. Because many of these programs have been shown to meet their basic goal of increasing access to psychiatric services (see Draine et al., 2007), we defined effectiveness in terms of the chief policy goal of recidivism reduction. Studies define recidivism differently, e.g., as re-arrest, revocation of community supervision (for any reason, i.e., a new crime or technical violation), and re-incarceration (for any reason). Given our interest in the link between mental illness and crime, and the larger policy priority of “preventing new crimes and new victims” (Keiser, 2009), we specifically searched for studies that focused on the outcome of re-arrest. We were also interested in studies that assessed whether a program’s effect on recidivism is mediated by mental health services or symptom reduction (i.e., when a program “works,” is that because it targets mental health?). For each program type, we identified a handful or small body of studies (the smallest for re-entry programs; the largest for jail diversion and mental health courts).

Third, we isolated the most rigorous study or studies for each program. We placed studies that randomly assign offenders to the program versus comparison condition at the top of the evidence hierarchy, given that experimental designs are the standard for drawing causal inferences about the effects of a program. For each program type, at least one experimental or quasi-experimental study was available. This allowed us to exclude single group, “pre-program, post-program” studies, which can inflate the apparent effects of a program (Weisburd, Lum, & Petrosino, 2001; cf. Pearson, Lipton, Cleland, & Yee, 2002). We also tried to omit studies that exclude offenders who drop out of the program, as this inflates apparent effectiveness (Lowenkamp, Latessa, & Holsinger, 2006).

The evidence is summarized in the last columns of Table 1 and in Table 2. Table 1 describes the most rigorous or “focal study” available for the program, including effects of the program on recidivism and the hypothesized mediator of symptom reduction. Because more than one rigorous quasi-experimental study had been published for two program types, Table 2 describes additional studies of jail diversion and mental health courts.

Table 2 Additional controlled studies of jail diversion and mental health court programsa
Full size table

What conclusions may be drawn from the evidence that we have identified and distilled in these tables? First, there is, at best, a mixed evidence that these programs as a whole reduce recidivism. Second, the evidence base seems the weakest for the mental health-based models (FACT, FICM) and for jail diversion programs, which vary substantially, but also tend to rely heavily on case management. As summarized by Morrissey et al. (2007), “the supporting evidence about the effectiveness of FACT in reducing arrests and keeping people out of jail is weak. Moreover, there is no compelling evidence that FICM can produce positive results at a reduced cost” (p. 537). Similarly, jail diversion usually increases time in the community (as it diverts individuals from incarceration), but often has little or no effect on rates of re-arrest. In fact, of participants in one jail diversion program, over one in five were re-diverted after a re-arrest within 18 months of their first diversion (Boccaccini, Christy, Poythress, & Kershaw, 2005). Third, the evidence for recidivism reduction is mixed, but not quite as weak, for criminal justice-based models that emphasize supervision by specialized courts or probation officers. Similarly, the one small study in our entire sample that included any emphasis on “criminal thinking” (Sacks et al., 2004)—an evidence-based correctional practice (Pearson et al., 2002)—looked promising.

Which Path Should Be Followed Now to Maximize Recidivism Reduction?

Possibility #1: Better Implement the Current Policy Model

Why are the contemporary programs for offenders with mental illness consistently and meaningfully not achieving their chief policy goal? One possibility is that programs vary in their fidelity to the basic policy model. Our review revealed no direct evidence for this model, i.e., that recidivism reduction is mediated by mental health services or symptom improvement. Still, this may be because programs often link offenders to mental health services that are ineffective or otherwise of low quality. If so, then they miss an essential “link” in the model, i.e., criminal justice involvement → provision of effective mental health services.

Although intuitively appealing, this possibility rests on little evidence. First, in experiments, even evidence-based mental health services (i.e., those that reliably affect clinical outcomes) have not affected criminal justice outcomes. Based on a sample of 223 patients with co-occurring disorders who were randomly assigned to ACT versus standard case management, Clark, Ricketts, and McHugo (1999) found no treatment-related difference in police contacts (80%) and arrests (44%) over a 3-year period. In another randomized controlled trial for patients with co-occurring disorders, Calsyn, Yonker, Lemming, Morse, and Klinkenberg (2005) found no treatment-related difference in arrests and incarcerations between those assigned to ACT, Integrated Dual Diagnosis Treatment (IDDT), or treatment as usual. Similar results were obtained for a sample of offenders with co-occurring disorders who were randomly assigned to IDDT or treatment as usual (Chandler & Spicer, 2006). Given such results, scholars have cautioned that positive outcomes observed for evidence-based mental services (e.g., reduced hospitalization, improved symptoms) will not necessarily extend to criminal behavior, and have called for “interventions that specifically target reduction of criminal behavior” (Calsyn et al., 2005, p. 245; see also Morrisey et al., 2007).

Second, there is no evidence for the current model’s implied link between symptom control or reduction and reduced recidivism. According to existing data, offenders who (for whatever reason) show symptom improvement during a program are no less likely to recidivate than those whose symptoms remain unchanged or worsen. Based on over 1,000 participants in a multi-site jail diversion study, Steadman, Dupius, and Morris (2009) found that no significant relationship between symptom reduction and the number of re-arrests over time. Similarly, based on approximately 360 participants in a study of specialty probation, Skeem et al. (2009) found that trajectories of symptom change were unrelated to the probability of arrest or revocation over a 1-year period.

Together, these studies cast doubt on the possibility that the problem lies with fidelity to the current policy model. Although some programs reduce recidivism, there is no evidence that they do so by linking individuals with evidence-based psychiatric treatment or by achieving symptom reduction.

Possibility #2: Explicitly Revisit and Expand the Current Policy Model

Given the available evidence, we believe that the most promising path toward improving outcomes for this population will require an explicit revision of the current policy model. If this model is inaccurate or incomplete, then even a program with excellent fidelity will not reduce recidivism. A viable explanation for the failure of modern programs to consistently meet their policy goal is that the criminalization hypothesis does not fully account for the link between mental illness and crime.

There is no evidence for the basic criminalization premise that decreased psychiatric services explain the disproportionate risk of incarceration for individuals with mental illness. Systems level data indicate that the probability of incarceration for people with mental illness cannot be predicted by (a) the closure of psychiatric inpatient beds (Erickson, Rosenheck, Trestman, Ford, & Desai, 2008; Steadman, Monahan, Duffee, Hartstone, & Robbins, 1984), (b) changes in the organization and financing of public mental health services (Norton, Yoon, Domino, & Morrissey, 2006), or (c) the availability of mental health services in the community (Fisher, Packer, Simon, & Smith, 2000).

In fact, there is little evidence that the risk of incarceration has uniquely increased for those with mental illness. Frank and Glied (2006) examined changes in the estimated living arrangements for people with serious and persistent mental illness (SPMI) in the USA from 1950 to 2000. During this period, the proportion of people with SPMI living in psychiatric institutions dropped by 23%, whereas the proportion living in correctional institutions rose only 4%. The rise in incarceration rates for those with SPMI follows a predictable pattern, remaining at 1% from 1950 to 1970, but rising to 3% by 1990 and 5% by 2000. As a function of “get tough on crime” policies, incarceration rates for the entire population—most of whom do not have SPMI—grew sharply in the 1980s and 1990s (Bureau of Justice Statistics, 2009). As Frank and Glied (2006) conclude, “it would be a mistake to attribute the increase in…incarceration among people with SPMI directly to the experience of deinstitutionalization” (p. 128); instead, the increase in this “undesirable circumstance” seems shared with the general population.

What is needed to shape more informative research and more effective interventions is an explicit conceptual framework that looks beyond mental illness as the principal cause of and solution to the problem of criminal justice involvement. Plausible alternatives to the criminalization hypothesis assume that the etiology of criminal behavior largely is shared by offenders with- and without-mental illness. In the following text, we describe these alternatives, review evidence bearing on competing theories, and offer an integrative conceptual model that attempts to outline a path for future research and policy development. Because different processes may be involved, our model explicitly distinguishes between recidivism that involves new crimes and “recidivism” that may not.

Available Alternative Theories and Relevant Evidence

Explaining Criminal Behavior

Theoretical Alternatives to Criminalization

There are two plausible general alternatives to the criminalization hypothesis. First, criminological models emphasize “the individual’s position in the social hierarchy” (Bonta, Law, & Hanson, 1998, p. 124). Formal criminological theories posit different mechanisms (e.g., inadequate or harsh parental discipline → low self control → crime; Gottfredson & Hirschi, 1990), some of which are elegantly applied by Silver (2006) to the link between mental illness and violence. Because specific theories lie beyond the scope of the present article, we focus on one broad criminological perspective. That is, people with mental illness “engage in offending and other forms of deviant behavior not because they have a mental disorder, but because they are poor. Their poverty situates them socially and geographically, and places them at risk of engaging in many of the same behaviors displayed by persons without mental illness who are similarly situated” (Fisher et al., 2006, p. 553). Poverty can force people to live in “settings that are rife with illicit substances, unemployment, crime, victimization, family breakdown, homelessness, health burdens, and a heavy concentration of other marginalized citizens” (Fisher & Drake, 2007, p. 546).

Second, social/personality psychology models focus on individual and proximate risk factors for offending like antisocial cognition and criminal associates (e.g., Andrews, in press; Andrews, Bonta, & Wormith, 2006; Gendreau & Goggin, 1997). A leading model suggests that criminal behavior largely is learned via early modeling and reinforcement patterns. Four major factors maintain ongoing criminal activity: “an established history of benefitting from criminal activity, a social environment that encourages and tolerates crime and criminals, personal attitudes and values supportive of criminal behavior, and a personality style that finds impulsive high-risk behavior rewarding” (Bonta et al., 1998, p. 138). Andrews et al., (2006) opine that “the predictive validity of mental disorders [for criminal justice involvement] most likely reflects antisocial cognition, antisocial personality pattern, and substance abuse” (p. 10). Reading between the lines, Andrews and colleagues may assume that a third variable associated with mental illness (e.g., adverse social environments) increases exposure to modeling and reinforcement patterns that “teach” or program antisocial behavior. As will be shown, there is indirect evidence for both social psychological and criminological alternatives to the criminalization hypothesis.

Summarizing Evidence for Theoretical Alternatives

These alternatives are consistent with three bodies of evidence indicating that major predictors of violence and recidivism are not unique to offenders with mental illness, but instead shared with general offenders. First, although the criminalization hypothesis posits that violence is typically the product of untreated psychosis or psychiatric deterioration in this population (see Torrey et al., 2002), available evidence suggests the opposite. As a whole, a large body of research indicates that “risk of violence is modestly elevated for people with mental disorder, particularly those who misuse substances” (Silver, 2006, p. 685). Still, most people with mental illness are not violent, most violent offenders are not mentally ill, and the strongest risk factors for violence (e.g., past violence) are shared by those with- and without- mental illness (see Link & Stueve, 1995; Monahan et al., 2001; Mulvey, 1994; Walsh, Buchanan, & Fahy, 2002). Moreover, the link between psychosis and violence is particularly weak among offenders (e.g., Bonta et al., 1998; Quinsey, Harris, Rice, & Cormier, 2006), perhaps because the base rate of violence is high and the strongest risk factors are well represented, leaving little room for the modest role that mental illness plays in other contexts (see Buchanan, 2008). Based on a meta-analysis of 204 diverse studies and samples, Douglas, Guy, and Hart (2009) found a small correlation between psychosis and violence (r = .16 or OR = 1.53). However, there was no meaningful correlation for offenders with mental illness (r = .00 or OR = 0.91) and general offenders (r = .01 or OR = 1.27).

Second, there is little evidence that offenders with mental illness recidivate because of (uncontrolled) symptoms or other clinical factors (e.g., Callahan & Silver, 1998; Monson, Gunnin, Fogel, & Kyle, 2001; Phillips et al., 2005). In a meta-analysis of 58 prospective studies of offenders with mental illness (70% with schizophrenia), Bonta et al. (1998) found that clinical variables (e.g., diagnoses, treatment history) did not meaningfully predict a new general offense (r = −.02) or a new violent offense (r = −.03). Instead, the strongest predictors of a new violent offense (r > .20) were antisocial personality, juvenile delinquency, criminal history, and employment problems.

Third, there are suggestions that this population’s disproportionate risk may be based on their having even more general risk factors for recidivism than their relatively healthy counterparts. The Levels of Services Inventory/Case Management Inventory (LS/CMI; Andrews, Bonta, & Wormith, 2004) assesses eight robust risk factors for recidivism. Based on a matched sample of 221 parolees with- and without-mental illness, Skeem et al. (2008) found that those with mental illness obtained significantly higher scores on the LS/CMI (η = .20), particularly on the antisocial pattern subscale (e.g., early or diverse criminal behavior, criminal attitudes, pattern of generalized trouble). Similarly, based on a sample of 600 probationers, Girard and Wormith (2004) found that those with mental health problems (n = 169) obtained higher scores on the LS/CMI than those without such problems. In turn, the LS/CMI predicts recidivism equally well for those with- and without-mental illness (Andrews et al., 2004; Girard & Wormith, 2004).

The latter findings are consistent with the social/personality perspective that these offenders are at risk not because they are mentally ill, but because they disproportionately experience key factors (e.g., antisocial pattern) that proponents believe establish and maintain ongoing criminal activity (Bonta et al., 1998). However, the social/personality model’s mechanism remains opaque, as there have been no direct investigations of whether disadvantaged environments or other third variables increase exposure of those with mental illness to modeling and reinforcement patterns that “teach” or program these key risk factors.

Similarly, only indirect evidence bears on the criminological perspective that those with mental illness offend because they are poor, which exposes them to risk factors and risky situations (see Draine, Salzer, Culhane, & Hadley, 2002). That is, offenders with mental illness tend to live in disadvantaged neighborhoods (Dickinger, Eno Louden, Robinson, Troshynski, & Skeem, 2008), be under- or unemployed (see Prins & Draper, 2009), have histories of victimization (Prins & Draper, 2009), abuse substances (Abram & Teplin, 1991; Abram et al., 2003), and associate with people who have criminal histories, drink heavily, and use drugs (Skeem, Eno Louden, Manchak, Vidal, & Haddad, 2008). Although each of these variables has been linked with criminal behavior, the extent to which they play a causal role has not been established.

Why the Criminalization Hypothesis Should Not Be Discarded

The data reviewed thus far provide robust, if indirect, support for criminological and social/personality models as alternatives to the criminalization hypothesis. That is, (a) incarceration rates for this population cannot be explained by the availability of psychiatric services, but instead seem to have risen alongside those of offenders without mental illness as a function of “get tough on crime” policies; (b) the strongest predictors of violence and crime are the same for offenders with- and without-mental illness; and (c) offenders with mental illness have more of these general risk factors than their relatively healthy counterparts.

Nevertheless, the criminalization hypothesis remains viable as a component of the policy model. Why?—chiefly because there is evidence that criminal behavior is directly attributable to mental illness for a small subgroup of offenders. Junginger, Claypoole, Laygo, and Cristiani (2006) conducted post-booking interviews with 113 arrestees with mental illness and co-occurring substance abuse disorder who were eligible for a jail diversion program. Raters reviewed interview data and police reports to reliably rate the probability that mental illness caused the index offense. Direct and indirect effects were defined as the influence of delusions or hallucinations, or any other symptom (e.g., confusion, depression, irritability etc.), respectively, on the offense. Of these offenders, 8% had been arrested for offenses that their psychiatric symptoms probably-to-definitely caused, either directly (4%) or indirectly (4%). The authors conclude: “persons with serious mental illness may be overrepresented in jails and prisons, but we can offer little evidence…that it was their illness that got them there” (p. 881). Still, mental illness got some of them there.

Similar results were obtained in a study of the crime patterns of “deeper end” offenders. Peterson, Skeem, Hart, and Vidal (2009) studied a matched sample of 221 parolees with- and without serious mental illness who had an average of three prior arrests. Based on the interview data and a review of parolees’ records, raters reliably classified each offender into one of five patterns of offending. The pattern of offending for the vast majority of parolees—mentally ill (90%) or not (68%)—was “reactive,” reflecting trait anger and impulsivity. Only 5% of parolees with mental illness manifested a pattern that was attributable to psychotic symptoms and only 2% fell in the disadvantaged or survival crime group. Thus, although most had patterns of offending similar to those without mental illness, a minority (7%) of the mentally ill sample clearly fit the criminalization hypothesis.

A similar process may describe the link between mental illness and violence for some psychiatric patients. Based on a sample of over 608 violent incidents that involved psychiatric patients enrolled in the MacArthur Violence Risk Assessment study, 11% were rated as having occurred while patients were delusional or hallucinating (Monahan et al., 2001).

Explaining Other Forms of “Recidivism”

Beyond explaining the criminal behavior of a small minority of offenders with mental illness, the criminalization hypothesis may also help explain “recidivism” that occurs without criminal behavior in the broader population. Criminal justice involvement may deepen via revocation or reincarceration, even in the absence of new crimes and new victims. If the goal is to end criminal justice involvement, the policy model must take this form of “failure” into account.

Applying the criminalization hypothesis, community supervision officers may attach a criminal label to deviant behavior by these offenders that permits (inappropriate) revocation of release. Although more research is needed, some evidence suggests that officers and judges apply lower thresholds for revoking community supervision, as a function of mental illness. Offenders, both with- and without-mental illness, are about equally likely to be re-arrested for a new offense (Bonta et al., 1998; Gagliardi, Lovell, Peterson, & Jemelka, 2004). However, those with mental illness are significantly more likely to commit technical violations and to have their community terms suspended or revoked (Eno Louden & Skeem, in press; Porporino & Motiuk, 1995).

These observations are consistent with findings that correctional officers respond conservatively to offenders with mental illness, perhaps out of fear or paternalism. Based on an experiment conducted with 264 probation officers who read case vignettes, Eno Louden, Gillig, and Skeem (2009) found that mental illness (particularly schizophrenia) increased officers’ perceptions of violence risk and promoted plans to keep the probationer under close surveillance and on a “short leash” (see also Callahan, 2004; Skeem, Encandela, & Eno Louden, 2003). Lynch’s (2000) ethnography suggests that reincarceration sometimes is inappropriately used for parolees in emotional crisis. In one case, a psychotic parolee who disclosed suicidal thoughts was arrested and “taken to the county jail for his safety” (p. 52). Similarly, based on a sample of over 200 probationers with mental illness, Solomon and Draine (1995) found that case managers often collaborated with probation officers to seek reincarceration on a technical violation (and jail-based treatment) for those who were perceived to be noncompliant with treatment. Together, these findings are consistent with the notion that some supervision “failures” reflect criminalization of mental illness rather than new crime.

Conceptual Framework for Expanding Research and Policy

Moderated Mediation Effect of Mental Illness on Criminal Behavior

The theories and evidence reviewed above can be integrated to outline a conceptual framework for future research, policy, and intervention. As shown in Fig. 1, the criminalization hypothesis and current policy focus suggest that there is a direct relationship between mental illness and criminal behavior. This model may apply to a small subgroup of offenders with mental illness (perhaps one in ten). Also shown in Fig. 1, the alternative models (e.g., criminological and social/personality) suggest that the relationship between mental illness and criminal behavior is fully mediated by a third variable (e.g., poverty, social learning) that establishes general risk factors for crime. This model may apply to the vast majority of offenders with mental illness. Based on available evidence, then, we propose that the effect of mental illness on criminal behavior is one of “moderated mediation” (Baron & Kenny, 1986). That is, whether the effect is mediated or direct varies across subgroups of offenders with mental illness; although the effect is usually indirect, it depends on the subgroup to which an offender belongs.

Fig. 1
figure 1

Moderated mediation effect of mental illness on crime behavior

Full size image

As explained later, a major task for future research will be to identify the specific moderator(s) that differentiates the subgroups for whom the effect is direct versus mediated. One simple variable that may mark a potential moderator is age of onset for criminal behavior (i.e., childhood/adolescence versus adulthood). Hodgins (2000) proposes that there are two types of offenders with mental illness; “early start offenders exhibit a stable pattern of antisocial behavior from a young age, while late start offenders exhibit antisocial behavior only after the onset of major mental disorder” (p. 91). As explained by Silver (2006), “from a theoretical standpoint, the heart of the distinction between early- and late-start offenders is that early start offenders are, from the beginning, more deeply embedded in and exposed to criminogenic risk factors both in themselves and in their social environments” (p. 700). For late starters, mental illness seems to play a more causal role in offending. There is some support for this distinction, particularly for violent behavior (e.g., Swanson et al., 2007; for a review, see Hodgins, 2008).

Partial Mediation of Other Forms of “Recidivism”

The effect of mental illness on criminal behavior may differ from its effect on “recidivism,” i.e., revocation and reincarceration without a new crime. Although more research is needed, available evidence suggests that the effect of mental illness on “recidivism” is partially mediated by system bias (officers’ stigma, close monitoring, paternalism, etc.). This process is depicted in Fig. 2.

Fig. 2
figure 2

Partially mediated effect of mental illness on other “recidivism”

Full size image

Using the Model to Revisit Program Effectiveness

The proposed model (a) expands explanatory mechanisms beyond the psychiatric realm to incorporate criminological and social/psychological factors, and (b) disaggregates recidivism to account for different processes that may explain criminal behavior versus “recidivism” for other reasons. This model may help explain why contemporary programs are not consistently reducing recidivism (see Tables 1, 2). On the one hand, because they are run by specialty officers, judges, and others with interests in mental health, some of these programs theoretically would reduce system bias that partially mediates the relationship between mental illness and “recidivism” (i.e., revocation and reincarceration without new crime).

On the other hand, the effect of these programs on the true recidivism target—criminal behavior—would be limited and could vary with group composition. Specifically, effective psychiatric treatment would reduce recidivism for the small subgroup of offenders for whom mental illness has a direct effect on criminal behavior. It is less likely that such treatment would reduce recidivism for the larger subgroup, where the effect of mental illness is fully mediated by general risk factors.

For this larger subgroup, it is possible that effective treatment would improve psychosocial functioning, which would promote a less criminal lifestyle, which would reduce criminal behavior (see Silver, 2006). However, this scenario seems unlikely to occur readily or often. First, it may be difficult to simply “undo” the causal process. From a criminological perspective, even if mental illness contributed to downward socioeconomic drift, it is unlikely that symptom improvement will reverse poverty or associated criminogenic factors that are more socioeconomic than medical (e.g., homelessness, victimization; Fisher & Drake, 2007). Second, the factors that originally caused criminal behavior may differ from those that maintain it. From a social/personality perspective, it is unlikely that symptom improvement will change such factors as “an established history of benefitting from criminal activity” and “personal attitudes and values supportive of criminal behavior” (Bonta et al., 1998, p. 138).

Anecdotes gathered from working with front line staff in programs across the country suggest another reason for variability in effectiveness. Sometimes, everyday practice may be more nuanced than the explicit policy focus on mental health would suggest. As shown later, to the extent that staff intuitively focus on changing an individual’s general risk factors for crime, programs may be more effective in reducing criminal behavior than those that bank more exclusively on psychiatric services.

Implications for Advancing Research and Practice

Having articulated a conceptual framework that is consistent with, but goes beyond, available evidence, we now outline how it can be further tested and applied to develop a more coherent and effective policy for supervising and treating subgroups of offenders with mental illness (i.e., developing real “evidence-based practice”). We also provide recommendations for practice, based on what is known now.

Priority #1: Identifying Offenders for Whom Mental Illness Directly Causes Criminal Behavior

Research and Policy Development

For perhaps one in ten offenders in this population, the covariation between their most visible status (mental illness) and present state (justice involvement) is not merely an illusory correlation. For this subgroup, the current policy emphasis on linkage with psychiatric services should go far in preventing new victims and new arrests. A major task for future research is to identify the moderator(s) (e.g., demographic, socioeconomic, clinical, criminal, or other factors) at work in the “moderated mediation” model. As noted earlier, one promising potential marker is late age of onset for criminal behavior. Once a method is identified for isolating this subgroup, we can determine whether effective psychiatric treatment alone reduces recidivism risk.

Current Practice

It is important not to lose sight of the fact that mental illness is strongly related to criminal behavior for subgroup of offenders…while keeping the subgroup’s size in perspective. Even among those with psychosis, symptoms directly cause crime for only a small fraction of offenders. Thus, there is an onus on practitioners to clearly demonstrate that an individual belongs to this subgroup. Until an evidence-based method is developed to do so, practitioners will have to articulate how an individual’s pattern of offending can be directly attributed to symptoms, and demonstrate this with data from that individual’s history (see Skeem & Mulvey, 2002).

Once identified, individuals in this subgroup should respond to effective and/or evidence-based psychiatric treatment with a reduction in recidivism. Still, all offenders with mental illness, including those in this subgroup, must have a basic level of “good enough” community supervision. Although officers who establish authoritarian relationships and rely almost exclusively on such negative pressures as threats of incarceration may be detrimental to all offenders, those with serious mental illness seem particularly vulnerable to such influences (see Skeem, Eno Louden, Polaschek, & Camp, 2007). If so, poor supervision practices (e.g., an exclusive focus on surveillance within the context of an authoritarian relationship or an exclusive focus on offender needs within the context of a permissive relationship) may undermine the positive effects of treatment.

Priority #2: For the Vast Majority, Expanding the Focus to Incorporate—or to Explicitly Identify—Evidence-Based Corrections

Contemporary policy is likely to have a limited effect on recidivism for the large subgroup of offenders for whom the relationship between mental illness and criminal behavior is indirect. As suggested earlier, it may be that the causal process cannot be reversed via the same meditational route, and/or the factors that caused criminal behavior differ from those that maintain it. These are testable propositions. However, from a practical standpoint, it may not matter that mental illness is a distal contributor to criminal behavior; the focus must be expanded beyond linkage with psychiatric treatment. Two directions hold particular promise for arriving at a coherent policy model for this group: (a) adapting existing evidence-based correctional programs to the abilities of these offenders and evaluating their effectiveness, and (b) isolating the ingredients of existing programs that reduce recidivism.

Research and Policy Development: Leveraging Evidence-Based Corrections

The time is ripe to apply evidence-based corrections to offenders with mental illness and evaluate their effectiveness. Cognitive behavioral group treatment programs like “Reasoning and Rehabilitation,” “Moral Reconation Therapy,” and generic versions that reduce criminal thinking and build pro-social skills have been shown to reduce general offenders’ recidivism (Aos, Miller, & Drake, 2006; Landenberger & Lipsey, 2005; Pearson et al., 2002). One of these programs has been adapted to address the cognitive limitations that some offenders with mental illness have (Young & Ross, 2007), but its effect on recidivism has yet to be tested. This is a clear direction for research and policy development.

These packaged programs are consistent with principles of intervention that embody the dominant model of “what works” for offenders; “Risk-Need-Responsivity” (RNR; Andrews, in press). RNR is underpinned by the social/personality theory outlined above. According to that theory, contingencies must be changed to increase rewards and satisfactions for alternatives to criminal behavior. For example, for an antisocial pattern, emphasis would be placed on building skills in problem solving and self-regulation and reinforcing use of these skills.

What are the “RNR” principles? Research indicates that offenders are less likely to recidivate when programs match the intensity of supervision and treatment to their level of risk for recidivism (R isk principle), target their criminogenic needs, or changeable risk factors for recidivism (N eed principle), and match modes of service to their abilities and styles (R esponsivity principle; see Andrews et al., 2006; Lowenkamp, Pealer, Smith, & Latessa, 2006; Lowenkamp et al., 2006). The last principle is particularly germane. “The most effective programs for reducing recidivism are those that target needs closely related to criminality” (Bonta et al., 1998, p. 138). Specifically, the effectiveness of correctional programs in reducing recidivism is positively associated with the number of criminogenic needs they target (i.e., dynamic risk factors for crime, like procriminal attitudes), relative to noncriminogenic needs (i.e., disturbances that impinge on an individual’s functioning in society, like depression; Andrews et al., 2006). Because mental illness is not a criminogenic need for this subgroup, it is important to target stronger risk factors for crime.

At the same time, in revising policy for this group, the role of psychiatric treatment should be contextualized, not jettisoned. We assume that all offenders with serious mental illness must have a basic level of “good enough” mental health services (see Prins & Draper, 2009). First, as suggested above, correctional programs are more effective in reducing recidivism when services are matched to the abilities, styles, and needs of offenders. Effective psychiatric treatment may complement correctional treatment by, for example, reducing hallucinations that interfere with an offender’s ability to attend to, and benefit from, cognitive behavioral sessions that target criminal thinking. Second, even if mental health services have no effect on recidivism (direct or indirect), they may achieve crucial public health outcomes for offenders with mental illness (e.g., reducing symptoms, substance abuse, and hospitalization). For example, an ACT team may not reduce recidivism, but is quite likely to reduce repeated hospitalizations.

Research and Policy Development: Isolating Effective Components of Existing Programs

Our review suggests that some programs for offenders with mental illness “work” to reduce recidivism, but there is no evidence that they do so for the reasons assumed. An important goal for future research is to identify the mechanisms by which programs reduce recidivism. Understanding what is critical to treatment and how it operates will help develop interventions that are fewer, more efficient, and more effective in respect of offenders with mental illness (see Kazdin, 2007). The existing wealth of program operationalizations only underscores the need to identify change mechanisms and bring greater parsimony to the field. This is essential for developing model programs that can be widely disseminated. At a local level greater than that exists now and in today’s economic environment, policymakers should insist on knowing why programs work because this will enable them to streamline programs while protecting their most essential elements.

For most offenders with mental illness, we suspect that existing programs work—when they work—in some of the same ways that programs for general offenders work. First, we have found that specialty probation reduces risk of recidivism less because of mental health service linkage or symptom reduction than because specialty officers more than traditional officers are likely to apply core correctional practices (Andrews, in press) like establishing firm, but fair and caring, relationships with offenders (Skeem et al., 2007), and using problem-solving strategies rather than threats of incarceration (Skeem, Manchak, Johnson, & Gillig, 2008). Similarly, the review above hints that existing programs may be effective when they involve ongoing specialty judicial or correctional supervision or some focus on criminal thinking. Second, given the “medicalization” of socioeconomic problems like unemployment and homelessness (Fisher & Drake, 2007), some psychiatric services (e.g., supported employment) may actually help mitigate the “‘criminogenic’ effects of unhealthy social environments” (see Fisher & Drake, 2009, p. 2) that affect all the offenders. Third, anecdotally, we have learned that staff members in mental health courts or other programs sometimes target factors that get an offender in trouble (e.g., hanging out with her/his drug dealing cousin). If so, then they are intuitively applying the RNR principle of targeting criminogenic needs. These hypotheses can be tested to make “active program ingredients” an explicit part of policy.

Current Practice

Much needs to be done to develop a specific evidence base for the large subgroup of offenders for whom the effect of mental illness on criminal behavior may be indirect and fully mediated. In the interim, two existing evidence bases can be referenced to tailor practice to individual offenders in this group, who differ substantially in their (a) degrees of risk for recidivism, based on core risk factors (e.g., antisocial personality pattern; criminal history), and (b) levels of psychopathology and functional impairment. These two dimensions correspond to evidence-based practices: (a) in corrections that have been shown to reduce recidivism risk for offenders, and (b) in mental health services that have been shown to achieve positive patient outcomes.

As shown in Fig. 3 (adapted from Prins & Draper, 2009), an offenders’ constellation of criminogenic risk and clinical impairment can inform sentencing and supervision in a manner that references the strongest evidence for addressing these often-different problems. For those at considerable risk for recidivism, evidence-based correctional principles should help reduce that risk. For those with pronounced clinical impairment, evidence-based mental health practices like IDDT, supported employment, supported housing, or even ACT should help reduce hospitalization, symptoms, or employment or housing problems (for a review, see Osher & Steadman, 2007). This model reserves the most intensive supervision and treatment resources for those with the greatest criminogenic risk and clinical need. For low-risk, low-need individuals, “good enough” supervision practices and psychiatric treatment in the community should suffice. Although these propositions rest on sound bases of evidence at present, their effect “in tandem” must be tested in future research.

Fig. 3
figure 3

Matching evidence bases to policy goals

Full size image

Priority #3: Assessing and Addressing System Bias

Thus far, we have presented priorities for better responding to individuals to reduce criminal behavior. These efforts must be complemented by an effort to assess and address the role that system bias can play in “recidivism” where there is no criminal behavior. The proposed model posits that the relation between mental illness and supervision failure is partially mediated by system bias.

Research and Policy Development

More intensive research is needed to increase the understanding of the specific reasons for “recidivism.” Offenders with mental illness appear particularly likely to incur technical violations that can lead to supervision failure (e.g., Eno Louden & Skeem, in press). To what extent is this because those with mental illness (a) have functional impairments that reduce their ability to adhere to such standard conditions of community release as maintaining employment or paying fines and fees (see Skeem & Eno Louden, 2006), (b) are required to abide by more conditions of release (e.g., mandated treatment) than those without mental illness, and/or (c) are subject to increased monitoring and control, which increases the likelihood that minor infractions will be detected (Skeem et al., 2008)? The answer to such questions can inform targeted policy reforms designed to reduce supervision failure, via modification of the conditions of release, stigma-sensitive training of supervision officers, or both.

Practice Implications

When they are behaving no worse than offenders without mental illness, it seems inappropriate to use incarceration to achieve social control over offenders with mental illness, regardless of whether this is motivated by fear or paternalism. It is important to remain mindful of our tendency to more closely watch offenders with mental illness and to more forcefully respond to their behavior. Even if evidence-based strategies in mental health and corrections are ideally matched to subgroups of offenders with mental illness, these individuals will continue to “fail” as long as we maintain an unusually high threshold for their success. We hope that the framework we have offered will facilitate continued advances in science and practice that inform one another to improve policy and outcomes for this large and important group.