The authors are affiliated with Policy Research Associates, Delmar, New York. Send correspondence to Ms. Morris, Policy Research Associates, 345 Delaware Ave., Delmar, NY 12054 (e-mail: lmorris@prainc.com).

In recent years the number of jail diversion programs for people with mental illness has dramatically increased throughout the United States (1). These programs share the common goal of circumventing, or significantly reducing, jail time through the linkage to community-based behavioral health services. Although these programs share this general vision, little is known about the process of deciding which individuals are diverted.

Few jail diversion studies provide information about individuals who are referred for diversion but who are ultimately not diverted. Instead, much of the literature on jail diversion has been descriptive in nature, including results from national prevalence surveys (2,3,4,5) and descriptions of jail diversion programs, program components, or program participants (6,7,8,9). Some recent articles report findings from single-site (10,11,12,13,14,15,16,17,18,19,20,21) and multisite (1,22,23,24) jail diversion outcome studies.

Only two studies have specifically addressed the decision-making process for diverting people with serious mental illness (25,26). Luskin's (25) study of a court-based diversion program found that a history of felony convictions, a current charge of a crime against a person, and being male decreased chances for diversion. In addition, Luskin found an interactive effect of age and gender, in that older males and younger females were more likely to be diverted, suggesting that youth signals danger for men but not for women. Interestingly, Luskin further suggested that the legal variables were ultimately less important than age and gender. Steadman and colleagues' (26) study of case processing in seven mental health courts showed that mental health court clients were more likely than individuals in the general criminal justice system to be older and white and to be women. Additionally, Steadman and colleagues found that age, gender, race, and the nature of the charges did not predict acceptance among cases reviewed by the court, suggesting that the overrepresentation occurred at the point of referral to the court and not at the point of the court's decision to accept or reject. Generally, the findings of jail diversion studies agree with those of Steadman and colleagues that overrepresentation of certain subpopulations exists, but it is still unclear where in the diversion process this disproportionate representation occurs.

The study reported here examined the process of determining who is diverted through an examination of program and court decisions related to jail diversion. First, we describe the volume of activities involved in determining who will be diverted and the characteristics of individuals considered for jail diversion by the programs. Second, we present the outcomes of program recommendation and court acceptance decisions. Third, we examine whether there are factors that distinguish between those who are recommended and rejected for diversion by program staff, as well as between those ultimately accepted and rejected by the court.

This study drew on data from 20 grantees funded in 2002, 2003, and 2004 through the Targeted Capacity Expansion (TCE) Jail Diversion Initiative of the Substance Abuse and Mental Health Services Administration (SAMHSA). Although many grantees operated multiple tracks, our study examined data from the postbooking tracks only, representing 18 grantees and 21 program tracks. The SAMHSA grantees entered data on all diversion determination activities into a Microsoft Access database created and supplied to grantees by the Technical Assistance and Policy Analysis Center for Jail Diversion, the coordinating center for this initiative. The data considered for this study were collected between February 2003 and July 2005.

The database was structured to reflect the underlying logic of jail diversion decision making. Jail diversion programs engage in a series of activities to determine whether individuals meet program criteria (for example, screening by a jail booking nurse, clinical and criminal justice assessments and record checks, and psychiatric evaluations). Formal eligibility criteria for diversion vary by program but always include psychiatric, criminal justice, and jurisdictional criteria. Many programs consider exceptions to formal criteria on a case-by-case basis. For individuals who meet program eligibility criteria and who agree to be considered, the program presents a diversion plan to the court, prosecutor, and public defender. The court then makes a determination about whether to accept the individual for diversion. Assuming that the court accepts the diversion plan and the person and his or her legal representative agree, the person is then diverted.

Following this logic, the database was organized by type of activity and captured each key decision. Activities were placed into one of four categories—initial screening, subsequent assessment, subsequent evaluation, and court decision. Within each category or activity, grantees recorded the relevant decisions: the eligibility decision as determined by the program (to recommend for diversion or not) and, if eligible, whether the individual agreed to proceed to court with the diversion plan; and the court decision to accept or not accept the diversion plan and, if accepted, whether the individual agreed to enter the diversion program. For each activity, grantees also recorded information on demographic characteristics (age, race or ethnicity, and sex), current criminal charge (most severe charge category and level), and reason or reasons rejected if applicable. No individual identifiers were entered into this database; therefore neither informed consent nor institutional review board approval was required.

Volume of activities

There were 42,518 jail diversion program activities—that is, screenings, assessments, and evaluations—during the study period, which resulted in 32,917 program decisions with an outcome of either rejection (N=30,916, or 93.9%) or recommendation for diversion (N=2,001, or 6.1%). Of the 2,001 cases forwarded to the court, 1,237 (64.6%) were accepted and 678 (35.4%) were rejected. Information was not provided about 86 court decisions. This study examined the 34,832 activities that resulted in a program recommendation or rejection (N=32,917) or a court acceptance or rejection (N=1,915).

Characteristics of referrals

Table 1 presents the characteristics of individuals for whom the programs made a diversion decision. This sample shows distinctly different trends in both demographic characteristics and charge information when compared with data for the general U.S. arrestee population as reported by the Bureau of Justice Statistics (27). The diversion group was older (mean age of 34 years) than the general arrestee population (mean age of 32). Our group also had more than twice as many women (27%) and more non-Hispanic whites (48%) than the general arrestee population (12% and 36%, respectively). In addition, our study group had nearly one-third fewer violent charges (9%) than all U.S. arrestees (25%).

Decision outcomes

As shown in Table 2, although the programs engaged in tens of thousands of activities, only 6.1% of their decisions resulted in a recommendation for diversion. Of the individuals who were recommended for diversion, 88.0% agreed to be considered by the court. Of the majority who were rejected by the program, the most common reasons indicated were based on legal (26%) or psychiatric (19%) criteria. Legal reasons included those related to the nature of the charge, criminal history, and perceived dangerousness. Psychiatric criteria included reasons related to diagnosis, psychiatric history, and civil commitment. Nearly one-third (28%) were not recommended for multiple reasons.

In comparison, two-thirds (65%) of the individuals proceeding to the court decision stage had their diversion plans accepted by the court, and nearly all (94%) of those accepted agreed to enroll in the jail diversion program (Table 2). Of those whose plans were not accepted by the court, the most common reasons were legal criteria (16%) or other reasons (14%, which were bond related or were related to having been sentenced or being enrolled in another program).

Factors related to decision making

In addition to determining the characteristics of people who were considered for jail diversion, we were also interested in exploring whether any of these characteristics were directly linked to program and court decision outcomes. The bivariate results in Table 3 show large differences between individuals who were recommended (divert) and those who were rejected (not divert) across all variables, including age, gender, race or ethnicity, charge type, and charge level. For court decisions, the bivariate results of those accepted (divert) and those rejected (not divert) show differences in gender, charge type, and charge level. Because these comparisons can be misleading unless the analysis controls for relationships between these variables, we proceeded with a multivariate approach.

Analytically the goal of the multivariate analysis was to develop two models: one predicting program recommendation for diversion (yes or no) and one predicting court acceptance (yes or no). Traditional logistic regression models were not appropriate because of the high level of program variability. Preliminary analyses that grouped grantees according to shared program characteristics (program criteria, size of the jail, screening procedures, length of time until diversion, and so forth) found no explanatory trends for this variability. Therefore, we implemented a conditional logistic regression approach (28), which controlled for the substantial program design differences, including those related to the exclusion of specific subpopulations, such as individuals with felonies, violent offenses, and men.

This approach was used to model both dependent variables—program recommendation for diversion and court acceptance. The independent main-effect variables included in both models were age (in years, centered for interpretive reasons [calculation in Table 4]), female, non-Hispanic white, violent (current offense), and felony (current offense). The following interactions were also included in the models: age (centered)-by-female, felony-by-female, and age (centered)-by-felony.

Multivariate model of program decisions

As shown in Table 4, there were several significant predictors of program eligibility, including three significant main effects. For the three significant main effects, only the impact of violent offenses can be interpreted as a true main effect because it was not involved in a significant interaction. The model found that people with violent offenses were significantly less likely to be recommended for diversion (odds ratio [OR]=.58, p<.001). The two other significant main effects (female OR=1.31, p<.001, and felony offenses OR=.19, p<.001) can be interpreted only by simultaneously examining the interaction effects.

All three interaction terms were found to be significant predictors of eligibility. For the age-by-female interaction (OR=1.01, p=.007), the calculated odds of eligibility for a woman, all else constant, more than doubled between the ages of 20 years (OR=1.02) and 60 years (OR=2.10) (Table 4). However, no significant change was found for men in the odds of being recommended for diversion during this same time span (at 20 years, OR=.94; at 60 years, OR=1.13). Thus there was an age effect for women (aging improved their odds of being recommended for diversion) but no age effect for men. Furthermore, these findings showed that for the age range of this study, women were always more likely than men to be recommended; thus we may conclude that there is a main effect of gender.

For the felony-by-female interaction (OR=4.13, p<.001), the model illustrated that, all else constant, male felons had significantly lower odds of being recommended for diversion (OR=.19) than male nonfelons (OR=1.00), female felons (OR=1.05), and female nonfelons (OR=1.31). For the final interaction effect, age-by-felony (OR=.98, p<.001), the calculated odds of eligibility for a felon, all else constant, dropped by half between the ages of 20 years (OR=.25) and 60 years (OR=.12). However, no significant change was found during this same time span for nonfelons (at 20 years, OR=.94; at 60 years OR=1.13). Thus there was an age effect for felons (aging decreased their odds of being recommended for diversion) but no age effect for nonfelons. Furthermore, on the basis of these findings across both gender and our age range, we may conclude that there is a main effect of felony in that felons were always less likely to be recommended than nonfelons.

Multivariate model for court decisions

As shown in Table 4, a felony main effect was the only significant predictor of court acceptance (OR=.59, p=.010). As with program recommendations, individuals with felonies were significantly less likely to be accepted into the diversion program.

A major finding of this study is the previously undocumented large number of program activities that precede enrollment in a jail diversion program. Jail diversion programs engage in a large number of activities to enroll a relatively small number of people. A large proportion of program resources were put into program recommendation decisions (94% of referrals rejected for diversion), and without careful accounting of these efforts, program performance may be underestimated.

A second major aim of our analyses was to address the question, Who is diverted? Having data for all individuals who were considered for diversion by both the programs and the courts allowed us to examine the diversion decision-making process one step earlier than most previous studies. Our examination yielded results similar to those of Steadman and colleagues' (26) mental health court study in that individuals referred for diversion were disproportionately older, female, and white compared with arrestees nationwide.

In-depth analyses that looked specifically at program recommendation decisions found that both legal and nonlegal characteristics were salient factors related to who was recommended for diversion by program staff. In terms of legal factors, it was not surprising that violent and felony offenses decreased the chance of eligibility, given that some programs exclude individuals on the basis of either or both of these criteria. However, legal factors were significant even among programs that did not have these exclusionary criteria. Just as significant, we found that certain nonlegal factors were also explanatory in program recommendation decisions. In particular, gender mattered outright and in interaction with other variables, with women having an advantage in eligibility for jail diversion. Combined, these results show that women and individuals with less serious charges are generally considered more appropriate for jail diversion.

An even more interesting finding of our analyses of the program recommendation decisions was the dynamic relationship between legal and nonlegal factors. Gender seemed to have an especially interactive effect with other variables, working with both age and charge to differentially affect eligibility determinations. Older women appeared to be especially appealing for diversion, whereas men with felonies were particularly unappealing. This finding can be seen as representing extremes in terms of perceptions of the highest risk (male felons) and lowest risk (older women) for diversion. For people with felony charges, the additional negative effect of increasing age on program recommendations is noteworthy and might be explained by their criminal histories.

The analysis of the program recommendation decision data, in combination with the comparison of the study sample to national arrestees, suggests that overrepresentation of specific subpopulations is introduced at the point of referral for program consideration and is further pronounced in the program decisions. One explanation for this finding is that referrals and program decisions are influenced by subjective factors related to public fear and general beliefs about who is "deserving" of diversion. A second explanation is that there is inherent disproportionate representation in the jail population related to which subpopulations experience mental illness; in other words, older people, whites, and women are more likely to be referred and diverted because they are more likely to be identified as having a serious mental illness upon entry into the criminal justice system.

Although this study cannot directly address the presence of subjective factors, the effect of inherent factors can be examined by considering jail prevalence studies. A review of such studies indicates the following: women entering jail are at least twice as likely as men to be identified as having a serious mental illness (29,30,31); overall younger people in jail are more likely to be identified as having a mental health problem (32); although there may be differential effects for specific diagnoses or gender, mental health problems are more likely to be identified among whites in jail than among African Americans or Hispanics (30,31,32). These findings suggest that inherent disproportionate representation partially accounts for our findings. Specifically, because having a serious mental illness was a requirement at all sites and women involved in the criminal justice system are more likely to have a serious mental illness, it makes sense that women are more likely than men to meet diversion criteria.

At best, however, the prevalence studies provide mixed support for the idea of disproportionate representation as an explanation for our age effects. On the one hand, the suggestion of these studies that younger individuals may be overrepresented among those with favorable diversion decisions supports our findings that younger felons have an advantage over older felons in terms of recommendation (although, overall, felons were disadvantaged for diversion). On the other hand, the prevalence studies are not consistent with our findings that individuals referred for diversion consideration were on average older than the national arrestee population and that at the program recommendation stage, older age gave women an advantage for diversion. Inherent disproportionate representation also does not appear to be a factor in the recommendation or acceptance decisions involving whites; however, it may explain the higher rate of whites referred for program consideration.

Although the specific reasons behind overrepresentation of certain demographic groups early in the diversion decision-making process may not be entirely clear, what is clear is that judges generally accept program recommendations. The continuing negative impact of felony charges at the court decision stage may be accounted for by program or staff views that determination based on certain legal criteria is the domain of the court. However, excluding people with serious charges may not be justified on empirical grounds, because defendants with violent charges have been shown to be as successful in diversion programs as those with nonviolent charges (24).

Limiting further analysis of the diversion decision-making process is this study's lack of clinical data (including current psychiatric diagnosis and symptomatology and substance use diagnoses), criminal histories, and other in-depth information. Collecting such data, however, was beyond the practical scope of the Targeted Capacity Expansion Initiative. Nonetheless, our large, if limited, data set of diversion referrals substantially improves our current understanding in this area.

One major finding of this study is the extremely large number of activities required to divert a small number of individuals into jail diversion programs for people with mental illness. Beyond this, our analyses highlight that factors other than formal criteria influence jail diversion decision making throughout the program determination process. Findings reveal disproportionate representation in terms of demographic characteristics as early as the referral phase and the additional influence of demographic factors in the program decision phase. It is unclear whether, and to what degree, this reflects subjective factors on the part of the decision makers or different distributions within the jail population in terms of factors related to diversion criteria. Type and level of current charges also influenced decision making in the early program determination phases, with charge level (felonies) having a continuing impact on court decisions. These findings encourage further examination of program-level decision making to ensure the inclusion of all clinically and legally appropriate individuals.

This article is based on work supported by grant 1-H79-SM54722-01 from the Center for Mental Health Services, Substance Abuse and Mental Health Services Administration (SAMHSA). The contents are solely the responsibility of the authors and do not necessarily represent the official views of SAMHSA or the other participants. The authors are grateful to Steven Banks, Ph.D., for statistical advice. The authors are also grateful to the grantee programs in this initiative, which include the following (major cities in parentheses): Birmingham, Alabama; Hartford, Connecticut; Hawaii County (Hilo and Kona), Hawaii; Dubuque County, Iowa; Cumberland County (Portland), Maine; St. Louis County, Missouri; New York City; Tulsa County, Oklahoma; Richland County (Columbia), South Carolina; Chesterfield County, Virginia; Jackson County, Missouri; New Britain and Bristol, Connecticut; Lancaster County (Lincoln), Nebraska; Bexar County (San Antonio), Texas; Shelby County (Memphis), Tennessee (two programs); Miami, Florida; Anchorage, Alaska; Yakima County, Washington; Orange County (Orlando), Florida; and East Baton Rouge Parish, Louisiana.

The authors report no competing interests.