COST-BENEFIT ANALYSIS OF UNITED STATES MENTAL HEALTH COURTS RELATIVE TO CRIMINAL COURTS FOR PARTICIPANTS WITH MENTAL ILLNESS

COST-BENEFIT ANALYSIS OF UNITED STATES MENTAL HEALTH COURTS

RELATIVE TO CRIMINAL COURTS FOR PARTICIPANTS WITH MENTAL ILLNESS

Chapter 1. Introduction

1.1 Introduction to Mental Health Courts (MHCs)

Mental Health Courts (MHCs) were established in the United States for individuals with

mental illness, mostly with serious mental illness (SMI), to be diverted from criminal courts for

offenses that otherwise could result in a person going to jail. Services provided to MHC

participants are supervision of the court including reviews of progress, identification of needs,

mental health treatment, case management, and supervision of compliance that is no longer than

the longest allowable jail sentence or period of probation based on the charge or offense (United

States Department of Justice, 2022). MHCs also support participants to stay in the community

and out of the criminal justice system. They assist in securing employment and housing as well

(Almquist & Dodd, 2009). In this dissertation, costs and benefits of MHCs as compared to

criminal courts are measured by variables in the data that relate to MHC services. These

variables are discussed in detail in the below research methods sections. The analysis includes

program costs and outcome benefits, but it is primarily focused on developing measures of

outcome benefits and reducing related uncertainty.

This dissertation focuses on societal-level cost-benefit analysis that includes costs and

benefits to both the taxpayer and the individuals who are study participants. All of the individuals

who are study participants are diagnosed with mental illness.

The goals of MHCs are to: 1. Increase the safety of the public; 2. Reduce jail recidivism;

3. Provide and mandate successful treatment; 4. Increase the quality of life of participants, 5.

Increase effective post-court treatment; and 6. Reduce justice system costs. (Dodd, 2009). The

population of individuals with SMI is most of what is examined in this dissertation, but some of

the reach of the MHCs is broader than SMI and includes individuals with other diagnoses of

mental illnesses. This scope of MHCs is discussed in detail under the data source subsection in

the research methods section.

MHCs differentiate from conventional criminal courts by five standardized characteristics.

They (1) have dockets that are specialized based on mental health status, (2) require voluntary

diversion of participants from conventional criminal courts, (3) often divert individuals with

mental illness from trial and incarceration to supervised community-based treatment, (4) use

sanctions and rewards to incentivize compliance, and (5) are less formal than traditional criminal

courts but include regular progress updates with a judge (VanGeem, 2015). In addition to being

voluntary, individuals with mental illness can only be diverted to MHCs where they are available

in terms of geography, acceptance of specific criminal charge, and acceptance of type of potential

criminal court (e.g. jail or prison related criminal court).

MHCs have the ability to ameliorate harm and increase access to treatment. MHCs act as

alternatives to criminal courts for individuals with mental illness, mostly SMI, who are entering or

engaged in criminal courts. MHCs involve a less criminalized model that incorporates

efficacybased mental health treatment. MHCs also include periodic mental health assessments and

decriminalized evaluations with a Judge (Substance Abuse and Mental Health Services

Administration, 2020). The relevant literature demonstrates both criminal justice effectiveness

(e.g., reduced recidivism in court appearances and convictions) and clinical effectiveness (e.g.,

reduced relapse and greater clinical functionality) among participants in MHCs (Sarteschi,

Volume 39, Issue 1, 2011; Bradley, 2015). Research is limited, however, on the cost-benefit

analysis needed to support policies to increase the adoption of MHCs.

The research scope includes individuals with mental illness in MHCs diverted from

jailbased criminal courts and does not include prison-based criminal courts, Drug Courts, Veteran

Treatment Courts, Juvenile Courts, or other criminal courts that may treat mental illness in unique

ways. These other courts treat different, although sometimes overlapping, subpopulations of

individuals in the criminal justice system. Their techniques for treatment may partially overlap

with MHCs but are not the same. Only individuals who are diverted to MHCs from jail-based

criminal courts (treatment group – 448 MHC participants) and individuals also with mental illness

at risk for jail in criminal courts (control group – 599 criminal court participants) are discussed.

Individuals who were offered MHCs and declined are not in the scope of the data.

Research shows that the percentage of these individuals varies by court, with an average of

93.9% enrolled and 6.1% declined (Steadman, Redlich, Griffin, Petrila, & Monahan, 2005). The

severity of legal charges, variation in knowledge of MHC proceedings and mental illness

characteristics are potential reasons for accepting or declining participation (Canada, Trawver, &

Berrenger, Deciding to Participate in Mental Health Court: Exploring Participant Perspectives,

2020).

Even when efficacy is proven, for continued investments and legislation in programs and

policies, projected economic contributions need to outweigh costs (DeMartini, Mizock, Drob,

Nelson, & Fisher, 2020). A positive economic contribution is helping society, a negative economic

contribution is a cost. This dissertation is an economic analysis of MHCs using data from the

“MacArthur Mental Health Court Study, California, Minnesota, Indiana, 2005-2008” which

includes variables that can be costed in the data (Steadman & Redlich, 2022). The treatment group

of MHC attendees and the control group of criminal court attendees are delineated under the

below data source section. The policy relevance section explores this critical social policy issue

from an economic public and societal perspective that is crucial to policy implementation.

There is a dearth of research on the economic analysis of diverting individuals with a

criminal charge who have mental illness to MHCs (Cowell, Hinde, Broner, & Aldridge, 2013).

The approximate annual incidence of those currently receiving treatment through MHCs is a

small percentage of individuals with mental illness in criminal courts (Treatment Advocacy

Center, 2016) (Substance Abuse and Mental Health Services Administration, 2020). There were

approximately 477 MHCs in 2020 in the United States (National Alliance on Mental Illness,

2024), and the number is growing (Burns, 2013).

1.2 The social policy problem

Incarceration of individuals with SMI in jails results in poor outcomes in health. 4.2% of

the general population has SMI, whereas 11% to 25% of individuals incarcerated in jails have

SMI (Winters, Greene-Colozzi, & Jeglic, 2017). In 2014, there were approximately 744,600

individuals in jail in the US (Treatment Advocacy Center, 2016). The ‘War on Drugs’

significantly increased the number of individuals in jails with SMI due to more stringent

punishment for petty, nonviolent crime (Pope, 2013).

The consequences are serious. Mental health services for incarcerated individuals are

uneven and insufficient (Pope, 2013). Individuals with SMI face unique challenges during

incarceration and upon community reentry. One study shows that jails may not properly screen

inmates for SMI, and that jail staff require additional training in screening for and managing SMI

including communication with outside providers (Scheyett, Vaughn, & Taylor, 2009).

Furthermore, access to necessary medication is oftentimes disrupted during incarceration

(Scheyett, Vaughn, & Taylor, 2009).

Jail sentences for comparable crimes tend to be longer for individuals with SMI and such

individuals face an increased risk of recidivism. Time served is harder, there are fewer

opportunities for constructive programming and there is more punishment. There is also a much

higher rate of suicide amongst inmates in jails with SMI (World Health Organization, 2022)

(Treatment Advocacy Center, 2016) (Kubiak, Comartin, Hanna, & Swanson, 2020). Challenges

confronting incarcerated individuals with SMI reentering the community from jails include:

finding employment, access to government benefits, housing and access to efficacy-based mental

health treatment (Baillargeon, Hoge, & Penn, 2010).

A lack of research and funding, together with historical increases in mandatory

sentencing, may be reasons why more eligible individuals with SMI are not participants in

MHCs (Fisher, Grudzinskas, Roy-Bujnowski, & Wolff, 2011). Furthermore, MHCs are not

currently available post-trial and post-conviction for the individuals with SMI currently

incarcerated in jails (Lamb & Weinberger, 2014). A study shows that MHCs (as compared with

criminal courts) decrease inpatient stays, outpatient days, emergency room visits, and substance

abuse, and increase group therapy hours (VanGeem, 2015).

1.3 Goals and accomplishments

This dissertation presents two separate studies. The first is an outcome regression

presented in Chapter 4. The second is a cost-benefit analysis presented in Chapter 5. The policy

problem that this dissertation addresses is the uncertainty about the economic impact on society

and the taxpayer of MHCs. In this research, statistically significant results are found that

demonstrate that MHCs have more benefits than costs when compared with treatment as usual in

criminal courts (TAU). Statistically significant results are also found by subgroups. Participants

with a medium crime severity at baseline and participants with a schizophrenic diagnosis have

the best net benefit-cost ratios out of all stratified groups.

This dissertation applies the cost-benefit analysis of several outcomes (all time periods

for these eight outcomes begin at baseline diversion from court): (1) Whether worked in the last

six months, (2) Days in jail in the last six months, (3) Number of serious crimes in the last 18

months, (4) Number of not serious crimes in the last 18 months, (5) Days in prison in the last 18

months, (6) Days homeless in the last six months, (7) Received disability benefits in the last six

months, and (8) Life satisfaction in the last six months. These outcomes are observed in clients

of MHCs and jail-based criminal courts, and analytical models are used to estimate the relative

costs and benefits of MHCs (Lee & Aos, 2011).

The quantitative data has been obtained and provides cost and survey data from the

MacArthur Mental Health Court Study (Policy Research Associates, Steadman, H., and Redlich,

A., 2022) (ICPSR study #38275) This dissertation is the first research study to compute the

costbenefit analysis of MHCs in the areas of geographic location, severity of target arrest, prior

criminal history, baseline violence, severity of mental illness, and co-occurring substance use

disorder. Specifically, there are MHCs (treatment group) and criminal courts (control group) in

four locations. While this dissertation provides findings from data that are dated from 2005-2008

and from only four locations, and hence may not be replicable current MHCs or other locations,

the dissertation discusses policy relevance for current states with or considering using MHCs as

a policy option.

Future work focusing on the cost-benefit analysis of individuals with mental illness and

MHCs, and other related groups, may be justified based on the findings of this research. This

future work may include subpopulations of individuals with mental illness that are not discussed

in this paper, such as individuals in other geographic regions and individuals stratified by the

type of charges accepted in different MHCs.

Chapter 2. Background - literature review supporting this study

2.1 Recent history of MHCs

The deinstitutionalization of state hospitals beginning in the 1950s into the 1970s

occurred without proper community supports and led to “subsequent criminalization” of

individuals with SMI (Lamb & Weinberger, Understanding and Treating Offenders with serious

mental illness in Public Sector Mental Health, 2017). Studies also cite a rapid growth in the

number of inmates with SMI since the 1980s as overcrowding in acute mental hospitals left

many individuals with partial or little treatment, and active untreated illness fueled mostly petty

nonviolent crimes (Lamb & Weinberger, 2017; Way, Sawyer, Lilly, Moffitt, & Stapholz, 2008).

The first MHC began in 1980 in Marian County, Indiana. While each varies in the

number of individuals with mental illness served, with the number of participants varying

between 10 and 200-300 men and women per year, they operate under similar procedures and

goals. On average, MHCs serve 75 individuals per year of which half are men. Differences by

MHC exist in ethnicity of attendees. Many of these differences are due to geographic and

sociodemographic variations among MHC locations (Substance Abuse and Mental Health

Services Administration, 2020). Jail sentencing is used less often in MHCs than in criminal and

Drug Courts; often due to the recognition that incarceration disrupts mental health treatment

(Desmond & Lenz, 2010).

According to VanGeem, individuals attending MHCs are primarily high school graduates

(80.3%), about half are not looking for employment or are disabled (50.6%), the majority are

living independently without support (81.6%), and many have a prior incarceration (65.5%)

(Canada, Serious Mental Illnesses and Treatment: Perspectives from Mental Health Court

Participants, 2012; VanGeem, 2015). Over half receive Supplemental Security Income or Social

Security Disability Income. The average yearly income is $5,369 with 88.6% living below the

federal poverty line (VanGeem, 2015). Across age groups, the offense category of individuals

with mental illness attending MHCs is about 70% for nonviolent offenses and 30% for violent

offenses (Fisher, et al., 2010).

MHCs differ in the defendants eligible for participation and other factors. MHCs vary by

severity of offense that is accepted (e.g., only misdemeanors, nonviolent felonies, other selected

felonies). Other variations include the population served, treatment modalities, incentives used

for participants, supervision, the process of referral, and procedures involving pleas (Canada,

Serious Mental Illnesses and Treatment: Perspectives from Mental Health Court Participants,

2012; VanGeem, 2015).

2.2 Prior findings on cost-benefit analyses related to MHCs

Of the 25 articles reviewed on economic cost-benefit analyses in the areas of criminal

justice and mental health, only two are a cost-benefit analysis of a MHC (see appendix A-1 for

the literature review search details). The first of these studies is the Steadman study, discussed in

detail below. The other study is of a MHC in a midwestern city that examines 12-month costs

and benefits with participants with SMI and non-violent felony convictions. The study compared

105 MHC enrollees to 45 participants eligible for MHC but not enrolled (control group). Total

savings for the one-year period of the study were $22,906 per successful MHC participant and

$7,612 per unsuccessful MHC participant as compared to control group participants (Kubiak,

Roddy, Comartin, & Tillander, 2015).

Only one other study similar to a cost-benefit analysis of MHCs is mentioned in the

costbenefit literature. It is a 2007 study conducted by RAND Corporation of three MHCs in

Pennsylvania. The study uses data from 365 MHC participants and creates a hypothetical

comparison group based on MHC and criminal court data in Allegheny County. It looks

specifically at government costs and savings, and finds that after one year MHC participants

versus the hypothetical comparison group save the government $1,804 per participant and after

two years save the government $9,584 per participant. Extensive independent variables were

included in this study: demographics, MHC records, claims data, justice center records,

employee salaries and other MHC operating costs (Ridgely, et al., 2007).

2.3 Prior findings on effectiveness of MHCs

The VanGeem study (VanGeem, 2015) literature review focuses on MHC completion and

the reduction of recidivism, and concurrent increases in mental health outcomes including

treatment in the community and mental health functioning. Research on MHCs shows that, in

comparison to both standard criminal court procedures and Drug Courts, MHCs are a promising

yet understudied modality for criminal justice involvement and outcomes. The data from the

literature review of the VanGeem study shows that MHCs perform better than criminal courts in

incarceration rates (52% decrease in average arrests based on the difference between 12 month

pre-arrest and 12 month post-arrest), reductions in jail days after discharge from MHC (33%),

and changes in access and usage of mental health services (22% more in treatment six-month

post-discharge from MHC and 73% decrease in average inpatient stays) (VanGeem, 2015).

MHCs also do better at crisis episode reduction (15% fewer for MHC participants), alcohol and

drug abstinence for participants (200% increase in abstinence), and decreasing participant visits

to the emergency room (43% decrease) (VanGeem, 2015). The type of studies that have

generated these outcomes are quasi-experimental pre-test/post-test research studies with a

treatment group and a control group.

2.4 The Steadman study

Cost-savings research on MHCs focuses on comparative costs and net taxpayer savings,

defined below within this section. The most relevant cost-savings study is the ‘Criminal Justice

and Behavioral Health Care Costs of Mental Health Court Participants: A Six-Year Study’ which

matched 296 MHC participants with 386 criminal court participants, all of whom had mental

illness at the time. The study compared data from baseline diversion to MHC (treatment group)

or arrest to criminal court (control group), to three years after MHC enrollment or arrest to

criminal court (Page 1100) (Steadman, et al., 2014). This dissertation reanalyzes the data used in

the Steadman et al. study.

The Steadman, et al. study (2014) utilizes the same baseline, six-month and

administrative data as is used in this dissertation, as well as data collected by the authors outside

of the timeframe of the MacArthur Mental Health Court Study. The outside data was collected on

study participants before and after the MacArthur Study period on criminal activity,

incarcerations, and mental health and substance abuse treatment. This outside data included a

timeframe of three years pre- and post-arrest, employing matching of the treatment and control

groups. The components of the MHCs were the same described in this proposal. Definition of

costs for the purpose of the Steadman paper was overall comprehensive costs to taxpayers for the

treatment group (MHC participants) and the control group (criminal court participants). It may

be inferred that outcomes for this Steadman, et al. study are the difference in the cost savings or

expenditures of the treatment and control group in taxpayer dollars.

Participants were compared on diagnosis, charges, gender, cost, alcohol use, drug use,

and prior psychiatric hospitalizations. Interviews were utilized in addition to secondary data

analysis. Covariates were age, gender, race-ethnicity, marital status, diagnosis, prior

hospitalizations, alcohol use, and drug use. (Redlich, Siyu, Steadman, Callahan, & Robbins,

2012).

The study was focused on comparative costs and net taxpayer savings. Comparative costs

are defined as whether variations in costs between the treatment group and control group exist

three years after participants target arrest. Net taxpayer savings refers to the outcome of MHCs

versus criminal courts per average (mean) participant on direct cost savings to taxpayers on a

yearly basis.

The added MHC treatment costs were found to exceed the estimated savings from

diversion to MHCs. The total annual costs for MHC participants averaged an additional $4,000

(valued in 2014 USD) during each follow-up year (Steadman, et al., 2014). In the Steadman

study, costs were approximately the same in each of the three years, and based on higher costs of

mental health treatment among the treatment group. The study reported that added costs were

due to added services such as a comprehensive and practical treatment plan and a need for

housing prior to MHC completion. The study concluded that for MHCs to contain costs, they

should not exclude populations who are high-need such as those with numerous prior

incarcerations. The study discusses all MHCs in the United States in its findings. However,

overall results were reported as inconclusive. Further research is needed on the value of MHCs.

2.5 Literature summary

In general, there are several factors when considering cost-benefit data on MHCs.

Standard jail costs may be underestimated for the subpopulation of individuals with mental

illness, particularly SMI, who are incarcerated. One study noted that criminal justice cost

reductions due to MHCs are possible, as well as improved functioning of participants (Steadman,

et al., 2014). Another consideration is whether MHCs reduce costs or shift costs to other sectors

or government levels. Almquist observed that: “Over time, mental health courts have the

potential to save money through reduced recidivism and the associated jail and court costs that

are avoided, and also through decreased use of the most expensive treatment options, such as

inpatient care (Page 6)” (Almquist & Dodd, 2009). Research papers have noted that MHCs are

implemented in part to reduce costs due to less days in jail by defendants with SMI. (Redlich,

Siyu, Steadman, Callahan, & Robbins, 2012).

Chapter 3. Theory

3.1 Overview

This section discusses the theoretical framework for this dissertation, which is costbenefit

analysis from the perspective of welfare economics theory and public finance theory. Cost-

benefit analysis is a method that identifies the costs and benefits of public projects. It was

developed for analysis of public projects. Costs and benefits are identified, then monetized, and

then prioritized based on project priority. “(Cost-benefit analysis) draws on the principles of

welfare economics and public finance, which provide the theoretical foundations for a general

framework within which costs and benefits are identified and assessed from society’s

perspective.” (Nas, 2016) Pp. 3.

3.2 Welfare economics theory

Two theorems comprise welfare economics theory:

(1) Competitive equilibrium is always Pareto-optimal.

Pareto-optimal is defined as: An economic state including private and public

stakeholders that makes no individual person better off without economically hurting

another individual person. The exceptions to the above statement (1) in welfare

economics theory are when private market distributions are considered to be unfair such

as in market failures in societal benefit (e.g. public goods needed), information gaps, and

externalities.

(2) A Pareto-optimal economic state of public and private spending is a perfect equilibrium

for an ideal economy so long as: if needed, based on prior market failures, there is a

redistributive resource allocation (Blaug, 2007).

3.3 Public finance theory

Public finance theory stipulates that the government should intervene by spending on

public goods and improving the capabilities of citizens. According to the theory, this is achieved

in two ways: correcting private market failures and achieving redistribution of goods.

Interventions are called for when market failures result in imperfect and unequal social welfare

outcomes, and when private market distribution is considered to be unfair. Interventions may be

related to market failures in societal benefit, information gaps, and externalities that result in

inequalities (Tandon, Fleisher, Li, & Yap, 2014).

Theoretically, public expenditures are justified to increase as long as the benefits to society

are monetarily greater than the economic expenses of increasing revenues. Spending by public

entities is justified to the point that public spending achieves no additional marginal benefit than

private spending. In regard to this proposal, an additional consideration based on public finance

theory is that spending on MHCs also be competitive in regard to the opportunity cost of other

public programs and agencies (Tandon, Fleisher, Li, & Yap, 2014).

3.4 How the theory informs the analysis

The conceptual theory is behind the cost-benefit analysis, including what costs and

benefits are included in the analysis. All direct costs and benefits must be able to be measured

based on their impact on the welfare of the economy as measured by the net impact on society.

The net impact on society includes the net economic impacts on both the taxpayer and the study

participants The theory tells us that the conceptual framework and model should be driven by

this consideration.

The relationship between the theory and the analysis is central to this dissertation. The

two theories are included because they not only complement each other, they complement all of

this research and inform the policy relevance section. The welfare economics theory works with

the analysis by establishing the economic conditions for current and additional spending on

public goods including MHCs. This spending includes how to correct past unfair market failures

including in societal benefit and externalities. Market failures in societal benefit are important to

this dissertation because individuals as a group who are eligible for MHCs may not be treated

fairly in the economy. Externalities are important because these same individuals may be the

victim of indirect costs of the economy such as difficulty in finding work due to mental illness

diagnosis.

The public finance theory works with the analysis by establishing the conditions for

which government should continue to spend on MHCs. This theory also describes the economic

condition that grounds all of the public spending on MHCs as compared to other government

expenditures. This condition is that spending on MHCs must meet all of the economic conditions

of public spending and, in addition, must also be competitive in regard to the opportunity cost of

other public spending. The threshold for allocating funds to MHCs is high because it must take

into account, at least in part, economic tradeoffs with other government priorities.

These two theories are needed together to best justify spending on MHCs because they

may make a case with this research for ongoing funding of MHCs and a redistribution of funds

to MHCs. They describe the economic conditions necessary for continued government spending

on MHCs. Together with this research, these two theories may also justify a one-time resource

allocation to fund MHCs.

This additional key consideration from the conceptual theory of a one-time resource

allocation to fund MHCs may be justified by current or historical unfair private market failures.

Unfair private market failures may be the result of modern-day discrimination, current medically

harmful treatment in jails and prisons, and the lasting effects of historical institutions that treated

individuals with mental illness in the United States of America. All of these factors may unfairly

negatively influence the economic participation and capabilities of individuals with mental

illness at risk of jail in the United States of America. These factors too add to the economic

analysis, even if their effects are only able to be discussed as indirect costs and benefits.

Chapter 4. Outcomes associated with MHC participation

4.1 Overview and research questions and hypotheses

The conceptual framework and model are discussed in appendix A-5.

a. Research questions and corresponding hypotheses

Research question 4.1: Is there a higher rate of work in the 6 months post diversion to court for

MHC compared to criminal courts participants?

Hypothesis 4.1: There is a higher rate of work for MHC participants.

Research question 4.2: Is there less of an incidence of going to jail in the 6 months post diversion

to court for MHC compared to criminal courts participants?

Hypothesis 4.2: There is less of an incidence of going to jail for MHC participants.

Research question 4.3: Is there less of an incidence of any serious crime in the 18 months post

diversion to court for MHC compared to criminal courts participants?

Hypothesis 4.3: There is less of an incidence of any serious crime for MHC participants.

Research question 4.4: Is there less of an incidence of any not serious crime in the 18 months

post diversion to court for MHC compared to criminal courts participants?

Hypothesis 4.4: There is less of an incidence of any not serious crime for MHC participants.

Research question 4.5: Is there less of an incidence of going to prison in the 18 months post

diversion to court for MHC compared to criminal courts participants?

Hypothesis 4.5: There is less of an incidence of going to prison for MHC participants.

4.2 Methods

4.2.1 Data source

The primary data source for this dissertation is the “MacArthur Mental Health Court

Study, California, Minnesota, Indiana, 2005-2008” obtained from the ICPSR study #38275.

Brandeis University IRB approval was secured prior to data receipt and analysis. The data was

collected through a baseline diversion to court survey, a six month post-diversion to court survey,

and administrative data that spans from 18 months pre-diversion to court 18 months

postdiversion to court. The software that houses the data is SPSS. Interviewers were trained to

administer the survey.

The surveys took place in many different locations, including jails, courts, personal

residences, and the locations of behavioral health services. Both the baseline and the six month

follow-up survey are structured in sections. There are 18 sections in the baseline survey, and 19

sections in the follow-up survey. The sections are similar across both surveys but not the same.

Participants are paid for their time. At the end of the baseline survey, participants are reminded

that they will be contacted for the six month follow-up survey. They are given voluntary choice

when contacted after six months about whether they want to participate in the follow-up survey.

4.2.2 Sample and assignment to group

The data source for all variables is The MacArthur Mental Health Court Study,

California, Minnesota, Indiana, 2005-2008 (Steadman & Redlich, 2022) (ICPSR study #38275).

The initial data consists of 1,047 study participants.

Baseline is defined as time of initial arrest and diversion to MHC (n= 409 individuals) or

criminal court (control group= 599 individuals). The treatment group sample consists of 409 men

and women who were eligible for MHCs in their location and opted for MHC participation

voluntarily at the potential time of arrest who were non-randomly assigned to the treatment

group and 39 individuals that moved from the criminal court to the MHC. The control group

consists of 599 criminal court participants with mental illness in the same geographic locations

and at the same time period as the treatment group participants for whom selection criteria was

the same as the treatment group. The control group participants did not have MHCs as an option

at the time of their arrest. The variables in the six-month survey, some of which comprise part of

the outcome sample, are missing 30.7% of respondents because only 69.3% of baseline

interviewees were reinterviewed. These variables are not imputed, and sample loss results.

The 18 month dependent variable timeframe allows for the treatment to be completed and

also begin to take effect post-treatment for the average (mean) participant in the treatment group.

The use of the administrative data allows for evaluation for one year longer than the short-term

dependent variables (6 month duration). This is good because time is needed for the treatment to

take effect.

While the majority of participants have serious mental illness (SMI), the breakdown is

unbalanced across groups (most severe is schizophrenic spectrum, least severe is depression and

then other):

Table 4.1: Number and percentage of participants by most serious diagnosis

and treatment group, initial sample.

Most serious

diagnosis

TAU

number

TAU

percent

MHC

number

MHC

percent

Total

number

Total

percent

Schizophrenic

90 15.0% 159 35.5% 249 23.8%

Dx Other Axis I 46 7.7% 31 6.9% 77 7.4%

Dx Bi-polar 131 21.9% 106 23.7% 237 22.6%

Dx Depression 262 43.7% 76 17.0% 338 32.3%

DX Other 70 11.7% 76 17.0% 146 13.9%

Total 599 100.0% 448 100.0% 1,047 100.0%

Notes: TAU denotes treatment as usual, MHC denotes mental health court. 39

TAU participants were reassigned to the MHC group. For the purpose of analysis,

these 39 participants are included in the MHC group.

Sixty-six percent of individuals in the initial MHC sample have SMI, and another 34

percent have illness that may or may not be considered SMI. For instance, SAMSHA includes

major depressive disorder as a SMI (Substance Abuse and Mental Health Services

Administration, 2023). Individuals coded as depression may or may not have major depressive

disorder.

The initial sample is from four sites – Indiana, Minnesota, and two from California. The

focus of the study is on analyzing the economic value of MHCs including as it relates to

decreasing recidivism and time in jail. The survey data was collected between 2005 and 2008

(Steadman & Redlich, 2022). The administrative data was collected from criminal justice and

court data for a period of 18 months prior to baseline and 18 months after baseline. See appendix

A-2 for a table: selected domains and variables.

Treatment as usual to Mental Health Court participants

Following Steadman et al. (Steadman, et al., 2014), the individuals (n=39) who began in

the treatment as usual and voluntarily opted into the MHC were assigned to MHC for data

analysis. These 39 individuals represent 8.7% of the MHC sample (N=448). This approach was

used because these individuals were very similar to the MHC group; they met the same

requirements as the MHC group including: voluntary choice to join MHC and voluntary choice

to stay in a MHC. Individuals who were offered MHCs and declined are not in the scope of the

data. According to literature, 93.9% of potential participants enrolled in MHCs, that is, only

6.1% declined (Steadman H. , Redlich, Griffin, Petrila, & Monahan, 2005).

4.2.3 Dependent variables

All of the dependent variables are not normally distributed. As a result, four of the

dependent variables were transformed from continuous variables to dichotomous variables (Days

in jail post six months, number of post18 serious crimes, number of post18 not serious crimes,

and number of days in prison post 18months – where post denotes the post-diversion to court

period). The fifth dependent variable, six-month days worked full-time or part-time, was already

a dichotomous variable. Table 4.2 presents the descriptive statistics and distributions for each

dependent variable before transformation.

Table 4.2: Description of the five preliminary variables that are recoded into my

dependent variables in the first set of regression models (before transformation to

dichotomous variables).

Count

Range Min. Max. Mean

S.D.

Skewness

Whether worked post 6 1 0 1 0.33 0.47 0.486

months

Days in jail post 6 months 726 184 0 184 61.37 63.32 0.854

Number of post 18

month serious crimes

895 5 0 5 0.36 0.72 2.315

Number of post 18

month not serious

crimes (Winsorized)

895 12 0 12 1.49 2.32 2.080

Number of days in

prison post 18 months

1,031 550 0 550 122.74 131.32 1.251

Notes: Baseline for all variables is the time of diversion to mental health court or criminal

court. Winsorized means that extreme outlier values that fall outside of the possible parameters

of the data are replaced with the closest rational value.

Table 4.3: Mean values and t-tests (and for the dichotomous variable a Pearson ChiSquare

test) by the treatment group for the five preliminary variables that are recoded into my

dependent variables in the first set of regression models (before transformation to

dichotomous variables).

Mean control

(TAU)

Mean

treatment

(MHC)

Statistical

significance

Days in jail post 6 months 84.90 37.47 0.001

Number of post 18 serious crimes 0.42 0.34 0.131

Number of post 18 not serious crimes 1.68 1.19 0.001

Number of days in prison post 18 months 126.75 116.25 0.197

Whether worked post 6 months (%) 32.7% 42.7% 0.005

Notes: MHC denotes mental health court; TAU denotes treatment as usual. Baseline for all

variables is the time of diversion to mental health court or criminal court. All statistics in this

table are weighted using inverse probability of propensity score weighting. The tests for

statistical significance reported in this table are a t-test for continuous variables and a Pearson

Chi-square two-sided test for the dichotomous variable.

The five dependent variables for the outcome regression are defined below. The

definitions of the dependent variables are in Table 4.4. See appendix A-6 for additional

regression analysis. See appendix A-7 for outcome regression histograms.

Table 4.4: Description of dependent variables in the outcome regressions.

Label Type of data When measured

Number

imputed

Number

Winsorized

Total

number

analyzed

Whether worked post 6

months, yes or no

Whether in jail post 6

months, yes or no

Whether serious crime

post 18 months, yes or no

Whether not-serious

crime post 18 months,

yes or no

Whether in prison post

18 months, yes or no

Survey data

Administrative

data

Administrative

data

Administrative

data

Post 6 months

Post 18 months

726

895

1,031

4.2.4 Independent variables

See appendix A-2, section four for the definition of the eighteen independent variables to

test the hypotheses. These eighteen independent variables are: geographic site (this variable may

also reflect differences between individual programs; sites are: Santa Clara (reference group),

site San Francisco CA, site Hennepin MN, site Marion IN), most serious target arrest charge

code, number of pre18 serious crimes (where pre denotes the pre-diversion to court period),

number of pre18 not serious crimes, baseline violence, diagnosis (diagnosis schizophrenic,

diagnosis other psychotic diagnosis, diagnosis bi-polar, diagnosis depression, diagnosis anxiety,

diagnosis adjustment, diagnosis personality, diagnosis pending, diagnosis other axis I), and

diagnosis cooccurring substance abuse disorder. All of these independent variables are from

administrative data except for baseline violence and diagnosis cooccurring substance abuse

disorder, which are from survey data. All of these independent variables are measured at baseline

or before baseline to baseline. Whether these independent variables are Winsorized or imputed

may be found later in this ‘Independent variables’ subsection.

Data Cleaning Required

This investigator inspected the raw data and discovered and corrected an error in data

transcribing (Policy Research Associates, Steadman, H., and Redlich, A., 2022) (ICPSR 38275).

Upon inspection, it was clear that the values for 28 of the variables were incorrect for some cases

because data from two rows (149 and 651) were omitted, resulting in a shift of all subsequent

data rows. The majority of these variables are independent variables. The investigator imputed a

zero for rows 149 and 651 for the missing data values so that the values for other participants

were corrected.

Examples of shifted data are below (Table 4.5). The appendix (A-4) shows an excerpt of

the actual data to explain the shift.

Table 4.5: Data correction example: Number of respondents by treatment arm and outcome with uncorrected

and corrected data. 'Mental Health Court outcome: three groups' by 'Study group.'

Treatment

arm

Total

Uncorrected Data

Graduated

Terminated Still in or opted

from MHC MHC

out

Corrected

Total Terminated Still in

from MHC MHC

Graduated

or opted

out

TAU 143 34 36 73 0 0 0 0

MHC 290 77 65 148 433 111 101 221

Total 433 111 101 221 433 111 101 221

Notes: MHC denotes Mental Health Court; TAU denotes treatment as usual.

An example of shifted data from the six-month dependent variable, ‘Whether worked in

last 6 months,’ is in Table # 4.6 below. Other shifted variables that have the same total number of

study participants and the same breakdown in the corrected data of TAU and MHC participants

(374 TAU participants and 352 MHC participants; 6M denotes the post-diversion to court six-

month follow-up survey) are ‘6M: Days in jail,’ ‘6M: Days in community,’ ‘6M: Days in

institution,’ ‘6M: Days homeless,’ and ‘6M: Days in psychiatric hospital.’ The twenty-two study

participants who got misaligned in this variable’s shift are all in the MHC group. Ten of these

study participants worked, and twelve of them did not work.

Table 4.6: Data correction example. Number of respondents by treatment arm

and 'Whether worked in the last six months' with uncorrected and corrected data.

Treatment arm

Uncorrected

Total Worked

Not

wor

ked

Corrected

Total Worked

Not

worked

TAU 396 139 257 374 129 245

MHC 330 102 228 352 112 240

Total 726 241 485 726 241 485

Notes: MHC denotes Mental Health Court; TAU denotes treatment as usual.

Treatment of missing data

The percentage of MHC (n=448) and TAU (n=599) are, respectively, 42.7% MHC and

57.3% TAU with no missing data (see appendix A-3 for Missingness by data type). MHC-only

variables, none of which are dependent variables, only have values for the 42.7% (n=448) of

total study participants (n=1,047) who are MHC participants and not for the 57.3% of the total

participants who are TAU participants. These MHC-only variables do not need to be imputed

Missing values were discovered for key variables with 15.0% or less of the sample missing. For

these variables (Number of pre18 serious crimes, Number of pre18 not serious crimes, Baseline

violence, Years of education), missing data is handled by series mean imputation. The cutoff for

series mean imputation is 15.0%; such that variables with 15.0% missingness or more are

removed. One potential dependent variable is removed due to too much missing data: ‘Days in

Jail Post 18 – Valid – 688 (34.3% missing).’Analysis was not completed using this variable.

Below is the series mean imputation on the four independent variables that are imputed.

Table 4.7: Imputation of variables with missing values using the series mean

imputation technique.

Number of replaced missing

Number of valid

Variable values cases

Number of pre18 serious crimes 152 1,047

Number of pre18 not serious crimes 152 1,047

Baseline violence 2 1,047

Years of education 4 1,047

Winsorizing

Table 4.8: Values of outliers and imputation for variables that required Winsorizing.

Value that

Variable Value of outliers was imputed

Pre-baseline to baseline 6 month nights in

jail

Eighteen extremely high

outliers ranging from 185 to 465

184

Number of post 18 not serious crimes One extremely high outlier

which is 47

During period, approximate nights

homeless

Six extremely high outliers

ranging from 186 to 589

184

Approximate nights in hospital

One extremely high outlier

which is 329

184

Precorrections 2

Twenty-five extremely high

outliers ranging from 551 to 883

548

4.2.5 Outcome regression analysis approach

To conduct the outcome analyses, the investigator developed a propensity score to

balance the samples and conducted logistic regression to test the hypotheses.

4.2.5.1 Propensity score matching

The propensity score method – inverse probability of treatment weighting (IPTW) was

used and participants in both the treatment and control groups who are closest to the mean of

participants are weighted more and those who are farthest away from the mean of participants are

weighted less. This weighting method balances the samples on pre-baseline to baseline and

baseline characteristics. The weight variable is rescaled for all further analysis so that it is a

normalized weight with a mean equal to one across all study participants. The process to

normalize the weight does not change its value, but rather is a linear transformation. IPTW

involves determining a propensity score through logistic regression, and then computing inverse

probability weights on each participant (treatment group: 1/propensity score and control group:

1/(1-propensity score)). The weight balances the data sample for statistical analyses; that is, all

covariates in the weighted sample are statistically similar across the treatment and control

groups.

Covariates used to balance the treatment group and control group samples were measured

at baseline and many reflected prior 6 months to baseline. Of these covariates, mental health

diagnosis, developmental disability diagnosis, and geographic site location are from

administrative data. Of these covariates, the following are from survey data: race and ethnicity,

education, age, sex, cooccurring substance use diagnosis, receipt of disability benefits, receipt of

mental health and substance use disorder treatment, work, nights in jail, alcohol and drug use,

nights homeless, nights in a hospital, anger, impulsiveness, number of arrests, and number of

days in prison. See appendix A-2 for a complete listing of these variables.

Crosstabs for categorical variables and independent samples t-tests for continuous

variables were created for each individual covariate by the treatment and control groups. Pearson

Chi-square two-sided p-value tests are performed on each of the categorical variables – both

before and after propensity score weighting. Four of these categorical variables violated the

assumptions of the Pearson Chi-square tests because of their high percentage of cells (over 20%)

with an expected count less than 5. These four variables are analyzed using Fisher’s Exact Test

(two-sided). Independent samples t-tests for t-test significance of two-sided p-value are

performed on each of the continuous variables – both before and after propensity score

weighting.

The Steadman et al. study matched on diagnosis, charges and gender. Steadman et al. also

matched on four additional variables: “prior cost (years 1–3)…alcohol use in the past 30 days,

drug use in the past 30 days, and prior psychiatric hospitalizations” (Steadman, et al., 2014) (p.

1102) to eliminate bias. This is similar to the variables used in this dissertation to balance the

sample through propensity scoring. However, this dissertation matches on many additional

variables.

4.2.5.2 Outcome regression models

In all five outcome regression models, the assumptions of linearity, independence and

equality of variance are met.

Binary logistic regression models in outcome regressions (five total)

Of the five dependent variables in the outcome regressions, two cover the six months

after the inception of MHC (treatment group) or criminal court (control group) and three cover

the eighteen months after the inception of MHC (treatment group) or criminal court (control

group). All of the data in this regression analysis is propensity score weighted using an inverse

probability of treatment weighting. The reference group for site, or geographic location, is Santa

Clara.

Impact percentage

The regression results discuss the impact percentage. The impact percentage is the odds

ratio minus one. For example, the results later in this dissertation give an odds percentage for

MHCs for worked post 6 months of 1.577. This generates an impact percentage of 58%. That

means that an offender who accessed the MHC was 58% more likely to be working at six months

than a statistically comparable offender without MHC. Exact percentages can be positive or

negative. This analysis focuses on the impact percentages for independent variables that are

statistically significant in the regression models.

4.3 Findings

4.3.1 Propensity score findings

Before inverse probability of treatment weighting (IPTW), 15 of the 38 covariates have

statistically significant different means across the treatment and control groups (See Chart 1

below). After weighting, none of the covariates have statistically significant different means

across the treatment and control groups. These results may be found in table 4.9 and table 4.10

below.

The following graph shows the propensity score distribution among the treatment and

control groups. There is significant overlap in this graph between the treatment and control

groups. The amount of overlap shows that the propensity scores may balance the sample well.

Confirmation that the propensity scores balance the sample well is confirmed by further analysis

detailed in this section.

Table 4.9: Balance of covariates by the treatment group of study participants for categorical variables before and after

applying the propensity score weight.

Inverse probability of propensity score

Covariate

Mean

control

(TAU) n

= 599

Mean

treatment

(MHC)

n=448

Pearson Chi-

Square two-sided

p-valuea

(categorical

variables)

Mean

control

(TAU) n

= 599

Mean

treatment

(MHC)

n=448

Pearson Chi-

Square two-sided

p-value

(categorical

variables)

Dx Schizophrenic

Dx Other Psych Dx

Dx Bi-polar

Dx Depression

Dx Anxiety

Dx Personality

Dx Pending

Dx Other Axis I

Dx Substance Abuse

Hispanic

Black

Asian

American Indian

White

Race: Other

Sex (Female)

15.0%

7.7%

25.2%

48.6%

11.9%

1.3%

6.3%

3.8%

18.7%

17.0%

33.4%

1.7%

2.8%

44.4%

3.7%

36.7%

35.5%

7.4%

25.2%

21.0%

5.4%

6.5%

7.4%

3.8%

32.4%

10.5%

36.6%

2.9%

1.8%

47.5%

2.9%

42.0%

<0.001

0.849

0.996

<0.001

0.515

0.970

<0.001

0.003

0.279

0.178

0.270

0.313

0.492

0.086

24.6%

7.9%

26.4%

36.6%

9.0%

3.8%

6.7%

4.4%

22.8%

16.1%

33.3%

1.7%

2.5%

47.5%

4.2%

39.5%

24.4%

8.4%

28.2%

34.9%

9.9%

3.6%

6.5%

4.4%

24.3%

14.9%

33.1%

1.9%

2.5%

47.8%

5.0%

40.1%

0.944

0.755

0.524

0.342

0.582

0.846

0.881

0.979

0.569

0.581

0.931

0.830

0.988

0.490

0.573

0.830

Were you receiving disability benefits

during the last six months

26.4% 46.2% <0.001 34.3% 34.3% 0.907

Original sample weighted sample

In past 6 mos, received tx for mental

or emotional problems

In past 6 mos, received alcohol,

drug, or detox tx

In last 6 mos, received both MH and

SA tx at the same time

In the past six months, did you work

either full time or part time.

74.1%

43.9%

28.2%

53.3%

82.4%

39.1%

27.9%

37.3%

0.003

0.193

0.548

<0.001

78.5%

42.0%

28.4%

48.1%

79.1%

40.6%

28.3%

47.8%

0.208

0.539

0.279

0.314

Site: San Francisco, CA 24.1% 24.1% 0.921 20.7% 19.6% 0.666

Site: Hennepin, MN 23.9% 23.4% 0.870 23.9% 26.5% 0.346

Site: Site Marion, IN 18.7% 22.1% 0.175 20.7% 19.0% 0.496

Covariate

Dx Adjustment

Dx DDMR

Hawaiian Native

Alaskan Native

Mean

control

(TAU) n

= 599

0.7%

0.0%

0.5%

0.2%

Mean

treatment

(MHC)

n=448

1.3% 0.9%

0.4%

0.0%

Fisher's Exact

Test (two-sided)

(because Pearson

Chi-Square

assumptions are

violated)

0.341

0.033

Mean

control

(TAU) n

= 599

0.8%

0.0%

0.6%

0.2%

Mean

treatment

(MHC)

n=448

0.8%

0.4%

0.6%

0.0%

Fisher's Exact

Test (two-sided)

(because Pearson

Chi-Square

assumptions are

violated)

0.500

0.499

Notes: TAU denotes treatment as usual, MHC denotes mental health court; Dx denotes diagnosis; ID denotes identification; MH

denotes mental health; SA denotes substance abuse; pre18 denotes the 18 months before diversion to court; mos denotes month(s);

DDMR denotes developmental disabilities and mental retardation.

a: The Pearson Chi-Square two-sided p-value is used except when necessary assumptions for this test are violated. In these

instances, Fisher's Exact Test (two-sided) is used. The variables that use Fisher's Exact Test are dx adjustment, dx DDMR, Hawaiian

native, and Alaskan native.

Table 4.10: Balance of covariates by the treatment group of study participants for continuous variables before and

after applying the propensity score weight.

Inverse probability of propensity

Original sample score weighted sample

Covariate

Mean

control

(TAU) n

= 599

Mean

treatment

(MHC)

n=448

t-test

significance

two-sided

p-value

(continuous

variables)

Mean

control

(TAU) n

= 599

Mean

treatment

(MHC)

n=448

t-test

significance

two-sided

p-value

(continuous

variables)

Years of education

Age

During period, approximate nights

in jail

Have you at any time needed

hospitalization (etc.) for mental

problems

Any alcohol past 30 days?

Any illegal drugs past 30 days?

During period, approximate nights

homeless

During period, approximate nights

in hospital

Anger 4 questions (Range 4-12)

Impulsiveness (Range 10-40)

Number of pre18 arrests

11.64

36.588

27.94

1.70

7.9

10.89

23.66

4.56

7.1

24.09

2.26

11.614

37.527

39.53

1.70

5.87

8.83

22.68

9.37

6.79

23.59

2.25

0.847

0.138

0.001

0.954

0.001

0.006

0.763

<0.001

0.038

0.146

0.914

11.62

36.73

31.02

1.71

6.94

9.95

23.74

6.19

6.99

23.89

2.26

11.56

36.67

33.47

1.71

7.01

10.03

22.83

6.41

6.97

24.13

2.26

0.645

0.920

0.480

0.989

0.913

0.919

0.775

0.871

0.895

0.478

0.963

Number of days in prison pre 72.72 77.18 0.486 72.45 72.58 0.983

18months

Notes: TAU denotes treatment as usual, MHC denotes mental health court; Dx denotes diagnosis; ID denotes

identification; MH denotes mental health; SA denotes substance abuse; pre18 denotes the 18 months before diversion to

court; mos denotes month(s); DDMR denotes developmental disabilities and mental retardation.

4.3.2 Correlations and descriptive findings

Correlations between dependent variables and independent variables are completed as a

precursor to regression analysis. They show significant relationships between variables that are

not causal. The regression modeling was not changed based on these results.

The below correlations are between the dependent variables and the key independent

variables for hypothesis testing. They are based on the data that is weighted from the propensity

score analysis. Significance is determined by bivariate correlations in SPSS. The Spearman

coefficient correlation is reported. See the below table for details on these bivariate correlations.

Table 4.11: Correlations between the dependent variables and the independent variables

used in regression models.

Independent Variable

Whether

worked

post 6

months

Whether

days in jail

post 6

months

Whether

post18

serious

crimes

Whether

post18 not

serious

crimes

Whether

days in

prison post

18 months

MHC Group 0.056 -0.441*** -0.043 -0.190*** -0.127***

Site Santa Clara, CAa N/A N/A N/A N/A N/A

Site San Francisco, CA -0.225*** 0.187*** 0.042 -0.124*** 0.153***

Site Hennepin, MN 0.154*** -0.045* -0.034 -0.078* -0.124

Site Marion, IN 0.051 -0.294*** -0.039 -0.059 -0.335***

Most Serious Target

Arrest Charge Code

Scaleb

0.020 0.025 0.013 0.035 -0.039

Number of Pre-18

Serious Crimes

-0.040 0.059 0.287*** -0.172*** -0.054

Number of Pre-18

NotSerious Crimes

-0.046 0.053 -0.111*** 0.225*** .093***

Baseline Violence -0.044 0.002 .078* 0.010 0.068*

Dx Schizophrenic -0.156*** -0.002 0.015 -0.077 -0.04

Dx Other Psych Dx 0.025 0.039 -.068* 0.143*** 0.015

Dx Bi-polar 0.016 -0.085* 0.085** -0.049 -0.004

Dx Depression 0.123*** 0.000 -0.006 0.042 0.055

Dx Anxiety 0.151*** -0.003 -0.059 0.018 -0.04

Dx Adjustment 0.029 0.037 -0.036 -0.025 -0.049

Dx Personality -0.063 0.079* -0.046 -0.019 -.085**

Dx Pending 0.044 -0.138*** -0.043 -0.048 -0.146***

Dx Other Axis I 0.052 -0.06 0.008 0.061* -0.002

Dx Substance Abuse 0.183*** -0.032 0.026 0.054

* P-value <.05

** P-value <.01

*** P-value <.001

Notes: Dx denotes diagnosis; MHC denotes mental health court; TAU denotes treatment and

usual, MH denotes mental health, SA denotes substance abuse. The reference site is Santa Clara,

CA. All data in this table uses the Spearman coefficient correlation (two-tailed) and are

propensity score weighted. a: Santa Clara is the reference site for the other three geographic

sites. b: This scale is from 1 (most serious charge) to 10 (least serious charge). All 1,047 study

participants are included on this scale.

Correlations between dependent variables are performed after the transformation of

dependent variables to dichotomous variables. As shown in the table below, there are a number of

significant but weak correlations between the dependent variables. All correlations are small

except for one moderate correlation between ‘in prison post-18 months’ and ‘serious crime

post18 months’ (r2 = .375). These findings suggest that these five dependent variables largely

represent unique domains of outcomes.

Table 4.12: Correlations between dependent variables.

Whether

Whether Whether not- Whether worked Whether

serious serious in prison post 6 in jail

post crime post crime post post 18

months 6 months 18 months 18 months months

Whether worked post 6 Correlation 1 months Significance NA

726

Whether in jail post 6 months Correlation -.090 * 1

Significance 0.015

726 726

-0.077

0.055

628 895

Whether not -serious crime

post 18 months

Correlation 0.014 .155 -.083 * 1

Significance 0.718 0.000 0.013

628 628 895 895

Whether in prison post 18

months

Correlation -.076 * .220 *** .203*** .375 *** 1

Significance 0.043 0.000 0.000 0.000

713 713 885 885 1,031

* P-value <.05

** P-value <.01

Whether serious crime post 18 Correlation 0.072 1 months Significance 0.070

628

***

*** P-value <.001

Notes: NA denotes not applicable. Significant correlations in blue. Correlations are Pearson Correlations. All significance

tests are two-tailed.

Multicollinearity was addressed for every independent variable in these outcome

regression models. All Pearson Correlations between independent variables are below 0.400.

There are no independent variables that cannot be in the models together due to high correlation.

All of the eighteen independent variables are the same for all five outcome regression models

The average length of time in treatment for the MHC group in this sample is 448 days; with the

shortest length being 27 days and the longest length being 1,161 days. The average time that the

criminal court (control) group is followed is not in the data, but the control group is included in

both of the surveys and in the administrative data spanning three years per participant.

4.3.3 Regression findings

Regression findings - independent discussion of the five regression models

The five outcome regression models are discussed in detail separately, and then as a

group. All of these results are weighted using the IPTW method.

Model # 4. 1.: Whether worked full-time or part-time post 6 months

Of study participants, 38.2% have any days worked full-time or part-time during the 6

months in this timeframe. Table 4.13 shows results that the odds of working in the six months

post-diversion increases by 58% if a participant is in the MHC group. The odds of working

decreases by 77% if participating in a court in San Francisco, CA as opposed to the reference

group (Santa Clara, CA). If a participant had a diagnosis of depression at baseline, the odds

increases by 61%. The Nagelkerke R Squared is 0.157, suggesting that the independent variables

explain 15.7% of the variance in the dependent variable in this model. There is predictive

capacity in the regression model, as the omnibus test is significant at the <.001 level.

Table 4.13: Logistic regression results for model 1: Whether worked post 6 months.

Independent variable

MHC Group (ref: TAU)

Site Santa Clara, CAa

Site San Francisco, CA

Site Hennepin, MN

Site Marion, IN

Most Serious Target Arrest Charge

Code Scaleb

Number of Pre-18 Serious Crimes

Number of Pre-18 Not-Serious

Crimes

Baseline Violence

Dx Schizophrenic

Dx Other Psych Dx

Dx Bi-polar

Dx Depression

Dx Anxiety

Dx Adjustment

Dx Personality

Dx Pending

Dx Other Axis I

Dx Substance Abuse

Constant

Exp(B)

OR Significance

Impact

(%)

95%

(lower

bound)

1.577 0.005 58%

1.144 2.175

1.000 NA 0% NA

0.234 0.000 -77%

0.127 0.431

1.367 0.149 37% 0.894 2.091 1.227

0.378 23% 0.778 1.935 0.936 0.068

-6% 0.873 1.005

1.032 0.785 3% 0.825 1.290

0.964 0.459 -4% 0.875

1.062

0.929 0.708 -7% 0.634 1.363

0.630 0.063 -37% 0.387 1.026

1.512 0.194 51% 0.810 2.821

1.030 0.896 3% 0.660 1.607

1.607 0.037 61% 1.030 2.507

1.517 0.119 52% 0.898 2.564

1.768 0.465 77% 0.384 8.142

0.843 0.727 -16% 0.323 2.200

1.174 0.689 17% 0.536 2.574 1.384 0.398

38% 0.651 2.940

0.816 0.318 -18% 0.548 1.216

0.784 0.569

Notes: Dx denotes diagnosis; OR denotes odds ratio; MHC denotes mental health court; NA

denotes not applicable. Blue cells denote statistically significant value. All statistics in this

table are weighted using inverse probability of propensity score weighting. The model fit

statistic for this model is the Nagelkerke R Square (0.157). The omnibus test is significant at

the <.001 level.

a Santa Clara is the reference site for the other three geographic sites. b

This scale is from 1 (most serious charge) to 10 (least serious charge).

2.: Whether in jail post 6 months

Of study participants, 74.2% have any days in jail during the 6 months in

this timeframe. Table 4.14 shows results that the odds of spending time in jail in the six months

post-diversion decreases by 97% if a participant is in the MHC group. Part of this finding may be

because some participants in the MHC group are choosing MHC participation instead of jail

time, while the

TAU participants do not have this option. However, this finding is still seen as very important.

The odds of spending time in jail decreases by 54% if participating in a court in Hennepin, MN,

and increases by 157% if participating in a court in San Francisco, CA and 89% if participating

in a court in Marion, IN, all as opposed to the geographic reference group (Santa Clara, CA). The

odds of spending time in jail in the six months post-diversion increases by 16% for participants

who have one or more pre-18-month not serious crime. The most important clinical factors

include depression (odds decreases by 57%) and substance abuse at baseline (odds increases by

186%). The Nagelkerke R Squared is 0.507, suggesting that the independent variables explain

50.7% of the variance in the dependent variable in this model. There is predictive capacity in the

regression model, as the omnibus test is significant at the <.001 level.

Table 4.14: Logistic regression results for model 2: Whether in jail post 6 months.

95% 95% CI CI

Independent variable

Exp(B)

OR Significance

Impact

(%)

(lower

bound)

(upper

bound)

MHC Group (ref: TAU) 0.030 0.000 -97% 0.015 0.058

Site Santa Clara, CAa 1.000 NA 0% NA NA

Site San Francisco, CA 2.574 0.027 157% 1.115 5.940

Site Hennepin, MN 0.461 0.010 -54% 0.255 0.832

Site Marion, IN 0.114 0.000 -89% 0.059 0.219

Model # 4.

Most Serious Target Arrest Charge

Code Scaleb

1.013 0.797 1% 0.920 1.115

Number of Pre-18 Serious Crimes

Number of Pre-18 Not-Serious

Crimes

Baseline Violence

Dx Schizophrenic

0.995 0.976 -1% 0.701 1.412

1.159 0.045 16% 1.003 1.340

0.759 0.310 -24%

0.446 1.293

1.077 0.818 0.570 2.036

Dx Other Psych Dx 0.872 0.760 -13% 0.362 2.101

Dx Bi-polar

Dx Depression

Dx Anxiety

Dx Adjustment

0.557 0.051 -44% 0.309 1.002

0.432 0.007 -57% 0.235 0.795

1.527 0.239 53%

300%

0.755 3.089

4.001 0.361 0.204 78.571

Dx Personality 0.974 0.973 -3% 0.204 4.639

Dx Pending 0.738 0.578 -26% 0.253 2.152

Dx Other Axis I

Dx Substance Abuse

Constant

1.390 0.579 39% 0.434 4.450

2.862 0.001 186% 1.532 5.347

71.627 0.000

Notes: Dx denotes diagnosis; OR denotes odds ratio; MHC denotes mental health court; NA

denotes not applicable. Blue cells denote statistically significant value. All statistics in this

table are weighted using inverse probability of propensity score weighting. The model fit

statistic for this model is the Nagelkerke R Square (0.507). The omnibus test is significant

at the <.001 level. a Santa Clara is the reference site for the other three geographic sites. b

This scale is from 1 (most serious charge) to 10 (least serious charge).

3.: Whether serious crime post 18 months.

Of study participants, 26.4% have one or more serious crimes during the 18 months in

this timeframe. Table 4.15 shows results that the odds of serious crime in the first eighteen

months post-diversion to MHC or criminal court decreases by 40% if participating in a court in

Hennepin, MN, and decreases by 55% if participating in a court in Marion, IN, as opposed to the

geographic reference group (Santa Clara, CA). For each pre-18-month serious crime that a

participant had, the odds of post-18 month serious crime increases by 115 %. If an individual

reported violence at baseline, they had 60% greater odds of post-18 month serious crime. Finally,

if a participant had a diagnosis of bipolar disorder at baseline, the odds of post-18

month serious crime increases by 80%. The Nagelkerke R Squared is 0.180, suggesting that the

independent variables explain 18.0% of the variance in the dependent variable in this model.

There is predictive capacity in the regression model, as the omnibus test is significant at the

<.001 level.

Table 4.15: Logistic regression results for model 3: Whether serious crime post 18 months.

Independent variable

Exp(B)

OR Significance

Impact

(%)

95%

(lower

bound)

95%

(upper

bound)

MHC Group (ref: TAU) 0.845 0.307 -15% 0.612 1.167

Site Santa Clara, CAa 1.000 NA 0% NA NA

Site San Francisco, CA 0.788 0.354 -21% 0.476 1.304

Site Hennepin, MN

Site Marion, IN

0.602 0.029 -40% 0.382 0.950

0.453 0.002 -55% 0.274 0.750

Model # 4.

Most Serious Target Arrest Charge

Code Scaleb

1.007 0.863 1% 0.935 1.084

Number of Pre-18 Serious Crimes

Number of Pre-18 Not-Serious

Crimes

Baseline Violence

Dx Schizophrenic

Dx Other Psych Dx 0.460 0.056 -54% 0.207 1.018

Dx Bi-polar

Dx Depression

Notes: Dx denotes diagnosis; OR denotes odds ratio; MHC denotes mental health court; NA

denotes not applicable. Blue cells denote statistically significant value. All statistics in this

table are weighted using inverse probability of propensity score weighting. The model fit

statistic for this model is the Nagelkerke R Square (0.180). The omnibus test is significant

at the <.001 level. a Santa Clara is the reference site for the other three geographic sites. b

This scale is from 1 (most serious charge) to 10 (least serious charge).

4.: Whether not-serious crime post 18 months

Of study participants, 62.7% have one or more not serious crimes during the 18 months

in this timeframe. Table 4.16 shows results that the odds of not serious crime in the first eighteen

months post-diversion to MHC or criminal court decreases by 60% if a participant is in the MHC

group. The odds of not serious crime decreases by 70% if participating in a court in San

Francisco, CA, decreases by 48% if participating in a court in Hennepin, MN, and decreases by

2.149 0.000 115% 1.767 2.614

0.913 0.064 -9% 0.829 1.005

1.600 0.014 60% 1.102 2.325

1.071 0.796 7% 0.636 1.803

1.801 0.016 80% 1.116 2.908

1.253 0.356 25% 0.776 2.023

Dx Anxiety 0.719 0.309 -28% 0.381 1.358

Dx Adjustment 0.235 0.321 -77% 0.013 4.095

Dx Personality 0.394 0.137 -61% 0.115 1.346

Dx Pending 0.842 0.727 -16% 0.319 2.218

Dx Other Axis I 1.290 0.520 29% 0.594 2.801

Dx Substance Abuse 0.749 0.163 -25% 0.499 1.124

Constant 0.315 0.009

50% if participating in a court in Marion, IN, all as opposed to the geographic reference group

(Santa Clara, CA). For each pre-18-month serious crime that a participant had, the

odds of post18 month not serious crime decreases by 22%. For each pre-18-month not serious

crime that a participant had, the odds of post-18 month not serious crime increases by 42%.

Finally, if a participant had a diagnosis of other psychotic diagnosis at baseline, the odds of post-

18 month not serious crime increases by 244%. The Nagelkerke R Squared is 0.220, suggesting

that the independent variables explain 22.0% of the variance in the dependent variable in this

model. There is predictive capacity in the regression model, as the omnibus test is significant at

the <.001 level.

Model # 4.

Table 4.16: Logistic regression results for model 4: Whether not-serious crime post 18

months.

Independent variable

Exp(B)

OR Significance

Impact

(%)

95%

(lower

bound)

95%

(upper

bound)

MHC Group (ref: TAU) 0.399 0.000 -60% 0.295 0.541

Site Santa Clara, CAa 1.000 NA 0% NA NA

Site San Francisco, CA 0.296 0.000 -70% 0.183 0.476

Site Hennepin, MN 0.518 0.002 -48% 0.341 0.786

Site Marion, IN 0.504 0.003 -50% 0.320 0.793

Most Serious Target Arrest Charge

Code Scaleb

Number of Pre-18 Serious Crimes

0.994 0.863 -1% 0.930 1.063

0.784 0.007 -22% 0.657 0.934

Number of Pre-18 Not-Serious

Crimes

Baseline Violence

Dx Schizophrenic

Dx Other Psych Dx

Dx Bi-polar

Dx Depression

1.421 0.000 42% 1.277 1.582

1.079 0.683 8%

11%

0.749 1.554

1.112 0.667 0.685 1.804

3.444 0.002 244% 1.574 7.535

0.870 0.539 -13%

12%

0.557 1.357

1.116 0.633 0.711 1.751

Dx Anxiety 1.248 0.415 25% 0.733 2.125

Dx Adjustment 0.544 0.441 -46% 0.116 2.559

Dx Personality 0.806 0.652 -19% 0.316 2.056

Dx Pending 1.022 0.958 2% 0.460 2.269

Dx Other Axis I 1.750 0.176 75% 0.778 3.933

Dx Substance Abuse 0.928 0.701 -7% 0.634 1.359

Constant 2.726 0.016

Notes: Dx denotes diagnosis; OR denotes odds ratio; MHC denotes mental health court; NA

denotes not applicable. Blue cells denote statistically significant value. All statistics in this

table are weighted using inverse probability of propensity score weighting. The model fit

statistic for this model is the Nagelkerke R Square (0.220). The omnibus test is significant

at the <.001 level. a Santa Clara is the reference site for the other three geographic sites.

b This scale is from 1 (most serious charge) to 10 (least serious charge).

5.: Whether in prison post 18 months

Of study participants, 88.3% have any days in prison during the 18 months in this

timeframe. Table 4.17 shows results that the odds of spending time in prison in the first eighteen

months post-diversion to MHC or criminal court decrease by 64% if a participant is in the MHC

group. The odds of spending time in prison in this timeframe decreases by 67% if participating in

a court in Hennepin, MN, and decreases by 94% if participating in a court in Marion, IN, both as

opposed to the geographic reference group (Santa Clara, CA). For each pre-18-month not serious

crime that a participant had, the odds of post-18 month time in prison increases by 34%. If an

individual reported violence at baseline, they had 60% greater odds of post-18 month time in

prison. In terms of diagnoses, if a participant had a diagnosis of other psychotic disorder, the

odds of post-18 prison time decreases by 64%. Similarly, the odds decrease by 48% for

depression, 91% for adjustment disorder, 85% for personality disorder, and 65% for a pending

disorder. The Nagelkerke R Squared is 0.323, suggesting that the independent variables explain

32.3% of the variance in the dependent variable in this model. There is predictive capacity in the

regression model, as the omnibus test is significant at the <.001 level.

Table 4.17: Logistic regression

results for model 5: Whether in

prison post 18 months.

Independent variable

MHC Group (ref: TAU)

Site Santa Clara, CAa

Exp(B)

OR Significance

Impact

(%)

95%

(lower

bound)

95%

(upper

bound)

0.357 0.000 -64% 0.223 0.571

1.000 NA 0%

92%

NA NA

1.923 0.249 0.632 5.850

0.331 0.002 -67% 0.162 0.674

0.058 0.000 -94% 0.028 0.119

0.962 0.441 -4% 0.873 1.061

0.838 0.202 -16% 0.639 1.099

1.343 0.002 34% 1.113 1.619

1.868 0.034 87% 1.047 3.333

Model # 4.

Site San Francisco, CA

Site Hennepin, MN

Site Marion, IN

Most Serious Target Arrest Charge Code Scaleb

Number of Pre-18 Serious Crimes

Number of Pre-18 Not-Serious

Crimes

Baseline Violence

Dx Schizophrenic 0.549 0.061 -45% 0.293 1.027

Dx Other Psych Dx

Dx Bi-polar

Dx Depression

Dx Anxiety 1.001 0.997 0% 0.492 2.037

Dx Adjustment

Dx Personality

Dx Pending

Dx Other Axis I

Dx Substance Abuse Constant 0.664

93.251

0.200

0.000

-34% 0.355

1.242

Notes: Dx denotes diagnosis; OR denotes odds ratio; MHC denotes mental health court; NA

denotes not applicable. Blue cells denote statistically significant value. All statistics in this

table are weighted using inverse probability of propensity score weighting. The model fit

statistic for this model is the Nagelkerke R Square (0.323). The omnibus test is significant

at the <.001 level. a Santa Clara is the reference site for the other three geographic sites.

b This scale is from 1 (most serious charge) to 10 (least serious charge).

0.359 0.032 -64% 0.141 0.914

0.905 0.751 -9% 0.490 1.674

0.518 0.049 -48% 0.270 0.996

0.092 0.011 -91% 0.015 0.574

0.154 0.000 -85% 0.060 0.394

0.347 0.018 -65% 0.145 0.831

0.347 0.051 -65% 0.120 1.006

4.4 Discussion of outcome regression findings

Across the two six-month dependent variables, there are three overlapping significant

independent variables. The first of the two six-month dependent variables is related to work and

the second is related to recidivism, so shared independent variable significance is not necessarily

expected.

Most notably is MHC group. Being a participant in the MHC group, as opposed to

treatment as usual, increases the odds of working post six months by 58% and decreases the odds

of spending time in jail post six months by 97%. These results suggest that MHCs are decreasing

time in corrections and concurrently increasing important community involvement. There is a

body of literature that shows that mental health outcomes among individuals with serious mental

illness are improved by employment. These outcomes include increases in self-esteem (Luciano,

Bond, & Drake, 2014). The recovery process from serious mental illness includes more

participation in employment (along with education and community involvement) (Drake &

Whitley, 2014).

Being in the geographic region of San Francisco, CA decreases the odds of working

during this timeframe by 77% and increases the odds of spending time in jail during this

timeframe by 157%. Of the four geographic sites, San Francisco has the highest per capita

income. The decreased odds of work may be due to less entry-level employment in San

Francisco than in the other three sites. The increases in jail time in San Franisco may be due to

less opportunities for community recovery (e.g. decreased odds of working).

Having a diagnosis of depression increases the odds of working over this time frame by

61% and decreases the odds of spending time in jail over this time frame by 57%. The diagnostic

category of depression is one of the less involved diagnoses (as compared to the diagnostic

categories of schizophrenia and bipolar disorder). For this reason, study participants with the

diagnosis of depression may have better outcomes than other diagnoses.

Across the three 18-month dependent variables, there are seven independent variables

that have at least two shared significant values. These independent variables are MHC group, two

of the three geographic sites, number of pre18 serious crimes, number of pre18 not serious

crimes, baseline violence, and the diagnosis of other psychotic diagnosis.

Being in the MHC group significantly decreases the odds of not serious crime and time in

prison by almost the same amount (60% for whether not serious crime and 64% for whether time

in prison). This suggests that there may be a relationship between these two dependent variables,

where less not serious crime may result in less time in prison. It is also possible that less time in

prison leads to less not serious crime.

The analysis shows a consistent impact of geography on crime and whether participants

have days in prison. The geographic sites of Hennepin, Minnesota and Marion, Indiana result in

less crimes and prison times. This suggests that jurisdictions with less per capita income, higher

rates of persons living in poverty, and higher percentages of persons with a disability under 65

years of age result in less crimes and prison times.

Similarly, prior crime seems to be predictive of future crime and incarceration. There may

be a link between prior crime and prison time. A participant may be serving time due to a prior

crime, or there may be less forgiveness of current crimes because of a history of prior crimes.

Self-reported baseline violence is predictive of increased serious crime and prison time.

Participants reporting violence may have a higher propensity to already be involved in the

criminal justice system. These participants may also be more likely to have committed past

serious crimes which is predictive of future serious crimes.

Finally, the mental health diagnosis of other psychotic diagnosis (in addition to

schizophrenia and some bipolar disorder diagnoses) is predictive of much more (244%) not

serious crime and much less (-64%) time in prison. Even with much more not serious crime, the

decrease in time in prison may be due to this crime being treated effectively in MHC and a weak

negative correlation between not serious crime and serious crime.

The cross-model story that these regression results tell cumulatively is compelling.

Summarized below are results related to being in the MHC group as opposed to the treatment as

usual criminal court group from these outcome regressions. The following describes four key

cross-model observations.

The odds of working in the six months post-diversion increases by 58% if a participant is

in the MHC group as opposed to the treatment as usual group at baseline. These individuals are

less likely to be incarcerated in this timeframe, and they are also less likely to have the criminal

charge resulting in their diversion to court on their record. They are more likely to have received

rehabilitative services post diversion to court including treatment for mental or emotional

problems, treatment for alcohol, drug and detox, MHC case management, and non-criminalized

court appearances.

The odds of spending time in jail in the six months post-diversion decreases by 97% if a

participant is in the MHC group as opposed to the treatment as usual group at baseline. This

statistic is high impact. Particularly because the data used in this binary regression analysis is

weighted to control for pre-diversion biases. A foundational goal of MHCs is to reduce jail

recidivism. This result confirms this goal is successful.

The odds of not serious crime in the first eighteen months post-diversion to MHC or

criminal court decreases by 60% if a participant is in the MHC group as opposed to the treatment

as usual group at baseline. This result is another high impact regression result that speaks to

decreased recidivism. This result may also be related to the decreased jail and prison results

because less not serious crime results in less incarceration. However, it may also be true that less

incarceration results in less not serious crime.

The odds of spending time in prison during the first eighteen months post-diversion to

MHC or criminal court decreases by 64% if a participant is in the MHC group as opposed to the

treatment as usual group at baseline. This result may show that MHCs reduce sentencing for

crimes because prisons are often post-conviction and for longer periods of time than jails.

The hypotheses for the first (4.1), second (4.2), fourth (4.4) and fifth (4.5) research questions in

the outcomes associated with MHC participation analysis are supported by this CBA. The

findings related to the hypothesis for the third research question (4.3) are not significant. For

MHC participants, MHCs significantly increase work, decrease jail, decrease not serious crime,

and decrease prison. There is no significant impact related to serious crime.

4.4.1 Limitations of the outcome chapter relevant to both analyses chapters

MHCs may be better off than what the dissertation found.

The first limitation is that there is no basis by which to generalize the findings to all

MHCs currently operating. The data used in this dissertation is only from four jurisdictions in

three states and was from 2005-2008 when MHCs may have operated differently. Further,

treatment of SMI in jails may also be different over this time period and different jurisdictions.

Hence, this research should be replicated with other sets of MHCs. Generalizability is a

limitation because these four jurisdictions are not necessarily representative of other

jurisdictions. Furthermore, the data analyzed in this study does not include information on how

the MHCs operated, such as specific mental health treatments and sanctioning guidelines. Not all

jurisdictions offer MHCs. Different populations might be arraigned in different jurisdictions.

Only individuals with mental illness in jurisdictions where a MHC accepts their criminal charges

are eligible for MHC. Criminal charge eligibility varies by MHC. Results presented in this

dissertation could not be generalizable to other jurisdictions.

See below for a table that looks at the similarity of these four MHC jurisdictions to the

United States. This below table shows that there is variability in many baseline demographics

between the four geographic sites. San Francisco and Santa Clara, California have per capita

incomes that are higher than the national average. These sites also have less poverty and less

persons with a disability under 65 as a percent of their total population than the national average.

Marion City, Indiana, however, has a per capita income that is far less than the national average,

along with comparatively much higher percentages of the population living in poverty and

persons under the age of 65 with a disability.

Table 4.18: Comparison of selected demographics in 2018 to 2022 from the United

States of America and the geographic regions of the four MHC jurisdictions

researched in this dissertation.

United States of America

Female

persons

(percent)

50.4%

Income (per

capita income

computed to

2023 USD$

from

2022USD$)

$44,562

Persons in

poverty

percent

Persons

with a

disability

under 65

percent

11.5% 8.9%

San Francisco city,

California

48.5% $93,081 10.5% 6.1%

Santa Clara (reference site) 47.3% $84,331 8.0% 4.6%

Hennepin County,

Minnesota

50.2% $59,615 10.8% 7.1%

Marion city, Indiana 54.5% $24,524 28.1% 18.6%

Average (pooled study site

geographic region data)

49.7% $68,547 13.3% 8.4%

Notes: MHC denotes mental health court. NA denotes not applicable.

The below table shows that all of the above selected demographics from the United States

of America fall within the confidence interval bounds of the same demographics from the pooled

geographic study site data. This result suggests that the pooled geographic study site data from

this study is at least partially representative of the entire United States of America. However, the

geography variable used in this dissertation may be indicative of MHC programmatic differences

in addition to geographic differences. Its reliability in this geographic comparison may be

limited.

Table 4.19: Confidence interval for comparison of selected demographics of the pooled

study sites compared with the United States of America.

United

States of Average (pooled geographic

America study sites data)

Confidence

interval

lower bound

Confidence

interval

upper bound

Female persons (percent) 50.4% 45.1% 55.1%

Income (per capita income computed to $44,562 $23,110 $114,251

2023 USD$ from 2022USD$

Persons in poverty percent 11.5% -0.4% 29.1%

Persons with a disability under 65 percent 8.9% -1.1% 19.3%

The longitudinal data spans at most from 18-month pre-baseline to 18-month

postbaseline (with the exception of a few variables, including one that denotes the participant’s

length of stay in MHC). The second limitation is that the time horizon is narrow, and policy

makers may wish to know two, five and ten year post-baseline projections, which are not

possible from these data.

The third limitation is that the assumptions used in extrapolating six month variables to

eighteen months for analysis may be wrong. The fourth limitation is that there are differences

between individual MHCs. Not all differences may be controlled for, including in the propensity

score analysis. Differences that may not be able to be controlled for include the makeup of

participants with different mental health diagnoses, severity of criminal charges, different uses of

rewards and sanctions, plea requirements of defendants, and utilization of community-based

resources. The fifth limitation is that the statistical method inverse probability of treatment

weighting only balances on the covariates used in the propensity score analysis. There may be

selection bias not measured in this analysis.

MHCs may be worse off than what the dissertation found.

The sixth limitation is that participation in MHCs is voluntary. Participation in TAU

criminal courts is mandatory. Only those who would have qualified for a MHC based on

diagnosis and criminal charges and were not eligible for a MHC are members of the TAU

criminal court participant sample. Propensity score analysis is used to mitigate selection bias,

however it may be present due to this limitation.

Can not tell whether MHCs may be better off or worse off than what the dissertation found.

The seventh limitation is that this study mainly utilizes survey data (selection and survey

biases). Most of the data is self-report interview data. There may have been a bias on behalf of

the interview participant to report information that suggests that the participant is doing well in

treatment and in court. There may also be a bias because some of the requested self-report

interview data may violate the terms of the MHCs and criminal courts (e.g. ‘Any illegal drugs

past 30 days?’) Underreporting in these areas could result. There may also be an underreporting

bias in certain variables based on the social desirability hypothesis.

The eighth limitation is that the data used in this dissertation is from 2005-2008. MHCs

have evolved over the last sixteen years. In discussing current policies and practices based on

this analysis, the age of the data needs to be considered. While the tenets of MHCs have stayed

the same and the same types of individuals are diverted, MHCs have grown in number and

evolved as individual courts. While this is a limitation, it is not considered to be a major

limitation. The data is still relevant today. The MHC delivery system has not changed much over

the years. Also, the most recent published study of this type, by Steadman et al., utilized this

same data. This study builds off of the Steadman et al. study.

The ninth limitation is that the six-month survey is missing 30.7% (321 participants) of

respondents because only 69.3% (726 participants) of baseline interviewees were reinterviewed

in the follow up sample. Sample loss is a resulting limitation from this reinterview rate because

the reinterview sample size is smaller than the entire study population (1,047 participants). The

tenth limitation is that there are 39 individuals who began in the treatment as usual group and

voluntarily opted into the MHC group. These individuals are included in the MHC group for

analysis (selection bias). The eleventh limitation is that the valid number of participants changes

in the weighted data when using SPSS. The twelfth limitation is that the potential dependent

variable: ‘Days in Jail Post 18 – Valid – 688 (34.3% missing),’ is removed due to too much

missing data. No dependent variables were imputed. Analysis was not completed using this

variable.

Chapter 5. Cost benefit analysis

5.1 Overview of approach and research questions and hypotheses

The cost-benefit analysis examines the direct costs and benefits of participation in MHCs

as opposed to criminal courts. Outcome benefits and program costs are compared across the

treatment and control groups by two separate analyses. The first is an economic analysis of eight

outcome variables that comprise the new variable, ‘Net economic contribution to society per

participant.’ See appendix A-2, section one entitled ‘Outcome variables (long-term and shortterm

dependent variables)’ for the definitions of these eight variables. See table 5.3 below for the

costing of these eight variables. This analysis calculates the average net benefit-cost ratio per

participant for MHCs as opposed to criminal courts. This analysis is also stratified to look at the

benefits and costs by mental health diagnosis at baseline and severity of crime at baseline.

The second analysis is a cost regression that uses these same eight outcome variables to

create a new composite dependent variable which is also entitled the ‘Net economic contribution

to society per participant.’ This cost regression uses the same independent variables as the

outcome regression discussed in chapter 4 of this dissertation.

Research questions and hypotheses

Research question 5.1: Do the total benefits outweigh the total costs of MHCs compared to

criminal courts for society?

Hypothesis 5.1.a.: Benefits outweigh costs for MHCs compared to criminal courts for society.

Hypothesis 5.1.b.: MHC participants will utilize more mental health treatment than criminal

court participants.

Research question 5.2: Do the total benefits outweigh the total costs of MHCs compared to

criminal courts based on the severity of mental illness of the population as determined by mental

health diagnosis at baseline?

Hypothesis 5.2.a.: Targeting individuals with high severity mental illness as determined by

mental health diagnosis at baseline are more likely to have benefits that outweigh costs in MHCs

as compared to criminal courts.

Hypothesis 5.2.b.: Overall benefits outweigh the total costs for MHCs compared to criminal

courts for people who have high severity mental illness based on mental health diagnosis at

baseline.

Research question 5.3: Do the total benefits outweigh the total costs of MHCs compared to

criminal courts based on the severity of crime of the population as determined by severity of

crime at baseline?

Hypothesis 5.3.a.: Targeting individuals with high severity of crime as determined by severity of

crime at baseline are more likely to have benefits that outweigh costs in MHCs as compared to

criminal courts.

Hypothesis 5.3.b.: Overall benefits outweigh the total costs for MHCs compared to criminal

courts for people who have high severity of crime based on mental health diagnosis at baseline.

5.2 Methods

5.2.1 Data source

As with the outcome regression, the primary data source for this dissertation is the

“MacArthur Mental Health Court Study, California, Minnesota, Indiana, 2005-2008” obtained

from the ICPSR study #38275. See Chapter 4, section 2.1 for the description of the data source.

The software used for analysis was SPSS.

5.2.2 Sample

The sample of 619 valid participants (319 TAU and 300 MHC) used in the cost-benefit

analyses (CBA) is derived from the 1,047 study participants who have no missing values across

the eight outcome variables that comprise the ‘Net economic contribution to society per study

participant.’ Inverse probability of treatment weighting (IPTW) is used for all regression analyses

in this CBA as was done for outcome analyses.

The below table shows that of the 619 valid participants, there are more males as a

percentage in both the MHC and the TAU group. By percentage, the MHC group has more

females than the TAU group. There are variations by race or ethnicity. The MHC group has a

higher percentage of participants who identify as white or black. The TAU group has a higher

percentage of individual who identify as Hispanic. Each study participant has only their most

serious diagnosis at baseline in this table. The percentage of a diagnosis of a schizophrenic

disorder as reported in this table for the MHC group is almost three times the TAU group

Table 5.1: Number and percentage of the 619 valid participants in the CBA by gender and

treatment group, race or ethnicity and treatment group, and mental health diagnosis and treatment

group.

TAU TAU MHC MHC Total Total

number percent number percent number percent

Breakdown by gender (males and females)

Male 197 61.8% 169 56.3% 366 59.1%

Female 122 38.2% 131 43.7% 253 40.9%

Total 319 100.0% 300 100.0% 619 100.0%

Breakdown by race or ethnicity

Hispanic 61 19.1% 32 10.7% NA NA

Black 99 31.0% 105 35.0% NA NA

White 146 45.8% 154 51.3% NA NA

Other 26 7.8% 22 7.0% NA NA

Total NA NA NA NA NA NA

Breakdown by most important mental health diagnosis

Dx schizophrenic 40 12.5% 109 36.3% 149 24.1%

Dx bipolar and other psychotic

diagnoses

114 35.7% 99 33.0% 213 34.4%

Dx anxiety, adjustment, personality,

depression, pending, or other axis one

165 51.7% 92 30.7% 257 41.5%

Total 319 100.0% 300 100.0% 619 100.0%

Notes: TAU denotes treatment as usual, MHC denotes mental health court. The previously mentioned

participants who transitioned from TAU to MHC are included in the MHC group. In the breakdown by

race or ethnicity, about 4% of respondents list more than one race or ethnicity. For this reason, these

descriptive statistics sum to above 100% and the corresponding total numbers and total percentages

are not reportable.

The below table shows the 619 valid participants by geography and treatment group. San

Francisco and Santa Clara vary the most in regard to TAU and MHC participant percentages.

Santa Clara has a higher percentage of TAU participants as compared with MHC participants.

San Francisco has a higher percentage of MHC participants as compared with TAU participants.

Table 5.2: Number and percentage of the 619 valid participants in the CBA by geography

and treatment group.

Geography TAU

number

TAU

percent

MHC

number

MHC

percent

Total

number

Total

percent

Notes: TAU denotes treatment as usual, MHC denotes mental health court. The previously

mentioned participants who transitioned from TAU to MHC are included in the MHC

group.

a Santa Clara is the reference site for the other three geographic

sites.

5.2.3 Dependent variables

Net economic contribution to society per study participant is the only dependent variable

in this CBA section. The net economic contribution to society of MHCs is defined as the

difference between costs and benefits for the MHC group and criminal court group across eight

outcome variables. This composite variable measures the cost of each study participant across

each of these eight outcome outcomes. These eight outcomes are all delineated below in Table

Site Santa Clara, CAa 120 37.6% 95 31.7% 215 34.7%

Site San Francisco, CA 37 11.6% 56 18.7% 93 15.0%

Site Hennepin, MN 92 28.8% 76 25.3% 168 27.1%

Site Marion, IN 70 21.9% 73 24.3% 143 23.1%

Total 319 100.0% 300 100.0% 619 100.0%

5.3. As shared above, these eight outcome variables are defined in appendix A-2, section 1.

entitled ‘outcome variables (long-term and short-term dependent variables)’. The 619 valid study

participants each have no missing values for all of these eight outcomes. All of these costs are

summed across the eight outcomes to comprise the ‘Net economic contribution to society per

study participant.’ The net economic contribution per MHC participant is the difference in the

costs of these eight outcomes by treatment and control group. If the difference is positive, it is a

contribution. If the difference is negative, it is a cost.

The dissertation analyzes the direct costs and benefits to society of MHCs as opposed to

treatment as usual in criminal courts. Costs and benefits are computed. Program costs and

outcome benefits are at the societal level. They include costs and benefits to the taxpayer, costs to

victims, and costs to study participants. If costs for variables in this study are only available for

past years or are most accurate for past years, the inflation rate of the United States is used to

compute these variables to 2023 USD. The source of the inflation rate is the consumer price

index of the United States as reported by the World Bank (Macrotrends, 2023). Every year’s

consumer price index is reported individually from 2003 through 2022, and accordingly

compounded to 2023 USD using Microsoft Excel.

Table 5.3 below further defines all of the outcome variables and program costs for this

cost-benefit analysis. There are eight outcome variables. Five of these outcome variables are

short-term (6 months post diversion to court) and three are long-term (18 months post diversion

to court). For both the cost regression and the net economic contribution to society, these five

short-term outcome variables are all extrapolated to 18 months by multiplying them by three.

These outcome variables are: Whether in jail post 6 months, Whether worked post 6 months,

Days homeless post 6 months, Received disability benefits post 6 months, Life satisfaction post 6

months, Whether in prison post 18 months, Whether serious crime post 18 months, and Whether

not-serious crime post 18 months.

Of the eight outcome variables that comprise this dependent variable, only one is

Winsorized, which is: 'Number of post18 not serious crimes (Winsorized).' None of these

variables are imputed. The six month variables are from survey data and the eighteen month

variables are from administrative data. In the instance of ‘Whether serious crime post 18

months,’ costing overlap exists with the jail and prison variables. In this instance, adjudication

and sanctioning costs are removed from the ‘Whether serious crime post 18 months’ variable.

The program cost variables are in the categories of treatment costs in the below table.

Four of these variables measure costs across both MHC and criminal court participation. These

four are: 6M: Community supervision yes no, 6M: Days in psychiatric hospital, 3.3 In past 6

mos, received tx for mental or emotional problems, and 3.4 In past 6 mos, received alcohol, drug,

or detox tx. The other two measure costs that are only associated with MHC participation. These

two are: Length of stay in MHC (days), and Real number of court appearances.

The variable name and the time and duration of measurement are included in the chart.

The specific computed 2023 dollar amount for the costs of these variables and the sources of the

costing for these variables are listed in the below chart. The mean value from the literature is

used in determining cost estimates for these variables. For MHCs, where costs exceed benefits in

these variables in relation to criminal courts, there is a net cost; and where benefits exceed costs

in these variables in relation to criminal courts, there is a net contribution to society.

A dollar value is placed on all of these eight outcome variables and six program cost

variables that can be directly costed based on the literature. These variables that can be directly

costed are computed and discussed independently and in summation. There are a small number

of costs that are not directly costed due to the limitations of the data. These costs are, for all

groups, costs of lost productivity during confinement including lost wages and costs to families

(e.g. cost of losing a caretaker). These costs are, for MHC-only, the costs of MHC-only sanctions

and rewards (Kubiak, Roddy, Comartin, & Tillander, 2015). Important variables that are not

directly costed will be considered in the intangible costs discussed in the limitations and policy

relevance sections.

Table 5.3: Cost-benefit analysis - costing details. The outcome variables, which are the first eight variables listed in this

table, comprise the net economic contribution to society per study participant composite variable. The variables that

comprise the costs of the program are defined after these outcome variables.

Average (mean if

Units - Originally applicable) cost or

Time and Dichotomous reported source benefit per individual

duration of or year USD and Computed to (computed to Costing

variables measure continuous date 2023USD$ 2023USD$)

1. Outcome variables - long-term outcome variables - long-term outcome variables used in net economic contribution to

society per study participant composite variable. Both MHC and criminal court.

Cost of serious crime post 18

months using costs from Miller

and McCollister

Baseline to

18 months

postbaseline

Continuous

variable: #

serious

crimes per 18

months

Low estimate

originally reported

$10,854 in 2017

USD; High

estimate originally

reported $87,909

in 2017 USD

(Subtracted are

adjudication and

sanctioning costs

(less for overlap

with jails and

prisons))

Low estimate:

Cost per

serious crime

s $13,234 in

2023 USD;

High

estimate is

$107,183 in

2023 USD

(Includes

costs related

to quality of

life

estimates).

$31,559 per serious

crime

Baseline to

Cost of not-serious crime post

18 months using costs from

Miller

18 months

postbaseline

Continuous

variable: #

not serious

crimes per 18

months

Originally reported

$1,916 in 2017

USD

(Subtracted are

adjudication and

sanctioning

costs (less for

overlap with

jails and

prisons))

Cost estimate $2,336 per notis

$2,336 in serious crime

2023 USD

Cost of days in prison post 18

months using costs from Carey

& Waller (low estimate)

Cost of days in prison post 18

months using costs from

McCollister (high estimate)

Baseline to

18 months

postbaseline

Baseline to

18 months

postbaseline

Continuous

variable: #

days in prison

per 18

months

Continuous

variable: #

days in prison

per 18

months

Originally reported

$84 in 2010 USD

Originally reported

$95 in 2016 USD

Low estimate: $116 per day in

$115 in 2023 prison USD

High estimate:

$117 in 2023

USD

2. Short-term outcome variables - used in net economic contribution to society per study participant composite

variable.

Both MHC and criminal court.

Cost of whether worked post 6

monthsa using costs from The

United States Department of

Baseline to 6

months

postbaseline

Dichotomous

variable: No

or yes (0 or 1)

Originally reported

$4,095 in 2024

USD

Low estimate: $5,720 per average

$4,095 in worked in six months

2024 USD

Baseline to

Labor (low estimate) any time

worked over

six months

Cost of whether worked

post 6 monthsa using

costs from The United

States Bureau of Labor

Statistics (high estimate)

6 months

postbaseline

Dichotomous

variable: No

or yes (0 or 1)

any time

worked over

six months

Originally

reported

$6,417 in 2020

USD

High

estimate:

$7,346 in

2023 USD

Cost of days in jail post

6 months using costs

from McCollister (low

estimate)

Baseline to

6 months

postbaseline

Continuous

variable: #

days in jail

per 6 months

Low estimate

originally

reported $87 in

2016 USD

Low

estimate:

Cost per day

of jail is

$107 in

2023 USD

$115 day

in jail

Cost of days in jail post

6 months using costs

from Carey & Waller

(high estimate)

Baseline to

6 months

postbaseline

Continuous

variable: #

days in jail

per 6 months

High estimate:

Originally

reported

$90 in 2011

USD

High

estimate:

Cost per

day of jail

is $123 in

2023 USD

Cost of 6M: Days homeless

using costs from McCollister

(low estimate)

Baseline to

6 months

postbaseline

Continuous

variable: #

days

homeless per

6 months

Low estimate

originally reported

$36 in 2016 USD

Low estimate:

Cost per day

homeless is

$44 in 2023

USD

$103 per day

homeless

Baseline to

Cost of 6M: Days

homeless using

costs from Baggatt

Scott (high estimate)

Baseline to

6 months

postbaseline

Continuous

variable: #

days

homeless per

6 months

High estimate

Originally

reported

$141 in 2020

USD

High

estimate: Cost

per day

homeless is

$161 in 2023

USD

Cost of 6M:

Received disability

benefits (Y/N)

using costs from

Council on Aging

(two estimates are

the same)

Cost of 6M:

Received disability

benefits (Y/N) using

costs from Social

Security (two

estimates are the

same)

6 months

postbaseline

Baseline to

6 months

postbaseline

Dichotomous

variable: No

or yes (0 or

Dichotomous

variable: No

or yes (0 or

Originally

reported

$5,484 in

2023 USD

(for six

months)

No imputation $5,484 per six

needed: monthsb

$5,484 in

2023 USD

(for six

months)

Cost of 6M: Life

satisfaction using

costs from The

United States

Department of

Baseline to

6 months

postbaseline

Dichotomous

variable: (0)

Terrible to

mixed; (1)

Satisfied to

Originally

reported $177

in 2024 USD

Low estimate: $248 per average

$177 in 2024 increase in labor

USD (Based productivity due to

on labor high life satisfaction

productivity in six months

Baseline to

Labor (low

estimate). delighted changes)

Cost of 6M: Life

satisfaction using

costs from The

United States

Department of

Labor (high

estimate).

Baseline to

6 months

postbaseline

Dichotomous

variable: (0)

Terrible to

mixed; (1)

Satisfied to

delighted

Originally

reported

$318 in 2024

USD

High

estimate:

$318 in 2023

USD (Based

on labor

productivity

changes)

3. Costs of the program (treatment costs) - Both MHC and Criminal Court.

Cost of 6M: Days in

community supervision yes no

using costs from McCollister

(low estimate)

Baseline to

6 months

postbaseline

Continuous

variable: #

days in

community

supervision

per 6 months

Originally reported

$4 in 2016 USD

Low estimate:

$5 in 2023

USD

$8 per day for

community

supervision

Cost of 6M: Days in

community supervision yes no

using costs from Carey &

Waller (high estimate)

Baseline to

6 months

postbaseline

Continuous

variable: #

days in

community

supervision

per 6 months

Originally reported

$8 in 2010 USD

High

estimate: $11

in 2023 USD

Cost of 6M: Days in psychiatric

hospital using costs from Choi

et al. (low estimate)

Cost of 6M: Days in psychiatric

hospital using costs from

McCollister (high estimate)

Baseline to

6 months

postbaseline

Baseline to

6 months

postbaseline

Continuous

variable: #

days in

psychiatric

hospital per 6

months

Continuous

variable: #

days in

psychiatric

hospital per 6

months

Originally

reported: $767 in

2022 USD

Originally

reported: $999 in

2016 USD

Low estimate:

$828 in 2023

USD

High

estimate:

$1,233 in

2023 USD

$1,031 per day in

psychiatric hospital

Cost of 3.3 In past 6 mos,

received tx for mental or

emotional problems using costs

from Zhu et al. (low estimate)

Baseline to

6 months

postbaseline

Dichotomous

variable: No

or yes (0 or 1)

Originally

reported: $83 in

2022 USD

Low estimate:

$89 in 2023

USD per 45

minute

$100 per 45 minute

individual

psychotherapy

session. The six

individual

psychotherapy

session

month estimate for

one day per week of

this treatment is:

Cost of 3.3 In past 6 mos,

received tx for mental or

emotional problems using costs

from McCollister (high

estimate)

Baseline to

6 months

postbaseline

Dichotomous

variable: No

or yes (0 or 1)

Originally

reported: $90 in

2016 USD High

estimate: $111

in 2023 USD

per 45 minute

individual

psychotherapy

session

$2.607.

Cost of 3.4 In past 6 mos,

received alcohol, drug, or detox

tx using costs from McCollister

(low estimate)

Baseline to

6 months

postbaseline

Dichotomous

variable: No

or yes (0 or 1)

Originally reported

$83 in 2016 USD

Low estimate:

$102 in 2023

USD

$112* per visit or

session. The six

month estimate for

one day per week of

this treatment is:

$2,924

Cost of 3.4 In past 6 mos,

received alcohol, drug, or detox

tx using costs from Carey &

Waller (high estimate)

Baseline to

6 months

postbaseline

Dichotomous

variable: No

or yes (0 or 1)

Originally reported

$90 in 2010 High

estimate:

$122 in 2023

USD

4. Costs of the program (treatment costs) - MHC-only.

Cost of Length of stay in MHC

(days) using costs from

Massachusetts Executive

Office of Health and Human

Services (low estimate)

Baseline to

MHC

completion

termination

Continuous

variable: #

days total in

MHC

Originally reported

$3 in 2023 USD*

Low estimate:

$3 in 2023

USD

$7 per day in MHC

(As measured by an

average of Drug

Court and Medicaid

case management

Cost of Length of stay in MHC

(days) using costs from Carey

& Waller (high estimate)

Baseline to

MHC

completion

termination

Continuous

variable: #

days total in

MHC

Originally reported

$8 in 2010 USD

High

estimate: $11

in 2023 USD

costs) cd

Cost of Real number of court

appearances using costs from

Carey et al. (low estimate)

Cost of Real number of court

appearances using costs from

Carey et al. (high estimate)

For total

term of

MHC

participation

For total

term of

MHC

participation

Continuous

variable: #

court

appearances

per total

MHC days

Continuous

variable: #

court

appearances

per total

MHC days

Originally reported

$74 in 2004 USD

Originally

reported: $121 in

2004 USD

Low estimate:

$117 in 2023

USD

High

estimate:

$192 in 2023

USD

$155 per court

appearance

Notes: MHC denotes mental health court; MH denotes mental health; SA denotes substance abuse; tx denotes treatment; BW

denotes bench warrant. Blue denotes dependent variable.

a. See sheet supporting these figures entitled 'Notes on work.'

b. This dollar figure is for Supplement Security Income only. May need to factor in other types of disability benefits.

c. Cost per session of case management to Medicaid in Massachusetts in 2023 USD is $18.21 (15 minutes, non-master’s level

counselor) and $28.94 (master’s level clinician). Assuming a mean average between these two costs, the average in 2023

USD is $23.58 per session. Assuming one session per week, the cost per day in 2023 USD is $3.37.

d. Assume one meeting per week based on the following in the best practices section: “Participants meet individually with a

clinical case manager or comparable treatment professional at least weekly during the first phase of mental health court.

(Michigan Assoc. of Treatment Court Professionals, 2021).”

5.2.4 Independent variables

The independent variables used in this CBA are the same as those used in the outcome

regression section. See Chapter: 4.4 (Outcomes associated with MHC participation; Independent

variables). These independent variables are only used in the cost regression analysis in this

section.

5.2.5 Net economic contribution to society approach

This is the first method of this CBA chapter. The net economic contribution to society

approach uses a Microsoft Excel spreadsheet analysis. This approach calculates the net

benefitcost ratios for MHC participation as opposed to TAU participation in criminal court. This

analysis is independent from the cost regression analysis.

Net benefit-cost equation:

1. The average net cost of MHC per MHC participant over the first 18 months from

diversion to court is (x). (Variables listed in sections three and four of table 5.1 above)

2. The average net benefit of MHC per MHC participant over the first 18 months from

diversion to court is (y). (Variables listed in sections one and two of table 5.1 above)

3. The average net benefit-cost ratio is (y/x).

Net program costs: The difference in the direct costs of implementing MHCs as compared to

criminal courts for the average MHC participant for the 619 valid participants in the CBA. Net

program costs are averaged to be the same per MHC participant in the benefit-cost analyses. Net

outcome benefits: The difference in savings of implementing MHCs as compared to criminal

courts for the average MHC participant for the 619 valid participants in the CBA.

Net costs and benefits take into account:

Net benefit-costs (computed dollar value to 2023 USD) = benefits of the intervention – costs

of the intervention = total net economic contribution to society (If dollar value is positive, then

benefits outweigh the costs. If dollar value is negative, then costs outweigh the benefits) (Office

of the Associate Director for Policy and Strategy, 2022).

The net economic contribution to society is calculated as a benefit-cost ratio using the

above equation in this analysis.

5.2.5.1 Stratified by mental health diagnosis approach

The stratification is completed in an Excel spreadsheet analysis. A separate benefit-cost

ratio analysis is completed on each stratified group; namely the mental health diagnosis at

baseline and the crime severity at baseline. Stratification by mental health diagnosis is achieved

through sorting the 619 valid participants into three groups: (1) diagnoses of anxiety, adjustment,

personality, depression, pending, and other axis 1 (n = 257 study participants); (2) diagnoses of

bipolar, other psychotic dx (n = 213 study participants); and (3) diagnosis of schizophrenic (n =

149 study participants). The net economic contribution to society for each of these stratified

groups is then calculated for each strata based on participation in the MHC group or criminal

court group. Net program costs are the same per participant as in the overall benefit-cost ratio.

The benefit-cost ratio is calculated for each of the three mental health diagnosis groups. These

three mental health diagnosis categories are not severity measures.

5.2.5.2 Stratified by crime severity at baseline approach

A new variable is constructed to measure crime severity. This variable is entitled:

‘Precorrections 2 pre 18 months (Winsorized).’ This composite variable is a combination of

prediversion six months days in jail and pre-diversion 18 months days in prison. In the instance

of the ‘Precorrections 2 pre 18 months (Winsorized)’ variable, the pre-diversion six month jail

variable is a six month variable that is multiplied by two instead of three to extrapolate to

prediversion 18 months because the most jail time pre-diversion 18 months to court is likely in

the six months right before diversion to court. See appendix table A-8 for supporting statistics for

computation of corrections cost for severity score analysis.

Stratification by crime severity is achieved through sorting the 619 valid participants into

three groups: (1) Low severity at baseline (Precorrections 2 18 months (Winsorized)– 207 study

participants who have an average of 7.1 days in corrections in the 18 months prior to diversion to

court); (2) Medium severity at baseline (Precorrections 2 18 months (Winsorized) - 206 study

participants who have an average of 68.2 days in corrections in the 18 months prior to diversion

to court); and (3) High severity at baseline (Precorrections 2 18 months (Winsorized) - 206 study

participants who have an average of 305.3 days in corrections in the 18 months prior to diversion

to court). The net economic contribution to society for each of these stratified groups is then

calculated for each strata based on participation in the MHC group or criminal court group. The

benefit-cost ratio is calculated for each of the three crime severity groups.

5.2.6 Cost regression approach

This is the second method of this CBA chapter. This cost regression uses a composite

continuous dependent variable, ‘Net economic contribution of MHC participation per

participant’,’ and the same 18 independent variables used in the first set of regression analyses

(See Appendix A-2, section 4. Independent variables for hypothesis testing for a listing of these

independent variables within the table: Selected domains and variables). Net program costs are

the same per participant as in the overall benefit-cost ratio. The ‘net economic contribution to

society per participant’ dependent variable is determined by a summation of the individual

participant benefits of the previously discussed eight outcomes variables listed under the

dependent variable section above (Section 5.2.4 Dependent variables). Each of the participants in

this analysis have a cost for each of the eight outcome variables that is summed to arrive at the

‘net economic contribution to society per participant’ dependent variable. This dependent

variable has a cost associated with each study participant. The cost regression analysis uses the

same independent variables as those listed in Chapter 4, section 4.2.4 Independent variables. All

of the cost regression results are weighted using the IPTW method. The regression is performed

in SPSS.

5.3 Cost benefit findings

5.3.1 Net benefit-cost ratio findings

These are findings from the first analysis in this CBA chapter, which is net benefit-cost

ratio analysis for MHC participation as opposed to TAU participation. This analysis is an Excel

spreadsheet analysis. If a net benefit-cost ratio exceeds one, then the program is beneficial.

Table 5.4 shows the finding that, without considering program costs, the net economic

contribution to society per MHC participant as compared to TAU participant over the first 18

months of court is $15,484. This dollar figure is the average economic cost per participant

without program costs of MHC (-$35,565) minus the average economic cost per participant

without program costs of TAU (-$51,049).

TAU ( n=319) MHC ( n=300)

-$51,049 -$35,565

Table 5.4: Net economic contribution per MHC participant over the first 18 months

from diversion to court.

Average economic contribution per

participant without program costs

Net economic contribution to society per $15,484

MHC participant over first 18 months

Notes: MHC denotes mental health court; TAU denotes treatment as usual in criminal

court. Negative costs are negative economic contributions.

Table 5.5 below delineates program costs, and shows calculations in Microsoft Excel that

demonstrate that the program costs of MHC participation as opposed to TAU participation per

MHC participant over the first 18 months post-diversion to court are $10,655. Program costs are

estimated over 18 months because this is the time length for the composite outcome variable

used in all analyses in this CBA chapter, and this is the timeframe for the benefit-cost ratio

calculations. Each participant has the same average program cost. There are two reasons for why.

The first is that consistency is needed for the 18 month time interval for the analysis. The second

is that the data does not contain the variables needed for individual participant cost analysis.

Table 5.5: Program costs summary - Statistical support to analysis tables (per TAU and MHC participants; this analysis

uses only the 619 valid participants with no missing data in the benefits analysis).

Average Average cost

over 18 cost over 18

Shared program inputs: TAU criminal months by months by court and MHC services TAU group

MHC group Difference

Cost estimate

(in 2023

USD) Unit of measurement

6M: Community Supervision yes no $8 per day $2,559 $3,088 $529

6M: Days in psychiatric hospital $1,031 per day $8,052 $11,449 $3,397

3.3 In past 6 mos, received tx for mental

or emotional problems

$2,607 per six months (one day

per week)

$6,521 $6,757 $236

3.4 In past 6 mos, received alcohol, drug,

or detox tx

$2,924 per six months (one day

per week)

$4,370 $4,719 $349

Subtotal $21,501 $26,013 $4,511

MHC-only services

Length of stay in MHC (days) $7

per day

$3,227 $3,227

Real number of court appearances $155 per court appearance NA $2,916 $2,916

Subtotal $0 $6,144 $6,144

Total $21,501 $32,157 $10,655

Notes: TAU denotes treatment as usual; MHC denotes mental health court; CC denotes criminal court; 6M denotes 6 months; mos.

denotes month(s). Extrapolation is achieved through using mean values of the original unstandardized variable to arrive at 18

month standardized values. All of this analysis uses weighted data. Negative costs are negative economic contributions.

The below table 5.6 ‘Net benefit-cost ratio’ shows that the average net benefit-cost ratio

is 1.45. For the average participant, MHC as opposed to TAU participation has more benefits

than costs.

The 95% confidence interval of the average economic contribution per participant (with

both MHC and TAU participants) in the benefit cost ratio is: lower bound - $46,614, and upper

bound - $40,563. If the same standard error used in this first calculation ($1,539) is applied to the

average net benefit, then the corresponding 95% confidence interval for the average net benefit

of MHC per MHC participant over the first 18 months from diversion to court is: lower bound

$12,406 and upper bound $18,562. The resulting low and high average net benefit-cost ratios are

1.16 to 1.74.

Table 5.6: Net benefit-cost ratio.

TAU MHC

(n=319) (n=300)

The average net cost of MHC compared to TAU

per participant over the first 18 months from

diversion to court is (x) (see costing worksheet

draft for details):

$10,655

The average net benefit of MHC per MHC

participant over the first 18 months from

diversion to court is (y):

$15,484

The average net benefit-cost ratio is (y/x): 1.45

The 95% confidence interval of the average economic contribution per

participant (with both MHC and TAU participants) in the benefit cost ratio is:

lower bound - $46,614, and upper bound - $40,563. If the same standard

error used in this first calculation ($1,539) is applied to the average net

benefit, then the corresponding 95% confidence interval for the average net

benefit of MHC per MHC participant over the first 18 months from

diversion to court is: lower bound $12,406 and upper bound $18,562. The

resulting low and high average net benefit-cost ratios are 1.16 to 1.74.

Notes: MHC denotes mental health court; TAU denotes treatment as usual in

criminal court. Results are standardized to 18 months, which is the first 18

months after diversion to a MHC or criminal court. The unstandardized

dependent variables are either 18 months or 6 months in duration; both from

the abovementioned time of diversion. The two dependent variables that are

6 months in duration are extrapolated through using mean values of the

original unstandardized variable to arrive at 18 month standardized values.

5.3.2 Stratification Findings

The two areas for stratifying study participants by treatment group and another group are

mental health diagnosis at baseline and crime severity at baseline (crime severity is measured by

days in jails and prisons in the 18 months pre-baseline to baseline to court diversion).

The average net cost of MHC (MHC denotes mental health court) per MHC participant

over the first 18 months from diversion to court is (x) (see costing worksheet draft for details):

$10,655. This cost is the same for all benefit-cost ratios in this dissertation. The average net

benefit of MHC per MHC participant over the first 18 months from diversion to court is (y):

$15,484 for the non-stratified overall benefit-cost ratio. The non-stratified overall benefit-cost

ratio is 1.45 for the 619 valid study participants in the second set of regressions.

5.3.2.1 Findings stratified by mental health diagnosis

This section discusses stratification by mental health diagnosis measured at baseline

diversion to court. Mental illness or serious mental illness with diagnoses of anxiety, adjustment,

personality, depression, pending, and other axis one, for MHC participants as compared to TAU

(TAU denotes treatment as usual) participants, has the second most net benefits in this strata.

This group in this strata has a benefit-cost ratio (1.94) that is higher than the non-stratified

overall benefit-cost ratio (1.45). The average net benefit of MHC per MHC participant over the

first 18 months from diversion to court is (y): $20,664 for these diagnostic categories.

Serious mental illness with diagnoses of bipolar and other psychotic disorder, for MHC

participants as compared to TAU participants, have the least net benefits in this strata – this

group is the only group in this strata that has a benefit-cost ratio (1.28) that is lower than the

nonstratified overall benefit-cost ratio (1.45). The average net benefit of MHC per MHC

participant over the first 18 months from diversion to court is (y): $13,614 for these diagnostic

categories.

Serious mental illness with diagnosis of schizophrenic disorder (includes both

schizophrenia and schizoaffective disorder) for MHC participants as compared to TAU

participants, have the most net benefits in this strata – this group has a benefit-cost ratio (2.08)

that is higher than the non-stratified overall benefit-cost ratio (1.45) and is the highest in this

strata. The average net benefit of MHC per MHC participant over the first 18 months from

diversion to court is (y): $22,184 for this diagnostic category.

Table 5.7: Net benefit-cost ratios for stratified variable: 'mental health diagnosis.'

Overall net

benefit-cost ratio

Net benefit-cost ratio -

diagnostic category

one (anxiety,

adjustment,

personality,

depression, pending,

other axis one

diagnoses)

Net benefit-cost

ratio - diagnostic

category two

(bipolar and other

psychotic

diagnoses)

Net benefit-cost

ratio - diagnostic

category three

(Schizophrenic

diagnoses)

TAU MHC

(n=319) (n=300)

TAU MHC

(n=165) (n=92)

TAU MHC

(n=114) (n=99)

TAU MHC

(n=40) (n=109)

The average net cost of MHC compared

to TAU per participant over the first 18

months from diversion to court is (x) (see

costing worksheet draft for details):

$10,655 $10,655

$10,655

The average net benefit of MHC per

MHC participant over the first 18 months

from diversion to court is (y):

$15,484 $20,664 $13,614 $22,184

The average net benefit-cost ratio is (y/x): 1.45 1.94 1.28 2.08

Notes: MHC denotes mental health court; TAU denotes treatment as usual in criminal court. Results are standardized to 18

months, which is the first 18 months after diversion to a MHC or criminal court. The unstandardized dependent variables are either

18 months or 6 months in duration; both from the abovementioned time of diversion. The two dependent variables that are 6

months in duration are extrapolated through using mean values of the original unstandardized variable to arrive at 18 month

standardized values. "Serious mental illness (SMI) commonly refers to a diagnosis of psychotic disorders, bipolar disorder, and

either major depression with psychotic symptoms or treatment-resistant depression; SMI can also include anxiety disorders, eating

disorders, and personality disorders, if the degree of functional impairment is severe" (Evans et al., 2016, Disparities Within

Serious Mental Illness, Page 1).

5.3.2.2 Findings stratified by crime severity at baseline

This section discusses stratification by crime severity measured by number of days in

jails and prisons in the 18 months pre-baseline to baseline diversion to court. Low crime severity

MHC participants as compared to TAU participants have the second most net benefits in this

strata – this group in this strata has a benefit-cost ratio (1.62) that is higher than the non-stratified

overall benefit-cost ratio (1.45). The average net benefit of MHC per MHC participant over the

first 18 months from diversion to court is (y): $17,252 for the stratified low crime severity

treatment group benefit-cost ratio.

Medium crime severity MHC participants as compared to TAU participants have the

most net benefits in this strata – this group has a benefit-cost ratio (2.05) that is higher than the

non-stratified overall benefit-cost ratio (1.45) and is the highest in this strata. The average net

benefit of MHC per MHC participant over the first 18 months from diversion to court is (y):

$21,791 for the stratified medium crime severity treatment group benefit-cost ratio.

High crime severity MHC participants as compared to TAU participants have the least

net benefits in this strata – this group is the only group in this strata that has a benefit-cost ratio

(0.94) that is lower than the non-stratified overall benefit-cost ratio (1.45), and is the only group

in this entire stratification analysis of mental health severity and crime severity that does not

have a benefit-cost ratio above 1. The average net benefit of MHC per MHC participant over the

first 18 months from diversion to court is (y): $9,995 for the stratified high crime severity

treatment group benefit-cost ratio.

Table 5.8: Net benefit-cost ratios for stratified variable: 'Precorrections 2 pre 18 months (Winsorized).'

Overall net

benefit-cost ratio

Net benefit-cost

ratio - low crime

severity at

baseline

Net benefit-cost

ratio - medium

crime severity at

baseline

Net benefit-cost

ratio - high crime

severity at

baseline

TAU MHC

(n=319) (n=300)

TAU

(n=120)

MHC

(n=87)

TAU

(n=92)

MHC

(n=114)

TAU

(n=107)

MHC

(n=99)

The average net cost of MHC compared to TAU

per participant over the first 18 months from

diversion to court is (x) (see costing worksheet

draft for details):

$10,655 $10,655

The average net benefit of MHC per MHC

participant over the first 18 months from

diversion to court is (y):

$15,484 $17,252 $21,791 $9,995

The average net benefit-cost ratio is (y/x): 1.45 1.62 2.05 0.94

Notes: MHC denotes mental health court; TAU denotes treatment as usual in criminal court. Results are standardized to 18

months, which is the first 18 months after diversion to a MHC or criminal court. The unstandardized dependent variables are

either 18 months or 6 months in duration; both from the abovementioned time of diversion. The two dependent variables that

are 6 months in duration are extrapolated through using mean values of the original unstandardized variable to arrive at 18

month standardized values.

5.3.3 Cost regression findings

These are findings from the second analysis in this CBA chapter, which is a cost

regression. The dependent variable in this linear regression model is: ‘Net economic contribution

to society per study participant.’ The most important finding from this regression is the amount

that the dependent variable changes based on whether a participant is in the MHC or TAU group.

The average economic cost per participant over the duration of time in court is -$52,456. This

cost regression model finds that there is an average net economic contribution of $20,676 for

MHC participants as opposed to TAU participants.

All cost regression results are weighted using inverse probability of treatment weighting.

See appendix table A-9 for the histogram of the dependent variable. This histogram is left

skewed. Skewness is -1.697, which is between -3 and 3. This suggests that the dependent

variable is not too skewed for the regression analysis. The Adjusted R Squared is 0.160,

suggesting that the independent variables explain 16.0% of the variance in the dependent

variable in this model.

While this cost regression analysis is run with all of the same independent variables as

the outcome regression analysis, only the treatment or control group variable has findings that

are important for this research. For this reason, only this treatment group variable is reported in

the below table.

Table 5.9: Cost regression coefficients - dependent variable 'Total net economic

contribution to society per participant.'

95% CI 95% CI

Independent Unstandardized Standard (lower (upper variable Beta

error Significance bound) bound) MHC Group $20,676 $4,794 0.000

$11,261 $30,090

Constant (dependent -$52,456 $12,718 0.000 -$77,431 -$27,481

variable)

Notes: Blue cells denote statistically significant value. All statistics in this table are weighted

using inverse probability of propensity score weighting. The model fit statistic for this model

is the Nagelkerke R Square (0.323). The omnibus test is significant at the <.001 level.

5.4 Discussion of cost benefit findings

Benefits outweigh costs for MHCs as opposed to criminal courts. This result is consistent

across the net benefit-cost ratio findings and the cost regression findings discussed in this CBA

chapter. Schizophrenic disorders have the most net benefits in the stratification by mental health

diagnoses. This result suggests that the most serious diagnosis upon diversion to court has the

most net benefits by diagnostic category. All of the results by diagnosis have a net benefit-cost

ratio for MHC participation that is greater than one (the break-even point). What follows is that

on average participants in MHCs as stratified by mental health diagnoses in this analysis may

save more money in the first 18 months post diversion to court than they cost.

The low and medium crime severity groups in the stratification have benefit-cost ratios

for MHC participation that are greater than one (the break-even point) and the benefit-cost ratio

of the unstratified sample. These results suggest that participants with low and medium crime

severities, and particularly medium crime severities which have the greatest net benefit-cost ratio

out of all crime severity levels, may be the best stratified samples to target for MHC

participation. The high crime severity stratified sample did not return investment in the first 18

months post diversion to court based on the benefit-cost ratio, but may do so after. This stratified

sample may be more likely to be involved in the traditional criminal justice system including

days in jails and prisons for both the treatment and control group for the first 18 months

postdiversion to court. However, this sample may show greater savings in a longer timeframe

costbenefit analysis study.

See the appendix for Table A-10 for the 'Net economic contribution to society per

participant' compared with 'Cost regression' based on the average benefit per participant and

standard deviation. The number of valid TAU participants in this analysis is 319. The number of

valid MHC participants in this analysis is 300.

The hypotheses for the first and second CBA research questions are supported by this

CBA. The hypotheses for the first research question (5.1) are fully supported. Benefits outweigh

costs for MHCs compared to criminal courts. This is true even though MHC participants utilize

more mental health treatment than criminal court participants. The hypotheses for the second

research question (5.2) are partially supported. The highest severity diagnosis (a diagnosis of a

schizophrenic disorder) is shown in the analysis to have the greatest benefits as opposed to costs

for all stratified mental health groups. Based on the data, it is clear that a diagnosis of a

schizophrenic disorder is the highest severity as a category based on the three mental health

diagnostic categories. The severity ordering of the other mental health diagnoses, however, is

unclear. Benefits outweigh costs for all mental health diagnostic categories in this analysis.

The hypotheses for the third CBA research question are not supported. Targeting

individuals with high severity of crime as determined by severity of crime at baseline is less

likely than other crime severity levels to have benefits that outweigh costs in MHCs as compared

to criminal courts. Overall benefits do not outweigh the total costs for MHCs compared to

criminal courts for participants who have high severity of crime at baseline.

5.4.1 Strengths

A major strength of this research is the comprehensiveness of the analysis. Several

recognized problems with the data are corrected statistically. One bias related to unbalanced

samples is identified and addressed statistically. Inverse probability of treatment weighting

(IPTW) is used to balance the MHC group and criminal court group samples at baseline, which

reduces selection bias.

Another major strength is that this research utilizes multiple quantitative analysis models

that use different but complementary statistical techniques. Quantitative analysis includes

outcome regression models, a cost regression model, and a financial analysis of the net economic

contribution to society per study participant. All of these models include study participant data

that spans up to three years. The results are discussed independently and in summation. The

results from these different quantitative analysis models are not conflicting but complementary.

5.4.2 Cost benefit analysis limitations

After all of these limitations in this section and in the above limitations section 4.4.1

entitled ‘Limitations of the outcome chapter relevant to both analyses chapters’ are considered,

there are no foreseeable differences between the MHC group and the control group at baseline.

MHCs may be better off than what the dissertation found.

The first limitation is that it is not possible to tangibly cost quality adjusted life years

(QALYs) of crime. There is too much uncertainty about how to cost QALYs to include them as a

tangible cost. In the literature, they are discussed as tangible and intangible costs. For more

information, see the appendices A-11 Quality of life literature summary, A-12 Quality of life

costing summary, and A-13 Qualitative discussion of cost to the victim.

Further explanation of costing changes in QALYs based on crime estimations in the

literature

Miller estimates the cost of QALYs to victims and their family’s at 85% of the total costs

of violent crime, or on average $77,055 2017USD$ (average QALY estimate for a violent crime)

out of $91,110 2017USD$ (average total estimate for a violent crime). Miller’s methodology is

delineated in the ‘Qualitative discussion of cost to the victim’ subsection of this dissertation.

From an intangible qualitative perspective, a percentage or dollar amount cannot be placed on

these costs of QALYs, but indirect benefits and costs are evident.

Additional evidence to Miller include Florence et al., who finds that there are significant

decreases in the value of a statistical life year (VSLY) for illegal opioid use. This analysis is

costing the value of life lost of the drug user. These authors cost a VSLY at $517,324, a reduced

quality of life at $390,003. Most significantly, these authors cost the annual reduced quality of

life lost due to illegal opioid use at $390 billion and the annual actual life lost at $481 billion.

McCollister et al. uses Neumann et al.’s quality adjusted life year costs (low projection in

2016USD$ = $50,000; medium projection in 2016USD$ = $100,000; high projection in

2016USD$ = $200,000.) and applies them to QALYs costing methodology that is similar to

Miller.

According to Robinson and Hammitt in ‘Skills of the Trade: Valuing Health Risks

Reductions in Benefit-Cost Analysis,’ “The QALY is a nonmonetary measure that integrates the

duration and severity of injury or illness. QALYs were originally developed for use in ranking

and prioritizing public health problems and in analyzing the cost-effectiveness of heath policy

and medical treatment decisions (Zeckhauser and Shepard, 1976). They are also widely used to

compare health status across individuals or population groups. In these contexts, QALYs are

generally not assigned a monetary value, but monetization is needed to apply these estimates in

benefit-cost analysis” (2013, pages 119-120).

The second limitation is that it may be hard to quantify cost-benefit analyses with regard

to direct savings to all specific agencies involved in MHCs. Agencies oftentimes like to know

their own direct costs and benefits for program and social policy decision-making.

MHCs may be worse off than what the dissertation found.

The third limitation is that each participant has the same average program cost in the

CBA. This includes the same program cost for each mental health diagnostic group and crime

severity group. This limitation may result in higher net benefit-cost ratios for the highest cost

mental health diagnostic groups and crime severity groups because their program costs may be

higher than the average program cost. The fourth limitation is that MHCs shift costs to other

programs such as Medicaid and mental health services. Whether this cost-shifting happens is not

tangibly quantifiable.

Can not tell whether MHCs may be better off or worse off than what the dissertation found.

The fifth limitation is that the cost sample is 619 participants. Sample loss is a resulting

limitation from both of these sample sizes that are smaller than the entire study population (1,047

participants). The sixth limitation is that there may be costs to the defendant that are not tangibly

quantifiable. The seventh limitation is that there is a high degree of uncertainty in the following

areas that are costed: government disability benefits support (costed by ‘6M: Received disability

benefits (Yes or no)’), life satisfaction (costed by ‘6M: Life satisfaction’), and housing (costed by

‘6M: Days homeless’).

The eighth limitation is that the costs and benefits used in this dissertation are

comprehensive, but there are costs and benefits that are not directly costed due to the limitations

of the data. Nontangible outcomes need to be considered if they represent indirect costs and

benefits of MHCs to society. Nontangible outcomes include indirect costs and benefits to

victims, families, communities and MHC participants. Specifically, they include keeping families

together, the good of individuals with mental illness being in the community, single parents being

with their children, caretakers being with those in need, and the dignity of work.

These areas are partially costed but are not fully captured in the costing variables already used.

The ninth limitation is that in the cost regression, the diagnoses findings are hard to interpret

because everybody had at least one diagnosis.

Chapter 6. Policy relevance

6.1 Overview - combined analyses discussion

The overall findings of this dissertation are that Mental Health Courts (MHCs) have more

net economic contributions than costs to society as compared with criminal courts. From a

costbenefit perspective, MHCs are good and useful.

This dissertation highlights a need to give serious consideration to the expansion of

MHCs. The outcome regression models results show how impactful MHCs are compared to

criminal courts in the areas of increasing employment, decreasing time in jail, decreasing

recidivism for not-serious crimes, and decreasing time in prison. While many assumptions were

made, this study finds substantial net economic contributions to society for MHC participants as

opposed to criminal court participants over the first 18 months post-diversion to court. The

theory, cost-benefit analysis from the perspective of welfare economics theory and public finance

theory, as applied to this analysis, supports a recommendation of more MHCs.

Previous studies find that MHCs are efficacious in reducing recidivism (VanGeem,

2015). In regard to the overall findings, this dissertation is different than Steadman et al..

Steadman et al. did not find that overall MHCs as opposed to TAU had more economic benefits

than costs. This dissertation, however, also builds off of the Steadman et al. study by looking at

participants with a cooccurring substance use disorder and by crime severity. Steadman et al.

argues that to contain costs these participants should be included in MHCs. The study presented

in this dissertation differs from the Steadman et al. study by (1) looking at societal value in

addition to taxpayer savings, (2) focusing on a three year study period instead of a six year study

period, (3) stratifying by mental health diagnosis and crime severity, (4) employing a different

matching technique to balance the sample at baseline, and (5) completing financial benefit-cost

ratio analyses in addition to regression modeling (Steadman, et al., 2014).

6.2 Audiences

The audiences for these policy recommendations are key government policy decision

makers at the levels of district court jurisdictions, state court jurisdictions, and the federal

government agencies of the Bureau of Justice Assistance (BJA) and the Substance Abuse and

Mental Health Services Administration (SAMHSA). Also included in the audience are crosslevel

government actors including academic experts, politicians and nonprofit decision makers.

At the district court and state court jurisdiction levels, the selected audiences are the

“police and sheriffs’ officials, judges, prosecutors, defense counsel, court administrators, pretrial

services staff, and corrections officials; mental health, substance abuse treatment, housing, and

other service providers; and mental health advocates, crime victims, consumers, and family and

community members” (Thompson, Osher, & Tomasini-Joshi, 2007) (Page 1). A special focus is

given to the judiciary, who in many cases leads the implementation of MHCs. There may also be

a MHC planning committee or advisory group at the court level mostly or completely comprised

of the individual actors listed in this section. The audience for this policy relevance section may

include individual actors and groups of actors that may have competing interests.

The federal government selected audiences are BJA and SAMHSA. Specifically, BJA

partially funds some MHCs. BJA partners with district court and state courts to fund the creation,

implementation, and expansion of Mental Health Courts (Bureau of Justice Assistance, 2024).

BJA is the best funding opportunity for MHC startups (Justice Center: The Council of State

Governments, 2024).

In the recent past, SAMHSA has funded grants to individual courts (Executive Office of

the Trial Court, 2024). Specific centers at SAMHSA involved in funding MHCs are the Center

for Substance Abuse Treatment and the Center for Mental Health Services (Substance Abuse and

Mental Health Services Administration, 2024).

Cross-agency and government level actors are important audiences as well. In addition to

academic experts, included are politicians at all levels of government, foundations, and other

private non-profit stakeholders (e.g., outpatient mental health providers and insurers, community

health centers, and victims advocacy groups).

6.3 Recommendations

MHC expansion should target subgroups

Policy actors should give serious consideration to the expansion of MHCs. The policy

actors who implement the expansion of MHCs should first consider for inclusion in these courts

subgroups that have the best benefit-cost ratio. Recommendations are based on stratified analysis

by diagnosis of mental illness at the time of diversion to court and severity of crime based on

incarceration in the 18 months prior to diversion to court.

Participants with a diagnosis of schizophrenia have the highest MHC group benefit-cost

ratio out of all stratified diagnostic groups. Their average net benefit-cost ratio is 2.08 – meaning

that over the 18 months of the quantitative analysis, their tangible economic contributions are

more than twice their costs. Specifically, their average net MHC benefits are 2.08 times costs

over the same 18 month period. This benefit-cost ratio far exceeds the threshold of one needed

for a program to be beneficial. There may not be a standard benefit-cost ratio threshold for

government investment. A recommendation is that these individuals with a diagnosis of

schizophrenia as compared to other mental health diagnoses should be prioritized for MHC

expansion and existing courts. This prioritization is particularly important if MHCs are faced

with more potential participants than they can fund. The other two stratified subgroups based on

diagnoses are both above 1.00, and thus the recommendation is to expand for all mental health

diagnostic subgroups in this study (the diagnoses of anxiety, adjustment, personality, depression,

pending, and other axis 1 as a stratified group have a benefit-cost ratio of 1.94, and the diagnoses

of bipolar and other psychotic disorder as a stratified group have a benefit-cost ratio of 1.28).

Participants with a medium level of crime severity are high-impact. They have the highest MHC

benefit-cost ratio out of all levels of crime severity. Their average net benefit-cost ratio is 2.05.

Specifically, their average net MHC benefits are 2.05 times costs over the same 18 month period.

A recommendation is that individuals in this crime severity level should be prioritized for MHC

expansion and existing courts. This prioritization is particularly important if MHCs are faced

with more potential participants than they can fund.

Low-crime severity participants also have a very favorable net benefit-cost ratio at 1.62.

The question then becomes what to do with the high-crime severity level whose net benefit-cost

ratio is 0.94. The recommendation is to keep these individuals in MHCs for the following

reasons: (1) they may need more time to show a positive benefit-cost ratio return due to many jail

and prison days in the first 18 months post-diversion to court; (2) mental health services

administered in the first 18 months post-diversion to court may pay off later and if MHC services

are removed from the benefit cost-ratio, the incarceration costs would likely be less for the MHC

group; (3) rehabilitation to community living may have the greatest long-term economic

contributions per participant for this group because they may be the most costly to incarcerate,

and (4) the intangible benefits may be large for this group (e.g. decreasing recidivism for highest

severity MHC offenders may substantially decrease costs to the victims, and the value in keeping

families together for what might otherwise have been a long criminal sentence of incarceration).

MHCs and the 1999 Olmstead Act

MHC policymakers are recommended to consider the case for the application of the 1999

Olmstead Act to MHC expansion. The Olmstead Act is based on the 1999 U.S. Supreme Court’s

decision that found that “the unjustified segregation of people with disabilities is a form of

unlawful discrimination under the Americans with Disabilities Act (ADA)…The Court held that

states are required to provide community-based services for people with disabilities who would

otherwise be entitled to institutional services when: a) such placement is appropriate; b) the

affected person does not oppose such treatment; and c) the placement can be reasonably

accommodated, taking into account the resources available to the state and the needs of other

individuals with disabilities.” (U.S. Department of Health and Human Services, 2024)

The contribution that this research makes to the enforcement of the 1999 Olmstead Act is

primarily in furthering the understanding of the costs of the reasonable accommodations required

for participants with mental illness to receive community-based treatment in MHCs instead of

criminal court participation. This understanding helps to address the foundational question:

Should individuals with mental illness be prosecuted in criminal courts or diverted to MHCs?

Criminal court participation much more frequently results in institutionalized treatment in jails

and prisons than MHC participation. Oftentimes, for new policies and programs to be enacted in

the government at the district court, state, and federal levels, financial benefits have to be

projected to outweigh costs in addition to proven clinical efficacy. This policy recommendation

may help to better align the treatment of individuals with mental illness (who as participants in

MHCs are mostly are diagnosed with serious mental illnesses) in the criminal justice system with

the implementation of the Americans with Disabilities Act.

Access to MHCs

The final recommendation is that district court, state and federal policymakers consider

the concept of whether MHCs serve the public best as voluntary for states and courts (the current

model), or whether MHCs should be available to every person with mental illness, or specifically

serious mental illness only, at the time of arrest for diversion from criminal courts.

Chapter 7. Conclusion

The comprehensiveness of this quantitative analysis and the utilization of multiple

quantitative analysis models that confirm the findings strengthen the applicability of the findings

to the policy recommendations. They also provide a solid foundation for future research. The

limitations make generalizability to other MHCs not possible. The limitations also include

sample loss and the potential for selection bias that could not be fully resolved. These limitations

need to be considered for the applicability of findings. Not every benefit and cost could be

directly costed. Future research may want to delineate costs by government agencies and other

stakeholders.

This cost-benefit analysis advances prior research and adds to the current literature on

MHCs by examining MHC costs and benefits from a government value and societal value

perspective. The government value perspective considers public sector resources, including

expenditures such as current criminal court and MHC operation. The societal value perspective

considers all costs and benefits to society, including all tangible and intangible costs and benefits

to society. The societal perspective goes beyond value to the taxpayer to include, for instance, the

economically tangible outcome of MHC participation on work. Not all outcomes can be

monetized, but they are still important to consider.

This dissertation presents a story of the promising cost-benefit analysis of MHCs for men

and women with mental illness, most of whom have serious mental illness.

COST-BENEFIT ANALYSIS OF UNITED STATES MENTAL HEALTH COURTS RELATIVE TO CRIMINAL COURTS FOR PARTICIPANTS WITH MENTAL ILLNESS

Students also viewed