Factors That Influence Program Completion_ Predicting Kentucky

CHAPTER I

INTRODUCTION

What society should do with the criminal population is a matter that every

generation of society’s leaders is called to answer. The response to this question can fall

somewhere between two extremes. On one end, we recognize offenders as victims of

underlying societal and personal troubles and therefore worthy of remaining a part of

society because we ought to fix the underlying problem. On the other end, we define

offenders by their behaviors and recognize them as criminal offenders and nothing more.

Many court jurisdictions choose to approach the drug addicted offender with a

specialized treatment-based program that defers a prison or jail sentence with the

opportunity to remain a part of society. This program, called drug court, brings together

the supervision the “criminal” warranted by his or her behaviors and the treatment the

“victim” needs to alleviate the social and personal issues. Success under the drug court

model is living a clean and sober life free of criminal behaviors.

Almost 30 years ago Miami-Dade Drug Court became the first treatment-based

drug court in the nation (Hoffman, 2000; Whiteacre, 2008). Motivated by concerns of

ineffective justice processing and system overload (Meithe et al., 2000), Miami-Dade

Drug Court redesigned case processing for the drug addicted offender. The thinking

behind this change was that the drug addicted and drug abusing population was a major

contributing factor in overloaded dockets. The then Florida Attorney-General, Janet

Reno, believed that this particular type of offender was stuck in a revolving door in and

out of the courts. If the drug addiction could be alleviated, then the criminal activities due

to and in support of the addiction would cease. The goal was to reduce “substance abuse

and criminal behavior while freeing the court and corrections systems to handle other

cases” (National Institute of Justice, 2006, p. 1).

After the emergence of the Miami Dade Drug Court, the drug court concept

gained popularity and began to spread across the country. According to the National

Association of Drug Court Professionals (NADCP) (2011), the number of drug courts in

operation today is estimated at more than 2,500. In some jurisdictions, drug court is

further specialized into family drug courts, juvenile drug courts, and veteran drug courts

(NADCP, 2011).

Drug court is a non-traditional approach to criminal supervision. The goal of drug

court is to reduce the burden of drug and drug related cases on the court system by using

mandated treatment and close supervision by the drug court judge (Hoffman, 2000;

NADCP, 2011; NIJ, 2006). The drug court model was theorized to do this by creating a

specialized caseload for drug offenders with a low risk of committing further criminal

acts and with low intensity drug histories (Marlowe et al., 2006). This new case

processing approach brought together criminal supervision and drug rehabilitation into a

single venue to address drug addiction as the root cause of criminal behavior (NIJ, 2006).

If drug addiction can be stopped, then the crime associated with the drug addiction will

stop as well. The Office of Justice Program (2004) publication, titled Defining Drug

Courts: The Key Components, serves as the guiding doctrine for the key components of

drug courts. In a section titled Purpose, this document defined the mission of drug court:

“the mission of drug courts is to stop the abuse of alcohol and other drugs and related

criminal activity” (p. 1).

Since the first drug court emerged in the late 1980s, the drug court movement

grew, both in terms of scope and number of courts in operation. In scope, the targeted

population has increasingly included offenders whose “substance abuse and criminal

activity may be more serious and pose a greater threat to society…” (OJP, 2006; p. 2)

than the population for which the original drug court model was formed (Leukfeld et al.,

2004; Marlow et al., 2006; Miethe, Lu, & Reese, 2000; Peyton & Grosswieler, 2001). To

examine the current core program requirements for drug court (frequent drug testing,

frequent contacts with supervisor staff, mandated treatment, and frequent contact with the

judge), one might think there is some confusion between level of supervision and level of

risk if the program is targeted for low-level and low-risk drug offenders. For example,

frequent contacts with justice supervision staff, either judge or probation or parole

officer, frequent drug testing, geographic travel restrictions, and even curfew, were once

the domain of intense supervision probation or parole type programs and reserved for the

highest risk offenders. Marlowe, Festinger, Lee, Dugosh, and Benasuitti (2006), in their

study of drug court participant risk and level of judicial supervision, suggest that the risk

principle in drug court may be applied improperly. They assert that “intensive

interventions such as drug court are believed to be best suited for offenders who are high

risk and have more severe criminal propensities or drug use histories but may be

ineffective or contraindicated for offenders who are low risk” (p. 54).

As drug courts grew in number, the drug court movement professionalized and

national level organizations formed. In the mid 1990s, the National Association of Drug

Court Professionals formed with initial membership that included the original drug court

innovators (NADCP, 2011). The National Drug Court Institute, which is concerned with

research and scholarship, appeared toward the end of 1997 and publishes a bi-annual

journal about drug courts. Additionally, the Congress of State Drug Court Associations

formed in the latter part of the 1990s to assist in drafting legislation and securing funding

at the state level. By 2007, the National Association of Drug Court Professionals

recognized 2,147 drug courts in existence with an additional 3,204 other problem-solving

courts (Huddleston, Marlowe, & Casebolt, 2008). As of December of 2009, there were

2,663 drug courts in operation (NADCP, 2011).

After almost 30 years in operation and the growth of drug court across the nation,

the research on its effectiveness at reducing both crime and addiction remains mixed.

While some research finds that drug court results in reductions in recidivism and relapse

(Belenko, 2001; GAO, 2011; Gottfredson & Exsum, 2002; Spohn, Piper, Martin &

Frenzel, 2001) and generates cost savings from “avoided law enforcement efforts,

judicial case processing, and victimization resulting from future criminal activity”

(Huddleston, Marlowe & Casebolt, 2008; p. 6), not all literature is supportive of the drug

court concept. For example, Hoffman (2001) and also Cissner and Rempel (2005) argued

that what drug courts do to produce reductions in recidivism and relapse is not clearly

understood. Belenko (2001) and Cissner and Rempel (2005) as well as Hoffman (2000)

argued that issues with data quality, small sample sizes, defining units of measure and

what time frames are measured often plague the research.

Moreover, judicial commentary not supportive and highly critical of the drug

court concept exists (see Boldt, 2010; King and Pasquarella, 2009; Hoffman, 2000;

Hoffman, 2001). In his law review article titled The Drug Court Scandal (2000), the

Honorable Morris Hoffman described drug court as a “fundamentally unprincipled” (p.

14) “half-crime approach” (p. 14) that suffers from “doctrinal schizophrenia” (p. 14) and

is nothing more than an “appeasement of two powerful political forces—the law

enforcement community and the treatment community” (p. 13). In essence, Hoffman

viewed drug court as a “political fad” (p. 39) that has lost sight of its purpose and argued

that drug courts are a type of “social tinkering” (p. 14) that falls outside the scope of the

judicial branch’s reach. Hoffman (2000) wrote:

…When we succumb to the very human temptation to do more—to fill the

void that is so achingly apparent in so many of the dysfunctional people we see

every day—we not only risk being wrong, we risk being imperial….The moral

authority of our most cherished institutions comes from their voluntary nature: the

value of advice from a priest, a teacher, or a loved one depends in large part on

the fact that we are free to ignore it. But judges’ pieces of “advice” are court

orders, enforceable ultimately by the raw physical power of imprisonment. It is

precisely because of the awesomely enforceable nature of our power that we must

be so circumspect in exercising them. It is one thing for a co-worker, family

member, doctor, or a clergyman to confront someone about a perceived drug

problem; it is quite another thing for a judge to compel drug treatment. Drug

courts not only fail to recognize this important institutional distinction, but their

very purpose is to obliterate it (p. 15).

Concerns with jurisprudence considered, drug courts have flourished yet the field

still lacks a clear understanding of exactly what works, for whom, or even why drug court

might be successful (Marlowe, DeMatteo & Festinger, 2004). Drug courts continue to

grow in number, perhaps, as suggested by Hoffman (2000) because they are appealing to

both law enforcement and treatment communities. Understanding what impact drug court

programming has on both recidivism and relapse is imperative to understanding if drug

courts work for their intended purposes and how they work. The field needs to

understand the factors associated with completing the program to understand why drug

courts may produce reductions in recidivism and relapse. With this knowledge drug court

operators can make informed decisions to assess, improve, and manage their drug courts

(Cheesmam, Rubio & VanDuisend, 2004).

The Drug Court System

At the core of the current drug court model is a fundamental change in the way

the justice system operates in terms of both relationships between justice actors and roles

of the justice actors. The breakdown of traditionally adversarial relationships is an

important factor in the drug court model (OJP, 2006). Rather than each fulfilling their

traditional role (e.g., a prosecuting attorney focused on conviction or a probation/parole

officer focused on criminal supervision compliance), court room working group members

come together as a team on the core premise of helping the offender begin a prosocial

lifestyle. Traditional courtroom working group members come together to form a drug

court team, which changes the work dynamics. Members of the drug court team can

include the prosecutor, defense attorneys, probation/parole officers, treatment social

workers, and even bailiffs. Under the drug court model, the prosecutor focuses less on

proving guilt and more on the individual’s wellbeing. Moreover, the label of “offender”

is changed to “participant” or “client.” Most importantly, the judge’s role “is expanded to

respond to each participant’s positive efforts, as well as to their noncompliance” (Hora,

2002, p. 1473) and it is this relationship that “identifies the judge’s role as key to program

success” (NIJ, 2006, p. 9).

Beyond breaking down adversarial roles and expanding the role of the judge, this

justice supervision adds a treatment requirement and individualized program plans

intended to improve participant life skills (NADCP, 2011). Drug court and treatment staff

identify areas of concern (e.g., employment, education, housing, health care) and work

with the offender toward those goals. Sometimes this includes helping participants divest

themselves of antisocial relationships. It is not unheard of for a drug court judge to order

a “drug court divorce,” which is when an offender is ordered to abstain from his or her

known criminal associates even if that person is a spouse, a parent, or other family

member. Obtaining employment, finishing a GED, enrolling in college, completing

community service, paying child support, and paying all court obligations are all

important parts of a prosocial lifestyle included in drug court programming. Failing to

complete a number of the tasks, which are normally not a part of criminal justice

supervision, may result in a drug court sanction.

Drug Court Operations and Process

In 1997, the Drug Courts Program Office, Office of Justice Programs (OJP),

published a document to establish a guiding doctrine of drug court. In most cases, drug

court is operated at the local or county level. In some instances, a drug court may include

multiple counties and jurisdictions, and for a few states, programs are operated under a

unified state model. Local differences in resources and funding necessitates designing the

drug court to fit the needs of the population and within limits of resources of the

community. As a result, a highly structured and regimented service delivery or program

model does not exist. Rather, best practices were outlined and issued at a national level.

Table 1 shows the key components for the drug court model. These key components

stress the integration of treatment and judicial case processing and outline the

fundamental elements necessary for the drug court programming to be effective.

Table 1

Drug Court Ten Key Components (OJP, 2004)

1. Drug court integrates alcohol and other drug treatment services with justice

system case processing.

2. Using a non-adversarial approach, prosecution and defense counsel promote

public safety while protecting participants’ due process rights.

3. Eligible participants are identified early and promptly placed in drug court.

4. Drug court provides access to a continuum of alcohol, drug and other related

treatment and rehabilitative services.

5. Abstinence is monitored by frequent alcohol and other drug testing.

6. A coordinated strategy governs drug court responses to participant’s compliance.

7. Ongoing judicial interaction with each drug court participant is essential.

8. Monitoring and evaluation measure the achievement of drug court goals and

gauge effectiveness.

9. Continuing interdisciplinary education promotes effective drug court planning,

implementation, and operations.

10. Forging partnerships among drug court, public agencies, and community-based

organizations generates local support and enhances drug court effectiveness.

Kentucky Drug Court

The current study examines Kentucky Drug Court, which is organized and

managed at the state level, but executed locally. According to the Kentucky Drug Court

mission statement (2008), their mission is the protection of public safety, the reduction of

continued criminal offending, and long term positive lifestyle changes for the drug

addicted offender.

The mission of the Kentucky Drug Court is to protect public safety and reduce the

recidivism rate of drug-addicted offenders through an integrated approach that

involves court supervision, substance abuse treatment services, education,

employment, and personal accountability, resulting in positive and long lasting

life changes. (Kentucky Drug Court, 2008).

Kentucky Drug Court operates on a unified state model, which means that the

state standardized certain elements across all drug courts. Each drug court is provided

oversight by the Administrative Office of the Court and follows the same structured

model for operations, case management, and information management. The first state-

modeled drug court was in Fayette County, Lexington, Kentucky, in 1996. Since then,

the drug court programs grew in number and, as of 2009, through local, state and federal

funds, 115 of the 120 counties in the state are serviced by a drug court. There are 83 drug

courts to cover the 115 counties throughout the state.

Entry into drug court. Defendants enter into drug court through one of four

tracks: diversion, probation, contempt, or family. Regardless of the track, the process for

moving a case to drug court is the same. This process consists of five steps: (1) referral to

drug court, (2) an eligibility review to ensure that both the individual and the case are

appropriate for drug court participation, (3) an addiction assessment to ensure the issues

surrounding the addiction can be supported by drug court, and in some instances, a case-

file review by a drug court panel, (4) voluntarily acceptance of participation in drug

court, and then (5) initial intake upon entrance into the program.

First, a referral occurs after an arrest and can occur at any number of court-related

events, such as at an initial hearing for a drug related crime, during an informal

discussion with a judge, or upon the suggestion of attorneys. Referrals can also come

through post-sentencing channels, such as probation hearings, or directly from a judge

during a court hearing. In any event, a referral to drug court is a formal court docketed

motion set by a judge. Second, the referral is provided to drug court staff members, who

then conduct an eligibility assessment. Eligibility criteria for Kentucky Drug Court are

generally standardized through the state. With few exceptions, certain crimes are

excluded from consideration. Offenders holding crimes of violence, sex offenses, and

drug manufacturing charges are ineligible for participation. Some drug courts support

both misdemeanor and felony level offenders while others accept only felony offenders.

Third, once an individual is determined eligible for drug court, an addiction assessment is

conducted. Kentucky Drug Court utilizes the Addiction Severity Index (ASI). The

information obtained from the addiction assessment is not entered into the management

information system (and therefore not available for analysis in this study). Some drug

courts convene a panel or a “drug court team” to make the acceptance/rejection decision.

According to senior drug court leadership (Neal, 2010), review of the participant referral

by a drug court team ensures the participant level of needs and risk matches the available

drug court resources. A participant’s case file is generated in the management

information system (MIS) when all steps in the review process are favorable to drug court

participation and the individual agrees to participate.

Program requirements. Kentucky’s felony drug court takes a minimum of 18

months to complete while misdemeanor drug court takes 15 months. The drug court is

organized into four phases of decreasing restrictions and supervision, and increased

incentives with progression through the phases. Stabilization of drug use is the goal of

Phase I, which is programmed to be a minimum of 30 days. If participants enter drug

court with documented clean time, such as from a treatment facility or another drug court,

this time may be reduced at the discretion of the drug court judge and staff. Phase II

places emphasis upon education about recovery and beginning a prosocial life. This

phase is designed to last no fewer than 240 days. Phase III is thought of as a self-

motivational phase where the participants experience fewer restrictions than Phase II but

are not completely free of supervision. Each drug court has a fourth phase, which most

term Aftercare. Drug courts vary with regard to the components of this fourth phase, but

in most programs, judicial supervision is minimal and allows participants to begin

making choices for their own purposes rather than for drug court requirements.

The requirements for each phase are generally the same; however the intensity of

each requirement varies by phase and individual participant need. According to the

Kentucky Drug Court participant handbook, the standard requirements are as follows:

 Submit to random drug/alcohol screens

 Attend clinical and educational contacts

 Attend drug court sessions with the judge

 Obtain and/or maintain court approved full-time or full-time equivalent

employment, training, or education

 Obtain and/or maintain court approved housing

 Make arrangements for payments of court obligations

 Make an individual contact with drug court staff

 Show an appropriate understanding of substance abuse treatment and

recovery

 Enroll and attend a self-help program, such as a twelve-step program

 Remain drug-free for at least 30 consecutive days

In addition to the common program requirements, some drug courts may require

the following:

 Obtain a twelve-step sponsor

 Submit to employment, home, school visits by drug court staff

 Attend and comply with certain types of counseling; e.g. domestic

violence, anger management, money management, vocational

rehabilitation,

 Submit to curfews

 Submit to geographic travel restrictions

 Comply with all medical and/or mental health referrals and/or treatment

Drug court programming is intensive, especially during this first phase, and

requires a high level of commitment to continue through all four phases. Phase I

requirements include a minimum of eight drug court contacts per week, three drug

screens, three clinical or educational contact hours, one drug court session with the drug

court judge, and one weekly contact with a case manager. Those requirements are

independent of any additional elements that personalize the drug court programming

(e.g., outpatient groups, mandatory issue-specific groups or twelve-step groups). Keeping

track of individual requirements, ensuring adequate transportation, arranging for daycare

if needed, all the while trying to remain drug and alcohol free and adjusting to a new

lifestyle is daunting. The intensity of the requirements decrease as one moves through

the phases, but even so, not everyone successfully completes the program. So what can

drug court staff do to ensure each participant receives the support they need to increase

the likelihood of graduating? Determining which factors influence a participant’s

completion in drug court is the knowledge gap that the current research addresses.

CHAPTER II

LITERATURE REVIEW

The National Association of Drug Court Professionals (NADCP) concludes that

“…Drug Courts work. Better than jail or prison. Better than probation and treatment

alone. Drug Courts significantly reduce drug use and crime and are more cost-effective

than any other proven criminal justice strategy” (NADCP, 2011). However, declaring that

“drug court works” may be premature. Reviews of available research (Belenko, 2001;

Cissner & Rempel, 2005; Goldkamp, White, & Robinson, 2001; Shaffer et al., 2010;

Turner et al., 2002) point to methodological shortcomings, including a lack of statistical

rigor and generalizability and also inadequate data quality and quantity that interfere with

a firm conclusion about the effectiveness of the drug court program.

Recidivism

The literature on drug court outcomes suggests that drug courts may produce

moderate reductions in criminal activity both during program participation and after

program completion (Belenko, 2001). The Government Accountability Office (GAO)

(2011), in a report that summarized findings from reviews of drug courts that received

federal grants, concluded that drug court participants were rearrested less than

comparison groups by a rate of six to 26 percentage points. Beyond this recent finding,

obtaining an actual measure of drug court’s effect on recidivism is problematic, as there

is no standard definition of recidivism among drug court outcome studies. Recidivism

can be measured as new arrests (Belenko, 2001; Gottfredson & Exum, 2002; Roman,

Townsend & Bhati, 2003; Spohn et al, 2001), new convictions (Spohn et al., 2001), or

even new court appearances (Miethe et al., 2000). Moreover, the timeframe in which the

recidivism is measured also varies. Recidivism can be measured during program

participation (Belenko, 2001; Gottfredson & Exum, 2002), post program completion

(Miethe et al., 2000; Spohn et al., 2001), or measured in terms of time to recidivism

(Spohn et al., 2001).

The Spohn et al. (2001) study presented the most comprehensive look across all

these different methods for measuring recidivism. This study, titled Drug Courts and

Recidivism: The Results of an Evaluation Using Two Comparison Groups and Multiple

Indictors of Recidivism, matched drug court participants (N = 285) to traditionally

processed offenders (N = 194), and offenders assigned to a diversion program (N = 232).

Spohn et al. (2001) included 12 different measures of recidivism, in which drug court

participants consistently performed better than traditionally processed felony drug

offenders. However, when compared to the diversion group, drug court participants

consistently performed worse. For example, 42.1% of the drug court group was rearrested

during the 12 month follow-up period, whereas 60.8% of the traditionally processed

felony offender group and 28.9% of the diversion group experienced rearrest. The drug

court group showed “substantially fewer total arrests than the traditionally adjudicated

offenders” (p. 160), but more than two times the number of total arrests than the

diversion group. However, once level of risk was considered, the differences between

drug court and diversion program disappeared while differences between traditionally

processed offenders and drug court remained. These results led the authors to conclude

that “drug court is an effective intervention” (p. 171). The authors suggested “that the

substance abuse treatment and intensive judicial supervision offered through the drug

court is effective in preventing or delaying a return to substance abuse and criminal

behavior” (p. 171).

Gottfredson and Exum (2002) conducted a quasi-experimental and randomized

design between Baltimore Drug Treatment Court and traditional probation/parole

services. This study followed 235 participants through the first year of their participation

in either the drug court program or traditional case processing. They found a 16%

reduction in rearrest for drug court participants at one year following assignment in the

study. Moreover, the frequency of new arrests and new charges for drug court

participants were also significantly reduced. Drug court participants showed an average

number of rearrests of 0.9 and an average of 1.6 new charges while the traditionally

processed offenders showed an average of 1.3 rearrests and 2.4 new charges. This

suggests that drug court participants are less likely to experience any new arrest than the

traditionally processed offenders and when a new arrest occurs, drug court participants

receive significantly fewer charges.

A different report by Roman, Townsend, and Bhati (2003) provided a “general

estimate of recidivism among a nationally representative sample” (p. 1) of more than

2,000 drug court graduates from 95 drug courts. The authors found that the one-year post

graduation recidivism rate was 16.4% and the two-year post graduation recidivism rate

almost doubled to 27.5%. However, this study did not include any comparison group, so

the meaning of these findings in relation to other correctional programs, such as

probation or parole or other traditionally type of case processing, was left unexamined.

Additionally, this report defined recidivism as “any arrest for a serious offense resulting

in the filing of a charge” (p. 1). By this definition, arrests that do not result in the filing of

a charge are excluded. This definition appears narrowly focused on court case processing

rather than the phenomenon of rearrest as used in the Spohn et. al (2001) and the

Gottfredson and Exum (2002) studies.

Another issue with understanding drug court’s effect on post-program recidivism

is that most other evaluation studies do not include post-program recidivism, and of those

that do, not all apply statistically rigorous methods. Belenko (2001) in his “critical

review of 37 published and unpublished evaluations of drug court” (p. 1) found that only

six of these 37 studies included analyses of post program recidivism. Of those six studies,

four found lower rates of recidivism, but only two of those four studies included test of

statistical significance. One study that applied tests of statistical significance found lower

rates of recidivism for drug court participants, but the difference was not statistically

significant. The remaining study Belenko reviewed found a statistically significant

increase in recidivism rates for drug court participants. This study (Miethe el al., 2000) is

discussed below.

Miethe et al. (2000) reviewed the Las Vegas Drug Court and explored the theory

of reintegrative shaming as it applied to the drug court context. In this study, the authors

gathered arrest records and conducted courtroom observations on both drug court and

non-drug court drug-related offenders. Recidivism for this study is defined as subsequent

court appearances for an offense during 1997 in Clark County, Nevada. They found that

drug court participants’ (N = 301) recidivism rates were 10 percent higher than a control

group’s (N = 301). The drug court group showed a recidivism rate of 26% whereas the

control group showed 16%. The authors suggested that higher recidivism rates for this

particular drug court may, in fact, show that drug court sessions may be stigmatizing

rather than reintegrative in orientation. This conclusion is based on a “wide disparity

between its organizational rhetoric and actual practices” (p 536) noted from three months

worth of courtroom observations and interviews with drug court participants. The

authors cited three main reasons this particular drug court experienced an increased risk

of post-program recidivism: the fact that drug court sessions with the judge served as a

public degradation ceremony, the failure of drug court staff to follow through with

reintegrative efforts post-graduation, and the failure to “increase offenders’

embeddedness in social institutions and interdependencies through repeated contact with

court officials, and in particular the judge…” (p. 538).

Relapse

Studies examining drug court’s effect on relapse are sparse. Relapse generally

refers to a return to drug use, but, similar to the problem with the definition of recidivism,

exactly how to measure a return to drug use is problematic. Cissner and Rempel (2005)

suggest that this may be “primarily due to the inherent difficulties in locating both drug

court participants and comparison group membership for follow-up interviews and

urinalysis testing” (p. 6).

Although infrequent, studies that examine the relationship between drug court

and relapse generally find that drug court produced reductions in drug use. For example,

the Government Accountability Office (2011), in a review of 32 drug court evaluations

and 11 cost-benefit studies from federally funded drug courts, found that “drug-court

program participants were less likely than comparison group members to use drugs” (p.

1) citing 56% of drug court graduates compared to 76% for non-drug court graduates.

This report examined both self-reported drug use and positive urinalysis results.

Another study, Gottfredson et al. (2005), in a continuation of their quasi-

experimental design to study Baltimore City Drug Court, relied on self-reported drug use

as an indicator of relapse. These authors used a variety of tracking mechanisms and

located 72% of study participants for a three-year post completion assessment.

Researchers conducted follow up interviews with 157 study participants. Relying upon

self-reported drug use scales, these authors concluded that drug court participants rated

lower on substance abuse and addiction measures than the traditionally processed

offenders. Specifically, drug court participants used fewer kinds of drugs, scored lower

on both alcohol addiction and drug addiction severity scales, and reported less cocaine

use. However, caution with these findings is warranted as the authors relied upon self-

disclosure of drug use behaviors to gauge relapse rather than drug screens. Using self-

disclosure as a method for determining drug use could be a challenge for researchers as

study participants may not be truthful about their drug use, especially if study participants

are embarrassed about a return to drug use and fear being viewed as failures.

Program Completion Outcomes

Post program completion research suggests that drug court may produce

reductions in recidivism and relapse, but tells us little about the characteristics of those

who complete the program. As researchers try to answer questions about how and if drug

courts work, an understanding of who successfully completes and who fails to

successfully complete the program is necessary. If studies used to determine the impact

of drug court rely upon drug court graduates, then it is necessary to understand if there

are predictable differences between those who graduate drug court and those who do not.

Of the available research specifically on drug court completion outcomes, this

review of literature focused on 14 key studies directly exploring program completion

outcomes, Belenko’s review of 37 drug court evaluations, and other studies that indirectly

explore drug court outcomes. See Table 2 in this literature review and Table A1 in

Appendix A for a details on the 14 key studies.

In general, these studies find that general criminogenic factors may play a role in

program completion outcomes, but, at times, show mixed and sometimes contradictory

findings. For example, age is often found to be related to completion outcomes. One

study found that younger participants experienced increased odds of graduation (Senjo &

Leip, 2001a), while another study found that older participants were more likely to

graduate (Hepburn & Harvey, 2007). So, we may be able to say that certain factors, age

for example, are predictive of completion outcomes, but we don’t know exactly how the

factors impact completion outcomes because the effects vary across studies. Moreover,

some authors found that the ability to predict program outcomes is limited because of

interaction effects among variables. The discrepancies across studies may be due to

differences in populations under study, variable definitions, locations, or even timeframes

of study. The following sections provide an in-depth look at variables commonly

included in drug court program completion studies.

Table 2

Key Studies

Author

Year

Study

Comparisons

Sample Design

Location

Method

Used

Anspach,

Ferguson &

Phillips

2004

Part of larger

evaluation;

Graduates vs.

non-graduates

Not described

(assumed all)

191

Statewide,

Maine

Path

Analysis

Boles et al.

2007

Drug court vs.

standard services

Not described

(assumed all)

684

Sacramento,

Bivariate

Logistic

and

Linear

regression

Butzin, Saum &

Scarpetti

2007

Drug court vs.

standard services

Not described

(assumed all)

116

New Castle

County,

Delaware

Bivariate,

Logistic

regression

Evans, Li &

Hser

2009

Graduates vs.

non-graduates

Multi stage:

Purposeful

geographic

selection; followed

by undescribed type

of random selection

926

Statewide,

California

Bivariate,

Logistic

regression

Hepburn &

Harvey

2007

Track 1 vs.

Track 2

(Mode entry)

All

510

Maricopa

County, AZ

Bivariate,

Logistic

regression

Hickert, Boyle,

& Tollefeson

2009

Graduates vs.

non-graduates

Not described

(assumed all)

288

Salt Lake

City, UT

Bivariate

logistic

regression

(enter,

forward,

backward)

Hiller, Knight

& Simpson

1999

Graduates vs.

non-graduates

Not described

(assumed all)

326

Dallas

County, TX

Bivariate

logistic

regression

(stepwise)

Marlowe et al.

2003

Bi-weekly vs as

needed groups

(status hearings)

All, solicited all new

misdemeanor

participants to

participate in study;

followed by random

assignment to

groups

197

Wilmington,

ANOVA,

chi-

square,

GEE

Mullaney &

Peat

2008

Part of

evaluation;

Graduates vs.

non-graduates

Random sample, 50

cases from each year

241

Undisclosed

“County

Adult Drug

Court”

Percent,

count

Table 2 (continued)

Author

Year

Study

Comparisons

Sample Design

Location

Method

Used

Peters, Haas &

Murrin

1999

Graduates vs.

non-graduates

All

Escambia

County, FL

Cox

regression

(forward

stepwise)

Schiff & Terry

1997

Graduates vs.

non-graduates

All drug court

admissions from

first year of

operation; those who

agreed to participate

in study

418

Broward

County, FL

Bivariate,

Logistic

regression

Sechrest &

Shicor

2008

Graduates vs.

non-graduates

All

102

Riverside

County, CA

Bivariates

Senjo & Leip

2001

Graduates vs.

non-graduates

Systematic random

sampling of

recovery center

records

100

Broward

County, FL

Bivariate,

Logistic

regression

Shaffer et al.

2010

Graduates vs.

non-graduates

All

302

Akron, OH

Bivariate,

Logistic

regression

Participant Characteristics

Sex. Most of the studies reviewed here find no difference between men and

women on program completion outcomes (Evans, Li & Hser, 2009; Hepburn & Harvey,

2007; Marlowe et al., 2003; Mullaney & Peat, 2008; Sechrest & Shicor, 2008; Senjo &

Leip, 2001b). In a review of eight drug court programs, Belenko (2001) observed that in

some studies, women show more positive outcomes while other studies show evidence

that men are more likely to complete. These differences were expressed through the use

of descriptive statistics, leaving no indication if these findings were statistically

significant.

One line of thought about sex differences in completion outcomes is that women

have a more difficult time successfully completing drug court than men because women

are more likely than men to experience gender-specific issues that interfere with their

ability to meet drug court requirements (Neal, 2010). For example, Neal discussed that, in

Kentucky Drug Court, she observed that women are more likely than men to be the

primary caregiver of minor children. Care giving may pose difficulties in making

necessary arrangements to meet the rigors of drug court programming.

Studies find that sex interacts with other variables of study commonly included in

drug court outcome studies. Belenko (2001) observed sex differences with regard to drug

of choice, with males significantly more likely to indicate a preference for cocaine,

alcohol, or marijuana than females. Shaffer et al. (2010), in a study of drugs of choice,

arrived at this same finding. Some studies (Hickert, Boyle & Tollefson, 2009; Hiller,

Knight, & Simpson, 1999) found that harder drugs, such as cocaine, and alcohol, were

negatively related to completion outcomes. This suggests that sex may influence

program completion as an interaction with other variables.

Age. Evidence for the influence of age on program completion outcomes is also

not consistent enough to draw a general conclusion. A number of studies (DeMatteo et

al., 2009; Hepburn & Harvey, 2007; Shaffer et al., 2010) found that age is not related to

completion outcomes. In contrast, other studies found that age is a significant predictor

but within these studies the directionality may differ. For example, Senjo and Leip

(2001a) found that older participants were less likely to graduate. Other studies (Cissner

& Rempel, 2005; Hickert, Boyle & Tollefson, 2009; Young and Belenko, 2002) found

that the likelihood of graduation increased with age. One study, Hepburn and Harvey

(2007), found that increased age was a significant factor in longevity in the program at 90

days retention, only to be dropped from significance at 180 days retention and not related

to successful program completion.

Similar to the situation with sex, some studies found that age interacted with other

drug court and offender characteristics (Shaffer et al., 2010; Hepburn & Harvey, 2007;

Rempel & DeStefano, 2001). Several studies found an interaction between age and

drugs. When comparing alcohol and marijuana users, one study found that those with

alcohol as their drug of choice were more likely to graduate than those listing marijuana,

but marijuana users tended to be younger (Shaffer et al., 2010). In modeling for

completion outcomes, Rempel and DeStefano (2001) found that age moderated the

effects of race on program completion. This study found “that black participants [were]

significantly older and Latino participants significantly younger than average” (p. 106).

In effect, age may hold both direct and indirect influence on program completion

outcomes.

Race. Findings from studies exploring race and program completion are mixed,

but generally find more positive completion outcomes for whites when compared to other

racial groups. A number of studies showed that whites experience higher graduation

rates than non-whites (Belenko, 2001; Hepburn & Harvey, 2007; Schiff & Terry, 1997;

Senjo & Leip, 2001a, 2001b). One of those studies (Senjo & Liep, 2001a) found that race

was the best predictor of program completion. Caution is warranted in assuming a direct

effect between race and successful program completion. Belenko (2001) acknowledged

that race can be a factor that influences program outcomes; however, he suggested that

these differences can be accounted for by other factors such as employment and drug of

choice, or with age as discussed earlier. Belenko (2001) discussed this interaction in the

context of the Roanoke, Virginia, drug court. The Roanoke, Virginia drug court found

that race was related to completion outcomes. Belenko speculated, but did not test, that

this observation could be accounted for by the fact that non-whites also had lower

employment rates than whites. Dannerbeck et al. (2006) found that race and drugs of

choice were related in that African Americans were more likely to report use of cocaine

and cocaine use was associated with a lower likelihood of graduating. Contrary to these

findings, many studies found no significant relationship between race and program

completion outcomes (Evans et al., 2009; Peters, Haas & Murrin, 1999). Although the

majority of studies reviewed here find that whites show more positive completion

outcomes, many studies also find interactions between race and other variables causing a

lack of firm conclusions about the relationship to completion outcomes.

Marital status. The majority of research that explored marital status found that

marital status has little influence on drug court completion outcomes; however, Hepburn

and Harvey (2007) found that marital status was associated with an increased likelihood

of retention at 90 days, although this relationship disappeared at 180 days retention and

was not associated with completion status. Other studies found no relationship between

marital status and drug court completion (Shaffer et al., 2010; Mullany & Peat, 2008;

Senjo & Leip, 2001a; DeMatteo et al., 2009; Hiller, Knight, & Simpson, 1999).

Current criminological theory may provide some support for understanding this

factor. In theory, marriage could either help or hinder a participant’s progress toward

completion. In routine activity theory (Cohen and Felson, 1979), a spouse may function

as a protective factor, preventing deviance from program rules by serving as a capable

guardian or handler. Another way to explain a positive social impact from a significant

other is suggested by theories of informal social controls (Hirschi, 1969; Sampson &

Laub, 1993). A spouse may be a person whom the participant looks up to and admires,

and whose opinion is valued. Participants are therefore constrained from committing acts

that threaten successful program completion due to the bond with their spouse. If this

were the case, then marriage may increase the likelihood of graduation.

However, this same relationship, if the spouse is also criminal or deviant, could

serve negatively to impact drug court outcomes as suggested by Sutherland’s differential

association theory (1937) and Akers’ (1985) social learning theory. In this view, the

spouse may promote definitions of acceptable behavior as those that violate program

rules.

Employment. Employment as a predictor of drug court program completion also

lacks consistency within the literature, and fails to be included as a regular variable of

interest (Senjo & Leip, 2001b). Studies that examined employment found that

employment increases the likelihood of graduation (Belenko, 2001; Hickert et al., 2009;

Shaffer et al., 2010; Mullany & Peat, 2008; Roll, Prendergast, Richardson, Burdon &

Ramirez, 2005). One study (Roll et. al., 2005) found that employment increased

graduation fourteen-fold. Evans, Li, and Hser (2009) found that employment problems at

the time of intake decreased the likelihood of graduation. Similarly, Hiller, Knight, and

Simpson (1999) found that unemployment within 30 days of adjudication to the treatment

program was associated with program dropout. Other studies, however, found that

employment was not related to program completion, but these studies cited limitations

from lack of variance or severely unequal group sizes (Rempel & DeStefano, 2001;

Sechrest & Shicor, 2001). For example, Hickert et al. (2009) discovered that employment

predicted successful completion in bivariate analysis, but failed to predict in multivariate

analysis. The authors suggested that lack of variation, meaning high rate of

unemployment across the entire sample, could account for this finding.

Employment is also found to interact with other variables. Hepburn and Harvey

(2007) found that employment interacted with other individual level characteristics to

produce a greater likelihood of successful completion. Specifically, they found that

employed participants who were married and obtained at least a high school diploma

were more likely to experience positive outcomes than the converse.

Theoretically, the influence of employment on program completion outcomes

could be either positive or negative. The rigors of drug court are intense and require

persistence and dedication to meet all the requirements. Anything that interferes with

meeting these requirements may negatively influence completion outcomes. This

includes fitting drug court around a work schedule or vice versa. Drug court

programming requires frequent drug testing, sessions with drug court staff and sessions

with the judge; all of which may require a flexible work schedule. If an employer is not

accommodating, a participant may have trouble meeting these requirements. On the other

hand, employment may serve as a protective factor. If an employer is supportive of drug

court participation, then holding a job may increase the likelihood of graduation. In this

scenario, social control theories (Hirschi, 1969; Sampson & Laub, 1993) might suggest

that a participant seeks to comply with requirements to ensure the relationship with the

employer is not jeopardized. Another issue to consider is that if employment is a program

requirement, then maintaining employment is a necessary condition of graduation (i.e.,

with participants required to attain or maintain employment). This would result in

naturally higher rates of graduation for those indicating any sort of employment.

Education. Level of educational attainment, although not always included in

drug court studies, demonstrates a positive relationship with drug court performance

(DeMatteo et al., 2009; Shaffer et al., 2010; Mullany & Peat, 2008; Schiff & Terry,

1997). Hickert et al. (2009) found a 15% increase in likelihood of graduation for every

increase in grade level. Even when measured dichotomously (i.e. not graduated,

graduated high school), graduation from high school demonstrated a positive influence on

completion rates (Hepburn & Harvey, 2007; Hickert et al., 2009). Other studies (Senjo

and Leip, 2001a; Hepburn & Harvey, 2007; Hiller et al., 1999), on the other hand, found

that education was not a significant predictor of graduation.

Similar to previously mentioned variables of interest, education shows interaction

effects with other commonly included variables of study. Shaffer et al. (2010) found

statistically significant differences between drugs of choice (crack/cocaine, marijuana,

and alcohol) and education level, and showed that those who completed high school were

more likely to prefer alcohol.

Mental illness. Research on mental illness and drug court completion is less

prevalent than research on other individual level characteristics. The presence of any type

of mental illness is not a commonly included variable of study. In the literature that

includes mental illness as a variable of study, the relationship between mental illness and

program completion is mixed. Cissner and Rempel (2005) found that those without a

diagnosed mental illness are more likely to graduate than those with a dual diagnosis.

Hickert et al. (2009) found that depression is associated with an increased probability of

dropout. Evans et al. (2009) showed that psychiatric conditions, as defined by the

Addiction Severity Index, were more prevalent among dropouts. Hiller et al. (1999)

found that the presence of depression, anxiety, and hostility was associated with program

dropout. To the contrary, other studies found no relationship between mental illness and

outcomes. In a study that used positive drug screens as a measure of drug court

performance, DeMatteo et al. (2009) found that antisocial personality disorder did not

differentiate between types of drug court performers (optimal performers, responders,

non-responders, and the noncompliant). Moreover, Cosden et al. (2006) found that

psychological problems had no significant impact on program completion. More studies

are needed on the relationship between mental illness and program completion.

Family and social supports. The role of family and social supports in drug court

completion outcomes is also relatively unexamined. However, those studies examining

these aspects found certain elements of a participant’s social setting matter for drug court

performance. Hickert et al. (2009) observed that participants whose free time is mostly

spent around their family are significantly less likely to dropout than those who spend

time with their friends or alone. Additionally, they found that caring for children did not

appear to have any influence on likelihood for graduation, but those living alone and

those living in socially isolated neighborhoods are less likely to graduate. Conflicting

with that finding on social isolation, Rempel and DeStefano (2001) found that social

isolation was not significantly related to program outcomes. Remepl and DeStefano

found that general social connectedness, measured in terms of having a stable residence,

being employed, or in school at the time of intake, positively influenced program

retention and completion. This concept, not unlike Sampson and Laub’s theory on

informal social controls (1993), suggests that social interactions can be protective factors.

On the other hand, support for a learning model also exists, which suggests that not all

social interactions are protective in nature. Possessing a deviant peer network was shown

to predict program dropout (Hiller et al., 1999). In this case, although a participant is

socially connected, the antisocial peer influence decreased the likelihood of program

completion.

Problem Behaviors

Drugs. Drug of choice’s impact on program completion is not firmly established

in the literature, but harder drugs such as cocaine and heroin are generally found to

negatively impact program completion (Hickert et al., 2009). Findings for other

substances, such as marijuana and alcohol, vary by study. For example, marijuana or

cocaine have shown a negative effect on graduation outcomes (Belenko, 2001; Hickert et

al., 2009; Hiller et al., 1999; Shaffer et al., 2010), and program dropout was predicted by

cocaine dependence (Belenko, 2001; Hiller et al., 1999; Saum, Scarpitti, & Robbins,

2001; Taxman & Bouffard, 2002). Hiller et al. (1999) found no difference between

dropouts and completers for alcohol, opioids, and marijuana. Shaffer et al. (2010) failed

to find support for a hypothesis that crack/cocaine preference was negatively related to

graduation. Hickert et al. (2009) observed that indicating a stimulant as a drug of choice

resulted in a decreased likelihood of successful program completion.

An issue central in determining factors related to program outcomes is that drug

of choice is also found to be related to other factors including age, risk level, and race. In

their study of the influence of drug of choice on program completion, Shaffer et al.

(2010) found that when comparing alcohol and marijuana users, those with alcohol as

their drug of choice were more likely to graduate than those listing marijuana; however,

this relationship disappeared after controlling for factors such as level of risk. Drug of

choice was shown to interact with factors such as age, level of risk, and employment.

Level of risk was the only factor significantly related to the program outcomes in

multivariate logistic regression. The authors used the Level of Service Inventory-Revised

(LSI-R) as a composite risk score, and found that the majority of alcohol abusers were

identified as low risk, while the majority of marijuana and crack/cocaine users’ risk levels

were higher. However, another study (Dannerbeck et al., 2006) found a significant

relationship between race and drug of choice: African Americans were more likely to be

cocaine users, and cocaine use was associated with non-completion.

Crime. Criminal history is found to be related to both program completion

outcomes and post program outcomes. A common finding, for example, is that the more

extensive the criminal history, the more likely the participant is to drop out of drug court

programming (Evans et al., 2009; Hickert et al., 2009; Mullany & Peat, 2008). Using the

Lifestyle Criminality Screening Form (LCSF), Hiller et al. (1999) demonstrated that

dropout is associated with higher scores on a criminality classification index. Other

studies found that the amount of prior jail time predicted program completion (Caulkins

& Chandler, 2006; Cosden et al., 2006). Cissner and Rempel (2005) observed that those

with no prior criminal record are more likely to graduate than those with a prior criminal

record. Hickert et al. (2009) found that receiving additional criminal charges prior to

intake is associated with dropout. In a similar vein, Evan et al. (2009) observed that,

within 30 days prior to assessment for program participation, dropouts experienced arrest

more frequently than those successfully completing the program. These studies, when

taken together, suggest that those with heavier involvement in the criminal lifestyle are

less likely to complete drug court programming.

Criminal history is also shown to be related to other variables of interest. The

most prevalent interaction with criminal history appears to be drug of choice. Stoops,

Staton, Mateyoke-Scrivner and Leukefeld (2005) found that criminal behavior interacted

with drugs of choice. In this study, methamphetamine users were significantly more

likely to report stealing, selling, or buying items worth more than $50 and less likely to

report weapon charges, violations of probation, or charges of non-support than

participants not indicating methamphetamine use.

Shaffer et al. (2010) also showed that criminal history interacts with drug of

choice. These authors found that participants who indicated crack/cocaine as the most

problematic drug demonstrated, on average, significantly more felony arrests than those

indicating alcohol. Similarly, Senjo and Leip (2001a) found that participants charged

with a cocaine drug crime experienced poorer completion outcomes than those not

charged with a cocaine related offense. In this case, the only charge examined was any

drug charge related to cocaine. This left all other drug-charge types unexamined. Two

studies, Sechrest and Shicor (2008) and Shaffer et al. (2010), found no differences

regarding charge type when charge was defined in terms of drug sale or drug possession.

Note that this definition did not include the drug associated with the charge as the Senjo

and Leip (2001a) study did. This leads to questions about possible interaction effects

between drug of choice and charge type. Shaffer et al. (2010) also explored the impact of

drug of choice and included prior charge information in terms of the mean number of

prior felonies and mean number of prior misdemeanors, and mean number of prior

juvenile arrests. They found that those who preferred cocaine showed significantly more

felony arrests than those who preferred marijuana, and that those who preferred

marijuana had a significantly higher number of juvenile arrests.

When considered as a whole, charge type, criminal history, and drug of choice

appear to be wrapped around each other as a product of risk. It appears that participants

with involvement in the criminal justice system at a young age, those with prior felony

arrests, and those with arrests involving drugs, specifically cocaine or crack, show

increased likelihood of poor program outcomes. This suggests that participants who are

heavily involved in the criminal lifestyle or involved for extended periods of time are less

amenable to drug court programming. Considering that drug court was originally

intended as a low-risk, diversionary program, but has evolved to include higher risk and

post-convictions offenders, these findings are not surprising.

Sanctions. The impact of receiving sanctions on drug court completion outcomes

is relatively unstudied, and of those studies that address the relationship between

sanctions and completion outcomes, the findings are inconsistent and sometimes

contradictory. Belenko (2001), after a review of multiple program evaluation studies,

concluded that not receiving jail sanctions was significantly related to positive

completion outcomes. This may suggest that sending a participant to jail as punishment

was not helpful in producing desired outcomes. On the other hand, Goldkamp et al.

(2001) found that the use of jail sanctions was not associated with either an increase or

decrease in graduation, suggesting that jail sanctions are neither hurtful or helpful toward

program completion. Marlow et al. (2004) and Cissner and Rempel (2005) found that

drug court participants who stay in treatment longer show more positive drug court

completion outcomes.

One study, Anspach et al. (2004), found that sanctions interacted with other

individual and drug court variables including receiving rewards. In this statewide

evaluation of Maine Drug Court, sanctions were identified as an intervening variable

between taking prescription medications and program graduation. Using a path analysis

technique, this study showed that participants “taking prescription medications are more

likely to receive” (p. 23) incarceration as a sanction and participants who received an

incarceration sanction were significantly less likely to graduate. They also found that

sanctions interact with rewards. As the number of rewards increased, the odds of a jail

sanction decreased, which increased the likelihood of graduation (Anspach et al., 2004).

At least theoretically, sanctions may impact completion outcomes in either

direction. Deterrence theory suggests that sanctions or the threat of sanctions should deter

non-compliance. A labeling perspective suggests that a jail sanction may serve to

increase the likelihood of further non-compliance and eventually lead to program

termination. Brown et al. (2010) found that receiving a jail sanction within the first 30

days of treatment predicted treatment dropout, and Anspach (2004) found that receiving

more jail sanctions decreased the likelihood of graduation.

The manner in which sanctions are applied may also be of concern. How, when,

how often, and why sanctions are issued are a matter of program operations and staff

discretion, which may be a reason why these factors are not well documented. Neal

(2010), a senior Kentucky Drug Court administrator, suggested that a judge’s views about

justice and punishment may influence the use of incarceration and the use of alternative

sanctions. If correct, attempting to gauge the impact sanctions have on completion

outcomes will be confounded by factors associated with the drug court judge. This means

that a judge’s view on the role of punishment is intertwined with the use and type of

sanction. This equally applies to drug court staff since they often recommend sanctions to

the judge. To point, some judges may be more likely to apply traditional sanctions while

other judges may be more willing to use non-traditional or creative sanctions. Anecdotal

stories from Kentucky Drug Court staff about creative sanctions include horse stall

cleaning, local animal shelter duty, roadside garbage clean up, repainting drug court

office walls, and community landscaping projects. Sanctioning preferences of drug court

judges and how sanctioning style influences drug court outcomes needs to be studied

further to draw firm conclusions.

Drug Court Variables

An unavoidable problem with drug court research is an inability to generalize

results. The problem stems from different operating environments and variations in the

delivery of services within and between drug courts. Each drug court operates within a

community context, and each community possesses different political environments, key

leadership, and service options. There is no standardized programming for drug courts

across the United States. The lack of standardized operations translates into a wide

variation in programming components, and type and intensity of components among drug

courts (Cissner & Rempel, 2005; Marlow et al., 2004). Therefore, generalizing from one

drug court to the next is problematic.

Status hearings. Drug court status hearings appear to play a role in program

completion, but the relationship may not be as clear as some of the individual level

variables discussed previously. Belenko (2001) and Goldkamp, White, and Robinson

(2001) found that an increased number of court appearances, also called status hearings in

some locations, are associated with an increased probability of graduation. The more one

attends court, the higher the probability of graduation. However, Cissner and Rempel

(2005) suggested that the context of the hearing may also play an important role. They

found that status hearings with positive feedback from the judge increased program

retention significantly more than status hearings with fewer instances of positive

feedback. Marlowe et al. (2006) found that status hearings, when matched with client

risk, have a positive impact on graduation outcomes. This suggests that frequent contact

with the judge may not be necessary for everyone; that those with less risk may perform

well with few contacts.

Rewards and positive comments during status hearings appear to also play a role

in program outcomes. As described previously in this literature review, in drug court, the

judge’s role is expanded to include oversight of positive performance, not just negative.

A judge may issue a positive remark or tangible reward/award for good performance.

Anspach, Ferguson and Phillips (2004) found that rewards, which were often issued in

status hearings, are positively related to the odds of graduation. In this study, participants

identified as high risk on the LSIR were shown to have better graduation rates when

assigned to bi-weekly status hearings with the judge.

Treatment. A drug court’s influence on treatment outcomes and the effect of

treatment on drug court outcomes is worthy of much study. Prior studies on drug court

and treatment found that participation in drug court increases the amount of time a client

remains in drug treatment and also found that the more time a client spends in treatment

the more likely a positive treatment outcome (Cissner & Rempel, 2005; Marlow et al.

2004;). These findings suggested that judicial supervision may increase the likelihood of

a positive treatment outcome. Other related studies found that drug court participants

stay in community-based treatment longer than those in treatment who are under a

traditional probation model (Belenko, 2001; Marlow et al., 2004). A consistent finding

across drug court research was that drug court participants experience in-program relapse

less frequently than traditional probation and parole (Belenko, 2001; Gottfredson &

Exum, 2002; Marlow et al., 2004). Receiving treatment during the first year of drug court

increased the odds of graduation (Goldkamp et al., 2001) and attendance in treatment

within the first 30 days of program participation increased the chance of graduation

(Cissner & Rempel, 2005). These findings suggest that getting into treatment, getting

into treatment early, and staying in treatment may increase the likelihood of graduation.

However, exploring the relationship between treatment and drug court completion

outcomes may not be straightforward because motivation for treatment may also matter.

One study, Evans, Li, and Hser (2009), found that overall low motivation in treatment is

associated with lower likelihood of drug court graduation. Specifically, the authors found

that low levels of desire for help and readiness for treatment are significantly related to

program dropout. Simply sitting time in treatment may not be sufficient to produce a

positive influence on completion outcomes. Rather, participants need motivation and

readiness for treatment.

Time in operation. Another factor to consider when exploring completion

outcomes is the length of time a drug court has been in operation (Belenko, 2001).

Belenko noted that in one drug court, Polk County, Iowa, evaluators observed graduation

rates increase after two years in operation; while in another study on the Orange County,

California, drug court showed a slight decrease in graduation rates after the first two

years of operation. Either of these findings could be the result of a number of factors,

such as drug court staff establishing a working rhythm, increased community support and

community resources, or even changes in judicial leadership for the drug court. As a

result, Belenko (2001) urged caution when examining or evaluating drug courts in early

implementation phases.

Summary of Literature Review

Some authors appear to be comfortable stating that drug courts work at reducing

recidivism and relapse better than traditional correctional sanctions (see Marlow et al.,

2004 for discussion), but exactly who makes it through the program, and how this relates

to post program outcome studies is relatively unknown. The literature on correlates of

drug court completion outcomes is focused in two inter-related areas: individual-level

characteristics and program-level characteristics. Correlates of program termination or

graduation do not appear to differ from traditional correlates of crime. Drug court

completion correlates include sex, age, race, marital status, employment, education, drug

of choice, criminal history, sanctions, and to a lesser extent mental health status, family

and social supports, and certain program-level characteristics, such as length of time a

program is in operation.

In this literature review, a number of methodological and operational factors were

shown to interfere with forming a definitive answer about which factors most influence

completion outcomes. First, interaction effects are noted among factors that the literature

shows are associated with program completion. What works, for whom, and under what

circumstances has yet to be determined (Goldkamp et al., 2001). Second, wide variation

in the operations among drug courts exists making comparisons difficult. Several

prominent authors (see Belenko, 2001; Marlow et al., 2004) take care to note that

findings from one drug court cannot be generalized to all drug courts or used to drawn

inferences to other drug courts. And lastly, the effect of program-level characteristics on

individual-level characteristics, and vice versa, is not well studied.

Review of Existing Outcome Methodology

To determine which factors influence drug court completion outcomes,

researchers use a variety of methods. Outcome measures generally center on completion

status as a binary outcome, which is typically some expression of graduation and

termination. After a review of the literature, the most common analytical strategy used to

explore completion outcomes beyond descriptive statistics is logistic regression. Logistic

regression is the desired statistical method for this type of outcome-based study as it

regresses an independent variable on a binary dependent variable to produce an estimate

of the odds, or “the relative probability of falling into one of two categories” (Menard,

1995).

Other statistical techniques center on testing group differences. These methods

include chi-square (Boles, Young, Moore & DiPirro-Beard, 2007; Butzin, Saum &

Scarpitti, 2002; Evans et al., 2009; Peters, Haas, & Murrin, 1999) which was used to

determine if significant differences exists between attributes in categorical variables.

Cluster analysis (DeMatteo et a., 2009) was used to type subclasses of drug abusers while

t-tests (Evans et al., 2009; Shaffer et al., 2010) and ANOVA (Boles et al, 2007; Shaffer

et al, 2010; Gottfredson and Exum, 2002; Evans et al., 2009) were used to determine if

differences between groups existed. Correlation techniques were also employed in select

studies (Cosden et al., 2006; Hiller et al., 1999) to measure the extent of relationships

with a binary outcome variable. Cosden et al. (2006) and Hiller et al. (1999) used Pearson

correlations techniques on the binary outcome variable for univariate descriptive analysis

to determine suitability for the multivariate model.

Another method for studying completion outcomes included the use of a cluster

analysis technique to develop typologies of drug abusers within drug court (DeMatteo et

al., 2009). These authors classified drug offenders into “types” of drug abusers. The

authors believed that a subgroup of offenders existed who reach abstinence early in

programming and remain abstinent thereafter. They suggested that classifying drug users

into “types” was important for targeting resources in an informed capacity rather than

blanketing all participants with restrictive programming. The authors found support for a

typology of drug abusers. Within their typology of drug abusers, one group, the “optimal

performers,” showed significantly higher graduation rates than the other clusters. The

types were optimal performers, or those with consistently drug-free screens, responders,

or those who started out with positive drug tests but became clean shortly after entering

the program, non responders, or those with persistently positive drug screens with no

improvement, and the noncompliant, meaning those who frequently fail to even complete

the drug screens. The authors suggested that these optimal performers would have also

been successful with less intense and less costly supervision methods.

Anspach et al. (2004) used path analysis to explore completion outcomes. This

evaluation study on Maine’s drug court used a path analysis to “differentiate clients who

successfully completed these drug court programs from those clients who were expelled.”

(p. 23), but only reported the results of the program elements concerning compliance,

rewards, and taking prescriptions (Anspach et al., 2004). Path analysis was chosen,

according to the authors, because it allowed for the control of both cause and effect

variables, as well as intervening and mediating variables.

Studies using the logistic regression method generally begin the analysis with

bivariate correlations to determine which variables perform well enough to contribute to

model performance. Hosmer and Lemeshow (2010) argue that this is an acceptable

technique for this type of modeling effort involving a binary outcome as removing

unproductive variables reduces noise within the model and therefore increases the overall

model accuracy and stability. Across the studies outlined in this paper that employed a

logistic regression method, most models consisted of traditional demographic variables:

age, sex, race, education level and employment status. In these studies, variable

exclusion was based on a failure of bivariate analyses to demonstrate statistical

significance, with the logic that non-statistically significant variables do not enhance the

ability to predict. Hosmer and Lemeshow (2010) discussed a growing trend to include all

“scientifically relevant variables” into the model regardless of the observed relationship

with the outcome variable. The reasoning behind that practice is to capture confounding

effects of the variables of interest. The authors reject this practice by arguing that over

fitted and numerically unstable models may result.

Caveats about drug court research in general were provided in Cissner and

Rempel (2005), Goldkamp et al. (2001), Marlow et al. (2004), and Turner et al. (2002),

and are discussed briefly in the literature review. The main concerns center on a lack of

proper study design and the insufficient use of statistics. Moreover, concerns about the

reliability of data were raised. Factors such as inconsistent record keeping and changes to

information systems contribute to data quality issues. Some drug courts do not maintain

or have not maintained computerized information systems, which leads to concerns about

the availability of data. All these issues considered, the core issue regarding drug court

studies is generalizability.

The concerns with generalizability may be well-founded if the purpose of a study

is to infer about drug court or drug court participants as a whole. However, the purpose of

many of these studies was not to infer the findings to other drug courts but to analyze a

particular phenomenon for a particular drug court. Even if a study design used rigorous

statistical methods, findings from one drug court may not be generalizable to another

drug court, as drug courts are organized and operated at the local level where operating

environments and access to resources differ across drug courts. These differences lead to

distinct program requirements for individual drug courts. In as much, this current study

seeks to identify factors that are predictive of drug court completion outcomes, specific to

the State of Kentucky Drug Court.

Kentucky

According to the U.S. Census Bureau (2010), Kentucky is a southern state, but is

bordered by states classified as part of the Midwest (Indiana, Missouri, Illinois) and states

considered part of the eastern region (Virginia, West Virginia) with the eastern half of

Kentucky situated in the Appalachian Mountain region. Kentucky is known for the blue

grass that grows through most of the state, its horse racing, most notably for the

Kentucky Derby at Churchill Downs, and for its bourbon distilleries. According to the

2010 U.S. Census (U.S. Census, 2010), Kentucky’s total population is 4,339,367 which is

only 7.4% growth from the 2000 census. Kentucky is 87.8% white, shows a high school

completion rate (for those 25 and older) of 81%, and 17.7% of the population lives below

the poverty line. All these indicators show that Kentucky is less diverse, less educated,

and has a higher rate of poverty than the national average. According to the U.S.

Department of Agriculture (2012), Kentucky is approximately 42% rural and 58% urban.

Kentucky ranks high on a number of health and social related concerns. This state

is ranked 5th in the nation for the percent of adults considered obese and is similarly

ranked for levels of physical inactivity (Center for Disease Control, 2012). The data show

that 25% of Kentucky residents are smokers. This is more than six and a half percentage

points higher than the national current smoker percentage according to the Centers for

Disease Control (CDC) (2012). The unemployment rate for Kentucky is 9.1% as of

December 2011 (WorkforceKentucky.gov, 2012) which shows that this is higher than the

national unemployment rate of 8.5% according to the Bureau of Labor Statistics (2012).

All told, these data portray Kentucky as a relatively unhealthy state.

Kentucky has a sizeable drug problem, most acutely in the eastern portion of the

state. The eastern section of Kentucky is an area recognized by the National Drug

Intelligence Center (NDIC) as a High Intensity Drug Trafficking Area (HIDTA).

Specifically, 27 of Kentucky’s eastern counties are part of the Appalachia HIDTA. NDIC

originally identified this high traffic area in response to the cultivation and distribution of

that region’s marijuana cash crop, but now includes prescription drugs. According to the

NDIC assessment (2007), “the Appalachia region consistently sustains high levels of

outdoor cannabis cultivation because of its favorable climate and rich soil” (p. 3). NDIC

identified a high poverty rate of the region as a contributing factor to the high intensity

designation. Marijuana production is a means of supplemental income. This report also

indicated that in some communities, “cultivation is often a multigenerational trade, since

young family members are introduced to the trade by other members who have produced

the drug for many years” (p. 3).

Moreover, a report from the Kentucky State Epidemiological Outcomes

Workgroup, released December of 2011, stated that the medical and psychosocial burden

of illicit use of prescription drugs is “particularly acute in Kentucky” citing increasing

rates of illicit use of opiate based drugs and increased prescription rates for hydrocodone,

and oxycodone. This report also included staggering statistics that Kentucky experienced

a 260% increase in fatal drug overdoses from 1999 to 2008, that Kentucky experienced a

900% increase in treatment admission for opiate based substances, and that fatal drug

overdoses surpassed suicide mortality in 2005. All told, Kentucky has a significant drug

problem.

Current Study

This current study seeks to identify factors related to successful completion in the

Kentucky Drug Court program for cases closed between January 1, 2007, and August 24,

2010, using data available within the management information system. This dissertation

used logistic regression to explore and predict completion outcomes. Diagnostic analyses

were run prior to the multivariate model to determine which independent variables are

meaningfully related to the dependent variable and to identify any independent variables

that may be significantly related to each other.

Prior research described in the literature review shows that completion outcomes

are affected by individual-level characteristics. These factors include sex (Belenko, 2001;

Hickert et al., 2009; Hiller et al., 1999), age (Cissner & Rempel, 2005; DeMatteo et al.,

2009; Hepburn & Harvey, 2007; Hickert et al., 2009; Shaffer et al., 2010; Senjo & Leip,

2001a; Young & Belenko, 2002), race (Belenko, 2001; Dannerbeck et al., 2006; Hepburn

& Harvey, 2007; Rempel et al., 2003; Schiff & Terry, 1997; Senjo & Leip, 2001a; Senjo

& Liep, 2001b), marital status in terms of treatment retention (Hepburn & Harvey, 2007),

employment (Belenko, 2001; Hickert et al., 2009; Shaffer et al., 2010; Mullany & Peat,

2008; Roll et al., 2005), education (DeMatteo et al., 2009; Hepburn & Harvey, 2007;

Hickert et al., 2009; Shaffer et al., 2010; Mullany & Peat, 2008; Schiff and Terry, 1997),

drug of choice (Belenko, 2001; Hickert et al., 2009; Hiller et al., 1999; Shaffer et al.,

2010; Miller &Shutt, 2001), criminal history (Evans et al., 2009; Hickert et al., 2009;

Mullany & Peat, 2008), mental illness (Hickert et al., 2009; Hiller et al., 1999), familial

status and social supports (Hickert et al., 2009; Hiller et al., 1999; Rempel & DeStefano,

2001). Research also shows that drug court program delivery elements such as quantity of

status hearings (Goldkamp et al., 2001; Marlowe et al., 2006) quality of status hearings

(Cissner & Rempel, 2005), sanctions (Belenko, 2001), treatment participation (Cissner &

Rempel, 2005; Marlow et al., 2004), and how long a drug court has been in operation

(Belenko, 2001) can influence completion outcomes. The extent to which these factors

individually contribute to program completion is difficult to ascertain because of the

interaction between all elements (Belenko, 2001; Cosden et al., 2006; Hickert et al.,

2009). The inconsistencies among findings within the literature may be driven by the

fact that each drug court operates in a different environment, offers and/or requires

different services which are provided by different providers and such services are

managed and deliver services differently. As a result, most drug courts are not directly

comparable to each other. One drug court’s set of best predictors may not be the same as

another’s.

Hypotheses

The current research seeks to identify factors that influence program completion

outcomes for the State of Kentucky’s drug court program. Three hypotheses are tested in

this study.

Hypothesis one. Characteristics about a participant predict completion outcomes.

Specifically, sex, age, race, marital status and education level can be used to predict

program completion outcomes.

Consistent with prior studies and theory as described in the literature review,

participants who possess certain characteristics indicative of distractions from or barriers

to drug court compliance are expected to show a decreased likelihood of graduating.

Females, older participants, non-whites, participants who are married, and those without

a high school diploma are expected to show less favorable outcomes. The assumption of

drug court leadership (Neal, 2010) is of interest in this study. Neal speculated that

females are more likely to be primary caregivers of minor children (not tested in this

study) and that primary care giving for minor children adds to the difficulty of

completing drug court programming; therefore females are expected to show less

favorable outcomes than males. The logic used for this increased-responsibilities

argument can be applied to married participants; that these characteristics indicate

increased responsibilities that lie outside the drug court influence and therefore increase

the level of difficulty in meeting program requirements. For example, being married may

carry the responsibility to care for children and a spouse in addition to the work needed to

meet drug court requirements. Participants without a high school diploma are expected to

show less favorable outcomes, as the less educated are likely to experience difficulty

obtaining and then maintaining quality employment; thereby also experiencing

challenges with access to resources to support program requirements or comply with

money-based program requirements (e.g., vehicle or other transportation, money for fees,

fine, or other payment court obligations, or money to maintain stable housing).

Hypothesis two. Problem behaviors leading into program participation and

punishments while in drug court (sanctions) predict program outcomes. Specifically,

drugs of choice, crime types, and in-program punishments predict program completion

outcomes.

Drugs of choice are not well researched in the literature, although harder drugs

have been found to result in less favorable outcomes. As such, participants indicating

drugs of choice including cocaine and crack, as well as opiates such as heroin, are

expected to show decreased odds of graduation. The number and type of crimes a

participant holds is also of interest for this study. A greater number of charges and certain

charge types may indicate a deeper level of criminal lifestyle and therefore a riskier

participant. Participants carrying multiple charges and charge types that suggest a deeper

level of criminal lifestyle, such as drug manufacturing and crimes against a person, are

expected to show a decreased likelihood of graduation.

Sanctions are also of interest. The type of sanctions one receives while in drug

court is not well documented in the literature, but some research suggests that not

receiving jail sanctions may be related to an increased likelihood of graduation (see

Belenko, 2001). However, receiving a sanction indicates some sort of issue with program

compliance. The noncompliance itself, rather than type of sanction, may drive program

termination leaving any influence sanction has on completion outcomes driven by the

non-compliance rather than sanction type. In this scenario, no difference in completion

outcomes is expected for any of the sanction types. However, given that severe sanctions

may be documented more reliably than other sanction types, participants who received a

sanction involving incarceration are expected to show decreased odds of graduating.

Moreover, the number of sanctions a participant receives may be indicative of a level of

program compliance or may represent a measure of willingness of drug court staff to

issue punishments for non-compliance. In either case, a greater number of sanctions is

expected to result in a lower likelihood of successfully completing the drug court

program.

Hypothesis three. Characteristics about the drug court program predict

completion outcomes. Neglected in the literature is how the drug court itself may

influence completion outcomes. In this study, only two elements of the drug court

programming were available for study: the track through which a participant entered drug

court and how long the drug court was in operation at the time of entry. Participants

entering drug court through the diversion track are expected to show a greater likelihood

of graduating than those on the probation track. This is expected as those on the diversion

track have more to lose from failing to complete drug court. Those on diversion track risk

imposition of both the conviction and the jail or prison sentence while those on the

probation track are already convicted and risk only the imposition of the sentence. The

relationship between the length of time a drug court is in operation and completion

outcomes is not well documented in the literature. Belenko (2001) only briefly discusses

it by providing anecdotal evidence that time in operation may influence outcomes in

either direction. The findings here will be a unique and important addition to the

literature.

CHAPTER III

METHODS

The purpose of this study is to identify factors related to drug court program

completion outcomes using data obtained from Kentucky Drug Court. The analytic

strategy uses a cross-sectional study design with logistic regression for the statistical

method. The variables of study are limited to those collected by drug court staff and

contained within the computerized case file and information management system. SPSS

(v.19) was used as the statistical tool for this research.

Data Collection

The data collection for this study occurred as part of routine program management

for the Commonwealth of Kentucky’s Drug Court. Drug court staff enters participant

data into a custom-developed Management Information System (MIS) as part of case

management practices. Participant records are updated on a regular basis by drug court

staff with information relating to progress in the program. The data used for this study

were pulled from MIS upon the researcher’s request. After discussion with drug court

staff regarding reliability of the data, drug court staff suggested that data from 2007 and

later be used. Staff agreed to pull from January 1, 2007, to the date of the request (August

24, 2010), which resulted in access to records that were closed during a three year and

nine month time frame. Drug court staff suggested the 2007 time frame as that is the year

when most drug court staff was trained on using the MIS. Data prior to 2007 are

considered more likely to be incomplete and unreliable for research purposes. This data

pull resulted in a total of 3,621 unique participant records from 83 drug courts within the

Commonwealth of Kentucky.

Records were included in the study if the record showed:

 the participant was 18 years of age or older (excludes juveniles),

 the participant participated in the felony, adult drug court (excludes

juvenile, family and misdemeanor drug court participants),

 the participant successfully passed through the assessment and screening

process (excludes records of those not eligible or those that did not begin

the program),

 the record has a date of entry documented (excludes cases where date of

entry could not be determined), and

 the record resulted in a date closed between January 1, 2007, and August

24, 2010.

The data were provided to this researcher in multiple Excel workbooks. Kentucky Drug

Court staff provided a spreadsheet for participant level information and one spreadsheet

for each of the major variables types where a one-to-many relationship exists. A one-to-

many relationship occurs when one participant record contains multiple entries. In this

study, a participant may have more than one drug of choice, charge, and sanction records

in the database; therefore, Kentucky Drug Court staff pulled these variables separately.

The data were imported into SPSS (v.19) files, restructured on participant ID, and then

merged on participant ID to form a flat file. Variables for completion status, race, sex,

marital status, education, track, drug of choice, charge, and sanctions were collapsed to

accommodate the logistic regression analysis. The variables of interest were explored for

model selection, and those variables that demonstrated a meaningful relationship to the

dependent variable were included in the logistic regression.

Discussed in the literature review were other individual level characteristics,

specifically employment, presence of mental illness and presence of family and social

supports. Although this data may be collected by drug court staff during the eligibility

review, this data is not recorded in a manner that allows for inclusion in this study.

Employment information is collected at the time of entry, however, data is not complete

enough to determine if the participant was employed at the time of entry or if the

employment is gained while in the program. The presence of mental illness and presence

of family and social supports is captured during the assessment process; however this

information is not entered into the MIS.

Dependent Variable

Graduation status is the primary focus of the study; therefore this variable was

recoded into a binary variable containing the values of not graduated and graduated (0,1).

This coding scheme showed that 29.8% of the total population of study graduated.

Completion status originally contained three values: terminated, administrative discharge,

and successful completion. Administrative discharge accounted for only 5.5% percent of

the records and was included in the “did not graduate” category. Administrative

discharge may occur when a participant is dismissed from the program, but not through a

non-compliance issue. This program completion status can be used when a participant

becomes injured or ill and unable to meet requirements. Participants discharged through

this outlet are eligible for drug court in the future whereas participants who are

terminated are ineligible for future participation.

Independent Variables

Variables included in this dataset fall into one of three domains and follow the

organization of the hypotheses. First are the data that inform upon the individual.

Demographic variables including sex, age, race, marital status, and education level are

used to explore participant characteristics. Second are the variables that demonstrate the

problem behaviors leading into drug court participation. Variables for problem behaviors

include drugs of choice, the charges an individual carries, and sanctions. Thirdly,

program variables, meaning those variables that reflect characteristics about the drug

court program and do not change given an individual’s choices or performance while in

the program include the track through which a participant entered the program, and how

long the drug court was in operation at the time the individual entered drug court.

Since the data for this study originated from the Kentucky Drug Court’s MIS,

some data required recoding for research purposes. In most cases, the data provide

showed far more categorical “types” of phenomenon than usable for study. See Appendix

B for recoding and classification schema. In the sections that follow are descriptions and

discussions of each of the variables included in the study.

Participant characteristics. Table 3, included below, shows the details of each

participant characteristic studied. Participant characteristics in this study are limited to

those contained within Kentucky Drug Courts’ MIS. Sex, age, race, marital status and

education are included.

Sex. The sex variable is limited to the categories of male and female. The original

sex variable allowed for unknown (n= 2) and other (n=28) categories. Kentucky Drug

Court staff indicated that unknown or other can be and are used in cases where a

participant verbally indicates transgender/transsexual for their sex. Kentucky Drug Court

staff confirmed they have had transgender/transsexual participants. Given the infrequent

occurrence of “other” and only two observations of “unknown,” these cases were deleted

to protect from unintentional identification of an individual participant. In the study

population, 62.1% of the population is male and 37.9% female.

Age. The age variable is continuous and represents the age of the participant on

the day they entered drug court. This variable was computed by subtracting date of birth

from the date of entrance and is documented in years. The median age for this drug court

population is 29 with the youngest at 18 years of age and the oldest at 69 years of age.

More than half of the population is less than 30 years of age. Logistic regression makes

no assumption about distribution of the variable (Mertler & Vanatta, 2005; Pampel,

2000); therefore no recoding or transformations of the age variable was necessary.

Race. Kentucky Drug court allows for 10 distinct racial classes. Nearly 97% of

the Kentucky Drug Court population under study indicated a race of either white or

black/African American, leaving approximately three percent spread across the remaining

eight classes. Leaving race in the original categories will result in violations of the cell

count rules for bivariate and multivariate analyses such as chi-square and logistic

regression, which require no fewer than five counts per cell (Mertler & Vanatta, 2005).

Appropriate binning to best capture the effect of race on program completion becomes

complicated in this situation. To use only white and non-white may miss differences

experienced by races included in the non-white category. However, given so few

observations in the data, a dichotomous (white, non-white) variable was the solution

selected. Schiff and Terry (1997) also collapsed race in this manner with similar

justification; “because of the lack of sufficient numbers in each category” (footnote, p.

303). A white/non-white classification method was also used in other drug court

outcomes studies, namely Butzin et al. (2002), Goldkamp et al., (2001), Hepburn and

Harvey (2007), Hickert et al. (2009), Peters et al. (1999), Schiff and Terry (1997), and

Senjo and Leip (2001b). Refer to Table B1 for details on the coding of race.

Marital status. Kentucky Drug Court documents marital status as divorced,

married, never married, other, separated, single, or widowed. To ensure cell frequencies

were sufficient for statistical analyses, the categories for marital status were collapsed

into married or not married.

Marital status was changed to system missing in situations where marital status

was listed as “other.” This was done as no operational definition of “other” could be

identified by Kentucky Drug Court staff. However, one staff member (Hardin County

Drug Court, 2010) suggested that this other category may include participants who were

engaged, homosexual couples in significant relationships but unable to legally marry, or

those who were still legally married but living apart. This affected less than one percent

of the population of study. This classification method showed that more than 75.9% of

the study population showed not married and 19.2% indicated they were married. Data

on marital status were missing for 4.9% of the records. Refer to Table B2 for details on

the coding of marital status for this study.

Education. The education variable represents the highest level of education

achieved on the day of intake. According to the 2010 U.S. Census, Kentucky showed a

high school graduation rate of slightly more than 74% for those aged 25 and older. This is

more than six percentage points lower than the national rate of 80%. Kentucky Drug

Court documents 20 distinct educational values and places emphasis on documenting the

level of high school last completed if a participant has not graduated high school.

Kentucky Drug Court also documents high school equivalencies such as a GED or

alternative school completion. This level of detail is remarkable and is worth exploring;

however, a broad range of groups such as this causes cell values to drop below the five

observation threshold. Education is grouped into two categories of less than high school

and at least high school. This method is consistent with the majority of drug court studies

reviewed here that included an education variable. In the current data, 29.6% indicate less

than high school while 62.5% show at least a high school diploma or equivalent. Data on

education level was missing for 7.9% of the records. Refer to Table B3 for details on the

coding of education for this study.

Table 3

Participant Characteristics

Percent Participants

Sex

Male 62.1

Female 37.9

Age

Mean 30.66

Median 29

Mode 22

Min/Max 18/69

Race

White 85.0

Not White 15.0

Marital Status

Married 19.2

Not married 75.9

Missing 4.9

Education Level

Less than high school 29.6

High School 62.5

Missing 7.9

Problem behaviors. The problem behaviors leading into drug court participation

include involvement with drugs and crime. This study also examines the number and

types of sanctions one receives while in drug court. A participant must have committed

some type of crime and indicate a problem with some type of drug. The number and

types of sanctions are indicative of some problem behavior while in the drug court

program.

Drugs. The MIS for Kentucky Drug Court allows for more than one drug of

choice to be entered. Knowing the number of drugs an individual finds problematic

allows us to determine if trouble with more than one drug, also termed polysubstance

abuse, results in worse program outcomes. For this study, the number of drugs of choice

was counted.

Discussion surrounding the number of drugs of choice is sparse within drug court

literature, perhaps because a drug of choice is operationalized in most studies as the

“one” drug most problematic. Of the key studies on drug court completion outcomes,

none included a count of multiple drugs of abuse. However, Brown and colleagues

(2010) in their study on the impact of jail sanctions on treatment outcomes studied

polysubstance abuse (abusing more than one drug). Using the Cox proportional hazards

model to determine factors that predict time to treatment failure, they found that

polysubstance abuse was a statistically significant predictor. This finding suggested that

abuse of more than one drug is a treatment hazard, meaning it increases the risk of

treatment failure. Since prior research suggests that treatment can influence drug court

completion outcomes, the number of drugs listed as a drug of choice is included for

study.

Kentucky Drug Court allows the specific drug of choice to be selected from a

drop down list and manually entered into the MIS. The original data file contained 47

unique entries for drug of choice. Several of these were misspellings while others were

the same drug but called something slightly different. All drugs of choice were collapsed

to seven categories to ensure the case to variable ratio remains appropriate for a logistic

regression. Categories were selected by this researcher and drug court staff after

reviewing the drugs appearing in the data file, their frequencies, and drug court

information needs. For example, although both cocaine/crack and methamphetamine

could fit into a category for “stimulants,” Kentucky Drug Court staff suggested that

cocaine/crack and methamphetamine should be in distinct categories.

The final drug of choice variable consists of seven dummy coded variables (0,1)

with zero indicating the absence of and one indicating the presence of the particular drug

type. The final groupings of drug types are methamphetamine and other stimulants,

cocaine and crack, marijuana, opiates, alcohol, sedatives/downers and “other” drugs.

“Other” drugs included inhalants, PCP, and LSD. Refer to Table 4 for percent of

participants indicating each drug of choice and Appendix B, Table B4 for detailed coding

information.

A conceptual issue with drug of choice as a variable in this study is that an actual

measure of a participant’s “drug of choice” is not available. Drug of choice, as a concept,

suggests a single drug that is most problematic or most preferred. However, in Kentucky

Drug Court, a participant may have more than one drug of choice documented. Moreover,

the MIS places the drugs of choice in alphabetical order rather than in order of relative

importance. There is no way to determine which drug was most problematic or preferred.

Therefore, all drugs of choice listed for a participant are assumed to be equally

problematic. The most common drug of choice listed is opiates, with 43.4% of all

participants indicating some type of an opiate. This was followed closely by marijuana

with 42.8% of participants.

Table 4

Drugs of Choice

Percent Participants

Opiate 43.4

Marijuana 42.8

Alcohol 32.0

Cocaine/Crack 27.8

Sedative/Downer 24.2

Meth or other stimulant 18.3

Other 3.0

Note: Values total more than 100%; a participant may indicate more than one drug of choice.

Crime. The type of crime a participant was convicted/stands accused of at the

time of entry into drug court is included as a variable in this study. The charge a

participant carries may be important to understand as some research shows that charge

has a relationship to program completion outcomes (Evans et al., 2009; Mullany & Peat,

2008; Peters et al., 1999; Senjo & Leip, 2001a). However, operational definitions of

charge or charge type differs between studies. For example, Senjo and Leip (2001a)

found that participants charged with a cocaine related crime show poorer outcomes than

those not charged with a cocaine related charge. In this case, the only charge examined

was any cocaine related charges, leaving all other non-cocaine charges and all other non-

drug charges unexamined. Sechrest and Shicor (2008) and Shaffer et al. (2010) find no

differences with regard to charge type at time of admission when charge was defined in

terms of any drug sale or drug possession. Only one study reviewed here provided an

examination of charge types beyond drug or drug related crimes. Anspach (2004), in an

evaluation of Maine’s adult drug treatment court, reviewed charge type by crimes against

a person, property related, drug related, motor vehicle related, and probation violation

related. However, these analyses provided only percent discharged from drug court, did

not include tests of statistical significance, and did not include classification information

on exactly what charges were included in the categories.

Kentucky Drug Court provided the data on participant charges, which resulted in

480 unique offenses. Given the narrow range of charge types reviewed in prior research

and the lack of analysis when a broader range of charges was included, using the prior

research to frame charge classification in this study is problematic. The classification

method used in this study centers on the grouping of similar offenses with regard to

qualities of the harm caused. For example, drug possession differs from the sale,

manufacturing, or trafficking of a drug in that the quantity of drug on the person is

minimal (i.e., for personal consumption). Selling, manufacturing, or trafficking a drug

implies either a quantity beyond sufficient for individual use with some evidence that the

use is intended for others. Additionally, selling, manufacturing, or trafficking of drugs

suggests an increased level of criminal involvement. Increased level of criminal

involvement is a key factor in assessing level of risk, which is also shown to impact drug

court outcomes (Marlowe et al., 2003; Marlowe et al., 2006; Spohn et al., 2001).

The final classification method resulted in eight dummy coded variables (0,1)

with zero indicating the absence of and one indicating the presence of the particular

charge type. The final categories are drug sale/trafficking/manufacturing, drug or drug

paraphernalia possession, vehicle or traffic related (excludes driving under the influence),

charges relating to the administration of justice, charges relating to public order, crimes

against a person, property crimes, and any charge of driving under the influence (DUI).

Possession charges were the most common, indicated by slightly more than 37.4% of

participants. This was closely followed by property crimes indicated by 27.5% of

participants. Table 5 shows a breakout of charge type by percent participants. See

Appendix B, Tables B5-B12 for details on the coding of criminal charges.

Table 5

Criminal Charges

Percent Participants

Drug Possession 37.4

Property 27.5

Administration of Justice 15.4

Drug Sale/Traffic/Manufacturing 14.4

Person 6.8

DUI 5.9

Public Order 5.7

Vehicle/Traffic 4.7

Note: Values total more than 100%; a participant may indicate more than one drug of choice.

Sanctions. Table 6 shows the types of sanctions received and the percent of

participants who received the sanction. Sanctions in drug court refer to the official

responses to infractions of program rules. In other words, sanctions represent

punishments for non-compliance. Sanctions are issued by the drug court judge, most

often during drug court proceedings in the courtroom. Sanctions can also originate by

suggestion from drug court staff. In Kentucky Drug Court, there is no standard guideline

for issuing sanctions, leaving the potential for each drug court and each case specialist to

hold different sanctioning practices. This may affect both the frequency of sanctioning

and the type of sanctioning.

In addition to different sanction practices between and within drug courts, drug

court staff indicated that this data point may not be reliable because of differing data

entry practices. Informal discussions with multiple drug court staff members in different

drug courts revealed that some staff do not enter sanctions into the database, while others

selectively enter sanctions. When discussing the reliability of the data, several drug court

staff indicated that data entry of sanctions and rewards, although important to drug court

progress, are not a priority. For example, one drug court staff indicated that they only

enter major sanctions or sanctions when another agency, such as a treatment agency, a

jail, or some other organization supporting community service sanction, is involved. If

this is a common practice among drug courts, minor sanctions or those under the

complete control of drug court, such as increased homework, changes in curfew,

increased drug testing, or phase demotions are underrepresented in the data. Further

study on program non-compliance, formal and informal responses to non-compliance,

and documentation of these responses is warranted, although outside the scope of this

research.

Even though unreliable as a measure of all sanctions, the variables collected for

this study may be good indicators of the more intensive sanctions such as jail and

additional treatment. Since sanctions are found to influence drug court completion

outcomes (Anspach, 2004; Marlowe et al., 2006) and associated treatment completion

outcomes (Brown et al., 2010), the number of sanctions will be examined to determine if

they impact completion outcomes. For the current study, this variable is continuous and

represents the total count of sanctions received while in the drug court program. The

mean number of sanctions for the population of study is 3.7, the median is three, and the

most frequently observed count is zero.

Kentucky Drug Court documents 16 unique sanction types. To keep the case to

variable ratio within acceptable limits for bivariate and logistic regression analyses, this

variable was collapsed into six dummy coded variables (0,1) with zero indicating the

absence of and one indicating the presence of the particular sanction type. The final

categories are incarceration/detention, treatment or treatment related, community service,

phase demotion or suspension, increase in program elements, and other. See Table B13

for details on coding of sanctions for this study.

Incarceration or detention includes any type of sanction that involves confinement

to a correctional facility or house. The incarceration or detention sanction type does not

include mandated in-patient or residential treatment, as those were placed in the treatment

or treatment related category. A community service sanction is any type of sanction

where the participant was required to perform some type of work within the community,

which may include activities such as cleaning the drug court office, working at the local

animal shelter, or road cleanup crew. Phase demotion or suspension includes anytime a

participant is either demoted in their program phase status, for example, from phase II

down to phase I, or when a participant is suspended from program participation.

Suspension can include situations where an individual is suspended in his or her current

phase status rather than advanced to the next phase. An increase in program elements can

include actions such as additional assignments, earlier curfew, or an increase in the

frequency of drug tests. According to drug court staff, sanctions included in the “other”

category may include sanctions such as increased number of drug court groups, an essay

for the judge, or cleaning of the drug court office area. However, these kinds of sanctions

may also be included in the increase in program elements or community service

categories. This issue suggests that the sanction categories may not be mutually

exclusive. Moreover, usage of the “other” category may vary widely across drug courts

and deserves further attention.

The most frequently recorded sanction listed is for incarceration or detention, with

70.4% of participants having at least one such sanction documented. The second most

commonly recorded sanction is for community service with 27.3% of participants. The

large gap from the most prevalent and the second most prevalent sanction could suggest

that incarceration/detention sanctions are better documented than the other sanctions

types.

Table 6

Sanction Types

Percent Participants

Incarceration/Detention 70.4

Community Service 27.3

Other Sanction 19.7

Treatment/Treatment related 18.4

Increase Program Elements 13.9

Suspension/Demotion 10.9

Note: Values total more than 100%; a participant may indicate more than one drug of choice.

Drug court variables. Drug court variables used in this study include the track

through which the participant entered drug court, how long a program was in operation

when the participant entered, and the number of months the participant spent in the

program.

Track. “Track” refers to the route through which an individual enters drug court.

Kentucky Drug Court documents track in one of four categories: probation, diversion,

contempt, or family. A participant enters through the probation track generally when

other probation efforts have failed or when the judge sends an individual directly to drug

court rather than traditional probation. Non-compliance while on traditional probation,

frequently caused by multiple positive drug screens, may result in a referral to drug court

as a last chance effort to remain out of prison. In this case, the judge, in consultation with

defense and prosecuting attorney, may offer drug court as a one-time alternative to

prison. In both of these scenarios, participants enter drug court as a form of supervision

more intense than traditional probation.

In contrast to the probation track, participants may also enter drug court through

the diversion track, meaning the charge(s) they stand accused of will be dropped upon

successful completion of the program. If a participant successfully completes drug court

programming through the diversion track, then a conviction is avoided.

Two other tracks are possible for Kentucky Drug Court: contempt and family.

Entry into drug court through the contempt track may occur when offered by a judge in

relation to a charge for contempt of court. The family track, as a route of entry into drug

court, may occur when a participant in a family court action experiences issues with drug

or drug related charges and the family court and criminal court judge agrees that drug

court is an appropriate option. Both of these alternative tracks are infrequently used for

felony adult drug court, and comprise less than one percent of the total population under

study. Since these routes of entry are infrequent and to ensure that cell values are

appropriate for analysis, these values were coded as system missing.

Track type will be used in this analysis to determine if the route through which

one enters drug court bears any influence on completion outcomes. As described in the

literature review, the original drug court model was designed as a diversionary program,

but now often includes those on probation. Some authors have suggested that drug court’s

scope has expanded to included offenders with higher risk levels and those already in the

system (for discussion see Leukfeld, 2004; Marlow et al., 2006; Miethe et. al, 2009;

Petyton & Grosswieler, 2001) leaving the current population served very different than

the model originally intended. Initial descriptive statistics for Kentucky Drug Court

shows that only 27% of the population under study came into drug court on the diversion

track. Slightly more than 72% enter through the probation track and less than one percent

through the contempt and family drug court track. This may suggest that Kentucky Drug

Court services a population of higher risk than the original model intended. Also, given

that higher risk individuals often show poorer completion outcomes, the track an

individual come through while a participant in drug court may prove useful to

understanding the completion outcomes.

Time program operational. This variable represents the number of months the

specific drug court program was in operation when the participant entered the program.

This variable was calculated by subtracting the date the program was implemented from

the date the participant began the drug court program. As discussed in the literature

review, Belenko (2001) observed that the time a program is in operation is important in

understanding outcomes. Program success may be dependent on allowing time for a

program and staff to work through implementation issues and develop the necessary

community relationships.

Months in program. The number of months an individual participated in the drug

court program was calculated using the date of entry and the date of last status change.

This variable showed a median of nine months in program for those who did not graduate

from drug court and a median of 21 months in program for those who graduated. This

finding should be self-explanatory as those who do not graduate the drug court program

will have less time in program. Since graduation is a function of time in the program this

variable is not included in the logistic regression analysis. However, this variable may be

important when examining differences within the outcome groups.

Data Summary

The original dataset contained 3,621 records. After data cleaning, the final dataset

contains 3,497 unique records. Several records were deleted because the information

showed that they were outside the requirement of the data pull. For example, 41 records

were deleted as they showed dates and/or descriptions of codes of ineligibility or non-

acceptance, or no time in the program was indicated. The focus of the research is to

examine drug court participants, not those that were not eligible for participation. Ten

cases were deleted because of errors in the dates of entry. Forty-three records that showed

an age less than 18 upon entry into drug court were deleted, as this research was limited

to adult drug court. In isolated instances, individuals younger than 18 may be allowed to

begin adult drug court; however, those cases represent unique case scenarios and are

therefore excluded from the dataset. Thirty records where an individual participant’s

identity could potentially be identified through bivariate analyses were deleted. This was

limited to the “unknown” and “other” responses for the sex variable. A total of 124

records, or slightly less than 3.5% of the total dataset, were deleted.

Descriptors of the population under study shows roughly 62% are male and 38%

female. The drug court population in this study is listed as 85% white and 15% non-

white. The median age of the population is 29 years old and more than half of the

population of study was less than 30 years of age at the time of intake. Roughly a third of

the population under study showed an education level of less than high school. Seventy-

two percent of the population under study entered drug court through the probation track.

The outcome variable, which represents the completion outcome, shows graduation rate

of slightly less than 30%.

In contrast to the drug court population, the Commonwealth of Kentucky

population statistics show roughly 51% male, 89% white, a median age of 36.5 and a

80.3% completion rate for high school (U.S. Census Bureau, 2010). This general

comparison suggests that the drug court population under study is younger, slightly more

racially diverse, and less educated than the general Kentucky population.

Logistic Regression

Logistic regression is selected as the analytic strategy for this research. The

dependent variable for this study, program completion outcome, is categorical with

values of graduated and did not graduate. Logistic regression is similar to both the

multiple regression and the discriminate analysis techniques in many ways, but best fit

the data at hand and research goals. Multiple regression uses two or more continuous

independent variables to predict the value of a continuous dependent variable (Pampel,

2000). Discriminate analysis, on the other hand, seeks to predict group membership

within a categorical dependent variable from multiple independent variables. Logistic

regression is often used as an alternative to both multiple regression and discriminate

analysis as it carries properties of each of the techniques. Logistic regression regresses

independent variables on a categorical dependent variable to predict group membership

(Mertler & Vanatta, 2005). The current study seeks to predict program completion

outcomes, either a participant graduated or did not graduate from the program, by using a

combination of continuous, ordinal, and categorical variables, leaving logistic regression

as the most appropriate statistic for this research.

Methods of logistic regression. There are three main types of logistic regression:

enter, forward, and backward. An “enter” logistic regression enters all variables into

model simultaneously (Field, 2005). Using the enter method for logistic regression, a

researcher can examine the individual contribution of each variable while holding all

other variables constant (Hickert et al., 2009). The enter method for logistic regression is

limited, however, in that it does not allow for the examination of relationships between

predictors. A variation of the enter method that allows for a limited examination of how

the predictors may be related to each other is a blockwise enter method. With a blockwise

enter method the researcher forces entry of the variables in specified groups and in a

specified order based on a priori decision criteria (Field, 2005; e.g., theory, past research,

time causal ordering). Entering variables in blocks allows researchers to assess the

contribution of each group (block) of variables and also assess changes in contributions

of individual variables when other variables are introduced into the model.

A brief discussion about stepwise methods is warranted before a description of

forward stepwise logistic regression is offered. The term stepwise, in relation to statistics,

refers to entering of variables in “steps,” and generally refers to a variation in statistical

analysis methods where a computer algorithm selects the order of variable entry into the

model (Menard, 2010). Stepwise is contrasted with the more traditional approach of using

some a priori decision criteria selected by the researcher. The use of stepwise methods is

contentious (see Hosmer & Lemeshow, 1989; Mundry & Nunn, 2009), but considered

permissible in the absence of prior precedence or theory, or when the purpose of the

research is the identification of predictors (Hosmer & Lemeshow, 1989; Menard, 2010;

Tabachick & Fidell, 2007). Some authors argue that stepwise methods result in over

fitted models that cannot be generalized beyond the dataset used to generate the model

(Hosmer & Lemeshow, 1989; Mundry & Nunn, 2009); however, this criticism is not

applicable when the research effort is not attempting to draw inferences beyond the

dataset at hand, such as the case with the current study. Additionally, Mundry and Nunn

(2009) suggest that stepwise methods should never be used for null hypothesis

significance testing as it greatly increases the probability of a type I error. In the

following description of forward logistic regression, the term stepwise refers to the

computer algorithm selected entrance criteria.

A forward stepwise method enters predictors into the model one at a time, in the

order of importance. The algorithm selects the order of entry based upon the variable’s

contribution to “how well the model fits the observed data” and excludes unproductive

variables (Field, 2005, p. 226). The most important predictor is entered at step one, the

second most important at step two along with the variables from the previous step, and so

on until all productive variables are entered into the model. A forward entry allows a

researcher to examine relationships among the predictor variables by examining

individual contributions to model fit as each variable is entered into the model.

In this study, the logistic regression analyses were run in two major ways, first,

through the variable selection method recommended by Hosmer and Lemeshow (1989),

Menard (2010) and Tabachick & Fidell (2007) and second, without using variable

selection. Using variable selection methods, an enter method logistic regression, Model I,

was conducted to test the hypotheses. A second model, a forward entry stepwise logistic

method, was used to identify the order of importance and potential relationships among

predictor variables. A third model excluded unproductive or problematic variables to

assess model performance in the absence of these variables. The second major way the

multivariate analyses were run was with all variables under study, with one variable

identified as having problems with multicollinearity excluded. This fourth model was run

with all variables under study to assess the variable selection process and to explore any

potential interactions between predictors.

Assessing assumptions. Logistic regression is considered more flexible than

either the multiple regression or the discriminate analysis methods as it does not make

assumptions about distribution, random sampling, a linear relationship between variables,

and homogeneity of the variance (Hosmer & Lemeshow, 2000; Mertler & Vanatta, 2005;

Tabachnick & Fidell, 2007). Although assumptions are few, concerns with logistic

regression models include the case to variable ratio, multicollinearity, and outliers. These

areas of concern were assessed and identified through a series of diagnostic tests and are

discussed below.

Case to variable ratio. To ensure sufficient number of cases relative to the

number of independent variables, unproductive variables were removed for the multiple

logistic regression. With a large enough sample size, such as observed in the data for this

research, a statistically significant finding could mean relatively little for identifying

relationships between variables (Tabachnick & Fidell, 2007). In instances such as this,

the relationship may be significant, but not meaningful or productive. In this case,

measures of effect should be used in combination with tests of statistical significance

(Tabachnick & Fidell, 2007).

In this study, productivity of the variable means that the independent variable is

not only related to the dependent variable in a statistically significant way (α = .10), but

demonstrates at least a small strength of relationship (Pearson’s r greater than or equal to

.1 or -.1) as defined by Cohen (1988). Menard (2010) and Cox (1970) suggest that linear

methods can be applied directly to binary variables when coded in a 0,1 format. Cox

(1970) argues that the binary variable can be treated “just as if they were quantitative

observations” (p. 16), but further explains that this type of method is limited for binary

data analysis. In this manner, this linear method was used as a diagnostic tool only, and

was followed by the more appropriate method for assessing the predictive relationships of

binary variables, which is the logistic regression method. The use of bivariate correlations

as an initial diagnostic tool for subsequent logistic regression models was used in other

drug court studies, namely Cosden et. al (2006) and also Hiller et. al (1999). Herein the

combination of statistical significance and sufficient strength of relationship is termed

meaningfully related. See Appendix C, Table C1 for the correlation matrix details on all

variables, Table 8 for correlation results between the predictors and the outcome variable,

and Table 9 for a listing of meaningfully related variables.

As logistic regression relies on a goodness-of-fit test to provide a measure of how

well the model fits the data, expected cell frequency counts should not drop below a

count of five (Mertler & Vanatta, 2010; Tabachnick & Fidell, 2007). Therefore, the

variables race, marital status, education, drugs of choice, charge type, and sanction type

were collapsed into fewer categories. Category selection was guided by theory, reviews

of methods used in prior research, and frequencies observed within the data. The

unknown and other categories in the sex variable were deleted, in part because of the

potential to violate the cell count criteria, but also to protect against the identification of

an individual participant. These methods for ensuring sufficient case to variable ratio and

expected cell frequency counts are suggested by Tabachnick and Fidell (2007), Mertler

and Vanatta (2005), and also discussed by Menard (2010) in terms of model

specification, variable selection, and model building. Appendix B shows the details of all

variable recoding.

Multicollinearity. Correlations (Pearson’s r), tolerance and variance inflation

factor (VIF) scores were calculated to assess issues of collinearity among the independent

variables using a multiple regression method. Variables with a VIF score greater than ten

and tolerance scores less than .10 indicate a potential problem with multicollinearity

(Pallant, 2007). Although multiple regression is an inappropriate statistic for a binary

dependent variable, this method is suggested by Menard (2010) who stated that tests for

collinearity are

typically not available in logistic regression software, but can easily be obtained

by calculating a linear regression model using the same dependent and

independent variables as you are using in the logistic model. Since the concern is

with the relationship among the independent variables, the functional form of the

model for the dependent variable is irrelevant to the estimation of collinearity. (p.

127)

Pearson’s r was used to assess any multicollinearity indicated by the tolerance and

VIF scores. A Pearson’s r of +/- .5 or greater was used to identify other highly related

variables. Only one variable, the number of drugs of choice, exceeded these thresholds

and was therefore removed from variable selection. See Appendix C, Table C2 for VIF

and tolerance scores, and Table C1 for the correlation matrix for predictor variables.

Outliers. Outliers for each logistic regression model were identified with

parameters set to list cases where residuals exceeded three standard deviations and for

which the model did not predict well. After deleting unproductive variables and the

variable showing multicollinearity, no outliers were found in any of the logistic

regression models.

Exclusion of irrelevant variables. Correlation coefficients were used to assess the

relatedness of the independent variables to the dependent variables (discussed previously

under case-to-variable ratio). Only variables that were meaningfully related to the

dependent variable were included. Excluding unproductive variables increases the

efficiency and overall fit of the model (Menard, 2010). As discussed earlier, productivity

or the meaningfulness of the relationship between variables was assessed with a relaxed

significance level (α = .10) as suggested by Menard (2010) and a strength of relationship

where r was at least .1 or -.1 or considered at least small using Cohen’s (1988) effect size

criteria.

Methods Summary

Bivariate analyses are used to describe the population under study, select

variables for multivariate analyses and test for relationships between predictor variables.

Multivariate analyses run for this study include enter and forward logistic regression

methods. This study uses the logistic regression methods to determine which factors are

related to drug court completion outcomes. Given that prior research finds relationships

among predictor variables and some authors argue that variable selection process may

exclude relevant variables, this study also uses bivariate and multivariate methods to

assess problematic variables and also assess the variable selection process.

CHAPTER IV

FINDINGS AND ANALYSIS

Bivariate Analyses

The variables included in this study were limited to those previously identified as

related to program outcomes and those with a theoretical relationship to program

completion. Table 7 shows each variable under study and the number and percent of

participants for both graduates and non-graduates of the drug court program.

Table 7

Independent Variables by Graduation Status

Variable

Graduates

Non-graduates

Sex

Male

610

58.5

1560

63.5

Female

432

41.5

895

36.5

Race

White

927

89.0

2046

83.3

Non-white

115

11.0

409

16.7

Marital Status

Not Married

742

71.2

1911

77.8

Married

266

25.5

3.3

407

137

16.6

5.6

Missing

Education

Less than high school

215

20.6

819

33.4

At least high school

779

74.8

1408

48.6

Missing

4.6

228

9.3

Drugs of Choice (% yes)

Meth or other stimulant

263

25.2

376

15.3

Cocaine/Crack

251

24.1

720

29.3

Marijuana

408

39.2

1087

44.3

Opiates

402

38.6

1117

45.5

Alcohol

308

29.6

810

33.0

Sedatives

221

21.2

627

25.5

Table 7 (continued)

Variable

Graduates

Non-graduates

Drugs of Choice(% yes)

Other

2.3

3.4

Charges (% yes)

Sale/Trafficking/

217

20.8

287

11.7

Manufacturing

Vehicle/Traffic

4.4

120

4.9

Drug Possession

469

45.0

840

34.2

Administration of Justice

8.0

457

18.6

Public Order

4.4

152

6.2

Person

4.7

190

7.7

Property

192

18.4

769

31.3

DUI

7.6

126

5.1

Drug Court Track

Diversion

354

34.0

593

24.2

Probation

685

65.7

1852

75.4

Missing

0.3

0.4

Sanctions

Incarceration / Detention

613

58.8

1848

75.3

Treatment

140

13.4

504

20.5

Community Service

270

25.9

685

27.9

Suspension/Demotion

116

11.1

264

10.8

Increase Program Elements

162

15.5

323

13.2

Other

137

13.1

553

22.5

The overall graduation rate in this study for Kentucky Drug court is 29.8%. Table

7 shows the frequencies for the independent variables by graduation status. Looking at

the frequencies, graduates appear to be slightly more female (41.5% versus 36.5%) as

well as slightly more white (89% versus 83.3%), married (25.5% versus 16.6%) with at

least a high school education (74.8% versus 48.6%) when compared to non-graduates.

The frequencies also suggest some patterns for problem behaviors. Those who

graduate from drug court were more likely to indicate methamphetamines or other

stimulants (25.2% versus 15.3%) as a drug of choice, although in turn, were less likely to

indicate cocaine/crack, marijuana, opiates, alcohol, sedatives, and other drugs. Graduates

were almost twice as likely as non-graduates (20.8% versus 11.7%) to carry a charge

related to the sale, trafficking, or manufacture of drugs. They were also more likely to

have charges of drug possession (45% versus 34.2%) and DUI (7.6% versus 5.1%). On

the other hand, those who graduated from drug court were less likely than non-graduates

to carry charges against the administration of justice (8% versus 18.6%), public order

(4.4% versus 6.2%), person (4.7% versus 7.7%), and property (18.4% versus 31.3%).

Those who successfully completed drug court were more likely than non-graduates to

enter into the program on a diversion track (34% versus 24.2%) with non-graduates

therefore more likely to enter from probation. While in the program, those who

graduated were less likely to indicate receiving an incarceration/detention sanction

(58.8% versus 75.3%) and a treatment related sanction (13.4% versus 20.5%) than non-

graduates. It should be noted that some of these differences are marked while others are

slight. In addition, there seem to be minimal if any differences in the percent of

graduates versus non-graduates by gender; charges related to traffic violations; and

sanctions of community service, suspension/demotion, or increased program elements.

Variable selection. To determine which variables to include in the multivariate

analyses, bivariate analyses were run to identify variables that are meaningfully related to

program completion outcomes. A variable is considered meaningfully related to

graduation status and included in later multivariate models if it met two criteria. First,

the observed relationship must be statistically significant at p = ≤ .100 in bivariate

correlation tests. As suggested by Menard (2010), a relaxed p-value to reach statistical

significance was used. The second criteria for variable selection is that the size of the

observed relationship must be at least small, showing a Pearson’s r greater than or equal

to +/- .100 as defined by Cohen’s (1988) measure of effect size. This was done for

diagnostic purposes and to accommodate for the power of the statistics associated with a

large number of observations (N= 3,497). Moreover, one variable indicating issues with

multicollinearity (VIF > 10, tolerance <.1, and r = +/-.5) was excluded. Table 8 shows

the results of variable selection analyses.

Results from the variable selection process show that 13 of the 31 original

variables are meaningfully related to drug court completion outcomes. Although prior

research and theory was used to guide variable selection, bivariate correlation analyses

were run to exclude variables not significantly related to the dependent variable to

identify the most parsimonious set of predicator variables. Using the criteria to identify

meaningful relationships (p < .10 and r ≥(+/-) .1), these bivariate analyses show that

graduates and those who fail to graduate do not greatly differ in terms of race, gender,

most drugs of choice and most sanctions.

Bivariate relationships between predictors. As several studies discussed in the

literature review found significant relationships between independent variables, specific

tests of relationships were conducted to determine if these same relationships are present

for Kentucky Drug Court. To test these relationships, chi-square tests of independence

were run between categorical variables while t-tests were used to test for differences

regarding the continuous variable for age. The threshold for statistical significance is α =

.05.

Table 8

Variable Selection

Variable

Pearson’s r

Sig.

Sex

-.047

.005

Age*

.216

.000

Race

.072

.000

Marital status*

.101

.000

Education*

.150

.000

Number drugs of

choice**

-.119

.000

Meth or other stimulant*

.117

.000

Cocaine/crack

-.054

.002

Marijuana

-.047

.005

Opiate

-.064

.000

Alcohol

-.034

.046

Sedative/downer

-.046

.006

Other

-.030

.080

Number charge types

-.016

.395

Drug sale/traffic/manu*

Manufacturing

.119

.000

Drug possession*

-.102

.000

Vehicle/traffic

-.010

.547

Admin justice*

-.135

.000

Public order

-.035

.038

Person

-.055

.001

Property*

-.132

.000

DUI

.048

.005

Number of sanctions*

-.166

.000

Any incarceration*

-.165

.000

Any treatment sanction

-.084

.000

Any community service

-.020

.227

Any suspension sanction

.006

.742

Any increase program

Elements

.032

.061

Any other sanction*

-.108

.000

Track

-.101

.000

Months in operation*

-.102

.000

Time in program

.538

.000

*p < .10 and r ≥(+/-) .1 and included in logistic regression model;

**removed for multicollinearity

Within the literature on drug court outcomes, a commonly cited relationship

between predictor variables is between drug of choice and participant characteristics. In

this current study, significant gender differences were found for marijuana, opiates,

alcohol, and sedatives. Chi-square tests, using the continuity correction for two-by-two

tables, showed that males were significantly more likely to indicate marijuana, χ2 (1, N

=3,497) = 79.785, p = .000, and alcohol, χ2 (1, N =3,497) = 42.016, p = .000, as a drug of

choice than females. Females were significantly more likely to indicate opiates, χ2 (1, N

=3,497) = 8.756, p = .003, and sedatives, χ2 (1, N =3,497) = 5.077, p = .024 than males.

Drugs of choice also showed racial differences. Chi-square tests, using the

continuity correction for two-by-two tables, showed that whites were significantly more

likely to indicate methamphetamines or other stimulants, χ2 (1, N =3,497) = 82.868, p =

.000, opiates, χ2 (1, N =3,497) = 214.185, p = .000, and sedatives, χ2 (1, N =3,497) =

52.535, p = .000, as a drug of choice than non-whites. Non-whites were significantly

more likely to indicate cocaine, χ2 (1, N =3,497) = 44.425 p = .000, marijuana, χ2 (1, N

=3,497) = 19.819, p = .000, and alcohol, χ2 (1, N =3,497) = 19.960 p = .000, than whites.

Significant age differences were found for drug of choice. Independent samples t-

tests show that participants indicating cocaine or crack, t(3495) = -6.745, p = .000, and

alcohol, t(3495) = -2.156, p = .031, as a drug of choice were significantly older than

participants who did not. Participants indicating marijuana, t(3495) = 9.705, p = .000,

opiates, t(3495) = 6.420, p = .000, sedatives, t(3495) = 7.909, p = .000, and other drugs

t(3495) = 3.896, p = .000, were significantly younger than those who did not.

Significant relationships are also found between drugs of choice and charges. The

results of this current study showed that those indicating methamphetamine or other

stimulant, χ2 (1, N =3,497) = 11.659, p = .001, and those indicating opiates, χ2 (1, N

=3,497) = 6.664, p = .010, were more likely to carry a charge relating to the sale,

trafficking, or manufacturing of drugs. Participants indicating cocaine were more likely

to carry a charge relating to the crime against a person than those not indicating cocaine,

χ2 (1, N =3,497) = 3.865, p = .049.

That certain drugs may be associated with a deeper involvement in criminal

activity is further supported by other relationships with charge type. The tests of

relationship here show that cocaine is the only drug of choice to show a significant

relationship to charges related to crimes against a person. Methamphetamines or other

stimulant as a drug of choice is significantly related to five charge types, more so than

any other drug of choice; sale, trafficking, or manufacturing, χ2 (1, N =3,497) = 8.326, p

= .004, vehicle or traffic, χ2 (1, N =3,497) = 5.325, p = .021, drug possession, χ2 (1, N

=3,497) = 18.945, p = .000, administration of justice, χ2 (1, N =3,497) = 9.571, p = .002,

and property crime, χ2 (1, N =3,497) = 6.003, p = .014. The less “hard” drug types,

specifically marijuana and alcohol are also related to the less intense charge types, such

as possession and property crimes. However, offense severity is not directly explored in

this study.

Multivariate Analyses: Logistic Regression

Several multivariate analyses were run using the variables that met selection

criteria and excluding those that did not, as delineated in Table 9. First, Model I used a

blocked enter logistic regression method to assess the predictive utility of the individual

variables and to explore the predictive utility of each set of predictors to test the three

hypotheses. Second, a forward stepwise logistic regression was run to determine the

importance of each predictor to model performance. Third, the results of these logistic

regression analyses warranted follow up analyses, which included an additional forward

stepwise logistic regression to assess problematic variables identified in Model I and

Model II. A fourth full model using forward stepwise logistic regression method was run

to assess the variable selection process and determine the presence of potential interaction

effects.

Table 9

Variables Included and Excluded

Included

Excluded

Marital status

*Number drugs of choice

Age

Race

Education level

Sex

Methamphetamine or other stimulant

Cocaine/crack

Charge for sale, trafficking or manufacturing

Marijuana

Charge for drug possession

Opiate

Charge related to the administration of justice

Alcohol

Property crimes

Sedative

Number of sanctions

Other drug

Received any incarceration sanction

Number of charges

Received any other type of sanction

Any vehicle or traffic related charge

Drug court track

Any charge against to public order

Number of months program in operation

Any crime against a person

Any DUI

Any treatment or treatment related

sanction

Any community service sanction

Any suspension sanction

Any sanction that increased program

elements

*removed due to multicollinearity

Model I: Enter logistic regression. A binary logistic regression model using a

three-block enter method was run to isolate the effects of problematic behaviors and

program performance from characteristics about the individual. At block one, participant

characteristics, marital status, age, and education, were entered. This block was

statistically significant, χ2(3, 3,070) = 232.640, p = .000 and showed a total percent

correctly classified at 69.4%. The Hosmer and Lemeshow test of model fit showed a poor

model fit, χ2(8, 3,070) = 25.604, p = .001. Nagelkerke’s R square showed that these

variables accounted for 10.3% of the total variance. Table 10 shows the results of block

one.

Table 10

Model I: Block 1

Independent Variables

S.E

Sig.

Wald

Exp(B)

Marital status

0.456

0.098

0.000

21.598

1.578

Age

0.051

0.005

0.000

119.760

1.053

Education

0.681

0.094

0.000

52.926

.1975

At block two, variables related to problem behaviors were entered to see if they

provided predictive utility beyond participant characteristics. Methamphetamine as a

drug of choice, charges for the sale, trafficking, and manufacturing drugs, charges for

possession, charges related to the administration of justice, property charges, the number

of sanctions, ever received a sanction for incarceration, and ever received an “other”

sanctions were entered. As expected, this block was statistically significant, χ2(8, 3070) =

262.184, p = .000, and the model performed better and gained overall good model fit

when variables representing participant problem behaviors were entered. The model was

statistically significant, χ2(11, 3,070) = 494.824, p = .000 and correctly classified 72.9%

of the cases. The Hosmer and Lemeshow test of model fit resulted in a good model fit,

χ2(8, 3,070) = 2.915, p = .940. Nagelkerke’s R square showed the variance predicted by

the model more than doubled to 21.0% with the addition of these variables. Table 11

shows the results of the entering block two. The predictors that influenced the odds of

graduating the greatest, identified by Exp(B), are level of education, carrying a charge

related to the sale, trafficking or manufacturing of drugs, and carrying a charge related to

the administration of justice.

Table 11

Model I: Block 2

Independent Variables

S.E

Sig.

Wald

Exp(B)

Marital status

0.396

0.104

0.000

14.578

1.486

Age

0.042

0.005

0.000

74.517

1.043

Education

0.755

0.098

0.000

59.335

2.127

Methamphetamine or other

stimulant

0.472

0.105

0.000

20.203

1.603

Sale/trafficking/

manufacturing

0.502

0.117

0.000

18.275

1.652

Drug possession

0.332

0.093

0.000

12.641

1.393

Administration of justice

0.983

0.143

0.000

47.533

0.374

Property

-0.330

0.110

0.003

8.930

0.719

Number of sanctions

-0.052

0.016

0.001

11.435

0.949

Any incarceration sanction

-0.478

0.109

0.000

19.127

0.620

Any other sanction

-0.435

0.123

0.001

12.449

0.647

At block three, drug court variables were entered into the model. The number of

months the drug court was in operation when the participant entered drug court and the

track through which a participant entered were added. At block three, only modest gains

in overall model performance were achieved. The overall model was statistically

significant, χ2(13, 3,070) = 565.125, p = .000, and this block was statistically significant

χ2(2, 3,070) = 70.301, p = .000. The addition of these two variables increased the percent

of cases correctly classified to 73.1%; however, the Hosmer and Lemeshow test of model

fit showed a poor model fit, χ2(8, 3,070) = 20.792, p = .008. Nagelkerke’s R square

showed that these variables accounted for 23.7% of the total variance. At block three, the

model correctly predicted not graduating for 89.9% of the cases, but only correctly

predicted 35.6% of those that graduated. See Table 12 for details of Model 1, block

three. The predictors that influenced the odds of graduating the greatest, identified by

Exp(B), did not change with the addition of these two variables.

The results for this model show that all variables significantly predict graduation

outcomes, but earning at least a high school diploma slightly more than doubles the odds

of graduation over those who do not have at least a high school diploma (p = .000, B =

0.717, ExpB = 2.049). Participants who carry a charge related to the sale, trafficking or

manufacturing of drugs show an 67.7% increase odds of graduation (p = .000, B = 0.517,

ExpB = 1.677), and those carrying a charge related to the administration of justice show a

53.9% reduction in the likelihood to graduate than those not carrying such a charge (p =

.000, B = 0.878, ExpB = 0.461). Other factors positively related to graduation include

age, marital status, indicating methamphetamine as a drug of choice, and carrying a

charge related to drug possession. Other factors negatively related to graduation include,

carrying a property crime charge, increases in the number of sanctions, receiving a

sanction related to incarceration, entering drug court through the probation track, and

increases in the time a drug court is in operation.

Table 12

Model I: Block 3

Independent Variables

S.E

Sig.

Wald

Exp(B)

Marital status

0.285

0.106

0.007

7.204

1.330

Age

0.049

0.005

0.000

93.039

1.050

Education

0.717

0.100

0.000

51.813

2.049

Methamphetamine or other

stimulant

0.397

0.107

0.000

13.743

1.487

Sale/trafficking/

manufacturing

0.517

0.119

0.000

18.719

1.677

Drug possession

0.413

0.095

0.000

18.717

1.511

Administration of justice

-0.878

0.144

0.000

37.348

0.461

Property

-0.273

0.112

0.015

5.964

0.761

Number of sanctions

-0.063

0.016

0.000

15.760

0.939

Any incarceration sanction

-0.494

0.111

0.000

19.864

0.610

Any other sanction

-0.441

0.125

0.001

12.391

0.643

Track

-0.499

0.099

0.000

25.447

0.607

Months program in operation

-0.005

0.001

0.000

26.323

0.995

Model II: Forward stepwise logistic regression. Model II used a forward entry

stepwise method, and resulted in final model performance identical to Model I, which is

expected as the same variables were entered with unproductive variables omitted. The

distinguishing feature and added valued of Model II is that having the order of entry

determined by each predictor’s individual contribution to model performance allows

researchers to examine the relative importance of each variable. Of these 13 variables

entered, age, having received a sanction for incarceration and the number of months the

program was in operation were entered first, suggesting that these three variables are the

most important predictors of drug court completion outcomes. Exp(B) results show that

for every year increase in age of the participant at the time of entrance, the odds of

graduating increase by 5.0%, that having received a sanction involving incarceration

decreases the odds of graduating by 39.0%, and that for every month longer in operation

the odds of graduating decreased by 0.5%. Carrying a charge related to the administration

of justice was entered fourth and showed that the odds of graduation decreased by 58.4%

for those carrying this charge type. Table 13 shows the results of the final step of the

forward stepwise logistic regression in order of entry into the model.

Table 13

Model II Results

Independent Variables

S.E

Sig.

Wald

Exp(B)

Age

0.049

0.005

0.000

93.039

1.050

Any incarceration sanction

-0.494

0.111

0.000

19.864

0.610

Months program in

operation

-0.005

0.001

0.000

26.323

0.995

Administration of justice

-0.878

0.144

0.000

37.348

0.416

Education

0.717

0.100

0.000

51.813

2.049

Drug possession

0.413

0.095

0.000

18.717

1.511

Number of sanctions

-0.063

0.016

0.000

15.760

0.939

Sale/trafficking/

manufacturing

0.517

0.119

0.000

18.719

1.677

Table 13 (continued)

Independent Variables

S.E

Sig.

Wald

Exp(B)

Track

-0.499

0.099

0.000

25.447

0.607

Methamphetamine or

other stimulant

0.397

0.107

0.000

13.743

1.487

Any other sanction

-0.441

0.125

0.001

12.391

0.643

Marital status

0.285

0.106

0.007

7.204

1.330

Property

-0.273

0.112

0.015

5.964

0.761

Follow up Analyses

Analysis of the performance within Model I and the order of importance indicated

by Model II suggest the presence of significant relationships between the drug court

variables and the other predictor variables. Two sets of follow up tests of relationships

were conducted. Since the drug court variables proved problematic, the first set of follow

up tests focused on bivariate relationships between track and the variables significantly

related to track and the length of time drug court was operational. A third forward

stepwise model without track and time in operation was run to confirm the findings from

the first two. The second set of follow up analyses was broader in focus and consisted of

a fourth logistic regression using a full forward stepwise model to explore the possibility

that the previous models excluded important and relevant variables.

Chi-square tests of relationships using the continuity correction for 2x2 tables

were run to explore track and variables previously found related to track. Results show

that the track through which one enters drug court is significantly related to carrying

charges against the administration of justice, χ2 (1, 3,497) = 46.541, p = .000, carrying

charges related to drug possession, χ2 (1, 3,497) = 7.276, p = .007, indicating

methamphetamine or other stimulant as a drug of choice, χ2 (1, 3,497) = 68.605, p = .003,

and receiving an “other” type sanction, χ2 (1, 3,497) = 15.003, p = .000. Those carrying

charges related to the administration of justice and those carrying drug possession

charges were disproportionately among those entering drug court on the probation track,

while those indicating methamphetamine as a drug of choice, and those having received

an “other” sanction type were disproportionately among the diversion track. Half of these

variables pull in the opposite direction as would be expected if they were to support the

relationship between track and completion outcomes. Specifically, methamphetamine or

other stimulant shows a positive relationship to completion outcomes while being

overrepresented in the probation track, with the probation track showing a negative

relationship with completion outcomes. Receiving an “other” type sanction shows a

negative relationship to completion outcomes while disproportionately among those who

entered through the diversion track, but those on the diversion track demonstrate an

increase in odds of graduation. The opposing relationships may be a contributing factor in

the decreased model fit observed in block three of Model I.

Since age was found related to the track through which one entered drug court and

also the most important contributing factor in overall model performance, age was further

assessed for relationships with variables related to track. Results from an independent

samples t-test show that those entering through the probation track are, on average,

significantly older than those entering through diversion. Those entering through

probation track showed a mean age of 31.2 years and diversion 29.2 years, t(2,860) = -

20.708, p = .000. Married participants, t(3,324) = -9.958, p = .000, participants who

completed a high school degree, t(3,219) = -6.160, p = .000, those carrying charges

related to the sale, trafficking or manufacturing of drugs, t(3,495) = -3.709, p = .000 and

carrying charges related to drug possession, t(3,495) = -5.875, p = .000, are significantly

older. Participants carrying charges related to the administration of justice, t(3,495) =

2.868, p = .004, property crimes, t(3,495) = -6.503, p = .000, having received a sanction

of incarceration, t(3,495) = 5.978, p = .000, and an “other sanction,” t(3,495) = 3.598, p =

.000, and those on the diversion track, t(3,495) = -6.161, p = .000 are significantly

younger.

It is worth noting that of the predictor variables related to track, indicating

methamphetamine or other stimulant as a drug of choice is the only variable that failed to

show significant differences with age, t(3,495) = -1.798, p = .072. Participants carrying a

charge related to drug possession were, on average, older (31.79 years) than those who

did not (29.98 years), t(3,495) = -5.875, p = .000. Participants carrying a charge related

to the administration of justice were, on average, younger (29.68 years) than those not

carrying such a charge (30.83 years), t(3,495) = 2.868, p = .004. Participants with an

“other” type sanction are, on average, younger (29.60 years) than those who do not

(30.92 years), t(3,495) = 3.598, p = .000. The opposing nature of the relationships

between track and variables related to track and completion outcomes suggests the

presence of noise within the model, which may explain the observed reduction in overall

model fit while still contributing to overall model performance.

An independent samples t-test was run between the time a drug court was in

operation and the track through which one entered drug court to determine if these two

variables are related to each other. The results were statistically significant, with those

entering through the probation track, on average, having entered drug court in programs

that were in operation significantly longer, t(2,860) = -20.708, p = .000. On average,

drug courts were in operation for 74.4 months for those entering through the probation

track, while only 46.5 months for those on the diversion track. These relationships may

help explain the decreased odds of graduating for those entering through the probation

track. Age shows a small, but statistically significant positive correlation with time in

operation (r = 0.073, p = .000), which is in the opposite direction if age were to support

explanation of a decrease in the odds of graduation the longer a program is in operation.

These results for these two relationships also suggest that including length of time a drug

court is in operation produces noise within the model.

An additional forward entry stepwise logistic regression, Model III, was run

without the drug court variables to assess the suspicion of contributing to noise within the

model. Of these 11 variables, age, having ever received a sanction for incarceration, and

carrying a charge related to the administration of justice (which was previously fourth)

were the first three variables entered, suggesting that these three variables are the most

important predictors of drug court completion outcomes. The only change in the top three

predictors from Model II to Model III is the absence of the drug court variable for time

program is in operation. See Table 14 for results of this confirmatory model.

Table 14

Model III: Confirmatory Stepwise Model

Independent Variables

S.E

Sig.

Wald

Exp(B)

Age

0.042

0.005

.000

74.195

1.043

Any incarceration sanction

-0.480

0.108

.000

19.793

1.619

Administration of justice

-1.009

0.141

.000

51.191

0.365

Education

0.751

0.097

.000

60.265

2.118

Property

-0.354

0.109

.001

10.584

0.702

Methamphetamine or other

stimulant

0.490

0.104

.000

22.340

1.632

Any other sanction

-0.427

0.122

.000

12.292

0.653

Marital Status

0.388

0.103

.000

14.273

1.474

Sale/traffic/manufacture

0.481

0.116

.000

17.115

1.617

Number of sanctions

-0.050

0.015

.001

10.559

0.952

Drug Possession

0.295

0.092

.001

10.253

1.343

Note: the variables are listed in order of importance, highest to lowest, in contributing

to overall model performance.

Because of overarching concerns with interaction effects between predictors and

the exclusion of potentially useful variables during variable selection process, a full

stepwise logistic regression model was run with 30 of 31 variables under study included.

The variable for the number of drugs of choice was excluded due to multicollinearity.

Table 15 shows the results of the full forward stepwise model.

As expected, with all variables under study entered, the model was statistically

significant, χ2(20, 2,484) = 557.544, p = .000 and correctly classified 75.2%. This

exploratory model retained 20 of the 30 variables entered and correctly classified not

graduating 90.1% and 42.2% for those that graduated. Nagelkerke’s R squared shows that

this model accounted for 28.3% of the total variance. These figures show improvement in

overall classification accuracy and the amount of variance explained over the previous

models. However, the Hosmer and Lemeshow test of model fit showed a poor model fit,

χ2(8, 2,484) = 27.007, p = 0.001 and across the majority of variables, the standard errors

increased in comparison to the standard errors from both Model II and Model III. Of

these 30 variables, age, the number of sanctions, and carrying a charge related to the

administration of justice were the first three variables entered, suggesting that these three

variables are the most important predictors of drug court completion outcomes.

The results from this model show that the top five predictors across all models are

relatively stable. Age, and carrying a charge related to the administration of justice are

consistently within the top three predictors. Variables for education and carrying

property crime charges changed positions in order of importance between Model III and

Model IV, but both consistently remained in the top five predictors. Nevertheless, this

exploratory model elevated the importance of the number of sanctions to the second most

important predictor whereas it was ranked 7 of 13 variables in Model II and 10 of 11 in

Model III. Moreover, eight of the previously excluded predictors emerged as contributing

to model performance (race, indicating drugs of choice for alcohol or opiates, carrying

charges related to crimes against a person or DUI, sanctions involving an increase in

program elements, suspension, or community service). One predictor that was previously

found meaningfully related and predictive of completion outcomes was dropped from the

model (carrying a charge related to drug possession). Table 16 describes the state of the

variables when examined across all the models. These results suggest the presence of

interaction effects among predictor variables.

Table 15

Model IV: Full Model, Forward Stepwise

Independent Variables

S.E.

Sig.

Wald

Exp(B)

Age

.051

.006

.000

77.450

1.053

Number of sanctions

-.145

.024

.000

37.809

.865

Charge administration of justice

-.887

.149

.000

35.213

.412

Charge property crime

-.531

.119

.000

19.886

.588

Education

.658

.112

.000

34.551

1.931

Increase programming sanction

.843

.156

.000

29.053

2.323

Months in operation

-.003

.001

.016

5.782

.997

Methamphetamine/other stimulant

.523

.126

.000

17.351

1.688

Charge crime against person

-.660

.202

.001

10.723

.517

Other sanction

-.498

.148

.001

11.261

.608

Incarceration sanction

-.366

.127

.004

8.295

.694

Drug of choice alcohol

-.283

.110

.010

6.640

.753

Track

-.421

.118

.000

12.809

.656

Community service sanction

.376

.131

.004

8.222

1.456

Drug of choice opiate

-.389

.105

.000

13.685

.678

Marital Status

.313

.122

.011

6.542

1.368

Suspension sanction

.417

.169

.014

6.092

1.517

Race

.345

.159

.030

4.726

1.412

Charge sale/traffic/manufacture

.348

.131

.008

6.994

1.416

Charge DUI

.446

.190

.019

5.548

1.563

Table 16

Comparison of Model Predictors

Included in all models

Excluded in all models

Emerged

in full model

Dropped

in full model

Age

Sex

Race

Drug possession

Marital Status Education

Drug of choice marijuana

Drug of choice opiate

Charges Admin justice

Charges public order

Drug of choice alcohol

Charges property crime

Drug of choice other

Suspension sanction

Number of sanctions

Number of charges

Increase program elements

Other sanction

Charges vehicle/traffic

Charges DUI

Incarceration sanction

Drug of choice cocaine/crack

Crime against person

Track

Community service

sanction

Months in operation

Charges sale/trafficking/

manufacturing

Results of Hypothesis Tests

To test the hypotheses in this study, the variables were examined two ways. First,

bivariate relationships between individual predictors and the outcome variable were

examined. Second, individual contributions to model performance were assessed. To

reject the null hypothesis, at least one variable tested in each hypothesis must be

meaningfully related, reliably predict program completion outcomes, and contribute to

overall model performance when in the presence of the other variables. Bivariate

correlations were used to identify meaningful relationships and used for variable

selection. Multivariate logistic regression was used to assess the ability to predict drug

court completion outcomes and individual contribution to model performance. In the

following sections, the results of the each hypotheses are described, followed by the

results for each predictor under study.

Hypothesis one. Reject the null hypothesis. Participant characteristics predict

Kentucky Drug Court completion outcomes, with age and level of education among the

strongest of all predictors studied. Not every participant characteristic tested in this study

was meaningfully related to drug court completion outcomes. Only age, marital status,

and education showed a meaningful relationship to the outcome variable and were

therefore entered into the logistic regression model. Several of the results are not in the

expected direction of the predictions offered and an increased-responsibilities argument

was not supported. Race and sex failed to show a relationship with the outcome variable,

while older and married participants showed an increased likelihood of graduation. The

prediction that not having at least a high school diploma is associated with a decrease in

the odds of graduation is supported.

Participant characteristics alone predicted drug court completion outcomes, but

not as well as when other variables were included. The results with only participant

characteristics entered (Model I, block 1) shows a poor model fit and approximately 10%

of the variance explained. All three variables remained significant predictors when the

blocks for problem behaviors and drug court variables were entered. These additional

variables resulted in greatly improved model performance and variance explained. Level

of education produced the strongest influence over changes in the odds of graduation,

even after the addition of the other variables under study. Refer to Table 10 to view the

results of the model with only participant characteristics involved, Table 11 for results

when problem behaviors were entered, and Table 12 for results when the drug court

variables were entered. In Model II, a forward stepwise logistic regression with all 13

variables entered, level of education showed the strongest influence on the odds of

graduation and age was the most important individual contributor for predicting

completion outcomes. Refer to Table 13 to view results of the forward stepwise logistic

regression model, which shows the variables in order of importance. Model IV, a full

forward stepwise logistic regression including 30 out of all 31 variables, showed that

participant characteristics of age, education, marital status and race all predict completion

outcomes.

Sex. Although initial examination of graduation status by sex appears to show a

slight differences between graduates and non-graduates (41.5% of graduates were female

compared to 36.5% of non-graduates), the tests for variable selection showed that sex

was not meaningfully related to program completion outcomes (p = .005, r = -.047). Sex

was therefore not included in the logistic regression analysis used for hypothesis testing.

Model IV, which included all variables of interest without respect to variable selection

criteria, showed that sex was not a significant predictor of completion outcomes.

Age. Initial analysis of data between graduates and non-graduates showed that the

mean age of graduates, 33.5 years, was older than non-graduates, 29.4 years. Age was

found meaningfully related to program completion outcomes in variable selection tests (p

= .000 , r = .216) and a significant predictor in all models. In all stepwise logistic

regression models, age was found to be the most important predictor, evidenced by being

entered into the model first. The results show that for every one year increase in age the

odds of graduation increase by 5.0% (p = .000, B = .049, ExpB = 1.050). The older the

participant, the more likely he or she is to graduate.

Race. Although initial examination of graduation status by race appears to show a

slight difference between graduates and non-graduates (89.0% of graduates were white

compared to 83.3% of non-graduates), initial tests for variable selection showed that race

was not meaningfully related to program completion outcomes (p = .000 , r = .072).

Race was therefore not included in the logistic regression analysis used for hypothesis

testing. However, in Model IV, which included all variables of interest without respect to

variable selection criteria, race emerged as a significant predictor of completion

outcomes. Although race was retained as a statistically significant predictor, it was

among the least important of predictors (entered 28 of 30 variables.)

Marital status. Initial examination of graduation outcomes by marital status

shows some differences between graduates and non-graduates (25.5% of graduates were

married compared to 16.6% of non-graduates). Marital status was found meaningfully

related to program completion outcomes (p = .000 , r = .101) and results from the

multivariate analyses showed that marital status reliability predicted program comes (p =

.000, B = .285, ExpB = 1.330). These results show that being married increased the odds

of graduation by 33.0%. The results of Model II, which included only meaningfully

related variables, and Model IV, which included all variables without regard to any

statistical variable selection, showed that marital status is one of the least important

predictors, entered at step 12 of 13 steps and 26 of 30 steps respectively.

Education. Initial examination of graduation status by education shows a striking

difference between graduates and non-graduates in education level. Of graduates, 74.8%

indicated having earned at least a high school diploma or equivalent compared to 48.6%

of non-graduates. Education was found meaningfully related to program completion

outcomes in variable selection tests (p = .000 , r = .150) and also reliability predicted

program completion outcomes (p = .000, B = 0.717, ExpB = 2.049). Results show that

participants indicating the completion of at least a high school diploma or equivalent

were 104.9% more likely to graduate from drug court than those with less than a high

school diploma. Level of education, of all the predictor variables, exerts the strongest

change in the odds of graduation and was also among the top contributing variables, in

the forward stepwise models. However, this variable may be a proxy for graduation, as

participants who do not possess at least a high school diploma or its equivalent at the time

of entrance are required to work on education while in the drug court program.

Hypothesis two. Reject the null hypothesis. Knowing participants’ problem

behaviors leading into the drug court program is useful for predicting completion

outcomes. Initial examination of completion outcomes by these problem behaviors shows

slight differences between graduates and non-graduates for most of the variables under

study. Indicating methamphetamine or another stimulant was shown to be meaningfully

related and reliably predict outcomes, along with possessing charges for the sale,

trafficking or manufacturing of drugs, drug possession, charges related to the

administration of justice, and property crimes. Having received a sanction involving

incarceration or detention, or an “other” type sanction was also shown to reliably predict

completion outcomes. Carrying charges relating to the sale, trafficking, or manufacturing

of drugs and carrying charges related to the administration of justice exerted the second

and third strongest effects on changes in the odds of graduation.

Similar to hypothesis one, several of the results are not in the expected direction.

The prediction that those with harder drugs, specifically cocaine or crack and opiates,

was not supported as a relationship with completion outcomes was not found. This study

failed to support predictions regarding a decreased likelihood of graduation for

participants carrying crimes against people or crimes for the sale, trafficking or

manufacturing of drugs. The results show a relationship between carrying a property

crime charge and completion outcomes in some analyses. Carrying a charge for the sale,

trafficking or manufacturing of drugs resulted in an increase in the odds of graduation.

The prediction regarding the number of drugs of choice could not be tested due to

problems with multicollinearity. The results of this study support the prediction that jail

sanctions are associated with decreased odds of graduation.

Adding variables for problem behaviors into the model (Model I, block 2)

resulted impressive gains in model performance. The amount of variance explained with

these additional variables more than doubled to 21.0% and resulted in a good model fit.

Overall model accuracy increased by 3.5 percentage points by adding these variables.

According to the forward stepwise model, receiving a sanction involving incarceration

and carrying a charge related to the administration of justice were among the most

important predictors. Specific findings for these variables are outlined below.

Drugs. Only one drug of choice, methamphetamine and other stimulants, was

found meaningfully related to completion outcomes (p = .000 , r = .117). Results of the

multivariate analyses show that participants indicating methamphetamine or other

stimulant as a drug of choice are 48.7% more likely to graduate from drug court than

those who did not (p = .000, B = 0.397, ExpB = 1.487). This variable was entered 10 of

13 in the forward stepwise logistic regression model. The remaining drug of choice types,

cocaine or crack (p = .002 , r = -.054), marijuana (p = .005 , r = -.047), opiates (p =

.000 , r = -.064) alcohol (p = .046 , r = -.034), sedatives (p = .006 , r = -.046), and other

types of drug (p = .080 , r = -.030) were not meaningfully related to drug court

completion outcomes and therefore not included in the multivariate analyses used for the

tests of hypotheses.

The number of drugs of choice, although it showed a meaningful relationship with

the completion outcome (p = .000 , r = .119), was not included in the multivariate

analyses due to evidence of multicollinearity. The number of drugs of choice showed a

variance inflation factor greater than ten and a strong relationship to the sedative as a

drug of choice (p = .000, r = .604). The number of drugs of choice was therefore not

included in the multivariate analyses.

Crime. Four of the original nine charge types studied here were found

meaningfully related to drug court completion outcomes. Carrying a charge against the

administration of justice (p = .000 , r = .135), the sale, trafficking, or manufacturing of

drugs (p = .000 , r = .119), drug possession (p = .000 , r = .102), and property crime (p

= .000 , r = -0.132) were meaningfully related to drug court completion outcomes. All

four of these variables also reliably predicted completion outcomes. Carrying a charge

relating to the administration of justice decreased the odds of graduating by 58.7% when

compared to those who do not (p = .000, B = -0.878, ExpB = 0.413). Across all four

models, this variable is the strongest predictor of the charge types, the fourth most

important variable in contributing to overall model performance, and the fourth strongest

influence on odds of graduation. Carrying drug possession charge type increased the

odds of graduating by 51.1% over those who do not (p = .000, B = .413, ExpB = 1.511)

and was entered sixth of 13 variables in the forward stepwise logistic regression. Charges

relating to the sale, trafficking or manufacturing of drugs resulted in a 67.7% increase in

the likelihood of graduating drug court over to those who did not (p = .000, B = 0.517,

ExpB = 1.677), revealing that this charge type exerts the second strongest influence on

changes in the odds of graduation. This charge type was entered eighth of the 13

variables in Model II, the forward stepwise model. Carrying charges related to property

crimes resulted in a 23.9% decrease in the odds of graduating (p = .015, B = -0.273,

ExpB = 0.761). The forward stepwise logistic regression showed that the property crime

charge type is the least important of the 13 variables in contributing to overall model

performance.

Charges relating to vehicle or traffic offenses (p = .547 , r = -0.010), charges

relating to crimes against public order (p = .038 , r = -0.035), charges relating to crimes

against a person (p = .00 , r = -0.055), and DUI charges (p = .005 , r = 0.048) failed to

meet the inclusion criteria; therefore they were excluded from the logistic regression

analyses for the test of hypotheses.

The number of charges a participant possess at the time of entry was not found

meaningfully related (p = .395 , r = -.016) and therefore excluded from the multivariate

analyses for the test of hypotheses.

Sanctions. Three of the seven variables related to drug court sanctions showed

meaningful relationships to drug court completion outcomes. Having received a sanction

involving incarceration or detention (p = .000, r = -0.165) or having received an “other”

type sanction (p = .000, r = -0.108), and the number of sanctions a participant received

while in drug court (p = .000 , r = -0.166) were meaningfully related to the outcome. All

three of these variables reliably predicted drug court outcomes. Having received a

sanction involving incarceration or detention significantly decreased the odds of

graduating by 39.0% compared to those who did not receive such a sanction (p = .000, B

= -0.494, ExpB = 0.610). The results of the forward stepwise logistic regression show

that having received a sanction involving incarceration or detention is the second most

100

important variable in contributing to overall model performance. Having received an

“other” type sanction decreased the odds of graduation by 35.7% compared to those who

did not (p = .000, B = -0.441, ExpB = 0.643). Each additional sanction received

decreased the odds of graduation by 6.1% (p = .000, B = -0.063, ExpB = 0.939).

The remaining sanction types, treatment or treatment-related sanctions (p = .000,

r = -0.084), community service sanctions (p = .227, r = -0.020), suspension or demotion

sanctions (p = .742, r = 0.006), and increased program elements (p = .060, r = 0.032)

failed to show meaningful relationships with drug court completion outcomes and were

excluded from multivariate analyses to test the hypotheses.

Hypothesis three. Reject the null hypothesis. Both the track through which a

participant enters drug court and the time a program is operational are meaningfully

related to completion outcomes, reliability predict completion outcomes, and increase,

although slightly, overall model performance. The prediction that entering through the

diversion track results in an increase in the odds of graduating is supported. The results

show a negative relationship between in the amount of time a drug court is in operation

and completion outcomes. The forward stepwise logistic regression model shows that the

number of months the program is in operation is the third most important variable

contributing to model performance.

While this null hypothesis is rejected, including these variables in the presence of

participant characteristic and problem behaviors resulted in only modest gains in model

performance and ultimately a poor fitting model. Nagelkerke’s R shows that the amount

of variance explained increased by less than three percentage points, and overall model

accuracy increased by approximately one percentage point. These findings suggest that

101

although these two variables reliability predict completion outcomes, they negatively

influence the overall fit of the model to the data, possibly resulting from relationships or

interactions among other predictor variables. Results of the specific tests for these two

variables are outlined below.

Track. Track was found related to program completion outcomes in variable

selection tests (p = .000, r = -0.101) and demonstrates that entering drug court through

the probation track decreased the probability of graduating by 39.3% when compared to

those entering through the diversion track (p = .000, B = -499, ExpB = 0.607) and was

entered ninth of 13 variables in the forward logistic regression model, Model II.

Time program in operation. Time in operation was shown to be meaningfully

related to completion outcomes (p = .000 , r = -0.100), and the likelihood of graduating

decreased by 0.5% for every month increase a program has been in operation (p = .000, B

= -0.005, ExpB = 0.995). The forward stepwise logistic regression results show that the

time the drug court has been in operation is the third most important variable contributing

to model performance.

Summary

Based on results across models, the top predictors for Kentucky Drug Court

completion outcomes are age, level of education, carrying charges related to the

administration of justice, the number of sanctions one receives and receiving a sanction

for incarceration. To arrive at this conclusion, the results from all models and the

consistency of the results across all models were explored. Two sets of analyses were run.

First, bivariate analyses were run to describe the population under study and assess all

variables of interest for inclusion into the model. For variable inclusion the tests for

102

statistical significance relied upon a relaxed alpha value, α = .10, supplemented by a

Pearson’s r value of at least +/- .1. This method was used to identify variables that were

meaningfully related to the outcome variable. Bivariate tests of relationships among

predictor variables were also run to determine if Kentucky Drug Court data also showed

relationships between predictor variables as found in previous studies and also test for

multicollinearity. Variables tested in this step showed significant relationships, but only

one variable showed signs of multicollinearity. When assessing analyses between

predictor variables, a Pearson’s r of +/- .5, VIF >10 and tolerance score <.1 were used as

the criteria for multicollinearity. One variable, the number of drugs of choice, was

removed for that reason.

Second, after variable selection was complete, multivariate analyses were run to

determine a set of predictors for drug court completion outcomes. Variables that were

meaningfully related and not collinear with other predictors were entered into an enter

method logistic regression model. Model I suggested issues with the inclusion of the drug

court variables, time in operation and track. To determine the importance of each

predictor, a forward stepwise logistic regression model, Model II, was run. Since the

results of Model I showed poor overall performance at the final block and suggested that

the drug court variables created excessive noise within the model, an additional forward

stepwise logistic regression, Model III, was run without the drug court variables.

Identical to Model I, block II, this model showed a good model fit. This confirmed that

the drug court variables are problematic. The level of importance among the variables

between Model II and Model III did not change significantly, showing that these

variables did not interact in a major way for the top performing variables. A fourth

103

stepwise logistic regression with all variables under study was run to examine the

possibility that important and relevant variables were excluded as the result of the

variable selection process. With all variables entered, 20 of the 30 variables were retained

and resulted in increased in model accuracy and variance explained. The full forward

stepwise model resulted in a poor fitting model, which suggests that the model is noisy.

Eight previously excluded variables emerged as significant contributors to model

performance, and one variable was dropped as a predictor. The top predictors were

generally the same as in previous models, but with the number of sanctions greatly

elevated in the order of importance. This suggests that reducing the variables during the

bivariate selection process did not influence the level of importance across the top

predictors in a major way. However, variable selection practices excluded predictors that

may be important to understanding drug court outcomes, and included a predictor that

may not be as important to predicting outcomes as thought given the results of the

bivariate analyses. These results point to interaction effects between predictor variables.

104

CHAPTER V

DISCUSSION

Understanding drug court completion outcomes is important for understanding if,

how, and for whom drug court may work for reducing both recidivism and relapse. Post

program completion studies that seek to understand the impacts of drug court but are

based on graduates include only a select group of individuals. It is quite possible that

those who successfully pass through drug court differ in very significant ways from those

who do not. After all, those who graduate show evidence of success at staying sober and

behaving in socially acceptable ways, otherwise, they would not graduate. Using

Kentucky Drug Court data, all of the hypotheses tested in this study were supported.

There are qualities about participants, their problem behaviors, and drug courts that

differentiate between graduates and those who fail to graduate.

Summary of Findings

The purpose of this study is to identify factors related to drug court program

completion outcomes using data obtained from Kentucky Drug Court. The research

question for this study is answered and, in general, is consistent with results of prior

studies. Some findings, however, were unexpected while others are new additions to the

literature. Of the original 31 variables included for study, 13 were meaningfully related

and predictive of completion outcomes. Of participant characteristics, age, education, and

marital status were related to and predictive of completion outcomes. Methamphetamine

105

or other stimulant for a drug of choice was the only drug of choice types related to and

predictive of completion outcomes. Several charge types were found related to and

predictive of completion outcome, specifically, charges relating the sale, trafficking, or

manufacturing of drugs, drug possession, administration of justice, and property crimes.

Having received a sanction involving incarceration or “other” types, as well as the

number of sanctions, were also related and predicted outcomes. Both drug court

variables, the track through which on entered drug court and how long a drug court was

in operation, were also related and predictive, although problematic when entered into

models with other drug court variables.

Factors predictive of drug court completion outcomes were found to be related to

each other, causing the nature of these relationships to be complex. For example, age was

found to be the most important variable contributing to model performance across all

models. Age was also found significantly related to a number of other predictor variables,

time in operation and the track through which one enters drug court. The drug court

variables individually contributed to the prediction of outcomes, but when included with

other drug court variables, caused model performance to drop and ultimately poor fitting

models. Tests of relationships revealed that age was significantly related to drug court

program variables, but in the opposite direction if they were to support the observed

relationship between drug court program variables and drug court completion outcomes.

Tension between predictor variables was also found for relationships between charge

type, sanctions, and track and completion outcomes. Given these findings, and

considering that including drug court variables resulted in decreased model performance,

drug court variables were determined to add noise into the model. Drug court completion

106

outcomes, when identifying a set of predictors, are better predicted in the absence of

these variables.

Because of the complexity of the relationships between the variables under study,

the suspicions of interaction effects, and the exclusion of potentially relevant variables,

an exploratory forward stepwise model was run. The changes in the order of importance

and changes in odds ratios between all models confirmed the presence of interaction

effects and also pointed to the presence of mediating and/or moderating variables, further

demonstrating the complexity surrounding both prediction techniques and factors leading

to successful completion of drug court.

However complex, some stability in predictors was found. Across all four of the

models run for this research, the top predictors were relatively stable and include mainly

participant characteristics and problem behaviors; age, education, carrying charges

related to the administration of justice or property crimes. However, two sanction-related

variables fluctuate in levels of importance. Both receiving an incarceration sanction and

the number of sanctions received arrived within the top two predictors but in different

models, suggesting that they are both important to understanding outcomes, but that they

also interact with other variables.

Response to the Research Question

Prior studies suggest that demographic factors such as race, sex, marital status,

age, education, and employment are related to completion outcomes. Of those factors, the

current study found that marital status, age, and education predicted program completion.

Analyses show that being married, being older, and having at least a high school diploma

makes graduation predictably more likely. That race and sex were not found related to

107

program completion in bivariate tests is an operationally significant finding. This study

found that sex was neither meaningfully related to nor predictive of completion outcomes

in any of the analyses. These findings are contrary to the increased-responsibilities

argument offered in the statement of the hypotheses. This shows that there are no direct

detectable patterns or relationships between sex and outcomes, which suggests that

programming may be equally effective between males and females. This may also

suggest that Kentucky Drug Court has effectively adapted programming around

perceived barriers associated sex. On the other hand, that race was not meaningfully

related in bivariate tests of relationships but emerged as a significant predictor in the full

forward stepwise model, although not among the strongest of predictors, suggests that

race may play an important part in understand outcomes for Kentucky Drug Court, but

through the interactions with other variables.

The findings regarding participant characteristics and drug court completion

outcomes could support a social bond perspective (Sampson & Laub, 1993). This

perspective suggests that patterns in criminality and conformity change throughout the

life course. Important to this study, a social bond perspective suggests that criminal

trajectories can be interrupted. One way this perspective sheds light on the findings is

that those who do well in drug court may have more social capital than those who do not

do well. The bonds associated with marriage and education may be important enough to

the participant that the costs and consequences of failing drug court extend beyond just

conviction and incarceration, but ultimately include the loss of the bond. In essence, both

marriage and education represent stakes in conformity that the participant may not want

to risk losing.

108

Another way Sampson and Laub’s adult social bond perspective explains the

findings is that drug court set the context for changes in criminal trajectories; drug courts

are turning points. Perhaps those who enter drug court who are also married or educated

are at a threshold between conformity and criminality but currently on the trajectory of

criminality. Drug court’s influence for these participants is toward conformity, possibly

helping participants see the value of those adult social bonds.

Perhaps both marriage and education are an artifact of age, therefore also

explaining why age is the single most important predictor of completion outcomes. It

could be that selection process for drug court results in the selection of those who are in

the process of aging out of criminal behaviors anyway. This explanation is in line with

Moffitt’s (1993) theory regarding adolescent-limited offending. However, in Moffitt’s

theory, life course persistent offenders comprise a minority of the individuals involved in

criminal behaviors. In this study, the overall graduation rate is less than 30%. If Moffitt’s

theory is a viable explanation for the findings, this suggests that an age-crime curve for

drug court participants is wider than a general age-crime curve and/or that drug court also

serves a large proportion of life course persistent offenders.

The current research also finds that criminal charges, sanctions, and drugs of

choice show utility for predicting completion outcomes, but further study is needed as

these relationships may be associated with level of risk, which was not measured here.

Drugs of choice inform upon drug using behavior, charge types inform upon criminal

behavior, and sanctions provide a look into a participant’s ability to comply with program

rules. These are the concerns of drug court: drugs, crime, and compliance. These

109

behaviors are at the core of the drug court program. That they are related to each other

and to drug court outcomes is not a surprise.

The lack of significant findings across the majority of drugs of choice suggests

that Kentucky Drug Court is effective for managing a broad range of drug addictions, or

perhaps considering that the overall graduation rate is less than 30%, that Kentucky Drug

Court is equally ineffective across the majority of the drug types. That methamphetamine

or other stimulants was the only significant drug of choice predictor for completion

outcomes suggests that methamphetamines or other stimulants pose a unique concern for

Kentucky Drug Court. However, this relationship is positive, meaning that those

indicating methamphetamine or other stimulant show increased likelihood of graduating.

Only 18.3% of the population under study indicated a problem with this drug type,

making it the second least prevalent drug type indicated. Tests of relationships with

methamphetamine or other stimulant as a drug of choice failed to show a significant

relationship with age, which excludes an aged-based explanation for this specific finding.

Test of relationships, however, showed a significant relationship between

methamphetamine or other stimulant as a drug of choice and the sale, trafficking, and

manufacturing of drugs. This finding is not unexpected as the use of methamphetamine

could logically be associated with the sale, trafficking, and manufacturing of

methamphetamine. The interesting observation is that both these variables show a

positive relationships with completion outcomes, which could be seen as contrary to the

risk principle if one views the sale, trafficking and manufacturing of drugs a crime

indicative of a deeper involvement in crime than other charge types such as traffic and

motor vehicle offenses, and property crimes. Maybe participants indicating this drug type

110

are actually less risky and less involved in a criminal lifestyle than previously thought.

Alternatively, maybe Kentucky Drug Court’s programming is more effective for those

with problems with methamphetamine or other stimulant than the other drug types.

Further investigation is warranted to explain these findings.

This study also explored various charge types to determine if certain kinds of

charges are associated with completion outcomes. That carrying a charge related to the

administration of justice is the strongest predictor among all the charge types and in the

top three strongest of all predictors while charges relating to the sale, trafficking, and

manufacturing is among the bottom half in variable importance is unexpected and counter

to arguments regarding level of risk. As discussed in the previous paragraphs, carrying a

charge related to the sale, trafficking, and manufacturing of drugs is associated with a

62.6% increase in the odds of program graduation. When considering that charge types

may represent a relative level of risk, crimes such as the sale, trafficking, and

manufacturing of drugs or crimes against a person could be indicative of a higher degree

of risk. Subsequently, these charge types should be strongly and negatively associated

with outcomes. Risk principle would also suggest that crimes against the administration

of justice are relatively minor. After all, one could argue that the kinds of bad acts in that

charge type do not result in physical, financial, or emotional harm to a person or the

propagation of other criminal behaviors. However, this study found the opposite: that

carrying charges related to these bad acts results in worse outcomes. Individual crimes

such as probation/parole violations, fleeing or evading a police officer, failing to appear

in court, and tampering with evidence are included in the category for crimes against the

administration of justice. This may suggest that a pattern of behavior relating to

111

violations against the administration of justice or general issues with authority continues

within drug court, and that this type of behavior poses a greater risk to program

performance than the behavior involved in crimes against people or property.

The influence of sanctions also proved worthy of note. First, receiving a sanction

of incarceration significantly changed the odds of program completion, but receiving a

sanction related to treatment did not. Participants may receive treatment as a program

requirement and treatment as a form punishment differently. Distinguishing the impact of

treatment as a programming requirement from treatment as a sanction would be

challenging. Given that receiving an incarceration sanction resulted in a decreased odds

of graduation, perhaps incarceration is accepted and received as punishment while

treatment as a form of punishment confuses the purpose. No predictions were made on

the effects of sanctions involving treatment, as it is unstudied in the literature. Treatment

as a form of punishment warrants further study.

The fact this current study is not the first to find a negative relationship between

outcomes and receiving a jail sanction (see Belenko, 2001) may suggest that

incarceration is contraindicated for the drug-addicted population. Although Walter

Reckless’s containment theory was met with much criticism (see Cullen & Agnew,

2003), maybe these concepts or parts of the theory warrant a revisit. Drug court

participants arrive in drug court with behavioral evidence that their “controls” are

compromised; they are involved in both crime and drug abuse. Drug court can be viewed

as a reinforcing the outer control system and, when taken out of this reinforcing system

and placed into a criminogenic one, incarceration “pushes” a participant back on the

trajectory of criminality. Those who do not receive incarceration do not experience that

112

“push,” also explaining why those who do not receive this sanction are more likely to

graduate.

Nearly 20% of the population under study received a sanction documented as

“other” and receiving a sanction documented as “other” decreased the odds of graduating

from drug court by 35.0%. “Other” is not defined and therefore cannot be explained.

Perhaps these other sanctions are creative or non-traditional sanctions that do not fit well

into the other categories. If this is the case, then these creative sanctions, when assessed

across all Kentucky Drug Courts, may not be effective at deterring future non-compliance

or that these sanctions may be stigmatizing. Future research should attempt to flesh out

what qualities about this “other” type of sanction may produce the decreases in likelihood

of graduation.

There are two issues to take into account about the sanction variables. First, data

entry of sanctions is generally inconsistent, but appears to be more consistent for severe

sanctions. Comparing sanctions involving incarceration with all the other sanction types

may not be appropriate. Second, the incarceration sanction variable may represent official

reactions to severe non-compliance while the other type sanctions represent official

reaction to less severe non-compliance. If this is the case, then the findings suggest that

severe non-compliance coupled with a severe sanction results in negative outcomes.

What would happen if severe non-compliance was met with something other than a

severe sanction? If the goal of drug court is to keep a participant out of jail and prison

because the drug abuse is the root cause, and drug court is offered in the hopes that the

resultant negative consequences of jail and prison never occur, then using jail as a

sanction may defeat the purpose.

113

Studying the influence of the drug court itself, in terms of track and time in

operation, is a new addition to the literature. That probation is associated with a

decreased likelihood of graduation when compared to diversion may suggest that the

threat of a prison or jail stay may not be effective at producing compliance. Since those

on the diversion track risk receiving both a conviction and the imposition of a prison or

jail term, while those on the probation track already experienced the conviction, it is

possible that avoiding the conviction is important enough to encourage compliance.

However, those on the diversion track may also carry a lower level of general risk to

begin with, suggesting that they are more successful because they are less risky. This

may also suggest that this population could perform just as well with less intense

supervision. This finding warrants further investigation to determine causal factors

between track and completion outcomes.

This study also found that the longer a drug court is in operation the lower the

likelihood of graduating. This begs the question, what about drug court’s operations over

time might decrease the odds of graduation? It is possible that as drug courts opened,

staff and leadership were hesitant to terminate a participant and over time, this hesitancy

declined? Perhaps, as noted in the literature review, the drug court increasingly began to

accept higher risk participants, who are more likely to be terminated. In this respect, the

decrease in graduation rates could be due to the increase in acceptable level of risk for

entry into drug court. However, further study is needed to explore the operational impacts

of the drug court program on completion rates and possible threshold effects of time in

operation.

114

Policy and Program Implications

Caution should be used when assessing the findings from this study for program

and policy implications on Kentucky Drug Court. The statistics used were very powerful,

resulting in very small relationships and small changes in the odds of graduating being

statistically significant. The most relevant example of this is for the relationship between

age and completion outcomes. Of the variables that were meaningfully related to drug

court completion outcomes, age showed the strongest relationship in bivariate analyses

yet this relationship is considered small when examining the effect size (r =.216). Age

was also considered the most important predictor in contributing to model performance,

evidenced by first entry in the forward entry stepwise models. Moreover, bivariate

correlations between other variables under study and completion outcomes showed that

some variables were statistically significant but the size of the relationships were so small

that they were not practical for further analyses.

Since the top predictors were generally stable across all the analyses, Kentucky

Drug Court could focus more detailed analysis efforts on those areas related to age,

education, carrying charges related to the administration of justice , the number of

sanctions, and sanctions involving incarceration. Kentucky Drug Court already frames

programming around many of these factors. Examining these areas to determine the exact

nature of the relationship to outcomes and subsequently, how programming could be

modified to address them, should be not be overly difficult. Theory can be useful for

explaining the observations, which is why theory is vitally important to program

operations and why the lack thereof in drug court research is a concern. If age is the

factor and the “why” is social capital and social bonds, programming and policy ought to

115

focus on those things that influence social capital and social bonds. There could be a

number of reasons why younger participants would have more difficult time completing

the program, for example finding a job when they have few job skills, displaying a poor

attitude, or having difficulty with transportation. A focus on education, job skills training,

mass transit familiarization, car pooling, or even drug court mentoring programs could be

solutions. The same reasoning could apply to marital status and education. Drug courts

could focus attention on interpersonal relationships by encouraging prosocial

relationships and formal education. Kentucky Drug Court shows evidence of already

considering these types of factors. The “drug court divorce” mentioned previously is an

example of discouraging a “low” quality bond with known antisocial individuals.

Additionally, formal education requirements are already built into Kentucky Drug Court

requirements.

Drugs of choice, charge types, and sanctions appear to be wrapped around the

concept of risk. This is not a new finding for the corrections discipline. Probation and

parole organizations have attempted to gauge level of risk using various measures, for

example the Level of Service Inventory-Revised (LSI-R). The LSI-R shows predictive

capability for drug court (see Shaffer, 2010), and Kentucky Drug Court should consider a

criminal risk measure as part of their intake procedures in addition to the Addiction

Severity Index. Kentucky Drug Court should also consider entering this information into

the management information systems and make it available for future studies. This may

aid in the selection and screening process to eliminate individuals with criminal histories

or social circumstances the drug court may have difficulty supporting.

116

Drug court staff should address sanctioning practices and data entry of sanctions.

It is no surprise that the more sanctions one receives, the less likely graduation. That

receiving a sanction of incarceration translates into worse outcomes is also not a surprise.

If the “why” of this finding suggests data entry bias, then the data lacks reliability and is

not valid as a measure of all sanctions. If however, sanctions involving incarceration are

generally reliable at data input, but measure official response to severe non-compliance,

then the findings suggest that jail may not be an effective deterrent for this population.

Maybe those participants who receive a jail sanction are more risky than participants who

do not receive a jail sanction or perhaps this type of sanction is not effective at forcing

compliance. If so, program non-compliance should be addressed in some way other than

jail or home detention. Treatment, for example, could be a viable alternative, but this too

could confuse the purpose of the action, as quite possibly punishing someone with

treatment may not “feel” like punishment. Further study is needed in this area.

Framing programming around the drug court variables studied in this research

could be problematic. Drug court cannot change the length of time it has been in

operation and cannot change the track through which a participant enters drug court.

However, drug court staff can try to be mindful to changes over time in the management

of participants and mindful of the varying needs for participants between those on

probation and diversion. Perhaps by tracking their sanctioning behaviors, monitoring

stock and flow rates through the program, and assessing outcomes a program can identify

successful strategies for helping participants reach graduation.

117

Limitations

Although this study found significant predictors of program completion outcomes,

the study suffers from several limitations. First, the variable selection process eliminated

potentially important variables. In this study, sex and race were eliminated from

multivariate analyses for the tests of hypotheses, as they failed to show a meaningful

relationship to graduation. However, both race and gender are often found related to drug

court outcomes, leaving their absence in the multivariate model questionable. When

entered into a full forward stepwise model, Model IV, race and seven other variables

emerged as important predictors. When using bivariate tests of relationship alone,

valuable information on underlying phenomenon affecting completion outcomes is

missed and resultant programming or policy changes potentially misguided. On the other

hand, the variable selection was necessary to reduce the amount of noise and build the

most parsimonious model (Menard, 2010), which is evidenced in this study as the two

drug court variables were problematic and increased standard errors when using all

variables without regard to variable selection.

The power of the statistics is also a concern. The statistics in this study were very

powerful, finding even trivial relationships and differences statistically significant. If not

careful, basing operational decisions on statistical significance alone may translate into

wasted resources. To accommodate the very large sample size, considerations of effect

size supplemented tests of statistical significance for variable selection. Even after using

statistical significance in combination with the effect size for variable selection, small

changes in the odds ratios were statistically significant. Modifying program features or

118

rewriting policy would not be wise if the problems it was designed to solve were quite

small and only affected a small segment of the population.

Uneven group size in the dependent variable is another potential limitation.

Seventy percent of the total population under study failed to complete drug court, leaving

30% in the category for graduation. Across all models, the percent correctly predicted for

graduates was between 35.6% and 42.2% while percent correctly predicted for non-

graduates ranged between 89.9% and 91.0%. With significantly more observations, the

ability of the statistics to detect relationships increases, thus lending to better prediction

for non-graduates.

Several limitations related to the variables exist. Kentucky Drug Court allows for

multiple drugs of choice to be entered and does not indicate which one is most

problematic. As a result, there is no way of knowing whether a listed drug of choice was

casually used (drug abuse) or if the drug caused significant life problems due to trying

quit or trying to continue use (drug addiction). Moreover, potentially important variables

such as a measure of risk, employment and the presence of mental illness were not

included in the study. These variables, either directly or indirectly, may be important to

understanding completion outcomes as discussed in the literature review. Place or

location may also play an important role in drug court, but was not included as part of

this study. Parts of eastern Kentucky lie in an area identified as a high intensity drug

trafficking area. It is possible that, across the entire state, certain places have significantly

different patterns for drugs of choice or charge types, for example. This would suggest

that program needs may also vary.

119

Future Studies

Drug courts are unique to their operating environments and therefore research on

the geography of drug courts would be a useful addition to the literature. The current

study assumes spatial continuity of the findings across the entire state. What drug court

outcomes look like for drug courts located in the high intensity drug trafficking area will

be of interest for Kentucky Drug Court leadership. Do these drug courts experience worse

outcomes? Are drugs of choice the same across the state? Exploring regional variation in

strongest predictors of outcomes would be a useful addition to the drug addiction and

geography literature.

Future studies for drug court outcomes should include a general measure of risk,

such as gained through the use of the LSI-R. As found in this study, drugs of choice,

charge type, and sanctions appear to center around the concept of risk. Developing a risk

profile or some construction of a risk indicator would allow researchers to tease out the

effects of demographic and program progress variables on completion outcomes while

controlling and assessing level of risk.

Future direction in drug court research could also include the use of data mining

methods. The results in this study show that identifying drug court completion outcomes

is not a straightforward and simple task and highlight the importance of moving beyond

traditional bivariate tests of relationships and a priori identification of meaningful

predictors. Future research on drug court outcomes should include robust techniques that

are capable of identifying interaction effects among predictor variables and compensating

for unbalanced group sizes in the dependent variable. Such methods may uncover

previously unidentified relationships and be a useful addition to the literature. In this

120

manner, identifying factors related to completion outcomes would serve as the grounding

point for future study or theory development. Alternatively, these results also highlight

the need to move cautiously toward these data mining methods, as the models produced

by including all variables, even those with trivial or no bivariate relationships, can be

noisy and poor fitting.

Summary

The core criticisms of drug court as a correctional program surround a few key

issues. First, many authors argue that it was founded without theoretical justifications.

Second, drug courts scope widened over the years. Lastly, the number of drug courts

quickly rose in the absence of quality assessments. Placed in the wider context of crime

and criminality, this study offers a theoretical base for Kentucky’s Drug Court, although

theory generation was not the purpose of this study. Considering the findings in this study

and Kentucky Drug Court’s operations, the theory that could support why traditional

correlates of crime are found related to drug court outcomes and why drug courts may be

a useful correctional program is Sampson and Laub’s (1993) social bond theory; the

salient concepts being social capital and turning points. Drug court takes individuals who

are on a path of criminality and attempts to redirect. Kentucky Drug Court programming

attempts to redirect by supporting, encouraging, and in some cases requiring the

generation of turning points. These include, but are not limited to, completing high

school, getting a job, becoming engaged in prosocial activities and surrounded by

prosocial peers, and getting and staying sober.

The lack of underlying theory may have contributed, at least partially, to the

second criticism, that the scope of drug courts target population widened. Drug courts

121

were initially designed as a diversionary program for the low risk drug offenders, but as

evidenced in Kentucky Drug Court, now frequently serve those who may be higher risk

and probationers. This study focused on only felony drug court. Kentucky also runs

misdemeanor drug courts, which generally follow very similar programming as the

felony drug courts. These two populations are treated very differently by justice policies

but within drug court, they are managed the same.

Criticisms on the lack of quality assessments and drug court research in general

center on lack of statistical rigor, data quality, units of measure, small sample sizes, and

time frames. This current study, although including multiple indicators of statistical

significance, suffered from issues with data quality and units of measure. These issues,

however, may stem from the fact that Kentucky Drug Court management information

system was designed for operational purposes, not necessarily for the express purpose of

supporting scholarly research. This study utilized a large number of observations

captured from a period spanning three years. A greater number of observations is

generally a more desirable situation than statistical analysis with few observations,

however, caution must be used in these circumstances as small and trivial relationships

are statistically significant. Small and trivial, yet statistically significant relationships

between the predictors and also between the predictors and the outcome adds complexity

and noise in statistical models.

Considering that drug courts are popular and continue to grow in number despite

of lack of quality assessments, perhaps we are seeing a shift in the way criminal justice

policy makers perceive the role of punishment. The body of literature on drug court,

including the works critical of drug court suggests this: judicial leadership, relying on

122

instinct and experience rather than scholarly education, came up with a good idea and

acted upon it. It “worked.” Others saw that it “worked” and the idea grew unconstrained

and without proper strategic guidance and planning. However, the results from this study

and others like it potentially confirm the instincts of judicial leadership—that something

other than jail and prison could move an individual out of a criminal trajectory. Over

time, however, as drug court grew in popularity and in number, it also grew is scope and

began serving other types of offenders for which it never intended to serve (i.e., high risk

and probationers). Placing drug court in a proper theoretical perspective will allow for a

more clear and attainable strategic plan, support the development of measurable goals,

and allow drug court leadership to manage program scope.

Scope-creep withstanding, many drug court researchers and practitioners assert

that drug courts “work” for this more risky population too. As the numbers of traditional

drug courts and other specialized drug courts continue to grow, and, as with Kentucky

Drug Court, misdemeanor and felony level participants are managed much the same,

perhaps criminal justice scholars and policy makers need to take a pause and think

through how, why, and who we punish. Discussions on the purpose of punishment, what,

and who society is actually punishing, and if some people or some acts are more or less

deserving of punishment, need to be reinvigorated.

Factors That Influence Program Completion_ Predicting Kentucky

Students also viewed