EVALUATING AN INTENSIVE SUPERVISION PROBATION PROGRAM IN A COLLEGE CITY: REVISITING THE DEBATE BETWEEN REHABILITATION AND CONTROL By Chelsea Breanne Diem A THESIS Submitted to Michigan State University in partial fulfillment of the requirements for the degree of MASTER OF SCIENCE Criminal Justice 2011 ABSTRACT EVALUATING AN INTENSIVE SUPERVISION PROBATION PROGRAM IN A COLLEGE CITY: REVISITING THE DEBATE BETWEEN REHABILITATION AND CONTROL By Chelsea Breanne Diem This study evaluated an intensive supervision probation (ISP) program that heavily emphasized both treatment and surveillance by requiring its offenders to attend repeat motion hearings in front of the sentencing judge to prove their compliance in the treatment-oriented conditions of the program. The experimental group consisted of 77 ISP participants and was matched based on propensity scores and a kernel matching algorithm with 148 regular probationers. Methods of analysis include the average treatment effect to determine the relationship between the program status of the offender (ISP verses regular probation) and likelihood of re-arrest and a survival analysis to determine the relationship between program status and time until re-arrest. Results indicate that the ISP did not reduce the likelihood of rearrest for the ISP participants as compared to regular probationers. Findings also show that the ISP did not reduce the time until failure for the ISP participants, before or after inverse probability of treatment weighting. Ideally, additional data such as results of motion hearings and severity of re-arrest would be conducted before presenting finalized results on the program‟s effectiveness. DEDICATION I dedicate this piece of work to five people: Above all, to Dr. Christopher Melde, who somehow always managed to find a way to explain a complicated concept and who tirelessly listened to me toss around ideas and thoughts. I could only hope to be half of the criminal justice professional that Dr. Melde is someday. I also dedicate this to my family – my mother, Michelle, who took the brunt of my stress-related frustrations with patience and love; my father, Kenneth, who provided unconditional support throughout this writing process; my sister, Kylie, for distracting me when I needed to talk about anything other than schoolwork; and to my brother-in-law, Brian, who helped me “find my thoughts” when I misplaced them. iii ACKNOWLEDGMENTS I would like to acknowledge several people for helping me in this creation. Dr. Steven Chermak inspired me to go after what I wanted and provided support in my most difficult times. Dr. Jesenia Pizarro showed me how persistence and perseverance can influence one‟s work and I will always be grateful to her. Dr. Beth Huebner provided statistical support to me, even though we had never met and she had no reason to do so. My office mates deserve the greatest thanks, as they dealt with me throughout the year: Michael Rossler, Jaemia Pratt, and Benjamin Rosek. And to my best friends, for loving me with or without this thesis: Jennifer Adams, Rachael McCabe, Amber Bahnweg, Lisa Sutton, Erika Smith, my second mom, Dianne Reeves, and my favorite internship supervisor, Matthew Brundage. Tim Homberg also deserves appreciation, as he kept me sane throughout my graduate school career. To the person I could burn the most aggression with through sweat, tears and laughter, my running partner, Emily Dawson: I would not have survived these past two years without you. Last but not least, I owe this entire thesis to the Judge, Court Administrator, and Probation Officers at the court which provided this data; without their efforts to reduce the occurrence of a potentially-fatal crime, this evaluation would not have existed. iv TABLE OF CONTENTS LIST OF TABLES ..................................................................................................................... vii LIST OF FIGURES ................................................................................................................... viii INTRODUCTION ..................................................................................................................... 1 INTENSIVE SUPERVISION PROBATION PROGRAMS ..................................................... 4 Characteristics of ISPs ................................................................................................... 4 Goals and Outcome Measures of ISPs ........................................................................... 5 THEORETICAL FRAMEWORK ............................................................................................. 7 ISPS AND RECIDIVISM.......................................................................................................... 10 Surveillance versus Treatment Studies .......................................................................... 10 Coercive Treatment Effects ........................................................................................... 11 DESISTANCE FROM CRIME ................................................................................................. 14 Time to Failure Studies .................................................................................................. 15 Limitations of Research ................................................................................................. 16 CURRENT STUDY‟S ISP ........................................................................................................ 17 Implementation and Target Population .......................................................................... 17 A Typical ISP Sentence ................................................................................................. 17 Regular Probation .......................................................................................................... 19 DATA COLLECTION METHODS .......................................................................................... 20 Dependent Variables and Outcome Measures ............................................................... 21 Independent Variables ................................................................................................... 21 DESCRIPTION OF POPULATION ......................................................................................... 23 ISP/Experimental Group ................................................................................................ 23 Regular Probation/Comparison Group .......................................................................... 23 DATA ANALYSIS METHODS ............................................................................................... 26 LIKELIHOOD OF RE-ARREST METHODS .......................................................................... 27 Propensity Scores, Kernel Matching, and Average Treatment Effect Methods ............ 27 TIME UNTIL FAILURE METHODS....................................................................................... 30 Survival Analysis Methods ............................................................................................ 30 Kaplan-Meier Estimates and Nelson-Aalen Estimates Methods ................................... 31 Cox Proportional Hazards Model Methods ................................................................... 32 v RESULTS .................................................................................................................................. 33 Bivariate Statistics ......................................................................................................... 33 LIKELIHOOD OF RE-ARREST RESULTS ............................................................................ 35 Propensity Scores, Kernel Matching, and Average Treatment Effect Results .............. 35 TIME UNTIL FAILURE FINDINGS ....................................................................................... 40 Kaplan-Meier Estimates and Nelson-Aalen Estimates Results .................................... 40 Cox Proportional Hazards Results ................................................................................. 44 DISCUSSION AND CONCLUSIONS ..................................................................................... 48 WORKS CITED ........................................................................................................................ 53 vi LIST OF TABLES Table 1. Descriptive Statistics.................................................................................................... 25 Table 2. Bivariate Statistics on Recidivism within 17 Months as Dependent Variable ............ 34 Table 3. Bias Diagnostics for ISP Participants versus Regular Probationers ............................ 38 Table 4. Cox Proportional Hazards Models ............................................................................... 47 vii LIST OF FIGURES Figure 1. Propensity Score Distribution for ISP Participants versus Regular Probationers ...... 36 Figure 2. Kaplan Meier Survival Estimates of Unweighted Data ............................................. 41 Figure 3. Kaplan Meier Survival Estimates of Weighted Data ................................................. 42 Figure 4. Nelson-Aalen Estimates of Unweighted Data ........................................................... 43 Figure 5. Nelson-Aalen Estimates of Weighted Data ................................................................ 44 viii Introduction Scholars have debated the role of probation for decades. The supervision philosophies related to sentencing tend to reflect the political nature of the times. In the 1960s and early 1970s, the goal of sentencing was largely focused on rehabilitating offenders; this switched to a more crime-control philosophy in the late 1970s and 1980s (Fulton, Stone, & Gendreau, 1994; Farabee, 2005; Caplan, 2006; Brown, 2007; Andrews, Zinger, Hodge, Bonta, Gendreau, & Cullen, 1990; Tonry & Lynch, 1996; Zimring & Hawkins, 1995). The movement from rehabilitation to control is often tied to Martinson‟s (1974) report on the failure of rehabilitation to reduce offender recidivism (Fulton et al., 1994; Farabee, 2005; Andrews et al., 1990); however, several factors aided in the transition. Caplan (2006) noted that around that time, the public began to advocate harsher punishments for criminals, state budgets were declining, and a general diminishing of faith in the treatment of offenders also led to the shift from rehabilitation to control. Similarly, Andrews and colleagues (1990) identified a shift in public opinion toward crime control and away from offender leniency and rehabilitation. Zimring and Hawkins (1995) noted that the shift was so effective partly due to the extent of the negative portrayal of rehabilitation. The emphasis on a punishment-oriented crime control philosophy helped to increase prison populations causing a need for the development of alternative sanctions for offenders (Byrne, 1986; Fulton et al., 1994; Tonry & Lynch, 1996; Petersilia, 1998). The „just deserts‟ philosophy further promoted the push for intermediate sanctions (Byrne, 1990; Fulton et al., 1994; Tonry & Lynch, 1996). Such sanctions provided a way for the court system to tailor punishment closer to the individual offenders‟ needs, allowing more sentencing options than traditional probation (which is often seen as too lenient) and prison (too harsh). One popular 1 intermediate sanction proposed was the idea of an intensive supervision probation or parole (ISP) program. These programs were not new in the 1980s – Brown (2007) noted that ISPs had roots dating back to the 1950s – but they provided a sentencing option that was still consistent with the „get-tough‟ policy of the time and could help reduce prison overcrowding (Byrne, 1986). The literature on ISPs tend to categorize the programs by their primary focus – those that emphasize treatment more heavily are referred to as treatment-oriented while those that emphasize strict control are referred to as surveillance-oriented. While it may be that ISPs focusing on treatment inherently have both treatment and surveillance components, program evaluations still tend to categorize these as treatment-oriented, furthering the dichotomy between treatment and control programs. Currently, there is still much debate as to whether ISPs are effective, and if so, which emphasis aids in the greatest reduction in recidivism. A myriad of studies have been conducted focusing on this debate, which provide the basis for this study. Unfortunately, however, “…published evaluations of programs that combine intensive supervision and rehabilitative treatment are few” (emphasis added; Lowenkamp, Flores, Holsinger, Makarios, & Latessa, 2010, p. 370). The current study helps fill this void by evaluating the impact of an ISP program that heavily emphasizes both supervision and rehabilitation. The ISP program began in a college city in 2006 and targeted high-risk drunk-driving offenders. The fundamental feature of the ISP program is its requirement that the participants attend repeated motion hearings in front of the judge to demonstrate their compliance in the treatment-oriented components of the ISP. Failure to demonstrate compliance can result in a 1-20 day segment of jail, depending on the severity of the violation. This is unlike other ISPs, as most programs do not require repeated accountability of treatment status. The research questions for this study, then, are as follows: Does this 2 particular ISP program, which heavily emphasized both rehabilitation and control, reduce the likelihood of re-arrest? Is the overall timing of re-arrest affected by participation in the ISP program? 3 Intensive Supervision Programs Characteristics of ISPs Intensive supervision probation programs differ in their required conditions, but some similarities characterize the programs. Common components include frequent and repeated contacts with probation or parole officers, smaller caseloads for officers, and drug and alcohol testing, all of which are considered control-oriented conditions (Fulton et al., 1994; Petersilia & Turner, 1991). There are supplemental conditions that different probation and parole departments have used in their ISP programs, some of which are treatment-oriented. For instance, Brown (2007) mentioned restitution, drug treatment, education, counseling, and mental health treatment as potential conditions for ISPs. Fulton and colleagues (1994) also discussed curfews, house arrest, and community service as typical ISP conditions. Similarly, Jolin and Stipak (1992) emphasized electronic monitoring as important components of an ISP. Petersilia (1998) touched on the emergence of technological advances in the 1980s, including voice verification systems, cheap on-site drug testing, and breathalyzers through the phone. Taxman, Byrne, and Thanner (2002) outlined two additional characteristics of an ISP in High Intensity Drug Trafficking Areas (HIDTA) in Washington, D.C., and areas of Maryland and Virginia. While the HIDTA incorporated drug testing, it emphasized graduated sanctions for violations. Most jurisdictions participating in the HIDTA created a behavioral contract, which informed offenders of the consequences for certain violations. For instance, in one jurisdiction, offenders who tested positive for alcohol or drugs would be ordered to increased contacts with probation or parole officers and increased drug testing upon the first violation. For the second positive test, the offenders would be ordered to attend self-help groups and report to court on the violation. Third positive tests would result in an increase in drug testing as well as an increase in 4 the level of treatment ordered by the court (Taxman et al., 2002). Graduated sanctions provide the criminal justice system with ways of enforcing compliance without immediately resorting to re-incarceration for non-compliance. Because technical violations of probation or parole are credited with adding to the prison overcrowding problem (Taxman et al., 2002; Byrne, Lurigio, & Baird, 1989; Caplan, 2006; Lowenkamp et al., 2010; Petersilia & Turner, 1993; Tonry & Lynch, 1996), it is important to have graduated sanctions for such violations, especially if the primary purposes of ISP is to reduce prison overcrowding and costs due to incarceration. Placement into an ISP varies by jurisdictions. Byrne (1986) discusses two contrasting approaches to ISP assignment, which he refers to as the administrative model and the judicial model. The administrative model encompasses those jurisdictions with which the decision for placement into an ISP lies with the probation department who prepare pre-sentence recommendations for the judge. This model is portrayed in Petersilia and Turner‟s (1991) study, in which placement into the ISP was based heavily on pre-sentence report recommendations. In the judicial model, judges are given the discretion to determine which offenders are appropriate for ISP programs (Byrne, 1986). Fulton and colleagues (1994) note that giving the discretion for placement into ISP programs to probation and parole departments is beneficial, as these departments are most qualified to make such recommendations. Goals and Outcome Measures of ISPs Although Byrne (1990) and Fulton and colleagues (1994) posited that agencies implementing ISPs need to focus on a primary purpose, most ISPs are expected to accomplish a variety of goals, some of which are complex and often conflicting. With such diverse goals, it is difficult for program evaluators to determine success (Fulton et al., 1994). For example, implementing a probation enhancement ISP to reduce prison costs is inherently illogical, as 5 enhancement ISPs are, by nature, not prison diversion programs (Petersilia & Turner, 1993). Thus, probation enhancement ISPs will not reduce the costs of incarceration. Without identifying goals and outcome measures, criminal justice agencies will continue to have difficulty evaluating ISPs. Suggestions have been made to consider other measures of success for ISPs. These range from providing intermediate sanctions to criminal justice agencies (Fulton et al., 1994) to minimizing risk to society and a strict adherence to punitive sanctions in the event of a violation (Harland & Rosen, 1987). Alternatively, an ISP in Oregon that focused on drug offenders had several additional goals such as reducing recidivism, reducing substance abuse, and stabilizing offenders‟ lives (Jolin & Stipak, 1992). Petersilia (1998) focused on cost savings as a primary purpose of ISPs, while Lowenkamp and colleagues (2010) emphasized additional goals such as reducing prison crowding and providing criminal justice agencies with greater sanctioning options. 6 Theoretical Framework Extant literature focuses on the debate between surveillance-oriented and treatmentoriented ISPs because the philosophies affect program components created by policy makers and job duties of employees. The discussion can be traced back to the debate between classical criminology and the positivist school (Byrne & Taxman, 2006). The classical school focused on free will and the deterrent effect that punishment has on offenders (Jeffery, 1959). This school of thought was developed in the eighteenth century in response to harsh and arbitrary punishments; its roots are traced back to Jeremy Bentham and Cesare Beccaria‟s belief in a human‟s rational decision-making process (Kennedy, 1983-1984). This belief led to the formation of deterrence theory, and the assumption that people weigh the cost-benefit of actions before deciding to act (Akers, 1990). According to Cullen and Gilbert (1982): “The goal of punishment is primarily the prevention of crime and only secondarily to exact retribution for the harm an offender has caused. To prevent crime, punishments should be just severe enough for the pain or unhappiness created by the punishment to outweigh the pleasure of happiness obtainable from the crime” (p. 30). Bentham and Beccaria emphasized three major facets of the deterrence theory, which are swiftness, severity, and celerity of punishment (Kennedy, 1983-1984). Thus, for a rational decision-maker to be deterred from crime, not only must the punishment for the crime be appropriately severe to outweigh the benefits of the crime, it should also be expected every time the crime occurs and administered swiftly upon the criminal act. Relatively recent developments in the deterrence theory have spurred a discussion regarding experiential effects as well – that is, it is possible that experiences with the criminal justice system with respect to the certainty, severity, and celerity of punishment may deter someone from crime (Paternoster, 1987; Pratt, Cullen, Blevins, Daigle, & Madensen, 2005; Minor & Harry, 1982). As the goals of a classical criminologist is to deter humans from crime, then classical criminology provides the basis for the 7 surveillance approach to an ISP (Farabee, 2005). According to Farabee (2005), this surveillance approach focuses on enhanced supervision through smaller caseloads and increased technology. The approach also promotes intensive monitoring of a probationer‟s whereabouts and is tailored to restrict freedom and deter a criminal from future offending. The positivist school came to light in the nineteenth century and emphasized the importance of studying crime empirically and scientifically (Jeffery, 1959; Cullen & Gilbert, 1982). The positivist school, unlike the classical criminology doctrine, focused on individualizing punishments, as supporters of the theory argued that crime is deterministic and offenders‟ motivation should be considered when punishing a criminal (Jeffery, 1959). It is important that the positivists emphasized determinism – that is, crime is “determined by factors largely outside the control of the individual” (Cullen & Gilbert, 1982, p. 33). Therefore, to eliminate crime the criminal justice system needs to target the underlying cause of the criminal behavior. Unlike the surveillance approach (which emphasizes control and intensive monitoring), Byrne and Taxman (2006) note that the treatment approach focuses on utilizing treatment programs to promote change in offenders, which will lead to a reduction in the event and likelihood of recidivism. The assumption of treatment based approaches, therefore, is that meaningful behavioral change is only likely when the underlying causes of crime are addressed. While there is not yet an agreement upon which emphasis better accomplishes goals of ISPs, the body of literature on the topic has expanded in the past two decades. Unfortunately, most ISP program creators and evaluators do not incorporate criminological theory into their programs and studies. However, it is imperative that the importance of doing so is stressed. Incorporating such theory may improve upon individual ISP programs through the use of existing knowledge of which theories work best to accomplish 8 certain goals. As mentioned above, swiftness, severity, and celerity of punishment are three important concepts studied in the literature on deterrence theory (Kennedy, 1983-1984). These concepts, among others, could be appropriately applied to ISPs by emphasizing certain probation conditions to enhance each component. 9 ISPs and Recidivism Surveillance versus Treatment Studies As noted earlier, there has not yet been a consensus reached on whether the surveillance or treatment philosophy is preferential in terms of accomplishing the goals of ISPs. In 2009, Drake, Aos, and Miller conducted a meta-analysis that consisted of 545 comparison-group evaluations based on programs in prevention efforts, adult corrections settings, and juvenile corrections settings, including ISPs. Findings indicated that overall, ISPs were associated with a 17.9% reduction in future crime only when they were treatment-oriented; there was no statistically significant reduction in recidivism when the ISPs were surveillance-oriented (Drake et al., 2009). Similarly, Petersilia and Turner‟s (1993) national ISP experiment found that although there were no significant differences found in recidivism rates between the ISP and regular probation participants, a higher level of participation in treatment led to a 10-20% reduction in recidivism (see Jolin & Stipak, 2007, and Petersilia & Turner, 1990, for comparable findings). Importantly, some studies have found positive effects for treatment-oriented probation terms for drunk-driving offenders (Taxman & Piquero, 1998; Deyoung, 1997), which is the population under study. In contrast, other studies have found no differences at all between treatment and surveillance approaches on re-arrest rates (Jolin & Stipak, 2007; Brown, 2007), while Brown (2007) found that treatment oriented program participants had significantly higher rates of program failure than those placed into control oriented ISPs. One possible explanation for the mixed findings among the literature is the inconsistency at which certain aspects of the programs are measured. For instance, some studies classified some programs as surveillance-oriented if they incorporated electronic monitoring of whereabouts, were placed on house arrest, or were ordered to have increased contact with 10 probation officers; studies classified the programs as treatment-oriented if there were high levels of participation in treatment or if there was simply a treatment condition ordered of the probationers. Arbitrary classifications of treatment or control programs may explain the mixed findings. While these studies provide some insight to the difference between treatment and surveillance approaches to probation, they fail to account for ISPs that heavily emphasize both treatment and supervision. As Brown (2007) noted in the explanation of her results, “programs that do not emphasize control, but rather place an emphasis on treatment, may not have the means available to ensure offender participation in treatment programs despite a treatment orientation” (p. 18). Therefore, a program that emphasizes both approaches may be better able to portray to offenders the importance of completing treatment through the incorporation of intensive monitoring to ensure attendance in treatment programs and strict sanctions upon failure to attend treatment. However, there is a debate among scholars on the ethicality of coercing offenders into treatment. Coercive Treatment Effects There are two arguments that conflict in terms of a court-ordered treatment program. Supporters of coercing offenders into treatment posit that the criminal justice system is a useful tool in initially getting people into treatment (Anglin, Brecht, & Maddahian, 1989; Platt, Buhringer, Kaplan, Brown & Taube, 1988; Taxman et al., 2002). Some authors note that criminal justice authorities can use the threat of violation sanctions to force an offender to attend their specific program (Anglin et al., 1989). Coercion to attend treatment programs encompasses the pressure to enter and remain in treatment, as well as the pressure to maintain changed behaviors once they are discharged from the treatment program (Platt et al., 1988). Hall (1997) 11 lists several more reasons to continue forced treatment, including the fact that a disproportionate amount of prisoners are drug-dependent, drugs and alcohol contribute to many of the crimes for which the inmates are incarcerated, and there is a high rate of relapse post-release, among others. Those who oppose using coercion to motivate an offender to attend a rehabilitation program argue that it may not produce any beneficial results, as the participants are not voluntarily putting themselves in a position to be helped (Polcin, 2001). Although admitting that substance users rarely join treatment programs without some coercion, Hartjen, Mitchell, and Washburne (1981) posit: “…the practice of coercing [offenders] to submit to programs of therapy via the threat of penal sanction is questionable because doing so lies outside the proper realm of criminal justice, as currently defined, and may be unethical in terms of the right of individuals to be protected from capricious judicial processes. It may also be impractical from the standpoint of successful treatment” (p. 21). Several empirical studies exist to examine the issues behind the two arguments. Anglin and colleagues (1989) studied the effect of criminal justice coercion on drug addicts in California. In the study, offenders were placed into three groups: Those recommended to treatment but not coerced, those with moderate criminal justice coercion, and those with high criminal justice coercion. Variables used to determine level of coercion were whether or not the offender was on probation, listed legal coercion as a primary or secondary reason to entering treatment, or were required to submit to drug testing while on probation. Results of the study showed that the offenders‟ behaviors after treatment was completed did not differ based on the level of coercion to which the offender was subjected. That is, the offenders‟ level of improvement in terms of reduction in drug use was not related to their coercion status (Anglin et al., 1989). One study showed that offenders who are coerced into treatment are more likely to stay in treatment, and for a longer period of time, than those who enter treatment voluntarily 12 (Farabee, Prendergast, & Douglas-Anglin, 1998). Attendance tenure is important, as Gendreau and Ross (1987) note that longer treatment tenure is associated with reduced drug use and reduced criminal activity. McFarlain, Cohen, Yoder, and Guidry (1977) conducted a study on variables associated with the retention of narcotics addicts in New Orleans area treatment programs, which showed that treatment participants under legal pressure were significantly more likely to remain in the program, but only for early stages of the treatment (see Rosenberg & Liftuk, 1976; Polcin, 2001 for similar results). As evidenced, although some studies found that coercing offenders into a treatment program may not produce ideal outcomes with respect to reduced substance use or recidivism, it does not appear to negatively affect these outcomes. 13 Desistance from Crime Thus far, the goal of reducing recidivism has focused solely on preventing offenders from being re-arrested in the future. However, there is a growing body of literature that emphasizes the importance of keeping offenders out of the criminal justice system for as long as possible. For example, Kazemian (2007) notes that once the onset of criminal offending has taken place, attempts should be made to decrease the intensity, length, and severity of offenders‟ criminal careers. DeJong (1997) emphasized the importance of studying the differences between those that are immediately re-arrested verses those who have a delayed re-arrest, and studying this could benefit criminal justice policy makers. In addition, Banks and Gottfredson‟s (2004) work on a drug court evaluation demonstrates the importance of time to failure, as differences between the experimental and control group were not apparent until four months after the follow-up period began. Such delayed effects would be missed in studies of recidivism that only examined the likelihood of re-arrest. Desistance, in this way, can be conceptualized as a process, as opposed to a dichotomy. Maguire and Raynor (2006) outlined several important facets to desistance from crime, noting that “Desistance is a difficult and often lengthy process, not an „event‟, and reversals and relapses are common” (p. 24). They also note that offenders require new knowledge, skills, and coping mechanisms to change behavior (Maguire & Raynor, 2006). Several events (often considered „transitions‟) are linked to desistance from crime, such as the maturation of offenders, marriage, and employment (Laub & Sampson, 2001), reduction in association with deviant peers (Warr, 1998; Giordano, Cernkovich, & Holland, 2003), and a reduction in the use of alcohol and/or drugs (Schroeder, Giordano, & Cernkovich, 2007; Hussong, Curran, Moffitt, Caspi, & 14 Carrig, 2004 note that an increase in substance use is associated with hindrance in desistance), among others. Time to Failure Studies Findings regarding time until failure for ISP programs are, again, generally mixed. One study compared residential and non-residential drug treatment programs for drug-involved probationers utilizing a survival analysis statistical method to examine the time until recidivism for the participants (Krebs, Strom, Koetse, & Lattimore, 2008). After matching the two groups based on propensity scores, results showed that probationers assigned to non-residential treatment programs were slower to recidivate than those probationers given no treatment options. However, probationers assigned to the residential treatment programs were not statistically different with regards to their time until recidivism than those probationers who were given no treatment options (for studies with similar results, see Guydish, Chan, Bostrom, Jessup, Davis, & Marsh, 2008; Perez, 2009). When the two treatment groups (residential treatment and nonresidential treatment) were compared, those probationers assigned to the residential treatment program recidivated more quickly than those assigned to the non-residential treatment program. Banks and Gottfredson (2003) focused on time until failure for drug court participants. Participants in this study who were sentenced through the drug court were assigned to either a program that had treatment-only, supervision-only, or a combination of supervision and treatment components. Those probationers ordered to the combination of supervision and 1 treatment program had the longest time until failure, but the results were not statistically significant when compared to the treatment-only group. Those in the treatment-only program had statistically significant longer time until failure than those in the supervision-only program, 1 Failure in this study was measured as the probationer‟s first rearrest; time until failure stopped being measured at that point for such probationers. 15 however. In addition, completing treatment was associated with a statistically significant increase in time until failure (Banks & Gottfredson, 2003). Limitations of Literature Thus far, research regarding ISPs categorizes the programs as focusing heavily on either treatment or surveillance. While it may be that ISPs focusing on treatment inherently have both treatment and surveillance components, program evaluations still tend to treat the emphasis as a dichotomy. Few programs utilize the surveillance components as leverage to complete treatment to the extent of the current study‟s ISP program. As stated earlier, the two research questions this paper attempts to answer are: Does this particular ISP program, which combines both rehabilitation and control, reduce the likelihood of re-arrest? How is the overall timing of rearrest affected by participation in the ISP program? Based on the above literature, it is hypothesized that this ISP, which introduces repeated motion hearings in front of a Judge to enforce treatment compliance, will reduce the likelihood of recidivism, as the punishment for violation is much more certain than the potential punishment for violation of regular probation. In addition, an ISP such as this study‟s program that enforces treatment participation should aid offenders to increase desistance from crime (through a reduction in substance use) and have a longer time to re-arrest than those who are not in such a program. 16 Current Study’s ISP Implementation and Target Population This study‟s ISP was implemented in 2006 in a college city. There were three people in charge of creating the program: a Judge, the Chief Probation Officer, and a Probation Officer with a Master‟s in Social Work and who was a licensed substance abuse counselor. The primary purpose of the ISP was to ensure completion of substance abuse treatment, with the intention that such treatment will help to reduce drinking and driving recidivism. The ISP targeted multipleoffense drunk drivers, as well as first-time offenders with circumstances surrounding their arrest that suggested a severe alcohol problem. Examples of these circumstances included getting into an accident while drinking and driving, having excessively high blood alcohol content (BAC) at the time of arrest, or driving while having children in the vehicle. Reliance on pre-sentence reports for placement in ISPs is typical of many such programs (Byrne, 1986; Fulton et al., 1994; Petersilia & Turner, 1991) and was the source of placement into ISP for this program. The presentence recommendations were based on information such as criminal history, BAC at the time of arrest, offense circumstances, and a substance abuse assessment. 2 A Typical ISP Sentence The ISP program under study incorporated the following: an 18 month term of probation, in which offenders meet twice per month with their probation officer (once individually and once with other ISP participants), completion of an intensive outpatient substance abuse program, 2 Per state law in this particular study, any person convicted of a drunk-driving offense must submit to a substance abuse assessment to determine needs and risk of recidivism. At this court, offenders were given the option to get a substance abuse assessment with the court‟s assessing agent or at an independent location. The court‟s assessing agent was contracted through an independent agency, and worked at the court part-time. These assessing agents utilize, at minimum, a NEEDS survey, which is a tool used to assess needs of substance users. 17 completion of 40 hours of community service, a 90 day license suspension period, and fines and costs. There is a 60 day daily breathalyzer requirement, which is sometimes replaced with a 60 3 day term of an alcohol tether . After the initial 60 days, the probationer is given a letter of the alphabet which corresponds to the timing for a random breathalyzer. In addition, 70 days of jail is ordered. This 70 day term is split up into segments of 10, 20, 20, and 20 days. The first 10 days of jail is served upfront, and the remaining three 20 day segments are postponed and reviewed at motion hearings. There is a requirement that offenders must have court appearances in the form of motions at three points in their probation term – after three months on probation, after nine months, and after 15 months of probation. At the motion, the offender is required to go in front of the judge on record and explain what he or she has done while on probation. The corresponding probation officer also attends the motion hearings, and is asked for their opinion on the offenders‟ progress while on ISP. If the offender is doing well and has completed all tasks asked of them, the corresponding 20 day segment of jail is permanently suspended. If the offender had not done what was ordered of them, they can be ordered at that time to serve any amount of that particular 20 day segment of jail. In consultation with the probation officer, the judge rules on the sanction to be imposed as a result of the violation. It is important to note that not all ISP conditions are identical. Each offender is interviewed as part of the pre-sentence report process and substance abuse assessment. The above description is the template for the ISP – however, in some instances, probationers are ordered to 90 days daily preliminary breath tests (PBTs) or tether (as opposed to the typical 60 days), probation terms are extended to 24 months (with four motions required at different 3 An alcohol tether (referred to as a SCRAM tether) measures an offender‟s blood alcohol concentration through perspiration. A local agency oversees the operation of the tether and reports any suspicious activity (including if the tether is removed from skin) to the Probation Officer. 18 intervals), and different types of substance abuse programs are required of different offenders, depending on needs. Vehicle immobilization conditions also vary from offender to offender (mostly due to outside factors, such as to whom the vehicle in question was registered), and upfront jail terms vary depending on the severity of the offense. In addition, mental health counseling is sometimes ordered to supplement the substance abuse counseling, as some probationers come to the probation department with existing mental health problems that may inhibit progress in the program. However, most of the probationers in this Court‟s ISP were ordered to the above typical conditions. Regular Probation The comparison group consisted of probationers who were higher-risk drunk-drivers but were sentenced prior to the creation of the ISP in 2006. These regular probation terms ranged from 12-24 months, and consisted of a substance abuse treatment program, vehicle immobilization (again, this varied based on the registration information), fines and costs, and community service. What is lacking in this program, as compared to the ISP, is that there is only one meeting per month with probation officer, there is no upfront jail, and the probationers are not periodically required to prove their compliance with court orders in order to avoid additional jail time through motion hearings in front of the sentencing Judge. Probation terms can still result in jail time, but this generally occurs only when repeated violations have taken place or the probation officer is recommending a probation revocation for the violations. Thus, the regular probation is largely treatment-oriented, in contrast to the ISP that combines treatment and intensive surveillance through the continued threat of punishment at motion hearings. 19 Data Collection Methods Because this ISP was implemented in 2006, and probation terms are typically 18 months, there were only 82 offenders who had completed the program at the time of this analysis; however, 5 of these offenders were dropped from the analysis due to missing data. Thus, 77 of the entire sample of 82 ISP probationers were included in this study. When the program was implemented, all probationers who qualified based on certain characteristics were sentenced to the program. Because of this full-implementation, the comparison group in this study consisted of all alcohol-related driving offenders sentenced to 12, 18, or 24 month terms of probation for drunk-driving offenses during the years 2004 and 2005. Offenders are typically sentenced to 6, 12, 18, or 24 month terms of probation, and as the target comparison group was higher-risk offenders, lower-risk cases sentenced to 6 months of probation were dropped from the comparison group (for an explanation on the importance of targeting the appropriate population, see Bonta, Wallace-Capretta, & Rooney, 2000). There were 168 participants in the comparison group; due to missing data, 20 of these original participants were dropped. The comparison group therefore consisted of 148 regular probationers. Court personnel were responsible for data collection. Resources utilized in the collection 4 of data included pre-sentence reports, pre-sentence face sheets , the court‟s database which contains information regarding the probationer‟s cases, and probation notes which were saved electronically. Data collection began in October 2010 and was finished in December 2010. The principle investigator was not given the data until all identifying information was deleted from the data file. 4 Pre-sentence face sheets contain demographic information regarding each offender. The information is obtained from pre-sentence questionnaires, which are filled out by the offender after a conviction is entered. Variables such as age, education, marital status, and charge were gathered from the pre-sentence face sheet and utilized in this study. 20 Dependent Variables and Outcome Measures Re-arrest data came from a state-wide database to which the court officials collecting data were given access. The state-wide database includes information that district and circuit courts around the specific state send in to the administrators. Although this database is very useful, it unfortunately does not have full participation in the state. That is, 22 counties in the state have not yet signed up to participate in the database. Therefore, information obtained solely from this database may exclude pertinent information regarding recidivism. Also, any probationer re-arrested out of the state will not show up on the re-arrest list. Recidivism was measured two different ways. The variable „Recidivism‟ is a dichotomous variable of whether or not the probationers were rearrested after their sentencing date within 17 months after their discharge date (0=no, 1=yes). The probationer sentenced most recently to the ISP only had 17 months post-discharge for this study to analyze; therefore, the dichotomous recidivism variable was restricted to this time frame so as to create equal analyses for all participants of the ISP and regular probation. The second measure of recidivism represents time to failure, and is measured by the amount of days from sentencing date to rearrest to the end of data collection. Thus, the time period for potential failure was longer for the comparison group (allowing for more re-arrests); however, the survival analysis used to analyze this data takes into consideration different follow-up times. If there was more than one re-arrest, the number of days between the separate re-arrests was also recorded. Independent Variables The following variables were gleaned from pre-sentence reports: the ten most recent offenses on the offender‟s criminal history (which were then organized into variables such as total number of prior offenses, total number of prior alcohol related offenses excluding driving 21 offenses, total number of prior alcohol related driving offenses, and total number of drug related offenses, all of which utilized a natural log to reduce skewness and kurtosis), and BAC at time of arrest. Pre-Sentence face sheets include mostly demographic information – variables include sex (0=male, 1=female), age (the natural log was taken to reduce skewness and kurtosis), education (responses ranged from “Did not complete high school” to “Completed post-graduate degree”), and marital status (“Married,” “Separated,” “Divorced,” or “Single”). In addition, variables relating to the instant offense are found on pre-sentence face sheets. These include the charge 5 associated with the instant offense (categorized as Operating While Intoxicated or Other ) and bond type information (this was dichotomized into personal recognizance bonds and non6 personal recognizance bonds ). For the years 2004 and 2005, from which the control group was selected, similar information was gathered through archived pre-sentence reports, pre-sentence face sheets, and probation face sheets. 5 Other included Operating While Intoxicated 2nd, Operating While Intoxicated 3rd, BAC Under (a charge given to minors who are arrested for driving with a BAC from 0.01 – 0.07) and Operating with Any Presence of Drugs. 6 Non-personal recognizance bonds included 10% bonds, surety, or cash bonds. 22 Description of Population ISP/Experimental Group There were a total of 77 probationers that completed the ISP by the time this study commenced and who did not present with missing data (see Table 1). The average age of an ISP participant in this sample was 28.0 years. Similar to the comparison group, a majority of the experimental group had completed some college (67.9%), while 10.3% were college graduates and 7.7% were high school graduates. Again, most of the probationers in the ISP were male (89.7%) and most were single (84.6%). The average BAC for an experimental group participant at the time of arrest was .16. Under half of the ISP participants were issued Personal Recognizance bonds (44.9%). Again similar to the comparison group, a large majority of the experimental group probationers had a charge of Operating While Intoxicated at the time of arrest (85.9%). The ISP probationers had an average of 2.7 prior offenses (disaggregated into an average of 0.7 prior alcohol offenses, 1.1 prior alcohol driving offenses and 0.2 prior drug offenses). Lastly, 20.5% of the experimental group recidivated in the follow-up time period. Regular Probation/Comparison Group As mentioned, all probationers assigned to 12, 18, or 24 month terms of probation for alcohol-related driving offenses during the years 2004 and 2005 were analyzed as part of the comparison group, minus those omitted due to missing data (refer to Table 1 for an overview of the descriptive statistics). There were a total of 148 probationers who fit these criteria. The average age among these probationers was 24.5 years old. A majority of the comparison group had completed some college (65.5%), while 11.5% of the group were college graduates and 23 10.1% were high school graduates. Most of the regular probationers were single (89.2%), and most were male (80.4%). The comparison group participants had an average BAC of .15 at the time of their arrest. Just under half of the group were given personal recognizance bonds (43.2%). A large majority of the comparison group (90.5%) were charged with Operating While Intoxicated at the time of arrest. With respect to criminal histories, the comparison group participants had an average of 2.6 prior criminal charges (the group had an average of 0.9 prior alcohol-related offenses, 0.5 prior alcohol-related driving offenses, and 0.3 prior drug-related offenses). Importantly, 20.3% of the comparison group participants recidivated at least after their sentencing date within 17 months of their discahrge date. 24 Table 1. Descriptive Statistics Comparison Group (N=148) Male Education Did not complete HS Completed GED Graduated HS Completed some college College graduate Completed some post-grad Completed post-grad work Marital Status Single Married Separated Divorced Charge Operating While Intoxicated Other Bond Type Personal Recognizance Non-Personal Recognizance Recidivism w/in 17 months No Yes Age BAC Prior Offenses Prior Alcohol Offenses Prior Alcohol-Driving Offenses Prior Drug Offenses Experimental Group (N=77) N 119 % 80.4 N 70 % 89.7 8 4 15 97 17 4 3 5.4 2.7 10.1 65.5 11.5 2.7 2 2 1 6 53 8 5 3 2.6 1.3 7.7 67.9 10.3 6.4 3.8 132 5 3 8 89.2 3.4 2 5.4 66 4 1 7 84.6 5.1 1.3 9 134 14 90.5 9.5 67 11 85.9 14.1 64 84 43.2 56.8 35 43 44.9 55.1 118 30 Mean 24.5 0.15 2.6 0.9 79.7 20.3 S.D. 7.4 0.05 2.4 1.3 62 16 Mean 28 0.16 2.7 0.7 79.5 20.5 S.D. 8 0.06 2 0.9 0.5 0.3 0.9 0.6 1.1 0.2 0.6 0.4 25 Data Analysis Methods In order to address the research questions, multiple statistical analyses were conducted. To begin, some bivariate statistics were run to determine the relationship between the independent variables and the dichotomous variable of recidivism. Then, propensity scores were calculated to determine the probability of assignment to the experimental group based on observed covariates. The experimental group participants were matched to the comparison group participants based on their propensity scores, utilizing a kernel matching algorithm. An independent samples t-test were used to determine the average treatment effect (ATE) with respects to the likelihood of recidivism as the outcome measure. Inverse probability of treatment weighting (IPTW) was then utilized in a survival analysis to determine the timing of recidivism at any given point during or after treatment through a Cox proportional hazards model. 26 Likelihood of Re-Arrest Methods Propensity Scores, Kernel Matching, and Average Treatment Effects Ideally, offenders in this city would have been randomly placed into either the regular probation or the ISP. Randomization of selection would have ensured that, with a large enough sample, the two groups would be identical with respect to demographic characteristics and other covariates that would possibly affect an offender‟s success (or failure) in either program. For this study, randomization was impossible, as the program had already been implemented for all high-risk offenders at the time this study commenced. When initial descriptive statistics were compared using independent samples t-tests with the status of control or experimental group as the dependent variable, the treatment and comparison groups were found to be significantly different on a number of variables. Analyses that fail to account for these differences may produce Type I or Type II errors, as observed differences in recidivism may not be due to effects of the ISP program, but to pre-existing differences between the probationers in each group. Therefore, counterfactual methods of analysis (Rosenbaum & Rubin, 1983) were utilized to address the issue of non-randomization of selection into the ISP program. To help account for these biases, and to make sure that any treatment effects were due to the treatment and not inherent differences between the participants, propensity score analyses were utilized. The propensity score, understood as the probability of the participant being selected into the treatment group, or the conditional probability of assignment to a particular treatment, given a set of observed covariates (Rosenbaum & Rubin, 1983; D‟Agostino, 1998; Rubin, 1997) is used to “reduc[e] the entire collection of background characteristics to a single composite characteristic that appropriately summarizes the collection” (Rubin, 1997, p. 757). 27 Inclusion of the propensity score has been found to produce unbiased estimates of treatment effects in observational studies (Caliendo & Kopeinig, 2008). There are different ways to analyze a propensity score, one of which is kernel matching. Kernel matching involves matching participants from the comparison group to participants in the experimental group based on their propensity score (Caliendo & Kopeinig, 2008). Kernel matching is different than other methods of matching such as nearest-neighbor matching, caliperradius matching, or interval matching. While other techniques match the propensity score of the experimental group participants to the propensity score of one (or a determined number) comparison group participant, kernel matching allows experimental participants to be matched with multiple comparison group participants. When utilizing kernel matching, a measure referred to as the bandwidth, or a smoothing parameter, must be designated (Sheather & Jones, 1991). The value of the bandwidth helps to determine the weight that is given to a matched pair in the sample, based on propensity scores. Once the propensity score for each participant is calculated, these scores can be analyzed to determine the comparability of treatment and comparison group participants. A bandwidth of .05 will only allow those participants in the experimental group to be compared with participants in the comparison group if their propensity scores are within .05 of each other. The closer the propensity scores are between the experimental and comparison group participants, the higher the weight given to the observation. If the propensity scores between cases exceed .05 in absolute value, the weight given to the matched pair is zero (that is, they are not considered a match). Participants that do not have a comparison within .05 of their own propensity score are omitted from the analysis. Therefore, when a bandwidth is smaller (such as .03 or .01), more participants are likely to be omitted from the study. As Smith and Todd (2001, p. 117) note, “In 28 general, increasing the bandwidth will increase the bias and reduce the variance associated with the estimator by putting a heavier weight on the information provided by more distant observations”. Kernel matching also utilizes the average of the weights given to each matched pair when there is more than one match within the .05 bandwidth of participants‟ propensity scores (Caliendo & Kopeinig, 2008). The major advantage, then, to kernel matching is that more information is used when creating weights for matched pairs, while a potential drawback is that “observations are used that are bad matches” (Caliendo & Kopeinig, 2008, p. 43). Caliendo and Kopeinig (2008) note the importance of choosing the appropriate matching algorithm when comparing propensity scores, and they emphasize that the different methods have different advantages and disadvantages. They write, “To give an example, if there are only a few control observations, it makes no sense to match without replacement. On the other hand, if there are a lot of comparable untreated individuals it might be worth using more than one NN [nearestneighbor] (either by sampling or KM [kernel matching]) to gain more precision in estimates” (Caliendo & Kopeinig, 2008, p. 45). Clearly, this study has almost twice the amount of untreated individuals in the control group than those who are treated in the experimental group, and thus kernel matching is an appropriate way to estimate causal effects. To conduct the hypothesis test to determine the relationship between ISP participation and the dichotomous recidivism variable, the average treatment effect (ATE) will be calculated. The ATE is the effect ISP would have had on probationers if they had been randomly selected into the ISP (Apel & Sweeten, 2010; Vella & Verbeek, 1999). 29 Time Until Failure Methods Survival Analysis To address the aforementioned issues of non-random assignment into ISP, the survival analysis will make use of inverse probability of treatment weighting to reduce the possibility of biased estimates resulting from confounding (however, the analyses will be run with unweighted data as well to compare and contrast the effects of weighting). Caliendo and Kopeinig (2008, p. 56) describe inverse probability of treatment weighting as “…the difference between a weighted average of the outcomes for the treated and untreated individuals, where units are weighted by the reciprocal of the probability of receiving treatment.” However, as with most statistical procedures, there are benefits and drawbacks to utilizing such weights. For example, although the inverse probability weights can address existing standardized biases even after propensity scores are calculated and can be utilized in survival analyses, they are not as robust in smaller samples (Lunceford & Davidian, 2004; for a review of inverse probability of treatment weighting, see Freedman & Berk, 2008). Calculating the inverse probability weights (IPTW) is done through the following formulas (see Figure 2 for the IPTW distribution): IPTW = 1 / (propensity score) for the experimental group IPTW = 1 / (1 - propensity score) for the control group (Freedman & Berk, 2008). The survival analysis utilized inverse probability of treatment weights to reduce bias between the two groups, so that test results are not as likely to be the result of confounding (Cole & Hernan, 2004). A survival analysis can describe data in terms of a hazard function (or the conditional failure rate) or the survivor function (the probability of surviving beyond time, t), among others (see Cleves, Roberto, Gould, & Marchenko, 2008 for a review). The survival analysis can also consider multiple failures, such as in the event that a probationer „failed‟ more 30 than once after they were sentenced to either regular probation or the ISP under question (a „failure‟ is a new arrest, and every time there is a new arrest, a new „failure‟ was recorded for each probationer). The survival analysis for nonparametric estimations (Kaplan & Meier, 1958) does not assume a normal distribution, as time until failure typically has an exponential distribution; that is, the event “has an instantaneous risk of occurring that is constant over time” (Cleves et al., 2008, p. 2). A sample is considered nonparametric when the covariates (or the variables predicting treatment) are unavailable, limited, or qualitative in nature (Cleves et al., 2008). The analysis can also take into consideration events that have not yet happened – it can analyze events that may have happened if the study had gone on longer than it did. This is referred to as right-censoring (or, right-truncation), and it is important in the current study in particular, as some of the probationers under scrutiny simply did not have as much time to fail as others. This right-censoring can be acknowledged and addressed through the use of either Kaplan-Meier estimation or the Nelson-Aalen estimator (Cleves et al., 2008). Kaplan-Meier and Nelson-Aalen Estimations Cleves and colleagues (2008) note that the use of either the Kaplan-Meier estimator or the Nelson-Aalen estimator depends on the desired function in the statistical analysis. In smaller samples, the Kaplan-Meier estimator is superior when calculating the survivor function, while the Nelson-Aalen estimator is superior when calculating the cumulative hazard function. These two estimators can be used to graphically portray the differences between the experimental and control groups with respect to either the survival function or the cumulative hazard function. The Nelson-Aalen estimate measures cumulative hazard, as mentioned. A cumulative hazard can be thought of as a sum of hazards over a time integral (Cleves et al., 2008). In this 31 study, a failure is repeatable; thus, “…it records the number of times we would expect (mathematically) to observe failures over a given period” (Cleves et al., 2008, p. 13). Thus, a cumulative hazard of .5 in the current study would mean that it is expected that a probationer would be re-arrested, on average, 0.5 times in the period studied. Cox Proportional Hazards Model To test the current study‟s hypothesis with respect to significance levels, the Cox proportional hazards method will be used. Cox proportional hazards regression has been found to be quite robust, as it requires few assumptions. There is no assumption regarding the shape of the distribution of the hazard over time; the only assumption with respect to the shape of distribution is that the shape is consistent for both the experimental and comparison group (Cleves et al., 2008), which was the case in this study. Cox proportional hazards models are also useful because they can account for censoring (Huebner, Varano & Bynum, 2007). Importantly, Cox proportional hazards model supports weighting, such as IPTW. 32 Results Bivariate Statistics To analyze the bivariate statistics, a chi-square test was run between the dependent, dichotomous variable of recidivism and the five nominal or ordinal-level independent variables of sex, education, marital status, charge, and bond type (see Table 2). Initial results show that only education is statistically significantly related to recidivism; that is, the higher one‟s education, the less likely they are to recidivate within the timeframe analyzed. An independent samples t-test was then run with the same dependent variable and the following continuous, independent variables: age, BAC at the time of arrest, prior offenses (natural log), prior alcoholrelated offenses (natural log), prior alcohol-related driving offenses (natural log), and prior drugrelated offenses (natural log). None of the relationships were statistically significant. To further explore the relationship between the independent variables and the dichotomous variable of recidivism, propensity scores were run and an ATE statistic was calculated. 33 Table 2. Bivariate Statistics on Recidivism in 17 Months as Dep. Var.: Total Population Recidivated Did Not Recidivate N % N Age BAC Prior Offenses Prior Alcohol Offenses Prior Alcohol-Driving Offenses Prior Drug Offenses 40 6 17.8 2.7 148 31 65.8 13.8 4 4 5 29 2 1 1 1.8 1.8 2.2 12.9 0.9 0.4 0.4 6 1 16 120 23 8 5 2.7 0.4 7.1 53.3 10.2 3.6 2.2 42 1 1 2 18.7 0.4 0.4 0.9 156 8 3 12 69.3 3.6 1.3 5.3 42 4 18.7 1.8 158 21 70.2 9.3 16 30 7.1 13.3 82 97 36.4 43.1 16 30 Mean 3.22 0.15 1.24 0.53 7.1 13.3 S.D. 0.26 1.06 0.63 0.58 62 118 Mean 3.26 0.07 1.09 0.43 27.6 52.4 S.D. 0.24 1.06 0.56 0.52 0.48 0.14 Sex Male Female Education* Did not complete HS Completed GED Graduated HS Completed some college College graduate Completed some post-grad Completed post-grad work Marital Status Single Married Separated Divorced Charge Operating While Intoxicated Other Bond Type Personal Recognizance Non-Personal Recognizance Program Status ISP Regular Probation % 0.44 0.34 0.44 0.14 0.42 0.29 chi2 statistic: * p < .05; ** p < .01; *** p < .001 t-test statistic: + p < .05; ++ p < .01; +++ 34 p < .001 Likelihood of Re-Arrest Results Propensity Scores and Kernel Matching In the current study, the propensity score was calculated based on the following confounders: Age (the natural log was taken to reduce skewness and kurtosis), sex, education, marital status, drunk driving charge, bond type, total number of prior offenses (again, the natural log was taken), total number of prior alcohol related offenses (natural log), total number of prior alcohol related driving offenses (natural log), total number of prior drug related offenses (natural log), and BAC at time of arrest. These variables were chosen due to availability during data collection and because these variables are used by Probation Officers in preparation of the presentence reports, which may or may not include recommending ISP to the sentencing Judge. Importantly, all variables were measured prior to treatment, and thus could not be impacted by the type of probation to which the offenders were assigned. The distribution of propensity scores is demonstrated in Figure 1. 35 Figure 1. Propensity Score Distribution for IPS Participants versus Regular Probationers I In this study, the propensity score utilized a bandwidth of .05. Preliminary analyses suggested that a smaller bandwidth (for example, a bandwidth of .03 or .01) led to the exclusion of too many participants from the study. When using the bandwidth of .05, one participant from the comparison group was excluded from the analysis, due to a significantly lower probability of treatment than the rest of the probationers. This can be seen graphically in Figure 1, as all of the propensity scores after dropping the one comparison group participant had propensity scores 36 within .05 of one another; therefore, at least one match could be made for every experimental and comparison group participant. When the propensity scores were analyzed, most of the variables‟ biases were reduced to 7 an acceptable level ; however, the variable „Bond Type‟ actually increased in standardized bias (see Table 2). This may be due, in part, to the fact that there were a limited number of variables available to predict the probability of treatment for the experimental and control group participants. Also, there may have been some underlying characteristic that the sentencing Judge looked at when determining the offenders‟ bond type and that information was unavailable for the current study. 7 An acceptable level is considered any absolute value under 20. 37 Table 3. Bias Diagnostics for ISP Participants versus Regular Probationers Unmatched Variables Age (ln) Male Education Marital Status Married Separated Divorced Operating While Intoxicated PR Bond BAC Prior Offenses (ln) Prior Alcohol Offenses (ln) Prior AlcoholDriving Offenses (ln) Prior Drug Offenses (ln) Reg. Prob. N=148 3.21 0.80 2.91 N=77 3.33 0.90 3.17 0.03 0.02 0.05 Matched 52.4 25.9 24.7 Reg. Prob. N=147 3.29 0.92 3.06 N=77 3.33 0.90 3.17 0.05 0.01 0.08 8.9 -5.7 9.6 0.05 0.00 0.10 0.05 0.01 0.08 1.8 9.0 -8.9 0.91 0.86 -14.9 0.90 0.86 -14.6 0.43 0.15 1.09 0.44 0.16 1.18 1.8 2.8 15.5 0.32 0.15 1.20 0.44 0.16 1.18 24.8 7.7 -5.0 0.49 0.38 -20.7 0.4 0.38 -2.8 0.32 0.69 100.7 0.71 0.69 -7.2 0.16 0.10 -20.4 0.10 0.10 0.2 IPS % Bias 38 IPS % Bias 17.8 -7.6 10.7 Once the propensity scores were calculated and the two groups were matched based on the designated bandwidth of .05, a t-test was run with the dichotomous variable of recidivism (0=no; 1=yes) as the dependent variable. The average treatment effect was 0.0006, indicating that the recidivism rate of the experimental group participants was .06% higher than the recidivism rate of those in the comparison group, although this was not significant; thus, had the results been significant, a negligible difference was found in recidivism between the two 8 groups . Inclusion in the ISP did not significantly affect whether or not the participants recidivated after their sentencing date. 8 The average treatment effect on the treated (ATT) statistic indicated a non-significant (p > .05) mean effect of -.04 for the ISP, whereby the mean effect of treatment on recidivism for the comparison group was 0.25 while the mean effect of treatment on recidivism for the experimental group participants was .0.21 (t = -0.55). 39 Time Until Failure Findings Kaplan-Meier and Nelson-Aalen Estimate Results To address the second research question regarding time until failure, the survival analysis was run. First, the Kaplan-Meier and Nelson-Aalen estimate graphs were created to graphically show the relationship between the treatment and comparison groups with respect to survival rates and cumulative hazard rates. As can be seen in Figure 2, the Kaplan-Meier graph for the unweighted data shows very similar survival rates for the regular probationers and ISP probationers. Most often, the two groups have identical survival rates; however, there are times in which the two groups switch with respect to which has a higher survival rate. The two groups begin with a probability of survival of 1.00, which means that at the time of sentencing, the two sets of probationers all are predicted to survive in the time analyzed without being re-arrested. As time continues, this probability lowers. 40 Figure 2. Kaplan Meier Survival Estimates of Unweighted Data 0.00 0.25 0.50 0.75 1.00 Kaplan-Meier Survival Estimates 0 500 1000 1500 Analysis Time in Days Regular Probationers 2000 2500 ISP Participants When the data is weighted, a slightly different pattern is found (see Figure 3). The graph indicates that, if the sample population would have been randomly placed into either ISP or regular probation, the ISP participants would have had a lower probability of survival than the regular probationers from the sentencing date until the end of their analysis time. That is, ISP participants would have had a higher probability of being re-arrested than the regular probationers almost immediately after the sentencing date. 41 Figure 3. Kaplan Meier Survival Estimates with Weighted Data 0.00 0.25 0.50 0.75 1.00 Kaplan-Meier Survival Estimates 0 500 1000 1500 analysis time Regular Probationers 2000 2500 ISP Participants Figure 4 provides the Nelson-Aalen graph for the unweighted data. Again, the ISP participants and regular probationers have similar cumulative hazards after sentencing date; the two groups tend to switch at certain points as to which has a higher cumulative hazard. Once weighting is introduced, however, the pseudo-experimental group of ISP participants would have had a higher cumulative hazard than the regular probationers throughout the time period studied (see Figure 5). 42 Figure 4. Nelson-Aalen Estimates of Unweighted Data 0.00 0.20 0.40 0.60 Nelson-Aalen Cumulative Hazard Estimates 0 500 1000 1500 Analysis Time in Days Regular Probationers 43 2000 ISP Participants 2500 Figure 5. Nelson-Aalen Estimates of Weighted Data 0.00 0.20 0.40 0.60 Nelson-Aalen Cumulative Hazard Estimates 0 500 1000 1500 Analysis Time in Days Regular Probationers 2000 2500 ISP Participants Cox Proportional Hazards Model Results Table 4 outlines three different models for the unweighted data. The unweighted analysis (model I) includes only the treatment status; model II introduces demographic controls, and model III includes demographic controls, covariates predicting treatment, and the treatment status. The weighted model includes only the treatment status. This is because the weighted data set used the covariates listed above (including demographic variables) to calculate the treatment weight; thus, it is a pseudo-random sample. In a random sample, demographics and the covariates in this study are considered controlled by the actual act of randomization of assignment. 44 The results for the cox proportional hazards model indicate that in the unweighted data set, participation in the ISP has a non-significant hazard ratio of 0.99 (see Table 4). This means that, had the results been significant, the ISP participants would have had a hazard rate (or a probability of failure) 1% less than the hazard rate of the comparison group. Thus, the ISP participants would have had a slightly slower time until failure than the comparison group. When the demographic characteristics are added to the analysis in model II, only education is a statistically significant variable. The hazard ratio for education is 0.79, indicating that as education increases by one unit, the hazard rate decreases by 21% as compared to those one unit lower on the education scale. An increase in education, therefore, results in a statistically slower time until failure. Again, treatment status is not significantly related to time until failure. The third model adds the covariates of charge, bond type, BAC, prior offenses, prior alcohol-related 9 offenses, prior alcohol-related driving offenses, and prior drug-related offenses . In this model, the significance of education is explained away and BAC (z-score) and prior offenses are significant. The hazard ratios for BAC (z-score) and prior overall offenses (natural log) are 1.34 and 2.04, respectively, indicating that a one-unit increase in the standard deviation of BAC from the mean was associated with a 34% increase in a hazard rate than those with a one unit decrease 10 in the standard deviation of BAC lower ; those with a higher BAC had a faster time until failure. With respects to prior offenses, those with a one unit higher natural log of prior offenses had a 204% increase in their hazard rate; thus, they had a much quicker time until failure than 9 It should be noted that the BAC variable was manipulated prior to running the cox proportional hazards model, as preliminary results indicated an outstanding standard error. Thus, the z-score for each BAC level was calculated and then included in Model III of the cox proportional hazards results. 10 The mean for BAC was .15; the standard deviation was .05. Thus, a „one-unit increase in the standard deviation from the mean of BAC‟ indicates a jump from .15 to .20 in BAC. 45 those with one unit less of the natural log of prior offenses. Again, it is important to note that the variable of treatment verses comparison group status was not significant in any of the models. In addition, only model III had a statistically significant model fit utilizing a chi-square test. When the weighted data was analyzed, treatment verses comparison group status was not significantly related to time until failure. 46 Table 4. Cox Proportional Hazards Models Age Male Education Marital Status Married Separated Divorced Operating Intox PR Bond BAC Prior Offenses Prior Alcohol Offenses Prior Alc. Driving Off. Prior Drug Offenses ISP Participation Model Fit 2 LR chi Log likelihood Model I Haz. Ratio S.E. ------------------------0.99 ----0.25 0.00 -413.07 Unweighted Data Model II Haz. Ratio S.E. 1.57 0.96 1.48 0.52 0.08 0.79* 0.42 0.67 0.24 ---- 1.34** 0.47 0.51 0.29 0.35 0.16 0.15 ----0.95 ----0.26 2.04* 0.82 0.85 1.05 0.99 0.66 0.24 0.31 0.39 0.28 Weighted Data Haz. Ratio S.E. 1.21 0.36 0.42 -423.00 * p < .05; ** p < .01; *** p < .001 47 0.75 0.48 0.43 0.86 0.60 S.E. 0.71 0.47 0.09 0.66 0.64 0.36 ---- 9.07 -403.15 ISP Participation Model Fit 2 Wald chi Log Pseudolikelihood Model III Haz. Ratio 0.88 1.30 0.82 26.22* -394.56 Discussion of Findings and Conclusion This study looked at a fundamentally distinct intensive supervision probation program in a college city. The program was unique in that it utilized a control-oriented condition to enforce compliance in treatment-oriented conditions; therefore, it was a program that heavily emphasized both treatment and surveillance. The probationers assigned to the ISP were required to submit to regular motion hearings in front of the sentencing judge to prove their compliance in their courtordered substance abuse treatment program. The probationers in the comparison group of regular probation were not required to repeatedly show their compliance in treatment components. The data was analyzed to determine the effectiveness of the program to reduce the likelihood of re-arrest amongst the participants, as well as the effectiveness of the program to increase time until re-arrest. Overall, findings were discouraging for this jurisdiction‟s ISP. Without weighting, the ISP program had no significant affect on either the likelihood or timing of recidivism as compared to the regular probation. Both the Kaplan-Meier and the Nelson-Aalen estimates indicate that there was no discernible difference between the ISP and regular probationers with respect to time until failure; and, once weighting was introduced, the ISP participants had a higher probability of failure (i.e., they would have been re-arrested more quickly) every day after their sentencing date and a higher expected number of failures at each day after sentencing than the regular probationers. The Cox proportional hazards model showed that only total number of prior offenses and BAC at the time of arrest were significantly related to re-arrest rates in the data prior to weighting. As participants had higher numbers of prior arrests and as participants had higher BAC at time of arrest, their hazard ratios increased, indicating a quicker time until 48 failure. The ISP program was not significantly related to time until failure in the Cox proportional hazards model in the unweighted or the weighted analyses. Explanations may vary for these findings. The propensity scores were based only on the few observed covariates that were available to data collectors at the time of this study. There may have been other important variables that predict treatment status for probationers but that were not systematically recorded at the court and were therefore not available for analysis such as employment status, race, severity of alcohol or drug problem, or others. It is also possible that those probationers in the ISP are more likely to be re-arrested, and likely to be re-arrested more quickly, than the regular probationers because they are being monitored more closely; thus, their chances of being caught committing a new crime may be higher. It is also possible that although, theoretically, the program may have increased the celerity of punishment for violation of the program, the punishment may have been too disruptive to be beneficial. If the probationers in the ISP were sent to jail repeatedly throughout probation for failing to attend treatment, they may have experienced situations that are conducive to future criminal behavior, such as the loss of employment that may lead to failure to pay fines and costs and a higher likelihood of theft-related activity. The jail terms may have also disrupted the probationers‟ family lives, which may lead them to cope with emotional problems through the use of alcohol or illegal substances. While these potential explanations are interesting, they cannot be substantiated or tested with the current data, and would require further research to validate. To explore the potential explanations for the findings further, in-depth data would have to be collected regarding probationers‟ experiences while in the ISP program or in the regular probation. For example, it would be interesting to see how many of the ISP probationers were 49 actually given jail time at each motion, and how many days they were ordered to serve at such motions. The discharge status should also be used to analyze how the likelihood and timing of re-arrest is affected by completion status of the ISP or regular probation. It is possible that ISP may have been beneficial, but only for those that completed the treatment program or for those that were never sentenced to any jail at their motion hearings. At the most basic level, a large limitation in this study is sample size, as only 77 ISP participants and only 148 regular probationers were included in the study. Although results may have been similar with larger samples, the results would have been more robust due to a larger sample size. And, although propensity score matching was utilized to reduce upfront biases, there was still some bias in existence between the two groups that may have affected the results. That is, it is possible that the ISP participants were different than the regular probation participants on unobserved covariates, and this could have resulted in Type II error. As mentioned earlier, the re-arrest database excluded information from 22 counties in the state of the database; thus, information regarding arrests taking place in those counties (and other states) was not considered in this study. Utilizing a different re-arrest database, such as the offenders‟ criminal histories in LEIN (law enforcement information network) may better encompass rearrest information, as the database covers national criminal justice jurisdictions. To address the issue of standardized bias, the study should be run in a jurisdiction that has not yet fully implemented such a program. Then, probationers could be randomly selected to either the ISP or the regular probation. This might provide more insight as to how the ISP program effects rearrest variables, as the groups would be identical due to randomization. It would be beneficial to understand if ISP participants were less likely to be re-arrested for another alcohol-related driving offense than regular probationers, as drunk-drivers are the 50 population under study. This study did not have the resources available to analyze the types of re-arrests offenders‟ had. One of the goals of the program was to reduce substance abuse, and it may be that the goals were reached with respect to reducing alcohol or drug-use related offenses. Practical implications of these findings could potentially be widespread. As the program requires a much higher frequency of contacts with probation officers, much more of the officers‟ time due to offender surveillance, and much more courtroom time due to repeated motions, the resources afforded the ISP are immense. If repeated studies of this nature show similar results as the current study, it may be futile to continue spending so many of the court‟s resources on the ISP program. In addition, jurisdictions around the country looking to implement a program to reduce a certain alcohol or drug-related crime should be cautious in framing the program identically to this study‟s ISP, as it may not produce beneficial results. However, conducting further research on this ISP may help to diagnose which component is unsuccessful, which could then be manipulated so as to produce better results in this and other jurisdictions. Theoretical implications are also important. This study only briefly mentioned criminological theory with respect to ISP program evaluations. While deterrence theory and the concept of desistance are current and important theories, there are many more that could be linked to an ISP program. Resources were not available for this study to analyze offenders‟ perceptions of the certainty, severity, and celerity of punishment while in the ISP, as compared to offenders‟ perceptions of these concepts while in the regular probation. It would be interesting to learn whether this ISP, as compared to regular probation, created a sense in which violating probation, or getting re-arrested, was outweighed by the cost punishment afforded them. It would also be beneficial to study the effects of the offenders‟ treatment programs on their substance use, so as to measure desistance from crime of offenders due to a reduction in 51 substance use. This would require further survey methods, but would allow researchers to fully understand how treatment programs can affect the time until failure for substance users. Aside from looking at other criminological theories, a more in-depth look into the current theories may be interesting. For example, relatively recent discussion has focused on the causal order of deterrence theory – that is, does potential punishment deter crime, or do the offenders‟ experiences in the past of punishment create a will to deter from crime? The concept of past experiences creating the perception of the cost of crime is referred to as the experiential effect (Paternoster, 1987; Pratt, Cullen, Blevins, Daigle, & Madensen, 2005; Minor & Harry, 1982). Applying this concept to the current study, offenders may have used their perceptions of punishment in the ISP to future decisions on whether or not to commit crime. That is, offenders in the ISP may not have had significantly different re-arrest rates and timing than the regular probationers because their experience in the ISP did not create a perception of certain, severe, and quick punishment for future offenses. This would take more survey and data analysis to confirm or reject; however, it is an important aspect to the deterrence theory that could be analyzed further. While the program under evaluation did not appear to have beneficial results with respect to the likelihood or timing of recidivism for its participants, these results are preliminary and require further exploration. The study does shed a light on the need for researchers and program creators to utilize criminological theories in the creation and evaluation of the programs. It also provides an example of a program that emphasized both treatment and surveillance and attempted to part from the heavily-dichotomized existing literature. Further research on combination programs may provide insight to which treatment components work best with which surveillance components to reduce recidivism and reach the goals of the program. 52 WORKS CITED 53 WORKS CITED Akers, R.L. (1990). Rational choice, deterrence, and social learning theory in criminology: The path not taken. The Journal of Criminal Law and Criminology, 81(3), 653-676. Andrews, D.A., Zinger, I., Hodge, R.D., Bonta, J., Gendreau, P., & Cullen, F.T. (1990). Does correctional treatment work? A clinically relevant and psychologically informed metaanalysis. Criminology, 28, 369-404. Anglin, M.D., Brecht, M.L. & Maddahian, E. (1989). Pre-treatment characteristics and treatment performance of legally coerced versus voluntary methadone maintenance admissions. Criminology, 27(3), 537-557. Apel, R.J. & Sweeten, G. (2010). Propensity score matching in criminology and criminal justice. Handbook of Quantitative Criminology, 5, 543-562. Banks, D. & Gottfredson, D. (2003). The effects of drug treatment and supervision on time to rearrest among drug treatment court participants. Journal of Drug Issues, 33, 385-412. Banks, D. & Gottfredson, D.C. (2004). Participation in drug treatment court and time to rearrest. Justice Quarterly, 21(3), 637-658. Bonta, J., Wallace-Capretta, S. & Rooney, J. (2000). A quasi-experimental evaluation of an intensive rehabilitation supervision program. Criminal Justice and Behavior, 27, 312319. Brown, K.L. (2007). Effects of supervision philosophy on intensive probationers. Justice Policy Journal: Analyzing Criminal and Juvenile Justice Issues and Policies, 4(1), 1-32. Byrne, J.M. (1986). The control controversy: A preliminary examination of intensive probation supervision programs in the United States. Federal Probation, 50(2), 96-108. Byrne, J.M. (1990). The future of intensive probation supervision and the new intermediate sanctions. Crime and Delinquency, 36(1), 6-41. Byrne, J.M., Lurigio, A.J. & Baird, C. (1989). The effectiveness of the new intensive supervision programs. Research in Corrections, 2(2), 1-64. Byrne, J.M. & Taxman, F.S. (2006). Crime control strategies and community change – reframing the surveillance vs. treatment debate. Federal Probation 70(1), 3-12. Caliendo, M. & Kopeinig, S. (2008). Some practical guidance for the implementation of propensity score matching. Journal of Economic Surveys, 22(1), 31-72. Caplan, J.M. (2006). Parole systems anomie: Conflicting models of casework and surveillance. Federal Probation, 70(3), 32-36. 54 Cleves, M., Roberto, G., Gould, W., & Marchenko, Y. (2008). An Introduction to Survival Analysis Using Stata (2nd Ed.) Texas: Stata Press. Cole, S.R. & Hernan, M.A. (2004). Adjusted survival curves with inverse probability weights. Computer Methods and Programs in Biomedicine, 75, 45-49. Cullen, F.J. & Gilbert, K.E. (1982). Reaffirming Rehabilitation. Ohio: Anderson Publishing Company. D‟Agostino, R.B., Jr. (1998). Tutorial in biostatistics: Propensity score methods for bias reduction in the comparison of a treatment to a non-randomized control group. Statistics in medicine, 17, 2265-2281. DeJong, C. (1997). Survival analysis and specific deterrence: Integrating theoretical and empirical models of recidivism. Criminology, 35(4), 561-576. Deyoung, D.J. (1997). An evaluation of the effectiveness of alcohol treatment, driver license actions, and jail terms in reducing drunk driving recidivism in California. Addiction, 92(8) 989-997. Drake, E.K., Aos, S. & Miller, M.G. (2009). Evidence-based public policy options to reduce crime and criminal justice costs: Implications in Washington State. Victims and Offenders, 4, 170-196. Farabee, D. (2005). Rethinking Rehabilitation: Why Can’t we Reform our Criminals? Washington, D.C.: AEI Press, American Enterprise Institute. Farabee, D., Prendergast, M. & Douglas-Anglin, M. (1998). The effectiveness of coerced treatment for drug-abusing offenders. Federal Probation, 62, 3-10. Freedman, D.A. & Berk, R.A. (2008). Weighting regressions by propensity scores. Evaluation Review, 32(4), 392-409. Fulton, B.A., Stone, S.B. & Gendreau, P. (1994). Restructuring Intensive Supervision Programs: Applying “What Works.” Kentucky: American Probation and Parole Association. Gendreau, P. & Ross, R. (1987). Revivification of rehabilitation: Evidence from the 1980s. Justice Quarterly, 4(3), 349-407. Giordano, P.G., Cernkovich, S.A., & Holland, D.D. (2003). Changes in friendship relations over the life course: Implications for desistance from crime. Criminology, 41(2), 293-328. Guydish, J., Chan, M., Bostrom, A., Jessup, M.A., Davis, T.B. & Marsh, C. (2008). A randomized trial of Probation Case Management for drug-involved women offenders. Crime & Delinquency (OnlineFirst). Doi 10.1177/0011128708318944. 55 Hall, W. (1997). The role of legal coercion in the treatment of offenders with alcohol and heroin problems. The Australian and New Zealand Journal of Criminology, 30(2), 103-120. Harland, A.J. & Rosen, C.J. (1987). Sentencing theory and intensive supervision probation. Federal Probation, 51(4), 33-42. Hartjen, C.A., Mitchell, S.M. & Washburne, N.F. (1981). Sentencing therapy: Some legal, ethical, and practical issues. Journal of Offender Counseling Services and Rehabilitation, 61, 21-39. Huebner, B.M., Varano, S.P., & Bynum, T.S. (2007). Gangs, guns, and drugs: Recidivism among serious, young offenders. Criminology and Public Policy, 6(2), 187-221. Hussong, A.M., Curran, P.J., Moffitt, T.E., Caspi, A., & Carrig, M.M. (2004). Substance abuse hinders desistance in young adults‟ antisocial behavior. Development and Psychopathology, 16, 1029-1046. Jeffery, C.R. (1959). Pioneers in criminology: The historical development of criminology. The Journal of Criminal Law, Criminology, and Police Science, 50(1), 3-19. Jolin, A. & Stipak, B. (1992). Drug treatment and electronically monitored home confinement: An evaluation of a community-based sentencing option. Crime and Delinquency, 38, 158-170. Kaplan, E.L. & Meier, P. (1958). Nonparametric estimation from incomplete observations. Journal of the American Statistical Association, 53, 457-481. Kazemian, L. (2007). Desistance from crime: Theoretical, empirical, methodological, and policy considerations. Journal of Contemporary Criminal Justice, 23(1), 5-27. Kennedy, K.C. (1983-1984). A critical appraisal of criminal deterrence theory. Dickinson Law Review, 88(1), 1-13. Krebs, C., Strom, K., Koetse, W. & Lattimore, P. (2008). The impact of residential and nonresidential drug treatment on recidivism among drug-involved probationers. Crime & Delinquency, 55(3), 442-471. Laub, J.H. & Sampson, R.J. (2001). Understanding desistance from crime. Crime and Justice, 28, 1-69. Lowenkamp, C.T., Flores, A.W., Holsinger, A.M., Makarios, M.D., and Latessa, E.J. (2010). Intensive supervision programs: Does program philosophy and the principles of effective intervention matter? Journal of Criminal Justice, 38(4), 368-375 56 Lunceford, J.K. & Davidian, M. (2004). Stratification and weighting via the propensity score in estimation of causal treatment effects: A comparative study. Statistics in Medicine, 23, 2397-2960. Martinson, R. (1974). What works? – Questions and answers about prison reform. The Public Interest, 35, 22-54. Maguire, M. & Raynor, P. (2006). How the resettlement of prisoners promotes desistance from crime: Or does it? Criminology and Criminal Justice, 6(1), 19-38. McFarlain, R.A., Cohen, G.H., Yoder, J. & Guidry, L. (1977). Psychological test and demographic variables associated with retention of narcotics addicts in treatment. Substance Use & Misuse, 12(2-3), 399-410. Martinson, R. (1974). What works? Questions and answers about prison reform. The Public Interest 35, 22-54. Minor, W.W. & Harry, J. (1982). Deterrent and experiential effects in perceptual deterrence research: A replication and extension. Journal of Research in Crime and Delinquency, 19, 190. Paternoster, R. (1987). The deterrent effect of the perceived certainty and severity of punishment: A review of the evidence and issues. Justice Quarterly, 4, 173-213. Perez, D.M. (2009). Applying evidence-based practices to community corrections supervision. Journal of Contemporary Criminal Justice, 25(4), 442-458. Petersilia, J. (1998). A decade of experimenting with intermediate sanctions: What have we learned? Federal Probation, 62(2), 3-9. Petersilia, J. & Turner, S. (1990). Intensive supervision for high-risk offenders: Findings from three California experiments. Santa Monica: RAND Corporation. Petersilia, J. & Turner, S. (1991). An evaluation of intensive probation and parole in California. Journal of Criminal Law and Criminology, 82, 610-658. Petersilia, J. & Turner, S. (1993). Intensive probation and parole. Crime and Justice, 17, 281335. Platt, J.J., Buhringer, G., Kaplan, C.D., Brown, B.S., & Taube, D.O. (1988). The prospects and limitations of compulsory treatment for drug addiction. The Journal of Drug Issues, 18(4), 505-525. Polcin, D.L. (2001). Drug and alcohol offenders coerced into treatment: A review of modalities and suggestions for research on social model programs. Substance Use & Misuse, 36(5), 589-608. 57 Pratt, T.C., Cullen, F.T., Blevins, K.R., Daigle, L.E. & Madensen, T.D. (2005). The empirical status of deterrence theory: A meta-analysis. In Francis T. Cullen, John Paul Wright, and Kristie R. Blevins (Eds.), Taking Stock: The Empirical Status of Criminological Theory – Advances in Criminological Theory, 15, 367-395. Rosenbaum, P.R. & Rubin, D.B. (1983). The central role of the propensity score in observational studies for causal effects. Biometrika, 70, 41-55. Rosenberg, C.M. & Liftuk, J.L. (1976). Use of coercion in the outpatient treatment of alcoholism. Journal of Studies on Alcoholism, 37(1), 58-65. Rubin, D.B. (1997). Estimating causal effects from large data sets using propensity scores. Annals of Internal Medicine, 127(8), 757-763. Schroeder, R.D., Giordano, P.C. & Cernkovich, S.A. (2007). Drug use and desistance processes. Criminology, 45(1), 191-222. Sheather, S.J. & Jones, M.C. (1991). A reliable data-based bandwidth selection method for kernel density estimation. Journal of the Royal Statistical Society, 53(3), 683-690. Smith, J.A. & Todd, P.E. (2001). Reconciling conflicting evidence on the performance of propensity-score matching methods. The American Economic Review, 91(2), 112-118. Taxman, F., Byrne, J. & Thanner, M. (2002). Evaluating the implementation and impact of a seamless system of care for substance abusing offenders – the HIDTA model. Washington D.C.: National Institute of Justice. Taxman, F.S. & Piquero, A. (1998). On preventing drunk driving recidivism: An examination of rehabilitation and punishment approaches. Journal of Criminal Justice, 26(2), 129-143. Tonry, M. & Lynch, M. (1996). Intermediate sanctions. Crime and Justice, 20, 99-144. Vella, F. & Verbeek, M. (1999). Estimating and interpreting models with endogenous treatment effects. Journal of Business and Economic Statistics, 17, 473-478. Warr, M. (1998). Life-course transitions and desistance from crime. Criminology, 36(2), 183216. Zimring, F.E. & Hawkins, G. (1995). Incapacitation: Penal Confinement and the Restraint of Crime. New York: Oxford University Press. 58