Family Conferencing in South Australia and Re-offending: Preliminary Results from the SAJJ Project YOUTH JUSTICE CONFERENCING AND RE-OFFENDING* Hennessey Hayes Kathleen Daly School of Criminology and Criminal Justice Griffith University Brisbane, Queensland 4111 Australia Address all correspondence to H Hayes: h.hayes@mailbox.gu.edu.au +61 7 3875 5781 (campus office) +61 7 3314 8594 (home office) + 61 4 00 832 573 (mobile) +61 7 3875 5608 (fax) This document is the manuscript that was submitted to the publisher in February 2003. Please refer to the published version: Hayes, H. and K. Daly (2003) ‘Youth justice conferencing and re-offending.’ Justice Quarterly 20(4): 725-64. * This research was funded in part by a grant from the Australian Research Council (1998- 99). We thank Brigitte Bouhours for research assistance, and Lorraine Mazerolle, Paul Mazerolle and the reviewers for their helpful comments on earlier versions. http://www.griffith.edu.au/school/ccj/kdaly_docs/kdpaper17.pdf 1 YOUTH JUSTICE CONFERENCING AND RE-OFFENDING Abstract The literature on restorative justice and re-offending consists largely of comparative analyses of justice system interventions and re-offending (e.g., comparisons of restorative justice conferencing and court by using experimental designs or by conducting meta-analyses) and suggests small but significant differences or no differences. We take an alternative approach in assessing the impact of restorative justice conferencing on re-offending, using data from the South Australia Juvenile Justice (SAJJ) Conferencing Project and police records. Drawing from conference observations and official police data, we explore the relative importance of conference dynamics and offender characteristics in predicting future offending. We find that in addition to well-known predictors of re-offending, such as previous offending and social marginality, when youthful offenders were observed to be remorseful and when the outcome was reached by genuine consensus, young people were less likely to re-offend. These findings suggest that when attention is focused on the claimed benefits of conferencing in its own right, it is possible to identify those elements of restorative processes that are associated with reductions in crime. The results from the SAJJ researchers' observations were generally confirmed in separate analyses of the coordinators' and police officers' conference observations. The SAJJ observational and interview data also show that of the five groups in the conference process -- coordinators, police officers, young people (offenders), victims, and the SAJJ observers -- the victims were least able to correctly predict a young person's post-conference offending and most likely to wrongly think the young person would re-offend. Despite this, close to 90% of victims recommended the government keep conferencing in the justice system. 2 YOUTH JUSTICE CONFERENCING AND RE-OFFENDING Any new criminal justice practice is confronted by a disarmingly simple question, "does it work"? When asking this question, policymakers and others want to know if a practice reduces re-offending or if it has some constructive impact on those caught up in the criminal justice system. Although criminologists have long recognized that research on crime control, justice practices, and re-offending is beset with conceptual, measurement, and interpretive problems, the politics of criminal justice demands evidence of this sort. A recent criminal justice innovation -- conferencing in the response to youth crime -- has been subject to intensive research, although the focus of most studies has been victims' and offenders' perceptions of what happened in a conference, more so than its longer-term effects such as victim recovery and crime reduction. This is understandable in that conferencing is a recent development, having emerged during the 1990s in New Zealand, Australia, North America, England, among other countries. This paper reviews the theoretical claims and empirical studies of conferencing and re-offending, considers problems of measuring re-offending, and presents new data on conferencing and re-offending from the South Australia Juvenile Justice (SAJJ) project. We focus largely, although not exclusively, on developments in Australia and New Zealand because these countries have utilized conferences longer than any other and within a legislative framework. YOUTH JUSTICE CONFERENCING Youth justice conferencing in Australia followed developments in New Zealand after passage of the Children, Young Persons and their Families Act in 1989. South Australia was the first Australian jurisdiction to establish a legislative basis for conferencing (1993) and to use conferences routinely in youth justice cases (1994). Interest in conferencing grew rapidly in the 1990s (see reviews in Bargen, 1996; Daly, 2001a), and by the end of the decade, all eight Australian states and territories established conferencing, and all but two jurisdictions (Victoria and the ACT) have legislated schemes (Daly & Hayes, 2001). Although there are significant differences in how conferences are organized and administered in Australia and New Zealand1, it is possible to characterize what happens, or 1 Variation in conferencing is greater if we include the police-run model of scripted conferencing, the prevalent model in North America and England (see Young, 2001). In Australia and New Zealand the prevalent model normally has at least two professionals present (a police officer and conference coordinator). 3 ideally, is supposed to happen when it is used as a diversion from court prosecution.2 Upon referral by a police officer or magistrate, a young person who has admitted to an offense, his/her parents or caregivers, the victim and his/her supporters, any other relevant parties to the offense, a police officer and conference facilitator (or coordinator/convenor) come together to discuss the offense. There are three phases in a conference, which normally ranges in length from 60 to 90 minutes: introductory, a discussion of the offense and its impact, and a discussion of the agreement. In the introductory phase, the facilitator introduces the participants and outlines what the aims and ground rules are for the conference, its legal status, and the expected roles and responsibilities of those in the room. Most participants receive this information prior to the conference, during the "intake" process. During the introduction, however, the facilitator emphasizes how the conference will proceed and what the conference should achieve – e.g., that everyone will be given an opportunity to discuss the offense and its impact, as well as negotiate a suitable response to it. In the second phase, the police officer reads an official account of the incident, and the young person (offender) describes what happened, as does the victim(s).3 There is a general discussion of the immediate effect of the offense on the young person, victim, and their supporters; and victims may ask offenders questions such as "Why me?" and "Will this happen to me again?" A typical outcome (although not universal) from the discussion phase is an apology by the offender to the victim. In the third phase, conference participants propose and discuss ways for the young person to "make up for the offense," and particular emphasis is given to negotiating an outcome that is agreeable to everyone in the room. In Queensland, for example, the most common way young offenders "make amends" is to offer a verbal apology. Other outcomes, such as direct restitution, are less common and occur in 2 Conferences are also used to provide pre-sentencing advice to judges and magistrates; currently, this option is utilized more frequently in New Zealand than Australia (see Consedine & Bowen, 1999). 3 The sequence of a police officer's reading of the case information and the offender's version of events varies within and across Australian and New Zealand jurisdictions. Victim presence in conferences also varies: in high-volume jurisdictions such as New Zealand and South Australia, victims are typically present in 50 to 60% of conferences, whereas in low-volume jurisdictions such as Queensland, victims are present in over 90% (Hayes et al. 1998). In South Australia when actual victims do not attend conferences, their views may be given by a family member (especially when victims are young), or they may be represented by a Victim Support Services worker. There is also significant jurisdictional variation in which people, at a minimum, must agree to an outcome. In South Australia, it is only the police officer and young person. See Daly and Hayes (2001) for a review of jurisdictional variation in Australia. 4 only about a quarter of all cases (Hayes, Prenzler & Wortley, 1998).4 During the conference, the facilitator is expected to lead the discussion and ensure that everyone has a chance to speak, and the police officer is expected to provide an authoritative legal presence and to describe the likely effects of future offending for the young person. Both professionals are to ensure that the outcome is appropriate and not excessive. While admitting to the offense (i.e., taking responsibility for wrongdoing) is the primary eligibility criterion for referral to a restorative justice conference, police officers consider other factors when exercising their discretion over how to proceed. In all Australian jurisdictions where conferencing is legislated, police officers should consider the nature of the offense and the young person's prior offending history. There is a broad range of offenses that may be conferenced. In some jurisdictions, conferencing tends to be reserved for less serious crimes (e.g., Western Australia). Other states, such as South Australia, tend to conference more serious crimes, including sexual offenses. Prior offending also is a factor. A conference is viewed as the preferred diversionary option for young offenders with either no prior offenses or only minor offenses on record. However, there remains considerable variation in the type of offenders referred to conference, with some jurisdictions referring serious repeat offenders (Daly & Hayes, 2001). THEORISING CONFERENCING AND RE-OFFENDING There are several ways of theorising the links between conferencing and re-offending, and they can be used in combination. The most developed is Braithwaite's (1989) theory of reintegrative shaming, which he developed before conferencing was introduced in New Zealand and which was originally based on the idea that some societies have more effective methods of social control than others because they "shame reintegratively" in that they "follow shaming ceremonies with ceremonies of repentance and reacceptance" (p. 74). Applying these ideas to conferencing, Braithwaite and Mugford (1994: 142) proposed that (1) an illegal act, not an offending person, should be denounced or "stigmatized" and (2) the people who would be most effective in inducing shame in youthful offenders would be significant others in their lives (for example, a parent or favoured aunt). Following Garfinkel (1956), they depicted contemporary criminal justice practices as status degradation 4 Some critics of restorative justice express concern over the potential for young offenders to receive harsher outcomes rendered by vengeful victims. Common outcomes in Queensland and South Australian conferences, however, would seem to challenge this claim (Hayes et al., 1998; Daly, 2003a). 5 ceremonies, which were about "the sequence disapproval-degradation-exclusion". They argued that reintegration ceremonies, which were about "the sequence disapproval- nondegradation-inclusion" (pp. 142) had a greater capacity to reduce re-offending. This comes about for a variety of reasons, but central among them is that shaming-reintegration ceremonies encourage "law-respecting, other-respecting, and self-respecting" identities in offenders (p. 148, see also, p. 141).5 The concept of reintegrative shaming was first linked to police-run conferencing in Wagga, Wagga, New South Wales in 1991; and it continues to enjoy popularity in police-run conferencing schemes in North America and England (see Young, 2001). Although the concept of reintegrative shaming is thought to be the theory underpinning restorative justice,6 the two should not be conflated. In general, reintegrative shaming focuses on how a conference may affect an offender. Restorative justice assumes a broader set of interactions between an offender and victim (and their supporters) where the recognition of the "other" is expected to encourage a more empathic orientation in the offender and a more sympathetic orientation by the victim to the offender's situation. This distinction is important in understanding the different emphases taken in research on conferencing. In Australia, for example, the Reintegrative Shaming Experiments (RISE) project is testing the theory of reintegrative shaming, but the South Australia Juvenile Justice (SAJJ) project is exploring the dynamics of "restorativeness" between victims, offenders, and their supporters. Both projects are also interested in the degree to which procedural justice features in conferences (RISE and SAJJ are discussed in more detail below). During the 1990s, arguments for the benefits of conference over court in reducing re- offending were augmented with Tyler's (1990) work on procedural justice. There is a good deal of fluidity in how procedural justice is defined and measured, owing to its popularity and applicability to many decision-making sites, the criminal justice system being just one. In brief, Tyler argued that the legal process itself was important to people, independent of its outcome. Indicators of procedurally just processes include suspects (or defendants) stating 5 Since this early work, Braithwaite (1999) developed a more comprehensive set of arguments on the relationship of reintegrative shaming, restorative justice, and crime reduction; and with other colleagues (Ahmed et al. 2001), he revised the theory to better articulate the relationship between the emotion of shame and the practice of shaming. 6 Restorative justice covers a diverse array of practices and organisational contexts. For a review of international debates on definition and variation in practice, see Miers (2001); for a review of its varied strands and antecedents, see Daly and Immarigeon (1998). 6 their case (having voice) and being treated with dignity and respect (standing), and decision- makers viewed as trustworthy and neutral. When legal processes are viewed as procedurally just (in other words, as "fair" and handled by competent and trustworthy authorities), there is an affirmation of the legal order and a commitment to law-abiding behavior. These and related arguments on defiance theory (Sherman, 1993) were taken up in the RISE project. In addition to the psychological dynamics of reintegrative shaming, RISE researchers hypothesized that elements of procedural justice would be more frequent in conferences than in court, and that this, in turn, would encourage even greater law-abiding behavior for conference than court offenders. A variant on the procedural justice and reintegration themes is seen in the development of family group conferencing in New Zealand (Maxwell & Morris, 1993). Growing out of Maori political challenges to the dominant white juvenile and child protection system in the 1980s came the idea that a Maori youth's family group should have greater voice and decision-making control than professionals. The implicit idea behind elevating lay actors over legal authorities or social workers is that a family social group not only had some degree of collective responsibility for the lawbreaker's behavior, it could also render better, perhaps more culturally appropriate decisions. Such decisions, in turn, might dissuade a youth from future offending, not because of his or her belief in the legitimacy of the (white) legal system, but because a family group has taken a more central role in the decision-making process. A final strand has featured for some time in research on reparation, diversion, and restitution schemes in North America, England, and Europe, which include mediated meetings between victims and offenders (see, e.g., Dignan, 1992; Marshall, 1992; Messmer & Otto, 1992; Schneider, 1986; Umbreit & Coates, 1993).7 Several claims are made for the enhanced benefits of alternative (that is, non court or non custody) responses to crime: (1) by diverting offenders from prosecution or from custody, fewer will be subject to the stigmatizing effects of the criminal process (or secondary deviance); (2) when offenders meet the people they victimized, they have a "more powerful reformative experience" (Marshall, 1992: 18); and (3) when offenders are given the chance to make up for their offense by paying money or doing community work, they may "re-establish individual self-esteem" and 7 Some argue against viewing restorative justice as confined to mediation, preferring instead to see it as a "fully- fledged systemic alternative" (Walgrave, 1999: 150). We bypass these and other definitional debates in the interest of broadly sketching the lines of argument in the field. 7 become "socially integrat[ed]" into their local community (Messmer & Otto, 1992: 2-3 ). This strand gives greater attention than the others to the benefits of mediation and negotiation in the justice process. While there is a rich set of claimed benefits of conferencing (or related reparative and mediation schemes of the 1980s and 1990s) over regular court processes, studies to date have mainly focused on participant satisfaction and perceived fairness of the process rather than its impact on re-offending. In fact, some argue that conferencing (or related schemes) were not introduced with the explicit intention of reducing re-offending, but with other aims and outcomes in mind (for example, bringing victims into the justice process, holding offenders accountable, rendering fairer justice and more satisfied participants); thus, they suggest that conferences should be assessed in those terms (see, e.g., Dignan, 1991: 37 on the Kettering scheme in England; Morris & Maxwell, 1997: 4 on family group conferencing in New Zealand). These scholars propose, further, that if the impact of alternative practices shows few or no differences in re-offending compared to a standard court process, this signals success, not failure in using a less intrusive and potentially less stigmatising form of legal intervention. Finally, some point out that while restorative language may exist in legislation setting forth alternatives, when such programs or practices are put into operation, they may be little different than regular court routines (for example, there may be low rates of victim participation, or there may be insufficient resources for adequate preparation and follow through of conference outcomes). Thus, we may expect to see no differences in re-offending (Morris & Gelsthorpe, 2000: 25-29; see also Walgrave, 1999). We review the empirical literature on conferencing and re-offending shortly, but before we do, we consider problems of measuring conferencing and re-offending. MEASURING THE IMPACT OF COURT AND CONFERENING ON RE-OFFENDING Three major problems face researchers in measuring the impact of court and conferencing on re-offending: problems of comparison, differences in when offenders take responsibility for their offending, and other temporal differences. Problems of comparison. In analyzing different types of legal interventions and re- offending, it is uncertain what should be studied or compared. For example, in youth justice, 8 if we compare the effect of court and conference interventions, we overlook other possible legal interventions such as formal cautions or no police action at all. Furthermore, we overlook the potential to investigate a new legal intervention in its own right. If we elect the comparative route, we run into problems of sample selection bias by the police or other officials who may consider some individuals to be "more appropriate" than others for the new alternatives. Although such problems can be addressed with randomized field experiments, these too are not without problems of generalizability. Differences in when offenders take responsibility. A central, if overlooked, problem in the court- conference comparative literature is that conferencing (or other diversion schemes) are available only to those youthful offenders who have admitted to an offense (or in New Zealand, who choose "not to deny"). If we ignore, for the moment, field experimental designs and focus instead on studies of conference and court in natural settings, we find that research on the effects of conferences are of offenders who have decided early on to admit responsibility for an offense. Members of the conference group may well differ from those in a court group in that the latter may not initially admit to the offense, although they may later admit after several court appearances. Therefore, the conference group is already distinctive from the court group in ways that are likely to be theoretically related to re-offending. If, hypothetically, members of the court and conference groups were similar in culpability for their offending, but those in the conference group make an admission at an earlier point in the legal process, we might assume that they have moved further along in a reformation process. This problem may be overcome in field experimental designs (such as the RISE project), where admitted offenders are randomly assigned to court or to a diversionary conference. However, and this is an essential point, to be eligible for random assignment to court or conference, an offender must first admit to the offense. Experiment- eligible offenders who are assigned to court may therefore differ from court offenders who do not admit early on. It is possible that the comparison of admitted court and conference youth in the RISE project partly explains the similarity in rates of re-offending for the two groups for three of the four types of offenses studied (see discussion below). Temporal differences in legal intervention. The temporal differences in taking responsibility for an offense affect the window of time for measuring re-offending. Again, let us focus on studies of conference and court in natural settings, not experimental field designs. Imagine for example, two male offenders who were both equally involved in a burglary early in January. One decides to admit to the offense soon after, and his case is diverted to 9 conference, while the other decides to plead not guilty. Based on data from South Australia, the conference case would likely be disposed of in 6 to 8 weeks, that is, by March; but if the case went to court, it would likely take longer, about 3 to 6 months, perhaps not until April. Let us imagine that the two youths are involved in a second burglary in the middle of March. For the court youth, the offense occurred during the pre-disposition period and would not count as an offense; but for the conference youth, it occurred after the conference and thus would count as an offense. With temporal differences in how conference and court cases are handled in natural settings, there are problems in comparing the two. These may be overcome with using field experimental designs, but as we suggest in the discussion of the RISE project below, this comes at a price of selecting the right start date for counting re- offending. To date, research on conferencing and re-offending has largely been of comparing conferences (or other interventions) with court. While comparative studies have value, they are limited, even with the use of randomised field experiments. We propose that research on re-offending may benefit from a focus on within-intervention variation with the aim of identifying variation in claimed features of the conference process with variation in offending. Scholars have long recognised the many problems of defining and measuring re- offending (Maltz, 1984). Despite these problems, Gendreau, Little and Goggin's (1996) meta-analysis of 131 studies of adult re-offending suggests general patterns. They identified re-offending predictors they termed "static" (the things that cannot be changed) and "dynamic" (the things that can be changed). The strongest static predictors were age/gender/race and criminal history, and the strongest dynamic predictors were "criminogenic need" factors, which they defined as antisocial personality, antisocial companions, interpersonal conflict, antisocial attitudes, and substance abuse (Gendreau et al., 1996: 583-584). Ideally, in analyzing conferencing and re-offending, one would want to determine the predictive power of static factors (age, gender, race, prior offense history) and dynamic factors such as an offender's remorse, positive victim-offender movement, the effect of perceived fairness of treatment, and other features of the conference process that may play a role in modifying an offender's future behavior. 10 RESEARCH ON RESTORATIVE JUSTICE, CONFERENCING, AND RE-OFFENDING Two recurring findings emerge from the literature on conferencing (and restorative justice generally). First, with some exceptions (early work in New Zealand by Maxwell & Morris, 1993), there are generally high levels of satisfaction with the process and outcomes by victims and offenders (Braithwaite, 1999; Daly, 2001a; Kurki, 2001; Marshall, 1999; Miers, Maguire, Goldie, Sharpe, Hale, Netten, Uglow, Doolin, Hallam, Enterkin & Newburn, 2001; Umbreit, Coates & Vos, 2001). Latimer, Dowden and Muise's (2001: 9, 11) meta- analysis of 22 studies found significantly higher levels of satisfaction for victims and offenders who participated in restorative processes compared to regular justice system practices. Satisfaction is a notoriously fuzzy concept with varied referents for victims and offenders, but a common denominator in the restorative justice literature is participants' sense of the fairness of the process and the outcome, and having a say in the decision-making process (van Ness & Schiff, 2001: 50; Umbreit et al., 2001: 131-2). This relates to a second recurring finding, especially in studies of conferencing in Australian states and territories: there are high levels of observed procedural justice (by researchers, police officers, and coordinators sitting in the conference) and perceived procedural justice by victims and offenders (Daly, 2001a). RISE researchers say that higher levels of perceived procedural justice are registered by conference than court offenders (Strang, Barnes, Braithwaite & Sherman, 1999), although Kurki's (2001) review of the RISE data suggests these conclusions are somewhat overdrawn in that court and conference differences vary by offense categories and are not consistently strong. How being "satisfied" with a justice activity or judging it to be "fair" affects offenders is uncertain. Latimer et al.'s (2001: 14) meta-analysis suggests that there are greater reductions in offending for those in "restorative justice programs" compared to others. However, mean effect size for the 32 tests examined was .07 (SD = .13), and effect sizes ranged from -.23 to .32. This seems illustrative of the variability in research findings that compare restorative approaches to conventional "controls" (e.g., Braithwaite, 1999; Dignan, 1991; McCold & Wachtel, 1998; McGarrell, 2001; Miers et al., 2001; Sherman, Strang & Woods, 2000). There are many ways of assessing the effects of restorative programs on re- offending. In this paper, we explore the variable effects of conferencing on re-offending, rather than compare conferencing with other interventions. 11 In Australia and New Zealand, two analytical approaches have been used. One approach, employed in the RISE project, utilises the comparative method in a field experiment design, in which admitted offenders are randomly assigned to court and conference (Sherman et al., 2000). A recent New South Wales study also used the comparative method in a retrospective study of official records (Luke & Lind, 2002). A second approach, employed in New Zealand research by Maxwell and Morris (2001) and in the SAJJ project results presented in this article, analyze variability in the conference process itself, including offenders' orientations to the conference and perceptions of how they were treated, in predicting re-offending. We review the findings from the three studies before turning to the SAJJ project. RISE study in Canberra. The RISE project gathered data on four types of offenses and offenders: drink drivers of all ages, violent offenders up to age 29, juvenile property offenders, and juvenile property offenders having organisational victims (shoplifters). Cases meeting the project criteria (including the offender having admitted to the offense) were randomly assigned to court or conference. The project study site was Canberra (the Australian federal capital), where data were gathered from mid-1995 to mid-2000; the Canberra conferencing model is police-run (see Strang et al., 1999). Sherman et al. (2000) report that for one group of offenders, those charged with violent offenses, there was a significant reduction in the average rate of offending in the post-referral period, compared to those who went to court. However, there were no significant court-conference differences in rates of re-offending for the drink driving, property, and shoplifting offenders. We note several problems with the RISE findings. One, discussed above and which features generally in randomized field experiments, is that members of the court and conference groups are similar in ways that are theoretically related to re-offending: they admitted to an offense early on. A second problem is that the observational window in measuring re-offending is 12 months after an offender was referred to court or a conference. In other words, the RISE findings are of the effects of referral to court or conference, not of the effect of the court or conference itself. The time between referral and intervention ranges from about 2 to 4 months, depending on the type of offense and site of disposition (Sherman 12 et al., 2000: 10, footnote 2).8 While the authors give good measurement reasons for this choice, this rationale comes at the price of making a sensible interpretation of the impact of court or conference on an offender's rate of re-offending. Specifically, during the 2- to 4- month window of time, an offender who was assigned to a conference may have offended several times, then attended the conference, and then not re-offended afterwards. The RISE recidivism study would count this conference a "failure" (in re-offending terms) when it was a success. It is essential, we think, to know the rates of offending for those offenders assigned to a court or conference, after they actually attended court or a conference. New South Wales study. Luke and Lind (2002) compared the effects of diversionary conferencing and youth court on re-offending in the Australian state of New South Wales. To control for the effects of prior record, they gathered official data (available up to 30 June 2001) only for first-time offenders (defined as having no prior proven court case), who appeared in the youth court for 12 months preceding (N = 5,516) and following (N = 3,830) the introduction of conferencing on 6 April 1998. They then gathered data only for first-time offenders, who were conferenced during the first 12 months of the diversionary scheme (6 April 1998 to 5 April 1999; N = 590). Several comparisons were made to separate selection effects from "treatment" effects. The results showed that the conference group had a predicted risk of re-offending and rate of re-appearance approximately 15-20% lower than the two court groups, after controlling for the effects of gender, age, offense type, Aboriginality. From these findings, they surmise that the conference experience had a crime reduction effect. New Zealand study. Maxwell and Morris' (2001) model of re-offending includes an offender's early life experiences, features of the conference itself, and post-conference life experiences. They conducted interviews with 108 young people (and their parents), who had taken part in conferences in New Zealand in 1990-91. (Conferences in New Zealand, like those in most Australian jurisdictions, are not run by the police, but by a facilitator with a police officer present.) The interviews were conducted during 1996-97, about 6 ½ years after the conference, and they focused on childhood experiences in school and families, and the offender's and their parent's recollection of what happened in the conference. Data on 8 For three of the four offense types, the time between referral and legal intervention was shorter for the court group than the conference group, a result that reflects RISE-eligible offenders having to admit the offense early on, coupled with the police officers being slow in organizing RISE conferences (Heather Strang 2002, personal communication). 13 reconvictions during the 6 ½ year period for the 108 youth were obtained from police records; these were organized into four categories, ranging from "not reconvicted" (29% of the sample) to "persistent reconvicted" (28%). The authors found that both early life experiences (e.g., poverty and parental neglect) and events after the family conference (e.g., access to training and employment, exposure to criminal associates) had a significant relationship to re-offending, especially for the persistent reconvicted group. However, they also found several conference factors that were related to reductions in re-offending. These included the young person's feeling remorseful, not being made to feel a bad person, feeling involved in the conference decision-making, agreeing with the conference outcome, and meeting the victim and apologising to him or her (Maxwell & Morris, 2001: 253). They argue that while early childhood interventions are likely to prove more effective than justice system interventions, "family group conferences can moderate the patterns of the past and can contribute to the prevention of offending" (p. 260). One potential problem with this study is that the interviews with the original sample of offenders (and their families) were conducted many years after the conference took place. THE SAJJ PROJECT, CONFERENCING, AND RE-OFFENDING In assessing the impact of conferences on re-offending, we asked whether observed behaviors or outcomes in conferences could predict re-offending, over and above those factors that are known to be highly predictive, such as a person's previous offending and social marginality. We approached the problem of predicting re-offending with the assumption that conferences are variable events, with varying degrees of procedural justice and restorativeness. Our approach is similar to Maxwell and Morris in that we are interested in explaining variability in re-offending from variability in the conference process, rather than comparing the different effects of court or conference on re-offending. Our approach differs, however, in that we utilise observational data gathered by SAJJ researchers at the time of the conference, rather than participants' memories of what occurred many years later. The SAJJ project had two waves of data collection in 1998 and 1999 (see Daly Venables, McKenna, Mumford & Johnston, 1998; Daly, 2001a for detailed reviews of the project's methodology, the instruments used, and the theoretical and research justification for item construction). In 1998, SAJJ researchers observed 89 conferences held in the metropolitan area of Adelaide, South Australia, and in several country towns. The observed conferences were selected on the basis of the offense category. SAJJ-eligible offenses were 14 personal crimes of violence and property offenses that involved personal victims or "community victims" (such as schools, churches, or housing trusts). Excluded were shoplifting cases, drug cases, and public order offenses. With 15% of conferences having multiple offenders, the conferences had 107 distinct offenders. The SAJJ researchers conducted in-depth interviews with offenders and victims in 1998, and again in 1999, one year later.9 For each conference, the police officer and the coordinator completed a self- administered survey, and a SAJJ researcher completed a detailed observational instrument. The SAJJ observational survey and the coordinator and police officer surveys contained many detailed items on what occurred in the conference (in particular, a variety of measures of procedural justice and restorativeness), global judgments of how well the conference went and how well it was managed, the professional relationships between the coordinator and police officer, the emotions expressed in the conference, and who was involved in fashioning the agreement and how it was decided. In this paper, we use the SAJJ observational data in analyzing re-offending; the police and coordinator observational data as a confirmatory analysis of the SAJJ observations; and the SAJJ researcher, police, coordinator, offender, and victim survey and interview data to compare predictions of re-offending. SAJJ observational data were keyed to primary offenders; therefore, the following analysis of re-offending is limited to the 89 primary offenders in the conferences. All independent variables are listed and defined in Table 2. These variables were selected from the SAJJ observational instrument and measure offender, offense, and conference characteristics, as well as restorativeness and procedural justice. Defining and measuring "restorativeness" and "procedural justice". The SAJJ observational instrument gathered detailed information from SAJJ researchers about the conference. The theory of restorative justice suggests that offenders and victims who meet to discuss offending and victimization and to negotiate outcomes will be "restored" by a fair and equitable process. SAJJ observers recorded who attended the conference, how it was managed, and how participants related to one another. Focusing their attention on young offenders and their relationships with other people in the room, observers noted several features of the conference that were indicative of restoration and procedural fairness. "Restorativeness" was measured by observing how young people behaved during the 9 Of the total of 107 young people and 89 primary victims (N = 196), SAJJ researchers interviewed 88% in Year 1; of that group, 94% were again interviewed in Year 2 (see Daly, 2001b for details on and analysis of interview 15 conference – i.e., whether they were defiant or remorseful, took responsibility for their actions, understood the impact of their offending, gave a clear story of the offence, were actively involved in the conference discussion, offered an apology, or assured the victim the offence would not happen again. They also observed how effective victims were at describing the impact of offending, whether victims understood offenders' situations, how the conference ended (e.g., "on a high"), and if there was "positive movement" or shared understandings between offenders and victims. "Procedural justice" was measured by observing how the conference was managed – whether participants treated one another respectfully, if the coordinator permitted everyone to "have a say", if the coordinator seemed impartial, if the coordinator negotiated the outcome well, if the offender appeared "powerless", if the offender understood the relationship between the offence and the outcome, or if the outcome was decided by "genuine consensus". These items were coded 1 when restorative and procedural justice elements were observed to a large degree (e.g., "mostly or fully") and 0 when these elements were observed to be less evident (e.g., "somewhat or not at all"; see Table 3). A striking finding from the SAJJ project is that the members of all five groups -- the SAJJ observer, police officer, coordinator, offender, and victim -- report very high levels of observed or perceived procedural justice. SAJJ observers, police officers, and coordinators rated conferences highly on all the measures of procedural justice. In the face-to-face interviews with offenders and victims after the conference (median time, 25 to 33 days), their response to items such as, "were you treated fairly," "were you treated with respect," "did you have a say in the agreement," among others, registered high levels of procedural justice, with 80 to 95% of victims and offenders saying they were treated fairly and had a say. However, compared to the high levels of observed and perceived procedural justice, there was a relatively lesser degree of "restorativeness" observed in the conference by the SAJJ researchers, police officers, and coordinators, or perceived by offenders and victims when asked in the interviews. These and other SAJJ results are discussed elsewhere (Daly, 2001b, 2002, 2003a), but we note them here to provide some context for the re-offending analysis. Defining and measuring re-offending. We chose to define re-offending operationally as any new official incident (which might involve multiple charges or counts), to which the police responded by arrest or apprehension, after the date of the SAJJ conference. These response rates). Of the 79 victims interviewed in 1998, 61 attended the conference and 18 did not. We refer to the former as the "conference victims" in this article. 16 incidents were dealt with by formal caution, conference, or court. Cases pending, but not finalized in court were included. Included were all violent offenses, property offenses, and driving offenses (normally these were drink driving). Excluded were breaches of good behavior bonds, which related to a previous sentence (or undertaking), because these did not reflect a fresh incident. An official incident can have multiple charges, but our count of pre- and post-conference incidents was of distinct incidents (that is, separated by several weeks or a month). We obtained data from the South Australian police on the type of offense, police actions taken (for example, formal caution, referral to family conference or to court), date of disposition, and outcomes (e.g., conference agreement and court sentence) for the total number of young people in the sample (N = 107) as of 21 March 1999.10 The post- conference window of time is thus 8 to 12 months.11 To operationalize offending, we chose to analyze participation, not incidence. We did so for statistical and theoretical reasons. In preliminary analyses of the effect of several key independent variables on the number of incidents post-conference, the correlation matrix showed that some key variables would have been too highly correlated if we used incidence as the dependent variable. Our theoretical rationale is two-fold. First, our analytic focus is on variation within conferences rather than differences between conferences and some other justice intervention. The latter approach is typically concerned with learning how two interventions affect real or estimated levels of offending (or incidence). We are asking instead: what are the variable effects, if any, of restorative justice conferences on re- offending behavior? Thus, our approach is concerned with learning if offender and conference characteristics are associated with any future involvement in crime (or participation). Second, we are interested to shed light on how a new criminal justice intervention affects the number of offenders in a community (that is, participation), rather than the number of criminal events in a community (that is, incidence; see Sherman et al., 10 Our thanks to Senior Sergeant Dave Wardrop for his assistance with the data collection, as well as his cooperation and counsel in helping us to decode the categories and meanings in official criminal histories. 11 The 89 conferences were sampled during a 4-month period, March through June 1998, the first conference observed was on 9 March and the last, on 25 June 1998. While the 8- to 12-month measurement window is somewhat small, it is wide enough to assess the potential impact of conferencing on re-offending. In fact, an analysis of days to first post-conference official incident showed that 50% of those who re-offended 8-12 months post-conference did so within 3 months (90.5 days), and 93% re-offended within 8 months (240 days). So, while this disparate follow-up period may pose a methodological challenge, this outcome is consistent with what other researchers have observed, that is, if re-offending is going to occur, it will happen shortly following an intervention (Maxwell & Morris, 2001: 245). 17 2000). For some datasets and research purposes, incidence may be the more appropriate dependent variable. For this dataset, an analysis of participation was the more sound approach statistically; and at this stage of investigating the effects of conferences on individual offenders, participation is the better measure. We note that with the exception of Maxwell and Morris (2001), no study has yet attempted to relate measures of what occurred in a conference to re-offending. Previous research has simply compared re-offending rates for court and conference groups, and researchers have only speculated on the reasons for the differences (if any) in re-offending for each group. Offending and re-offending in the SAJJ sample. Unless noted otherwise, our depiction of offenders is of the 89 primary offenders in the conferences, not the total of 107. The reason is that in recording conference observations, SAJJ researchers focused on the behavior and interactions of one designated primary offender when there were multiple offenders, as did the coordinators and police officers. For pre-conference activity, 43% of the young people had not had any official contact prior to the incident that led to the SAJJ conference; thus 43% were "first time" offenders.12 Some 27% had had one contact, and 30% had two or more contacts. For post-conference offending, 60% had no official contact following the conference, 17% had one contact only, and 23% had two or more contacts. The number of pre- and post-conference incidents ranged from none (except the SAJJ incident) to 17. One-third of the sample can be termed "experimenters" in that the SAJJ offense was their only offense; an additional 26% did not re-offend post-conference (we refer to this group as the "desisters").13 The "drifters", those with only post-conference offenses, represent only 9% of young offenders.14 Close to one-third are "persisters," who offended both pre- and post-conference. These proportions are summarized in Table 1. 12 The window of time for pre-conference offending could range from several months to 7 years, depending on the youth's age and how long they have lived in South Australia. In an analysis of the first address shown on the police file, over 80% had come to police attention in the previous 2 years. 13 We appreciate that the term "desister" may connote a fixed state when recent research suggests it is better understood as a temporal process (Laub & Sampson, 2001; Bushway, Piquero, Briody, Cauffman & Mazerolle, 2001). 14 We borrowed this term from the theoretical literature on delinquency to capture the notion of moral release (Matza, 1964). 18 Conferencing and re-offending: bivariate relationships We began by running a series of cross-tabulations between a participation measure of re-offending (our dependent variable, where 1 = re-offended post-conference), offender characteristics, and actions or behaviors occurring in conferences that may be linked to re- offending. Table 2 presents the independent variables in five groups: offender-related, offense-related, conference-related, measures of restorativeness, and measures of procedural justice.15 Phi coefficients were used to assess the strength of the association, and the asterisks indicate the level of statistical significance.16 Measures of restorativeness and procedural justice analyzed here are derived from the SAJJ observational data and do not represent how offenders perceived these features of the conference. For the offender-related variables, males comprise 76% of the sample, and Aboriginal youth, 12%.17 Our measure of social marginality (or residential instability) is the number of distinct residential addresses on file with the South Australian police; nearly 30% of the young people had lived at different addresses three or more times.18 From Table 2 one sees significant bivariate associations for re-offending and Aboriginality, social marginality, and prior offending. Some 64% of Aboriginal youth re-offended, whereas 37% of non- Aboriginal youth did; 72% of residentially unstable youth re-offended compared to 28% of more stable youth; and 55% of those who offended pre-conference re-offended compared with 21% with no prior offending. In light of previous research, we were initially puzzled by an apparent lack of association between sex and re-offending. When we examined sex and race together by crosstabulating a combined sex/race variable with re-offending, we found that re-offending was highest (and similar) for Aboriginal male and female youth (63% and 67%, respectively), 15 It was difficult to classify some measures. For example, "offender was defiant" can be a negative indicator of restorativeness (denying the harm to a victim) or an indicator of procedural justice (reacting against perceptions of being treated unfairly by a legal authority). 16 We concur with Sherman et al. (2000) on the interpretation of statistical significance. While it is conventional to regard effects at the .05 level as statistically significant, a more generous error level of .15 may be appropriate in the policy context. We have limited our interpretation of the bivariate and multivariate analyses to the .10 level. However, we would point out that the phi coefficients of .15 to .16 near statistical significance at the .10 level. 17 SAJJ sample demographics are nearly identical to those of all youth conferenced in 1998 (see Daly, 2001b: 22). 19 followed by non-Aboriginal males (40%), and non-Aboriginal females (28%). From this result, coupled with the low number of Aboriginal females in the sample, it was essential that our multivariate analysis use a set of combined, rather than separated sex and race variables, so we created a set of dummy variables for sex/race. The dummy variables were coded as follows: 1 for Aboriginal male or female, 0 otherwise; 1 for non-Aboriginal male, 0 otherwise. Non-Aboriginal female was the redundant category. For the offense-related variables, Table 2 shows no statistically significant association between re-offending and the type of offense (violent or property) or type of victim (individual or organization, although this neared significance), but does for victim-offender relationships. Those offenders who committed a SAJJ offense against someone they knew were less likely to re-offend. For the conference-related variables, victims were present in 74% of conferences, with an additional 6% attended by a representative of the Victim Support Services. The results show that conferences are calm events with low degrees of anger (10%) and moderate levels of crying (25%). There is no association between re-offending and whether a victim (or professional victim representative) was at the conference, and none for the presence of emotions displayed at the conference. The finding for victim presence is noteworthy in the light of criticisms levelled at conferencing schemes with low rates of victim participation (for example, in New Zealand), which commentators have assumed would result in diminished positive effects. SAJJ researchers judged 10% of conferences to be a "waste of time"; a significantly higher share of young people in such conferences re-offended 8 to 12 months later. The "waste of time" variable is a global measure that captures observers' overall impressions of conference success. Of the nine conferences judged as a waste of time, other observational variables show that for these conferences, the offenders were not at all remorseful, there was little or no positive movement between offenders and victims, and offenders did not understand the impact of the crime on the victim. In one conference involving two young men, the victim was particularly vindictive and lied about the extent of the harm. While these conferences may have been procedurally fair, they lacked elements of mutual understanding between the victim and offender. 18 The number of addresses in the police files did not match the number of arrests on record. In 49% of cases, there were more addresses than arrests on file because address information is recorded when any matter comes to police attention, including victimization, being a witness, and child protection matters. In 33% of cases there were fewer addresses than arrests and in 18% there were equal addresses and arrests on file. 20 Some measures of restorativeness are related to re-offending, and others not. (Note that phi coefficients may be identical in magnitude for some variables, but significant for one and not the other because the number of conferences is smaller for some variables, that is, those involving victims who were present at the conference.) When young people were observed to be remorseful and when SAJJ observers recorded the conference as having ended on a high, a positive note of repair and good will, youth in those conferences were less likely to re-offend. There is no association, however, between re-offending and an offender understanding the impact of the crime on the victim, or a victim being effective in describing the impact of the offense. Based on the expectations of restorative justice theory, we would have expected to see a stronger relationship of re-offending and variables tapping "movement and mutual understanding" between victims, offenders, and their supporters, but this did not emerge. Several measures of restorativeness neared statistical significance at the .10 level: when offenders took responsibility for the offense, were actively involved, and were not defiant, they were less likely to re-offend. For procedural justice, there are no significant associations for what we might term the standard or Tyler-based measures of procedural justice (such as the police officer treating the offender with respect, the coordinator permitting everyone to have a say, the coordinator being impartial, process of deciding outcome was fair) mainly because they had too little variability (ranging from 89 to 99%).19 This is an important finding. Standard measures of procedural justice may be better able to distinguish court and conference processes than they can variability among conferences. One conference-specific procedural justice measure is linked to re-offending. When the outcome was observed to be decided by genuine consensus, young people in those conferences were less likely to re-offend. Conferencing and re-offending: multivariate analyses These bivariate results helped us to identify the variables for our multivariate analyses. We used logistic regression to analyze the effects of offender-related, offense- related and conference-related characteristics, along with measures of restorativeness and procedural justice, on our measure of re-offending. We were interested to determine if, in addition to those offender-related variables known to be predictive of re-offending (not only 19 For example, SAJJ observers recorded the following: police officers treated the young person with respect (99% yes), the young person treated the police officer with respect (96% yes), the coordinator permitted everyone to have a say (98% yes), and the coordinator seemed impartial (93% yes). 21 from the literature, but also from our bivariate analysis of the data), other conference-related variables were predictive. We explored many combinations of variables, although we were limited by the number of variables we could include in the equation because of the number of cases. Table 3 shows the fruits of our analyses. We estimated two models: (1) a full model, which has the variables known to be predictive of re-offending (prior offending, sex/race- ethnicity, social marginality -- our control variables), together with key conference variables; and (2) a reduced model, which included only the controls. Examining the full model, we find that all but one of the control variables were significantly associated with re-offending. Non-Aboriginal males were more likely to re-offend, as were residentially unstable youth and those with pre-conference offending. Over and above these offender characteristics, there were two significant conference variables. When young people were observed to be remorseful and when they participated in conferences in which outcomes were decided by genuine consensus, they were less likely to re-offend. Overall, 80% of cases were correctly classified with this model, and a pseudo measure (Nagelkerke R2) of explained variation was 43%. With the reduced model, however, 72% of cases were correctly classified, and a pseudo measure of explained variation was 30%. We note that this is a stringent test of the effects of conferencing on re-offending: one might have assumed that the control variables, especially pre-conferencing offending and our measure of social marginality, would have overwhelmed any potential conference effects. The control variables did account for most of the variability in predicting re-offending, and as we shall see, the odds ratios for these variables are large. However, the statistical significance of the two conference variables, coupled with the improvement in explained variation, suggests that they have unique effects. The odds ratios (in order of magnitude) for the full model show the following. The odds of re-offending are nearly nine times greater for youths who are residentially unstable (three or more residential changes on file with the police) compared to those with only one or two residential movements. Compared to other offenders, the odds of re-offending are three times greater for non-Aboriginal males. Youth with a history of prior offending have an odds of re-offending about three times greater than those with no detected prior offending. However, when particular conference features are present (offender remorse and consensual decision making), the odds of re-offending are reduced. When young people show remorse in the conference, the odds of re-offending are reduced by about a third (0.33); and when 22 outcome decisions are arrived at by genuine consensus, the odds of re-offending are reduced by about a quarter (0.27).20 While the odds ratios derived from the beta coefficients in Table 3 do not permit us to say how much more likely one outcome is over another, the relative risk ratio does (Agresti, 1990: 17-18). Translating the odds ratios from the 2 X 2 contingency tables for the predictors in the full model to risk ratios (i.e., cross-classifying the predictors with re- offending), we find that Aboriginal males and females were 2.57 times as likely to re-offend as non-Aboriginal offenders. Non-aboriginal males were equally likely to re-offend compared to other offenders (i.e., this dummy variable is independent of re-offending in the bivariate analysis; relative risk ratio = 0.982; odds ratio = 0.944). Those with prior offending were 1.79 times as likely to re-offend as those with no prior offending, and residentially unstable offenders (i.e., those with three or more residential addresses on their police record) were 3.79 times as likely to re-offend as more residentially stable youth. Conference-related variables seemed to diminish the likelihood of re-offending. Young offenders who were observed to be remorseful were 0.55 times as likely to re-offend as unremorseful youth. Also, offenders whose conference agreements were observed to be consensual were 0.74 times as likely to re-offend as offenders whose conference agreements were not. The odds ratios show that offender characteristics play a marked role in predicting future behavior; however, what happens in a conference may also be important. This outcome is consistent with Maxwell and Morris' (2001) findings from New Zealand, which 20 Several reviewers suggested that our single-item measures may not have adequately captured procedural justice and restorativeness and that a scale might be preferable. There are two key points we would make on this. First, we are not analyzing whether the presence of procedural justice or restorativeness generally predicts re-offending. In fact, the procedural justice variables were so consistently high, with little variability, they would be poor predictors, whether as single items or constructed as a scale. The variable "consensus in deciding the outcome" is reflective more of the agreement in negotiation amongst the parties (in particular, the offender and the police officer) than it is "fairness" or "respect"; perhaps for this reason, it has more variability and is more strongly predictive. Second, scales may be appropriate for some statistical analyses but not others. We decided, however, to create two additive scales for restorativeness and procedural justice by summing across the items listed in Table 2 and averaging them (i.e., scale scores ranged from 0 to 1). The mean restorativeness score and associated Cronbach's alpha were .46 (SD .24) and .75, and the mean procedural justice score and associated alpha were .83 (SD .13) and .26. The very low alpha for procedural justice likely flows from the low variance for many of the scale items (see Table 2). We entered these variables into a revised logistic regression model. The scale measures improved the model only slightly over the reduced model in Table 3, with 74% of cases correctly classified and a pseudo-R2 of 34%. The betas for restorativeness and procedural justice were -.82 (p=.49) and –3.02 (p=.17), respectively. The associated odds ratios were .44 and .05, respectively. Other variables entered into the revised model (sex/race-ethnicity, pre-conference offending, residential mobility) retained significance as in Equation 1 of Table 3. This is an instance when particular variables such as remorse are predictive of future behaviour (i.e, not re-offending) whereas a scale of "restorativeness", which includes many types of behaviours and interactions between offenders and victims, is not predictive. 23 show that prior offending and negative life experiences bear the heaviest on re-offending, but that re-offending is less likely when young people feel remorseful, are involved in conference decision-making, and agree with the conference outcome. We do not suggest that conferences "produce" remorseful youth who are less likely to re-offend, or that consensual decision- making has a causal impact on re-offending. It may be that some youth are already prepared to be remorseful at the conference and to be agreeable to outcome negotiations. It is not possible in this quantitative analysis to disentangle the effects of offender characteristics (that is, already feeling remorseful or readiness to negotiate an outcome) and a conference context that may encourage or promote these sensibilities, although this could be pursued in a qualitative analysis of selected cases.21 However, to the extent that certain behaviors and decision-making processes in conferences are indicative of future behavior, research from New Zealand and ours from South Australia show strikingly similar patterns. Consistency and reliability Colleagues and reviewers have queried the inter-rater reliability of the SAJJ observations, and thus we pause to consider questions of consistency and reliability. A decision was taken early in the project that only one SAJJ researcher should attend each conference. Even with the presence of one researcher, the coordinator and police officer were concerned with the influence this might have on the conference; two or more researchers would be excessive. Moreover, for ethical reasons, we wished to minimise the level of intrusion that research observations presented. One method of increasing consistency of observation is by developing a shared understanding of what the observational items mean; this was accomplished in the pre-research period as the Project Director (Daly) worked intensively with the SAJJ researchers in fashioning the items used in the observational protocol.22 However, there is another way to test the SAJJ results: by comparing them with 21 In a separate analysis of the SAJJ data, Daly (2003b) finds that the conference had an impact on young persons' views of the legal system. Those young people who stayed out of trouble were more likely to have positive views of the legal system, saying this was a result of the conference. 22 See Daly et al. (1998) for a discussion of instrument development. Daly first worked with the police officers in identifying and finalizing observational items, and revising the items further when working with the coordinators. Next, she wrote memos about the SAJJ observational instrument (which built in part on the police and coordinator surveys), which became the basis for intensive work with the SAJJ researchers in finalizing the SAJJ observational instrument. The aim was to develop items that made sense to the police and coordinators, and to create similar observational items across the three groups, although the SAJJ observational protocol was more detailed. 24 the observations of the police officers and coordinators in each conference. For the 89 conferences, five SAJJ observers and ten coordinators completed the conference observation protocol or survey; a higher number of police officers (N = 21) participated in conferences and completed the survey, with six officers involved in just one SAJJ conference. In light of these numbers (especially for the police), coupled with the different perspectives and experiences that members of each group may bring to the conference process, this is a demanding test of consistency and reliability. Our analysis finds a good to high degree of consistency across the key variables between the pairs (Appendix, Table A) and a high level of association between re-offending and the remorse and consensus variables (Appendix, Table B). In running the logistical regression analysis with the police officer and coordinator judgments of offender remorse and degree of genuine consensus, we find that while the SAJJ model is replicated by the coordinator model, it is partly replicated by the police model (Appendix, Table C). Specifically, remorse and conference consensus predicted re-offending in the coordinator model, but remorse (not conference consensus) predicted re-offending in the police model (see discussion in the Appendix). Conditional probabilities of re-offending Logistical regression odds ratios can be helpful in demonstrating the relative impact of individual offender and conference characteristics on re-offending. But in addition, logistic regression allows us to calculate the conditional probability of re-offending for various case scenarios (Poulos & Orchowsky, 1994; Hosmer & Lemeshow, 1989), by solving for the conditional probability = 1 / 1 + e –logit, where logit is the linear regression equation or logit transformation (B0 + B1X1 + … + BiXi). Using the coefficients from the SAJJ equation, we examined the impact of youth characteristics and conference elements on the probability of re-offending, and found that offender remorse and consensual decision-making lowered the probability of further offending. This was true for youth who possessed characteristics associated with further offending ("high-risk") and for those who did not ("low-risk") (see Table 4). When a young person was an Aboriginal male or female, residentially unstable, had a record of prior offending, and was not observed to be remorseful and the conference decision was not observed to be consensual (high-risk youth, poor conference elements), the conditional probably of a further offense was very high at .94. But when offender remorse and consensual decision-making were present for high-risk Aboriginal youth, the probability 25 of further offending decreased from .94 to .68.23 For a non-Aboriginal male with no prior offending history, who was residentially stable, who was observed to be remorseful and whose conference decision was observed to be consensual (low-risk youth, good conference elements), the conditional probability of re-offending was .11. But when good conference elements were absent for this youth (low-risk youth, poor conference elements), the probability of re-offending increased from .11 to .48. The shift in conditional probabilities of re-offending is more dramatic for high-risk non-Aboriginal females and low-risk Aboriginal males and females. For high-risk non-Aboriginal females, when good conference elements were absent, the conditional probability of offending was .80; but when good conference elements were present, it was .35. For low-risk Aboriginal males and females, when good conference elements were absent, the conditional probability of re-offending was .52; but when the elements were present, the probability was .12. WHO PREDICTS RE-OFFENDING BEST? Pooling the 1998 interview data from victims and offenders, the survey data from police officers and conference coordinators, and the observational data from SAJJ researchers, we cross-classified the responses to questions about the likelihood of future offending with our data on post-conference offending. In the interviews with young people in 1998, we asked them, "Do you think it likely or unlikely that you will be involved in a serious offense in the future?," and in the interviews with victims who attended the conference, we asked, "Do you think it likely or unlikely that the offender will be involved in a serious offense in the future?" The coordinator and police survey item, and the SAJJ observer item was, "Do you think it likely or unlikely that the young person will be involved in a serious offense in the future, one that comes to the attention of the police?" All were asked to answer on a four-point scale ranging from very likely to very unlikely, with an additional category of "do not know" or "unsure". The results show that the conference victims were least accurate in predicting young people's future offending (Table 5). Including the "do not know" and "unsure" responses, 46% of victims were correct in predicting future offending (or non-offending). The better predictors were the police (65%), offenders (60%), SAJJ observers (57%), and coordinators (55%). (If we remove the "do not know/unsure" category, the police and coordinators are the 23 We included the combined sex/race-ethnicity variable for Aboriginal males and females in the conditional probability calculations because it neared significance at the .10 level. 26 best predictors, making accurate predictions for about three-fourths of offenders.) False positives, that is, wrongly predicting future offending, were more frequent for the victims (38%) than for other groups; and not surprisingly, false negatives (predicting non-offending, when it did occur) were most frequent for the offenders (25%). Across the five groups, the coordinators were most likely to reserve judgment on re-offending with 26% saying they did not know or were unsure, whereas the percentages for the SAJJ observers, police, offenders, and victims were consistently smaller (10 to 13%) Several implications can be drawn from this analysis. The coordinators, police officers, and SAJJ observers not only have a more accurate view of a young person's future behavior (law-abiding and otherwise), but also they are more optimistic of the potential for young people to change, as evidenced by their relatively low percent of false positives. Members of these groups appear to believe in the potential for conferencing to promote change and social reintegration. By comparison, the conference victims appear to be more skeptical in the potential for conferencing, and they have a more static and negative view of offenders. One of the hallmarks of restorative justice (and re-integrative shaming) -- that the process should strive to separate the "badness" of acts from the identities of individuals committing them -- may be especially difficult for victims to contemplate and accept. Nevertheless, nearly 90% of conference victims recommended the government keep conferencing in the justice system. SUMMARY AND DISCUSSION New justice interventions such as conferencing are often required to demonstrate their claimed benefits by showing reductions in re-offending. Whether this should be a requirement is a matter of debate among researchers in the conferencing and restorative justice areas. Advocates and commentators give a variety of reasons for why conferencing is likely to be more effective than regular court processes in reducing crime. These include the positive (crime reduction) impact of how decisions are made and who participates in making them; changes in the sources of knowledge about crime and its impact (that is, away from lawyers to actual victims, offenders, and their supporters); changes in ways of thinking about sanctions and their purposes; and a renewed interest in justice system interactions as a form of moral education, a dialogic encounter intended to persuade and to show the offender how crime hurts victims and other members of the community. Analyzing how these and other elements may affect re-offending should be a focus of research and policy. However, to date, 27 this has not been the case. Rather, the more frequent approach is to compare measures of re- offending for different kinds of legal interventions. While such an approach has its place and should be carried out, it has limitations. Specifically, the problems of comparing court and conference in natural settings include sample selection bias by the police (or other referring group); differences in offenders' orientations to more immediately admit to an offense (or to deny it); and temporal differences in the handling of court and conference cases, which give different windows of time for measuring re-offending. These problems of comparison may be overcome with randomized field experiments, but these too have problems. Specifically, we noted that for an offender to be eligible for an experiment, a typical protocol is that s/he must admit to the offense first, before being assigned to court or conference. Offenders who admit early on to an offense may differ in theoretically important ways from those who do not. On ethical grounds, randomised field experiments are obliged to use this protocol. Nonetheless, it comes at a price of screening out a segment of arguably the most interesting people to study: those who may have more entrenched patterns of denial and who would be most likely to re-offend. Moreover, a significant problem for randomized field experiments is when to begin the clock in counting re-offending. The RISE project started the clock at the date of referral, and while there may be good reasons for this choice, it comes at the price of making accurate claims. If claims are to be made about the different effects of court and conference, a logical time to start the clock is after an offender has experienced the conference or court encounter, not before it. We have emphasized the benefits of studying re-offending by focusing on the variable qualities and effects within a conference (or restorative justice) practice. This approach permits an analysis of the theorized linkages between conference processes and re-offending, which can only be surmised or inferred when using a conference-court comparison. It is important that researchers take a more in-depth look at the conference process (or other similar process) in its own right to determine whether any of the claimed elements are present and with what impact. In taking this approach, we find that conferences have variable degrees of claimed elements present, and that this variability can be linked to re-offending. Specifically, holding constant previous offending and indicators of marginality, we find that young people, who were observed to be remorseful and who were in conferences where outcomes were achieved by genuine consensus, were less likely to re-offend. When extending the SAJJ model to the two other observers in the conference room (the police officer and coordinator), we find that offender remorse remains highly predictive, although 28 conference consensus is predictive for one group, but not the other. While we would have preferred a perfect replication of the SAJJ model, we are satisfied that these results add credibility to the SAJJ model. Research findings from a single jurisdiction can be limited in generalizability, but our results from South Australia are similar to those of Maxwell and Morris (2001) for New Zealand, and this gives them added weight. Our interpretations are also similar. Offenders come into the justice system with varied degrees of negative life experiences, social marginality and disadvantage, and previous contact with the justice system. We would expect these factors to have a strong impact on predicting future offending, irrespective of the character of any justice system intervention. Over and above these factors, both the SAJJ project and Maxwell and Morris (2001) find that conference-specific factors are indicative of reduced re-offending, in particular, offender remorse and involvement in decision-making. Based on these statistical results, we cannot make a causal claim that conferences induce remorse or contrition or that consensually-based outcomes cause reductions in re-offending. To make such a claim would require an analysis of offenders' pre-conference orientations and the significance of the conference in the life of the young person and his or her supporters. Moreover, offenders, victims, and their supporters come to conferences with varying degrees of knowledge about the process and what they are expected to do; these factors likely affect the probability of good conference elements emerging (or not). Jurisdictions in Australia and New Zealand vary in pre-conference preparation by coordinators. Some jurisdictions (like Queensland) routinely have face-to-face pre-conference meetings with victims and offenders, whereas other jurisdictions (South Australia and the ACT) do not. We have tried to make a case for the benefits of studying within-conference variation for predicting re-offending, but we recognize its limitations. Specifically, policymakers may be more impressed with comparisons of different legal interventions, especially if a new one can be shown to outperform an older, more established form. In comparing the benefits of conference over court, however, researchers need to specify which behaviors and interactions occur in conferences more often than in court, which are then linked to reductions in re- offending. From the recidivism literature and our analysis, such reductions will be at the margins, with other variables such as pre-conference offending and social marginality having substantially greater influences. Our analysis shows that while about one-quarter of young people were changed by the conference process toward more law-abiding behavior, the victims attending conferences 29 were skeptical of the capacity for youthful offenders to change. Despite the fact that two- thirds of conference victims believed the young person who victimized them or their property would be in trouble again, close to 90% of them said the government should keep family conferencing. Victims therefore distinguish between seeing conferencing as a good thing, to be maintained, and holding ambivalent (and inaccurate) attitudes about the potential for youthful lawbreakers to change. These results lead us to speculate that in supporting a new justice idea in the youth justice area, evidence of reductions in re-offending may be more salient to policymakers or politicians than to their constituents. 30 REFERENCES Agresti, A. (1990). Categorical Data Analysis. New York: Wiley. Ahmed, E., Harris, N., Braithwaite, J., & Braithwaite, V. (2001). Shame Management through Reintegration. Cambridge: Cambridge University Press. Bargen, J. (1996). Kids, cops, courts, conferencing and children's rights: a note on perspectives. Australian Journal of Human Rights, 2, 209-228. Braithwaite, J. (1989). Crime, Shame and Reintegration. Cambridge: Cambridge University Press. Braithwaite, J. (1999). Restorative justice: assessing optimistic and pessimistic accounts. Crime and Justice: A Review of Research, 25, 1-127. Braithwaite, J. & Mugford, S. (1994). Conditions of successful reintegration ceremonies. British Journal of Criminology, 34, 139-171. Bushway, S., Piquero, A., Briody, L., Cauffman, E. & Mazerolle, P. (2001). An empirical framework for studying desistance as a process. Criminology, 39, 491-516 Consedine, J. & Bowen, H., eds. (1999). Restorative Justice: Contemporary Themes and Practice. Lyttelton, New Zealand: Ploughshares Publications. Daly, K. (2001a). Conferencing in Australia and New Zealand: variations, research findings, and prospects. In A. Morris and G. Maxwell (eds.), Restorative Justice for Juveniles: Conferencing, Mediation and Circles. Oxford: Hart Publishing. Daly, K. (2001b). South Australia Juvenile Justice (SAJJ) Technical Report No. 2: Research Instruments in Year 2 (1999) and Background Notes. Brisbane: School of Criminology and Criminal Justice, Griffith University. Available at http://www.aic.gov.au/rjustice/sajj/index.html Daly, K. (2002). Restorative justice: the real story. Punishment & Society, 4, 55-79. Daly, K. (2003a). Mind the gap: restorative justice in theory and practice. In A. von Hirsch, J. Roberts, A. E. Bottoms, K. Roach, & M. Schiff (eds.) Restorative Justice and Criminal Justice: Competing or Reconcilable Paradigms?. Oxford: Hart Publishing. Daly, K. (2003b). Making variation a virtue: evaluating the potential and limits of restorative justice. In E. Weitekamp & H. Kerner (eds.) Restorative Justice in Context: International Practice and Directions. Devon, England: Willan Publishing. Daly, K. & Hayes, H. (2001). Restorative justice and conferencing in Australia. Trends and Issues in Criminal Justice No. 186. Canberra: Australian Institute of Criminology. Available at http://www.aic.gov.au/publications/tandi/tandi186.html 31 Daly, K. & Immarigeon, R. (1998). The past, present, and future of restorative justice: some critical reflections. Contemporary Justice Review, 1, 31-45. Daly, K, Venables, M., McKenna, M., Mumford, L. & Christie-Johnston, J (1998). South Australia Juvenile Justice (SAJJ) Research on Conferencing, Technical Report No. 1: Project Overview and Research Instruments. Brisbane: School of Criminology and Criminal Justice, Griffith University. Available at http://www.aic.gov.au/rjustice/sajj/index.html Dignan, J. (1991). Repairing the Damage: An Evaluation of an Experimental Adult Reparation Scheme in Kettering, Northamptonshire. Sheffield, England: Centre for Criminological and Legal Research, University of Sheffield. Dignan, J. (1992). Repairing the damage: can reparation be made to work in the service of diversion? British Journal of Criminology, 32, 453-472. Garfinkel, H. (1956). Conditions of successful degradation ceremonies. American Journal of Sociology, 61, 420-424. Gendreau, P, Little, T. & Goggin, C. (1996). A meta-analysis of the predictors of adult offender recidivism: what works! Criminology, 34, 575-607. Hayes, H, Prenzler, T & Wortley, R. (1998). Making Amends: Final Evaluation of the Queensland Community Conferencing Pilot. Griffith University: Centre for Crime Policy and Public Policy. Available at http://www.gu.edu.au/school/ccj/amends.pdf . Harris, N. & Burton, J. (1996). The Reliability of Observed Reintegrative Shaming, Shame, Defiance and Other Key Concepts in Diversionary Conferences: RISE Working Paper Number 5. Canberra: Australian National University. Available at http://www.aic.gov.au/rjustice/rise/working/risepap5/risepap5.pdf Harris, N. & Burton, J. (1998). Testing the reliability of observational measures of reintegrative shaming at community accountability conferences and at court. Australian and New Zealand Journal of Criminology, 31, 230-241. Hesmer, D. & Lemeshow, S. (1989). Applied Logistic Regression. New York: John Wiley & Sons. Kurki, L. (2001). Evaluation of restorative justice practices. Paper presented to Cambridge Seminar on Restorative Justice, Toronto, May. Latimer, J., Dowden, C. & Muise, D. (2001). The Effectiveness of Restorative Justice Practices: A Meta-Analysis. Department of Justice: Research and Statistics Division. Available at http://canada.justice.gc.ca/en/ps/rs/rep/meta-b-e.html 32 Laub, J. & Sampson, R. (2001). Understanding desistance from crime. Crime and Justice: An Annual Review of Research. Chicago, Ill.: University of Chicago Press. Luke, G. & Lind, B. (2002). Reducing juvenile crime: conferencing versus court. Crime and Justice Bulletin: Contemporary Issue in Crime and Justice. New South Wales Bureau of Crime Statistics and Research. Available at http://www.lawlink.nsw.gov.au/bocsar1.nsf/files/CJB69.pdf/$file/CJB69.pdf Maltz, M. (1984). Recidivism. New York: Academic Press. Matza, D. (1964). Delinquency and Drift. New York: Wiley. Marshall, T. (1999). Restorative Justice: An Overview. London: Home Office. Marshall, T. (1992). Restorative justice on trial in Britain. In H. Messmer and H.-U. Otto (eds.) Restorative Justice on Trial. Rotterdam: Kluwer Academic Publishers. Maxwell, G. & Morris, A. (1993). Family, Victims and Culture: Youth Justice in New Zealand. Wellington: Institute of Criminology, Victoria University of New Zealand. Maxwell, G. & Morris, A. (2001). Family group conferences and reoffending. In A. Morris and G. Maxwell (eds.), Restorative Justice for Juveniles: Conferencing, Mediation and Circles. Oxford: Hart Publishing. McCold, P. & Wachtel, B. (1998). Restorative Policing Experiment: The Bethlehem, Pennsylvania Police Family Conferencing Project. Pipersville, PA: Community Service Foundation. McGarrell, E. (2001). Restorative Justice Conferences as an Early Response to Young Offenders. Washington DC: Office of Juvenile Justice and Delinquency Prevention, US Department of Justice. Messmer, H. & Otto, H. (1992). Restorative justice: steps on the way to a good idea. In H. Messmer & H.-U. Otto (eds.), Restorative Justice on Trial. Rotterdam: Kluwer Academic Publishers. Miers, D. (2001). An International Review of Restorative Justice. Crime Reduction Research Series Paper 10. London: Home Office. Available at http://www.homeoffice.gov.uk/rds/prgpdfs/crrs10.pdf Miers, D., Maguire, M., Goldie, S., Sharpe, K., Hale, C., Netten, A., Uglow, S., Doolin, K., Hallam, A., Enterkin, J. & Newburn, T. (2001). An Exploratory Evaluation of Restorative Justice Schemes. Crime Reduction Research Series Paper 9. London: Home Office. Available at http://www.homeoffice.gov.uk/rds/prgpdfs/crrs09.pdf 33 Morris, A. & Gelsthorpe, L. (2000). Something old, something borrowed, something blue, but something new? A comment on the prospects for restorative justice under the Crime and Disorder Act 1998. Criminal Law Review, Jan, 18-30. Morris, A. & Maxwell, G. (1997). Family group conferences and convictions. Occasional Papers in Criminology New Series No. 5. Wellington, New Zealand: Institute of Criminology, Victoria University of Wellington. Poulos, T. & Orchowsky, S. (1994). Serious juvenile offenders: predicting the probability of transfer to criminal court. Crime and Delinquency, 40, 3-17. Schneider, A. (1986). Restitution and recidivism rates of juvenile offenders: results from four experimental studies. Criminology, 24, 533-552. Sherman, L. (1993). Defiance, deterrence, and irrelevance: a theory of the criminal sanction. Journal of Research in Crime and Delinquency, 30, 445-473. Sherman, L., Strang, H. & Woods, D. (2000). Recidivism Patterns in the Canberra Reintegrative Shaming Experiments (RISE). Canberra: Centre for Restorative Justice, Research School of Social Sciences, Australian National University. Strang, H, Barnes, G., Braithwaite, J. & Sherman, L. (1999). Experiments in Restorative Policing: A Progress Report on the Canberra Reintegrative Shaming Experiments. Canberra: Australian Federal Police and Australian National University. Available at http://www.aic.gov.au/rjustice/rise/index.html Tyler, T. (1990). Why People Obey the Law. New Haven: Yale University Press. Umbreit, M. & Coates, R. (1993). Cross-site analysis of victim offender mediation in four states. Crime and Delinquency, 39, 565-585. Umbreit, M, Coates, R. & Vos, B. (2001). Victim impact of meeting with young offenders: two decades of victim offender mediation practice and research. In A. Morris and G. Maxwell (eds.), Restorative Justice for Juveniles: Conferencing, Mediation and Circles. Oxford: Hart Publishing. van Ness, D. & Schiff, M. (2001). Satisfaction guaranteed? The meaning of satisfaction in restorative justice. In G. Bazemore and M. Schiff (eds.), Restorative Community Justice. Cincinnati, OH: Anderson Publishing Company. Walgrave, L. (1999). Community service as a cornerstone of a systemic restorative response to (juvenile) crime. In G. Bazemore and L. Walgrave (eds.), Restorative Juvenile Justice: Repairing the Harm of Youth Crime. Monsey, NY: Criminal Justice Press. 34 Young, R. (2001). Just cops doing "shameful" business? Police-led restorative justice and the lessons of research. In A. Morris and G. Maxwell (eds.), Restorative Justice for Juveniles: Conferencing, Mediation and Circles. Oxford: Hart Publishing. 35 Table 1. Pre- and post-conference offending Official contact includes any incident that led to a formal caution, conference, or court disposition. (N=89) pre-conference contact: 0 (except SAJJ case) 43% 1 27 2+ 30 3+ 17 4+ 14 5+ 7 post-conference contact: 0 60% 1 17 2+ 23 3+ 18 4+ 11 5+ 10 Group Offended pre-SAJJ conference Offended post-SAJJ conference Percentage of primary offenders (N=89) Experimenters no no 34 Desisters yes no 26 Drifters no yes 9 Persisters yes yes 32 1 Table 2. Bivariate relationships between selected variables and re-offending Variable Categories Percent re- offending Phi Offender-related variables Sex (76% male) 1=male 0-female 43 33 .08 Racial-ethnic identity (12% Aboriginal) 1=Aboriginal 0=non-Aboriginal 64 37 .18* Number of addresses on police file (28% 3 or more) 1=3+ 0=1 to 2 72 28 .40** Pre-SAJJ conference offending (57% yes) 1=yes 0=no 55 21 .34** Offense-related variables Type of offense (44% violent) 1=violent 0=property 36% 44% -.08 Victim-offender relationship (52% known) 1=known 0=stranger 32% 50% -.18* Type of victim (80% personal) 1=personal 0=organization only 37% 56% -.16 Conference-related variables Victim or representative present (80% yes) Victim only present (74% yes) 1=yes 0=no 1=yes 0=no 41% 39% 41% 39% .02 .02 Conference was a waste of time (10% yes) 1=yes 0=no 78% 36% .26** Angry/aggressive remarks aimed at offender (10% yes) 1=yes 0=no 57% 39% .10 Crying by participants (25% yes) 1=yes 0=no 46% 39% .06 Measures of restorativeness Offender gave a clear story (71% mostly or fully) 1=mostly or fully 0=somewhat or not at all 43% 35% .08 Offender accepted responsibility (62% mostly or fully) 1=mostly or fully 0=somewhat or not at all 35% 50% -.15 Offender actively involved (78% mostly or fully) 1=mostly or fully 0=somewhat or not at all 36% 55% -.16 Offender was defiant (30% somewhat, mostly or fully) 1=somewhat, mostly or fully 0=not at all 52% 36% .15 YP was remorseful (49% mostly or fully) 1=mostly or fully 0=somewhat or not at all 27% 53% -.27** Offender offered spontaneous apology (41% mostly or fully)‡ 1=mostly or fully 0=somewhat or not at all 33% 46% -.13 Offender assured victim offense wouldn't happen again (58% mostly or fully) ‡ 1=mostly or fully 0=somewhat or not at all 34% 50% -.16 Offender understood impact of crime on victim (52% mostly or fully) ‡ 1=mostly or fully 0=somewhat or not at all 41% 41% .00 2 Table 2 cont. Victim effectively described impact of offense (72% mostly or fully) ‡ 1=mostly or fully 0=somewhat or not at all 39% 45% -.05 Victim understood offender's situation (34% mostly or fully) ‡ 1=mostly or fully 0=somewhat or not at all 46% 38% .07 Conference ended on a high (55% yes)† 1=yes 0=no 33% 50% -.18* Positive movement or mutual understanding between offender and victim expressed in words (31% mostly or fully) ‡ 1=mostly or fully 0=somewhat or not at all 41% 41% .00 Positive movement or mutual understanding between offender and victim expressed symbolically (20% mostly or fully) ‡ 1=mostly or fully 0=somewhat or not at all 36% 42% -.05 Measures of procedural justice Process of deciding outcome was fair (89% yes) 1=yes 0=no 39% 50% -.07 Conference coordinator negotiated outcome well (89% yes) 1=yes 0=no 41% 40% .00 Outcome decided by genuine consensus (64% yes) 1=yes 0=no (offender accepts/accepts with reluctance police officer's modification of outcome) 33% 53% -.19* Offender understood relationship between offense and outcome (80% mostly or fully) 1=mostly or fully 0=somewhat or not at all 37% 56% -.16 Offender appeared "powerless" in room full of adults (38% some/fair/to high degree) 1=some/fair/to high degree 0=not at all 47% 36% .11 Police officer treated the offender with respect (99% agree) 1=agree 0=disagree 41% 0% .09 Offender treated police officer with respect (96% agree) 1=agree 0=disagree 38% 100% -.26** Coordinator permitted everyone to have a say (98% agree) 1=agree 0=disagree 39% 100% -.18 Coordinator seemed impartial (93% agree) 1=agree 0=disagree 41% 33% .04 † "Ended on a high" refers to the SAJJ observer's determination that there was mutual good will and positive movement between the offender and victim by the end of the conference. ‡ Percentages are based on the 71 conferences for which a victim (or victim representative) was present. * p < .10 ** p < .05 3 Table 3. Logistic regression results for offender and conference characteristics and re- offending (1 = re-offending) Equation 1: Full model Predictor Beta Odds ratio Sex/race-ethnicity (1=Aboriginal male or female) 1.66 5.31 Sex/race-ethnicity (1=non-Aboriginal male) 1.52** 4.55 Pre-conference offending (1=yes) 1.09* 2.97 3 or more residential movements (1=yes) 2.17** 8.75 Offender showed remorse (1=yes) -1.12** .33 Genuine consensus in outcome decision (1=yes) -1.30** .27 Constant (B0) -1.59* N = 89 80% cases correctly classified Pseudo-R2 = 43% χ2 = 34.0 *** Equation 2: Reduced model Predictor Beta Odds ratio Sex/race-ethnicity (1=Aboriginal male or female) 1.49 4.45 Sex/race-ethnicity (1=non-Aboriginal male) 1.29* 3.63 Pre-conference offending (1=yes) 1.11** 3.04 3 or more residential movements (1=yes) 1.73** 5.67 Constant (B0) -2.65** N = 89 72% cases correctly classified Pseudo-R2 = 30% χ2 = 23.0 *** * p < .10 ** p < .05 4 Table 4. Conditional probabilities of re-offending for various case scenarios Predictor Logit B0 + B1 … + Bi Probability 1 / 1 + e -logit Aboriginal young people, male and female Scenario 1 – high-risk youth, poor conference elements 3.338 .94 Sex/race-ethnicity (1=Aboriginal male or female) Pre-conference offending (1=yes) Three or more residential movements (1=yes) Offender showed remorse (0=no) Genuine consensus in outcome decision (0=no) Scenario 2 – high-risk youth, good conference elements .922 .68 Sex/race-ethnicity (1=Aboriginal male or female) Pre-conference offending (1=yes) Three or more residential movements (1=yes) Offender showed remorse (1=yes) Genuine consensus in outcome decision (1=yes) Scenario 3 – low-risk youth, poor conference elements .079 .52 Sex/race-ethnicity (1=Aboriginal male or female) Pre-conference offending (0=yes) Three or more residential movements (0=yes) Offender showed remorse (0=no) Genuine consensus in outcome decision (0=no) Scenario 4 – low-risk youth, good conference elements -2.337 .12 Sex/race-ethnicity (1=Aboriginal male or female) Pre-conference offending (0=yes) Three or more residential movements (0=yes) Offender showed remorse (1=yes) Genuine consensus in outcome decision (1=yes) non-Aboriginal young people Scenario 5 – high-risk male, poor conference elements 3.184 .93 Sex/race-ethnicity (1=non-Aboriginal male) Pre-conference offending (1=yes) Three or more residential movements (1=yes) Offender showed remorse (0=no) Genuine consensus in outcome decision (0=no) Scenario 6 – high-risk female, poor conference elements 1.669 .80 Sex/race-ethnicity (0=non-Aboriginal female) Pre-conference offending (1=yes) Three or more residential movements (1=yes) Offender showed remorse (0=no) Genuine consensus in outcome decision (0=no) Scenario 7 – high-risk male, good conference elements .768 .65 Sex/race-ethnicity (1=non-Aboriginal male) Pre-conference offending (1=yes) Three or more residential movements (1=yes) Offender showed remorse (1=yes) Genuine consensus in outcome decision (1=yes) 5 Table 4 cont. Scenario 8 – high-risk female, good conference elements -.747 .35 Sex/race-ethnicity (0=non-Aboriginal female) Pre-conference offending (1=yes) Three or more residential movements (1=yes) Offender showed remorse (1=yes) Genuine consensus in outcome decision (1=yes) Scenario 9 – low-risk male, poor conference elements -.075 .48 Sex/race-ethnicity (1=non-Aboriginal male) Pre-conference offending (0=no) Three or more residential movements (0=no) Offender showed remorse (0=no) Genuine consensus in outcome decision (0=no) Scenario 10 – low-risk female, poor conference elements -1.59 .21 Sex/race-ethnicity (0=non-Aboriginal female) Pre-conference offending (0=no) Three or more residential movements (0=no) Offender showed remorse (0=no) Genuine consensus in outcome decision (0=no) Scenario 11 – low-risk male, good conference elements -2.491 .11 Sex/race-ethnicity (1=non-Aboriginal male) Pre-conference offending (0=no) Three or more residential movements (0=no) Offender showed remorse (1=yes) Genuine consensus in outcome decision (1=yes) Scenario 12 – low-risk female, good conference elements -4.006 .03 Sex/race-ethnicity (0=non-Aboriginal female) Pre-conference offending (0=no) Three or more residential movements (0=no) Offender showed remorse (1=yes) Genuine consensus in outcome decision (1=yes) 6 Table 5. Predictions of re-offending by SAJJ observers, police officers, coordinators, offenders, and victims SAJJ observers (N=89) Police officers (N=89) Coordi- nators (N=89) Offenders (N=93) Conference victims (N=61) Correctly predicted offending or non-offending 57% 65% 55% 60% 46% False positive (wrongly predicted offending) 17% 14% 10% 3% 38% False negative (wrongly predicted non-offending) 16% 11% 9% 25% 3% Do not know or unsure 10% 10% 26% 12% 13% Summary Correct prediction, all cases 57% 65% 55% 60% 46% Correct prediction, excluding do not know or unsure 64% 73% 74% 68% 53% Note: Predictions made by SAJJ observers, police officers, and coordinators were of the 89 primary offenders in conferences having multiple offenders. Predictions were made by all of the offenders interviewed in 1998 and the conference victims interviewed in 1998. 7 APPENDIX There was a generally good degree of consistency across the police, coordinators, and SAJJ observers for remorse (paired group comparisons ranged from 57% to 66% agreement, see Appendix, Table A) and an even higher level of consistency for judgements about how conferences ended, that is, "on a high" (paired group comparisons ranged from 73% to 75% agreement). The same is true for judgements about whether a conference was a "waste of time". Paired group comparisons for SAJJ observers, police officers, and coordinators showed that agreement rates ranged from 69% (between coordinator and police officer) to 90% (between SAJJ observer and coordinator). For characterising the outcome decision (which had three options, "genuine consensus", "acceptance", and "acceptance with reluctance"), the paired group comparisons ranged from 60% (between coordinator and police officer) to 71% (between the SAJJ observer and coordinator). In general we find that the police officers tended to view the conference and the offender's behaviour more optimistically and positively than the coordinators and researchers. When, in 1999, the results of the coordinator and police surveys were presented to these groups, they were initially surprised to see the higher ratings by the police for many items. If anything, the two groups would have expected the reverse. The explanation the police gave was that compared to what they saw in the field or in the station, the civility of the offender and other conference participants, coupled with the coolness of the deliberative process, was a decided improvement. In light of the high numbers in each group (5 researchers, 10 coordinators, and 21 police officers), the fact that the coordinators and police officers are criminal justice workers, not researchers, and several items asked them to judge their own behaviour, we are pleased with the levels of agreement across the three groups. For comparison, we summarize the inter-rater reliability of similar items in the RISE project. Harris and Burton (1996, 1998) assessed the agreement rates for observers (2 per conference or court case) in a sample of 45 RISE conference and court cases. Following each case, observers independently completed a "global observation instrument" consisting of 40 questions asking observers to rate various aspects of the case – e.g., "How much reintegrative shaming was expressed?", "How sorry/remorseful was the offender for their actions?" Item responses were given on an 8- point Likert scale ("none" to "very much"; "not at all sorry" to "very sorry"). Agreement rates across the 40 items ranged from 42% to 98% agreement. For "How much reintegrative shaming was expressed?", overall agreement (i.e., for conference and court cases) was 80%. 8 For "How sorry/remorseful was the offender for their actions", overall agreement was 73%. Agreement was much lower for items relating to the outcome decision. For "How much did the offender contribute to the conference/court outcome?", overall agreement was 53%. Similarly, for "How much was the offender coerced into accepting the conference/court case outcome?", overall agreement was 49%. Because our agreement rates are similar to those achieved in the RISE research for judgements about remorse and higher for judgements about outcome decisions, even across groups of observers from very different professional backgrounds, our view is that the SAJJ observations are reliable. An analysis of the phi coefficients of re-offending with the police and coordinator remorse and consensus variables showed that both were significantly related to re-offending for both groups (Appendix, Table B). We also ran the logistic regression analysis, using the police and coordinator remorse and genuine consensus variables (see Appendix, Table C). When police measures of remorse and consensus were entered, the model correctly classified 79% of cases and explained 52% of the variation in re-offending. When the coordinator measures were entered, the model correctly classified 78% of cases and explained 42% of the variation in post-intervention offending. In both the police model and coordinator model, remorse remained a significant predictor. The consensus variable neared statistical significance for the coordinators; thus, we are satisfied that the coordinator model replicates the SAJJ model. However, consensus was not a significant predictor of re-offending for the police in the logistic regression analysis. We suspect that one reason for this result is that the remorse variable was so highly predictive of re-offending for the police, it overwhelmed the effect of the consensus variable. 9 Appendix Table A. Paired group comparisons on key conference variables (remorse, consensus, waste of time, ended on high), percent with same judgments "To what extent was the YP remorseful for their actions?" (mostly/fully) coordinator police officer SAJJ observer coordinator - 66 57 police officer - - 64 SAJJ observer - - - "How would you characterise the outcome decision?" (genuine consensus) coordinator police officer SAJJ observer coordinator - 60 71 police officer - - 69 SAJJ observer - - - "The conference was largely a waste of time." (agree/strongly agree) coordinator police officer SAJJ observer coordinator - 69 90 police officer - - 70 SAJJ observer - - - "The conference ended on a 'high' – a positive note of repair and goodwill." (agree/strongly agree) coordinator police officer SAJJ observer coordinator - 75 73 police officer - - 75 SAJJ obsever - - - Appendix Table B. Bivariate relationships between selected variables and re-offending for coordinator and police Remorse Coordinator: YP was remorseful (54% mostly or fully) 1=mostly or fully 0=somewhat or not at all 27% 59% -.34** Police: YP was remorseful (56% mostly or fully) 1=mostly or fully 0=somewhat or not at all 20% 66% -.47** Genuine consensus Coordinator: Outcome decided by genuine consensus (51% yes) 1=yes 0=no (offender accepts/accepts with reluctance police officer's modification of outcome) 31% 50% -.19* Police: Outcome decided by genuine consensus (48% yes) 1=yes 0=no (offender accepts/accepts with reluctance police officer's modification of outcome) 30% 50% -.20* * p < .10 ** p < .05 10 Appendix Table C. Logistic regression results for offender and conference characteristics and re-offending (1 = re-offending) "Coordinator model" Predictor Beta Odds ratio Sex/race-ethnicity (1=Aboriginal male or female) 1.27 3.57 Sex/race-ethnicity (1=non-Aboriginal male) 1.22* 3.42 Pre-conference offending (1=yes) 1.40** 4.06 3 or more residential movements (1=yes) 1.73** 5.64 Offender showed remorse (1=yes) -1.35** .26 Genuine consensus in outcome decision (1=yes) -0.81 .45 Constant (B0) -1.67** N = 89 78% cases correctly classified Pseudo-R2 = 42% χ2 = 33.6 ** "Police model" Predictor Beta Odds ratio Sex/race-ethnicity (1=Aboriginal male or female) 1.49 4.44 Sex/race-ethnicity (1=non-Aboriginal male) 1.85** 6.36 Pre-conference offending (1=yes) 1.51** 4.54 3 or more residential movements (1=yes) 1.71** 5.54 Offender showed remorse (1=yes) -2.35** .10 Genuine consensus in outcome decision (1=yes) -0.48 .62 Constant (B0) -1.84** N = 89 79% cases correctly classified Pseudo-R2 = 52% χ2 = 43.7 ** * p < .10 ** p < .05 YOUTH JUSTICE CONFERENCING AND RE-OFFENDING* RESEARCH ON RESTORATIVE JUSTICE, CONFERENCING, AND RE-OFFENDING Table 1. Pre- and post-conference offending Summary "How would you characterise the outcome decision?" (genuine consensus) "The conference was largely a waste of time." (agree/strongly agree) "The conference ended on a 'high' – a positive note of repair and goodwill." (agree/strongly agree)