Skip to start of content

Reintegrative Shaming Experiments (RISE)

Recidivism patterns in the Canberra Reintegrative Shaming Experiments (RISE)

Lawrence W Sherman, Heather Strang and Daniel J Woods
Research School of Social Sciences, Australian National University

November 2000

Australian National University

Results

Youth violence3

This experiment found that diversionary conferences reduced offending rates by about 38 crimes per 100 offenders per year, relative to the effect of being sent to court.

The youth violence experiment consists of 100 cases involving 121 offenders. Because the last case came into RISE only in June 2000 and several offenders have not yet been disposed of or interviewed, it has been necessary to conduct most of this analysis on the basis of the first 89 cases, involving 110 offenders, rather than the full experimental sample. We have at least one year of followup and baseline data on all 110 offenders in these 89 cases.

Figure 1 shows that the officially detected rate of offending per offender in the year before the RISE date was .081 per month in the conference group, and .071 in the court group. On an annual basis, this amounts to just under one offence per offender per year, on average. Yet in the year after the RISE date, the officially detected offending rate for the conference cases fell to less than half of one offence per offender-or one offence every two years. The rate of offending by offenders assigned to court barely dropped at all. In terms of significance, the drop in offending rates by offenders sent to court was almost 62 percent likely to have been a chance effect, while the drop in offending rates for conference offenders was only 1 percent likely to have been due to chance. The difference of before-after difference between these two measures was only 16 percent likely to have been due to chance, or 84 percent likely to be a result that would be found repeatedly with a similar sample of cases. The measure Cohen's D shows that the effect size of this difference of differences is .26, which is a relatively substantial effect.

Figure 2 expresses the effects of diversionary conferences on offending rates in percentage terms. While the conference group's rate of offending fell by 49 percent, the court group's fell by only 11 percent. This difference produces a net reduction of 38 percent in the conference group relative to the change in the court group.

Drink driving experiment4

This experiment found that diversionary conferences resulted in a very small increase in all offences (6 crimes detected per 100 offenders per year for all offences; 4 crimes per 100 offenders per year for drink driving offences).

The experiment consisted of 900 cases, each with only one offender. Half were randomly assigned to court processing and half to diversionary conferencing. All cases had been assigned by December 1997 and had been treated by the end of 1998. The data presented in the Figures are based on the entire 900 case sample.

Figure 3 analyses all types of offences by the offenders apprehended for drinking and driving. It shows that there was an increase in offending by both groups in the year after RISE relative to the year before RISE. Both increases were relatively unlikely to be due to chance. Yet the difference between the differences was modest, if only 13 percent likely to be a chance result. The effect size of that difference was a mere .1, which is modest indeed. On the other hand, in relative terms, Figure 4 shows that the net increase in offending rates across all offences was 53 percent more for the conference cases than it was for the court cases. Yet this analysis is based on very low rates of offending-under one-tenth of one offence per year, or one offence every ten years. In comparison to the violent crime experiment, these offenders' mean rate of offending is one tenth the size.

Figure 5 shows that the effect of conferences on detected rates of drinking and driving was about the same as it was on all offences. The difference is that the detected rates of drinking and driving remained unchanged before and after assignment to court, while they doubled after assignment to conferences. But the base rates are so very low that they only amount to a net gain of about 4 offences per 100 offenders per year. While the difference of differences is significant, and while we can be confident that the difference is due to the randomly assigned treatment, the effect size is quite small. The major caveat here is that the low detected rates may reflect much higher rates of undetected drinking and driving. It is by that logic that we designed such a large sample, in order to detect any differences in low rates of detected offending. To the extent that the increase Figure 6 reports in detected drink driving indicates that the undetected rate has risen by the same magnitude, this could mean substantially more offending after conference compared with court.

It is important to note that these differences in reoffending behaviour relate only to the first year post-RISE and decay and vanish over the longer timeframe for which we have data in this experiment. There is also no difference between the court and conference groups in the percentage who had any second-time drink driving offence; that is, the small difference in offence rates is due to higher rates of third and fourth offences in the conference group. One interpretation which would explain such a pattern in the results is as follows. A reason why court cases should have lower reoffending than conference cases is that drivers' licences are suspended in court cases, but conferences do not have the power to do this. For first offences, these court-treated offenders typically had their licences suspended for three to six months. But in spite of the fact that they were put off the road for this length of the follow-up period, they were just as likely to commit a second offence as conferenced offenders. However, once a drink driver has two court convictions within a short space of years, magistrates normally suspend licences for twelve months. This means the incapacitative consequence of a second conviction (twelve months) was normally two-four times that of the first (three-six months). That is enough for the beneficial effects of a conference to be overwhelmed by the effect of the twelve month licence suspension in reducing reoffending. The fact that the difference does not hold up beyond the twelve month follow-up is also consistent with the interpretation that it is licence suspensions which help court to prevent reoffending. If this interpretation is right, drink driving conferences have no chance of preventing drink driving more than court unless conferences are given the same power to recommend licence suspensions. One of our future projects will be to discover whether this interpretation is correct through detailed decomposition of the results.

Juvenile property (shoplifting apprehended by store security officers)

This experiment found no differences in offending rates between court and conference groups on the basis of one-year before-after changes (though after-only analyses show significantly lower reoffending for the conference-assigned group than court-assigned group.)

This experiment consists of 117 cases involving 143 offenders. Because the last case came into RISE only in June 2000 and again several offenders have not yet been disposed of or interviewed, it has been necessary to conduct most of this analysis on the basis of the first 108 cases, involving 135 offenders.

Figure 7 shows changes in the rates of offending in one-year periods pre-RISE and post-RISE. While there may appear to be a difference between the groups, the difference is very small and the odds of chance effects are very large. Neither the within-group nor between group differences are likely to be anything other than chance effects. Thus even while Figure 8 appears to show a reduction in offending rates for offenders assigned to court, the magnitude of change in the rates from the year prior to the year after is so small that the effect is most likely due to chance.

Juvenile property-personal victims5

This experiment also shows no differences in offending rates according to whether offenders were assigned to court or conference.

This experiment consists of 173 cases involving 249 offenders. Again, because the most recent case came into the experiment in June 2000, a number of offenders have not been treated or interviewed. This analysis was therefore conducted on the basis of the first 162 cases, involving 238 offenders.

Figure 9 shows that none of the before-after differences are statistically discernible. All of them are at least 26 percent or more likely to be due to chance. Despite Figure 10 suggesting a greater decline in offending rates for court cases than for conference cases, this difference reflects such a small change in both groups that it is highly unlikely to be found in repeated tests with the same kinds of samples.

Why did results differ across experiments? baseline differences and similarities at random assignment

The primary hypothesis about why results differed across experiments is that restorative justice affects offenders charged with different kinds of offences differently. Put another way, the dynamics of each type of offence may create a different emotional climate and basis for legitimacy of legal intervention using court or conference processes. Yet as a competing explanation for these stark differences across experiments, the following section explores the thesis that the differences were due to the pre-existing differences in the characteristics of offenders in each experiment.

Differences between experiments

This section describes the findings in Figures 11-13, which indicate differences across experiments that are unlikely to be due to chance.

Figure 11 shows that the violence experiment differed from the others in that a slightly higher proportion of offenders in the conference group had smoked marijuana before RISE than we found in the court group. This reversed the ratio found in the other experiments. It seems hard to construct a theory that drug use would interact with restorative justice, but the difference is nonetheless unlikely to be due to chance.

Figure 12 shows that the violence experiment differed from the others in the ratio of males to females between court and conference groups. The conference group in the violence experiment was much more male than the court group-by a relative difference of almost 50 percent. This difference may have been a bias against showing the effectiveness of diversionary conferences, since males generally have higher rates of offending than females-although it is not clear that females have any lower rates of repeat offending once they have been apprehended for a violent offence. Whatever its effect on the analysis, the difference in reoffending between the violence experiment and the three others that failed to find beneficial effects of conferences is highly unlikely to have been due to chance.

Figure 13 shows that there is a difference in pre-RISE binge drinking by across the four experiments. In this case, however, there is no clear correlation between the violence experiment and the other three experiments. There is no way to explain the different effects of conferences with the distribution shown here.

Similarities across experiments

The remaining Figures display data for all four experiments that fail to shown any differences that are not due to chance. While in 22 significance tests, we would expect one difference to be significant at the 5 percent likelihood level, there were three such differences reported in the last section. The other 19 tests are reported here, none of which found any plausible explanations for the different effects of restorative justice for violence as compared to the other types of offences.

Figures 14, 14a, 15, and 15a show the distribution of offence types for the two experiments taking multiple types of offences; drink driving and shoplifting (security apprehensions) were only able to take one offence type each. Where the offence types were mixed, there was no difference in distributions by type between court and conference.

Figure 16 suggests a reactive measurement, in which people who had been through conference seem less likely to admit prior property crime than those who had gone to court. This consistent pattern (with a question not asked of drinking drivers) cannot explain why violence offenders reacted so well to restorative justice.

Figure 17, in contrast, appears to fit the pattern of different results across experiments: there are higher rates of prior violent crime in the conference cases for the violence experiment than in the other two. But the small size of these differences makes them likely to be due to chance. The higher base rates of prior violence in the property crime experiments than in the violence experiment is interesting, but might also be a reactive measure in relation to the type of offence. That is, persons charged with violence might be less likely to admit to prior violence than persons charged with property crime.

Figures 18-19 show no statistically discernible differences across experiments in percentage of offenders unemployed, or who were Aboriginal or Torres Strait Islander.

Differences within experiments

The remaining figures examine selected baseline characteristics within experiments, finding no differences in the following by experiment by Figure number: