Serious (Violent or Chronic) Juvenile Offenders: A Systematic Review of Treatment Effectiveness in Secure Corrections

1.1. Relevance of reviewing this subcategory of offenders

There are many studies showing that those juveniles responsible for violent offenses are at high risk of becoming chronic offenders, committing many types of offenses and likely to receive an institutional sentence. For example, Thornberry, Huizinga and Loeber (1995) reported results from the Programme of Research on the Causes and Correlates of Delinquency, which consists of three well co-ordinated longitudinal research projects: The Denver Youth Survey, the Pittsburgh Youth Study and the Rochester Youth Development Study. In total these three projects involved 4,500 inner-city youths, ranging in age, at the beginning of data collection, from 7 to 15 years old.

Chronic violent offenders constituted only 15% of the total sample in Rochester and 14% of the adolescent sample in the Denver study; however, they committed 75% of all the violent offenses reported in the Rochester study and 82% of all the violent offenses reported in the Denver study. Data from the Rochester and Denver studies indicated the criminal versatility of these violent offenders (i.e. they commit a wide array of other offenses including property crimes, public disorder, status offenses and drug sales). In conclusion the authors stated “If we do not successfully reach this small group, we will leave the vast majority of the violence problem untouched” (p. 220).

Similar results were obtained in the Cambridge longitudinal study (Farrington, 2003), where 73% of males convicted as juveniles between the ages of 10 to 16 were reconvicted between ages 17 and 24, in comparison with only 16% of those not convicted as juveniles (also see studies of Krohn et al., 2001, and Stattin & Magnusson, 1991, as quoted by Farrington, 2003). Violent juveniles in the Cambridge study were also criminally versatile: 55 of the 65 males with a conviction for violence also received a conviction for a non-violent crime. To a large extent, the frequent offenders were versatile and sooner or later committed a violent offense. Effective interventions with juveniles should therefore affect later offending rates in adulthood.

Finally, those juveniles with multiple convictions are more likely to receive further periods of incarceration. A twenty-state research programme sponsored by the Office of Juvenile Justice and Delinquency Prevention, ‘Juveniles Taken into Custody’, reported programmes that shared age 18 as the upper age of juvenile jurisdiction, permitting readmission rates to be calculated over a reasonable time period. Of the 8,057 youths released in 1992, 27% were readmitted within one year of their release. Male readmission rates were much higher than for females (28% and 16%, respectively), and there was a strong relationship between the number of prior correctional commitments and readmission rates (Krisberg & Howell, 1998).

1.2. Current doubts about the effectiveness of interventions with violent juvenile offenders

The challenges involved in the treatment of the violent delinquents have been widely reported. As Thornberry, Huizinga and Loeber (1995) point out, by the time most serious delinquents are identified and receive intensive treatment from the juvenile justice system, they are well into their delinquent careers. For example, the National Youth Survey in United States (Elliot, 1994; Elliot, Huizinga and Morse, 1986, quoted by Thornberry, Huizinga and Loeber, 1995) found a substantial gap between the peak ages of involvement in serious violence and processing by the juvenile justice system. In addition, the offenders enrolled in treatment programmes have a host of negative characteristics that reduce the likelihood of successful intervention. “These offenders are older, are heavily involved in delinquent careers, and are likely to have progressed along overt, covert and authority conflict pathways. They are likely to be involved in other forms of delinquency, to use drugs, and to exhibit other related “behavior problems”. They are likely to have multiple risks factors and social deficits […]. Given these limitations, our expectations of treatment programmes should be modest” (Thornberry, Huizinga and Loeber, 1995, p. 233).

Lipsey and Wilson (1998) highlighted the paucity of systematic reviews of interventions with different types of offenders, especially those most serious offenders who might be presumed to be among the most resistant to treatment. This includes juvenile offenders.

An underlying problem is the dearth of primary intervention research conducted specifically with serious juvenile offenders: Most of the samples are mixed including less serious offenders and not separately identified and analysed. In an attempt to clarify the situation in serious juvenile offenders, Lipsey and Wilson (1998) conducted a meta-analysis (not in the context of a systematic review) focusing on two basic questions:

Does the evidence indicate that intervention programmes generally are capable of reducing the reoffending rates for serious delinquents? And if so, what types of programmes are the most effective?

Lipsey and Wilson included 200 experimental or quasi-experimental studies (published between 1950 and 1995) that involved serious juvenile offenders to some degree (more stringent inclusion criteria produced a very small number of studies). The juveniles finally selected were those “reported to be adjudicated delinquents”. The juvenile samples were largely male and with an average age of 14 to 17 years old. They categorised the studies as non-institutionalised (N=117) and institutionalised (N=83).

With non-institutionalised juveniles, treatment effects were larger for juvenile samples with mixed priors (i.e., including some proportion of person offenses) than those with mostly property priors. The most effective interventions were a group composed of interpersonal skills training, individual counselling and behavioural programmes, while the less effective interventions were wilderness/challenge programmes, early release from probation or parole, deterrence programmes (shock incarceration), and vocational programmes (distinct from employment related programmes).

The results with institutionalised juveniles contrasted markedly with those for non-institutionalised juveniles: With the offenders in institutions, the treatment effects are much the same for a given programme whatever the sample characteristics such as age, gender, ethnic mix and history of prior offenses. Again, the most successful intervention was interpersonal skills training, followed by the teaching family home programme (Achievement Place Project). The least effective interventions were wilderness/challenge programmes, drug abstinence, employment related programmes and milieu therapy.

The mean effect sizes were similar for both non-institutional (r = .07) and institutional interventions (r = .05), and the difference was not statistically significant. Specifically, the intervention for the most effective treatment with institutionalised juveniles showed mean effect sizes of .17-.19. In terms of the equivalent recidivism rate differentials, these techniques had an impact on recidivism that was equivalent to reducing a .5 control group baseline to around .15-.17, which is a substantial reduction considering the challenge presented by this category of offender ¹ .

Although Lipsey and Wilson categorised interventions as either institutional or non-institutional, they included in the institutionalised category many programmes that were, in fact, residential community-based interventions, such as Achievement Place.

According to Andrews et al. (1990), treatment for delinquent behaviour is most effective when the juveniles to whom that treatment is administered have appreciable risk of actually reoffending (the ‘risk principle’). The contrary view, however, is often expressed: That the most serious cases will be least amenable to treatment. The authors’ meta-analysis supported the risk principle: For both groups of offenders, the average intervention programme produced a positive effect equivalent to about a 12% reduction in subsequent reoffense rates.

In spite of these results, it remains to be demonstrated what specific strategies are really promising in rehabilitating incarcerated juvenile offenders, and, as a subgroup, the incarcerated serious juvenile offender. Presently, we have some preliminary results which suggest that the efforts directed at juveniles are more promising that the ones directed at adults. Redondo et al. (1997), reported in the first meta-analysis of only European evaluations, that in terms of crime typology, the most effective interventions (criterion: General improvement) were obtained with offenders against persons (r = .419), and the least with sexual offenders (r = .085), and that juvenile centres (r = .257) and juvenile prisons (r = .193), were more effective than adult prisons (r = .119). In a second systematic review, Redondo, Sánchez-Meca and Garrido (1999) analysed the specific influence of 32 European treatment programmes (applied during the eighties) on recidivism. Important findings included: (1) behavioural and cognitive-behavioural programmes were the most effective, (2) treatments were more successful with juvenile offenders. The reason for this probably reflected the use of most successful techniques (behavioural and cognitive-behavioural) with juveniles; and (3) the greatest effectiveness was achieved with violent offenders (not sex offenders), which seems to confirm the risk principle (Andrews et al., 1990).

In an update of the European meta-analysis, Redondo, Sánchez-Meca and Garrido (2002), found that the largest effect sizes were obtained with adolescents (r = .35), although all of the age categories achieved significant positive results.

Outcome measures in this area of “violent offending” also pose a challenge to an investigator and reviewer and must therefore be considered in this review. As Serin and Preston (2001) stress, the definition of “violent offender” and the issue of measures of recidivism have yet to be clarified. It is necessary to specify with more detail the characteristics of the offenders enrolled into programmes and the quality of the reoffending, separating the new violent offenses from the general recidivism rate.

The role played by different moderating variables (e.g., prior offense history, chronicity of violent offending, age at intervention, booster programmes and gender) requires further investigation, as described by Lipsey and Wilson and the European meta-analysis.

In summary, many gaps remain in our knowledge about the treatment of serious delinquents:

The Lipsey and Wilson (1998) meta-analysis compared institutionalised and non-institutionalised treatment for serious delinquents, but they included in the institutionalised category many programmes that were in fact residential community-based interventions, like Achievement Place. We still do not know the effectiveness of secure corrections treatment per se, in comparison to that of the traditional juvenile prison and training schools as well as modern small units for some kinds of offenders (with individualised treatment as a philosophy in the programme intervention).

The role played by different moderating variables (for example: Prior offense history versus non prior history; violent non chronic offenders versus violent chronic offenders; intervention at an early age versus at a later age; programmes that count with booster treatment after leaving versus programmes do not count with it; male delinquents versus female delinquents) has to be more investigated, as we have presented above in the case of Lipsey and Wilson and the European meta-analysis.

The measurement of “violent offender” and the issue of the recidivism have not been clear and consistent. It is necessary to specify with more detail who are the subjects treated and the quality of the reoffending, separating the new violent offenses from the general recidivism rate.

It is important, given the nature of this issue, to include in this systematic review all eligible studies that can be found, independent of language.

2.1. General Objectives

To collect and assess the quality, in a systematic way, of the outcomes of empirical research regarding the effectiveness of treatment programmes implemented in secure corrections in order to decrease the reoffense rate and quality (i.e., type of offence) of serious (chronic or violent) delinquents (12-21 years old).

2.2. Specific Objectives

To identify quantitative published and unpublished studies (in different languages) relating to the evaluation of correctional intervention programmes for institutionalised serious (chronic or violent) juvenile offenders.

To analyse the effects of correctional intervention in serious (violent or chronic) juvenile offenders.

In the category of serious delinquents we have included violent or chronic offenders. As explained earlier, chronic and violent juveniles have many characteristics in common. Additionally, the majority of violent juveniles have previous criminal convictions and it is reported that almost half of persistent delinquents had committed violent offenses.

This review pays particular attention to the variability caused by moderating variables. These include: Type of treatment (theoretical framework of the treatment, focus), subjects or participants in the programs (age, type of offense, gender), the setting in which the intervention occurs (the stage in which the programme is applied, the regime of the participants, and the country), methodology (type of assignment of the participants to the groups, groups attrition, and the follow-up period), and extrinsic variables (publication year and source) (Lipsey, 1994; Sánchez-Meca, 1997).

3.1. Criteria for Inclusion and Exclusion of Studies for this Review

3.2. Search Strategy for Identification of Relevant Studies.

Several strategies were used to identify studies meeting the criteria of this review. In order to reduce potential biases, we searched for: (1) published and unpublished studies, (2) between 1970 and 2003, (3) studies in areas of criminology, psychology, sociology, social service, education and psychiatry, (4) from any country and written in one of the following languages: English, Spanish, French, Portuguese, German and Italian.

First, we did a hand search of a selection of specialised relevant journal contents that are held in our Universities. We reviewed by hand search seven non-English journals and 21 English journals (Appendix 1).

Second, we conducted a specific search of 13 available electronic databases relevant to the topic area. Searches were done online using available Madrid Autónoma and Valencia Universities (Spain) resources. A report about keywords and results by each database consulted is included in Appendix 2. The keywords used to search in these databases were:

Delinquen (cy, ts), criminal (s), convicted, offender(s), inmates

Delincuencia, delincuentes, criminales, encarcelados.

Institution (alized, al, alization), detention, facility(ies), prison (s,ers), incarceration (ed), hospital (s), borstal (s), correctional(s), reformatories.

Institución, institucionalizados, detención, detenidos, prisión (es, eros), encarcelamiento, hospitales, correccionales, reformatories.

Boy(s), girls(s), adolescent(ce,s), juvenile (es), youth, young.

Jóvenes, juvenil, adolescentes.

Treatment(s), program(s), therapy (ies), rehabilitation, intervention(s).

Tratamiento, programa, terapia, rehabilitación, intervención.

Aggression (ive), anger, violence, violent, serious, chronic, persistent.

Agresión, ira, violencia, violento, serio, crónico, persistente.

We used other methods to search studies for this review. These methods included: (1) broad searches of the Campbell Collaboration Social, Psychological, Educational & Criminological Trials Register (C2-SPECTR) developed by the UK Cochrane Centre and supervised by the University of Pennsylvania Graduate School of Education (Petrosino et al. 2000); (2) we checked all the citations from the meta-analysis of Lipsey (1999) and Lipsey & Wilson (1998) about serious juvenile offenders (this meta-analysis quoted 74 references about interventions with institutionalised offenders) and also some European meta-analysis (Redondo, Garrido & Sánchez-Meca, 1997; Redondo, Sánchez-Meca, and Garrido, 1999), where 15 studies were about juveniles; (3) we maintained postal and e-mail correspondence with different institutions that work in topics linked with crime in several countries (information about these communications is included in Appendix 3); (4) we distributed letters requesting information from participants and lecturers attending the 2002 Annual Meeting of the American Society of Criminology at Chicago (copy of this letter is included in Appendix 4). Finally, we did a World Wide Web search with two popular search engines: Altavista and Google (information about the results of this search is included in Appendix 5).

3.3. Search Results and Selection of Studies

The total number of references obtained from the search methods (excluding the Internet searches, which generated thousands of websites) was 1299. Most of them included abstracts. Although the majority of these references contained the keys words of our search, in general the studies were not evaluation reports.

Many of the studies identified were theoretical assessments about the quality of a particular approach or the best way to consider several issues of the facilities regimes. Several other studies were descriptions of young offenders in the transition process to live in freedom after secure care. Some other studies focused on the validation or application of psychological tests used with juvenile offenders and violent youths while they were incarcerated. Other studies discussed topics such as community and diversion programs, the prediction of recidivism or the description of profiles of serious offenders. A small group of studies dealt with the training of people who work with juvenile delinquents.

The 1,299 studies identified included 74 references from the Lipsey meta-analysis (Lipsey, 1999; Lipsey and Wilson, 1998), and 14 references from the European meta-analysis (Redondo, Sánchez-Meca and Garrido, 2002, 1999a).

In regard to the 74 references of Lipsey's meta-analysis, 38 of them were excluded of our review. Nineteen were excluded because these studies were done previously to 1970; one reference was about sexual offenders; 3 studies were about bootcamps; and 15 articles were programmes that included time spent in the community and not inside an institution.

From the 1,299 initial identified references, we selected 122 as potentially eligible to include them in this review. These 122 researches were assessed using an eligibility checklist. This universe of references was constituted by 36 studies from Lipsey's meta-analysis, 14 references form the European meta-analysis and 72 references were obtained through a search in several databases.

As result of the selection process, 11 out of the 36 Lipsey's studies were included in our review and 25 of them were excluded.

From the 25 excluded studies, 10 corresponded to studies where juveniles participants were not serious offenders (in accord to our criteria); five studies lacked available information about the outcome of recidivism; three studies lacked sufficient methodological rigor; two studies did not have a description of the offender type (it was not possible to know if they were serious offenders) and these studies did not observe our criteria of institutionalisation; in two cases there was no measure of recidivism and their participants were not serious offenders; in one study there was no outcome of recidivism and the methodology was poor; another study lacked methodological rigor, juveniles were not serious offenders and participants resided in the community. Finally, one study did not correspond to the application of a specific program; actually it assessed the effect of different placements on the juveniles.

In the European meta-analysis, of the 14 studies with juvenile participants, none of them corresponded to this review's criteria.

From the 72 remaining references identified in the database searches, 10 were selected for this review, one was impossible to obtain ³ , and 61 were excluded. Of the 10 selected studies, two were about the same research (Fagan, 1990a; Fagan, 1990b), so they were considered as a single study (comparison between one experimental and one control groups). Two other works shared data: Lukin (1981) and Jessnes (1975), and both of them were included as a single study as well. Another reference (Gordon, 1997) coincided with the same program and some participants from the study of Greenwood and Turner (1993), which was included in the Lipsey's meta-analysis. In this case, data from the Gordon's study were considered for this review because they included the data from Greenwood and Turner. A summary of the studies’ selection process is included in Appendix 6.

Finally, we found 17 reports that fit the criteria of our review. These 17 references allowed us to analyse 30 different studies (comparison between experimental and treatment groups).

3.4. Data Management and Extraction

The reviewers selected 17 research reports for this review and found 30 comparisons between treatment and control groups in these works. We used a coding protocol to register the extracted data from each comparison. This coding protocol was constructed on the basis of the literature reviewed about correctional intervention programmes for serious institutionalised delinquents. We have also taken into account previous experiences in other systematic reviews such as the Boot Camp review (MacKenzie, Wilson & Kider, 2001) and previous meta-analyses (Marín-Martínez, Hidalgo, López, López, Moreno, Redondo, Rosa and Sánchez-Meca, 2002).

Building on the work of Lipsey (1994) and Sánchez-Meca (1997) the coding instrument was divided in three variables groups:

Substantive variables: This category includes the characteristics that are the object of this review, such as special features about the research participants, treatment and context. Methodological variables: This group of variables referred to the design quality of the selected studies.

Extrinsic variables: This group of variables included study characteristics not directly linked with the object of this review, such as language used, published/not published document, etc; however, these can affect results.

Two trained psychologists completed the coding protocol for each study; the discrepancies were solved by mutual consent.

3.5. Statistical procedures and conventions

The most usual recidivism measure reported in the studies about the effectiveness of programmes for delinquents consists of classifying the sample subjects as recidivists versus non recidivists. This was the case in our meta-analysis, as only two studies reported continuous measures of recidivism. As our unit analysis was the comparison between a treatment group and a control (or comparison) group, we selected as the effect-size index the odds ratio and its translation to natural logarithm. To calculate the odds ratio from each study, we composed a 2x2 contingency table, with the cell frequencies indicating the number of recidivists and non recidivists in each group (treatment and control). Odds ratios over the value 1 (and log odds ratios over zero) meant a lesser recidivism probability in the treatment group than in the control group and, at the inverse, odds ratios under the value 1 (and log odds ratios under zero) represented a higher recidivism probability in the treatment group than in the control group. If the study reported separate data for general and serious recidivism (such as we have defined in section 3.1.4), then an odds ratio was calculated for each recidivism measure and separate meta-analyses were carried out for each one of them. When a study reported recidivism data for several follow up periods, an odds ratio was calculated only for the last follow up, because including effect estimates for different follow up periods from the same sample violates the independence assumption in the meta-analytic calculations. Thus, for each study the odds ratio was calculated from the completers, that is, from the subjects with data in the last follow up. Although the strategy of calculating an odds ratio only for the last follow up avoids dependence in the effect estimates, it can bring attrition in the data used to calculate the effect sizes and, as a consequence, the effect estimates can be biased. In order to assess whether attrition could affect the effect estimates, intent-to-treat analyses (analyses based on assignment to conditions even if the juveniles did not complete or receive assigned treatment) were also carried out. With this purpose, another odds ratio for each study was calculated assuming the ‘poorest scenario’, that is, assuming that all of the subjects lost in the last follow up, both in the control and in the treatment groups, had recidivated. Therefore, separate meta-analyses were carried out for the two different recidivism measures (general and serious recidivism) and for completers and intent-to-treat data.

Only two studies did not report dichotomous recidivism measures (both of them reported in Friedman & Friedman, 1970), but means and standard deviations on a continuous recidivism measure (e.g., number of offenses). In those cases, we calculated a standardized mean difference, d, and then it was translated into the log odds ratio, Lor, by Lor = 1.65d, in order to put all of the effect estimates in the same metric (Haddock, Rindskopf, & Shadish, 1998; Sánchez-Meca, Marín-Martínez, & Chacón-Moscoso, 2003).

The meta-analytic calculations were carried out assuming a random-effects model. Under this statistical model each effect estimate is weighted by its inverse-variance, being the variance equal to the sum of the within-study and the between-studies variances. With this weighting scheme, an average effect size and its 95% confidence interval were calculated for the different recidivism measures. The heterogeneity among the effect estimates was assessed by the between-studies variance (□²), the Q test, and the I² index (Higgins & Thompson, 2002; Huedo-Medina, Sánchez-Meca, Marín-Martínez, & Botella, 2006). Although the random-effects model was the model in our analyses, we also calculated average effect estimates from a fixed-effects model in order to examine the robustness of our results to changes in the statistical model.

When the heterogeneity Q test was statistically significant, mixed-effects analyses were carried out to search for moderator variables that could explain the variability among the effect estimates. For qualitative moderator variables, weighted analyses of variance were applied on the effect estimates, whereas the relationship between continuous moderator variables and the effect estimates was assessed by weighted regression models. Some of these analyses were crucial to assess the validity of our meta-analytic results:

To assess if publication bias might be a threat against the validity of our results: (a) we compared the average effect sizes for the published versus unpublished studies, and (b) we applied the Egger test (Rothstein, Sutton, & Borenstein, 2005).

To examine if attrition affected the effect estimates in our meta-analyses: (a) regression analyses assuming a mixed-effects model were carried out to test whether the treatment group attrition, the control group attrition, and the differential attrition, and (b) separate meta-analyses were carried out to compare the average effect size obtained from the completers with those obtained from the intent-to-treat data.

As we had included studies with random and non random assignment of the subjects to the groups, we compared the average effect size for both design types, in order to assess whether the remaining analyses could be done with all the effect estimates regardless of the subjects assignment rule.

Although all of the meta-analytic calculations were carried out with the log odds ratio, this effect-size index is difficult to interpret in a practical sense. To make easier the practical interpretation of the results, the average log odds ratios were translated into both odds ratios and correlation coefficients and, then, the correlation coefficients were translated into the BESD (Binomial Effect Size Display) to obtain an index of effectiveness in terms of differential percentage of success between the treatment and the comparison groups (Rosenthal, 1991). To accomplish this, first each average log odds ratio, Lor, was translated into a d index by: d = Lor/1.65, and then the d index was translated into a correlation coefficient, r, by: r=d/$$ 4 (Haddock et al., 1998; Rosenthal, Rosnow, & Rubin, 2000; Sánchez-Meca et al., 2003). Thus, an average log odds ratio of, for example, Lor = 0.211, became an odds ratio of or = e^0.211 = 1.235 and a correlation coefficient of r = 0.064, meaning that, on average, the subjects that received a treatment exhibited about a 6.4% less recidivism that those of control groups. On the other hand, this translation also enabled us to compare our results with those obtained in previous meta-analyses that had presented their effect estimates in terms of correlation coefficients. All of the meta-analytic calculations were carried out with the program Comprehensive Meta-analysis 2.0, CMA 2.0 (Borenstein, Hedges, Higgins, & Rothstein, 2005).

3.6. Criteria for determination of independent findings

The selected studies reported multiple outcomes in the same sample of subjects and several outcome measures at different time points (follow ups). In order to avoid statistical dependency in the data in these cases, we computed an effect size (ES) for each outcome measure, that is, for general and serious recidivism, and separate meta-analyses were carried out for each one of them. On the other hand, when the study reported data for multiple follow up periods, an effect estimate was calculated for only the last follow up. Therefore, each study contributed with only an effect estimate in the meta-analytic calculations.

4.1. Description of selected studies

This review included an analysis of 17 documents (eight journal articles, two books, one published governmental report, two unpublished governmental reports, three unpublished dissertations, one unpublished research report and one unpublished demonstration project report). In these 17 documents, we identified 30 comparisons between a treatment group and a control group. We named these comparisons “studies”.

For these 30 studies, we have only included groups with “n” (number of youths in the sample of each group) equal to or above five ⁴ . When the studies had information about more than one control or comparison group, we chose one of them in order to avoid the dependency in the data ⁵ .

Table 1 shows the main characteristics of the studies included in this meta-analysis. In general, the studies included in this review were published in the United States, with samples of male violent offenders, with a mean age of 16 years. Most of the programmes were non-behavioural interventions, followed by cognitive-behavioural and cognitive treatments, and the minority being the behavioural and educational programmes. Most of the programmes–focused on the individuals, only one study focused on the family, two of them applied multi-focused services, three were directed to groups of offenders and five were directed to peers. The participants in the studies lived at juvenile prisons, as well as in special training schools and centres of juvenile reform. Of the 30 studies, 13 of them were experimental studies, whereas 17 were quasi-experimental studies.

A total number of 7,509 juveniles were included in all the 30 selected studies (3,685 juveniles were in the treatment groups and 3,824 in the comparison groups). However, as we have included only the most serious offenders from the samples, the initial total population included in this review is smaller (6,658 juveniles) ⁶ . The sample sizes ranged from 5 to 660 juveniles. The last follow up in the studies was between 6 and 120 months. On average, the last follow up period for the 30 studies had a median value of 18 months, whereas the mean was 31.6 months (SD = 36.1). The global attrition was 17% for general recidivism and 30% for serious recidivism. Descriptive characteristics of continuous variables of the studies are included in the Table 2. Additionally, a descriptive summary of the studies included in this review and some comments about each one of them are included in the Appendix 7, as well as a summary of the variables database for this review can be found in the Table 13.

4.2. Results of the Meta-Analysis

The 30 studies included in this review reported measures about general recidivism, but only 15 of them analysed serious recidivism. Next, separate meta-analyses for both recidivism measures and for effect sizes based on completers and on intent-to-treat data are presented.

4.2.1. Effect Sizes for General Recidivism

Overall Results at the Last Follow-up for Completers Data

For this review, we defined general recidivism in a broad sense, including any official record obtained from the police or adult/juvenile justice courts, which involve any kind of new offenses with any kind of court response (parole, prison, etc.). The general recidivism definition for each study is detailed in Appendix 7, and the results obtained from the completers data is showed in Table 3. Figure 1 shows a forest plot for the effect sizes distribution in the metric of the odds ratio, with the values over 1 showing a lowest rate of general recidivism in the treatment group in comparison to the control group. Assuming a random-effects model, the average odds ratio was or₊ = 1.235, being positive in favour of the treatment groups and statistically significant (p = .006). Its translation into a correlation coefficient was r = .064, meaning that the subjects that received any intervention programme exhibited, in average, a 6.4% less recidivism into crime than those of the control groups. The average effect size obtained from the fixed-effects model was very similar (r = .069) to that from the random-effects model and the heterogeneity Q test was statistically significant (p = .037; I² = 34.10%). These results imply that the effectiveness of the applied treatments was heterogeneous and, as a consequence, we analysed the influence of moderator variables on the effect estimates by means of mixed-effects models. Because of the small number of studies included in the meta-analysis, only a few conceptually relevant moderator variables were examined.

OPEN IN VIEWER

1

Forest plot of the odds ratios obtained for general recidivism at the last follow up (completers data).

Overall Results for Intent-to-Treat Data

With the purpose of carrying out a sensitivity analysis, the odds ratio of each single study was re-calculated taking into account the initial sample size of both the treatment and the control groups. As the outcome was dichotomous (recidivism versus non recidivism into offending), we assumed that all the subjects that had been missed before the last follow-up had recidivated (an “intent-to-treat analysis”). With this strategy, we were recreating the poorest scenario in terms of effectiveness. If under this scenario the results were similar to those obtained with the completers data, then we will have a strong argument to dismiss possible biases due to attrition. Table 4 shows the results of our meta-analysis for intent-to-treat data. With both the fixed- and the random-effects model a statistically significant average odds ratio was obtained and the average correlation coefficient for the random-effects model was even higher than that obtained from the completers data (or+ = 1.307; r = .081). This result showed that assuming the worst scenario (that is to say, that all missing individuals both in the experimental and the control group-had recidivated), the intervention was still effective. In this way we can conclude that our results are not overly influenced by attrition. As a consequence the following analyses were carried out only with the completers data.

4.2.4. Serious Recidivism and Effect Size

With relation to serious recidivism, the 15 studies that were taken to analyse presented a significant mean log odds-ratio that supported the effectiveness of the treatment (or+ = 1.354; confidence limits: 1.074 and 1.708; r = .091) (Figure 2). Moreover, the ESs were homogeneous around the mean odds ratio[Q(14) = 10.585, p = .718], and the between-studies variance was 0 (as well as the I² index); as a consequence, the assumed statistical model in this case was the fixed-effects model. Therefore, we can affirm that the interventions reduced the serious recidivism of the offenders.

OPEN IN VIEWER

2

Forest plot of the odds ratios obtained for serious recidivism at the last follow up (completers data).

5.1. General recidivism

In general, the mean ES for general recidivism was positive in favour of the treatment groups. Thus, the evidence suggests that implementing programmes is better than not doing it. However, due to the available information from the studies and considering the number of studies analysed in this review, it is very difficult to discriminate among the main characteristics associated with effective programmes.

The previous meta-analyses of Lipsey and Wilson (1998) and Lipsey (1999) reported ESs of r = .05 for institutional interventions with institutionalised serious juveniles offenders. The present review reports an ES of r = .06, a value very similar to that obtained by Lipsey. The present conclusion is that the effectiveness of interventions with serious offenders is smaller.

Regarding the participants variables, similar to other meta-analysis such as Lipsey's, most of the studies have focused on male juveniles. Although a few analysed female evaluations produced slightly negative findings, these results are statistically unstable. This situation could be explained because of the reduced number of studies with female samples. The gender variable of the participants did not seem to have any effect on the magnitude of the ES. Additionally, the proportion of explained variance by this variable was small.

Although one of our main concerns in this review was to assess the differential effect of the correctional programs on the different kind of offenders (violent, chronic and mix), this variable did not show significant differences on treatment efficacy. After comparing violent, chronic and mixed samples, the data obtained showed no apparent interaction with the effectiveness of the programmes. This situation could be explained because the primary studies provided limited information about the samples. Although all the studies included in this review had the inclusion criteria of serious offenders, in some of them the total sample were constituted by violent, chronic or mix offenders, but in others only around 50% of the sample was composed of this kind of offenders. This condition could influence the results of this variable.

When it came to the type of interventions, this variable was the most important in theoretical terms. Data showed that in general the interventions are effective in favour of the treatment groups. However, with the available data we cannot affirm that the type of intervention has an influence on the ES. In spite of this fact, it is important to point out that two types of treatment presented significant ES, and these interventions included some cognitive components (cognitive – behavioural and cognitive programmes). These results are in agreement with the results of other meta-analyses.

Another important result is the focus of the intervention programme. The multi-focused interventions reported a significant ES. Although there were only two studies classified into the category of multi-focused treatment, it is important to consider this factor in future research, with its respective complete description in the studies. The result about the focus of the interventions can be an initial evidence of a promising component of treatment success.

In general, no single approach in the treatment of violent adolescents has been proven effective (for example, Lipsey and Wilson, 1998). In words of Tate, Reppucci and Mulvey (1995): “Service provision should be reconceptualized as an ongoing care model that emphasizes intervention in multiple spheres of an adolescent's life. The most promise lies in a comprehensive, long-term commitment, not in the development of any singular, more powerful approach” (p. 780). The results of this review suggest more attention in this sense.

It is important to point out that many studies did not have available information about the intensity and the magnitude of the intervention. The latter is important if we take into account the early age of onset in delinquent activities of the serious offenders and their long learning histories of illegal behaviour. For this reason it is important to propose the discussion about the efficacy of short vs. long intervention programmes. Because only a few of our reviewed studies reported information about the intensity and the magnitude of the programmes, the conclusions obtained in both issues are limited.

Other features in the literature associated with successful programmes such as the principles of risk, needs or responsivity (Andrews, 1995; Lösel, 1995; McGuire & Priestley, 1995) were not studied in our review. Again, in many studies there was not available information about these principles.

A very important methodological point is that it is difficult to find control groups that have no correctional program (the “treatment-as-usual” effect). This situation could be considered as one of the reasons that explains why we did not find higher ESs. For instance, in researches as the one of Cornish and Clarke (1975), the control group as well as the treatment group received a therapeutic community intervention with only some differences in the structure. In another example, Friedman and Friedman (1979), the control groups had an educational intervention. And in other studies as in Guerra and Slaby (1990) the youths in the control group did not have any cognitive intervention, but it was not possible to know if this group had received any other kind of intervention in the past. Thus, if the control group received some kind of intervention, this circumstance could influence the ES results.

From 17 articles or documents included in this review, 6 corresponded to experimental studies and 11 to quasi-experimental studies. From the 30 studies (comparisons) 13 were experimental and 17 quasi-experimental. As it is established in the results of this research, the experimental studies presented a nonsignificant small ES, whereas the quasi-experimental studies obtained a significant small ES. In spite of the absence of a statistically significant difference between the mean ESs of the experimental and quasi-experimental studies, our results should be interpreted very cautiously, because the experimental studies showed a nonstatistically significant mean ES. Also these data suggest the importance of more experimental research, with the aim of obtaining stronger conclusions about the effectiveness of the interventions with institutionalised young serious delinquents (violent and chronic).

The statement of Tate, Reppucci and Mulvey (1995) is still valid: “There is a clear need for methodologically sophisticated studies of treatment effectiveness that are more precise with regard to their definition of violence and that either exclusively target or conduct separate analyses for violent juveniles” (p. 780).

With respect to the presence of some possible bias in the calculation of ES, neither the attrition nor the publication bias seem to affect the results.

5.2. Serious recidivism

In spite of the fact that not all the included studies in this review had data about serious recidivism, the analysis of the 15 studies with this kind of results presented interesting data. The global ES for the serious recidivism was favourable to treatment groups with a statistically significant mean ES. These results indicate that rehabilitation programmes for serious offenders reduce serious recidivism to a higher degree than general recidivism. This is an important finding because the treatment for serious offenders has among its main objectives to reduce the recidivism and the dangerousness of the serious offenders. The data points out an important effect on both of these aims.