Introduction

Many youth involved in the juvenile justice system suffer from alcohol or drug use problems (McClelland et al. 2004; Robertson et al. 2004; Teplin et al. 2005). Given that substance use problems are highly correlated with criminal recidivism (e.g., van der Put et al. 2014), rehabilitative efforts aimed at reducing recidivism among youth may also need to address these concurrent drug use issues. Juvenile drug courts are specialized dockets designed for juvenile offenders with alcohol or other drug problems. Drug court programs aim to reduce criminal recidivism among drug-involved offenders by addressing substance use and abuse, and typically involve risk assessments, periodic interaction with judges, monitoring and supervision, incentives and sanctions, and referral to counseling and treatment services (Belenko and Dembo 2003). Although the therapeutic model of the juvenile drug court is similar to traditional juvenile courts’ rehabilitative mission, juvenile courts have historically shifted between these therapeutic models and more crime-control, punishment-oriented models (Feld 1988; Greenwood and Turner 2011; Lipsey et al. 2010).

The first drug court program opened in 1989 in Miami-Dade County, Florida, and by 2014, there were an estimated 2966 drug courts in the United States, 433 of which were juvenile drug courts (National Drug Court Resource Center [NDCRC] 2015). Despite this proliferation, prior reviews of research suggest that, although adult drug courts are effective in reducing recidivism (Mitchell et al. 2012a, b), juvenile drug courts may have only modest, if any, effects on recidivism (Latessa and Reitler 2015; Latimer et al. 2006; Mitchell et al. 2012a; Shaffer 2006; Stein et al. 2015). However, these reviews have not fully explored whether and how various characteristics of juvenile drug court programs may be associated with program effects. The current systematic review and meta-analysis, therefore, aimed to provide a comprehensive synthesis of the juvenile drug court effectiveness research, with particular emphasis on examining variability in effects across programs.

Juvenile drug court model

Compared to the punitive, adjudication-focused approaches common in traditional criminal courts, drug courts take a rehabilitative problem-solving approach to dealing with crime and substance use (Butts and Roman 2004a, b; Inciardi et al. 1996). Juvenile drug courts use a therapeutic jurisprudence model aimed at reducing recidivism and rehabilitating juvenile offenders with substance use problems. This integrated treatment and justice model recognizes that juvenile offenders with substance use problems face unique challenges and treatment needs (Belenko and Dembo 2003). Services include frequent judicial hearings in court where judges review juveniles’ progress, working with program staff and/or families to develop individualized treatment and rehabilitation plans, and providing incentives and sanctions based on juvenile behavior. The incentives and sanctions in a juvenile drug court are based on results from frequent drug tests, rewarding abstinence from drugs and punishing youth who use substances. In addition to this periodic judicial monitoring, youth are referred to substance use treatment services in the community. Juvenile drug court programs often last for 12 to 18 months, but can vary considerably in length given that program graduation often requires sustained abstinence from drugs and compliance with program requirements.

In 2003, the National Drug Court Institute (NDCI) and National Council of Juvenile and Family Court Judges (NCJFCJ) convened a workgroup of experts that outlined 16 strategies and recommendations for juvenile drug court implementation (U.S. Bureau of Justice Assistance 2003). These 16 strategies (see Table 1) were not intended to be research-based benchmarks, but, nonetheless, provide a useful framework for understanding some of the key issues in the implementation and operation of juvenile drug courts.

Table 1 The “16 Strategies” for juvenile drug courts (U.S. Bureau of Justice Assistance 2003)
Full size table

Prior reviews of juvenile drug court research

Several prior research reviews have arrived at different conclusions about the effectiveness of juvenile drug courts. For instance, whereas some narrative literature reviews have concluded that there is limited evidence of effectiveness (Belenko 2001; Roman and DeStafano 2004), others report that the research does demonstrate that drug courts might be effective (Henggeler 2007; van Wormer and Lutze 2011). Similarly, the conclusions from meta-analyses have been mixed. The largest meta-analysis to date, Mitchell et al. (2012a), synthesized findings from 34 juvenile drug court evaluations current through 2010. They found that juvenile drug courts were associated with significantly lower general recidivism, but found no evidence of an effect on drug recidivism or drug use. However, in a more recent meta-analysis of 31 juvenile drug court evaluations current through 2012, Stein et al. (2015) reported a statistically significant but small reduction in post-program recidivism. Older meta-analyses based on smaller numbers of studies have also reported either small beneficial effects (Shaffer 2006) or no significant effects on recidivism (Latimer et al. 2006; Utah Criminal Justice Center 2012).

These prior meta-analyses are not up-to-date with the most current research, however, and, thus, do not include evidence from recent evaluations (e.g., Latessa et al. 2013). Furthermore, these prior meta-analyses did not thoroughly investigate variability in the effects of juvenile drug courts, nor focus on how courts’ adoption of the 16 strategies (Table 1) might be associated with program effects.

Objectives

This meta-analysis sought to quantitatively synthesize findings from the evidence base of juvenile drug court research, including more recent studies, with particular emphasis on examining variability in effectiveness. Specifically, this meta-analysis examined: (1) the effects of juvenile drug courts on general recidivism, (2) the effects of juvenile drug courts on drug-related recidivism, (3) the effects of juvenile drug courts on drug use, and (4) variability in these effects across participant and drug court characteristics.

Methods

Inclusion and exclusion criteria

The population of eligible studies for this meta-analysis was experimental and controlled quasi-experimental evaluations of juvenile drug courts. To be eligible for inclusion, studies had to (1) evaluate a drug court program, defined as a specialized court designed to handle drug-involved cases that involves referring youth to treatment services, conducting regular drug screens, and involvement of a judge who actively monitors progress and sanctions prohibited behaviors; (2) include a comparison condition that was treated in the traditional fashion by the court system (e.g., probation with or without referral to treatment services); (3) measure criminal behavior (such as arrest or conviction) at least once after the start of the program; (4) report findings on a study sample of youth aged 18 years or under; (5) be published during or after 1989; (6) be conducted in the United States or Canada; and (7) use an appropriate research design.

Appropriate research designs included those where youth were randomly assigned to conditions, quasi-experiments that matched participants on at least one baseline measure of criminal offending or substance use, quasi-experiments that used statistical controls to adjust for baseline differences in participants’ offending or substance use, and quasi-experiments that provided enough information to permit calculation of effect sizes indexing baseline differences in participants’ offending or substance use. We excluded studies that compared one drug court program to another of similar intensity (i.e., treatment–treatment comparisons or dose–response evaluations). There were no other restrictions on eligibility.

Search strategy

A comprehensive search strategy was used to identify studies that met the aforementioned inclusion criteria. We included all studies reviewed in the most recent meta-analysis on juvenile drug court effectiveness (Mitchell et al. 2012a), which included literature between 1989 and August 2011. The Mitchell et al. (2012a) meta-analysis used the same eligibility criteria described above and used a comprehensive systematic literature search to identify studies; therefore, we used the reference list from this meta-analysis to identify literature between 1989 and August 2011. We then conducted our own literature search, designed to extend and update the body of research compiled by Mitchell et al. (2012a). This literature search was used to identify studies reported between August 2011 and December 2014. The following electronic databases were searched using the ProQuest platform: ERIC, International Bibliography of the Social Sciences, ProQuest Criminal Justice, ProQuest Education, ProQuest Family Health, ProQuest Health & Medical Complete, ProQuest Health Management, ProQuest Nursing & Allied Health, ProQuest Psychology, ProQuest Science, ProQuest Social Science, ProQuest Sociology, ProQuest Dissertations & Theses (US, UK, & Ireland), PsycARTICLES, PsycINFO, and Sociological Abstracts. We also conducted extensive supplementary searches of the following research registers and websites: Campbell Collaboration Library, Cochrane Collaboration Library, CrimeSolutions.gov, International Clinical Trials Registry, National Criminal Justice Reference Service, Center for Court Innovation, Chestnut Health Systems, Drug Court Clearinghouse, National Drug Court Institute, National Council of Juvenile and Family Court Judges, NPC Research, RAND Drug Policy Research Center, Reclaiming Futures, and the Urban Institute. We checked the bibliographies of all screened and eligible studies, as well as the bibliographies of prior narrative reviews and meta-analyses. We also conducted hand-searches of 2010–2014 conference proceedings from the American Society of Criminology, as well as manuscripts published in Drug Court Review and Juvenile and Family Court Journal.

Screening and coding procedures

Under the supervision of the first author, a team of master’s level research assistants conducted all eligibility screening and coding. First, all abstracts and titles were screened independently by two researchers; we retrieved the full text for any report deemed potentially eligible by at least one researcher. Next, all retrieved full-text reports were screened for eligibility independently by two researchers; the first author resolved any disagreements about eligibility. Finally, studies deemed eligible for inclusion were independently coded by two researchers and the first author resolved any coding disagreements. All data extraction followed a standardized coding protocol, with data entered directly into a FileMaker Pro database. The coding protocol was an abbreviated version of the one used in the Mitchell et al. (2012a) meta-analysis and provided detailed instructions for extracting data related to general study characteristics, participant groups, the drug court conditions, outcome measures, and statistical data needed for effect size calculations (coding protocol available upon request). Data collection and extraction was completed for all eligible studies, including those identified from the Mitchell et al. (2012a) meta-analysis, as well as the updated literature search.

Statistical procedures

Effect size metric

Most of the eligible studies reported binary measures for recidivism and substance use, so we used an odds ratio (OR) effect size to index the reported effects. Odds ratios were coded such that values greater than 1 indicated beneficial drug court effects relative to the comparison condition (e.g., lower recidivism, lower substance use). All analyses were conducted using the log odds ratio, with results translated back into the odds ratio metric for ease of interpretability. For the handful of studies that measured outcomes on a continuous scale (e.g., mean number of new arrests), we first computed the small-sample corrected standardized mean difference effect size (Hedges’ g, 1981), then used the Cox transformation to convert those to odds ratio effect sizes (Sánchez-Meca et al. 2003). We examined the distribution of effect sizes and sample sizes for outliers, but no outliers were identified.

Moderator variables

We coded a wide range of moderator variables from the study reports that described general study, method, drug court setting, and participant characteristics. General study and method characteristics included publication type (journal article vs. other), publication year, country, study design (randomized experiment vs. quasi-experiment), possible implementation problems (yes, no/unclear), and overlap between the period within which outcomes were assessed and the drug court treatment period (complete, partial, or no overlap). Defining this overlap period was complicated given that many youth may have received supplementary or ongoing services after the formal completion of the drug court; for our purposes, we simply defined overlap with the formal drug court treatment period (and did not consider overlap due to any ongoing or supplementary services).

Characteristics of the drug courts included year first opened, number of youth served per year, number of youth served in the most recent year, number of drug court phases, number of drug tests per week in the first phase, number of status hearings per month in the first phase, length of drug court (in months), method of disposition (pre-plea, post-plea, both), whether charges were dismissed upon graduation, whether violent offenders were excluded from participation, whether drug offenses were required for eligibility, explicit mention of dedicated drug court staff, provision of a written document of contingencies, explicit mention of a standardized risk assessment tool, referral of youth to brand-name substance use treatment providers, number of treatment providers used for referrals (single, multiple), number of substance use treatment modalities referred to (single, multiple), and whether psychiatric comorbidities were addressed in the treatment. We also coded whether the drug court adhered to the 16 strategies, assessing whether each strategy was explicitly mentioned in the program description, implied by the description of the program, explicitly not used based on the program description, or not mentioned/implied.

Finally, characteristics of the youth included the sex composition of the sample (percent male), racial/ethnic composition of the sample (percent Black, Hispanic, White), average age of participants, average number of prior arrests, and average number of prior drug arrests.

Missing data

When primary studies failed to include sufficient statistical information to estimate effect sizes, we contacted the study authors for that information. We did not impute missing effect sizes on any outcome variables but, rather, omitted them from any analysis involving those outcomes. Some studies also failed to provide information on key characteristics of the study methods, drug court programs, or outcome characteristics. We did not impute missing data on any of these other study characteristics; rather, we only present descriptive information for those studies with available data.

Analytic strategies

All analyses were weighted using random-effects inverse variance weights to ensure that each effect size’s contribution was proportionate to its statistical precision (Hedges and Olkin 1985; Lipsey and Wilson 2001). Only one effect size per participant sample was included in any given meta-analysis to ensure the statistical independence of the effect size estimates in each analysis. Several studies included two or more measures of recidivism, or measured outcomes at multiple follow-up points. To ensure the statistical independence of effect sizes within any given analysis, we conducted separate analyses by outcome type (general recidivism, drug recidivism, drug use) and follow-up period (during-program, post-program). For studies that reported multiple post-program effects for a given outcome, we first selected effects measured at the most frequently reported follow-up point (12–18 months for general recidivism; 6–12 months for drug recidivism), and when those were not available, we selected the first available follow-up point for that study.Footnote 1 If more than one effect size was reported within each of these categories, we used a set of decision rules to select the effect size to be used in the analysis. Namely, preference was given to effect sizes that were (1) general (i.e., covered all types of offenses as opposed to a specific offense type), (2) based on arrests, (3) dichotomous, (4) measured at the latest time point during a follow-up period, and (5) adjusted for other confounding characteristics (e.g., past arrest history, demographics).

Random-effects meta-analyses using the restricted maximum likelihood estimator for the random-effects variance component were used to estimate the mean effects for each outcome type at each follow-up period. Mixed-effects meta-regression models were then used to investigate variability in effects in relation to the moderator variables. We also used contour-enhanced funnel plots (Peters et al. 2008) to explore the possibility of bias resulting from the omission of small sample size studies with null or negative findings due to selective publication, reporting, or other forms of dissemination biases. None of the funnel plots (available upon request) indicated asymmetry, thus providing no indication of potential small-study bias.

Results

Literature search

We identified 7400 candidate reports in the updated literature search (6763 through database searching; 637 through other sources); 520 were duplicates that were dropped from consideration and 5704 were screened as ineligible at the abstract level (see Fig. 1). Of the 1176 articles retrieved in full text, 1144 were deemed ineligible. The final meta-analysis includes findings from 32 studies; these 32 studies reported findings for 46 independent samples comprised of 8738 juveniles (ESM 1 includes references to all the studies included in the meta-analysis).

Fig. 1
figure 1

Study identification flow diagram

Full size image

Description of included studies

Table 2 provides a brief summary of the 46 study samples included in the meta-analysis, and the left panel of Table 3 presents descriptive statistics for the key features of the studies, outcomes, and participants in those 46 samples. Most of the studies (89 %) were published in journal articles and all (100 %) were conducted in the United States. The methodological quality of the studies was generally poor—only three studies (7 %) randomly assigned participants to conditions, the average overall attrition rate was 0.18 (standard deviation (SD) = 0.24) and the average differential attrition between drug court and comparison groups was 0.06 (SD = 0.09). Although the drug court and comparison groups in the studies were matched well in terms of age, on average, groups were non-equivalent in terms of risk level, racial composition, and sex composition. All baseline difference effect sizes were coded such that positive values (g > 0, OR > 1) indicated the participants in the juvenile drug courts were at lower risk of recidivism. Thus, as shown in Table 3, compared to participants in the comparison conditions, the juvenile drug court participants tended to be at significantly lower risk, more likely to be White, and more likely to be female. This suggests that many of these studies may suffer from a selection bias that favors the juvenile drug court.

Table 2 Characteristics of the included studies
Full size table
Table 3 Key features of the study methods and participants, and bivariate associations with effect sizes
Full size table

Most of the effect sizes reported in the studies indexed differences on measures of general recidivism (72 %), and the average maximum length of follow-up was 18.5 months (SD = 12.8). The effect sizes reported in studies often involved outcomes measured over intervals that were completely overlapping with the drug court intervention period (17 %) or partially overlapping (43 %); only 39 % of the effect sizes were reported entirely in a post-program period.

The demographic composition of the study samples was predominantly male (M = 79 %) and White (M = 67 %), with an average age of 15.9 years (SD = 0.59). Few studies reported prior arrest history for participants; among those studies, youth in the drug courts had an average of 4.95 prior arrests (SD = 3.67; k = 17) and 1.21 prior drug arrests (SD = 0.53; k = 6) upon entry into the drug court.

The left panel of Table 4 presents descriptive statistics for the key features of the juvenile drug courts. On average, the drug courts served 16.5 youths per year (SD = 10.95), involved 3.6 phases (SD = 1.57), conducted urinalysis screens around 3.9 times per week in the first phase (SD = 3.36), had 2.5 status hearings per month in the first phase (SD = 1.38), and lasted 10.4 months (SD = 2.61). The method of disposition, and how charges were handled upon graduation, were poorly reported in most studies. Most of the drug courts explicitly excluded violent offenders (67 %) and very few (15 %) required youth to have a drug offense to be eligible for participation. Most studies reported that the drug court had dedicated staff (74 %) and reported using a risk assessment tool (61 %), but few studies (24 %) reported that the court provided youth with written documents explaining the contingencies of the program. Finally, most of the drug courts (80 %) referred youth to multiple substance use treatment providers (e.g., multiple community agencies offering treatment, which may have used different modalities or types of treatment) and most of the drug courts (70 %) referred youth to treatment providers using multiple levels of care (i.e., outpatient, intensive outpatient, inpatient).

Table 4 Key features of the juvenile drug courts, and bivariate associations with effect sizes (k = 46)
Full size table

Figure 2 shows the drug courts’ adherence to each of the 16 juvenile drug court strategies as reported in the studies (rated as explicitly used, implicitly used, explicitly not used, or unclear). The strategies most frequently reported explicitly were monitoring and evaluation (100 %), drug testing (91 %), family engagement (83 %), judicial involvement and supervision (80 %), and goal-oriented incentives and sanctions (78 %). The strategies least frequently reported explicitly in the reports were the use of culturally competent approaches (0 %), developmentally appropriate services (0 %), confidentiality (2 %), and a focus on strengths (9 %) (see also left panel of Table 5).

Fig. 2
figure 2

Reporting of adherence to “16 Strategies” for juvenile drug courts

Full size image
Table 5 Juvenile drug courts’ explicit adherence to “16 Strategies” and bivariate associations with effect sizes (k = 46)
Full size table

Overall effects of juvenile drug courts

We first conducted a series of meta-analyses to estimate the overall effects of juvenile drug courts, with analyses split by outcome type (general recidivism, drug recidivism, drug use) and outcome timing (during program, post-programFootnote 2).

General recidivism

Figure 3 shows results from the meta-analysis synthesizing findings from the 11 studies that measured general recidivism during the juvenile drug court program. Although the mean effect size was positive in direction (favoring the juvenile drug court groups), it was not statistically significant, thus indicating that, on average, these juvenile drug courts did not have more beneficial effects on general recidivism during the program than traditional juvenile justice system processing (OR = 1.18, 95 % CI [0.71, 1.98]). However, the heterogeneity across those effect sizes was statistically significant with the individual estimates ranging across a relatively broad range, as is evident in Fig. 3 (τ 2 = 0.29; I 2 = 67.4 %; Q 10 = 25.8, p = 0.004).Footnote 3

Fig. 3
figure 3

Forest plot of general recidivism effect sizes, during program. Notes: Odds ratios and 95 % confidence intervals shown for each included study. All effect sizes coded such that odds ratios greater than 1 indicate a beneficial effect of the drug court (i.e., lower recidivism). The boxes represent the proportionate weight of each individual effect size’s contribution to the mean effect size (larger boxes indicate larger inverse variance weights). The diamond represents the random-effects mean effect size and 95 % confidence interval

Full size image

Figure 4 shows results from the meta-analysis synthesizing findings from the 41 studies that measured general recidivism after the juvenile drug court program period. In these 41 studies, the average post-drug court follow-up period was 14.15 months (SD = 5.59). The mean odds ratio effect size was close to a value of 1, a mean effect that was not statistically significant (OR = 1.03, 95 % CI [0.82, 1.30]). On average, therefore, these drug courts did not reduce the recidivism of the participating juveniles after their involvement with the drug court ended, relative to traditional juvenile justice alternatives. These effect estimates, however, also showed statistically significant heterogeneity (τ 2 = 0.40; I 2 = 79.8 %; Q 40 = 187.3, p < 0.0001). As Fig. 4 shows, this variability ranged from odds ratios that were individually significant in favor of the comparison group to those individually significant in favor of the drug court group.

Fig. 4
figure 4

Forest plot of general recidivism effect sizes, post-program. Notes: Odds ratios and 95 % confidence intervals shown for each included study. All effect sizes coded such that odds ratios greater than 1 indicate a beneficial effect of the drug court (i.e., lower recidivism). The boxes represent the proportionate weight of each individual effect size’s contribution to the mean effect size (larger boxes indicate larger inverse variance weights). The diamond represents the random-effects mean effect size and 95 % confidence interval

Full size image

Drug recidivism

None of the studies included in the meta-analysis provided effect sizes for drug recidivism outcomes during the program period. Figure 5 shows results from the meta-analysis synthesizing findings from the 12 studies that measured drug recidivism after drug court participation. In these 12 studies, the average post-drug court follow-up period was 15.83 months (SD = 10.28). Although the mean effect size was positive in direction (favoring the juvenile drug court groups), and of a size with practical significance, it was not statistically significant (OR = 1.31, 95 % CI [0.78, 2.19]). Here also, the effect estimates showed statistically significant heterogeneity (τ 2 = 0.47; I 2 = 86.5 %; Q 11 = 111.4, p < 0.0001).

Fig. 5
figure 5

Forest plot of drug recidivism effect sizes, post-program. Notes: Odds ratios and 95 % confidence intervals shown for each included study. All effect sizes coded such that odds ratios greater than 1 indicate a beneficial effect of the drug court (i.e., lower recidivism). The boxes represent the proportionate weight of each individual effect size’s contribution to the mean effect size (larger boxes indicate larger inverse variance weights). The diamond represents the random-effects mean effect size and 95 % confidence interval

Full size image

Drug use

Figure 6 shows results from the meta-analysis synthesizing findings from the eight studies that measured drug use during the juvenile drug court program period. The mean effect size was negative in direction (favoring the comparison groups), but not statistically significant (OR = 0.70, 95 % CI [0.26, 1.91]). And, once again, the individual effect estimates showed statistically significant heterogeneity (τ 2 = 0.92; I 2 = 79.4 %; Q 7 = 27.7, p = 0.0002). None of the studies included in the meta-analysis provided effect sizes for drug use outcomes in the post-program period.

Fig. 6
figure 6

Forest plot of drug use effect sizes, during program. Notes: Odds ratios and 95 % confidence intervals shown for each included study. All effect sizes coded such that odds ratios greater than 1 indicate a beneficial effect of the drug court (i.e., lower drug use). The boxes represent the proportionate weight of each individual effect size’s contribution to the mean effect size (larger boxes indicate larger inverse variance weights). The diamond represents the random-effects mean effect size and 95 % confidence interval

Full size image

Exploring the variability in effects

Despite the discouraging finding of no significant mean effects in any of the relevant outcome domains reported above, the consistently significant heterogeneity statistics indicate that the effects of some of the juvenile drug courts were more positive than others. A particular goal of this synthesis was to examine this variability and explore the extent to which it was associated with identifiable characteristics of the respective drug courts or the juvenile participants whose outcomes were assessed in the available studies. To explore this variability, we conducted a series of meta-regressions to examine whether various methodological features, juvenile characteristics, or drug court characteristics were associated with the magnitude of the effects on the recidivism and drug use outcomes.

These analyses were conducted separately by outcome type (general recidivism, drug recidivism, drug use) and follow-up timing (during program, post-program). With so few effect sizes available for each outcome, it was not possible to estimate meta-regression models that included covariates to control for any potential confounders. Each meta-regression, thus, examined only the bivariate relationship between a selected moderator variable and the effect sizes for the respective outcome. Given the large number of significance tests used to examine these bivariate relationships, we used a Benjamini–Hochberg (1995) correction for multiple comparisons within each of the four outcome categories to account for potential inflation in estimates of statistical significance. The results from these moderator analyses should be interpreted cautiously, however, given the small number of effect sizes and the omission of covariates in each analysis.

Study methods and quality

We first examined whether the various methodological features of the studies were associated with the observed effects on recidivism and drug use outcomes. If so, those methodological features might be confounded with some of the substantive features of interest and, as such, would require special attention in the analysis. The right panel in Table 3 shows standardized regression coefficients from a series of bivariate meta-regression models predicting the effect sizes with each of the method quality characteristics in turn.

The results indicated that none of the method quality characteristics were significantly associated with effect size magnitude. For instance, there was no evidence that effects varied for studies using randomized versus quasi-experimental designs (for general recidivism post-program: b = −0.63, 95 % CI [−2.14, 0.87], β = −0.14).Footnote 4 Given that so few studies (k = 3) used randomized designs, we also conducted sensitivity analyses restricting the meta-analyses to studies using either randomized or quasi-experimental designs with individual matching. Restricting the meta-analyses to only include studies using these more rigorous designs yielded substantively similar results to those reported in Figs. 3, 4, 5, and 6 for general recidivism during program (OR = 1.15, 95 % CI [0.53, 2.49]), general recidivism post-program (OR = 1.04, 95 % CI [0.79, 1.36]), drug recidivism post-program (OR = 1.67, 95 % CI [0.89, 3.11]), and drug use during program (OR = 0.64, 95 % CI [0.14, 2.99]). Thus, despite the generally poor quality of the study designs present in this research, restricting the analyses to the more rigorous evaluations available still provided no clear pattern of evidence for the overall effectiveness of juvenile drug courts on recidivism or drug use.

There were also no statistically significant relationships between either overall or differential attrition rates with effect sizes for any of the outcomes. On the other hand, studies with possible implementation problems had significantly smaller effects on post-program general recidivism measures (b = −0.51, 95 % CI [−1.00, −0.02], β = −0.33). However, when each group was examined separately, the mean effect size was not significant among the studies reporting implementation problems (OR = 1.21, 95 % CI [0.92, 1.59]) or among studies without implementation problems (OR = 0.73, 95 % CI [0.48, 1.09]). Further, after applying the Benjamini–Hochberg correction for multiple comparisons, the meta-regression coefficient was no longer statistically significant. Thus, although the body of research synthesized in this meta-analysis suffers from poor methodological quality, there was no clear pattern of evidence that the variations in study quality indexed by these variables were associated with larger or smaller effects.

Juvenile characteristics

The bottom right panel of Table 3 shows standardized regression coefficients from another series of bivariate meta-regression models predicting effect sizes from various characteristics of the juvenile participants represented in the studies providing each outcome. Those regression analyses revealed no statistically significant relationships between the variability in juvenile drug court effects and the age, race, or sex composition of the participating juveniles (none of these effects were significant before or after the Benjamini–Hochberg correction). It is important to note that these demographic variables are at the aggregate study level (e.g., percentage of male participants), so they cannot provide much insight into variability in effects at the individual level (e.g., whether drug courts are more effective for males). It was not possible to reliably estimate the associations between juveniles’ average number of prior arrests (any arrest or only drug arrests) on any outcome except general post-program recidivism given the inconsistent reporting of prior arrest histories. For post-program general recidivism, those relationships were not statistically significant.

Drug court characteristics

The right panel of Table 4 shows standardized regression coefficients from bivariate meta-regression models that examined whether the characteristics of the drug courts themselves, including their adherence to the 16 strategies, were associated with effect size magnitude. Drug courts that used multiple treatment providers for youth referrals showed significantly smaller effects on post-program drug recidivism than courts that used a single provider (b = −1.18, 95 % CI [−2.10, −0.27], β = −0.55). Also, explicitly reported use of risk assessment tools was significantly related to larger effects on drug use outcomes (b = 1.57, 95 % CI [0.36, 2.79], β = 0.61). For that contrast, the mean effect sizes for each of these groups separately were also statistically significant (OR = 1.19, 95 % CI [0.56, 2.50] with risk assessment; OR = 0.25, 95 % CI [0.09, 0.64] without risk assessment). However, none of these effects remained statistically significant after applying the Benjamini–Hochberg correction for multiple comparisons.

The right panel of Table 5 shows results from the meta-regression models examining whether each of the 16 recommended juvenile drug court strategies for which use was explicitly reported was associated with effect size magnitude (vs. implicit or not reported). There was no clear pattern of evidence that the explicit use of any of the 16 strategies was associated with the effects of juvenile drug courts on recidivism or drug use outcomes. We conducted additional sensitivity analysis examining whether the explicit/implicit vs. no reporting of each of the strategies were associated with effects; those results were substantively similar and, thus, are not reported here. It is important to note, however, that several of the strategies initially showed significant negative relationships with effects. Namely, during-program general recidivism was higher in courts that explicitly reported using comprehensive treatment planning (b = −0.91, 95 % CI [−1.69, −0.13], β = −0.52) and focused on strengths (b = −1.29, 95 % CI [−2.30, −0.28], β = −0.66); post-program general recidivism was higher in courts that explicitly reported frequent judicial involvement (b = −0.70, 95 % CI [−1.37, −0.03], β = −0.34) and supervision and family engagement (b = −0.75, 95 % CI [−1.44, −0.07], β = −0.37); and during program drug use was higher in courts that explicitly reported comprehensive treatment planning (b = −1.57, 95 % CI [−2.79, −0.36], β = −0.61), a focus on strengths (b = −1.57, 95 % CI [−2.79, −0.36], β = −0.61), and educational linkages (b = −1.57, 95 % CI [−2.79, −0.36], β = −0.61). None of these effects remained statistically significant after applying the Benjamini–Hochberg correction however. These results should be interpreted cautiously given the small number of effect sizes in any given analysis, and even smaller subgroup sample sizes across the strategies.

Discussion

This meta-analysis synthesized findings from 46 controlled evaluation studies to examine the effects of juvenile drug courts on recidivism and drug use. The results showed no clear pattern of evidence indicating that the juvenile drug courts examined, on average, produced greater reductions in general recidivism, drug recidivism, or drug use than did the traditional juvenile court processes with which they were compared. The methodological quality of the included studies was low, however, with few studies using random assignment and many studies with high attrition rates and substantial baseline differences between drug court and comparison participants on prior risk, race, and sex. When the analysis was restricted to studies using more rigorous designs, the results were substantially the same. Thus, the overall conclusion from this meta-analysis is that the currently available research does not support the view that juvenile drug courts are generally more effective at reducing recidivism or drug use than traditional court processing. These findings are consistent with those reported in prior reviews of the juvenile drug court research, which have also concluded that drug courts have small to no effect on juveniles’ subsequent recidivism (Latessa and Reitler 2015; Latimer et al. 2006; Mitchell et al. 2012a; Shaffer 2006; Stein et al. 2015).

However, there was substantial and statistically significant variability in the effects of juvenile drug courts on all of the outcomes examined. That variability allows for the possibility that there are some configurations of juvenile drug courts that are effective. This meta-analysis, therefore, also explored the relationship of different features of the juvenile participants and the drug courts themselves to the different outcomes, including adherence to the 16 strategies recommended for juvenile drug courts. The aim of those analyses was to identify the distinctive characteristics of those drug courts that did show better outcomes than their comparison conditions in the respective studies. However, none of the participant or drug court characteristics that could be coded from the research reports showed a significant relationship with the recidivism or drug use effects reported. These findings are also consistent with those of prior meta-analyses (e.g., Mitchell et al. 2012a). Though there may indeed be certain forms of juvenile drug courts that are effective, the available research does not provide sufficient information to identify their distinctive characteristics so that they might be emulated elsewhere. Future evaluation studies should attempt to provide more extensive information about the nature of the drug courts and their juvenile participants in order to facilitate identification of any features that are reliably associated with better outcomes. Additional research is also needed to examine the potential effects of juvenile drug courts on other important outcomes, such as school engagement, truancy, and general measures of psychosocial adjustment.

Although our moderator analyses failed to identify drug court characteristics capable of differentiating those courts that produced larger or smaller effects, the overall null findings could be due to several factors about which we might speculate. As noted, for instance, the methodological quality of the currently available studies is quite low, so the null findings could simply be due to the poor baseline equivalence of the juveniles served by drug courts and those served by traditional juvenile courts. Large and variable selection bias in those comparisons could mask positive or negative effects and would increase the heterogeneity of the observed effects across studies. We attempted to investigate this possibility by analyzing only higher quality studies, but were limited by the paucity of such studies.

The descriptive data assembled in this meta-analysis suggest another possible explanation for the null findings. The fundamental logic of juvenile drug courts is that they will be more effective than traditional courts in reducing substance use among youth for whom substance use is problematic and that, in turn, will lead to less overall delinquent behavior. However, while the juvenile participants in most of the drug courts represented in this meta-analysis were drug-involved, there was little indication that very many of them would meet criteria for clinical substance use disorders (i.e., abuse or dependence). Adolescence is a period when experimentation with drugs is not unusual. Youth who do not have serious substance use problems may not benefit greatly from enforced treatment, and whatever drug involvement they have may not be strongly linked to other delinquent behaviors.

Moreover, the low levels of substance use disorders among participants could mean that many juveniles are being referred to inappropriate levels of treatment (Butts and Roman 2004b). Ideally drug court staff would create individualized and tailored treatment plans for each juvenile based on their specific risks and needs (Belenko and Dembo 2003), but the research reviewed here provided little indication that this was common practice. Thus, the services these juvenile drug courts offered may be mismatched with clients’ actual needs in ways that undermine their effectiveness. Whether such mismatches are a particular problem or not, the results of this meta-analysis raise general questions about the effectiveness of the treatment services used by juvenile drug courts. It is surprising how few of the available studies assessed drug use outcomes and, among those that did, how little indication there was of positive effects on those outcomes. Indeed, the increased monitoring and supervision of drug court participants might explain why juveniles in the drug courts had slightly higher rates of drug use compared to those in traditional court processing with presumably less intensive supervision. Nevertheless, if juvenile drug courts do not serve juveniles with serious substance use problems and/or do not provide them with effective substance use treatment, there is little reason to expect them to be more effective in reducing drug-related or general recidivism than traditional court processing.

Indeed, it is not apparent from the data compiled in this meta-analysis that the way traditional juvenile courts handle juveniles with drug involvement is very different from the way they are handled in drug courts. The traditional court comparison conditions described in many of the studies included services similar to those provided in the drug courts, such as judicial supervision and referral to community substance use treatment programs (see Table 1). The substantial similarity in this regard reflects an important difference between adult drug courts and juvenile drug courts, a point well made a decade ago by Butts and Roman (2004a).

The most important unresolved issue may be whether the juvenile justice system really needs juvenile drug courts. Adult drug courts were a significant innovation for the criminal justice system. They introduced a problem-solving approach to a system accustomed to fact finding and punishment... But this approach is not exactly revolutionary in the juvenile justice system. In fact, it is standard operating procedure in traditional juvenile courts. Justice experts even refer to [adult] drug courts as ‘juvenile courts for adults.’ (Butts and Roman 2004a, p. x–xi)

The contrast between juvenile drug court and traditional juvenile court processing thus may be relatively small in many juvenile justice systems. As such, the incremental benefits of handling drug-involved juveniles in specialized drug courts may be as generally negligible as they appear in this meta-analysis, despite the limitations of the research on which those results are based. Indeed, adult drug courts may be effective in reducing recidivism (Mitchell et al. 2012a, b), given that their therapeutic, rehabilitative model is so different from the punitive model used in traditional court processing; whereas, in contrast, most juvenile courts and juvenile drug courts use a therapeutic problem-solving model. An important direction for future research, therefore, is to examine the effects of juvenile drug courts relative to other therapeutic court models, such as mental health courts, which have a similar therapeutic mandate but may address other important comorbid health issues.

Nonetheless, the findings from this review must be interpreted cautiously. Its strengths include the comprehensive literature search used to identify the most current evidence on juvenile drug courts, the relatively large number of studies available for the quantitative synthesis, examination of effects on a range of recidivism and drug use outcomes, and exploration of potential effect size moderators related to the juvenile participants and drug court procedures. The primary limitation for drawing conclusions from this review, however, stems from the low methodological quality of the research evidence that was synthesized. Further, inconsistent reporting of the characteristics of the methods, participants, and court procedures in the available studies hampered our ability to conduct extensive moderator analyses.

While recognizing the shortcomings of the available evidence, the consistency of the overall null results across the outcomes examined and the moderator analyses conducted provides little or no support for the view that juvenile drug courts are generally more effective than traditional juvenile courts for reducing drug use or re-offense rates. The heterogeneity in the drug court effects leaves open the possibility that there may be particular drug court configurations that are especially effective but, if so, their characteristics cannot be identified from currently available research. The available evidence, therefore, neither shows overall average positive effects of drug courts on any of the relevant outcomes nor offers guidance about what might be effective drug court practice for obtaining such effects. To be most informative under these circumstances, any future research studies should use high quality controlled trials to study drug courts that have well-specified structures, procedures, and treatment protocols.