A mixed methods systematic review on the effects of arts interventions for children and young people at-risk of offending,or who have offended on behavioural,psychosocial,cognitive and offending outcomes: A systematic review

Types of studies

We included randomised and non-randomised controlled trials and quasi-experimental study designs. We did not include quantitative studies that did not employ a control or comparator group. We included qualitative studies that were conducted alongside intervention trials that investigated the experiences and perceptions of participants, and that offered insight into the barriers and facilitators associated with delivering and receiving arts interventions. We included qualitative and mixed methods studies that were focused on the delivery of an arts intervention. We included those which explored aspects of the process of intervention delivery from the perspectives of those delivering and those who were participants in the intervention and/or their carers/family members or significant agents (e.g., probation officers). We included studies from any global setting.

Types of participants

We included studies that included CYP (8–25 years) who were either identified as at-risk of offending behaviour (secondary populations) or already in the criminal justice system (tertiary populations).

Types of interventions

We included studies of interventions involving arts participation. Arts participation included involvement in artistic and creative activities. Studies which included arts participation as an intervention on its own or alongside other interventions, such as mentoring, were included. We included studies that used art as therapy (a form of psychotherapy) and as a medium to address emotional difficulties.

Examples of arts interventions in included studies are:

Music making

We included studies that compared arts interventions to either no intervention, usual care, other types of arts intervention or non-arts control. The intervention had to involve organised arts interventions targeted to the CYP population. We did not include associational studies between arts participation and offending behaviour.

Types of outcome measures

Briefly describe the types of outcome measures that will be included and excluded.

Duration of follow-up

We considered outcomes at the following time points: short-term immediately post-intervention to <3 months; medium-term 3 to <12 months post intervention, long-term >1 year post intervention. Where studies reported multiple follow-ups within a single time-point range we preferentially extracted as follows: short-term, the closest follow-up point to the end of the intervention. Medium- and long-term: the latest time point reported.

Types of settings

We included studies employing arts interventions in any setting including (i) for example, youth offender institutions (UK) or juvenile correctional facilities (USA), prisons, other residential settings; (ii) community and workplace settings; and (iii) schools.

Electronic searches

We consulted Campbell guidance on searching for studies (Kugley, 2016). Our search strategy included expert advice from information services experts at Brunel University London Library. We searched key databases including AMED (via EBSCOHost NHS Open Athens), Academic Search Complete; APA PsycInfo; CINAHL Plus; ERIC; SocIndex; SportDiscus (via EbscoHost), Medline (via Ovid), CENTRAL, Web of Science, Scopus, PTSDPubs and Performing Arts Periodicals Database (via ProQuest), Sage, the US National Criminal Justice Reference Service, the Global Policing and the British Library EThOS databases, and the National Police Library to January 2023.

We used a combination of controlled vocabulary, that is, medical subject headings (MeSH), and free text terms to identify published articles. In addition, we checked reference lists of reviews and retrieved articles for additional studies. Search strategies can be found in Supporting Information: Appendices 1–9. We included separate search strings for identifying quantitative and qualitative studies. To identify the population of interest we used the search filter proposed by the Canadian Health Libraries Association (CHLA, 2022). We used and adapted the Cochrane highly sensitive search filter to identify RCTs (Higgins, 2021), a validated filter for identifying non-randomised controlled studies (Waffenschmidt, 2020) and the University of Texas School of Public Health (University of Texas, 2022) filter for identifying qualitative studies which have been demonstrated to show good performance in sensitivity and specificity (Wagner, 2020).

Our searches were worldwide and included studies from any country. We agreed the provision and support for translation of potentially relevant papers into English.

Titles and abstracts were independently screened by two reviewers to identify potential sources of disagreement. These were discussed and reviewed by a third senior author in the team. Two reviewers then screened the full texts of potentially relevant studies and applied the inclusion and exclusion criteria, with recourse to a third reviewer for any records where there was uncertainty.

We checked the reference lists of relevant systematic reviews found in our searches.

Searching other resources

In searching other sources we sought expert advise. We worked with an expert Advisory Board convened by Campbell and the Youth Endowment Foundation for this purpose. We searched the WHO International Clinical Trials Registry Portal (ICTRP). We conducted a grey literature search of databases such as Arts and Humanities Citation Index and ProQuest using our search terms. In discussion and agreement with our Advisory Board, we conducted a selected website search including the UK's National Criminal Justice Arts Alliance and other websites with a specific focus on CYP and the criminal justice system. We conducted an Advanced Google Scholar search and sifted the first 100 returns using search terms from our search strategy as appropriate. We included reports that met our inclusion criteria. We used our inclusion and exclusion criteria to select grey literature. We revised our grey literature search alongside advice from experts on the project Advisory Board.

Description of methods used in primary research

The interventions involved organised arts interventions targeted to the population. We did not include associational studies between arts participation and offending behaviour. We included controlled study designs. Studies that compared arts interventions to either no intervention, usual care, or other types of arts intervention or non-arts control were included. We included studies comparing one type of arts intervention to another. Studies using any qualitative research method to examine the context, intervention assumptions, implementation process (including barriers and facilitators), and mechanisms of impact and outcomes were included. Qualitative studies evaluating how an arts intervention works could be conducted as independent studies or alongside controlled study designs.

Selection of studies

Two review authors independently assessed the titles and abstracts of potential included studies identified by the search strategy for their eligibility. We obtained the full text of studies we considered potentially eligible, or if the eligibility of a study was unclear from the title and abstract. We excluded studies that did not match the inclusion criteria (see ‘criteria for considering studies for this review’). We resolved disagreements between review authors regarding inclusion by discussion. If agreement could not be reached, a third review author assessed relevant studies, and a majority decision was made. We did not anonymise studies before the assessment.

Data extraction and management

Two review authors independently extracted data from all included studies using a standardised and piloted data extraction form. They resolved discrepancies and disagreements by consensus. In cases where consensus could not be achieved, a third review author assessed the article, and a majority decision was made.

We extracted the following data from quantitative studies included in the review:

Study characteristics (aims/objectives, study design, sample size, description of the sample, country, recruitment years and procedure, conflict of interest, funding source).

Characteristics of the participants (gender/sex, age, ethnicity, socioeconomic status, education, at risk or in contact/conflict with the criminal justice system).

Description of the interventions (experimental and control), context and setting, country/location, intervention assumptions/theoretical framing, implementation processes (human and financial resources), fidelity, dose, adaptation, reach, mechanisms of impact (participant response, mediators, unanticipated consequences).

Data collection methods including duration and timing of follow-up/outcome assessment.

Results as outcome measures of interest to this review, including details of measurement scales and analysis methods.

Risks and biases.

Discussion including interpretations by authors, limitations and implications.

We extracted the following data from qualitative studies in the review:

Study characteristics and context (aims/objectives study design, sampling approach, description of the sample, country, recruitment years and procedure, conflict of interest, funding source).

Characteristics of the participants (gender/sex, age, ethnicity, socioeconomic status, education, at risk or in contact/conflict with the criminal justice system).

Description of the interventions, context and setting, country/location, theoretical framing, implementation processes (human and financial resources), processes of impact (funding context, design and delivery model, participants' responses, unanticipated consequences).

Data collection methods.

Findings as qualitative themes/processes including analysis methods.

Methodological limitations.

Discussion including interpretations by authors and implications.

As arts interventions are complex we extracted detailed information regarding the intervention guided by the MRC guidance on process evaluations of complex interventions (Moore, 2015) and items on the TiDiER (Hoffman, 2014) and Cert (Slade, 2016) checklists framed by a focus on why, what, who, where, when, how much, how well, tailoring and modifications.

Assessment of risk of bias in included studies

Two reviewers independently assessed the risk of bias or study quality of included studies. We used the Cochrane Risk of Bias tool (Higgins, 2011) to evaluate included controlled trials. We assessed the following domains for each study: Random sequence generation; Allocation concealment; Blinding of participants and providers; Blinding of assessors; Incomplete outcome data; Selective reporting. For cross-over studies only we assessed the domain ‘free from carry-over effects’.

We took a risk-to-rigour approach to evaluate qualitative studies (Noyes, 2018) using the CASP tool for qualitative research (CASP UK) to appraise the rigour and significance of the sampling, data, collection, analysis and reporting of results.

Measures of intervention effect

For continuous outcome measures we expressed the size of the intervention effect using the mean difference (MD) when all studies utilised the same measurement scale, or the standardised mean difference (SMD) when studies used different scales, with 95% CIs. If pooling from different scales for which the direction of interpretation varied, we planned to normalise the direction of the scales to a common direction. To aid the interpretation of the pooled effect size, we planned to back-transform the SMD to the most commonly used outcome scale on the basis of the median standard deviation from trials using that scale when possible.

For dichotomous outcomes we planned to report the Relative Risk, Odds Ratio or Risk Difference where available from individual included studies. In the event that we pooled data in a new analysis, we planned to preferentially report the relative risk as the effect size of interest but to also report the risk difference.

Unit of analysis issues

Unit-of-analysis issues refer to issues regarding clustering (individuals randomised/allocated in clusters), cross-over designs, and studies with multiple outcome measurement time points.

For studies with more than two eligible active treatment groups that are included in a meta-analysis as separate interventions, we planned to divide the number of participants in the control group between active treatment groups, to avoid double counting (Higgins, 2021). For cluster RCTs, we planned to seek direct estimates of the effect from an analysis that accounted for the cluster design. When the analysis in a cluster trial does not account for the cluster design, we planned to use the approximately correct analysis approach, presented in the Cochrane Handbook (Higgins, 2021). For cross-over studies, we planned to only include data from the first phase of the study, when they are available due to the risk of carryover effects. However, as first-phase, or phase-by-phase data were not available for any of the included cross-over studies we took the decision to analyse these studies as presented.

Criteria for determination of independent findings

Where we identify multiple reports for a single study we only included data from that study once in any given analysis. Where a study reported multiple outcome domains with some conceptual overlap that fit one of our stated outcome domains, the research team agreed which of the measures conceptually best matched our outcome of interest and only included that measure. This decision was not made on the basis of the results of these outcomes.

Dealing with missing data

Where there were insufficient data presented in the study report to enter into an analysis, we requested the missing data from the study authors. We planned to preferentially calculate and extract effect sizes derived from intention-to-treat analyses. We evaluated the potential risk of bias introduced by missing data in our assessment of risk of bias, within the domain ‘Incomplete Outcome Data’ and planned to explore the impact of risk of bias through sensitivity analyses.

Assessment of heterogeneity

We planned to deal with heterogeneity by only combining studies that examine similar interventions. To estimate statistical heterogeneity, we would calculate the χ² statistic, the between-study variance (τ ²) and the proportion of this variance not due to sampling error (I²). We planned to use these measures, together with a visual inspection of the forest plots to form judgements about heterogeneity. If we identified substantial heterogeneity, we planned to report it and explore possible causes by prespecified subgroup analysis.

Assessment of reporting biases

We planned to consider the potential influence of small study biases on review findings. We planned to use funnel plots to visually explore small study biases where there were at least 10 included studies in a meta-analysis.

Data synthesis

We conducted separate analyses of the quantitative evidence for the following comparisons: Arts intervention versus no intervention or usual practice; Arts interventions versus non-arts control; Arts interventions versus other arts intervention.

We planned to pool studies of arts interventions in the primary analysis, including different types of arts, delivery modes and settings using a random-effects model to account for the anticipated heterogeneity between studies. For each comparison of interest, we planned to conduct separate analyses at short, medium and long-term follow-up. For the primary analysis, we planned to pool data from studies regardless of the specific population. Where there were inadequate data to enable statistical pooling we planned to conduct a narrative synthesis of the evidence. For head-to-head comparisons of different types of arts intervention, we would only pool studies if the intervention and comparators are conceptually similar.

In the event that we conducted a narrative synthesis we planned to separately synthesise studies within the comparisons outlined above, guided by the SWiM guideline (Campbell, 2020). We planned to first summarise all arts interventions in the primary synthesis, including different types of arts, delivery mode and setting and explore the potential heterogeneity of treatment effects between studies by considering intervention settings (custodial, community or school-based interventions) and population age (children and adolescents 8–18 years; young adults 18–25 years). We reported effect sizes for each reported outcome of interest with estimates of precision where available. We included all relevant studies for each comparison and outcome and documented the size and risk of bias of those studies in our reporting. We did not plan to present this synthesis in a tabular or graphical format.

See below for details of the synthesis of qualitative research.

Subgroup analysis and investigation of heterogeneity

Sensitivity analysis

When sufficient data were available, we planned to explore the impact of risk of bias for the primary analyses, by repeating the analyses and excluding studies rated at high risk of bias.

Treatment of qualitative research

We took a thematic approach to analysing and synthesising data from qualitative studies. This included line-by-line reading for extraction and preliminary coding, development of descriptive themes and refinement of analytical themes (Thomas, 2008). We conducted our thematic analysis with attention to the complexity of arts interventions for CYP at risk of offending or in the criminal justice system. We mapped themes from the findings of the qualitative studies to theoretical domains of complexity relating to the intervention itself, the population, the implementation of the intervention and the specific context that may impact the process of delivering and engaging with the interventions. Table 2 outlines the complexity framework for qualitative analysis. We took a reflexive approach and consulted with Advisory Board to seek advice about the relevance of themes for policy and practice.

Summary of findings and assessment of the certainty of the evidence

We used the GRADE system to rank the level of certainty of the evidence (Schünemann, 2020). The GRADE approach uses five considerations (risk of bias, consistency of effect, imprecision, indirectness, and publication bias) to assess the certainty of the body of evidence for each outcome, and uses the following criteria to describe the confidence in the evidence:

high – we are very confident that the true effect lies close to that of the estimate of the effect

We decreased the grade rating by one (−1), two (−2), or three (−3) levels, up to a maximum of −3, (or very low) for any criteria, based on the level of concern it raised.

To assess the methodological quality of the included qualitative studies, two review authors independently applied the CASP quality checklist for each qualitative study. The checklists were used to indicate if a specific study had been well-designed, appropriately carried out and properly analysed. We then employed the CERQual schema (Confidence in the Evidence of Reviews of Qualitative Research) for judging how much confidence could be placed in the overall review findings developed through the synthesis. CERQual has four components (methodological limitations, relevance adequacy and coherence) and uses the following criteria for judging confidence in the body of literature (Lewin et al., 2015).

Methodological limitations – the extent to which there are problems in the design or conduct of primary studies that contributed evidence to the review.

Relevance – the extent to which evidence in the primary study is applicable (perspective, population, phenomenon of interest, setting).

Coherence – the degree to which primary studies provide a convincing explanation for patterns.

Adequacy – the degree of richness and quantity/scope of data

Confidence was decreased if there were serious or very serious limitations in the design or conduct of the study, the evidence was not relevant to the study objectives, the findings/conclusions were not supported by the evidence, or the data were of inferior quality and inadequate in supporting the findings. Confidence was increased if the study was well-designed with few limitations, the evidence was applicable to the context specified in the objectives, the findings/conclusions were supported by evidence and provided a convincing explanation for any patterns found or the data supporting findings were rich and of high quality.

Results of the search

See Figure 1 for a summary of the search screening process. There were 54,598 initial records from database searches and 22 from other sources (reference searches and experts' contributions). After de-duplication 20,196 records were retained. Forty-four records of 43 studies met our eligibility criteria (3 quantitative, 38 qualitative and 2 mixed methods).

OPEN IN VIEWER

Figure 1

Flow diagram of the search screening process.

Included studies

Country of origin and setting

Study design

Participants

Interventions

Outcomes

Excluded studies

There are 76 Excluded studies. Reasons for exclusion included incorrect population, study design, intervention, outcome and/or comparator.

Quantitative studies

See Figures 2 and 3 for a summary of the risk of bias evaluation for quantitative studies. All 5 studies were rated at high risk of bias overall due to being at high risk of bias for multiple domains of the Cochrane Risk of Bias tool.

OPEN IN VIEWER

Figure 2

Risk of bias summary: Review authors' judgements about each risk of bias item for each included study.

OPEN IN VIEWER

Figure 3

Risk of bias graph: Review authors' judgements about each risk of bias item presented as percentages across all included studies.

Only two studies (Bittman, 2009; Tyson, 2002) described a random process of allocation, of which one (Tyson, 2002) reported a clear method of randomisation and was rated at low risk of bias on this domain. The remaining studies were at unclear risk (Bittman, 2009), or at high risk as they were not randomised (Anderson, 2010; Caulfield, 2022; DeCarlo, 2004). Of the randomised studies neither reported a process for ensuring concealed allocation and were judged at high risk of bias on this domain.

None of the studies attempted to blind participants or practitioners and all were rated at high risk of bias for this domain. Two studies (Anderson, 2010; Bittman, 2009) clearly did not ensure blinded assessors and were rated at high risk, and three (Caulfield, 2022; DeCarlo, 2004; Tyson, 2002) did not clearly report blinding of assessors and were rated at unclear risk. All studies were rated at high risk of bias for incomplete outcome data due to attrition levels and a lack of accounting for attrition in the analyses. All 5 studies were rated at unclear risk of bias for selective outcome reporting due to the lack of registration records of published protocols.

The single cross-over trial (Bittman, 2009) was rated at high risk of bias for carry-over effects as no washout period was observed, no baseline adjustment was applied in the analysis and there were signs of potential baseline imbalance on one outcome.

Qualitative studies – Quality appraisal

See Table 3 for CASP quality appraisal of individual qualitative studies and Table 4 for CERQual evidence profile for qualitative studies.

For the qualitative studies, the most frequent methodological weaknesses within the studies were limited discussion of recruitment strategies, missing detail about the process of data collection, a lack of rigour in data analysis, inadequate discussion of relationships between participants and researcher and incomplete information regarding ethical procedures, approvals or issues. The results of the quality checklist for qualitative studies varied with the best scoring (meeting 8 out of 8 criteria) in six sources (Gowland-Pryde, 2016; Hanrahan, 2017; Howard, 2022; Lea, 2019; Varley, 2019; Zlotowitz, 2016) and the worst scoring (meeting 3 out of 8 criteria) in six sources (Baker, 2007; de Roeper, 2009; Lotter, 2015; Massó-Guijarro, 2020; Winn, 2010, 2011).

The use of the CERQual schema for judging the confidence in the findings from the synthesis of qualitative evidence resulted in a judgement of very low confidence about the evidence for understanding micro, meso and macro-level processes influencing the successful design and delivery of arts-based interventions for at-risk and offending young people, and their impact on behavioural, psychosocial, cognitive and offending outcomes. Very low confidence judgements are due to major concerns with methodological limitations, coherence and adequacy, and moderate concerns about relevance.

Arts interventions versus no intervention or usual care

We requested additional data to allow the analysis of effect sizes from Anderson (2010); Bittman (2009); Caulfield (2022); DeCarlo (2004) but did not receive these data. Results for the primary outcomes for this comparison are presented in Summary of findings Table 1.

Arts interventions versus other types of arts intervention

Results for the primary outcomes for this comparison are presented in Summary of findings Table 2.

The only outcome where there was evidence comparing one arts intervention to another was psychological and emotional well-being.

Arts interventions versus non-arts intervention

Only one study compared an arts intervention to a non-arts intervention. DeCarlo (2004) compared RAP therapy to group psycho-educational therapy using the composite outcome, the RAP Therapy assessment scale. However, data were not reported in adequate detail to allow the analysis of effect sizes.

Micro-level experiences, barriers and facilitators

Micro-level processes refer to findings reported about individual participants in their setting and reflect a population complexity domain by identifying the relationship between CYP at-risk of offending or who have already offended, arts interventions, individual demographics, cultures, sociopsychological and cognitive factors.

The qualitative studies all reported that participants experienced a range of positive emotions through taking part in arts-based interventions. Some specific emotions were identified including feelings of hope and aspiration (Atherton, 2022), enjoyment (Gowland-Pryde, 2016), self-confidence or courage (Barrett, 2015; Caulfield, 2019; Gann, 2010; Lotter, 2015), pride in successfully creating an art form or gratitude at having access to art-based interventions (Daykin, 2017; Hadland, 2010; Seroczynski, 2011), and being valued, respected and/or praised for their involvement and creative outcomes (Bowey, 2006; Caulfield, 2022: Chong, 2020; Lazzari, 2005; Nicklin, 2017; Parker, 2018; Thompson, 2015, 2022). Positive emotional experiences of arts-based interventions generally underpinned and served to foster a sense of possibility or hope for the future, including able to manage and change more negative emotions and traumatic experiences associated with being at risk, having offended and/or being in the justice system (Caulfield, 2019; Fullchange, 2018; Hanrahan, 2017; Pope, 2022; Varley, 2019; Winn, 2010, 2011). Positive emotional experiences were also reported as encouraging development of a more positive attitude to learning, employment and skill development (Cesar, 2020; Tett, 2012) and acceptance of the potential benefits of creative activity (Caulfield, 2019).

In some studies, the elicitation of positive emotions was reported as enhancing self-reflection and a critical self-awareness for CYP defined as at-risk or offending. Participants were reported as being able to redefine a more (self) caring and accepting sense of identity that challenged and resisted established and negative mindsets through learning new skills and achieving successful arts-related outcomes (Atherton, 2022; Clennon, 2015; Gann, 2010; Hanrahan, 2017; Massó-Guijarro, 2020; Morgan, 2020; Nicklin, 2017; Parker, 2018; Tett, 2012; Thompson, 2015; Winn, 2010, 2011). It was suggested by study authors that such processes of self-determination could enable experiences of autonomy and empowerment through participation in art activities (Hickey, 2018; Lotter, 2015; Pope, 2022). The creative process itself was cited in some studies as important in allowing for experiences of self-expression and sometimes through the perception that participation in the arts allows for exploration through risk in a safe context (Atherton, 2022; Baker, 2007; Barrett, 2012; Cesar, 2020; de Roeper, 2009; Gowland-Pryde, 2016; Hickey, 2018; Lazzari, 2005; Lotter, 2015; Podkalicka, 2009; Varley, 2019).

Studies reported that arts-based interventions for CYP at-risk of offending or who have offended, contributed to the development of positive personal relationships with other CYP and adults involved in the arts intervention work. Arts-based interventions were reported to allow for supportive interactions that could lead to relationships of trust, reciprocity and care (Atherton, 2022; Caulfield, 2019; Cesar, 2020; Daykin, 2017; Hanrahan, 2017; Lazzari, 2005; Lea, 2019; Massó-Guijarro, 2020; Morgan, 2020; Nicklin, 2017; Parker, 2018; Zlotowitz, 2016). Arts interventions were also described as fostering more collaborative relationships based on sharing and/or listening, features that can support help-seeking behaviours. Arts-based interventions for CYP in this population groups were reported as helping participants to develop a sense of belonging and positive awareness of others including those in the justice setting and more broadly with families and communities.

Micro-level barriers to the successful implementation of arts interventions for CYP at-risk of offending or who have offended were identified to some extent. Where participants felt that they had no choice about participation or the arts activities and where arts interventions were not culturally relevant to them, implementation was challenging (Howard, 2022). Achieving positive outcomes from arts interventions was made difficult in situations where there were high levels of anxiety amongst participants generally (Flores, 2016) and where anxiety levels could be exacerbated in the context of having to learn something or perform (Fullchange, 2018; Hickey, 2018) There is also recognition that positive personal experiences happen only in the moment of taking part in arts and may not necessarily spill over into wider life (Cesar, 2020).

Meso-level processes influencing design and delivery

Meso-level processes refer to findings reported in the qualitative studies about community-level experiences and impacts as well as design and implementation of arts intervention for CYP at-risk of offending or who have offended. In the qualitative studies meso level findings also capture connections between micro and meso-level processes and reflect three complexity domains; contextual (residential/setting status and involvement, family and community support), intervention (providers, type, youth-focused, facilities and theoretical framing) and implementation (mode delivery, fidelity, adherence, local support structures).

Qualitative studies identified the significance of understanding the setting in which arts interventions were taking place for successful engagement and delivery and provided evidence of the contextual complexity of arts intervention work. Studies reported that successful arts interventions in prison settings were designed to recognise but disrupt or distract from the punitive/penal structures (Atherton, 2022; Baker, 2007; Hickey, 2018; Lazzari, 2005). Arts-based interventions were also reported as offering a more flexible environment than that of formal education, allowing participants to be creative, learn, achieve and interact with others in positive, relatively informal ways. The suggestion that this led to supportive and trusting relationships was made for interventions both in custodial and community (Caulfield, 2022; Morgan, 2020; Podkalicka, 2009) and educational settings (Jordan, 2015; Parker, 2018). Studies identified further contextual complexity in terms of the importance of ensuring that arts-based interventions included culturally relevant programmes. Culturally relevant features included art activities that reflected the interests and needs of CYP at-risk or offending and were meaningful to them such as music technology (Clennon, 2015), writing, poetry and music (Lea, 2019), songwriting, lyric writing and video recording (Massó-Guijarro, 2020) rap music composition (Baker, 2007; Hickey, 2018) and music making (Thompson, 2015). Studies also reported elements of implementation complexity as important in ensuring arts-based interventions elicited positive emotional and behavioural outcomes including enabling creativity as self-expression and/or risk-taking in a non-judgemental (safe) environment (Atherton, 2022; Baker, 2007; Barrett, 2012; Cesar, 2020; de Roeper, 2009; Gowland-Pryde, 2016; Hickey, 2018; Lotter, 2015; Podkalicka, 2009; Varley, 2019) and ensuring participants felt a sense of ownership and belonging in the arts intervention space (Bowey, 2006; Clennon, 2015; de Roeper, 2009; Pope, 2022). Intervention complexity was reported in studies including the importance of facilitator characteristics, attitudes and behaviours, expert arts instructions, leadership or facilitation, by a relatable and/or experienced ‘artist’ who was able to work the CYPs to co-produce art, take a position that respectfully challenging young people to be creative and supported them in their endeavours to engage in the process of producing art (Atherton, 2022; Caulfield, 2019; Clennon, 2015; Daykin, 2017; Gowland-Pryde, 2016; Howard, 2022; Lazzari, 2005; Pope, 2022; Tett, 2012). Additionally, complexity in barriers to design and implementation were reported to include negative attitudes of prison staff to arts interventions (Daykin, 2017; Howard, 2022).

Macro-level influences on experience, barriers and facilitators

Macro-level processes refer to findings reported about wider societal, economic and political drivers and impacts of arts interventions. Macro-level findings capture the interconnections between societal contexts in which arts interventions are designed and implemented and local experience of participants. Studies did not always refer directly to macro-level processes in reports of findings, but there was some, albeit limited commentary on the ways in which these shaped complexity in terms of experiences of CYP at-risk of offending or who had offended, the design and delivery of arts intervention, and their potential to impact on behavioural, psychosocial, cognitive and offending outcomes. We note here that the evidence for macro level process was extremely limited and restricted to four studies. Therefore, clear conclusions about them are not possible. One qualitative study from the USA reported the arts intervention as a response to a lack of policy advocacy for the arts for CYP at risk of offending or who had offended and associated limits to funding streams which impact negatively on prioritising of arts programmes in the sector (Cesar, 2020). One study highlighted the need to develop interventions with external partners in policy and practice to implement sustainable and successful arts programmes for offending CYP connected to the justice system (Caulfield, 2019). One study from South Korea (Chong, 2020) and another from the USA (Winn, 2010) identified the need to connect arts-based interventions to wider communities and national organisations for supporting CYP at-risk of offending CYP. The study implied that partnership working between the justice system, and education, housing and healthcare organisations could serve to support more effective and sustainable design and delivery of arts programmes to address potential and actual offending behaviours.

Quantitative evidence

We found insufficient evidence from quantitative studies to support or refute the effectiveness of arts interventions for CYP at-risk of offending or who have offended for any outcome. From the included studies there was only very low certainty evidence and no clear evidence of an effect of arts interventions for any of the included outcomes. We found no evidence for our primary outcome ‘Offending Behaviour’. The evidence base is poorly developed and there is no published evidence available.

Qualitative

The qualitative studies in this review showed that micro-level experiences of arts-based interventions for CYP at-risk of offending or who had offended which may influence the success of arts-based interventions broadly reflected individual positive personal experiences. These included positive emotions, the development of a sense of self, successful engagement in creative processes and practices, and the development of positive personal relationships with people involved in the intervention (other participants, prison staff and delivery experts) and other people in their lives, for example, family, friends and communities. Meso-level contextual processes influence the success of arts-based interventions and include the need for accessible delivery sites which can offer flexibility in timing and access to, and support for using appropriate space and facilities in which arts-based interventions can act as a creative alternative to restricted life in youth justice settings. Qualitative studies reported the need for arts-based interventions to be designed as culturally relevant (reflecting race/ethnic, national identify, gender and age characteristics, for example), youth-focused and theoretically framed with models or theories relevant to practice, consistent, regular and sustainable if they are to be successful in engaging CYP at-risk of offending or those who have offended. In addition, the role of supportive staff, family and community members, and expert delivery by professional artists to whom participants could relate were also reported as meso-level intervention processes important to the successful implementation of arts programmes for this population. Macro-level barriers to the implementation of arts interventions in youth justice settings were reported in very few studies but lack of policy advocacy for arts in youth justice settings is highly likely to mean that funding is limited and that arts interventions are not prioritised. There was reference in one study to the need to connect arts interventions to wider support in relation to education, housing, and healthcare. This is because it is likely that the most successful and sustainable arts interventions for CYP at-risk of offending or who have offended will be informed and supported by key personnel in schools, communities and healthcare.

Quantitative

All studies were small, conducted in a single centre, and at high risk of bias across multiple domains. Only two studies were described as randomised and, of those, only one described that process adequately. No study reported a process of allocation concealment. While blinding is clearly challenging in this field there was little detail reported on attrition or appropriate methods of analysis to account for it. No study was pre-registered with an available protocol, and so there is a risk of selective outcome reporting and other post hoc changes in study conduct. The reported details of interventions was often quite superficial, creating challenges for understanding what was done, and subsequently for replication.

Qualitative

Methodological weaknesses were found in most qualitative studies including limited discussion of recruitment strategies, scant detail about data collection, a lack of rigour in data analysis, no adequate discussion of relationships between participants and researcher, and a lack of information regarding ethical procedures, approvals or issues. The use of the CERQual schema for judging the confidence in the findings from the synthesis of qualitative evidence results in a judgement of very low confidence about the evidence for understanding micro, meso, and macro-level processes influencing the successful design and delivery of arts-based interventions for at-risk and offending young people and their impact on behavioural, psychological, cognitive and offending outcomes. Very low confidence judgements are due to major concerns with methodological limitations, coherence and adequacy, and moderate concerns about relevance.