History of Alcoholism and the Justice System

Aggression and Violent Behavior 65 (2022) 101761

Available online 30 June 2022 1359-1789/© 2022 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by- nc-nd/4.0/).

Developmental predictors of offending and persistence in crime: A systematic review of meta-analyses

Miguel Basto-Pereira a,*, David P. Farrington b

a William James Center for Research, ISPA-Instituto Universitário, R. Jardim do Tabaco 34, 1100-304 Lisboa, Portugal b Institute of Criminology, Cambridge University, Sidgwick Avenue, Cambridge CB3 9DA, United Kingdom

A R T I C L E I N F O

Keywords: Developmental predictors Offending Juvenile delinquency Meta-analyses Longitudinal studies

A B S T R A C T

Meta-analyses have provided major findings about developmental predictors of offending. However, there has been little focus on their relative ability to predict offending behaviour. Therefore, we conducted a systematic review of meta-analyses with two aims: 1) to summarize all well-established knowledge about developmental (explanatory) predictors of offending, and 2) to sort those predictors according to their effect size. The strongest predictors of general offending were related to family/parental dimensions. Delinquent peers, school/employ- ment problems, family problems, certain types of mental health problems, and alcohol/substance abuse were the most important predictors of persistence in crime. Our findings suggest the crucial role of family-related developmental predictors in preventing offending. The predictors of persistence in crime highlight the multi- systemic nature of persistent antisocial behaviour.

1. Introduction

A deep understanding of developmental factors that longitudinally predict offending and persistence in crime is particularly relevant in explaining offending and in addressing the causes of criminal behaviour effectively. Because our interest is in explanation rather than pure pre- diction, we focus on explanatory predictors, defined as those that are measuring an underlying construct that is different from antisocial behaviour. Thus, we exclude behavioural predictors such as previous offending, aggression or conduct disorder.

Over the last 50 years, multiple longitudinal studies have been initiated to advance knowledge about the factors predating or causing criminal behaviour (Farrington, 2013; Jolliffe et al., 2017). Different longitudinal studies have addressed distinct sets of different predictors, and these studies have found a multitude of important risk factors for delinquency and conduct disorder, such as poor parental supervision, impulsiveness, low IQ, family disruption, social inequality, school problems, and antisocial models (Farrington et al., 2017; Murray & Farrington, 2010).

The various longitudinal studies (e.g., Cambridge Study in Delin- quent Development; Pittsburgh Youth Study; Dunedin Longitudinal Study) conducted over the years have resulted in a new era of theories on developmental criminology (Farrington, 2006; Loeber, 2019; Moffitt,

2018; Wikstrom et al., 2012), with several practical implications for justice policies (e.g., Zane, 2021), assessment tools (e.g., Wormith, 2011), and more effective interventions (e.g., Tremblay et al., 2003).

These advancements have also led to some scientific consensus across studies and contemporary theoretical approaches. For example, nowadays it is known that juvenile delinquency is an important risk factor for adult criminal behaviour, although it is also known that most youth offenders will cease their criminal behaviour when entering adulthood (e.g., Farrington, 2003; Laub & Sampson, 2001; Moffitt, 1993, 2018). In addition, the most serious and chronic criminal careers are influenced by both environmental (e.g., antisocial peers) and indi- vidual/temperamental (e.g., impulsiveness) risk factors during child- hood (Cicchetti, 2016; Farrington, 2003; Laub & Sampson, 2001; Moffitt, 1993).

In contrast, there are still many controversial issues about criminal career development. Whereas we know that some of the risk factors that explain or predate delinquency are similar across longitudinal studies (e. g., antisocial models), the relative importance of each of these causes, the interaction of each with age or gender, or the processes explaining how each potential causal mechanism influences the decision to initiate, persist, or desist from a criminal career are still far from achieving sci- entific consensus (e.g., for a review, see Basto-Pereira & Maia, 2017 and Siegel, 2015). For example, Laub and Sampson (2001) argued that

* Corresponding author. E-mail address: [email protected] (M. Basto-Pereira).

Contents lists available at ScienceDirect

Aggression and Violent Behavior

journal homepage: www.elsevier.com/locate/aggviobeh

https://doi.org/10.1016/j.avb.2022.101761 Received 1 October 2021; Received in revised form 27 May 2022; Accepted 25 June 2022

Aggression and Violent Behavior 65 (2022) 101761

2

youths with delinquent patterns share similar childhood risk factors regardless of the seriousness or chronicity of criminal behaviour, and it is the strengthening of bonds with society later in life (e.g., entering the labor market, marriage) that is the main reason for the cessation of criminal careers. In opposition, for Moffitt (1993, 2018), the factors explaining persistence/desistance from crime during adulthood are mostly dependent on causal mechanisms already present during child- hood (e.g., neuropsychological problems, uncontrolled behaviour, inadequate parenting).

To overcome many of these controversial issues, meta-analyses of longitudinal studies appeared as a solution to provide a reliable and replicable strategy of summarizing results and identifying common patterns across studies. Consequently, in recent years, there have been an increasing number of meta-analyses in this field, as a result of the need to summarize the studies from multiple cohorts and provide solid evidence-based knowledge about the mechanisms underlying crime. Different dimensions of general offending have been tested across a set of meta-analyses of longitudinal studies, including child maltreatment (Braga et al., 2017), parental supervision (Flanagan et al., 2019), and individual/temperamental characteristics (e.g., intelligence; Ttofi et al., 2016). In addition, some meta-analyses have also reviewed the pre- dictors of persistence in crime among justice-involved youths.

The main aim of meta-analyses that analyse long-term longitudinal studies is to understand whether different types of social, psychological, or biological factors during development temporally predict offending or persistence in crime. Therefore, the intrinsic question in this type of study is often related to the level of significance: Do scientific studies indicate that factors during development increase the risk of later offending? A significance value below 0.05 is typically interpreted as a “yes”. However, p-values do not measure effect size (e.g., Wasserstein & Lazar, 2016). In the case of very small effect sizes, conclusions based on p-values might be misleading or deceptive for various reasons. First, predictors with very small effect sizes might be so close to zero that in practice their effect is irrelevant for interventions or public policies (Sullivan & Feinn, 2012; Szucs & Ioannidis, 2017). Second, criminal behaviour, like any other human behaviour, is an extremely complex phenomenon reflecting the interaction of a large and intricate network of societal, familial, and biological factors (Szucs & Ioannidis, 2017; Woods, 1988). In this article, we focus on effect sizes.

1.1. The current study

To advance knowledge about the mechanisms underlying criminal behaviour, there is a need to map the multitude of relationships pro- vided by meta-analytic studies and to refocus on what effect sizes across meta-analyses of longitudinal studies tell us about the causes of criminal behaviour. In other words, a deeper understanding of the most impor- tant mechanisms underlying offending reported across meta-analyses has several advantages. It enables us 1) to test the empirical validity of current theories of crime, 2) to know what needs to be tested in future meta-analyses, and maybe most crucially, 3) to identify the most important explanatory predictors of crime; these will contribute to developing more accurate risk assessment tools and more effective in- terventions to prevent offending and future recidivism.

In addition, previous research (Basto-Pereira et al., 2015) has noted significant differences between developmental predictors of youth offending in the community population when compared with predictors of recidivism among justice-involved youths. From a theoretical and legal point of view, youths previously exposed to the justice system are typically older and affected by a larger number of risk factors (Loeber & Farrington, 2012). In this regard, the first contact with the justice system often causes or aggravates the risks of labeling effects and deviant peer contagion (e.g., Bernburg et al., 2006; Dishion & Tipsord, 2011). Thus, predictors of general offending in community populations, particularly when measured during childhood, are normally more representative of the very early stages of a criminal career, while developmental

predictors of persistence in crime may involve the presence of multiple risk factors developed as a consequence of criminal career progression during adolescence in interplay with the consequences of early justice contact. For this reason, it is particularly relevant to study develop- mental predictors of offending in both cases, in the general community, and among justice-involved youths.

Therefore, focusing on addressing these concerns, we conducted a systematic review of meta-analyses to address these two main aims: 1) to summarize all the well-established knowledge about developmental predictors of offending and 2) to sort those predictors by their impor- tance (effect size) for predicting offending in the general population or persistence in crime among justice-involved youths.

2. Methods

2.1. Search process and eligibility criteria

Following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Moher et al., 2009), systematic literature searches were conducted in six major databases—Web of Sci- ence, Scopus, PsychINFO, PsychARTICLES, Scielo, and PubMed—to iden- tify meta-analyses analysing characteristics that longitudinally predict antisocial behaviour during adolescence and adulthood. In addition, a search was conducted manually in key journals that publish meta- analyses. Meta-analyses were searched from the very beginning of the databases until January 25, 2020. The following search terms and Boolean operators were used: meta-analysis AND delinquen* OR offend* OR violen* OR recid* OR crim* OR antisocial OR conduct problems OR disruptive OR rearrest OR reoffend* AND (None) OR predict* OR factors OR desist* OR persist*; this resulted in 50 search combinations.

For a study to be considered eligible, it must a) be a meta-analysis, b) analyse psychological, social, or biological characteristics during childhood or adolescence predicting antisocial outcomes (e.g., rearrests, convictions) later in life, c) analyse explanatory predictors of criminal behaviour, d) have diversified samples of offenders or community samples, and e) be published in English, Spanish, or Portuguese in peer- reviewed journals up to January 25, 2020. The following exclusion criteria were adopted: a) the outcome evaluated only a particular type of crime, b) the meta-analysis was conducted in a community or offending sample with specific characteristics (e.g., offenders with mental illness), c) longitudinal effect sizes were not reported, or could not be directly calculated using the data provided, d) the meta-analysis did not provide well defined and rigorous definitions of measures, outcomes, and in- clusion/exclusion criteria, e) there was a lack of explanatory predictors of crime tested, f) the meta-analysis did not examine predictors during childhood/adolescence, and g) the predictors of interest for this study were based on fewer than two studies.

2.2. Study selection and data collection

The study selection process was conducted in the following order: 1) removal of duplicates, 2) screening abstracts for exclusion of papers not fulfilling the eligibility criteria, and 3) all the papers that were not excluded after abstract-screening were read through carefully to ensure the exclusion of all studies that were not in compliance with the pre- established criteria.

Information was obtained from the selected meta-analyses on the following topics: a) the source (bibliographic reference), b) the partici- pants' genders, c) the types of predictors, d) the participants' ages when predictors were measured (childhood versus adolescence versus mixed), e) the number of studies analysed by predictors in each meta-analysis, f) the standardized mean effect sizes, g) the p-values, h) the types of outcome, and i) the participants' ages when outcomes were measured.

M. Basto-Pereira and D.P. Farrington

Aggression and Violent Behavior 65 (2022) 101761

3

2.3. Synthesis of results and analytic strategy

Detailed information about each predictor, from each meta-analysis included, was collected and described in detail. Mean effect sizes were converted to r metrics. Each predictor was placed in one of two tables: predictors of crime or predictors of persistence in crime during adult- hood. The outcome of persistence in crime includes meta-analyses assessing predictors of recidivism during adulthood among justice- involved youths, and meta-analyses assessing predictors of life-course persistent (versus adolescence-limited) trajectories of criminal behav- iour. The effect size of each longitudinal predictor of crime is presented separately for males and females if an included meta-analysis reports that effect size separately by gender (e.g., predictor: low-attain- ment–Females, predictor: low-attainment–Males).

Static predictors (e.g., gender; ethnicity), and behavioural predictors of crime, measuring a similar underlying construct to offending (e.g., previous offending, aggression or conduct disorder), were excluded from our analyses, which focused on explanatory predictors. Subsequently, the developmental predictors were separated into two different cate- gories: predictors of general offending and predictors of recidivism among youths with a history of offending. For reasons of simplicity and clarity, all predictive factors were coded in the risk direction. Informa- tion about reversed factors are provided in each table (e.g., Prosocial peer relations reversed to Low prosocial peer relations).

In a subsequent analysis, those predictors were ordered by their ef- fect size, from the larger to the smaller effects. Detailed information about each predictor was provided (e.g., bibliographic reference, num- ber of effect sizes included, etc.). Non-significant predictors, or pre- dictors with an effect size r smaller than 0.10, were excluded from these analyses because we aimed to identify the strongest explanatory pre- dictors of crime. Also, to avoid bias caused by a very small number of independent samples (including an overestimation of the real effect size), in this meta-synthesis of findings, all the predictors tested in less than five samples were excluded. According to Jackson and Turner (2017, p. 290): “5 or more studies are needed to reasonably consistently achieve powers from random-effects meta-analyses that are greater than the studies that contribute to them”. Lastly, predictors were analysed as major dimensions to add comprehensibility and interpretability to our analysis.

We have excluded predictors with small values of r from Tables 3 and 4 in order to highlight the strongest predictors. However, small values of r (e.g., r = 0.10) do not necessarily indicate weak relationships. For example, consider a 2 × 2 table relating a dichotomous risk factor to a dichotomous outcome such as delinquency. Assume that there are 100 people in the risk category out of a total of 400, and that 100 of the total number of people become delinquent. Now, if 33 out of 100 (33 %) in the risk category become delinquent, compared with 67 out of 300 (22.3 %) in the non-risk category, the product-moment correlation r (also called the phi correlation in a 2 × 2 table) would be 0.107. In general, an r value of 0.10 would correspond to an absolute difference of about 10 % between risk and non-risk categories in a 2 × 2 table (for all the relevant formulae, see Farrington & Loeber, 1989). However, the relative dif- ference is substantial; almost 50 % more of those in the risk category became delinquent, compared with those in the non-risk category (33 % compared with 22.3 %). This effect therefore cannot be considered insignificant.

3. Results

3.1. Selected meta-analyses

A total of 4095 articles were found. Of the 4095 articles, 3149 were duplicates. Titles and abstracts were screened for the remaining 946, and 869 were excluded from these initial screening, mainly because the articles found were not meta-analyses. Seventy-seven meta-analyses passed the initial screening and were retained for full-text reading.

Sixty-three meta-analyses were excluded for the following reasons: a) 21 meta-analyses did not examine predictors during childhood/adoles- cence, b) 14 meta-analyses did not test longitudinal predictors of general offending, c) 13 meta-analyses did not test causal, explanatory, dynamic predictors of crime, d) 10 meta-analyses evaluated only a particular type of crime or category of crime, e) two meta-analyses were conducted in a sample with specific characteristics (e.g., only individuals with psychi- atric diagnoses), f) in two studies, predictors of general offending were evaluated in samples mixing minors and adults, g) in one meta-analysis, the predictors of interest were tested with fewer than two studies. Fourteen meta-analyses of longitudinal studies assessing developmental predictors of general offending and/or persistence in crime during adulthood were included (see Fig. 1).

3.2. Study characteristics

This systematic review included 14 meta-analyses of longitudinal studies. Eleven meta-analyses tested developmental predictors of gen- eral offending (Braga et al., 2017; Braga et al., 2018; Derzon, 2010; Flanagan et al., 2019; Hoeve et al., 2012; Leschied et al., 2008; Portnoy & Farrington, 2015; Reaves et al., 2018; Spruit et al., 2016; Ttofi et al., 2016; Wilson et al., 2009), while three meta-analyses tested develop- mental predictors of recidivism (Assink et al., 2015; Cottle et al., 2001; Scott & Brown, 2018). The meta-analyses were published between 2001 (Cottle et al., 2001) and 2019 (Flanagan et al., 2019) in peer-reviewed journals. Thirteen meta-analyses examined our predictors of interest using gender-mixed samples, while one of the studies (Scott & Brown, 2018) conducted analyses separately for males and females. Twelve meta-analyses reported mean effect sizes using r or Cohen's d metrics, while two studies used different metrics, namely, the Odds-Ratio or OR (Ttofi et al., 2016) and Fisher's Z, which is similar to r (Cottle et al., 2001). For all the effect sizes provided, a conversion to r metrics was performed.

Eleven meta-analyses included only longitudinal designs, while three meta-analyses included and analysed separately studies with cross- sectional and longitudinal designs (Portnoy & Farrington, 2015; Spruit et al., 2016; Wilson et al., 2009). All the studies testing predictors of recidivism included only studies with longitudinal designs. The number of samples included in those meta-analyses ranged between 13 (Reaves et al., 2018) and 119 (Derzon, 2010); on average each meta-analysis included approximately 44 independent samples. Three longitudinal studies (Braga et al., 2018; Flanagan et al., 2019; Portnoy & Farrington, 2015) evaluated criminal and non-criminal forms of antisocial behav- iour together; we recalculated the mean effect size only for criminal behaviour.

Eight meta-analyses did not differentiate childhood from adolescent predictors of offending or recidivism, while six meta-analyses (Braga et al., 2017; Cottle et al., 2001; Leschied et al., 2008; Scott & Brown, 2018; Spruit et al., 2016; Wilson et al., 2009) analysed the impact of predictors on specific phases of life using a longitudinal design (child- hood or adolescence). All the meta-analyses testing predictors of offending used outcomes measured during adolescence or adulthood, while outcomes of recidivism among young offenders were always assessed during adulthood. For a detailed description of all the tested developmental predictors of offending and persistence in crime, see Tables 1 and 2.

Generally, the significance tests use two-tailed p-values, but one- tailed tests would be justified in the light of clear directional pre- dictors (based on risk factors). Therefore, the number of significant re- sults is underestimated.

3.3. Summary of the meta-findings

Tables 3 and 4 summarize all the well-established longitudinal pre- dictors of offending and persistence in crime and sort those predictors by their importance according to effect size. Since the objective was to

M. Basto-Pereira and D.P. Farrington

Aggression and Violent Behavior 65 (2022) 101761

4

summarize the most important predictors of antisocial outcomes, non- significant predictors, predictors with relatively small effect sizes (r < 0.10) and those tested in fewer than five independent samples were excluded from these tables.

Of the 14 meta-analyses included in this study, 112 longitudinal predictors were identified: 1) 53 predictors of general offending across 11 meta-analyses, and 2) 59 predictors of persistence in crime across three meta-analyses. The largest effect sizes for general offending during childhood and adolescence were the family structure (e.g., child involved in the child welfare system; marital status of the parents) during adolescence (r = 0.32; Leschied et al., 2008), lack of child-rearing skills (r = 0.26; Derzon, 2010), home discord (r = 0.26; Derzon, 2010), family structure during childhood/adolescence (r = 0.23; Leschied et al., 2008), and low level of parental knowledge (r = 0.22; Flanagan et al., 2019). Lack of parental management was the best predictor of general offending during childhood (r = 0.20; Leschied et al., 2008), and family structure was the best predictor of general offending during adolescence (r = 0.32; Leschied et al., 2008).

The most important longitudinal predictor of persistence among

juvenile justice youths was non-severe pathology, such as stress or anxiety (r = 0.30; Cottle et al., 2001), female education/employment (r = 0.25; Scott & Brown, 2018), male-lack of prosocial peer relations (r = 0.23; Scott & Brown, 2018), family problems (r = 0.22; Cottle et al., 2001), and alcohol/drug abuse (r = 0.21; Assink et al., 2015).

Education/employment problems (r = 0.25 for females; Scott & Brown, 2018; to r = 0.15, Assink et al., 2015), family problems (r = 0.22, Cottle et al., 2001; to r = 0.10 for males, Scott & Brown, 2018), and (delinquent) Peers (r = 0.20, Cottle et al., 2001; to r = 0.13 for females, Scott & Brown, 2018), were consistent predictors of persistence in crime across all the meta-analyses, always showing effect sizes r > 0.10 across all three meta-analyses.

In addition, dimensions of alcohol or substance abuse and specific dimensions related to mental health were statistically significant pre- dictors of persistence in crime. Specific dimensions of mental health, such as non-severe psychopathology (Cottle et al., 2001) and emotional and behavioural problems (Assink et al., 2015) were statistically sig- nificant predictors of crime with r > 0.15. The meta-analyses conducted by Scott and Brown (2018) addressing mental health as the presence of a

Fig. 1. Flow-diagram.

M. Basto-Pereira and D.P. Farrington

Aggression and Violent Behavior 65 (2022) 101761

5

Table 1 Childhood and adolescent predictors of general offending.

Reference Predictors k r Age period - Predictor Age period - Outcome

Braga et al. (2017) Maltreatment 7 0.11* Childhood Adolescence

Braga et al. (2018)** Maltreatment 8 0.14* Childhood/adolescence Adulthood

Derzon (2010)

Parent's education and expectations 3 0.30 Childhood/adolescence Adolescence/adulthood (Lack of) child rearing skills 13 0.26* Childhood/adolescence Adolescence/adulthood Home discord and instability 11 0.26* Childhood/adolescence Adolescence/adulthood Family stress 10 0.21* Childhood/adolescence Adolescence/adulthood Maltreated as child 8 0.21* Childhood/adolescence Adolescence/adulthood Other family deviance 9 0.19* Childhood/adolescence Adolescence/adulthood (Lack of) warmth and relationship 22 0.18* Childhood/adolescence Adolescence/adulthood (Inappropriate) discipline 9 0.17* Childhood/adolescence Adolescence/adulthood Parent antisocial behaviour 11 0.15* Childhood/adolescence Adolescence/adulthood Foster care 5 0.14* Childhood/adolescence Adolescence/adulthood Urban housing 9 0.13* Childhood/adolescence Adolescence/adulthood Family size 9 0.11* Childhood/adolescence Adolescence/adulthood Broken home 25 0.10* Childhood/adolescence Adolescence/adulthood Unwanted pregnancy 5 0.10 Childhood/adolescence Adolescence/adulthood Residential mobility 3 0.08* Childhood/adolescence Adolescence/adulthood Separated from parents 2 0.08* Childhood/adolescence Adolescence/adulthood Parent use and tolerate ATOD (alcohol, tobacco, and drug use of adolescents)

1 0.08 Childhood/adolescence Adolescence/adulthood

Young parent(s) 4 0.08 Childhood/adolescence Adolescence/adulthood (Lack of) supervision and involvement 10 0.06 Childhood/adolescence Adolescence/adulthood Parental psychopathology 4 0.02* Childhood/adolescence Adolescence/adulthood

Flanagan et al. (2019)**

Low level of parental knowledge a 8 0.22* Childhood/adolescence Adolescence/adulthood Low supervision a 18 0.18* Childhood/adolescence Adolescence/adulthood Child closure a 6 0.16* Childhood/adolescence Adolescence/adulthood Lack of parental rule setting a 4 0.12* Childhood/adolescence Adolescence/adulthood

Hoeve et al. (2012) Low attachment a 17 0.17* Childhood/adolescence Adolescence/adulthood

Leschied et al. (2008)

Family structure adolescence 19 0.32* Adolescence Adulthood Family structure total 36 0.23* Childhood/adolescence Adulthood Parent management middle childhood 8 0.20* Childhood Adulthood Adverse family environment adolescence 15 0.19* Adolescence Adulthood Internalizing concerns adolescence 24 0.14 Adolescence Adulthood Family structure middle childhood 5 0.13 Childhood Adulthood Parent management total 17 0.12* Childhood/adolescence Adulthood Internalizing concerns - Total 42 0.11* Childhood/adolescence Adulthood Adverse family environment total 35 0.11* Childhood/adolescence Adulthood Family structure early childhood 12 0.08* Childhood Adulthood Adverse family environment early childhood 9 0.08* Childhood Adulthood Adverse family environment mid childhood 11 0.08* Childhood Adulthood Parent mental health-Total 36 0.07* Childhood/adolescence Adulthood Social and interpersonal concerns middle childhood 7 0.07 Childhood Adulthood Parent mental health early childhood 21 0.07 Childhood Adulthood Parent mental health adolescence 15 0.07 Adolescence Adulthood Parent management adolescence 4 0.06 Adolescence Adulthood Internalizing concerns middle childhood 14 0.05 Childhood Adulthood Social and interpersonal concerns total 18 0.04 Childhood/adolescence Adulthood Social and interpersonal concerns early childhood 7 0.01 Childhood Adulthood

Portnoy and Farrington (2015)** Low resting heart rate a 6 0.07* Childhood/adolescence Adolescence/adulthood

Reaves et al. (2018) Negative school climate - Interpersonal relationships a 3 0.21* Childhood/adolescence Adolescence Negative school climate- Institutional environment a 16 0.14* Childhood/adolescence Adolescence

Spruit et al. (2016) Lack of sports participation a 8 0.07* Adolescence Adolescence

Ttofi et al. (2016) Low intelligence a 4 0.08 Childhood/adolescence Adolescence/adulthood

Wilson et al. (2009) Childhood violence exposure 3 0.15* Childhood Adolescence

Note. ** Using the data provided in the meta-analyses or direct contact with the author, overall effect sizes were recalculated including only longitudinal studies assessing childhood/adolescent predictors of general offending; k = Number of studies. *, p < .05. a = Reversed Protective Factor.

M. Basto-Pereira and D.P. Farrington

Aggression and Violent Behavior 65 (2022) 101761

6

Table 2 Childhood and adolescent predictors of persistence in crime.

Reference Predictor k r Age period - Predictor Age period - Outcome

Assink et al. (2015)

Alcohol/drug abuse 57 0.21* Childhood/adolescence Adulthood Sexual behaviour 7 0.20* Childhood/adolescence Adulthood Relationship 51 0.19* Childhood/adolescence Adulthood Emotional and behavioural problems 150 0.18* Childhood/adolescence Adulthood School/employment 63 0.15* Childhood/adolescence Adulthood Other 27 0.13* Childhood/adolescence Adulthood Family (problems) 273 0.12* Childhood/adolescence Adulthood Attitude 19 0.10* Childhood/adolescence Adulthood Physical health 14 0.04 Childhood/adolescence Adulthood Neighborhood 16 − 0.04 Childhood/adolescence Adulthood

Cottle et al. (2001)

Nonsevere pathology 7 0.30* Adolescence Adulthood Ineffective use of leisure timea 2 0.23* Adolescence Adulthood Family problems 5 0.22* Adolescence Adulthood Delinquent peers 7 0.20* Adolescence Adulthood Number of out-of-home placements 2 0.18* Adolescence Adulthood Low standardized achievement scorea 3 0.15* Adolescence Adulthood Substance abuse 6 0.15* Adolescence Adulthood Low full scale IQa 5 0.14* Adolescence Adulthood History of special education 2 0.13* Adolescence Adulthood Victim of abuse 5 0.11* Adolescence Adulthood Low verbal IQ scorea 4 0.11* Adolescence Adulthood Single parent 5 0.07* Adolescence Adulthood Low performance IQ scorea 2 0.31 Adolescence Adulthood Severe pathology 2 0.07 Adolescence Adulthood Low school attendancea 2 0.05 Adolescence Adulthood Parent pathology 3 0.04 Adolescence Adulthood Low school report of achievementa 6 0.03 Adolescence Adulthood History of treatment 2 0.02 Adolescence Adulthood Substance use 2 0.01 Adolescence Adulthood

Scott and Brown (2018)

Female Education/employment 8 0.25* Adolescence Adulthood Female-Lack of prosocial peer relations (outliers removed)a 4 0.15* Adolescence Adulthood Female problematic family circumstances and parenting 12 0.14* Adolescence Adulthood Female antisocial peer relations 12 0.13* Adolescence Adulthood Female education/school concerns (outlier removed) 5 0.10* Adolescence Adulthood Female substance abuse 16 0.05* Adolescence Adulthood Female mental health 5 0.04* Adolescence Adulthood Female-Low level of prosocial values and attitudesa 3 0.52 Adolescence Adulthood Female-Low of family relationships and supporta 4 0.38 Adolescence Adulthood Female-Personality – Low self-efficacy, positive problem solvinga 3 0.26 Adolescence Adulthood Female-Low of extracurricular activities and community supporta 6 0.16 Adolescence Adulthood Female Child abuse 4 0.1 Adolescence Adulthood Female-Low of education and employment opportunitiesa 3 0.06 Adolescence Adulthood Female poor use leisure/recreation (outlier removed) 9 0.05 Adolescence Adulthood

Male- Low education and employment opportunitiesa 3 0.32* Adolescence Adulthood Male-Low family relationships and supporta 4 0.27* Adolescence Adulthood Male- Lack of rejection or non-absence of substance usea 3 0.27* Adolescence Adulthood Male- Lack of prosocial peer relationsa 5 0.23* Adolescence Adulthood Male education/employment problems (outlier removed) 7 0.21* Adolescence Adulthood Male antisocial peer relations (outlier removed) 10 0.20* Adolescence Adulthood Male poor use leisure/recreation 10 0.16* Adolescence Adulthood Female - Lack of rejection or non-absence of substance usea 3 0.15* Adolescence Adulthood Male education/school concerns (outlier removed) 5 0.13* Adolescence Adulthood Male problematic family circumstances and parenting 12 0.10* Adolescence Adulthood Male substance abuse 16 0.08* Adolescence Adulthood Male- low extracurricular activities and community supporta 6 0.2 Adolescence Adulthood Male-low level of prosocial values and attitudesa 3 0.14 Adolescence Adulthood Male mental health 5 0.02 Adolescence Adulthood Male child abuse 4 0 Adolescence Adulthood Male-personality – Low self-efficacy, positive problem solvinga 3 − 0.01 Adolescence Adulthood

Note. ** Using the data provided in the meta-analyses or direct contact with the author, overall effect sizes were recalculated including only longitudinal studies assessing childhood/adolescent predictors of persistence. k = Number of studies. *, p < .05. a = Reversed Protective Factor.

M. Basto-Pereira and D.P. Farrington

Aggression and Violent Behavior 65 (2022) 101761

7

mental health problem or diagnosis (Yes/No) showed substantially lower effect sizes. Another important disparity between effect sizes was found for substance abuse. The dimension of alcohol/drug abuse (Assink et al., 2015) was an important longitudinal predictor of persistence in crime (r = 0.21), but predictors exclusively addressing substance abuse in the other two meta-analyses showed statistically significant but sub- stantially smaller effect sizes (r = 0.15, Cottle et al., 2001; r = 0.08 for males, r = 0.05 for females, Scott & Brown, 2018).

Finally, Scott and Brown's (2018) meta-analyses provided separate

analyses by gender, and these found education/employment problems as the most important predictors of recidivism among males (r = 0.21) and females (r = 0.25), followed by masculine antisocial peers (r = 0.20) and female problematic family circumstances/parenting (r = 0.14). Table 5 summarizes the concepts and effect sizes in each meta-analysis of persistence associated with each one of the five key categories identified.

Table 3 Most important predictors of general offending ordered by overall effect size.

Reference Predictor k r Type Age period -Predictor Age period – Outcome

Leschied et al. (2008) Family structure - Adolescence 19 0.32* General offending Adolescence Adulthood Derzon (2010) (Lack of) child rearing skills 13 0.26* General offending Childhood/adolescence Adolescence/adulthood Derzon (2010) Home discord and stability 11 0.26* General offending Childhood/adolescence Adolescence/adulthood Leschied et al. (2008) Family structure - Total 36 0.23* General offending Childhood/adolescence Adulthood Flanagan et al. (2019) Low level of parental knowledgea 8 0.22* General offending Childhood/adolescence Adolescence/adulthood Derzon (2010) Family stress 10 0.21* General offending Childhood/adolescence Adolescence/adulthood Derzon (2010) Maltreated as child 8 0.21* General offending Childhood/adolescence Adolescence/adulthood

Leschied et al. (2008) Lack of parent management middle childhood (supervision/discipline)

8 0.20* General offending Childhood Adulthood

Leschied et al. (2008) Adverse family environment adolescence 15 0.19* General offending Adolescence Adulthood Derzon (2010) Other family deviance 9 0.19* General offending Childhood/adolescence Adolescence/adulthood Derzon (2010) (Lack of) warmth and relationship 22 0.18* General offending Childhood/adolescence Adolescence/adulthood Flanagan et al. (2019) Poor supervisiona 18 0.18* General offending Childhood/adolescence Adolescence/adulthood Hoeve et al. (2012) Low attachment 17 0.17* General offending Childhood/adolescence Adolescence/adulthood Derzon (2010) (Inappropriate) discipline 9 0.17* General offending Childhood/adolescence Adolescence/adulthood Flanagan et al. (2019) Child closurea 6 0.16* General offending Childhood/adolescence Adolescence/adulthood Derzon (2010) Parent antisocial behaviour 11 0.15* General offending Childhood/adolescence Adolescence/adulthood Reaves et al. (2018) Negative school climate – Institutional environment 16 0.14* General offending Childhood/adolescence Adolescence Braga et al. (2018) Maltreatment 8 0.14* General offending Childhood/adolescence Adulthood Derzon (2010) Foster care 5 0.14* General offending Childhood/adolescence Adolescence/adulthood Derzon (2010) Urban housing 9 0.13* General offending Childhood/adolescence Adolescence/adulthood Leschied et al. (2008) Parent management total 17 0.12* General offending Childhood/adolescence Adulthood Leschied et al. (2008) Internalizing concerns – Total 42 0.11* General offending Childhood/adolescence Adulthood Leschied et al. (2008) Adverse family environment total 35 0.11* General offending Childhood/adolescence Adulthood Derzon (2010) Family size 9 0.11* General offending Childhood/adolescence Adolescence/adulthood Braga et al. (2017) Maltreatment 7 0.11* General offending Childhood Adolescence Derzon (2010) Broken home 25 0.10* General offending Childhood/adolescence Adolescence/adulthood

Notes. Including only dynamic predictors with k ≥ 5, r ≥ 0.10 and p < .05; k = Number of studies. *, p < .05. a = Reversed Protective Factor.

Table 4 Predictors of persistence in crime during adulthood.

Reference Predictor k r Outcome type Age period -Predictor Age period - Outcome

Cottle et al. (2001) Nonsevere pathology 7 0.30* Recidivism Adolescence Adulthood Scott and Brown (2018) Female - Education/employment 8 0.25* Recidivism Mostly adolescence Adulthood Scott and Brown (2018) Male-Lack of prosocial peer relations a 5 0.23* Recidivism Mostly adolescence Adulthood Cottle et al., 2001 Family problems 5 0.22* Recidivism Adolescence Adulthood Assink et al. (2015) Alcohol/drug abuse 57 0.21* Persistent Del Behav Childhood/adolescence Adulthood Scott and Brown (2018) Male-Education/employment problems 7 0.21* Recidivism Mostly adolescence Adulthood Scott and Brown (2018) Male-Antisocial peer relations 10 0.20* Recidivism Mostly adolescence Adulthood Assink et al. (2015) Sexual behaviour problem 7 0.20* Persistent Del Behav Childhood/adolescence Adulthood Cottle et al. (2001) Delinquent peers 7 0.20* Recidivism Adolescence Adulthood Assink et al. (2015) Relationship 51 0.19* Persistent Del Behav Childhood/adolescence Adulthood Assink et al. (2015) Emotional and Behavioural problems 150 0.18* Persistent Del Behav Childhood/adolescence Adulthood Scott and Brown (2018) Male-Poor use leisure/recreation 10 0.16* Recidivism Mostly adolescence Adulthood Assink et al. (2015) School/employment 63 0.15* Persistent Del Behav Childhood/adolescence Adulthood Cottle et al. (2001) Substance abuse 6 0.15* Recidivism Adolescence Adulthood Scott and Brown (2018) Female-Problematic family circumstances/parenting 12 0.14* Recidivism Adolescence Adulthood Cottle et al. (2001) Low full scale IQ 5 0.14* Recidivism Adolescence Adulthood Assink et al. (2015) Other 27 0.13* Persistent Del Behav Childhood/adolescence Adulthood Scott and Brown (2018) Female-Antisocial peer relations 12 0.13* Recidivism Adolescence Adulthood Scott and Brown (2018) Male-Education/school concerns 5 0.13* Recidivism Adolescence Adulthood Assink et al. (2015) Family (problems) 273 0.12* Persistent Del Behav Childhood/adolescence Adulthood Cottle et al. (2001) Victim of abuse 5 0.11* Recidivism Adolescence Adulthood Assink et al. (2015) Attitude 19 0.10* Persistent Del Behav Childhood/adolescence Adulthood Scott and Brown (2018) Male-Problematic family circumstances/parenting 12 0.10* Recidivism Adolescence Adulthood Scott and Brown (2018) Female-Education/school concerns 5 0.10* Recidivism Adolescence Adulthood

Notes. Including only explanatory predictors with k ≥ 5, r ≥ 0.10 and p < .05. Persistent Del Behav = Persistence in crime was assessed through persistent (versus adolescence limited) trajectories of criminal behaviour during adulthood. Persistent Del Behav = Persistent Delinquent Behaviour. a = Reversed Protective Factor.

M. Basto-Pereira and D.P. Farrington

Aggression and Violent Behavior 65 (2022) 101761

8

4. Discussion

This study addresses one of the major aims of developmental crim- inology, which is to evaluate the childhood and adolescent factors that precede or explain offending behaviour (Farrington et al., 2017; Loeber & Le Blanc, 1990). To our knowledge, this is the first systematic review of meta-analyses that maps all the well-established knowledge about the developmental predictors of offending and to sort those risk/protective factors according to their relative importance in predicting offending and persistence in crime. In addition, this study is particularly useful because the in-depth knowledge of these factors is crucial in developing better theories and more effective assessment tools, interventions, and justice policies.

We identified 11 meta-analyses addressing longitudinal predictors of general offending, most of them showing statistically significant pre- dictors; however, three meta-analyses did not present longitudinal predictors of offending with effect sizes equal or larger than r = 0.10. In addition, for most of the predictors that were assessed across meta- analyses addressing persistence in crime, many effect sizes were also small. These initial findings support the notion that complex events influenced by a variety of factors, such as criminal behaviour, result in a large network of statistically significant factors; however, some of those factors may have low theoretical and practical relevance (Orben & Przybylski, 2019). Therefore, this work is an opportunity to sort each one of the meta-analysed predictors by their effect size and identify major dimensions in criminal behaviour during child/adolescent development.

4.1. Developmental predictors of general offending and persistence in crime

The results of our systematic review of meta-analyses show that early family-related factors had some of the larger effect sizes in predicting general offending. Those family-related variables include family struc- ture, home discord, (lack of) child-rearing skills, family stress (Derzon, 2010), level of parental knowledge (Flanagan et al., 2019), parental management during middle childhood (supervision/discipline), and an adverse family environment during adolescence (Leschied et al., 2008).

Three meta-analyses also highlighted the effect of child (Braga et al., 2017; Derzon, 2010) and adolescent maltreatment (Braga et al., 2018) on later general offending. These findings clearly support previous psychobiological (e.g., Lee & Hoaken, 2007) and psychosocial (e.g., Kerig & Becker, 2015) approaches stressing the detrimental impact of child abuse and neglect on later delinquency. In addition, child maltreatment is exclusively (e.g., neglect) or often (sexual or physical abuse) perpetrated by family members (Langevin et al., 2019; Papalia et al., 2020). Moreover, children from dysfunctional families are particularly at risk of being victims of maltreatment (e.g., Stith et al., 2009).

Contrary to our expectations, dimensions like resting heart rate (Portnoy & Farrington, 2015) or child internalizing concerns (Leschied et al., 2008) showed small and/or non-significant effect sizes in pre- dicting general offending. Whereas family predictors among children and youths appear to be the most important predictors of general offending (Flanagan et al., 2019; Leschied et al., 2008), among adoles- cents with justice involvement, family problems are only one of the key predictors of persistence during adolescence and adulthood (Assink et al., 2015; Cottle et al., 2001; Scott & Brown, 2018). We identified five key developmental predictors: occupation (education/employment) problems (Assink et al., 2015; Scott & Brown, 2018), delinquent/anti- social peers (Assink et al., 2015; Cottle et al., 2001; Scott & Brown, 2018), specific dimensions related to mental health problems (Cottle et al., 2001), alcohol/drug abuse (Assink et al., 2015; Cottle et al., 2001), and family problems (Assink et al., 2015; Cottle et al., 2001; Scott & Brown, 2018). More primary research is needed comparing predictors of offending versus recidivism (see e.g., Farrington, 2020). Ta

bl e

5 D

efi ni

tio ns

o f t

he c

on st

ru ct

s as

so ci

at ed

w ith

k ey

d im

en si

on s

pr ed

ic tin

g pe

rs is

te nc

e in

c ri

m e

du ri

ng a

du lth

oo d.

A ut

ho rs

Fa

m ily

r

Em pl

oy m

en t

r Pe

er d

el in

qu en

cy

r A

lc oh

ol /

Su bs

ta nc

e A

bu se

r

M en

ta l H

ea lth

Sy

m pt

om s

r

A ss

in k

et a

l. (2

01 5)

Fa m

ily (

pr ob

le m

s)

“F ac

to rs

re la

tin g

to fa

m ili

al p

ro bl

em s,

su ch

a s h

av in

g cr

im in

al fa

m ily

m

em be

rs , l

ow p

os iti

ve p

ar en

tin g,

la rg

e fa

m ily

si ze

, h av

in g a

p oo

r r el

at io

n w

ith

pa re

nt s,

an d

pa re

nt al

c on

fli ct

” (

A ss

in k

et a

l., 2

01 5,

p .5

0)

0. 12

* Sc

ho ol

/e m

pl oy

m en

t “F

ac to

rs r

el at

in g

to e

du ca

tio n

an d

em pl

oy m

en t,

su ch

a s p

oo r

ac ad

em ic

ac

hi ev

em en

t, be

in g

a fr

eq ue

nt tr

ua nt

, ha

vi ng

a la

ck o

f i nt

er es

t i n

sc ho

ol ,

ha vi

ng a

n un

sta bl

e jo

b re

co rd

, a nd

n ot

be

in g

em pl

oy ed

” (A

ss in

k et

a l.,

2 01

5,

p. 50

)

0. 15

* R

el at

io ns

hi p

“F ac

to rs

re la

tin g

to th

e na

tu re

a nd

q ua

lit y

of r

el at

io ns

hi ps

w ith

p rim

ar ily

p ee

rs , s

uc h

as h

av in

g de

lin qu

en t p

ee rs

, e xp

er ie

nc in

g pe

er r

ej ec

tio n,

b ei

ng a

g an

g m

em be

r, ha

vi ng

p oo

r re

la tio

ns hi

p w

ith p

ee rs

, a nd

de

vi an

t p ee

r a ss

oc ia

tio ns

”. (A

ss in

k et

a l.,

20

15 , p

.5 0)

0. 19

* A

lc oh

ol /d

ru g

ab us

e “M

ai nl

y fa

ct or

s re

la tin

g to

a lc

oh ol

an

d dr

ug a

bu se

”, (

A

ss in

k et

a l.,

2 01

5,

p. 50

)

0. 21

* Em

ot io

na l a

nd

be ha

vi ou

ra l p

ro bl

em s

“F ac

to rs

r el

at in

g to

in

te rn

al iz

in g

an d

ex te

rn al

iz in

g pr

ob le

m s”

. (A

ss in

k et

a l.,

2 01

5,

p. 50

)

0. 18

*

Co tt

le

et a

l. (2

00 1)

Fa m

ily p

ro bl

em s

“( e.

g. , p

oo r

re la

tio ns

hi ps

w ith

in th

e fa

m ily

)” , (

Co tt

le e

t a l.,

2 00

1, p

.3 78

)

0. 22

* Lo

w s

ch oo

l r ep

or t o

f a ch

ie ve

m en

ta,

b 0.

03 ns

H

av in

g de

lin qu

en t

pe er

s 0.

20 *

Su bs

ta nc

e ab

us e

0. 15

* N

on se

ve re

p at

ho lo

gy b

(e .g

., st

re ss

, a nx

ie ty

), (

Co tt

le e

t a l.,

2 00

1,

p. 37

8)

0. 30

*

Sc ot

t a nd

Br

ow n

(2 01

8)

Pr ob

le m

at ic

fa m

ily

ci rc

um st

an ce

s an

d pa

re nt

in g

“I na

de qu

at e

su pe

rv isi

on , d

iffi cu

lty

co nt

ro lli

ng b

eh av

io ur

, i na

pp ro

pr ia

te

di sc

ip lin

e, in

co ns

ist en

t p ar

en tin

g,

fa m

ily su

bs ta

nc e

ab us

e, fa

m ily

cr

im in

al h

ist or

y” (

Sc ot

t & B

ro w

n,

20 18

, p .9

36 )

0. 14

* (F

)0

.1 0*

(M

)

Ed uc

at io

n/ em

pl oy

m en

t pr

ob le

m s

“U ne

m pl

oy ed

/n ot

se ek

in g

em pl

oy m

en t”

, ( Sc

ot t &

B ro

w n,

2 01

8,

p. 93

6)

0. 25

* (F

)0

.2 1*

(M

)

A nt

is oc

ia l p

ee r

re la

ti on

sh ip

s “D

el in

qu en

t i nfl

ue nc

es , g

an g

af fil

ia te

d/

in vo

lv ed

” (S

co tt

& B

ro w

n, 2

01 8,

p .9

36 )

0. 13

* (F

)0

.2 0*

(M

)

Su bs

ta nc

e ab

us e

“C hr

on ic

d ru

g us

e,

ch ro

ni c

al co

ho l

us e”

( Sc

ot t &

Br

ow n,

2 01

8,

p. 93

6)

0. 05

* (F

)0

.0 8*

(M

)

M en

ta l h

ea lt

h “M

en ta

l h ea

lth w

as

co de

d as

m en

ta l h

ea lth

pr

ob le

m s o

r di

ag no

se s

(y es

o r

no )”

( Sc

ot t &

Br

ow n,

2 01

8, p

.9 36

)

0. 04

* (F

)0

.0 2ns

(M )

Ed uc

at io

n/ sc

ho ol

c on

ce rn

sa

“L ow

( ac

ad em

ic )

ac hi

ev em

en t,

tr ua

nc y

at sc

ho ol

, c ur

re nt

sc ho

ol p

ro bl

em s”

(

Sc ot

t & B

ro w

n, 2

01 8,

p .9

36 )

0. 13

* (M

)0

.1 0*

(F

)

N ot

e. M

= M

al es

; F =

Fe m

al es

. a =

O nl

y sc

ho ol

-s pe

ci fic

d im

en si

on s w

er e

as se

ss ed

in th

is v

ar ia

bl e.

b =

O nl

y th

is v

ar ia

bl e

fr om

th is

d im

en si

on w

as e

va lu

at ed

a cr

os s m

or e

th an

fi ve

st ud

ie s (

an d

no t e

xc lu

de d

fo r e

lig ib

ili ty

re

as on

s) . *

p <

.0 5;

n s =

no n-

si gn

ifi ca

nt p

re di

ct or

.

M. Basto-Pereira and D.P. Farrington

Aggression and Violent Behavior 65 (2022) 101761

9

Families have a primary role in socialization and social learning, and most developmental theories of offending have recognized the critical role of families, particularly parents (e.g., child rearing skills, parental supervision, caring families) in preventing versus promoting offending (Farrington, 2006). It is possible that family problems not only predict offending, but also play an important role as a potential cause of later persistence in crime. In this regard, a series of systematics reviews have shown important links between early family risk factors and school dropout (Gubbels et al., 2019), unemployment (Bunting et al., 2018), mental health problems, including addiction (Rasic et al., 2014), and inadequate interactions with peers (Groh et al., 2014).

As the multiple systems of which a youth is part are contaminated by psychosocial problems (e.g., delinquent peers, lack of parental super- vision, mental health problems), the risk of recidivism appears to in- crease. Therefore, while parental training and family support are suggested as key components of interventions that prevent offending in the first place, multisystemic approaches may be a more adequate approach for youths with histories of criminal behaviour.

The predictive ability of mental health dimensions for persistence in crime substantially changes across meta-analyses, which indicates that inside the broader concept of mental health problems, some diagnoses and psychopathological symptoms might be more or less important in predicting persistence in crime. Interestingly, non-severe pathology, which is focused on symptoms of anxiety, stress, and other general psychopathological symptoms, is the most important predictor of persistence in crime found across all meta-analyses (Cottle et al., 2001). In addition, the Assink et al. (2015) meta-analysis found emotional and behavioural problems as one of most important predictors. However, in the opposite direction, the dimension of mental health assessed by Scott and Brown (2018) had small effects for females and did not even reach statistical significance in predicting persistence in crime for males.

Most of the current studies and developmental and life-course the- ories (DLC) of offending have neglected the role of mental health vul- nerabilities as important explanations for criminal career development (for a review of DLC theories, see Farrington, 2006). It would be important to understand, for example, if specific psychopathologies linked to high vulnerability to stress or anxiety are important predictors of relapse among justice-involved youths, and why. For example, is this mediated by emotion regulation deficits? More research is needed. Also, the long-term neurological and psychosocial impact of alcohol and substance abuse on the development of youthful criminal careers is underexplored across developmental theories of offending.

In contrast, most of the DLC theories take into account family dy- namics, school/employment problems, and antisocial models as central causes of youth antisocial behaviour (e.g., Farrington, 2006; Laub & Sampson, 2001; Moffitt, 2018; Thornberry & Krohn, 2005). Nonethe- less, the way each one of those theories operationalizes each one of these constructs may vary (e.g., informal social control, attachment, social learning). Thus, a deeper understanding of how each one of these risk factors leads to the development of criminal behaviour is an important line for future research.

Only one meta-analysis addressed the gender-specific roles of each of the tested predictors across longitudinal studies addressed our research questions. The way gender (and ethnicity) shapes predictors from crime during development is one of the most underexplored topics across meta-analyses. The important findings from Scott and Brown's (2018) meta-analysis suggest that there are similar effect sizes for males and females in the most important predictors of recidivism, supporting the hypothesis of gender neutrality for global risk factors. More primary research is needed comparing risk factors for males and females in relation to offending and recidivism.

There is also a lack of meta-analyses studying the longitudinal impact of childhood biological and temperamental characteristics on later offending or recidivism. From the few meta-analyses addressing indi- vidual characteristics versus offending behaviour, the meta-analysis conducted by Portnoy and Farrington (2015) shows a low resting

heart rate (r = 0.07) as a statistically significant predictor of later offending. Also, the meta-analysis conducted by Cottle et al. (2001) showed the role of low verbal IQ and low full-scale IQ as predictors of recidivism across a limited number of longitudinal studies. Some meta- analyses, not specifically addressing longitudinal predictors of crime during the developmental period (and for that reason not included in this systematic review), suggest an important role of other individual characteristics in general offending, such as low self-control (Vazsonyi et al., 2017) or low cognitive and affective empathy (Van Langen et al., 2014). The role of many of these individual characteristics in predicting childhood or adolescence in later offending or persistence in crime is still underexplored.

4.2. Limitations

This systematic review is not free of limitations. First, it includes only meta-analyses addressing explanatory predictors of crime evaluated in the first 18 years of life. This decision allows us to focus our discussion on early predictors of offending, but at the same time adult factors promoting changes in criminal patterns later in life are neglected. Because behavioural predictors such as conduct disorder were excluded from consideration, our focus is on explanation rather than pure pre- diction. Second, this systematic review includes only meta-analyses addressing longitudinal predictors of crime, excluding overall effect sizes from cross-sectional studies, and our discussion is focused on predictors tested across five or more independent samples. This decision allows us to guarantee that predictors precede offending outcomes and to focus on predictors that are well tested across multiple studies, reducing the risk of bias in our conclusions. Nonetheless, this decision is not free from consequences, since it also substantially reduced the number of studies included.

Third, measures of association depend partly on the true association and partly on the methods of measuring the predictor and outcome variables. For example, the product moment correlation r is based on the assumptions that the variables are measured on equal-interval scales (like height and weight), that they are normally distributed, and that they are linearly related. Most variables in the social sciences violate these assumptions. Therefore, differences in r values may reflect dif- ferences in measurement methods rather than differences in the true underlying association (unless all variables are measured in the same way to make them comparable). The same problem applies when OR and d values are converted into r values; the conversion formulae are based on assumptions that may be violated by the nature of the vari- ables. Nevertheless, large differences in r values probably reflect real differences in predictive efficiency.

Lastly, this work includes and compares overall effect sizes from meta-analyses that include studies from different years and use different inclusion/exclusion criteria and analytic strategies; this may have introduced some bias in our conclusions. Also, it would be desirable in future meta-analyses to investigate which variables predicted outcomes after controlling for (independently of) other variables. Nevertheless, the most important findings of this work are replicated across different meta-analyses, despite the bias introduced by methodological discrep- ancies across studies.

4.3. Final conclusions

Family factors (parental supervision/parental warmth, family structure) are the most important childhood and/or adolescent pre- dictors of general offending, followed by child maltreatment. Among adolescents already involved in the justice system, there are five major predictors of persistence in crime across meta-analyses, namely, edu- cation/employment problems, delinquent peers, family problems, alcohol/drug abuse, and specific forms of mental health problems.

Our findings support the crucial role of programs working with families, particularly parents, with the aim to prevent offending in the

M. Basto-Pereira and D.P. Farrington

Aggression and Violent Behavior 65 (2022) 101761

10

first place (see e.g. Farrington, 2021). Among juvenile justice youths, there is a constellation of long-term problematic factors explaining persistence in crime. Programs to prevent recidivism should evaluate and intervene in each of the above-identified factors (e.g., school failure, psychopathology, families, relationships with peers, addiction prob- lems) that could cyclically create the perfect conditions to recidivate. Since many of those predictors might be avoided or attenuated by a healthy family environment, programs promoting desirable parenting and strengthening families should be the top policy priority.

Declaration of competing interest

None.

Acknowledgements

We are grateful for the reviewers' valuable comments. Thank you for your help in improving our manuscript.

References

*Assink, M., van der Put, C., Hoeve, M., de Vries, S., Stams, G., & Oort, F. (2015). Risk factors for persistent delinquent behaviour among juveniles: A meta-analytic review. Clinical Psychology Review, 42, 47–61. https://doi.org/10.1016/J.Cpr.2015.08.002

Basto-Pereira, M., Começanha, R., Ribeiro, S., & Maia, Â. (2015). Long-term predictors of crime desistance in juvenile delinquents: A systematic review of longitudinal studies. Aggression and Violent Behavior, 25, 332–342. https://doi.org/10.1016/j. avb.2015.09.012

Basto-Pereira, M., & Maia, Â. (2017). Juvenile delinquency, crime and social marginalization: Social and political implications. Wagon Lane, UK: Emerald.

Bernburg, J., Krohn, M., & Rivera, C. (2006). Official labeling, criminal embeddedness, and subsequent delinquency: A longitudinal test of labeling theory. Journal of Research in Crime & Delinquency, 43, 67–88. https://doi.org/10.1177/ 0022427805280068

*Braga, T., Gonçalves, L. C., Basto-Pereira, M., & Maia, Â. (2017). Unraveling the link between maltreatment and juvenile antisocial behavior: A meta-analysis of prospective longitudinal studies. Aggression and Violent Behavior, 33, 37–50. https:// doi.org/10.1016/j.avb.2017.01.006

*Braga, T., Cunha, O., & Maia, Â. (2018). The enduring effect of maltreatment on antisocial behaviour: A meta-analysis of longitudinal studies. Aggression and Violent Behaviour, 40, 91–100. https://doi.org/10.1016/J.Avb.2018.04.003

Bunting, L., Davidson, G., Mccartan, C., Hanratty, J., Bywaters, P., Mason, W., & Steils, N. (2018). The association between child maltreatment and adult poverty - a systematic review of longitudinal research. Child Abuse and Neglect, 77, 121–133. https://doi. org/10.1016/J.Chiabu.2017.12.022

Cicchetti, D. (2016). Socioemotional, personality, and biological development: Illustrations from a multilevel developmental psychopathology perspective on child maltreatment. Annual Review of Psychology, 67(1), 187–211. https://doi.org/ 10.1146/Annurev-Psych-122414-033259

*Cottle, C. C., Lee, R. J., & Heilbrun, K. (2001). The prediction of criminal recidivism in juveniles: A meta-analysis. Criminal Justice and Behaviour, 28(3), 367–394. https:// doi.org/10.1177/0093854801028003005

*Derzon, J. H. (2010). The correspondence of family features with problem, aggressive, criminal, and violent behaviour: A meta-analysis. Journal of Experimental Criminology, 6(3), 263–292. https://doi.org/10.1007/S11292-010-9098-0

Dishion, T. J., & Tipsord, J. M. (2011). Peer contagion in child and adolescent social and emotional development. Annual Review of Psychology, 62, 189–214. https://doi.org/ 10.1016/j.addbeh.2012.06.020

Farrington, D. P. (2003). Developmental and life-course criminology: Key theoretical and empirical issues-The 2002 Sutherland award address. Criminology; An Interdisciplinary Journal, 41(2), 221–225. https://doi.org/10.1111/J.1745- 9125.2003.Tb00987.X

Farrington, D. P. (2006). Building developmental and life-course theories of offending. In F. T. Cullen, J. P. Wright, & K. R. Blevins (Eds.), Taking stock: The status of criminological theory (pp. 335–366). Transaction Publishers.

Farrington, D. P. (2013). Longitudinal and experimental research in criminology. Crime and Justice, 42(1), 453–527. https://doi.org/10.1086/670396

Farrington, D. P. (2020). Childhood risk factors for criminal career duration: Comparisons with prevalence, onset, frequency and recidivism. Criminal Behaviour and Mental Health, 30, 159–171. https://doi.org/10.1002/cbm.2155

Farrington, D. P. (2021). The developmental evidence base: Prevention. In D. A. Crighton, & G. J. Towl (Eds.), Forensic psychology (3rd ed., pp. 263–293). Wiley.

Farrington, D. P., & Loeber, R. (1989). Relative improvement over chance (RIOC) and phi as measures of predictive efficiency and strength of association in 2 X 2 tables. Journal of Quantitative Criminology, 5, 201–213.

Farrington, D. P., Gaffney, H., & Ttofi, M. M. (2017). Systematic reviews of explanatory risk factors for violence, offending, and delinquency. Aggression and Violent Behaviour, 33, 24–36. https://doi.org/10.1016/J.Avb.2016.11.004

*Flanagan, I. M. L., Auty, K. M., & Farrington, D. P. (2019). Parental supervision and later offending: A systematic review of longitudinal studies. Aggression and Violent Behaviour, 47, 215–229. https://doi.org/10.1016/J.Avb.2019.06.003

Groh, A. M., Fearon, R. P., Bakermans-Kranenburg, M. J., Ijendoorn, V., Steele, R. D., & Roisman, G. I. (2014). The significance of attachment security for children's social competence with peers: A meta-analytic study. Attachment & Human Development, 16 (2), 103–136. https://doi.org/10.1080/14616734.2014.883636

Gubbels, J., Van Der Put, C. E., & Assink, M. (2019). Risk factors for school absenteeism and dropout: A meta-analytic review. Journal of Youth and Adolescence, 48(9), 1637–1667. https://doi.org/10.1007/S10964-019-01072-5

*Hoeve, M., Stams, G. J. J. M., van der Put, C. E., Dubas, J. S., van der Laan, P. H., & Gerris, J. R. M. (2012). A meta-analysis of attachment to parents and delinquency. Journal of Abnormal Child Psychology, 40(5), 771–785. https://doi.org/10.1007/ S10802-011-9608-1

Jackson, D., & Turner, R. (2017). Power analysis for random-effects meta-analysis. Research Synthesis Methods, 8(3), 290–302. https://doi.org/10.1002/Jrsm.1240

Jolliffe, D., Farrington, D. P., Piquero, A. R., Loeber, R., & Hill, K. G. (2017). Systematic review of early risk factors for life-course-persistent, adolescence-limited, and late- onset offenders in prospective longitudinal studies. Aggression and Violent Behaviour, 33, 15–23. https://doi.org/10.1016/J.Avb.2017.01.009

Kerig, P. K., & Becker, S. P. (2015). Early abuse and neglect as risk factors for the development of criminal and antisocial behaviour. In J. Morizot, & L. Kazemian (Eds.), The Development of Criminal and Antisocial Behaviour: Theoretical Foundations and Practical Applications (pp. 181–199). Springer.

Langevin, R., Marshall, C., & Kingsland, E. (2019). Intergenerational cycles of maltreatment: A scoping review of psychosocial risk and protective factors. Trauma, Violence & Abuse. https://doi.org/10.1177/1524838019870917, 1524838019870917.

Laub, J. H., & Sampson, R. J. (2001). Understanding desistance from crime. Crime and Justice, 28, 1–69.

Lee, V., & Hoaken, P. N. S. (2007). Cognition, emotion, and neurobiological development: Mediating the relation between maltreatment and aggression. Child Maltreatment, 12(3), 281–298. https://doi.org/10.1177/1077559507303778

*Leschied, A. W., Chiodo, D., Nowicki, E., & Rodger, S. (2008). Childhood predictors of adult criminality: A meta-analysis drawn from the prospective longitudinal literature. Canadian Journal of Criminology and Criminal Justice, 50(4), 435–467. https://doi.org/10.3138/Cjccj.50.4.435

Loeber, R. (2019). Developmental pathways to conduct problems and serious forms of delinquency. In D. P. Farrington, L. Kazemian, & A. R. Piquero (Eds.), The Oxford handbook of developmental and life-course criminology (pp. 159–172). Oxford University Press.

Loeber, R., & Farrington, D. P. (2012). From juvenile delinquency to adult crime: Criminal careers, justice policy. New York, NY: Oxford University Press.

Loeber, R., & Le Blanc, M. (1990). Toward a developmental criminology. In M. Tonry (Ed.), Crime and justice: A review of research (Vol. 12, pp. 375–473). University of Chicago Press.

Moffitt, T. E. (1993). Adolescence-limited and life-course-persistent antisocial behaviour: A developmental taxonomy. Psychological Review, 100(4), 674–701. https://doi.org/ 10.1037/0033-295X.100.4.674

Moffitt, T. E. (2018). Male antisocial behaviour in adolescence and beyond. Nature Human Behaviour, 2(3), 177–186. https://doi.org/10.1038/S41562-018-0309-4

Moher, D., Liberati, A., Tetzlaff, J., & Altman, D. G. (2009). Preferred reporting items for systematic reviews and meta-analyses: The PRISMA statement. PLoS Medicine, 6(7), Article e1000097. https://doi.org/10.1371/journal.pmed.1000097

Murray, J., & Farrington, D. P. (2010). Risk factors for conduct disorder and delinquency: Key findings from longitudinal studies. Canadian Journal of Psychiatry. Revue Canadienne de Psychiatrie, 55(10), 633–642. https://doi.org/10.1177/ 070674371005501003

Orben, A., & Przybylski, A. K. (2019). The association between adolescent well-being and digital technology use. Nature Human Behaviour, 3(2), 73–182. https://doi.org/ 10.1038/S41562-018-0506-1

Papalia, N., Mann, E., & Ogloff, J. R. P. (2020). Child sexual abuse and risk of revictimization: Impact of child demographics, sexual abuse characteristics, and psychiatric disorders. Child Maltreatment, 26(1), 74–86. https://doi.org/10.1177/ 1077559520932665

*Portnoy, J., & Farrington, D. P. (2015). Resting heart rate and antisocial behaviour: An updated systematic review and meta-analysis. Aggression and Violent Behaviour, 22, 33–45. https://doi.org/10.1016/J.Avb.2015.02.004

Rasic, D., Hajek, T., Alda, M., & Uher, R. (2014). Risk of mental illness in offspring of parents with schizophrenia, bipolar disorder, and major depressive disorder: A meta- analysis of family high-risk studies. Schizophrenia Bulletin, 40(1), 28–38. https://doi. org/10.1093/Schbul/Sbt114

*Reaves, S., Mcmahon, S. D., Duffy, S. N., & Ruiz, L. (2018). The test of time: A meta- analytic review of the relation between school climate and problem behaviour. Aggression and Violent Behaviour, 39, 100–108. https://doi.org/10.1016/J. Avb.2018.01.006

*Scott, T., & Brown, S. L. (2018). Risks, strengths, gender, and recidivism among justice- involved youth: A meta-analysis. Journal of Consulting and Clinical Psychology, 86 (11), 931–945. https://doi.org/10.1037/Ccp0000343

Siegel, L. J. (2015). Criminology: Theories, patterns, and typologies (12th ed.). Cengage Learning.

*Spruit, A., van Vugt, E., van der Put, C., van der Stouwe, T., & Stams, G. J. (2016). Sports participation and juvenile delinquency: A meta-analytic review. Journal of Youth and Adolescence, 45(4), 655–671. https://doi.org/10.1007/S10964-015-0389-7

Stith, S. M., Liu, T., Davies, L. C., Boykin, E. L., Alder, M. C., Harris, J. M., Som, A., Mcpherson, M., & Dees, J. E. M. E. G. (2009). Risk factors in child maltreatment: A

M. Basto-Pereira and D.P. Farrington

Aggression and Violent Behavior 65 (2022) 101761

11

meta-analytic review of the literature. Aggression and Violent Behaviour, 14(1), 13–29. https://doi.org/10.1016/J.Avb.2006.03.006

Sullivan, G. M., & Feinn, R. (2012). Using effect size-or why the p value is not enough. Journal of Graduate Medical Education, 4(3), 279–282. https://doi.org/10.4300/ Jgme-D-12-00156.1

Szucs, D., & Ioannidis, J. P. A. (2017). Empirical assessment of published effect sizes and power in the recent cognitive neuroscience and psychology literature. PLoS Biology, 15(3), E2000797. https://doi.org/10.1371/Journal.Pbio.2000797

Thornberry, T. P., & Krohn, M. D. (2005). Applying interactional theory to the explanation of continuity and change in antisocial behaviour. In D. P. Farrington (Ed.), Integrated developmental and life course theories of offending advances in criminological theory (pp. 183–209). Transaction.

Tremblay, R. E., Vitaro, F., Nagin, D., Pagani, L., & Séguin, J. R. (2003). The Montreal longitudinal and experimental study: Rediscovering the power of descriptions. In T. P. Thornberry, & M. D. Krohn (Eds.), Taking stock of delinquency: An overview of findings from contemporary longitudinal studies (pp. 205–254). Kluwer Academic/ Plenum Publishers.

*Ttofi, M. M., Farrington, D. P., Piquero, A. R., Lösel, F., Delisi, M., & Murray, J. (2016). Intelligence as a protective factor against offending: A meta-analytic review of prospective longitudinal studies. Journal of Criminal Justice, 45, 4–18. https://doi. org/10.1016/J.Jcrimjus.2016.02.003

Van Langen, M. A. M., Wissink, I. B., Van Vugt, E. S., Van Der Stouwe, T., & Stams, G. J. J. M. (2014). The relation between empathy and offending: A meta-

analysis. Aggression and Violent Behaviour, 19(2), 179–189. https://doi.org/10.1016/ J.Avb.2014.02.003

Vazsonyi, A. T., Mikuška, J., & Kelley, E. L. (2017). It's time: A meta-analysis on the self- control-deviance link. Journal of Criminal Justice, 48, 48–63. https://doi.org/ 10.1016/J.Jcrimjus.2016.10.001

Wasserstein, R. L., & Lazar, N. A. (2016). The ASA statement on p-values: Context, process, and purpose. The American Statistician, 70(2), 129–133. https://doi.org/ 10.1080/00031305.2016.1154108

Wikstrom, P., Oberwittler, D., Treiber, K., & Hardie, B. (2012). Breaking rules: The social and situational dynamics of young people's urban crime. Oxford University Press.

*Wilson, H. W., Stover, C. S., & Berkowitz, S. J. (2009). Research review: The relationship between childhood violence exposure and juvenile antisocial behaviour: A meta- analytic review. Journal of Child Psychology and Psychiatry, and Allied Disciplines, 50 (7), 769–779. https://doi.org/10.1111/J.1469-7610.2008.01974.X

Woods, D. (1988). Coping with complexity: The psychology of human behaviour in complex systems. In L. P. Goodstrein, H. B. Andersen, & S. E. Olsen (Eds.), Tasks, Errors, and Mental Models (pp. 129–148). Taylor & Francis Inc.

Wormith, J. S. (2011). The legacy of D. A. Andrews in the field of criminal justice: How theory and research can change policy and practice. International Journal of Forensic Mental Health, 10(2), 78–82. https://doi.org/10.1080/14999013.2011.577138

Zane, S. N. (2021). Have racial and ethnic disparities in juvenile justice declined over time? An empirical assessment of the DMC mandate. Youth Violence and Juvenile Justice, 19(2), 163–185. https://doi.org/10.1177/1541204020962163

M. Basto-Pereira and D.P. Farrington