1 Faculty of Social and Behavioural Sciences
Graduate School of Childhood Development and Education
Early onset substance use and later
delinquent behaviour – a multilevel
meta-analysis
Research Master Educational Sciences Thesis 2
Student: Bennie Mooren Supervisor: dr. H.E. Creemers
Reviewers: dr. M. Hoeve and dr. J.J. Asscher Date: 15-8-2016
2
Abstract
There is a body of work on the relationship between substance use and delinquent
behaviour. However, there is no consensus on the longitudinal relationship between early onset substance use and delinquency. Therefore, we conducted a multilevel meta-analysis to examine this association. Furthermore, we determined to what extent the association was moderated by delinquency characteristics, early onset substance use characteristics, sample descriptors and research design characteristics. A total of 14 studies (N = 46212) reporting on 131 effect sizes were included. The 3-level meta-analysis showed a small but significant association r = .163. The period in which the delinquency was exhibited was a significant moderator of this association. The association was weaker for
adolescents older than 17 years compared to adolescents from the ages 15 or 16 years. Furthermore, a stronger association was found for marijuana initiates compared to alcohol initiates. In addition, higher attrition rates yielded a lower association between early onset substance use and delinquency. Likewise, a longer follow-up period
negatively influenced the longitudinal relationship between early onset substance use and delinquency. Implications of the present results for future research on this association and the prevention and intervention workers in this field are discussed.
3 Preface
Hereby, I will enlighten the thesis process and in particular my own role and the role of my co-author and other contributors. First, I started reading about the longitudinal relationship between substance use and delinquent behaviour, searching for a possibility to make a valuable contribution to the existing body of work. Finding myself struggling with the contradicting results from study, I decided to conduct a meta-analysis. Dr. Hanneke Creemers helped me with constructing and demarcate the research questions. Since I had no previous experience in conducting an extensive database search, drs. Janneke Staaks guided me in the computerized search and constructing the search strings. In order to acquaint myself with meta-analyses, I read several books and articles on this subject. Furthermore, I practiced with an example dataset following the manual written by drs. Mark Assink and Carlijn Wibbelink MSc. When I got stuck or had questions regarding the analysis in the computer program R, I could turn to drs. Mark Assink. I performed all the analyses by myself. The writing of this thesis was also done by me, but my supervisor dr. Hanneke Creemers provided me with extensive feedback. Not only on my academic writing, but also provided me guidelines and food for thought for the interpretation of my results. I would therefore thank all contributors for their support and guidance during the process of conducting and writing this thesis.
4
Substance use, and the use of alcohol, tobacco and cannabis in particular, is a common part of the development of adolescents in Western societies. Among adolescents, cannabis is the most widely used illicit drug in Western societies (Degenhardt et al., 2008), with onset of use peaking in middle and late adolescence (Vega et al., 2002). Although cannabis is the most widely used illicit drug, alcohol is the most consumed drug in countries across the world (Degenhardt et al., 2008). By the age of 15 years more than half of the European adolescents started to drink alcohol. Similar results have been found for the use of tobacco. Compared to Western countries, youth from Asian and African areas consume less often tobacco, alcohol or cannabis (Degenhardt et al., 2008). Hallucinogens, opioids and stimulant (e.g. ecstasy and cocaine) are used in very small percentages (0-1%) and use often starts in late adolescence or young adulthood (De Looze et al., 2014; Degenhardt, Lynskey, & Hall, 2000; Johnson & Gerstein, 1998).
Although substance use seems to become more and more normative during adolescence, particularly early onset substance use has been associated with various adverse outcomes. Definitions for early onset substance use range from substance use before the age of 12 to substance use before age 16 (Fergusson & Horwood, 1997; Kaplow, Curran, & Dodge, 2002). Early onset tobacco use has been related to an increased risk of cannabis use (Creemers et al., 2009; McCambridge & Strang, 2005), and has been associated with lower academic achievement and behavioural problems (Ellickson, Tucker, & Klein, 2001). Early onset cannabis use has been shown to be related to illicit drug use (Fergusson, Boden, & Horwood, 2008), poor school
5
Jacobson, & Kendler, 2009). The combination of early onset alcohol and cannabis use has been associated with elevated risk of having a criminal justice record and developing substance use dependence in adulthood (Green et al., 2015). Research has indicated that the rate of drug-related problems is peaking among those who initiated drug use at age 12 or younger, and decreases with increasing age of initiation (Anthony & Petronis, 1995). Furthermore, Anthony and Petronis (1995) stated that when children start using
substances before the age of 12 years, they have a 68% risk of developing a drug problem within seven years, while this risk is 43% when children initiated drug use at age 16 (Anthony & Petronis, 1995). Early onset of tobacco, alcohol, and marijuana use all have been found to be related with higher levels of mental health problems, such as depression (Pang, Farrahi, Glazier, Sussman, & Leventhal, 2014), suicidality (Miller et al., 2011), and conduct problems (Creemers et al., 2009).
Although early onset substance use has been associated with delinquent
behaviour, there is no consensus yet on the longitudinal relationship between early onset substance use and delinquent behaviour. Longitudinal research on this subject can provide insights in developmental patterns relating to early onset substance use and delinquency. Some studies found a significant positive association between early onset substance use and later delinquent behaviour (Ellickson, Tucker, Klein, & Saner, 2004; Gordon, Kinlock, & Battjes, 2004; White, Loeber, Stouthamer-Loeber, & Farrington, 1999). For instance, Ellickson and colleagues (2004) found that marijuana initiates in grade 7, 8, or 9 were more likely to exhibit delinquent behaviours in grade 10 compared to children who did not initiate marijuana use in grade 9 or earlier. In line with this finding, Gordon, Kinlock and Battjes (2004) found that early onset of substance use was
6
associated with higher levels of criminal involvement in a sample of adolescents that received an outpatient substance abuse treatment. However, other studies did not find a significant association (Becker et al., 2012; Hemphill et al., 2014; Hunter, Miles,
Pedersen, Ewing, & D’Amico, 2014; Marcus & Jamison II, 2013). Becker and colleagues (2012) found no prospective association between early onset alcohol use and delinquent behaviour over time. Similarly, Hunter and colleagues (2014) did not find significant associations between early onset marijuana and alcohol use and future reports of delinquency in a sample of 193 adolescents.
A meta-analysis on this longitudinal association is relevant for several reasons. First, research regarding this association shows mixed results. A meta-analysis could provide further insights in the longitudinal relationship between early onset substance use and later delinquent behaviour. Second, differences in study findings are often related to methodological differences among the included studies (Lipsey & Wilson, 2001). However, little is known about factors moderating the association between early onset substance use and delinquency. For example, it is yet unclear whether early onset substance use, defined as substance use before the age of 16 years (Anthony and
Petronis,1995), is differentially associated to specific forms of delinquency. In addition, it is not clear to what extent the use of specific substances (i.e. alcohol, tobacco or
marijuana use), when initiated early, is associated to delinquent behaviour later in life. We focus on alcohol, tobacco and cannabis, because other substances are only consumed by a small amount of the adolescents before the age of 16. Furthermore, some studies use official records of delinquency while others use self-reports (Pardini, 2006). Therefore, it is interesting to test if the relationship between early onset substance use and delinquency
7
is moderated by the type of measurement of delinquency. Additionally, research comparing boys and girls showed that prospective associations for substance use on delinquency were found for boys, but not for girls (Mason & Windle, 2002). Testing whether the relationship between early onset substance use and subsequent delinquent behaviour is moderated by gender, furthers our understanding in the generalisability of this association across gender. Lastly, ethnic minorities are often over-represented in crime statistics and delinquency research (Carson & Esbensen, 2014; Hawkins, Laub, Lauritsen, & Cothern, 2000). Yet, it is still unclear whether the relationship between early onset substance use and delinquent behaviour is different for juveniles from ethnic
minority groups. A meta-analytical approach can provide a better overview of the
influence of possible moderators than empirical studies (Fagard, Staessen, & Thijs, 1996; Ioannidis & Lau, 1999).
Therefore, the fist aim of this study was to determine the prospective association between early onset substance use and later delinquent behaviour. A second aim was to investigate whether the relationship between early onset substance use and delinquency is moderated by delinquency characteristics, substance use characteristics, sample
descriptors and research design descriptors. Determining the prospective association between early onset substance use and delinquent behaviour could provide further understanding of this relationship for children, parents, prevention and intervention workers, and scientists. Identifying possible moderators of this association could help developing and improving existing prevention and intervention programs for substance use and delinquency.
8
Sample of studies
For the selection of studies, several inclusion criteria were formulated. First, studies were eligible if their outcome measure was delinquency. Second, the studies had to focus on early onset substance use, specifically alcohol, tobacco and/or cannabis, before the age of 16 years (Anthony & Petronis, 1995). Studies reporting on the use of hallucinogens, opioids and stimulants, or on a composite measure of substance use including illicit substances other than cannabis, were not included in the present study. Third, only studies with a prospective or longitudinal design were included. Fourth, only studies that provided results on the bivariate association between early onset substance use and delinquency, or enough statistics to calculate this association, were included. Until February 2nd 2016, articles, book chapters, dissertations, and reports were searched in the following three electronic databases: PsycINFO, Web of Science and Medline. Five key concepts were made for the computerized search 1) population, 2) substance use, 3) delinquency, 4) longitudinal, and 5) early onset.
The following combinations of the five components were used in this search strategy: (“school age” OR “adolescence” OR “childhood development” OR “adolescent development” OR “juvenile”) AND (“addiction” OR “alcohol” OR “cannabis” OR “drug (ab)use” OR “hashish” OR “marijuana” OR “tobacco” OR “smoking” OR “substance use”) AND (“assault” OR “convict*” OR “crime” OR “criminal behavior” OR “delinq*” OR “gang” OR “incarcerat*” OR “homicide” OR “juvenile delinquency” OR “offend*” OR “violent crime”) AND (“follow up” OR “longitudinal” OR “prospective” OR
“systematic review” OR “meta-analy*”) AND (“early experience” OR “early onset” OR “childhood onset” OR “initial use”). To determine whether the retrieved studies could be
9
included in this meta-analysis, titles, abstracts, and if necessary, full article texts were read.
Additional to the computerized search, reference lists from primary studies, reviews and meta-analysis were checked to find relevant studies. Finally, several authors in the field were contacted to ask whether they had suggestions for studies that would be suitable for inclusion in this study. In total, the literature search strategy yielded 1896 studies. After screening of these studies, 41studies were selected based on title and abstract. Of those 41 studies, 19 studies reporting on 14 unique samples met the inclusion criteria and were eligible for further analysis (see Figure 1 for a flow chart of the search results and Table 1 for an overview of included studies and their characteristics).
Missing data
It can be challenging to detect and retrieve all the relevant studies concerning the research question, when conducting a meta-analysis. Furthermore, primary studies that report non-significant results are less often published compared to articles that report non-significant results, which is called the “file drawer problem” (Rosenthal, 1979). One method to handle the file drawer problem is to ask authors in the field about unpublished studies, doctoral dissertations, theses and studies that did not yield significant results. Another method is calculating the fail-safe N (Rosenthal, 1979). A third method to investigate the possibility of missing data, due to limitations in the search strategy, publication bias, or other causes is the funnel-plot-based trim and fill method (Duval & Tweedie, 2000a, 2000b). This can be done by using the “trimfill” function from the metafor package (Viechtbauer, 2010) in the R environment (R Core Team, 2016). The trim and fill method
10
uses the existing effect sizes to impute missing effect sizes, which then can be used to restore an asymmetric funnel plot into a symmetric funnel plot (Duval & Tweedie, 2000a, 2000b).
Coding of studies
Following guidelines of Lipsey and Wilson (2001) a coding system was
developed. The coding system was split up in two domains: study descriptors and effect sizes. The study descriptors provide information on study-level, while effect size
information provides information on the strength of the longitudinal relationship between substance use and delinquency. Study descriptors were divided into delinquency
characteristics, early onset substance use characteristics, sample descriptors, and research design descriptors.
For delinquency characteristics the following constructs were coded: mean age of delinquency measurement and the developmental period in which delinquency was measured (early adolescence (13-14y), middle adolescence(15-16y), late adolescence(17-18y) and adulthood (19+)). Furthermore, type of delinquency was categorised following Puzzanchera (2014) in violent crime (i.e., murder, nonnegligent manslaughter, forcible rape, robbery, aggravated assault), property crime (i.e., burglary, larceny-theft, motor vehicle theft, arson), and non-indexed crime (all other criminal behaviours). Studies that measured an overall measure of delinquency were not coded, since no clear difference in behaviours from the other categories could be established.
In addition, delinquency severity was coded (non-severe and severe). Severe delinquent behaviours were: murder, manslaughter, assault (with or without weapon),
11
kidnapping, torture, mayhem, rape/ sodomy/ unlawful sexual penetration, sexual abuse, aggravated robbery, possession and/or use of a firearm, gang involvement, and
aggravated burglary (Torbet, Gable, Montgomery, & Hurst, 1996). Similar to the type of delinquency, studies that measured an overall measure of delinquency were not coded in the severity category. All other criminal behaviours were coded as the rest category (non-severe delinquency).
Several items were coded for early onset substance use characteristics. First, the age of substance initiation was coded, using the mean age of the sample at the first
assessment of substance use. Several studies did not assess initiation of substance use at a specific measurement wave, but asked the participants whether or not they had initiated substance use before a specific age. For these studies, age of substance initiation was coded as the specific age used in the study.
In addition, type of substance (alcohol, tobacco, marijuana or a composite of those) was coded. There were no combinations of substances present in the final sample. Additionally, the level of substance use (experimental use, regular use, or frequent use) was coded. Experimental users were scored as using is in the past month or less frequent. Regular users were scored as using in the past week or until month. Frequent users were scored as more than once a week or when they reported that they used in frequently in a questionnaire.
Within sample descriptors, percentage of males in the sample, percentage of white participants, the attrition rate, and publication year were coded. Research design
descriptors consisted of the type of measurement of delinquency (i.e. self-reports, parental reports, or official records), and time to follow-up (in months). Additionally, to
12
assess the methodological quality of the included studies, a checklist was created based on the guidelines of Hayden, Windt, Cartwright, Côté and Bombardier (2013).
The relevant effect sizes from the included studies were numbered and coded. Relevant effect sizes were the ones that report on the relationship between early onset substance use and delinquent behaviour. When available, corresponding standard errors or confidence intervals for the effect sizes were coded.
Data analysis
To express the association between early onset substance use and subsequent delinquent behaviour all effect sizes were transformed to Pearson’s r, by using the formulas of Lipsey and Wilson (2001), Rosenthal (1994), and Cooper, Hedges, and Valentine (2009). All correlations were Fisher’s Z transformed to approximate a normal sampling distribution (Lipsey & Wilson, 2001). Usually studies report multiple effect sizes for an outcome measure, delinquency in this case (Hox, 2010). In order to deal with dependency of study results, a multilevel random effects model was used for the
calculation of combined effect sizes for the analyses (Hox, 2010; Van Den Noortgate & Onghena, 2003). The maximum likelihood multilevel approach is in general superior to the fixed-effects approach used in traditional meta-analysis (Van Den Noortgate & Onghena, 2003). A three level meta-analytical model was used to analyse the data, modelling three sources of variation: sampling variance of the observed effect sizes (Level 1), variance between effect sizes of the same study (Level 2), and variance between studies (Level 3) (Cheung, 2014). Using a three-level approach ensures that effect sizes extracted from the same study (i.e., dependent effect sizes) can be modelled. This way, all information provided by primary studies can be used, and maximum
13
statistical power can be achieved. For the statistical analyses the function “rma.mv” of the metafor package (Viechtbauer, 2010) in the R environment (R Core Team, 2016) was used. The syntax was written so that the three sources of variance were modelled
following the manual by Assink and Wibbelink (2015). First, an overall effect size was calculated. Second, the heterogeneity of the effect size was checked. After establishing significant variance within and between studies, moderator analyses were first conducted for each moderator separately in order to identify characteristics that might explain these differences. Second, moderators found to be significant at an alpha-level of .05 were examined simultaneously to address confounding.
Results
Descriptive statistics, central tendency and variability, and assessment of missing data
The present study included 19manuscripts, describing 14 studies (k) published between 1997 to 2016. The total sample consisted of N = 46212 participants, and the size of the samples described in the included studies ranged from 359 to 11064 participants. The mean age of the participants at the baseline measurement was 14.40. Studies were conducted in the USA (k = 11), Australia or New Zealand (k = 2), and Europe (k = 1). In total, the coded studies produced 131 separate effect sizes, each reflecting the association between early onset substance use and subsequent delinquent behaviour.
As presented in Table 2, the overall effect size of the association between early onset substance use and later delinquent behaviour was significant but small (r = .163), based on criteria for the interpretation of effect sizes as formulated by Cohen (1988). The results of the likelihood-ratio tests showed that there was significant variance between effect sizes from the same study (i.e. level 2 variance), and that there was significant variance between the effect sizes from different studies (i.e., level 3 variance; see Table
14
1). This variability stresses the importance of conducting moderator analyses to reveal which moderators account for the variability.
The computation of the fail-safe N showed that 141602 studies with mean effect sizes of zero needed to be added to reduce the combined significance level to an alpha level of .05. The fail-safe N exceeds Rosenthal’s benchmark of 105 (5n +10), suggesting that the findings are robust to the threat that excluded studies might have yielded a non-significant effect. However, the fail-safe N does not concentrate on the size of the estimated effect, but rather whether p-values reach a particular threshold. This method is therefore not considered the most appropriate manner to assess publication bias (Higgins & Green, 2011).
The asymmetrical distribution of the effect sizes, obtained by the trim and fill analysis suggested that bias was present in the sample of studies that was included in the present study (see Figure 2). The analysis suggested seven observed effect sizes to be missing on the left side (negative findings) and no effect sizes to be missing on the right side (positive findings; see Figure 2). Therefore, a “corrected” overall effect was
calculated for the association between early onset substance use and later delinquent behaviour (see Table 2). After the trim and fill analysis, the overall effect was
non-significant, r = .040 (SE = 0.048), t(137) = 0.843, p = .401. The results of the trim and fill method suggest that, in case of a true publication bias, the population effect would be smaller than the effect estimated based on the effect sizes reported in the literature, that is, a non-significant association between early onset substance use and later delinquent behaviour.
15
Results of the separate moderator analyses are described below and presented in Table 3, where moderators are classified into delinquency characteristics, early onset substance use characteristics, sample descriptors, and research design descriptors.
First, moderation by delinquency characteristics was assessed. No significant differences were found when we focussed on violent crime, property crime and non-index categories, suggesting that early onset substance use is not differently associated to types of delinquency. Likewise, no significant moderator effects were found for the severity of delinquency (i.e., severe v. non-severe), and for the age at which delinquency was assessed. When the age of delinquency was categorised in the developmental periods early adolescence (13-14 years), middle adolescence (15-16 years), late adolescence (17-18 years), and adulthood (19+), significant differences were found with the omnibus test. Since delinquency peaks at middle adolescence (Elliott, Huizinga, & Menard, 2012), this developmental period was chosen as the reference category. Follow-up contrasts showed no significant difference for the early adolescence group compared to the
mid-adolescence group. However, significant smaller associations were found for late adolescence (t(117) = -3.861, p < .001) and adulthood (t(117) = -3.092, p = .002), suggesting that the relationship between early onset substance use and delinquent behaviour declines after age 17.
Second, moderation by early onset substance use characteristics was assessed. Only substance type was a significant moderator. Stronger associations were found for studies looking only at the marijuana initiates compared to studies focussing on only alcohol (t(108) = 3.047, p = .003). There were no differences between studies that focused on alcohol and studies that focused on tobacco. Furthermore, no significant
16
differences were found for the age at first assessment of substance use, and for the level of substance use, indicating that the age at which substance use was measured does not influence the relationship with delinquency. Likewise, the association between early onset substance use and later delinquent behaviour is not determined by the fact that an adolescent is an experimenter, regular user or a frequent user.
Third, moderation by sample descriptors was assessed. Publication year, percentage of males in the sample and percentage of white participants in the sample were not found to be significant moderators. However, attrition was a significant moderator. In studies with higher attrition rates the association between early onset substance use and delinquency was smaller (t(117) = -3.265, p < .001).
Fourth, assessment of moderation by research design characteristics yielded one significant moderator. Studies with a larger period between the onset of substance use and measurement of delinquency found weaker associations between early onset substance use and delinquency (t(129) = -2.964, p = .004). Type of measurement of delinquency did not influence the association between early onset substance use and subsequent delinquency.
Attrition rate, time to follow up, and substance type were combined into one multivariate moderator analysis to address confounding (F(4,94) = 8.658, p < .001). Attrition rate, (t(94) = -2.938, p = .004) and type of substance (marijuana only use versus alcohol only use) remained the only significant moderators (t(94) = 2.473, p = .015)
Discussion
The present study aimed to provide insights in the longitudinal association between early onset substance use and subsequent delinquent behaviour, using a
meta-17
analytical approach. Furthermore, we examined whether this association was moderated by delinquency characteristics, early onset substance use characteristics, sample
descriptors, and research design descriptors.
Based on the results from 19 studies containing 131 effect sizes, a significant overall prospective association between early onset substance use and subsequent
delinquent behaviour of r = .163 was found. This indicates that early onset substance use is, to a small extent, associated with later delinquent behaviour. This is an interesting finding, since cross-sectional studies report that there is an established association
between substance use and delinquency, with Pearson’s r ranging from .35 to .80 (Arthur, Hawkins, Pollard, Catalano, & Baglioni, 2002; Barnes, Welte, & Hoffman, 2002;
Flannery, Williams, & Vazsonyi, 1999; Kessler, Davis, & Kendler, 1997; Wasserman, McReynolds, Lucas, Fishher, & Santos, 2002). This indicates that the co-occurrence of substance use and delinquent behaviour is common, but the prospective link is small. It is possible that this discrepancy in the strength of the association is due to the
interaction between genes and environment. Research on adolescent twins between the ages 13-18 years showed that genetics as well as environment have an influence on both conduct disorder symptoms and substance dependence vulnerability (Button et al., 2006). The phenotypic correlation between conduct disorder and dependence vulnerability was of .489, with higher correlations for monozygotic twins than for dizygotic adolescent twins, indicating the importance of genetics on the association. The bivariate analysis showed that the two traits shared half of their genetic influence in common, as well as all their shared environmental influences (Button et al., 2006). They state that conduct disorder symptoms and dependence vulnerability share genetic influences and that genes
18
and shared environment contribute to the co-occurrence of the two traits. The longitudinal relationship between conduct disorder and substance dependence vulnerability is therefore also influenced by the interaction between genes and
environment. It could be the case that this interaction of genes and environment is smaller over a longer period of time, and thereby explaining the smaller longitudinal association we found compared to reported cross-sectional associations.
Moderator analyses were carried out to identify factors that influence the prospective association between early onset substance use and subsequent delinquent behaviour. First, moderator analyses were carried out for delinquency characteristics. These analyses showed that the prospective association between early onset substance use and delinquency was not moderated by the type of delinquency and severity of delinquency. Both categories, non-severe and severe delinquency, had sufficient effect sizes, respectively 57 and 33, to find significant results if they were present. Other risk factors might interfere in the association between early onset substance use and
delinquency and our categorisation of type of delinquency and severity of delinquency. Early onset substance use does not seem to be associated to what kind delinquency the adolescent is exhibiting. It could be the case that the adolescent is using substances and portraying delinquent behaviour for the same reason, namely sensation seeking (Lynne-Landsman, Graber, Nichols, & Botvin, 2011a). Sensation seeking could cause the young adolescent to initiate substances and the same sensation seeking behaviour could result later in delinquent behaviours. We did find a significant moderator effect for the period in which delinquency was measured. The association between early onset substance use and delinquent behaviour is smaller for studies looking at late adolescence or adulthood
19
compared to mid-adolescence. Our findings suggest that early onset substance is especially related to the adolescence-limited trajectory of the developmental taxonomy theory of Moffit (1993). Additionally, mid-adolescence is the age category in which adolescents are more likely to exhibit novelty and sensation seeking behaviour, including trying substances and delinquency (Lynne-Landsman et al., 2011a).
Second, moderator analyses for early onset substance use characteristics showed an effect for substance type. Studies that focus on early onset marijuana use report a stronger association than studies that report on early onset alcohol use. Alcohol
consumption by young adolescents is more accepted in Western societies compared to early marijuana use. These marijuana initiates are often a more deviant group, with more risk factors for exhibiting delinquent behaviours, such as having deviant peers (Brook, Gordon, Brook, & Brook, 1989; Lynne-Landsman, Graber, Nichols, & Botvin, 2011b). However, this finding needs to be interpreted with caution since the studies focussed on early onset marijuana use, but did not necessarily report on alcohol use or other deviant behaviours at time of marijuana use initiation. Concurrent alcohol use or deviant
behaviour at the time of marijuana initiation could also be predictors of later delinquency. Interestingly, we did not find significant results for level of substance use, indicating that there were no differences between experimental users, regular users and frequent users. This finding suggests that the frequency of use is not so much of importance in the association, but the early onset substance use itself is of importance for later delinquent behaviour. On a methodological note, this finding might be explained by the small amount of effect sizes (only nine) for the high usage group.
20
Third, moderator analyses for sample descriptors showed that the effects of publication year, percentage of males and percentage of white participants in the sample were not significant. Our finding that there are no differences between boys and girls in the relation between early onset substance use and delinquent behaviour contrasts with the study of Mason and Windle (2002). However, our study is in line with recent research that also did not find gender differences (Dijkstra et al., 2015). We did not find a
significant moderating effect for ethnicity on the association between early onset substance use and delinquent behaviour. Even though that there is a body of work showing that ethnic minorities are overrepresented in delinquent behaviours, our finding suggests that early onset substance use is not differently associated with delinquency for natives or ethnic minorities. We did find a moderator effect of attrition rate. Attrition is one of the major methodological problems in longitudinal studies (Deeg, 2002;
Gustavson, von Soest, Karevold, & Røysamb, 2012). In the present study, the association between early onset substance use and subsequent delinquent behaviour decreased when attrition rates increased. Previous research showed that the most delinquent adolescents are the ones most likely to drop out from the study (Loeber & Farrington, 1994). This could also negatively influence the association between early onset substance use and subsequent delinquency. It is therefore for future research important to set up the study in such a way that retention rates stay as high as possible.
Fourth, moderator analyses for research design descriptors showed that the prospective association between early onset substance use and delinquent behaviour is moderated by the time between the onset of use and the measurement of delinquent behaviour. A longer time between the measurements is associated with a decline in the
21
longitudinal association between early onset substance use and delinquent behaviour. This could be related to the attrition finding, in the sense that adolescents have more opportunities to drop out the study when follow-up measurements lie far apart. An examination of the bivariate association of time between substance use onset and
delinquency measurement and the attrition, showing a small but significant correlation r = .254, p <.01, seems to confirm this explanation. In addition, the developmental period also seems to be related with the follow-up period: the longer the follow-up period the more likely adolescents are in their late adolescence or young adulthood when
delinquency was measured. Our findings suggest an association between attrition rate, developmental period and the follow-up period.
We did not find a significant moderator effect for the type of measurement of delinquency, indicating no difference in parent reports, self-reports or official records on the longitudinal association between early onset substance use and delinquency. Official record data is per definition only on delinquent behaviour for which the adolescent was charged. Self-reported data on delinquent behaviour can cover all delinquency, but is also more prone to subjectivity, since the adolescents decides what they tell and not tell. However, self-reported measures on delinquency have been shown to be reliable and valid (Elliott et al., 2012). A methodological explanation could be the small number of effect sizes for parental reports (only three), resulting in small power to detect
differences. This finding is therefore not surprising.
Methodological issues
An important limitation of the reported results is the fact that we examined zero-order associations between measures of early onset substance use and delinquent behaviour.
22
Different measures of substance use and delinquent behaviour can be expected to be inter-correlated to varying degrees. This implies that the effect sizes reported in the studies are not always independent of one another, such that, the prospective association between early onset substance use and delinquency can partially be explained by a not coded factor, such as socioeconomic status. Not all studies control in the same manner for confounding variables, which makes it impossible to obtain controlled-for associations in the same way for all studies. Therefore, we chose to focus solely on zero-order
associations in the present meta-analysis. This should be acknowledged when interpreting the results.
Second, the trim and fill analysis (Duval & Tweedie, 2000b) suggested that there was a risk of publication bias. When correcting for this publication bias, the overall effect size of early onset substance use and later delinquent behaviour decreased and was non-significant. This means that the population effect size is so small that there is no effect of early onset substance use on later delinquent behaviour. However, it must be noted that the trim and fill method is not a perfect method to account for missing data. It is noted that the method was originally developed for meta-analyses in which the effect sizes are independent of each other (Nakagawa & Santos, 2012). Furthermore, when effects sizes are heterogeneous, the performance of the trim and fill method is limited (Peters, Sutton, Jones, Abrams, & Rushton, 2007). Therefore, the results from the trim and fill method in the present study should be interpreted with caution.
A third methodological issue concerns the large heterogeneity and diversity in what studies on the longitudinal association between early onset substance use and delinquent behaviour report, and therefore what we were able to code in the analysis.
23
Several studies were excluded from the analysis as a result of not reporting enough information to calculate the effect size. In addition, some studies did not report on several moderators, causing that some moderator analyses were based on a relatively small amount of effect sizes, resulting in a reduction of statistical power.
Furthermore, our age categorisation for the developmental periods, early adolescence, mid-adolescence, late adolescence and adulthood is not the only categorisation mentioned in the literature. It is noteworthy that a different age categorisation could lead to other results.
Conclusion
The present meta-analysis contributes to the literature on the longitudinal association between early onset substance use and delinquent behaviour. Overall, the results showed that there is a small association between early onset substance use and subsequent
delinquent behaviour. The association between early onset substance use and delinquency was not moderated by delinquency type or severity. Furthermore, studies that reported on marijuana use show a stronger association between early onset substance use and
delinquency, compared to studies that reported on alcohol use. However, this finding must be interpreted with caution, since we focussed on zero-order correlations. We could therefore not account for other substance use and deviant behaviours. The longitudinal relationship is negatively influenced by higher attrition rates and a longer period between the assessments of substance use and delinquency. In addition, this association was smaller when offenders were measured in a later age period compared to mid-adolescence.
24
The findings of the present study provide starting points for more focused future research and the development and improvement of both prevention and intervention programs. Future research could focus on broadening the types of substance use including other substances, such as cocaine and amphetamines. Additionally, research specific on combinations of substances and delinquency could provide further insights since in the present study there were no studies focussing on combinations of substances. Implications for scientists and prevention workers who are developing and improving prevention and intervention programs, are that they should not focus extensively on minimizing the effects of early substance use on delinquent behaviour, since the
association found in this meta-analysis is small. When working on minimizing the effects of early onset substance use, they should focus more on marijuana than on alcohol or tobacco. Furthermore, it is noteworthy that an equal approach for boys and girls, and natives and ethnic minorities can be used, since no differences were found for these constructs on the longitudinal relationship between early onset substance use and delinquent behaviour. Additionally, no distinction has to made for the influence of early onset substance use for different types of delinquency or delinquency severity. It should be acknowledged that a prospective association between early onset substance use and delinquent behaviour is present, although it is relatively small.
25 References
Anthony, J. C., & Petronis, K. R. (1995). Early onset drug use and risk of later drug problems. Drug and Alcohol Dependence, 40(1), 9–15. Retrieved from
http://ovidsp.ovid.com/ovidweb.cgi?T=JS&CSC=Y&NEWS=N&PAGE=fulltext& D=psyc3&AN=1996-02300-002;
Arthur, M. W., Hawkins, J. D., Pollard, J. A., Catalano, R. F., & Baglioni, A. J. (2002). Measuring Risk and Protective Factors for Use, Delinquency, and Other Adolescent Problem Behaviors: The Communities That Care Youth Survey. Evaluation Review,
26(6), 575–601. http://doi.org/10.1177/0193841X0202600601
Barnes, G. M., Welte, J. W., & Hoffman, J. H. (2002). Relationship of Alcohol Use to Delinquency and Illicit Drug Use in Adolescents: Gender, Age, and Racial/Ethnic Differences. Journal of Drug Issues, 32(1), 153–178.
http://doi.org/10.1177/002204260203200107
Becker, S. J., Nargiso, J. E., Wolff, J. C., Uhl, K. M., Simon, V. A., Spirito, A., &
Prinstein, M. J. (2012). Temporal relationship between substance use and delinquent behavior among young psychiatrically hospitalized adolescents. Journal of
Substance Abuse Treatment, 43(2), 251–259.
http://doi.org/10.1016/j.jsat.2011.11.005
Brook, J. S., Gordon, A. S., Brook, A., & Brook, D. W. (1989). The consequences of marijuana use on intrapersonal and interpersonal functioning in black and white adolescents. Genetic, Social, and General Psychology Monographs, 115(3), 349– 369.
Button, T. M. M., Hewitt, J. K., Rhee, S. H., Young, S. E., Corley, R. P., & Stallings, M. C. (2006). Examination of the Causes of Covariation Between Conduct Disorder Symptoms and Vulnerability to Drug Dependence. Twin Research and Human
Genetics, 9(01), 38–45. http://doi.org/10.1375/twin.9.1.38
Carson, D. C., & Esbensen, F.-A. (2014). The mediating effects of delinquent attitudes on race, race heterogeneity, and violent offending. Journal of Crime and Justice, 37(1), 23–41. http://doi.org/10.1080/0735648X.2013.830394
Cheung, M. W. L. (2014). Modeling dependent effect sizes with three-level meta-analyses: A structural equation modeling approach. Psychological Methods, 19(2),
26 211–229. http://doi.org/10.1037/a0032968
Cohen, J. (1988). Statistical power analysis for the behavioral sciences. Statistical Power
Analysis for the Behavioral Sciences. http://doi.org/10.1234/12345678
Cooper, H., Hedges, L., & Valentine, J. (2009). The Handbook of Research Synthesis and
Meta-Analysis. Russell Sage Foundation. Retrieved from
https://books.google.com/books?id=LUGd6B9eyc4C&pgis=1
Creemers, H. E., Korhonen, T., Kaprio, J., Vollebergh, W. A. M., Ormel, J., Verhulst, F. C., & Huizink, A. C. (2009). The role of temperament in the relationship between early onset of tobacco and cannabis use: The TRAILS study. Drug and Alcohol
Dependence, 104(1-2), 113–8. http://doi.org/10.1016/j.drugalcdep.2009.04.010
De Looze, M., Van Dorsselaer, S., De Roos, S., Verdurmen, J., Stevens, G., Gommans, R., … Vollebergh, W. (2014). Gezondheid, welzijn en opvoeding van jongeren in
Nederland. Retrieved from
http://www.trimbos.nl/webwinkel/productoverzicht-webwinkel/psychische-gezondheid/af/~/media/files/inkijkexemplaren/af1009 hbsc 2009 inkijkexemplaar.ashx
Deeg, D. J. H. (2002). Attrition in longitudinal population studies: Does it affect the generalizability of the findings?1An introduction to the series. Journal of Clinical
Epidemiology, 55(3), 213–215. http://doi.org/10.1016/S0895-4356(01)00472-3
Degenhardt, L., Chiu, W. T., Sampson, N., Kessler, R. C., Anthony, J. C., Angermeyer, M., … Wells, J. E. (2008). Toward a global view of alcohol, tobacco, cannabis, and cocaine use: Findings from the WHO world mental health surveys. PLoS Medicine,
5(7), 1053–1067. http://doi.org/10.1371/journal.pmed.0050141
Degenhardt, L., Lynskey, M., & Hall, W. (2000). Cohort trends in the age of initiation of drug use in Australia. Australian and New Zealand Journal of Public Health, 24(4), 421–426.
Dijkstra, J. K., Kretschmer, T., Pattiselanno, K., Franken, A., Harakeh, Z., Vollebergh, W., & Veenstra, R. (2015). Explaining Adolescents’ Delinquency and Substance Use: A Test of the Maturity Gap: The SNARE study. Journal of Research in Crime
and Delinquency, 52(5), 747–767. http://doi.org/10.1177/0022427815582249
Duval, S., & Tweedie, R. (2000a). A Nonparametric “ Trim and Fill ” Method of Accounting for Publication Bias in Meta-Analysis. Journal of the American
27
Statistical Society, 95(449), 89–98. http://doi.org/10.2307/2669529
Duval, S., & Tweedie, R. (2000b). Trim and fill: A simple funnel-plot-based method of testing and adjusting for publication bias in meta-analysis. Biometrics, 56(2), 455– 463. http://doi.org/10.1111/j.0006-341X.2000.00455.x
Ellickson, P. L., Tucker, J. S., & Klein, D. J. (2001). - High-risk behaviors associated with early smoking: results from a 5-year follow-up.; - The Journal of adolescent health : official publication of the Society for Adolescent Medicine. Journal of
Adolescent Health, 28(6), 465–473.
Ellickson, P. L., Tucker, J. S., Klein, D. J., & Saner, H. (2004). Antecedents and outcomes of marijuana use initiation during adolescence. Preventive Medicine,
39(5), 976–984. http://doi.org/10.1016/j.ypmed.2004.04.013
Elliott, D. S., Huizinga, D. H., & Menard, S. (2012). Multiple Problem Youth:
Delinquency, Substance Use, and Mental Health Problems. Springer Science & Business Media.
Fagard, R. H., Staessen, J. A., & Thijs, L. (1996). Advantages and disadvantages of the meta-analysis approach. Journal of Hypertension. Supplement : Official Journal of
the International Society of Hypertension, 14(2), S9–12; discussion S13. Retrieved
from http://www.ncbi.nlm.nih.gov/pubmed/8934372
Fergusson, D. M., Boden, J. M., & Horwood, L. J. (2008). The developmental
antecedents of illicit drug use: Evidence from a 25-year longitudinal study. Drug
and Alcohol Dependence, 96(1-2), 165–177.
http://doi.org/10.1016/j.drugalcdep.2008.03.003
Fergusson, D. M., & Horwood, L. J. (1997). Early onset cannabis use and psychosocial adjustment in young adults. Addiction, 92(3), 279–296.
http://doi.org/10.1111/j.1360-0443.1997.tb03198.x
Flannery, D. J., Williams, L. L., & Vazsonyi, A. T. (1999). Who are they with and what are they doing? Delinquent behavior, substance use, and early adolescents’ after-school time. The American Journal of Orthopsychiatry, 69(2), 247–253.
Gillespie, N. A., Neale, M. C., Jacobson, K., & Kendler, K. S. (2009). Modeling the genetic and environmental association between peer group deviance and cannabis use in male twins. Addiction (Abingdon, England), 104(3), 420–9.
28
http://doi.org/10.1111/j.1360-0443.2008.02457.x
Gordon, M. S., Kinlock, T. W., & Battjes, R. J. (2004). Correlates of early substance use and crime among adolescents entering outpatient substance abuse treatment.
American Journal of Drug & Alcohol Abuse, 30(1), 39–59.
http://doi.org/10.1081/ADA-120029865
Green, K. M., Musci, R. J., Johnson, R. M., Matson, P. A., Reboussin, B. A., & Ialongo, N. S. (2015). Outcomes associated with adolescent marijuana and alcohol use among urban young adults: A prospective study. Addictive Behaviors, 53, 155–160. http://doi.org/10.1016/j.addbeh.2015.10.014
Gustavson, K., von Soest, T., Karevold, E., & Røysamb, E. (2012). Attrition and generalizability in longitudinal studies: findings from a 15-year population-based study and a Monte Carlo simulation study. BMC Public Health, 12, 918. Retrieved from
http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=3503744&tool=pmcentr ez&rendertype=abstract
Hawkins, D. F., Laub, J. H., Lauritsen, J. L., & Cothern, L. (2000). Race , ethnicity , and serious and violent juvenile offending. Juvenile Justice Bulletin, (June), 8.
Hayden, J. A., Windt, D. a Van Der, Cartwright, J. L., Côté, P., & Bombardier, C. (2013). Research and Reporting Methods Annals of Internal Medicine Assessing Bias in Studies of Prognostic Factors. Annals of Internal Medicine, 144(158), 280–286. http://doi.org/10.7326/0003-4819-158-4-201302190-00009
Hemphill, S. A., Heerde, J. A., Scholes-Balog, K. E., Smith, R., Herrenkohl, T. I., Toumbourou, J. W., & Catalano, R. F. (2014). Reassessing the Effects of Early Adolescent Alcohol Use on Later Antisocial Behavior: A Longitudinal Study of Students in Victoria, Australia, and Washington State, United States. Journal of
Early Adolescence, 34(3), 360–386.
Higgins, J. P. T., & Green, S. (2011). Cochrane Handbook for Systematic Reviews of
Interventions Version 5.1.0 [updated March 2011]. The Cochrane Collaboration
(Vol. 4).
Hox, J. J. (2010). Multilevel Analysis: Techniques and Applications, Second Edition
29
Hunter, S. B., Miles, J. N. V, Pedersen, E. R., Ewing, B. A., & D’Amico, E. J. (2014). Temporal associations between substance use and delinquency among youth with a first time offense. Addictive Behaviors, 39(6), 1081–6.
http://doi.org/10.1016/j.addbeh.2014.03.002
Ioannidis, J. P. A., & Lau, J. (1999). Pooling Research Results: Benefits and Limitations of Meta-Analysis. Joint Commission Journal on Quality Improvement, 25(9), 462– 469. Retrieved from https://www.readbyqxmd.com/read/10481815/pooling-research-results-benefits-and-limitations-of-meta-analysis
Johnson, R. A., & Gerstein, D. R. (1998). Initiation of use of alcohol, cigarettes,
marijuana, cocaine, and other substances in US birth cohorts since 1919. American
Journal of Public Health, 88(1), 27–33.
Kaplow, J. B., Curran, P. J., & Dodge, K. A. (2002). Child, parent, and peer predictors of early-onset substance use: A multisite longitudinal study. Journal of Abnormal
Child Psychology, 30(3), 199–216.
Kessler, R. C., Davis, C. G., & Kendler, K. S. (1997). Childhood adversity and adult psychiatric disorder in the US National Comorbidity Survey. Psychological
Medicine, 27(5), 1101–1119. Retrieved from
http://ovidsp.ovid.com/ovidweb.cgi?T=JS&CSC=Y&NEWS=N&PAGE=fulltext& D=med4&AN=9300515;
Lipsey, M. W., & Wilson, D. B. (2001). Practical meta-analysis. Applied Social Research
Methods Series, 49, 264. http://doi.org/10.1016/j.autneu.2007.06.087
Loeber, R., & Farrington, D. P. (1994). Problems and solutions in longitudinal and experimental treatment studies of child psychopathology and delinquency. Journal
of Consulting and Clinical Psychology, 62(5), 887–900.
http://doi.org/10.1037/0022-006X.62.5.887
Loeber, R., & Farrington, D. P. (2000). Young children who commit crime:
Epidemiology, developmental origins, risk factors, early interventions, and policy implications. Development and Psychopathology, 12(4), S0954579400004107. http://doi.org/10.1017/S0954579400004107
Lynne-Landsman, S. D., Graber, J. A., Nichols, T. R., & Botvin, G. J. (2011a). Is Sensation Seeking a Stable Trait or Does it Change Over Time? Journal of Youth
30
and Adolescence, 40(1), 48–58. Retrieved from
http://ovidsp.ovid.com/ovidweb.cgi?T=JS&CSC=Y&NEWS=N&PAGE=fulltext& D=psyc8&AN=2010-26384-005;
Lynne-Landsman, S. D., Graber, J. A., Nichols, T. R., & Botvin, G. J. (2011b). Trajectories of aggression, delinquency, and substance use across middle school among urban, minority adolescents. Aggressive Behavior, 37(2), 161–176. http://doi.org/10.1002/ab.20382
Lynskey, M., & Hall, W. (2000). The effects of adolescent cannabis use on educational attainment: a review. Addiction, 95(11), 1621–1630. http://doi.org/10.1046/j.1360-0443.2000.951116213.x
Marcus, R. F., & Jamison II, E. G. (2013). Substance Use in Adolescence and Early Adulthood: Which Best Predicts Violence in Early Adulthood? Journal of Child &
Adolescent Substance Abuse, 22(1), 38–57.
http://doi.org/10.1080/1067828x.2012.729909
Mason, W. A., & Windle, M. (2002). Reciprocal Relations Between Adolescent Substance Use and Delinquency: A Longitudinal Latent Variable Analysis[1].
Journal of Abnormal Psychology, 111(1), 63–76.
http://doi.org/10.1037/0021-843X.111.1.63
McCambridge, J., & Strang, J. (2005). Age of First Use and Ongoing Patterns of Legal and Illegal Drug Use in a Sample of Young Londoners. Substance Use & Misuse,
40(3), 313–319. http://doi.org/10.1081/JA-200049333
Miller, M., Borges, G., Orozco, R., Mukamal, K., Rimm, E. B., Benjet, C., & Medina-Mora, M. E. (2011). Exposure to alcohol, drugs and tobacco and the risk of subsequent suicidality: Findings from the Mexican Adolescent Mental Health Survey. Drug and Alcohol Dependence, 113(2-3), 110–117.
http://doi.org/10.1016/j.drugalcdep.2010.07.016
Moffitt, T. E. (1993). Adolescence-limited and life-course-persistent antisocial behavior: a developmental taxonomy. Psychological Review, 100(4), 674–701. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/8255953
Nakagawa, S., & Santos, E. S. A. (2012). Methodological issues and advances in biological meta-analysis. Evolutionary Ecology, 26(5), 1253–1274.
31 http://doi.org/10.1007/s10682-012-9555-5
Pang, R. D., Farrahi, L., Glazier, S., Sussman, S., & Leventhal, A. M. (2014). Depressive symptoms, negative urgency and substance use initiation in adolescents. Drug and
Alcohol Dependence, 144, 225–30. http://doi.org/10.1016/j.drugalcdep.2014.09.771
Pardini, D. A. (2006). The Callousness Pathway to Severe Violent Delinquency.
Aggressive Behaviour, 32(6), 590–598. http://doi.org/10.1002/ab.20158
Peters, J. L., Sutton, A. J., Jones, D. R., Abrams, K. R., & Rushton, L. (2007). Performance of the trim and fill method in the presence of publication bias and between-study heterogeneity. Statistics in Medicine, 26(25), 4544–4562. http://doi.org/10.1002/sim.2889
Puzzanchera, C. (2014). Juvenile arrests 2012. Juvenile Offenders and Victims: National
Report Series, 1–12.
R Core Team. (2016). R: A language and environment for statistical computing. Vienna: R Foundation for Statistical Computing. Retrieved from https://www.r-project.org/ Rosenthal, R. (1979). The file drawer problem and tolerance for null results.
Psychological Bulletin, 86(3), 638–641. http://doi.org/10.1037/0033-2909.86.3.638
Rosenthal, R. (1994). Parametric measures of effect size. In H. Cooper & L. V. Hedges (Eds.), The handbook of research synthesis (pp. 231–244). Russel Sage Foundation. Torbet, P., Gable, R., Montgomery, I., & Hurst, H. (1996). State Responses to Serious
and Violent Juvenile Crime. Retrieved from
https://books.google.com/books?hl=nl&lr=&id=2lU3r-EMhz4C&pgis=1 Van Den Noortgate, W., & Onghena, P. (2003). Multilevel Meta-Analysis: A
Comparison with Traditional Meta-Analytical Procedures. Educational and
Psychological Measurement, 63(5), 765–790.
http://doi.org/10.1177/0013164403251027
Vega, W. A., Aguilar-Gaxiola, S., Andrade, L., Bijl, R., Borges, G., Caraveo-Anduaga, J. J., … Wittchen, H.-U. H. U. (2002). Prevalence and age of onset for drug use in seven international sites: Results from the International Consortium of Psychiatric Epidemiology. Drug and Alcohol Dependence, 68(3), 285–297.
http://doi.org/10.1016/S0376-8716(02)00224-7
32
Statistical Software, 36(3), 1–48. Retrieved from
http://www.scopus.com/inward/record.url?eid=2-s2.0-77958110812&partnerID=tZOtx3y1
Wasserman, G. A., McReynolds, L. S., Lucas, C. P., Fishher, P., & Santos, L. (2002). The Voice DISC-IV With Incarcerated Male Youths: Prevalence of Disorder.
Journal of the American Academy of Child & Adolescent Psychiatry, 41(3), 314–
321. http://doi.org/10.1097/00004583-200203000-00011
White, H. R., Loeber, R., Stouthamer-Loeber, M., & Farrington, D. P. (1999).
Developmental associations between substance use and violence. Development and
Psychopathology, 11(4), 785–803. http://doi.org/10.1017/S0954579499002321
Wibbelink, C. J. M., & Assink, M. (2015). Manual for conducting a three-level meta-analysis in R. Manuscript in Preparation.
33 Table 1 Characteristics of Included Studies
Reference Year Sampl
e ID #
ES N %Male %White Substance
M Age EO SU M Age Delinquency Type delinquency Type of measurement
Brook & Balka 1999 1 2 133
2 46.47 100.00 c 14 19.00 non severe self reports
Brook & Brook 2003 2 1 222
6 51.71 100.00 c 15,2 17.20 non severe self reports
Chung 2002 3 1 423 47.52 72.58 b 13 13.00 both self reports
Ellickson, McGuigan 2000 4 2 312
8 999.00 29.00 g 14 18.00 both self reports
Ellickson, Tucker & Klein 2001 4 20 432
7 52.00 68.00 a 14 18.00 non severe self reports
Ellickson, Tucker & Klein 2003 5 10 336
9 52.00 32.00 b 14 23.00 both self reports
Ellickson, Tucker, Klein
& Saner 2004 4 3
155
7 52.00 33.70 c 14 16.00 non severe self reports
34 Table 1 (continued)
Reference Year Sampl
e ID #
ES N %Male % White Substance
M Age EO SU M Age Delinquency Type delinquency Type of measurement
Fergusson & Horwood 1999 6 3 990 49.64 999.00 a 10 16.00 non severe parental report
Green et al 2016 7 3 608 52.60 87.20 f 15 22.00 both official records
Hemphill et al 2014 8 10 185
8 48.00 21.87 b 13 15.00 non severe self reports
Hill et al 1999 3 2 808 51.00 54.00 b,c 10,3 18.00 severe self reports
Lynne-Landsman et al 2011 9 15 293
1 50.00 93.00 g 12 14.00 non severe self reports
Marcus et al 2013 10 20 110
64 51.00 26.70 g 14,8 21.80 severe self reports
Mason et al 2010 3 4 808 54.00 53.00 b 10,7 18.00 both self reports
35 Table 1 (continued)
Reference Year Sampl
e ID #
ES N %Male % White Substance
M Age EO SU M Age Delinquency Type delinquency Type of measurement Murphy et al 2012 11 6 898
4 51.00 33.14 b 15 14-23 both self reports
Pedersen et al 2010 12 1 135
3 999.00 999.00 c <16 27.00 non severe official records Shope, Waller et al 2001 13 12 440
3 47.00 16.00 a,b,c 15,62 23.70 both official records
White, Loeber, Farrington 1999 14 8 506 100.00 57.50 b,c 13,25 14.25 severe self reports Note: Sample ID = the unique sample number; # ES = number of effect sizes per study; %Male = percentage male in the sample; %
White= percentage of white participants; M Age EO SU = mean age assessment of early onset substance use; M Age delinquency = mean age assessment of delinquency.
a = tobacco b = alcohol; c= marijuana; d=tobacco&alcohol; e= tobacco&marijuana; f=alcohol&marijuana; g=all substances. <16 = substance use initiation before 16years
36
Table 2 Results for the Overall Prospective Association between Early Onset Substance Use and Delinquent Behaviour. Early onset substance
use
#ES Mean r(SE) 95% CI Sig.
mean r (p) % Var. at level 1 Level 2 variance % Var. at level 2 Level 3 variance % Var. at level 3
Before trim and fill analysis
131 0.163(0.032) (0.099; 0.226) <.001*** 1.77 .006*** 30.51 .013*** 67.72
After trim and fill analysis
138 0.040(0.048) (-0.054;0.134) .401 0.68 .006*** 11.44 .044*** 87.88
Note. #ES = number of effect sizes; SE = standard error; CI = confidence interval; Sig = significance; Mean r = mean effect size (r); % Var. = percentage of variance explained; Level 2 variance; variance between effect sizes form the same study; Level 3 variance = variance between studies.
* p < .05.
** p < .01.
37
Table 3 Results for Continuous and Categorical Moderators (Bivariate Models). Moderator variables # Studies # ES Intercept (95% CI)/mean r (95% CI) β (95% CI) F (df1, df2)a pb Level 2 variance Level 3 variance Delinquency characteristics Delinquency type 7 84 1.898 (2,81) .156 0.001*** 0.003*** Violent crime index (RC) 6 40 0.094(0.048;0.139)*** Property crime index 4 10 0.127(0.076;0.179)*** 0.034(-0.003;0.071) Non-index 6 34 0.115(0.070;0.159)*** 0.021(-0.009;0.051) Delinquency severity 7 67 0.399(1,65) .530 .002*** 0.006*** Non-severe(RC) 4 22 0.140(0.074;0.206)*** Severe 8 45 0.128(0.066;0.189)** -0.012(-0.051;0.027) Age of delinquency 13 125 0.149(0.087;0.212)*** -0.005(-0.016;0.007) 0.688(1,123) 0.40 .005*** .011***
38 Table 3 (continued) Moderator variables # Studies # ES Intercept (95% CI)/mean r (95% CI) β (95% CI) F (df1, df2)a pb Level 2 variance Level 3 variance Delinquency period 12 121 5.849(3,117) <.001** .004*** .013*** Middle adolescence (RC) 5 21 0.234(0.151;0.318)*** Early adolescence 2 17 0.200(0.074;0.327)** -0.034 (-0.151; 0.083) Late adolescence 6 46 0.115(0.036;0.194)** -0.119(-0.181;-0.058)*** Adulthood 4 37 0.056(-0.045;0.157) -0.178(-0.293;-0.064)** Early onset substance use characteristics Age at first assessment 14 131 0.162(0.099;0.225)*** 0.003 (-0.014;0.019) 0.108 (1,129) .743 .006*** .012***
39 Table 3 (continued) Moderator variables # Studies # ES Intercept (95% CI)/mean r (95% CI) β (95% CI) F (df1, df2)a pb Level 2 variance Level 3 variance Type of substance 12 111 8.344(2,108) <.001*** .004*** .011*** Alcohol (RC) 7 51 0.158(0.092;0.224)*** Tobacco 4 31 0.143(0.072;0.215)*** -0.015 (-0.063;0.034) Marijuana 9 26 0.226(0.159;0.293)*** 0.068(0.024;0.112)** Level of substance use 14 131 1.093 (2,128) .338 .006*** .013*** Experimental (RC) 9 52 0.155 (0.088;0.221)*** Regular 11 70 0.173(0.108;0.239)*** 0.019(-0.019;0.056) High 2 9 0.125(0.037;0.214)** -0.029(-0.096;0.037) Sample descriptors Publication year 14 131 0.1632 (0.0993; 0.2271)*** -0.0026 (-0.0104;0.0052) 0.436 (1,129) 0.51 .0058*** .0130*** Percentage of males 13 120 0.156(0.092;0.220)*** 0.000(-0.000;0.0001) 1.572 (1,118) .212 .006*** .012***
40 Table 3 (continued) Moderator variables # Studies # ES Intercept (95% CI)/mean r (95% CI) β (95% CI) F (df1, df2)a pb Level 2 variance Level 3 variance Percentage white 12 119 .150(0.076;0.225)*** 0.001(-0.001;0.003) 0.966 (1,117) .328 .006*** .014*** Attrition 13 119 0.156(0.051;0.262)** -0.009(-0.014;-0.003)** 10.658 (1,117) .001*** .005*** .035*** Research design descriptors Type of measurement of delinquency 14 131 0.018 (2,128) 0.982 .006*** .014*** Self-report (RC) 12 112 0.161(0.088;0.235)*** Parental report 1 3 0.171(0.038;0.304)* 0.009(-1.04;0.123) Official records 3 16 0.170(0.022;0.319)* 0.009(-0.157;0.175) Time to follow up 14 131 0.175(0.099;0.252)*** -0.001(-0.002;-0.0005)** 8.787(1,129) 0.004** 0.005*** 0.019***
41
Note: # Studies = number of studies; # ES = number of effect sizes; mean r= mean effect size (r); CI = confidence interval; β = estimated regression coefficient; Level 2 variance = variance between effect sizes from the same study; Level 3 variance = variance between studies.
a Omnibus test of all regression coefficients in the model. b p-Value of the omnibus test
43 Figure 1 Flowchart
PRISMA 2009 Flow Diagram
Records identified through database searching (k = 2335) Sc reen in g Inc lude d Eli gib ilit y Ide nti fic at io n
Additional records identified through other sources
(k = 3)
Records after duplicates removed (k = 1896)
Records screened (k = 1896)
Records excluded based on title and abstract (k = 1855)
Full-text articles assessed for eligibility
(k = 41 )
Excluded:22
Full text not available: 2
Included other illicit drugs in analyses: 1 No bivariate results (i.e. only
multivariate): 14
Calculation of effect sizes not possible based on reported results: 3
No longitudinal relationship reported: 1 No empirical study (i.e. literature review): 1
Studies included in quantitative synthesis
(meta-analysis) (k = 19 )
44 Figure 2. Trim and fill plot.
