All data on adolescents in SE and RYC were collected in the fall of 2008. Trained interviewers administered anonymous questionnaires in classes (SE) or living rooms (RYC) and underlined the anonymity of the respondents when introducing and clarifying the questionnaire. The data on students in mainstream education were collected in a similar way in the fall of 2007 [15].

The association between substance use and RYC, SEB and SEL respectively and the role of background factors was investigated by using logistic regression analysis. In the first model, the association between substance use (daily use of tobacco, heavy weekly alcohol drinking, lifetime prevalence of cannabis use and lifetime prevalence of hard drug use) and group membership (RYC, SEB, SEL, with mainstream education being the reference group) was tested. In a next step, this association was adjusted for the background factors such as gender, ethnic background, age and marital status of parents. Finally, to investigate a possible interaction between these background characteristics and group membership, the following two-way interaction terms were included: gender (0 = girls, 1 = boys) by respectively RYC, SEB or SEL, age (0 = adolescents in the age of 12–13 years, 1 = adolescents aged 14–16 years) by RYC, SEB or SEL, ethnic background (0 = Dutch, 1 = ethnic minority) by RYC, SEB or SEL and marital status of parents (0 = nondivorced parents, 1 = divorced parents) by RYC, SEB and SEL. In each of these analyses RYC, SEB, or SEL was coded as 1 and mainstream education was coded as 0.

All analyses were performed taking into account the fact that the data are clustered, i.e., adolescents from the same class or RYC group were drawn as a single group which results in dependence between observations (adolescents from the same class/group share several characteristics, like having the same teacher or living in the same region). This will influence standard errors and p values. Therefore, robust standard errors were calculated by using the Huber/White/sandwich estimator as implemented in STATA version 9.1 [18].

Substance use rates are particularly high among adolescents living in RYC institutions. Especially, daily tobacco use was highly prevalent, with more than 20 times as many daily smokers among adolescents aged 12–13 years in RYC as compared to those in mainstream education. Being in RYC also increased the risk of lifetime prevalence of cannabis use, hard drug use and—to a smaller degree—heavy weekly drinking in comparison with those in mainstream education. In addition, elevated substance use rates among adolescents who are in SE due to behavioral problems (SEB) were found for all investigated types of substance use. Daily tobacco use and lifetime prevalence of hard drug use, among adolescents in SE for learning disabilities (SEL) were also higher compared to those in mainstream education, although differences were less pronounced as for those in SEB or RYC. Differences in substance use rates between the three risk groups and mainstream education were particularly large in the youngest age category (12–13 years old).

Background factors, including age, ethnic background and family situation, partly explained the differences in substance use between mainstream education on the one hand and SE and RYC on the other hand, but all differences between the groups remained substantial and significant. Four significant interactions were found in the relationship between SE/RYC and substance use that were in line with the risk paradox: the elevated risk of substance use in SE/RYC was particularly noticeable among adolescents who are normally at lower risk for substance use (i.e., in SEB, girls for heavy alcohol drinking and cannabis use, and in RYC, ethnic minority adolescents for heavy weekly alcohol use and adolescents with a stable family situation for daily tobacco use).

Our results are consistent with findings of previous studies that showed high prevalences of substance use among adolescents who attend SE schools [4, 9]. As in our study, SE students in both New Zealand and US reported elevated rates of substance use compared to students in mainstream education. In addition, previous studies that found high substance use rates in RYC-settings [1, 21] are confirmed by our results. For example, Vaughn and colleagues [21] investigated substance use among 17 years old adolescents who live in a RYC-setting and found that almost half of these adolescents (46%) had ever used marijuana. Although the present study investigated a younger group, the number of adolescents who used cannabis was even higher (almost 70% of the 16 years old in RYC had used cannabis at least once). In addition, the percentage of heavy alcohol users (consuming more than ten alcoholic drinks per week for boys and more than eight alcoholic drinks per week for girls) was much higher in SEB and RYC compared to mainstream education. About 40% of the adolescents aged 16 years in SEB and RYC and SEL reported to drink more than ten or eight glasses of alcohol per week. This is comparable to the average amount of alcoholic beverages Dutch adults’ drink [22].

Though the present study found an increased risk for substance use among adolescents with learning disabilities, this risk was not as elevated as compared to adolescents in RYC or SEB. Adolescents in SEL are not able to attend mainstream education because they have difficulties with their learning abilities, whereas adolescents in SEB attend SE due to a whole range of problems: emotional disturbance, behavioral problems and, or family problems. Apparently, learning problems as such are less strongly associated with substance use.

In line with previous studies on substance use, this study found that ethnic minorities [21], younger adolescents [1] and adolescents in intact families [3, 14] were less likely to use substances. As shown in Table 1, the percentage of broken families was much higher in RYC and, to a lesser extent in SE, as compared to mainstream education. Therefore, the high prevalence of substance use in RYC and SE can partially be attributed to the high number of adolescents with an unstable family situation.

Several interaction effects between risk factors were found, that are in line with the assumptions of the risk paradox : the elevated risk of substance use among adolescents in RYC/SE was in some cases particularly marked for those who would normally be at lower risk for substance use (i.e., girls, ethnic minorities and adolescents with a stable family situation). We can think of two explanations for these findings. First, those normally at lower risk (e.g., children from intact families) are generally living in more protective environments. The threshold for entering the RYC-setting might accordingly be higher—due to more resources in these families for coping with their child’s problems without professional help—and therefore, higher levels of problem behavior in the children who are finally brought into these RYC-settings may prevail. Thus, elevated levels of problem behavior in children in RYC originating from these settings are to be expected. Moreover, Loeber and Keenan [13] noted in their review report that despite the lower prevalence of disruptive disorders in girls compared to boys, girls who do suffer from conduct disorder are at higher risk for developing comorbid conditions like substance use. They hypothesized that since girls have a higher threshold for problematic behavior or physical disorders than boys, they consequently require more harmful risk factors to develop problem behavior than boys. Following this hypotheses, it can be theorized that girls have a higher threshold for attending SEB schools and therefore have more complex and serious behavioral problems including the use of cannabis and excessive use of alcohol if they do attend SE schools.

Second, we can hypothesize that being in SE-settings has a levelling effect on the problem behavior of all children entering these settings—due to the effects of peer influence within these groups, and due to being in these settings as such. These effects may be particularly elevated in the groups normally at lower risk as well, because entering these settings brings them into contact with higher risk groups. Ethnic minority children may serve as an example in this respect. It is a well-known fact that ethnic minority children are usually at (far) lower risk for drinking alcohol—due to the protective effect of restrictive parenting and cultural values forbidding alcohol in these groups [23]. Entry to a group of high-risk peers—who tend to drink heavily—may have a particularly strong effect on children originating from an ethnic minority background. Future studies analyzing peer processes in these settings are needed to test these assumptions.

A major strength of the current study is that the methodology was the same for RYC, SE and mainstream education. This makes the comparison of high risk groups with adolescents in mainstream education more valid. Possible limitations have to be taken into account. First, analyses were based on self-report data which could result in over- or under-reporting of substance use. But as the administration of the questionnaires in school classes and living rooms occurred under the supervision of a trained interviewer and anonymity was assured, self-report is assumed to be reliable [16]. Second, there is a possible under-representation of substance users due to the number of adolescents—in particular in RYC and SEB—that refused to cooperate. It can be expected that the number of substance users is especially high among these refusers. Third, since the study is based on a cross-sectional design, inferences on causal relationships cannot be made. Fourth, other non-assessed factors can possible also account for much of the elevated risk (family problems, deviant peers, delinquency, school problems) and therefore have to be taken into consideration in future studies.

In conclusion, this study found high rates of substance users among adolescents in RYC institutions and SE, in particular SEB, compared to adolescents in mainstream education. These results confirm that adolescents in these settings are not only at higher risk for psychosocial problems, but also at higher risk for developing substance use disorders. These differences are only partially explained by other risk factors that are elevated in this group (e.g., coming from broken families). Acknowledging these risks is of major importance for prevention and treatment programmes directed at these groups, as high levels of substance use may have a considerable impact on the success of these programmes. Results of this study have important implications for health education and intervention programs for adolescents in RYC institutions and in adolescents in SE. It is important to develop substance use prevention programmes that target high-risk adolescents in SE and RYC.