During adolescence, several neurobiological alterations take place, to some extent driven by pubertal changes and hormones (Giedd, 2008; Blakemore et al., 2010; Peters et al., 2015). While some brain areas, such as the visual cortex and motor cortex, are already fully developed in childhood (Gazzaniga et al., 2002), the fine-tuning of other brain regions, such as the prefrontal cortex, is still in progress during adolescence (Luna et al., 2004; Blakemore and Choudhury, 2006). The prefrontal cortex plays a major role in the regulation of behavior. In the present study, we assessed elements of behavioral control including cognitions that are assumed to be involved in the regulation of behavior (Moffitt et al., 2011), such as the inhibition of responses (e.g., response inhibition) and delay of gratification (Krueger et al., 1996), as well as traits and personality characteristics, such as acting without thinking (Barratt, 1983; Evenden, 1999). We assessed two elements of behavioral control; self-reported behavioral control (i.e., effortful control) and cognitive control (i.e., response inhibition and working memory, Peeters et al., 2015).

Several studies from different disciplines (cf. Verdejo-García et al., 2008) have ascertained the role of behavioral control in the onset and continuation of risk behavior. The use of substances among adolescents, for instance, has been linked to problems with delay of responses (Romer et al., 2010), inhibition problems (Fernie et al., 2013), self-reported impulsivity (Krank et al., 2011; White et al., 2011), and effortful control in particular (Piehler et al., 2012). Behavioral control appears to play a vital role in initiating alcohol use (Squeglia et al., 2014; Peeters et al., 2015) and the development of problematic drinking behavior in adolescents (Nigg et al., 2006). Problems with inhibition, both assessed on a cognitive as well as on a behavioral level increase the chance of early initiating of alcohol use and problem drinking in adolescents. Whelan et al. (2012) found reduced activity in brain regions important for cognitive control even among adolescents who just initiated alcohol use, suggesting that vulnerabilities in neural circuits underlying cognitive control might precede the initiation of drinking behavior in adolescents (Wetherill et al., 2013).

Altogether, these results suggest that relatively early weaknesses in behavioral control might place adolescents at risk for early initiation of risk behavior and the development of related problems. Adolescents with behavioral control problems are more likely to act without thinking and less likely delay response, receiving decreased attention for the negative consequences of behavior and increasing their involvement in risk behavior. This reasoning is in line with Krueger et al. (1996) and Tarter et al. (2003) who assumed that early weaknesses in behavioral control underlie the development of externalizing psychopathology later in life.

Adolescent risk taking behavior is sometimes conceptualized as irrational and impulsive; however, studies suggest little to no differences in risk evaluation and perception between adults and adolescents (Gerrard et al., 1996; Steinberg and Cauffman, 1996; Spear, 2000). This suggests that adolescents, just like adults, are aware of the consequences involved in risk taking behavior. One possible explanation for the observed difference in risk taking behavior among adolescents and adults might thus be that the expected rewarding value of (some) risk behaviors is greater for adolescents than for adults (Van Leijenhorst et al., 2010a,b; Spear, 2011). The gains might simply be much higher for adolescents than for adults when engaging in risky behaviors (Crone and Dahl, 2012). In addition, adolescents might be more sensitive to rewards than adults. Indeed, Chein et al. (2011) found increased activity in reward related brain systems in adolescents while performing a risk taking task; however, this increase was only observed when adolescents completed the task in the presence of peers. This reward sensitization was present not only at a neurocognitive level (e.g., neural activation), but also at a behavioral level (e.g., task performance), resulting in more risk taking behavior by adolescents, as observed by peers. Adults did not reveal this heightened activity or increased risk taking while performing the task with peers (Chein et al., 2011). This study suggested that the peer presence increased the reward for engaging in risky behaviors among adolescents, but not for adults. It seems that the presence of peers changes the perception with respect to the anticipated reward when engaging in risk behavior (Spear, 2011), a change only observed in adolescents but not in adults.

Besides differences in reward perception, the neurological response observed among adults and adolescents is different when faced with the same rewarding stimuli (Ernst et al., 2005; Galvan et al., 2006; Van Leijenhorst et al., 2010a,b). Van Leijenhorst et al. (2010a), for instance, found that adolescents compared to adults and children revealed a heightened neurological response toward rewards in a decision making task. Participants could choose either a low-risk gamble (lower risk and lower reward) or a high-risk gamble (higher risk and higher reward). While performing the task, brain activation was assessed using fMRI techniques. The results indicated that during adolescence, reward related systems show a peak in activation in response to risky decisions, with a possible high rewarding outcome.

With respect to risk taking behavior in real life like substance use, individual variability among adolescents in reward sensitivity either due to heightened neurological responses or higher expected rewarding value of engagement in these risk behaviors might explain why some adolescents more than others engage in risky behaviors like substance use. Taking risk might be more rewarding for some individuals in certain situations compared to different individuals in different situations (Dawe et al., 2004; Galvan et al., 2007). Heightened reward sensitivity might contribute to more risk taking behaviors in reward sensitive adolescents, as shown in a study by van Hemel-Ruiter et al. (2015) who found that adolescents (12–18 years) who scored high on reward sensitivity drank more heavily compared to adolescents who were less sensitive to reward. Moreover, Xiao et al. (2013) found differences in reward sensitivity assessed with the Iowa Gambling Task (IGT; a measure assumed to assess variation in reward sensitivity, Franken and Muris, 2005; Cauffman et al., 2010) between adolescent binge drinking and abstainers, such that binge drinkers were more sensitive for reward than the abstainers. Altogether, these results suggest that individual differences in reward sensitivity are directly associated with risk behaviors, such as alcohol use. Moreover, the increased sensitivity to reward observed in adolescents relative to adults and children affects the level at which behavioral control must be deployed (Somerville et al., 2011).

Neurodevelopmental models suggest that the peak in risk behavior in mid adolescence can be explained by the interplay between not yet fully developed cognitive control functions and increased neural responses toward reward (e.g., dual system models/imbalance model, cf. Spear, 2000; Galvan et al., 2006; Casey et al., 2008; Giedd, 2008; Somerville et al., 2011; Smith et al., 2013; Silverman et al., 2015). To this end it is hypothesized that adolescents with relatively weak behavioral control at age 11 and high reward sensitivity at age 16 are at the greatest risk for risk behaviors at age 16, such as alcohol use, cannabis use, and smoking. To the best of our knowledge, this is the first longitudinal study that examined this interaction looking at both self-reported behavioral control as well as cognitive control. Although these measures all tap in the same underlying construct, namely behavioral control, they might be differently related to specific risky behaviors like substance use (Verdejo-García et al., 2008; Janssen et al., 2015) and interact differently with reward sensitivity.

Participants in the present study were selected from a larger longitudinal population study that included 2230 Dutch adolescents who enrolled the study at age 11, and they were followed at least to age 25. For a detailed description of the inclusion criteria and selection process, please see de Winter et al. (2005) and Huisman et al. (2008). Mean age of the population sample was 11.09 years at baseline (SD = 0.59, 50.8% girls), 13.56 years at T2 (SD = 0.53), and 16.13 years at T3 (SD = 0.73). Response rates were 2149 (96.3, 51.2% girls) at T2 and 1816 (81.6, 52.3%) at T3 for the population sample.

At the third wave, a focus sample of 744 adolescents was selected and invited to participate in a number of laboratory tasks. In total, 715 (96.1%) adolescents (49.1% boys) agreed to participate in this experimental session (Mean age T1 = 11.10, SD = 0.55, T3 = 16.11, SD = 0.59). Adolescents with increased risk for mental health problems (e.g., high frustration/fearfulness and low effortful control, and family risk parental depression, anxiety, addiction, psychoses, antisocial behavior, single parent household) were oversampled in this focus cohort, resulting in a group of 66% adolescents being at risk and 34% adolescents being randomly selected from the population sample (N = 715). Information on mental health problems assessed with the Youth Self-Report Scale for the focus sample and for the total TRAILS sample (including simple t-test) are provided in Table 1. Adolescents in the focus cohort scored significantly higher on Attentional Deficit Hyperactivity Disorder (ADHD) and Oppositional Disorders, in line with Krueger et al. (1996) and Tarter et al. (2003). Due to a strong overlap between these subscales of the YSR and aspects of behavioral control (e.g., ADHD and effortful control Pearson correlation = -0.48, p < 0.001; oppositional problems and effortful control Pearson correlation = -0.34, p < 0.001) we decided not to include these subscales as confounding variables. The experimental protocol was approved by the Central Committee on Research Involving Human subjects (CCMO). Written informed consent was obtained from the participants. Assessment took place under the guidance of a TRAILS research assistant who received extensive training to ensure a standardized procedure for all participants. Assessment took place at different locations (depending on the residence of the participants). At each location, the experimental room was sound proof, and it had blinded windows (for a detailed description of the procedure of the experimental session, see Bouma et al., 2009).

In this study, we used a focus cohort, which provided the possibility to include additional measures that might be relevant for this specific group and age. This resulted in varying availability of measures. Measures of behavioral control for instance were available at wave 1, however, not at wave 3, while measures of reward sensitivity were only assessed within the focus cohort and therefore only available at wave 3.

First, descriptive statistics and Pearson correlations among the study variables are provided. Second, simple path analyses were used to examine the unique effects of effortful control, cognitive control, and reward sensitivity on the three risk behaviors. For each risk behavior, we selected adolescents who indicated that they had not used the substance in question at baseline. This resulted in three different data sets, one including only non-drinkers at baseline (N = 489), one including only non-cannabis users (N = 699), and one including only non-smokers at baseline (N = 615). Table 2 includes an overview of the sample size and demographic information for each data set separately. In the second step, interaction variables were created between reward sensitivity and cognitive control and between reward sensitivity and effortful control using centered variables. Each interaction was entered in a separate model in order to maintain a clear interpretation of each interaction effect. Cannabis use and smoking both revealed a skewed distribution with many zeros; therefore, we used a Zero-Inflated Poisson (ZIP) model as a traditional Poisson model is not sufficient when standard deviations are bigger than the mean (over-dispersion; cf. Peeters et al., 2012). The ZIP model allowed us to interpret the continuous part of the model (adolescents who used cannabis/cigarettes) while accounting for the many zero’s. We controlled for gender and used Maximum likelihood with robust standard errors (MLR) as estimation method to account for non-normality of the data in all analysis. FIML was used to deal with missing data. Analyses were completed in Mplus version 7.3 (Muthén and Muthén, 1998). Model fit measures were not informative because all possible paths in the model were estimated (e.g., full model).

For each risk behavior (i.e., alcohol, cannabis, smoking), we selected the non-users at baseline resulting in three different samples with non-users at baseline (either non-drinker, non-cannabis user or non-smoker at wave 1). Measures of behavioral control at wave 1 therefore preceded risk behavior. Table 2 includes information on mean age, percentage boy/girl, the three measures of behavioral control and for reward sensitivity for the three subsamples separately and for the total sample. We further looked at the use of other substances in the specific subsamples (alcohol, cannabis, and smoking behavior). In the alcohol sample only one adolescent reported cannabis use. Ninety-three percent reported that they never smoked a cigarette (1.5% adolescents indicated cigarette use more than once). In the cannabis subsample, 69.6% reported no alcohol use at baseline (around 15% of the drinkers reported that they only drank alcohol once in their lives). 87.9% reported no smoking behavior (4.5% of the smokers reported cigarette use more than once). In the smoking subsample, 73.8% reported no alcohol use at baseline (around 12% of the drinkers reported that they only drank once in their lives). Only three (0.5%) adolescents reported cannabis use, of whom 1 reported cannabis use more than once at baseline.

In Table 3, descriptive statistics for the three risk behaviors, alcohol, cannabis, and smoking, are presented. All three risk behaviors revealed a positive association with each other. Furthermore, Pearson correlation revealed poorer effortful control, working memory performance, and response inhibition for boys. Alcohol and cannabis use were both higher among boys; however, smoking behavior appeared to be higher among girls. T-test supported this assumption [t(684) = 2.847, p < 0.005]. In addition, weaker effortful control at T1 was associated with more alcohol use, cannabis use, and smoking behavior at T3. Reward sensitivity at T3 was positively associated with alcohol use at T3, however, no significant correlation was found with cannabis use or smoking. Working memory and response inhibition correlated positively with each other. A negative correlation was found between working memory and effortful control, suggesting poorer working memory functioning is associated with relatively weaker effortful control skills (note that higher scores on inhibition and working memory indicate poorer functioning).

In Tables 4–6, the main unique effects of effortful control, cognitive control, and reward sensitivity on risk behavior are presented. We included all three behavioral control measures in the same model to analyze their unique contribution to risk behavior. With respect to alcohol, effortful control at T1 significantly predicted T3 alcohol use (β = -0.15, SE = 0.05). That is, adolescents with relatively poor effortful control at T1 increased stronger in their alcohol use between T1 and T3 compared to adolescents with relatively good effortful control. For reward sensitivity and cognitive control, no main effect on alcohol use was found.

With respect to cannabis use, a similar pattern for effortful control was observed in that effortful control at T1 predicted a stronger increase in cannabis use between 11 and 16 years of age (β = -0.57, SE = 0.12). Weaker effortful control skills at age 11 predicted cannabis use at age 16. In addition, working memory functioning at age 11 predicted cannabis use at age 16 (β = 0.43, SE = 0.15). Adolescents with relatively weaker working memory skills at age 11 used more cannabis use at age 16 (note higher scores on working memory indicate poorer functioning). In contrast to what was expected, response inhibition was a significant predictor of cannabis use, with those having relatively good inhibitions skills progressing more heavily in the use of cannabis compared to those with weaker inhibition skills (β = -0.08, SE = 0.03). Additional analysis revealed that response inhibition was only a significant predictor of cannabis use when controlling for other measures of behavioral control and not when analyzed alone (β = -0.36, SE = 0.40). In addition, gender was a significant predictor of cannabis use at age 16, with boys more likely using cannabis compared to girls.

With respect to smoking behavior, none of the hypothesized main effects were significant. Only gender was a significant unique predictor.

The interaction effect with reward sensitivity was examined for all three measures of behavioral control (Tables 4–6). A significant interaction was found between reward sensitivity and effortful control for alcohol use (β = -0.57, SE = 0.15) and cannabis use (β = -0.28, SE = 0.13). These interaction effects are illustrated in Figures 1 and 2. The interaction effect reveals that adolescents with relatively poor effortful control at age 11 and high levels of reward sensitivity at age 16 are the heaviest drinkers and cannabis users at age 16 (controlled for previous use). For adolescents with good effortful control at baseline, the level of reward sensitivity in mid adolescence does not appear to influence the amount of alcohol or cannabis that is consumed in mid adolescence. In contrast, for smoking no main effect of response inhibition on smoking behavior was observed; however, the interaction between response inhibition and reward sensitivity predicting smoking behavior was significant (Figure 3; β = 2.67, SE = 1.28). When response inhibition was relatively good, adolescents who were less reward sensitive smoked less at age 16 (reversed effect).

Bedsides the strengths of the study, such as a large sample size and the use of different measures of control, some limitations should be mentioned. First, both effortful control and cognitive control were assessed only at wave 1 at age 11. This allowed us to look at weaknesses in behavioral control before initiation of alcohol. Yet, it can be argued that levels of behavior control at this stage of live are not indicative of levels of control during mid adolescence when the peak in risk taking is observed. After all, the ongoing maturation of the prefrontal cortex during adolescence is assumed to play a vital role in explaining the risk behavior (Steinberg, 2007; Casey and Jones, 2010). At the same time, recent studies have suggested (Forbes and Dahl, 2010; Peters et al., 2015) that the onset of puberty entails hormonal changes that underlie structural brain maturation and influence cognitive processing associated with reward, motivation, and risk taking behavior. Moreover, individual differences in cognitive control might already be visible at this early age, reflecting a general pattern of growth that is not age specific (Spear, 2000; Casey et al., 2008). In addition, longitudinal assessment of cognitive control in the TRAILS study does reveal correlation between cognitive control at ages 11 and 19 (Boelema et al., 2015). Nevertheless, for future research, it would be interesting to include measures of cognitive control in mid adolescence when the peak in risk taking behavior is observed.

Second, the measures of cognitive control in the present study were designed to reveal neurocognitive abnormalities in complex cognitive functioning (de Sonneville, 1999). This measure might be particularly relevant for clinical populations, but it might be less sensitive when it involves detecting differences in functioning in a relatively normal sample of adolescents (Boelema et al., 2015). This might explain why no predictive effects of cognitive control on alcohol were found. Other tasks, such as the Self Ordered Pointing Task might be better in detecting working memory difficulties in non-clinical populations (see for instance Peeters et al., 2015). Third, the BGT has been originally developed to assess decision-making behavior under arousing circumstances (reward-based decision-making). It is possible that the BGT task does not assess reward sensitivity but rather decision making in arousing situations. Nevertheless, the IGT, a similar decision-making task as the BGT, assesses the extent to which immediate rewards are weighted in relation to long term consequences (Ernst et al., 2003; Bechara, 2005), which can be interpreted as a measure of sensitivity to reward (e.g., for some, an immediate reward might not outweigh long term consequences while for others, immediate reward is much more appealing). Similar gambling tasks have been used to examine reward processing at a neuropsychological level (Van Leijenhorst et al., 2010a). In addition, relatively poor performance on the IGT (predominantly preference for immediate gains) has been associated with self-reported reward sensitivity (Davis et al., 2007). A limitation related to the BGT is that only the first 71 cards were used instead of all 100 cards as in Bowman and Turnbull (2004). Nevertheless, the gambling ratio in the first and last blocks in our study revealed similar results (more gambling in the first block, and less gambling options in the last block) as the task used in Bowman and Turnbull (2004). Lastly, since temperament and parental psychopathology were selection criteria for this subsample (cf. Bouma et al., 2009), generalizability of results to other adolescent populations might be restricted. It should be noted that 34% of this sample was selected from the normal population, resulting in a sample slightly oversampled with adolescents at risk for behavioral and mental health problems. Simple t-test revealed only significant differences for ADHD and Oppositional Disorder assessed with the Youth Self-Report Scale (see Table 1 for more details).