Predictors of binge drinking in adolescents: ultimate and distal factors - a representative study

Analysis of ultimate and distal predictors of binge drinking in a representative sample of German adolescents

The matter concerns a representative survey of 9^th graders in Germany conducted in 2007/2008. In the year 2006, there were 910,000 9^th graders in Germany. The goal was to survey 50,000 adolescents from different regions. The basis for the selection of the regions was the federal classification of rural districts and independent cities (urban districts), which total 440. The urban districts contain cities of each size (below 100,000 and up to 3.3 million (Berlin) inhabitants). The number of inhabitants in the rural districts also varies from about 50,000 to over 600,000. Therefore, the rural and urban districts (i.e., regions) were sorted into classes of region size from which the random drawing then took place. The classes of region size were: Western Germany (urban districts): cities with more than 500,000 inhabitants, cities with more than 100,000 inhabitants, cities with fewer than 100,000 inhabitants; Western Germany (rural districts): districts with more than 100,000 inhabitants, districts with fewer than 100,000 inhabitants; Eastern Germany (former GDR) (urban districts): cities with more than 100,000 inhabitants (there are only two cities with more than 500,000 inhabitants), cities with fewer than 100,000 inhabitants; Eastern Germany (former GDR) (rural districts): districts with more than 100,000 inhabitants, districts with fewer than 100,000 inhabitants; special case: Berlin. With the knowledge about the number of 9^th graders in each class of region size (from the official education statistics) and the goal of 50,000 adolescents to be questioned, it was possible to calculate how many adolescents per class of region size had to be included. Note that classes were drawn by chance, but students were not. The number of 50,000 students refers to a goal of 2,500 classes. It was known that about 20 students per class could be retrieved and used for data analysis. The number of 2,500 classes was chosen in such a way that for every region in Germany that was supposed to be represented in the survey, a sufficient number of classes was evident. The goal was to match the distribution of the 9^th graders in the classes of region size (in the population) to the same percentage in the sample. It was assumed that every 2^nd student (in large cities, every 6^th student) in a drawn region would be questioned. Thus, we could calculate how many regions had to be drawn out of every class of region size. These steps resulted in 61 regions. Which region was chosen to take part was then drawn by chance in order to secure a representative sample. At the Criminological Research Institute of Lower Saxony, the sample was drawn stratified by school type (on the basis of school lists provided by the local education authorities). A master list in which all school classes (9^th grade) of one region were consecutively sorted was used. Then all directors of the drawn schools were informed in writing about the survey and asked for the participation of their 9^th-grade school classes. If the directors agreed to the survey, information material including consent forms for parents were sent to the schools. On an appointed day, the written survey was administered without the students whose parents refused participation, who themselves refused to participate, or who were otherwise busy or absent during the survey. The survey at the school was carried out by trained external study assistants - not by the employees of the schools - in order to preserve reliability and validity.

The research project was granted by the Federal Ministry of the Interior in Germany; thus, a statement of an ethics committee was not necessary. Instead, the survey was audited by each Ministry of Education of every German state (Bundesland) and additionally of every state responsible for data protection. The survey then actually took place only in those states in which the survey was permitted after following this procedure. A further ethics committee was not included since the data protection matters were covered by the above described procedure and no other intervention besides filling out an anonymous questionnaire was applied. One manuscript based on this data set has already been published. That study concerns epidemiological data on binge drinking according to differences in urban and rural areas and concerning migration background [4].

The item assessing heavy episodic drinking (binge drinking) was derived from the representative survey of adolescents of the German Federal Center for Health Education [14]. Binge drinking is defined as the consumption of five or more standard drinks at one drinking opportunity. The adolescents were asked a) if they had consumed alcohol in the last 30 days (30-day-prevalence) and if yes, b) on how many days they had consumed 5 or more standard alcoholic drinks in a row. The answer categories were a) yes/no and b) not on one day, on one day, on two days, (...), on 20 or more days.

The following paragraphs describe the variables that were chosen to operationalize the constructs of Petraitis et al. [16] (see Table 2).

The following potential predictors were measured:

1. Acknowledgment of success/rewards by parents: The students were asked whether they achieved something to be proud of in the last 12 months in the areas sports, music, friends, family, school, computer games, society, or work and from whom they received the acknowledgment. A sum score of fatherly and motherly acknowledgment across the eight areas was built. The items were constructed by the Criminological Research Institute of Lower Saxony.

2. Parental warmth in childhood: A scale based on the concept of parental style by Baumrind [21] (translated by Wilmers et al. [22]) was used. It consists of six items exploring parental warmth in childhood for mother and father separately. Cronbach's alphas were .86 (motherly warmth) and .90 (fatherly warmth). A sum score was used for parental warmth.

3. Parental control/supervision in adolescence: A scale based on the concept of parental style by Baumrind [21] (translated by Wilmers et al. [22]) was used. It consists of three items exploring parental control and supervision in adolescence in the last 12 months for mother and father separately. Cronbach's alphas were .76 (motherly control) and .80 (fatherly control). A sum score was used for parental control.

4. Parental separation events: The students were asked whether their parents were separated or divorced or whether their mother or father had died. If one of the items was answered yes, the student received a 'positive' parental separation score. The items were constructed by the Criminological Research Institute of Lower Saxony.

5. Cultural communication in the family: A scale of two items developed by Kunter et al. [23] based on the theory of cultural capital by Bourdieu [24] was used. These items explore whether it is usual for the student's family to talk about political or social questions and whether it is usual to talk about books, movies, or TV broadcasts. Cronbach's alpha of the scale was .77. The sum score of the two items was used.

6. Number of friends: The students were asked about the number of friends with whom they spend time outside of school. The item was constructed by the Criminological Research Institute of Lower Saxony.

7. Number of delinquent friends: The number of friends known by the student with at least one delinquent behavior in the last 12 months was assessed. Five different delinquent behaviors were listed (e.g., selling illicit drugs). A sum score of delinquent friends was built. The item was constructed by the Criminological Research Institute of Lower Saxony.

8. Deviant/Assimilated behavior of one's own group of friends: Different delinquent behaviors in the group of friends of the student, including dealing drugs, were assessed with five items, and socially acceptable adolescent group behaviors that don't break laws were assessed with two items. The items were formulated in the "we"-perspective meaning the student is engaged actively or passively in the behavior himself. A sum score for deviant behavior and a sum score for assimilated behavior were used. The items were constructed by Wetzels et al. [25].

9. Smoking parents: We asked whether the student's mother or father regularly smokes. If one or both parents engage in smoking, the student received a 'positive' value on this variable.

10. Community/neighborhood cohesion: A scale developed by Sampson et al. [26] (translated by Oberwittler [27]), consisting of five items, was used. A sample item is "People in my neighborhood help each other." Cronbach's alpha of the scale was .78. The sum score was used for the analysis.

11. Community/neighborhood/school safety: A scale developed by Wilmers [22], consisting of five items, was used. A sample item is "How safe do you feel when you are at home in your apartment?" Cronbach's alpha of the scale was .75. The sum score was used for the analysis.

12. Welfare status: The students were asked whether their parents or they themselves lived on social welfare (unemployment pays "Hartz IV" welfare aid according to German social legislation). If they answered yes (versus no or I don't know) the student received a 'positive' welfare status score. The item was constructed by the Criminological Research Institute of Lower Saxony.

13. Violence level in the school: The construct was assessed with two items developed by Wilmers et al. [22] asking for violence in the school and fights and trouble among the students. A sum score of the two items was used.

14. Willingness of teachers to intervene during violent conflicts: The construct was assessed with two items developed by Olweus [28] asking whether teachers intervene when students fight violently and whether teachers prefer to look the other way if brawling among students occurs. A sum score of the two items was used.

15. Violence/problems at school - aggressive behavior of teachers: Three items describing verbal assaults and violent behavior of teachers against the students were used. The student was asked whether he had ever experienced one or more of those behaviors. If the student answered yes, he received a 'positive' score. The items were constructed by the Criminological Research Institute of Lower Saxony.

16. Non-profit volunteer activities: The students were asked for six different non-profit volunteer activities (e.g., working as a trainer for children) concerning their current involvement. An involvement score was built across the six areas. Past involvement was not counted. The item was constructed by the Criminological Research Institute of Lower Saxony.

17. Religiosity: The construct was assessed with a single item by Wetzels et al. [29]: "How important is religion for you personally?" which could be answered on a scale with five levels.

18. School commitment: Bonding to school was assessed with two items asking how much a student likes to go to school and how strongly he agrees with the statement that he really likes his school. A sum score of the two items was used. They were constructed by the Criminological Research Institute of Lower Saxony.

19. Social integration in school: The extent to which a student is integrated and accepted at school was assessed with two items asking for a self-rating of one's popularity with other students and the self-rated estimation of having lots of friends at school. Both items were rated on a 4-point graduated scale; a sum score of the two items was used. They were constructed by the Criminological Research Institute of Lower Saxony.

20. Social Desirability/Conventional Values: The construct was assessed with the revised version of the Social Desirability Scale by Crowne & Marlowe [30]. The German version of the scale was developed by Lück & Timaeus [31]. The scale consists of four items, with a 4-point graduated rating and is intended as a sum scale. However, reliability analysis in this sample showed an unsatisfying internal consistency with Cronbach's alpha = .20. A subsequent factor analysis showed that the four items loaded on two separate factors. Therefore, the four items were dichotomized, and an index with possible values from 0 to 4 was built. Higher values indicate higher social desirability.

21. Planned type of school leaving certificate: A single item with three answer categories was used to assess the planned type of school leaving certificate. According to the German school system, it was possible to choose between special school/secondary general school certificate (9 years) "Hauptschulabschluss," secondary modern school certificate (10 years) "Realschulabschluss," or general qualification for university entrance/hiqh school diploma "Abitur." The item was constructed by the Criminological Research Institute of Lower Saxony.

22. Absenteeism/Truancy: Students were asked to indicate whether the item "I have so far never been truant a whole day" was true for them. All students who did not check the item received a 'positive' truancy score. The item was constructed by Wilmers et al. [22].

23. Hedonistic reasons for truancy: Those students who admitted to having been absent without excuse (truancy) at least for one school lesson or one school day in the last half year were asked for the reasons. Two reasons displaying hedonistic attitudes were used for the analysis: "because I wanted to sleep in" and "because I was not in the mood for school." Students answering yes to one or both items received a 'positive' hedonistic values score. The items were constructed by Wilmers et al. [22].

24. Attention deficit disorder: The presence of an attention deficit disorder, which has high impulsivity as a diagnostic criterion, was asked with a single item developed by the Criminological Research Institute of Lower Saxony. The student had to answer whether a psychologist or a doctor had ever diagnosed an attention deficit disorder.

25. Risk-taking behavior: Risk-taking was assessed based on the concept of Grasmick et al. [32] with a four-item scale in a German translation by Wilmers et al. [22]. The internal consistency measured by Cronbach's alpha was satisfying (α = .85). A sum score across the four items was used for analysis.

26. School grades: A mean school grade was computed for the three self-stated school grades in Math, German, and History. The item assessing the school grades was constructed by the Criminological Research Institute of Lower Saxony.

27. Self-esteem: The construct was assessed with a scale developed by Ravens-Sieberer et al. [33] and is part of the KINDL questionnaire, which assesses health-related quality of life in children and adolescents with a total of six dimensions. The dimension self-esteem consists of four items showing a Cronbach's alpha of .61. A sum score across the four items was used for analysis. A sample item is: "In the last week, I was proud of myself."

28. Mental well-being/mood: The construct was assessed with a scale developed by Ravens-Sieberer et al. [33], and is also part of the KINDL questionnaire. The dimension mental well-being/mood consists of four items showing a Cronbach's alpha of .56. A sum score across the four items was used for analysis. The following is a sample item: "In the last week, I felt lonely."

29. School anxiety: The construct was assessed with a scale developed by Wilmers et al. [22] consisting of five items, for example: "I often cannot fall asleep because I am worried about school." The internal consistency measured with Cronbach's alpha was .79. A sum score across the five items was used in the analysis.

30. Suicidal thoughts: This aspect was assessed with a single item asking how often the student had already thought about suicide. The item was constructed by the Criminological Research Institute of Lower Saxony.

31. Mandatory repetition of school year: An item assessing a German specificity of the school system was included. It is possible that because of weak academic skills, a student is forced to repeat a whole school year. This aspect was assessed with a single item developed by the Criminological Research Institute of Lower Saxony: "Did you ever have to repeat a class?"

A total of 3,052 classes (9^th grade) were drawn. For 921 classes, the directors/main class teachers refused to participate. 2,131 classes participated. Actually, the 2,131 classes included 50,708 students, but 6,098 of them did not participate (reasons, for example: parents' refusal or absenteeism). Therefore the total sample size was 44,610 students (return rate 88%). Figure 1 comprises a detailed flow-chart of the sample record.

The return rates (students, without director refusal) differed between the school types in that grammar/secondary schools as well as private/not state-run schools had the highest return rates (92.0/92.8) and special schools the lowest (75.5). Furthermore, the return rates differed across the classes of region size. In the large cities, the return rate was lower in comparison with rural districts and urban districts with fewer than 500,000 inhabitants. In spite of the varying return rates in the different classes of region size, the final sample represented the proportions of the population very well (e.g., students living in cities with more than 100,000 inhabitants in Western Germany: 12.04% in the sample and 11.68% in the population). The proportion of students in the 9^th grade in every class of region size in Western and Eastern Germany was compared to their proportion in the sample. With those two percentages for each category, the reliability can be seen and rated. The proportions never differed more than 0.36% between population and sample in the different classes of region size except for Berlin where the difference was 0.62%.

To address the varying return rates, weighting factors were calculated so that the proportion of school forms in the sample corresponded to that in the population, and in the same manner, the proportion of regions with different sizes in the sample corresponded to that in the population. The two weighting factors were multiplicatively connected when the data from the total sample were analyzed. Thereby the imbalances regarding the school forms were eliminated, as were the much smaller imbalances regarding the classes of region size.

The sample can be characterized as follows: 51.3% of the sample was male, the mean age was 15.3 (SD 0.7) years. The percentage of adolescents with a migration background was 27.4%, whereby students with a Turkish migration background constituted the largest group (6.0%; more than 2,600 students) followed by emigrants from the former Soviet Union states (5.8%; more than 2,500 students). A total of 12.2% lived in large cities with more than 500,000 inhabitants including Berlin, whereas the majority lived in rural districts (68.8%). The migration background varied between 39.9% in large cities with more than 500,000 inhabitants and 23.9% in rural districts.

After operationalization of the six defined variable groups by Petraitis, the resulting 31 variables were analyzed concerning multicollinearity. The goal was to get a lean but well operationalized model. We determined that variables with a medium (r > .5) or even high (r > .7) correlation with other variables needed to be reduced. Taking only the sample size into account, it would have been possible to include a large number of predictors. According to Altman [34], the number of independent variables used should not exceed the square root of the sample size (here n = 44,610; potential predictors > 200). However, we decided that a lean model was still a priority. Correlation coefficients were computed according to the measurement level of the variables, for example, Pearson's r for metric variables, Cramer's V (Phi) for categorical variables, etc. In consequence of the multicollinearity analysis, three pairs of variables showed a medium or high correlation: a) parental warmth in childhood with parental control/supervision in adolescence (r = .737). We decided to keep the variable parental warmth in childhood because more single items were used for the sum score of the scale than in the parental control construct and therefore variance in the variable parental warmth was higher. As a consequence, parental control in adolescence was not included as a predictor in the multivariate analysis. b) Absenteeism/Truancy with Hedonistic reasons for truancy (Phi = .516). Since hedonistic reasons for truancy is logically subordinate to absenteeism, and furthermore, hedonistic reasons were the most frequently named reasons for truancy, the item hedonistic reasons for truancy was omitted from the multivariate analysis and the item Absenteeism/Truancy was kept in the analysis. c) Number of delinquent friends with Deviant behavior in one's own group of friends (r = .601); The variable deviant behavior in one's own group of friends was kept in the analysis because the variable was comprised of a sum score built across several items, therefore having more variance than a single item like number of delinquent friends, which was omitted from the multivariate analysis.

The remaining variables were included as predictors in a multiple binary logistic regression analysis with binge drinking as the dependent variable. The independent variables were included in the regression equation by the enter method. As a measure for explained variance of the model, R² according to Nagelkerke, was used. Statistical analysis was performed with PASW 18.0. Because of the sample size, the level of significance was set to p < .001 [35];

however, statistical significance is not equivalent to clinical relevance, especially in large samples [36‐38]. Therefore, Odds Ratios and their confidence intervals were also used for interpretation of results. Even though in general the use of Odds Ratios in comparison to other effect sizes has been discussed in the literature, it has been explicitly suggested that for logistic regressions [39], the interpretation of Odds Ratios in the sense of risk can be done safely when effect sizes are not large. In these cases, the interpretation of Odds Ratios is unlikely to lead to qualitatively different judgments about the study results [40]. Since the largest effects in our study showed a risk of increasing or decreasing below 40%, the over interpretation of effects by interpreting Odds Ratios in the sense of relative risks was relatively small according to the table in the publication of Davies et al. [40]. There is no published predefined level of Odds Ratio clearly indicating relevance of a predictor for all kinds of studies. Rather, clinical relevance has to be defined by experts, which in this case were the study authors [36‐38]. We decided to interpret a predictor as clinically relevant in our study if the Odds Ratio was higher than 1.2 or smaller than 0.8 in combination with a p-value below .001. Predictors which change the risk to an OR of at least 1.1 resp. 0.9 at a p-level of < .001 are further interesting to consider being on the threshold to clinical relevance.

Missing values were evident in less than 2% of the cases across the chosen variables. Given the sample size of 44,610, they could have been ignored. However, we chose to impute the missing values conservatively in order to have the full sample included in the regression analysis. This means that if a student did not answer a certain item, the item was given the zero or "no" value; for example, if the item for parental separation was not answered, it was counted as "no" for parental separation events. This handling was used for the variables parental warmth and control, parental separation events, cultural communication in the family, number of (delinquent) friends, smoking parents, neighborhood cohesion and safety, living on welfare, violence level in school, willingness of teachers to intervene, aggressive behavior of teachers, volunteer activities, school commitment, social integration in school, planned school leaving certificate, hedonistic reasons for truancy, ADHD, risk-taking behavior, self-esteem, mental well-being, and mandatory repetition of school year. Only items for which social desirability could have been a reason for the missing value - because the item asked, for example, for something that was inconsistent with conventional norms - were missing values imputed with the mean value of the students who answered the item. This worked only when a metric variable was evident. This latter imputation method concerns the variables deviant/assimilated behavior in one's own group of friends, social desirability, average school grades, and school anxiety.

In contrast to anxiety, which was not relevantly or not positively associated with binge drinking in our study and in previous studies [41], the presence of depression and emotional problems are known to be positively associated with binge drinking [42] as we discovered also in our study, in which the presence and frequency of suicidal thoughts were strong predictors of binge drinking. Studies that explicitly included suicidal thoughts as predictors of substance use support our results also, even though they were not specifically aimed at binge drinking [43, 44].

Concerning the factors associated with school and academic achievement, the literature shows a clear picture. As in our study, binge drinking has been associated with academic failure (worse school grades and mandatory repetition of a school year) [45, 46], which could be the consequence of memory and neurocognitive deficits on the one hand, which have shown to be side effects and direct consequences of binge drinking [47‐50]. On the other hand, binge drinking can also be a coping mechanism for dealing with academic failure in comparison to other peers [51, 52].

Other predictors were significant but still surprising. For example, the finding that living on welfare is rather a protective factor for binge drinking even though it is known that living on welfare or poverty in general is a risk factor for many other health factors such as obesity, diabetes, stress, or kidney disease [53‐57]. However, comparable to our study, Bellis et al. [58] found in their sample of 11,000 students in England in the same age group as investigated here that children with greater expendable incomes report more unsupervised, frequent, and heavy drinking. The same was found for Spanish adolescents [59]. Also, another study showed that early alcohol use by adolescents is more frequent in higher-income households [60]. The most obvious explanation for this effect may be that binge drinking causes financial expenditures that can rarely be afforded by teenagers with a smaller budget.

There have been three studies identifying religiosity or spirituality as a protective factor for binge drinking [61‐63] as our data also suggest. However, they were all carried out in the U.S. where religion plays a far greater role in daily life than in Germany, especially for younger people. They also showed that a protective effect is evident only for heterosexual adolescents [62] and when religion reaches a certain degree of personal importance in the life of the individual student.

The thesis that binge drinking is a social act or even constitutes a spare-time activity is supported by the finding that social integration and number of friends is also positively associated with engaging in binge drinking in our study, suggesting that heavy alcohol consumption is something that students do together as a collective [see also [64, 65] and that binge drinking does not represent a consumption pattern of boredom in Germany as found in England [58]. Our results are supported by a recent study on 26,000 Spanish adolescents for whom social integration in a group of friends who frequently go out and get drunk is predictive of habitual excessive drinking [59].

Concerning intrapersonal factors, Wechsler et al. [65] found, as we did, that risk-taking behavior is positively associated with binge drinking in U.S. college students. Also, Neumark et al. [66] showed that different risky behaviors are associated with binge drinking in adolescents from different cultural/religious backgrounds.

The risk of adolescent binge drinking due to parental separation has not been explicitly shown so far. Only in adults a person's own divorce has been shown to be a significant risk factor for binge drinking [67, 68]. As we have shown, a risk behavior such as binge drinking in adolescents can already be a consequence of experiences with parental divorce and therefore of a change or even loss in the own family - and social support system. Thus, binge drinking after parental separation could be interpreted as a coping mechanism for this critical life event.

Since there seem to be associations between delinquent behavior, deviant behavior in a person's own peer group, and truancy/absenteeism, these concepts and their roles in the literature are discussed together. These variables are among the strongest predictors of binge drinking in our study. Also, other authors have found that engaging in deviant behaviors (together with antisocial peers [69]) foretells binge drinking in adolescents [51, 52, 69] and is additionally often associated with truancy and "hanging around during the day" [51, 59, 66, 70]. Remarkably, the association of binge drinking with truancy has been shown in different European countries and provides support for our results in Germany. It is also known from tobacco research that the number of friends and their substance-related behavior, even if it is against the social/legal norms, have strong impacts on the consumption behavior of the adolescent himself [71].

The relatively strong impact of violence at school, in particular, in the form of aggressive behavior of teachers - be it verbally or other - was rather surprising. There have been no published studies known to the authors thus far that have investigated this construct and its role in consumption behavior. Thus, this finding is new and adds an additional piece of knowledge to understanding excessive consumption behaviors in adolescents for the scientific community. There are different possible ways to interpret this result: 1. Adolescence as a developmental phase is characterized by a high vulnerability for being hurt. In the case of low frustration tolerance, a slight or offensive behavior by someone higher in the hierarchy could result in problematic behavior such as binge drinking, through which the person regains his/her strength and reputation in the peer group (compensation mechanism). 2. It is also plausible that teachers know which adolescents engage more in deviant behavior (like truancy) or problematic risky behavior (like binge drinking) than others because of the students' academic achievements. It could be that either consciously or unconsciously, the teachers react more aggressively toward those adolescents.

Trying to integrate the results into the theory of triadic influence suggested by Petraitis et al., who developed the model for different substances ("Illicit Substance Use"), it can be seen that almost all intrapersonal distal factors are important for predicting binge drinking, but not all ultimate intrapersonal factors are. It has to be taken into account that only part of the suggested ultimate intrapersonal factors could be operationalized. Concerning the attitudinal/environmental factors, all ultimate factors named by Petraitis showed predictive power also for binge drinking, even though the policy factors were not operationalized since they are the same for all analyzed adolescents in the sample because of the federal German alcohol policy. The distal attitudinal/environmental factors named by Petraitis could not all be proven to be predictive of binge drinking: Whereas commitment to school and religion, as well as hedonic values and tolerance of deviance showed statistical significance, the commitment to conventional values did not serve as a predictor, and engagement in non-profit volunteer activities showed a different direction of association, as expected. For the social/interpersonal distal factors, the model of Petraitis was basically supported for binge drinking; for the ultimate factors, however, only part of the constructs are relevant for binge drinking. Whereas home strain, parental separation, and unconventional values and behaviors of parents were statistically significantly associated with binge drinking, factors such as parental warmth or acknowledgment of success by parents did not have significant predictive power. It has to be taken into account that the chosen and analyzed items are the best approximation to the theoretical model of Petraitis that was possible for us. The study was not designed in the first place to prove this specific model, and therefore not all influence factors could be operationalized. However, out of every distal and ultimate category of the model, several variables were available to operationalize the constructs.

It is true that the cross-sectional study presented is based on a representative sample, however, there are limitations concerning the composition of variables and the study design. Variables asking for adolescents' beliefs referring to Petraitis et al. proximal influence factors were not assessed. Therefore the indisputable impact of these variables on binge drinking cannot be described in the context of our variables. Furthermore the study is limited on 9^th graders. Thus we cannot compare our results with younger or older adolescents. Our survey is a cross-sectional study, for this reason we have no information on changes in binge drinking behaviors and their influencing factors.

The prevalence of binge drinking in the surveyed sample of 9^th graders lies, on the one hand, in the expected range (20% to 60%) that is reported by other large studies [12, 13, 15]. On the other hand, a representative investigation in Germany of 12-to-17-year-olds reported much lower levels of Binge Drinking just as was found by a review in the US investigating Binge Drinking in 12-to-20-year-olds [5, 14]. An explanation could be that in our study, only 9^th graders with a mean age of 15 years were included, and the standard deviation of the currently surveyed age group was a lot smaller (SD = 0.7). Obviously, this age group is a lot more engaged in substance consumption activities than 12-year-olds. Still, the percentages of 56.9% for males and 47.5% for females are likely to be valid and probably not overestimated. The analyses of the study were based on a large representative sample of adolescents, suggesting validity of the data. However, it has to be taken into account that the adolescents who refused to participate probably engage at least to the same percentage as their participating colleagues in alcohol consumption, maybe even more, since they were unwilling to disclose this behavior. It could therefore even be possible that the number of consumers or specifically binge drinkers is slightly underestimated. The fact that the prevalence of binge brinking is higher in boys than in girls was also found in other studies e. g. [72]. This representative sample supports that difference in consumption patterns.

Whereas some of the influence factors for binge drinking are not surprising since they are known predictors of substance abuse in general (e.g., risk-taking behavior), there are two points that could be targeted in interventions that should not focus on adolescents alone: 1. training teachers in positive, reassuring behavior and constructive criticism which has also been proofed as effective pedagogic strategy in intervention studies (for example [73]) and 2. a focus on high risk students either because they lack coping strategies when in negative mood or because of their low academic achievements and absenteeism from school. An interesting fact is also the protective value of religiosity, which so far has not been the focus of prevention strategies especially not in a country like Germany.

Further research on Binge Drinking in adolescents should try to include all facets of Petraitis' model as predictors or be theoretically based on alternative models. Furthermore, research should focus on adolescents of different age groups and aim for longitudinal studies.