Substance Use & Misuse, 50:376–386, 2015 C 2015 Informa Healthcare USA, Inc. Copyright ISSN: 1082-6084 print / 1532-2491 online DOI: 10.3109/10826084.2014.984847
ORIGINAL ARTICLE
Subst Use Misuse Downloaded from informahealthcare.com by University of Missouri Kansas City UMKC on 04/01/15 For personal use only.
Development of Risk Perception and Substance Use of Tobacco, Alcohol and Cannabis Among Adolescents and Emerging Adults: Evidence of Directional Influences Dennis Grevenstein, Ede Nagy and Henrik Kroeninger-Jungaberle Institute of Medical Psychology, University Hospital Heidelberg, Heidelberg, Germany Social Cognitive Theory (Bandura, 1977), the Health Belief Model (Rosenstock, Strecher, & Becker, 1988; Strecher, Champion, & Rosenstock, 1997), or the Theory of Planned Behavior (Ajzen, 1991). The concept of risk perceptions is commonly used in many areas of health psychology, including vaccination (Brewer et al., 2007) and sexually transmitted diseases (Ijadunola, Abiona, Odu, & Ijadunola, 2007). Risk perceptions have been linked to substance use as well (Thornton, Baker, Johnson, & Lewin, 2013), notably to alcohol consumption (Chomynova, Miller, & Beck, 2009; Lundborg & Lindgren, 2002; Miller, Chomcynova, & Beck, 2009; Sj¨oberg, 1998), smoking (Borrelli, Hayes, Dunsiger, & Fava, 2010; Gerking & Khaddaria, 2012; Song, Glantz, & Halpern-Felsher, 2009; Viscusi, 1991) and cannabis use (Apostolidis, Fieulaine, Simonin, & Rolland, 2006; Kilmer, Hunt, Lee, & Neighbors, 2007; Piontek, Kraus, Bjarnason, Demetrovics, & Ramstedt, 2013). Thus, risk perceptions are an important aspect of health behavior and specifically substance use in adolescence (Larsman, Ekl¨of, & T¨orner, 2012; Millstein & Halpern-Felsher, 2002). As a result, a lot of effort has been put into the creation of prevention programs to positively influence adolescent risk perceptions regarding the damaging outcomes of substance use (Soole, Mazerolle, & Rombouts, 2008). For example, warning labels on cigarette packages are supposed to inform users and increase their risk perception (Strahan et al., 2002). These labels, even though they may be accurate, essentially operate as fear appeals. They aim to persuade people by inducing fear of negative outcomes, ultimately leading to self-protective action (Floyd, Prentice-Dunn, & Rogers, 2000; Rogers, 1983). More positively framed, these practices should enable users to make informed choices with regard to their own substance use and its potentially damaging effects (Resnicow et al., 2002; Turoldo, 2009). The core problem with information based campaigns is that they have been
Background: While several studies have investigated the relationship between risk perception and substance use, surprisingly little is known about mutual influences between both variables over time. Objectives: The present study aimed to explore two different hypotheses separately for tobacco, alcohol and cannabis: influences from risk perception on behavior (motivational hypothesis) and influences from behavior on risk perception (risk reappraisal hypothesis). Methods: A prospective and longitudinal cross-lagged panel design was used with substance use and risk perception measured five times over the course of 10 years. Participants were 318 German youths aged 14–15 at the beginning of the study. Risk perception and substance use frequency were measured using self-reports. Results: Structural equation modeling indicated significant influences of risk perception on substance use behavior for all substances, which supports the motivational hypothesis. Changes in risk perception predict changes in future substance use of tobacco, alcohol and cannabis. Specifically for cannabis, influences of substance use on risk perception can also be shown, thus, supporting the risk reappraisal hypothesis. Conclusions: While there is support for the rationale behind adequate risk perception as a goal of preventive interventions, the possibility of risk reappraisal should not be neglected, especially regarding illicit substances. Keywords alcohol, tobacco, cannabis, risk perception, cross-lagged, longitudinal
Perceptions of risk and vulnerability have been discussed as a major part of decision making processes (Slovic, 1987; Weinstein, 1984). Assessing the consequences of any decision as well as one’s own vulnerability to problematic outcomes is also a central aspect of several formalized models in health psychology, including the
Address correspondence to Dennis Grevenstein, Institute of Medical Psychology, University Hospital Heidelberg, Bergheimer Str. 20, 69115 Heidelberg, Germany; E-mail: [email protected]
376
Subst Use Misuse Downloaded from informahealthcare.com by University of Missouri Kansas City UMKC on 04/01/15 For personal use only.
RISC PERCEPTION AND SUBSTANCE USE
shown to be little effective (Foxcroft & Tsertsvadze, 2011; Thomas, McLellan, & Perera, 2013). This may in part be due to the common difficulty of transforming intentions into actual behavior (Webb & Sheeran, 2006). Moreover, experimental studies have presented evidence that in some cases, negative or threatening information may in fact have opposite effects, such as defensive responses (Glock & Kneer, 2009), psychological reactance (Erceg-Hurn & Steed, 2011), or even completely adverse effects, such as a reduced risk perception (Myers, 2014). Pure knowledge about dangers and risk, as it is often provided by prevention programs or warning labels, was often found to have very little predictive value for actual risk behavior (Rosendahl, Galanti, Gilljam, & Ahlbom, 2005). Therefore, it is important to show that a change in general risk perceptions can manifest in actual behavior. There are several variations of conceptualizing risk perceptions (Brewer et al., 2007). Participants are commonly asked to rate risks either with regard to themselves or with regard to people in general. Many studies have shown differential associations between perceived risk and risk behavior, depending on the risk target. Perceived personal risk is commonly found to be smaller than perceived general risk, which was described as a self-serving bias (Weinstein, 1984). Additionally, personal risk perception was sometimes found to be positively correlated to risk behavior, such as alcohol use (Sj¨oberg, 1998). Even though subjectively perceived personal susceptibility may be a stronger predictor of risk behavior than a general assessment of risks, general risk perceptions are still a major target of public health interventions. The step from general risk perceptions, such as the knowledge about smoking being a risk factor for lung cancer, to a change in health behavior, that is quitting smoking, is rarely tested. The question arises whether risk perceptions are the cause of substance use or maybe its result. As described earlier, the former interpretation is most common. When people learn about risks, they are motivated to change their behavior or engage in preventive action to avoid negative outcomes, which has been referred to as the motivational hypothesis (Brewer, Weinstein, Cuite, & Herrington, 2004; Weinstein, 1993). An increase of risk perceptions should also lead to an increase of safety behavior or to a decrease of risky, possibly harmful behavior. There is however, another possibility. Risk behavior might influence the perception of how hazardous a certain behavior might be. Two possible outcomes can be the result of this reappraisal process. First, people who increasingly perform health threatening behaviors, might accurately realize that they are endangering their health. For example, when you start drinking alcohol on a regular basis, you may experience negative consequences and those could make you assess the riskiness and dangers of alcohol differently, which would ultimately lead to an increase of risk perception. The second possible outcome would be a systematic decrease of risk perceptions. Over the course of adolescence, which is characterized by several developmental tasks (Havighurst, 1972), young people make their own experiences with psychoactive substances. During
377
this process, substance use can be an important part of adolescent development (Silbereisen, Noack, & Reitzle, 1987) and can represent, for instance, an individual’s quest for autonomy from the (adult or peer) mainstream, achieving peer-group acceptance, or the development of coping strategies (Hurrelmann & Quenzel, 2012). The experiences made, may therefore be positive, rather than negative, which could lead to a decrease in risk perception. This could also be important with regard to illicit substances such as cannabis, which may be perceived as more risky due to its illicit nature, rather than in accordance with its scientifically proven danger. Additionally, there is the possibility of cognitive dissonance (Festinger, 1957). In this paradigm, people experience a feeling of unease or dissonance when their attitudes and behaviors do not match. If a person smokes cigarettes, but knows that smoking tobacco may actually increase her chance of getting lung cancer, she will experience dissonance. To resolve the situation she can either change her behavior (and quit smoking) or change her attitude (and “adjust” her perception of risks). The same process could explain why awareness of health risks tends to wear out. Often, people become aware that they are at risk, but fail to take pre-cautions. Even if you know that your consumption of alcohol has reached an unhealthy level, you may not easily be able to change your behavior at a given moment and subsequently lose interest in the issue or actively deny the risk. Therefore, if a person’s risk perception decreases after changing her behavior, it may be just the result of people’s need to justify their actions in order to make them feel at ease. Any effect of changes in people’s behavior on their risk perception can be termed the risk reappraisal hypothesis (Brewer et al., 2004). There have been general methodological problems associated with risk perception research. Weinstein, Rothman, & Nicolich (1998) discuss three major aspects: (1) The use of a longitudinal design is necessary to study change as implied by the motivational or self-protective hypotheses. A cross-sectional correlation cannot tell anything about the process of change and may in fact document the relative accuracy of risk perceptions. (2) Controlling for prior behavior is necessary to differentiate change and stability in prospective designs. (3) One has to acknowledge that the relationship between behavior and risk perception may not be constant, but is rather likely to change over time. This is an important issue since substance use is an age dependent phenomenon and serves multiple purposes over the course of adolescence (Silbereisen et al., 1987; Viscusi, 1991). The following study uses structural equation modeling to fulfill the aforementioned requirements. A longitudinal cross-lagged panel design was used which followed participants from the age of 14 years into young adulthood at the age of 24. Models were computed separately for tobacco, alcohol, and cannabis. Both possible directions of influence are of interest. Risk perceptions influencing subsequent substance use would reflect the motivational hypothesis. Influences of substance use behavior on later risk perception would reflect the risk reappraisal hypothesis.
378
D. GREVENSTEIN ET AL.
METHODS
Subst Use Misuse Downloaded from informahealthcare.com by University of Missouri Kansas City UMKC on 04/01/15 For personal use only.
Study Sample
The following research is part of a ten-year-longitudinal study of drug use patterns (RISA) conducted in the Rhine/Neckar metropolitan region in the South of Germany from 2003 to 2012. The study was approved by the ethics committee of the University Hospital Heidelberg (No. 218/2005). Informed consent and written permission from legal guardians were obtained. Participants were 318 students with a mean age of 14 at the beginning of the study. The sample included 164 female (51.6%) and 154 male (48.4%) participants. The study comprised 14 data collection events. 65.4% of the participants (n = 208) grew up in a traditional family, which was defined as living with both biological parents up to the age of 18 years. Level of education spanned equally across the three-tier German school system. The sample was ethnically diverse. Of all participants, 54.1% (n = 172) were of German nationality, 15.7% (n = 50) did not possess German nationality, while 30.2% (n = 96) did not provide that information. These data are comparable to the official census which denotes 19.3% of all students in southwest Germany having a migration background (Statistisches Landesamt Baden-W¨urttemberg, 2014). The sample can be characterized as rural or sub-urban with participants living in smaller cities up to 100,000 inhabitants. Sample attrition amounted to n = 134 (42.1%) over the course of ten years with signs of systematic drop out. At age 14–15, in comparison to participants remaining in the study until age 24, those who dropped out consumed moderately more tobacco, Ms = 3.59 vs. 2.70 (SDs = 2.49 vs. 2.11), t(248.14) = 3.30, p = .001, Cohen’s d = 0.39, more cannabis, Ms = 1.49 vs. 1.26 (SDs = 1.00 vs. 0.70), t(214.03) = 2.27, p = .024, d = 0.27 and had a lower tobacco risk perception Ms = 3.55 vs. 3.88 (SDs = 1.16 vs. 1.09), t(308) = −2.61, p = .009, d = 0.29. There have been no signs of systematic dropout with regard to socio-demographic variables. Despite noticeable sample attrition over the course of ten years, participant dropout was comparable to other studies on adolescents’ development (Honkinen et al., 2009). Measures
Socio-Demographic Variables Participants were asked to provide information on several demographic variables, including gender, age, nationality, type of school, and family situation. Subsequently, gender and family situation were chosen as covariates in the models, as neither nationality nor type of school were correlated with substance use or risk perception. Risk Perception Risk perception was measured using a single item: “how dangerous do you think is the consumption of this substance for people in general?” Answers were given separately for tobacco, alcohol and cannabis on 6-point scales with the end points marked (1) “harmless” and (6) “very dangerous.”
Substance Use Frequency The scale was adapted from the national survey on drug use among adolescents (BZgA, 2004). It is similar to the brief self-report drug use frequency measure provided by O’Farrell, Fals-Stewart, and Murphy (2003). Substance use frequency was measured using a single item question: “how often have you used this substance in the last 6 months?” Answers were given separately for tobacco, alcohol, and cannabis on a 7-point scale with the following options: (1) “not used in last 6 months,” (2) “1–2 times in the last 6 months,” (3) “3–5 times in the last 6 months,” (4) “1–3 times a month,” (5) “1–2 times a week,” (6) “several times a week” and (7) “several times a day.” Statistical Analysis
The descriptive data analysis was carried out using SPSS 20. Mplus 5.21 (Muth´en, 1998–2007) was used for Structural Equation Modeling (SEM). Using SEM (Hoe, 2008; Kline, 2011), all the study variables can be investigated at the same time in the same model. Therefore, it is possible to model interconnections and mutual influences of the variables as well as the development of variables over time while controlling for individual differences in prior behavior and initial covariation between variables. Most important are the diagonal (longitudinal, cross-lagged) paths from one type of variable to another type of variable at the next time point. Vertical (cross-sectional) paths between variables and horizontal (autocorrelative, longitudinal) paths within a variable are needed to control for statistical covariation. Thus, the diagonal, cross-lagged paths are partial regressions that allow estimating the unique predictive influence of a variable at a given time. We included cross-sectional covariation between risk perception and substance use at the beginning and at the end of the study to control for covariation between both types of variables. To model the longitudinal aspect, every variable at a given point in time was regressed on every variable at the preceding point in time. Covariates were also controlled for at the beginning of the study. The models included covaration paths with gender and family setting for both, substance use and risk perception, at the first data point. Controlling for covariates is only needed once at the beginning as all subsequent paths are partial regressions. To estimate whether the model accurately represented the empirical data, goodness-of-fit was evaluated by (1) the—ideally non-significant—χ 2 test (Bentler & Bonett, 1980) and as low as possible a χ 2 /df ratio, ideally as low as 2 (Tabachnick & Fidell, 2007). The χ 2 test is highly sensitive to sample size and therefore it is in most cases significant; (2) the comparative fit index (CFI) with values of .90/.95 and above indicating appropriate/good model fit (Bentler, 1990; Hu & Bentler, 1999); (3) the root mean square error of approximation (RMSEA) with values of .00–.05/.06–.08/.09–.10 indicating good/reasonable/poor model fit (Browne & Cudeck, 1993). Deviation from close fit (
ORIGINAL ARTICLE
Subst Use Misuse Downloaded from informahealthcare.com by University of Missouri Kansas City UMKC on 04/01/15 For personal use only.
Development of Risk Perception and Substance Use of Tobacco, Alcohol and Cannabis Among Adolescents and Emerging Adults: Evidence of Directional Influences Dennis Grevenstein, Ede Nagy and Henrik Kroeninger-Jungaberle Institute of Medical Psychology, University Hospital Heidelberg, Heidelberg, Germany Social Cognitive Theory (Bandura, 1977), the Health Belief Model (Rosenstock, Strecher, & Becker, 1988; Strecher, Champion, & Rosenstock, 1997), or the Theory of Planned Behavior (Ajzen, 1991). The concept of risk perceptions is commonly used in many areas of health psychology, including vaccination (Brewer et al., 2007) and sexually transmitted diseases (Ijadunola, Abiona, Odu, & Ijadunola, 2007). Risk perceptions have been linked to substance use as well (Thornton, Baker, Johnson, & Lewin, 2013), notably to alcohol consumption (Chomynova, Miller, & Beck, 2009; Lundborg & Lindgren, 2002; Miller, Chomcynova, & Beck, 2009; Sj¨oberg, 1998), smoking (Borrelli, Hayes, Dunsiger, & Fava, 2010; Gerking & Khaddaria, 2012; Song, Glantz, & Halpern-Felsher, 2009; Viscusi, 1991) and cannabis use (Apostolidis, Fieulaine, Simonin, & Rolland, 2006; Kilmer, Hunt, Lee, & Neighbors, 2007; Piontek, Kraus, Bjarnason, Demetrovics, & Ramstedt, 2013). Thus, risk perceptions are an important aspect of health behavior and specifically substance use in adolescence (Larsman, Ekl¨of, & T¨orner, 2012; Millstein & Halpern-Felsher, 2002). As a result, a lot of effort has been put into the creation of prevention programs to positively influence adolescent risk perceptions regarding the damaging outcomes of substance use (Soole, Mazerolle, & Rombouts, 2008). For example, warning labels on cigarette packages are supposed to inform users and increase their risk perception (Strahan et al., 2002). These labels, even though they may be accurate, essentially operate as fear appeals. They aim to persuade people by inducing fear of negative outcomes, ultimately leading to self-protective action (Floyd, Prentice-Dunn, & Rogers, 2000; Rogers, 1983). More positively framed, these practices should enable users to make informed choices with regard to their own substance use and its potentially damaging effects (Resnicow et al., 2002; Turoldo, 2009). The core problem with information based campaigns is that they have been
Background: While several studies have investigated the relationship between risk perception and substance use, surprisingly little is known about mutual influences between both variables over time. Objectives: The present study aimed to explore two different hypotheses separately for tobacco, alcohol and cannabis: influences from risk perception on behavior (motivational hypothesis) and influences from behavior on risk perception (risk reappraisal hypothesis). Methods: A prospective and longitudinal cross-lagged panel design was used with substance use and risk perception measured five times over the course of 10 years. Participants were 318 German youths aged 14–15 at the beginning of the study. Risk perception and substance use frequency were measured using self-reports. Results: Structural equation modeling indicated significant influences of risk perception on substance use behavior for all substances, which supports the motivational hypothesis. Changes in risk perception predict changes in future substance use of tobacco, alcohol and cannabis. Specifically for cannabis, influences of substance use on risk perception can also be shown, thus, supporting the risk reappraisal hypothesis. Conclusions: While there is support for the rationale behind adequate risk perception as a goal of preventive interventions, the possibility of risk reappraisal should not be neglected, especially regarding illicit substances. Keywords alcohol, tobacco, cannabis, risk perception, cross-lagged, longitudinal
Perceptions of risk and vulnerability have been discussed as a major part of decision making processes (Slovic, 1987; Weinstein, 1984). Assessing the consequences of any decision as well as one’s own vulnerability to problematic outcomes is also a central aspect of several formalized models in health psychology, including the
Address correspondence to Dennis Grevenstein, Institute of Medical Psychology, University Hospital Heidelberg, Bergheimer Str. 20, 69115 Heidelberg, Germany; E-mail: [email protected]
376
Subst Use Misuse Downloaded from informahealthcare.com by University of Missouri Kansas City UMKC on 04/01/15 For personal use only.
RISC PERCEPTION AND SUBSTANCE USE
shown to be little effective (Foxcroft & Tsertsvadze, 2011; Thomas, McLellan, & Perera, 2013). This may in part be due to the common difficulty of transforming intentions into actual behavior (Webb & Sheeran, 2006). Moreover, experimental studies have presented evidence that in some cases, negative or threatening information may in fact have opposite effects, such as defensive responses (Glock & Kneer, 2009), psychological reactance (Erceg-Hurn & Steed, 2011), or even completely adverse effects, such as a reduced risk perception (Myers, 2014). Pure knowledge about dangers and risk, as it is often provided by prevention programs or warning labels, was often found to have very little predictive value for actual risk behavior (Rosendahl, Galanti, Gilljam, & Ahlbom, 2005). Therefore, it is important to show that a change in general risk perceptions can manifest in actual behavior. There are several variations of conceptualizing risk perceptions (Brewer et al., 2007). Participants are commonly asked to rate risks either with regard to themselves or with regard to people in general. Many studies have shown differential associations between perceived risk and risk behavior, depending on the risk target. Perceived personal risk is commonly found to be smaller than perceived general risk, which was described as a self-serving bias (Weinstein, 1984). Additionally, personal risk perception was sometimes found to be positively correlated to risk behavior, such as alcohol use (Sj¨oberg, 1998). Even though subjectively perceived personal susceptibility may be a stronger predictor of risk behavior than a general assessment of risks, general risk perceptions are still a major target of public health interventions. The step from general risk perceptions, such as the knowledge about smoking being a risk factor for lung cancer, to a change in health behavior, that is quitting smoking, is rarely tested. The question arises whether risk perceptions are the cause of substance use or maybe its result. As described earlier, the former interpretation is most common. When people learn about risks, they are motivated to change their behavior or engage in preventive action to avoid negative outcomes, which has been referred to as the motivational hypothesis (Brewer, Weinstein, Cuite, & Herrington, 2004; Weinstein, 1993). An increase of risk perceptions should also lead to an increase of safety behavior or to a decrease of risky, possibly harmful behavior. There is however, another possibility. Risk behavior might influence the perception of how hazardous a certain behavior might be. Two possible outcomes can be the result of this reappraisal process. First, people who increasingly perform health threatening behaviors, might accurately realize that they are endangering their health. For example, when you start drinking alcohol on a regular basis, you may experience negative consequences and those could make you assess the riskiness and dangers of alcohol differently, which would ultimately lead to an increase of risk perception. The second possible outcome would be a systematic decrease of risk perceptions. Over the course of adolescence, which is characterized by several developmental tasks (Havighurst, 1972), young people make their own experiences with psychoactive substances. During
377
this process, substance use can be an important part of adolescent development (Silbereisen, Noack, & Reitzle, 1987) and can represent, for instance, an individual’s quest for autonomy from the (adult or peer) mainstream, achieving peer-group acceptance, or the development of coping strategies (Hurrelmann & Quenzel, 2012). The experiences made, may therefore be positive, rather than negative, which could lead to a decrease in risk perception. This could also be important with regard to illicit substances such as cannabis, which may be perceived as more risky due to its illicit nature, rather than in accordance with its scientifically proven danger. Additionally, there is the possibility of cognitive dissonance (Festinger, 1957). In this paradigm, people experience a feeling of unease or dissonance when their attitudes and behaviors do not match. If a person smokes cigarettes, but knows that smoking tobacco may actually increase her chance of getting lung cancer, she will experience dissonance. To resolve the situation she can either change her behavior (and quit smoking) or change her attitude (and “adjust” her perception of risks). The same process could explain why awareness of health risks tends to wear out. Often, people become aware that they are at risk, but fail to take pre-cautions. Even if you know that your consumption of alcohol has reached an unhealthy level, you may not easily be able to change your behavior at a given moment and subsequently lose interest in the issue or actively deny the risk. Therefore, if a person’s risk perception decreases after changing her behavior, it may be just the result of people’s need to justify their actions in order to make them feel at ease. Any effect of changes in people’s behavior on their risk perception can be termed the risk reappraisal hypothesis (Brewer et al., 2004). There have been general methodological problems associated with risk perception research. Weinstein, Rothman, & Nicolich (1998) discuss three major aspects: (1) The use of a longitudinal design is necessary to study change as implied by the motivational or self-protective hypotheses. A cross-sectional correlation cannot tell anything about the process of change and may in fact document the relative accuracy of risk perceptions. (2) Controlling for prior behavior is necessary to differentiate change and stability in prospective designs. (3) One has to acknowledge that the relationship between behavior and risk perception may not be constant, but is rather likely to change over time. This is an important issue since substance use is an age dependent phenomenon and serves multiple purposes over the course of adolescence (Silbereisen et al., 1987; Viscusi, 1991). The following study uses structural equation modeling to fulfill the aforementioned requirements. A longitudinal cross-lagged panel design was used which followed participants from the age of 14 years into young adulthood at the age of 24. Models were computed separately for tobacco, alcohol, and cannabis. Both possible directions of influence are of interest. Risk perceptions influencing subsequent substance use would reflect the motivational hypothesis. Influences of substance use behavior on later risk perception would reflect the risk reappraisal hypothesis.
378
D. GREVENSTEIN ET AL.
METHODS
Subst Use Misuse Downloaded from informahealthcare.com by University of Missouri Kansas City UMKC on 04/01/15 For personal use only.
Study Sample
The following research is part of a ten-year-longitudinal study of drug use patterns (RISA) conducted in the Rhine/Neckar metropolitan region in the South of Germany from 2003 to 2012. The study was approved by the ethics committee of the University Hospital Heidelberg (No. 218/2005). Informed consent and written permission from legal guardians were obtained. Participants were 318 students with a mean age of 14 at the beginning of the study. The sample included 164 female (51.6%) and 154 male (48.4%) participants. The study comprised 14 data collection events. 65.4% of the participants (n = 208) grew up in a traditional family, which was defined as living with both biological parents up to the age of 18 years. Level of education spanned equally across the three-tier German school system. The sample was ethnically diverse. Of all participants, 54.1% (n = 172) were of German nationality, 15.7% (n = 50) did not possess German nationality, while 30.2% (n = 96) did not provide that information. These data are comparable to the official census which denotes 19.3% of all students in southwest Germany having a migration background (Statistisches Landesamt Baden-W¨urttemberg, 2014). The sample can be characterized as rural or sub-urban with participants living in smaller cities up to 100,000 inhabitants. Sample attrition amounted to n = 134 (42.1%) over the course of ten years with signs of systematic drop out. At age 14–15, in comparison to participants remaining in the study until age 24, those who dropped out consumed moderately more tobacco, Ms = 3.59 vs. 2.70 (SDs = 2.49 vs. 2.11), t(248.14) = 3.30, p = .001, Cohen’s d = 0.39, more cannabis, Ms = 1.49 vs. 1.26 (SDs = 1.00 vs. 0.70), t(214.03) = 2.27, p = .024, d = 0.27 and had a lower tobacco risk perception Ms = 3.55 vs. 3.88 (SDs = 1.16 vs. 1.09), t(308) = −2.61, p = .009, d = 0.29. There have been no signs of systematic dropout with regard to socio-demographic variables. Despite noticeable sample attrition over the course of ten years, participant dropout was comparable to other studies on adolescents’ development (Honkinen et al., 2009). Measures
Socio-Demographic Variables Participants were asked to provide information on several demographic variables, including gender, age, nationality, type of school, and family situation. Subsequently, gender and family situation were chosen as covariates in the models, as neither nationality nor type of school were correlated with substance use or risk perception. Risk Perception Risk perception was measured using a single item: “how dangerous do you think is the consumption of this substance for people in general?” Answers were given separately for tobacco, alcohol and cannabis on 6-point scales with the end points marked (1) “harmless” and (6) “very dangerous.”
Substance Use Frequency The scale was adapted from the national survey on drug use among adolescents (BZgA, 2004). It is similar to the brief self-report drug use frequency measure provided by O’Farrell, Fals-Stewart, and Murphy (2003). Substance use frequency was measured using a single item question: “how often have you used this substance in the last 6 months?” Answers were given separately for tobacco, alcohol, and cannabis on a 7-point scale with the following options: (1) “not used in last 6 months,” (2) “1–2 times in the last 6 months,” (3) “3–5 times in the last 6 months,” (4) “1–3 times a month,” (5) “1–2 times a week,” (6) “several times a week” and (7) “several times a day.” Statistical Analysis
The descriptive data analysis was carried out using SPSS 20. Mplus 5.21 (Muth´en, 1998–2007) was used for Structural Equation Modeling (SEM). Using SEM (Hoe, 2008; Kline, 2011), all the study variables can be investigated at the same time in the same model. Therefore, it is possible to model interconnections and mutual influences of the variables as well as the development of variables over time while controlling for individual differences in prior behavior and initial covariation between variables. Most important are the diagonal (longitudinal, cross-lagged) paths from one type of variable to another type of variable at the next time point. Vertical (cross-sectional) paths between variables and horizontal (autocorrelative, longitudinal) paths within a variable are needed to control for statistical covariation. Thus, the diagonal, cross-lagged paths are partial regressions that allow estimating the unique predictive influence of a variable at a given time. We included cross-sectional covariation between risk perception and substance use at the beginning and at the end of the study to control for covariation between both types of variables. To model the longitudinal aspect, every variable at a given point in time was regressed on every variable at the preceding point in time. Covariates were also controlled for at the beginning of the study. The models included covaration paths with gender and family setting for both, substance use and risk perception, at the first data point. Controlling for covariates is only needed once at the beginning as all subsequent paths are partial regressions. To estimate whether the model accurately represented the empirical data, goodness-of-fit was evaluated by (1) the—ideally non-significant—χ 2 test (Bentler & Bonett, 1980) and as low as possible a χ 2 /df ratio, ideally as low as 2 (Tabachnick & Fidell, 2007). The χ 2 test is highly sensitive to sample size and therefore it is in most cases significant; (2) the comparative fit index (CFI) with values of .90/.95 and above indicating appropriate/good model fit (Bentler, 1990; Hu & Bentler, 1999); (3) the root mean square error of approximation (RMSEA) with values of .00–.05/.06–.08/.09–.10 indicating good/reasonable/poor model fit (Browne & Cudeck, 1993). Deviation from close fit (
