UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Development and prediction of adolescent smoking and drinking Maggi, Stefania 2005

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
831-ubc_2005-104139.pdf [ 6.02MB ]
Metadata
JSON: 831-1.0099804.json
JSON-LD: 831-1.0099804-ld.json
RDF/XML (Pretty): 831-1.0099804-rdf.xml
RDF/JSON: 831-1.0099804-rdf.json
Turtle: 831-1.0099804-turtle.txt
N-Triples: 831-1.0099804-rdf-ntriples.txt
Original Record: 831-1.0099804-source.json
Full Text
831-1.0099804-fulltext.txt
Citation
831-1.0099804.ris

Full Text

D E V E L O P M E N T A N D P R E D I C T I O N O F A D O L E S C E N T S M O K I N G A N D D R I N K I N G by STEFANIA MAGGI B.A. (Honours), The University of Padua, Italy, 1993 M.A., The University of Toronto, 1996 A THESIS SUBMITTED IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in THE FACULTY OF GRADUATE STUDIES (Health Care and Epidemiology) THE UNIVERSITY OF BRITISH COLUMBIA January 2005 © Stefania Maggi, 2005 Abstract This study examines the development and etiology of cigarette smoking and alcohol drinking among Canadian children and adolescents. The National Longitudinal Survey of Children and Youth (NLSCY) was employed first to identify the developmental trajectories of smoking and drinking and then to identify patterns of early prediction of occasional smoking, daily smoking, and infrequent drinking. Utilizing data from participants between 10 and 17 years of age, growth mixture modelling was used to identify sub-groups of children and adolescents following similar developmental trajectories in the acquisition of smoking (N=2,886) and drinking (N=2,i8i). Results of this portion of the study indicate that there is no one 'universal' way in which adolescent smoking and drinking develops over time; rather, acquisition of smoking and drinking is better characterized by a number of distinct developmental pathways. Employing a different NLSCY sample of 1,414 (for smoking) and 1,422 (for drinking) 14- and 15-year-old participants, multinomial and logistic regression analyses were used for the data reduction and modelling of the early predictors of smoking and drinking. Results of these analyses indicate that while the early pattern of prediction of occasional smoking was found to be associated with family interpersonal factors such as parenting and family functioning, the early pattern of prediction of daily smoking was found to be associated with individual level factors such as hyperactivity/inattention and poor health. This led to the development of the 'smoking as coping' model where two causal pathways are hypothesized: the self-medication pathway explaining daily smoking and the social-compensation pathway explaining occasional smoking. In contrast, we found that infrequent drinking could not be predicted by any individual or interpersonal risk factors, suggesting that it may simply be a normative behaviour among young adolescents. Interestingly, we found that children who did not drink at all were more likely to" have had asthma and to be physically aggressive. These results contradict the dominant view where substance use, even in moderation, is believed to be associated with negative predictors. T a b l e o f C o n t e n t s Abstract i i Table of Contents iii List of Tables v List of Figures vi Preface vii Acknowledgements x Chapter I: Identifying Developmental Trajectories of Smoking Behaviours from Late Childhood to Early Adolescence • l 1.1 Introduction i 1.2 Method 4 1.2.1 Participants 4 1.2.2 Measures 7 1.2.3 Analytical approach 8 1.3 Results 10 1.3.1 Trying smoking 10 1.3.2 Frequency of smoking 11 1.3.3 Intensity of smoking 12 1.3.4 Gender, trying, frequency, and intensity of smoking 13 1.3.5 Trying, frequency and intensity of smoking 14 1.4 Discussion 14 Chapter II: Exploring the developmental course of early drinking behaviours of Canadian children and adolescence 20 2.1 Introduction 20 2.2 Methods 23 2.2.1 Participants 23 2.2.2 Measures 26 2.2.3 Analytical approach 27 2.3 Results 28 2.3.1 Tried drinking 28 2.3.2 Frequency of drinking 28 2.3.3 Number of times being drunk 30 2.4 Discussion 3 1 Chapter III: Theories of normative risk-taking and problem behaviours: searching for the early predictors of smoking and drinking 37 iii 3. i Introduction 37 3.1.1 Risk-taking as normative behaviour 37 3.1.2 Risk-taking as a manifestation of problem behaviours 40 3.1.3 The gap between theory and prevention practices 45 3.1.4 Recent advances in research and theory 48 3.1.5 Rationale and purposes of the present study 49 3.2 Method 53 3.2.1 Data Source: the National Longitudinal Survey of Children and Youth 53 3.2.2 Data Collection 53 3.2.3 Participants 55 3.2.5 Predictors 57 3.2.6 Analytical Plan 57 3.2.7 Data reduction procedure 59 3.3 Results 62 3.3.1 Early Predictions of Occasional and Daily Smoking 63 3.3.2 Early Prediction of Infrequent Drinking 7 1 3.4 Summary of Results 74 3.5 Discussion 76 Chapter IV: Conclusion 86 4.1 Conclusion and recommendation for future work 86 Chapter V : Limitations 89 5.1 Limitations 89 5.1.1 Limitations of the NLSCY 89 5.1.2 Limitations associated with analytical strategies 91 References 93 Appendix I: Content of the first assessment of the National Longitudinal Survey of Children and Youth conducted i n 1994/95 1 Q 4 Parents/Other Family Members 104 Family 104 Community 1 Q 4 School 105 Appendix II: Predictors F r o m W h e n Participants Were 8-9 Years of Age 107 Appendix III: Predictors of smoking that were eliminated through the multinomial regression data reduction procedure 110 Appendix IV: Predictors of drinking that generated n o n significant results and were eliminated H 3 Appendix V : Procedure and results of attrition analysis 116 iv L i s t o f T a b l e s Table 1.1 Socio-demographic characteristics of the 10- and 11-year-old participants (n=3,434) in 1994/95 5 Table 1.2 Sample sizes per age group across the four cycles of assessments and questions about smoking behaviours 6 Table 2.1 Sample sizes per age group across the four cycles of assessment and questions about drinking 25 Table 3.3.1 Summary of 18 multinomial regression analyses that were conducted separately for each sub-category of predictor for occasional and daily smoking 64 Table 3.3.2 Model summary of the stepwise backward logistic regression analysis for occasional smoking (n=ii2i) 67 Table 3.3.3 Summary of the final model obtained with logistic egression for occasional smoking (n=H2i) 68 Table 3.3.4 Model summary of the stepwise backward logistic regression analysis for daily smoking (n=i223) 69 Table 3.3.5 Summary of the final model obtained with logistic regression for daily smoking (n=i223) 70 Table 3.3.6 Summary of 18 binomial logistic regressions that were conducted separately for each sub-category of predictor for infrequent drinking 72 Table 3.3.7 Model summary of the stepwise backward logistic regression analysis for less than monthly drinking (n=i38o) 73 Table 3.3.8 Summary of the final model obtained with logistic regression for infrequent drinking (n=i38o) 74 Table 3.4.1 Summary of the results obtained for occasional, daily smoking, and infrequent drinking 75 List of Figures Figure 1.1 Changes in the probability of having tried smoking between 10-11 and 16-17 years of age (N=2,886) 11 Figure 1.2 Changes in the frequency of smoking between 10-11 and 16-17 years of age (n=28o) 12 Figure 1.3 Changes in the intensity of smoking between 12-13 and 16-17 years of age (n=26o) 13 Figure 2.1 Changes in the probability of having tried drinking between 10 and 15 years of age (N=2,i8i) 28 Figure 2.2 Changes in the frequency of drinking between 10-11 and 16-17 years of age (n=427) 29 Figure 2.3 Changes in the number of times of having been drunk in the past 12 months (n=478) 30 Figure 3.1 Smoking as coping model 83 vi Preface This thesis was motivated by the author's interest in understanding the psychosocial mechanisms involved in the initiation of smoking and drinking. It investigates the development of cigarette smoking and alcohol drinking from late childhood to adolescence. There are two fundamental beliefs that have inspired this work and that throughout the three studies presented here gradually transformed themselves into two plausible hypotheses. The first hypothesis is that health behaviours such as smoking and drinking are gradually acquired over time according to specific developmental pathways. This hypothesis derives from the analysis of recent research that indicates that adolescents seem to follow different pathways when becoming smokers. Such evidence comes from both qualitative (Lovato et al., 2002) and quantitative studies (Chassin et al., 2000; Colder et al., 2001) and suggests that the development of smoking behaviours in younger adolescents could not be adequately described by one single model. The author argues that rather than a 'universal' developmental pathway leading to substance use there are different pathways that adolescents may follow in the acquisition of smoking and drinking. If there are different ways in which adolescents acquire smoking and drinking, then the adolescent population would be better described as heterogeneous, that is, made up of different sub-groups (or sub-populations) of adolescents. There are important implications of the idea that there may be different pathways leading to the acquisition of adolescent smoking and drinking. First, it would suggest reconsidering the current theoretical approach to the study of adolescent health behaviours where general theories are favoured over substance-specific and behaviour-specific theories (Igra & Irwin, 1996). Second, it would indicate that current prevention and intervention strategies need to incorporate this emerging evidence so that programs can be better targeted to specific sub-groups of adolescents rather than aiming at the entire adolescent population. For example, current prevention and intervention programs target all adolescent smokers virtually in the same way with little distinction between occasional smokers and regular or daily smokers (USDHHS, vii 1994)- The fact that there are differences among adolescent smokers in how they acquire such behaviour indicates that certain strategies may be more effective with one group of smokers (i.e., the occasional smokers) than another group of smokers (i.e., the daily smokers). Therefore, the second hypothesis investigated in this series of studies is closely associated with the first hypothesis and maintains that different causal pathways may be involved in the initiation of smoking and drinking, and consequently, more specific theories of adolescent substance use (as opposed to general and 'universal' theories) need to be developed and tested. In order to address the first of these two hypotheses, data were analysed from the National Longitudinal Survey of Children and Youth (NLSCY) to identify the developmental trajectories of smoking and drinking from late childhood to early adolescence. Chapter I is a longitudinal investigation of the developmental trajectories of smoking and Chapter II, also a longitudinal study, describes the search for the developmental trajectories of drinking. In both studies, a semi-parametric modelling approach (growth mixture modelling) was employed with the intent of identifying groups of adolescents sharing similarities in their development and acquisition of smoking and drinking. These two studies are the first of their kind conducted with a Canadian national sample, and they constitute a valuable contribution to understanding the development of smoking and drinking among Canadian children and adolescents. Chapter I and Chapter II provide important information regarding the prevalence of smoking and drinking at different ages (between 10 and 17 years of age) and the stability of these behaviours in late childhood and adolescence. Based on the findings that smoking and drinking reach their highest prevalence at age 14-15 and remain stable thereafter, the work presented in Chapter III focuses on the search for early predictors of smoking and drinking among 14- and 15-year-old adolescents. Chapter III addresses two fundamental questions: what are the early predictors of smoking, and of drinking? Are occasional smoking and daily smoking two steps on the same continuum of substance use or do they constitute two 'independent' phenomena? The answers to these questions have important implications. First, a better understanding of the factors involved in the onset of smoking and drinking may lead to improved viii prevention and intervention strategies aimed at young adolescents. More importantly, new prevention strategies specifically designed for late childhood, before the onset of smoking and/or drinking, can be identified and tested. Second, if in fact there are differences in the etiology of occasional smoking and daily smoking, prevention efforts can be differentially directed to these two distinct phenomena, thus increasing their efficacy. To address these questions, data from the NLSCY were analysed and the associations investigated between a large number of predictors and three separate outcomes: occasional smoking, daily smoking, and infrequent drinking. The predictors were assessed in 1994-95 when the participants were 8 and 9 years old and the outcomes were measured six years later when participants were 14 and 15 years old. Chapter III describes the theoretical rationale for conducting this study and reports the detailed analyses that were conducted, first, to reduce the large number of predictors, and second, to identify the models that best describe the patterns of association found between early predictors and the three outcomes. Finally, in Chapter III, the results obtained are discussed with some interpretations offered of the emerging patterns of prediction. In Chapter IV, the three previous chapters are summarized, the implications of the main findings of this thesis are discussed, and recommendations for future research are offered. Finally, Chapter V highlights some of the most important limitations of the three studies. ix Acknowledgements There are several people who have contributed to this work in different ways. My greatest gratitude goes to Prof. Clyde Hertzman who has had an immense influence on my thinking. His mission to identify and intervene on the early predictors of child health and developmental outcomes has made me understand the importance of investigating the casual roots of behaviours such as smoking and drinking. I feel honoured and privileged that Prof. Hertzman has opened his heart and his mind to my work. I am also very grateful that Prof. Hertzman has always been supportive and encouraging especially when I was facing those many challenging times that doctoral students are doomed to face. My thanks go to my husband Dr. Amedeo D'Angiulli who has patiently waited for this work to be finished so that we can move forward with our life, but more importantly, he can have his wife back to her normal state of mind. Thanks go to my son Leonardo whose smile has the power to remind me what really matters in life. I thank Dr. Tracy Vaillancourt for co-authoring a manuscript based on Chapter I and Dr. Anne George for co-authoring a manuscript based on Chapter II. Their contribution to the manuscripts has been invaluable and I thank them for greatly improving the quality of the present work. I thank Joshua Fagbemi for his help with statistical analysis and modelling, Lee Grenon, James Croal and Darren Lauzon for the support they provided at the Research Data Centre at the University of British Columbia in Vancouver where the confidential data used in this thesis are stored, and Bette Shippam for editorial advice. Thanks also go to my thesis committee members, Drs. Bruno Zumbo, James Frankish, Dafna Kohen, and Kim Shonert-Reichel, for showing encouragement, for providing me with insightful comments, and for believing in my work. Finally, I would like to thank the following agencies that have invested in me by providing financial support to conduct this work as part of my doctoral studies: Human Early Learning Partnership (H.E.L.P.), Canadian Institutes for Health Research (CIHR), Social Sciences and Humanities Research Council (SSHRC), Michael Smith Foundation for Health Research (MSFHR), and the Killam Trust. C h a p t e r I: I d e n t i f y i n g D e v e l o p m e n t a l T r a j e c t o r i e s o f S m o k i n g B e h a v i o u r s f r o m L a t e C h i l d h o o d t o E a r l y A d o l e s c e n c e 1 1 . 1 Introduction An extensive body of literature has been written over the past several years attempting to understand the mechanisms that lead adolescents to become habitual smokers. At present, it is well established that smoking acquisition is a developmental process. Several studies have shown that after initiation, most adolescents who continue to smoke tend to gradually increase their intake over time (USDHHS, 1994). But while smoking acquisition has been described as a monotonic process (i.e., gradual increase over time), further research is indicating that, in fact, smoking may not become a stable behaviour across the lifespan (Moss et al., 1992; Statistics Canada, 2000). To illustrate, although some have found that the likelihood of becoming a daily and heavy smoker increases during the transition out of high school (Johnston, O'Malley, & Bachman, 1994), others have shown that this trend is often interrupted by marriage and parenthood (Yamaguchi & Kandel, 1985). The conventional quantitative approach to the study of smoking behaviours has been to examine general patterns that adequately describe population trends (while treating individual variability as "noise" rather than something of interest from the standpoint of public health). Similarly, at a theoretical level, the focus has been on the development of general models (Flay et al., 1983; Leventhal & Cleary, 1983; Mausner, 1973; Stern et al., 1987) which share the commonality of attempting to describe (and in some cases explain) smoking behaviours in "universal" terms. Indeed, these models are generally based on the identification of similarities among developing smokers rather than on variations in patterns of smoking acquisition over time. For instance, Flay et al.'s (1983) model, perhaps the most accepted and widely referenced 1 Chapter I is a version of a manuscript submitted for publication to the journal, Health Psychology, S. Maggi, C. Hertzman & T. Vaillancourt. Changes is smoking behaviours from late childhood to adolescence: A group-based study of Canadian youth. model for smoking acquisition, suggests that adolescent smoking unfolds in a stage-like fashion: After trying cigarette smoking, there is a transitional period characterized by instability and low frequency of smoking (experimentation), followed by the establishment of a more stable pattern of behaviour (i.e., stable smoking or abstinence). Despite the fact that a significant body of research has applied the concept of smoking acquisition as a staged process, there is scant empirical evidence to support its validity. For example, in their comprehensive review of the literature, Mayhew, Flay and Mott (2000) concluded that whereas many studies found certain specific factors to be consistently associated with a given stage of the smoking acquisition process, only a very limited number of these factors were exclusively associated with one stage (i.e., they tended to predict two or more stages). This result might be due to the lack of precision of the theoretical definitions of the stages and the lack of validation of the measures for the theoretical stages. However, it is also possible that the stages are in fact adequately defined, but serve to describe the process of smoking acquisition of only a specific sub-group of adolescents, rather than the whole adolescent population. That is, there may be different ways in which adolescents become smokers that are not accounted for by the model. To address some of these theoretical limitations a number of researchers explored alternative methodological strategies for the analysis of longitudinal data. For instance, modeling techniques that complement population modeling methods (e.g., hierarchical modeling) and individual methods (e.g., latent growth modeling) such as growth mixture modeling (Nagin, 1999; Jones, Nagin, & Roeder, 2001) have recently been used to study the development of cigarette smoking in adolescence (Chassin et al., 2000; Colder et al., 2001). The appeal of growth mixture modeling is that population level inferences can be drawn from a person-centred analytical strategy that is based on the identification of similarities among individuals within the same population. In other words, growth mixture modeling allows the investigator to describe different patterns, rather than a general pattern of development within the population. The advantage of identifying different developmental patterns of smoking within the population of adolescents who smoke is that more precise theories can be tested, predictors of specific 2 developmental patterns of smoking can be identified, and prevention programs can be better targeted to reach high-risk sub-groups of adolescents. Recent research on the smoking behaviours of early adolescents indicates that, in fact, adolescents do seem to follow different pathways when becoming smokers. Lovato and colleagues (2002) conducted in-depth interviews with 14 to 17 year-old Canadian adolescents (N=35) and found that they described their early smoking behaviours in dissimilar ways. A number indicated that they slowly and gradually became daily smokers over the years (consistent with Flay et al.'s [1983] model), while others indicated that they alternated daily smoking with non-smoking for several months (a pattern of smoking not described by Flay et al. [1983]). Results from recent quantitative studies on developmental trajectories of smoking are consistent with the preliminary results found by Lovato and her collaborators (2002). Using a cohort sequential design, Chassin and colleagues (2000) investigated the developmental trajectories of smoking of a sample of Midwestern adolescents. Participants who in 1980-1983 were in 6 t h to 12 th grade were tracked in 1987 and 1993 (N=8,556). The authors identified one group of abstainers (59%) and five distinct groups of smokers: 1) the 'early stable' group (12%) characterized by an early onset which persisted over time, 2) the 'late stable' group (16%) characterized by a consistent smoking pattern that began later in young adulthood, 3) the 'erratic' group (2%) characterized by a small number of participants whose smoking increased and decreased several times during the course of the study, 4) the 'experimenters' group (6%) characterized by an early onset that did not increase substantially, and 5) the 'quitters' group (5%) characterized by a gradual increase in smoking during adolescence that desisted by the time they reached the age of 30. Also, in a longitudinal study of adolescents aged 12 to 16 (N=502) from the Kansas City metropolitan area, Colder et al. (2001) studied the developmental trajectories of smoking in younger adolescents and found that smoking behaviour could not be adequately described by one single model. Rather, this study ascertained that already at such a young age, there were distinct groups of smokers who followed dissimilar developmental pathways. Specifically, Colder and colleagues (2001) established that while participants were fairly homogeneous in their smoking 3 behaviours until the age of 13, there was considerable diversity following this age with two groups of adolescents remaining fairly stable over time (stable puffers and stable light smokers) and three groups of adolescents gradually increasing the intensity of their smoking at different rates (early rapid escalators, late moderate escalators, and late slow escalators). The ability to identify distinct developmental trajectories of smoking in late childhood has critical practical implications, particularly for policy and prevention. Although it has been argued that intervening at a younger age may lead to greater success in the prevention of smoking behaviours, surprisingly little research devoted to understanding the developmental course of smoking among children has been conducted to date. In fact, longitudinal studies of smoking typically start when participants are 14-15 years old, and evidence about smoking in younger adolescents is primarily derived from cross-sectional studies which do not permit the examination of intra-individual changes over time (for reviews of longitudinal studies see Mayhew, Flay & Mott, 2000; USDHHS, 1994). The purpose of the present study is to examine smoking behaviours from late childhood to adolescence using a national sample of Canadian youth, in order to contribute to a better understanding of the developmental course of smoking acquisition from its very inception. Specifically, by following participants forward in time since they were as young as 10 and 11 years old, the intent was to map the smoking acquisition process as it began and to provide critical information concerning the significant changes associated with its development over a period of six years. 1 . 2 M e t h o d 1.2.1 Participants Participants were drawn from the National Longitudinal Survey of Children and Youth (NLSCY) (detailed description of the sampling and data collection procedures can be found in Human Resources Development Canada & Statistics Canada, 1996), a national study intended to develop policy recommendations and program development on critical factors affecting the development of children. As such, the NLSCY addressed a broad range of factors that are known to be relevant to child development. The NLSCY began surveying 15,579 households with children between o and 11 years of age in 1994/95 who are being followed up regularly at two 4 years intervals. Among the 15,579 households assessed in 1994/95, 3,434 of these households had at least one 10- and 11-year old child living within. Table 1.1 shows the univariate socio-demographic characteristics of the 3,434 10- and 11-year-old children who were assesses in the first cycle of the NLSCY in 1994-95. Table 1.1 Socio-demographic characteristics of the 10- and 11-year-old participants (11=3,434) i n 1994/95 Birth Country Conversational Language Canada English French PMK* 89% 86% 27% Spouse of PMK 73% 73% 23% Child 96% 79% 25% Parental Education - Highest level of schooling < secondary secondary > secondary college or university degree PMK 18% 19% 28% 35% Spouse of PMK 20% 17% 23% 38% Ethnicity of the child European descent 81% American Indians, Metis, or Inuit/Eskimo 4% Chinese 0.6% Blacks or South Asians 0.4% Other 14% Household Income (in $CAN) < 7,500 2% 7,501-17,500 11.7% 17,501-45,000 44-3% >45,ooi 42% t Socio-demographic characteristics are asked as part of the general household questionnaire administered to the person most knowledgeable (PMK) of the child, which in 98% of the cases is the mother. Table 1.1 indicates that the study participants are for the most part born in Canada, of European descent, and with an household income in the medium to high range. Recent data from Census 2001 indicate that only about 13% of the Canadian population is visible minorities and that the majority of Canadians are white and of European descent (Statistics Canada, 2004), indicating that the ethnic origins of the 3,434 participants may not be significantly different from those of the general Canadian population. 5 As of 2004, the NLSCY has released a total of four assessments for analysis. Not all participants participated in all four assessments. To ensure that the probabilistic models described below were, as much as possible, adequate representations of the 'true' (i.e., observed) trajectories, we decided to impose a strict inclusion criterion. Namely, participants had to have responded to at least three of the four assessments. Table 1.2 shows the sample sizes for each of the three outcome variables analyzed below. Table 1 . 2 Sample sizes per age group across the four cycles of assessments and questions about smoking behaviours "Have you ever tried cigarette smoking, even just a puff?" (TRIED SMOKING) Cycle 1, Cycle 2± Cycle 2 Cycle 3 Cycle 4 Age group 10-11 12-13 14-15 16-17 NI 2,778 2,789 2,756 1,344 "If you smoke, how often do you smoke cigarettes?" (FREQUENCY OF SMOKING) Cycle 1, Cycle 2± Cycle 2 Cycle 3 Cycle 4 Age group 10-11 12-13 14-15 16-17 n 138 223 225 236 "On the days that you smoke, about how many cigarettes do you usually smoke?" (INTENSITY OF SMOKING) Cycle 1, Cycle 2± Cycle 2 Cycle 3 Cycle 4 Age group 10-11 12-13 14-15 16-17 n i6_H 168 259 122 t This is the number of participants who were asked the question about trying smoking. However, the following questions on frequency and intensity of smoking were only asked of participants who answered 'yes' to question of trying smoking, thus, the smaller number of participants who answered to the questions of frequency and intensity of smoking. Also note that while both questions of frequency and intensity of smoking are asked to the participants having tried smoking, the response rate to these questions varies from one age group to the other. tt Given the small number of participants in the 10-11 years old age-group, trajectories for intensity of smoking were derived only from participants between 12 and 17 years of age. *In 1994-95, cycle 1 of the NLSCY was conducted. During cycle 1, families with children up to 11 years of age were recruited. Two years later, cycle 2 was conducted and new families with children up to 13 years of age were added to the original cohort. For this reason, participants who were 10-11 years old were identified in cycle 1 and cycle 2 and selected for inclusion in the analysis, provided that they had completed at least three assessment cycles. The sample sizes reported in Table 1.2 refer to the number of participants who responded to each of the three questions of smoking pertinent to this study (tried smoking, frequency of 6 smoking, and intensity of smoking). Note that while a larger number of participants were asked about trying smoking, a smaller number of participants were asked about frequency and intensity of smoking. This is because the latter was only asked of participants who answered 'yes' to the question of trying smoking. The overall response rate for all three questions ranged from a low of 75% in Cycle 3 to a high of 88% in Cycle 2. However, the use of a strict inclusion criterion such as the one described above, led to a further reduction in the sample size for the questions on frequency and intensity of smoking. The socio-demographic characteristics of this sub-sample of participants are similar to those observed among the 3,434 to- and n-year old participants from the first cycle of the NLSCY. 1.2.2 Measures The NLSCY was designed to address a broad range of issues relevant to child and adolescent development, inlcuding smoking behaviour. However, the main purpose of the survey was not to conduct an in-depth analysis of the development of adolescent smoking, and only a limited number of questions were posed to the participants about this topic. There are only three questions that have been consistently asked across the four survey cycles, and they are about l) trying, 2) frequency, and 3) intensity of smoking. These questions were answered by children 10 years of age and older as part of an age specific questionnaire on peers, family, self-esteem, lifestyle and health related behaviour. The age specific questionnaire was self-administered at the participants' homes. Trained interviewers visited and encouraged the participants to find a quiet and private place in their homes where they could respond to the questionnaire. To assure confidentiality, the participants placed the completed questionnaire in an envelope, sealed it, and returned it to the interviewer. The interviewer was not allowed to open the questionnaire until having returned to the research facilities. 7 The primary reason for the interviewers to visit at the participant's home was to ensure that the questionnaires were self-completed. Moreover, interviewers were trained to address questions participants might have about specific items in the questionnaire. This paper reports trajectories of the changing probability of trying smoking and the frequency of smoking from age 10 to 17. Because only 16 participants from the 10-11 years old age group responded to the question of intensity of smoking, trajectories for intensity were derived only from participants between 12 and 17 years of age. Trying smoking was measured with the question, 'Have you ever tried cigarette smoking, even just a few puffs?' generating a binary (yes/no) outcome. Frequency of smoking was measured with the question, 'If you do smoke, how often do you smoke cigarettes?' This question generated an ordinal outcome with 5 levels (0= I don't smoke; 1= a few times a year, 2=at least once or twice a month, 3= at least once or twice a week, 4=every day). Intensity of smoking was measured with the question, 'On the days that you smoke, about how many cigarettes do you usually smoke?' This question generated a continuous outcome with values between 1 and 25. 1.2.3 Analytical approach In order to identify distinct developmental trajectories of smoking, a group-based approach known as growth mixture modelling was employed. In growth mixture modeling the goal of the analysis is to group individuals into categories (i.e., groups) based on similarities in the changes of any given behaviour over time (Nagin, 1999). Growth mixture modeling uses "mixtures of suitably defined probability distributions (...), identifies distinct groups of developmental trajectories within the population" (Nagin, 1999; pp. 139), and differs from hierarchical and latent curve methodologies that model population variability with multivariate continuous distribution functions. Instead, growth mixture modeling uses a multinomial modeling strategy designed to identify homogenous clusters of developmental trajectories. Individuals are assigned posterior probabilities to belong to any of the identified groups and are categorized within the one group for which they had the highest probability of membership. In other words each individual is assigned only to one of the groups identified. 8 While this approach identifies distinct groups within a population, its purpose is not that of providing precise sub-categorizations within a population per se, but rather its purpose is to direct the attention to differences in the developmental patterns of specific behaviours. For this reason growth mixture modeling was selected as the most appropriate strategy for investigating whether there is a unique process of smoking acquisition from late childhood to adolescence or whether there are different ways in which adolescents acquire smoking over this period of time. Using the SAS program's PROC TRAJ module (Jones, Nagin, & Roeder, 2001) growth mixture models for smoking were estimated. In order to select the best model both statistical and "public health" criteria were used. Conventionally, the model that produces the largest BIC (Bayesian Information Criterion, the standard index for model fit) (Schwarz, 1978) is selected as the most adequate. In growth mixture modeling, the number of groups to be tested for each model is specified a priori by the analyst. To begin, a model was tested with one group only, followed by a model with two groups, a model with three groups and so forth until the model with the largest BIC (statistical criterion) was identified. With this procedure the ideal number of groups is empirically derived rather than determined by pre-specified theories. In some cases it is possible that two models (with different numbers of sub-groups) produce virtually identical BIC. To guide the selection of the appropriate model in such a case, both statistical and public health criteria can be considered. From a statistical point of view, parsimony is favoured and the model with the least number of groups may be selected. However, parsimony may be overruled in the case in which the additional group that is identified by the procedure directs the attention to the existence of a high-risk group for which public health intervention may need to be planned ("public health" criterion). Growth mixture modeling allows the inclusion of covariates in the estimate of the models. However, the differences between trajectories with respect to specific variables such as socio-demographic characteristics can also be investigated with chi-square analyses. It was decided to conduct growth mixture modeling without factoring in the effects of covariates, and to conduct a posteriori chi-square analyses on the trajectories identified. Thus, chi-square analysis was performed to explore associations between the trajectories for trying, frequency and 9 intensity of smoking and gender. Furthermore, the trajectories identified were cross-tabulated for trying, frequency, and intensity of smoking with each other in order to identify associations between these different aspects of smoking behaviour. The investigation of associations between socio-demographic characteristics and trajectory membership is also of theoretical and practical interest. Thus, chi-square analyses investigating associations between socio-economic characteristics and trajectories for trying, frequency and intensity of smoking were conducted. Chi-square analyses revealed that there may be differences in the age at which the participants tried smoking among different ethnic groups, but not at the level of the country of birth, language, parental education, and household income. Unfortunately, the results of these analyses could not be released because some of the cells in the cross-tabulations had frequencies smaller than 5. Similarly, chi-square analyses investigating the association between frequency and intensity of smoking and socio-demographic characteristics did not produce significant results and could not be released because of the small frequencies appearing within some of the cross-tabulations. 1 .3 R e s u l t s 1.3.1 Trying smoking The question about trying smoking produced a binary response where o=no (did not try smoking) and i=yes (tried smoking). Due to the binary nature of this question, a logit module was utilized to model the participants' answers. A model that identified three distinct trajectories for trying smoking was selected as the best among those tested (BIC= -4497; N=2,886).2 The difference in the BIC compared to the four-group model and the two-group model was 14 (BIC=-4511) and 663 (BIC=-5i99) units, respectively. As shown in Figure 1.1, for 40.5% of the participants the probability of having tried smoking was virtually o until the age of 14-15 ("late onset"); the probability of having tried smoking rapidly increased after the age of 10 for 49.3% of the participants ("middle onset") who by the age of 15 were very likely to have tried smoking; and 2 Every trajectory that was identified by the analyses conducted in Chapter I and Chapter II was assigned a label. Labels were chosen for consistency with labels used in other studies, or because of their intuitive appropriateness in describing the trajectories. 10 for 10.2% of the participants the probability of having tried smoking was already high at the age of 10 ("early onset"). Figure 1.1 Changes i n the probability of having tried smoking between 10-11 and 16-17 years of age (N=2,886). 1.3.2 Frequency of smoking A model that identified five distinct trajectories of change in the frequency of smoking was selected among those tested (BIC = -1101; n=28o). The difference in BIC from the four-group and six-group models was 4 (BIC=-iios) and 27 (BIC=-ii28) units, respectively. It is notable that in the six-group model, the additional group identified was composed of participants who at age 16-17 smoked but at a frequency of non daily. While this group of 'occasional smokers' may be of 'clinical' interest, the fact that the five-group model (the one without the occasional smokers) proved statistically superior indicates that in our sample there were not enough participants following a developmental pattern leading to occasional smoking that was distinct from the pattern that led to frequent smoking. As shown in Figure 1.2, 71.9% of the smoking participants ("early frequent smokers" and "late frequent smokers") who responded to the question on frequency of smoking reported daily smoking by the age of 16 and 17. In other words, about 38% of respondents gradually increased their frequency of smoking, especially after the age of 14 ("late frequent smokers") while 34% 11 showed the sharpest increase at an earlier age, namely after 12 years of age ("early frequent smokers"). —•— Late infrequent 0 = never experimenters (16.1%) smokers 10-11 12-13 14-15 16-17 Age Figure 1.2 Changes in the frequency of smoking between 10-11 and 16-17 years of age (11=280). The remaining 28.1% of participants at ages 16 and 17 reported that they did not smoke at the time they were surveyed hence they were labelled as "experimenters". However, those participants fall into three different groups, each describing a different trajectory of change in their smoking patterns: 16.1% reported smoking at least once or twice a year around 14-15 years of age ("late infrequent experimenters"); 6.8% reported smoking at least once or twice a year at 10 and 11 years of age only ("early infrequent experimenters"); and 5.2% reported smoking at least weekly between 10 and 15 years of age ("early frequent experimenters"). 1.3.3 Intensity of smoking The best model identified two distinct trajectories of change in the intensity of smoking (BIC = -1490; n=26o). The difference in BIC from the three-group and one-group models was 8 (BIC=-i498) and 80 (BIC=-i57o) units respectively. Figure 1.3 shows that the vast majority of the participants (97.7% - "late slow escalators") were fairly light Oess than 5 cigarettes per day) and stable smokers until the age of 14-15. 12 30 -, a> ti a> i_ 20 A a> .a 10 E z - Late slow escalators (97.7%) - Early rapid escalators (2.3%) 12-13 14-15 A g e 16-17 Figure 1.3 Changes in the intensity of smoking between 12-13 and 16-17 years of age(n=26o) After 14-15 years of age, however, the 'late slow escalators' gradually increased the amount of cigarettes they smoked per day such that, by the time they were 16-17 years old, they reported to be smoking about 10 cigarettes per day. Unlike the 'late slow escalators', the remaining 2.3% of participants followed a trajectory characterized by a sharp and gradual increase in the amount of cigarettes smoked ("early rapid escalators"). These participants, by the time they reached 16-17 years of age, reported smoking more than a pack a day. 1.3.4 Gender, trying, frequency, and intensity of smoking Gender differences are often observed in smoking behaviours. For instance, in the Canadian population the prevalence of current smoking among females aged 15 to 19 is approximately between 2% and 9% greater than the prevalence of current smoking in males of the same age (Gilmore, 2002). In the present study, it was found that while males and females were equally represented in the initial sample of 2,886 (49.3% males and 50.7% females), more females continued to smoke than males throughout the study (39.8% males and 60.2% females). In addition, chi-square analysis between gender and trying smoking indicated that females are 13 more likely than males to fall into the "middle onset" group, that is, they are more likely to have tried smoking between the ages of 11 and 13 (x2 =8.750; df = 2; p < . o s ) 3 . It is significant, however, that no other gender differences were found for frequency (x2 =5.030; df = 4; p > . o s ) and intensity (x2 =.039; df = 1; p > . o s ) of smoking. 1-3-5 Trying, frequency and intensity of smoking Two additional questions were addressed in the present study. These questions are relevant both at the practical and theoretical levels: whether the age at which adolescents first try smoking is associated with specific changes in the frequency and intensity of smoking, and whether different trajectories of frequency of smoking are associated with changes in the intensity of smoking. Chi-square analyses revealed that the "middle onset" group was most likely to fall into the "late frequent smokers" trajectory, and that the "early onset" group was most likely to fall into the "late infrequent experimenters" group (x2 =11.432; df = 4; p < . o s ) . To clarify, participants who tried smoking at the age of 12 and 13 were more likely to become daily smokers by the time they reached 16-17 years of age, while those who tried smoking at the age of 10 and 11 were more likely to smoke infrequently at the age of 14-15 and to have stopped smoking altogether by the age of 16-17. No other dissimilarities were found in the association between frequency and intensity of smoking (x2 =4.785; df = 4; p>.05) and between trying smoking and intensity (x2 =.634; df = 1; p > . o s ) . 1.4 Discussion By utilizing growth mixture modelling, it was possible to identify distinct ways in which Canadian children and adolescents acquire smoking over time. Specifically, it was determined that participants were grouped into three categories depending on the age they tried smoking for the first time (early onset, middle onset, and late onset) and that girls were more likely than boys to start smoking between 11 and 13 years of age. In addition, five different developmental trajectories were identified in the frequency of smoking, but it was also found that, except for a 3 Note that when cross-tabulations have one or more empty cells the degrees of freedom change accordingly. Because in these cross-tabulations there was at least one empty cell, the Research Data Centre did not release the frequency tables for these comparisons. 1 4 small proportion of participants who had smoked more than 10 cigarettes since the age of 12-13, the vast majority of participants reported never smoking more than 10 cigarettes per day throughout the duration of the study. The largest variability in the changes of smoking behaviour over time was observed at the level of the frequency with which participants reported smoking. Specifically, three patterns of smoking experimentation were identified which led to non-smoking at age 16-17, and two patterns of smoking acquisition which led to daily smoking at age 16-17. Further, in two of the three patterns of experimentation (early infrequent experimenters and late infrequent experimenters), smoking was described as a fairly sporadic behaviour with participants reporting that they smoked somewhere between a few times a year and once or twice a month, but then desisted smoking altogether by the time they reached 16-17 years of age. In contrast, the third pattern of experimentation described the behaviour of participants who smoked more frequently (between once or twice a month and at least once a week) - although never daily - between 10-11 and 14-15 years of age, but ceased smoking completely by the time they reached 16-17 years of age (early frequent experimenters). The two patterns of smoking acquisition Gate frequent smokers and early frequent smokers) are similar in that frequency of smoking gradually increased over time for both of them, but are different in terms of the age at which these increases took place. The early frequent smokers reported smoking at least once or twice a month since the age of 12-13 and continued to do so until 16-17 years of age, at which time they reported daily smoking. Conversely, the late frequent smokers did not start smoking until they were 14-15 years old, but progressed more rapidly to daily smoking so that by the time they were 16-17 years old, they were smoking daily. In summary, the results indicate that among the participants, daily smoking was acquired gradually over the course of several years (i.e., from smoking once or twice a year at age 10-11 to daily smoking at age 16-17). i n addition, the results indicate that the amount (i.e., number of cigarettes) smoked per day also gradually increased over time (from less than 5 a day at age 12-13 to about 10 at age 16-17). As well, the results show that several routes to non-smoking by age 16-17 exist and as such the gradual, monotonic process of acquiring long-term 15 smoking behaviour does not necessarily apply to 'acquiring' non-smoking behaviour. In other words, by studying the longitudinal trajectories of smoking behaviours between 10 and 17 years of age it was possible to differentiate patterns of experimentation from patterns of acquisition— the first with non-smoking as an outcome at age 16-17 and the second with daily smoking as the outcome at age 16-17. The fact that it was possible to differentiate between trajectories of smoking acquisition and trajectories of smoking experimentation provides further support to the notion that smoking is a complex and multifaceted behaviour and that it may be necessary to move away from the idea of a general, universal theory of smoking acquisition in order to more fully understand it. Furthermore, the results of the present study seem to signify that Flay et al.'s model (1983) may be more adequate in describing the process of smoking acquisition than the process of smoking experimentation. In other words, while experimentation has been generally considered as one of the stages in the smoking acquisition process leading to regular smoking, the results indicate that experimentation can, in fact, also be described as a distinct process in itself that does not necessarily lead to smoking acquisition. Along with redefinitions of the theories of smoking acquisition, new theories characterising the process of smoking experimentation as a distinct phenomenon from smoking acquisition need to be developed and tested. Prevention specialists can take advantage of the theoretical distinction between smoking acquisition and smoking experimentation by designing programs that are targeted to different sub-groups of adolescents displaying diverse levels of risk of becoming daily and heavy smokers. Studies have shown that male (Distefan et al., 1998; Flint, Yamada & Novotny, 1998) Caucasian adolescents (Choi et al., 1997; Distefan et al. 1998; Flint, Yamada & Novotny, 1998; Pierce et al., 1996) who experimented with smoking and had a weak commitment to not smoke in the future were more likely to become regular smokers (Choi et al., 1997; Choi et al., 2001; Distefan et al., 1998). Having parents and family members who smoked (Choi et al., 1997; Flay, Hu, Richardson, 1998; Flint, Yamada & Novotny, 1998; Pierce et al., 1996) and having a 'best friend' who smoked, were also associated with the progression from experimentation to regular 16 smoking (Choi et al., 1997; Distefan et al., 1998; Flint, Yamada & Novotny, 1998; Pierce et al., 1996). These studies are examples of investigations of some of the most immediate precursors of smoking in which the focus is on finding association between 'proximal' predictors and smoking behaviour. However, proximal predictors do not adequately explain the long-term causes and origins of behaviours in general (Petraitis, Flay, & Miller, 1995), and smoking in particular. Consequently, little is known about the more 'distal' predictors and/or the predisposing factors that may be associated with smoking. The value of investigating these predisposing factors is that it would offer an additional opportunity for prevention work with younger children, long before smoking takes place, thus intervening early by 'modifying' the developmental pathway of smoking acquisition that the child was likely to undertake during adolescence. While looking ahead and investigating whether the trajectories of smoking identified in the present study are predictive of adult smoking, the value to proceed backwards in search of the predictors of these early trajectories of smoking is recognized. A full understanding of the development of smoking behaviours from childhood to adolescence is critical for the advancement of health promotion and intervention in this field. While the present study provides important information regarding the early development of smoking, there are some limitations. For example, the NLSCY is not a survey designed specifically to study adolescent smoking behaviours and some important variables such as susceptibility, attitudes and beliefs were not included in the questionnaire. In addition, the sample size used for frequency and intensity of smoking was considerably smaller than the sample size used for trying smoking. One example of how the results of the present study may have been affected by the size of the sample is related to the fact that the model of frequency of smoking in which a sub-group of occasional smokers was identified was too unstable to reach statistical significance. Research has shown that underreporting of smoking and/or misclassification of smokers and non-smokers is not uncommon among adolescents when self-reports are used to assess smoking status (Dolcini et al, 2003). In addition, research has also indicated that this 17 misclassification may be even greater in the case of occasional smokers (Fergusson 6k Horwood, 1995; Maggi, Linn & Marion, in press). In the present study, about 10% of the participants reported to take up smoking over the course of the six years of the study. A national survey of tobacco use among Canadian adolescents estimates that about 12% of adolescents between 15 and 19 years of age are regular smokers and about 11% of them are occasional smokers (Health Canada, 2001). It is possible that the inability to identify a distinct developmental trajectory characterized by occasional smoking is in fact the consequence of greater underreporting of smoking by adolescents who smoke occasionally. Despite these limitations, the present study generated a number of interesting results. For example, participants who tried smoking after the age of 12 were more likely to become daily smokers than those who tried smoking before the age of 12. This result can at first appear as counterintuitive in that it does not replicate findings from previous studies indicating that the younger the age at onset the more likely a youth is to progress to regular smoking (USDHHS, 1994). However, age at onset is typically asked retrospectively and not enough prospective studies investigating the process of smoking acquisition have been conducted from childhood to adolescence. Consequently, strong empirical basis is lacking for establishing a clear association between age at onset and development of regular smoking. In addition, it is possible to attempt an explanation for the meaning and implications of such a result. For example, in Canada at approximately 12-13 years of age, most children make the transition from elementary to secondary school. The transition to secondary school is known to be a critical time, especially in the domains of psycho-social development and peer-relationships (Brown, 1990; Brown, Dolcini, & Leventhal, 1997; Buhrmester, 1996; Dishion, Capaldi, & Erickson, 1968; Harter, 1990; Yoerger, 1999) which, not surprisingly, is strongly associated with smoking uptake (Conrad, Flay, & Hill , 1992; Evans et al., 1978; Petraitis, Flay, & Miller, 1995). The type of circumstances associated with the onset of smoking during the transition from elementary to secondary school may be such that adolescents will find themselves smoking at higher frequency than the adolescents who tried smoking at a younger age. Moreover, it is possible that smoking is used by adolescents as a coping strategy associated 18 with increasing academic and social demands, or to adjust to the changes in their social networks of peers within the school context. Conversely, adolescents who had tried smoking before their entry into secondary school may have done so for 'experimental' (i.e., 'just to try it') purposes rather than to use cigarettes as a coping strategy. It is also significant that the majority of the participants that smoked could be considered 'light' smokers in that they smoked less than 1 0 cigarettes per day. Thus, it is possible that the motivation for trying smoking around the time of entry into secondary school and the social context in which it occurs may be different from the motivation and the social context of trying smoking when it occurs at an earlier age. This and other similar hypotheses certainly need to be investigated further in future research. In conclusion, the results of the present study contribute to the understanding of smoking acquisition and maintenance in several important ways. First, the research outcomes contribute to the advancement of developmental theories of smoking acquisition and introduce the concept of smoking experimentation as a distinct phenomenon from smoking acquisition. Second, the findings underscore the importance of studying smoking behaviours in younger age groups than previously studied. Third, they direct attention to some specific practical considerations for prevention and intervention planning. Finally, the results of the present study highlight the need for more longitudinal studies that focus on the broader developmental context of children and adolescents in an attempt to understand the development of smoking behaviours. Indeed, only with repeated (and frequent) measurements over time during the most critical times for child and adolescent development will it be possible to fully understand the processes by which some adolescents become regular and daily smokers while others (despite years of engagement with cigarette smoking) become non-smokers. 1 9 Chapter II: Exploring the developmental course of early drinking behaviours of Canadian children and adolescence 4 2.1 Introduction Alcohol use is a prevalent behaviour in Western societies. While alcohol can be consumed in moderation and within the appropriate social settings, it is too often abused by adults, youth and children alike. The abusive and excessive consumption of alcohol is responsible for a wide range of well known consequences to both individual health and the well being of society as a whole (Canadian Center on Substance Abuse, 1996; Single et al., 1999; Single, Maclennan, & Macneil, 1994; USDHHS, 2000). Many efforts have been made to identify who is more likely to abuse alcohol and what factors may be involved in the development of alcoholism and other forms of alcohol abuse. For example, research has shown that in Canada, adult men are more likely to drink than women, 80.6 and 68.4 percent, respectively (Single, Maclennan, & Macneil, 1994) and that the percentage of Canadians who drink one to six drinks per week gradually increases from 27% at age 15-17 to 46% at age 55-64 (Statistics Canada, 2000). The prevalence of drinking among adolescents also has increased in Canada: the percentage of adolescents 12 to 19 years old who engage in episodes of heavy drinking (five or more drinks per occasion) has risen from 30% in 1993 to 42% in 1999 (Statistics Canada, 2000). Furthermore, the frequency with which adolescents drink also has increased in recent years in several Canadian provinces (Statistics Canada, 2000), with the exception of British Columbia where alcohol use among youth has slightly decreased between 1992 and 2003 (McCreary Centre Society, 2004). Poulin and Elliott (1997) found that while in Nova Scotia the prevalence of exclusive alcohol drinking (i.e., alcohol as the only substance of use) decreased between 1991 and 1996, the 4 Chapter 2 is a version of a manuscript submitted for publication to the journal Addiction, S. Maggi, C. Hertzman & M . A. George. Exploring the developmental course of early drinking behaviours of Canadian children and adolescents. 20 prevalence of alcohol drinking in conjunction with tobacco and cannabis use has significantly increased in the same time period. Similarly, Adlaf et al. (2000) found that between 1993 and 1996 the proportion of students who reported alcohol use in Ontario increased significantly. This rising trend in the prevalence of frequent and heavy drinking among Canadian youth is of great public health concern. Several studies have shown that there is a strong association between the age of onset of drinking and the development of later drinking problems (e.g. Hawkins et al., 1997; Rachel et al., 1982; York, 1999; Grant & Dawson, 1997). It has been argued that prevention efforts aimed at delaying the onset of drinking could result in a reduced prevalence of subsequent alcohol related problems. Additional important information that could inform prevention programs can be derived from an understanding of the different ways in which adolescent drinking behaviours evolve over time. Research conducted with American (Colder et al., 2002; Stice, Myers, & Brown, 1998; Wills et al., 1996) and European adolescents (Donato el al., 1995; Choquet & Ledoux, 1985; Kokkevi & Stefanis, 1991) indicates that there may in fact be distinct patterns of adolescent drinking. In some European countries wine drinking during meal times in the family context is a relatively common behaviour, especially among male adolescents. In contrast, alcohol drinking with peers outside of the family context is a less prevalent behaviour, but when it occurs, larger quantities of alcohol are consumed (Donato et al., 1995). North American adolescents also have different patterns of alcohol consumption. Of major public health concern for the United States are two distinct phenomena: heavy drinking where adolescents drink large quantities of alcohol at frequent intervals, and binge drinking where adolescents (generally college students) drink excessive quantities of alcohol on isolated occasions (U.S.D.E., 1993). Recent studies indicate that not only are there distinct ways in which adolescents engage in drinking, but also that there are different ways in which drinking behaviour develops over time. Several studies have been conducted on the developmental trajectories of drinking, but these primarily focus on the transition from late adolescence to adulthood (e.g., Casswell, Pledger, & Pratap, 2002; Chassin, Pitts & Prost, 2002; Guo et al., 2000; Hil l et al., 2000; Muthen 21 & Muthen, 2000). To our knowledge, only two studies (Colder et al., 2002; Wills et al., 1996) have been conducted to date that used a group-based approach and focused on the identification of developmental trajectories of drinking from late childhood to early adolescence. Wills and colleagues (1996) employed cluster analysis to group 1,702 participants (students attending public school in a New York State school district) based on similarities of smoking, drinking and marijuana use over a three-year period (from grade 7 to grade 9). They found five clusters of substance use: 50% of the participants were 'stable nonusers'; 26% were 'minimal experimenters' in that they engaged in substance use no more than once or twice in the past; 14% were 'late starters' in that they increased substance use in the third year of assessment; and 10% of the participants followed one of two patterns of 'escalation' characterized by frequent substance use since the beginning of the study and a gradual increase in use in the following years. In another study with participants from the United States, Colder and colleagues (2002) identified classes of drinking behaviours of 1,918 adolescents from grade 7 through grade 12. They depicted five distinct classes of behaviour based on the quantity and frequency of drinking: the 'occasional very light drinkers' drank very seldom and consumed small amounts throughout the study; the 'occasional heavy drinkers' drank about three or four times a month and about 3 drinks per occasion; the 'escalators' continued drinking infrequently throughout the study but significantly increased the number of drinks per occasion; the 'rapid escalators' increased both the frequency and amount of drinking over time; and the 'heavy drinkers with declining frequency' considerably decreased the frequency in which they drank, but the amount they drank on each occasion remained fairly stable over time. These two studies reveal that there are different ways in which young adolescents drink alcohol over time, and that these patterns are observable starting in late childhood. While Canada shares many cultural characteristics with the United States, it is also known to vary in several ways (e.g., socio-demographic composition and religiosity). Thus, it is not apparent whether the American patterns of drinking found by these two studies also apply to Canadian adolescents. While extensive monitoring of alcohol and other substance use among 22 adult and adolescent populations is in place in Canada, little has been done to identify the different developmental patterns for drinking during the early adolescent years. In other words, current knowledge about the early development of Canadian adolescents' drinking behaviours is limited and for the most part based on comparisons with studies conducted outside the Canadian context. Given the current limited knowledge, longitudinal studies based on Canadian national samples that aim to identify the developmental course of drinking behaviours among children and young adolescents are greatly needed. In the present study, using a national sample of Canadian children and adolescents the intent was to identify the different ways in which Canadian youth engage in drinking behaviours from late childhood to early adolescence. Specifically, we employed the National Longitudinal Survey of Children and Youth (NLSCY) to distinguish 'sub-groups' of adolescents clustered according to similarities in their evolving drinking behaviours over a 6-year period between 10-11 and 16-17 years of age. Rather than assuming that there is a 'universal' way in which drinking behaviours develop over time, the motivating hypothesis was that the adolescent population, when it comes to health behaviours in general and drinking in particular, may be more aptly described in heterogeneous terms and that there are a number of sub-groups of adolescents that display distinct developmental patterns of drinking. Thus, a growth mixture modelling approach was employed to describe the different pathways adolescents follow with respect to drinking from late childhood to early adolescence. 2.2 M e t h o d s 2.2.1 Participants Participants were extracted from the National Longitudinal Survey of Children and Youth (NLSCY) (for details on the sampling procedure and data collection see Human Resources Development Canada & Statistics Canada, 1996). The NLSCY is a Canadian national survey designed to develop policy recommendations and program development on critical factors associated with child and adolescent development. Thus, the measures included in the NLSCY address a wide range of factors known to be associated with developmental outcomes. The NLSCY began surveying 15,579 households with children between 0 and 11 years of age in 23 1994/95- Since 1994/95, follow up assessments are being conducted at two year intervals. Table 1.1 (see Chapter 1) shows the univariate socio-demographic characteristics of the 3,434 10- and 11-year-old NLSCY participants assessed in 1994/95. Table 1.1 indicates that generally, the NLSCY sample is composed for the most part of participants who were born in Canada, of European descent, and with a household income in the medium to high range. In 1994/95 there were 3,434 children between the ages of 10 and 11. The socio-demographic characteristics of the NLSCY participants can be considered representative of the general Canadian population. For example, recent data from the 2001 Census indicate that only 13% of the Canadian population is comprised of visible minorities and that the majority of Canadians are Caucasian and of European descent (Statistics Canada, 2004). In 2004, four follow up assessments had been conducted and released for analysis. Children who were 10-11 years old in 1994/95 were 16-17 years old in 2000/01. To be included in the present study, children and adolescents had to have responded to at least three out of the four assesments. However, in the first assessment conducted in 1994/95 the question "How many times in the past 12 months have you been drunk?" was not asked. Thus, to be included in the study children and adolescents must have provided answers to this question in at least two out of the three assessments. This strict criterion for inclusion allowed for a more accurate estimate of the probabilistic models that are described below and to ensure that the developmental trajectories of drinking resulting from the analysis were as close as possible to the observed (i.e., real) evolving drinking behaviours. Table 2.1 illustrates the sample sizes for each of the three outcome variables analyzed below. The sample sizes reported in Table 2.1 refer to the number of participants who were asked each of the three questions of drinking pertinent to this study (tried drinking, frequency of drinking, and intensity of drinking) and for which valid answers were provided by the participants. Note that while a larger number of participants were asked the question about having tried drinking, a smaller number of participants were asked about the frequency with which they drank and the number of times they got drunk (i.e., intensity). The reason for this discrepancy is that all children 10-11 years and older are asked the question about trying drinking 24 but only participants who answered 'yes' to that question were prompted to respond to the following questions on frequency and intensity of drinking. The overall response rate for all three questions ranged from a low of 75% in the third assessment (conducted in 1998/99) to a high of 88% in the second assessment (conducted in 1996/97). However, the use of a strict criterion of inclusion such as the one described above led to a further reduction in the sample size, particularly for the questions on frequency and intensity of drinking. The socio-demographic characteristics of of the sub-sample of participants included in the present study are similar to those observed in the 3,434 10- and 11-year-old participants. Table 2.1 Sample sizes per age group across the four cycles of assessment and questions about drinking "Have you ever had a drink of alcohol?" (TRIED DRINKING) Cycle 1, Cycle 2± Cycle 2 Cycle 3 Cycle 4 Age group 10-11 12-13 14-15 16-17 N 2160 2158 2153 N.A.* "How often do you drink?" (FREQUENCY OF DRINKING) Cycle 1, Cycle 2± Cycle 2 Cycle 3 Cycle 4 Age group 10-11 12-13 14-15 16-17 n 226 390 392 351 "How many times in the past 12 months have you been drunk?" (NUMBER OF TIMES BEING DRUNK) Cycle 1, Cycle 2± Cycle 2 Cycle 3 Cycle 4 Age group 10-11 12-13 14-15 16-17 n N . A . n 170 438 389 +Only 50 participants responded to this question in Cycle 4 and none of them were 16-17 years old. nThis question was not asked in 1994-95. ±In 1994-95, cycle 1 of the NLSCY was conducted. During cycle 1, families with children up to 11 years of age were recruited. Two years later, cycle 2 was conducted and new families with children up to 13 years of age were added to the original cohort. For this reason, participants who were 10-11 years old 25 were identified in cycle 1 and cycle 2 and selected for inclusion in the analysis, provided that they had completed at least three assessment cycles. 2.2.2 Measures The purpose of the NLSCY is to address a wide range of issues relevant to child and adolescent development, but not specifically to conduct an in-depth analysis of the development of adolescent drinking. Consequently, only a limited number of questions were asked of the participants about this topic. Three questions were consistently asked across the four follow up assessments, concerning trying, frequency and intensity of drinking. Children 10 years and older answered these questions as part of an age specific self-administered questionnaire on peers, family, self-esteem, lifestyle and health related behaviours. Participants were visited at their homes by trained interviewers whose primary role was to encourage finding a private place where participants could complete the questionnaires and to address questions about specific items in the questionnaire. Once the questionnaire was completed, the participants sealed and placed them in envelopes that were then returned to the interviewer. Results pertaining to the questions about trying drinking ('Have you ever had a drink of alcohol?'), frequency of drinking ('How often do you drink alcohol?') and intensity of drinking measured as the number of times being drunk in the past 12 months ("In the past 12 month, how many times have you been drunk?') are discussed in the present study. In previous studies composite measures of drinking were generally used and were calculated by combining frequency of drinking with number of drinks per occasion or number of times being drunk (Casswell, Pledger, & Pratap, 2002). We argue that combining different aspects of drinking into a single measure may result in the loss of critical information about how each aspect of drinking develops over time. In fact, Casswell, Pledger, and Pratap (2002) found that frequency of drinking and number of times beingdrunk followed different trajectories during the transition from late adolescence to young adulthood. 26 2.2.3 Analytical approach In order to identify different developmental patterns of drinking a person-centered approach known as growth mixture modelling was employed. Growth mixture modelling can be used to study the development of those behaviours for which people tend to follow distinct trajectories or pathways. In fact, the goal of growth mixture modeling is to cluster people in different groups based on similarities in the way they change over time (Nagin, 1999). To identify the patterns of drinking for trying, frequency and number of times being drunk, the SAS program's PROC TRAJ module (Jones, Nagin, & Roeder, 2001) was utilized. The conventional approach to model selection within growth mixture modelling suggests that the model that produces the largest BIC (Bayesian Information Criterion, the standard index for model fit) (Schwarz, 1978) be selected. The number of groups (i.e., patterns) to be tested is specified a priori by the analyst. The process began by specifying a model with one group only, then a model with two groups, a model with three groups and so forth until the largest BIC was obtained. This procedure allows the analyst to select the optimum number of groups based on empirical observation. When two models produced virtually identical BICs, parsimony was favoured, and the model with the fewest number of groups was selected. The investigation of associations between socio-demographic characteristics and trajectory membership is also of theoretical and practical interest. Thus, chi-square analyses investigating associations between socio-economic characteristics and trajectories for trying, frequency of drinking, and number of times being drunk were conducted. Chi-square analyses revealed that there may be differences in the age at which the participants tried drinking among different ethnic groups. However no differences were observed at the level of the country of birth, language, parental education, and household income. Unfortunately, the results of these analyses could not be released because some of the cells in the cross-tabulations had frequencies smaller than 5. Similarly, chi-square analyses investigating the association between frequency of drinking, number of times being drunk and socio-demographic characteristics did not produce significant results and could not be released because of the small frequencies appearing within some of the cross-tabulations. 27 2.3 Results 2.3.1 Tried drinking A series of logit models were tested for the answer to the question "Have you ever had a drink of alcohol?" that produced binary data in the form of i=yes and o=no. A model that identified two trajectories for the changes in the probability of having tried drinking was selected (BIC =-3440; N=2,i8i). The difference in the BIC between the models with one and three trajectories models was of 163 and 11 units, respectively. Figure 2.1 Changes in the probability of having tried drinking between 10 and 15 years of age (N=2 , i8i) As shown in Figure 2.1, 46.5% of the participants were very unlikely to have tried drinking before 14-15 years of age ('late onset') while 53.5% of participants were very likely to have tried drinking by the age of 12-13 ('early onset'). 2.3.2 Frequency of drinking A series of models were tested to identify trajectories in the frequency of drinking. Frequency of drinking was measured with an ordered-categorical variable where o=never and 28 5=every day. The model that best fit the data identified five distinct trajectories for frequency of drinking (BIC = -1734; n=427). The difference in the BIC between the models with four and six trajectories was 3 and 50 units, respectively. Figure 2 shows that 22.6% of the participants were less than monthly drinkers and established this pattern of drinking after 14-15 years of age ('regular less than monthly drinkers'); 20% of participants drank more than once or twice a month and started to do so around 12-13 years of age ('early regular more than monthly drinkers'); 33.1% of the participants were monthly drinkers at 16-17 years of age but had been gradually increasing the frequency of their drinking since they were 10-11 years old ('regular monthly drinkers'); 15.4% of participants drank more than once or twice a month but did not start drinking until they were 14-15 years old ('late regular more than monthly drinkers'); and 8.9% of the participants did not engage in drinking until they were 14-15 years old, but at 16-17 years of age reported that they did not drink at all ('former infrequent drinkers'). •m— Regular less than monthly drinkers: established their drinking pattern late (22.6%) • — Early regular more than monthly drinkers: established their drinking pattern at 12-13 (20%) •B— Former infrequent drinkers (8.9%) •X— Regular monthly drinkers: established their pattern of drinking at 10-11 (33.1%) •6— Late regular more than monthly drinkers: established their pattern of drinking at 14-15 (15.4%) Figure 2.2 Changes in the frequency of drinking between 10-11 and 16-17 years of age (n=427) 0 = never 1 = once or twice a year 29 2-3-3 Number of times being drunk A series of censored normal models were tested to identify trajectories in the intensity of drinking that produced ordered-categorical data where o=never and 2=12 or more. Four trajectories were identified that best described changes in the number of times being drunk in the past 12 months (BIC= -182; n=478). The difference in the BIC between the models with three and five trajectories was 386 and 555 units, respectively. Figure 2.3 shows that all of the 12-13 year old participants who reported drinking had been drunk between 1 and 12 times in the past 12 months. Figure 2 . 3 Changes in the number of times of having been drunk in the past 12 months ( n = 4 7 8 ) About 48% of the participants at 16-17 years of age reported having been drunk more than 12 times Gate escalators); 9.3% of the participants reported having been drunk more than 12 times in the past 12 months since they were 14-15 years old (early escalators); 37.4% reported that they had been drunk between 1 and 12 times in the past twelve months since they were 12-13 years old (stable drinkers); and 5.4% reported that they had been drunk between 1 and 12 times 30 when they were 12-13 and 14-15 years old, but said they had never been drunk in the past 12 months when they were 16-17 years old (former drinkers). 2.4 Discussion The purpose of this study was to describe the development of early drinking behaviours among a national sample of Canadian adolescents. The findings indicate that Canadian adolescents do not follow a 'universal' developmental trajectory for drinking behaviours, but rather different sub-groups of adolescents tend to follow distinct pathways. Because of the different measures of drinking used to identify trajectories, a direct comparison with the results obtained by Wills et al. (1996) and Colder et al. (2002) may not be entirely valid 5. Notwithstanding that concern, it appears that there might be some differences in the results obtained here, compared with previous studies. For example, according to Wills and colleagues (1996) only about 24% of their participants had either used substances or showed an increase in substance use between grade 7 and grade 9. In contrast, the results identified a much larger proportion of participants who had consumed alcohol at least once in the past. It was found that, while the probability of having tried drinking by the age of 10-11 was 'none to low' for virtually all study participants, 12-13 year old participants tended to be grouped into two distinct sub-groups: one sub-group (made up of about 46% of the sample) who were still extremely unlikely to have tried drinking, and another, larger sub-group (about 53%) who were much more likely to have tried drinking. These results also indicate that after 12-13 years of age, the probability of having tried drinking sharply increased for both sub-groups of participants, but the difference between one sub-group and the other remained fairly constant over time (i.e., they followed parallel trajectories). 5 Wills et al. (1996) identified trajectories for the combined used of smoking, drinking and marijuana, and Colder et al. (2002) used a measure of drinking where frequency and number of drinks were recoded into one single composite measure for drinking. Because we found that there are different ways in which frequency of drinking and number of times being drunk developed over time, we recommend that measures of frequency and number of times being drunk be kept separate in the analysis. 31 It is possible that there are different influences and mechanisms at play for those adolescents who have tried drinking by age 12-13 versus those who first tried drinking at 14-15 years of age or later. It could be hypothesised that children who live in family environments where adults consume alcohol regularly may be more likely to have tried alcohol at a younger age. This would likely come about because alcohol was more readily accessible in their homes or because parents may have allowed their children to have 'a taste' of alcohol under their supervision. A growing body of research is finding that access, primarily through licensed premises, to alcohol products in previous years is associated with consequent frequency of drinking among young adults (Casswell & Zhang, 1997; Wechsler et al., 2000; Williams & Lillis, 1988; O'Malley & Wagenaar, 1991; Wagenaar, 1993; Draper, 1995; Adlaf, Ivis & Smart, 1997). However, relatively unexplored is the possibility that there are marked differences in the context in which trying drinking occurs at different ages and what effect these contexts may have on the frequency of drinking in following years. The author is currently undertaking research that aims at exploring the role of individual, family, neighbourhood, and school predictors of the onset of drinking and drinking behaviours of 14-15 year old Canadians. In addition, as the NLSCY study participants continue to be followed over time, future research will be conducted on the association between the age at onset of drinking and problem drinking in late adolescence and young adulthood. Another pertinent finding of the present study is that by age 16-17 about 92% of the participants who had ever consumed alcohol reported drinking at least once or twice a month and seemed to have established this pattern of drinking by 14-15 years of age. Interestingly, it was not possible to identify a sub-group of'high risk' adolescents (i.e., those drinking weekly or daily) as a distinct group from the rest of the participants. These results are in partial contrast to those obtained by Colder and colleagues (2002) who found that one of the two most dominant patterns of drinking was a progressive increase in the frequency and number of drinks consumed per occasion, such that by grade 12 a considerable proportion of their subjects reported drinking between three and twenty times in the past month. While a sub-group was not identified which drank considerably more frequently than monthly, it is not possible at this time to predict how 32 likely our participants are to develop patterns of problem and heavy drinking in the future. It is possible that among these participants the trajectory leading to problem drinking may require a much longer period of time to become fully manifested (i.e., it has a longer latency time). Alternatively, the sample size was not large enough to capture low prevalence behaviours such as alcohol abuse in late childhood and early adolescence, which may be especially infrequent for that specific age group. While marked similarities were observed in the patterns of drinking among the study participants at 16-17 years of age, it is noteworthy that the differences in the drinking patterns were much more noticeable when the participants were younger. In fact, five sub-groups were identified of adolescents who followed distinct developmental patterns of the frequency of drinking between 10-11 and 14-15 years of age. Some may argue that, because by 16-17 years of age adolescents' drinking behaviours tend to become more similar, the findings reported here lose relevance. However, from a developmental perspective, this result is of critical importance. In fact, it is intriguing that different pathways lead to similar behaviours in that it contradicts the principle that there are universal directions governing human development (for a discussion on how development is theorized in psychology see Root, 1999). The most direct implication of this finding is that researchers should consider directing their efforts to the search for those factors that are associated with each different pathway of alcohol drinking. One limitation of the present study is that it did not have a large enough sample to analyze the associations between the predictors investigated in the NLSCY and the different developmental pathways that we identified. While with this study we could not address questions relevant to specific sub-groups of the population, the results obtained here can generate valid insights applicable to the general Canadian population. In addition, results from the present study can be used to develop specific hypotheses regarding the developmental course of drinking behaviours that should be investigated in future studies. In light of the results obtained in the present study, a number of hypotheses regarding the function of alcohol drinking in the lives of our study participants can be proposed. For example, it is significant that not only did most participants report drinking once or twice a 33 month, but also that by age 16-17 the majority of participants who had tried alcohol had been drunk more than 12 times in the past year (57.2%), and more than a third had been drunk between one and 12 times in the past year. In total, about 92% of the participants who reported drinking had been drunk at least 12 times in the past year. Thus, for the most part, it is logical to infer that nearly every time our study participants used alcohol it resulted in them 'getting drunk'. This result is important in that it indicates that 16-17 year olds do not necessarily know how to drink in moderation. This raises the question as to whether our study participants are generally unable to drink responsibly, whether the explicit intent of their drinking is to 'get drunk'. In addition, because in the present study actual number of drinks consumed per occasion was not measured, these findings raise the question as whether participants 'pretend' to be drunk while in fact they might have been consuming small quantities of alcohol. If it were determined that adolescents are, in general, unable to recognise the physiological responses of alcohol intake, then interventions could be taught as to how to become more aware of their bodies' reactions to alcohol and, consequently, how to regulate their behaviour so that they develop into moderate and responsible drinkers. Clearly, historical, religious, and philosophical beliefs cause the promotion of moderation, rather than abstinence, controversial. However, promotion of moderate and responsible drinking may offer an opportunity for engaging in discussion with adolescents for whom total abstinence is unrealistic. In the case in which there is an explicit intent to 'get drunk' or to 'pretend' to be drunk it would be worthwhile to investigate the purpose that 'being drunk' fulfills. While it is not difficult to generate plausible, common sense hypotheses of why adolescents want to 'get drunk' (e.g., rebelliousness, need for de-legitimating one's action, facilitation of social interactions), the majority of the theoretical models that have been put forward have focused primarily on the phenomena of alcohol abuse and alcoholism (for a discussion on the psychological theories of drinking and alcoholism see Leonard & Blane, 1999). Theories of adolescent alcohol use are generally non-specific in that they encompass a wide range of 'problem behaviours' such as smoking, marijuana and other drug uses (the problem behaviour theory by Jessor and Jessor 34 (1980) and its later developments are perhaps the most prominent examples), and the consumption of alcohol is viewed as either a strategy to cope with stress and emotional discomfort or as the result of a personal (including genetic) predisposition. In contrast, other theoretical approaches have attempted to explain adolescent alcohol use within the context of 'normal' development and have acknowledged that drinking in adolescence may in fact have some 'adaptive' functional meanings, such as enhancing self-esteem, sense of self-worth, and sense of belonging to the peer group (Hall, 1904; Addams, 1910; Erikson, 1963; Baumrind, 1987; Silbereisen, Eyferth, & Rudinger, 1986; Bateson, 1991). In the present study, it was found that while there are different ways in which adolescents drink alcohol, the majority of the study participants drank at some time between 10-11 years and 16-17 years. It seems that the 'norm' was to consume alcohol and, more importantly, to 'get drunk'. For this reason, explanatory theories of drinking as a 'normative' behaviour may be better suited to these results. The long and short term consequences of alcohol use are primarily discussed within the field of physical and mental health. However, the reasons for using alcohol, particularly among the adolescent population, go far beyond the health realm, and take on complex social and emotional meanings. These social and emotional meanings are currently under-investigated and need to be explored and examined more fully. For example, we found that there were different developmental pathways leading to the same frequency of drinking behaviour at age 16-17. It would be worthwhile investigating whether the reasons for wanting to 'get drunk' are the same for all adolescents or if they differ depending on the adolescents' specific developmental trajectory of drinking behaviour. When investigators have fully explored the social and emotional meanings of adolescent alcohol drinking we will be in a better position to distinguish between emergent social drinking and emergent problem drinking. In addition, researching of the social and emotional meaning of adolescent alcohol use can provide insights into the phenomenon of 'getting drunk' or 'pretending' to be drunk. Specifically, whether or not these behaviours are associated with problem drinking in consequent years, and therefore, constitute a public health issue for Canadians and needs to be addressed. 3 5 Some may argue that individual differences can only be effectively addressed with individual level intervention methods. However, public and population health approaches to health promotion do in fact already recognize the value of individual differences such as age, gender and ethnicity. By acknowledging that, within the population of adolescents who drink, there are different sub-groups of adolescents that follow different developmental pathways public and population health prevention approaches could better define targets and strategies for program delivery, thereby increasing the probability of success. 36 Chapter III: Theories of normative risk-taking and problem behaviours: searching for the early predictors of smoking and drinking In the previous two chapters, the developmental course of smoking and drinking among a national sample of Canadian children and youth was explored. Unlike the previous two chapters where smoking and drinking behaviours were 'described' as they changed over time, this third chapter attempts to 'explain' why adolescents smoke and/or drink at 14 and 15 years of age. In fact, the purpose of this chapter is to investigate the early patterns of prediction associated with smoking and drinking. In particular, the hypothesis of a 'dose-response' relationship between predictors of occasional smoking and daily smoking is explored, and it is investigated whether smoking and drinking can be 'explained' with similar patterns of predictors, or rather, whether they are better 'explained' by distinct patterns of predictors. The hypotheses tested in this third chapter are grounded in the theoretical background that is discussed in the Introduction that follows. 3.1 Introduction 3.1.1 Risk-taking as normative behaviour Historians have speculated that adolescence is an 'invention' of industrialized societies (Galambos & Leadbeater, 2002) and that it was 'created' so that children could be trained in those technical skills that became necessary to succeed in the highly specialized contemporary labour market (Kett, 1977). Adolescence is commonly defined as the transitional period between childhood and adulthood during which adolescents spend most of their time attending school. The 'invention' of adolescence as a prolonged transitional stage in virtually all contemporary western societies has corresponded in time with the disappearance of formal rites of passage that used to define the movement from childhood to adulthood. 37 Because, in contemporary societies, the transition from childhood to adulthood is no longer sharply defined, adolescents are deemed to experience the transitory nature of their stage and to undergo critical changes (Feldman & Elliot, 1990; Hattie, 1992). Developmental psychologists have characterized adolescence as a time of rapid physical, cognitive and socio-emotional changes resulting in sexual maturity, identity formation and emancipation from childhood dependency (for a general discussion of adolescent psychology see Feldman & Elliot, 1990). Particularly salient is the identity crisis experienced by most adolescents. It is through such a crisis that adolescents question the values of the dominant adult culture eventually resulting in their emancipation and the re-definition of themselves as independent individuals (Baumrind, 1987; Harter, 1990, Hattie, 1992). Risk-taking behaviours play a significant role during adolescence, particularly with respect to identity formation - offering adolescents with a range of experiences that are dangerous, but that can lead to positive gains such as enhanced self-confidence, increased stress tolerance, and overall, a greater sense of belonging (Bateson, 1991; Baumrind, 1987; Silbereisen & Reitzle, 1991). Bateson (1991) points out that "in many of our discussions we have confused the risks inherent in normal development with psychopathology" (p. 353). She talks about 'calibration' where adolescents try to calibrate their behaviours so that their individual experiences match the characteristics of the environment as they perceive it. Thus, through the process of learning to calibrate their behaviours in the social context adolescents necessarily have to engage in some risk-taking in that the outcomes of their behaviours are not always predictable. Bateson also believes that preventing risk-taking behaviours might be wrong because this could "interfere with the process of learning calibration during maturation" (p. 354). Like Bateson (1991), other theorists have argued that total disengagement with risk-taking behaviours could lead to detrimental long-term effects on adolescent social and identity development. For example, Erikson (1959) talked about 'foreclosed identity' as a likely consequence among people who have not undergone the necessary process of emancipation and have remained dependent, and unable to take responsibility for their own actions. Others have 38 argued that risk-taking behaviours are functional to the needs of the developing adolescents particularly as they relate to peer-affiliation, self-concept and identity development (Addams, 1909; Baumrind, 1987; Erikson, 1963; Hall, 1904; Silbereisen & Reitzle, 1991). A vast body of literature in social psychology points to the benefits that derive from a sense of belonging to a group of peers. For example, one of the greatest benefits of feeling part of a group of peers consists of the interactions that take place between the members of the group (Sherif, 1961; Tarrant, 2002). That is, the interactions between group members are at the same time the source and the consequence of the sense of belonging. Research has also indicated that adolescents tend to perceive risk-taking behaviours as 'functional' in that taking certain risks may help adolescents fulfill some of their developmental needs, especially with respect to the sense of acceptance and belonging to the peer group and to the process of identity formation. Green (1997) in an investigation of children's views of accidents found that stories about avoidance and consequences of accidents were "one resource that children draw on to create and demonstrate their social competence" (p. 459). Plumridge and Chetwynd (1999) reported that young recreational injection drug users talked about the fact that injecting provided them with an increased sense of self-will and self-control. Pavis, Masters and Cunningham-Burley (1996) and Pavis, Cunningham-Burley, and Amos (1997) found that it was not uncommon for study participants to report that they smoked or drank to be 'sociable', to be 'part of the group', or to enhance their mood and cope with stresses. Delorme et al. (2003) reported that in several instances their study participants felt happy and relaxed, accomplished and mature, and/or sociable and confident after smoking their first cigarette. While it appears that adolescents perceive benefits associated with risk-taking behaviours when they occur in the context of the peer group, the processes by which these behaviours produce such benefits are largely unknown (Tarrant, 2002). Some have argued that there is need for more studies to focus on the social context in which risk-taking behaviours take place (Baumrind, 1987; Beck & Summons, 1987; Pavis, Cunningham-Burley, & Amos, 1997). F ° r example, Baumrind (1987) claims that in order to gather a fair understanding of risk-taking behaviours, they need to be studied within the context 39 of adolescent development and the youth culture. Her argument is that adolescent development "... is necessarily conditioned by secular events" (p. 94), and that the characteristics of contemporary societies greatly influence the values and meanings of youth cultures. According to this framework, it is important to understand the interplay between the individual, the family and the peer group, but it is also of critical importance to place both the individual and his/her social interactions within a broader historical and socio-political context. 3.1.2 Risk-taking as a manifestation of problem behaviours While for the most part risk-taking behaviours are normal expressions of a developmental stage and are deemed to be limited to the adolescent years, in some cases those behaviours are precursors of more problematic behaviours, long lasting addictions or damaging life-styles. Virtually all of the well established theories of risk-taking and substance use are in fact theories of problem behaviour. An implicit assumption of most theories of substance use is that adolescents engage in behaviours like smoking and drinking because they possess some 'negative' personal and/or social characteristics (e.g., social nonconformity, alienation or rebelliousness, poor self-esteem and/or self-efficacy), and/or because they come from disadvantaged environments (e.g., poor families and neighbourhoods, parents with inadequate parenting skills, etc.). Petraitis, Flay and Miller (1995) eloquently discussed 14 of the most established theories of adolescent substance use, which they grouped into four major categories: cognitive-affective theories, social learning theories, conventional commitment and social attachment theories, and theories with a focus on intrapersonal characteristics. Below, these four categories of theories are summarized. Cognitive-affective theories such as the Theory of Reasoned Action (Ajzen & Fishbein, 1980) and the Theory of Planned Behaviour (Ajzen, 1985,1988) propose that the primary cause of substance use such as smoking and drinking is the adolescent expectation of the consequences of using specific substances. In essence, adolescents' positive cognitions and expectations about 40 smoking cigarettes or drinking alcohol are what cause adolescents to smoke or drink. According to the Theories of Reasoned Action and Planned Behaviour the key to preventing smoking and drinking would be in the modification of beliefs and attitudes about these specific substances. Social learning theories of substance use are more concerned with what causes certain adolescents to create positive beliefs and attitudes (Akers, 1977; Bandura, 1977,1986; Sutherland, 1939). Social learning theories assume that substance-specific beliefs, attitudes and cognitions are formed according to the beliefs, attitudes and cognitions of those people who are considered role models by the adolescents. In terms of prevention applications, proponents of the social learning theories argue that a key to successful prevention lies in diminishing the salience of'substance-using role models' and increasing the importance of'non-substance-using role models'. This way, beliefs, attitudes and cognitions can be formed in reference to more positive role models. Conventional commitment and social attachment theories assume that the primary cause of behaviours such as smoking and drinking is the emotional closeness and attachment to peers who smoke or drink (Elliott, Huizinga, & Ageton, 1985; Elliott, Huizinga, & Menard, 1989; Hawkins & Weis, 1985). According to the conventional commitment and social attachment theories, adolescents who are less committed to conventional society and its institutions such as school and religion, and who have weak bonds with conventional role models such as parents and teachers, are more likely to establish attachments with peers who are involved with substance use. Within the conventional commitment and social attachment theories, there are two specific theories that are worth presenting: the social control theory (Elliott, Huizinga, & Ageton, 1985; Elliott, Huizinga, & Menard, 1989) and the social development model (Hawkins & Weis, 1985). These two theories are particularly relevant to the present study because they are concerned with explaining the actual roots of risk-taking behaviours. That is, these theories propose that the causes of risk-taking behaviours may be present long before these behaviours become manifest and measurable. The social control theory proposes that there are three possible causes of weak commitment to conventional society and institutions and to conventional role models: a) when 41 adolescents perceive a discrepancy between their aspirations and the opportunities for their aspirations to be realized; or, when they feel the relationship with their parents is not a close one, adolescents may feel 'strain'. This strain causes adolescents to grow apart from their parents who normally oppose engagement with substance use, and to create attachments with peers who use substances; b) adolescents who live in disorganized neighbourhoods where crime and unemployment may be common, and attend schools with limited resources are less likely to feel committed to conventional society, and consequently, are more likely to engage in substance use. This is because disorganized schools and neighbourhoods are manifestations of social disorganization that could represent, to adolescents, the failing of established institutions (Kornhauser, 1978); c) adolescents may become involved with substance use as a consequence of attachment to substance using peers, if they have not been properly socialized by their parents to meet conventional standards of behaviour. The social development model also assumes that adolescents not committed to conventional society and institutions and more attached to substance using peers are more likely to use substances themselves. However, proponents of this model (Hawkins & Weis, 1985) attribute different causes for the lack of conventional commitment than those postulated by the social control theory. The social development theory postulates that, depending on the developmental stage, parents, peers and the school influence the individual's behaviour in different ways. For example, adolescents who, in earlier years, did not have positive interactions with their parents and/or teachers and who lacked interpersonal and academic skills were more likely to become attached to and be more influenced by peers who use substances. The emphasis on interpersonal and academic skills of the social development model is of great importance in that, unlike previous theories, it implies that prevention efforts should focus on the improvement of such skills long before the substance use has become manifest. It is also important because in addition to recognizing the importance of social institutions, school and neighbourhoods, it underscores the role that individual differences play in substance use. Consequently, conventional commitment and social attachment theories (i.e., social control theory and social development theory) imply that prevention of behaviours such as smoking and 42 drinking should focus on improving academic achievement and interpersonal skills, quality living in neighbourhoods and communities, and the quality of the relationship between parents and children. Because of these characteristics, these theories can be considered 'holistic' models in that they attempt to explain the interplay between several domains that are relevant to adolescents, such as peers and the family, school and neighbourhoods, and individual level factors such as social and academic skills. Theories with a focus on intrapersonal characteristics. While in the theories so far discussed there is a greater emphasis on the adolescents' social setting (e.g., family and peer relationships), theories with a focus on intrapersonal characteristics place equal emphasis on both the adolescents' social context and the adolescents' individual characteristics. For example, these theories assume that adolescents respond differently to similar social settings depending on their personal characteristics such as self-esteem and coping skills. In the paragraphs that follow the multistage social learning model (MSLM - Simons, Conger & Witbeck, 1988), one of the theories with a focus on intrapersonal characteristics, is discussed in greater detail because of its relevance to the present work. 1 Simons, Conger and Witbeck (1988) tried to address the fundamental question of why adolescents tend to affiliate with peers who use substances. By means of the MSLM, the authors propose that there are three stages associated with the involvement of adolescents with substance use. The first stage describes the causes of the initial involvement with substances and postulates that it particularly affects adolescents who lack long term aspirations related to family, religion and education. This stage also describes those who have poor parental supervision, discipline and support, and who have parents who use substances themselves. The model assumes that adolescents with such characteristics will be more likely to start using substances such as cigarettes, alcohol and other drugs. In the second stage, adolescents who have previously used substances and who lack social and interpersonal skills will tend to affiliate with peers who are deviant and who are substance users. The third stage is concerned with the causes of the escalation from initial use to more regular use and abuse. The third stage of this model postulates that adolescents who are emotionally distressed, lack social and coping skills, and have parents 43 and peers who are substance users will be more likely to continue their engagement with substances and to develop regular patters of use and abuse. The MSLM is important in that it offers a distinction between the numerous causes associated with different levels of engagement with substance use. However, empirical studies have provided only minimal supporting evidence to some of the key predictors of this model (Dembo et al., 1990; Kandel, Kessler, & Margulies, 1978). The majority of the longitudinal studies that measured anxiety and depressive feelings could not find a direct relationship with substance use (Block, Block, & Keyes, 1988; Brook et al., 1990; Kaplan et al., 1986; Kellam, Brown, & Fleming, 1982; Lerner & Vicary, 1984; Pedersen, 1991; Shedler & Block, 1990; Simcha-Fagan & Schwartz, 1986; Teichman, Barnea & Rahav, 1989; White, Pandina, & LaGrange, 1987). However, none of these studies differentiated between experimental and regular substance use. Thus, it may not be possible to draw conclusions with respect to the empirical support, or the lack of, to the MSLM. In addition to the four categories of theories described above, there have been attempts to incorporate these different models into complex and comprehensive theories of substance use and problem behaviour. Among the most commendable of such attempts is Jessor's problem behaviour theory (Jessor, Donovan & Costa, 1991; Jessor & Jessor, 1977) which attempts to explain the causes of several adolescent behaviours that tend to be disruptive to the society as a whole and the adolescents themselves (e.g., substance use, sexual activity, criminal behaviour). In fact, problem behaviour theory argues that adolescents who have the tendency to engage in one type of problem behaviour (e.g., alcohol drinking) are also likely to engage in other types of problem behaviour (e.g., criminal conduct) and that adolescents who have more distant relationships with their parents tend to be more influenced by their peers and more likely to display problem behaviours. Proponents of problem behaviour theory argue that the causes of problem behaviours cannot be simply explained by the sum of a number of risk factors. Rather, the causes of problem behaviours are better understood by a 'covariance' of risk factors where moderating and mediating effects determine the associations between the risk and protective factors and the problem behaviours. This way of conceptualizing the interplay between risk 44 factors is one of the appealing aspects of problem behaviour theory. However, one limitation of this theory is that its proponents have not provided an operational definition of what constitutes problem behaviour. For example, when do risk taking behaviours stop being normative and become problem behaviours? Is the distinction between normative risk taking behaviours and problem behaviours one of frequency or the behaviour, intensity, or associated with its consequences? Implicitly, problem behaviour theory doesn't appear to be concerned with the distinction between normative risk taking behaviours and behaviours that instead are expressions of more serious and long lasting dysfunctional patterns. 3.1.3 The gap between theory and prevention practices The discussion so far presented indicates that in the past few decades, several theories of substance use and risk-taking behaviours have been put forward by several theorists. Despite the fact that, taken together these theories emphasise the importance of a number of factors at the individual level, the family level, the neighbourhood level, and the broader societal level, prevention practices are known to target almost exclusively individual level variables. For example, an extensive review of tobacco prevention programs revealed that regardless of their strategy (e.g., media campaign, skill training, school interventions, peer-led programs) prevention programs tend to be focused on one or more of the following individual level factors: attitudes, beliefs, substance specific cognitions, resistance/refusal skills, self-efficacy and self-esteem (USDHHS, 1994). The fact that considerable efforts are made to prevent risk-taking behaviours by intervening in individual processes and by promoting individual change is likely to be a consequence of the pervasiveness of the epidemiological model that dominates this field of study, and according to which individuals tend to be studied in isolation from their social context (Rhodes, 1997). It is generally acknowledged that theories of substance use, problem behaviour, prevention, and intervention practices are markedly disjointed. As Petraitis, Flay and Miller (1993) have pointed out, in the current state, existing theories of substance use present a major flaw in that they do not provide the "appropriate foundation of effective prevention programs" 45 (p. 67). Despite the fact that theories of substance use have touched on the most important factors associated with such behaviours, they have not elaborated enough on the processes by which these different factors contribute to the etiology and maintenance of substance use. The theories that have elaborated the most on such processes (e.g., theory of reasoned action and planned behaviour, and social learning theories) tend to emphasise individual level factors such as intentions, beliefs and attitudes. Not surprisingly, these are also the theories most commonly used as the basis for smoking prevention programs. However, because attitudes, beliefs and intentions are more proximal predictors in that they are less temporally distanced from the manifestation of specific behaviours and, at the same time, they are more temporally distanced from their underlying causes (Petraitis, Flay & Miller, 1993), significant progress has yet to be made in the development of etiological theories that can form a strong basis for prevention work. One could argue that research that focuses on the 'ultimate' predictors of smoking and drinking (i.e., exogenous factors that put adolescents at higher risk for substance use), as opposed to proximal and distal predictors, has the potential to significantly contribute to a greater understanding of the causal mechanisms involved in substance use. Ultimate predictors can also be conceptualized as predisposing factors that are present before substance use has become manifest, for example, the socialization processes between parents and children from birth to adolescence, or the children's interactions with physical environments such as schools and neighbourhoods. Once it is understood how predisposing factors may contribute to the causes of substance use behaviours, the origins and mechanisms involved in distal and proximal predictors of maintenance or discontinuation of these behaviours may also be better understood. To provide an analogy, predisposing factors are like the beginning of a story without which it is not possible to understand the complex unfolding of events which occur thereafter. As the processes involved in normative risk-raking behaviours may be fundamentally different from the mechanisms responsible for risk-taking as the precursor of problem behaviours, it becomes imperative for researchers (and practitioners) to make a clear distinction between normative risk-taking and risk-taking as expressions of problem behaviours. 46 This distinction, however, is rarely made. In the literature, 'risk-taking' and 'problem behaviours' are often used interchangeably to characterize adolescents who at any time during their adolescent years have smoked, drank or used other illicit substances. Individual differences in frequency, intensity, and duration of such uses, and, more importantly, in the developmental changes of these behaviours over time are rarely taken into account. Risk-taking behaviours such as smoking and drinking have generated a particularly rich body of research not only because they tend to be highly prevalent in adolescence (Glantz & Pickens, 1992; Kandel & Yamaguchi, 1985; Statistics Canada, 2000; Windle, 2003) but also due to the health consequences associated with these behaviours (Surgeon General Report, 1992; Rehm et al., 2003). Researchers recognize that there are adolescents who engage in such behaviours only temporarily (i.e., they are 'experimenters'), or on an occasional basis (i.e., they are non-daily users, or 'occasional' users) and that only a proportion of the adolescent population will actually develop nicotine or alcohol dependence in later years (Canadian Tobacco Use Monitoring Survey, 2003; USDHHS, 1994 - for a discussion on the developmental course of smoking and drinking see Chapter I and Chapter II). However, since the distinction between normative risk-taking and problem behaviours is rarely made, it is difficult to isolate the predictors of smoking and drinking as part of normal development from smoking and drinking as symptoms (or manifestations) of problem behaviours. In addition, because the majority of current smokers are also current drinkers or binge drinkers (e.g., Bobo & Husten, 2000; Bien & Burge, 1990; DiFranza & Guerrera, 1990; Jackson, et al., 2002), it is also difficult to isolate the factors that uniquely predict smoking from those factors that instead uniquely predict drinking (clearly smoking and drinking may also share common predictors). As prevention programs tend to be substance specific, the ability to isolate unique predictors of specific substance use becomes of great practical importance. While different substances may share common causes, mediating and moderating factors may act differentiy in determining the etiology of specific behaviours. A better understanding of the processes and mechanisms that govern substance specific mediators and moderators could become critical information for planners of prevention programs. 47 Furthermore, if the purpose is to understand the etiology of substance use as a manifestation of problem behaviours, it is imperative that an operational distinction is made between such behaviours and substance use as part of normal adolescent risk-taking. This distinction would allow the creation of more precise theories that could in turn be used to inform prevention programs targeting high-risk population rather than the general adolescent population of substance users. The discouraging results in prevention, particularly in the prevention of smoking (Ellickson, Bell, & McGuian, 1993; Vartiainen, et al., 1990), have led some researchers to conclude that in order to be successful, prevention needs to start at earlier ages, before adolescents have even tried smoking. It has also been argued that in fact, the theories that are used as the basis for prevention initiatives do not adequately integrate the current knowledge of the factors that are involved in adolescent risk-taking (Petraitis, Flay & Miller, 1995). Both standpoints, however, share the common underlying conviction that all substance use behaviours are problematic and therefore need to be eradicated. Nonetheless, as previously discussed, for most adolescents, risk-taking such as adolescent substance use is part of normal development and as such, it may fulfill important developmental needs. Acknowledging the social and emotional benefits of risk-taking may allow researchers to better understand the difference between those behaviours that are part of normal development versus those that instead are expressions of problem behaviours. 3.1.4 Recent advances in research and theory Some progress in research and theory development is currently underway. First, researchers are increasingly aware of the limitations of cross-sectional designs and of the need for longitudinal studies on the developmental course of behaviours such as smoking and drinking (e.g., Mayhew, Flay, & Mott, 2000; Chassin et al., 2000; Colder et al., 2000). Second, more qualitative work is being conducted that aims at unveiling the processes involved in risk-taking behaviours in the context in which they occur (Pavis, Masters & Cunningham-Burley, 1996; Pavis, Cunningham-Burley, &Amos, 1997; Delorme, 2003; Green, 1997; Plumridge &Chetwyn, 48 1999)- Third, recent advances in statistical modelling have provided researchers with more alternatives for the analysis of longitudinal data (Nagin, 1999), particularly with respect to the identification of high-risk subgroups of adolescents. One example is the growing body of literature wherein person-centered approaches are used to study the developmental trajectories of smoking and drinking in adolescents (see Chapter I and Chapter II; Chassin et al., 2000; Colder et al., 2000; Colder et al., 2002; Wills et al., 1996; Stice, Myers, & Brown, 1998). These studies indicate that the adolescent population is composed of different subgroups, each of which displays distinct developmental behavioural patterns. One of the values of this new line of research is that it highlights the importance of conceptualizing adolescents not necessarily as a homogeneous population, but rather as one that displays heterogeneous patterns of behaviour and for which differences between groups (rather than similarities) need to be investigated. In this context, theories of problem behaviour and theories of normative risk-taking behaviours may become equally valuable in that both can adequately explain the behaviours of different subgroups of adolescents: theories of problem behaviour may be better suited to explain the behaviours of adolescents who, for example, use substances regularly, in large quantities, or for prolonged periods of time (i.e., the 'high-risk' group), whereas theories of normative risk-taking behaviours may be better suited to explain the behaviour of adolescents who use substances moderately, temporarily, and/or occasionally. 3.1.5 Rationale and purposes of the present study The purpose of this study is to investigate the early patterns of prediction associated with smoking and drinking, that is, the predictors that are most closely associated with the actual causes of these behaviours. Here, the hypothesis of a 'dose-response' relationship between predictors of occasional smoking and daily smoking is explored, and it is investigated whether smoking and drinking can be 'explained' with similar patterns of predictors, or rather, whether they are better 'explained' by distinct patterns of predictors. This study follows two previous studies that aimed at identifying the developmental trajectories of smoking and drinking between 10-11 and 16-17 years of age (see Chapter I and 49 Chapter II). These studies found that there are different developmental patterns of use for both smoking and drinking in that there are distinct ways in which different subgroups of adolescents engage in smoking and drinking over time. It would be ideal to investigate these early patterns of prediction using the same participants that were included in the previous two studies where developmental trajectories of smoking and drinking were identified. Such an approach would allow the identification of early prediction of developmental trajectories rather than the identification of patterns of prediction of 'end point' behaviours. Unfortunately, the number of participants eligible for inclusion in the previous two studies was limited to a few hundred, and missing data at the level of the predictors would also further reduce the potential for obtaining statistically valid models of prediction. For these reasons, the results of the two studies described above were used to inform the decision as to what 'end point' of these behaviours was worth investigating. The main criterion that we used in making such a decision was that of the 'stability' of behaviour. In other words, by looking at the trajectories of smoking and drinking we identified the age at which, for the most part, the trajectories seemed to have become stable. For example, it was found that the differences between groups in terms of frequency and intensity of both smoking and drinking tended to decrease over time and that at age 14-15, participants tended to be grouped into three major categories: those who drank or smoked frequently, those who drank or smoked occasionally and those who did not smoke or drink at all. Furthermore, at 14-15 years of age, the highest prevalence rates of any smoking and drinking behaviours were observed. Finally, the difference between the smoking and drinking behaviours at 14-15 years of age and at 16-17 years of age was insignificant. These results lead to the conclusion that among the study participants, and for the period of time under study, smoking and drinking behaviours tended to become 'stable' by age 14-15. It also led to the belief that smoking and drinking at age 14-15 was a good indicator of these behaviours at age 16-17. While similarities were found in smoking and drinking, some critical differences in these behaviours were identified. First, drinking turned out to be more prevalent than smoking: at age 50 14-15> 98.6% of the participants had tried drinking at least once while 81.4% had tried smoking at least once. This result is similar to what was reported for American adolescents in that the prevalence of nicotine dependence appeared lower compared to that of alcohol dependence (Anthony & Echeagaray-Wagner, 2000). Second, nearly all the times participants engaged in drinking, they reported to have gotten drunk; on the other hand, a very small proportion of participants who smoked were heavy smokers. Finally, virtually none of the participants who drank reported drinking more frequently than once or twice a month. On the other hand, about 38% of participants who smoked reported smoking daily. These results revealed that while drinking was overall a more prevalent behaviour, participants tended to smoke more frequently. The findings of the two previous studies discussed above provided the rationale for studying smoking and drinking separately: while smoking and drinking tend to be concurrent behaviours (Bobo & Husten, 2000; Bien & Burge, 1990; DiFranza & Guerrera, 1990; Jackson, Sher, Cooper & Wood, 2002) there might be unique predictors of each of these, and specific theoretical models for smoking and for drinking may have better explanatory power than a non-substance-specific theory. As previously discussed, the focus is on the ultimate predictors that are also 'early' predictors of smoking and drinking in that they are temporally distanced from the time these behaviours start manifesting themselves. In addition, it is argued that if infrequent substance use is one 'step' below frequent use on the continuum for 'problem behaviour' then it should be found that the same factors predict both infrequent and frequent substance use. However, the strength of the association between the predictors and infrequent substance use should be weaker than the strength of association between the predictors and frequent substance use. In other words, if infrequent use and frequent use are two incremental steps on the same continuum, a 'dose-response' relationship should be observed between substance use behaviours and their predictors. On the contrary, if infrequent substance use and frequent substance use are two separate phenomena, then, for the most part, different factors should predict each of these two behaviours. The expectations are that infrequent substance use and frequent substance use are, in fact, two independent phenomena and that they are predicted by different factors. 51 In order to test the hypothesis that infrequent use and frequent use are two separate phenomena data were used from the National Longitudinal Survey of Children and Youth (NLSCY). Fourteen and fifteen year old participants were identified in 2000/01 when the most recent follow-up was conducted. This allowed us to identify ultimate predictors at the youngest ages possible using participants for whom valid observations were available from the first assessment conducted in 1994/95, when participants were 8-9 years old and when smoking and drinking were not yet manifest behaviours. It is worth noting that the participants included in the present study are not the same participants who were included in the two previous studies (described in Chapter I and Chapter II) who in 2000/01 were 16 and 17 years old. Preliminary descriptive analysis of the participants included in the present study revealed that only 14 participants reported drinking on a weekly basis in 2000/01 when they were 14-15 years old. This result is similar to that obtained in the previous study (Chapter II) where it was found that virtually none of the participants were weekly drinkers by the age of 16 and 17. Because of the very low prevalence of weekly drinking among the study participants, the hypothesis that infrequent drinking and frequent drinking are two distinct phenomena could not be explored. However, it was still be possible to explore the association between infrequent drinking and a number of early predictors as described below. To summarize, the purposes of the present study were to: 1. Test the hypothesis of a 'dose-response' type of relationship between predictors of occasional smoking and daily smoking; 2. Quantify the degree of association between smoking and drinking and a number of individual, family, and neighbourhood predictors; 3. Compare and contrast the predictors of smoking and drinking; 4. Propose theoretical frameworks for the early prediction of occasional and daily smoking; 5. Propose a theoretical framework for the early prediction of infrequent drinking. 52 3.2 Method 3.2.1 Data Source: the National Longitudinal Survey of Children and Youth The National Longitudinal Survey of Children and Youth (NLSCY) is a joint initiative by Human Resources and Development Canada (HRDC) and Statistics Canada. The purpose of the NLSCY is to "develop information for policy analysis and program development on critical factors affecting the development of children" (Statistics Canada, 1995, pg. 1). The NLSCY began surveying households with children between o and 11 years of age in 1994/95. Eligible households were selected from participants in the Labour Force Survey and the National Population Health Survey. The Labour Force Survey and the National Population Health Survey are two Canadian national surveys on labour market, and physical and mental health, respectively. A total of 15,579 households were surveyed in 1994/95. Once the households were identified, up to four children per household were randomly selected to be followed longitudinally in subsequent years. Households and their children were followed up at two year intervals. Four follow ups to date have been completed the most recent one for which data are available was conducted in 2000/01. 3.2.2 Data Collection In 1994/95, the most knowledgeable person of the child (PMK) was asked to respond to a questionnaire regarding household composition, general health information of the parents, neighbourhood safety, and social support. In addition the PMK was asked several questions specific to the child's health and behaviour. In 98% of the cases the PMK was the mother of the child. The data collection procedure with the PMK relied heavily on computer assisted interviewing (CAPI) technology, and occurred either over the telephone or by face-to-face interviews. Interviews were conducted by interviewers who were trained to ensure that they had adequate understanding of the survey concepts. 53 At the end of each interview, the PMK of school age children also were asked to provide written permission to allow for information to be collected from teachers and principals. Teachers and principals were mailed a questionnaire about the children's school related behaviours, the school's characteristics, and the disciplinary climate of the school. Teachers and principals were expected to respond to the questionnaire and mail it back to Statistics Canada. Finally, interviewers who conducted interviews directly at the respondent's house were expected to complete a questionnaire on their perceptions of the neighbourhood in which the respondent lived. Information collected in 1994/95 from the PMK, the teacher, and the interviewer was used to identify the predictors for this study. Because of a high non-response rate associated with the teachers' and principals' questionnaires (only 10% of the eligible participants had complete teacher and principal reports) it was decided not to include those variables among the predictors tested in the present study. In 2000/01, the children for whom the PMK provided information in 1994/95 also completed an age specific questionnaire. Age specific questionnaires were available for participants aged 10-11,12-13,14-15, and 16-17 years. The present study used responses that participants provided to the age specific questionnaire at 14-15 years. The self-administered questionnaire covered a variety of topics including smoking and drinking and was used to measure the outcomes for this study. The interviewer travelled to the participants' houses and delivered the questionnaire to the participants in person. The interviewer also instructed the participants as to how to fill in the questionnaire and encouraged them to find a quiet and private place within the home where they could self-administer the questionnaire. To ensure confidentiality, the participants placed the completed questionnaire in an envelope, sealed it, and returned it to the interviewer. The interviewer was not allowed to open the questionnaire until having reached the research facilities. 54 3-2.3 Participants In 1994/95 3,514 8- and 9-year-old children were assessed. Of these children, 2,014 were also assessed in 2000/01. The reason for this difference in the number of participants from 1994/95 to 2000/01 is that in several cases information regarding the outcomes was not available. For example, some participants only responded to the survey in 1994/95 and either did not respond to any of the other three follow-ups, or did not participate in the 2000/01 follow up, only. In some cases, even if they did take part in the 2000/01 survey follow up, incomplete responses to the age specific self-administered questionnaire were provided. In other instances, a small number of participants moved permanently from the country and could not be located. Overall the response rate to the self administered questionnaire for 14-15 years old was 77% in 2000/01. See Table 1.1 from Chapter I for the socio-demographic characteristics of the participating household in 1994/95. In 2000/01, out of the 2,014 who were assessed in 1994-95,1,414 participants provided answers with respect to smoking and 1,422 participants provided answers with respect to drinking. This represents about 70% of the available participants but only 40% of the initial sample of 3,514 that was assessed in 1994/95. While 'sample mortality' (the phenomenon by which participants drop out of a study) is typical in long term longitudinal studies such as the NLSCY, there may be some limitations associated with the design and methods of administration of the NLSCY that are also responsible for this high reduction of sample participants from one follow up to the next. Thus, the question that necessarily arises when such a low response rate is experienced is whether the participants who dropped out of the study are at all similar to the participants who are included in the analysis. To address this issue we conducted an attrition analysis where the participants who did not respond to the survey in 2000-01 (non-respondents - i.e., those for whom we did not have outcome data) were compared to the participants who provided answers in 2000-01 (respondents - i.e., those for whom we did have outcome data). Results of this attrition analysis are presented in Appendix V. 55 3-2.4 Outcomes In the present study, 'normative risk-taking' is defined as occasional or moderate uses of alcohol and cigarettes (i.e., occasional smoking and infrequent drinking) and 'problem behaviour' as regular and frequent uses of these substances (i.e., daily smoking and weekly drinking) at age 14-15. Although somewhat arbitrary, these definitions may in fact be a good approximation of these behaviours in mid adolescence. The majority of adolescents in the study at age 14-15 were non-smokers, non-drinkers, occasional smokers (Chapter I) and/or infrequent drinkers (Chapter II). To identify non-smokers/non-drinkers, occasional smokers/infrequent drinkers and daily smokers/weekly drinkers, the question was asked: "Which of the following best describes your experience with smoking cigarettes/drinking alcohol?" Participants who responded "I don't smoke/drink" were coded as 'o' and were used as referents in the logistic models described below. Participants who responded that they smoked/drank but not daily (for smoking) and less than weekly (for drinking) were coded as '1'. Finally, participants who reported smoking daily and/or drinking weekly were coded as '2'. As previously noted only 14 participants reported that they drank more than once or twice a month. Thus in the present study, the association between predictors and weekly drinking could not be explored. However, the associations between predictors and infrequent drinking were investigated. It is important to note that a purpose of the present study was to identify unique predictors of drinking and smoking. The ideal strategy to address this question is to conduct analyses separately for smoking and drinking, with participants who either only smoke or only drink. However, because virtually all adolescents who smoked also reported drinking (only 30 moderate smokers and 13 daily smokers did not drink at all), there were not enough participants who either only smoked or drank. Thus, it was decided to 'control' for smoking in the analysis with drinking as the outcome, and to 'control' for drinking in the analysis with smoking as the outcome. This way it was still possible to identify unique predictors of smoking and drinking with an adequate sample size (see Analytical Plan for more details). 56 3-2.5 Predictors The general purpose of the NLSCY is to collect information about critical factors affecting the development of children in Canada. The NLSCY is a study of general issues associated with child development and, as such, it was designed to examine a broad range of factors that influence, both positively and negatively, children's development with regard to the characteristics of the individual, the family, the neighbourhood and the school. The process involved an expert panel of academics and researchers in the areas of child development, as well as federal and provincial officials who provided advice on the design and on the content of the survey. The selection of the most appropriate measures to be included in the NLSCY was based on a number of criteria. The measures had to be concise, suitable for use in a household survey, and compatible with studies conducted both within Canada and abroad. In addition, scales used to measure constructs had to have adequate psychometric properties even when the scales had been adapted or modified6. Finally, scales had to be available both in English and French. A complete list of the content of the NLSCY is included as Appendix I. For a detailed description of each question and scales of Cycle 1 of the NLSCY please refer to "Overview of Survey Instruments for 1994/95 Data Collection Cycle 1" (Statistics Canada, 1995 - Catalogue No. 95-02). 3.2.6 Analytical Plan The sampling procedure used in the NLSCY was designed with the intent to select a representative sample of Canadian children and consequently, to allow inferences that are generalizable at the population level. For this reason, Statistics Canada strongly recommends the use of statistical weights when conducting analysis using NLSCY data. 6 In some instances questions were modified or added to a scale, or a given scale had never been used for Canadian children. Consequently, there was the concern that the factorial structure of the scales used in the NLSCY might have been different from those found in the literature. Thus, a new factor analysis was performed On all scales to determine the factors inherent to each scale, scales scores were calculated based on the factors identified, and psychometric properties of the scales were measured. 57 Generally, there are two types of weights that can be used in longitudinal surveys to adjust for complex sampling procedures and loss to follow-up. The first type of weight is based on recorded variables: for example, the socio-economic status of the participants, their language or employment status, and is used to compensate for the loss to follow-up that affects most longitudinal surveys where only a selected sub-group of the initial sample consistently responds to the survey follow-ups. The second type of weights is calculated to compensate for 'errors' associated with sampling procedures such as selection bias, or clustering sampling with probability proportional to size (Winship & Radbill, 1994). In other words, these design weights are used to assure that the proportional contribution of specific geographical areas at the national and provincial level is correctly reproduced within the survey sample. In the case of the NLSCY the weights are complex in the sense that they combine the two types of weights described above. While Statistics Canada strongly recommends that weighted data always be used when conducting analysis on the NLSCY, the use of weights based on recorded individual characteristics is not always recommended. This is particularly true when the purpose of the analysis is the exploration of the associations between any given outcome and the participants' characteristics that are used to calculate the weights. For example, Winship and Radbill (1994) argue that the use of non-weighted data is preferred when the researcher's interest is estimating relationships through multivariate analysis and regression analysis (rather than estimating prevalence) because such data produces less biased ordinary least square (OLS) estimates and smaller standard errors than the weighted OLS estimates. In other words, they recommend that, rather than using weighted data, the participant characteristics that have a theoretical interest for the researchers be included in the modelling and not be used to calculate the weights. The primary objective of this study is to explore associations between outcomes and predictors, including some of those predictors that Statistics Canada used to calculate their weights (e.g., socio-economic status and employment status). In addition, the weights provided by Statistics Canada are a complex combination of design weights and weights based on recorded 58 variables. Therefore, following recommendations by Winship and Radbill (1994), the analysis described below was conducted with non-weighted data. 3.2.7 Data reduction procedure The number of predictors relevant to the present study and included in the NLSCY was considerably large. There are several problems associated with investigating too many predictors at once. First, there is the argument that the investigation is not theoretically grounded and that the researcher is undertaking a 'fishing expedition'. A second argument, related to the previous one, is that by investigating too many predictors at once, the researcher is bound to find significant results that are likely to be obtained by chance rather than because of 'true' associations. To address these arguments, predictors were selected that had been investigated in previous studies on smoking and drinking and that were hypothesised as important by previous theorists. Extensive reviews on smoking and drinking such as those provided by the United Stated Department of Health and Human Services (1994) and the U.S. Department of Education (1993) were used as guiding documents in the selection of the predictors. These predictors include factors at the individual level, the family level and the neighbourhood level. Appendix II shows the predictors (and specific questions) that were selected for inclusion in the analysis. Clearly, the list of predictors was still considerably large and needed to be further reduced. Thus, a rigorous process of elimination was applied based on empirical criteria that consisted in three phases. Phase 1: Grouping of predictors Predictors were grouped according to characteristics of the children, the family, and the neighbourhood. Within each of these broader categories sub-categories of predictors were identified on the basis of conceptual similarities. The sub-categories generated are as follows: child demographics, child behaviour, child health, literacy and mathematics skills, extracurricular activities, peer relationships, relationships with adults, mother's depression and isolation, parenting and family functioning, mother's socio-demographic characteristics, 59 mother's smoking, mother's drinking, father's socio-demographic characteristics, father's smoking, father's drinking, family socio-demographic characteristics, siblings, and neighbourhood characteristics. It is worth recalling that only children 10 years and older were asked to respond to self-administered age appropriate questionnaires. Thus, because in 1994/95 our study participants were 8 and 9 years old, none of the predictors analysed here were obtained from the children's self-reports. Phase 2: Data reduction process Two separate data reduction processes were conducted, one for smoking and one for drinking. Smoking To test the 'dose-response' hypothesis, a series of multinomial regression analyses was conducted separately for each sub-category of predictors. In these analyses, participants who did not smoke at all were assigned a value of 'o' and were used as the referent. Participants who reported to be occasional smokers were assigned a value of '1,' and participants who reported to be daily smokers were assigned a value of'2'. Eighteen multinomial regression models were tested. Several predictors were removed as a result of this initial analysis. Three empirical criteria were used for the elimination of predictors: a) Predictors that were collinearly associated with each other were eliminated automatically by the program STATA.8 (Stata Corporation, 2003) that was used to conduct this part of the analysis. STATA.8 eliminates variables above the threshold for co-linearity according to the Variance Inflation Factor (VIF). This procedure eliminates only one of the variables that are collinearly associated, that is, the variable that explains the least of the variance; b) Predictors were eliminated that generated 'true' non significant values (i.e., sample size was adequate to estimate statistically meaningful relative risks) at both levels of the outcome. Predictors had to be non-significant both at the level of non daily and daily smoking; 60 c) It was planned to eliminate predictors that could not provide meaningful results because thie analysis was based on too small a sample (i.e., lack of power). None of the predictors were eliminated at this stage due to lack of power. Appendix III shows the predictors of smoking that were eliminated as part of this process and the reason for their elimination. Results of the multinomial logistic regression analyses revealed that three of the several predictors examined had a 'dose-response' relationship with the outcome (see the Results section below for details). These results led to the conclusion that while in some instances they do share common predictors, there are several variables that predict either occasional or daily smoking, but not both. Consequently, it was decided to treat occasional and daily smoking as two separate outcomes and to conduct binomial logistic regression analysis in the remaining modelling phase of the analysis. Drinking A series of logistic regression analyses was conducted separately for each sub-category of predictors described above where participants who did not drink at all were coded as o, and participants who drank infrequently were coded as 1. As for smoking, 18 separate logistic regression models were tested, one for each sub-category. Several predictors were removed as a result of this initial analysis. Two empirical criteria were used for the elimination of predictors: a) Predictors were eliminated that generated 'true' non significant values (i.e., sample size was adequate to estimate statistically meaningful relative risks); b) It was planned to eliminate predictors that could not provide meaningful results because the analysis was based on too small a sample (i.e., lack of power). None of the predictors was eliminated at this stage due to lack of power. Appendix IV shows the predictors of drinking that generated non-significant results and that were eliminated as part of this process. Phase 3: Modelling In this next phase of the analysis, three separate modeling exercises (one for each outcome - occasional, daily smoking, and infrequent drinking) were conducted. Models were 61 tested that included the predictors that were not eliminated as a result of the previous data reduction procedure. A stepwise logistic regression procedure was used for the identification of predictive models of non daily smoking, daily smoking, and infrequent dinking. Employing a stepwise procedure can provide an effective way to screen a large number of predictors and to fit several logistic regression equations simultaneously (Hosmer & Lameshow, 2000). In stepwise logistic regression, first a fit for the 'intercept only model' is calculated and a value for its log-likelihood is provided. Log-likelihood is the statistic used to assess the significance of models that follow binomial distributions such as those of logistic regressions. A backward (as opposed to forward) stepwise procedure was selected where all predictors were entered at once and then removed one at a time from the model. The backward stepwise procedure was the method most consistent with the process of elimination that had been adopted so far. Following recommendations by Hosmer and Lameshow (1990), a PE = .20 was specified as the 'alpha' level judging the importance of variables to be included in the model. Because the choice of PE determines how many variables eventually are included in the model, the value that the researcher specifies can have important implications for the results of the analysis. Research has indicated that a PE = .05 is too stringent and that important variables may be excluded from a model (Bandel & Afifi, 1977; Costanza & Afifi, 1979; Lee & Koval, 1997). Consequently, Hosmer and Lameshow (1990) highly recommend a PE in the range between .15 and .20. 3 . 3 R e s u l t s The analytical plan outlined above was applied to address three of the five purposes of the present study: 1) test the hypothesis of a 'dose-response' relationship between predictors of non-daily and daily smoking; 2) quantify the degree of association between smoking and drinking and a number of individual, family, and neighbourhood predictors; and 3) compare and contrast the predictors of smoking and drinking. The fourth and fifth purposes of the present study were to propose theoretical frameworks of early predictions of occasional smoking, daily smoking, and infrequent drinking, and which are addressed in the Discussion. 62 3-3-1 Early Predictions of Occasional and Daily Smoking The first part of the analysis was conducted with the dual purpose to test the hypothesis of a 'dose-response' relationship between predictors of occasional and daily smoking and to reduce the number of predictors on the basis of empirical findings. Descriptive analyses revealed that of the 1,257 participants who were assessed in 1994-95,14° (11%) had become occasional smokers, and 157 (12%) had become daily smokers in 2000-01. In total, about 23% of the participants reported smoking either occasionally or daily. The prevalence of smoking found here is similar to that observed in other national studies where about 22.5% of teens aged 15-19 reported themselves as current smokers in 2001 (Health Canada, 2001). Whether participants had ever tried drinking was also included in each of the 18 multinomial regression models to control for the effect of drinking as described in the Analytical Plan. In all cases, drinking was a significant predictor of both occasional and daily smoking. However, the results associated with this predictor are not discussed here because they could be misleading. In fact, participants' drinking behaviours were assessed at the same time as the outcome, and thus, drinking cannot be considered as one of the early predictors that were measured six years prior to the assessment of the outcomes. The only reason for having included drinking in these models was to control for its effect in the analysis and to identify predictors that were uniquely associated with smoking. Table 3.3.1 reports the results of the multinomial regression analysis for variables predicting occasional and daily smoking. The results in bold indicate those variables that were found to be predictive of both non-daily and daily smoking. 63 Table 3.3.1 Summary of 18 multinomial regression analyses that were conducted separately for each sub-category of predictor for occasional and daily smoking Predictor n RRR1 S.E. 95% CI p Occasional Smoking Daily How often do you check his/her homework or provide help with homework? 703 •837 .066 .716-.979 .027 About how many close friends does he/she have? 1171 .769 .103 .591-1.001 .051 How often does he/she hang around with kids you think are frequently in trouble? * 1171 .748 .106 •566-.988 .041 Intact Family Status t 1397 .007 .002 .003-.013 .000 Hostile/Ineffective Parenting 1348 1.080 .030 1.022-1.142 .006 Family Functioning* 1348 •955 .019 •918-.994 .026 moking Hyperactivity/Inattention 1321 1.123 •034 1.057-1.192 .000 Gender 1405 .674 .124 .469--969 •033 How often has he/she taken part in music, dance, art or other non-sport activities? * 1353 1.293 .123 1.072-1.558 .007 How often does he/she hangs around with kids you think are frequently in trouble? * 1171 .625 .084 .480-.815 .001 Would you say that your child health is? * 1386 1.614 .182 1.293-2.015 .000 Social Support 1370 .918 .030 .860-.980 .010 Mother: number of cigarettes smoked per day 401 1.038 .018 1.002-1.075 •034 Mother: how often has drank alcohol 404 1.289 .146 1.032-1.609 .025 Father: how often has drank alcohol 542 1.270 .130 1.039-1-553 .020 Father: number of times had 5 or more drinks at once 542 1.008 .003 1.001-1.014 .018 64 Table 3.3.1 Continued Income Adequacyt+ 1397 1.565 •342 1.018-2.404 .041 Socioeconomic Statusm 1397 .556 .106 .383-.809 .002 Annual Household Income 1397 .980 .009 .962-.999 .044 Intact Family Status* 1397 .005 .001 .002-.009 .000 Older Siblings 1405 .358 .185 .129-.987 .047 Younger Siblings 1405 •347 .187 .120-.998 .050 1 Relative Risk Ratios + An intact family consists of a married or common-law couple where all children are the natural and/or adopted offspring of both members of the couple. A score of 1 means that the child is member of a single parent family, is a foster child or does not live with a parent; a score of 2 means the child is not a member of an intact family but is in a couple census family; a score of 3 means that the child is from and intact family. n Income Adequacy is a measure of household income proportional to household size. t + t Socio-economic status was derived for each household in the sample and the result assigned to each selected child in that household. It was derived from five sources: the level of education of the PMK, the level of education of the spouse/partner, the prestige of the PMK's occupation, the prestige of the occupation of the spouse/partner, and household income. *This variable is reverse coded where higher scores mean less frequency or less positive outcomes. Table 3.3.1 indicates that there are six factors that are significantly associated with occasional smoking. These factors are: parental involvement with homework, number of close friends, family status, involvement with 'troubled' friends, hostile and ineffective parenting, and family functioning. Table 3.3.1 indicates that there are several factors that are significantly associated with daily smoking. These factors are: gender, having siblings, family status, annual income, socio-economic status, social support, hyperactivity and inattention, participation in extracurricular activities, involvement with 'troubled' friends, health status, income adequacy, maternal smoking and drinking behaviours, paternal drinking behaviours. Results of the multinomial logistic regressions reported above indicate that two out of the seven predictors that yielded significant results at the level of occasional smoking were also significantly associated with daily smoking. These predictors are: involvement with 'troubled' friends and family status. The relative risk ratios associated with these variables were larger for daily smoking than for occasional smoking, but the confidence intervals for these values 65 overlapped. This indicates that while these two variables may be significantly associated with both daily and occasional smoking, their association with the outcomes cannot be characterized in terms of a 'dose-response' relationship. It is important to remember that these associations reflect the results of analysis conducted separately for each sub-category of predictors. For example, involvement with 'troubled' friend was one of the predictors that belonged to the 'Peer Relationships' sub-category. On the other hand, family status belonged to the sub-category of 'Family Socio-Demographic Characteristics'. Thus, these finding are better interpreted as preliminary results. In fact, while these findings suggest that that both occasional and daily smoking may be predicted by these two variables, the interplay between these and the other significant predictors presented in Table 3.3.1 may be different for occasional smoking and daily smoking. Following these preliminary analyses, two binomial logistic regressions were conducted (one with occasional smoking as the dependent variable and one with daily smoking as the dependent variable) where all the significant predictors summarized in Table 3.3.1 were included in the regression models. Maternal smoking and drinking and paternal smoking and drinking were reported by only a very small proportion of participants. Specifically, 28% of maternal smoking and drinking and 38% of paternal smoking and drinking were reported by the PMK in 1994/95. In other words, mothers only responded to questions about their own smoking in 28% of the cases and to questions about their child's father's smoking in 38% of the cases. This response rate is very low, and the associations found between such behaviours and the participants' smoking behaviours may be biased. For this reason, despite the fact that a significant association was found between parental smoking and drinking and adolescent smoking, it was decided not to include these predictors in the next phase of the analysis. It is worth pointing out that a low response rate such as the one observed here may be indicative of the fact that parental smoking and drinking may be considered sensitive topics by the parents, and consequently, they may choose not to report these behaviours to the interviewer. It is possible that the low response rate associated with reports of 6 6 parental smoking and drinking may be reflective of the fact that the strategy employed in the NLSCY to collect data is not the most appropriate in the case of sensitive topics. The outcomes for smoking were dichotomised so that in one case non-smoking = o and occasional smoking = 1 (daily smokers were excluded from the comparison), and in the other case non smoking = o and daily smoking = 1 (occasional smokers were excluded from the comparison). Then, two separate logistic regression models were tested as previously described in the Analytical Plan. Whether participants had ever tried drinking was also included in regression models for smoking to control for the effect of drinking. The summary statistics of the models tested in the stepwise backward logistic regression for occasional smoking are presented in Table 3.3.2. Table 3 . 3 . 2 Model summary of the stepwise backward logistic regression analysis for occasional smoking ( n = H 2 i ) % Correct -2 Log Cox & Snell Nagelkerke Chi-Square* Sig. Classification Likelihood R Square R Square (df) Step 1 88.7 683.624 .092 .184 5.700(8) .681 Step 2 88.7 683.834 .092 .182 5.624(8) .689 Step 3 88.7 684.693 .091 .180 8.449(8) •391 Step 4 88.7 685.850 .091 .179 6.519(8) .589 Step 5 88.7 687.050 .090 .177 13.012 .111 tHosmer and Lemeshow Test Step 1 = How often do you check his/her homework or provide help with homework? About how many close friends does he/she have? How often does he/she hang around with kids you think are frequently in trouble? Intact Family Status; Hostile/Ineffective Parenting; Family Functioning; Has ever tried drinking. Step 2 = How often do you check his/her homework or provide help with homework? About how many close friends does he/she have? How often does he/she hang around with kids you think are frequently in trouble? Hostile/Ineffective Parenting; Family Functioning; Has ever tried drinking. Step 3 = How often do you check his/her homework or provide help with homework? About how many close friends does he/she have? Hostile/Ineffective Parenting; Family Functioning; Has ever tried drinking. Step 4 = How often do you check his/her homework or provide help with homework? About how many close friends does he/she have? Hostile/Ineffective Parenting; Family Functioning; Has ever tried drinking. Step 5 = About how many close friends does he/she have? Hostile/Ineffective Parenting; Family Functioning; has ever tried drinking. 67 The percent of correct classification of the model presented in Table 3.3.2 is estimated as 88.7%. However, the positive predictive value and negative predictive value of the model was estimated as 0% and 100%, respectively. This indicates that the overall percent of correct classification overestimates the predictive value of the model and this is possibly due to the fact that only about 13% of the participants were occasional smokers. This low prevalence of occasional smoking has resulted in an unbalanced comparison of two groups, where one (the non-smokers) is overrepresented than the other (the occasional smokers). Table 3.3.3 summarizes the results related to the model in step 5, that is, the model that the statistical procedure has identified as the one best describing the patterns of association found in the data. Table 3 . 3 . 3 Summary of the final model obtained with logistic egression for occasional smoking ( n = H 2 i ) Predictor OR S.E. 95% CI P About how many close friends does he/she have? .838 .121 .662-1.062 .144 Family Functioning* •954 .020 •916-.993 .020 Hostile/Ineffective Parenting 1.067 .026 1.050-1.122 .000 Has ever tried drinking** 9.292 .291 5.251-16.445 .000 *This variable is reverse coded where higher scores mean less frequency or less positive outcomes. **This variable was included in the model to control for the effect of drinking and better identify the early predictors that are associated with smoking. However, because drinking was assessed at the same time as smoking (i.e., when participants were 14-15 years old), drinking cannot be treated as one of the ultimate predictors of smoking investigated in this study. Thus, this study does not further discuss the results associated with drinking as a predictor. Table 3.3.3 indicates that the model that best predicts occasional smoking includes three predictors. These predictors are: family functioning, number of close friends, and hostile and ineffective parenting. The summary statistics of the models tested in the stepwise backward logistic regression for daily smoking are presented in Table 3.3.4 68 Table 3.3.4 Model summary of the stepwise backward logistic regression analysis for daily smoking (n=i223) % Correct Classification -2 Log Likelihood Cox & Snell R Square Nagelkerke R Square Chi-Square* (df) Sig-Step 1 88.1 716.748 .152 .288 6.313(8) .612 Step 2 88.1 716.749 •152 .288 6.314(8) .612 Step 3 88.1 716.811 .152 .288 5-773(8) •673 Step 4 88.1 717.017 • 152 .287 5-077(8) •749 Step 5 88.1 717.346 •152 .287 4.567(8) .803 Step 6 88.1 718.872 .150 .285 3-458(8) .902 tHosmer and Lemeshow Test Step 1 = How often does he/she hang around kids you think are frequently in trouble? Intact Family Status, Has ever tried drinking, Gender, Older siblings of the child in the household, Younger siblings of the child in the household, Social Support, Socio-economic Status, Annual Household Income, Hyperactivity/Inattention, Has taken lessons in music, dance, art or other non sport activities, Would you say your child health is? Income Adequacy; Step 2 = How often does he/she hang around kids you think are frequently in trouble? Has ever tried drinking, Gender, Older siblings of the child in the household, Younger siblings of the child in the household, Social Support, Socio-economic Status, Annual Household Income, Hyperactivity/Inattention, Has taken lessons in music, dance, art or other non sport activities, Would you say your child health is? Income Adequacy; Step 3 = How often does he/she hang around kids you think are frequently in trouble? Has ever tried drinking, Gender, Younger siblings of the child in the household, Social Support, Socio-economic Status, Annual Household Income, Hyperactivity/Inattention, Has taken lessons in music, dance, art or other non sport activities, Would you say your child health is? Income Adequacy; Step 4 = How often does he/she hang around kids you think are frequently in trouble? Has ever tried drinking, Gender, Younger siblings of the child in the household, Social Support, Socio-economic Status, Annual Household Income, Hyperactivity/Inattention, Has taken lessons in music, dance, art or other non sport activities, Would you say your child "health is? Income Adequacy; Step 5 = How often does he/she hang around kids you think are frequently in trouble? Has ever tried drinking, Gender, Social Support, Socio-economic Status, Annual Household Income, Hyperactivity/Inattention, Has taken lessons in music, dance, art or other non sport activities, Would you say your child health is? Income Adequacy; Step 6 = How often does he/she hang around kids you think are frequently in trouble? Has ever tried drinking, Gender, Social Support, Socio-economic Status, Annual Household Income, Hyperactivity/Inattention, Would you say your child health is? Income Adequacy; The percent of correct classification of the model presented in Table 3.3.4 is estimated as 88.1%. However, the positive predictive value and negative predictive value of the model was estimated as 2.6% and997%, respectively. As for the discussion of the percent correct classification in the model for occasional smokers, the overall percent of correct classification 69 overestimates the predictive value of the model for daily smokers. With daily smokers being underrepresented in the sample (only 12% of the participants were daily smokers) the comparison conducted by the analytical procedure is between two groups that are not equally represented thus undermining the sensitivity of the model. Table 3.3.5 summarizes the results related to the model in step 6, that is, the model that the statistical procedure has identified as the one that best describes the patterns of association found in the data. Table 3 . 3 . 5 Summary of the final model obtained with logistic regression for daily smoking (11=1223) Predictor OR S.E. 95% CI P Gender* •544 .203 •338-.771 .003 Socio-economic Status •633 .200 .440-.969 .022 How often does he/she hangs around with kids you think are frequently in trouble? n .762 .124 .600-.982 .028 Social Support •947 .036 .885-1.021 •137 Annual Household Income .985 .010 .968-1.007 .114 Hyperactivity-Inattention 1.097 .027 1.037-1-155 .001 Would you say that your child health is? n 1-353 .116 1.082-1.709 .009 Income Adequacy 1-373 •235 .829-2.129 .177 Has ever tried drinkingm 13-968 •304 7.609-25.114 .000 The referent in gender was female. +tThis variable is reverse coded where higher scores mean less frequency or less positive outcomes. n t This variable was included in the model to control for the effect of drinking and better identify the ultimate predictors that are associated with smoking. However, because drinking was assessed at the same time as smoking (i.e., when participants were 14-15 years old), drinking cannot be treated as one of the ultimate predictors of smoking investigated in this study. Thus this study will not further discuss the results associated with drinking. Table 3.3.5 indicates that the model that best predicts daily smoking includes eight predictors. These predictors are: gender, engagement with 'troubled' friends, socio-economic 70 status, annual income, social support, hyperactivity/inattention, health status, and income adequacy. 3.3.2 Early Prediction of Infrequent Drinking The number of participants who reported to drink more often than once or twice a month was too small to explore the 'dose-response' hypothesis that was investigated with respect to smoking. Thus, multinomial logistic regression analyses as a strategy for data reduction could not be conducted. Instead, a total of 18 separate binomial logistic regressions were conducted, one for each of the sub-categories of predictors described in Appendix II. Whether participants had ever tried smoking or not was also included in each of the 18 regression models to control for the effect of smoking as described in the Analytical Plan. In some cases, smoking was significantly associated with drinking. However, as previously mentioned, the results associated with this predictor could be misleading and will not be further discussed. About 55% of the participants reported drinking either once or twice a month or once or twice a year. On the other hand only 1.2% reported drinking weekly. According to a 1995 national survey by Health Canada, 72% of Canadians aged 15 and over said they drank in the past year (Health Canada, 1995). This prevalence of drinking was about 5% less than that observed in previous years indicating that adolescent drinking may be gradually decreasing (Health Canada, 1995). The finding that 55% of study participants reported drinking in 2000-01 is consistent with the hypothesis that adolescent drinking is gradually decreasing. Table 3.3.6 summarizes the results of the 18 logistic regression models that were conducted with infrequent drinking as the outcome. 71 Table 3.3.6 Summary of 18 binomial logistic regressions that were conducted separately for each sub-category of predictor for infrequent drinking Predictor n OR S.E. 95% CI P Has child ever had asthma?+ 1378 1-585 .177 1.121-2.241 .009 Physical Aggression 1313 .906 •045 .830-.989 .027 Province of Residence*" Quebec Saskatchewan Manitoba Alberta British Columbia 1397 3.206 2.247 2.235 2.183 2.181 •837 .672 .674 .670 .653 1.922-5.350 1.250-4.038 1.238-4.037 1.196-3.985 1.212-3.925 . O O O .007 .008 . O i l .009 Mathematics Computation Test 699 1.088 .030 1.026-1.154 .005 In the last 12 months, outside of school hours, how often has he/she taken part in any club, groups, or community programs with leadership, such as Brownies, Cubs or Church group?f t 1345 1.207 •057 1.080-1.348 .001 The referent for this variable was o=yes. f This variable is reverse coded where higher scores mean less frequency or less positive outcomes. "The referent for this variable is New Foundland Table 3.3.6 indicates that there are five predictors that are significantly associated with infrequent drinking: asthma, physical aggression, province of residence, involvement with extra-curricular community programs, and the Mathematics Computation Test. It is worth pointing out that the response rate obtained for the Mathematics Computation Test was much lower than expected. Mathematics tests were completed by only 50.5% of the eligible children. Because of such a high non-response rate it is possible that the results obtained with respect to this variable may be biased. Statistics Canada (Special Survey Division, 1998) conducted a study to assess the impact that the low response rate had on the results. The distributions of some of the characteristics observed in household interviews were compared for cases where there was a response to the mathematics test versus non-response. Results of this study indicated that there were differences between the distributions of respondents and non-respondents with respect to the following variables: household income, failed a grade, how well child is doing in reading, how well the child is doing in Math, how well the child is doing in composition, how well the child is doing in general, received help outside the school, contacted 72 by school regarding behaviour, child looks forward to going to school, important that child has good grades, how far it is hoped that the child will go in school. Because of the bias associated with responses to the Mathematics Computation Test, it was decided to eliminate this predictor despite the fact that we obtained significant results in the preliminary analysis. Next we conducted one logistic regression where all the predictors reported in Table 3.3.8 (except Mathematics Computation Test) were entered in the model, and a stepwise backward procedure was conducted to identify the model with the ideal number of predictors. Table 3.3.7 presents the model summary statistics of the logistic regression for infrequent drinking. Table 3 . 3 . 7 Model summary of the stepwise backward logistic regression analysis for less than monthly drinking (11=1380) % Correct -2 Log Cox & Snell Nagelkerke Chi-Square+ Sig. Classification Likelihood R Square R Square (df) Step 1* 65.7 1616.997 •175 • 233 5-436(8) .710 *The model converged at the first step of the procedure. fHosmer and Lemeshow Test Step 1 = Has ever tried smoking, Physical Aggression; Province of Residence; In the last 12 months, outside of school hours, how often has he/she taken part in any club, groups, or community programs with leadership, such as Brownies, Cubs or Church group? Has child has ever had asthma? The percent of correct classification of the model presented in Table 3.3.7 is estimated as 65.7%. However, the positive predictive value and negative predictive value of the model was estimated as 27.8% and79.2%, respectively. While infrequent drinkers are equally represented in the sample as non-drinkers (about 55% of the participants were infrequent drinkers) the model does more adequately predict non-drinkers than infrequent drinkers. Table 3.3.8 summarizes the results associated with the model tested in step 1. 73 Table 3 . 3 . 8 Summary of the final model obtained with logistic regression for infrequent drinking ( n = i 3 8 o ) Predictor OR S.E. 95% CI P Physical Aggression .921 •035 .860-.986 .019 Province of Residence Nova Scotia New Brunswick •565 •471 .302 •398 .313-1.020 .216-1.028 .058 •059 Has child has ever had asthma? f 1.518 •171 1.086-2.122 .015 Province of Residence+tt Ontario 1-515 •255 .919-2.497 .103 In the last 12 months, outside school hours, how often has he/she taken part in any club, groups or community programs with leadership, such as Brownies, Cubs, or church groups? n 1-137 .058 1.016-1.273 .026 The referent for this variable was o=yes t1This variable is reverse coded where higher scores mean less frequency or less positive outcomes. t t +The referent for this variable is New Foundland Table 3.3.8 indicates that there are five predictors associated with infrequent drinking. These predictors are: physical aggression, province of residence, drinkers. In addition, participants who had not had asthma, had infrequently taken part in extra-curricular community programs, and who lived in the province of Ontario, were more likely to become infrequent drinkers. 3.4 Summary of Results The main purpose of the present study was to identify the early predictors of occasional smoking, daily smoking and infrequent drinking. In the Introduction it was argued that while adolescent smoking and drinking may share common predictors, these behaviours may also be associated with unique predictors. In addition, it was hypothesized that while smoking and drinking tend to co-occur the causal pathways that describe the etiology of smoking may be different from those that describe the etiology of smoking. In order to explore this hypothesis, separate analyses were conducted to investigate the associations between the predictors listed in Appendix II and three separate outcomes: occasional smoking, daily smoking and infrequent 74 drinking. Table 3.4.1 summarizes the results of these analysis and shows the predictors that were found to be significantly associated with occasional smoking, daily smoking, and drinking. Table 3 . 4 . 1 Summary of the results obtained for occasional, daily smoking, and infrequent drinking Occasional Daily Smoking Infrequent Drinking Number of Close Friends Gender Physical Aggression Family Functioning Socio-economic Status Asthma Hostile/Ineffective Parenting Annual Household Income Participation in community Social Support programs with leadership Hyperactivity/Inattention Province of Residence Child Health Income Adequacy The results summarized in Table 3.4.1 indicate that occasional smoking, daily smoking and infrequent drinking cannot be predicted by the same combination of factors. In fact, it was found that occasional smoking, daily smoking and infrequent drinking did not have any predictors in common. These results lead to the following conclusions: 1) occasional smoking and daily smoking cannot be explained as two incremental steps on the same behavioural continuum. In other words, the frequency with which participants smoke cigarettes is associated with different etiologies and underlying causal mechanisms; 2) infrequent drinking cannot be explained with the same causal pathways as those used to explain occasional or daily smoking. For example, it was found that while the early predictors of occasional smoking are concerned with interpersonal factors (number of close friends, hostile/ineffective parenting and family functioning), the early predictors of daily smoking are primarily associated with individual level factors (gender, hyperactivity/inattention, child health) and socio-demographic characteristics of the family (income adequacy, socio-economic status, annual household income). On the other hand, the early predictors of infrequent drinking are associated with individual level factors (physical aggression, asthma), extra-curricular activities (participation in community programs with leadership) and child demographics (province of residence). 75 3.5 Discussion The present study was motivated by the hypothesis that different etiological pathways are associated with different ways in which adolescents engage in smoking and drinking. It was argued that the causal roots of smoking may be different from the casual roots of drinking. More importantly, it was argued that occasional smoking and daily smoking may have different etiologies, that is, there are unique causes associated with occasional smoking that are different from the causes associated with daily smoking. To investigate these hypotheses, a sub-sample from the NLSCY was selected for which both predictors and outcomes were available. The predictors were assessed in 1994-95 when the participants were 8 and 9 years old while smoking and drinking were assessed six years later when the participants were 14 and 15 years old. Because of the temporal distance between the predictors and the outcomes and since the predictors were assessed before smoking and drinking became manifest, the investigation is about the search for the predisposing factors of smoking and drinking. Predisposing factors (as opposed to distal and proximal predictors) are those closely associated with the actual roots of the behaviour, and thus, are the most informative type of predictors on the causal pathways (etiology) of the behaviours under investigation (Petraitis, Flay, & Miller, 1995). The results of the present study indicate that the etiologies of smoking and drinking are not as closely related as has been hypothesised by some theorists (e.g., Jessor & Jessor, 1980). For example, the findings suggest that, while smoking and drinking tend to co-occur, the underlying causal pathways that are associated with adolescent smoking are not the same as those associated with adolescent drinking. More importantly, the findings indicate that not only are there differences in the etiology of different substance use (i.e., smoking and drinking), but that there can be important differences in the causal pathways associated with the frequency with which adolescents engage with specific substances (i.e., occasional smoking and daily smoking). This is an important result in that it forces the reconsideration of the current approach to the study of substance use, wherein the tendency is to develop general theories. That is, favoured theories are those that can be applied to a number of different behaviours (e.g., 76 problem behaviour theory) or substance specific theories where the frequency with which adolescents engage in the behaviour is seen as a series of incremental steps on a continuum of substance use (e.g., stages of smoking acquisition). The findings of the present study cannot be adequately explained by any of the theories of substance use presented in the Introduction because they are either too general or do not adequately differentiate between frequent and infrequent substance use. Thus, an alternative interpretation of these results is proposed and, when appropriate, similarities and differences with the theoretical models discussed in the Introduction are indicated. A model of 'smoking as coping' is proposed to explain the mechanisms involved in the etiology of smoking behaviour in early adolescence. In the Introduction, it was argued that frequency of use could be employed as an operational criterion to differentiate between normative behaviour and problem behaviour. For example, occasional smoking was used as an example of normative behaviour and daily smoking as an example of problem behaviour. However, the results obtained in the present study forced the reconsideration of the validity of such an argument. Only 11% of the participants reported to be occasional smokers indicating that occasional smoking among 14 and 15 years olds is not a highly prevalent behaviour. If occasional smoking were part of normal adolescent development, one would expect to see a much higher prevalence rate than that observed in the present study. In addition, the patterns of association that emerged in the data suggest that the causes of occasional smoking are attributable to interpersonal problems that the child may be experiencing with his/her parents. The result that occasional smokers tended to have issues with their parents in their childhood is consistent with the social control theory (Elliott, Huizinga, & Menard, 1989; Elliott, Huizinga, & Ageton, 1985), and the social development model (Hawkins & Weis, 1985), both of which argue that adolescents who feel distanced from their parents may perceive 'strain' that in turn is the cause of adolescent substance use. However, results suggest that the attachment to substance using peers is not necessarily what causes adolescents to smoke. 77 In fact, it was found that children who lived in families with more interpersonal problems but at the same time had more close friends were less likely to become occasional smokers. This pattern of association describes a child whose primary developmental challenges may be associated with how his/her parents interact with him/her, and whose family may be experiencing interpersonal problems. The fact that having several friends reduces the risk of these children becoming occasional smokers speaks to the function of the peer group in facing the developmental challenges associated with the relationship with parents. For example, having several close friends may effectively compensate for the lack of a warm and close relationship with parents. Adolescents who do not have several close friends could perceive smoking as a means through which they can make more friends, consequently compensating more effectively for the lack of a positive relationship with their parents. In this instance, the interest that adolescents have in smoking is not in smoking as a 'substance' but rather in smoking as a 'context'. In other words, adolescents may consciously use smoking to 'cope' with some developmental challenges that they experience at the interpersonal level. As previously discussed, there is growing evidence that adolescents are conscious of the positive consequences associated with smoking in that it enhances the sense of belonging and acceptance in the peer group (Delorme et al., 2003; Pavis, Cunningham-Burley, & Amos, 1997; Pavis, Masters & Cunningham-Burley, 1996). For example, it was found that number of close friends was inversely associated with the likelihood of becoming smokers (i.e., individuals who have more close friends are less likely to become occasional smokers than individuals with fewer number of friends). These findings go against most theories of substance use behaviours (e.g., conventional commitment and social attachment theories, social development model, problem behaviour theory) where friends are seen as negative influences on the adolescent behaviours promoting onset or escalation of substance use. On the other hand, this result, provides support for the view that children are more likely to 'choose' as friends other children with whom they share similarities rather than being passively influenced by substance using peers (Botvin et al., 1992; Conrad, Flay, & Hill , 1992; Fergusson & Horwood, 1995; Kandel, 1980; Newcomb & Bentler, 1989; Stanton, Silva & 78 Oei, 1991). Such a finding has important implications at the prevention level where much emphasis is placed on teaching children refusal skills to resist peer influences. In fact, it suggests that a shift in perspective is needed if one's goal is to prevent occasional smoking in adolescence. More specifically, it may be recommended that prevention programs be delivered earlier than when they are typically implemented and that they focus on increasing the opportunities for children to create as many close friendships as possible. This way one may be able to prevent occasional smoking, and at the same time promote the adoption of adequate compensatory strategies that those children experiencing difficulties with their parents need in order to cope with their problems. If the concept of 'smoking as coping' is correct, then one might expect that young occasional smokers are less likely to show signs of physical dependence, but more likely to show signs of psychological dependence. Johnson and colleagues (2002) demonstrated that there were five distinct aspects of tobacco dependence including social, pleasurable, empowering, emotional, and full-fledged or physical dependence. They also found that several participants reported high levels of social and sensory reinforcement associated with cigarette smoking in the absence of physical dependence. Similarly, Shiffman et al. (1995) identified a proportion of the smoking population who smoked but did not appear to physically dependent on nicotine. Shiffman and colleagues (1995) labelled these smokers as "chippers" (i.e., occasional or non-regular smokers): they smoked less than 5 cigarettes a day but not on a regular basis alternating periods of daily smoking and abstinence. Furthermore, in a recent study, Johnson and colleagues (in press) found that while some participants who reported high levels of physical dependence also reported high levels of psycho-social dependence (i.e., social and sensory reinforcement), others reported high levels of physical dependence but very low levels of psycho-social dependence. The recent developments of research on different types of tobacco dependence seem to be consistent with the finding of the present study that the causal pathway associated with occasional smoking is not the same as that associated with daily smoking. However, further 79 research is required to explicitly link the results of the present study with the research on the development of tobacco dependence. While the results of the present study lead to reconsidering the concept of occasional smoking as a normative behaviour, they reinforce the assumption that daily smoking at age 14 and 15 is in fact adequately described as a problem behaviour. First, daily smoking was as uncommon as occasional smoking with about 12% of the participants being daily smokers. Second, daily smoking was more common among females, and among participants who had either cognitive7 and/or health problems (i.e., hyperactivity and inattention, and overall physical health). In addition, participants with more disposable income (income adequacy can be considered an indirect measure of disposable income) were also more likely to become daily smokers, probably because they could afford to buy cigarettes. On the other hand, participants from higher socio-economic status, higher annual household income, and/or greater social support were less likely to have become daily smokers. Children from higher socio-economic families, not only benefit from having more educated parents, but also tend to live in better housing conditions and in better neighbourhoods. However, none of the neighbourhood predictors were found to be significantly associated with daily smoking (or even occasional smoking and infrequent drinking), suggesting that neighbourhood characteristics maybe less influential in determining the onset of smoking (and infrequent drinking) and that they may play more important roles in the maintenance and/or progression of such behaviour later during adolescence. At the same time, children whose families have more social support around them are likely to 'cope' better with the challenges associated with poor health and cognitive function, and are less likely to become daily smokers. Given these premises, children who experience health and cognitive challenges, but at the same time, do not benefit from a higher socio-economic status and adequate social support, need to adopt a coping strategy that allows them to face their specific developmental challenges. Findings from this study suggest that smoking may be one of these strategies. 7 Hyperactivity and inattention are a type of cognitive function associated with working memory and are generally called 'executive functions'. For simplicity, here we will refer to hyperactivity and inattention as 'cognitive function'. 80 It is worth pointing out that none of the predictors of daily smoking is related to interpersonal or inter-relational factors indicating that the origins of daily smoking may be better found within intra-individual level factors such as cognitive/executive functions and physical health. One hypothesis is that children who experience cognitive and health problems may perceive smoking as a way to 'self-medicate' their problems, and are thus, more interested in smoking as a 'substance' rather than smoking as a 'context'. Children may be 'aware' of the self-medicating properties of cigarette smoking because they have observed that their parents use cigarette smoking while growing up. This hypothesis is supported by the preliminary analysis reported in this study where it was found that the number of cigarettes mothers smoked per day was significantly associated with an increase in the risk for becoming daily smokers. Maternal smoking was not purposefully included in the final analysis because the response rate associated with this variable was too low, and therefore, would lead to biased associations. However, the hypothesis that children (particularly girls) exposed to maternal smoking during childhood may assimilate the concept of smoking as a self-mediation strategy is worth investigating further. The idea of smoking as self-medication is not new in the literature. Tobacco use has been found to be associated with several mental health conditions such as depression and schizophrenia among adults (Dalack et al., 1998; Tanskanen et al., 1999) and several investigators have speculated that smokers use nicotine because of its properties of mood-regulation. For example, Coger, Moe, and Serafetinides (1996), Levin et al. (1996), and Pomerleau et al. (1995) have suggested that adult individuals with attention-deficit/hyperactivity disorder (ADHD) employ nicotine to enhance cognitive function, and research indicates that nicotine enhances performance in attentional tasks among adults with ADHD (Conners et al., 1996). Furthermore, recent research findings indicate that females may take significant advantage of nicotine in vigilance tasks indicating a sex-dependent nicotine effect on attention (Trimmel & Wittberger, 2004). To summarize, these results are very consistent with the self-medication theory. However, the stunning aspect of these results is that an association was found between deficits in cognitive function and daily smoking in children as young as 8 and 9 years old. Despite the fact 81 that measures of prevalence of daily smoking at age 8-9 were not available for the study participants, the investigation of developmental trajectories of smoking behaviours presented in Chapter I indicate that virtually none of the participants was a daily smoker at 10-11 years of age. Thus, it can be inferred with a certain degree of confidence that the deficit in cognitive function and the poor health status were present before the onset of daily smoking. One of the most important conditions for establishing a causal relationship is the temporal sequencing of the events. In the case of these results, it can be assumed that the children's cognitive function and health conditions may be important contributors to the causal pathways leading to daily smoking. The fact that in preliminary analysis it was found that maternal smoking was also significantly associated with an increased risk for becoming daily smokers may indicate that children, through observational learning and modelling, have come to know that smoking can function as self-medication for their problems. This result is consistent with the social learning theory (Akers, 1977; Bandura, 1977,1986; Sutherland, 1939) that states that paternal attitudes, beliefs, and behaviours related to smoking play a critical role in forming the adolescent's attitudes, beliefs and behaviours about smoking. One other novel finding of the present study is that overall health status was found to be significandy associated with daily smoking; where the poorer the health status, the higher the likelihood of becoming daily smokers. Generally, health status has been investigated as a consequence of smoking, but here it is suggested that health status may in fact be one important causal factor. These results need to be replicated because of the important implications they might have on the epidemiological research of the health consequences of smoking. If health status is one of the causes of smoking as well as a consequence, appropriate methodologies need to be employed to isolate the effects from the causes of smoking. In addition, there is need to conduct more investigations in order to determine what specific health problems are predictive of daily smoking. For example, are the specific health problems that affect the child's general health status related to mental health, or to other physical conditions or physical disabilities? Is there an association between hyperactivity and inattention and other specific health conditions? 82 The results indicate that the 'smoking as coping' model may also be appropriate to describe the causes of daily smoking in that adolescents may take up smoking to face some of their developmental challenges that are specifically related to cognitive and health problems. However, the underlying causes of daily smoking seem to be associated with critical developmental issues that, if neglected, could lead to several problems throughout childhood and adulthood. For example, adults who in their childhood suffered from ADHD are also more likely to suffer from antisocial, depressive, and anxiety disorders, school failure, occupational problems and traffic accidents (Faraone et al., 2000). In addition, while smoking adolescents may improve their cognitive/executive functions at the same time they are likely to worsen their overall already precarious health. For these reasons, daily smoking can also be adequately described as 'problem behaviour'. In summary, a 'smoking as coping' model is proposed that hypothesizes two distinct causal pathways, one leading to occasional smoking and one leading to daily smoking. The causal pathway leading to daily smoking is that of 'self-medication' where children who experience cognitive and health problems during childhood are more likely to take up smoking because of the medicating properties of nicotine. On the other hand the 'social compensation' pathway is that which leads to occasional smoking and describes children who experience difficulties with their parents during their childhood and 'choose' to smoke because of the benefits associated with social context in which smoking occurs. Figure 3.1 summarizes the 'smoking as coping' model and the two casual pathways proposed here. Self-medication causal pathway Cognitive/executive functions and health problems • Takes up smoking for the medical properties of nicotine • Becomes daily smoker Social-compensation causal pathway Family related interpersonal issues • Takes up smoking for the benefits associated with the social context of smoking • Becomes occasional smoker Figure 3 . 1 Smoking as coping model 83 The results of the present study suggest that there may be important moderating or mediating factors that are involved in the causal pathways hypothesized by the 'smoking as coping' model. For example, it was found that number of close friends functioned as a protective factors against occasional smoking and that socio-economic status and social support functioned as protective factors against daily smoking. By proposing the 'smoking as coping' model, a hypothesis of the mechanisms by which risk and protective factors co-act in determining the risk of becoming occasional or daily smoker is proposed. However, these hypotheses, being based on investigations of predisposing factors, are primarily concerned with the actual roots of the behaviour, that is, with the causes responsible for the onset of smoking that are present before smoking has become manifest. As pointed out in the Introduction, understanding the predisposing factors of the behaviour is like having a clear understanding of the beginning of a story that develops into a complex interplay of events. Distal and proximal predictors then become critical in enriching one's understanding of how behaviours become stable life styles versus how they happen to disappear shortly after their appearance. Future research needs to be conducted to link the current knowledge about distal and proximal predictors of smoking to the causal pathways hypothesized in the smoking as coping model. In the Introduction, it is argued that occasional smoking and infrequent drinking could be conceptualized as two types of normative risk taking behaviour, while daily smoking would be more properly conceptualized as problem behaviour. Unlike occasional smoking, the findings of the present study do reinforce the belief that infrequent drinking is in fact a normative behaviour among young adolescents. It was found that about 55% of the participants reported drinking infrequently (i.e., less than weekly) indicating that this behaviour is in fact very common among 14-15 years old. Furthermore, of all the predictors tested, only one risk factor was identified: participants who lived in Ontario were more likely to become infrequent drinkers compared to children living in New Foundland. The reason why children living in Ontario were more likely to become infrequent drinkers is not easy to explain. Similarly, the reason why children living in New Brunswick and Nova Scotia were less likely to become infrequent drinkers is also difficult to 84 interpret. The reasons why there might be these inter-provincial differences are not obvious and are worth further investigation. In addition to provincial differences, it was found that children who have had asthma and who scored higher in physical aggression were less likely to become infrequent drinkers. These results are quite intriguing in that those two 'protective' factors are indicative of the fact that the child is experiencing some problem, either at the level of physical health (i.e., the child has asthma) or at the level of behaviour and mental health (i.e., the child is physically aggressive). In other words, the child is facing some developmental challenges that 'impede' him/her from experimenting with infrequent drinking. For example, asthma can greatly impact the range of activities that a chid can engage in and may have repercussions on the ability of the child to have an adequate social life in childhood and adolescence potentially interfering with education, physical activity, socialization, and self-esteem (Stone, 2002). In addition, children who are aggressive may be less likely to develop those pro-social skills that are necessary to 'fit in' the peer group with whom infrequent drinking is likely to occur. In fact, children who are physically aggressive tend to affiliate more with friends who are also aggressive (Deptula & Cohen, 2004), thus showing the tendency to affiliate with 'deviant' peers rather than 'normal' peers. In other words, children who engage in infrequent drinking tend to be healthy children who do not face critical developmental challenges, while those who do not engage in infrequent drinking may be experiencing health or behavioural problems. For these reasons, it is believed that infrequent drinking is adequately described as a normative behaviour. In summary, in the present study, early predictors of occasional smoking, daily smoking, and infrequent drinking were identified. It was demonstrated that occasional smoking, daily smoking, and infrequent drinking could be predicted as early as age 8 and 9 and by a number of different predictors. It was proposed the 'smoking as coping' model where two different causal pathways are hypothesised, one for occasional smoking and one for daily smoking. Finally, it is argued that given its high prevalence among the study participants and because no risk factors were identified in association with this behaviour, infrequent drinking may be adequately described as normative risk taking behaviour. 85 C h a p t e r I V : C o n c l u s i o n 4.1 Conclusion and recommendation for future work The present study has generated many new and interesting findings about the development and the predisposing factors of smoking and drinking among Canadian children and adolescents. First, it is rather surprising that as a combined result of the variable reduction procedure and statistical modelling only a very limited number of predictors yielded a significant association with the three outcomes investigated in this research (i.e., occasional smoking, daily smoking, and infrequent drinking). As illustrated throughout the method sections, this cannot be simply due to sampling errors or measurement issues associated with the survey design and the sampling strategy. Compared with other national studies, the NLSCY provided rather consistent and reliable prevalence estimates and meets the validity (internal, at least) criteria as well as the tolerable attrition rate which characterizes current longitudinal surveys. Thus, in so far as it meets the general standards, one can infer that NLSCY gives a reliable picture of the development and the etiology of smoking and drinking behaviour in Canadian adolescents. Second, the results disagree with most theories of adolescent substance use which place a main explanatory causal role either in the peers or in the social influence processes outside the family. Presumably, the critical element of difference here is that these other perspectives, despite claiming to be concerned with the actual roots of the behaviour, were rarely investigated among young children and therefore, before the manifestation of the outcome. That is, peer influences were investigated at times when the behaviours had already started or shortly before the onset of the behaviour. As previously discussed, the role of distal and proximal factors in the maintenance of smoking and drinking behaviours may be better explained if the casual roots of these behaviours can first be isolated and understood. The present study indicates that peer influence, as traditionally defined as 'peer pressure' (either direct or indirect) may not be causally 86 associated with the onset of smoking and drinking, but rather, it may play important roles in the maintenance of such behaviours. In fact, the only significant association with peer related factors was found for occasional smoking where having several close friends was inversely related with the likelihood of becoming occasional smoker, thus suggesting the positive influence of peers on adolescent behaviours rather than negative influence or peer pressure. Another new and important result of the present study is that while the predisposing factors of occasional smoking could be attributed to family interpersonal factors such as parenting and family functioning, the predisposing factors of daily smoking could be found within individual level factors such as hyperactivity and inattention and poor health. This led to developing the 'smoking as coping' model where two causal pathways are hypothesized: the self-medication pathway and the social-compensation pathway. To this respect the greatest element of novelty of the findings is that the self-medication model, a model that has been broadly investigated among adolescents and adults, could be so easily applied to interpret the patterns of results found in our 8 and 9 year old participants. The data is the first to suggest that the ultimate causes of daily smoking may be so closely related to the 'intention' to self-medicate problems associated with cognitive/executive functions and other health related problems. In addition, the data is the first to show that occasional smoking and daily smoking cannot be predicted with the same pattern of association and that these behaviours may be caused by quite distinct factors. There are several implications of these findings. First, results of the present study re-direct the attention to the importance of understanding childhood as the 'building blocks' for adolescent health behaviours. One concept that is implicit in most of the theories of substance use behaviour is that adolescence substance use is the primary predictor for later adult behaviour. As such, adolescence substance use is conceptualized as the 'precursor' of adult behaviour and health outcomes and research has traditionally focused on investigating adolescence as detached from its previous developmental phases of infancy and childhood. Breaking with this limited view of adolescence as a transitory phase detached from previous developmental phases, the findings indicate that the intra-personal and inter-personal situations 87 of the individual during childhood may be the primary factors responsible for the adolescents' decisions associated with the health behaviours acted in adolescence. Second, the finding that each of the three outcomes investigated could be predicted by very different patterns of association, lends support to the call to develop more specific theories of adolescent substance use that take into account both the particular substance that is being used and the frequency with which the substance is used at any given time. Third, prevention and intervention planners, as well as policy decision makers, could potentially benefit greatly from the results of the present study. The critical element that could change the way prevention of smoking and drinking is currently planned has to do with the fact that younger children should be targeted and that more general developmental issues can be addressed rather than through health education where awareness about health and social consequences of substance use is promoted. For example, to prevent daily smoking, school based approaches may be implemented for children who have problems at the level of cognitive/executive functions who can be identified and provided with support and special education. One advantage of this approach is that one would get 'more for your buck' in that intervention would have an impact in reducing the risk for children to become daily smokers but at the same time, children would also improve in many more developmental outcomes including school achievement that ultimately influences their chances to do well in life throughout adulthood. 8 8 C h a p t e r V : L i m i t a t i o n s 5.1 Limitations The three studies presented in this thesis have several limitations are worth discussing. Some of these limitations are inherent to the NLSCY while others are associated with the analytical approaches used to conduct analysis of the data. These limitations are discussed below under separate headings. 5.1.1 Limitations of the NLSCY One of the limitations of the NLSCY is associated with the sampling procedure that was used to identify the initial sample of participants in 1994-95. Participants from another national survey, the Labour Force Survey, were approached in order to identify households containing at least one child in the o to 11 age range. The Labour Force Survey is a national survey designed to collect information from a representative sample of Canadians on basic demographics and labour market information; consequently, information about household members was readily available. This procedure, while economic, may be associated with the risk of selection bias in that more 'compliant' households may have been selected for participation in the NLSCY. It is possible that people who adhere to national surveys have different characteristics from the people who refuse to participate in such surveys. While the large NLSCY sample may have increased the chances of reaching a representative portion of the Canadian population, it is true that the NLSCY recruited participants among a pre-selected sub-group of the population and that the biases associated with the sampling procedure used in the Labour Force Survey may in fact be perpetuated in the NLSCY. A second limitation of the NLSCY is associated with the large loss to follow-up that occurred throughout the study, particularly between Cycle 1 and Cycle 2. There may be several reasons for this drastic reduction in the sample size (about 40% of the participants were lost to follow up between 1994-95 and 2000-01). For example, the survey may have required a greater 89 time commitment than families had initially expected. Telephone interviews may have also been less effective that face to face interviews in collecting complete information from the participants. In fact, it is reasonable to assume that it may be more difficult for participants, particularly those with young children at home, to find the time and the right moment to stay on the phone for a long time. On the other hand, while more expensive, a home visit from the interviewer may have resulted in more complete assessments. However, some of the greatest problems with missing data in the NLSCY are associated with the adolescent questionnaires. The interviewers hand the questionnaires to the adolescent participants at their home and instruct them as to how the questionnaire needs to be filled in. This procedure has serious limitations when used with adolescent participants. For example, for the most part, home may not be a safe place for adolescents to respond to delicate questions such as those associated with their involvement with sexual activity, smoking, and drugs and alcohol use. It is not surprising that several of the self-administered questionnaires are filled in incompletely by the NLSCY participants. In addition, once completed, each participant placed the completed questionnaires in an envelope and returned it to the interviewer who did not open it until back at the office. If the interviewer had the ability to go over the questions with the participant, a greater response rate may have been achieved. Alternatively, a web-based questionnaire or school questionnaires may have produced a greater response rate than the home based self-administered questionnaires. Because of the large loss to follow-up we conducted an attrition analysis that revealed that for the most part, non-respondents were similar to respondents. However, the attrition analysis also identified some differences between non-respondents and respondents. Particularly relevant to the present study is the fact that non-respondents were more likely to have mothers who reported frequent and heavy drinking. Given the suggesting evidence of the link between parental alcohol problems and adolescent heavy drinking (U.S.D.E., 1993J it is likely that some of the non-respondents may be adolescents who also drink frequently and/or heavily. This would explain why in our study we could not identify a sub-group of adolescents who drank more frequently than once a month and therefore we could not investigate the early patterns of prediction for adolescent heavy drinking. 90 5.1.2 Limitations associated with analytical strategies This study has some limitations that are associated with the type of analytical strategy used to conduct the analyses. First, growth mixture modelling is a method of approximations of unknown distributions, and as such is aptly suited for generating hypothesis of population heterogeneity than describing, in precise terms, how the population can be divided into discrete sub-groups or categories. Aware of this limitation, growth mixture modelling was used in Chapter I and Chapter II to hypothesise different developmental trajectories of smoking and drinking, and it was argued that these hypotheses need to be tested in future research. In addition, the results obtained with growth mixture modelling were used to develop the rational for focusing on the smoking and drinking behaviours of 14-15 years old participants and to investigate the association between these behaviours and several early predictors, as presented in Chapter III. The analytical procedure for the investigation of the early predictors of smoking and drinking also has some limitations that are worth discussing. As for any regression techniques, logistic regression measures the degree of association between predictors and outcome. That is, logistic regression does not have the ability to establish causality. While in the Discussion some hypothesis of causality are proposed within the 'smoking as coping' model, care was taken to present the results of the regression analysis in terms of associations. Finally, it is arguable that the procedure used to initially conduct data reduction also has some limitations. For example, while theoretically relevant predictors were selected for inclusion in the analysis, these were still too many and needed to be reduced. Instead of further reducing the number of predictors according to specific theoretical models, it was preferred to impose an empirical process of elimination that consisted in conducting several separate logistic regression analyses, one for each of the categories of predictors identified in Chapter III. This procedure has the draw back that does not allow testing for interaction effects between predictors from different categories, and it could increase the chances for significant results to be obtained by chance (i.e., Type I error). However, as shown throughout Chapter III, significant results were obtained for 91 the most part with a 'p' value smaller than .05, and consequently, it is unlikely that greater than usual Type I error has been incurred. Finally, the analysis conducted here only tested for main effects and did not include interaction terms. This was a conscious decision made to facilitate interpretation of results, especially because of the large number of predictors that were initially investigated. It is true that by simplifying the interpretation of the results the richness of information that would have derived through the investigation of interactions has been lost. This limitation is inherent to the exploratory nature of this work that aimed to generate hypotheses based on well documented empirical regularities. It is responsibility of future research to investigate the validity of these hypotheses by further investigating the relationship that exists between one predictor and the other, thus testing models of interaction/moderation as well model of mediation effects. 9 2 R e f e r e n c e s Addams, J . (1910). The spirit of youth and the city streets. New York: Macmillan Adlaf, E., Ivis, F., & Smart, R. (1997). Ontario student drug use survey. Toronto: Addiction Research Foundation. Adlaf, E., Paglia, A., Ivis, F. J., & Ialomiteanu, A. (2000). Nonmedical drug use among adolescent students: highlights from the 1999 Ontario Student Drug Use Survey. Canadian Medical Association Journal, 162(12), 1677-1680. Ajzen, I. (1985). From intention to action: A theory of planned behavior. In J . Kuhland & J. Beckman (Eds.), Action-control: From cognition to behavior (pp. 11-39). Heidelberg: Springer. Ajzen, I. (1988). Attitudes, personality, and behavior. Homewood, IL: Dorsey Press Ajzen, I., & Fishbein, M . (1980). Understanding the attitudes and predicting social behavior. Englewood Cliffs, NJ: Prentice-Hall Inc. Akers, R L. (1977). Deviant behavior: A social learning approach. Belmount, CA: Wadsworth. Anthony, J. C , & Echeagaray-Wagner, F. (2000). Epidemiologic analysis of alcohol and tobacco use: Patterns of co-occurring consumption and dependence in the United States. Alcohol Research & Health, 24(4), 201-208. Bandura, A. (1977). Self-reinforcement: The power of positive personal control. In P. G. Zimbardo & F. L. Ruch (Eds.), Psychology and Life (9th ed.). Glenview, IL: Scott, Foresman. Bandura, A. (1986). Social foundations of thought and action: A social cognitive theory. Englewood Cliffs, NJ: Prentice Hall Bateson, M . C. (1991). Reflections on risk and calibration. In L. P. Lipsitt & L. L. Mitnick (Eds)., Self-regulatory behaviors and risk taking: Causes and consequences. Norwood, NJ: Ablex Publishing Corporation. Baumrind. D. (1987). A developmental perspective on adolescent risk-taking in contemporary America. In C. E. Irwin (Ed.), Adolescent social beahvior and health. New directions for child development (pp. 93-125). Social and Behavioral Science Series, No. 37, Fall. San Francisco: Jossey Bass. Beck, K. H., &T. G. Summons. (1987). The social context of drinking among high school drinking drivers. American Journal of Drug and Alcohol Abuse, 13(1-2), 181-198. Bien, T. H., & R. Burge. (1990). Smoking and drinking: A review of the literature. International Journal of Addiction, 25(12), 1429-1454. Block, J., Block, J. H., & Keyes, S. (1988). Longitudinally foretelling drug usage in adolescence: Early childhood personality and environmental precursors. Child Development, 59(2), 336-355-93 Bobo, J . K. & Husten, C. (2000). Sociocultural influences on smoking and drinking. Alcohol Research & Health, 24, 225-32. Botvin, G. J., Dusenbury, L., Baker, E., James-Ortiz, S., Botvin, E. M. , & Kerner, J . (1992). Smoking prevention among urban minority youth: Assessing effects on outcome and mediating variables. Health Psychology, 11(5), 290-299. Brook, J . S., Whiteman, M . , Gordon, A. S., & Brook, D. W. (1990). The psychosocial etiology of adolescent drug use: A family interactional approach. Genetic, Social, and General Psychology Monographs, 116(2), 113-267. Brown, B. B. (1990). Peer groups and peer cultures. In S. S. Feldman, & G. R. Elliott (Eds.), At the threshold: The developing adolescent. Cambridge, MA: Harvard University Press. Brown, B. B., Dolcini, M . M . , & Leventhal, A. (1997). Transformations in peer relationships at adolescence: Implications for health-related behaviour. In J. Schulenberg, J . Maggs, & K. Hurrelmann (Eds.), Health risk and developmental transitions during adolescence. Cambridge, England: Cambridge University Press. Buhrmester, D. (1996). Need fulfillment, interpersonal competence and developmental contexts of early adolescent friendship. In W. Bukoski, A. Newcomb, & W. Hartup (Eds.), The company they keep: Friendship in childhood and adolescence (pp. 158-183). New York: Cambridge University Press. Canadian Center on Substance Abuse. (1996). The cost of substance abuse in Canada. Ottawa, Canada. Casswell, S. & Zhang, J . F. (1997). Access to alcohol from licensed premises during adolescence: A longitudinal study. Addiction, 92, 737-745. Casswell S., Pledger, M. , & Pratap, S. (2002). Trajectories of drinking from 18 to 26 years: identification and prediction. Addiction, 97,1427-1437. Chassin, L., Pitts, S. C , & Prost, J . (2002). Binge drinking trajectories from adolescence to emerging adulthood in high-risk samples: Predictors and substance abuse outcomes. Journal of Counselling and Clinical Psychology, 70(1), 67-78. Chassin, L., Presson, C. C , Pitts, S. C , & Sherman, S. J . (2000). The natural history of cigarette smoking from adolescence to adulthood in a Midwestern community sample: Multiple trajectories and their psychosocial correlates. Health Psychology, 19, 223-231. Choi, W. S., Gilpin, E. A., Farkas, A. J., & Pierce, J . P. (2001). Determining the probability of future smoking among adolescents. Addiction, 96, 313-323. Choi, W. S., Pierce, J . P., Gilpin, E. A., Farkas, A. J., & Berry, C. C. (1997). Which adolescent experimenters progress to established smoking in the United States. American Journal of Preventive Medicine, 13, 385-391. Choquet, M. , & Ledoux, S. (1985). Self-reported alcohol consumption among high school students in France. Epidemiological approach of the alcoholisation process and its evolution. Drug and Alcohol Dependence, 16,133-143. Coger, R. W., Moe, K. L., & Serafetinides, E. A. (1996). Attention Deficit Disorder in adults and nicotine dependence: Psychobiological factors in resistance to recovery?" Journal of Psychoactive Drugs, 28(3), 229-240. 94 Colder, C. R., Campbell, R. T., Ruel, E., Richardson, J. L., & Flay, B. R. (2002). A finite mixture model of growth trajectories of adolescent alcohol use: Predictors and consequences. Journal of Counselling and Clinical Psychology, 70(4), 976-985. Colder, C. R., Mott, J., Levy, S., & Flay B. (2000). The relation of perceived neighborhood danger to childhood aggression: A test of mediating mechanisms. American Journal of Community Psychology, 28(1), 83-103. Colder, C.R., Metha, P., Balanda, K., Campbell, R.T., Mayhew, K. P., & Stanton, W. R. (2001). Identifying trajectories of adolescent smoking: an application of latent growth mixture modeling. Health Psychology, 20(2), 127-135. Colder, C. R., & O'Connor, R. (2002). Attention biases and disinhibited behavior as predictors of alcohol use and enhancement reasons for drinking. Psychology of Addictive Behavior, 16(4), 325-332. Conners, C, Casat, C, Gualtieri, C, Weller, E., Reader, M., Reiss, A., Weller, R., Khayrallah, M., & Ascher, J. (1996). Bupropion hydrochloride in attention deficit disorder with hyperactivity. Journal of the American Academy of Child and Adolescent Psychiatry, 34(10), 1314-1321. Conrad, K., Flay, B. R., & Hill, D. (1992). Why children start smoking cigarettes: predictors of onset. British Journal of Addictions, 87(12), 1711-1724. Costanza, M. C, & Afifi, A. A. (1979). Comparison of stopping rules in forward stepwise discriminant analysis. Journal of the American Statistical Association, 74, 777-785. Dalack, G. W., Healy, D. J., & Meador-Woodruff, J. H. (1998). Nicotine dependence in Schizophrenia: Clinical phenomena and laboratory findings. American Journal of Psychiatry, 155(11), 1490-1501. Delorme, D. E., Kreshel, P. J., & Reid, L. N. (2003). Lighting Up: Young adults' autobiographical accounts of their first smoking experiences. Youth and Society, 34(4), pp. 468-496. Dembo, R., Williams, L., La Voie, L., Getreu, A., Berry, E., Genung, L., Schmeidler, J., Wish, E. D., & Kern, J. (1990). A longitudinal study of the relationships among alcohol use, marijuana/hashish use, cocaine use, and emotional/psychological functioning problems in a cohort of high-risk youths. International Journal of Addiction, 25(11), 1341-1382. Deptula, D., & Cohen, R. (2004). Aggressive, rejected, and delinquent children and adolescents: A comparison of their friendships. Aggression and Violent Behavior, 9, 75-104. DiFranza, J. R., & Guerrera, M. P. (1990). Alcoholism and smoking. Journal of Studies on Alcohol, 51,130-135. Dishion, T. J., Capaldi, D. M., & Yoerger, K. (1999). Middle childhood antecedents to progressions in make adolescent substance use: An ecological analysis of risk and protection. Journal of Adolescent Research, 14,175-205. Distefani, J. M., Gilpin, E. A., Choi, W. S., & Pierce, J. P. (1998). Parental influences predict adolescent smoking in the United States, 1989-1993. Journal of Adolescent Health, 22, 466-474. 95 Dolcini, M . M . , Adler, N . E., Lee, P., & Bauman, K. L. (2003). An assessment of the validity of adolescent self-reported smoking using three biological indicators. Nicotine & Tobacco Research, 5(4), 473-483. Donato, F., Monarca, S., Chiesa, R., Feretti, D., Modolo, M . A., & Nardi, G. (1995). Patterns and covariates of alcohol drinking among high school students in 10 towns in Italy: A cross-sectional study. Drug and Alcohol Dependence, 37,59-69. Draper R. (1995). Trends in alcohol policy in Canada. Proceeding of the 34 t h International Congress on Alcoholism and Drug Dependence, Calgary 518-521, Calgary. Ellickson, P., Bell, R., & McGuigan, K. (1993). Preventing adolescent drug use: Long-term results of a junior high program. American Journal of Public Health, 83, 856-861. Elliott, D. S., Huizinga, D., & Menard, S. (1989). Multiple problem youth: Delinquency, substance use, and mental health problems. New York: Springer-Verlag. Elliott, D. S., Huizinga, D., & Ageton, S. S. (1985). Explaining delinquency and drug use. Beverly Hills: Sage. Erickson, E. H. (1968). Identity, youth, and crisis. New York: Norton. Erikson, E. (1959). Identity and the life cycle. Selected Papers. With a Historical Introduction by David Rapaport. New York: Int. Univ. Press. Erikson, E. (1963). Childhood and society. New York: Norton. Evans R. I., Rozell R. M . , Mittlemark M . B., Hansen A., Bane A., & Havis J . (1978). Deterring the onset of smoking in children: Knowledge of immediate physiological effect and coping with peer pressure, media pressure, and parent modeling. Journal of Applied Social Psychology, 8,126-135. Faraone, S. V., Biederman, J., Spencer, T., Wilens, T., Seidman, L. J., Mick, E., & Doyle, A. E. (2000). Attention-deficit/hyperactivity disorder in adults: An overview. Biological Psychiatry, 48(1), 9-20. Feldman, S., & Elliott, G. (1990). Adolescence: Path to a productive life or a diminished future? Carnegie Quarterly, 35,1-13. Fergusson, D. M . , & Horwood, L. J. (1995). Transitions to cigarette smoking during adolescence. Addictive Behavior, 20(5), 627-642. Flay, B. R., d'Avernas, J . R., Best, J. A., Kersell, M . W., & Ryan, K. B. (1983). Cigarette smoking: Why young people do it and ways of preventing it. In P. J . McGrath, & P. Firestone (Eds.), Pediatric and adolescent behavioral medicine (pp. 97-127). New York: Springer Publishing Company. Flay, B. R., Hu, F. B., & Richardson, J. (1998). Psychosocial predictors of different stages of cigarette smoking among high school students. Preventive Medicine, 27, A9-A18. Flint, A. J., Yamada, E. G., & Novotny, T. E. (1998). Black-White differences in cigarette smoking uptake: Progression from adolescent experimentation to regular use. Preventive Medicine, 27, 358-364. 96 Galambos, N . L., & Leadbeater, B. J . (2002). Transitions in adolescent research. In W. W. Hartup & R. K. Silbereisen (Eds.), Growing points in developmental science: An introduction (pp. 287-306). New York: Psychology Press. Gilmore, J . (2002). Report on smoking in Canada, 1985 to 2001. Statistics Canada, Catalogue 82F0077XIE. Glantz, M . D. & Pickens, R. W. (1992). Vulnerability to drug abuse: introduction and overview. In M . D. Glantz, & R. W. Pickens (Eds.), Vulnerability to Drug Abuse (pp. 1-14). Washington, DC: American Psychological Association. Grant, B. F., & Dawson, D. A. (1997). Age at onset of alcohol use and its association with DSM-IV alcohol abuse and dependence: results from the National Longitudinal Alcohol Epidemiologic Survey. Journal of Substance Abuse, 9,103-110. Green, J. (1997). Risk and the construction of social identity: Children's talk about accidents. Sociology of Health & Illness, 19(4), 457-479. Guo, J., Collins, L. M . , Hil l , K. G., & Hawkins, D. J . (2000). Developmental pathways to alcohol abuse and dependence in young adulthood. Journal of Studies on Alcohol, 61(6), 799-817. Hall, G. S. (1904). Adolescence: Its psychology and its relations to physiology, anthropology, sociology, sex, crime, religion, and education. New York: Appleton-Century-Crofts. Harter, S. (1990). Self and identity development. In S. S. Feldman, & G. R. Elliott (Eds.), At the threshold: The developing adolescent (pp. 352-388). Cambridge, MA: Harvard University Press. Hattie, J . (1992). Self-concept. Hillsdale, NJ: Erlbaum. Hawkins, J . D., Graham, J . W., Maguin, E., Abbott, R., Hill , K. G., & Catalano, R. F. (1997). Exploring the effects of age of alcohol use initiation and psychosocial risk factors on subsequent alcohol misuse. Journal of Studies on Alcohol, 58, 280-290. Hawkins, J . D., & Weis, J . G. (1985). The social development model: An integrated approach to delinquency prevention. Journal of Primary Prevention, 6, 73-97. Health Canada. (1995). Canada's alcohol and other drugs survey. Catalogues No. H39-338/1995E. Health Canada. (2001). Quitting smoking among adults. CTUMS (Canadian Tobacco Use Monitoring Survey), Annual, February-December 2001. Health Canada. (2003). Canadian tobacco use monitoring survey (CTUMS). Health Promotion Directorate, Statistics Canada. Special Surveys Group, http://www.hc-sc.gc.ca/hecs-sesc/tobacco/research/ctums/#2Q03. Hill, G. K., White, H. R., Chung I., Hawkins, D. J., & Catalano, R. F. (2000). Early adult outcomes of adolescent binge drinking: Person- and variable-centred analysis of binge drinking trajectories. Alcoholism: Clinical and Experimental Research, 24(C), 892-901. Hosmer, D. W., & Lemeshow, S. (2000). Applied logistic regression. New York: Wiley. 97 Human Resources Development Canada and Statistics Canada. (1996). Growing up in Canada: National longitudinal survey of children and youth. Ottawa: Statistics Canada. Igra, V., & Irwin, C. E., Jr. (1996). Theories of adolescent risk-taking behavior. In R. J . DiClemente, W. B. Hansen, & L. E. Ponton (Eds.), Handbook of adolescent risk behavior (pp. 35-51). New York: Plenum Press. Jackson, K. M . , Sher, K. J., Cooper, M . L. & Wood, P. K. (2002). Adolescent alcohol and tobacco use: Onset, persistence and trajectories of use across two samples. Addiction, 9X50). 517-531-Jessor, R., & Jessor, S. (1980). A social-psychological framework for studying drug use. In Theories on drug abuse: Selected contemporary perspectives (NIDA Research Monograph 30) (pp. 102-109). Washington, DC: U.S. Government Printing Office. Jessor, R., & Jessor, S. L. (1977). Problem behavior and psychosocial development. New York: Academic Press. Jessor, R., Donovan, J. E., & Costa, F. M . (1991). Beyond adolescence: Problem behavior and young adult development. Cambridge, England: Cambridge University Press. Johnson, J. L., Bottorff, J. L., Moffat, B., Ratner, P. A., Shoveller, J . A., & Lovato, C. Y. (2002). Tobacco dependence: Adolescents' perspectives on the need to smoke. Social Science & Medicine, 56,1481-1492. Johnson, J . L., Ratner, P. A., Tucker, R. S., Zumbo, B., Bottorff, J . L., Prkachin, K. M. , & Shoveller, J . (in press). Development of a multi-dimensional measure of tobacco dependence in adolescence. Addictive Behaviors. Johnston, L. D., O'Malley, P.M., & Bachman, J . G. (1994). National survey results on drug use from the monitoring the future study, 1975-1993: Vol. II. College students and young adults [National Institute on Drug Abuse]. NIH Publication No. 94-3810. Washington, DC: U.S. Government Printing Office. Jones, B. L., Nagin, D. S., & Roeder, K. (2001). A SAS procedure based on mixture models for estimating developmental trajectories. Sociological Method Research, 29, 374-393. Kandel, D. B. (1980). Developmental stages in adolescent drug involvement. In D. J. Lettieri, M . Sayers, & H. Wallenstein (Eds.), Theories on drug abuse. Washington, DC: U.S. Department of Health and Drug Services and National Institute on Drug Abuse. Kandel, D. B., & Yamaguchi, K. (1985). Developmental patterns of the use of legal, illegal and medically prescribed psychotropic drugs from adolescence to young adulthood. In C. L. Jones & R. J . Battes (Eds.), Etiology of drug abuse: Implications for prevention (pp. 193-235). Washington, DC: U.S. Government Printing Office Kandel, D. B., Kessler, R. C , & Margulies, R. Z. (1978). Antecedents of adolescent initiation into stages of drug use: A developmental analysis. In D. B. Kandel (Ed.), Longitudinal research in drug use: Empirical findings and methodological issues (pp. 13-40). Washington, DC: Hemisphere Publishing Corp. Kaplan, H. B., Martin, S. S. & Robbins, C. (1986). Escalation of marijuana use: Application of a general theory of deviant behavior. Journal of Health and Social Behavior, 27, 44-61. 98 Kellam, S. G., Brown, C. H. , & Fleming, J . P. (1982). developmental epidemiological studies of substance use in Woodlawn: Implications for prevention research strategy. NIDA Research Monographs, 41, 21-33. Kett, J . F. (1977). Rites of passage: Adolescence in America, 1790 to the present. New York: Basic Books. Kokkevi, A., & Stefanis, C. (1991). The epidemiology of licit and illicit substance use among high school students in Greece. American Journal of Public Health, 81, 48-52. Kornhauser, R. R. (1978). Social sources of delinquency. Chicago, IL: University of Chicago Press. Lee, K. I., & Koval, J . J . (1997). Determination of the best significance level in forward stepwise logistic regression. Communications in Statistics: Simulation and Computation, 26(2), 559-575-Leonard, K. E. & Blane, T. (1999). Psychological theories of drinking and alcoholism. The Guilford Press: New York. Lerner, J . V., & Vicary, J . R. (1984). Difficult temperament and drug use: Analyses from the New York longitudinal study. Journal of Drug Education, 14,1-8. Leventhal, H . & Cleary, P. D. (1983). The smoking problem: A review of the research and theory in behavioral risk modification. Psychological Bulletin, 88, 370-405. Levin, E. D., Conners, C. K., Sparrow, E., Hinton, S. C , Erhardt, D., Meek, W. H., Rose, J . E., & March, J . (1996). Nicotine effects on adults with attention-deficit/hyperactivity disorder. Psychopharmacology, 123, 55-63. Lovato, C. Ys, Johnson, J . L., Maggi, S., Baillie, L. E., Ratner, P. A., & Shoveller, J . A. (2002). Early patterns of smoking acquisition as described by adolescents. Unpublished manuscript. Maggi, S., Linn, G., & Marion, S.A. (in press). Are questions from the Italian National Health Survey adequate to measure prevalence of smoking among teens? Substance Use & Misuse. Mausner, B. (1973). An ecological view of cigarette smoking. Journal of Abnormal Psychology, 81,115-126. Mayhew, K. P., Flay, B. R., & Mott, J . A. (2000). Stages in the development of adolescent smoking. Drug Alcohol Dependence, 59(1), 61-81. McCreary Centre Society. (2004). Healthy youth development: Highlights from the 2003 Adolescent Health Survey III. ISBN 1-895438-55-1. Moss, A. J., Allen, K. F., Giovino, G. A., & Mills, S. L. (1992). Recent trends in adolescent smoking, smoking-uptake correlates, and expectations about the future. US Department of Health and Human Services, Public Health Services Center for Disease Control and Prevention National Center for Health Statistics. Advance Data 221. Muthen, B. O. & Muthen, L. K. (2000). The development of heavy drinking and alcohol related problems from ages 18 to 37 in a U.S. national sample. Journal of Studies on Alcohol, 61(2), 290-307. 99 Nagin, D. S. (1999). Analyzing developmental trajectories: A semi-parametric, group-based approach. Psychological Methods, 4(2), 139-157. Newcomb, M . D., & Bentler, P. M . (1989). Substance use and abuse among children and teenagers. Washington, DC: American Psychological Association, 44, 001-007. O'Malley, P. M . & Wagenaar, A. C. (1991). Effects of minimum drinking age laws on alcohol use, related behaviours and traffic crash involvement among American youth: 1979-87. Journal of Studies on Alcohol, 52, 278-490. Pavis, S., Masters, H., & Cunningham-Burley, S. (1996) Lay concepts of positive mental health and how it can be maintained. Edinburgh: University of Edinburgh. Pavis, S., Cunningham-Burley, S., & Amos, A. (1997). Alcohol consumption and young people: Exploring meaning and social context. Health Education Research: Theory & Practice, 12(3), 311-22. Pedersen, W. (1991). Mental health, sensation seeking and drug use patterns: a longitudinal study. British Journal of Addiction, 86(2), 195-204. Petraitis, J., Flay, B. R., & Miller, T. Q. (1995). Reviewing theories of adolescent substance use: Organizing pieces in the puzzle. Psychological Bulletin, 11X1), 67-86. Pierce, J . P., Choi, W. S., Gilpin, E. A., Farkas, A. J., & Merritt, R. K. (1996). Validation of susceptibility as a predictor of which adolescents take up smoking in the United States. Health Psychology, 15, 355-361. Pieterse, M . E., Seydel, E. R., DeVries, H. , Mudde, A. N. , & Kok, G. J . (2001). Effectiveness of a minimal contact smoking cessation program for Dutch general practitioners: A randomized controlled trial. Preventive Medicine, 32(2), 182-90. Plumridge, E., & Chetwynd, J . (1999). Identity and the social construction of risk: Injecting drug use. Sociology of Health & Illness, 21(3), 329-343. Pomerleau, O. F., Downey, K. K., Stelson, F. W., & Pomerleau, C. S. (1995). Cigarette smoking in adult patients diagnosed with Attention Deficit Hyperactivity Disorder. Journal of Substance Abuse, 7(3), 373-378. Poulin, C , & Elliott, D. (1997). Alcohol, tobacco and cannabis use among Nova Scotia adolescents: Implications for prevention and harm reduction. Canadian Medical Association Journal, 156(10), 1387-1393. Rachel, J . V., Guess, L. L., Hubbard, R. L., & Maisto, S. A. (1982). Alcohol misuse by adolescents. Alcohol Health & Research World, 6, 61-68. Rehma, J., Rehnd, N. , Roome, R., Monteirof, M . , Gmelg, G., Jerniganh, D., & Fricki, U. (2003). The global distribution of average volume of alcohol consumption and patterns of drinking. European Addiction Research, 9(4), 147-156. Rhodes, T. (1997). Models of outreach among injecting drug users. London: Centre for Research on Drugs and Health Behaviour. Root, M . (1999). Philosophy of social science. Blackwell: Oxford UK & Cambridge USA. Schwarz, G. (1978). Estimating dimensions of a model. Annals of Statistics, 6,461-464. 100 Shedler, J., & Block, J. (1990). Adolescent drug use and psychological health: A longitudinal inquiry. American Psychologist, 45(5), 612-630. Sherif, M . (1961). Conformity-deviation, norms and group relations. In I. A. Bass, & B. M . Berg (Eds.), Conformity and deviation (pp. 159-198). New York: Harper Shiffman, S., Paty, J . A., Gnys, M . , Kassel, J . D., & Elash, C. (1995). Nicotine withdrawal in chippers and regular smokers: subjective and cognitive effects. Health Psychology, 14, 301-309. Silbereisen, R., Eyferth, K. & Rudiger, G. (Eds.). (1986). Development as action in context: Problem behaviour and normal youth development. New York: Sringer-Verlag. Silbereisen, R. K., & Reitzle, M . (2001). Psychologie 2000 - Bericht ueber den 42. Kongress der Deutschen Gesellschaft fuer Psychologie in Jena 2000. Lengerich: Pabst. Simcha-Fagan, O. & Schwartz, J . E. (1986). Neighborhood and delinquency: An assessment of contextual effects. Criminology, 24(4), 667-701. Simons, R. L., Conger, R. D., & Witbeck, L. B. (1988). A multistage social learning model of the influences of family and peers upon adolescent substance abuse. Journal of Drug Issues, 18, 293-315. Single, E., Ashley, M . J., Bondy, S., Rankin, J., & Rehm, J . (1999). Evidence regarding the level of alcohol consumption considered to be low-risk for men and women. Canberra, Australian Commonwealth Department of Health and Aged Care: 1999. Single, E., Maclennan A., & Macneil P. (1994). Horizons 1994: Alcohol and other drug use in Canada. Canadian Centre for Substance Abuse. Cat. No. H39-307/ 1994E. Stanton, W. R., Silva, P. A., & Oei, T. P. (1991). Change in children's smoking from age 9 to age 15 years: The Dunedin study. Public Health, 105(6), 425-433. Statistics Canada. (2000). Canadian tobacco use monitoring survey, 1999. Catalogue No 82M009XCB. Statistics Canada. (2004). Census 2001. Catalogue No. 92-376-XIE. Stern, R. A., Prochaska, J . O., Velicer, W. F., & Elder, J . P. (1987). Stages of adolescent cigarette smoking acquisition: measurement and sample profiles. Addictive Behaviours, 12, 319-329-Stice, E., Myers, M . & Brown, S. A. (1998). A longitudinal grouping analysis of adolescent substance use escalation and de-escalation. Psychology of Addictive Behaviors, 12(1), 14-27. Stone, K. D. (2002). Atopic diseases of childhood. Current Opinions in Pediatrics, 14(5), 634-646. Sutherland, E. H. (1939) Principles of criminology. Philadelphia: Lippincott. Tanskanen, A., Viinamaki, H., Koivumaa-Honkanen, H-T., Hintikka, J., Jaaskelainen, J., & Lehtonen, J . (1999). Smoking and depression among psychiatric patients. Nordic Journal of Psychiatry, 53, 45-48. 101 Tarrant, M . (2002). Adolescent peer groups and social identity. Social Development, 11,110-123. Teichman, M . , Barnea, Z. & Rahav, G. (1989). Sensation seeking, state and trait anxiety, and depressive mood in adolescent substance users. International Journal of Addiction, 24(1), 87-99. Trimmel, M . , & Wittberger, S. (2004). Effects of transdermally administered nicotine on aspects of attention, task load, and mood in women and men. Pharmacology & Biochemistry of Behavior, 78(3), 639-645. U.S.D.E. (1993). Youth & alcohol: Selected reports to the Surgeon General. U.S. Department of Education (1993) PH334. United States Department of Health and Human Services (USDHHS). (2000). Tenth special report to the US Congress on alcohol and health from the Secretary of Health and Human Services. Washington, National Institutes of Health, 2000 (NIH Publications, No. 00-1583.2000). United States. Public Health Service. Office of the Surgeon General Smoking and Health in the Americas. (1992). Surgeon General's report. National Center for Chronic Disease Prevention and Health Promotion (U.S.). Office on Smoking and Health. USDHHS. (1994). Preventing tobacco use among young people. U.S. Department of Health and Human Services, Public Health Services, Center for Disease Control and Prevention, National Center for Chronic Disease Prevention and Health Promotion, Office on Smoking and Health. Vartiainen, E., Pallonen, U., McAlister, A.L., & Puska, P. (1990). Eight-year follow-up results of an adolescent smoking prevention program: The North Karelia Youth Project. American Journal of Public Health, 80(1), 78-79. Wagenaar, A. (1993). Research affects public policy: the case of legal drinking age in the U.S. Addiction, 88, 75S-81S. Wechsler, H. , Kuo, M. , Lee, H., Dowdall, G. W. (2000). Environmental correlates of underage alcohol use and related problems of college students. American Journal of Preventive Medicine, 91, 24-29. White, H. R., Pandina, R. J., & LaGrange, R. (1987). Longitudinal predictors of serious substance use and delinquency. Criminology, 3, 715-40. Williams, T. P. & Lillis, R. P. (1988). Long-term changes in reported alcohol purchasing and consumption following and increase in New York State's purchase age to 19. British Journal of Addiction, 83, 209-217. Wills, T. A., McNamara, G. M . , Vaccaro, D., & Hirky, A. E. (1996). Escalated substance use: a longitudinal grouping analysis from early to middle adolescence. Journal of Abnormal Psychology, 105(2), 166-180. Windle, M . (2003). Alcohol use among adolescents and young adults. Alcohol Research and Health, 27(1), 79-85. Winship, C , & Radbill, L. (1994). Sampling weights and regression analysis. Sociological Methods and Research, 23(2), 230-257. 102 Yamaguchi, K. & Kandel, D. (1985). On the resolution of role incompatibility: A life event history analysis of family roles and marijuana use. American Journal of Sociology, 90,1284-1325-York, J. (1999). Clinical significance of alcohol intake parameters at initiation of drinking. Alcohol, 19(1), 97-99-103 Appendix I: Content of the first assessment of the National Longitudinal Survey of Children and Youth conducted in 1 9 9 4 / 9 5 Parents/Other Family Members Demographics sex marital status relationships-everyone to everyone else country of birth/citizenship/immigration ethnicity language religion Education highest grade level diploma/degree current attendance Labour Force main activity employment hours worked; wage rate shifts/weekend work Income sources (household) amount (household, respondent) Physical Health restriction of activities chronic conditions general health smoking alcohol consumption maternal history Mental Health (respondent) depression Family Family Functioning Marital Satisfaction Social Support (respondent) Housing owned subsidized condition number of bedrooms Community Neighbourhood satisfaction safety social cohesion problems 104 volunteering neighbourhood observation (by interviewer) problems land use condition of buildings School Teacher child's achievement child's behaviour teaching practices parental involvement teacher demographics Principal students at the school disciplinary problems parental involvement principal demographics Math Skills Test Child Demographics Child Health general health height and weight level of activity health status injuries chronic conditions/activity limitations health care use medications traumatic event Perinatal Information mother's prenatal health delivery details mother's/child's postnatal health breast-feeding Temperament Education level school type/language absenteeism behavioural problems achievement attitudes/expectations special education moves Literacy reading encourage writing homework Activities preschool extracurricular TV, video games Responsibilities Child Behaviour Hyperactivity-inattention Prosocial behaviour Emotional disorder-anxiety 105 Physical aggression Indirect aggression Property offences Positive interaction sleep/patterns/feeding feelings/actions difficult behaviour Motor and Social Development Parenting positive interaction parenting techniques basic care Relationships Family and Custody History custody of child previous and/or subsequent unions of parents separations siblings living outside of the home Child Care types hours summer care Receptive vocabulary 10-11 year olds Friends and Family School Homework Feelings and Behaviour Puberty Smoking, Drinking & Drugs Activities Self-esteem 106 Appendix II: Predictors F r o m When Participants Were 8-9 Years of Age List of predictors measured in 1994/95 when participants were 8-9 years old. These predictors were grouped in the following sub-categories and were included in the data reduction process 1. Child Behaviour Hyperactivity-inattention Prosocial behaviour Emotional disorder-anxiety Physical aggression Indirect aggression Property offences Positive interaction 2. Child Demographics Gender Living in urban-rural area ' Province of residence 3 . Literacy & Mathematics How often do you check his/her homework or provide help with homework? How often does he/she read for pleasure? How often does he/she talk about a book with family or friends? How often does he/she go to the library including the library school? Mathematics Computation Test 4 . Activities In the last 12 months, outside of school hours, how often has (...) taken part in any clubs, groups or community programs with leadership, such as Brownies, Cubs or church groups? Taken part in unorganized sports or physical activities? Taken part in any sport which involved coaching or instruction? Taken lessons in music, dance, art or other non-sport activities? Played computer or video games? How often does he/she play alone (e.g., riding a bike, doing a craft or hobby, playing ball)? About how many days a week on average does he/she watches TV or videos at home? On those days, how many hours on average does he/she spend watching TV or videos? 5. Peer Relationships About how many days a week does he/she do things with friends? About how many close friends does he/she have? How many of his/her close friends do you know by sight and by first and last name? When it comes to meeting new children and making new friends is he/she shy? How often does he/she hang around with kids you think are frequently in trouble? How well has he/she gotten along with other children such as friend and classmates (excluding brothers and sisters)? How well has he/she gotten along with his/her brother and/or sister? 1 0 7 6. Relationships with adults During the past six months, how well has he/she gotten along with his/her parents? Since starting school in the fall, how well has he/she gotten along with his/her teachers at school? 7 . Health status Would you say child's health is: excellent, good or fair Has child ever had asthma? Does child have bronchitis? Does child have kidney disease? Does child have emotional problems? Has child had any worry or unhappiness? 8. Mother depression and isolation Mother's depression Social support 9. Parenting and family functioning Hostile ineffective parenting Consistent parenting Punitive-aversive parenting Family functioning 10. Mother characteristics Age Years of education of mother Mother's working status (part time, full time) Mother unemployed Mother's Immigrations status English as the first language Marital status of the mother 11. Mother smoking Smokes cigarettes How many cigarettes per day 12. Mother drinking Drank in the past year How often drank alcohol Number of times had 5 or more drinks at once Highest number of drinks at once 13. Father characteristics Age of spouse/partner Years of education of father Father's working status (part time, full time) Father unemployed Gender of spouse/partner Marital status of the spouse/partner Father's Immigrations status English as the first language 14. Father smoking Smokes cigarettes How many cigarettes per day 108 15. Father drinking Drank in the past year How often drank alcohol Number of times had 5 or more drinks at once Highest number of drinks at once 16. Family Socio-Demographic Characteristics Income adequacy Socioeconomic status Annual household income Blended family status Step family status Intact family status Single-parent family 17. Siblings Older siblings of the child in household Younger siblings of child in household Siblings same age as child in household Number of siblings at home 18. Neighbourhood Neighbourhood safety Type of neighbours Neighbourhood problems How would you rate the traffic? Is there litter around this area? Do people loiter or hang out? Are people shouting or fighting? Are intoxicated persons visible? How would you characterize land use? What conditions are the buildings in? 109 Appendix III: Predictors of smoking that were eliminated through the multinomial regression data reduction procedure 'True' non significant results 1. Child Behaviour Prosocial behaviour Emotional disorder-anxiety Physical aggression Indirect aggression Property offences Positive interaction 2. Child Demographics Living in urban-rural area Province of residence 3 . Literacy & Mathematics How often does he/she read for pleasure? How often does he/she talk about a book with family or friends? How often does he/she go to the library including the library school? Mathematics Computation Test 4 . Activities In the last 12 months, outside of school hours, how often has (...) taken part in any clubs, groups or community programs with leadership, such as Brownies, Cubs or church groups? Taken part in unorganized sports or physical activities? Taken part in any sport which involved coaching or instruction? Played computer or video games? How often does he/she play alone (e.g., riding a bike, doing a craft or hobby, playing ball)? About how many days a week on average does he/she watches TV or videos at home? On those days, how many hours on average does he/she spend watching TV or videos? 5. Peer Relationships About how many days a week does he/she do things with friends? How many of his/her close friends do you know by sight and by first and last name? When it comes to meeting new children and making new friends is he/she shy? How well has he/she gotten along with other children such as friend and classmates (excluding brothers and sisters)? How well has he/she gotten along with his/her brother and/or sister? 6. Relationships with adults During the past six months, how well has he/she gotten along with his/her parents? Since starting school in the fall, how well has he/she gotten along with his/her teachers at school? 7. Health status Has child ever had asthma? Does child have bronchitis? Does child have kidney disease? Does child have emotional problems? Has child had any worry or unhappiness? 110 8. Mother depression and isolation Mother's depression 9. Parenting and family functioning Consistent parenting Punitive-aversive parenting 10. Mother characteristics Age Years of education of mother Mother's working status (part time, full time) Mother's Immigrations status English as the first language Marital status of the mother 11. Mother drinking Drank in the past year Number of times had 5 or more drinks at once Highest number of drinks at once 12. Father characteristics Age of spouse/partner Years of education of father Father's working status (part time, full time) Gender of spouse/partner Marital status of the spouse/partner Father's Immigrations status English as the first language 13. Father smoking How many cigarettes per day 14. Father drinking Highest number of drinks at once 15. Family Socio-Demographic Characteristics Blended family status Step family status 16. Siblings Siblings same age as child in household Number of siblings at home 17. Neighbourhood Neighbourhood safety Type of neighbours Neighbourhood problems How would you rate the traffic? Is there litter around this area? Do people loiter or hang out? Are people shouting or fighting? Are intoxicated persons visible? How would you characterize land use? What conditions are the buildings in? 111 Dropped due to collinearity 1. Mother characteristics Mother unemployed 2. Mother smoking Smokes cigarettes 3. Mother drinking Drank in the past year 4. Father characteristics Father unemployed 5. Father smoking Smokes cigarettes 6. Father drinking Drank in the past year 112 Appendix IV: Predictors of drinking that generated non significant results and were eliminated 1. Child Behaviour Hyperactivity-inattention Prosocial behaviour Emotional disorder-anxiety Indirect aggression Property offences Positive interaction 2. Child Demographics Gender Living in urban-rural area 3. Literacy & Mathematics How often do you check his/her homework or provide help with homework? How often does he/she read for pleasure? How often does he/she talk about a book with family or friends? How often does he/she go to the library including the library school? Mathematics Computation Test 4. Activities In the last 12 months, outside of school hours, how often has (...): Taken part in unorganized sports or physical activities? Taken part in any sport which involved coaching or instruction? Taken lessons in music, dance, art or other non-sport activities? Played computer or video games? How often does he/she play alone (e.g., riding a bike, doing a craft or hobby, playing ball)? About how many days a week on average does he/she watches TV or videos at home? On those days, how many hours on average does he/she spend watching TV or videos? 5. Peer Relationships About how many days a week does he/she do things with friends? About how many close friends does he/she have? How many of his/her close friends do you know by sight and by first and last name? When it comes to meeting new children and making new friends is he/she shy? How often does he/she hang around with kids you think are frequently in trouble? How well has he/she gotten along with other children such as friend and classmates (excluding brothers and sisters)? How well has he/she gotten along with his/her brother and/or sister? 6. Relationships with adults During the past six months, how well has he/she gotten along with his/her parents? Since starting school in the fall, how well has he/she gotten along with his/her teachers at school? 113 7. Health status Would you say child's health is: excellent, good or fair Does child have bronchitis? Does child have kidney disease? Does child have emotional problems? Has child had any worry or unhappiness? 8. Mother depression and isolation Mother's depression Social support 9. Parenting and family functioning Hostile ineffective parenting Consistent parenting Punitive-aversive parenting Family functioning 10. Mother characteristics Age Years of education of mother Mother's working status (part time, full time) Mother unemployed Mother's Immigrations status English as the first language Marital status of the mother 11. Mother smoking Smokes cigarettes How many cigarettes per day 12. Mother drinking Drank in the past year How often drank alcohol Number of times had 5 or more drinks at once Highest number of drinks at once 13. Father characteristics Age of spouse/partner Years of education of father Father's working status (part time, full time) Father unemployed Gender of spouse/partner Marital status of the spouse/partner Father's Immigrations status English as the first language 14. Father smoking Smokes cigarettes How many cigarettes per day 15. Father drinking Drank in the past year How often drank alcohol Number of times had 5 or more drinks at once Highest number of drinks at once 114 16. Family Socio-Demographic Characteristics Income adequacy Socioeconomic status Annual household income Blended family status Step family status Intact family status Single-parent family 17. Siblings Older siblings of the child in household Younger siblings of child in household Siblings same age as child in household Number of siblings at home 18. Neighbourhood Neighbourhood safety Type of neighbours Neighbourhood problems How would you rate the traffic? Is there litter around this area? Do people loiter or hang out? Are people shouting or fighting? Are intoxicated persons visible? How would you characterize land use? What conditions are the buildings in? 115 Appendix V : Procedure and results of attrition analysis Respondents and non-respondents were compared across the same predictors that were investigated to identify the early predictors of smoking and drinking as described in the 'Analytical Plan' (see page67) and listed in Appendix II. Specifically, two series of binary logistic regressions were conducted: in the first series participants who responded to the question about smoking (i.e., respondents = 1) were compared to participants who did not respond to the question about smoking (i.e., non-respondents = o); in the second series participants who responded to the question about drinking (i.e., respondents = 1) were compared to participants who did not respond to the question about drinking (i.e., non-respondents = o). Results from the attrition analysis revealed that, for the most part, respondents and non-respondents shared more similarities than differences. In fact, only five of the several predictors that were investigated generated significant results. Specifically, participants who were non-respondents for both questions about smoking and drinking tended to come from non-intact families, have a mother who drank more often, and who would drink higher quantities of alcohol more often. Furthermore, these non-respondents' mothers would report that they knew fewer of their children's close friends by name or sight, and that their children tended to read less often for pleasure. In sum, the results of the attrition analysis revealed that, despite representing only 40% of the original sample, the participants that were included in the present study, for the most part, are similar to the participants who dropped out of the NLSCY sometime between 1994-95 and 2000-01. 116 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.831.1-0099804/manifest

Comment

Related Items