Social contexts of regular smoking in adolescence: towards a multidimensional ecological model

Using data from the Longitudinal Study of Adolescent Health (Add Health), this study investigates factors at the individual, family, peer, school, neighborhood, and state levels that are important for subsequent adolescent cigarette regular smoking after controlling for the baseline smoking behavior and individual characteristics. Our analysis shows that time spent with peers and smoking of significant others (i.e., peers and families) are associated with higher risk of an adolescent becoming a regular smoker. Protective factors that may mitigate or prevent adolescents from becoming regular smokers were closeness with parents, attending a private school, attending a school with strong concentration of Hispanic students, self-perceived fitting at school, and living in a neighborhood where neighbors watch out for one another. Because we controlled for the baseline smoking behavior and a wealth of individual factors, these associations may represent causal effects of social contexts on adolescent smoking as a risky behavior that can have a lifelong harmful impact. A take home message from this study is that social contexts are important for understanding individual smoking behavior in adolescence. An ecological model that incorporates overlapping and interacting social systems is more helpful than single-system models to reveal etiology of adolescent smoking. More studies are needed to implement these ecological models to study human behavior.

ft IP] A INSTITUTE OF PUBLIC & INTERNATIONAL AFFAIRS THE UNIVERSITY OF UTAH Social Contexts of Regular Smoking in Adolescence: Towards a Multidimensional Ecological Model Ming Wen, Ph.D. Assistant Professor Department of Sociology The University of Utah Heather Van Duker, B.S. Statistician Intermountain Injury Control Research Center Department of Pediatrics The University of Utah Lenora M. Olson, Ph.D. Associate Research Professor Intermountain Injury Control Research Center Department of Pediatrics The University of Utah Acknowledgement This research was partially supported by the Primary Children's Medical Center Founda­tion. An earlier version of this paper was presented at the Population Association of America AnnualMeetingin 2006 in Los Angeles, CA. September 19, 2006 Working Paper 2006-09-14 http://ipia.utah.edu/workingpapers.html Institute of Public and International Affairs The University of Utah 260 S. Central Campus Drive, Room 215 Salt Lake City, Utah 84112Social Contexts of Regular Smoking in Adolescence: Towards a Multidimensional Ecological Model Key word: adolescent smoking, social capital, family, neighborhood, school Tables: 5 Abstract Using data from the Longitudinal Study of Adolescent Health (Add Health), this study investigates factors at the individual, family, peer, school, neighborhood, and state lev­els that are important for subsequent adolescent cigarette regular smoking after con­trolling for the baseline smoking behavior and individual characteristics. Our analysis shows that time spent with peers and smoking of significant others (i.e., peers and families) are associated with higher risk of an adolescent becoming a regular smoker. Protective factors that may mitigate or prevent adolescents from becoming regular smokers were closeness with parents, attending a private school, attending a school with strong concentration of Hispanic students, self-perceived fitting at school, and liv­ing in a neighborhood where neighbors watch out for one another. Because we con­trolled for the baseline smoking behavior and a wealth of individual factors, these as­sociations may represent causal effects of social contexts on adolescent smoking as a risky behavior that can have a lifelong harmful impact. A take home message from this study is that social contexts are important for understanding individual smoking behav­ior in adolescence. An ecological model that incorporates overlapping and interacting social systems is more helpful than single-system models to reveal etiology of adoles­cent smoking. More studies are needed to implement these ecological models to study human behavior. Author Contact Information Ming Wen, Ph.D. Assistant Professor Department of Sociology The University of Utah [e-mail] Contexts of Regular Smoking in Adolescence: Towards an Ecological Model BACKGROUND Adolescent smoking is an important health issue. Not only is nicotine one of the most addictive substances (Kandel and Merrick 2003) but approximately 80% of adults who use tobacco initiated cigarette smoking before age 18 years in the United States (Center of Disease Control and Prevention 2000). Although tobacco use may not be immediately life-threatening to adolescent smokers, cigarette smoking is a serious risk factor for poorer perceived health, mental problems, and a variety of physiological symptoms in later adulthood (Bjarnason et al. 2003). With more than one-third of high school students reporting smoking at least once a month and 17% reporting smoking a cigarette almost every day (U.S. Department of Health and Human Services 1996; U.S. Department of Health and Human Services 1998), regular teen cigarette smoking is a significant public health concern in our society (Center of Disease Control and Prevention 2000). Theoretical Perspectives An extensive amount of research has theorized and tested numerous pathways linking social context to adolescent smoking behavior. The social learning theories emphasize the importance of learning process which includes social interactions with influential others such as parents, peers, schoolmates, and neighbors. As a result, adolescents are typically protected from smoking if the socializing agents discourage smoking and cigarettes are less accessible (Crawford 2001; Eisenberg and Forster 2003; Foshee and Bauman 1992; Kirby 2002; Ma et al. 2003). 1 The socialization and social development theories argue that in addition to the direct influence of influential agents' behavior, influential agents also affect the quality and content of the adolescent's socializing process. In this pathway, parental control, parent-child closeness, parent-child communication, and parental involvement in the child's daily life are all important factors exerting daily influences on child development; and indeed, these factors have been frequently linked to adolescent risk behavior including cigarette smoking (Barnes and Farrell 1992; Johnson and Hoffmann 2000; Kandel et al. 2004; Kirby 2002; Shakib et al. 2003). Complementarily, the socioeconomic resources perspectives argue that low parental socioeconomic status (SES) puts adolescents at risk (Soteriades and DiFranza 2003; Thomson, Hanson and McLanahan 1994) while increased household income is protective, perhaps due to its linkage with parental education, positive parenting styles, and other advantaged life circumstances. Parental education is a protective factor against children's deviant behaviors because parental education is positively linked to parental expectations on children's academic achievement and it probably contributes to more constructive and efficient parenting styles. In addition, the social strain theories maintain that exposure to stressors should be linked to increased smoking (An et al. 1999; Johnson and Hoffmann 2000). In this pathway, stressors including parental conflict and separation, as an additional feature of family processes and not necessarily related to family SES, can act as a stressor (Vandewater and Lansford 1998). These stressors have been positively linked to adolescent smoking likely through raising depressive symptoms and rebelliousness in adolescents (Kirby 2002). Another identified stressor is attending an academically 2 competitive school. The competitive school environment which maintains high school- level academic expectations is a marker of the school's competitiveness and can result in a stressful school environment (Johnson and Hoffmann 2000). Clearly, these theories are not mutually exclusive and can be applied to more than one life domain. An individual's learning and socializing activities are not confined to family settings but routinely operationalized in all forms of social relationships. Resources available to the adolescent are not necessarily Significant Life Domains of Adolescents For a typical adolescent, family, peer, school, and neighborhood constitute the key life domains whereby they learn, socialize, and conduct daily activities. Each of these domains may exert unique impacts on adolescent behavior. Peer influence on adolescent well-being has been well documented (Adam and Chase-Lansdale 2002; Brook et al. 2003; Duncan, Duncan and Strycker 2002; Duncan et al. 1995; Gerard and Buehler 1999; Harris, Duncan and Boisjoly 2002; Kobus 2003). For example, research on the role of social influences found that teenagers who perceived high prevalence rates of smoking were at an increased risk of initiating this behavior (Botvin et al. 1992; Chassin et al. 1984; Sussman et al. 1994). And non-smokers are at greater risk for transitioning to tobacco use if they have smoking friends (Kobus 2003). School environment is another important domain for adolescents. They spend so much time in school, and most of their social contacts and peer groups are established, connected, and maintained in school settings (Feld 1981). School climate may structure norms, views, beliefs, and values which inevitably impact individuals' exposure to deviant or positive behaviors and internalize normative standards by which teenagers gauge what 3 is common, socially acceptable, and of little or great consequence for their future (Harris, Duncan and Boisjoly 2002). For example, "school climate," reflected in aggregate academic expectations and smoking prevalence, exerts contextual effects on adolescent smoking, after controlling for individual-level confounders (Ellickson et al. 2003; Johnson and Hoffmann 2000). Neighborhood environment is also a highly relevant social domain for adolescents who spend much of their after-school time in their local communities. Neighborhood quality is typically predictive of school cultural climate and academic performance. Poor neighborhoods deficient in social and economic resources lack capacity to maintain quality schools. Conversely, poor school quality may worsen neighborhood conditions when relatively better off residents who may have sufficient individual resources decide to move to better neighborhoods with better schools. Neighborhood environment may affect youth behaviors through other pathways as well. There is a rapidly expanding literature that addresses the relationship between place of residence and health and well­being (Kawachi and Berkman 2003). A small portion of this work is focused on youth development. Existing evidence generally points to the importance of neighborhood on youth development. For example, neighborhood contexts have been linked to youth alcohol and drug problems (Duncan, Duncan and Strycker 2002), propensity toward problem behaviors, including aggressive behavior (Kowaleski-Jones 2000), lower intelligence scores and problem behaviors (Brooks-Gunn et al. 1993), psychological problems (Aneshensel and Sucoff 1996), and crime and violence (Sampson, Morenoff and Raudenbush 2005). However, there is little evidence on the effects of neighborhood environment on adolescent smoking. As smoking (compared to diet for example) is a 4 visible behavior in the neighborhood particularly among adolescents, neighborhood context would have a real chance to have an impact (Ross 2000). Moreover, policies related to tobacco consumption may also play a role in individual smoking behavior. For example, if a school has stringent rules against teacher and student smoking and has the means and determination to implement these rules, then students of that school may feel it less tempting to try smoking because of the consequential punishment. Similarly, if a state has resources to disseminate the health impact of smoking or has policies against cigarette vending machines accessible to minors, teenagers in that state may feel more discouraged to give smoking a try. How these policies, compared to more proximate factors such as individual and family characteristics, play a role in adolescent smoking is not well known. Nevertheless, it is possible that more proximate factors matter more for individual behaviors and outcomes, as shown in the vast literature of neighborhood contextual effects on health (Kawachi and Berkman 2003). Indeed, other than the contextual factors, many individual factors have been found to be strong predictors of adolescent smoking (Kandel et al. 2004). For instance, poor school performance, low self-esteem, work for pay, and perceived physical maturity are positively associated with smoking (An et al. 1999; Carvajal et al. 2000; Resnick et al. 1997). In contrast, immigrant status has been shown to be strongly protective for both adults and adolescents against poor health practices including smoking. Multidimensional Ecological Model Although published studies regarding pathways and factors of adolescent smoking are numerous, few studies have simultaneously examined the impact of a wide range of risk or protective factors at multiple levels to adjudicate among different theories of the 5 causes of adolescent smoking. The reasons an adolescent chooses to engage in smoking are multifactorial. As has been argued previously (Bronfenbrenner 1979; Kawachi and Berkman 2003; Susser 1994; Susser 1994; Wilcox 2003), we need to recognize and analyze the contributions of overlapping and interacting contexts and explore a multidimensional ecological model of health and behavior while taking various social settings into account. A central premise of this ecological model, readily accepted in developmental psychology, is that individuals can not be studied without a consideration of the multiple ecological systems in which they operate (Brooks-Gunn et al. 1993). Using a nationally representative longitudinal sample of adolescents, this study aims to investigate factors at the individual, family, peer, school, neighborhood, and state levels that are important for adolescent cigarette smoking. In essence, we attempt to identify a fuller model of adolescent regular smoking that incorporates factors across all key life domains of a typical adolescent. METHOD Data We used data from waves 1 and 2 of the National Longitudinal Study of Adolescent Health (Add Health). More detailed information on the Add Health has been published elsewhere (Harris et al. 2003; Resnick et al. 1997; Udry 2003). Briefly, the sampling frame consists of all high schools in the United States with an enrollment of over 30 students and the junior high and middle schools that sent their graduates to those high schools in the 1994-1995 school year. From this frame, 134 schools (80 high schools and 54 middle or junior high schools) were selected to be in the Add Health sample. Using appropriate sample weights, 18,924 students from these schools make up a representative 6 sample of adolescents enrolled in the seventh through twelfth grades in the United States. Approximately one year after the first interview, members of the original sample who had not yet graduated were reinterviewed. Students completed 14,738 surveys, and sample weights could be constructed for 13,570. This study uses data for adolescents who completed both waves of the survey. After imputing missing values for explanatory variables (see the imputation method below), we have 13,552 cases for our analysis. Dependent variable The dependent variable was measured at wave II in 1996. We used a dichotomous measure of regular smoking. Responses were coded 1 if they ever smoked at least 1 cigarette every day for 30 days by wave II. Independent variable All independent variables were measured at wave I in 1994. We studied a variety of social context variables at the individual, peer, parental, family, school, neighborhood, and state levels as described below. The Appendix provides detailed description of all the measures used in this study. Table 1 illustrates sample statistics of a selected set of variables. Individual level Individual demographic and baseline measures controlled in the analyses include age, sex, ethnic/racial background, immigrant status, self-perceived physical maturity, average grade in math and English, self-perceived intelligence, self-esteem, work hours per week, earnings per week during the school year, and weekly allowance. Additionally, we controlled for reported smoking behavior at wave I (ever smoked at least 1 cigarette 7 every day for 30 days by wave I) to adjust for its relationship to smoking behavior at wave II. Peer level Measures of peer influence factors include how frequently the adolescent spends time with their three best friends during the week, and how many of their three best friends smoke1. Parent or family level Corresponding to the guiding theories of adolescent behavior, we include measures of family SES (i.e., household income, parent's education), family structure (intact vs. nonintact), household members' smoking, the mother's expectations for the adolescent regarding educational attainment, parental conflicts, parental control, parent-child closeness, parent-child communication, and an item indicating how many days per week the parent has dinner with the adolescent as a proxy for parental involvement in the adolescent's life. School Level Measures related to the school context are reported by the school administrator, by the adolescent, and collected by Add Health investigators from public sources. These measures include urbanicity of the school (urban, suburban, or rural), whether the school is private or public, percent of Hispanic, black, and Asian students attending the school, percent of students who expect to go to college (as a proxy for average school academic aspiration), a school average student smoking rate, an average score of students' rating on how well they get along with teachers and other students at their school, an average score of students' rating on how well the adolescent feels that they fit in at school, percent of 8 students smoking, percent of teachers that smoke at school, whether or not the school has a no smoking policy for the teachers, whether or not the school has a no smoking policy for the students, and severity of punishment for a violation of the no smoking policy. Neighborhood Level Neighborhood variables include both Census-based measures of neighborhood SES and self-perceived neighborhood environment. We include neighborhood median household income and percent of residents age 25 or above with college degree as measures of neighborhood SES. Perceived measures tap neighborhood social capital (i.e., neighbors watch out for each other), attachment (i.e., happy to live in the neighborhood), and use of a recreation center in the neighborhood as a proxy for neighborhood physical condition. State Level State level variables include the percent of people who smoke in the state, state restrictions on cigarette vending machines accessible to minors, whether the state has marketing restrictions for tobacco, and whether the state has a program for the dissemination of information about tobacco. Imputation for missing values To prevent the false loss of data, missing independent variable values were imputed using IVEware (Raghunathan et al. 2001). Rather than falsely cause a reduction in standard error estimates using mean substitution, IVEware uses regression imputation where imputed values are sampled from their appropriate sampling distributions. For example, binary variables are sampled from the binomial distribution, and count variables 9 are sampled from the Poisson distribution. The complete code of imputation is available upon request2. Cases missing the dependent variable were dropped. Statistical modeling Weighted logistic regression models were fit to test the risk or protective factors related to adolescent regular smoking. The modeling method has taken the complex survey design into account, generating findings that are nationally representative of adolescents in the US between 1994 and 1996. Because the primary sampling unit is the school, we used robust standard errors to control for intra-school correlations. RESULTS Table 1 selectively presents sample statistics of several variables. About 18% and 20% of students smoked at least 1 cigarette every day for 30 days in wave I (1994) and wave II (1996) respectively. These figures are close to other national estimates of adolescent regular smoking (U.S. Department of Health and Human Services 1996; U.S. Department of Health and Human Services 1998). We have a roughly balanced sample of males and females. In addition, about half of the sample is white, 27% of respondents are foreign-born, 23% of students have parents who graduated from college, and 40% of respondents have at least one household member smoking. Table 2 tests how individual, peer, and parent or family characteristics contribute to the likelihood of adolescent regular smoking, after controlling for the baseline smoking behavior which has been routinely reported as the most powerful predictor of subsequent smoking behavior (Ellickson et al. 2003). Model 2.1 (Table 2, Model 1) includes demographics and individual-level variables that have been previously reported as salient predictors of adolescent risky behavior. Except for length of living in current residence, 10 all of these variables are significant in this model. We find that girls are more likely than boys to smoke; white adolescents are more likely to smoke than black, Asians/Pacific Islander, and Hispanic adolescents; and US born adolescents are more likely than the foreign born to smoke. Self-perceived physical maturity, earnings and allowance increase smoking, whereas higher grade-point-average (GPA) and self worth protects against smoking. Model 2.2 adds peer characteristics to the model. Adolescents who spend time with peers more frequently and who have close friends who smoke are more likely to take up the smoking habit and become regular smokers. Model 2.3 adds family structure and SES variables. Unexpectedly, neither the family structure nor the SES variables exert prospective effect on regular smoking. Model 2.4 adds additional measures of parent or family characteristics, testing family-level models of adolescent smoking. Family members' smoking is significantly associated with higher likelihood of adolescent smoking, and parent-child closeness is highly protective against adolescent regular smoking, after controlling for individual-level key predictors. Model 2.5 thus provides the model that includes the significant predictors at the individual, peer, parent or family levels. The first model in Table 3, Model 3.1, adds school demographics to Model 2.5. School urbanicity has no effect on adolescents' regular smoking. However, attending a private school and having a higher percentage of Hispanic students at school is significantly protective. Model 3.2 tests the additional effect of school academic climate. Neither percent of students who expected to go to college nor school attendance rate exhibits significant effects on adolescent smoking behavior. Model 3.3 tests the effects of school social capital and smoking prevalence and shows that self-perceived fitting at 11 school (e.g. feel comfortable at school, get along well with other students and teachers at school, etc.) has a protective effect on smoking, but percent of students who smoke or teachers' smoking at school does not affect adolescent smoking in the presence of other variables. Model 3.4 shows that school policies related to cigarette consumption do not offer additional help to curb adolescent smoking. Model 3.5, the last model in Table 3, thus presents the final model that includes all significant predictors from the individual level to the school level. Table 4 examines the additional effects of neighborhood characteristics. Perceived neighborhood social capital, measured by neighbors' watching out for one another is a protective factor over and above risk or protective factors of other life domains. None of the other three neighborhood variables is a salient predictor of regular smoking in adolescence. Lastly, Table 5 presents results of examining the effects of state-level factors in addition to the impacts of more proximate factors. These higher level contextual factors are not significant after controlling for previously found significant predictors of smoking. Thus, the series of modeling, aimed to identify a fuller model of adolescent smoking, results in Model 4.3 (Model 3, Table 4) as the most parsimonious and effective model including key predictors of adolescent regular smoking across individual-, peer-, family-, school-, and neighborhood-levels. In summary, according to Model 4.3, girls are 21% more likely than boys to smoke; blacks are 125% less likely than whites to be regular smokers; every $100 more earnings per week is associated with 11% higher risk of smoking ; one level higher in frequency of hanging out with friends corresponds to 24% higher risk; one more close friend smoking 12 leads to 74% higher risk; having a household member smoking increases risk by 48%; a level higher in parent-child closeness corresponds to 1% lower risk of smoking; attending a private school makes it 34% less likely to smoke; 20% higher in Hispanic population is associated with 28% lower risk4; one unit higher in the level of self-perceived fitting at school are associated with 2% reduction in the risk of smoking; and lastly, one level higher in neighborhood social capital decreases 17% of the subsequent smoking in adolescence. DISCUSSION A large body of published studies on the factors of adolescent risk behaviors including smoking are available, however there is a paucity of research that has simultaneously examined the impact of a wide range of risk or protective factors at multiple life domains to adjudicate among different theories about the causes and pathways of adolescent risk behaviors. A unique contribution of this study is to implement a multidimensional ecological model, testing the relative contributions of overlapping social contexts of adolescent smoking, and identifying a fuller model explaining the etiology of adolescent regular smoking. Setting out from this ecological perspective, guided by previously developed theoretical frameworks, we tested an unusually large number of covariates as potential risk or protective factors of adolescent regular smoking. As expected, we found peer, family, school, and neighborhood are all important life domains exerting significant impacts on subsequent regular smoking after controlling for the baseline smoking behavior. This result points to the need of continued expansion of the explanatory 13 variables measured in empirical works to advance our understanding of how social contexts contribute to adolescent development (Brooks-Gunn et al. 1995). Among the four aforementioned theoretical frameworks of adolescent behavior, this research strongly supports the social learning perspectives. Our study found that both peer and household members' smoking were significant predictors of adolescent regular smoking, and their effects were independent of each other. By contrast, school-level and state-level prevalence of smoking were not significant factors. It is plausible that only the most influential social agents' behaviors matter for adolescents' behavior. Close friends and household members typically constitute the most accessible behavioral models of an adolescent and weaker ties may play less a role in influencing youth behavior. In this vein, because the less direct effects of more remote socializing agents, such as loosely connected neighbors or schoolmates, objective prevalence of smoking of broader social contexts may be less effective in shaping behaviors than perceived prevalence of smoking. In fact, one study recently showed that actual school-level prevalence of smoking did not predict individuals' subsequent smoking whereas perceived prevalence of smoking was associated with increased risk of smoking (Ellickson et al. 2003). The authors argued that their findings supported the significance of the selected environment (peer cigarette use) and the constructed environment (perceived school prevalence of smoking) but pointed to a weaker effect of the imposed environment (school-level actual prevalence of smoking). Our research did not distinguish the actual environment from the perceived, but we did touch the distinction between the selected environment versus the imposed environment. While we found that peer (selective environment) smoking was more important than school-level actual prevalence of smoking (imposed environment), a result consistent with 14 the Ellickson and colleagues' study (2000), we also found that family member (imposed environment) smoking was one of the strongest predictors of adolescent smoking. So perhaps it matters less whether the environment is selected or imposed than whether the ties are close enough to exert a real influence on the adolescent's behavior. In addition, cigarette smoking of close friends and family members also makes it immediately easier for adolescents to access cigarettes. The measure of frequency of hanging out with friends taps the extent to which the adolescent is attached to his/her peer groups. Although higher frequency may signal the popularity of the adolescent and abundant social support he or she may receive from the peers of the same age, our data showed that this peer behavior proved a risk factor of regular smoking in adolescence. This finding may speak to the potential downside of social capital. Although the evidence of protective effect of social capital is mounting (Kawachi and Berkman 2000; Putnam 2000), it is becoming widely accepted that social capital is neither an unqualified social good nor a panacea and at times it can be harmful (Kawachi and Berkman 2000; Kawachi and Berkman 2001; Portes 1998; Portes and Landolt 1996; Woolcock 1998). In fact, it is not difficult to imagine the downside of social capital, including its coercive aspects, the inhibition of individual expression and freedom, and the situations where social capital of the group (e.g., criminal gangs) may provide resources for its members but meanwhile be disruptive of social cohesion of the larger community. How exactly these processes unfold in real settings of peer interaction that matters to adolescent behavior and development is largely unknown and warrants further investigation. 15 Also related to the social capital theory, the socialization and social development theories, often applied to family settings, were tested and supported as well. We found that parent-child closeness was strongly protective against adolescent regular smoking. By contrast, other parenting styles and parental academic expectations do not make a difference in subsequent adolescent regular smoking. From a family perspective, therefore, it seems that to reduce adolescent smoking the most effective way would be to eliminate or reduce adult smoking in the family and boost parent-child bonding. A neighborhood-level social capital measure, neighborly watching out for one another, also exhibited strong protective impact, net of the effects of all other proven predictors of smoking in adolescence. The conflicting effects of social capital, lodged in different life settings (i.e., peer, family, neighborhood), suggest the need to investigate relative contribution of social capital of different yet overlapping life domains to adolescent risk behavior. To our knowledge, few published articles have addressed this question. The other two theories that motivated this study are the socioeconomic resource perspectives and the social strain theories. Despite intuitively appealing, these theories failed to predict adolescent subsequent smoking after controlling for sociodemographic factors, peer and family smoking, and baseline smoking behavior. Neither parental education nor household or neighborhood income showed any effect. In fact, weekly earnings are positively related to smoking behavior, a finding consistent with previous evidence that adolescent disposable income is a risk factor of smoking (Soteriades and DiFranza 2003) but contradictory to what the socioeconomic resource perspectives would predict. That is, it seems that adolescents smoke not because they lack money but because, in part, their disposable income allows them to afford cigarettes. 16 We tested the social strain theory by examining the effects of parental conflicts, school competitiveness, and neighborhood poverty. Contrary to what have been previously reported (Chuang et al. 2005; Johnson and Hoffmann 2000; Vandewater and Lansford 1998), we did not find evidence that adolescents smoke out of stress due to, say, parental conflicts, attending a school that is academically too competitive, or living in a neighborhood that is highly deprived. One possible reason why these factors did not show effects in this study compared to previous studies is that we have simultaneously examined a wider spectrum of explanatory factors some of which are either stronger predictors of smoking or mediators of effects of social stressors. On the other hand, the social strain variables we used can at best indirectly tap stress exposure among adolescents. Given that stress has been consistently linked to a wide range of health- related outcomes (Thoits 1995), more studies are needed to better capture stress exposure and vulnerability and their impacts on smoking among adolescents to further explore the social strain theory. The last and most broad context we tested is state-level policies regarding cigarette consumption. These policies do not seem to affect individual smoking behavior. However, we only examined a limited number of policies and other policies may have stronger effects. For example, as earlier ecological data from Canada showed, the relationship between alcohol price and consumption was very close, and as price fell consumption rose (Seeley 1960). Similar trends may be observed for cigarette consumption. On the other hand, findings from a recent study suggested that state-level higher cigarette excise taxes did not seem effective in checking the initiation of smoking among adolescents (DeCicca, Kenkel and Mathios 2002). Perhaps the effects of more 17 macro-level policies-by nature an ecological concept-are more readily manifested in aggregate patterns, making ecological design a sensible choice for analyzing policy impacts. The research found several patterns that are not easily explainable. Racial difference in adolescent regular smoking needs to be investigated. Obviously, it is not because whites are more deprived than blacks that they are more likely to smoke. The traditional models that explain racial/ethnic disparities in health-related health outcomes do not seem to apply in this case considering that white adolescents on average are more resourceful than black adolescents at least in terms of economic resources. The protective effect of the concentration of Hispanic students is also intriguing, especially given that this effect is independent of school prevalence of smoking and individual immigrant status. Future investigations are needed to explore these issues. This work is not without limitation. Despite the longitudinal design of this study, it is still possible that peers are formed based on common interests and habits, so it may well be the tendency to smoke that forms peer groups not the other way around. Perhaps both pathways are operating, however; that is, peers are formed based on traits and traits of peers are also affecting other peers. Future work is needed to sort out these dual directional pathways. Our neighborhood measures may not be as rich as used in some studies focusing on neighborhood effects, as neighborhood measures are limited in the Add Health. We used one item, neighborly watching out for one another, to tap neighborhood social capital, and we found significant effect of this measure on adolescent smoking. All other neighborhood variables were not significant (we have tested all the available data on 18 neighborhood contexts). Alternative neighborhood measures, such as those tapping subcultural orientation (Wen and Christakis 2006), may exhibit stronger neighborhood effects. None the less, it is also possible that neighborhood effects are just weaker for adolescent smoking, although they may be salient to other individual health and behavioral outcomes (Cagney, Browning and Wen 2005; Haan, Kaplan and Camacho 1987; Ross 2000; Sampson, Morenoff and Gannon-Rowley 2002; Wen, Browning and Cagney 2003; Wen, Cagney and Christakis 2005). How neighborhood contexts affect adolescent behavior including smoking has not been thoroughly examined (Ross 2000), a question that warrants more attention in the future. That being said, our findings provide longitudinal and nationally representative evidence of the importance of a host of social factors and pathways to adolescent smoking. As far as we know, this is the first attempt of fitting a fuller comprehensive ecological model of adolescent regular smoking in a longitudinal setting, using large-scale national data. The incredibly rich information on adolescent well-being and social settings provided by the Add Health affords an excellent opportunity of testing such a model. Controlling for individual characteristics, our study shows that frequent hanging out with peers and smoking of significant others (i.e., peers and families) are closely associated with higher risk of smoking. Meanwhile, closeness with parents, attending a private school, attending a school with strong concentration of Hispanic students, a sense of fitting in and belonging at school, and living in a neighborhood where neighborhood watch out for one another are all factors preventing adolescents from becoming regular smokers. Because we have controlled for the baseline smoking behavior along with a wealth of individual factors, these associations may possibly represent causal effects of 19 social contexts on adolescent smoking as a risky behavior that can have a lifelong harmful impact. Nevertheless, the causality issue can never be settled without an appropriate experimental design. A take home message from this study is that social contexts at the peer-, family-, school-, and neighborhood-levels are important for understanding individual smoking behavior in adolescence. A multidimensional ecological model that incorporates overlapping and interacting social systems is more helpful than single-system models to reveal etiology of adolescent smoking. More studies are needed to implement these ecological models to study human behavior. For now, the most significant contextual factors seem to be the behaviors of significant or admired others (peer or family member smoking) and the relational assets or deficits in the family (closeness with parents), among peers (frequency of hanging out with friends), at school (perceived fitting well at school), or in the neighborhood (neighborly watching out for one another). Different forms of social capital could thus be possible entry points of fruitful interventions tackling adolescent smoking. 20 References Adam, Emma K., and P Lindsay Chase-Lansdale. 2002. "Home sweet home(s): parental separations, residential moves, and adjustment problems in low-income adolescent girls." Developmental Psychology 38:792-805. An, Lawrence C., Patrick M. O'Malley, John E. Schulenberg, Jerald G. Bachman, and Lloyd D. Johnston. 1999. "Changes at the high end of risk in cigarette smoking among US high school seniors." American Journal of Public Health 89:699-705. Aneshensel, Carol S., and Clea A. Sucoff. 1996. "The neighborhood context of adolescent mental health." Journal of Health and Social Behavior 37:293-310. Barnes, Grace M., and Michael P. Farrell. 1992. "Parental support and control as predictors of adolescent drinking, delinquency, and related problem behaviors." Journal of Marriage and the Family 54:763-776. Bjarnason, Thoroddur, Aleksandra G. Davidaviciene, Patrick Miller, and Alojz Nociar. 2003. "Family structure and adolescent cigarette smoking in eleven European countries." Addition 98:815-824. Botvin, Gilbert J., Elizabeth M. Botvin, Eli Baker, and Linda Dusenbury. 1992. "The false consensus effect: predicting adolescents' tobacco use from normative expectations." Psychological Reports 70:171-178. Bronfenbrenner, Urie. 1979. The Ecology of Human Development: Experiments by Nature and Design. Cambridge, MA: Harvard University Press. Brook, David W., Judith S. Brook, Elizabeth Rubenstone, Cheshu Zhang, Merrill Singer, and Michael R. Duke. 2003. "Alcohol use in adolescents whose fathers abuse drugs." Journal of Addictive Diseases 22:11-34. Brooks-Gunn, Jeanne, Brett Brown, Greg J. Duncan, and Kristin A. Moore. 1995. "Child development in the context of family and community resources: An agenda for national data collections." in Integrating Federal Statistics on Children: Report of a Workshop, edited by Commission on Behavioral and Social Sciences and Education: The National Academy of Sciences. Brooks-Gunn, Jeanne, Greg J. Duncan, Pamela K. Klebanov, and Naomi Sealand. 1993. "Do neighborhoods influence child and adolescent development." American Journal of Sociology 99:353-95. Cagney, Kathleen A., Christopher R. Browning, and Ming Wen. 2005. "Race and self-rated health at older ages: What difference does the neighborhood make?" Journal of Gerontology: SOCIAL SCIENCES 60B:S193-S194. Carvajal, Scott C., Dawn E. Wiatrek, Richard I. Evans, C. Raymond Knee, and Susan G. Nash. 2000. "Psychosocial determinants of the onset and escalation of smoking: Cross-sectional and prospective findings in multiethnic middle school samples." Journal of Adolescent Health 27:255-65. Center of Disease Control and Prevention. 2000. "Youth risk behavior surveillance--United States-- 1999." MMWR Morbidity Mortality Weekly Report 49:9-10. Chassin, Laurie, Clark C. Presson, Steven J. Sherman, Eric Corty, and Richard Olshavsky. 1984. "Predicting the onset of cigarette smoking in adolescents: a longitudinal study." Journal of Applied Social Psychology 14:224-243. 21 Chuang, Ying-Chih, Susan T. Ennett, Karl E. Bauman, and Vangie A. Foshee. 2005. "Neighborhood influences on adolescent cigarette and alcohol use: Mediating effects through parent and peer behaviors." Journal of Health and Social Behavior 46:187-204. Crawford, M. A. 2001. "Cigarette smoking and adolescents: Messages they see and hear." Public Health Reports 116:203-15. DeCicca, Philip, Donald Kenkel, and Alan D. Mathios. 2002. "Putting out the fires: Will higher taxes reduce the onset of youth smoking?" Journal of Political Economy 110:144-169. Duncan, Susan C., Terry E. Duncan, and Lisa A. Strycker. 2002. "A multilevel analysis of neighborhood context and youth alcohol and drug problems." Prevention Science 3:125-33. Duncan, Terry E., Elizabeth Tildesley, Susan C. Duncan, and Hyman Hops. 1995. "The consistency of family and peer influences on the development of substance use in adolescence." Addiction 90:1647-1660. Eisenberg, Marla E., and Jean L. Forster. 2003. "Adolescent smoking behavior." American Journal of Preventive Medicine 25:122-128. Ellickson, Phyllis L., Chloe Bird, Maria Orlando, David Klein, and Daniel McCaffrey. 2003. "Social context and adolescent health behavior: Does school-level smoking prevalence affect students' subsequent smoking behavior." Journal of Health & Social Behavior 44:525-535. Feld, Scott. 1981. "The focused organization of social ties." American Journal of Sociology 86:1015­35. Foshee, Vangie, and Karl E. Bauman. 1992. "parental and peer characteristics as modifiers of the bond-behavior relationship: An elaboration of control theory." Journal of Health and Social Behavior 33:66-76. Gerard, Jean M., and Cheryl Buehler. 1999. "Multiple risk factors in the family environment and youth problem behaviors." Journal of Marriage and the Family 61:343-361. Haan, Mary, George A. Kaplan, and Terry Camacho. 1987. "Poverty and health: prospective evidence from the Alameda County Study." American Journal of Epidemiology 125:989-998. Harris, Kathleeen Mullan, Greg J. Duncan, and Johanne Boisjoly. 2002. "Evaluating the role of "nothing to lose" attitudes on risky behavior in adolescence." Social Forces 80:1005-1039. Harris, Kathleen Mullan, Francesca Florey, Joyce Tabor, Peter S. Bearman, Jo Jones, and J. Richard Udry. 2003. The National Longitudinal Study of Adolescent Health: Research Design [WWW document]. URL: http://www.cpc.unc.edu/projects/addhealth/design. Johnson, Robert A., and John P. Hoffmann. 2000. "Adolescent cigarette smoking in U.S. racial/ethnic subgroups: Findings from the National Education Longitudinal Study." Journal of Health and Social Behavior 41:392-407. Kandel, Denise B., Gebre-Egziabher Kiros, Christine Schaffran, and Mei-Chen Hu. 2004. "Racial/ethnic differences in cigarette smoking initiation and progression to daily smoking: A multilevel analysis." American Journal of Public Health 94:128-135. Kandel, I., and J. Merrick. 2003. "Smoking and adolescence." Int J Adolesc Med Health 15:365-8. Kawachi, Ichiro, and Lisa F. Berkman. 2000. "Social cohesion, social capital, and health." Pp. 174-190 in Social Epidemiology, edited by L. F. Berkman and I. Kawachi. New York: Oxford University Press. -. 2000. Social cohesion, social capital, and health. New York: Oxford University Press. -. 2001. "Social ties and mental health." Journal of Urban Health 78:458-467. 22 -. 2003. Neighborhoods and Health. New York: Oxford University Press. Kirby, James B. 2002. "The influence of parental separation on smoking initiation in adolescents." Journal of Health and Social Behavior 43:56-71. Kobus, Kimberly. 2003. "Peers and adolescent smoking." Addiction 98:37-55. Kowaleski-Jones, Lori. 2000. "Staying out of trouble: Community resources and problem behavior among high-risk adolescents." Journal of Marriage and Family 62:449-464. Ma, Grace X., Steve Shive, Patricia Legos, and Yin Tan. 2003. "Ethnic differences in adolescent smoking behaviors, sources of tobacco, knowledge and attitudes toward restriction policies." Addictive Behaviors 28:249-68. Portes, Alejandro. 1998. "Social Capital: Its Origins and Applications in Modern Sociology." Annu. Rev. Sociol. 24:1-24. Portes, Alejandro, and P. Landolt. 1996. "The Downside of Social Capital." American Prospect 26:18­21. Putnam, Robert D. 2000. Bowling Alone: The Collapse and Revival of American Community: Simon & Schuster Paperbacks. Raghunathan, Trivellore E., James M. Lepkowski, John Van Howewyk, and Peter Solenberger. 2001. "A multivariate technique for multiply imputing missing values using a sequence of regression models." Survey Methodology 27:85-89; The software is available at: http://www.isr.umich.edu/src/smp/ive/. Resnick, Michael D., Peter S. Bearman, Robert W. Blum, Karl E. Bauman, Kathleen M. Harris, Jo Jones, Joyce Tabor, Trisha Beuhring, Renee E. Seiving, Marcia L. Shew, Marjorie Ireland, Linda H. Bearinger, and Richard J. Udry. 1997. "Protecting adolescents from harm: findings from the National Longitudinal Study of Adolescent Health." Journal of the American Medical Association 278:823-832. Ross, Catherine E. 2000. "Walking, exercising, and smoking: Does neighborhood matter?" Social Science & Medicine 51:265-274. Sampson, Robert J., Jeffrey D. Morenoff, and Thomas Gannon-Rowley. 2002. "Assessing "neighborhood effects": social processes and new directions in research." Annual Review of Sociology 28:443-78. Sampson, Robert J., Jeffrey D. Morenoff, and Stephen Raudenbush. 2005. "Social anatomy of racial and ethnic disparities in violence." American Journal of Public Health 95:224-232. Seeley, John R. 1960. "Death by liver cirrhosis and the price of beverage alcohol." Canadian Medical Association Journal 83:1361-1366. Shakib, Sohaila, Michele Mouttapa, Anderson Johnson, Anamara Ritt-Olson, Dennis Trinidad, R., Peggy E. Gallaher, and Jennifer B. Unger. 2003. "Ethnic variation in parenting characteristics and adolescent smoking." Journal of Adolescent Health 33:88-97. Soteriades, Elpidoforos S., and Joseph R. DiFranza. 2003. "Parent's socioeconomic status, adolescents' disposable income, and adolescents' smoking status in Massachusetts." American Journal of Public Health 93:1155-60. Susser, Mervyn. 1994. "The logic in ecological: I. The logic of analysis." American Journal of Public Health 84:825-9. -. 1994. "The logic in ecological: II. The logic of design." American Journal of Public Health 84:830-5. 23 Sussman, Steve, Clyde W. Dent, Lou A. McAdams, Alan W. Stacy, Dee Burton, and Brian R. Flay. 1994. "Group self-identification and adolescent cigarette smoking: a 1-year prospective study." Journal of Abnormal Psychology 103:576-580. Thoits, P.A. 1995. "Stress, coping, and social support processes: Where are we? What next?" Journal of Health and Social Behavior:53-79. Thomson, Elizabeth, Thomas L. Hanson, and Sara S. McLanahan. 1994. "Family structure and child well-being: Economic resources vs. parental bahaviors." Social Forces 1:221-242. U.S. Department of Health and Human Services. 1996. "Trends in the Well-Being of America's Children and Youth: 1996." Office of the Assistant Secretary for Planning and Evaluation. -. 1998. "Tobacco use among high school students-United States, 1997." in Morbidity and Mortality Weekly Report. Udry, J. Richard. 2003. The National Longitudinal Study of Adolescent Health (Add Health), Waves I & II, 1994-1996; Wave III, 2001-2002 [machine-readable data file and documentation]. Chapel Hill, NC: Carolina Population Center, University of North Carolina at Chapel Hill. Vandewater, Elizabeth A., and Jennifer E. Lansford. 1998. "Influences of family structure and parental conflict on children's well-being." Family Relations 47:323-330. Wen, Ming, Christopher R. Browning, and Katheleen A. Cagney. 2003. "Poverty, Affluence and Income Inequality: Neighborhood Economic Structure and Its Implications for Self-Rated Health." Social Science & Medicine 57:843-860. Wen, Ming, Kathleen A. Cagney, and Nicholas A. Christakis. 2005. "Effect of specific aspects of community social environment on the mortality of individuals diagnosed with serious illness." Social Science & Medicine 61:1119-34. Wen, Ming, and Nicholas A. Christakis. 2006. "Effect of community distress and subcultural orientation on mortality following life-threatening disease in the elderly." Sociology of Health and Illness 28:558-582. Wilcox, Pamela. 2003. "An ecological approach to understanding youth smoking trajectories: problems and prospects." Addiction 98:57-77. Woolcock, Michael. 1998. "Social capital and economic development: toward a theoretical synthesis and policy framework." Theory and Society 27:151-208. 24 Table 1: Sample Statistics of Selected Variables Frequency Percent Regular smoker at wave I No 11055 81.6 Yes 2497 18.4 Regular smoker at wave II No 10825 79.9 Yes 2727 20.1 Sex Male 6961 51.4 Female 6591 48.6 Race/Ethnicity White 7283 53.7 Black 2866 21.2 Am. Indian 136 1.0 Asian/Islander 916 6.8 Hispanic 2183 16.1 Other 168 1.2 US born No 3614 26.7 Yes 9938 73.3 Number of best friends that smoke 0 7669 56.6 1 2775 20.5 2 1539 11.4 3 1569 11.6 At least one household member smokes No 8197 60.5 Yes 5355 39.5 Parent education Less than High School 2252 16.6 High School or equivalent 4306 31.8 Some college 3896 28.8 College graduate 1856 13.7 Post graduate 1242 9.2 Median Min Max Age 15 11 21 N=13,552 25 Table 2: Logistic Regression of Individual, Peer, and Parent or Family Characteristics on Adolescent Regular Smoking (1) (2) (3) (4) (5) Regular smoking at 17.012*** 9.972*** 9.983*** 9.594*** 9.766*** Wave I (14.535 - 19.912) (8.476 - 11.732) (8.481 - 11.751) (8.144 - 11.303) (8.324 - 11.458) Age 1.047** 1.019 1.021 1.011 1.011 (1.003 - 1.093) (0.975 - 1.066) (0.977 - 1.067) (0.965 - 1.060) (0.967 - 1.057) Sex 0.862* 0.831** 0.822** 0.839** 0.825** (0.741 - 1.002) (0.710 - 0.972) (0.705 - 0.960) (0.717 - 0.983) (0.704 - 0.966) Black 0.387*** 0.438*** 0.423*** 0.458*** 0.432*** (0.301 - 0.498) (0.339 - 0.565) (0.328 - 0.546) (0.356 - 0.588) (0.335 - 0.556) American Indian 0.983 0.932 0.926 1.012 0.948 (0.556 - 1.740) (0.496 - 1.753) (0.499 - 1.720) (0.538 - 1.906) (0.504 - 1.785) Asian/Pacific Islander 0.614*** 0.686** 0.708** 0.788 0.704** (0.445 - 0.846) (0.502 - 0.936) (0.518 - 0.969) (0.588 - 1.057) (0.533 - 0.931) Other 0.729** 0.790* 0.781* 0.844 0.814 (0.565 - 0.941) (0.614 - 1.018) (0.606 - 1.008) (0.657 - 1.086) (0.632 - 1.049) Hispanic 0.932 1.020 1.009 1.065 0.996 (0.402 - 2.161) (0.430 - 2.420) (0.423 - 2.403) (0.442 - 2.564) (0.414 - 2.395) US born 1.285** 1.222* 1.192* 1.179 (1.035 - 1.596) (0.998 - 1.497) (0.974 - 1.460) (0.968 - 1.437) Residential tenure 1.000 (0.982 - 1.018) Physical maturity 1.152*** 1.130*** 1.130*** 1 123*** (1.077 - 1.232) (1.058 - 1.207) (1.058 - 1.206) (1.051 - 1.200) Average grade of math 0.752*** 0.811*** 0.816*** 0.829*** 0.805*** and English (0.684 - 0.828) (0.736 - 0.893) (0.741 - 0.898) (0.752 - 0.913) (0.737 - 0.880) Self-perceived 0.893*** 0.913** 0.911** 0.918** intelligence (0.828 - 0.963) (0.846 - 0.984) (0.842 - 0.985) (0.849 - 0.992) Self-esteem 0.857** 0.909 (0.754 - 0.974) (0.806 - 1.025) Hours worked per week 1.006 (0.997 - 1.015) Earnings per week 1.001* 1.002** 1.001** 1.001** 1.001** (1.000 - 1.003) (1.000 - 1.003) (1.000 - 1.003) (1.000 - 1.002) (1.000 - 1.003) Allowances per week 1.008** 1.006 (1.000 - 1.016) (0.998 - 1.014) Frequency of hanging out 1 237*** 1.238*** 1 237*** 1.242*** with friends (1.133 - 1.350) (1.136 - 1.350) (1.133 - 1.350) (1.139 - 1.355) Peer smoking 1 797*** 1 795*** 1.762*** 1 777*** (1.677 - 1.925) (1.673 - 1.925) (1.644 - 1.888) (1.656 - 1.907) Family structure 0.898 (intact vs. non-intact) (0.770 - 1.046) Household income 0.999 (0.998 - 1.001) Parental education 0.938 (0.814 - 1.081) Family members' smoking 1.502*** 1.537*** (1.290 - 1.749) (1.323 - 1.786) Mother's educational 0.977 expectations (0.920 - 1.038) Parental conflicts 1.058 (0.967 - 1.158) 26 Parental control 1.027 (0.981 - 1.076) Parent-child closeness 0.985** 0.983*** (0.971 - 0.998) (0.973 - 0.993) Frequency of parent-child 1.022 having dinner together (0.869 - 1.202) Parent-child communication 0.979 (0.914 - 1.049) N=13,552 Odds rations are presented with 95% confidence intervals in parentheses * significant at 10%; ** significant at 5%; *** significant at 1% 27 Table 3: Logistic Regression of Additional School Characteristics on Adolescent Regular Smoking (1) (2) (3) (4) (5) Regular smoking at wave I 9.767*** 9.763*** 9.484*** 9.543*** 9 527*** (8.324 - 11.459) (8.316 - 11.460) (8.086 - 11.122) (8.134 - 11.195) (8.124 - 11.173) Age 1.006 1.007 1.000 1.012 1.009 (0.961 - 1.052) (0.962 - 1.054) (0.950 - 1.052) (0.966 - 1.059) (0.965 - 1.055) Sex 0.831** 0.829** 0.824** 0.823** 0.825** (0.709 - 0.973) (0.708 - 0.971) (0.702 - 0.968) (0.701 - 0.966) (0.704 - 0.967) Black 0.484*** 0.459*** 0.451*** 0.442*** 0.445*** (0.347 - 0.677) (0.358 - 0.590) (0.350 - 0.581) (0.346 - 0.566) (0.346 - 0.572) American Indian 1.178 1.180 1.142 1.131 1.145 (0.541 - 2.566) (0.531 - 2.624) (0.518 - 2.517) (0.514 - 2.485) (0.513 - 2.558) Asian/Pacific Islander 0.861 0.878 0.906 0.892 0.891 (0.586 - 1.264) (0.642 - 1.201) (0.668 - 1.228) (0.653 - 1.218) (0.656 - 1.212) Other 1.196 1.187 1.216 1.214 1.207 (0.862 - 1.659) (0.856 - 1.646) (0.881 - 1.677) (0.880 - 1.675) (0.874 - 1.666) Hispanic 1.170 1.176 1.155 1.144 1.142 (0.464 - 2.951) (0.472 - 2.931) (0.457 - 2.917) (0.452 - 2.896) (0.453 - 2.879) Average grade of math and English 0.808*** 0.808*** 0.825*** 0.823*** 0.824*** (0.740 - 0.882) (0.739 - 0.882) (0.753 - 0.903) (0.752 - 0.901) (0.754 - 0.902) Earnings per week 1.001** 1.001** 1.001** 1.001** 1.001** (1.000 - 1.003) (1.000 - 1.003) (1.000 - 1.003) (1.000 - 1.003) (1.000 - 1.003) Frequency of hanging out 1 239*** 1 239*** 1 239*** 1.240*** 1.240*** with friends (1.135 - 1.352) (1.135 - 1.352) (1.136 - 1.352) (1.137 - 1.353) (1.136 - 1.352) Peer smoking 1 771*** 1 773*** 1.740*** 1 743*** 1.746*** (1.651 - 1.899) (1.651 - 1.903) (1.623 - 1.865) (1.625 - 1.870) (1.627 - 1.872) Family members' smoking 1.510*** 1.517*** 1.506*** 1.502*** 1.506*** (1.298 - 1.757) (1.306 - 1.762) (1.294 - 1.753) (1.290 - 1.750) (1.295 - 1.751) Parent-child closeness 0.982*** 0.982*** 0.985*** 0.985*** 0.986*** (0.972 - 0.993) (0.972 - 0.992) (0.975 - 0.996) (0.975 - 0.996) (0.975 - 0.996) Suburbana 1.050 (0.850 - 1.297) Rurala 1.001 (0.782 - 1.282) Private schoolb 0.729** 0.689** 0.763* 0.756* 0.746* (0.534 - 0.995) (0.501 - 0.948) (0.560 - 1.039) (0.558 - 1.024) (0.545 - 1.022) Percent of Hispanic students 0.305*** 0.317*** 0.304*** 0.294*** 0.283*** (0.167 - 0.557) (0.177 - 0.569) (0.175 - 0.529) (0.170 - 0.506) (0.163 - 0.491) Percent of Black students 0.842 (0.546 - 1.298) Percent of Asian students 1.088 (0.466 - 2.536) Percent of students who expect to go to college 1.459 (0.506 - 4.208) School attendance rate 1.009 (0.923 - 1.103) People get along in schoolc 0.987 (0.936 - 1.041) Self-perceived fitting at schoold 0.978*** 0.977*** 0.978*** (0.963 - 0.993) (0.962 - 0.993) (0.962 - 0.993) Percent of students smoking 1.398 28 (0.640 - 3.051) Percent of teachers smoking 1.003 (0.992 - 1.014) Policy against teacher smoking 0.932 (0.794 - 1.094) Policy against student smoking 1.017 (0.632 - 1.639) Severe punishment for smoking 1.035 (0.958 - 1.118) N=13,552 Odds ratios are presented with 95% confidence intervals in parentheses * significant at 10%; ** significant at 5%; *** significant at 1% a. urban=reference group b. public school=reference group c. An average score for each school on how well students at the school feel they get along with their teachers and other students at school. d. An average score for each school on the extent to which students feel they fit in at school. 29 Table 4: Logistic Regression of Additional Neighborhood Characteristics on Adolescent Regular Smoking (1) (2) (3) Regular smoking at Wave I 9.509*** 9 577*** 9.577*** (8.108 - 11.152) (8.177 - 11.217) (8.180 - 11.214) Age 1.008 1.009 1.009 (0.964 - 1.055) (0.965 - 1.055) (0.965 - 1.056) Sex 0.825** 0.833** 0.829** (0.704 - 0.967) (0.709 - 0.978) (0.706 - 0.973) Black 0.458*** 0.452*** 0.445*** (0.357 - 0.589) (0.351 - 0.582) (0.346 - 0.573) American Indian 1.162 1.150 1.148 (0.513 - 2.633) (0.515 - 2.566) (0.513 - 2.571) Asian/Pacific Islander 0.875 0.887 0.881 (0.640 - 1.196) (0.654 - 1.202) (0.648 - 1.198) Other 1.214 1.205 1.202 (0.882 - 1.671) (0.872 - 1.664) (0.871 - 1.659) Hispanic 1.142 1.154 1.137 (0.453 - 2.881) (0.470 - 2.837) (0.455 - 2.845) Average grade of math and English 0.823*** 0.827*** 0.826*** (0.753 - 0.900) (0.756 - 0.904) (0.755 - 0.904) Earnings per week 1.001** 1.002** 1.001** (1.000 - 1.003) (1.000 - 1.003) (1.000 - 1.003) Frequency of hanging out with friends 1.238*** 1.244*** 1.242*** (1.135 - 1.350) (1.140 - 1.356) (1.138 - 1.354) Peer smoking 1.750*** 1.746*** 1.744*** (1.631 - 1.878) (1.628 - 1.872) (1.626 - 1.870) Family members' smoking 1.520*** 1.506*** 1.503*** (1.303 - 1.774) (1.296 - 1.751) (1.293 - 1.747) Parent-child closeness 0.986*** 0.985*** 0.986** (0.975 - 0.996) (0.975 - 0.996) (0.976 - 0.997) Private schoola 0.732** 0.746* 0.746* (0.538 - 0.995) (0.544 - 1.024) (0.545 - 1.022) Percent of Hispanic students 0.284*** 0.284*** 0.280*** (0.164 - 0.490) (0.163 - 0.496) (0.161 - 0.490) Self-perceived fitting at schoolb 0.977*** 0.978*** 0.980** (0.962 - 0.993) (0.962 - 0.994) (0.965 - 0.996) Neighborhood median household incomec 1.000 (1.000 - 1.000) Percent of residents age 25 or above with college degreec 0.888 (0.332 - 2.379) Neighbors watch out for each otherd 0.832* 0.854* (0.685 - 1.010) (0.709 - 1.028) Happy to live in the neighborhoodd 1.051 (0.981 - 1.126) Neighborhood recreational facilitye 0.889 (0.751 - 1.052) N=13,552 Odds ratios are presented with 95% confidence intervals in parentheses * significant at 10%; ** significant at 5%; *** significant at 1% a. public school=reference group b. An average score for each school on the extent to which students feel they fit in at school. c. Measures of neighborhood socioeconomic status d. Measures of neighborhood social capital e. A measure of neighborhood recreational facility as a proxy of neighborhood physical condition 30 Table 5: Logistic Regression of Additional State Characteristics on Adolescent Regular Smoking (1) (2) (3) (4) Percent of smokers in the state 1.003 (0.976 - 1.030) Policy against cigarette vending 0.992 machines accessible to minors (0.870 - 1.132) No marketing restrictions for tobacco 0.981 (0.851 - 1.130) State program for dissemination of 1.033 information about tobacco (0.893 - 1.194) N=13,552 Odds ratios are presented with 95% confidence intervals in parentheses * significant at 10%; ** significant at 5%; *** significant at 1% a. All the models controlled for individual, peer, family, school, and neighborhood variables that are included in Model 4.3 (Table 4, Model 3). Model 4.3 is used as the baseline model for models that include variables at the state level and are presented in Table 5. 31 Appendix: Description of Measures Individual-level variables: Regular smoking at Wave I & II: "Have you ever smoked cigarettes regularly, that is, at least 1 cigarette every day for 30 days?" (yes/no) Race/Ethnicity was captured by five racial/ethnic groups: Non-Hispanic whites, Non-Hispanic blacks, Hispanics, American Indian, Asian/Pacific Islander, and Others Physical Maturity: "How advanced is your physical development compared to other boys or girls your age?" (5 levels) Self-Perceived Intelligence: "Compared with other people your age, how intelligent are you?" (6 levels) Self-Esteem was evaluated by the individual's agreement to the statement "You have a lot of good qualities." (5 levels) Frequency of hanging out with friends: "During the past week, how many times did you just hang out with friends?" (0 to 3) Peer smoking: "Of your 3 best friends, how many smoke at least 1 cigarette a day?" (0-3) Parent- or family-level variables: Family member smoking: "Are there any cigarette smokers in your household?" (yes/no) Mother's expectation for adolescent of college graduation: "On a scale of 1 to 5, where 1 is low and 5 is high, how disappointed would your mom be if you did not graduate from college?" (5 levels) Parent's level of education, as reported by parent on the parent questionnaire, is regrouped into five categories: "Less than high school," "High school or equivalent," "Some college," "College graduate," and "Post graduate." Parental conflict: "How much do you fight or argue with your current (spouse/partner)?" (4 levels) Parental control: "Do your parents let you make your own decisions about the time you must be home on weekend nights, the people you hang around with, what you wear, how much television you watch, which television programs you watch, what time you go to bed on week nights, and what you eat?" This variable is a sum of the number of decisions the adolescent's parents let them make about their daily lives. A higher score indicates more decisions being made by the adolescent. The coefficient of alpha was 0.60 32 Parent-child closeness is a summed score combining the adolescent's perception of how close they are to their parents, as well as how much they perceive their parents care about them. The coefficient of alpha was 0.85. Parental involvement in the adolescent's life was captured by an item capturing how many times per week the adolescent eats dinner with parents. "On how many of the past 7 days was at least one of your parents in the room with you while you ate your evening meal?" (0-7) We then constructed a dichotomized variable to indicate if the adolescent eats dinner with the parents for 5 days or more as a proxy for parental involvement in the adolescent's life. Other parental involvement variables are not reliable as they have too many missing values. Parent-child communication is measured by a sum score of responses to several items asking whether the adolescent talked with the mother or the father about someone s/he is dating, or a party s/he went to, about a personal problem you were having, about other things s/he was doing in school. The coefficient alpha was 0.61. Household income: "About how much total income, before taxes did your family receive in 1994? Include your own income, the income of everyone else in your household, and income from welfare benefits, dividends, and all other sources." Family Structure was captured by a dichotomized indicator of intact family structure versus non-intact family structure. School-level variables (perceived or actual): School average academic aspiration was measured by percent of students who expect to go to college. Perceived school conflict was measured by the mean score of students' responses to the item "Since school started this year, how often have you had trouble getting along with other students?" Self-perceived fitting at school was captured by a scale that comprises 7 items. The items ask the adolescent whether she or he gets along with the teachers, gets along with other students, feel close to people at school, feel like she or he is part of school, is happy to be at school, believes that the teachers at school treat students fairly, and feel safe at school. The coefficient of alpha was 0.77. Percent of students smoking was measured by the mean score of students' responses to the item "During the past twelve months, how often did you smoke cigarettes?" Percent of teachers smoking was measured by the school administrator's estimate of the percentage of the full-time classroom teachers smoking at school. Policy against teacher smoking (yes/no) 33 Policy against student smoking (yes/no) Severe punishment for smoking was indicated by discipline policy for student smoking at school on 1st offense that range from verbal warning to expulsion. Neighborhood-level variables (perceived or actual): Neighborhood social capital was captured by the item "People in this neighborhood look out for each other" (yes/no) and the item "On the whole, how happy are you with living in your neighborhood?" (5 levels). Neighborhood physical condition was tapped by the item "Do you use a physical fitness or recreation center in your neighborhood?" (yes/no) State-level variables: Policy against cigarette vending machines where minors may access them (yes/no) Policy against marketing for tobacco (yes/no) State has a program for dissemination of information about tobacco (yes/no) 34 1 Based on data from the in-school social network file, we also constructed prevalence rates of smoking among all friends who either nominated the adolescent as a friend or are nominated by the adolescent. We then tested this variable to examine the effect of peer smoking. The results are similar to those using the number of smokers among three best friends. We chose not to use this variable in our final analysis because more than 30 percent of the values needed to be imputed. 2 Totally there were four variables that have 10% or higher missing values for which we had to impute. They are people's getting along at school (28%), parental conflict (14%), parental education (13%), and household income (23%). 3 The coefficient of earnings per week in Model 4.3 (Table 4, Model 3) is ln (1.001) equal 0.001 for $1. For $100, the coefficient would be 0.1. The odds ratio is thus e(0.1)=1.11. So every $100 more earnings per week are associated with 11% higher risk of smoking. 4 The coefficient of percent of Hispanic students in Model 4.3 (Table 4, Model 3) is ln (0.28) equal -1.27 for 100 percent. For 20 percent, the coefficient would be a fifth of -1.27 equal -0.25. The odds ratio is thus 1/e(-0.25)=1/0.78=1.28. So 20% higher in Hispanic population is associated with 28% lower risk. 35