In press Social Problems PEER CONTEXT AND THE CONSEQUENCES OF ADOLESCENT DRINKING Robert Crosnoe Chandra Muller Department of Sociology and Population Research Center University of Texas at Austin Kenneth Frank Department of Counseling, Educational Psychology, and Special Education Michigan State University Running Head: Adolescent Drinking Direct correspondence to the first author at Department of Sociology and Population Research Center, University of Texas at Austin, 1 University Station A1700, Austin, TX 78712-1088 (crosnoe@mail.la.utexas.edu). The authors acknowledge the support of the Spencer Foundation and grants from the National Institute of Child Health and Human Development (R01 HD40428-02, PI: Chandra Muller) and the National Science Foundation (REC-0126167, Co-PI: Chandra Muller and Pedro Reyes) to the Population Adolescent Drinking Research Center, University of Texas at Austin. Opinions reflect those of the author and not necessarily those of the granting agencies. This research used data from Add Health, a program project designed by J. Richard Udry, Peter S. Bearman, and Kathleen Mullan Harris and funded by a grant P01-HD31921 from the National Institute of Child Health and Human Development, with cooperative funding from 17 other agencies. Special acknowledgment is due Ronald R. Rindfuss and Barbara Entwisle for assistance in the original design of Add Health. Persons interested in obtaining data files from Add Health should contact Add Health, Carolina Population Center, 123 W. Franklin Street, Chapel Hill, NC 27516-2524 (www.cpc.unc.edu/addhealth/contract.html). The authors would like to thank Roz King for helpful comments on an earlier version of this manuscript. 2 Adolescent Drinking Peer Context and the Consequences of Adolescent Drinking ABSTRACT Research has focused extensively on adolescent drinking, particularly on peer influences and other predictors of this behavior. This study shifts the focus by examining how drinking is associated with other individual trajectories during adolescence and how these associations vary by peer context. With the National Longitudinal Study of Adolescent Health, we found that adolescent engagement in drinking predicted declining academic achievement and escalating emotional distress. These associations, however, varied by level of drinking in the peer context. In general, person-context mismatches exacerbated the risk of drinking or weakened the protective nature of not drinking. Both the micro-level and macro-level peer context (e.g., friends, school drinking rates) were important for achievement, while the macro-level context was important for emotional distress. This research suggests new ways of thinking about the role of peers in adolescent problem behavior. 3 Adolescent Drinking Peer Context and the Consequences of Adolescent Drinking INTRODUCTION [Heading Level 1] Recent research has documented the long-term health benefits of moderate alcohol intake (Mukamal et al. 2003), but these benefits in later life are countered by two more problematic dynamics. First, through multiple mechanisms, alcohol use can have severe consequences for family and community life, economic productivity, and public expenditure. Second, during the early life course, the numerous social and psychological risks of alcohol use outweigh its eventual health benefits (Elliot, Huizinga, and Ageton 1985; Hawkins, Catalano, and Miller 1992; Schulenberg and Maggs 2002). The intense empirical focus on the etiology of adolescent alcohol use, especially peer influences, is related to both dynamics. Given the potential harm of adolescent drinking and given that adolescence is the foundation for long-term trajectories, understanding the etiology of adolescent drinking points us to methods for addressing this complex phenomenon (Chassin, Pitts, and DeLucia 1999; Johnston, O’Malley, and Bachman 1996; Schulenberg and Maggs 2002). This etiological focus can be enhanced, however, by broader consideration of the role of drinking in adolescent society. This study attempts such an investigation by approaching some old problems in new ways. It does so by applying key concepts of the life course paradigm to experiences within a specific stage of the life course. First, recognizing the overlap among different life course trajectories, this study examines how drinking is related to other primary markers of adolescent functioning. Specifically, is adolescent drinking associated with declining academic performance and escalating emotional distress? Like 4 Adolescent Drinking past studies, therefore, this study is concerned with the connections between drinking and other domains, but, unlike most of the literature, it posits drinking as predictor rather than outcome. Second, recognizing the context-specific nature of the overlap among life course trajectories, this study examines whether the potential consequences of adolescent drinking vary by peer context. Specifically, are the consequences of drinking greater in cases of mismatch between adolescents and their peer contexts (e.g., drinking in contexts characterized by low rates of drinking)? Thus, like past research, we consider the importance of peers, but, unlike much of this literature, we posit the peer context as a moderator rather than a predictor. The potential significance of this study is both conceptual and methodological. Conceptually, we view drinking as one thread in a tapestry of personal attributes and as a phenomenon that varies qualitatively across social settings. Such an approach promotes understanding of why drinking is problematic for the individual life course and the larger society and offers a more nuanced picture of adolescent social life. Methodologically, this study takes a multi-level view of peer context—micro (e.g., close friends) and macro (e.g., school culture)—and draws on the National Longitudinal Study of Adolescent Health (Add Health). This approach allows us to compare two common conceptualizations of the peer group, use the reports of adolescents and their friends, address selection issues, and generalize to a national context. Alcohol Use During Adolescence [Heading Level 2] As discussed above, adolescent alcohol use is viewed as a major societal problem, and, as such, it has generated a great deal of attention in both public and scholarly arenas (Hawkins et al. 1992). Two general themes have emerged from this rich empirical 5 Adolescent Drinking activity. First, drinking is especially risky and more indicative of general adjustment problems during this stage because adolescents are psychologically less equipped to drink responsibly than adults (Newcomb and Bentler 1989; Schulenberg et al. 1999). Second, during adolescence, drinking is a largely social phenomenon. In this stage of life, drinking is less stigmatized than other forms of substance use and may be a high-status activity in some settings. Consequently, adolescents may drink as a way of integrating themselves into groups and gaining status (Crosnoe 2002; Schulenberg et al. 1999). This study integrates these two general themes. Although our focus is on one stage of the life course, we draw on the basic concepts of the life course paradigm to craft our approach. Adolescent Drinking and Other Domains of Functioning [Heading Level 2] According to the life course paradigm (Elder 1998), individual lives can be viewed as consisting of multiple, intertwined trajectories in various domains of adjustment and functioning. A key challenge of life course research, therefore, is to examine how different trajectories are intertwined, or connected to each other. For example, understanding alcohol use requires the examination of how it overlaps with other individual attributes and behaviors. In general, past research examining the overlap among domains of adolescent functioning has treated drinking as an outcome. For example, adolescents are more likely to use alcohol or other substances when they have problems in personal relationships, struggle in school, engage in delinquent acts, or suffer emotional problems (Jessor, Donovan, and Costa 1991; Keefe and Newcomb 1996; Mensch 1988; Schulenberg et al. 1994). Less often has research in this area reversed the focus, examining adolescent drinking as a predictor. Studies with this focus have revealed, as expected, that 6 Adolescent Drinking adolescent drinking is a risk factor that increases the likelihood of negative outcomes during adolescence and later stages of the life course (Chassin et al. 1999; Galambos and Silbereisen 1987; Garmezy and Masten 1986; Schulenberg and Maggs 2002). The present study takes this less common avenue, examining the potential consequences of adolescent drinking for two key domains of adolescent functioning. First, academic achievement, a marker of success in a conventional institution, is the foundation of long-term status attainment. Adolescent drinking may impair academic performance by disrupting cognitive functioning, downgrading the importance of academic responsibilities, harming relationships with adults (e.g., teachers, parents) that influence academic performance, and exposing young people to non-conventional norms of behavior (Crosnoe 2002; Jessor et al. 1991). Second, emotional distress is a marker of mental health and is also related to socioemotional adaptation to both intimate and institutional settings in social life. Adolescent drinking may contribute to emotional distress by generating conflict in the family, exposing adolescents to stressful situations, and impairing coping skills (Hussong et al. 2001; Jessor et al. 1991). Following this, the first goal of this study is to examine whether adolescent drinking lowers academic achievement and increases emotional distress. In doing so, we apply a longitudinal framework that addresses the potential bi-directionality of the associations between alcohol use and these other domains, and we draw on a source of data that allows these results to be generalized to a national context. Drinking and the Peer World [Heading Level 2] In addition to pointing research towards the overlap among various domains of adolescent functioning, the life course paradigm also emphasizes that such overlap may 7 Adolescent Drinking vary in magnitude and meaning across social contexts (Elder 1998). In other words, primary social contexts influence adolescent functioning directly but also moderate the associations among various individual characteristics. As described above, the peer context is one of the most commonly studied social contexts in the literature on adolescent drinking (Costa, Jessor, and Turbin 1999). In general, such research has focused heavily on the direct influence of peers, reporting a strong association between peer drinking and adolescent drinking even after taking selection forces into account (Aseltine 1995; Kandel 1996; Schulenberg et al. 1999). The intertwining paths of the life course approach just outlined, however, suggest a reconceptualization of the role of peers that involves moderation rather than direct influence. Following this, the second goal of this study is to examine whether the longitudinal associations between adolescent drinking and two domains of functioning (academic achievement, emotional distress) vary by level of drinking in the peer context. By taking this approach, we are essentially examining the implications of a person-context mismatch. A great deal of research suggests that individual functioning is hampered by such mismatches. For example, student learning is compromised when the characteristics of students do not align with the structure of educational institutions, mental health is compromised when the physical attributes of adults are undervalued in their primary groups, and the socioemotional functioning of young people is impaired when individuals’ demographic profiles are underrepresented in their immediate social settings. Such mismatches, which, by definition, identify non-normative circumstances, increase the likelihood of being stigmatized and of negative self-evaluations and decrease perceptions of safety, security, and belongingness (Eccles et al. 1993; Johnson, Crosnoe, 8 Adolescent Drinking and Elder 2001; Ross 1994). Thus, if an individual has some potentially problematic attribute or engages in some potentially problematic behavior, the negative implications of this characteristic or behavior may be compounded if that characteristic is also uncommon in their primary groups. In this study, the person-context mismatch involves drinkers in peer contexts with low rates of drinking and non-drinkers in peer contexts with high rates of drinking. In the former case, we expect that the potential negative consequences of adolescent drinking will be greater because, in addition to drinking, adolescents will also be out of step with the normative structure of their primary social setting. In the latter case, we expect that the potential protection of avoiding risk behaviors will be weaker because such students could be marginalized within their primary social setting. In pursuing this goal, we bridge different studies and disciplines by conceptualizing the peer context on two levels. The micro-level context concerns adolescents’ inner-circles of friends, the other young people with whom they enjoy personal connections and sustained interaction. The significance of this level concerns direct behavioral modeling and the power that comes with emotional involvement (Crosnoe 2000; Hartup and Stevens 1997). To gauge peer dynamics on this level, we measure the rate of drinking among close friends. In contrast to these more personal relationships, the macro-level context encompasses the larger social world of the high school—the normative and behavioral climate constructed and supported by other students in the school (Brown 1990; Cleveland and Weibe 2003). To capture this level, we measure the level of drinking in the school. Thus, our treatment of person-context mismatch concerns how well students “fit” with their friendship groups and the larger 9 Adolescent Drinking peer culture of their schools and the power of such mismatches to condition the developmental significance of drinking. METHODS [Heading Level 1] Sample [Heading Level 2] This study drew on Add Health, an ongoing nationally representative study of young people (7-12th graders in 1994). Add Health was created with a stratified sampling design beginning with a representative sample of schools. Using the Quality Education Database, an exhaustive list of American high schools, as the sampling frame, 80 high schools were stratified by region, urbanicity (defined by U.S. Census Bureau), sector (e.g., private, Catholic, public), racial composition, and size and then selected randomly within strata. Attempts were made to match each of these high schools to one of their feeder schools, typically a middle school. Feeder schools were selected with a probability based on their student contribution to the high school, so that the feeder school that was the origin school of the largest proportion of the high school student body was most likely to be selected as a study school. Because some high schools were comprehensive (e.g., contained 7 –12th grades), they were their own feeder schools and were not matched to an external feeder school. The final sample included 132 high schools and middle schools from the 80 original strata (see Bearman, Jones, and Udry 1997 for more on the Add Health sampling process). From September 1994 through April 1995, all available students in these 132 schools (n = 90,118) completed the brief, paper and pencil In-School Survey, designed primarily as a guide for the later In-Home Interviews and to identify possible subjects for 10 Adolescent Drinking planned oversamples. A representative sample (n = 20,745 in 132 schools) of this InSchool population, selected evenly across high school – feeder school pairs, then participated in the In-Home Interview at Wave I (April – December 1995). Wave I, which included a much broader battery of items as well as information from parents, school administrators, and the Census, is the core Add Health data set. The Add Health team attempted to follow up this Wave I sample again from April through August of 1996, although they chose not to follow up the portion of the sample that had graduated from school after Wave I. Thus, Wave I seniors, as a group, did not participate in Wave II. With this exclusion, plus the expected attrition, 14,736 adolescents from Wave I also participated in the Wave II In-Home Interview (Bearman et al. 1997). The sample for the present study was created by imposing four filters on the InHome sample. Below, we discuss the motivation for each of these selection filters in detail and also compare the Wave I sample members who were and were not targeted for exclusion based on the independent application of each selection filter on key study variables. We then present descriptive statistics that demonstrate how the cumulative, as opposed to independent, application of these often overlapping selection filters altered the composition of the study sample. First, our longitudinal design required that adolescents participate in both waves of the In-Home Interview (71% of the original Wave I In-Home sample). This filter eliminated all Wave I seniors as well as any adolescent who dropped out of the sample after initial participation. Because the adolescents who participated in Waves I and II differed on demographic characteristics (e.g., older, male, more likely to be White, higher parent education) and on the developmental markers of interest in this study (e.g., higher 11 Adolescent Drinking achievement, lower emotional distress and alcohol use) compared to the adolescents who participated in Wave I only, the application of this selection filter biased the sample. Second, we only included adolescents who had a valid sampling weight (92% of the Wave I sample). Because Add Health oversampled some groups (e.g., twins, disabled adolescents), sampling weights, which adjust the sample for this unequal probability of selection, are required to make the raw data nationally representative. Different weights, which have been made available with the Add Health data, are necessary for different combinations of data. Because of our longitudinal framework, we have used the longitudinal sampling weight. A small portion of the In-Home sample was not assigned this weight because they attended one of two schools for which weights could not be calculated or because they were late additions to the special oversamples (refer to Chantala and Tabor 1999 for more on weighting in Add Health). Our analysis of the group of excluded adolescents, both for this study and for our prior research with Add Health, revealed that their exclusion does not substantially alter sample characteristics. Third, because of the need to use parent-reported data to create some measures, we included only the adolescents who had at least one parent interviewed at Wave I (85% of the Wave I sample). The adolescents excluded for this filter did come from more advantaged backgrounds, as might be expected. They also had lower academic achievement and higher emotional distress and alcohol use. Fourth, the Add Health peer network data was a vital resource for this study. Consequently, we included all adolescents for whom we could construct valid network alcohol measures, as described above (70% of the Wave I sample). Of the Wave I adolescents targeted for exclusion by this filter, over four-fifths did not nominate any 12 Adolescent Drinking friends with most of the remainder nominating friends outside the Add Health sample. This selection filter introduced the most bias to our study sample. These excluded adolescents, who were generally social isolates or who were part of out-of-school networks, differed in expected ways from the adolescents who had valid network data with one important exception. The two groups did not differ on alcohol use. The application of these filters resulted in a study sample of 7,758 adolescents in 122 schools (out of 20,745 adolescents in 132 schools in Wave I). As we have made clear above, this study sample did not have the same characteristics as the original Wave I sample. To characterize the cumulative bias introduced by the systematic application of these filters, Table 1 includes descriptive statistics for each stage of the selection process. Because the samples created by the addition of each new filter were not mutually exclusive, we could not calculate whether differences among them were statistically significant, but the general trends in this table are informative. The selection process did bias the study sample towards greater social advantage and social adjustment. This bias must be remembered in the interpretation of results, but we argue that these selection criteria were necessary and that this bias was balanced by the benefits of using multilevel, multi-context data from a national sample of adolescents and their friends. [Table 1 About Here] Measures [Heading Level 2] The two dependent variables drew on information from the Wave II In-Home Interview, the other individual-level variables on information from the Wave I In-Home Interview (adolescent or parent), and the school-level variables on the reports of school administrators (at the same time as the Wave I In-Home Interview) or the aggregation of 13 Adolescent Drinking responses from the In-School Survey (which occurred a year before the Wave I In-Home Interview). Descriptive statistics for all study variables can be found in the Appendix. Academic Achievement [Heading Level 3] In both Waves I and II, adolescents reported their grades in four academic subjects (math, science, English, and social studies) in the past year. These responses, ranging from 1 (A) to 4 (D or F), were averaged and then converted to a standard four point grade point scale for each year. This composite was somewhat skewed, with a third of the sample reporting a grade point average of 3.0 or better. Adolescent Emotional Distress [Heading Level 3] The Center for Epidemiologic Studies-Depression (CES-D) scale was used to measure distress. This scale, which was created to gauge the presence and extent of depressive symptomatology (e.g., depressed affect, feelings of guilt, hopelessness, helplessness, and worthlessness), correlates with other distress scales, is high in construct validity, and has demonstrated reliability across multiple populations (Radloff and Locke 1986). Add Health contained a modified version of the CES-D scale, including 15 of the 20 original items (Resnick et al. 1997). Adolescents were asked how often they had felt certain things during the past week. Examples included “You felt that you could not shake off the blues, even with help from your family and your friends” and “You felt lonely”. Responses ranged from 0 (never or rarely) to 3 (most or all of the time). The Alcohol Use [Heading Level 3] Adolescents reported how often they drank alcohol in the past year, with responses collapsed into a scale ranging from 0 (none) to 7 (everyday or almost 14 Adolescent Drinking everyday). As reported in past research (Chassin et al. 1999), the significance of alcohol use often takes a non-linear form, with some increases in alcohol use (e.g., from none to some, crossing the threshold to frequent use) more important than others. Our analyses revealed the same non-linear pattern. As a result, this continuous scale of alcohol use would have obscured true effects. To avoid this problem, we could have used a quadratic or logged term for alcohol use or dummy variables designating different levels of drinking. Our analytical checks revealed the same pattern of results for all three approaches, increasing our confidence in our findings on the non-linear effects of alcohol use on different adolescent outcomes across different peer contexts. Of the three, we chose the categorical approach and, following past research, broke the alcohol use scale into three dummy variables. Non-drinkers, the reference category, did not drink alcohol in the last year. Occasional drinkers drank alcohol more than once but less than once a month in the past year. Frequent drinkers drank alcohol two or three times a month or more. We preferred the categorical approach because these designations of alcohol use are more accessible to general audiences than the quadratic or logged approaches. We should reiterate, however, that all three approaches produced the same results. Friends’ Alcohol Use [Heading Level 3] Each adolescent was asked to list up to five female and five male friends. If a listed friend was also in Add Health, then we used information from the In-School Survey on that friend’s alcohol use. Thus, this peer measure does not refer to a respondents' estimation of the level of alcohol use among friends, but rather to the mean of the self-reported alcohol use (0 = none in past year, 7 = everyday or almost everyday in past year) of all listed friends in the sample. As discussed above, using friends’ reports 15 Adolescent Drinking does bias the sample in terms of average characteristics and in meaning. Because of this application of the peer selection filter, this study essentially focuses on the peer contexts of adolescents with peers at school and cannot make conclusions about adolescents who were isolated from peers. We argue, however, that this less representative sample is still informative. First, it helped us to attend to students, and student functioning, across an array of social contexts. Second, the use of the network data, based on the direct reports of friends, had certain advantages. This method reduced the error of ego-based measures on peer characteristics, which might be inaccurate because adolescents do not have full information on their friends’ behavior or because they overestimate the extent to which their friends are similar to them (Kandel 1996; Wilcox and Udry 1986). Of course, perceptions are often more important than reality in guiding our behavior, and so these ego-based measures are certainly useful for the study of peer dynamics. Network data, however, allows for a more accurate description of the use of alcohol in the peer context, which, beyond perceived alcohol use, helped us to tap the opportunity structure for adolescents’ own drinking. School Level of Drinking [Heading Level 3] In the original In-School Survey, all adolescents in each school reported their level of alcohol use (0 = never in the past year, 6 = nearly everyday in past year). The mean of all responses within a school was the average level of drinking in that school. Individual-Level Controls [Heading Level 3] Analyses controlled for five demographic factors: gender (1 = female), race (African-American, Hispanic-American, Asian-American, and Other, with non-Hispanic White as the reference category), parent education (self-reported years of education for 16 Adolescent Drinking mother and father, averaged), family structure (1 = two biological parents living together with adolescent, 0 = other), and age (in years). We also controlled for number of friends in order to gauge the differences between having few or many friends who drank (regardless of level) and for alcohol use of parents (parent report of self and partner use, 0 = never in past year, 6 = nearly everyday, slightly different from student measure), an important predictor of adolescent drinking and problem behavior (Chassin et al. 1999). School-Level Controls [Heading Level 3] School-level analyses controlled for four school-level factors: average parent education in the school (based on the school mean of parent education, as reported by the adolescent), minority representation (percent of student body that was non-White), school size (based on the total enrollment of the school given by school administrator, logged), and school level (1 = middle school, 0 = high school or comprehensive school). Plan of Analysis [Heading Level 2] Our primary goals of this study were to examine the consequences of adolescent drinking and whether these consequences varied by peer setting, and our analytical strategy to pursue these goals consisted of three steps for each outcome. First, we estimated a base model in which the outcome of interest was regressed on all of the individual-level controls, a prior measure of the outcome, and the alcohol use dummy variables. This model gauged the degree to which level of alcohol use was associated with a change in the outcome over one year. Second, we added to this base model the measure for the rate of drinking among friends and a set of interaction terms between the adolescent alcohol use dummy variables and the measure for friends’ drinking. This model gauged the degree to which drinking among friends moderated the associations of 17 Adolescent Drinking adolescent drinking with the two outcomes. Third, we added to the base model the school-level controls, the measure for the rate of drinking in school, and interaction terms for the adolescent alcohol use dummy variables with the school measure. This model gauged the degree to which the school rate of drinking moderated the associations of adolescent drinking with the two outcomes. Add Health data collection was school-based, which means that the adolescents sampled were not statistically independent observations (e.g., students were likely to be more similar to those in their school than to those in other schools). Failing to account for this within-school clustering has negative implications for statistical inference. To correct for this problem, we included random effects for schools in our models with the mixed procedure in SAS, a restricted form of multi-level modeling (Bryk and Raudenbush 1992; Singer 1998). This procedure corrected the design effects of Add Health. It also allowed us to distinguish the variation in the outcomes that occurred among students within a school from the variation that occurred among students in different schools, the estimation of school effects on individual behavior, and the estimation of cross-level interactions, all of which were important aspects of this study. RESULTS [Heading Level 1] Three Types of Adolescent Drinkers [Heading Level 2] Past research suggests qualitative differences between non-drinkers, occasional drinkers, and frequent drinkers (Chassin et al. 1999), and preliminary analyses support this suggestion (Table 2). Beginning with demographic characteristics, frequent drinkers were older and more likely to be male than other adolescents. The three groups of 18 Adolescent Drinking adolescents did not differ in parent education (an indicator of socioeconomic status), but the non-drinkers were less likely to be White than others. Next, descriptive statistics also reveal other key differences. Compared to the other groups, non-drinkers had friends who drank the least, while frequent drinkers had friends who drank the most. Frequent drinkers attended schools in which drinking was most common among students. Moreover, academic achievement decreased at each step between non-drinking and frequent drinking, while emotional distress increased with each step. [Table 2 About Here] Thus, in general, drinkers had different demographic profiles than non-drinkers, were part of peer contexts in which drinking was more common, and had more problematic patterns of functioning than drinkers. Among drinkers, however, different levels of drinking also entailed different characteristics, lifestyles, and environments. To investigate these descriptive patterns more closely, we turned to multivariate analysis. Recall that our two general goals were to examine the potential negative consequences of adolescent drinking and to investigate whether these consequences were context-specific. In the following sections, we pursue these two goals for two separate outcomes: academic achievement and emotional distress. Adolescent Drinking and Academic Trajectories [Heading Level 2] To begin, Model 1 in Table 3 presents the results relevant to the question of whether adolescent drinking was associated with declining academic achievement. Controlling for level of academic achievement in Wave I, occasional and frequent drinking predicted lower achievement in Wave II (b = -.06, p < .001 for occasional; -.07, p < .001 for frequent). Switching the reference category for the alcohol use dummy 19 Adolescent Drinking variables revealed that the associations with achievement of occasional and frequent drinking did not differ significantly from each other. We should also note here that parents’ alcohol use also did not predict Wave II achievement. [Table 3 About Here] Next, Model 2 presents results relevant to the question of whether this association between drinking and achievement varied by peer context, with the friendship group as the marker of peer context. Such variation did occur, but only for frequent drinking. The interaction between friends’ drinking and occasional alcohol use was not significant, indicating that the decline in academic achievement associated with moderate drinking was the same whether the adolescent had friends who drank or not. Among frequent drinkers, however, this decline did vary by the level of drinking among friends (b = -.04, p < .05). To interpret this interaction, we wrote out multiple equations—varying the values of frequent alcohol use (1, 0) and friends’ alcohol use (one standard deviation above and below the mean) while assigning all other variables to their sample means. Doing so revealed that the academic achievement of frequent drinkers, but not nondrinkers, varied by the level of drinking among their close friends. While non-drinkers generally had higher achievement (predicted g.p.a. = 2.82), the frequent drinkers who were members of friendship groups high in drinking had achievement closer to the nondrinkers (2.77) than did the frequent drinkers who were members of friendship groups low in drinking (2.68). Finally, Model 3 used the student body as a whole, and not just the friendship group, as the marker of the peer context. In this case, the peer context measure interacted with occasional drinking and not frequent drinking (b = .13, p < .01). Writing out the 20 Adolescent Drinking equations for this interaction term as described above, revealed that both non-drinkers and occasional drinkers drove this interaction. Non-drinkers did slightly better academically in schools where drinking was less common overall (estimated g.p.a. = 2.78) than in schools where it was more common (2.72), while occasional drinkers did slightly worse in schools in which drinking was less common (2.67) than in schools in which it was more common (2.70). We also performed additional analyses of these three models in which we controlled for adolescents’ emotional distress and friends’ academic achievement and in which we combined the friendship group and school models into a single comprehensive model. These additional analyses, not presented here, generated no different inferences. Thus, in terms of changes in academic performance over time, the key distinction seems to have been between drinking and not drinking. Those who drank, no matter what the level, experienced declines in their achievement level across a one-year period, although the context of their alcohol use moderated its academic risk status. In general, occasional and frequent drinking coincided with academic decline, but this decline was slightly worse (in other words, the risk was greater) among those who drank on a regular basis outside of a peer context, either the friendship group or the student body as a whole, engaged in drinking. This finding suggests that person-context mismatches add to the negative consequences of adolescent drinking. Of course, these associations were not great in magnitude. For example, the association with achievement for both levels of drinking represented only about 10% of a standard deviation in achievement, or the difference of a tenth of a letter grade. Still, what we measured here was not an association between adolescent drinking and 21 Adolescent Drinking academic achievement but an association between drinking and a one-year change in achievement. The persistence of these associations despite this more conservative framework and the control for a host of individual and social factors suggests that, while modest, they are meaningful. The fact that we have controlled for previous grades implies that peer dynamics affect change in behaviors, and, therefore, these small effects can accumulate over time. Adolescent Drinking and Emotional Distress [Heading Level 2] To adapt our two general research goals to the second adolescent outcome, we asked whether increasing alcohol use was associated with escalating emotional distress and whether this association varied by peer context. Table 4 presents the results of three multi-level models relevant to these questions. [Table 4 About Here] Model 1 in Table 4 indicates that drinking in Wave I was a risk factor for emotional distress in Wave II (b = .04, p < .001 for occasional; .05, p < .001 for frequent). Switching the reference category for the alcohol dummy variables revealed no significant differences between the two categories of drinkers in emotional distress. Model 1 in Table 4 also reveals that drinking among parents was not significantly associated with changes in emotional distress over time. Model 2 in Table 4 includes interactions between the alcohol use dummy variables and the friendship group markers of the peer context. Unlike for academic achievement, the rate of drinking among close friends did not moderate the associations between adolescent drinking and emotional distress. Model 3 in Table 4 replaces the friendship group marker and interactions with the school marker of peer context. Level 22 Adolescent Drinking of drinking interacted significantly with the average rate of alcohol use in the student body (b = -.05, p < .05 for occasional; b = -.08, p < .01 for frequent). These interactions did not change when controlling for friends’ level of emotional distress and adolescents’ own academic achievement or when estimated simultaneously with the friendship group markers and interactions (not shown). To interpret these interaction terms, we again wrote out multiple equations— varying the values of occasional and frequent alcohol use (1, 0) and school level of alcohol use (one standard deviation above the mean and below the mean) while assigning all other variables in the model to their sample means. In both cases, the significant interaction term was driven by the non-drinkers. Emotional distress was generally higher for occasional and frequent drinkers (predicted distress = .60), but the emotional distress of non-drinkers, although always lower, differed by the rate of alcohol use in their schools. Specifically, non drinkers were more emotionally distressed when attending schools characterized by high levels of alcohol use (.58) than when attending schools low in alcohol use (.53). Thus, drinkers, of all levels, were at a greater risk of increasing emotional distress over a one-year period. Nevertheless, like academic achievement, this risk was moderated by drinking norms in the peer context. Unlike academic achievement, however, this moderation largely occurred because of person-context mismatches involving non-drinkers rather than drinkers. Also unlike academic achievement, macrolevel peer contexts, as opposed to micro-level, were more likely to moderate the associations between adolescent drinking and emotional distress. 23 Adolescent Drinking CONCLUSION [Heading Level 1] The issue of underage alcohol use has been studied extensively, but the ways in which it has been studied have been less extensive. We know a great deal about who drinks, the reasons for drinking, and how personal relationships influence drinking but far less about the social psychological implications of drinking and how such implications vary across interpersonal contexts. Guided by the life course paradigm, this study explored these under-studied phenomena by asking why drinking is problematic and how this problematic nature depends on the peer context in which it occurs. To begin, we explored the consequences of adolescent drinking. Like past studies, we found that adolescent drinking was associated with declining academic performance and escalating emotional distress. These findings help to fill in the widespread assumptions about the dangers of early drinking. While it may indeed be part of growing up, adolescent drinking is problematic because it is related to other individual trajectories. The implications of this overlap can be seen in the short-term—undermining school success, risking mental health problems. Even though this study did not investigate the long-term consequences of adolescent drinking, we argue that these findings hint at how adolescent drinking may disrupt the transition to adulthood and, therefore, have lasting consequences. For example, problems in school or emotional difficulties could decrease the likelihood of attending college, with far-reaching effects. Because the adult life course is, in many ways, predicated on what occurs during adolescence, we need to understand adolescent drinking and its implications for postadolescence experiences. Future research can do more to explore the long-term consequences of adolescent drinking than we were able to do here by utilizing more 24 Adolescent Drinking extensive longitudinal data, modeling growth curves across multiple time points, and exploring the degree to which different behavioral trajectories intertwine over time and across life stages. Because our models revealed strong lagged effects for achievement and distress (e.g., prior achievement strongly predicting later achievement) during a relatively short time span, modeling associations among trajectories across broader periods of time will likely reveal how different experiences develop over time in opposition to or in complement with each other and, ultimately, which of these experiences has the most lasting effect. We also investigated how the consequences of adolescent drinking varied by peer context. Like past studies, we found that adolescents who associated with peers who drank had more problematic trajectories during this time period. Unlike past studies, we also explored whether peer factors moderated the association between adolescent drinking and other behaviors and, in doing so, uncovered a complex pattern of risk. In general, mismatches between adolescents and their peer contexts were problematic. For academic achievement, drinkers generally had lower achievement than non-drinkers, but this was especially true when they were members of friendship groups in which drinking was uncommon. For emotional distress, non-drinkers typically had lower distress, but this was less true when they were part of friendship groups or attended schools in which drinking was normative. Thus, the burden of being out of step with peers appeared to add to the risk of drinking and subtract from the protection of not drinking. These results suggest that the peer orientation of young people and their sensitivity to social conformity complicate our understating of the nature of risky health behavior. Drinking is generally problematic, but even more so when it defies prevailing 25 Adolescent Drinking norms. Not drinking is generally positive during this stage, but less so when drinking is valued. The social meaning of drinking, which varies from context to context, matters. This conclusion is closely related to the work of Terri Moffit, Avshalom Capsi, and colleagues (see Moffitt 1997) on adolescent problem behavior. According to their life course perspective, the substance use of some adolescents is indicative of psychopathology and, as such, likely leads to major problems in the long term. In the short term, however, these youth attain a social status in schools that leads other, less troubled adolescents to mimic their behavior. These latter adolescents typically age out of their early problem behavior. Clearly, these two patterns of drinking have different meanings and, therefore, different consequences. As echoed by our study, the circumstances in which adolescent drinking occurs are required to decipher the meaning of this behavior. The significance of the social meaning of drinking, which our findings illustrate, brings up three related issues. First, we took a social approach to the interaction between adolescent behavior and the peer context, focusing on the developmental risks of personcontext mismatches. Yet, a more psychological phenomenon could also be at work here. Young people who drink despite prevailing norms against drinking may have underlying dispositions (e.g., alcoholism, severe maladjustment) that are strong enough to break social norms (Shedler and Block 1990). For example, the literature on social settings of drinking, typically not involving adolescents or peers, suggests that solitary, private drinking is more indicative of psychopathology than social, public drinking (Mayer et al. 1998). In all likelihood, both social and psychological (or biological) mechanisms are at work here. We have not been able to tease them apart, although controlling for prior 26 Adolescent Drinking adjustment should partly do this. Future research could help to shed light on this process by examining the factors that select young people into person-context mismatches. Second, by necessity, our selection process eliminated adolescents without friends. By doing so, however, we likely eliminated an intriguing group: socially isolated drinkers. In many ways, these young people represent the most problematic adolescent population. Given the social nature of drinking, those who drink outside of social groups are more likely to have the internal dispositions to drink discussed above. Yet, the focus of this study was person-context mismatches, not solitary vs. social drinking. We were interested in those adolescents who had friends but were out of step with their friends, a focus which sheds light on interpersonal and developmental processes just as it informs knowledge on drinking. Again, future research can build on what we have done here by comparing these two related but different phenomena. Third, our analyses revealed a clear pattern in which the association between drinking and achievement was closely related to both intimate and larger peer contexts and the association between drinking and emotional distress was only related to the larger peer culture of the school. This pattern suggests that mismatches between adolescents and their close friends interfere with their day to day behavior, but that consequences of mismatches between adolescents and their general social settings are also manifested in emotional maladjustment. The mechanisms behind these two different scenarios (e.g., the micro-context of behavioral trajectories, the macro-context of psychological trajectories) need to be studied more extensively. This basic pattern, however, demonstrates the value of taking multiple perspectives on the peer world. 27 Adolescent Drinking The findings of this study can be extended and deepened in multiple ways beyond those suggestions we have already made. Drinking is one social problem with important health consequences, yet there are others, such as drug use and cigarette smoking, that may also be responsive to social context. Furthermore, in studying social context, we have focused on two levels of interpersonal context (e.g., the friendship group, the peer culture of the school), but more refined measures of peer group and other within-school environments are likely to better inform the processes that contribute to adolescent development. Of course, social structural, as opposed to interpersonal, contexts may also play a role in shaping adolescent behaviors. For example, the trends we have observed could vary by race, ethnicity, or gender. Finally, the findings on the school peer group, which suggest that the type of school that an adolescent attends can have behavioral implications, hints at potential institutional effects that need to be further explored. The school is the logical starting point for adolescents, but other institutions during this unique stage of life (e.g., workplace, church) may also shed light on drinking behavior. This study contributes important new information about the contextual nature of adolescent life. Drinking is an individual risk behavior, but one that occurs within a social environment. Health-related risk behaviors, and the complications associated with them, are major problems facing adolescents, their families, their schools, and the other institutions that serve them, but to understand the complex patterns in these behaviors we need to explore more fully the settings in which they occur, are maintained, or are discouraged. This study takes a first step in that direction. This lesson, no doubt, also applies to other social problems at other stages of the life course. 28 Adolescent Drinking Table 1. Descriptive Statistics for Each Stage of the Sample Selection Process Means Wave I Filter 1 Filter 1-2 Filter 1-3 Filter 1-4 Female .51 .51 .51 .51 .53 White .50 .51 .52 .54 .56 Age (years) 16.16 15.82 15.82 15.75 15.72 Parent education 5.41 5.44 5.45 5.45 5.55 Wave I alcohol use 1.09 1.01 1.01 1.00 1.00 Wave I academic achievement 2.75 2.76 2.77 2.78 2.82 Wave I emotional distress .49 .49 .48 .47 .47 Wave I number of friends 2.98 3.07 3.04 3.02 3.26 n 20,745 14,736 13,568 11,927 7,758 Note: Filter 1 excluded adolescents in the Wave I In-Home sample who did not participate in the Wave II In-Home Interview, Filter 2 those who did not have a valid sampling weight, Filter 3 those who did not have a parent interviewed, and Filter 4 those who did not have valid friendship information. 29 Adolescent Drinking Table 2. Descriptive Statistics by Drinking Status Three Levels of Alcohol Use None Occasional Frequent Sociodemographic Characteristics Gender (female) White Parent education Age (years) Peer Context Friends’ alcohol use School level of alcohol use 53%a 52%b 5.57 (2.24) 15.38c (1.56) 56%a 61%a 5.57 (2.16) 16.00b (1.37) 46%b 64%a 5.50 (2.17) 16.36a (1.40) .80c (1.00) .84c (.34) 1.30b (1.15) 1.00b (.32) 1.81a (1.26) 1.06a (.30) Adolescent Behavior Academic achievement (Wave II) 2.92a 2.77b 2.64c (.73) (.76) (.75) Emotional distress (Wave II) .42c .51b .57a (.36) (.39) (.24) n 4,212 2,277 1,214 Note: Standard deviations presented in parentheses below means. Statistics with different subscripts differ significantly (p < .05), as determined by one-way ANOVA for means and cross-tabulations for frequencies. A represents the highest (or most positive) mean or frequency with B and C representing statistics in descending order. The statistical tests for the school-level factors are to be interpreted with caution because, unlike the multivariate analyses presented later, they do not control for the multi-level nature of the data. Occasional drinkers drank alcohol more than once a year but less than once a month. Frequent drinkers drank alcohol more than once a month. 30 Adolescent Drinking Table 3. Results of Individual-Level and Multi-Level Models for Academic Achievement Model 1 Model 2 Model 3 Individual-Level Factors Gender (female) .09*** .09*** .09*** (.01) (.01) (.01) African-American -.15*** -.15*** -.14*** (.02) (.02) (.02) Hispanic-American -.08** -.08** -.07** (.03) (.03) (.03) Asian-American .02 .02 .03 (.04) (.04) (.04) Other -.07 -.07 -.06 (.04) (.04) (.04) Parent education .03*** .03*** .03*** (.00) (.00) (.00) Family structure (two-parent) .07*** .07*** .07*** (.01) (.01) (.01) Age (in years) .01* .01* .01 (.01) (.01) (.01) Number of friends .00 .00 .00 (.00) (.00) (.00) Parental alcohol use .01 .01 .01 (.01) (.01) (.01) Prior academic achievement .60*** .60*** .60*** (.01) (.01) (.01) Occasional alcohol use -.06*** -.08*** -.18*** (.02) (.02) (.04) Frequent alcohol use -.07*** -.14*** -.13* (.02) (.03) (.06) Friends’ alcohol use ---.00 --(.01) Individual-Level Interactions Occasional * friends’ alcohol use --.01 --(.01) Frequent * friends’ alcohol use --.04* --(.02) 31 Adolescent Drinking Table 3 (continued) Model 1 Model 2 Model 3 School-Level Factors Mean parent education --- --- Minority representation --- --- School size (log) --- --- Middle school --- --- School level of alcohol use --- --- .01 (.02) -.00 (.00) -.03 (.02) -.09* (.04) -.07 (.05) Cross-Level Interaction Terms Occasional * school level of alcohol use --- --- Frequent * school level of alcohol use --- --- .13** (.04) .06 (.06) n = 6,818 *** p < .001, ** p < .01, * p < .05 Note: Unstandardized coefficients presented with standard errors in parentheses. 32 Adolescent Drinking Table 4. Results of Individual-Level and Multi-Level Models for Emotional Distress Model 1 Model 2 Model 3 Individual-Level Factors Gender (female) .06*** .06*** .06*** (.01) (.01) (.01) African-American .01 .01 .01 (.01) (.01) (.01) Hispanic-American .05** .05** .05** (.01) (.01) (.02) Asian-American .07** .07** .06** (.02) (.02) (.02) Other .05* .05* .04* (.02) (.02) (.02) Parent education -.01*** -.01*** -.01*** (.00) (.00) (.00) Family structure (two-parent) -.04*** -.04*** -.04*** (.01) (.01) (.01) Age (in years) .01* .01* .01 (.02) (.02) (.00) Number of friends .00 .00 .00 (.00) (.00) (.00) Parental alcohol use -.01 -.01 -.01 (.00) (.00) (.00) Prior emotional distress .53*** .53*** .53*** (.01) (.01) (.01) Occasional alcohol use .04*** .04*** .08*** (.01) (.01) (.02) Frequent alcohol use .05*** .06*** .12*** (.01) (.02) (.03) Friends’ alcohol use ---.00 --(.01) Individual-Level Interaction Terms Occasional * friends’ alcohol use ---.00 --(.01) Frequent * friends’ alcohol use ---.01 --(.01) 33 Adolescent Drinking Table 4 (continued) Model 1 Model 2 Model 3 School-Level Factors Mean parent education --- --- Minority representation --- --- School size (log) --- --- Middle school --- --- School level of alcohol use --- --- .00 (.01) .00 (.00) -.01 (.01) .02 (.02) .06* (.02) Cross-Level Interaction Terms Occasional * school level of alcohol use --- --- Frequent * school level of alcohol use --- --- -.05* (.02) -.08* (.03) n = 7,338 *** p < .001, ** p < .01, * p < .05 Note: Unstandardized coefficients presented with standard errors in parentheses. 34 Adolescent Drinking Appendix. Descriptive Statistics for Study Variables Mean (SD) Key Study Variables Adolescent alcohol use None --Occasional --Frequent --Academic Achievement Wave I 2.82 (.76) Wave II 2.83 (.75) Emotional distress Wave I .47 (.38) Wave II .49 (.39) Friends’ alcohol use 1.12 (1.15) School level of alcohol use .92 (.34) Individual-Level Characteristics Gender (female) --Race Non-Hispanic White --African-American --Hispanic-American --Asian-American --Other --Family structure (two-parent) --Parent education 5.55 (2.90) Age (years) 15.72 (1.53) Number of friends 3.26 (2.72) Parental alcohol use -.01 (.80) Frequency 55% 30% 15% ----- --------- 53% 56% 20% 15% 6% 3% 56% --------- 35 Adolescent Drinking Appendix (continued) Mean (SD) School-Level Characteristics Mean parent education Minority representation School size Middle school n 4.65 (.72) 46.50 (33.58) 1032.46 (784.32) --7,758 Frequency ------24% 7,758 36 Adolescent Drinking REFERENCES [Heading Level 1] Aseltine, Robert. 1995. “A Reconsideration of Parental and Peer influences on Adolescent Deviance.” Journal of Health and Social Behavior 3:103-121. Bearman, Peter, Jo Jones, and J. Richard Udry. 1997. “The National Longitudinal Study of Adolescent Health: Research Design.” Carolina Population Center, University of North Carolina at Chapel Hill. WWW document: http://www.cpc.unc.edu/ projects/addhealth/design.html. Retrieved September, 15, 2002. Brown, B. Bradford. 1990. “Peer Groups and Peer Cultures.” Pp. 171-196 in At the Threshold, edited by Shirley Feldman and Glen Elliott. Cambridge, MA: Harvard. Bryk, Anthony S. and Stephen W. Raudenbush. 1992. Hierarchical Linear Models: Applications and Data Analysis Methods. Newbury Park, CA: Sage. Chantala, Kim and Joyce Tabor. 1999. “Strategies to Perform a Design-Based Analysis Using the AddHealth Data.” Carolina Population Center, University of North Carolina at Chapel Hill. WWW document: http://www.cpc.unc.edu/projects/ addhealth/strategies.html. Retrieved December 1, 2002. Chassin, Laurie, Steven C. Pitts, and Christian DeLucia. 1999. “The Relation of Adolescent Substance Use to Young Adult Autonomy, Positive Activity Involvement, and Perceived Competence.” Development and Psychopathology 11:915-932. Cleveland, Hobart H. and Richard P. Wiebe. 2003. “The Moderation of Adolescent-toPeer Similarity in Tobacco Use by School Levels of Substance Use. Child Development 74:279-91. Coleman, James. 1961. The Adolescent Society. New York: Free Press of Glencoe. 37 Adolescent Drinking Costa, Frances M., Richard Jessor, and Mark S. Turbin. 1999. “Transition into Adolescent Problem Drinking: The Role of Psychosocial Risk and Protective Factors.” Journal of Studies on Alcohol 60:280-490. Crosnoe, Robert. 2000. “Friendships in Childhood and Adolescence: The Life Course and New Directions.” Social Psychology Quarterly 63:377-391. Crosnoe, Robert. 2002. “Academic and Health-Related Trajectories in High School: The Intersection of Gender and Athletics.” Journal of Health and Social Behavior 43:317335. Eccles, Jacquelynne S., Carol Midgely, Allan Wigfield, Christy M. Buchanan, David Reuman, Douglas MacIver. 1993. “Development During Adolescence: The Impact of Stage-Environment Fit on Young Adolescents' Experiences in Schools and in Families.” American Psychologist 48:90-101. Elder, Glen H. Jr. 1998. “Life Course and Human Development.” Pp. 939-991 in Handbook of Child Psychology, edited by William Damon. New York: Wiley. Elliot, Delbert, David Huizinga, and Suzanne Ageton. 1985. Explaining Delinquency and Substance Use. Beverly Hills: Sage. Galambos, Nancy L. and Ranier K. Silbereisen, 1987. “Substance Use in West German Youth: A Longitudinal Study of Adolescents' Use of Alcohol and Tobacco.” Journal of Adolescent Research 2:161-174. Garmezy, Neil and Ann Masten. 1986. “Stress, Competence, and Resilience: Common Frontiers for Therapist and Psychopathologist.” Behavior Therapy 17:500-521. Hartup, Willard and Nan Stevens. 1997. “Friendships and Adaptation in the Life Course.” Psychological Bulletin 121:355-370. 38 Adolescent Drinking Hawkins, J. David, Richard Catalano, and Janet Miller. 1992. “Risk and Protective Factors for Alcohol and Other Drug Problems in Adolescence and Early Adulthood: Implications for Substance Use Prevention.” Psychological Bulletin 112:64-105. Hussong, Andrea M., Richard E. Hicks, Suzanne A. Levy, and Patrick J. Curran. 2001. “Specifying the Relations Between Affect and Heavy Alcohol Use Among Young Adults.” Journal of Abnormal Psychology 110:449-461. Jessor, Richard, John Edward Donovan, and Frances Marie Costa. 1991. Beyond Adolescence: Problem Behavior and Young Adult Development. Cambridge: Cambridge University. Johnson, Monica K., Robert Crosnoe, and Glen H. Elder, Jr. 2001. “Student Attachment and Academic Engagement: The Role of Ethnicity.” Sociology of Education 74:31840. Johnston, Lloyd D., Patrick M. O’Malley, and Jerald G. Bachman. 1996. National Survey Results on Drug Use from the Monitoring the Future Study, 1975-1995, vol. 2, Secondary School Students. National Institute of Health Publication No. 96-4139. Washington D.C.: Government Printing Office. Kandel, Denise. 1996. “The Parental and Peer Contexts of Adolescent Deviance: An Algebra of Interpersonal Influences.” Journal of Drug Issues 26:289-315. Keefe, Keunho and Michael D. Newcomb. 1996. “Demographic and Psychosocial Risk for Alcohol Use: Ethnic Differences.” Journal of Studies on Alcohol 57:521-534. Mayer, Randall R., Jean L. Forster, David M. Murray, and Alexander C. Wagenaar. 1998. “Social Settings and Situations of Underage Drinking.” Journal of Studies on Alcohol 59:207-215. 39 Adolescent Drinking Mensch, Barbara S. 1988. “Dropping Out of High School and Drug Involvement.” Sociology of Education 61:95-113. Moffitt, Terrie E. 1997. “Adolescent-Limited and Life Course Persistent Offending: A Complementary Pair of Developmental Theories.” Pp. 11-54 in Developmental Theories of Crime and Delinquency: Advances in Criminological Theory, edited by Terence P. Thornberry. New Brunswick, NJ: Transaction. Mukamal, K.J., Kate M. Conigrave, Murray A. Mittleman, Carlos A. Camargo Jr., Meir J. Stampfer, Walter C. Willett, Eric B. Rimm. 2003. “Roles of Drinking Pattern and Type of Alcohol Consumed in Coronary Heart Disease in Men.” New England Journal of Medicine 348:109-118. Newcomb, Michael D. and Peter M. Bentler. 1989. “Substance Use and Abuse Among Children and Teenagers.” American Psychologist 44:242-248. Radloff, Lenore S. and Locke, Ben Z. 1986. “The Community Mental Health Assessment Survey and the CES-D Scale.” Pp. 177-189 in Community Surveys of Psychiatric Disorders, edited by Myrna Weissman, Jerome Meyers, and Catherine Ross. New Brunswick, NJ: Rutgers. Resnick, Michael D., Peter S. Bearman, Robert W. Blum, Karl E. Bauman, Kathleen M. Harris, Jo Jones, Joyce Tabor, Trish Beuhring, Renee E. Sieving, Marcia Shew, Marjorie Ireland, Linda H. Bearinger, J. Richard 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. 1994. “Overweight and Depression.” Journal of Health and Social Behavior 33:63-78. 40 Adolescent Drinking Scheier, Lawrence M., Gilbert Botvin, and Eli Baker. 1997. “Risk and Protective Factors As Predictors of Adolescent Alcohol Involvement and Transitions in Alcohol Use: A Prospective Analysis.” Journal of Studies on Alcohol 58:652-667. Schulenberg, John and Jennifer Maggs. 2002. “A Developmental Perspective on Alcohol Use and Heavy Drinking during Adolescence and the Transition to Young Adulthood.” Journal of Studies on Alcohol 63:54-70. Schulenberg, John, Jennifer Maggs, Ted Dielman, Sharon Leech, Deborah Kloska, Jean Shope, and Virginia Laetz. 1999. “On Peer Influences to Get Drunk: A Panel Study of Young Adolescents.” Merrill-Palmer Quarterly 45:108-142. Schulenberg, John, Jerald G. Bachman, Patrick M. O’Malley, and Lloyd D. Johnston. 1994. “High School Educational Success and Subsequent Substance Use: A Panel Analysis Following Adolescents to Young Adulthood.” Journal of Health and Social Behavior 35:45-62. Shedler, Jonathan and Jack Block. 1990. “Adolescent Drug Use and Psychological Health: A Longitudinal Inquiry.” American Psychologist 45:612-630. Singer, Judith D. 1998. “Using SAS Proc Mixed to Fit Multilevel Models, Hierarchical Models, and Individual Growth Models.” Journal of Educational and Behavioral Statistics 24:323-355. Wilcox, Steven and J. Richard Udry. 1986. “Autism and Accuracy in Adolescent Perceptions of Friends' Sexual Attitudes and Behavior.” Journal of Applied Social Psychology 16:361-374. 41