Abstract
Previous research on pubertal timing has either evaluated contextual predictors of early puberty or negative adjustment outcomes associated with off-time development, especially early maturation. In this study, we integrated these 2 lines of research by evaluating the moderating influence of early childhood household risk on associations between early puberty and 8th-grade substance use in a longitudinal sample of 1,070 participants. We determined trajectories of early childhood household risk using group-based trajectory analysis. Rates of early maturation were higher but not significantly so in groups with high household risk. Early timing was associated with higher rates of substance initiation only among individuals with a history of high household risk.
Keywords: substance use, pubertal timing, family conflict, relationship quality, trajectory analysis
A central theme in the study of human development is the effect on behavior and adjustment of the interaction between an individual and his or her environment over the life course. Of particular interest is the study of turning points or specific events in one's life that alter future development (Graber & Brooks-Gunn, 1996; Pickles & Rutter, 1994). The transition of puberty is an example of a formative event in one's life that marks the physical transition from childhood to adolescence. If this transition occurs early, it can be particularly deleterious to the course of an adolescent's behavior and adjustment over time. Early pubertal timing (i.e., experiencing puberty earlier than other same age, same gender adolescents) has been linked to a range of negative adjustment outcomes, including higher rates of both internalizing and externalizing problems (Ge, Conger, & Elder, 2001; Graber, Lewinsohn, Seeley, & Brooks-Gunn, 1997; Hayward, Killen, Wilson, & Hammer, 1997; Stice, Presnell, & Bearman, 2001; Tremblay & Frigon, 2005). Specifically, several recent studies have found that early pubertal timing was associated with substance use (i.e., tobacco, alcohol, marijuana) among both male and female adolescents (Dick, Rose, Viken, & Kaprio, 2000; Ge et al., 2006; Graber et al., 1997; Lanza & Collins, 2002; Tschann et al., 1994; Westling, Andrews, Hampson, & Peterson, 2008; Wichstrom, 2001). Moreover, earlier initiation of substance use, that is, initiation in the middle school years, has been identified as a risk factor for escalation to substance abuse (Anthony & Petronis, 1995; Federman, Costello, Angold, Farmer, & Erkanli, 1997; Grant & Dawson, 1997). As such, early pubertal timing can be conceptualized as a turning point that increases the risk of future substance use and abuse.
Extant research on substance initiation, use, and abuse has also focused on aspects of the family environment as risks for the development of substance use. Demographic characteristics of families (i.e., socioeconomic status, family structure), relationship quality within families, as well as parent's own substance use problems are all aspects of the home environment that have been shown to be associated with increased risk for earlier initiation of substance use and substance abuse among adolescents (Andrews, Hops, & Duncan, 1997; Mayes & Suchman, 2006). Similarly, the timing of puberty, itself, is influenced by an array of biological, behavioral, and social-contextual factors, including genetic influences, nutrition and dieting, exercise, sexual abuse, family context such as father absence, and quality of parent–child relationships (for recent reviews, see Belsky et al., 2007; Ellis, 2004; Graber, 2003).
In sum, experiencing the transition of puberty earlier than one's peers has the potential to negatively alter an individual's developmental trajectory. However, there is evidence that one's developmental trajectory may affect the likelihood of experiencing the pubertal transition earlier. Belsky, Steinberg, and Draper's evolutionary theory of socialization (BSD) highlights the importance of early childhood socialization experiences on subsequent somatic development (Belsky, Steinberg, & Draper, 1991). BSD posits two developmental pathways associated with pubertal development. The first pathway is characterized by inadequate resources, marital discord, and poor parent–child relationship quality during early childhood. As hypothesized by BSD, children raised under these conditions would develop a mistrustful and opportunistic interpersonal orientation, viewing the world as unpredictable. It would be evolutionarily advantageous to experience earlier pubertal maturation under such conditions as this would increase the amount of time that an individual is capable of reproducing and hence maximize the likelihood that an individual passes on their genes. However, the second pathway is characterized by adequate resources, spousal harmony, and positive parent–child relationship quality. Parental investment in the well-being of the offspring leads to a secure and trustful interpersonal orientation. Under such conditions, BSD hypothesized that pubertal development would be later, with more emphasis on forming long-term pair bonds. Both pathways predict that contextual factors of early childhood not only affect the timing of puberty but also influence postpubertal behavior.
The possibility that the effect of early puberty on one's future behavior may depend on one's prior developmental history is a consideration that must be addressed in assessments of the causal influence of early puberty on future substance use. To date, the evaluation of early pubertal timing effects on adolescent behaviors has been conducted separately from evaluations of the precipitating social/contextual antecedents of early pubertal timing (e.g., Andersson & Magnusson, 1990; Belsky, 2007; Belsky et al., 2007; Ellis, 2004; Ge, Brody, Conger, Simons, & Murry, 2002; Graber, 2003; Lynne, Graber, Nichols, Brooks-Gunn, & Botvin, 2007; Stattin & Magnusson, 1990). We are aware of only one article in which both social/contextual antecedents of the timing of puberty as well as concurrent associations between completion of puberty and negative adjustment outcomes were evaluated (Tremblay & Frigon, 2005). However, this article evaluated these associations via two separate sets of analyses, finding that significant predictors of earlier pubertal maturation also predicted negative adjustment outcomes. As such, true integration of these two lines of research did not occur given that both sets of analyses were conducted independently of one another. Although the work of Tremblay and Frigon (2005) was an important step forward, in the current investigation we advance the field by using statistical techniques based on growth curve modeling to gain a more accurate understanding of links between early childhood household risk, pubertal timing, and substance use.
In addition, previous research has evaluated links between puberty and social context in longitudinal research (Andersson & Magnusson, 1990; Lynne et al., 2007; Stattin & Magnusson, 1990). However, these studies evaluated the role of social contexts that were concurrent with or subsequent to pubertal timing as potential mediators of links between pubertal timing and negative adjustment outcomes. The current study differs in that precipitating social contexts are evaluated. The stage termination model of pubertal timing effects states that early maturers may have more negative adjustment outcomes because they are perceived as older on the basis of their physical appearance and are thus exposed to more adult-like situations at an earlier age through association with older/deviant peers (Brooks-Gunn, Petersen, & Eichorn, 1985). Although this model is influential in elucidating mechanisms that underlie pubertal timing effects, it does not address the role of social/contextual factors as precipitating influences on the likelihood of experiencing early pubertal timing or how to interpret pubertal timing effects in light of such associations. In the current investigation, we address this gap in model testing.
Many of the antecedents of early timing, especially social-contextual factors such as family resources and conflict, also predict adolescent behaviors such as substance initiation and use (Mayes & Suchman, 2006). As such, the interpretability of pubertal timing effects on substance use is difficult to discern from the effects of precipitating social/contextual factors. The primary aim of this study is to integrate these two literatures and thus examine the association between early pubertal timing and substance initiation during adolescence while taking into account prior developmental history regarding childhood antecedents of earlier maturation and substance initiation.
Early Pubertal Timing and Substance Use
Girls tend to experience the onset of the physical signs of puberty (e.g., breast development, pubic hair, growth spurt) approximately 1.5 years before boys experience comparable changes. These physical signs of maturity are often accompanied by increased responsibility and exposure to more adult-like situations as girls, their peers, and adults around them respond to these cues for maturation. Overall, there is substantial evidence that more negative adjustment outcomes are associated with early puberty in girls than in boys.
Specifically, a surge of recent studies have reported links between earlier pubertal maturation in girls and numerous negative adjustment outcomes, including alcohol, tobacco, and/or substance use (e.g., Dick et al., 2000; Graber et al., 1997; Stice et al., 2001; Westling et al., 2008; Wichstrom, 2001). For example, among twin adolescent girls who were highly discordant in their onset of puberty, higher rates of tobacco and alcohol use were found among early maturing girls compared with their on-time/late maturing sisters as well as their on-time/late maturing peers (Dick et al., 2000). Using the National Longitudinal Study of Adolescent Health data for middle school age girls, Lanza and Collins (2002) found that early maturing girls reported levels of substance use that were three times greater than their on-time/late maturing counterparts. Stice et al. (2001) also found higher rates of substance use disorders as well as comorbid substance use and depressive disorders among early maturers in a study of young adolescent girls. Similarly, in an evaluation of data from the Oregon Adolescent Depression Project examining both girls and boys, Graber et al. (1997) found higher rates of substance use disorders during mid-adolescence among early maturing girls but not boys. Ge et al. (2006) evaluated perceptions of substance use, future intentions, and willingness to use substances, as well as actual reports of substance use among African American boys and girls in early adolescence. They found that early pubertal timing was more strongly associated with more favorable social images of substance users and greater intentions and willingness to use substances among early maturing girls compared with early maturing boys. However, both genders reported similar levels of actual substance use. In contrast, Tschann et al. (1994) found that early maturation was a risk factor for tobacco, alcohol, and marijuana use for both boys and girls during middle school; Westling et al. (2008) also demonstrated that early maturation predicted subsequent alcohol use for both boys and girls. Finally, Wichstrom (2001) found that early pubertal maturation was more strongly associated with alcohol use among Norwegian boys compared with girls. Thus, the results of the aforementioned studies suggest that early maturation is a potential risk factor for both boys and girls.
The Effect of High Risk Home Environments on Substance Use
High risk home environments (i.e., household risk) are characterized by less resources (e.g., low parental education, low socioeconomic status, father absence) and more conflict (e.g., poor relationship quality, parental substance use problems). Previous studies have found that high risk home environments are associated with an increased risk of substance use and abuse during adolescence among both boys and girls (Andrews et al., 1997; Hawkins, Catalano, & Miller, 1992; Mayes & Suchman, 2006). Research on the role of family resources in adolescent substance use has yielded mixed results. There is evidence that households characterized by higher socioeconomic status and parental education are associated with greater experimentation with alcohol and marijuana use. However, more extreme poverty and low parental education are associated with substance use and abuse (Hawkins et al., 1992; Mayes & Suchman, 2006). Although there is some evidence that children of divorced parents are at higher risk for substance use, this effect does not appear to be independent of household conflict. Household conflict is an independent and stronger predictor of adolescent substance use compared with household structure (Hawkins et al., 1992). High levels of marital conflict as well as parent–child conflict are positively correlated with adolescent substance use (Hawkins et al., 1992). In addition, Walden, Iacono, and McGue (2007) found that parental substance use was associated with greater increases in adolescent substance use with similar effects for both genders.
Findings from studies investigating gender differences in the prevalence of substance use are mixed and vary by substance and severity of use (Andrews, 2005). Specifically, some previous research has indicated that male adolescents tend to engage in higher rates of illicit drug use (e.g., marijuana use) compared with female adolescents and are at higher risk for substance dependence in young adulthood (Andrews, 2005; Mayes & Suchman, 2006). Other studies show that girls report either higher or comparable rates of substance use (e.g., cigarettes, alcohol, marijuana use) compared with boys by mid-to-late adolescence (Andrews, Tildesley, & Hampson, 2009; Westling et al., 2008). In the 2007 Youth Risk Behavior Surveillance System report, no gender difference was found for current alcohol use among high school students, although male adolescents had higher rates of current cigarette use (Centers for Disease Control and Prevention, 2008). Given the previous findings of gender differences regarding onset of pubertal timing, pubertal timing effects, and prevalence of substance use, it is important to evaluate these constructs among both male and female adolescents.
The Effect of High Risk Home Environments on Pubertal Timing
As indicated, a range of bio-behavioral factors have been identified as antecedents of and influences on the onset and progression of puberty. In particular, evaluations of social influences on pubertal onset have revealed that puberty may occur earlier among girls who grow up in more conflict ridden home environments (Belsky et al., 2007; Ellis, McFadyen-Ketchum, Dodge, Pettit, & Bates, 1999; Graber, Brooks-Gunn, & Warren, 1995; Moffitt, Caspi, Belsky, & Silva, 1992) or who live in households in which a step-father is present (Ellis & Garber, 2000) or a father is absent (Surbey, 1990). Surbey (1990) evaluated the association between family structure and pubertal timing in a retrospective study and found that father absence in the childhood years was linked with earlier maturation in college-age women. Prospective studies of the association between household conflict and pubertal timing found that low warmth in family relations and to a lesser extent conflict in middle childhood or early adolescence were associated with earlier maturation in girls across families regardless of father absence (Graber et al., 1995; Moffitt et al., 1992). Subsequently, Ellis et al. (1999) extended these findings by demonstrating that lack of affection in parent–child relationships, as well as father absence, assessed at 4–5 years of age also predicted earlier maturation in girls. In a study of girls whose mothers had a history of depression, Ellis and Garber (2000) found that maternal mood disorders were associated with earlier maturation in girls, but the association was mediated by poor quality of family relationships and father absence, which were both independent predictors of earlier maturation in girls. Most recently, Belsky et al. (2007) examined parenting effects on pubertal onset in the National Institute of Child Health and Human Development Study of Early Child Care and Youth Development. Although their analyses excluded non-White participants, the study is unique in its use of annual physical examinations to assess pubertal maturation and the examination of both girls and boys. Results of this study revealed that greater maternal harsh control predicted earlier pubertal timing in girls but not boys.
Thus, across several studies, family structure and conflict have been identified as predictors of earlier maturation in girls. Notably, most studies that have demonstrated these effects have been conducted with girls only or have only found effects for girls and not for boys. In the present investigation, we extend previous research by evaluating the association of both family structure and conflict within the family on pubertal timing in both boys and girls. In addition, this is perhaps the first study to examine household risk associated with low family resources and parental substance use as antecedents of early pubertal timing.
Current Study
This investigation is the first step in integrating research that has evaluated social and contextual influences on pubertal timing with research that has evaluated pubertal timing effects on adolescent adjustment. The contribution of the current study is enhanced by the following: (a) The sample includes both male and female adolescents, (b) household risk (e.g., demographics, relationship quality, and parental substance use) is assessed via multiple reporters (parent and child), and (c) the sample has been followed longitudinally from early childhood (first grade) through adolescence. There are two preliminary objectives and one primary objective in this study. The first preliminary objective evaluates the possibility of distinct groups within the population that differ in onset and rate of change in household risk across childhood (Grades 1–6). Group-based trajectory analysis provided a statistical method for longitudinal examination of this objective. The second preliminary objective is to replicate previous research on social/contextual influences of pubertal timing by evaluating whether an individual's developmental history of household risk influenced the likelihood of early pubertal maturation for the sample overall as well as separately by gender. It was expected that higher household risk would predict earlier maturation in girls. On the basis of the literature, we did not have a hypothesis for boys.
The primary objective of this study is to evaluate the association between early pubertal maturation and substance use in the eighth grade and to determine whether this association differs on the basis of one's developmental history of household risk (Haviland & Nagin, 2005). This primary objective addresses the possibility that one's developmental history of household risk during childhood may influence the association between early pubertal maturation and substance use in adolescence. The results of this analysis will potentially contribute information regarding the mechanisms underlying the well-established association between early pubertal timing and substance use by linking two disparate literatures, one that assessed antecedents of pubertal timing and the other that assessed negative adjustment associated with pubertal timing. Potential gender differences are also evaluated.
Previous research has found that early pubertal timing accentuated the association between childhood behavioral and emotional problems and subsequent behavioral and emotional adjustment in adolescence (Caspi & Moffitt, 1991; Ge, Conger, & Elder, 1996). Given these previous findings, it was expected that both history of household risk and pubertal timing would predict substance initiation and that the influence of early pubertal timing on substance initiation would be more pronounced among individuals with a history of household risk. Both actual reports of substance use in eighth grade (tobacco, alcohol, marijuana, and inhalant use) as well as subjective norms of use were used as outcome measures in this study. Subjective norms theoretically tap into adolescents’ normative beliefs regarding expectations for their behavior on the basis of a reference group that can be an individual's friends, family, or larger community (Ajzen & Fishbein, 1973) and have been found to be associated with subsequent substance use by the adolescent (e.g., Andrews, Hampson, & Barckley, 2008).
Method
Design and Participants
Data for this study were collected as part of the Oregon Youth Substance Use Project (OYSUP), an ongoing longitudinal investigation examining the etiological factors predictive of substance use (see Andrews, Tildesley, Hops, Duncan, & Severson, 2003). OYSUP is based on a cohort-sequential design (Schaie, 1965, 1970), wherein five grade cohorts (Grades 1–5) were followed for eight annual assessments over a 9-year period, with a 1-year break (between Time [T]4 and T5) in assessments due to a funding gap. In cohort-sequential designs, grade will always be confounded with either differences between cohorts or differences between time of assessment. An evaluation of cohort differences at each grade level regarding household risk revealed that Cohort 1 reported slightly more household risk in fourth grade compared with Cohorts 2 and 3. There were no other significant cohort differences in household risk. As such, developmental changes across grades were evaluated collapsing across time of measurement, and cohort was controlled for in all subsequent analyses. We selected a sample of 2,127 students in 15 elementary schools from a single school district in western Oregon using stratified random sampling (by school, grade, and sex) holding the sampling proportions at the population means from which this sample was drawn. This ensured that there were equal proportions of male and female adolescents within each grade recruited to participate in the first assessment and that grade level and school were not confounded. Parents of the selected students were sent a letter followed by a phone call describing the project and soliciting participation. Recruitment and assessment materials were printed in both Spanish and English. In addition, a Latina research assistant recruited Hispanic American parents and conducted assessments in Spanish.
Parents of 1,075 students consented to their child's participation (50.7%). This participation rate was similar to other epidemiological community-based studies (e.g., 52%: Jessor & Jessor, 1977; 60.1%: Lewinsohn, Hops, Roberts, Seeley, & Andrews, 1993) in which active informed consent is required. However, low rate of consent does have the potential to bias the sample in that those who consent to participate may not represent the population from which they were drawn. Andrews et al. (2003) explored this potential form of sample bias for the OYSUP sample and found that the baseline participants were statistically comparable with elementary school students in the district from which they were drawn on demographics and socioeconomic status. In addition, the OYSUP sample reported comparable levels of substance use compared with public school students from the same region of Oregon (Oregon Department of Human Services, 2000), with the exception of inhalants. The OYSUP sample was slightly different from the population from which it was drawn in that there were higher reported rates of inhalant use, and participants were slightly higher in verbal academic achievement compared with the school district average.
Of the 1,075 students whose parents consented, five students were absent on the day of assessment resulting in a final sample of 1,070 children. Approximately 215 students per grade (first through fifth) participated with an even distribution by gender (50.3% female, n = 538). The average age of participants at the first assessment was 9.0 years (SD = 1.45). Seventy-one percent of mothers and 66% of fathers had more than a high school education, and 7% of mothers and 11% of fathers had not graduated from high school. The majority of the sample was European American (85.8%), with 7.1% Hispanic American, 1.1% African American, 2.2% Asian American, 2.4% American Indian or Alaskan Native, and 1.7% other or of mixed race or ethnicity. A little under half of the sample (40%) was eligible for a free or reduced lunch.
One hundred and thirty-eight children (12.8% of the total sample) who participated in the T1 assessment did not participate in the eighth assessment (T8). A probit regression was conducted predicting attrition in the eighth assessment from baseline demographic characteristics and baseline reports of substance use and intentions to use substances in the future. A joint test of the significance of these covariates revealed that the cohort at T1 was the only significant predictor of attrition at T8 (Z = 3.22, p = .001), such that older cohorts were more likely to discontinue their participation or be lost to follow-up. There were no significant differences between those missing data at T8 compared with the baseline sample regarding demographic characteristics, substance use, or intentions to use substances in the future.
Procedure
At T1, participants were assessed at their school during regular school hours. For all other annual follow-up assessments (T2–T8), the assessment location varied depending on where the student currently attended school. Individuals who moved outside of the school district but were still within driving range of our laboratory were assessed at the laboratory. Participants who did not live within driving range were assessed via telephone only if they were in the fourth grade or higher. Younger participants who moved (at T2 and T3) were not assessed until they entered fourth grade.
For all assessments, first through third graders were assessed via an individual interactive structured interview (Andrews et al., 2003). Fourth through eighth graders answered a written questionnaire in group sessions. Elementary school students (fourth and fifth graders) had a trained monitor read the questions aloud to the group while another monitor answered questions for individual participants as they arose. Middle school students (sixth through eighth graders) read the questions to themselves while a trained monitor was available to answer individual questions. In circumstances in which a middle school student could not read the questionnaire on his or her own, the monitor read it to him or her. All items were comparable across grade levels.
Measures
Early pubertal timing
Early pubertal timing is both a biologically and socially constructed variable based on the development of secondary sex characteristics relative to other same age, same gender peers. Importantly, early pubertal timing encompasses both pubertal onset and stage of pubertal development but goes beyond these constructs by embedding this information in a social context—classifying individuals into either early maturers or on-time/late maturers on the basis of their pubertal stage relative to peers. In this study, early pubertal timing was calculated on the basis of parent report of physical development. These reports were collected at fourth and fifth grades for girls and were collected at sixth and seventh grades for boys, given that girls tend to develop external signs of pubertal maturation about 1.5–2 years earlier than boys on average. Parents completed the Pubertal Development Scale (Petersen, Crockett, Richards, & Boxer, 1988) to assess their child's growth spurt, body hair, and skin changes (acne) as well as voice change and facial hair for boys, and breast growth and onset of menstruation (yes or no) for girls. Response options included the following: 1 = not yet started, 2 = barely started, 3 = definitely underway, and 4 = seems complete. Parent's responses on these items were averaged to create a continuous measure of pubertal status (level of physical development at a particular point in time) in which higher values indicated higher levels of physical maturation. Girls with average pubertal status scores greater than 2 in the fourth grade as well as girls with average scores greater than 2.5 in the fifth grade fell within the upper quartile on this measure of pubertal status and were classified as early maturers. Boys with average pubertal status scores greater than 2 in the sixth grade as well as boys with average scores greater than 3 in the seventh grade fell within the upper quartile on this measure of pubertal status and were classified as early maturers. The final indicator of early pubertal maturation was a dichotomous variable in which 1 represented early maturers, and all other children were coded zero.
Early childhood household risk
In the current study, we operationalized household risk on the basis of the contextual indicators outlined in BSD as important in the prediction of pubertal timing (Belsky et al., 1991). It was the sum of eight dichotomous variables in which each variable was coded as 1 for presence of the risk factor and coded as 0 for absence of the risk factor. Hence, on a scale of 0–8, higher values indicated the presence of greater household risk. This variable was created for each grade from first through sixth and was normally distributed at each grade, means ranging from 2.53 (SD = 1.63) to 2.88 (SD = 1.71). The eight variables used to compose this construct assess household resources, structure, and conflict. These three categories were chosen to reflect the three main categories of household risk identified by BSD and previous research as influential for pubertal timing and adolescent substance use. Parental survey responses were based largely on mother report (93%–96% of the sample within each grade). Of the sample, 75% had mother reported data for every assessment throughout the study. Father report was substituted for missing data when available.
Household resources were assessed via two variables: parent education and family income. Low parent education was a dichotomous variable that was assessed at each grade to capture changes in parent education during the course of the study. Parents with a high school diploma/equivalency or less were coded 1. Parents with any additional training beyond high school, including vocational training and college, were coded 0. Low family income was approximated by a single dichotomous variable that was coded 1 for participants who were eligible for free or reduced lunch and 0 for participants who were not eligible.
The stability of household structure was assessed via a single variable that tapped into family structure at each grade. The biological or adoptive parent of the participating child reported their current relationship status was with the child's other biological or adoptive parent. Individuals who reported that they were currently married to the participating child's other parent, or that they were currently living with the child's other parent as if married, were coded as zero. All other response options were coded as 1. Changes in this variable across Grades 1–6 provided information regarding the stability of family structure.
Household conflict was assessed via five variables that tapped into the degree of conflict present in relationships within the household between the parents and the participating child. The first variable, parent's report of relationship quality with spouse or partner, consisted of nine items from the Dyadic Adjustment Scale (Spanier, 1976; e.g., “How often do you discuss divorce or separation?” “Do you kiss your mate?”). Cronbach's alphas ranged from .82 to .88 across Grades 1–6. Response options ranged from 0 (never) to 5 (all of the time). Items were averaged such that higher values represented worse parent report of relationship quality with their partner. The upper 25% on this averaged composite variable were coded as 1. All others were coded 0.
The second variable, parent's report of relationship quality with child, consisted of seven items from the Conflict Behavior Questionnaire (CBQ; Prinz, Foster, Kent, & O'Leary, 1979; e.g., “My child and I joke around often”; “My child and I have big arguments about little things”). Cronbach's alphas ranged from .56 to .69 across Grades 1–6. Response options were as follows: 0 = mostly true, 1 = mostly false. These items were averaged such that higher values represented a more conflictual relationship. The upper 25% on this variable was coded as 1. All others were coded 0.
The third and fourth variables in this category were the children's report of their relationship quality with their mother and their father, using the CBQ. Separate variables were created for the mother and the father. These variables consisted of seven items from the CBQ (e.g., “My mom/dad and I joke around often”; “My mom/dad and I have big arguments about little things”). Cronbach's alphas ranged from .53 to .76 across Grades 1–6. Response options were as follows: 0 = mostly true, 1 = mostly false. Items were averaged such that higher values represented worse child report of relationship quality with either their mother or their father. The upper 25% on these composite variables were coded as 1. All others were coded 0.
The fifth variable assessed symptoms of dependence as a result of substance use, using items adapted from the National Household Survey of Drug Abuse (U.S. Department of Health and Human Services, Substance Abuse and Mental Health Services Administration, Office of Applied Studies, 2001). These items were similar to symptoms of dependence specified in the Diagnostic and Statistical Manual of Mental Disorders (4th ed., text rev.; American Psychiatric Association, 2000). The participating parents reported on 15 items assessing problems associated with their use of a specific substance, including use of alcohol, marijuana, and other illegal drugs (e.g., “Ignored the family”; “Had to get emergency medical help”; “Had trouble with the police”; “Had arguments or fights with family or friends”), as well as the same 15 problems that occurred as a result of their spouse or partners’ substance use. Response options were as follows: 1 = more than 6 years ago, 2 = within the last 6 years. Because of the response options, it is possible that at each annual assessment, some of the parental responses could be referring to the same event.
Preliminary analyses indicated that reports of any problems associated with either marijuana use or other illegal drug use were low at each time point (18%–23% for marijuana use and 15%–21% for other drug use), whereas reports of problems associated with alcohol use were much more prevalent (43%–47%).1 Analyses also revealed moderate to substantial overlap of problems associated with alcohol and marijuana or other illegal drugs. Given the degree of overlap in problems as well as the overlap in response options across times of assessment, a single dichotomous variable was created to indicate any problem due to severe alcohol use, marijuana use, or other illegal drug use across Grades 1–6 (coded as 1). All other responses were coded 0.
Adolescent substance use outcomes
In eighth grade, adolescents reported on their own use of cigarettes, alcohol, marijuana, and inhalants and the extent of their heavy drinking. Adolescents also reported on their subjective norms.
We assessed cigarette use in the past year using the following response options: 0 = never, 1 = once, 2 = a couple of times, 3 = some each month, 4 = some each week, and 5 = some each day. Alcohol, marijuana, and inhalant use in the past year were assessed separately for each substance with response options ranging from 0 (never) to 5 (some each day). Binge drinking was also assessed via adolescent self-report of how many times in the past year they have gotten really drunk from too much alcohol so that they fell down or got sick. Response options were as follows: 1 = never, 2 = 1 or 2 times, 3 = 3 or 4 times, and 4 = 5 or more times.
The adolescent's subjective norms regarding the use of each substance (alcohol, cigarette, marijuana, and inhalant use) were assessed by summing perceptions of how many kids in their school or neighborhood have tried the substance and perceptions of the number of their friends that use the substance. Response options for each of these items ranged from 0 (none) to 4 (all).
Analysis Plan
A group-based approach to analyzing developmental trajectories was used to address the preliminary objective of this study (Jones & Nagin, 2007; Nagin, 2005). This approach is a type of finite mixture modeling that identifies groups of individuals following approximately the same developmental trajectory over a specified period of time for the outcome of interest (e.g., household risk). This statistical methodology assumes that individual variation in the outcome of interest over time can be adequately represented by a finite number of different polynomial functions of time. Given the approximately normal distribution of the household risk variables at each point in time, in this study we estimated a censored normal group-based trajectory model using full information maximum likelihood (FIML) estimation.
Household risk was estimated as the latent variable , specific to individual i, in group j, at time t. It is important to note that a separate set of parameters, θj, was estimated for each group in the trajectory model, j = 1, 2, . . ., J. Group membership probabilities (πj), on the basis of the group specific sets of parameters, θj, were used to approximate the proportion of individuals within the population who fall in each trajectory group.
Posterior probabilities of group membership were used to classify individuals into a specific trajectory group on the basis of their data and the group membership probabilities.
Individuals were classified as belonging to a particular trajectory group if their posterior probability for that group was greater than 60%. Benefits of this analytical technique include the ability to identify groups of individuals within the population that would not necessarily have been predicted a priori and robust parameter estimates despite missing data due to the use of FIML to estimate the parameters of the model. FIML uses all individuals who provide data for at least one time point in the estimation of the group-based trajectories. As such, the trajectory model is based on the full sample as opposed to only individuals with complete data over time.
We evaluated model fit using the Bayesian information criterion to determine the number of latent trajectory groups that best represented the data, in which higher values indicated better model fit. However, it is important to emphasize that the number of groups identified as best representing the data is not immutable, and assignment to a particular trajectory group does not necessarily mean that a particular individual will follow that group's trajectory in lock step. These trajectory groups serve as a useful heuristic device in describing developmental patterns in the data.
The validity of assigning individuals to groups on the basis of their highest posterior probability was evaluated by Nagin (2005) via five simulated data sets in which true group membership and model parameters were known. On the basis of these simulations, Nagin put forth diagnostic guidelines for evaluating the adequacy of a model to identify distinct groups of trajectories and classify individuals to a trajectory group. First, the average posterior probability (Ave. PP) must be 0.70 or greater for all groups. This indicates that, on average, the individuals assigned to a particular trajectory group had a 70% or greater probability of belonging to that group on the basis of their individual data. The second diagnostic is the odds of correct classification for group j (OCCj).
Values of 5 or greater for all OCCj indicate high accuracy in individual assignment to trajectory groups. Finally, comparing the correspondence between the model estimation of the proportion of the population that follows a particular trajectory group (πj) with the proportion of the sample assigned to a particular trajectory group on the basis of their highest posterior probability (Pj) provides a third diagnostic. The values of πj and Pj would be identical if all individuals assigned to a trajectory group had a posterior probability of 1 (i.e., Ave. PP = 1.0 for all groups). Assignment uncertainty introduces error resulting in less correspondence between the values πj and Pj. Closer correspondence between these estimates indicates better model fit.
In this study, we used group-based trajectory analysis as a method to create balance within each trajectory group between early maturing adolescents and on-time/late maturing adolescents on covariates measured prior to puberty that are related to pubertal timing, similar to propensity score matching using subclassification (Rubin, 1997). Randomized experiments are the ideal methodology for evaluating the impact of an independent variable on a dependent variable through randomly assigning participants to levels of the independent variable. In this study, we attempt to create the same balance on covariates that would be obtained through random assignment by comparing individuals who vary in pubertal timing within trajectory group.
An advantage of using group-based trajectory analysis to create balance on time-varying covariates is the ability to model longitudinal change over time in these covariates and to statistically determine the number of different trajectory groups, rather than arbitrarily subclassifying individuals into quartiles. A benefit of blinded randomized trials is the clear separation of design (e.g., random assignment of individuals to levels of the independent variable) and analysis of the effect of the independent variable on the dependent variable that reduces experimenter bias. In the current study, we assume this benefit by distinctly separating the statistical balancing procedure (e.g., design) from the evaluation of pubertal timing effects on substance use.
However, creating balance on covariates that occur prior to maturation using group-based trajectory analysis does have two important assumptions (Haviland & Nagin, 2005). First, it is assumed that the estimated effect of pubertal timing on adolescent substance use within trajectory group is not confounded by uncontrolled covariates. This is also known as the ignorability assumption (Rosenbaum & Rubin, 1983b; Rubin, 1997). To estimate the true causal effect of pubertal timing on substance use, ideally we would observe the same individual as an early maturer and as an on-time/late maturer to estimate the difference in substance use associated with early pubertal timing. However, because pubertal timing is a static variable, this is not possible. Thus, for any one individual, we only observe one of these two conditions, and the other condition can be thought of as the missing counterfactual. Under the ignorability assumption, it is assumed that by comparing individuals within trajectory group, the missing counterfactual for early maturing individuals can be estimated by the expected outcome of the on-time/late maturing individuals and vice versa.
However, all statistical procedures that aim to create balance in covariates within observational studies are limited in that balance can only be ensured for measured covariates, whereas randomized controlled trials create balance on both measured and unmeasured covariates. This leaves open the possibility of unmeasured confounding variables that would violate the ignorability assumption. Sensitivity analyses were conducted to determine how strong an unmeasured covariate would have to be to reduce the findings of this study to nonsignificance (Rosenbaum & Rubin, 1983a).
The second assumption associated with using group-based trajectory analysis as a balancing procedure is that the variables that comprise the household risk variable are assumed not to differ between early maturers and on-time/late maturers when evaluated within trajectory group. To create balance, the individuals within a trajectory group should have comparable levels of the variables that comprise the household risk variable. Furthermore, groups of individuals, such as early versus on-time/late maturers, should also not differ on the variables that compose the household risk variable when evaluated within trajectory group. This results in reduced bias in pubertal timing effect estimates. However, if the pubertal timing groups differ on the variables that compose the household risk variable within trajectory group than balance is not fully achieved, and pubertal timing effects may be confounded. The validity of this assumption was evaluated via logistic regressions with each individual household risk variable predicting pubertal timing within household risk trajectory group. None of these associations were statistically significant, validating the second assumption.
A cross-tabulation of household risk trajectory group membership and early pubertal timing was conducted to determine whether different developmental trajectories of household risk were associated with higher proportions of early maturers among boys, girls, and the total sample. Finally, the association between early pubertal maturation and eighth-grade substance use and the moderating effect of one's developmental history of household risk on this association was evaluated via a series of analyses of covariance (ANCOVAs). The interaction between household risk and pubertal timing provided an overall test to determine whether pubertal timing effects differed depending on one's history of household risk in childhood. The dependent variables assessed were the individual substance use outcomes. The control variables included cohort and ethnicity. The independent variables were early pubertal timing, gender, and household risk trajectory group membership. In preliminary analyses, to investigate gender differences, we also examined the three-way interaction of gender, household risk, and pubertal timing.
Missing data
There is a certain amount of planned missingness in these data because of the cohort-sequential sampling design of the study. For example, the data available for second grade only includes individuals who were in either first or second grade during the first assessment. Participants who were in Grades 3–5 at first assessment are missing data in the second grade by design. The missingness (of data from participants in a given grade) in a cohort-sequential design is assumed to be missing completely at random. Excluding missing data attributed to the study design, the missing data for household risk variables ranged from 2% to 27% across Grades 1–6. These missing data were assumed to be missing at random. As discussed previously, the FIML iterative parameter estimation employed in the group-based trajectory analysis used the full sample, reducing bias associated with only evaluating individuals with complete data across time (Graham, 2008). As such, all participants have estimated posterior probabilities of trajectory group membership.
Results
Descriptive Analyses
Although rates of substance use are fairly low during middle school, many students had experimented with substances at some point in their lifetime. Of the sample, 22% reported having ever tried cigarettes, 25% reported having ever drunk an entire drink of alcohol, 7% reported binge drinking in the past year, 17% reported having ever tried marijuana, and 6% reported having ever tried inhalants. In eighth grade specifically, average levels of substance use were low to moderate, depending on the outcome of interest (see Table 1; average levels were used in all analyses). There were a few gender differences regarding adolescent self-report of substance use and subjective norms of substance use (see Table 1). In general, female adolescents reported slightly higher levels of tobacco use and binge drinking in the past year. Female adolescents also reported slightly higher subjective norms regarding tobacco, alcohol, marijuana, and inhalant use among individuals in their neighborhood/school and who were their friends.
Table 1.
Means, Standard Deviations, and Sample Sizes for Eighth-Grade Adolescent Tobacco, Alcohol, and Other Substance Use
Male adolescents |
Female adolescents |
Total sample |
|||||||
---|---|---|---|---|---|---|---|---|---|
Substance use outcome | M | SD | n | M | SD | n | M | SD | n |
Cigarettes | |||||||||
Past year | 0.19 * | 0.76 | 376 | 0.35 * | 0.96 | 397 | 0.27 | 0.87 | 773 |
Subjective norms | 1.20 *** | 1.06 | 378 | 1.55 *** | 1.13 | 397 | 1.38 | 1.11 | 775 |
Alcohol | |||||||||
Past year | 0.57 | 0.94 | 375 | 0.78 | 0.98 | 397 | 0.68 | 0.96 | 772 |
Binge drink until sick or fall | 1.07 * | 0.35 | 378 | 1.12 * | 0.39 | 396 | 1.10 | 0.37 | 774 |
Subjective norms | 1.43 *** | 1.29 | 378 | 1.85 *** | 1.34 | 396 | 1.64 | 1.33 | 774 |
Marijuana | |||||||||
Past year | 0.23 | 0.73 | 375 | 0.29 | 0.82 | 397 | 0.26 | 0.78 | 772 |
Subjective norms | 1.14 *** | 1.24 | 378 | 1.47 *** | 1.24 | 397 | 1.31 | 1.25 | 775 |
Inhalants | |||||||||
Past year | 0.04 | 0.33 | 378 | 0.09 | 0.44 | 397 | 0.07 | 0.39 | 775 |
Subjective norms | 0.43 *** | 0.69 | 378 | 0.68 *** | 0.84 | 397 | 0.56 | 0.78 | 775 |
Note. Values in bold mark significant gender differences.
p < .05.
p < .001.
Trajectories of Household Risk
We conducted an evaluation of household risk using group-based trajectory analysis. Six distinct developmental trajectories of household risk were found (see Figure 1). All six trajectories of household risk were relatively stable across elementary school, indicating that household risk, as operationalized in this study, did not appear to increase or decrease significantly over early childhood.2
Figure 1.
Final six-group trajectory model of early childhood household risk.
As can be seen in Figure 1, the six groups of household risk were as follows: none (10%), low (24%), moderate (31%), moderate high (20%), high (14%), and very high (2%). The percentages reported for each of these groups refer to the proportion of the population that was expected to fall within each of these trajectory groups. Following model estimation, the posterior probability for each participant was estimated to determine his or her individual probability of membership within each of the six trajectory groups, on the basis of his or her own data. Participants were assigned to trajectory groups if their posterior probability was greater than 60%. Average posterior probabilities for being assigned to the appropriate group ranged from 0.79 to 0.93, all greater than the threshold of 0.70. The lowest odds of correct classification was 8.9, which is greater than the 5.0 minimum acceptable threshold. In addition, the correspondence between πj and Pj was within 2 percentage points for all groups (see the Appendix). These diagnostics indicate excellent model fit and accuracy of individual assignment to trajectory groups.
Table 2 provides descriptive statistics within trajectory group as well as for the total sample regarding gender, race/ethnicity, pubertal timing, and each of the eight individual variables that compose the household risk composite variable. As shown, demographics within each trajectory group were relatively similar to that of the total sample. Notably, although not significantly different, there was a greater percentage of female adolescents within the very high household risk trajectory group, and the percentage of early maturers within each trajectory group increased slightly as the level of household risk increased.3
Table 2.
Sample Descriptive Statistics
Household risk trajectory group |
|||||||
---|---|---|---|---|---|---|---|
Variable type | None | Low | Moderate | Moderate high | High | Very high | Total sample |
Demographic variables | |||||||
Gender (% female) | 42 | 53 | 50 | 54 | 55 | 71 | 50 |
Race/ethnicity (% Caucasian) | 89 | 87 | 85 | 87 | 80 | 91 | 85 |
Pubertal timing (% early) | 22 | 22 | 25 | 26 | 27 | 27 | 24 |
Early childhood household risk variables | |||||||
% qualified for free or reduced lunch | 0 | 12 | 36 | 65 | 88 | 100 | 41 |
% parent with high school education or less | 0 | 20 | 32 | 46 | 62 | 80 | 35 |
% household structure | 0 | 18 | 40 | 63 | 82 | 100 | 44 |
% poor parent relationship quality | 0 | 9 | 30 | 29 | 43 | 100 | 24 |
% poor parent relationship quality with child | 0 | 5 | 8 | 15 | 31 | 80 | 14 |
% poor child relationship quality with mother | 0 | 10 | 11 | 9 | 27 | 40 | 15 |
% poor child relationship quality with father | 0 | 8 | 13 | 21 | 39 | 80 | 19 |
% parental substance use problems | 0 | 35 | 61 | 72 | 81 | 100 | 53 |
Table 2 also illustrates descriptive differences within each trajectory group regarding the percentage of individuals who were exposed to each of the eight household risk variables that composed the household risk composite variable. Clear increases in exposure to household risk were evident between individuals assigned to each of the six trajectory groups.
Household Risk, Pubertal Timing, and Adolescent Substance Use
The final objective of this study was to evaluate the association between early pubertal maturation and substance use in the eighth grade and to determine whether this association differed on the basis of one's developmental history of household risk for boys, girls, and the total sample (Haviland & Nagin, 2005).
Interactions with gender
Unfortunately, given that so few individuals followed the highest trajectory of household risk (n = 21), it was not feasible to evaluate gender differences for this group. As such, evaluations of gender differences were conducted for a reduced household risk trajectory variable in which individuals within the two highest trajectory groups of household risk (high and very high) were combined into a single group. There were no significant three-way interactions between gender, reduced household risk trajectory group, and pubertal timing on any of the substance use outcomes evaluated in this study, nor for the subjective norms of substance use. The lack of significant three-way interactions indicated that there were no differences between boys and girls in associations between early pubertal timing and household risk trajectory group on eighth-grade substance use.
Given these findings, ANCOVAs controlling for gender, cohort, and race/ethnicity were used to assess associations between all six groups of childhood household risk, early pubertal maturation, and eighth-grade substance use. Significant interactions between household risk trajectory group and pubertal timing were decomposed through an analysis of simple main effects. Bonferroni-corrected pairwise comparisons were used for all follow-up analyses.
Cigarettes
A significant interaction between pubertal timing and household risk trajectory group was found regarding the adolescent's own cigarette use in the past year, F(5, 577) = 5.34, p < .001, η = .04. Follow-up Bonferroni-corrected pairwise comparisons revealed that early pubertal maturation was associated with higher levels of eighth-grade adolescent cigarette use only among individuals within the moderate, moderate high, and very high household risk trajectory groups (see Table 3). The averages reported in Table 3 indicate that early maturers within the very high household risk trajectory group reported smoking cigarettes or cigars a couple of times in the past year. This is in contrast to both early and on-time/late maturers from the other household risk trajectory groups whose reports of cigarette smoking were near zero.
Table 3.
Pubertal Timing × Household Risk Trajectory Group Interactions
Cigarette |
Alcohol |
Marijuana |
|||
---|---|---|---|---|---|
Past year |
Subjective norms |
Binge drink |
Past year |
Subjective norms |
|
Household risk trajectory group | M (SE) | M (SE) | M (SE) | M (SE) | M (SE) |
None | |||||
Early maturers | 0.08 (0.20) | 0.99 (0.25) | 1.06 (0.08) | 0.16 (.017) | 0.90 (0.28) |
On-time/late maturers | 0.23 (0.12) | 0.99 (0.14) | 1.08 (0.05) | 0.23 (0.10) | 0.93 (0.16) |
Total sample | 0.16 (0.12) | 0.99 (0.15) | 1.07 (0.05) | 0.19 (0.10) | 0.92 (0.16) |
Low | |||||
Early maturers | 0.09 (0.15) | 1.38 (0.19) | 1.03 (0.06) | 0.19 (0.13) | 1.39 (0.21) |
On-time/late maturers | 0.10 (0.08) | 1.18 (0.10) | 1.05 (0.03) | 0.12 (0.07) | 0.98 (0.11) |
Total sample | 0.10 (0.08) | 1.28 (0.10) | 1.04 (0.03) | 0.15 (0.07) | 1.19 (0.12) |
Moderate | |||||
Early maturers | 0.46* (0.14) | 1.19 (0.17) | 1.05 (0.06) | 0.13 (0.11) | 1.28 (0.19) |
On-time/late maturers | 0.14* (0.07) | 1.29 (0.08) | 1.05 (0.03) | 0.14 (0.06) | 1.25 (0.10) |
Total sample | 0.30 (0.08) | 1.24 (0.09) | 1.05 (0.03) | 0.14 (0.06) | 1.27 (0.11) |
Moderate high | |||||
Early maturers | 1.01*** (0.16) | 1.91 (0.20) | 1.36** (0.06) | 0.71** (0.13) | 1.96 (0.22) |
On-time/late maturers | 0.26*** (0.10) | 1.67 (0.12) | 1.13** (0.04) | 0.30** (0.08) | 1.57 (0.13) |
Total sample | 0.63 (0.09) | 1.79 (0.11) | 1.24 (0.04) | 0.50 (0.08) | 1.77 (0.13) |
High | |||||
Early maturers | 0.44 (0.19) | 1.84 (0.23) | 1.14 (0.07) | 0.61 (0.15) | 1.77 (0.25) |
On-time/late maturers | 0.60 (0.11) | 1.80 (0.14) | 1.13 (0.05) | 0.45 (0.10) | 1.74 (0.16) |
Total sample | 0.52 (0.11) | 1.82 (0.13) | 1.13 (0.04) | 0.53 (0.09) | 1.75 (0.15) |
Very high | |||||
Early maturers | 2.46*** (0.59) | 4.52*** (0.73) | 2.50*** (0.24) | 2.48*** (0.49) | 6.07*** (0.82) |
On-time/late maturers | 0.08*** (0.30) | 1.55*** (0.37) | 0.99*** (0.12) | –0.02*** (0.25) | 0.82*** (0.41) |
Total sample | 1.27 (0.33) | 3.03 (0.41) | 1.74 (0.13) | 1.23 (0.28) | 3.45 (0.46) |
Total sample | |||||
Early maturers | 0.76*** (0.12) | 1.97*** (0.14) | 1.36*** (0.05) | 0.71*** (0.10) | 2.23*** (0.16) |
On-time/late maturers | 0.23*** (0.06) | 1.41*** (0.08) | 1.07*** (0.03) | 0.20*** (0.05) | 1.22*** (0.08) |
Note. Asterisks mark significant differences within trajectory group.
p < .05.
p < .01.
p < .001.
In addition, a significant interaction between pubertal timing and household risk trajectory group was found regarding subjective norms of cigarette use, F(5, 578) = 2.86, p = .015, η = .02. Early maturers also reported significantly higher subjective norms of cigarette use compared with on-time/late maturers in the very high household risk trajectory group only (see Table 3). Early maturers in the very high household risk trajectory group reported that most of their friends and other kids in their school and neighborhood smoked cigarettes or cigars. All other adolescents reported between none and some cigarette use among friends and peers.
Alcohol
There were no significant effects of pubertal timing, household risk trajectory group, or the interaction on alcohol use in the past year. However, a significant interaction between pubertal timing and household risk trajectory group was found regarding binge drinking until sick or falling down, F(5, 577) = 7.71, p < .001, η = .06. Follow-up Bonferroni-corrected pairwise comparisons revealed that early pubertal maturation was associated with higher levels of eighth-grade adolescent binge drinking until sick or falling down only among individuals within the moderate high and very high household risk trajectory groups (see Table 3). The averages reported in Table 3 indicate that early maturers within the very high trajectory group reported binge drinking until sick or falling down between two and three times in the past year. This was in contrast to both early and on-time/late maturers from the other household risk trajectory groups, whose reports of binge drinking until sick or falling down were near zero.
The interaction between pubertal timing and household risk regarding subjective norms of alcohol use was not significant. However, significant main effects of both pubertal timing, F(1, 577) = 11.73, p = .001, η = .02, and household risk trajectory group, F(5, 577) = 4.76, p < .001, η = .04, were found regarding subjective norms for alcohol consumption among friends and other kids from an adolescent's school and neighborhood. Early maturers (M = 2.27, SD = 0.18) reported that more of their friends and peers had drank an entire drink of alcohol compared with on-time/late maturers (M = 1.58, SD = 0.09). In addition, individuals from the moderate high (M = 2.18, SD = 0.14), high (M = 1.99, SD = 0.17), and very high (M = 2.79, SD = 0.51) household risk trajectory groups had significantly higher subjective norms regarding drinking an entire drink of alcohol compared with the none (M = 1.44, SD = 0.18), low (M = 1.62, SD = 0.13), and moderate (M = 1.52, SD = 0.12) household risk trajectory groups. Notably, adolescents in the very high household risk trajectory group reported the highest consumption of alcohol.
Marijuana
A significant interaction between pubertal timing and household risk trajectory group was found regarding the adolescent's own marijuana use in the past year, F(5, 577) = 4.79, p < .001, η = .04. Follow-up Bonferroni-corrected pairwise comparisons revealed that early pubertal maturation was associated with higher levels of eighth-grade adolescent marijuana use only among individuals within the moderate high and very high household risk trajectory groups (see Table 3). The averages reported in Table 3 indicate that early maturers within the very high trajectory group reported using marijuana a couple of times in the past year, with an average of 3–5 times in the past month. This was in contrast to both early and on-time/late maturers from the other household risk trajectory groups, whose reports of marijuana use were near zero.
Similarly, a significant interaction between pubertal timing and household risk trajectory group was found regarding subjective norms of marijuana use among friends and other kids from the adolescent's school/neighborhood, F(5, 578) = 6.57, p < .001, η = .05. Follow-up Bonferroni-corrected pairwise comparisons revealed that early pubertal maturation was associated with higher subjective norms of marijuana use only among individuals within the very high household risk trajectory group (see Table 3). Early maturers in the very high household risk trajectory group reported that nearly all of their friends and peers used marijuana, whereas all other adolescents reported that none to only some of their friends used marijuana.
Inhalants
There was no significant interaction between pubertal timing and household risk trajectory group on inhalant use in the past year, and there was not a significant main effect of pubertal timing. However, a significant main effect of household risk trajectory group was found for adolescent inhalant use, F(5, 578) = 3.79, p = .002, η = .03. Although reports of inhalant use were quite low in the eighth grade overall, individuals from the high household risk trajectory group (M = 0.32, SD = 0.06) reported significantly higher levels of inhalant use in the past year compared with individuals in the none (M = 0.07, SD = 0.06), low (M = 0.06, SD = 0.04), moderate (M = 0.04, SD = 0.04), and moderate high (M = 0.10, SD = 0.05) household risk trajectory groups. The practical significance of this difference was very slight in the eighth grade, but it did highlight the additional risk for future use of more serious substances among individuals who lived in households characterized by high levels of household risk in early childhood. There was no reported inhalant use among the small sample of individuals in the very high trajectory class. There were no significant effects of pubertal timing, household risk trajectory group, or the interaction on subjective norms of inhalant use.
Sensitivity Analysis
In this study, we attempted to reduce bias in the estimation of pubertal timing effects on substance use by evaluating pubertal timing effects within household risk trajectory group, thus recreating the benefits of random assignment (i.e., individuals differ only in the timing of puberty as if pubertal timing were randomly assigned). The main limitation of trying to recreate balance in an observational study is that balance can only be evaluated for the measured variables included in the study, whereas random assignment creates balance on both observed and unobserved confounding variables. By definition, a confounding variable must be associated with both the independent and dependent variables to introduce bias. For example, a common genetic factor that both influences pubertal timing and substance use could bias the results of this study.
Sensitivity analyses were conducted on the statistically significant substance use outcomes (e.g., cigarette use, binge drinking, and marijuana use) to evaluate what the strength of the association between an unmeasured confound and pubertal timing would have to be to render the significant associations between pubertal timing and substance use nonsignificant (Love, 2008; Rosenbaum & Rubin, 1983a). For the purposes of the sensitivity analyses, all early maturers were 1:1 matched to an on-time/late maturer within trajectory group. Prior to matching, the proportion of early maturers and on-time/late maturers who reported using a particular substance was estimated within each trajectory group for the full sample (e.g., 5% of on-time maturers and 13% of early maturers reported marijuana use in the low household risk trajectory group). Matches were made to ensure that the proportions of early and on-time maturers who reported substance use were maintained within each trajectory group for the matched sample. Individuals were matched to maximize concordance, creating a more conservative estimate of the sensitivity of the pubertal timing effect to unmeasured confounds.
The results of these sensitivity analyses show that an unobserved confound, such as genetics, would not only need to be a near perfect predictor of substance use but would also need to produce between a 1.7 and 2.24 fold increase in the odds of being an early maturer to account for the associations observed in this study. To aid in interpretation of this sensitivity analysis, a recent evaluation of genetic contributions to pubertal development found heritability to be approximately .40 for female adolescents and .74 for male adolescents at 12 years of age. By 14 years of age, heritability increased to .68 for female adolescents and to .92 for male adolescents (Dick, Rose, Pulkkinen, & Kaprio, 2001). Although this is a strong association, particularly for male adolescents, it is important to emphasize that this same genetic factor must also be a near perfect predictor of substance use to account for the observed associations in this study. It is highly unlikely that any confounding variable would meet the criteria established in this sensitivity analysis, thus increasing the confidence that the observed pubertal timing effects in this study are not confounded by an unmeasured variable.
Discussion
In this study, we expand the etiological literature by investigating the impact of early pubertal timing on substance use accounting for the youth's developmental history of household risk characterized by low resources and high conflict. Moderating effects of different developmental trajectories of childhood household risk on the association between pubertal timing and substance use in adolescence were shown, integrating two previously separate yet related lines of research. The significant interactions found in this study between early pubertal timing and etiological predictors of early pubertal timing on the prediction of adolescent substance use highlight the importance of interactions that take place across development between individuals and their environment that ultimately shape future adjustment.
Specifically, early pubertal timing was only associated with eighth-grade cigarette use, serious binge alcohol consumption, and marijuana use among those individuals with a history of high levels of household risk throughout early childhood, and this finding did not differ by gender. These results provide extremely important information regarding our conceptualization of the risk associated with early pubertal maturation. Only 9% of the total sample consisted of early maturers who were from households characterized by high levels of early childhood household risk. This represents 69 early maturers, in contrast to the 123 early maturers who were not exposed to high levels of household risk in early childhood. In particular, the most extreme reports of cigarette, alcohol, and marijuana use occurred among early maturers in the very high household risk trajectory group, which composed only 2% of the sample (21 individuals). One of the implications of this finding is that the significant association between negative adjustment outcomes and early pubertal maturation found in previous studies may be attributed to a small subset of early maturers who have a history of household risk during their early childhood. Such outcomes include depressive disorders and symptoms (e.g., Ge et al., 2001; Graber et al., 1997; Hayward et al., 1997; Stice et al., 2001), conduct disorders (e.g., Graber et al., 1997), and eating disorders and symptoms (e.g., Graber, Brooks-Gunn, Paikoff, & Warren, 1994; Graber et al., 1997; Stice et al., 2001).
In fact, the majority of early maturers in this study did not differ from on-time/late maturers regarding substance initiation in the eighth grade. If this is indeed the case, perhaps it would be best to conceptualize early pubertal timing as another stressor that, when combined cumulatively with stressors of early childhood and adolescence, places individuals at a unique risk for negative adjustment. Caspi and Moffitt (1991) discussed the possibility that early pubertal timing was a stressful transition that accentuated preexisting individual differences in behavioral problems. They found that early maturing girls with a childhood history of behavior problems evinced greater increases in behavior problems in adolescence compared with all other girls. Similarly, Ge et al. (1996) found that early maturing girls reported amplified effects of symptoms of distress and father hostility in early adolescence on emotional distress in late adolescence compared with on-time and late maturing girls. In the current study, we replicated previous findings for girls and extended them to boys by finding that early maturers with a preexisting risk toward substance use reported amplified or accentuated vulnerability associated with early puberty compared with all other individuals. In this study, we expand on previous findings by evaluating the role of contextual indicators of household risk in childhood rather than by evaluating accentuation or amplification of preexisting vulnerabilities toward behavioral problems/emotional adjustment on later adjustment.
It is important to note, however, that we only evaluated eighth-grade substance use in this study. It is possible that early maturation could still confer risk for substance initiation at later ages among individuals with a developmental history of low or moderate household risk. Notably, in this study we also evaluated subjective norms of substance use among friends and peers that have been used in previous research as an indicator or precursor of future substance use (Andrews et al., 2003; Ge et al., 2006). Similar to the findings for substance use, early pubertal maturation was associated with higher subjective norms among individuals within the very high household risk trajectory group only. The findings regarding subjective norms also suggest that early pubertal maturation may only confer risk for negative adjustment among a small subset of early maturers who have a history of household risk during their early childhood. Future research should evaluate pubertal timing effects on substance use through late adolescence and early adulthood among individuals with different histories of household risk in early childhood. Likewise, future research should also evaluate other negative adjustment outcomes associated with early pubertal maturation (e.g., depression, eating disorders, delinquency) to determine whether similar associations between household risk and pubertal timing are conferred for other outcomes.
Moreover, additional mediating mechanisms could potentially explain the associations demonstrated here. For instance, externalizing behaviors often co-occur with substance use, and early maturing adolescents are more likely to demonstrate externalizing behaviors than on-time or late maturing adolescents (Cota-Robles, Neiss, & Rowe, 2002; Ge et al., 2002, 2001; Graber et al., 1997; Lynne et al., 2007; Obeidallah, Brennan, Brooks-Gunn, & Earls, 2004; Williams & Dunlop, 1999). Future research should evaluate potential moderated mediation effects (MacKinnon, 2008). For example, it would be informative for future research to evaluate adolescent aggression as a mediator of the moderated association between household risk and pubertal timing on substance use and other adjustment outcomes.
Many of the main effects in this study should not be interpreted in light of the significant interaction effects. However, the main effects found in this study do support previous research on early pubertal timing as a risk for future substance use. In this study, we found that early pubertal maturation was associated with eighth-grade cigarette, alcohol, and marijuana use. We also evaluated early childhood household risk on adolescent substance use and found, as expected, that households characterized by low resources and high conflict were associated with higher reports of adolescent cigarette, alcohol, marijuana, and inhalant use in the eighth grade. In addition, findings indicate that female adolescents were more vulnerable (in terms of higher substance use) to higher levels of household risk compared with male adolescents. Although the aforementioned findings are important extensions of previous work, the primary objective of this study was to evaluate the moderating influence of early childhood household risk on the association between early pubertal timing and substance initiation in early adolescence.
Furthermore, early pubertal maturation was not significantly associated with exposure to early childhood household risk or individual components of household risk (e.g., poor parent–child relationship quality). However, the percentage of early maturers within each trajectory group did slightly increase as household risk level increased. Early pubertal timing was based on parent report of their adolescent's physical development and was objectively defined as the most physically developed upper quartile of the sample. It would be interesting to determine whether adolescent perceptions of pubertal timing, in addition to the objective indicator of pubertal timing used in this study, differed on the basis of their history of early childhood household risk. Future research should evaluate perceptions of early pubertal timing for differences in associations with negative adjustment outcomes based on one's developmental history of household risk.
The construct of pubertal timing was based on a well validated measure of the development of secondary sex characteristics (Pubertal Development Scale; Petersen et al., 1988) rather than initial changes in gonadotropins that mark the onset of puberty. This was done in part to be consistent with existing literature on the assessment of pubertal timing and also because pubertal timing is a socially and biologically defined construct. As such, it is not the onset of puberty that is of interest but the onset of puberty in relation to peers. In fact, the stage termination model of pubertal timing suggests that adolescents who look older than their peers on the basis of secondary sex characteristics may be placed in more adult-like situations at an earlier age, leading to higher rates of substance use and other problem behaviors (Brooks-Gunn et al., 1985). Although this study did not find statistically significant associations between early childhood household risk and pubertal timing, results of this study do highlight the importance of evaluating one's developmental history of household risk when interpreting pubertal timing effects.
Strengths and Limitations
An advantage gained by using group-based trajectory analysis as opposed to simple mean analyses is that trends in the data appear that would not have necessarily been predicted a priori. This is an innovative technique that provides new insights into human development and allows for a more accurate, albeit still limited, statistical representation of longitudinal data. One specific limitation of this technique is that some information regarding individual variation in trajectories of household risk is lost by aggregating individuals into groups with posterior probabilities that range from 0.60 to 1.00. It is important to emphasize that the classification of individuals to groups does not mean that all individuals follow the group's trajectory in lock step. In this study, we used group-based trajectory analysis as a heuristic device to classify individuals on the basis of their own longitudinal data. The diagnostic evaluations of the six-group model of household risk provide strong evidence that this model adequately represents the individual variation in the sample. In addition, a sensitivity analysis revealed the findings in this study to be robust to changing the classification threshold to 0.50 or greater rather than 0.60.4 A strength of using group-based trajectory analysis is that it allowed for the evaluation of patterns of change in household risk overtime. The stability of household risk observed in this study lends some validity to research in which household risk exposure is only evaluated at a single time point in childhood. Other strengths of this study included a longitudinal sample followed from Grades 1–8 and multiple reporters for many of the constructs of interest.
The construct of household risk was a sum of eight indicators of household risk assessed from both the participants and their parents. This measure assumed equal weight for each of the eight variables. As such, this study is limited in that the cumulative risk factor obscures which specific risk factors a child experiences. The decision to create a cumulative risk index was based on previous research that emphasized that the accumulation of risks is an excellent predictor of problem behavior (Sameroff, Gutman, & Peck, 2003). However, it would be useful for future research to account for perceived severity of each indicator and weight the presence of that indicator by the corresponding report of the severity of that risk.5
Family resources were assessed, in part, on the basis of student eligibility for free or reduced price lunch as an indicator of family income. This is potentially problematic in that parents may under-report income to the school system for the purposes of determining lunch eligibility. In addition, parent-report of relationship quality with the child did have relatively low reports of internal validity on the basis of Cronbach's alphas despite the use of a well-validated measure. This may be due to reduced variability in responses during the early grades.
This study was limited in that the sample was not representative of the general population. As such, future research will determine whether these findings generalize to other populations. On a similar note, the current study is limited in that the sample did not include enough minority participants to allow for meaningful evaluation of race/ethnicity differences. Ethnic differences in the onset of pubertal timing have consistently revealed that African American adolescents tend to mature earlier than their White adolescent counterparts (Herman-Giddens et al., 1997). In addition, there is evidence of differences in prevalence rates of substance use on the basis of race/ethnicity, with White adolescents reporting higher rates of substance use compared with Black adolescents (Mayes & Suchman, 2006). Future studies should evaluate these associations among minority adolescents.
Conclusion
We combined two previously separate lines of research regarding pubertal development using innovative statistical analyses. These results shed light on the dynamic, multidimensional process of human development and open the door to many new research questions regarding interconnections between household risk, pubertal timing, and negative adjustment in adolescence. Moreover, these results identify unique subgroups of individuals (i.e., early maturers with high household risk) who are high risk for early substance initiation and hence more likely to have long term difficulties. Subsequent studies of household risk and puberty may be fruitful in identifying protective factors that can be enhanced via prevention programming for high risk youths.
Acknowledgments
The content of this article is solely the responsibility of the authors and does not necessarily represent the official views of the National Institute on Drug Abuse, the National Institute of Mental Health, or the National Institutes of Health. The project was supported by National Institute on Drug Abuse Grant RO1DA10767 (awarded to Judy A. Andrews). Article preparation was supported by National Institute of Mental Health Grants T32MH018834 (awarded to Nicholas Ialongo) and RO1DA10767.
Appendix.
Diagnostics of Assignment Accuracy
Assigned group | πj | Pj(%) | Ave. PP | OCC |
---|---|---|---|---|
None | 10 | 100 (11) | 0.92 | 103.5 |
Low | 24 | 210 (24) | 0.83 | 15.5 |
Moderate | 31 | 268 (31) | 0.80 | 8.9 |
Moderate high | 19 | 149 (17) | 0.79 | 16.0 |
High | 14 | 130 (15) | 0.85 | 34.8 |
Very high | 2 | 21 (2) | 0.93 | 651.0 |
Note. πj = group membership probability (in percentage); Pj = proportion of sample classified in group based on posterior probability; Ave. PP = average posterior probability; OCC = odds correct classification.
Footnotes
Three continuous variables were created for each grade. These variables were a sum across items for either alcohol, marijuana, or other illegal drug use. To evaluate the degree of overlap between each yearly assessment, we estimated correlations separately for each substance across time, as well as separately within each time point among the three substances. Correlations among alcohol problems, marijuana problems, and other illegal drug use problems at each grade were moderate to high, ranging from .38 to .72. Parents who reported problems due to illegal drug use also tended to report problems associated with marijuana use and alcohol use. Correlations between problems due to marijuana use and problems due to other illegal drug use were stronger than correlations between problems due to alcohol use and either of the aforementioned illegal substance variables. This is in part because of the higher prevalence of reports of problems associated with alcohol use.
The lack of change within trajectory group is attributable to the stability of the constructs used to compose this measure. Two of the eight dichotomous variables that make up the household risk composite are constant across all grades (parental substance use problems and eligibility for free or reduced lunch). An evaluation of the consistency of classification as high risk (the upper quartile) versus low risk (the lower three quartiles) for the remaining six variables reveals relatively stable classification across time. Consistent classification as low risk across time ranged from 95% to 96% for parent education, 66% to 84% for parent relationship quality, 76% to 92% for parent report of relationship quality with child, 77% to 90% for child report of relationship quality with mother, 77% to 87% for child report of relationship quality with father, and 92% to 96% for household structure.
Given that previous research has evaluated the individual impact of stressors such as parent–child conflict and household structure on the probability of early maturation (Belsky et al., 2007), a cross-tabulation analysis was conducted to determine whether there were significant associations between early maturation and the eight individual household risk variables. None of the individual variables were significantly associated with higher rates of early pubertal maturation for the overall sample or separately by gender.
The sensitivity of the results to changes in the posterior probability classification threshold was evaluated by classifying individuals to trajectory groups on the basis of a posterior probability of .50 or greater. The results were the same with only a few exceptions. First, the Bonferroni-corrected follow-up tests for past year cigarette use within the moderate trajectory class and past year marijuana use within the moderate high trajectory class were no longer significant. Second, the interaction term did not reach significance for subjective norms of cigarette use, but the Bonferroni-corrected follow-up tests were still significant. All of the other effects remained significant and maintained the direction of the effect reported in this study. Most importantly, the findings for the very high trajectory group remained the same. This provides some evidence of the robustness of the findings.
ANCOVAs controlling for gender, race/ethnicity, and cohort were conducted to evaluate the associations between the individual indicators of household risk and the eighth-grade substance use outcomes and subjective norms. Of the demographic characteristics, parental education was significantly associated with cigarette use and subjective norms, and inhalant use (ps ranged from .006 to .027) and eligibility for free or reduced lunch were significantly associated with cigarette use and subjective norms, binge drinking, and marijuana use (ps ranged from .007 to .040). Household structure was significantly associated with subjective norms of cigarette use, alcohol use and subjective norms, binge drinking, marijuana use and subjective norms, and inhalant use and subjective norms (ps ranged from <.001 to .048). Of the five household conflict variables, parent report of relationship quality with each other was significantly associated with cigarette use and subjective norms, alcohol use, marijuana use and subjective norms, and inhalant use (ps ranged from .001 to .048). Parent report of relationship quality with the child was associated with cigarette use, alcohol use, and marijuana use and subjective norms (ps ranged from .014 to .035). Child report of relationship quality with the mother was significantly associated with subjective norms of cigarette use, subjective norms of marijuana use, and inhalant use and subjective norms (ps ranged from .021 to .043). Child report of relationship quality with dad was significantly associated with binge drinking, marijuana use, and inhalant use (ps ranged from .010 to .038). Parental substance use problems was significantly associated with subjective norms of marijuana use (p = .038) and inhalant use (p = .008).
Contributor Information
Sarah D. Lynne-Landsman, Department of Mental Health, Johns Hopkins Bloomberg School of Public Health
Julia A. Graber, Department of Psychology, University of Florida
Judy A. Andrews, Oregon Research Institute, Eugene, Oregon
References
- Ajzen I, Fishbein M. Attitudinal and normative variables as predictors of specific behaviors. Journal of Personality and Social Psychology. 1973;27:41–57. [Google Scholar]
- American Psychiatric Association . Diagnostic and statistical manual of mental disorders. 4th ed. Author; Washington, DC: 2000. text rev. [Google Scholar]
- Andersson T, Magnusson D. Biological maturation in adolescence and the development of drinking habits and alcohol abuse among young males: A prospective longitudinal study. Journal of Youth and Adolescence. 1990;19:33–41. doi: 10.1007/BF01539443. [DOI] [PubMed] [Google Scholar]
- Andrews JA. Substance abuse in girls. In: Bell DJ, Foster SL, Mash EJ, editors. Handbook of behavioral and emotional problems in girls. Kluwer Academic/Plenum Press; New York, NY: 2005. pp. 181–209. [Google Scholar]
- Andrews JA, Hampson S, Barckley M. The effect of subjective normative social images of smokers on children's intentions to smoke. Nicotine & Tobacco Research. 2008;10:589–597. doi: 10.1080/14622200801975819. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Andrews JA, Hops H, Duncan SC. Adolescent modeling of parent substance use: The moderating effect of the relationship with the parent. Journal of Family Psychology. 1997;11:259–270. doi: 10.1037/0893-3200.11.3.259. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Andrews JA, Tildesley EA, Hampson SE. Prevalence and transitions in substance use across the school years. 2009. Manuscript in preparation.
- Andrews JA, Tildesley E, Hops H, Duncan S, Severson HH. Elementary school age children's future intentions and use of substances. Journal of Clinical Child and Adolescent Psychology. 2003;32:556–567. doi: 10.1207/S15374424JCCP3204_8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Anthony JC, Petronis KR. Early-onset drug use and risk of later drug problems. Drug and Alcohol Dependence. 1995;40:9–15. doi: 10.1016/0376-8716(95)01194-3. [DOI] [PubMed] [Google Scholar]
- Belsky J. Experience in childhood and the development of reproductive strategies. Acta Psychologica Sinica. 2007;39:454–468. [Google Scholar]
- Belsky J, Steinberg LD, Draper P. Childhood experience, interpersonal development, and reproductive strategy: An evolutionary theory of socialization. Child Development. 1991;62:647–670. doi: 10.1111/j.1467-8624.1991.tb01558.x. [DOI] [PubMed] [Google Scholar]
- Belsky J, Steinberg LD, Houts RM, Friedman SL, DeHart G, Cauffman E, Susman E. Family rearing antecedents of pubertal timing. Child Development. 2007;78:1302–1321. doi: 10.1111/j.1467-8624.2007.01067.x. [DOI] [PubMed] [Google Scholar]
- Brooks-Gunn J, Petersen AC, Eichorn D. The study of maturational timing effects in adolescence. Journal of Youth and Adolescence. 1985;14:149–161. doi: 10.1007/BF02090316. [DOI] [PubMed] [Google Scholar]
- Caspi A, Moffitt TE. Individual differences are accentuated during periods of social change: The sample case of girls at puberty. Journal of Personality and Social Psychology. 1991;61:157–168. doi: 10.1037//0022-3514.61.1.157. [DOI] [PubMed] [Google Scholar]
- Centers for Disease Control and Prevention Youth Risk Behavior Surveillance—United States, 2007. Morbidity & Mortality Weekly Report. 2008;57:1–131. [PubMed] [Google Scholar]
- Cota-Robles S, Neiss M, Rowe DC. The role of puberty in violent and nonviolent delinquency among Anglo American, Mexican American, and African American boys. Journal of Adolescent Research. 2002;17:364–367. [Google Scholar]
- Dick DM, Rose RJ, Pulkkinen L, Kaprio J. Measuring puberty and understanding its impact: A longitudinal study of adolescent twins. Journal of Youth and Adolescence. 2001;30:385–399. [Google Scholar]
- Dick DM, Rose RJ, Viken RJ, Kaprio J. Pubertal timing and substance use: Associations between and within families across late adolescence. Developmental Psychology. 2000;36:180–189. [PubMed] [Google Scholar]
- Ellis BJ. Timing of pubertal maturation in girls: An integrated life history approach. Psychological Bulletin. 2004;130:920–958. doi: 10.1037/0033-2909.130.6.920. [DOI] [PubMed] [Google Scholar]
- Ellis BJ, Garber J. Psychosocial antecedents of variation in girls’ pubertal timing: Maternal depression, stepfather presence, and marital and household risk. Child Development. 2000;71:485–501. doi: 10.1111/1467-8624.00159. [DOI] [PubMed] [Google Scholar]
- Ellis BJ, McFadyen-Ketchum S, Dodge KA, Pettit GS, Bates JE. Quality of early family relationships and individual differences in the timing of pubertal maturation in girls: A longitudinal test of an evolutionary model. Journal of Personality and Social Psychology. 1999;77:387–401. doi: 10.1037//0022-3514.77.2.387. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Federman EB, Costello JE, Angold A, Farmer EMZ, Erkanli A. Development of substance use and psychiatric comorbidity in an epidemiologic study of White and American Indian young adolescents: The Great Smoky Mountain Study. Drug and Alcohol Dependence. 1997;44:69–78. doi: 10.1016/s0376-8716(96)01317-8. [DOI] [PubMed] [Google Scholar]
- Ge X, Brody GH, Conger RD, Simons RL, Murry VM. Contextual amplification of pubertal transition effects on delinquent peer affiliation and externalizing behavior among African American children. Developmental Psychology. 2002;38:42–54. doi: 10.1037//0012-1649.38.1.42. [DOI] [PubMed] [Google Scholar]
- Ge X, Conger RD, Elder GH. Coming of age too early: Pubertal influences on girls’ vulnerability to psychological distress. Child Development. 1996;67:3386–3400. [PubMed] [Google Scholar]
- Ge X, Conger RD, Elder GH. Pubertal transition, stressful life events, and the emergence of gender differences in adolescent depressive symptoms. Developmental Psychology. 2001;37:404–417. doi: 10.1037//0012-1649.37.3.404. [DOI] [PubMed] [Google Scholar]
- Ge X, Jin R, Natsuaki MN, Gibbons FX, Brody GH, Cutrona CE, Simons RL. Pubertal maturation and early substance use risks among African American children. Psychology of Addictive Behaviors. 2006;20:404–414. doi: 10.1037/0893-164X.20.4.404. [DOI] [PubMed] [Google Scholar]
- Graber JA. Puberty in context. In: Hayward C, editor. Gender differences at puberty. Cambridge University Press; New York, NY: 2003. pp. 307–325. [Google Scholar]
- Graber JA, Brooks-Gunn J. Transitions and turning points: Navigating the passage from childhood through adolescence. Developmental Psychology. 1996;32:768–776. [Google Scholar]
- Graber JA, Brooks-Gunn J, Paikoff RL, Warren MP. Prediction of eating problems: An 8-year study of adolescent girls. Developmental Psychology. 1994;30:823–834. [Google Scholar]
- Graber JA, Brooks-Gunn J, Warren MP. The antecedents of menarcheal age: Heredity, family environment, and stressful life events. Child Development. 1995;66:346–359. doi: 10.1111/j.1467-8624.1995.tb00875.x. [DOI] [PubMed] [Google Scholar]
- Graber JA, Lewinsohn PM, Seeley JR, Brooks-Gunn J. Is psychopathology associated with the timing of pubertal development? Journal of the American Academy of Child & Adolescent Psychiatry. 1997;36:1768–1776. doi: 10.1097/00004583-199712000-00026. [DOI] [PubMed] [Google Scholar]
- Graham JW. Missing data analysis: Making it work in the real world. Annual Review of Psychology. 2008;60:1–28. doi: 10.1146/annurev.psych.58.110405.085530. [DOI] [PubMed] [Google Scholar]
- Grant BF, Dawson DA. Age at onset of alcohol use and its association with DSM–IV alcohol abuse and dependence: Results from the national longitudinal alcohol epidemiologic survey. Journal of Substance Abuse. 1997;9:103–110. doi: 10.1016/s0899-3289(97)90009-2. [DOI] [PubMed] [Google Scholar]
- Haviland AM, Nagin DS. Causal inferences with group based trajectory models. Psychometrika. 2005;70:557–578. [Google Scholar]
- Hawkins JD, Catalano RF, Miller JY. Risk and protective factors for alcohol and other drug problems in adolescence and early adulthood: Implications for substance abuse prevention. Psychological Bulletin. 1992;112:64–105. doi: 10.1037/0033-2909.112.1.64. [DOI] [PubMed] [Google Scholar]
- Hayward C, Killen JD, Wilson DM, Hammer LD. Psychiatric risk associated with early puberty in adolescent girls. Journal of the American Academy of Child & Adolescent Psychiatry. 1997;36:255–262. [PubMed] [Google Scholar]
- Herman-Giddens ME, Slora EJ, Wasserman RC, Bourdony CJ, Bhapkar MV, Koch GG, Hasemeier CM. Secondary sexual characteristics and menses in young girls seen in office practice: A study of pediatric research in office settings network. Pediatrics. 1997;99:505–512. doi: 10.1542/peds.99.4.505. [DOI] [PubMed] [Google Scholar]
- Jessor R, Jessor SL. Problem behavior and psychosocial development: A longitudinal study of youth. Academic Press; New York, NY: 1977. [Google Scholar]
- Jones BL, Nagin DS. Advances in group-based trajectory modeling and an SAS procedure for estimating them. Sociological Methods & Research. 2007;35:542–571. [Google Scholar]
- Lanza ST, Collins LM. Pubertal timing and the onset of substance use in females during early adolescence. Prevention Science. 2002;3:69–82. doi: 10.1023/a:1014675410947. [DOI] [PubMed] [Google Scholar]
- Lewinsohn PM, Hops H, Roberts RE, Seeley JR, Andrews JA. Adolescent psychopathology: Prevalence and incidence of depression and other DSM–III–R disorders in high school students. Journal of Abnormal Psychology. 1993;102:133–144. doi: 10.1037//0021-843x.102.1.133. [DOI] [PubMed] [Google Scholar]
- Love TE. Spreadsheet-based sensitivity analysis calculations for matched samples. Center for Health Care Research & Policy, Case Western Reserve University; Cleveland, OH: 2008. Available from http://www.chrp.org/propensity. [Google Scholar]
- Lynne SD, Graber JA, Nichols TR, Brooks-Gunn J, Botvin GJ. Links between pubertal timing, peer influences, and externalizing behaviors among urban students followed through middle school. Journal of Adolescent Health. 2007;40:181.e7–181.e13. doi: 10.1016/j.jadohealth.2006.09.008. [DOI] [PMC free article] [PubMed] [Google Scholar]
- MacKinnon DP. Introduction to statistical mediation analysis. Erlbaum; New York, NY: 2008. [Google Scholar]
- Mayes LC, Suchman NE. Developmental pathways to substance abuse. In: Cicchetti D, Cohen DJ, editors. Developmental psychopathology: Risk, disorder, and adaptation. 2nd ed. Wiley; Hoboken, NJ: 2006. pp. 599–619. [Google Scholar]
- Moffitt TE, Caspi A, Belsky J, Silva PA. Childhood experience and the onset of menarche: A test of a sociobiological model. Child Development. 1992;63:47–58. doi: 10.1111/j.1467-8624.1992.tb03594.x. [DOI] [PubMed] [Google Scholar]
- Nagin DS. Group-based modeling of development. Harvard University Press; Cambridge, MA: 2005. [Google Scholar]
- Obeidallah D, Brennan RT, Brooks-Gunn J, Earls F. Links between pubertal timing and neighborhood contexts: Implications for girls’ violent behavior. Journal of the American Academy of Child & Adolescent Psychiatry. 2004;43:1460–1468. doi: 10.1097/01.chi.0000142667.52062.1e. [DOI] [PubMed] [Google Scholar]
- Oregon Department of Human Services . Public Schools Drug Use Survey: Key findings report. Author; Salem, Oregon: 2000. [Google Scholar]
- Petersen AC, Crockett L, Richards M, Boxer A. A self-report measure of pubertal status: Reliability, validity, and initial norms. Journal of Youth and Adolescence. 1988;17:117–133. doi: 10.1007/BF01537962. [DOI] [PubMed] [Google Scholar]
- Pickles A, Rutter M. Statistical and conceptual models of “turning points” in developmental processes. In: Magnusson D, Bergman LR, Rudinger G, Torestad B, editors. Problems and methods in longitudinal research: Stability and change. Cambridge University Press; New York, NY: 1994. pp. 133–165. [Google Scholar]
- Prinz RJ, Foster S, Kent RN, O'Leary D. Multivariate assessment of conflict in distressed and nondistressed mother–adolescent dyads. Journal of Applied Behavior Analysis. 1979;12:691–700. doi: 10.1901/jaba.1979.12-691. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Rosenbaum PR, Rubin DB. Assessing sensitivity to an unobserved binary covariate in an observational study with binary outcome. Journal of the Royal Statistical Society, Series B. 1983a;45:212–218. [Google Scholar]
- Rosenbaum PR, Rubin DB. The central role of the propensity score in observational studies for causal effects. Biometrika. 1983b;70:41–55. [Google Scholar]
- Rubin DB. Estimating causal effects from large data sets using propensity scores. Annals of Internal Medicine. 1997;127:757–763. doi: 10.7326/0003-4819-127-8_part_2-199710151-00064. [DOI] [PubMed] [Google Scholar]
- Sameroff A, Gutman LM, Peck SC. Adaptation among youth facing multiple risks: Prospective research findings. In: Luthar SS, editor. Resilience and vulnerability: Adaptation in the context of childhood adversities. Cambridge University Press; New York, NY: 2003. pp. 364–391. [Google Scholar]
- Schaie KW. A general model for the study of developmental problems. Psychological Bulletin. 1965;64:92–107. doi: 10.1037/h0022371. [DOI] [PubMed] [Google Scholar]
- Schaie KW. A re-interpretation of age-related changes in cognitive structure and functioning. In: Goulet LR, Baltes PB, editors. Life-span developmental psychology: Research and theory. Academic Press; San Diego, CA: 1970. pp. 485–507. [Google Scholar]
- Spanier GB. Measuring dyadic adjustment: New scales for assessing the quality of marriage and similar dyads. Journal of Marriage and the Family. 1976;38:15–28. [Google Scholar]
- Stattin H, Magnusson D. Paths through life: Vol. 2. Pubertal maturation in female development. Erlbaum; Hillsdale, NJ: 1990. [Google Scholar]
- Stice E, Presnell K, Bearman SK. Relation of early menarche to depression, eating disorders, substance abuse, and comorbid psychopathology among adolescent girls. Developmental Psychology. 2001;37:608–619. doi: 10.1037//0012-1649.37.5.608. [DOI] [PubMed] [Google Scholar]
- Surbey MK. Family composition, stress, and the timing of human menarche. In: Ziegler TE, Bercovitch FB, editors. Socioendocrinology of primate reproduction. Wiley-Liss; New York, NY: 1990. pp. 11–32. [Google Scholar]
- Tremblay L, Frigon J. Precocious puberty in adolescent girls: A biomarker of later psychosocial adjustment problems. Child Psychiatry and Human Development. 2005;36:73–94. doi: 10.1007/s10578-004-3489-2. [DOI] [PubMed] [Google Scholar]
- Tschann JM, Adler NE, Irwin CE, Millstein SG, Turner RA, Kegeles SM. Initiation of substance use in early adolescence: The roles of pubertal timing and emotional distress. Health Psychology. 1994;13:326–333. doi: 10.1037//0278-6133.13.4.326. [DOI] [PubMed] [Google Scholar]
- U.S. Department of Health and Human Services, Substance Abuse and Mental Health Services Administration, Office of Applied Studies . National Household Survey on Drug Abuse, 2000. Author; Rockville, MD: 2001. [Google Scholar]
- Walden B, Iacono WG, McGue M. Trajectories of change in adolescent substance use and symptomatology: Impact of paternal and maternal substance use disorders. Psychology of Addictive Behaviors. 2007;21:35–43. doi: 10.1037/0893-164X.21.1.35. [DOI] [PubMed] [Google Scholar]
- Westling E, Andrews JA, Hampson SE, Peterson M. Pubertal timing and substance use: The effects of gender, parental monitoring and deviant peers. Journal of Adolescent Health. 2008;42:555–563. doi: 10.1016/j.jadohealth.2007.11.002. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Wichstrom L. The impact of pubertal timing on adolescents’ alcohol use. Journal of Research on Adolescence. 2001;11:131–150. [Google Scholar]
- Williams JM, Dunlop LC. Pubertal timing and self-reported delinquency among male adolescents. Journal of Adolescence. 1999;22:157–171. doi: 10.1006/jado.1998.0208. [DOI] [PubMed] [Google Scholar]