DEVELOPMENT AND CORRELATES OF ALCOHOL USE FROM AGES 13–20

Abstract

This study examined alcohol use development from ages 13–20 years. The sample comprised 256 youth (50.4% female; 51.2% White, 48.8% African American) assessed annually for 6 years. A cohort-sequential latent growth model was used to model categorical alcohol use (non-use vs. use). Covariates included gender, race, income, parent marital status, risk taking, spiritual beliefs, parent alcohol use, family alcohol problems, family cohesion, friends’ alcohol use, and normative peer use. The alcohol use trajectory increased steadily with age. Risk taking, friends’ alcohol use, and normative peer use were positively associated with higher initial rates of alcohol use. Initial parent alcohol use and positive change in parents’ and friends’ alcohol use over time were related to an increase in alcohol use from ages 13–20 years.

Adolescent alcohol use continues to be a concern among researchers, preventionists, and public health experts. A large number of individuals begin experimenting with alcohol in adolescence (Pagan, Rose, Viken, Pulkkinen, Kaprio, & Dick, 2006). Adolescent alcohol use is particularly serious because of its risks for negative outcomes during adolescence and throughout the life course. Adolescent alcohol use and early onset of alcohol use are related to negative young adult and adult outcomes such as alcohol abuse, delinquency, accidents, homicides, sexually transmitted diseases (Peterson, Hawkins, Abbott,&Catalano, 1994; Pitkanen, Lyyra, & Pulkkinen, 2005), emotional distress, and ineffective social skills (Ellickson, Tucker, & Klein, 2003; Wechsler, Lee, Kuo, Seibring, Nelson, & Lee, 2002).

Population levels of alcohol use change across different age groups over time. Prospective longitudinal studies that follow the same respondents over time are particularly valuable in documenting these patterns of change. Developmental studies have emphasized the importance of growth models based on representative samples of youth (Schulenberg, O’Malley, Bachman, Wadsworth, & Johnston, 1996) to determine trajectories of alcohol use and to identify factors influencing those trajectories (Barnes, Reifman, Farrell, & Dintcheff, 2000). To date, most of these studies have focused on early to mid-adolescence. While there is a body of literature on alcohol use in late adolescence and early adulthood among high-risk samples (e.g., children of alcoholics; Chassin, Flora, & King, 2004; King & Chassin, 2007), less data exist on alcohol use trajectories during this period among general-population samples.

The developmental period from mid- to late-adolescence and early adulthood is a critical one because this is when population base rates of alcohol use tend to increase. Melchior, Chastang, Goldberg, and Fombonne (2008) found that alcohol use rates among French youth increased dramatically between ages 16–21, with more than 90% of 16-year-olds having used alcohol. The Monitoring the Future Study indicated that about 80% of American adolescents have used alcohol by the end of high school (Johnston, O’Malley, & Bachman, 2001). In late adolescence and early adulthood, individuals also may move from experimentation to more established and problematic alcohol use patterns (Pagan et al., 2006; Schulenberg et al., 1996). Most alcohol research on late adolescence and young adulthood has been conducted with college students, who may not be representative of the young adult population and whose reports of prior use are typically retrospective rather than prospective.

Prior research suggests that individual, family, and peer factors may negatively or positively influence youth alcohol use. Such factors include personal religious/spiritual beliefs, personal risk-taking or sensation-seeking behavior, family alcohol problems, family cohesion, parental alcohol use, peer alcohol use, and association with deviant peers.

Adolescents with stronger religious or spiritual beliefs tend to use less alcohol (Cotton, Zebracki, Rosenthal, Tsevat, & Drotar, 2006; Geppert, Bogenschutz, & Miller, 2007; Wallace & Williams, 1997). Adolescents who generally take risks to achieve stimulation (Hampson, Andrews, & Barckley, 2008; Roberti, 2003; Zuckerman, 1994) also tend to engage in more risky health behaviors, such as alcohol use; risk taking increases in adolescence.

Family alcohol problems and parental alcohol use relate positively to adolescent alcohol use (Chassin, Pitts, DeLucia, & Todd, 1999; Dielman, Butchart, & Shope, 1993), both directly and indirectly, by affecting family relations influencing alcohol use (Patterson, Reid, & Dishion, 2002). Family cohesion and family climate play a protective role in suppressing alcohol consumption among youth, and parenting behaviors influence alcohol use through adolescence and into young adulthood (Fischer, Forthun, Pidcock, & Dowd, 2007; Ham & Hope, 2003).

Peer alcohol use and associations with deviant peers also positively influence adolescent alcohol use (Bray, Adams, Getz, & McQueen, 2003; Nasim, Belgrave, Jagers, Wilson, & Owens, 2007). Bray et al. (2003) found that, from grades 7–11, initial levels of peer alcohol use were related to changes in adolescent alcohol use. Since peer influences increase with age, peers’ alcohol use would be expected to relate to alcohol use in later adolescence and early adulthood.

The purpose of the present study is to document the development and significant correlates of alcohol use through adolescence and young adulthood (ages 13–20 years). A cohort-sequential latent growth model (LGM) was specified to examine alcohol use over time, controlling for gender, race, and family income. Covariates included parent marital status, personal risk taking and religiosity, parent alcohol use, family alcohol problems, family cohesion, peer alcohol use, and perceived peer alcohol use norms. Based on prior research, it was hypothesized that:

1. alcohol use would follow a linear trajectory from ages 13–20 years;

2. no significant effects of gender, race, parent marital status, or income would be found on the alcohol use intercept or slope;

3. less religiosity and greater risk taking would relate to more alcohol use;

4. parent alcohol use and family alcohol problems would be positively related, and family cohesion would be negatively related, to increased alcohol use; and

5. peer alcohol use would be positively related to personal alcohol use.

METHOD

Participants and Procedures

Families in Portland, Oregon, were randomly recruited via telephone using a computer-assisted telephone interviewing system. Of those eligible, 75% agreed to participate. Quotas were established to ensure a final target youth sample with age (9-, 11-, and 13-year-old cohorts), gender (male and female), ethnicity (African American and White), and neighborhood groups equally represented. Recruitment details are presented in Duncan, Strycker, Duncan, He, and Stark (2002).

For the present investigation targeting adolescence to young adulthood, analyses were restricted to the oldest two cohorts at age 13 years or older (study years 3–8). The sample comprised 256 youth from two age cohorts (n = 132 13-year-olds and n = 124 15-year-olds) assessed at six time points (hereinafter referred to as T1–T6). Using a cohort-sequential design, the two age cohorts were combined to represent an 8-year trajectory from ages 13–20. The sample was 50.4% female, 48.8% African American, and 51.2% White.

Prior to the main study, focus groups were conducted to discuss the assessment approach and a pilot study was conducted to determine the feasibility and validity of the assessment process. Assessments took place in participants’ homes under the supervision of a research assistant. Research assistants were trained to follow a standard assessment protocol at all their visits, and served primarily to supervise the completion of questionnaires and to ensure that family members completed the questionnaires separately and confidentially. Families were initially paid $130 for completing the assessment, with a $20 bonus if all eligible family members participated, and each participating child was paid $5. Family payments were increased by $10 each year (T2–T6). Most youth and families (69.9%) completed all six assessments; at least two assessments were available for 96.1% of youth and families.

Measures

Adolescent Alcohol Use

Youth were asked separately for beer, wine/wine coolers, and hard liquor, “How often do you drink (beer/wine/liquor) now?” Responses ranged from 0 = have never used at all to 8 = use 2–3 times per day or more. The maximum frequency among the three alcohol types was retained. This variable was highly skewed, and therefore recoded to form a binary construct with 0 = not currently using alcohol and 1 = currently using alcohol.

Parental Marital Status

Parents/guardians were asked their marital status, scored as either 0 = single/not living in a committed relationship or 1 =married/living in a committed relationship.

Family Household Income

Parents were asked their total household income last year before taxes. Responses were on an 11-point scale that ranged from 1 = under $5,000 to 11 = $90,000 or more. Mean values were used in two-parent households.

Youth Demographic Variables

Youth were asked to identify their sex and race. Sex was coded 0 = male and 1 = female. Race was coded 0 = African American and 1 = White.

Youth Religious/Spiritual Beliefs

Youth were asked about the importance of religious or spiritual beliefs in their lives. Responses ranged from 0 = no religious or spiritual beliefs to 5 = religious or spiritual beliefs are very important.

Youth Risk Taking/Sensation Seeking

Youth were asked how much they agreed or disagreed with risk-taking statements such as “I would do almost anything on a dare,” “I like to do dangerous stuff just for fun,” and “I like to take chances more than other kids my age.” Responses, on a 5-point scale, ranging from 1 = strongly disagree to 5 = strongly agree, were averaged. Reliability analyses showed a coefficient alpha of .72 and 1-year test-retest of r = .69.

Parent/Guardian Alcohol Use

Parents/guardians reported on their own alcohol use, defined as maximum use of beer, wine, and hard liquor. In two-parent households, maximum values were used. Responses ranged from 1 = never used to 9 = use 2–3 times per day or more.

Family Alcohol Problems

Parents/guardians were asked, “Have any members of your family ever had a serious problem with drinking?” for each of the following: brothers or sisters, mother or father, grandparents, children, and partner. Responses were 0 = no problems or 1 = problems. Maximum values were used in two-parent households. The maximum value across family members was used to create the family alcohol problem construct coded 0 = no family history of problem drinking and 1 = family history of problem drinking.

Family Cohesion

Family cohesion was measured using the 10-item FACES scale (Olson, Portner, & Lavee, 1985). Items included “Family members ask for each other’s help” and “Family members feel very close to each other.” Respondents described how often these occur in the family on a scale ranging from 1 = hardly ever to 5 = almost always. Reliability analyses showed a coefficient alpha of .88 for parent/guardian and youth scores.

Peer Alcohol Use

Youth were asked, “How often in the past year did your friends drink alcohol?” Responses ranged from 0 = did not use at all to 8 = used 2–3 times per day or more.

Alcohol Use Norms among School Peers

Youth were asked, “How many of the kids at your school do you think drink alcohol?” with alcohol defined as beer, wine, or hard liquor. Responses were on a 5-point scale ranging from 1 = very few to 5 = almost all.

Table 1 presents descriptive statistics for all baseline (T1) measures and change scores (computed by subtracting T1 values from T6). The change scores represent observed change in the predictors over the study period. Although measures were available at each time point, the use of individual predictors across time would have resulted in overly complex models. Therefore, in addition to baseline measures, change scores also functioned as correlates of adolescent alcohol use trajectories (see Kessler and Greenberg (1981) for details on change scores).

Table 1.

Descriptive Statistics for All Baseline Measures

	Mean (SD) or %
% female youth	50.4
% African-American youth	48.8
% two-parent household income	59.8
Parent-reported household income (%)
Under $5,000	4.7
$5,000 to $9,9999	9.8
$10,000 to $19,999	9.4
$20,000 to $29,999	12.1
$30,000 to $39,999	12.5
$40,000 to $49,999	11.3
$50,000 to $59,999	10.2
$60,000 to $69,999	5.1
$70,000 to $79,999	6.3
$80,000 to $89,999	4.7
$90,000 or greater	14.1
T1 Measures
Family alcohol problems	29.3
Parent alcohol use (Scale: 1–9)	4.27 (2.28)
Family cohesion (Scale: 1–5)
Parent report	3.77 (0.61)
Child report	3.44 (0.73)
Peer alcohol use (Scale: 0–8)	1.28 (1.79)
Peer alcohol use norms (Scale: 1–5)	2.27 (1.17)
Youth risk taking (Scale: 1–5)	2.37 (0.88)
T1 to T6 Change Scores (T6 minus T1)
Parent alcohol use	0.25 (1.80)
Family cohesion
Parent report	−0.18 (0.52)
Child report	−0.13 (0.95)
Peer alcohol use	2.52 (2.64)
Peer alcohol use norms	1.36 (1.47)
Youth risk taking	−0.10 (0.88)
Youth religious/spiritual beliefs	−0.44 (1.46)

Analytic Model

Recent modeling techniques incorporate both group- and individual-level characteristics to describe change over time. These random coefficient methods (e.g., multilevel models, hierarchical linear models) are able to describe an individual’s developmental trajectory and to capture individual differences in the trajectories over time. The latent growth model (LGM; Meredith &Tisak, 1990; Rao, 1958; Tucker, 1958) is a particularly flexible random coefficient model that was developed within the structural equation modeling framework. LGM approaches can accommodate many complexities in describing change over time, such as growth mixture (e.g., Muthén, 2001) and piecewise growth (e.g., Duncan & Duncan, 2004) models.

LGM techniques were originally developed for normally distributed continuous outcome measures. Muthén (1996, 2001) has extended LGM approaches to accommodate binary and ordered categorical outcomes via the probit model within the random effects model framework. The present study models youth alcohol use as a binary outcome, coded 0 = not currently using alcohol and 1 = currently using alcohol.

Cohort-Sequential LGM

The concept of “convergence,” first proposed by Bell (1953), maximizes the advantages of the longitudinal study design while minimizing its problems (e.g., subject attrition, costs). With this method, multiple independent cohorts assessed at staggered but partially overlapping time points are linked to create a common trajectory. Fewer repeated measures are required than in a traditional longitudinal design to cover the period under study. This “cohort-sequential” design (Nesselroade & Baltes, 1979) has gained popularity in the LGM context (Duncan, Duncan, & Strycker, 2006; Meredith & Tisak, 1990), and has been applied to numerous research areas (e.g., Baer & Schmitz, 2000; Caskie, Schaie, & Willis, 1999; Tildesley & Andrews, 2008; Watt, 2008).

In the present study, the two staggered overlapping age cohorts were linked to construct a trajectory of alcohol use from ages 13–20 years (Figure 1). Across 6 years of data collection, cohort 1 represents the 13-year-old cohort with data for ages 13, 14, 15, 16, 17, and 18 years and cohort 2 represents the 15-year-old cohort with data for ages 15, 16, 17, 18, 19, and 20 years. In the figure, F1 represents Factor 1, the intercept or initial level of alcohol use, and F2 represents Factor 2, the slope or change in alcohol use across time. The factor loadings of the intercept are constant for individuals across time and are fixed at 1. The slope factor loadings are fixed at 0 to 7 to represent linear growth from ages 13–20 years with the intercept located at age 13. The slope mean (M_s) and variance (D_s), intercept variance (D_i), and correlation between the intercept and slope (R_is) are estimated parameters. The mean of the intercept (M_i) is fixed at 0 since location parameters are captured by the thresholds associated with the binary dependent variables (Muthén & Asparouhov, 2002).

Figure 1.

The cohort-sequential LGM for change in youth alcohol use from ages 13–20 years. Notes: Over 6 years of data collection, cohort 1 represents the 13-year-old cohort with data for ages 13, 14, 15, 16, 17, and 18 years; cohort 2 represents the 15-year-old cohort with data for ages 15, 16, 17, 18, 19, and 20 years. Boxes with dotted lines represent the “missing” data by design for each cohort. F1 represents Factor 1, the alcohol use intercept, and F2 represents Factor 2, the alcohol use slope. The mean of the slope, M_s, and variances of the intercept, D_i, and slope, D_s, as well as the correlation, R_is, between the intercept and slope, are estimated. The mean of the intercept, M_i, is fixed at 0. Thresholds for each observed variable are estimated and constrained to be equal across time and cohort. A linear growth trajectory is specified by fixing the factor loadings on the slope from 0–7 across ages 13–20 years.

For the binary dependent variables, threshold parameters (τ) are specified to represent category intervals. For each two-level ordered binary dependent variable, there is one threshold parameter. The threshold parameters allow translation of the binary scale into a continuous scale and replace the measurement errors (e) typically displayed for continuous outcomes. The thresholds were fixed to be equal across time to ensure measurement invariance.

The same developmental model is assumed for each cohort to test the feasibility of a common trajectory across the 8-year age span. The cohort-sequential models were estimated within a multiple-group design to impose the cross-cohort equality constraints necessary to evaluate invariance. All free parameters were constrained to be equal across groups except for the scale factors associated with non-overlapping (across cohorts) dependent variables. Scale factors account for possible differences in variance across groups (see Muthén and Christoffersson (1981) for a technical description of factor scores) and the first scaling factor was fixed at unity for a reference point.

A two-step process was used to test the LGM. First, an unconditional model was specified to identify the functional form of the growth parameters and to evaluate the amount of variation in the intercept and slope. Next, a conditional model introducing hypothesized predictors of the intercept and slope was estimated, and backwards elimination was used to derive the most parsimonious model. Due to the categorical nature of the data, models were estimated using a weighted least squares robust estimator. Model fit was assessed using a chi-square goodness of fit test; comparative fit index (CFI; Bentler, 1990), with values greater than .90 indicating an acceptable model fit; and the root mean square error of approximation (RMSEA; Steiger & Lind, 1980), with values less than .05 indicating acceptable model fit.

All LGM models were conducted with Mplus software version 5.1 (Muthén & Muthén, 2004), which uses the expectation maximum algorithm to account for missing data in dependent outcome measures and list-wise deletion for independent predictors. The amount of missing data on the alcohol use outcome variable was 0.0%, 4.3%, 9.4%, 14.8%, 23.0%, and 26.6% at T1 through T6, respectively. Missing data for the predictors was minimal at the T1 assessment, with an average of 0.5% missing data across all predictors, but increased over time, with an average of 4.0% missing at T2, 8.6% at T3, 12.3% at T4, 22.2% at T5, and 26.6% at T6. A multiple-imputation strategy was used to address missing data for the predictors. Multiple imputations are preferred to single imputation, which systematically underestimates variance (Rubin, 1987). NORM software (Schafer, 1997) was used to impute 10 complete datasets; pooling of the parameter estimates was accomplished with Mplus imputation algorithms. Rubin (1987) showed that an estimate based on five imputed datasets with 50% missing information has only a 5% greater standard deviation then an estimate based on an infinite number of datasets. Unless missing data rates are exceptionally high, there is little to be gained with more than 5–10 datasets (Schafer, 1999). Missing data patterns at follow-up were not associated (p < .05) with baseline measures of the study outcome or demographic characteristics, indicating that the missing at random (MAR) assumption was tenable (Rubin, 1987).

RESULTS

Unconditional Linear Model

The model depicted in Figure 1 was specified (χ²[30, N = 256] = 25.42, p = .670, CFI = 1.00, RMSEA < .001). Next, a 1-degree-of-freedom Wald test was used to assess the appropriateness of cross-cohort equality constraints imposed on the mean slope, slope variance, intercept variance, and slope and intercept covariance. The non-significant Wald tests, along with adequate fit statistics, suggested that a common linear trajectory in alcohol use across the three cohorts was tenable. Finally, a higher-order quadratic term was tested, but degradation in model fit and lack of significant variance in the rate of change over time did not support the inclusion of the non-linear trend. The linear trend was also evidenced by examination of the proportion of youth alcohol use and by visual inspection of change in youth alcohol use across time, cohort, and the binary alcohol-use categories (Figure 2). The linear growth model resulted in a significant mean slope (M_s = 1.46, SE = .228, t = 6.38, p < .001), indicating an increase in alcohol use from ages 13–20. Significant intercept variance (D_i = 0.75, SE = .138, t = 5.45, p < .001) and slope variance (D_s = 2.14, SE = .495, t = 4.32, p < .001) indicated that adolescents differed in their initial alcohol use at age 13 (intercept) and in individual growth from ages 13–20 (slope).

Figure 2.

Proportion of alcohol use by study cohort. Over 6 years of data collection, cohort 1 represents the 13-year-old cohort with data for ages 13, 14, 15, 16, 17, and 18 years; cohort 2 represents the 15-year-old cohort with data for ages 15, 16, 17, 18, 19, and 20 years.

Conditional Linear Model

Significant variance in the intercept and slope parameters warranted examination of predictors to help explain differences in initial and changing alcohol use from ages 13–20. T1 measures were modeled as predictors of the intercept, and T1 measures and change scores (T6 minus T1 for measures of ordinal level or continuous scale) were used as predictors of the slope. Initially, all predictors were entered into the model. Predictors considered least influential (based on p values) on the intercept and slope were removed one at a time and the model re-estimated. This iterative backwards-elimination process continued until all remaining predictors showed a significant relationship (p < .05) with either the intercept or slope factor. To control for the effects of gender, race, and family income, these variables remained in the model regardless of p value. The final, parsimonious model showed acceptable fit (χ²[124, N = 256] = 117.16, p = .876; CFI = 0.99; RMSEA = .003) and retained five significant predictors of the intercept or slope (see Table 2).

Table 2.

Intercept and Slope Parameter Estimates for the Final Conditional Model

	Coefficient	SE	t Value	p Value
Intercept
Race	0.40	0.22	1.80	.071
Sex	0.13	0.17	0.72	.469
Income	−0.04	0.04	−1.25	.213
T1 youth risk score	0.17	0.06	2.89	.004*
T1 normative peer use	0.21	0.08	2.59	.010*
T1 friend alcohol use score	0.16	0.06	2.57	.010*
Slope
Race	−0.62	0.48	−1.29	.198
Sex	−0.04	0.36	−0.12	.904
Income	0.08	0.08	1.03	.305
T1 normative peer use	−0.41	0.17	−2.36	.018*
T1 parent alcohol use score	0.10	0.05	2.02	.044*
Parent alcohol change score	0.15	0.06	2.30	.021*
T1 friend alcohol use score	−0.12	0.14	−0.88	.380
Friend alcohol use change score	0.19	0.06	3.12	.002*

^*Significant at p < .05 or less.

Baseline risk taking, friends’ alcohol use, and peer alcohol normative use explained significant variation in initial alcohol use (F1); risk-taking youth, those whose friends use more alcohol, and youth who perceived greater use of alcohol by peers in their school showed higher alcohol-use rates at age 13. Peer alcohol normative use and parent alcohol use at T1 also accounted for significant variation. in the slope factor (F2) but in different directions; higher alcohol use by parents at baseline was associated with increased youth alcohol use over time, whereas higher perceived use of alcohol among peers at T1 was related to less of an increase in youth alcohol use over time. Positive change in both parents’ and friends’ alcohol use over time (change scores) was significantly related to increases in youth alcohol use from ages 13–20. Race, sex, and family income at T1 were not significantly associated with the intercept or slope factors. However, race showed a trend (p = .071), with the intercept indicating higher alcohol use by White vs. African American participants at age 13.

DISCUSSION

The present study employed a cohort-sequential design and LGM analyses to document change in alcohol use spanning 8 years from adolescence to early adulthood. Results support the accelerated design procedure as an efficient method for examining longitudinal data (McArdle & Hamagami, 1992; Raudenbush & Chan, 1992).

An upward linear trend in alcohol use was found between the ages of 13–20 years, with greater proportions of alcohol users as the sample aged. This finding is consistent with prior studies on adolescent alcohol use (e.g., Bray et al., 2003). After controlling for demographic variables of gender, race, and income, individual risk taking, friends’ alcohol use, and normative peer use emerged as significant positive predictors of initial alcohol use (age 13). Youth with higher levels of risk taking/sensation seeking, whose friends used alcohol, and who perceived higher use of alcohol by peers in their school were more likely to be alcohol users at age 13. Parent alcohol use at T1, and positive change in parents’ and friends’ alcohol use over time were significantly and positively related to the slope of alcohol use over the 13- to 20-year age span; those individuals whose parents used more alcohol when they were 13 and whose parents and friends increased in their alcohol use over time showed a steeper alcohol use trajectory from ages 13–20.

These findings support prior research demonstrating the relationship between risk taking and substance use in adolescence and young adulthood (Hampson et al., 2008). Future research should examine whether risk taking is mediated by other factors such as deviant peers (Romer & Hennessy, 2007). In addition, it would be helpful for future studies to specifically assess cause-and-effect questions about risk taking and early substance use. For example, to what degree does longevity of usage and amount of alcohol consumption impact the degree and type of risk taking? Findings also underscore the importance of peer influences (friends’ use and normative peer use) on alcohol use (Bray et al., 2003), both in initial use and in changing use during adolescence and young adulthood. Such findings point to the need for intervening with peer groups to reduce the incidence and escalation of alcohol use during later adolescence, and for developing interventions to help protect individuals against negative peer influences (Nasim et al., 2007).

It is not clear why results showed a negative effect of T1 normative peer use on the alcohol slope. Given that normative peer use had a significant positive effect on initial alcohol use (age 13), this finding may be due to a ceiling effect or indicate that normative peer use (perceived use of alcohol by peers in one’s school) is more influential on alcohol use at younger ages. That is, among 13-year-olds, higher perceived use of alcohol by school-age peers was positively related to alcohol use. However, as participants aged, the initial perception of higher normative peer alcohol use was related to less of an increase in alcohol use over time. Alternatively, increases in friends’ and parents’ use over time were positively related to increased alcohol use during later adolescence and early adulthood.

In this study, parent use of alcohol at the initial time point (when youth were 13) and increasing parent use over time emerged as significant correlates of adolescent and young adult alcohol use. Importantly, higher levels of parent alcohol use and increasing parental use of alcohol over time were associated with increasing youth alcohol use from ages 13–20 years. This result would be expected based on social learning theories stating that youth alcohol use behavior is often acquired from role models such as parents (Andrews, Hops, & Duncan, 1997) and on previous research indicating that most youth first use alcohol when parents or family members are present (Strycker, Duncan, & Pickering, 2003). The current study supports recent research pointing to the importance of parent behavior on youth alcohol use even into young adulthood (Fischer et al., 2007; Ham & Hope, 2003), and demonstrates that both parents’ and friends’ alcohol use can encourage or discourage youth alcohol use.

Because no significant differences in alcohol use patterns were found between girls and boys or across ethnicities, multiple-sample analyses were not employed. Instead, these demographic variables were controlled for in the model. No significant effects were found for gender, race, family income, or parent marital status on the intercept or slope of youth alcohol use. Research on the effects of demographic variables on alcohol use has produced mixed results. For example, Johnston et al. (2001) found that boys generally use more alcohol than girls, possibly with a diminishing difference over time. Studies examining the relationship between alcohol use and socioeconomic status also have produced equivocal findings, making it difficult for clear trends to emerge (Oesterle, Hill, Hawkins, & Abbott, 2008). Past studies have often shown that alcohol use is more prevalent among White than among African-American youth (Barnes et al., 2000), but most of these studies have focused on younger age groups rather than later adolescence and early adulthood. It is possible that boys and girls, and African-American and White young people, become more similar in their alcohol use patterns as they enter later adolescence and young adulthood.

This study has strengths and limitations. Despite fairly high participation rates, the study is limited by its agreement rate bias (75%) and in terms of the covariates used. Genetic and other personal and social contextual factors may need to be considered when studying alcohol use during this developmental period (Conger, 1997; Gottfredson & Koper, 1996). The study focused on use vs. non-use among a general population sample; a focus on alcohol abuse and/or high-risk samples may yield different results. Also, an additive model is presented here, and it is possible that the covariates may exert their effects in a more interactive manner. Strengths of the study include the randomly recruited sample, a cohort-sequential LGM to determine growth in alcohol use from ages 13–20, and use of parent and youth report data. Study participants are representative of the socioeconomic status of the county and state from which they were recruited. A particular strength of this study is its focus on the entire adolescent period reaching into young adulthood. Most studies of this age group have been conducted with college students, limiting their generalizability. Future research on late adolescence and young adulthood should continue to collect data from different and representative samples of young people.

The World Health Organization (WHO) estimates that alcohol-use disorders affect 76.3 million people worldwide (WHO, 2004). Adolescence is a critical period for the onset and development (increase) of alcohol use, and early adulthood is a time when alcohol use patterns may become more established. Future research should continue to: focus on trajectories of alcohol use across childhood, adolescence, and early adulthood; document patterns of change across genders and ethnicities; and identify the risk and protective factors influencing these trajectories so that appropriate prevention and intervention strategies can be targeted to individuals and groups most at risk.