Perceived Friends’ Use as a Risk Factor for Marijuana Use Across Young Adulthood

Abstract

Perceived social norms of substance use are commonly identified as a risk factor for use. How the strength of association between perceived friends’ use and substance use may change across development has not yet been documented. The current analysis considers how the associations between perceived friends’ marijuana use and participants’ own any marijuana use in the past year changes from ages 18 to 30 using longitudinal data from the U.S. national Monitoring the Future study from 1976 to 2014 (N=30,794 people). Time-varying effect modeling (TVEM) was used to examine the associations between perceived friends’ use of marijuana and participants’ own annual marijuana use by age, as well as the extent to which these time-varying associations were moderated by sex, race/ethnicity, and parental education. Associations between perceived friends’ use and own marijuana use increased with age. In addition, the association between perceived friends’ use and own marijuana use significantly varied by demographic groups, such that it was significantly greater for men from ages 19 to 24 and from ages 27 to 30, compared to women; for whites, compared to other race/ethnicities, across all ages; and for individuals whose parents attended college, compared to those whose parents had a high school education or less, across all ages. Results suggest that perceived friends’ marijuana use becomes an even more important marker for increased marijuana use as people age through young adulthood. Therefore, the role of peers in substance use remains crucial beyond adolescence and should be incorporated into intervention strategies for young adults.

Marijuana is the second most widely used drug in the U.S., following only alcohol in its prevalence (Johnston, O’Malley, Bachman, Schulenberg, & Miech, 2015; Substance Abuse and Mental Health Services Administration, 2014). In 2014, 65% of young adults aged 29/30 in the U.S. reported using marijuana in their lifetime (Johnston et al., 2015). Marijuana has negative consequences including lung infections, increased heart rate, hallucinations, and addiction (National Institute on Drug Abuse, 2015), and adolescent use in particular appears to have especially high risks for addiction (Anthony, Warner, & Kessler, 1994; Anthony & Petronis, 1995) and impaired cognitive functioning (Meier et al., 2012). Differences in marijuana use by age have been documented such that marijuana use tends to be most prevalent among those in their late teens and early twenties, followed by decreases into young adulthood (Chen & Jacobson, 2012; Johnston et al., 2015).

As prevalence of marijuana use changes across young adulthood, the importance of various risk factors may also change. Understanding how the strength of the association between various risk and protective factors and risk behaviors across young adulthood may vary with age will indicate which risk factors are most salient at which developmental time periods and aid in the creation of interventions that target or address the most critical factors for individuals at particular ages (Coyle & DiClemente, 2014; Vasilenko & Lanza, 2014).

One of the most consistent predictors of marijuana use is the presence of marijuana-using peers (e.g., Hawkins, Catalano, & Miller, 1992; Kandel & Davies, 1991; Kandel, Kessler, & Margulies, 1978). Perceived descriptive norms—such as the level of marijuana use in which an individual believes her/his peers engage—are theoretically and empirically linked to an individual’s own behavior. The theories of reasoned action (Ajzen & Fishbein, 1980; Fishbein & Ajzen, 1975) and planned behavior (Ajzen, 1985) describe the importance of attitudes and group norms on a range of behaviors. Social norms theory (see Berkowitz, 2004; Perkins, 2003), widely used in substance use research, contends that perceptions of peer behavior have direct impact on individual behavior, regardless of how accurate the perceived friends’ use is.

Empirical support for the association between marijuana use among peers and individuals’ marijuana use has focused almost exclusively on adolescents and college students. There is evidence that having a greater number of marijuana-using friends in high school predicts an individual’s own use (Keyes et al., 2011; Skinner & Cattarello, 1989; White et al., 2006) and that peer use and individual use are related among college students (Kilmer et al., 2006; Neighbors, Geisner, & Lee, 2008; Page & Roland, 2004; Simons, Neal, & Gaher, 2006). More generally, normative beliefs of substance use rates have been a central target for numerous adolescent and young adult interventions (Larimer & Cronce, 2002; Larimer & Cronce, 2007).

Less research has examined the importance of peer factors across young adulthood; existing research has found weaker support for peer influences on marijuana use than on cigarette use and alcohol use (Andrews, Tildesley, Hops, & Li, 2002). Still, the few studies available show that peer marijuana use remains an important predictor of individuals’ marijuana use in the mid-twenties (Andrews et al., 2002; Dishion & Owen, 2002; Kandel, 1984; Windle & Wiesner, 2004). Studies that do look at changes by age often break age into arbitrary groups that may obscure more nuanced results and make it difficult to identify critical periods. For example, a study by Dull (1983) examined correlations between friends’ drug use and individual’s own drug use and found no differences in the strength of the correlation for those aged 17–25 and 31–40. This suggests that the centrality of friends’ drug use as an explanatory variable remains into adulthood and additional research should explore these associations. In a study of general risk-taking, Gardner and Steinberg (2005) found variation in the effects of peers by age in the laboratory when comparing the performance of three age-defined groups (with mean ages 14, 19, and 37). A more nuanced look at the association of perceptions of peer behavior with risk behaviors across age, including findings beyond the laboratory and influences specific to marijuana use, is needed.

The Current Study

Perceived social norms of substance use are commonly identified as a risk factor for marijuana use among adolescents and college students. However, the extent to which perceived friends’ use is a stronger or weaker risk factor for marijuana use at different ages has not yet been documented. Typical analytic models for measuring the influence of social norms have constrained researchers to assuming that risk factors have particular effects that remain the same over time. However, an examination of changing associations between risk factors for substance use and behaviors may elucidate age-appropriate targets of behavioral interventions. Therefore, the current analysis considers how the associations between perceived friends’ marijuana use and own marijuana use change from ages 18 to 30. This is the first study to document the age-related changes in the associations between perceived friends’ marijuana use and individuals’ marijuana use across young adulthood. Research questions for the current study are: (1) How do associations between perceived friends’ marijuana use and own marijuana use vary across young adulthood? (2) Are these effects moderated by (i) sex, (ii) race/ethnicity, or (iii) parental education?

Method

Sample

This study uses longitudinal data from Monitoring the Future, an ongoing study of adolescents and young adults (Johnston et al., 2015). Each year since 1976, nationally representative samples of about 16,000 high school seniors have been surveyed in their classrooms. Approximately 2,400 individuals are randomly selected from each senior year cohort for biennial follow-up via mailed questionnaires. The biennial follow-ups begin one year after high school completion for one random half of the selected cohort and two years after high school for the other half, so that surveys are completed at modal ages 19/20, 21/22, 23/24, 25/26, 27/28, and 29/30 (Johnston et al., 2015).

Monitoring the Future study data collected in 1976 to 2014 were used in this study. Participants were high school seniors from 1976 to 2013 who were selected to participate in the longitudinal study. Respondents included in this analysis were followed longitudinally from age 18 to as old as age 30, and provided data on perceived friends’ use of marijuana and their own marijuana use for at least one follow-up time point. The question about perceived friends’ use of marijuana was asked on four of the six randomly-distributed questionnaire forms. The question was introduced on one form in 1976, one additional form in 1989, and two additional forms in 1990. Compared to those who provided at least one follow-up wave of data, those who were selected for follow-up but provided only age 18 data were more likely to be male; be white; have parents with a high school education or less; report past-30-day use of cigarettes, alcohol and marijuana; and report that they perceived more of their friends were marijuana users. Though statistically significant, the effects of attrition are small in magnitude (i.e., R²=.01 for sex, R²=.02 for race/ethnicity, and R² <.005 for all other variables). Analytic Ns range from 30,013 to 30,794, depending on the model (with 169,158 to 173,427 person-waves). The analytic sample is 55% female, 75% White, 8% African American, 8% Hispanic, and 9% Other races; 69% of the sample report having one or both parents with at least some college education. Age in years was based on the sample modal age at each data collection point, resulting in coverage of every age from 18 to 30.

Due to the study design with recruitment in high school, there are fewer measurements at older ages because participants have not yet aged. In order to investigate potential cohort effects we compared individuals who completed up to half of the study ages to those who completed more than half of possible ages. The association of marijuana use and perceived friends use at age 18 was quite stable comparing more recent cohorts (ages 18–24 at last time point; r=0.62) to older cohorts (ages 25–30 at last time point; r=0.65). We also examined correlations among cohorts recruited before and after the policy changes toward legal medical marijuana use (first passed in 1996 in California) and recreational marijuana use (first passed in 2012 in Washington and Colorado). Results suggest that there are not strong cohort differences for the correlations between perceived friends’ use and own marijuana use (e.g., among age 18 cohorts 1976–1995: r=0.66 vs. 1996–2013, r=0.62; 1976–2011: r=0.64; 2012–2013, r=0.62).

Measures

Demographics

Variables assessed at age 18 included sex (male=1, female=0), race/ethnicity (coded as White [reference group], Black, Hispanic, and Other [including those who selected more than one option and those who chose not to answer]), and parental education (as a proxy for socioeconomic status; coded for the highest level of schooling completed by mother or father as 1=some college or more or 0=high school education or less).

Marijuana Use

Annual marijuana use at each wave was assessed with the question, “On how many occasions (if any) have you used marijuana (grass, pot) or hashish (hash, hash oil) during the last 12 months?” Response options ranged from 1=“None” to 7=“40 or more occasions.” For analysis, the responses were dichotomized to indicate 1=any use and 0=no use in the past 12 months.

Perceived Friends’ Use of Marijuana

Perceived friends’ marijuana use at each wave was assessed by asking, “How many of your friends would you estimate smoke marijuana (pot, weed) or hashish?” with response options 1=None, 2=A few, 3=Some, 4=Most, and 5=All. Variables were centered within each modal age to reflect average perceived friends’ use relative to others of the same modal age. In addition, to examine the moderating effects of perceived friends’ use by sex, race/ethnicity, and parent education, interaction terms were created by centering perceived friends’ use within each modal age and pertinent demographic subgroup (e.g., modal age and male, modal age and female), to reflect average perceived friends’ use relative to others of the same modal age and demographic subgroup.

Plan of Analysis

Time-varying effect modeling (TVEM) is a regression-based method that allows for the flexible modeling of the relationship of covariates to an outcome over continuous time, without the assumption of a parametric form (e.g., linear, quadratic) (Lanza, Vasilenko, Liu, Li, & Piper, 2014; Tan, Shiyko, Li, Li, & Dierker, 2012). Data from longitudinal panel studies are appropriate for estimating coefficients as a continuous, flexible function of age (e.g., Evans-Polce, Vasilenko, & Lanza, 2015; Schuler, Vasilenko, & Lanza, 2015). TVEM assumes only that the relationship changes over time/age in a smooth way. The intercept and slope coefficients are estimated as continuous functions of time. Figures represent the coefficient functions, along with point-wise confidence intervals, across ages 18 to 30. This permits hypothesis testing of regression coefficients to be conducted across continuous time; for specific time points when the confidence intervals do not contain the value under the null hypothesis, the coefficient is significant at p<.05. The SAS %TVEM macro (v3.1.0) (Li et al., 2015), available for download at methodology.psu.edu, was used to estimate all models. The method=P-spline argument was used to automatically select the optimal number of knots (corresponding to smoothness) for each coefficient function and to adjust standard errors for the fact that the repeated measures data were not independent. This model selection procedure relies on information criteria, which balance model fit with parsimony, to avoid overfitting the data.

In the current analysis, we first show the course of marijuana use and perceived friends’ use of marijuana from ages 18 to 30, specifying intercept-only models for a binary and a continuous outcome, respectively. Next, we examine the time-varying association of perceived friends’ use and marijuana use over time (Research Question 1). Finally, we use separate models to examine the potential time-varying moderating effects of sex, race/ethnicity, and parent education on the dynamic link between perceived friends’ use and marijuana use over time (Research Question 2).

Results

Marijuana Use and Perceived Friends’ Use by Age

The intercept-only models for marijuana use and perceived friends’ use are presented in Figure 1. Figures 1a and 1b show how the estimated prevalence of marijuana use and the level of perceived friends’ use (before centering) change across age from ages 18 to 30. The prevalence of marijuana use was relatively stable from ages 18 to 20, with a past 12-month prevalence of about 43% at age 18 and 40% at age 21. After age 21 marijuana use declined more rapidly, to about 22% at age 30. Perceived friends’ use decreased from an average of about 2.5 (between “a few” and “some” friends) at age 18 to about 2 (“a few friends”) by age 30.

Figure 1. Time-varying coefficient functions representing the estimated prevalence of past 12 month marijuana use and mean perceived friends’ marijuana use from ages 18 to 30.

Note: — Coefficient function across age - - - - 95% confidence intervals of the coefficient function

Association of Perceived Friends’ Use and Marijuana Use by Age

To address Research Question 1, we modeled the association between perceived friends’ use and marijuana use as a non-parametric function of age from 18 to 30. Results are shown in Figure 2. The solid line represents the estimated strength (odds ratio, OR) of the association between perceived friends’ use and use across age, and the dotted lines represent 95% point-wise confidence intervals; during ages where the confidence interval does not include 1, there is a statistically significant association between the two variables. There is a significant, positive, and increasing effect of perceived friends’ use on marijuana use across age through age 30. At age 18, a one-unit increase in perceived friends’ marijuana use (e.g., from “a few friends” to “some friends” using) is associated with 4.2 times greater odds of marijuana use, and this odds ratio increases to about 6.5 times at age 29.

Figure 2. Time-varying coefficient function representing the increase in odds of past 12 month marijuana use associated with a one-unit increase in perceived friends’ marijuana use from ages 18 to 30.

Note: — Coefficient function across age - - - - 95% confidence intervals of the coefficient function. Periods where confidence intervals do not include 1 indicate a statistically significant effect of perceived friends’ marijuana use on own use. Specifically, at all ages odds ratios indicate that greater friends’ use is associated with greater odds of marijuana use.

Sex Differences

To address Research Question 2, we examined the main effects and moderating effects of (i) sex, (ii) race/ethnicity, and (iii) parent education by perceived friends’ use predicting marijuana use. Results for sex (Research Question 2i) are shown in Figures 3 and 4. Figure 3a shows the main effect of sex on marijuana use. Specifically, men had significantly greater odds of marijuana use than women at all ages, although the magnitude of the difference varied slightly across young adulthood. In particular, sex differences are smallest around ages 20 to 23 (OR=1.15) and somewhat larger at both older and younger ages (OR about 1.3 at ages 18 and 30).

Figure 3. Sex differences in (a) the odds of past 12 month marijuana use and (b) the association between perceived friends’ marijuana use and the odds of past 12 month marijuana use from ages 18 to 30.

Note: — Coefficient function across age - - - - 95% confidence intervals of the coefficient function. Periods where confidence intervals do not include 1 indicate where gender differences are statistically significant. Specifically, odds ratios greater than 1 indicate that males have greater marijuana use (Figure 3a) or a stronger association between perceived friends’ use and their own marijuana use (Figure 3b), compared to females.

Figure 4. Time-varying coefficient function representing the association between perceived friends’ marijuana use and the odds of past 12 month marijuana use from ages 18 to 30 by sex.

Note. At each age, the value represents the odds ratio for persons of that sex, reflecting the increase in odds of marijuana use corresponding to a one-unit increase in perceived friends’ use of marijuana. Boxes indicate periods where the difference between males and females is statistically significant. Specifically, males and females significantly differed from ages 19 to 24 and 28 to 30.

Figure 3b shows the test of the interaction of sex by perceived friends’ use on own marijuana use, to examine potential moderating effects of sex across age. The interaction is significant at ages when the confidence band does not contain 1.0. There is a significant interaction from age 19 to 24 and again from age 28 to 30. To probe these interactions, the association between perceived friends’ use and marijuana use was calculated for males and females at all ages and plotted in Figure 4. From ages 19 to 24 (where Figure 3b indicates a significant interaction), the effect of perceived friends’ use on odds of marijuana use was significantly stronger for males compared to females. At about ages 25–26, the sex difference in the association between perceived friends’ use and own marijuana use was not significant. From age 27 to 30, the effect of perceived friends’ use on marijuana use was again significantly stronger among men.

Race/Ethnicity Differences

Results for race/ethnicity (Research Question 2ii) are shown in Figures 5 and 6. The main effects of Black, Hispanic, and Other race (compared to the reference group White) on marijuana use across young adulthood are shown in Figure 5a. Blacks, Hispanics, and young adults of Other races had significantly lower odds of marijuana use than Whites from age 18 to 29. The difference between White and Black young adults remained significant through age 30, but was greatest at the youngest ages (e.g., OR=0.54 at age 18), whereas the difference between Hispanic and White young adults remained fairly constant at about OR=0.8.

Figure 5. Race/Ethnicity differences in (a) the odds of past 12 month marijuana use and (b) the association between perceived friends’ marijuana use and the odds of past 12 month marijuana use from ages 18 to 30.

Note: — Coefficient function across age - - - - 95% confidence intervals of the coefficient function. Periods where confidence intervals do not include 1 indicate a statistically significant difference between each racial/ethnic group compared to White participants. Specifically, odds ratios lower than 1 indicate that individuals of that racial/ethnic group have lower marijuana use (Figure 5a) or a weaker association between perceived friends’ use and their own marijuana use (Figure 5b), compared to White young adults.

Figure 6. Time-varying coefficient functions representing the association between perceived friends’ marijuana use and the odds of past 12 month marijuana use from ages 18 to 30 by race/ethnicity.

Note. At each age, the value represents the odds ratio for persons of that race/ethnicity, reflecting the increase in odds of marijuana use corresponding to a one-unit increase in perceived friends’ use of marijuana. For all age periods, the effect for Black and Hispanic participants was statistically significantly different from White participants. Boxes indicate periods where the difference between the Other race/ethnicity group statistically differed from White young adults. Specifically, White and Other groups were statistically different at ages 18–19 and 23–30.

Examination of the moderating effect of race/ethnicity by perceived friends’ use on the odds of marijuana use is shown in Figure 5b. Significant interactions were evident across all ages for Black compared to White and Hispanic compared to White individuals. There was also evidence of significant differences between Other race and White participants for ages 18–19 and 23–30, but not for ages 19–23. To probe these interactions, the association between friends’ use and marijuana use was calculated for White, Black, Hispanic, and Other participants at all ages and plotted in Figure 6. Across all ages that indicated significant interactions in Figure 5b, the effect of perceived friends’ use on the odds of marijuana use was significantly stronger for Whites as compared to Black and Hispanic individuals. For individuals of Other Races, from about age 20 to 22, the race/ethnicity difference in the association was not significantly different from Whites; however, for ages 18–19 and then again 23–30, the effect of perceived friends’ use on the odds of marijuana use was significantly stronger for Whites.

Parent Education Differences

Results for parent education (Research Question 2iii) are shown in Figures 7 and 8. The main effect of parent education on the odds of marijuana use (Figure 7a) across young adulthood shows that respondents whose parents had at least some college education had significantly higher odds of marijuana use from age 18 to 29 compared to those whose parents had less education. This time-varying association peaked around age 23, decreased through age 29, and was no longer significant by age 30.

Figure 7. Parent education differences in (a) the odds of past 12 month marijuana use and (b) the association between perceived friends’ marijuana use and the odds of past 12 month marijuana use from ages 18 to 30.

Note: — Coefficient function across age - - - - 95% confidence intervals of the coefficient function. Periods where confidence intervals do not include 1 indicate where differences are statistically significant for participants with college educated compared to non-college educated participants. Specifically, odds ratios greater than 1 indicate that young adults whose parents attended college have greater marijuana use (Figure 7a) or a stronger association between perceived friends’ use and their own marijuana use (Figure 7b), compared to those whose parents had a high school education or less.

Figure 8. Time-varying coefficient function representing the association between perceived friends’ marijuana use and the odds of past 12 month marijuana use from ages 18 to 30 by parent education.

Note. At each age, the value represents the odds ratio for persons with parents with that level of education, reflecting the increase in odds of marijuana use corresponding to a one-unit increase in perceived friends’ use of marijuana. Differences between participants whose parents did and did not attend college were significant at all ages

Figure 7b shows the test of the interaction of parent education by perceived friends’ use on the odds of marijuana use across the entire age range. Significant interaction effects were evident from age 18 to 30, with the strongest interaction at ages 28–30. To probe the interaction, the association between friends’ use and marijuana use across age was calculated for participants whose parents had at least some college education and for those whose parents did not have college education, and plotted in Figure 8. Individuals whose parents had at least some college education had consistently stronger associations between perceived friends’ use and marijuana use.

Discussion

Current policy changes shifting toward legalization of medical and recreational marijuana use in the U.S. have brought renewed focus to the correlates and consequences of marijuana use (Hasin, Wall, et al., 2015; Hasin, Saha, et al., 2015; Miech et al., 2015; Monte, Zane, & Heard, 2015). In this context, it is particularly important to consider how known risk factors for use contribute to our understanding of the behavior and how it unfolds across young adulthood, during times of peak lifetime marijuana use (Johnston et al., 2015). As the political and legal landscape for marijuana use continues to change, it will be important to continue to monitor what risk factors are most relevant for marijuana use, and at what ages. Peer factors for marijuana use are among the strongest identified risk factors, and the current study suggests that understanding peer substance use remains important throughout young adulthood. The current study shows that the association between perceived friends’ marijuana use and an individual’s own use of any marijuana in the past year actually strengthens with age and is strongest around age 28. Furthermore, there was evidence of moderation by sociodemographic factors. Interactions indicated that the association between perceived friends’ use and own use tended to be stronger for men than for women, for Whites than for other racial/ethnic groups, and for those with higher parental education than for those whose parents did not have college education. These results suggest that perceived peer substance use may be an especially strong correlate for own marijuana use for men, Whites, and those from higher socioeconomic backgrounds. The potential mechanisms underlying these associations require future research, but may include differences in social contexts, disposable income, or acceptability of substance use.

Such homophily in marijuana use between an individual and his/her friends can result from both selecting friends with similar substance use and socialization over time (Kandel, 1985). In adolescence and young adulthood, both selection and socialization operate as influences to increase similarity in the substance use of peers (Dishion & Owen, 2002; Osgood et al., 2013; Patrick, Schulenberg, Maggs, & Maslowsky, 2016). Some have concluded that, by the end of young adulthood, marijuana users may have settled into friendships with other users that reinforce the use for all members of the peer group so that, in young adulthood, similarity between peers’ and own use of marijuana may be due primarily to choosing friends with similar substance use (Andrews et al., 2002), rather than socialization within existing friendship groups.

The increasing association between perceived friends’ use and own use of marijuana may also be due to a stronger commitment to marijuana behavior as integral to one’s identity. Marijuana users in their late twenties may be more committed to marijuana use and therefore their perceived social friends’ use may be a stronger correlate of behavior. This is consistent with prior work documenting drug use is associated with stronger intimacy in adult male friendship networks (Kandel & Davies, 1991). Skinner and Cattarello (1989) used cross-sectional data from high school students in Monitoring the Future to show that group norms (e.g., time spent with people using marijuana) had an increasing effect on marijuana use as behavioral commitment increased. Future research could address the behavioral commitment of young adult users in their late-twenties or thirties to examine the role of commitment to marijuana use in identity formation and peer substance use. Future research should also closely examine the effects of peers on individuals’ substance use and other risk behaviors past age 30.

The current study’s focus on friends across young adulthood is relatively unique. The majority of research on the decline in the prevalence of substance use in general, and marijuana use in particular, has pointed to social roles and responsibilities of adulthood, including marriage and parenthood (Bachman et al., 2002; Bachman, Wadsworth, O’Malley, Schulenberg, & Johnston, 1997; Chen & Kandel, 1998; Homish, Leonard, & Cornelius, 2007; Leonard & Rothbard, 1999; Leonard & Homish, 2005; Oesterle, Hawkins, & Hill, 2011; Schulenberg & Maggs, 2002; Schulenberg et al., 2005). The current results suggest that, in addition to the important family and social role influences during young adulthood, understating the roles of peer influence and peer selection remains important for understanding the persistence or desistence of substance use into the late twenties. Future examination of the potential interaction of social roles and peer use may be warranted.

Limitations and Future Research

Several limitations of the current study merit mentioning. First, bias may have been introduced from both attrition and the fact that the sample was drawn from high school seniors, therefore excluding those who dropped out of high school. Second, given the structure of the data, there are fewer measurements available for the older ages because more recent cohorts have not aged through the full range of 18 to 30. Third, no objective measures of peer substance use were available. Fourth, the direction of effects is not known. Marijuana-using young adults may be more likely to perceive that their friends use or attract marijuana-using friends; perceiving that friends use may also lead to an increase in marijuana use. That is, some combination of peer influence and selection likely accounts for the association between friends’ use and own marijuana use. Fifth, the measure of perceived friends’ use was rather general, not specifying how many friends or what types of friends were used to estimate levels of perceived marijuana use. Finally, TVEM is a relatively new statistical technique in the behavioral sciences and is actively being extended. The current version uses all available data for an analysis; the integration of multiple imputation of missing data with TVEM is an important area of future work. This version also does not have the capability of weighed estimation; this is a feature that is anticipated to be available in the coming year.

There is a long history of prevention addressing peer influence among adolescents, largely focused on substance use initiation (e.g., Botvin, 1986; Ellickson, Bell, & McGuigan, 1993; Hansen, Johnson, Flay, Graham, & Sobel, 1988; Pentz et al., 1989). More recently, intervention strategies for college drinking have emerged that address social influences (Larimer & Cronce, 2002; Larimer & Cronce, 2007). Less research has focused on the role of peers in marijuana use, and in particular the continued role that peers play in substance use throughout young adulthood. Theories and intervention strategies that acknowledge the roles of peers in substance use beyond adolescence are needed.

Future research directions should include replication of these results, which are based on any marijuana use in the past year, and extensions to examine heavier and more recent use. Historical changes should also be considered because the meaning and influence of perceived friends’ use may change as the social and legal contexts shift towards legal medical and recreational use, and these effects may differ for those under and over the legal age of 21. An additional focus on injunctive norms (e.g., perceived approval/disapproval of marijuana use) (Bachman, Johnston, & O’Malley, 1998; LaBrie, Hummer, & Lac, 2011; Napper, Kenney, Hummer, Fiorot, & LaBrie, 2016; Neighbors et al., 2008) and potential overestimation of friends’ substance use among young adults would also be informative. College students tend to overestimate the degree to which their peers use drugs, including marijuana (Arbour-Nicitopoulos, Kwan, Lowe, Taman, & Faulkner, 2010; Perkins, Meilman, Leichliter, Cashin, & Presley, 1999), but research should examine the extent to which this overestimation applies to young adults, more broadly. Finally, there is heterogeneity in the types of trajectories of marijuana use across young adulthood (Ellickson, Martino, & Collins, 2004; Jackson, Sher, & Schulenberg, 2008; Kandel & Chen, 2000; Schulenberg et al., 2005; Windle & Wiesner, 2004). Different individuals have different pathways of increasing and decreasing marijuana use, which may mean that the associations between peer norms and substance use would also differ across these groups. Future research should examine these classes or mixtures of individuals with different trajectories, and the extent to which they have different underlying risk factors. Statistical analyses, like those used in the present study to examine nuanced associations between risk factors and behaviors across age, could be used to answer these and additional questions in addictions research about the strength of the association between risk factors and behaviors across the life course.

APPENDIX

Time-varying effect modeling (TVEM) is a regression-based method that allows for the flexible modeling of the relationship of covariates to an outcome over continuous time, without the assumption of a parametric form (e.g., linear, quadratic) (Tan, Shiyko, Li, Li, & Dierker, 2012). TVEM assumes only that the relationship changes over time in a smooth way. The intercept and slope coefficients are estimated as continuous functions of time. Figures represent the coefficient functions, along with point-wise confidence intervals. TVEM can accommodate different numbers and spacing of measurement occasions, so data are arranged in a person-wave format (i.e., one record for each measurement occasion for each person). Options in the TVEM macro allow the modeling of normal, logistic, and Poisson outcomes. The SAS %TVEM macro (v3.1.0) (Li, Dziak, Tan, Huang, Wagner & Yang, 2015) is available for download at methodology.psu.edu.

In general, the TVEM model for a continuous outcome, where i refers to the individual and t the time of assessment, is specified by

$$y_{\mathit{it}}={\beta }_0(t)+{\beta }_1(t)x_{\mathit{it}}+{\epsilon }_{\mathit{it}}$$

and the model for a binary outcome is specified by

$$\mathit{ln}\left(\frac{P(y_{\mathit{it}}=1)}{1-P(y_{\mathit{it}}=1)}\right)={\beta }_0(t)+{\beta }_1(t)x_{\mathit{it}}$$

Equations for each of the models in this study are as follows:

• Figure 1a. Estimated Prevalence of Marijuana Use by Age

  $$\mathit{ln}\left(\frac{P(\mathit{MJ}\_\mathit{US}{E}_{\mathit{it}}=1)}{1-P(\mathit{MJ}\_\mathit{US}{E}_{\mathit{it}}=1)}\right)={\beta }_0(t)$$
• Figure 1b. Estimated Mean of Perceived Friends’ Marijuana Use by Age.

  $$\mathit{Friends}\_\mathit{Us}{e}_{\mathit{it}}={\beta }_0(t)+{\epsilon }_{\mathit{it}}$$
• Figure 2. Time-varying coefficient function representing the increase in odds of past 12 month marijuana use associated with a one-unit increase in perceived friends’ marijuana use from ages 18 to 30

  $$\mathit{ln}\left(\frac{P(\mathit{MJ}\_\mathit{US}{E}_{\mathit{it}}=1)}{1-P(\mathit{MJ}\_\mathit{US}{E}_{\mathit{it}}=1)}\right)={\beta }_0(t)+{\beta }_1(t)\mathit{PerceivedFriendsUs}{e}_{\mathit{it}}$$
• Figures 3 & 4. Sex differences in (a) the odds of past 12 month marijuana use and (b) the association between perceived friends’ marijuana use and the odds of past 12 month marijuana use from ages 18 to 30.

  $$\begin{array}{l}\mathit{ln}\left(\frac{P(\mathit{MJ}\_\mathit{US}{E}_{\mathit{it}}=1)}{1-P(\mathit{MJ}\_\mathit{US}{E}_{\mathit{it}}=1)}\right)\\ ={\beta }_0(t)+{\beta }_1(t)\mathit{PerceivedFriendsUs}{e}_{\mathit{it}}+{\beta }_2(t)\mathit{Mal}{e}_i\\ +{\beta }_3(t)\mathit{PerceivedFriendsUs}{e}_{\mathit{it}}\times \mathit{Mal}{e}_i\end{array}$$

  In this model, the exponentiated intercept $e^{{\beta }_0\left(t\right)}$ represents the odds of marijuana use over time when all other predictors are 0 (i.e., female with average perceived friends’ use of marijuana). Exponentiated slopes $e^{{\beta }_1\left(t\right)}$ and $e^{{\beta }_2\left(t\right)}$ represent the time-varying association of perceived friends’ use of marijuana and the time-varying association of sex (male) with the odds of marijuana use. Finally, exponentiated slope $e^{{\beta }_3\left(t\right)}$ is the time-varying association of the interaction of perceived friends’ use and sex on the odds of marijuana use.

• Figures 5 & 6. Race/Ethnicity differences in (a) the odds of past 12 month marijuana use and (b) the association between perceived friends’ marijuana use and the odds of past 12 month marijuana use from ages 18 to 30.

  $$\begin{array}{l}\mathit{ln}\left(\frac{P(\mathit{MJ}\_\mathit{US}{E}_{\mathit{it}}=1)}{1-P(\mathit{MJ}\_\mathit{US}{E}_{\mathit{it}}=1)}\right)\\ ={\beta }_0(t)+{\beta }_1(t)\mathit{PerceivedFriendsUs}{e}_{\mathit{it}}+{\beta }_2(t)\mathit{Blac}{k}_i\\ +{\beta }_3(t)\mathit{PerceivedFriendsUs}{e}_{\mathit{it}}\times \mathit{Blac}{k}_i+{\beta }_4(t)\mathit{Hispani}{c}_i\\ +{\beta }_5(t)\mathit{PerceivedFriendsUs}{e}_{\mathit{it}}\times \mathit{Hispani}{c}_i+{\beta }_6(t)\mathit{OtherRace}{s}_i\\ +{\beta }_7(t)\mathit{PerceivedFriendsUs}{e}_{\mathit{it}}\times \mathit{OtherRace}{s}_i\end{array}$$
• Figures 7 & 8. Parent education differences in (a) the odds of past 12 month marijuana use and (b) the association between perceived friends’ marijuana use and the odds of past 12 month marijuana use from ages 18 to 30.

  $$\begin{array}{l}\mathit{ln}\left(\frac{P\left(\mathit{MJ}\_\mathit{US}{E}_{\mathit{it}}=1\right)}{1-P\left(\mathit{MJ}\_\mathit{US}{E}_{\mathit{it}}=1\right)}\right)\\ ={\beta }_0\left(t\right)+{\beta }_1\left(t\right)\mathit{PerceivedFriendsUs}{e}_{\mathit{it}}+{\beta }_2\left(t\right)\mathit{ParentEducatio}{n}_i\\ +{\beta }_3\left(t\right)\mathit{PerceivedFriendsUs}{e}_{\mathit{it}}\times \mathit{ParentEducatio}{n}_i\end{array}$$