Prevention of Alcohol Use in Older Teens: A Randomized Trial of an Online Family Prevention Program

Abstract

This study examines effects of a randomized controlled trial for an online, family-based prevention program for older teens, Smart Choices 4 Teens, on alcohol use and related outcomes. Families (N=411; teen age M=16.4, SD=0.5) were randomly assigned to the intervention or control condition in 2014–2015. Both intent to treat (ITT) and dosage models were conducted. ITT models: At the 6-month follow-up, teens in the experimental condition reported fewer friends who had been drunk, and parents in the experimental group reported more communication about social host laws. At the 12-month follow-up, parents in the experimental condition reported consuming fewer drinks than parents in the control group. Dosage models: At the 6-month follow-up, dosage was inversely related to teen drinking in the past six months or 30 days, frequency of teen drinking during the past six months and 30 days, drinks consumed by teens over the past six months, teen drunkenness and binge-drinking during the past 30 days, teen reported communication about safe drinking and positively related to parent and teen reported communication about social host laws. At 12 months, dosage was inversely related to teen alcohol use, frequency of teen drinking over the past 30 days, drinks consumed by teens over the past six months and 30 days, and teen drunkenness over the past six months. Results suggest that Smart Choices 4 Teens is beneficial for families. Dissemination and implementation strategies that motivate completion of program content will improve outcomes related to older teens’ alcohol use.

The increasing prevalence of alcohol use during the teen years makes it an important public health issue. National estimates show that 19.9% of teens ages 14–15 report having drank alcohol in the past year, which increases to 38.8% at 16–17 years old (SAMHSA, 2017). Delaying the onset of alcohol use during the teen years can reduce the risk of alcohol problems in adulthood (Dawson, Goldstein, Chou, Ruan, & Grant, 2008; Grant & Dawson, 1997; Hingson & Zha, 2009). Thus, late adolescence is a critical period for parental involvement in decreasing rates of alcohol initiation, regular use, as well as alcohol-related problems. However, parental roles become more complex as teens seek and gain more autonomy (Kerr, Stattin, & Burk, 2010; Pettit, Keiley, Laird, Bates, & Dodge, 2007). Prevention strategies to help parents guide teens through this period are therefore timely.

Parental influence

Alcohol use and related problem behaviors remain strongly influenced by parents even into late adolescence (ages 15–17) and young adulthood (ages 18–25) (Abar & Turrisi, 2008; Patock-Peckham & Morgan-Lopez, 2007; Reifman, Barnes, Dintcheff, Farrell, & Uhteg, 1998; Turner, Larimer, & Sarason, 2000; Turrisi, Wiersma, & Hughes, 2000). However, few intervention programs are targeted at older teens, even though studies of college students and teens preparing for college indicate that interventions aimed at preventing alcohol use among older teens who have not started drinking, or reducing alcohol use in those who have, are still relevant for this age group (Turrisi, Abar, Mallett, & Jaccard, 2010b; Turrisi, Jaccard, Taki, Dunnam, & Grimes, 2001).

Several aspects of the parent-teen relationship have been shown to be important for older teen alcohol use. One such aspect is parental monitoring, which has been identified as a key influence on teen drinking. Higher levels of parental monitoring in the last year of high school were associated with less college drinking (Arria et al., 2008). A second aspect, parenting style, has also been shown to be important, as the authoritative parenting style (high responsiveness and warmth) has been shown to be protective for teen drinking and peer use (Bahr & Hoffmann, 2010; Piko & Balazs, 2012).

Parent communication regarding teen alcohol use is also important (Mares, van der Vorst, Engels, & Lichtwarck-Aschoff, 2011; Martyn et al., 2009; Shin & Hecht, 2013; Shin, Lee, Lu, & Hecht, 2016; Windle et al., 2008). For example, more frequent communication about teen alcohol use has been associated with less positive expectancies about alcohol (Miller-Day & Kam, 2010), less excessive drinking and alcohol-related problems (Mares et al 2011), and a lower probability of becoming a regular drinker (Shin et al 2016). However, the frequency of communications about alcohol may not accurately reflect the influence of parent-teen communications regarding teen alcohol use. Some studies have found that more frequent communication about alcohol was related to higher rates of alcohol use and related problems (Koning, van den Eijnden, Glatz, & Vollebergh, 2013; Spijkerman, van den Eijnden, & Huiberts, 2008; van der Vorst, Engels, Meeus, Dekovic, & Van Leeuwe, 2005), suggesting the possibility that in some families, communication occurs more frequently when parents believe their teen is already using alcohol and experiencing unfavorable consequences.

Parent-teen communication about teen alcohol use includes many different content areas, such as the rules and expectations for teen use, potential negative consequences, safe drinking behavior, and influence of media on drinking (Ennett, Bauman, Foshee, Pemberton, & Hicks, 2001). Most studies combine these different topic areas (Mares et al., 2011; Miller-Day & Kam, 2010; Shin & Hecht, 2013; Shin et al., 2016), but work on older teens (aged 16–19) suggests that differentiating between the various strategies may be useful (Bourdeau, Miller, Vanya, Duke, & Ames, 2012). Further, as youth progress through adolescence, parents adjust the ways in which they communicate with their teens regarding alcohol and their messages around alcohol to be effective with their older teens (Bourdeau et al., 2012). Based on Social Learning Theory (Bandura, 1977), parents also communicate their views on drinking through modeling their own alcohol use. Studies have found that parental drinking is related to more risky drinking among youth (Latendresse et al., 2008; White, Johnson, & Buyske, 2000).

Peer influence

Peer influence is especially salient during the teen years and has been linked to alcohol use. A systematic review of longitudinal studies found that having friends who engage in deviant behavior or alcohol use was prospectively related to alcohol use in adolescents (Leung, Toumbourou, & Hemphill, 2014). Similarly, having delinquent friends is related to perceived peer approval and use of alcohol, which subsequently predicts teens’ own alcohol use initiation (Trucco, Colder, & Wieczorek, 2011). Peer drinking also predicts changes in drinking from adolescence into young adulthood (Cruz, Emery, & Turkheimer, 2012). A study of Hispanic adolescents found that peer influences were more consistently related to substance use, attitudes, and behaviors than parental influences, with the exception of parental injunctive norms (Parsai, Voisine, Marsiglia, Kulis, & Nieri, 2009). However, parental and peer influences on teen drinking have been measured in different ways with different results. For instance, one study found that parental warmth or control did not moderate peer influence (Trucco et al., 2011). In contrast, steady strong parental disapproval of substance use reduced risk for heavy drinking regardless of affiliation with drinking peers (Martino, Ellickson, & McCaffrey, 2009). Both parental and peer influences were significantly related to substance use when considered simultaneously in models (Mayberry, Espelage, & Koenig, 2009). Parenting (i.e., monitoring and relationship quality) may also be important in influencing affiliation with deviant peers (Van Ryzin, Fosco, & Dishion, 2012).

Family-based Programs

Family-based approaches to preventing adolescent alcohol and other drug (AOD) use can be effective throughout adolescence, even though most AOD prevention strategies target the individual only (Kumpfer, Alvarado, & Whiteside, 2003). A meta-analysis showed the effectiveness of family-based prevention strategies, with effects persisting 48 months later (Smit, Verdurmen, Monshouwer, & Smit, 2008). Similarly, a systematic review found that while effects of family-based preventive interventions were small, they tended to be consistent and lasting into medium- and longer-term (Foxcroft & Tsertsvadze, 2011).

However, family-based prevention programs are frequently focused on younger adolescents. For example, Family Matters (FM) (Bauman, Foshee, Ennett, Hicks, & Pemberton, 2001), a family-based prevention program focused on building communication skills using vignettes and exercises, has shown effectiveness for preventing alcohol and tobacco use in teens 12–14 (Bauman et al., 2002; Bauman, Foshee, Ennett, Pemberton, et al., 2001). FM has also lowered the prevalence of smoking and drinking alcohol at 3- and 12-month follow-ups, with effect sizes at 3-months being larger for alcohol (Bauman et al., 2002). An exception to this younger focus is the Parent Handbook (PH), a parent-targeted intervention that focuses on alcohol use for incoming college students. PH was effective in lowering drinking and drunkenness for teens and their peers, as well as reducing drinking-related harmful consequences (Turrisi et al., 2001). Hypothesized mediators, such as favorable attitudes toward drinking, alternatives to drinking, and beliefs about alcohol-related behavior also were impacted by the program in the desired direction (Turrisi, Abar, Mallett, & Jaccard, 2010a).

Online Interventions

Prior studies suggest that online intervention approaches to prevention are appealing and acceptable to parents and teens (Brown, Dunn, & Budney, 2014; Widman, Golin, Kamke, Massey, & Prinstein, 2017). Most adults (89%) use the Internet (Center, 2017b) and 92% of U.S. adolescents ages 13–17 use the Internet daily (Center, 2015b). Both parents and teens search for health information online (72% of adults and 31% of teens 12–17) (Center, 2014; Lenhart, Purcell, Smith, & Zickuhr, 2010). Benefits of online programs include standardization of delivery, ability to engage users with interactive online features, and ability to monitor program usage. Several randomized controlled trials have shown online interventions for adolescent alcohol use to be effective (Doumas, Esp, Turrisi, Hausheer, & Cuffee, 2014; Rooke, Thorsteinsson, Karpin, Copeland, & Allsop, 2010; Spijkerman et al., 2010; Voogt, Kleinjan, Poelen, Lemmers, & Engels, 2013); however, few studies have examined the effectiveness of online interventions for teens and their parents together, despite evidence of the utility of family-based programs.

The Current Study

This paper examines outcomes for alcohol use and related behaviors for families who participated in an online, interactive, family-based prevention program for older teens (16–17) in a randomized controlled trial. In the experimental condition, parents and teens navigated the online program separately and had off-line semi-structured discussions together using realistic but hypothetical scenarios involving teens and alcohol use, designed to promote teen skill building. The program, Smart Choices 4 Teens, had three components for building knowledge and skills: parent-teen general communication, teen alcohol use, and teen relationships with romantic partners. The focus of this paper is on alcohol-related outcomes and the analyses tested the primary hypothesis that teens in the experimental condition would decrease alcohol use and alcohol-related problems at posttest compared to teens in the control condition. The secondary hypotheses were that compared to teens in the control condition, teens in the experimental condition would have better outcomes related to perception of peer drinking, attitudes towards alcohol, parent-teen communication about alcohol and parent drinking.

Analyses of these hypotheses were performed from an Intent-to-Treat (ITT) approach. Concerns have been raised about using the ITT approach in community-based prevention trials (Gross & Fogg, 2004; Horvitz-Lennon, O’Malley, Frank, & Normand, 2005), so analyses of program exposure also were performed, testing the hypothesis that greater program exposure will be related to decreased alcohol use and alcohol-related problems. Examining alternative models such as dosage models have been proposed in trials with little control over program exposure and where substantial portions of participants do not appear motivated to fully adhere to the intervention implementation protocol, as this non-adherence can bias results (Gross & Fogg, 2004; Horvitz-Lennon et al., 2005).

In comparison to treatment and medical trials that typically aim to improve an existing problem or condition (e.g., alcoholism, diabetes), participants may lack motivation to complete preventive interventions because the intervention addresses illnesses or discomforts that have not yet happened (Coie et al., 1993; Gross & Fogg, 2004; Ten Have et al., 2008). Consistent with this, low rates of adherence in family-based prevention programs are common, often with most families dropping out before completing the intervention (Ingoldsby, 2010). High rates of nonadherence in prevention trials can introduce Type 2 error by underestimating the intervention’s effectiveness (Fergusson, Aaron, Guyatt, & Hébert, 2002; Gross & Fogg, 2004; Gupta, 2011). Heterogeneity can also be a problem if participants who were noncompliant are grouped together with compliant participants in analysis, providing little information about treatment efficacy (Gupta, 2011).

In addition, potential participants likely want to know the possible benefits of an intervention if they were to follow the intervention protocol (Gross & Fogg, 2004; Pocock & Abdalla, 1998). One proposed solution to this problem is to follow other analysis approaches based on treatment actually received or adherence to protocol, such as our analysis of dosage, alongside ITT analyses (Gross & Fogg, 2004; Horvitz-Lennon et al., 2005; Ten Have et al., 2008).

Methods

Participants

Participants were recruited for a large, randomized controlled trial designed to test a family online intervention to address older teens’ alcohol use and romantic relationships from November 2014 to November 2015. Parent-teen dyads were recruited nationally from two online panel vendors (see Wang-Schweig et al., 2017). Panel vendors are companies involved in recruiting participants into a panel to participate in web-based surveys (Berrens, Bohara, Jenkins-Smith, Silva, & Weimer, 2003; Craig et al., 2013). The vendors match panelists to a targeted audience for data collection. Market research firms typically maintain these panels (e.g., Qualtrics Online Sample) (Baker et al., 2010). The term “online panel” includes both nonprobability-and probability-based panels or panels created through recruitment of participants through random sampling methods (Baker et al., 2010; Hays, Liu, & Kapteyn, 2015; Heeren et al., 2008). Studies suggest that online panels produce samples that are comparable to nationally representative samples in terms of alcohol use and demographic make-up (Heeren et al., 2008).

Panel members provided demographic characteristics when joining the panel, which were used to determine eligibility for the trial (Wang-Schweig et al., 2017). As shown in the CONSORT diagram (see Supplemental Table 1), panel vendors provided contact information for 1,531 adult panelists via a secure shared website. Among these, 559 were found to be eligible, i.e., parent with a teen aged 16–17; English speaking; a device (tablet or computer) compatible with viewing the online intervention; and both parent and teen verified via phone after passing a screener. The research team performed a second level of verification, which entailed calling and speaking to both parent and teen and verifying their eligibility and interest in participating. We were unable to contact 607 panelists to complete this second level of verification, 365 were found to be not eligible, and 148 declined to participate. Among the 559 eligible panelists, 411 (73.5% of eligible adults) completed baseline online surveys (and their teens completed separate online surveys) and enrolled in the study.

After both parent and teen completed the baseline survey, participants were randomly assigned at the family-level to either the intervention or control condition using a computer-generated program. Families in both the control and intervention condition received links to resources, including hotline numbers and websites providing information about teen alcohol and drug use, sexual behavior, suicide, and other health issues. The websites provided contact information for crises, as well as information about the specific health issue (e.g., the Mothers Against Drunk Driving website provided information about rates and consequences of impaired driving). Control families also had access to an 800 number throughout the duration of the project for contact with the research team. No control families contacted staff for support regarding content provided in the resource information. Automated emails and texts were sent to invite families to complete the follow-up surveys at six and 12 months after baseline. One week after follow-up survey invitations were sent, reminder emails and texts were sent to families who had not completed the surveys, with the emails/texts repeated after another week of nonresponse. One week later, separate phone calls were made to parents and teens if they had still not completed. Participants could complete the surveys up to two months after surveys were initiated. Of the 411 enrolled dyads, 315 teens (76.6%) and 364 parents (88.6%) completed the six-month follow-up surveys, and 311 teens (75.7%) and 354 parents (86.1%) completed the 12month follow-up surveys.

Among enrolled families at baseline, parents were on average 43.7 years of age (SD = 6.7) and their teens were 16 or 17 years of age (M = 16.4, SD = 0.5). Mothers comprised 84.7% of the parents and slightly more than half (55.3%) of the teens were females. Household size ranged from two to 12 (M = 4.5, SD = 1.6). About one-tenth (9.5%) of the teens were Hispanic/Latino. Teens reported the following race/ethnicities: 72.5% White, 1.9% Asian, 11.7% African-American, 1.0% Native American, 8.3% Multi-racial, 2.7% some other race, and 1.9% unreported. No significant differences were found between the experimental and control conditions on demographic characteristics at baseline.

Survey Procedures

Parents were first sent an email invitation from the online panel vendor. Parents who responded to the invitation were contacted by telephone by study staff to confirm eligibility. Both parent and teen needed to agree to participate and to complete confidential online baseline surveys for the family to be enrolled. Participants were informed that all surveys were confidential and would not be shared with anyone, including other family members. One eligible parent was included in the study, and they provided consent for themselves and for their teen. Teens provided their assent to participate. Each participant received $30 for baseline surveys, $40 for 6-month follow-up surveys, and $50 for 12-month follow-up surveys. All procedures and consent/assent forms were approved by the Institutional Review Board.

Intervention

The intervention, Smart Choices 4 Teens, is an interactive, online family prevention program with three sequential components (general parent-teen communication, teen alcohol use, and teen romantic relationships). The program was developed by adapting two evidence-based prevention programs: Family Matters (FM) ((Bauman, Foshee, Ennett, Hicks, et al., 2001) and Parent Handbook (PH) (Turrisi et al., 2001) to provide an age appropriate program that would appeal to both older teens and their parents. FM is a family-based prevention program that provides strategies for developing communication skills through vignettes and exercises, and has demonstrated effectiveness in preventing alcohol and tobacco use in teens ages 12–14 (Bauman et al., 2002; Bauman, Foshee, Ennett, Pemberton, et al., 2001). PH is an intervention targeted at parents of incoming college students and has shown effectiveness in lowering drinking and drunkenness for teens and their peers, as well as reducing drinking consequences (Turrisi et al., 2001).

The two programs were adapted in a series of steps: 1) The research team together with an expert panel began by developing material drawn from findings in the literature. This expert panel (including the original developers of FM and PH) reviewed the entire program content from both FM and PH to determine what remained valid for this age range. 2) A set of focus groups (three parent groups and three teen groups conducted separately) reviewed the content and the specific exercises for families. 3) Based upon findings from step two, changes were made to program content, with insight from the expert panel. 4) A second set of focus groups (three parent and three teen groups conducted separately) focused specifically on the changed materials but also provided another overall assessment of the entire program. 5) The research team and expert panel again reviewed program materials in light of the focus group findings and made final adjustments in the program materials.

The first major change that was made was targeting the intervention at both parents and teens. Although FM is for parents and teens, many activities are just for parents, while PH is a parent program. Second, the format was changed to be presented online instead of via booklets. Third, the interactive nature of the interventions was increased. While FM includes parent/teen activities, much of the information is presented in didactic format to parents only and PH was a booklet with no activities included. Fourth, changes were made to content to be age-appropriate, as FM is targeted at parents and teens ages 12–14, e.g., content and activities regarding rule-setting were changed to focus on discussing parental expectations for teen behavior, and scenarios of fictional teens in realistic situations, used to facilitate discussion between parents and teens, were adapted from FM to be age-appropriate for older teens. Finally, additional content was added regarding social host liability laws, and a component regarding romantic relationships was added.

The Communication component provided a foundation for the other components by providing needed communication skills through videos and interactive activities. The Alcohol Prevention component’s goal was to prevent or reduce teen alcohol use. It began with an introduction to the module, which provided an overview and statistics about teen alcohol use, information about peer pressure, and consequences of drinking. This component then presented several activities and videos focusing on social host laws in each state, physical and social consequences of drinking, signs of alcohol poisoning, a BAC calculator activity, myths about sobering up, parental influences important to address teen alcohol use, refusal skills, and indicators of problem drinking. Prior to completing the Alcohol Prevention component, families had to have completed the Communication component. The Romantic Relationships component included activities and videos focusing on such topics as decisions to initiate romantic relationships and decision making around sex.

For each component, Smart Choices 4 Teens required that parents and teens go through the activities online separately and then come together to complete a discussion activity at the end of each component. Families had to indicate that they had completed the discussion to move on to the next component. In addition, as the program is family-based, families could only progress to the next component if their other participating family member had also completed that component. For example, if a teen had completed the Communication component, but their parent had not, the teen would not be able to begin the Alcohol component until their parent had also completed the Communication component. On average, the Communication component took families 20.29 minutes to complete (21.83 minutes for parents, 18.75 minutes for teens), excluding the time for discussion, while the Alcohol component took 15.94 minutes (16.34 minutes for parents, 15.53 minutes for teens), and the Relationship component took 19.75 minutes (19.79 minutes for parents, 19.71 minutes for teens).

The Alcohol Prevention component is the focus of this paper. In this component, the discussion activity described four real-life scenarios pertaining to teen drinking. Two scenarios portrayed peer pressure (e.g., pressure to drink at a party) and two portrayed problems related to drinking situations (e.g., a friend gets ill from drinking). Families selected one scenario from each category to discuss off-line. Families were prompted to download a tailored discussion guide and engage in a discussion. The discussion was focused on possible decisions for the teen in the scenario so as to make the discussions less threatening to the teen participant.

Measures

Overall model predictors.

The variable indicating intervention condition was defined as 1 = experimental condition, 0 = control condition. Dosage was defined based on the family member who had made the most progress in the program. As the program is family-based, the person furthest along in the program represents the maximum exposure within the family unit. Because our primary focus was alcohol-related outcomes as a result of completing the alcohol component, we constructed the dosage variable to consider whether families at least completed the alcohol component (0 = no exposure, 1 = only communication component completed, 2 = completed at least up to alcohol component).

Primary study outcomes.

Primary outcomes assessed were related to teen alcohol use. All primary outcome measures are teen reports.

Teen alcohol use.

Alcohol items were adapted from the National Longitudinal Study of Adolescent Health (Harris et al., 2008) and asked at each time point. Teens first reported whether they had consumed alcohol over the past six months and past 30 days (1 = Yes, 0 = No).

Quantity and frequency.

Teens reported their frequency of drinking over the past 30 days and past six months (0 = Never, 1 = once a month, 2 = 2–3 times a month, 3 = once a week, 4 = 2–3 times a week, and 5 = daily or almost daily). The variable was recoded to more closely approximate the actual number of times used over the time period. For example, 2–3 times a month in the last 30 days was recoded to 2.5, whereas 2–3 times a month over the past six months was recoded to 15 (2.5 * six months). Teens also reported the usual amount drank on drinking days (0 = less than one to 7 = more than six).

Problem drinking.

To assess binge drinking, teens reported the number of times they had consumed four or more drinks of alcohol (for females) or five or more drinks (for males) over a two to three-hour period (0 = never, 1 = once a month, 2 = 2–3 times a month, 3 = once a week, 4 = 2–3 times a week, and 5 = daily or almost daily). To assess drunkenness, teens reported the number of times they had gotten “drunk or very, very high” on alcohol over the time period on the same scale as for binge drinking.

Secondary outcomes.

Secondary outcomes assessed peer drinking, alcohol attitudes, parent-teen communication about alcohol, and parent drinking. All measures are teen-reported unless otherwise stated.

Peer drinking.

To assess peer drinking, teens were asked to think of their five closest friends and respond to three questions (created for the study) regarding these friends’ drinking: 1) Do any of these five friends drink? (1 = Yes, 0 = No), 2) How many of these friends drink? and 3) How many of these friends have gotten drunk over the past six months?

Alcohol attitudes.

Teen attitudes regarding alcohol were measured by three constructs: 1) positive alcohol attitudes, 2) approval of drinking, and 3) alcohol norms. Positive alcohol attitudes were assessed by teens’ responses to 14 items adapted from Turrisi and colleagues (Turrisi et al., 2010a; Turrisi & Jaccard, 1992; Turrisi, Jaccard, & McDonnell, 1997; Turrisi et al., 2000). Example items are: Getting drunk will result in something bad happening to me (this is a reversed item), A few drinks make it easier to talk to people (1 = strongly disagree to 5 = strongly agree). Items were reversed as needed so that higher scores indicated more positive attitudes towards drinking. Items were averaged at each time point to create scales for each time (α = .89 at baseline, .90 at 6-month follow-up, and .90 at 12-month follow-up).

Approval of drinking at was assessed at each time point by three items adapted from Baer and colleagues (Baer, 1994; Baer, Stacy, & Larimer, 1991; Core Institute, 1999): how strongly teens approved of 1) drinking one or two drinks on one occasion, 2) drinking four or more drinks on one occasion, and 3) driving after drinking alcohol. Response options were 1 = strongly disagree to 5 = strongly agree. Items were averaged at each time point (α = .62 at baseline, .68 at 6-month follow-up, and .65 at 12-month follow-up) so that a higher score indicated greater approval.

Alcohol norms were assessed by asking teens what percentage of teens they thought consumed no alcoholic beverages at all, adapted from Baer and colleagues (Baer, 1994; Baer et al., 1991; Core Institute, 1999). A higher percent indicated lower levels of perceived teen use.

Communication regarding teen alcohol use.

Both parents and teens responded to items regarding their alcohol communication at each time period (baseline, six months, twelve months). All parent-teen alcohol communication items were adapted from Turrisi et al. (2007) and Reimuller et al. (2011). Three constructs were assessed: 1) Consequences and Expectations: alcohol communication regarding consequences and parental expectations about teen alcohol use (12 items, e.g., how drinking could get the teen in trouble with a parent or the law; drunk driving consequences), 1 = never to 5 = very often. Items were averaged to create separate parent and teen scales (parent report: α = .95 at baseline, .97 at 6-months, .97 at 12 months; teen report: α = .96 at baseline, .96 at 6 months, and .97 at 12 months). 2) Safe drinking strategies: two items that consisted of teen drinking was acceptable in moderation, and teen drinking was acceptable under supervision, 1 = never to 5 = very often. Items were averaged to create separate parent and teen scales at each time point (parent report: r = .72, p < .001 at baseline, r = .72, p < .001 at 6months, .74, p < .001. at 12 months; teen report: r = .73, p < .001 at baseline, r = .72, p < .001 at 6 months, and r = .78, p < .001 at 12 months); and 3) Both parents and teens responded to one question as to how often parents had discussed social host laws with them (1 = never to 5 = very often).

Parent drinking.

Parents reported the usual amount drank on drinking days in the past six months (0 = less than one to 7 = more than six).

Selection model variables.

In addition to demographic variables, other baseline parentteen relationship and communication variables were considered in selection models. Since the program as a whole also included sexual behavior as a target behavior, we included baseline sexual behaviors in our selection models to account for selection bias in completing the program. We wanted to account for the possibility that families with teens who were already engaging in sexual behavior may have been more likely to complete the program. All selection model variables were assessed at baseline.

Positive evaluation of parent-teen relationships.

Since most standard parent-child relationship measures assess general aspects of the parent-teen relationship, we developed four items to assess the specific concepts taught in the Communication Module that reflect their evaluation of their relationship: usually when my parents/teen and I talk… 1) my parent/teen seems to hear what I say, 2) I am satisfied with our discussions, 3) I can share things that are bothering me/My teen can share things that are bothering him/her, and 4) I go to them for help when I need advice about something important/He or she goes to me for advice when he/she needs something important. Response options were 1 = Never to 5 = Very often. Items were averaged to create separate scales for parents’ and teens’ report (Teen report: α = .84; parent report: α= .82).

Truthfulness with parent/teen.

Using modified items from the evaluation from the Parent Handbook, parents and teens responded to two items regarding their truthfulness with each other (Turrisi et al., 2001): 1) I am honest with my parent/teen, and 2) I keep my promises to my parent/teen, 1=Never to 5=Very often (parents: r = .45; p < .001; teens: r = .68, p < .001).

Negative communication style in parent-teen relationship.

Both parents and teens responded to the item: My parent/teen and I disagree so much that I cannot get my ideas across. Teens also responded to an item: My parent lectures me rather than talks with me. All response options were from 1 = never to 5 = very often. Items were created to assess the specific concepts taught in the Communication Module, as standard communication measures used in prior studies tend to be more general. The two teen items and one parent item were averaged to create a negative communication scale (α = .68).

Time spent together.

Both parents and teens were asked to think about a typical day in their family, and answered the question “How much time do you and your teen spend doing things together?” (0 = None to 6 = More than 2 hours).

Time spent talking.

Both parents and teens reported on the amount of time they spend talking to each other on a typical day (0 = None to 6 = More than 2 hours).

Teen sexual activity.

Teens responded as to whether they had ever had sex (0 = No, 1 = Yes).

Demographics.

Teens’ gender, ethnicity, and age were assessed through teen self-reports and used as controls. In addition, parents’ self- reports of parental gender and income (1 = $20,000 or less to 10 = more than $300,000) were used as controls.

Analyses

Descriptive analyses were conducted first to describe rates and mean scores of outcomes overall and by condition across three assessments to check the randomization. Analyses examining outcomes in relation to intervention condition were then conducted in SPSS 21 (IBM, 2012). All models controlled for baseline values of dependent variables, teen gender, teen age, and teen ethnicity. For ITT models, condition was entered as a predictor (1 = Intervention, 0 = Control). For ITT models, to examine the five primary outcomes (i.e., alcohol use, drinking frequency, drinking quantity, drunkenness, and binge drinking) over the past six months and past 30 days, at both follow-up time points (six and 12 months), 20 tests were conducted. For the 10 secondary outcomes (e.g., parent-teen communication), 26 tests were conducted, including both parent and teen reports of communication, and at both follow-up time points. Due to limitations of adjusting for multiple comparisons, such as substantial reduction in statistical power and unacceptable levels of Type II error (Garamszegi, 2006; Nakagawa, 2004; Perneger, 1998; Rice, 1989), we present effect sizes for all tests in tables 2–5 and Supplemental tables 2–21, as recommended by several authors (Jennions & Møller, 2003; Nakagawa, 2004; Stoehr, 1999) to allow comparison of results across studies.

Dosage models examined dosage as a predictor for the same dependent variables as in ITT models, and at both follow-up time points. The entire sample was included in dosage models, with the control group assigned a dosage of zero. A dummy variable indicating condition was entered into dosage models (1 = Control, 0 = Intervention). Dosage models also included an inverse Mills’ ratio as a covariate in models (see next paragraph, Selection model).

Dichotomous outcomes were examined using multiple logistic regression, while continuous outcomes were examined using multiple linear regression. Listwise deletion was used for missing cases.

Selection model.

Because families chose how far they continued in the program, we conducted a probit analysis predicting whether or not a family completed the alcohol component from baseline measures (see Measures, Selection model variables). Out of the predictors, only the teen’s White ethnicity was significant (b = −.21, z = −2.23, p < .05). Based on this analysis, an instrumental variable, inverse Mills’ ratio (IMR) representing the underlying selection processes was included as a covariate in our primary dosage analyses (Heckman, 1979). The IMR (nonselection hazard) was calculated in Stata using the two-step procedure described in Heckman (1979). When the IMR is significant in a model, this indicates that the predicted probability of completing the component is associated with the outcome. That is, the same factors that predispose families to complete (or not complete) the component are related to the outcomes. In models where dosage is significant, this indicates that even accounting for factors that are related to completing the program, that dosage is still related to outcomes.

Alternative approaches exist for dealing with varying levels of adherence in intervention studies, such as propensity scores and the complier average causal effect (CACE) estimation method. A propensity score is the probability of being treated based on an individual’s background characteristics (D’Agostino, 1998; D’Agostino & D’Agostino, 2007). The CACE method estimates causal effects of treatments for compliers (Jo, 2002a, 2002b; Jo, Ginexi, & Ialongo, 2010). However, a major problem in using this approach is dealing with missing compliance information, which can lead to biased estimates (Jo, 2002a). Use of selection models (Heckman, 1979) provides several benefits, including ease of use and wide use in research (a 700% increase in use over the last decade) (Certo, Busenbark, Woo, & Semadeni, 2016; Tucker, 2010). Selection models also offer an advantage over propensity scores in that they can deal with selection on unobservable factors, yet propensity score matching requires self-selection of participants to be explained completely by observable factors (Tucker, 2010).

Potential bias due to attrition.

To determine possible bias due to attrition at follow-up, we used the same baseline variables used in the selection model to predict completion of six- and 12-month follow-up surveys by parents and teens. Only negative parent-teen communications significantly predicted less parent completion of six-month follow-up surveys (b = −.53, p < .05), while only parents’ report of positive evaluation of the parent-teen relationship predicted less teen completion of six-month follow-up surveys (b = −.51, p < .05). Fewer parents and teens completed the 12-month follow-up surveys if the teen reported ever having had sex (parents: b = −.76, p < .05; teens: b = −.79, p < .05). These potential biases were accounted for in dosage analyses through the inclusion of the inverse Mills ratio as a covariate in dosage models.

Results

Descriptive Statistics

Among families assigned to the experimental condition, 14.1% had no exposure to the program. About one third (36.4%) completed only the communication module (which was the first component). About half (49.5%) of the families completed the alcohol module. Table 1 presents means, standard deviations, and rates for the overall sample and by condition at baseline. No significant differences in outcomes variables were found across conditions at baseline.

Table 1.

Means, Standard Deviations, and Rates Overall and by Condition at Baseline

	Overall (N=411)		Control (N=205)		Experimental (N=206)		χ2 or t
Variable	M or %	SD	M or %	SD	M or %	SD
Alcohol use
Past 6 months alcohol use	25.70%	--	27.90%	--	23.50%	--	1.00
Past 6 months drinking frequency	2.74	10.91	2.94	14.06	2.55	6.35	0.35
Past 6 months drinking quantity	0.54	1.33	0.59	1.40	0.50	1.25	0.66
Past 6 months drunkenness	0.26	0.82	0.24	0.74	0.29	0.89	−0.67
Past 6 months binge drinking	0.11	0.44	0.10	0.37	0.13	0.50	−0.69
Past 30 days alcohol use	11.80%	--	10.50%	--	13.10%	--	0.66
Past 30 days drinking frequency	0.23	0.82	0.22	0.89	0.25	0.73	−0.43
Past 30 days drinking quantity	0.27	0.99	0.27	1.05	0.27	0.93	−0.08
Past 30 days drunkenness	0.07	0.34	0.07	0.35	0.08	0.34	−0.46
Past 30 days binge drinking	0.07	0.40	0.06	0.39	0.08	0.41	−0.36
Peer drinking
Friends drink	47.20%	--	47.00%	--	47.50%	--	0.01
Number friends drink	3.01	1.44	2.87	1.37	3.13	1.50	−1.15
Number friends drunk past 6 months	2.29	1.66	2.28	1.69	2.30	1.63	−0.09
Alcohol attitudes
Positive alcohol attitudes	2.08	0.76	2.09	0.76	2.08	0.76	0.21
Approval of drinking	1.86	0.77	1.85	0.77	1.87	0.77	−0.25
Alcohol norms	39.43	22.12	39.67	22.69	39.20	21.59	0.21
Parent-teen alcohol communicatio
Teen report - alcohol communication consequences/expectations	3.14	1.12	3.18	1.10	3.10	1.14	0.74
Parent report - alcohol communication consequences/expectations	3.23	1.06	3.26	1.07	3.19	1.05	0.67
Teen report - alcohol communication safe drinking	2.13	1.28	2.17	1.28	2.08	1.29	0.70
Parent report - alcohol communication safe drinking	1.77	1.02	1.79	1.11	1.75	0.92	0.33
Teen report - discussion of social host laws	1.21	0.73	1.23	0.78	1.19	0.67	0.56
Parent report - discussion of social host laws	1.46	0.96	1.46	1.00	1.45	0.92	0.15
Parent drinking quantity	1.18	1.37	1.20	1.40	1.17	1.34	0.22

Intervention Effects

Intent-to-treat (ITT) models.

No statistically significant intervention effects were found for primary outcomes but were found for secondary outcomes (see Supplemental Tables 2–13).

Six-month follow-ups.

Specifically, teens assigned to the experimental condition reported fewer friends who had been drunk in the past six months as compared to teens in the control condition. Also, parents in the experimental group reported higher levels of communication about social host laws.

12-month follow-ups.

At the 12-month follow-up, parents in the experimental condition reported consuming fewer drinks than parents in the control group.

Dosage model.

Significantly better outcomes for families with more program dosage were reported at both the six and 12-month follow-ups (Tables 2, 3, 4, and 5 and Supplemental Tables 14–21).

Table 2.

Dosage model at six-month follow up for primary alcohol use outcomes regarding the past six months

	Alcohol use (N=287)			Drinking frequency (N=267)				Drinking quantity (N=284)				Drunkenness (N=286)				Binge drinking (N=290)
	b	SE	eβ (Odds ratio)	b	SE	β	t	b	SE	β	t	b	SE	β	t	b	SE	β	t
Dosagea	−.91	.33	.40**	−3.60	.99	−.42	−3.62***	−.30	.13	−.22	−2.33*	−.16	.09	−.17	−1.73†	−.07	.06	−.14	−1.33
Condition	−1.31	.56	.27*	−6.02	1.78	−.40	−3.39**	−.39	.23	−.17	−1.74†	−.15	.16	−.09	−.90	−.06	.10	−.06	−.59
Baseline	2.23	.33	9.30***	.07	.04	.11	1.89†	.50	.04	.55	11.14***	.54	.05	.54	10.80***	.42	.05	.43	8.00***
Teen gender	−.30	.33	.74	−1.47	.92	−.10	−1.60	−.07	.12	−.03	−0.62	−.03	.08	−.02	−.32	−.03	.05	−.04	−.68
Teen age	.35	.32	1.41	.55	.94	.03	.58	.06	.12	.03	.52	.05	.09	.03	.59	.05	.05	.05	.90
Teen White	−.29	.35	.75	−.13	1.01	−.01	−.13	.08	.13	.03	.65	.05	.09	.03	.61	.07	.05	.07	1.27
Inverse Mills Ratio	1.96	4.62	7.09	23.68	13.60	.11	.174†	4.70	1.73	.13	2.72**	1.85	1.23	.07	1.50	.86	.74	.06	1.17

Note. For the condition variable, 1 = control, 0 = experimental condition.

^aDosage effect sizes (Cohen’s d): alcohol use d = −.51 frequency d = −.45, quantity d = −.28, drunkenness d = −.21, binge drinking d = −.15

^†p < .10

^*p < .05

^**p < .01

^***p < .001

Table 3.

Dosage Model at Six-Month Follow Up for Primary Alcohol Use Outcomes Regarding the Past 30 Days

	Alcohol use (N=283)			Drinking frequency (N=283)				Drinking quantity (N=283)				Drunkenness (N=283)				Binge drinking (N=293)
	b	SE	eβ (Odds ratio)	b	SE	β	t	b	SE	β	t	b	SE	β	t	b	SE	β	t
Dosagea	−1.01	.42	.37*	−.47	.13	−.38	−3.60***	−.22	.11	−.18	−1.94†	−.17	.06	−.30	−2.90**	−.15	.06	−.27	−2.40*
Condition	−1.26	.65	.28†	−.78	.23	−.35	−3.35**	−.31	.20	−.14	−1.55	−.23	.10	−.22	−2.17*	−.16	.11	−.16	−1.41
Baseline	2.91	.46	18.36***	.52	.07	.41	7.69***	.62	.05	.61	12.78***	.61	.07	.46	8.74***	.27	.06	.24	4.35***
Teen gender	−.16	.41	.85	−.05	.12	−.02	−.38	−.03	.10	−.02	−.32	−.08	.05	−.08	−1.46	−.06	.06	−.06	−1.01
Teen age	0.59	.41	1.80	.13	.12	.06	1.07	−.03	.11	−.01	−.24	.02	.05	.01	.28	.01	.06	.01	.13
Teen White	−.40	.44	.67	−.15	.13	−.06	−1.16	−.02	.11	−.01	−.22	.07	.06	.06	1.21	.08	.06	.07	1.33
Inverse Mills Ratio	--	--	--	−.44	1.78	−.01	−.25	2.51	1.50	.08	1.67†	1.26	.79	.08	1.59	2.35	.85	.15	2.76**

Note. For the condition variable, 1 = control, 0 = experimental condition. For the alcohol use model, the Inverse Mills Ratio was dropped from that model because of computational difficulties.

^aDosage effect sizes (Cohen’s d): alcohol use d = −.55 frequency d = −.43, quantity d = −.23, drunkenness d = −.35, binge drinking d = −.28

^† p < .10

^*p < .05

^**p < .01

^***p < .001

Table 4.

Dosage Model at 12-Month Follow Up for Primary Alcohol Use Outcomes Regarding the Past Six Months

	Alcohol use (N=282)			Drinking frequency (N=265)				Drinking quantity (N=278)				Drunkenness (N=280)				Binge drinking (N=283)
	b	SE	eβ (Odds ratio)	b	SE	β	t	b	SE	β	t	b	SE	β	t	b	SE	β	t
Dosagea	−.55	.32	.58†	−1.99	1.32	−.19	−1.51	−.45	.16	−.30	−2.84**	−.26	.13	−.21	−2.00*	−.07	.06	−.13	−1.13
Condition	−.71	.54	.49	−2.54	2.34	−.13	−1.08	−.72	.28	−.27	−2.55*	−.27	.23	−.12	−1.18	−.05	.11	−.05	−.48
Baseline	1.85	.31	6.37***	.31	.10	.19	3.01**	.44	.05	.44	8.13***	.69	.07	.48	9.23***	.41	.06	.38	6.74***
Teen gender	−.23	.30	.80	−.75	1.17	−.04	−.64	.01	.14	.01	.10	.00	.12	.00	−.03	.03	.06	.03	.60
Teen age	.05	.31	1.05	.24	1.19	.01	.20	−.08	.15	−.03	−.54	.00	.12	.00	.00	.04	.06	.04	.65
Teen White	.17	.34	1.19	.53	1.27	.03	.41	−.11	.16	−.04	−.69	.19	.13	.08	1.52	.03	.06	.03	.47
Inverse Mills Ratio	2.65	4.43	14.21	21.19	17.29	.08	1.23	1.98	2.14	.05	.92	1.79	1.74	.05	1.03	−.06	.85	.00	−.07

Note. For the condition variable, 1 = control, 0 = experimental condition.

^aDosage effect sizes (Cohen’s d): alcohol use d = −.30, frequency d = −.19, quantity d = −.35, drunkenness d = −.24, binge drinking d = −.14

^†p < .10

^*p < .05

^**p < .01

^***p < .001

Table 5.

Dosage Model at 12-Month Follow Up for Primary Alcohol Use Outcomes Over the Past 30 Days

	Alcohol use (N=278)			Drinking frequency (N=277)				Drinking quantity (N=277)				Drunkenness (N=276)				Binge drinking (N=283)
	b	SE	eβ (Odds ratio)	b	SE	β	t	b	SE	β	t	b	SE	β	t	b	SE	β	t
Dosagea	−.99	.38	.37**	−.97	.29	−.39	−3.40**	−.24	.11	−.21	−2.20*	−.12	.07	−.17	–1.57	−.07	.05	−.17	−1.49
Condition	−1.27	.59	.28*	−1.46	.50	−.34	−2.91**	−.35	.19	−.17	–1.79†	−.09	.13	−.08	−.72	−.04	.09	−.05	−.44
Baseline	2.28	.43	9.75***	.47	.14	.19	3.26**	.59	.05	.61	12.56***	.71	.09	.44	8.04***	.30	.05	.34	6.13***
Teen gender	.16	.38	1.18	.13	.26	.03	.51	−.07	.10	−.03	−.71	−.03	.07	−.02	−.45	−.01	.05	−.01	−.22
Teen age	.60	.38	1.82	.13	.26	.03	.48	.04	.10	.02	.40	.11	.07	.09	1.59	.08	.05	.09	1.66†
Teen White	.20	.45	1.22	.12	.28	.02	.41	−.01	.11	.00	−.10	.12	.07	.09	1.65	.02	.05	.02	.36
Inverse Mills Ratio	.82	5.51	2.28	.60	3.87	.01	.16	.19	1.49	.01	.12	−.57	1.02	−.03	−.56	.95	.68	.08	1.39

Note. For the condition variable, 1 = control, 0 = experimental condition.

^aDosage effect sizes (Cohen’s d): alcohol use d = −.55, frequency d = −.41, quantity d = −.27, drunkenness d = −.19, binge drinking d = −.18

^†p < .10

^*p < .05

^**p < .01

^***p < .001

Six-month follow-ups.

At the six-month follow-up, higher program dosage was significantly related to the primary outcomes related to alcohol consumption including a lower likelihood of teens’ reporting any alcohol use in the past six months or past 30 days, less frequent drinking during the past six months and past 30 days, fewer drinks consumed over the past six months, and less drunkenness and binge-drinking during the past 30 days (see Tables 2 and 3). Greater program dosage was also significantly associated with secondary outcomes, including teen report of less communication about safe drinking strategies and both parent and teen reports of greater communication about social host laws (Supplemental Tables 16–17).

12-month follow-ups.

Primary outcomes related to both alcohol consumption and problems were retained at the 12-month follow up. Specifically, greater program dosage was significantly related to lower likelihood of any alcohol use, less frequent drinking over the past 30 days, fewer drinks consumed over the past six months and past 30 days, and less drunkenness over the past six months (see Tables 4 and 5).

Changes in rates of drinking.

The rate of teens’ past six-month drinking among families with no program exposure increased from 25.7% at baseline to 27.3% at 12 months (a 6.2% increase), while the rate decreased from 22.2% to 16.1% for those who had at least completed the alcohol component (a 27.5% decrease). A similar pattern was found for past 30-day drinking, with an increase from 11.1% to 16.0% (a 44.1% increase) in teens with no exposure, and a decrease from 10.3% to 6.9% (a 33.0% decrease) for those completing the alcohol component.

Discussion

Results from the Smart Choices 4 Teens program indicates efficacy for prevention of alcohol use and alcohol-related problem behaviors for older teens when parents and teens completed more of the program content and specifically the Alcohol Prevention component. Given that delay of onset of drinking and heavy drinking are related to better outcomes for teens during their adult years (Dawson et al., 2008; Grant & Dawson, 1997; Hingson & Zha, 2009), efforts to engage parents and teens in changing teens’ behaviors and attitudes towards alcohol use should produce beneficial health outcomes in the future. Not only are health benefits important reasons for delaying and reducing alcohol use among teens, but other social benefits (e.g., jobs, school achievements) may also be expected to improve with the delay in alcohol use and decrease in heavy drinking (Balsa, Giuliano, & French, 2011).

The intervention also showed efficacy for parent-teen communication and parents’ own drinking. Parents in the experimental group reported drinking fewer drinks themselves and discussed social host laws more often. Less parent drinking may reduce the modeling of drinking behavior, which is important as parent drinking may motivate teen drinking (Alati et al., 2014; Armstrong et al., 2013). Families who received more of the program communicated more about social host laws but less about safe drinking strategies. This is consistent with the goals of the intervention. “Safe” drinking strategies such as allowing or “teaching” teens to drink under supervision has been shown to increase risky drinking (Kaynak, Winters, Cacciola, Kirby, & Arria, 2014; Livingston, Testa, Hoffman, & Windle, 2010; McMorris, Catalano, Kim, Toumbourou, & Hemphill, 2011). Of particular importance is that the dosage effects were noted across different alcohol involvement indicators and some of these impacts persisted at the 12-month follow-up period, showing that the program may have sustained effects.

Although fewer significant effects were found when performing a strict ITT comparison of control and experimental conditions, there were significant effects in the desired direction for secondary outcomes. This suggested that the positive impact on alcohol-related behaviors was quite robust, especially since half of the experimental group had no exposure to the alcohol component. As noted earlier, there is a growing concern about using the ITT approach in prevention trials, as low rates of adherence to the intervention protocol are common in familybased prevention programs and can bias results (Gross & Fogg, 2004; Horvitz-Lennon et al., 2005; Ingoldsby, 2010). Therefore, we conducted recommended analysis of dosage that considers nonadherence, in addition to ITT analyses (Gross & Fogg, 2004; Horvitz-Lennon et al., 2005; Ten Have et al., 2008), finding support for intervention effects when program adherence increased.

A clear implication of the results is that strategies must be identified to promote completion of the program if it is to be disseminated wide-scale. An advantage of online programs is low-cost scalability compared to programs that require group facilitators. In this trial, intervention use depended on parents’ and teens’ intrinsic motivations as monetary compensation was linked to completion of surveys, not use of the intervention. The selection model results suggest that non-White teens are less likely to complete the Alcohol component of the intervention. Future studies should focus on determining how to engage minority families at higher levels. Experimental group families who had not logged onto the website in two weeks (n=53) were called to encourage participation. During these informal conversations, over one third of these families (37.7%) revealed that lack of time was a barrier to program completion. Marketing family-based prevention programs to families must compete with other activities in families’ lives. There must be sufficient reason for families to invest time in a program such as Smart Choices 4 Teens and the evidence that the program is effective provides one such motivation. Still, evidence alone may not be enough. Some changes to the program structure may be needed to make the program briefer without harming efficacy to achieve higher adherence. Some parents reported that they did not think their teens “needed” a prevention program because they believed their teens were not engaged in this behavior. In these cases, families may need more knowledge about the value of prevention before temptations to drink may occur, even if there are not problems at present.

Limitations of the study should be noted. Participants were part of an online panel and families that agree to participate in such panels may differ from families who do not. However, samples obtained via web panels are similar to nationally representative samples in regards to alcohol use and demographic background (Heeren et al., 2008), with samples obtained that are diverse in terms of socio-economic status and ethnicity (Rothman, In press), as is the current sample. Considering that nearly all adults (87% of adults 30–49) and teens (92% of teens aged 13–17) use the Internet (Center, 2015a, 2017a), web panels can provide low cost national samples not from a university setting. As noted above, failure to complete the entire Smart Choices 4 Teens program was another limitation. Use of self-report measures of the alcohol outcomes also were open to social desirability biases and demand effects. Although self-report limitations are often exaggerated (Chan, 2008), we used established measures to avoid these errors where possible. Another limitation was the lower reliability of some items (approval of drinking, negative communication style). Listwise deletion in analyses is another limitation. This method can lower statistical power and could cause bias in results if missing data is not random. However, because the study uses longitudinal data, our primary source of missing data is due to participant attrition at follow-up waves. Due to this, instead of imputing entire surveys for missed waves of data collection, we prefer the selection model approach because it takes into account attrition biases.

Rates of past month alcohol use were slightly lower than national rates (SAMHSA, 2017). Although participants were able to complete the online surveys at any location with Internet, participants could have completed the surveys on their home computer. Therefore, it is possible that underreporting of drinking behaviors occurred if teens were worried about their parents seeing their responses. However, national surveys with higher rates also used methods subject to the same issues. For example, the National Survey on Drug Use and Health is administered by interviewers at participants’ homes, with participants responding to questions read aloud by the interviewer and on computers. Monitoring the Future follow-up surveys are conducted through mailed questionnaires to participants, which may be completed at home with parents present. In addition, several studies have found that adolescents (ages 13 to college-aged) who completed surveys online reported similar or higher rates of substance use than when using paper and pencil or in-person methods (Bates & Cox, 2008; Khadjesari et al., 2009; Kypri, Gallagher, & Cashell-Smith, 2004; Miller et al., 2002; Pedersen, Grow, Duncan, Neighbors, & Larimer, 2012; Wang et al., 2005). Follow-up rates present another limitation, although our rates are comparable to those of other family-based intervention studies (79%−89%) (Bauman et al., 2002; Spoth, Trudeau, Guyll, Shin, & Redmond, 2009). Bias may be possible due to this attrition, but this was accounted for in dosage models through the inclusion of the inverse Mills ratio as a covariate.

Overall, the findings suggest that the Smart Choices 4 Teens intervention is efficacious for preventing teen alcohol use for families who complete the alcohol component materials. The online format of the intervention has advantages for wider dissemination, but it needs to be convenient and feasible for busy families (the lack of need to attend an in-person class is an advantage). Future analyses will examine specific mediators of dosage effects to determine how the intervention affects outcomes and will examine ways to disseminate the program cost effectively and in such a way to increase adherence.