Social Mediation of Persuasive Media in Adolescent Substance Prevention

Abstract

Social commentary about prevention messages may affect their likelihood of acceptance. To investigate this possibility, student participants (N = 663) viewed three anti-marijuana advertisements, each followed immediately by videotaped discussions involving four adults or four adolescents using either extreme or moderate language in their positive commentaries. The commentaries were expected to affect participants’ perceptions of the extent to which the ads were designed to control their behavior (perceived control), which was hypothesized to inhibit persuasion. Two indirect effects analyses were conducted. Marijuana attitudes and usage intentions were the outcome variables. Both analyses revealed statistically significant source by language interactions on participants’ perceived control (both p < .02). Further analyses revealed significant indirect effects of language extremity on attitudes and intentions through perceived control with adult, but not peer sources (both p < .05). These perceptions were associated with more negative marijuana attitudes and diminished usage intentions when adults used moderate (vs. extreme) language in their favorable ad commentaries (both p < .05). The findings may facilitate development of more effective prevention methods that emphasize the importance the role of perceived control in persuasion, and the impact of interpersonal communication variations on acceptance of media-transmitted prevention messages.

The mass media have been called upon increasingly to draw attention to factors that undermine health and to encourage behaviors that promote well-being. The focus of these efforts ranges from HIV/AIDS prevention to smoking avoidance to eating the daily requisite number of vegetables. Although some campaigns have resulted in beneficial outcomes, they often fail (Crano, Siegel, & Alvaro, 2012; Hornik et al., 2001; Noar, 2006). Research on the interpersonal mediation of media effects suggests media campaigns may be improved through a better understanding of the ways the mass public processes media-based persuasive communications. Nearly 70 years ago, Lazarsfeld and colleagues (1948) argued that the mass media operated indirectly via authoritative commentators (opinion leaders) who transmitted their acceptance or rejection of communicated information to their audiences: “Influences stemming from the mass media first reach ‘opinion leaders’ who, in turn, pass on what they [see,] read and hear to those of their every-day associates for whom they are influential” (Lazarsfeld, Berelson, & Gaudet, 1948, p. 151; also Katz & Lazarsfeld, 2006; Lowery & DeFleur, 1995).

In this two-step flow of communication model, the actions of authoritative sources help determine the persuasive success or failure of a communication. Media effectiveness is viewed as dependent in part on receivers’ responses to factors that affect their perceptions of these sources, and secondarily on the communication’s more direct effects. Message variables that affect opinion leaders’ reactions also influence opinion followers, but the theory’s shift in focus from direct to mediated persuasion has important implications for the ways campaigns are designed and assessed. In practice, some campaign designers appear implicitly to have recognized the wisdom of the mediated, two-step flow model. Campaigns that encourage parents to talk with their children about the dangers of substance misuse, for example, are consistent with this model, assuming children see their parents as authoritative opinion leaders (Huansuriya, Siegel, & Crano, 2014). Conversely, prevention ads that urge friends to talk with friends about substance related dangers may not prove as successful if the communicators (i.e., the assumed opinion leaders) who act as mediators between the persuasive message and the mass audience are not viewed as worthy of close consideration (Flanagan, Elek-Fisk, & Gallay, 2004). Considering features that lend to the source’s authority, along with the source’s presentation style, may prove an important addition to the two-step flow model when applied to mass persuasion campaigns. Messages often are attributed to spokespersons that campaign organizers assume will be accepted by their targeted audience. The validity of this assumption may have much to do with a campaign’s success or failure.

Language Extremity and Source Status

Language extremity and status, two factors that may influence the persuasive impact of opinion leaders, are the focus of the current investigation. Using the insights of the two-step flow model, both of these variables are hypothesized to influence audience members’ perceptions of the persuasive intent of media-based communications. Perception of persuasive intent has been shown to attenuate message acceptance (Blakenship, Wegener, & Murry, 2012; Brock, 1967; Crano & Alvaro, 2013; Lienemann & Siegel, 2016; Quick & Stevenson, 2007; Quinn & Wood, 2004; Tormala, 2008). Investigating this factor thus may broaden our understanding of factors that may be employed advantageously to achieve positive prevention outcomes.

Extremity

Language extremity refers to the degree to which the wording of an appeal is explicit, directive, and commanding (Dillard & Shen, 2005). Such language often backfires in persuasion, as extremity implies dominance, which can arouse resentment and resistance (Siegel & Burgoon, 2002). At the core of the language extremity construct is the idea that extreme and intense messages convey the impression that sources believe message recipients are subject to their control. This impression arouses resistance, or reactance, because it threatens recipients’ freedom of choice (Bessarabova, Fink, & Turner, 2013). Moderate, implicit language may foster positive responses, and thereby enhance influence. Psychological reactance theory (PRT: Brehm, 1966) affords a useful framework for language extremity effects. It suggests messages posing a threat to receivers’ freedom arouse reactance. A growing body of reactance research has emerged in prevention and health promotion (Erceg-Hurn & Steed, 2011; Hall et al., 2016; Rhodes, Ralston, & Bigsby, 2016; Richards & Banas, 2015). It implies that improperly constructed messages give rise to reactance, which attenuates and sometimes reverses intended effects (Dillard & Shen, 2005; Grandpre, Alvaro, Burgoon, Miller, & Hall, 2003; Miller, Burgoon, Grandpre, & Alvaro, 2006; Quick & Stephenson, 2007).

Reactance or resistance may be especially relevant in contexts involving social mediation of persuasive communications. Reactions to a persuasive message that stimulates positive cognitive responses, which Petty and Cacioppo (1986) viewed as essential for message acceptance and attitude change, may be positively affected by mediated opinion leaders who are accepted by an audience; conversely, acceptance may be blocked or reversed by interactions with mediated sources who generate resistance (Knowles & Linn, 2004; Petty, Ostrom, & Brock, 1981). Thus, an individual’s positive response to a prevention message may be undone in response to an opinion leader’s commentary, which motivates a reevaluation and of a heretofore cogent and credible communication. The converse also may hold. An opinion leader may neutralize negative reaction that surfaces as a result of a receiver’s initial exposure to a communication (Miller, Lane, Deitrick, Young, & Potts, 2007).

Source status

Research on resistance has demonstrated the powerful effects of the message source in persuasion. Brehm and Brehm (1981) used the term “social power” to describe a source’s capacity to reward or punish. They suggested that messages emanating from sources of high social power and low legitimacy would arouse greater resistance than messages from low-power sources. Thus, the reactance model supports the hypothesis that peer sources will elicit less resistance than adults, because they are unlikely to have the power to impose sanctions for lack of compliance (thereby arousing reactance). In the current research, we assume that adults, who generally hold coercive power over adolescents, will be viewed as more authoritative than peers, and thus may generate greater resistance, and consequently less compliance, than peers. However, under appropriate conditions, authoritative sources (adults) may be as influential as peers as sources of substance preventive communications. Insofar as adults may be more likely than adolescents to advance prosocial (anti-substance use) norms, identifying factors that enhance their effectiveness is a laudable ambition. Social psychological research on unexpected source-statement incongruities provides insight into this possibility.

Unexpected source-statement combinations

Research on the combination of source and language extremity suggests that persuasive message effects may be amplified when their sources advocate views that are unexpectedly at odds with their established (or expected) positions. For example, Baum and Groeling (2009, p. 251) found that unexpected support from a member of an opposing political party, which could involve a real cost to the message source, was significantly more influential than praise from in-party members, which they labeled “cheap talk.” Similarly, Koeske and Crano (1968) found unattributed statements rated as not credible were significantly more persuasive when attributed to a source from whom the message was unexpected and potentially costly. This study did not find an enhancement of source credibility for sources stating a position at odds with their self-interest; the persuasive effect operated through an enhancement of message credibility, not source credibility. Arguably, these effects occur because an unexpected or costly paring of source with statement attenuates perceptions of the message’s intent to control. Such attributions render the incongruous source’s communication more acceptable and influential than the “cheap talk” emitted by communicators who expectedly parrot the company line. Research on language expectancy theory is consistent with this possibility: an extreme statement that is not expected is considerably more influential than the same statement when it is expected, and vice versa (see Brooks, 1970; Burgoon & Siegel, 2004).

Hypotheses

Considerations of language extremity, source status, and source-statement incongruity give rise to the following hypotheses: Message source (peer/adult) and language (moderate/extreme) will interact, affecting participants’ perceptions of the extent to which the communication was designed to control their behavior (hereafter, ad control) (H1). When the message source is a peer, language will not significantly affect ad control; with adult sources, moderate language will lead to lower perceptions of ad control than extreme language. We hypothesize perceptions of ad control will predict marijuana outcomes (attitudes and intentions to use) over and above the direct effect of message features (source and language (H2). A measure of actual post-treatment substance use was not feasible, but behavioral intentions have been widely and successfully used as predictors of health-relevant behavior (e.g., Albarracin, Johnson, Fishbein, & Muellerleile, 2001; Sheppard and colleagues, 1988; Sutton & White, 2016; Tyson & Covey, 2014), and have been adopted in the current research. Finally, we hypothesize an indirect effect of the interaction between experimentally manipulated message features on marijuana outcomes through perceived ad control (H3).

Method

Overview

Anti-substance advertisements broadcast in an earlier national anti-substance campaign were used as stimuli. They were chosen by a group of six expert evaluators from a larger pool of 50 available ads developed by the Partnership for a Drug Free America (PDFA). Following individual analysis of all 50 ads, two senior investigators agreed on three main ad types — animated/humorous, dramatic/off-screen narration, and slice of (real) life. Six researchers viewed all ads and independently nominated the three most persuasive within each ad type. Following discussion, rankings were compared and one ad was chosen from for each ad type.

All participants were exposed to all three anti-marijuana ads. Ad order was counterbalanced. Immediately following each ad, four videotaped confederates, adolescents or adults, discussed it using either extreme or moderate language. Outcome measures assessed subjects’ perceptions of the extent to which the ads were designed to control their behavior (ad control), their attitudes toward marijuana, and their usage intentions (see Figure 1).

Figure 1. Hypothesized model of ad effects on marijuana outcomes (attitudes and intentions).

Participants

Data were collected from 663 students attending 13 different schools in Arizona (N_elementary = 257, N_{middle school} = 205, N_{high school} = 201). Five of the 13 schools were located in two urban settings, Phoenix and Tucson, and provided 282 students. The other eight schools were located in three rural cities, Tombstone, Wilcox, and Benson, and accounted for the remaining 381 students. Participants were not asked to disclose their age; however, their grade level was recorded. More than 60% of the sample was drawn from grades five through seven, and approximately 35% from grades 8 through 12. Gender was fairly evenly divided, with 48.4% female participants. The majority of participants self-identified as Caucasian (69.8%), with 22.9% identifying as Hispanic, 3.9% as Black, 2.7% as Asian, and 9.7% as Other. Sixty participants (9%) selected more than one race. The majority of participants reported never having used marijuana (81.6%). Of those who said that they had used marijuana, 50 (7.5%) reported using marijuana in the past 30 days.

Consent was obtained from school administration and students’ parents/guardians. Assent was obtained from all participants, who were randomly assigned to one of the experimental conditions formed by the factorial combination of the independent variables. Data were reasonably complete, with no more than 10 missing participants in any analysis.

Procedure

Laptop computers were placed in students’ schools (e.g., libraries or classrooms). The investigators supervised all sessions. Up to 25 participants were seated individually at computers, and after providing assent, completed a computerized questionnaire that gathered socio-demographic data. They were randomly assigned by computer algorithm to experimental conditions, and received the specific videotaped anti-drug messages for that condition. After viewing all three compound videos, participants completed items assessing the outcome variables. Study duration averaged 17 minutes.

Experimental manipulations

Each anti-drug video was composed of two elements: the ad, and a short appended video of adolescent or adult actors who used either extreme or moderate language when commenting on the ad that they had ostensibly viewed. All sources agreed with the content of the ads, but their scripted language varied in extremity. The commentators used either extreme, highly directive, explicit language when discussing the messages (e.g., “People don’t understand – you can’t do this sort of stuff!” and “Only complete idiots would do drugs.”), or moderate and non-extreme language (“It’s your health, your body, your brain.” and “When it comes right down to it, it’s your decision.”). Participants witnessed the assigned videotaped discussion group’s commentaries immediately after each of the three ads they viewed. For each participant, the source and language manipulations were consistent across the three communications (i.e., those in the peer source, extreme language condition viewed videos of the same peer sources using extreme language in response to all three ads).

Outcome Measures

All outcome measures were collected via computer after participants had been exposed to all three of the ads (with commentaries) they had been randomly assigned to receive.

Perceived ad control

This variable was assessed with the following item: “The commercials in the videos I just watched were trying to tell me what to think.” Responses were indicated on five-point Likert-type scales ranging from “strongly disagree” to “strongly agree.” Higher scores indicate greater agreement.

Attitudes toward marijuana

Participants’ attitudes toward marijuana were assessed with the following items: “Marijuana is “bad,” “stupid,” “cool,” and “fun. Responses ranged from (1) “strongly agree” to (5) “strongly disagree.” Items were scored so that higher scores indicated more positive attitudes toward marijuana use (α = .91). A composite score was computed by averaging the four items. Mean scores ranged from one to five. Higher scores represented more negative attitudes toward marijuana.

Marijuana usage intentions

This measure was assessed with the following items: “Do you intend to smoke marijuana in the next year” and “Do you think you might try marijuana in the future?” Responses, on five-point scales, could range from “definitely yes” to “definitely no.” Higher scores indicated stronger intentions to use marijuana (α = .88, r = .79, p < .001). A composite score was computed by averaging the two items: mean scores ranged from one to five.

Covariates

Participants were asked their school grade and sex. In addition, as an indicator of friends’ norms, they reported whether they thought their friends (a) disapproved of marijuana use, and (b) would be bothered by their use. These binary items were summed to form an index score (KR-20 = .77). These three variables were entered as covariates in all analyses.

Results

Data were collected across 13 schools. Students within schools were assigned randomly to experimental conditions. To rule out school-specific bias, we calculated an intraclass correlation between school and lifetime marijuana use to determine the association between usage and attending a specific school. The relation (ICC = .04) indicated a weak school-usage association, and so the individual respondent was used as the unit of analysis (Crano, Brewer, & Lac, 2015b). Differences in the socio-demographic factors among the experimental conditions also were tested. This analysis disclosed no significant differences among conditions on grade level, sex, or friends’ perceived marijuana usage norms (for descriptive statistics, see Table 1). We had planned to study effects of usage variations in the analyses, but the number of users afforded insufficient power to detect moderated mediation effects. However, the percentages of users in each condition did not differ, χ²(1, N = 122) = .57, p = .45 , and nonusers, χ²(1, N = 541) = .15, p = .70.

Table 1. Descriptive Statistics for Full Sample and by Condition.

	Full Sample	Adult/Explicit	Adult/Implicit	Peer/Explicit	Peer/Implicit
	M (SD)	M (SD)	M (SD)	M (SD)	M (SD)
Ad Control	3.14 (1.23)	3.32 (1.20)	2.93 (1.25)	3.16 (1.26)	3.15 (1.19)
Attitudes	1.66 (1.01)	1.74 (1.08)	1.55 (0.91)	1.56 (0.85)	1.49 (0.83)
Intentions	1.58 (0.93)	1.83 (1.22)	1.57 (0.96)	1.69 (1.01)	1.54 (0.81)

Analytic Approach

Listwise deletion was used across all analyses, resulting in the removal of 10 participants who declined to respond to measures of attitudes toward marijuana. The hypotheses relating perceived ad control with susceptibility or resistance to persuasion were tested via indirect effects modeling. Manipulated variables’ impacts were hypothesized as due in part to their indirect effects through participants’ perceptions of the degree to which the communications were designed to control their behavior (perceived ad control). The three covariates - grade, sex, and friends’ marijuana norms - were entered in all analyses.

Hayes’ (2009) PROCESS Model 8 (version 2.13; SPSS 21) was used to examine the indirect effect of the interaction between source and language treatments through perceived ad control on marijuana-related attitudes (Analysis 1) and marijuana usage intentions (Analysis 2). Hayes’ (2009) bootstrapping approach was used to create empirical representations of the underlying sampling distributions of the indirect links by repeatedly resampling the obtained data, defined 95% bias corrected confidence intervals. The attitude and intention data were sampled 10,000 times to create the distributions.

Analysis 1 – Marijuana Attitudes

Before considering the full indirect effect of the interaction of source and language on marijuana attitudes through perceived ad control, we examined the conditional effects of source and language variables on perceived ad control. A moderated mediation analysis (Hayes’ Model 8) disclosed a significant interaction of source with language on perceived ad control, t(646) = 2.15, p = .032, 95% CI[.035, .781], as predicted (H1). A simple slopes analysis of the interaction was conducted using Hayes’s PROCESS Model 1. Within peer source conditions, the analysis disclosed no significant effect of language extremity on perceived ad control t(646) = .05, p = .96, 95% CI[-.256, .270]. However, perceived ad control was significantly attenuated when adult sources used moderate, as opposed to extreme language in their commentaries, t(646) = -2.97, p = .003, 95% CI[-.665, -.136], supporting H2. A graphic representation of this interaction is presented in Figure 2.

Figure 2.

Model of the interaction between source (adult or peer) and language (extreme or moderate) on participants’ perceptions of ad control. Standard error estimates for error bars were derived using ANCOVA.

The analysis then turned to the link between perceived ad control and participants’ attitudes toward marijuana. As predicted (H3), perceived ad control was significantly associated with participants’ attitudes toward marijuana, t(645) = 2.11, p = .036, 95% CI[.002, .050]: Higher levels of perceived ad control were associated with more positive attitudes toward marijuana over and above the effects of grade, sex, friends’ marijuana norms, and the source by language interaction.

The full model disclosed no significant direct effects of peers’ language variations on subjects’ attitudes toward marijuana, t(645) = .963, p = .34, 95% CI[-.058, .169]. However, using bootstrapped confidence intervals, the analysis indicated a significant indirect effect of language on marijuana attitudes through perceived control when the source was an adult, B = -.089, SE = .045, t(645) = -1.99, p = .047 95% CI[-.176, -.001], but not a peer, B = -.033, SE = .036, t(645) = -.918, p = .359, 95% CI[-.103, .038], supporting H4. When the source was an adult, moderate (vs. extreme) language resulted through perceived ad control in significantly less positive attitudes toward marijuana (see Figure 3 for depiction of the full model).

Figure 3.

Test of the indirect effect of the language by source interaction on marijuana attitudes through perceptions of ad control. Weights and standard errors derived using PROCESS Model 8.

Analysis 2 – Marijuana Usage Intentions

A second analysis was conducted to examine the conditional effects of the source and language manipulations on marijuana usage intentions through perceived ad control (see Figure 4). Results for the interaction of source and language extremity on perceived ad control were identical across analyses because the same sample was used in Analysis 1 and Analysis 2. As with Analysis 1, a moderated mediation analysis (Model 8) disclosed a significant interaction of the two manipulated variables on perceived ad control, t(646) = 2.15, p = .032, 95% CI[.035, .781] (H1). A simple slopes analysis was conducted to examine this result. In the peer source condition, there was no effect of language extremity differences on perceptions of ad control, t(646) = .05, p = .96, 95% CI[-.256, .270]. However, participants’ perceptions that the ads were meant to control their actions were significantly weaker when adults used moderate, rather than extreme language, t(646) = -2.97, p = .003, 95% CI[-.665, -.136], supporting H2 (see Figure 2).

Figure 4.

Test of the indirect effect of the language by source interaction on marijuana intentions through perceptions of ad control. Weights and standard errors derived using PROCESS Model 8.

The analysis then turned to the link of perceived ad control with intentions to use marijuana. Perceptions that the ads were controlling were significantly associated with usage intentions, t(645) = 2.45, p = .014, 95% CI[.014, .126]. As predicted (H3), higher levels of perceived control were associated with stronger intentions to use marijuana over and above the effects of grade, sex, friends’ marijuana norms, and the interaction of source with language.

Analysis of the direct effects of source and language on intentions revealed modest language effects for peer sources, t(645) = -1.687, p = .092, 95% CI[-.316, .024]: moderate (vs. extreme) language was associated with somewhat, but not significantly lower intentions to use marijuana, B = -.146, SE = .087. However, participants’ intentions to use marijuana were significantly lower when adult sources used moderate as opposed to extreme language in their ad assessments, B = -.227, SE = .095; t(650) = -2.382, p = .018, 95% CI[-.414, -.040].

Bootstrapped confidence intervals again were calculated to examine indirect effects, and they replicated results found in the analysis of participants’ attitudes toward marijuana. The analysis revealed a significant indirect effect of language on marijuana usage intentions through perceived control with adult sources, B = -.028, SE = .015, 95% CI[-.068, -.006], but not with peer sources, B = .001, SE = .010, 95% CI[-.021, .022], supporting H4 for marijuana intentions.

Discussion

Systematic evaluations and meta-analyses assessing changes in health behavior provide evidence that some measure of success may be expected from mass media campaigns (Derzon & Lipsey, 2002; Snyder, 2007). In reviewing a decade of health media campaign research, for example, Noar (2006) suggested that future research consider the effects of interpersonal communication about mass media campaign messages; he argued that such communication may affect campaign outcomes (also see Boulay, Storey, & Sood, 2002; Hafstad & Aaro, 1997; Hafstad, Aaro, & Langmark, 1996). Our results support the wisdom of Noar’s ideas, and indicate that responses to discussions about campaign messages may help determine the success or failure of campaigns. Few studies have has addressed this issue systematically, despite widespread recognition that interpersonal communication about media campaigns may build on general message awareness to promote attitude and behavior change (de Vreese & Boomgaarden, 2006; Durkin & Wakefield, 2006; Flay 1987; Morton & Duck 2001).

This analogue experiment was designed to build upon these insights by incorporating insights from Lazarsfeld and colleagues’ (1948) two-step flow of communication model with research that points to adolescents’ strong reactance to manipulation (e.g., Crano, Alvaro, & Siegel, 2015a; Grandpre et al., 2003; Knowles & Linn, 2004). In a persuasive media context, this integration of Lazarsfeld’s model with resistance to others’ attempted control suggests that authoritative (opinion) leaders’ power to influence their adolescent audiences may be enhanced by moderating the extremity of the language they use in promoting their positions. Using extreme language to characterize a media communication led to stronger perceptions of attempted control, which weakened the anti-marijuana messages’ sought-for effects. Authoritative opinion leaders (adults) who used moderate, non-directive language when commenting on a series of anti-marijuana messages generated significantly lower perceptions of persuasive intent in their adolescent audience. In turn, these lower perceptions of attempted control were associated with participants’ marijuana related attitudes and usage intentions: adolescents exposed to adults’ moderate and positive evaluations of the media messages expressed lower perceptions of the source’s intent to persuade, and these perceptions were associated significantly with less positive attitudes toward marijuana and to lower intentions to engage in its use.

It must be stressed that all participants viewed identical marijuana-prevention ads; only the source and language used by the adults or peer sources who commented on them differed. Consistent with the two-step flow model, these findings suggest that the ways prevention ads are characterized by authoritative message sources (or opinion leaders, in Lazarsfeld’s terms) may significantly affect audience members’ ad perceptions and evaluations. The language used in describing the ads, that is, became an integral feature of the ads themselves.

Participants’ relatively flat response to peer commentators, irrespective of language extremity, was consistent with two-step flow logic. Peers generally are not viewed as having coercive power, and thus, the extremity of their language is less relevant than in contexts involving adults, who often do hold this power. As such, extremity variations in the language peers used in proscribing substance use was not expected to have a differential effect on perceived control, and it did not.

These results foster potentially useful suggestions for developing substance prevention interventions, and for designing research on the role of ad discussion in response to mass media campaign communications. First, it may be that encouraging authoritative sources to talk to children about substance use is not sufficient to deter initiation, as differences in the manner in which the advice is delivered may lead to widely different outcomes. For example, our analyses suggest that research on interventions informing parents about what to say about substance use, and how to say it, is vital. Missteps in such discussions may lead to failure or, in the worst case, may exacerbate the very problems they were intended to prevent. On the other hand, if done properly, the intervention of knowledgeable trusted, and skilled communicators could enhance the impact of media-based prevention campaigns significantly (Huansuriya et al., 2014).

This view is consistent with research on other health relevant issues (e.g., Fasula & Miller, 2006; Whitaker, Miller, May, & Levin, 1999). Whitaker and colleagues, for example, found that parent-teen discussions about sexuality and sexual risk behaviors were related to increased likelihood of teen-partner discussions about sexual risk and teens’ condom use, but only if teens perceived parents as open, skilled, and comfortable in having such discussions. Similarly, Fasula and Miller (2006) found that discussions about sex between African-American or Hispanic adolescents and their mothers positively affected preventive behaviors, but only when the mothers’ communications were perceived as open, comforting, and understanding. Arguably, parents and other respected message sources can enhance the effects of media based prevention communications, but they also can reduce or reverse their message’s intended outcomes if their presentations are enacted unwisely. Prevention campaigns based on the implications of the two-step flow model might be especially effective if opinion leaders were the focus of the preventive communication. Such an approach could enhance the effectiveness of parents or other responsible adults as prevention models, and might indirectly affect the adolescent audience as well. When parents are targeted by a persuasive message, for example, their children who “overhear” the preventive material prove less resistant to the appeal (Crano et al., 2007).

Limitations

A limitation of the research is that only a single item was used to measure perceptions of ad control, thus increasing the susceptibility of this measure to response instability. Generally, single-item measures are less stable than more reliable multi-item indicators, and enhance the likelihood of measurement error. Error tends to contribute to Type II error, and so is seen as possibly masking stronger findings; however, it is never a positive feature, even with supportive results. A more extensive measure would have been a better option. Another potential limitation is that the sample was drawn from a single state, and thus the results might not generalize to the wider population of adolescents. Although generalization is never assured, the sample was very large, diverse, and drawn from rural and urban schools. Because the number of users in the sample was small, there was insufficient power to detect moderated mediation effects of usage status. However, the percentages of users in each condition did not differ significantly, so this factor is unlikely to have affected the results. A final caveat: Even though the research involved use of a true experimental design, causal interpretation of results of moderated mediation analyses must be approached with caution. The experimentally induced interaction of source with language strongly supported our contention that these factors combine in unique, predictable ways to affect participants’ perceptions of the intent to control, and of their subsequent attitudes and usage intentions. The mediation of the experimentally manipulated factors of attitudes and intentions through perceived control, however, does not enjoy the same causal priority as the experiment, and hence must be construed carefully. Although a causal interpretation of the mediation effect is not inconsistent with the data, it is not certain.

Conclusion

The pattern of results, together with complexities suggested by past research (Durkin & Wakefield, 2006; Gunter et al., 2006; Kelly & Edwards, 1992), support the proposition that analyses of discussions of mass media campaigns should be concerned with more than main effects (Crano et al., 2015a). In the current research, discussion of the ads was itself mediated. As such, the study’s outcomes are most applicable to mediated contexts in which individuals or groups of individuals offer their comments on the proffered campaign messages. Our goal was to examine theoretical predictions within the framework of the two-step flow of communication model in a controlled setting. The approach was well suited to such an endeavor, and it took advantage of a large adolescent sample. Examining and being cognizant of the influence of commentaries on mass media prevention ads promises to open fruitful avenues for research, which may benefit considerably from considering media effects models that suggest a role for interpersonal, media-related discussion (Lazarsfeld et al., 1948; McCombs & Shaw, 1972). To date, such models have driven little systematic research in substance use prevention. We believe these findings may stimulate more intensive analysis of the interpersonal factors that affect ad acceptance of mass media prevention communications.