Attitude Ambivalence, Friend Norms, and Adolescent Drug Use

Abstract

This study assessed the moderating effects of attitudinal ambivalence on adolescent marijuana use in the context of the theory of planned behavior (TPB). With data from the National Survey of Parents and Youth (N=1,604), two hierarchical multiple regression models were developed to examine the association of ambivalent attitudes, intentions, and later marijuana use. The first model explored the moderating effect of ambivalence on intentions to use marijuana; the second tested the moderation of ambivalence on actual marijuana use 1 year later. Results across both analyses suggest that ambivalence moderated the association of friend norms and subsequent adolescent marijuana use: friend norms were better predictors of marijuana intentions (β=0.151, t=2.29, p=0.02) and subsequent use when adolescents were attitudinally ambivalent about marijuana use (β=0.071, t=2.76, p= 0.006). These results suggest that preventive programs that affect the certainty with which adolescents holds pro- or antimarijuana attitudes may influence the likelihood of their resistance to, initiation, or continuance of marijuana use.

National surveys indicate that nearly half the students in the USA will have tried marijuana at least once before the end of high school (Johnston et al. 2010; National Institute on Drug Abuse 2006). These findings are troublesome. Adolescent marijuana use has been associated with a range of harmful consequences, including delinquency, developmental and cognitive impairment, deficits in learning ability, increased risk of contracting sexually transmitted diseases, respiratory problems, poorer academic achievement and higher school dropout, greater delinquency and affiliation with delinquent peer groups, more positive expectancies regarding outcomes of marijuana, and future polydrug use (Block and Ghoneim 1993; Boyer et al. 1999; Brook et al. 2002; Lac et al. 2009; Lundqvist 2005; Lynskey and Hall 2000; Lynskey et al. 2003; Skenderian et al. 2008).

Considerable research on adolescent substance misuse has been focused on the determinants of usage, such as parental involvement, personality characteristics, and the social environment (Lac and Crano 2009; Penning and Barnes 1982; Petraitis et al. 1995; Siegel et al. 2009). Less frequent is research on adolescents’ attitudes toward illicit substances and how qualities of these attitudes relate to behavior. Psychologists have long been at odds over the relationship between attitudes and behaviors (Crano and Prislin 2006; Eagly and Chaiken 1995; Fazio and Zanna 1978; McGuire 1985; Petty and Cacioppo 1986; Wicker 1969; Zanna et al. 1980), but today the consensus is that attitudes can cause actions, but a number of variables moderate the attitude–behavior relation (Ajzen and Fishbein 1977; Cacioppo et al. 1997; Crano 2012; Fishbein and Ajzen 1975; Petty and Krosnick 1995; Priester and Petty 1996, 2001). Of the variables identified as potential moderators of the attitude– behavior link, attitudinal ambivalence is one of the most promising, but has received perhaps the least attention (Conner and Armitage 2008). Considerable research suggests that ambivalence moderates the attitude-behavior bond (Cooke and Sheeran 2004). However, the question whether this observation holds in the case of adolescent substance misuse, and how it might operate, remains unanswered.

To shed light on this issue, this research was designed to investigate the effects of attitude ambivalence. It is framed within the context of a widely used predictive model which links attitudes and subsequent behaviors, the theory of reasoned action (TPB; Ajzen 1988, 1991; Ajzen and Cote 2008). The research was designed to explore the relative importance of attitudinal ambivalence by considering its effects on a topic of importance to many adolescents, the initiation and continued use of marijuana. It is reasonable to assume that ambivalence plays an important role in many adolescent nonusers’ decisions to initiate or avoid marijuana use. Youth identified in earlier research as vulnerable nonusers, for example, would seem in many instances to exemplify the category of ambivalent (Crano et al. 2008). If so, then preventive models that capitalize on a clearer understanding of the role of ambivalence in attitude–behavior consistency may have greater chances of success. The research is designed to assess the importance of the ambivalence factor within the framework of the TPB, and thereby provide potentially useful insights into prevention.

Attitudes, Actions, and Ambivalence

Most operationalizations of attitude are developed in such a way that it is not possible to hold both positive and negative evaluations simultaneously (Conner and Armitage 2008). Attitudes usually are defined as falling along a unidimensional continuum: people may hold positive or negative evaluations of an object, but not both. According to this view, a person’s attitude toward marijuana can range from extremely negative to extremely positive, but it would not be possible for the same person to hold both positive (e.g., marijuana is “cool”) and negative (e.g., marijuana is dangerous) evaluations of the drug concurrently. It might be argued that the midpoint of this theorized unidimensional drug evaluation continuum indicates concurrent positive and negative evaluations, but mid-range responses also can indicate either no opinion or indifference. Given the inherent limitations of the unidimensional continuum, it is difficult to differentiate those holding positive and negative beliefs about marijuana from those who are indifferent to it.

A multidimensional view of attitudes has characterized research on ambivalence, which suggests that people can hold positive and negative attitudes toward an object simultaneously (Cacioppo et al. 1997; Priester 2002; Priester and Petty 1996, 2001; Thompson et al. 1995). According to the ambivalence school of thought, attitudes fall along a bidimensional continuum: people can hold both positive and negative feelings about an object concurrently. Attitude ambivalence thus refers to the conflict of simultaneous evaluations (Armitage and Conner 2004; Conner and Armitage 2008; Conner and Sparks 2002).

Ambivalent attitudes are thought to cause evaluative tension as a result of conflict between the negative and positive features of the attitude object (Brown and Farber 1951; Kaplan 1972; Priester 2002). For example, if adolescents believe they will have more friends if they use marijuana, but also fear the cognitive consequences of the drug, they will experience cognitive conflict. This conflict results in aversive evaluative tension (Hass et al. 1992) of which the individual may or may not be consciously aware. In either case, almost all drive-reduction theories hold that this tension is experienced as undesirable or unpleasant, and that accommodations will be made to reduce it if possible (Preister 2002). If one the components of the attitude (say, the positive evaluation) were to change, bringing it into congruence with the other (the negative evaluation), ambivalence would be reduced. This change could be the result of a conscious decision on the part of the individual, but it need not be so. Thus, evaluative tension reduction may be compatible with the two major conceptualizations of ambivalence, felt and potential.

Measurement of Ambivalent Attitudes

There is disagreement about the most appropriate way to measure attitude ambivalence (Conner and Sparks 2002; Priester and Petty 1996; Thompson et al. 1995), but most ambivalence research uses either the felt or the potential ambivalence method. Using bipolar scales, felt ambivalence typically is assessed by respondents indicating the degree of conflict they feel when evaluating an attitude object. A strength of this approach is that it seems to tap the discomfort experienced when evaluations conflict. However, this method has been criticized on the grounds that extraneous factors may affect felt ambivalence (Bassili 1996) and people’s cognitive access to this type of information is far from established (Nisbett and Wilson 1977; Zajonc 1980). An alternate approach involves measurement of potential ambivalence, in which separate items are used to tap the positive and the negative evaluations of an attitude object (Armitage and Arden 2007; Kaplan 1972; Newby-Clark et al. 2008). These positive and negative evaluations are combined to yield a single ambivalence score.

Behavioral Intentions and the Theory of Planned Behavior

TPB describes the network relations relating attitudes to intentions and behavior. It posits that behavioral intentions (the action tendency to engage in a behavior) are the best predictors of action; they are a function of attitudes (evaluative beliefs about a person or object), subjective norms (estimates of important others’ approval or disapproval ofan action), and perceived behavioral control (beliefs about one’s ability to overcome obstacles in enacting a targeted behavior). In the TPB, the more positive a person’s attitude toward a behavior, the more likely is the person to intend to act on that attitude. A second determinant of intention is the subjective norm, the social pressure on the actor to engage or refrain from enacting the implicated behavior. Subjective norms are the perceived opinions of important others, individuals or groups. In this research, subjective norms refer to the adolescent respondents’ estimates of their friends’ beliefs. The final determinant of intention, perceived behavioral control (PBC), theoretically may affect behaviors either directly or indirectly through its effect on intention. Along with PBC, subjective norms and attitudes combine interactively to determine behavioral intentions and subsequent behavior; as well, intentions mediate the relationship between attitudes, PBC, subjective norms, and behavior.

Theory of Planned Behavior and Ambivalence

TPB holds that strong attitudes are more likely to guide behavior than weak attitudes (Converse 1995), but that intentions completely mediate the relation between attitudes and behaviors. Krosnick and Petty (1995) defined strong attitudes as those that fall at either end of the evaluative continuum; however, in the mid-range, this definition conflates those who are indifferent with those who are ambivalent. Other research has conceptualized strong attitudes as unambivalent (Conner et al. 2003). According to this definition, low attitudinal ambivalence should lead to a stronger relationship between attitudes and behavioral intentions, which in turn should enhance attitude-consistent behavior (Crano and Prislin 2006).

Moore (1973, 1980) found stronger relationships between attitudes and behaviors in unambivalent (vs. highly ambivalent) respondents. Following this line of research, Sparks et al. (1992, 2001) showed that higher levels of ambivalence attenuated the relationship between attitudes and intentions. These studies have been criticized for employing cross-sectional designs that did not allow researchers to distinguish the direction of effects (Conner et al. 2003). However, in a longitudinal study, Armitage and Conner (2000) demonstrated the moderating effect of ambivalence on the relationship between attitudes and behavioral intentions. Conner et al. (2003) followed this research with an investigation of the healthy eating behaviors of adults in the UK. The researchers predicted that ambivalence toward eating a healthy diet would weaken the relationship between attitudes and intentions, attitudes and behavior, and PBC and behavior. Results supported these expectations, suggesting that ambivalence may be a significant moderator in the TPB. This study made no predictions regarding the influence of ambivalence on subjective norms. It could be that the effects of subjective norms in adults differ from those experienced by adolescents, because in adolescence, peer influence, a major source of subjective norms for adolescents, assumes greater importance in developing attitudes and behavior (Hartup 2005), as group behavior is viewed as crucial in adolescence (e.g., Palmonari et al. 1989, 1990).

Peer group membership is beneficial for social development and feelings of self-worth (Cottrell 2007; Palmonari et al. 1990). Adolescents often strongly identify with their peer groups (Gavin and Furman 1989) and typically report higher levels of group identification than adults (Liebkind 1982). Taken together, this line of research indicates that subjective norms from important groups (or friends’ norms) will play a significant role in TPB research involving adolescents. In a study of subjective norms in the TPB, Johnston and White (2003) found that group norms had a more powerful effect on subjective norms than did information from individuals. Further, Sparks and Shepherd (1992) found that social identity processes played an influential role in the behavioral intentions. These results suggest that adolescents’ drug use behavior may be directly and strongly influenced by important groups (or by subjective norms, using TPB terminology).

Current Study

The purpose of this study is conceptually to extend the findings of Conner et al. (2003) to a consideration of substance misuse in youth and to test the moderating effect of attitudinal ambivalence on subjective norms. The research thus makes use of an adolescent sample, and is focused on adolescent drug use (vs. the focus of Conner et al. (2003), the healthy eating behavior of adults). It extends the earlier study of Conner and associates by including prediction and analysis of the moderating role of potential ambivalence on subjective norms in a large, national sample of adolescents. This study makes use of data collected for the National Survey of Parents and Youth (David et al. 2010), an evaluation of the effects of the National Youth Anti-Drug Media Campaign (see www.mediacampaign.org).

Relatively few studies have used the TBP to predict adolescent marijuana use (Lac et al. 2009), but the research by Priester (2002) on the association of ambivalence and alcohol use is highly relevant. Priester found that participants who were ambivalent about alcohol consumption and safe sex practices had significantly less attitude–behavior congruence than participants of low ambivalence. Based on Priester’s findings and the research reviewed to this point, it is reasonable to expect that ambivalent attitudes will moderate the relationship between subjective norms and behavioral intentions, and subsequent behaviors.

If ambivalent attitudes cause evaluative tension (Priester 2002), then subjective norms may prove a viable means of reducing ambivalence, because in the absence of concrete information, people look to others for information on how to behave (Festinger 1954; Suls and Wheeler 2000). Following this line of reasoning, adolescents with ambivalent attitudes should feel enhanced evaluative tension, and this state should motivate them to reduce ambivalence by acting on the subjective norms of important groups in guiding ideas and behaviors. Friends are among the most important identity groups for adolescents (Tarrant 2002). Aligning their attitudes with the views of their identity group might allow adolescents to reduce ambivalence about important issues. Subjective norms also may influence behavioral intentions and future behaviors. These considerations lead to the expectation that attitude ambivalence will moderate the relation between subjective norms and behavioral intentions. Friend norms should be stronger predictors of intentions for adolescents holding highly ambivalent attitudes toward marijuana (H1). Furthermore, attitude ambivalence is expected to moderate the relationship between subjective norms and behavior. Friends’ norms will more strongly predict future use for adolescents whose attitudes toward marijuana use are highly ambivalent (H2).

Method

Data for this secondary analysis were collected and archived in the National Survey of Parents and Youth (NSPY), a 4-year panel survey conducted in concert with the National Youth Anti-Drug Media Campaign (David et al. 2010). The sampling methodology applied in the NSPY was comprehensive and designed to develop a nationally representative sample (see Crano et al. (2008a, b)). Nonsensitive data were collected via computer-assisted personal interviews. Sensitive data (drug-relevant perceptions and behaviors) were collected via audio computer-assisted self-administered interviews: respondents listened to items using headphones and responded via touch-sensitive screens. Respondents were interviewed four times at approximately yearly intervals from November 1999 to June 2004, and received $20 for each interview. Given the longitudinal nature of the data, the NSPY provides an opportunity to test the predictive validity of the TPB and the moderating role of attitudinal ambivalence over time.

The overall cross-sectional response rate for all youth (ages 9–18) at each round, defined as the product of (a) the percent of sampled households that were eligible, (b) the eligible households that completed the screening roster, (c) eligible households selected for follow-up, and (d) completion rate of youth in the round, was 64 % in round 1. Follow-up response rates for eligible participants were 86.3, 92.3, and 93 % in rounds 2, 3, and 4, respectively.

Respondents

Only respondents aged 12–16 years at round 1 were used in the analysis. Nine to 11 year olds were excluded because they answered different, abbreviated, surveys. To maximize sample size, data from rounds 1–2 were used in the present study; thus, older respondents (17–18 year olds) at round 1 also were excluded because they would have aged out of the study before completing round 2 of data collection.

The first analysis is a cross-sectional investigation of factors in round 1 that affect behavioral intentions. In the second analysis, behavior at round 2 is predicted from the TPB variables at round 1. Only respondents with complete data on the main dependent variables (intentions at R1 and marijuana use at R2) were included in the analyses. Additionally, only those respondents who had never used marijuana at round 1 were used in the analyses (165 sample respondents who reported use at R1 were excluded from the analyses, as the research is concerned with marijuana initiation and 189 respondents who were missing data on intentions at R1 and marijuana use at R2 were excluded from the analysis). After satisfying inclusion requirements, the final sample size was 1,604 adolescents.

Measures

Attitudes Toward Drug Use

A single item from round 1 was used to measure respondents’ attitudes toward marijuana use, “Your using marijuana, even once or twice, or the next 12 months, would be?” 1 (extremely bad) to 7 (extremely good).

Subjective Norms

Two items were used to measure subjective norms, both of which focused on the friends of the respondent. In round 1, respondents answered five-point Likert items (strongly approve–strongly disapprove) tapping how important friends would expect them to behave in regard to drug use, “How do you think your close friends would feel about you using marijuana even once or twice over the next 12 months.” The five-point response option ranged from (strongly approve to strongly disapprove). A second item stated, “When it comes to drug use, I want to do what my close friends want me to do.” A five-point (strongly agree– strongly disagree) response option was provided. The two questions measuring friend norms (approval of use and desire to follow friends’ opinions) were multiplied to create a friend norm variable, following procedures typically used to create the subjective norms variable in TPB research (e.g., Fishbein and Ajzen 1975).

Perceived Behavioral Control

Five five-point Likert scales were combined in round 1 to measure perceived behavioral control regarding marijuana use: “How sure are you that you could say no to marijuana if you really wanted to, if … you are at a party where most the people are using it? A close friend suggests you use it? You are home alone and feelingsad or bored? You are on school property and someone offers it? You are hanging out at a friend’s house and someone offers it?” Response options ranged from (not at all sure I can say no) to (completely sure I can say no). Internal consistency of this scale was acceptable (a=0.93).

Intentions

Respondents answered one question measuring their intentions to use marijuana at round 1: “How likely is it that you will use marijuana, at least once or twice, over the next 12 months” Response options ranged from 1 (I definitely will not) to 4 (I definitely will).

Behavior

Marijuana was measured at round 2 (the second survey was administered approximately 1 year after the first measurement round). Participants were asked “Have you ever, even once, smoked marijuana?” Those responding yes were asked, “How long has it been since you last smoked marijuana?” Respondents answering no received a score of 1; other answers were scored as follows: 2 (yes, more than 12 months ago), 3 (yes, more than 30 days but within the last 12 months), or 4 (yes, during the last 30 days).

Attitude Ambivalence

At round 1, respondents answered eight questions that were used to create an index of attitude ambivalence toward marijuana use: “How likely is it that the following would happen to you if you used marijuana, even once or twice, over the next 12 months, I would … (1) upset my parents?, (2) get in trouble with the law?, (3) lose control of myself?, (4) start using stronger drugs?, (5) be more relaxed?, (6) have a good time with my friends?, (7) feel better?, and (8) be like the coolest kids?” Responses could range from 1 (very unlikely) to 5 (very likely). The first four items were recoded so that higher scores indicated more positive beliefs about marijuana use.

For each respondent, a standard deviation was computed across all eight items. This measure constitutes the study’s operationalization of ambivalence. It is conceptually similar to the potential ambivalence measure of Kaplan (1972), which used a combination of scores on univalent positive and negative scales to indicate ambivalence, but it avoids the pitfalls of collapsing separate measures of positive and negative scores into a separate index, an issue raised earlier by Ullrich et al. (2008, p. 774), who argued that the combination of positive and negative evaluations “confounds the effects of the index and its components.” The present research, which adopts an alternative to ambivalence measurement, avoids the statistical problems identified by Ullrich et al. Our standard deviation approach provides a clear and useful conceptual analog to the more commonly used measures avoids the potential statistical hitches involved in the more typical operational definitions of potential ambivalence and provides a more certain measure of this construct.

Results

Analysis Plan

Two separate hierarchical multiple regression analyses were performed: the first was concerned with prediction of behavioral intentions, the second with the predictive validity of behavioral intentions for marijuana use. In the first step of both regression models, main effects of attitude, norms, PBC, and ambivalence were entered, followed at step 2 by the interactions of the main variables from TPB with ambivalence (see Table 1). Significant interactions were followed with simple slopes analyses (Aiken and West 1991).

Table 1.

Regression models for intentions at R1 and marijuana use at R2

	Intentions—R1		Marijuana use—R2
	β	t	β	t
Age	0.074	3.31*	0.112	4.62*
Ambivalence	0.038	1.70	0.054	2.26**
Attitude	0.268	11.72*	0.140	5.46*
PCB	−0.182	−8.02*	−0.008	−0.31
Intentions			0.152	5.60***
Ambivalence×attitude	−0.020	−0.79	0.002	0.07
Ambivalence×norms	0.151	2.29**	0.071	2.76***
Ambivalence×PCB	0.034	1.44	0.001	0.01
Ambivalence× intentions			−0.005	0.89

PCB perceived behavioral control

^*p<0.001,

^**p<0.05,

^***p<0.01

Background Variables

A number of demographic variables also were collected in the NSPY, including age of the respondents (M=13.53, SD=1.45), sex (there were 791 females and 813 males) and race/ethnicity (1,093 respondents were Caucasian; there were 220 African American, 66 Asian, and 225 Hispanic respondents). The average grade point average (on a four-point scale) was a 2.71 with a standard deviation of 0.91. The inclusion of these variables as covariates in the hierarchical regressions did not significantly change the results; however, age significantly predicted both behavioral intentions (β=0.074, t=3.31, p=0.001) and behavior (β=0.112, t=4.62, p<0.001), and so was included in the first step of the analyses.

Behavioral Intentions

Hierarchical regression of age, attitude, friend norms, PBC, ambivalence (step 1), and the interaction of ambivalence with these variables (step 2) on intentions to use marijuana was statistically significant at step 1, R²=0.23, F(5, 1,598)= 95.23, p<0.001. Higher perceived behavioral control to resist marijuana use was negatively related to intentions to use marijuana (β=−0.182, t=−8.02, p<0.001). More positive marijuana use attitudes were associated with stronger intentions to use the substance (β=0.268, t=11.72, p< 0.001), and the more that friends were thought to approve of use, the stronger were marijuana intentions (β=0.230, t= 9.95, p<0.001). These main effects were qualified by the hypothesized ambivalence by friend norms interaction at step 2 (R²=0.234, ΔR²=0.004, F(3, 1,595)=2.95, p= 0.032, β=0.151, t=2.29, p=0.022). No other interactions were statistically significant. Analyses of simple slopes (Fig. 1) revealed that the relationship between friend norms and behavioral intentions was stronger under high (β= 0.286, t=8.16, p<0.001) than low ambivalence (β=0.171, t=5.16, p<0.001), as predicted (H1).

Fig. 1.

Behavioral intentions as function of friend norms, moderated by ambivalence

Behavior—Marijuana Use at Round 2

Hierarchical regression of age, attitudes, friend norms, PBC, behavioral intentions, ambivalence (step 1), and the interaction of ambivalence with these variables (step 2) on marijuana use was statistically significant at step 1, R²=0.098, F(6, 1,597)=28.80, p<0.001. Intention to use marijuana at round 1 was significantly associated with marijuana use 1 year later, at round 2 (β=0.152, t=5.60, p<0.001). More positive marijuana attitudes among the originally abstinent respondents at round 1 were associated with later (round 2) marijuana use (β=0.073, t=2.82, p=0.005). Respondents’ friends’ approval of use at round 1 also was associated with round 2 marijuana use (β=0.140, t=5.46, p<0.001). Finally, greater attitude ambivalence at round 1 was significantly associated with marijuana use at round 2 (β=0.054, t= 2.26, p=0.024). These effects were qualified by significant interaction of friend norms and ambivalence at Step 2 (R²= 0.10, ΔR²=0.007, F(4, 1,593)=3.02, p=0.017), no other interaction was statistically significant. Analyses of simple slopes (Fig. 2) for the significant interaction between ambivalence and friend norms (β=0.071, t=2.76, p=0.006) showed that the relationship between friend norms and marijuana use was statistically significant under high (β= 0.234, t=5.97, p<0.001) but not low ambivalence (β= 0.056, t=1.53, p=0.13), predicted by hypothesis 2.

Fig. 2.

Marijuana use as a function of friend norms, moderated by ambivalence

Discussion

The purpose of this study was to test the moderation of attitudinal ambivalence in the TPB on adolescent marijuana use, replicating and extending the findings of Conner et al. (2003), using an ambivalence indicator that was not susceptible to the methodological weaknesses identified by Ullrich and Krueger (2010) in their earlier critique of potential ambivalence measures. The secondary data analysis performed on the NSPY involved two separate hierarchical multiple regression models: the first explored the moderating effects of ambivalence on intentions to use marijuana; the second tested the moderation on actual marijuana use 1 year later. Results across both regressions suggest that ambivalence at the very least provides partial moderation of the TPB for adolescent marijuana use.

For the cross-sectional analysis of behavioral intentions, adolescents’ attitudes, PBC, and friend norms directly predicted intentions to use, as the model suggests. For attitudes, the more favorable adolescents’ attitudes toward marijuana use, the more likely they were to intend to use the drug in the future. This result replicates a typical finding in TPB— attitudes should predict behavioral intentions. For PBC, the more adolescents felt that they could refuse marijuana the less likely they were to intend to use the drug in the future— again consistent with the expectations of the TPB. Friends’ (subjective) norms significantly predicted behavioral intentions: the more their friends approved of marijuana use, themore the respondents intended to use marijuana. The unique contribution of the present research is the finding that the direct relationship between subjective norms and intentions was moderated by ambivalence, as seen in the statistically significant norm-intention interaction. The interaction indicates that friend norms are more predictive of intentions under conditions of high than low attitudinal ambivalence. Ambivalence did not moderate the relationships between attitudes and behavioral intentions and PBC and intentions. These are important results as well, as they suggests that friends’ opinions regarding usage played a significant role in adolescents’ behavioral intentions to use marijuana only when adolescents held ambivalent attitudes about marijuana. Effects of friends’ opinions were significantly attenuated when the adolescent held unambivalent marijuana attitudes.

In the usage analysis of year2, behavioral intentions, friend norms, ambivalence, and attitudes all directly predicted marijuana use. For behavioral intentions, the more adolescents indicated that they intended to use marijuana, the more likely they were to do so at year2; a result that replicates the typical TPB finding (e.g., Armitage and Conner 2000; Rivis and Sheeran 2001). The analysis also uncovered a direct relation between attitudes and marijuana use: more favorable adolescent evaluations of marijuana in year1 were associated with greater use in year2. Unique to this analysis was a significant main effect for ambivalence, which indicated that greater ambivalence at year1 was associated with great use at year2. This association probably reflects the temporal dependence of adolescents’ attitudes toward marijuana. In early adolescence, most hold negative attitudes toward marijuana, but as time passes, the valence of the attitude becomes more positive (Johnston et al. 2010). This change in attitude is probably associated with, or anticipated by, a growing ambivalence. For friend norms, higher friends’ approval of use at year1 was associated with greater use at year2. The friends’ norms finding replicated results from the cross sectional behavioral intention regression analysis of the year1 data. This main effect must be interpreted with caution, owing to the statistically significant interaction of norms with ambivalence.

As shown, ambivalence moderated the direct relationship of friend norms with behavior. This interaction implies that friend norms were significantly stronger predictors of drug use when adolescents held ambivalent attitudes toward marijuana. This is a noteworthy result. It suggests that adolescents are most likely to be influenced by peers when they are ambivalent about the issue in question. When they are unsure of their beliefs (i.e., when they hold both positive and negative attitudes toward an object or behavior), adolescents will look to similar others to determine how to behave.

This finding has important practical implications for drug prevention. Adolescents who hold both positive and negative opinions about initiation are most likely to be influenced by close friends. Friends’ influence under high ambivalence affects both attitudes and actions. These results suggest that a potentially profitable prevention target is the perception of normative beliefs and attitudes. Considerable research suggests that adolescents strongly overestimate the extent to which their friends engage in illicit substance use (e.g., Keyes et al. 2011; Crano et al. 2008; Zhao et al. 2006). Correcting misperceptions of this sort may have two important benefits: first, corrections may reduce general ambivalence regarding drug use, thereby reducing peer influence; second, successfully correcting misperceptions may lead the adolescent to the conclusion that the assumption that one’s friends are all positively predisposed to marijuana use is incorrect. Blanton and colleagues’ deviance regulation model builds on this logic, and has produced provocative results suggesting the potential of norm-based approaches that capitalize on variations in normative perceptions to attenuate substance misuse (Blanton and Burkley 2008; Blanton and Christie 2003).

An alternative to the standard approach, which takes advantage of adolescents’ relative ambivalence toward marijuana, also is suggested in the study’s analyses. Although adolescents’ attitudes toward marijuana have received considerable attention in the prevention literature, much less research has focused on how the stability or ambivalence of these attitudes may relate to behavior. Capitalizing on the results of the present analyses, we suggest two potentially profitable possibilities. The first is that prevention models be designed make greater use of information based on hard-won scientific knowledge to facilitate adolescents’ adoption of proper attitudes and consequent behaviors. This approach simultaneously reduces ambivalence and provides a strong knowledge base for antidrug attitudes and behaviors. Research demonstrating the harms of marijuana misuse for adolescents is copious and one-sided, and should be a major emphasis in prevention. Presenting information that is truthful, credible, not exaggerated, not readily falsified, and persuasive thus would seem an easy chore, but this assessment is not borne out in practice (Crano 2010; Erceg-Hurn 2008; Fishbein et al. 2002; Hornik et al. 2008; Skenderian et al. 2008). Much greater attention to fundamental principles of persuasion must be paid if prevention efforts are to succeed (Crano et al. 2013).

A second prevention-relevant suggestion arising from the current findings is that the information provided in prevention messages should be designed specifically to attenuate ambivalence, thereby rendering adolescents less susceptible to their peers’ mistaken ideas about the positive outcomes accruing to marijuana use. Univalent negative attitudes toward use of the drug would likely trump friends’ positive normative affirmations, as may be inferred from earlier research (e.g., Crano et al. 2008a, b).

An apparent limitation of this study is its reliance on a secondary data source in testing ambivalence effects. Insecondary analyses, one almost never controls the questions posited, and this was true in the present context. As well, relying on longitudinal data collected by another source provides no control over measures (leading us to use single-item measures of attitudes, intentions, and marijuana use), or missing data and participant drop out. On the positive side, the dropout rate was not severe, and the use of longer, more reliable measures likely would have strengthened results. In addition, the limitations imposed by the secondary nature of the data forced the creation of a reliable and, we believe, valid alternative to the measurement model for potential ambivalence. The new measure avoided the statistical issues raised by Ullrich et al. (2008), and produced results consistent with expectations based on earlier results that used the approach that was called into question. This conceptual replication of earlier results, which suggests the validity of earlier findings, would not have occurred if a more standard approach to measuring potential ambivalence was adopted, so the limitation imposed by the secondary analytic aspect of the research ultimately may have proved an advantage. Another limitation concerns effect sizes, which in our study were small. Due to the design of our study (non-experimental secondary longitudinal data analysis), we would not expect large effect sizes and the fact that we found significant effects suggest that ambivalence is important to adolescent marijuana use. In addition, if these small effects were extrapolated to the population of US adolescents, it becomes clear that interventions influencing ambivalence might affect thousands of potential marijuana initiators. Even so, future experimental studies designed specifically to examine ambivalence in adolescent marijuana use are clearly indicated.

The results provide insight into future research directions that may result in more positive preventive outcomes than we have come to expect. Research directed specifically to reducing adolescents’ attitudinal ambivalence about marijuana use has great potential to affect initiation and we are hopeful that future research will examine this possibility in developing antimarijuana prevention models.