Impulsivity Moderates the Effects of Movie Alcohol Portrayals on Adolescents’ Willingness to Drink

Abstract

This study examined impulsivity as a moderator of adolescents’ reactions to positive vs. negative portrayals of drinking in American movie clips. Impulsivity, along with willingness and intentions to drink in the future, were assessed in a pretest session. In the experimental sessions, adolescents viewed a series of clips that showed drinking associated with either positive outcomes (e.g., social facilitation) or negative outcomes (fights, arguments). A third group viewed clips with similar positive or negative outcomes, but no alcohol consumption. All participants then responded to an implicit measure of attentional bias regarding alcohol (a dot probe), followed by explicit alcohol measures (self-reports of willingness and intentions to drink). Hypotheses, based on dual-processing theories, were: a) high-impulsive adolescents would respond more favorably than low-impulsive adolescents to the positive clips, but not the negative clips; and b) this difference in reactions to the positive clips would be larger on the willingness than the intention measures. Results supported the hypotheses: Adolescents high in impulsivity reported the highest willingness to drink in the positive-clip condition, but were slightly less willing than others in the negative-clip condition. In addition, results on the dot probe task indicated that reaction times to alcohol words were negatively correlated with changes in alcohol willingness, but not intention; i.e., the faster their response to the alcohol words, the more their willingness increased. The results highlight the utility of a dual-processing perspective on media influence.

Adolescents learn a lot about their social worlds from Hollywood, including which kinds of social behaviors are expected / encouraged and which ones are not. One social behavior frequently portrayed in movies is drinking. Surveys have indicated that > 80% of American movies include some drinking (Dal Cin, Worth, Dalton, & Sargent, 2008), and the amount of drinking in movies appears to be increasing (Bergamini, Demidenko, & Sargent, 2013). Even PG13-rated movies contain significant amounts of alcohol use (in fact, about as much as R-rated movies), and that includes depictions of heavy use: e.g., > 15% of G/PG movies have some portrayal of intoxication (Dal Cin et al., 2008).

The impact of this learning experience has been demonstrated in a number of longitudinal survey studies showing positive prospective links between amount of movie alcohol exposure (MAE) and (early) initiation and escalation of use among adolescents (Sargent, Wills, Stoolmiller, Gibson & Gibbons, 2006; Wills, Sargent, Gibbons, Gerrard & Stoolmiller, 2009), both of which are predictors of alcohol problems later in life (NSDUH, 2013). Until fairly recently, however, there had been few experimental studies designed to examine this important MAE → use relation. Even though the methodology in the longitudinal work has been quite sophisticated (e.g., the “Beach” method used by Sargent et al., 2008), experimental studies in controlled settings are needed to increase confidence in the causal inferences about the MAE / use relation proposed in many of these longitudinal studies.

Experimental Studies

Valence

The vast majority of alcohol portrayals in movies are positive (Bahk, 2001; Smith et al., 2013), which is one reason why earlier lab studies focused on the issue of portrayal valence. Although the results have been somewhat mixed (e.g., de Graaf, 2013; Kulick & Rosenberg, 2001), a majority of studies have suggested that positive portrayals--in which the drinking facilitates socializing and/or sex, for example--tend to have more impact in terms of alcohol use. Moreover, these results are generally consistent with those from studies showing that positive more than negative affect (created by pleasant vs. unpleasant visual stimuli) is associated with drinking and increases alcohol craving in problem drinkers (Mason, Light, Escher & Drobes, 2008). Even violent media has more impact (on aggressive thoughts and behavior) when there is a positive outcome of some kind—i.e., the violence is rewarded as opposed to punished (Carnagey & Anderson, 2005).

Imitation

A series of experimental studies on MAE effects has been conducted in the Netherlands in the context of a “bar lab” (see Engels et al., 2009). This kind of setting allows for unobtrusive manipulation of relevant variables (including showing movies or TV shows) in a controlled fashion that can imitate in situ experiences. These studies have shown that watching movies or TV alcohol advertisements is associated with increased drinking in the same setting (Koordeman, Anschutz & Engels, 2015; Koordeman, Kuntsche, Anschutz, van Baaren, & Engels, 2011). This research illustrates one proposed route of media influence, consistent with social cognitive theory (Bandura, 2001), which is modeling. Although informative, however, these studies did not examine what is potentially a more lasting route to behavior, especially for those under legal drinking age, which is through attitude change.

Cognitive mediation

Several studies have examined the impact of movie portrayals of alcohol use on changes in adolescents’ relevant cognitions, including alcohol expectancies (beliefs about the effects that alcohol consumption will have) and intentions to drink (Koordeman, Anschutz & Engels, 2014). Positive portrayals have been shown to lead to more favorable alcohol expectancies; and favorable expectancies are associated with greater expectations of using.¹ Relatively few studies have assessed behavioral intentions to drink as a reaction to MAE; and those that have used intention measures have shown inconsistent effects. Kulick and Rosenberg (2001), for example, reported that positive movie portrayals were associated with more favorable alcohol expectancies, but there was no effect on intentions. Slater et al. (1997) found positive effects on drinking intentions after exposure to TV alcohol ads for White but not Hispanic students. Setodji, Martino, Scharf and Shadel (2013) found that media smoking portrayals increased intentions to smoke, but only when participants were with friends, not when they were alone. Finally, in van Hoof, de Jong, Fennis and Gosselt (2009), viewing alcohol portrayals in soap operas actually led to lower intentions to drink. The lack of consistent intention effects may have to do with the nature of the construct itself (what it assesses) and the type of information processing it involves. It also raises questions as to whether media alcohol portrayals actually affect adolescents’ intentions to drink. To address this issue, we relied on the literature in dual processing.

A Dual-Processing Perspective on Media Influence

Dual-processing theories maintain that decision making involves two different modes of information processing, referred to variously as Systems 1 and 2, reflective / impulsive, and analytic / heuristic (Evans, 1984; see Sherman, Gawronski, & Trope, 2014, for a more complete discussion of dual processing). The analytic system processes information in a more reasoned (though not necessarily rational) manner. It is central to expectancy-value theories, such as the theories of Reasoned Action (Fishbein & Ajzen, 1975) and Planned Behavior (Ajzen, 1991), which maintain that behavior is the result of a planful and reasoned process that leads to a decision to engage or not engage in a behavior—an intention. As the name implies, processing in the heuristic system is based more on heuristics (cognitive “short cuts” that facilitate information processing), which means it is less deliberative, often more impulsive, and usually quicker (truncated) (Stanovich, West, & Toplak, 2014). Heuristics include images, especially images of individuals who perform a behavior (e.g., actors or actresses). For these reasons, a number of researchers, primarily those in media and communication, have suggested that media effects often involve heuristic processing more than analytic processing (Gerbner, Gross, Morgan & Signorelli, 1986; Shrum, Burroughs, & Rindfleisch, 2004).

The Prototype-Willingness Model

The current study draws on the literature in dual processing, and one dual-processing theory in particular, the Prototype-Willingness model (PWM; Gerrard, Gibbons, Houlihan, Stock & Pomery, 2008; Gibbons, Gerrard & Lane, 2003; Gibbons, Stock, Gerrard & Finneran, 2015). The model focuses on psychosocial factors related to health decision making and health behavior, especially among adolescents. Consistent with its dual-processing perspective, the PWM maintains there are two decision paths to behavior. The reasoned path involves more analytic processing (reasoning), including more consideration of consequences. This deliberation results in a decision to engage or not engage in a behavior, i.e., a behavioral intention (BI). The social reaction path involves more heuristic processing that leads to a proximal antecedent to behavior that is related to, but distinct from, BI and that is behavioral willingness (BW). BW is defined as an openness to risk opportunity. Unlike a plan or a goal state—the definition of BI (Ajzen, 1991; 1996)--BW is more reactive than proactive. It is also more affected by social factors than BI and tends to be more impulsive. A central heuristic in the PWM is the image or prototype associated with the behavior (a “typical” drinker, for example); the more favorable the image, the greater the willingness to engage.

BW versus BI

Generally speaking, BW is a better predictor of adolescent risk behavior than is BI (Abedini et al., 2014; Andrews, Hampson & Peterson, 2011; Pomery, Gibbons Reis-Bergan & Gerrard, 2009; Rivis, Sheeran & Armitage, 2010). Few adolescents intend to have risky sexual encounters, for example, just as few adults intend to drink a lot and then drive, but, the percent who are willing to engage in these behaviors, and eventually do, is higher than the percent intending to do them. In fact, several studies have shown that many adolescents are willing to engage in risky behaviors that they have little or no intention or expectations of engaging in (Gibbons, Gerrard, Reimer & Pomery, 2006). A primary reason for this is that, consistent with its heuristic nature, BW is associated with less consideration of consequences, especially negative consequences; Gerrard, Gibbons, Houlihan, Stock & Pomery, 2008; Stock, Litt, Arlt, Peterson & Sommerville, 2013). Moreover, a number of studies have shown that BW is also more influenced than BI by social factors, including social influence (Gibbons et al., 2004), social comparison (Stock, Gibbons, Beekman & Gerrard, 2015), and social images or prototypes (see Todd, Kothe, Mullan, & Monds, in press; and van Lettow et al., in press, for meta-analytic reviews of the PWM). Roberts, Gibbons, Gerrard and Kingsbury (2014), for example, found that subliminal presentations of bikinied women in provocative poses to single men led to an elevation in their willingness to engage in risky sex, but had no effect on their intentions.

The PWM and media influence

Applications of the prototype model to the study of media influences on behavior have been based on an assumption that media influence works in a similar fashion to social influence. In this perspective, actors are seen as “superpeers,” who can exert significant (“peer”) influence on adolescents (Brown, Halpern & L’Engle, 2005). In other words, media effects should primarily follow the social reaction path to action, with actors portraying prototypes (usually positive) associated with various risky behaviors, like smoking, sex, and drinking. These PWM applications have shown that the prospective relation between MAE and (changes in) alcohol use is mediated by prototypes and willingness: More MAE → more favorable images of drinkers and more BW, which then → increases in alcohol consumption (Dal Cin, Worth, Gerrard, Gibbons & Sargent, 2009; Gibbons et al., 2010). Consistent with the media as social influence perspective and the prototype model, in the current study, it was assumed that positive MAE portrayals would have more of an effect on BW to drink than on BI to drink.

Impulsivity

Self-control, which is the ability to regulate one’s actions, thoughts, and emotions, has two dimensions: good self-control (also termed planfulness) and impulsivity (sometimes referred to as poor self-control). Adolescents who are high in impulsivity tend to be impulsive, distractible, and have difficulty with delay of gratification. As might be expected, among adolescents, impulsivity is positively related to substance use and abuse, and other risky behaviors (Lejuez et al., 2010; Wills & Stoolmiller, 2002). Impulsive adolescents are also more susceptible to social influence, especially with regard to substance use (Gibbons et al., 2006). That being the case, one might expect that BW would be higher among individuals who are high in impulsivity, and it is (e.g., Gerrard et al., 2006). Wills et al. (2010), for example, found that adolescents who were high in impulsivity were generally more willing to drink; they were also more responsive to alcohol ads (again, in terms of willingness to drink), and more responsive to peer influence (on drinking). Similarly, van Hemel-Ruiter, de Jong & Wiers (2011) reported that adolescents who drank alcohol responded more favorably to alcohol photos if they were low in “inhibition capacity” than high on this dimension (inhibition capacity is another term for self-control; thus, low “capacity” means high impulsivity). Consistent with a dual-processing perspective, van Hemel-Ruiter et al. suggested that alcohol expectancies (which involve more analytic processing--because they are cognitions about specific future outcomes) are more predictive of alcohol use for high self-control adolescents, whereas appetitive valence (including visual appeal and affect; i.e., heuristic cues) is a better predictor for those who are high in impulsivity. In short, reactions to media portrayals may vary as a function of impulsivity.

Explicit vs Implicit Measures of Alcohol Attitudes

Most studies of MAE have included explicit measures of alcohol attitudes (e.g., direct questions about alcohol expectancies). The current study also included an implicit (indirect) measure intended to assess attentional bias. The dot probe task (MacLeod, Mathews, & Tata, 1986) involves a brief presentation of two stimuli (e.g., words) side by side, followed by the appearance of a dot behind the spot where one of the stimuli was presented. Participants report as quickly as possible where the dot is; thus, reaction time is an indication of which word the person was focusing on. Drinkers, for example, respond more quickly when the dot is behind an alcohol-related stimulus (Townshend & Duka, 2001), smokers when it is behind a smoking stimulus (Ehrman et al., 2002). These types of implicit measures involve rapid responding that is likely to reflect heuristic processing. In contrast, explicit measures are more likely to involve analytic processing (Stanovich, West & Toplak, 2014; Wiers et al., 2002).

Summary

Experimental studies of MAE effects on drinking have suggested that positive portrayals are associated with more positive alcohol expectancies, especially for inexperienced drinkers; but, relatively few experimental studies have examined cognitive factors other than expectancies. In fact, several recent reviews of the MAE literature have reached some consensus on needed areas of focus for future MAE research; those are: individual differences (e.g., self-control) as moderators of reactions to movie portrayals (Koordeman, Anschutz & Engels, 2012; Meier, 2010), clarification of the effects of positive vs. negative alcohol portrayals (Koordeman et al., 2015), experimental studies that control for baseline measures of drinking, attitudes, etc. (Wills et al., 2010); and examination of MAE effects on cognitions, especially among adolescents (Meier, 2010). These issues were examined in an experimental study with adolescents.

Overview

In pretest sessions (T1), adolescents completed measures of impulsivity, previous drinking, and alcohol BW and BI. In the experimental session (T2), all participants first viewed three control clips with no or minimal alcohol. They then viewed a series of movie clips that depicted alcohol consumption leading to either positive or negative consequences, or had no drinking. After viewing the clips, they were presented with an implicit measure, the dot probe task, and then explicit measures of alcohol prototypes as well as BW and BI. More favorable responses to the positive portrayals were anticipated on the willingness measure at T2 (controlling for T1), more so among adolescents who were high in impulsivity. These differences were not expected on the BI measures. Also, because BW is associated with avoidance of consideration of negative consequences, it was not expected to be elevated among the impulsive adolescents in the condition in which negative consequences were featured. There was no clear basis for a prediction in the no-alcohol condition. We also examined three sets of possible mediators of the MAE effects on BW: a) alcohol images (prototypes), assuming MAE → more favorable images → more BW; b) reaction times to the alcohol words in the dot probe: i.e., heuristic (quicker) response would be related to more BW; and c) reactions to the clips: i.e., the more enjoyable or exciting they were thought to be, the greater the increase in BW.

Method

Participants

The sample was composed of 143 students (77 females) ages 14 to 16 (M = 14.4; SD = .53) from two high schools in or near a university town (there were no differences in results due to site). Most were 9^th-graders (the grade that was recruited), but a few 10^th graders also participated. Overall, the group tended toward higher SES, typical of a college town; e.g., 73% indicated at least one parent had a college degree, 49% indicated both of their parents did. The sample was 89% White, 2% Black, 4% Hispanic, and 5% Asian.

Measures

(Pretest and Experimental session administrations are indicated by T1 and T2)

Impulsivity. (T1)

The impulsivity measure (Wills, Gibbons, Gerrard & Brody, 2000) was derived from the Self-control Rating Scale (Kendall & Wilcox, 1979). It has two subscales, each with five items: impulsivity (e.g., “I get carried away by new and exciting ideas, but I don’t think of the possible problems”) and distractibility (e.g., “I like to switch from one thing to another”). The subscales correlated at r = .54, so they were combined to form an impulsivity scale (10-item α = .81).

Drinking experience. (T1)

This item asked about previous use of alcohol (defined as beer, wine, or liquor), followed by five options from I have never tried alcohol in my life to I usually drink more than a few times per week.

BW. (T1, T2)

At T1, participants were asked: “Suppose that sometime in the next year you were with a group of friends some place and there was some alcohol you could have if you wanted. Under these circumstances, how willing would you be to do each of the following: Have a sip, Drink a whole drink? Drink more than one drink?” These were followed by 4-point scales from Not at all willing to Very willing. The three items were aggregated. In order to assess current BW (and BI), the term “next year” was dropped from the T2 items; αs: T1 = .87, T2 = .90. These three BW items have been used in previous alcohol studies and shown to have good reliability and predictive validity vis a vis alcohol consumption (Dal Cin et al., 2009; Gerrard et al., 2006; Gibbons et al., 2015).

BI. (T1, T2)

At T1, the stem for the BI items was the same as that for the BW items, followed by: “…how much (under these circumstances) would you intend (plan) to drink …?” (same 3 items; scale = Not at all to A lot; α = .76). At T2, the drinking scenario was described, followed by two items: How much would you intend to drink (4 points from none to more than one drink) and then How likely is it that sometime in the next year you will drink alcohol at a party? (not at all to very). Likelihood measures, like this, are often used to assess BI (Armitage & Conner, 2001).

Manipulation check (T2)

After each clip, participants were asked about the overall tone of the clip from negative to positive (7-point scale).

Mediators (T2)

After each clip, participants were asked how exciting it was, how much they enjoyed it, how much drinking there was, and how likeable was the main character (all 7-point scales); and, they were asked three questions assessing their recall of what happened in each clip (e.g., “What was the name of the blackjack dealer?” “What did the characters toast?”). In addition, in order to determine whether changes in prototypes of same-age drinkers mediated the movie effects, prototypes were assessed at T1 and again at T2 (after all of the clips had been viewed) with the usual stem (Gibbons et al., 2015): “Think about the type of person your age who drinks alcohol…” This was followed by five adjectives (e.g., smart, cool), each with a 4-point scale from not at all to very. These five adjectives were aggregated (T1 and T2 αs = .71 and .53). Finally, at the end of the session, they were asked if they had seen each one of the clips before.

Dot probe. (T2)

A total of 65 risk, reward, alcohol, and neutral words were presented, each paired with another word. For risk and reward (12 for each), the accompanying word was an antonym (e.g., for risk: “penalty” / “bonus;” “hurt” / “help;” for reward: “succeed” / “fail” and “lively” / “dull”). Alcohol words (N = 11) were paired with neutral words (“booze” / “fence”); neutral words (30) were paired with other neutral words (“salad” / “print”).

Procedure

Recruitment

9^th-graders at local high schools were contacted by their teachers about participating in a study that concerned media and health behavior. They were then given informed consent letters to take home to their parents. If a parent signed the letter, then assent letters were given to the students; both signatures were required for participation. The sample reflected students’ class schedules (pretest and experimental sessions occurred mostly during study periods), and whether they remembered to bring the consent form home and then back to school. There were no other criteria for participation, besides year in school, availability, and consent/assent. Students received a $10 Amazon gift card for participating. The study was approved by the University IRB. It was conducted in 2013 / 2014.

T1 and T2

The pretest sessions occurred two to four weeks prior to the lab sessions. The latter were run in groups of 3 to 5. At that time, students were told: a) the study concerned reactions to different kinds of movies; b) they would see a series of 7 clips, each followed by a series of questions; and c) at the end, they would answer a few questions about their reactions to what they had seen and about their health behavior. Students had their own laptops with headphones so as not to disturb each other. They were randomly assigned to one of the three movie conditions.

The film clips

The clips ranged from 115 to 170 sec in length. All participants first saw three control clips, one each from: Spiderman, Soul Food, and Under the Tuscan Sun. The first two had no drinking, the third had a brief amount of drinking (positive in tone). They then viewed four more clips. In the experimental conditions, those four clips contained significant amounts of drinking that led to either positive consequences (enjoyment, social facilitation) or negative consequences (fights, social embarrassment, etc.). The Negative condition included clips from: 500 Days of Summer, 21, The Wedding Singer, and Why Did I Get Married? The Positive condition clips came from: What Happens in Vegas, Glory Road, He's Just Not That into You, and 500 days of Summer (a different scene than in the negative clip). The third group saw the three control clips, and four other clips that had no alcohol, but depicted positive (e.g., flirting) and negative (fighting) outcomes similar to those in the alcohol clips. The two negative clips came from: Why Did I Get Married? and 21; the two positive clips were from 500 days of summer and The Wedding Singer. The order of clip presentation was the same within each condition. On average, each movie had been seen by 19% of the sample; range = 5% for Glory Road to 65% for Spiderman.

The manipulation check indicated that participants rated the positive clips as having a more positive overall tone than the no-alcohol clips, and the no-alcohol clips were seen as more positive than the negative clips (all ps < .01).

Results

Means and Correlations

Means and correlations for the primary measures are presented in Table 1. BW and BI were correlated with each other; T1: r = .51; T2: r = .61 (both ps < .001). They were also both correlated with previous drinking and prototype at T1, with the BW correlations being somewhat stronger than the BI correlations: rs for previous drinking = .56 (BW) and .33 (BI; both ps < .001), for prototype, rs = .31 (BW) and .20 (BI; ps < .001 and .02). Finally, BW was correlated with impulsivity (r = .29, p < .001), but BI was not (r = .13, NS). As expected, the difference between these latter two correlations was significant: t = 2.00, p < .05.

Table 1.

Means and Correlations of Primary Variables

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
1. Gender	--
2. Imp	.10	--
3. Past drinking	.08	.24	--
4. T1 BWa	.11	.29	.53	--
5. T1 BIa	−.03	.13	.31	.51	--
6. T2 BWa	.11	.39	.36	.77	.35	--
7. T2 BIa	.13	.36	.15	.48	.18	.62	--
8. Dot probeb	.23	.08	.09	.04	.04	−.08	−.04	--
Means:	1.55	2.38	1.34	−.05	−.01	−.04	−.03	6.19b
Scale:	-	1 – 4	1 – 5	1 – 4a	1 – 4a	1 – 4a	1 – 4a

Notes: Gender coded: male = 0, female = 1; Imp = Impulsivity (continuous); BW = behavioral willingness to drink; BI = behavioral intention to drink. Dot Probe = reaction time to alcohol words in the dot probe task.

^aStandardized;

^bScale = 300 – 1000, log transformed. High scores reflect slow reaction times.

For all correlations: |r| > .16, p < .05; |r| > .23, p < .01

T1 (Pretest) Measures

Sixty percent of the participants reported that they watched at least 3 – 5 movies per month; 24% said 6 or more. The percent of students who reported having tried the following substances at least a few times in their lifetime was: alcohol = 38%; marijuana = 10%, cigarettes = 7%; only 2% reported having had 3 or more drinks at least once in the last 3 months. This amount of alcohol consumption is typical of this age (Monitoring the Future;Johnston et al., 2014). ANOVAs on the pretest BW and BI items revealed no effect of Condition, nor an Impulsivity × Condition interaction for either measure (all ps > .17), indicating there were no random assignment problems.

T2 (Posttest) Measures

BW

ANCOVAs were first conducted on the 3-item T2 BW index. These analyses included the T1 BW index and previous drinking as covariates; and Movie Condition, Impulsivity, which was split at 60% low / 40% high,² and then the Impulsivity × Condition interaction term as predictors. However, examination of the index indicated that few participants (18%) reported any willingness to have > one drink (the 3^rd item in the index), which is typical for this age (Gibbons et al., 2010; Litt, Stock & Gibbons, 2015), and there were no significant effects on this item.³ Therefore, the item was dropped and then the ANCOVA was redone with just the two-item index (“sip” and “one drink;” αs: T1 = .86, T2 = .89).These initial analyses indicated that T2 BW responses in the Negative condition were different from those in both the Positive and No-alcohol conditions, whereas the latter two were similar to each other. Consequently, the latter two movie conditions were combined for analyses.

Results of this analysis are presented in Table 2. As expected, there were main effects of the T1 BW covariate (p < .0001), Condition (p < .05), and a marginal positive effect of previous drinking (p = .08). In spite of the high stability in BW (T1/T2 autocorrelation: r = .77), the anticipated Impulsivity × Condition interaction was significant: F (1, 135) = 15.71, p < .001.⁴ As can be seen in the Table, the High Impulsive, Positive/No-alcohol cell had the largest T2 adjusted mean (2.03). That mean was higher than: a) the two Low-impulsive cell means (both ts [135] > 2.45, ps < .02); b) the High-impulsive, Negative cell (1.46; t = 4.30, p < .001); and c) the mean of the other three cells combined (M₍₃₎ = 1.55; t[135] = 4.80, p < .001); those three cell means did not differ from each other (all ts < 1.61, ps > .10). In other words, relative to those low in Impulsivity, High Impulsive adolescents had higher adjusted BW in the Positive / No-alcohol movie condition, but slightly lower adjusted BW in the Negative movie condition.

Table 2.

T2 Means for BW and BI Adjusted for T1 Means

Movie Cond:	Negative		Positive/No Alcohol
Imp Level:	Low Imp	High Imp	Low Imp	High Imp
(N)	(27)	(20)	(60)	(34)
BW:	1.70a	1.46a	1.52a	2.03b
(SE)	(.10)	(.12)	(.07)	(.09)
BI:	1.55a	1.91a	1.65a	1.77a
(SE)	(.12)	(.13)	(.08)	(.10)

Notes: Movie Cond = movie condition; Imp = Impulsivity (split at 60% / 40%); BW = behavioral willingness to drink; BI = behavioral intention to drink. Means are the average of the two items for T2 BW and BI, adjusted for the average of the T1 BW and BI. Means without common superscripts differ at p < .01; all other means are not significantly different (p ≤ .10).

BI

Means on the two BI items were similar to one another, and they were also combined into an index and standardized. As with the BW measure, the main effect of the T1 covariate was significant (p < .0001). However, in contrast to BW, the main effect of Impulsivity on BI was marginal (p = .06), and neither the Condition main effect nor the Condition × Impulsivity interaction was significant (ps > .28; see Table 2). The pattern was similar (nonsignificant) across both BI items, when analyzed separately. Thus, MAE had no apparent impact on adolescents’ intentions to drink, regardless of their level of Impulsivity.

BW vs. BI

ANOVA

To further examine change in BW versus BI over time (i.e., in the same analysis), a 2 × 2 × 2 × 2 repeated measures ANOVA was conducted that included BW (2-item) and BI from T1 and T2 as dependent measures, Condition and Impulsivity as between-participant factors, and Time (T1 vs. T2) and Type of Measure (BW vs. BI) as within-participant factors (see Table 3). Given the wording differences between BW and BI, they were both standardized before analysis. Results revealed a main effect of Impulsivity: F = 8.01, p < .005, as high Impulsive participants had higher BW / BI than did those low in Impulsivity. Most important, the expected 4-way interaction (F = 4.88, p < .03) reflected results from the individual ANCOVAs: BI remained fairly consistent--there were no significant changes from pretest to posttest for any cell (all ps > .44); in contrast, BW increased significantly, but only for the High Impulsive participants in the Positive/No-alcohol conditions (M = .37, t = 2.63, p < .01); no other group reported a significant change in BW (all ts < 1.71, ps > .08).⁵

Table 3.

Standardized Means and SDs for T1 and T2 BW and BI

Movie Cond:	Negative				Positive/No Alcohol
Imp Level:	Low Imp		High Imp		Low Imp		High Imp
(N)	(27)		(20)		(60)		(34)
	Mean	(SD)	Mean	(SD)	Mean	(SD)	Mean	(SD)
BW (T1):	−.21	(.94)	.43	(1.16)	−.17	(.87)	−.01	(1.05)
BW (T2):	−.10	(.86)	.11	(1.11)	−.29	(.72)	.36	(1.19)
Δ:	.11		−.32		−.12		.37
t:	.68		1.71		1.19		2.63a
BI (T1):	−.28	(.95)	.19	(1.00)	−.02	(1.00)	.09	(1.08)
BI (T2):	−.31	(.77)	.33	(1.06)	−.11	(1.00)	.13	(.91)
Δ:	−.03		.14		−.09		.04
t:	.19		.75		.83		.28

Notes: Movie Cond = movie condition; Imp = Impulsivity (split at 60% / 40%); BW = behavioral willingness to drink; BI = behavioral intention to drink. Δ: change from T1 to T2; t is for change from T1 to T2.

^ap < .01

Mediation

To examine mediation, the BW ANCOVAs were repeated with each one of the potential mediators included (separately and then in combination) as a covariate (cf. Lee, Chassin, & MacKinnon, 2010). As mentioned earlier, the prototypes were related to BW as anticipated, but they were not differentially affected by the clips (none of the main effects or interactions was significant, ps > .70), and so they did not mediate the MAE effects. Among the clip evaluation items, the only one that was significantly related to T2 (Δ) BW was exciting —the more exciting the adolescents thought the clip was, the more they increased their BW. However, that did not change the primary interaction (Impulsivity × Condition), which actually got slightly stronger when this covariate was included. Thus, the exciting rating, as well as the other clip evaluation measures, including the evaluation of tone (manipulation check) —by themselves or aggregated--did not mediate the effect of movie condition on change in BW.

Dot Probe

Reaction time scores on the dot probe were skewed on either end of the distribution (Kurtosis = 12.0), which is typical of this measure (Eberhardt, Goff, Purdie, & Davies, 2004), so they were trimmed at a minimum of 300 ms and a maximum of 1000 ms, and then log-transformed (Ratcliff, 1993). Our hypothesis was that because of its implicit nature, responses on the dot probe (reaction time to alcohol-related words) would be more reflective of change in BW than BI. As expected, the correlation between reaction time and T2 BW, with T1 BW partialled, was significant: r = −.19, p < .03; thus, the quicker the response time (i.e., the more participants were focusing on alcohol words), the more their willingness to drink increased. The same partial correlation for BI was nonsignificant: r = −.04 (p = .68). The second analysis examined reaction time to the alcohol words as a mediator of the MAE → Δ BW relation; it included the same BW ANCOVA as before, this time with reaction time as a covariate. The ANCOVA revealed a significant effect of reaction time (consistent with the partial correlation), but there was no evidence that reaction time mediated the movie effect, and the primary interaction remained significant. Thus, participants’ attentional bias toward alcohol was related to change in their BW but not their BI, and that was true regardless of which clips they had seen or their level of impulsivity. Unexpectedly, the relation between reaction to the risk words and Δ BW was also significant: r = −.22, p = .01 (that was not the case for Δ BI, r = −.09, p = .31), suggesting that those who were thinking about risk-taking right after the clips were also more likely to increase their alcohol BW.

Discussion

MAE had the anticipated impact on the adolescents who were high in impulsivity. After viewing clips that clearly showed negative consequences associated with drinking, their willingness to drink remained relatively unaffected (in fact, it actually declined slightly). In contrast, when they viewed clips with positive alcohol outcomes, they reported more BW to drink than did their low impulsive counterparts. The clips were chosen because the consequences portrayed were typical of those seen in many movies: social facilitation (a good time) in the positive condition and social disruption (arguments, fights, break-ups) in the negative. Thus, the results are consistent with previous studies showing more favorable reactions to positive alcohol portrayals (Smith et al., 2013). In this case, however, the favorable reactions occurred only among the group shown to be most responsive to media effects—i.e., those high in impulsivity. In fact, it occurred only among those in the top 40% of the impulsivity distribution, which probably reflects the fact that this sample was young, and so a relatively high level of impulsivity may be necessary for the effect (increase in drinking BW) to occur.

Dual Processing

BW vs. BI

As in previous studies, BW and BI were correlated, and they were affected differently by the experimental manipulation. Intentions to drink did not appear to be systematically influenced by the clips, whereas willingness to drink was increased for some of the adolescents, as predicted. In the long run, it is certainly possible that observation of attractive individuals engaging in a particular behavior in an enjoyable social setting will increase the likelihood of adolescents’ plans or intentions to perform that behavior over time. But development of such intentions is more likely after multiple MAE exposures and more time to formulate the plans. Brief presentations (like clips) trigger heuristic processing (“drinking socially is fun,” “cool people drink”) that would be reflected in changes in BW more than BI. Moreover, these changes were on the lower levels of risk—i.e., the “have a sip,” and, to a lesser extent, the whole drink items, but not the heavier drinking item (“more than one drink”)-- as one might expect at this young age. In fact, this is consistent with a developmental pattern outlined in the PWM regarding substance use: low levels of risk willingness in early adolescence → higher BW levels later in adolescence, which eventually develop into intentions for young adults. Finally, and again consistent with the PWM, these changes in BW did not occur, even for the high-impulsive adolescents, when they were reminded of potential negative consequences of the behavior.

Explicit vs. implicit measures

Reaction times on the implicit measure of attentional bias, the dot probe, correlated with changes in alcohol willingness for all adolescents. The more they were thinking about alcohol after watching the clips, the more they increased their willingness to drink. The correlation between this implicit measure and the more explicit measure (change in BW) is consistent with that found in other alcohol studies: r ~ .20 (Wiers, van de luitgaarde, vanden Wildenberg, & Smulders, 2005). This kind of implicit measure is particularly useful for identifying short-term memory cognitions (Reich, Below, & Goldsman, 2010), such as those likely produced by brief movie scenes. Moreover, research by Stacy and colleagues has shown that implicit measures can effectively predict alcohol use (among adolescents as well as adults) in concert with explicit measures (Wiers et al., 2002). Ames, Sussman, Dent and Stacy (2005) also showed that implicit cognitions mediated the relation between impulsivity and alcohol use and alcohol problems. We do not believe the impulsive adolescents in this study are at high risk for problematic drinking. However, we do believe that certain implicit measures can provide a useful addition to more traditional explicit measures when examining media effects on drinking and other kinds of risky behaviors that could lead to problems later on, especially among adolescents who are high in impulsivity.

Media

Along the same lines, the fact that the implicit measure was only weakly related to intentions to drink is consistent with a dual-processing perspective. Brief exposures to the drinking scenes are not likely to afford the time necessary to develop an intention to act. But they can alter the kind of reactive, heuristic-based responding that is associated with willingness. The heightened alcohol willingness, in turn, can be seen in the attentional bias (toward alcohol) that was demonstrated, once again, in a rapid processing situation—the dot probe. Especially within movies, heuristics are numerous (conformity/peer pressure, focus on impulse, association with attractive persons), just as they comprise a large portion of the take-home messages from many substance-related (and other) advertisements. It may change with continued exposure over time (i.e., years of watching positive depictions of drinking), but the type of impulsive responding that characterizes much adolescent risk behavior can be prompted by relatively few exposures (Gerrard et al., 2006), especially in the media, and that can be detected by willingness measures. Thus, although there is good reason to assess the long-term effects of MAE on adolescents’ intentions to drink, we would suggest that examinations of short-term exposures focus more on willingness measures.

Limitations and Future Directions

There are several limitations of the current study to be noted, some of which suggest possibilities for future research. First, although the pattern of means was as expected among the High-impulsive group in the negative and positive conditions, their BW mean was also fairly high in the No-alcohol condition. All adolescents did view some drinking in one of the control clips, Under the Tuscan Sun, and, although that segment was very short, it was definitely positive. In the absence of any portrayal of negative consequences, it is possible that the High impulsive group may have been reacting to that exposure in the No-alcohol Condition; but, we don’t know for sure. Second, the wording change from T1 to T2 for the BI construct required us to use a standardized version of that item. That may have contributed somewhat to the lack of effect (systematic change) for this construct.

Third, although there is considerable value in experimental designs, there are also drawbacks associated with brief presentations, like those used here. We did not identify a mediator of the MAE effects, for example. It is possible that the brief presentations did not allow for the hypothesized mediation process (MAE → reaction to the movie → change in BW) to occur. Of course, it is also possible that we did not assess the right mediator. In this respect, participants’ “exciting” evaluations did relate to their changes in BW, but that didn’t vary as a function of valence of the clips or level of impulsivity. Similarly, Dal Cin et al. (2009) and Gibbons et al. (2010) found that the effects of long-term MAE on willingness and then on actual drinking several years later were mediated by same-age drinker prototypes; that was not the case here. It is quite possible that short-term exposures are insufficient to change adolescents’ perceptions of a very diverse group of people—those who drink. One characteristic of the image (prototype) that can be modified in future media studies, however, is age. As in most previous prototype studies, we assessed same-age (i.e., adolescent) images; but the vast majority of drinking in movies is done by adults. Perhaps MAE can alter images of adult drinkers, which then could affect BW to drink. More research on mediation is definitely called for; that research should include additional implicit measures as well as other types of drinker prototypes. Finally, there are many other factors besides those assessed or manipulated in this study (e.g., peer pressure, expectancies) that affect adolescent drinking behavior. Future studies should include such measures in addition to those assessed in this study.

Conclusion

Brief portrayals of alcohol consumption can have an effect on adolescents’ willingness to drink, but there are important qualifiers of the effect. First, it appears to be true more so for adolescents who are high in impulsivity than those who are low in this dimension. Second, although the movie effect can be seen in heuristic measures, such as willingness, there was little evidence of an effect on the more analytic measures, intentions to drink. Finally, the effect is stronger for positive portrayals (e.g., alcohol facilitating social interaction) than negative portrayals (fights, arguments). Even high impulsive adolescents did not respond favorably to the alcohol portrayals when they were clearly accompanied by negative consequences. This suggests that interventions that highlight or direct adolescents’ attention to negative consequences of their risk behavior—especially social consequences (Kingsbury, Gibbons, & Gerrard, 2015)--may be effective. Finally, the results on both the explicit and implicit measures provide further evidence of the utility of a dual-processing approach to the study of media.