The Effect of Message Frame in Anti-Smoking Public Service Announcements on Cognitive Response and Attitude Toward Smoking

Abstract

This study investigated whether and how message frames in anti-smoking public service announcements (PSAs) affect individuals' cognition and attitude toward smoking. Individuals in a sample of 315 participants were randomly assigned to one of three experimental framing conditions: (a) health consequence, (b) secondhand smoke, and (c) industry manipulation. Each participant viewed four PSAs in a random order within a particular message frame. The study found strong evidence for the application effect in framing. The accessibility effect in framing was found to be conditional on message frame. Individuals' cognition on health consequence of smoking and on industry manipulation predicted their attitude toward smoking, but not cognition on secondhand smoke. The three frames also led to different patterns of affective responses that can be a basis for persuasion. Implications for message framing effect and anti-smoking campaigns were discussed.

There has been evidence for the effectiveness of anti-smoking media campaigns (for reviews see Hornik, 2002; Wakefield, DPhil, Nichter, & Giovino, 2003). This success is reflected in the facts the smoking rate among adults in the United States is decreasing (22.1% in 2003), and that smoking rates among youth have declined considerably since 1997. However, the smoking rate among adults is falling too slowly to meet the government's goal of 12% or less by 2010 (Centers for Disease Control and Prevention [CDC], 2004). In addition to tobacco companies' marketing activities, this “insufficient” success of anti-smoking campaigns could be due to (a) ineffective anti-smoking messages (e.g., Farrelly et al., 2002) and/or (b) anti-smoking messages that arouse reactance (boomerang effects) among recipients (e.g., Grandpre, Alvario, Burgoon, Miller, & Hall, 2003). To increase campaign success, it is important to understand the impact of message features and how messages are processed in order to develop effective health messages (Fishbein & Cappella, 2006).

Scholars in public health and health communication have examined individuals' responses to different types of anti‐smoking messages and have identified certain features that enhance message effectiveness. There is evidence that anti-smoking ads that activate strong negative emotions are better received and associated with decreased intention to smoke (Biener, 2000), while messages that simply present health consequences of smoking might not work (Goldman & Glantz, 1998). In addition to emotionality, thematic content of anti-smoking ads is also important (Terry-Mcelrath et al., 2005). For example, messages that dispose the manipulative and misleading nature of the tobacco industry (for more sophisticated audiences) and those that focus on the negative consequences of secondhand smoke were found to be more effective (Goldman & Glantz, 1998). Norm-based messages were also found to be effective for young adolescents (Pechmann & Knight, 2002).

On the other hand, there are also inconsistencies regarding the effectiveness of different types of anti-smoking ads. Wakefield et al. (2003) pointed out that there is no single recipe for anti-smoking advertising that reduces youth smoking, and that the population of youth might not be homogeneous regarding which types of ads are the most effective. Moreover, the effectiveness of message themes might have been confounded with that of other message features (e.g., message sensation value), particularly in studies that have used actual ads aired in states and nationally (e.g., Pechman & Knight, 2002; Terry-Mcelrath et al., 2005). It is fruitful, then, to explore and better understand the psychological mechanisms that underlie the effectiveness of message themes while controlling for other message features. Such is the goal of this article. Using message framing theory as the framework, individuals' affective and cognitive responses to anti-smoking ads presented in three message frames (i.e., themes) are investigated as the mediating mechanisms to message effects.

Message frame means a central organizing idea or story line that provides meaning to an unfolding strip of events (Gamson, 1992; Pan & Kosicki, 2005; Price, Tewksbury, & Powers, 1997;). The analysis of Farrelly, Niederdeppe, and Yarsevich (2003) showed that this “frame as story line” approach has been adopted in the design and production of anti-smoking ads. Farrelly et al. (2003) identified several major themes or message frames in anti-smoking PSAs aired in media campaigns. This article investigates the impact of three message frames: health consequence, secondhand smoke, and industry manipulation. The consequence frame in anti-smoking PSAs focuses on the short- and long-term health consequences of smoking, such as bad breath, yellow teeth, disease and death, and so on. The secondhand smoke frame emphasizes dangers of secondhand smoke on nonsmokers, as both short- and long-term impacts. The industry manipulation frame documents how the tobacco industry denies the health risks and addictive nature of the product, lies, and targets teens in advertising. These message frames were selected for investigation in this study because (a) they are the most studied and found to be effective and (b) they are more widely used (see analysis by Farrelly et al., 2003). First, hypotheses are derived based on framing theory (Pan & Kosicki, 2005). Second, data from a quasi-experimental study are analyzed to test the hypotheses and to assess the impact of message frame in anti-smoking PSAs on recipients' cognition, affect, and attitude toward smoking.

MESSAGE FRAMING THEORY

Frame

Scholars tend to agree that frames reside in two arenas: the discourse and the message recipients (Pan & Kosicki, 2005). Frames that reside in the discourse are linguistic devices. Message framing thus involves a message source placing a target (e.g., an object, an event, and/or an issue) in an interpretive framework and laying out the basic premises upon which arguments regarding the target might be developed. It consists of highlighting some aspects of a perceived reality and making them more salient in a communicative context in such a way as to promote a particular problem definition, causal interpretation, moral evaluation, and/or treatment recommendation. Such frames that reside in the discourse can be labeled as message frames.

Frames that reside in the message recipients are endogenous in the individual—they mean a recipient adopting an interpretive framework for thinking and/or talking about a target. Such a frame involves activating a relevant cognitive structure in one's memory (and possibly forming a new one), and invoking some values or principles as a premise to organize the recipient's responses to a target in an uncertain situation while excluding other potentially applicable values and principles. In this sense, frames are constructs or cognitive nodes stored in one's memory and can be information about social objects and their attributes; goals, values, and motivations; and affective or emotional states (Price & Tewksbury, 1997; see also Dillard, Solomon, & Samp, 1996; Nabi, 2003). Such frames endogenous in the message recipient can be called psychological frames. They are conceived as structured cognitive devices that individuals use to make sense of information in the stimuli. The likelihood of such a frame being activated is determined by its accessibility in one's memory (Higgins, 1996; Pan & Kosicki, 2005). The accessibility of a psychological frame can be chronic, which is mainly determined by dispositional individual differences; or situational, which is determined by its recency (i.e., the last time it was activated) and frequency (i.e., how often it is activated).

Framing Effect

Originating in prospect theory (Tversky & Kahneman, 1981), the framing effect in persuasion can be conceptualized as a process of social influence that connects message frames and psychological frames (Pan & Kosicki, 2005). As a process, framing effect thus impacts both how a message is processed and how the message impacts persuasive outcomes. Price and Tewksbury (1997) decompose framing effect into two parts: an application effect and an accessibility effect. The application effect occurs at the initial stage of message processing, where the message frames activate certain knowledge structures (i.e., psychological frames), making them more likely to be applied in response to the message. Once activated, the psychological frames (i.e., structured cognitive devices) become more accessible, making them more likely to be used in subsequent evaluations of the message and message advocacy. This is called the accessibility effect.

In this perspective, therefore, framing effect can be defined as such a social influence process: exposure to a message frame → activation of corresponding interpretative principle → processing the message within this framework → message impact. In other words, the framing effect is caused by message frames, but mediated by psychological frames. Although in a different area of research, the relational framing theory (Dillard et al., 1996) has similar conceptualization of how frames in the discourse influence individuals' perceived frames, which further impact relational outcomes. In addition, Fishbein and Yzer (2003) also suggest that media messages can strengthen the association between beliefs about smoking and attitudes toward smoking in health campaigns, even without changing the means of the beliefs. Guided by this conceptualisation of message framing effect, the following hypotheses were advanced:

• H1: Message frames in PSAs predict themes in cognitive responses (the application effect) such that:

• H1a: The consequence frame leads to more thoughts related to health consequences of smoking.

• H1b: The secondhand smoke frame leads to more thoughts related to secondhand smoke.

• H1c: The industry manipulation frame leads to more thoughts related to the tobacco industry.

• H2: Themes in cognitive response activated by the message frame predict attitude toward smoking (the accessibility effect) such that:

• H2a: Among respondents exposed to the consequence frame, thoughts on consequences of smoking predict attitude, but not thoughts on secondhand smoke or tobacco industry.

• H2b: Among respondents exposed to the secondhand smoke frame, thoughts on secondhand smoke predict attitude, but not thoughts on consequence or tobacco industry.

• H2c: Among respondents exposed to the industry manipulation frame, thoughts on the tobacco industry predict attitude, but not thoughts on consequence or second hand smoke.

In addition to cognition, affective responses to anti-smoking PSAs might provide another means of attitude change (Dillard & Peck, 2001). A recent content analysis (Rhodes, Roskos-Ewoldsen, Hull, & Monahan, 2007) shows that the anti-smoking PSAs are relatively high in message sensation value. There has been evidence that message sensation value predicts emotional responses and processing of anti-smoking messages (e.g., Niederdeppe, 2005; see also Morgan, Palmgreen, Stephenson, Hoyle, & Lorch, 2003). There is also strong evidence that the discrete emotions can be the basis for attitude change (Dillard & Meijnders, 2002). Because there is no clear theoretical guidance regarding how message frames in anti-smoking PSAs might impact discrete emotions, a research question was asked:

• RQ: How might the three message frames impact attitude via discrete emotions?

METHOD

Stimuli Messages

The stimuli messages were selected from a collection of anti-smoking PSAs. Four PSAs were selected from each of the three message frames. A team of three researchers participated in the process of message selection to ensure that possible overlap between the frames was nonexistent or minimal. The appendix presents synopses of these anti-smoking PSAs.

Experimental Design

The study was a 3 (message frame) × 4 (message sequence) between-subjects factorial design. The participants were randomly assigned to one of the three message frames and watched four PSAs within that particular frame, which were presented in one of the following sequences: 1, 2, 3, 4; 2, 4, 1, 3; 3, 1, 4, 2; or 4, 3, 2, 1.

Participants

Participants were 315 students recruited from the research subject pool from introductory communication classes at the University of Georgia. They ranged in age from 17 to 24 years (M = 19.78, SD = 1.71), with 87% describing themselves as white/Caucasian, 8% as being of Asian descent, 2% as Hispanic descent, 2% as African descent, and 1% as other. Sixty-two percent reported their sex as female and 38% as male. There were slight fluctuations in the actual sample size in data analyses due to missing values. In addition to easy access and availability, the use of a college students sample can also be justified: (a) College students do smoke a lot (smoking rate being 28.5%; see Wechsler, Rigotti, Gledhill-Hoyt, & Lee, 1998), and (b) they are an important target audience for anti-smoking messages. It should be noted that the college population has been a secondary, not primary, target of PSAs in the industry manipulation frame.

Procedure

When the participants arrived at the laboratory, they were informed that their task would be to evaluate some television PSA messages. They were randomly assigned to one of the 12 experimental conditions. Participants signed and returned consent forms to the experimenter before they received the questionnaire. They then watched the messages projected onto a large screen. After participants had watched the first message, they provided affective responses on close-ended scales, and listed whatever thoughts that came to their minds when they were watching the PSA. They were asked who they were angry at (the PSA, the message source, the experimenter, smokers, or the tobacco companies) if they experienced anger at all when they were watching the PSA. The participants then rated the message sensation value (Morgan et al., 2003) of each message. The second message was then played and the same procedure was repeated for the subsequent messages. After all four PSAs had been viewed and responses to the messages completed, participants reported their attitude toward smoking. Finally they answered questions regarding their demographic information, and questions regarding their smoking behavior. An entire session lasted about 35 min. The participants were thanked and their questions answered, if they had any, before they left the laboratory.

Measures

Discrete emotions

Emotional reactions to the PSAs were assessed after each message using on a 5-point response scale where 0 = none of this feeling and 4 = a great deal of this feeling (Dillard & Peck, 2001). The scales and their corresponding items were: surprise (surprised, startled, and astonished; α =. 85); disgust (sickened, disgusted, and revolted; α = .83); fear (fearful, afraid, and scared; α = .93); sadness (sad, dreary, and dismal; α = .81); happiness (happy, content, and cheerful; α = .76); anger (irritated, angry, annoyed, and aggravated; α = .88).

Cognitive responses

Three coders received 5 hr of training before they started coding the first fifth of the data. Reliabilities were checked for this portion of the data only. The remainder of the data was divided among the three researchers and coded by one person only. Coding took place in four steps. First, the coders segmented the data into psychological thought units (roughly, independent clauses). In some cases, the grammatical subject was provided to some units. For example, the response “Smoking is a selfish and disgusting thing to do around other people” were coded as two units, as the grammatical subject “smoking” is added to the second unit, “(Smoking) is a disgust thing to do around other people.” The coders agreed upon 93–96% of the thought units. Guetzkow's U was around .02 for all three pairs of coders.

Second, affective responses were culled from the open‐ended cognitive response data. Coders used a list of feeling terms compiled by Shaver, Schwartz, Kirson, and O'Connor (1987), supplemented by emotional words used by the participants in their responses to the messages. These supplemented terms tended to be in more colloquial forms, for example, “grossed out,” “(feel) good,” etc. A unit was classified as affective whenever those words appeared. The intercoder reliability (Krippendorf'sα) for this step was 1.00.

Third, coders evaluated the relevance of the cognitive responses to the message. The purpose of this step was to screen irrelevant cognitions and thereby reduce the level of noise in the data. Relevant thoughts were those related to the task (i.e., research participation), message (i.e., content or delivery), the particular health problem addressed in the message (i.e., smoking), or the message advocacy. Irrelevant thoughts were unassociated with the task or the messages. For example, “I'm hungry” was coded as irrelevant. Krippendorf's α for this step was .92.

Finally, the relevant cognitive thought units were coded into different topic areas, or cognitive themes: health consequence, secondhand smoke, and tobacco industry. The themes in cognitive responses are identified by (1) the unit's focal center, which refers to the grammatical subject phrase of the unit, which may be defined in terms of the source and subject of a message; and (2) the content subject of a unit, which highlights whom or what the message is about (Samp & Solomon, 2005). For example, the unit “Smoking is really bad (for one's health)” is coded as having a consequence frame due to its focal center; the unit “That (seeking profit) is what tobacco companies do” is coded as having an industry frame due to its content subject. Krippendorf's α for this step was .81. The total number of cognitive responses with a particular focal center or content subject indexes the theme in cognitive response.

Message sensation value

The perceived video/audio features of the PSA messages were measured by 10 features of the scale of message sensation value (Morgan et al., 2003): cuts, special visual effect, slow motion, unusual colors, intense images, sound saturation, sound effects, acted out (vs. talking head), unexpected format, and twist ending. The music judgment included in Morgan et al. was not retained because it did not predict perceived message sensation value in their investigation. For each dimension, 0 = “lack of this feature” and 1 = “this feature is present.” Morgan et al. (2003) provided empirical evidence that the subjective message sensation value is associated with the formal and content features of the PSA messages. These 10 dimensions of subjective message sensation value were summed to create a scale of message sensation value. Because these message features were considered as formative indicators (i.e., causes) of message sensation value, not reflective indicators, scale reliability was irrelevant (Diamantopoulos & Winklhofer, 2001); hence, Cronbach's α was not computed.

Attitude toward smoking

Five 1- to 5-point semantic differential items were used to measure attitude toward smoking. The word pairs were: good/bad, wise/foolish, favorable/unfavorable, desirable/undesirable, beneficial/detrimental. These five items were averaged into an index for attitude toward smoking (Cronbach's α = .94).

Smoking behavior

Three questions were asked to measure smoking behavior. The first question asked which term best describes the participant: non-smoker (0), occasional smoker (1), or regular smoker (2). The second question asked how often the participant smokes (0 = never, 1 = rarely, 2 = sometimes, 3 = everyday). The third question asked on average how many cigarettes the participant smokes everyday (0 = none, 1 = 1–10, 2 = more than 10, 3 = more than a pack). The z scores of these three items were averaged into an index for smoking behavior (Cronbach's α = .73).

Strategies for Data Analyses

Two procedures were selected to take care of the interdependence due to repeated measures: The multivariate General Linear Model (GLM) procedure was selected to estimate the impact of message frames on cognitive and affective responses because that procedure yields direct mean comparisons across the three message frame conditions and standardized effect sizes. Pairwise comparisons were carried out via the Bonferroni method to adjust for possible Type I error inflation due to multiple tests. Two-level models, with individuals at the higher level and measurement occasions (i.e., message nested within frame) as repeated measures clustered within individuals, were also estimated via the Linear Mixed Model (LMM) procedure in SPSS to explore the impact of cognition and affect on attitude toward smoking. The advantage of the two-level model is that it controls for (a) random variance at the individual level, (b) random variance at the message level, and (c) the covariance across the two levels (i.e., intraclass correlation). Assuming alpha = .05, with a sample size of 1,252 (observations), the statistical power to detect an effect size equivalent to r = .10 exceeded .99.

Data analyses were also carried out for smokers versus nonsmokers separately. The differences in means of cognitive and affective responses and their impact on attitude toward smoking were very similar across the two groups. Hence, results were collapsed across the two groups for ease of presentation and interpretation.

RESULTS

Effects of Message Frame on Cognition

Hypothesis 1 predicts that message frames dictate the themes in one's cognitive responses to the anti-smoking PSAs. A multivariate GLM procedure, with message theme as a between-subjects factor, and message sequence, message sensation value, sex, age, and smoking behavior as covariates, was used to compare the themes in cognitive responses. The multivariate F (Pillai's trace) was significant for message frame: F(6, 2448) = 5.58, p < .001, η² = .22. This means that in general, message frames predicted themes in cognitive responses.

Univariate analyses showed that the main effect of message frame was significant on all three themes in cognitive responses. The top panel in Table 1 presents the marginal means for themes in cognitive responses within each message theme. For the consequence theme, the main effect of message frame was significant: F(2, 1223) = 8.92, p < .001, η² = .10. Individuals who watched the PSAs within the consequence frame had more thoughts on health risks associated with smoking (M = .41, SD = .64) than those in the secondhand smoke frame (M = .05, SD = .22) or the industry manipulation frame (M = .14, SD = .22). The main effect of message frame was also significant on the secondhand smoke theme: F(2, 1223) = 38.30, p < .001, η² = 17. Individuals who were exposed to the secondhand smoke PSAs had more thoughts related to secondhand smoke (M = .54, SD = .89) than those in the consequence frame (M = .03, SD = .29) or the industry manipulation frame (M = .01, SD = .07). Message frame also significantly predicted the industry manipulation theme in cognitive responses: F(2, 1223) = 84.83, p < .001, η² = 25. Those who were shown the industry manipulation PSAs had more thoughts on tobacco companies (M = .83, SD = 1.03) than either the consequence frame (M = .04, SD = .39) or the secondhand smoke frame (M = .01, SD = .07).

TABLE 1 Marginal Means of Cognitive and Affective Responses as a Function of Message Frame

	Message Frames
Message Frame	Consequence	Secondhand Smoke	Industry manipulation
Cognitive themes
Consequence	.41 ^a	.07 ^c	.14 ^b
Secondhand smoke	.02 ^b	.54 ^a	.01 ^b
Industry manipulation	.04 ^b	.01 ^b	.83 ^a
Discrete emotions
Anger	.96 ^c	1.29 ^b	1.73 ^a
Surprise	2.07 ^a	1.69 ^b	1.59 ^b
Fear	1.22 ^a	1.17 ^a	.51 ^b
Sadness	1.58 ^a	1.12 ^b	.90 ^c
Disgust	1.76 ^a	1.07 ^c	1.37 ^b
Guilt	.50 ^a	.26 ^b	.41 ^a
Happiness	.09 ^b	.41 ^a	.20 ^b
Note. Sex, age, message sequence, message sensation value, and smoking behavior controlled for.
^a,b,cMarginal means in each row with different superscripts are significantly different at p < .05 (a > b > c); means with the same superscripts are not significantly different from each other.

It should be noted that the observations (i.e., cognitive response to each PSA) within the same message frame were not independent of each other, which could introduce biases to the GLM results. To explore that possibility, three two-level models were also estimated via the LMM procedure in SPSS, using message frame to predict each of the theme in cognitive responses, with message theme and sequence as between-subjects factors, and message sensation value, sex, age, and smoking behavior as covariates. The results were consistent with the GLM results. These results demonstrated strong evidence for H1 and each of its three components.

Effects of Message Frame on Affect

Part of the research question involves the impact of message frame on individuals' emotional responses. A multivariate GLM procedure, with message theme as a between-subjects factor, and message sequence, message sensation value, sex, age, and smoking behavior as covariates, was used to compare the seven emotions. The multivariate F (Pillai's trace) was significant for message Frame: F(14, 2456) = 51.23, p < .001, η² = .22. This means that overall, the three message frames led to different patterns of affective responses among the recipients.

Univariate analyses showed that the main effect of message frame was significant on all seven discrete emotions. The bottom panel in Table 1 presents the marginal means for the seven emotions within each message theme. The main effect of message frame on anger was significant: F(2, 1233) = 46.46, p < .001, η² = .07. The industry manipulation frame led to the strongest anger (M = 1.73, SD = 1.21), followed by the secondhand smoke frame (M = 1.29, SD = 1.16), and the consequence frame led to the least amount of anger (M = .96, SD = 1.05). When individuals were asked whom they were angry at, smokers and tobacco companies were the two main targets (within the corresponding message frame). The message, message source, or the experimenter did not seem to be the target of anger from the participants.

The main effect of message frame on surprise was significant: F(2, 1233) = 21.15, p < .001, η² = .03. The consequence frame led to more surprise (M = 2.07, SD = 1.15) than the secondhand smoke frame (M = 1.69, SD = 1.18) and the industry manipulation frame (M = 1.59, SD = 1.14), which were not significantly different regarding amount of this affect they activated. The effect of message frame on fear was also significant: F(2, 1233) = 51.66, p < .001, η² = .08. The consequence frame led to most fear (M = 1.22, SD = 1.32), but not significantly more than the secondhand smoke frame (M = 1.17, SD = 1.15). The industry manipulation frame elicited significantly less fear (M = .51, SD = .84). The impact of message frame on sadness was also significant: F(2, 1233) = 50.56, p < .001, η² = .08. The consequence frame led to significantly stronger sadness (M = 1.58, SD = 1.02) than the secondhand smoke frame (M = 1.17, SD = 1.02), which led to significantly more sadness than the industry manipulation frame (M = .51, SD = .91). The impact of message frame on disgust was also significant: F(2, 1233) =36.42, p < .001, η² = .06. The consequence frame (M = 1.76, SD = 1.34) caused significantly higher level of disgust than the industry manipulation frame (M = 1.37, SD = 1.12), which aroused more disgust than the secondhand smoke frame (M = 1.07, SD = 1.07). The effect of message frame on guilt was also significant: F(2, 1233) = 11.88, p < .001, η² = .02. The consequence frame (M = .50, SD = 97) and the industry manipulation frame (M = .41, SD = .76) led to significantly more guilt than the secondhand smoke frame (M = .26, SD = .69), but were not significantly different from each other.

The anti-smoking PSAs led to less happiness than the negative emotions across the three message frames. There was a significant main effect of message frame as well: F(2, 1233) = 12.32, p < .001, η² = .02.The secondhand smoke frame led to the highest level of happiness (M = .46 SD = .78) than either the consequence frame (M = .09, SD = .59) or the industry manipulation frame (M = .20, SD = .65), which were not significantly different from each other.

Seven two-level models were also estimated via the LMM procedure in SPSS, using message frame to predict each of the seven emotions, with message theme and sequence as between-subjects factors, and message sensation value, sex, age, and smoking behavior as covariates. The results were consistent with the GLM results. These results showed that message frame did arouse different patterns of affective responses. All three frames caused more negative emotions than happiness. The consequence frame caused more surprise, fear, sadness, and disgust. The secondhand smoke frame aroused strong anger, fear, sadness, disgust, and happiness. The industry manipulation frame elicited strong anger and disgust.

Impact of Affect and Cognition on Attitude toward Smoking

Hypotheses 2 and part of the RQ concern how the cognitive and affective processes that mediate the impact of the PSAs on attitude toward smoking would differ across the three message frames. Recall that the measures for cognitive and affective responses were repeated across measurement occasions defined by PSA (four in total) nested within each message frame. Three two-level models were estimated using the LMM procedure in SPSS, with individuals at the higher level and measurement occasions as repeated measures clustered within individuals, to predict attitude toward smoking using individuals' cognitive and affective responses within each message frame, while controlling for age, sex, message sequence, and smoking behavior. Table 2presents the fixed-effects parameter estimates from the three two-level models.

TABLE 2 Using Cognitive and Affective Responses to Predict Attitude Within Each Message Frame

Fixed-Effects Parameters (β)	Message Frame
	Consequence	Secondhand Smoke	Industry Manipulation
Thoughts on:
Consequence	−.10∗∗	−.04	−.07∗
Secondhand smoke	−.05	−.06	.01
Tobacco industry	−.01	−.03	−.09∗∗
Anger	.01	.00	.00
Surprise	−.03	−.02	.01
Fear	−.07∗	−.01	−.03∗
Sadness	.00	−.02	−.02
Disgust	.01	−.03∗	.01
Guilt	−.01	−.04∗	.01
Happiness	.01	−.05∗	−.03
Note. Sex, age, message sequence, message sensation value, and smoking behavior are controlled for. Values are maximum likelihood estimates, and only fixed-effects parameters (β terms) are presented.
∗ p < .05, ∗∗ p < .01.

Themes in cognitive response

Within the consequence frame, the corresponding theme in cognitive response had a significant negative impact (β = −.10, p < .01) on attitude toward smoking, while the impact of other two themes (β = −.05 for secondhand smoke theme, and β = −.01 for industry manipulation theme) did not reach significance. Comparison of the three coefficients following the Cohen and Cohen (1983) procedure (pp. 53–55) showed that the impact of consequence theme was significantly larger than that of the other two themes. This provided evidence for H2a. Within the secondhand smoke frame, none of the three themes in cognitive response was a significant predictor of attitude toward smoking. Therefore, H2b was not supported. Within the industry manipulation frame, the impact of the corresponding theme on attitude was negative and significant: β = −.09, p < .01. However, the consequence theme was also a significant predictor: β = −.07, p < .05. Comparison of the two coefficients following the Cohen and Cohen (1983) procedure showed that the two were not significantly different from each other. Hence, H2c did not receive support either, although the impact of the secondhand smoke theme was non-significant within this message frame.

Discrete emotions

The impact of discrete emotions on attitude toward smoking varied across the three message frames. Fear significantly predicted attitude toward smoking within the consequence frame (β = −.07, p < .05) and the industry manipulation frame (β = −.03, p < .05). None of the other emotions was a significant predictor. Within the secondhand smoke frame, three emotions significantly predicted attitude: disgust (β = −.03, p < .05), guilt (β = −.04, p < .05), and happiness (β = −.05, p < .05).

DISCUSSION

The majority of studies on anti-smoking PSAs focus on message effectiveness but pay less attention to the mechanisms that might underlie message impact. Setting out to better understand processing of anti-smoking PSAs, this study investigated the message framing effect, as well as how message frames might impact attitude toward smoking via affect. Data from a quasi-experimental study provided strong evidence for the application effect in framing. The accessibility effect in framing was found to be conditional on message frame. Individuals' cognition of health consequence of smoking and of industry manipulation predicted their attitude toward smoking, but not cognition of secondhand smoke. The three frames also led to different patterns of affective responses that can be basis for persuasion.

The effects seemed small. However, recall that the key variables were individuals' responses to each individual message, not to four messages combined, and that each PSA lasted for only 30 s. Another reason was that only cognitions with a clear theme, not all responses, were included in the data analyses. In addition, the participants, who were participating to fulfill class requirements or for a small proportion of extra credit, probably were not very motivated.

The Application Effect in Message Framing

Hypothesis 1 concerns the application effect in framing such that the message frames in the anti-smoking PSAs would positively predict the themes in cognitive responses. The results from repeated-measures GLM analyses provided evidence for such application effect: Individuals listed more thoughts on consequence within the consequence frame, more thoughts on secondhand smoke within the corresponding message frame, and more thoughts on the tobacco industry within the industry manipulation frame. This is consistent with studies on framing effects that used news stories that are in print, longer, and possibly more involving than 30-s-long television PSAs (Price et al., 1997).

The individuals listed considerable (although significantly less) thoughts on consequence of smoking within the secondhand smoke and industry manipulation frames. This pattern was absent when it comes to the cognitive themes on secondhand smoke or industry manipulation; that is, individuals listed very few thoughts on secondhand smoke or industry manipulation within noncorresponding message frames. There are two possible explanations. The first explanation lies in the overlap between the frames of consequence and secondhand smoke, and between the frames of consequence and industry manipulation: (a) The harms secondhand smoke does to nonsmokers are exactly what smoking does to smokers; and (b) what the tobacco industry has been lying about is the health consequences of smoking. The second explanation lies in the chronic psychological frame within the individuals due to anti-smoking messages that they had been exposed to outside of this study, which tend to have an emphasis on the consequences of smoking (e.g., Rhodes et al., 2007). In other words, the psychological frame on consequence of smoking is more recent and frequent, that is, more accessible than the psychological frames on secondhand smoke and on industry manipulation. The cues related to smoking within the frames of secondhand smoke and industry manipulation might have been enough to activate thoughts on the health consequences of smoking, whereas there were not enough cues to increase the accessibility of thoughts on secondhand smoke or thoughts on industry manipulation when the message was not presented within the corresponding frame.

The Accessibility Effect in Message Framing

Hypothesis 2 predicted the accessibility effect in framing such that within each message frame, the corresponding theme in cognitive response would significantly predict attitude toward smoking, but not the other two themes. Evidence for this hypothesis was mixed at the best. There was evidence for the accessibility within the consequence frame: The corresponding theme in cognitive response predicted attitude toward smoking, but not the other two themes. The corresponding theme predicted attitude within the industry manipulation frame as well, but so did the consequence frame. None of the themes was a significant predictor within the secondhand smoke frame. These results did not fully support H2, indicating that the accessibility effect in framing might not be very robust, instead might be conditional on message frame: There was significant accessibility effect under the consequence frame, but none under the secondhand smoke frame, and mixed under the industry manipulation frame. This is consistent with the Price et al. (1997) claim that the accessibility effect might be subtle and more difficult to detect.

Although inconsistent with H2c, the main effect of consequence theme on attitude within the industry manipulation message frame is actually congruent with what the theory would predict. Recall that in addition to thoughts on the tobacco industry, individuals also listed considerable thoughts on health consequences associated with smoking (see Table 1). This means that the psychological frame of consequence was also activated and accessible (although less so than the psychological frame of industry manipulation), which makes it more likely to be utilized in the process of attitude formation/change.

The null effect within the secondhand smoke frame might be due to the lack of correspondence between the message advocacy and measured outcome. The advocacy in anti-smoking PSAs with a secondhand smoke frame seems to focus on not to smoke in public places or around other people (presumably nonsmokers). The frame of secondhand smoke, therefore, might significantly predict one's attitude toward smoking in public places or around nonsmokers, but not attitude toward smoking per se. Preliminary analysis of the thought-listing data revealed some evidence for this explanation. A typical response to the “Driver” PSA (see the Appendix for a synopsis) was “he (the smoker) should have asked (for permission to smoke).” And “he should not have smoked around his own kid” was a very common response to the “Numbers” PSA.

It should also be noted that although it was not clearly explicated, the accessibility effect in message framing seems to have two components as well: (a) The psychological frame corresponding to the message frame becomes more accessible as a result of framing, and (b) the more accessible psychological frame, in turn, is more predictive of attitude. The impact of psychological frames on attitude might be indirect and a function of their accessibility. It is not surprising, therefore, that the effect of theme in cognitive response on attitude might not be detected. On the other hand, the impact of theme in cognitive responses on attitude is inferential and indirect evidence for the accessibility effect. More direct evidence should come from the impact of psychological frames on the accessibility of beliefs and/or attitude, which can be measured as the inverse of reaction time to belief statements (agree vs. disagree) or attitudinal objects (like vs. dislike) (Arpan, Rhodes, & Roskos-Ewoldsen, 2007). Future studies that directly measure accessibility of cognition and attitude should help further our understanding of the accessibility effect in the process of message framing.

The Role of Discrete Emotions

Framing effect is considered as mainly cognitive in nature, and the role of discrete emotions tends to be ignored in the literature. This study suggested that different message frames could lead to very different patterns of affective responses, which could be the basis of attitude change. Overall, the three message frames caused stronger negative emotions than happiness, probably due to the unpleasant topic they addressed. Message frames that concerned risks associated with smoking aroused stronger fear. Consistent with the fear appeal literature, fear was found to significantly predict attitude toward smoking within two of the three message frames. Its null effect within the secondhand smoke frame was probably due to the lack of correspondence issue discussed in the previous section.

The secondhand smoke frame and the industry manipulation frame led to stronger anger. However, inconsistent with the literature on anger and reactance (Dillard & Shen, 2005), anger was not found to be detrimental to persuasive efforts. Smokers and tobacco companies being the target of anger, instead of the message or message source, explains the absence of reactance. On the other hand, anger toward tobacco companies or smokers does not seem to directly facilitate anti-smoking persuasive efforts either. It is possible that the impact of anger toward tobacco companies or smokers might be disassociated with attitude toward smoking, but rather it could be associated with attitude toward tobacco companies, which in turn predicts smoking intentions and behavior (see Hersey et al., 2005). This could explain why anger toward tobacco companies was not more strongly associated with attitudes toward smoking. This possibility needs to be further explored in future studies with appropriate design and method.

Disgust, guilt, and happiness were found to be significant predictors of attitude toward smoking only within the secondhand smoke frame. There exists the possibility that these emotions could have been the outcome while attitude toward smoking had been the cause. Analysis of the cognitive responses revealed that this was particularly likely for happiness: Those who voiced support for the Clean Air Act that bans smoking in public places also reported that they were “glad” or “happy” that such PSAs were produced and aired. This possibility is less likely for guilt and disgust: Smokers felt stronger guilt and disgust than nonsmokers, whereas these two emotions were negatively associated with attitude toward smoking. The possibility of sampling error, however, cannot be ruled out as an explanation for these effects.

The interpretation of these results regarding emotions, however, needs to be qualified by the fact that only attitude toward smoking was used as the outcome variable when testing the applicability effect in framing. This was considered as appropriate because (a) behavior change theories (e.g., theory of reasoned action) suggest that attitudes toward the behavior are comprised of beliefs about the consequences of performing that behavior and their relative weight, and (b) testing the applicability effect in framing requires a uniform outcome variable; otherwise, comparison among the psychological frame would become impossible. However, attitude toward smoking might not be the most appropriate target for beliefs about secondhand smoke. This also might not be appropriate for anti-industry messages. Future studies that have appropriate outcome measures should provide empirical test of such possibilities.

Implications for Anti-Smoking PSAs

There has been evidence for the effectiveness of anti-smoking PSAs in public health campaigns (Farrelly et al., 2002; Niederdeppe, Farrelly, & Haviland, 2004; Sly, Trapido, & Ray, 2002). It is less clear how these anti-smoking PSAs might be persuasive. Results from the current study suggested that when presented in different message frames (Farrelly et al., 2003), anti-smoking PSAs might lead the train of thoughts that individuals generate and emotions that they experience, which then predict potential persuasion outcomes.

The message frames shaped the themes in cognitive responses to anti-smoking PSAs, and cognition on consequence of smoking and on tobacco companies' unethical practice negatively predicted attitude toward smoking. This adds to the evidence from survey research (e.g., Niederdeppe et al., 2004; Sly et al., 2002) that anti-smoking campaigns can be successful by changing individuals' cognition. Such success, however, might be dependent upon the accessibility of cognition, which is a function of its recency and frequency (Higgins, 1996; Pan & Kosicki, 2005). This indicates that repeated and frequent exposure to multiple anti-smoking PSAs might be necessary to change one's attitude toward smoking, or for one to remain a nonsmoker (Sly et al., 2002). Single-shot or sporadic efforts are less likely to be effective because they have little impact on the recency and frequency of cognition on smoking consequence or industry manipulation.

Although the secondhand smoke frame successfully activated the corresponding cognitive theme, cognition on secondhand smoke did not impact one's attitude toward smoking. The open-ended responses suggest that it is possible that this message frame might reinforce or increase acceptance of smoking as long as it is not in public places or around other people. At least, data from the current study indicate that this message frame is not optimal if the goal is to change attitude toward and behavior of smoking.

In addition, information on risks associated with smoking tends to elicit fear among the recipients, which has a unique contribution to persuasion above and beyond the impact of cognition. This suggests that in combination with message framing, fear appeal can be an effective means of message design. On the other hand, anger and reactance are considered as obstacles in anti-smoking campaigns (Grandpre et al., 2003). The current study shows that message framing can be effective in channeling affective responses as well, and that if anger is directed to be against the tobacco industry or smokers, instead of the message and its source, the threat of boomerang effect might be reduced or avoided. It is also noteworthy that the impact of message frame on cognition and affect appears to be comparable among both smokers and nonsmokers. This implies that message frames might be an economic means of message targeting and reaching smokers and nonsmokers with the same messages.

Limitations and Directions for Future Research

Conclusions from the study and any generalization that might be drawn are qualified by the limitations in the study. First, only 12 anti-smoking PSAs within three message frames were utilized as stimuli in this study, while there are hundreds more anti-smoking PSAs; also, the three frames used in this study are not the only ones available (see Farrelly et al., 2003). Whether the findings in this study can be generalized to other message frames, or other health topics, remains an empirical question that can only be answered by future research. Second, the participants were all young adults enrolled in introductory communication classes and overwhelmingly white. Third, there is a potential mismatch between the messages' targeted audiences and the college sample used in this study. For example, the industry manipulation PSAs were targeted to 12- to 17-year-olds, whereas some PSAs in the other two frames were targeted more broadly to adults. This would seem to be consequential for their expectation of effects, and could explain why angry emotions didn't seem to be salient within the industry manipulation frame. The idea behind these anti-industry campaigns is that they tap into teen angst and channel that toward tobacco companies. Presumably this mechanism would be weaker among older teens and young adults. Future research with better match between the targeted audiences and samples would provide further insight if this might be the case. In addition, the study took place in a laboratory setting and the PSAs were not located within some form of regular TV programming; instead, they were shown four in a row. These features limit the external validity.

On the other hand, there are strengths in the study as well. The PSAs were professionally produced and used in public health campaigns, which increases the external validity of the study. The fact that the obtained results were above and beyond such factors as message sequence, smoking behavior, message sensation value, and demographic variables demonstrates internal validity of the study. The multiple testings (the two-level models were estimated within each message frame to test H2) might have inflated Type I error, particularly regarding the impact of the discrete emotions because they are significant at p < .05 level; however, that should be less of a concern regarding the effect of cognitive themes, which were significant at the p < .01 level. In addition, the two-level models took into consideration interdependence in data that arose from repeated measures, which should enhance the internal validity as well.

CONCLUSION AND SUMMARY

A quasi-experimental study was conducted to investigate the message framing effect in anti-smoking PSAs. Data from 12 PSAs within three message frames yielded strong evidence for the application effect in framing: Message frames significantly predicted themes in cognitive response. The accessibility effect was rather subtle: There was mixed evidence that the themes in cognitive response predicted attitude toward smoking. The message frames also impacted persuasion via emotion in a different pattern. The findings might have important implications for message framing effect, as well as for public anti-smoking campaigns.

ACKNOWLEDGMENTS

The author thanks Elizabeth Bigsby and Leslie Abbott for their assistance in data collection. This study was supported by a Junior Faculty Research Grant from the University of Georgia Research Foundation.

APPENDIX: SYNOPSES OF ANTI-SMOKING PSAs

PSAs Within the Consequence Frame

“Ghost”—Watched by the mother, a toddler is taking his first steps. The grandpa opens his arms when the toddler walks toward him. The kid walks right through him as he disappears. The mother says, “I wish your grandpa could see this.” And the following text appears: “Be there tomorrow. Stop smoking now.”

“Krystell”—Krystell, age 9, talks about how she does not want to be like her mom when she grows up while her friends do, because her mother started smoking when she was 10, and died from lung cancer. The images of her mother suffering from lung cancer in the hospital are shown as Krystell talks to the audience.

“Shower”—A man is taking a shower. He was 39 when he got throat cancer from smoking cigarettes and almost died. Now there is a permanent hole in his throat. He has to keep the hole clean. He talks how nothing will be the same, even the simplest things.

“Tar lung”—A girl sits on the sofa and lights up a cigarette. As she inhales, the smoke travels to her lung and forms a film of tar. A healthy lung is placed in a rectangular pan and cut open before a jar of tar is poured onto it, illustrating how much tar will be in one's lung from smoking for 1 year.

PSAs Within the Secondhand Smoke Frame

“Driver”—Two people are driving in a car when the man lights a cigarette. The female driver sneers at him and then begins to drive wildly. When he asks why she does this, the woman responds that because he is risking her life by exposing her to secondhand smoke, she is returning the favor.

“Millie”—A woman, Millie, describes her reaction to learning that she had two smoking-related inoperable tumors in her lungs. She discusses how she was particularly devastated because she never smoked. She developed cancer by inhaling secondhand smoke.

“Numbers”—A man is presented as a decent human being, a respected colleague, and a good husband and a loving father. He puts 38 people at danger because he smokes, including his own wife and daughter. Text appears that secondhand smoke kills 53,000 people a year.

“Waitress”—A group of waitresses talk about how their risks of heart disease and lung cancer are increased because of their exposure to secondhand smoke, and plead with the audience to support the law on smoke-free bars and restaurants.

PSAs Within the Industry Manipulation Frame

“Billboard”—Two men dismantle a roadside billboard in Senegal, Africa, just before an entourage of vehicles speeds past and then the men begin to remount the billboard in its original place. The announcer then notes that Marlboro kept the billboard down “until the President of the United States left” and that “what the tobacco industry says they don't do, they just don't do in the US.”

“Congressional testimony”—Tobacco executives testify about the relationship between smoking and disease and answer accusations about the marketing of cigarettes to youth. Apparently they were all lying.

“Cowboy”—Matt McLaren, brother of Wayne McLaren, one of the Marlboro cowboys who died of lung cancer, talks about how the tobacco industry used his brother in ads to create an image that smoking makes you independent. He tells the audience not to believe that.

“Insurance company”—In a black-and-white setting, tobacco executives are taking, laughing, and arguing in smoke filled rooms. The narrator points out that the same tobacco executives who deny that smoking is harmful own an insurance company that gives discounts to nonsmokers.