Social Diffusion of Campaign Effects

Abstract

This study examines the role of campaign conversation in the generation and diffusion of campaign effects. Based on Hornik’s social diffusion model of campaign influence, this study tests whether campaign conversation, along with campaign exposure, can be a process through which a campaign affects a person’s health perceptions using secondary analyses of the Truth^SM campaign data. The results of a multilevel modeling of the Legacy Media Tracking Survey (LMTS) data (n = 10,357) show that the effect of the Truth^SM campaign activity in the media on individuals’ smoking-related beliefs was conveyed through campaign exposure and campaign conversation. Theoretical and practical implications for campaign planning and evaluation are discussed in this article.

Keywords

social diffusion interpersonal communication exposure antismoking campaign

Numerous studies have reported that health campaigns can affect the public’s health perceptions or behaviors (e.g., Biglan, Ary, Smolkowski, Duncan, & Black, 2000; COMMIT Research Group, 1995; Farrelly, Davis, Haviland, Messeri, & Healton, 2005; Farrelly, Healton, et al., 2002; Flynn et al., 1994; Porto, 2007; for review, see Flay, 1987; Hornik, 2002; Noar, 2006). According to Snyder et al.’s (2004) meta-analysis, mass media health campaigns have a small but significant effect on people’s health behavior change. Although there is a general consensus regarding the existence of campaign effects, no such consensus exists with regard to the process of campaign effects. In fact, little research has been conducted to examine the processes whereby mass media health campaigns influence individuals’ health perceptions and behaviors (Rimal, Flora, & Schooler, 1999).

Traditionally, campaign effects have been modeled as a result of individuals’ direct exposure to campaign messages. In this traditional individual exposure model, campaigns can affect people’s health perceptions and behaviors if individuals are exposed to (e.g., watching, listening, or reading) campaign messages.

An individual’s exposure to a campaign, however, may not be the only way that the campaign can influence the individual’s behavior. Hornik (Hornik, 2002; Hornik & Yanovitzky, 2003) suggested that campaign effects can also occur via institutional and social routes. The institutional diffusion model suggests that campaigns can ultimately affect individual behavior through institutional change, such as changes in public policy. According to this model, a health communication campaign can produce its effects by affecting mass and elite opinions and generating new policy initiatives (e.g., smoking ban). The institutional diffusion model has received considerable scholarly attention. For example, Yanovitzky (Yanovitzky, 2002; Yanovitzky & Bennett, 1999) provided support for the model by showing that the impact of media news coverage of drunk driving on individuals’ drunk-driving behavior is mediated by relevant legislation actions.

The social diffusion model of campaign influence suggests that campaigns can ultimately affect people’s behaviors by generating interpersonal communication within social networks and, accordingly, by changing social norms. In other words, this model suggests that campaign-generated discussion can affect people’s normative perceptions and, ultimately, health behaviors. Yanovitzky and Stryker (2001) provided empirical evidence supporting this model by showing that news coverage of youth binge drinking affects the binge-drinking behavior of youth by changing their perception of the social acceptability of the behavior. Yanovitzky and Stryker, however, failed to test one of the key constructs of this model (i.e., interpersonal communication) in their study. Although they showed that increased news coverage of binge drinking affects the perceived peer approval of the behavior, they were not able to determine whether the source that causes changes in normative perceptions is interpersonal conversation. A test of the interpersonal communication component, however, is critical in the test of the social diffusion model, because the social diffusion model suggests that the impact of a campaign is mediated by a person’s interpersonal communication with others.

In this light, this study examines the role of interpersonal communication in the generation and diffusion of campaign effects. More specifically, this study tests whether campaign conversation can be another way that a campaign affects a person’s health perceptions or behaviors using secondary analyses of the Truth^SM campaign data. It tests whether campaign conversation, along with individual exposure, can convey the impact of the Truth^SM Campaign on individuals’ beliefs about smoking. By examining the role of campaign conversation in the diffusion of campaign effects, this article sheds light on how mass media health campaigns can influence individuals’ health perceptions and behaviors.

Rationale

Campaign Activity

Campaign activity can be defined as the net sum of campaign outputs generated by campaign producers. Campaign activity is different from campaign exposure in a sense that campaign activity refers to the total campaign outputs present in the media environment, whereas campaign exposure refers to the proportion of the outputs that impinged on audiences. In other words, campaign activity exists in the environment regardless of how much receptivity (i.e., exposure) there has been to the campaign. Thus, although campaign exposure is measured at the individual level, campaign activity is assessed at the campaign producers’ level. More specifically, campaign activity can be measured by obtaining information about whether and, if so, how broadly and frequently a campaign message is placed in the media environment. The extent to which campaign activity induces campaign exposure depends on campaign message placement, such as where campaign messages were placed, at what time, and with what frequency, and message features, such as message content or style (Southwell, 2005).

The difference between campaign activity and campaign exposure has generally been ignored in prior research; instead, campaign exposure has often been used as a proxy for campaign activity. However, overlooking the distinction between campaign activity and campaign exposure can lead to underestimation of campaign effects because exposure is not the only way campaign activity exerts its effects.

Campaign evaluation is often conducted by comparing campaign effects among individuals with varying levels of campaign exposure. However, focusing on the exposure-effect relationship may result in underestimation of campaign effects because those who were not exposed to a campaign can still be influenced by the campaign through other paths, such as conversation. If this is the case, a group of individuals who reported a lower level of exposure (IV) may be able to report high levels of engagement in healthy behaviors (DV), which would result in smaller differences in the outcome between higher and lower exposure groups, that is, underestimation of campaign effects. Distinguishing campaign activity from individual exposure can help us to avoid the potential underestimation of campaign effects by granting us an opportunity to think about other processes through which campaign activity affects audiences’ health perceptions or behaviors.

In fact, Hornik (Hornik, 2002; Hornik & Yanovitzky, 2003) suggested that campaign activity can influence a person’s health behavior not only through exposure but also through conversation. I will start with a brief description of the traditional individual exposure model and then turn to the social diffusion model.

The Individual Exposure Model

Traditionally, campaign effects have been believed as a consequence of individuals’ exposure to campaign messages (Hornik, 2002; Hornik & Yanovitzky, 2003). Exposure, one of the crucial concepts in understanding the process of mass media campaign effects, is referred to “the extent to which audience members have encountered specific messages or classes of messages/media content” (Slater, 2004). Exposure is typically assessed by self-reported measures which assess the availability of a message in the audience’s memory. In this sense, self-reported exposure measures encoded exposure (Southwell, Barmada, Hornik, & Maklan, 2002), that is, exposure that has left a minimal memory trace. Exposure is typically assessed by global self-reports, recall, or recognition measures (Slater, 2004).

The traditional individual exposure model suggests that campaign activities affect a person’s behavior via the person’s exposure to campaign messages (Hornik & Yanovitzky, 2003). Put differently, the model suggests that campaign activities induce a person’s exposure to campaign messages, which, in turn, leads to a behavior change. In addition, the model proposes that an individual’s exposure to campaign messages can induce behavioral changes through its impact on the individual’s attitudes, social norms, and efficacy based on the integrative model of behavior prediction (Fishbein, 2000; Fishbein & Yzer, 2003).

The Social Diffusion Model

Interpersonal communication can be another way campaign activity influences individuals’ perceptions or behaviors (Hornik, 2002; Rimal & Flora, 1998; Southwell & Yzer, 2007; Valente & Saba 1998, 2001). It has been suggested that campaigns affect audience members by generating relevant conversation. Also, campaign-generated interpersonal communication is believed to affect audiences via secondary diffusion and social influence processes (Hornik & Yanovitzky, 2003).

Conversation can affect people’s perceptions or behaviors by secondarily diffusing campaign messages, so that those who were not directly exposed to a campaign message can be indirectly exposed to the message while communicating with others who have been exposed to the message. Boulay, Storey, and Sood (2002) demonstrated the social diffusion function of interpersonal conversation using data collected for an evaluation of a mass media campaign implemented in Nepal to promote family planning. Boulay et al. (2002) found that 40% of those who were not directly exposed to the campaign were indirectly exposed to the message via interpersonal communication.

Conversation is also believed to affect individuals’ responses to a campaign by delivering relevant normative information in a social network (Hornik, 2002; Hornik & Yanovitzky, 2003; Lapinski & Rimal, 2005; Rimal & Real, 2005; Yanovitzky & Stryker, 2001). People’s conversation with others can affect their normative perceptions by helping them to understand the nature and strength of others’ beliefs and others’ expectations of them (e.g., whether others expect them to hold similar beliefs; Lapinski & Rimal, 2005; Real & Rimal, 2007; Rimal & Real, 2003).

The idea that interpersonal conversation can convey mass media campaign messages and relevant normative information dates back to Katz and Lazarsfeld’s (1955) two-step flow theory. According to the theory, mass media information flows through two steps: The first step is from mass media to opinion leaders, and the second is from opinion leaders to individuals. In this model, opinion leaders are believed to interpret and relay mass media information to follower individuals because they have greater access to mass media information. Thus, the two-step flow theory suggests that opinion leaders are intermediaries that connect mass media information to individuals, implying that conversations with opinion leaders can be a way mass media campaigns influence people’s health perceptions or behaviors.

Based on the secondary diffusion and social influence function of conversation, the social diffusion model posits that campaign activity can influence a person’s perception or behavior, not only because the person is exposed to the campaign but also because the person engages in campaign-related conversation. The model suggests that campaign activity generates relevant conversation, which, in turn, affects a person’s normative perceptions and, ultimately, behaviors (Hornik, 2002; Hornik & Yanovitzky, 2003). Therefore, the social diffusion model suggests that the impact of campaign activity on a person’s behavior is conveyed through the person’s social interaction or conversation with others.

By combining the two models (i.e., the individual exposure model and the social diffusion model), this study proposes that campaign activity influences individuals’ health perceptions through two processes: Campaign exposure and campaign conversation. In this study, the dual process model is tested in the context of the Truth^SM Campaign. The Truth^SM Campaign has been a major national antismoking campaign implemented in the United States to counter the influence of tobacco marketing targeting youths (Evans, Wasserman, Bertolotti, & Martino, 2002; Farrelly, Davis, Yarsevich, et al., 2002). Launched in 2000, the campaign has induced a significant decrease in smoking prevalence among its primary target audience: The 12- to 17-year-olds (Farrelly et al., 2005). This study tests whether the impact of the Truth^SM campaign activity on individuals’ beliefs about smoking was conveyed by individuals’ exposure to and the conversation about the campaign.

Hypotheses

The individual exposure model suggests that campaign activity affects individuals’ smoking beliefs through campaign exposure, whereas the social diffusion model suggests that campaign activity affects smoking beliefs through campaign conversation. Thus, this study hypothesizes that the Truth^SM campaign activity will have indirect effects on individuals’ smoking beliefs through exposure (Hypothesis 1) and conversation (Hypothesis 2).

Hypothesis 1 (H1): The Truth^SM campaign activity will influence the level of campaign exposure, which in turn will affect smoking beliefs.

Hypothesis 2 (H2): The Truth^SM campaign activity will affect the level of campaign-related interpersonal communication, which in turn will affect smoking beliefs.

In addition, campaign exposure and campaign conversation are believed to be nonindependent of each other. More specifically, campaign exposure is expected to generate campaign conversation. In their review of the role of interpersonal communication in mass media campaigns, Southwell and Yzer (2007) noted that conversation has been posited as an outcome of media campaigns in various campaign studies. In fact, previous research (Hindin et al., 1994; Mohammed, 2001; Rimal et al., 1999; Schuster et al., 2006; Shefner-Rogers & Sood, 2004; Sood & Nambiar, 2006) reported a positive relationship between individuals’ exposure to a campaign and their engagement in relevant conversation. Hindin et al.(1994) reported that those who were exposed to a family planning campaign in Ghana were more likely to talk to a partner or a service provider about using family planning. Mohammed also found that, of those who listened to Twende na Wakati, a Tanzanian entertainment-education program promoting family planning and HIV prevention, 81% discussed HIV prevention with others, whereas, of those who did not listen to the program, only 65% discussed HIV prevention. Based on consistent findings reported in past studies, this study hypothesizes that campaign exposure will be positively related to campaign conversation.

Hypothesis 3: Campaign exposure will be positively associated with campaign-generated interpersonal communication.

Figure 1 visualizes the posited relationships among campaign activity, campaign exposure, campaign conversation, and smoking beliefs. Although not presented in Figure 1, the following covariates were included in the model to control for their effects on smoking beliefs: Demographics (age, gender, and race), general television use, respondents’ smoking status, the extent of peers’ smoking, academic performance, family history of smoking-related death, and population density of each area. Past research on smoking shows that one’s smoking status can influence the person’s beliefs about the health effects of smoking. For example, Brownson et al. (1992) reported that smokers are less likely to perceive that smoking is harmful than nonsmokers. Also, smokers with quitting intentions and smokers without quitting intentions are expected to have different beliefs about smoking. Emery et al. (Emery, Gilpin, Ake, Farkas, & Pierce, 2000) have shown that hardcore smokers who are defined as heavy smokers with weak quitting histories and no quitting intentions have weaker antismoking beliefs compared to nonhardcore smokers. Smokers with and without quitting intentions are expected to show different smoking beliefs because they belong to different stages of change, that is, the contemplation stage and the precontemplation stage, respectively. According to the transtheoretical model (Prochaska & DiClemente, 1983), individuals in the precontemplation stage can be uninformed or underinformed about the consequences of their behavior (Prochaska, Redding, & Evers, 2008). Thus, smokers without quitting intentions can have weaker antismoking beliefs compared to smokers with quitting intentions. In addition, it has been found that friend smoking and poor academic achievement are related to a person’s transition to regular smoking (Tucker, Ellickson, & Klein, 2003). In addition, family history of heart disease has been found to be negatively associated with smoking when some attitudinal factors were controlled (Hunt, Davison, Emslie, & Ford, 2000). Finally, past research (Ennett, Flewelling, Lindrooth, & Norton, 1997) has shown that population density is related to low smoking rates among students when some school-level variables were controlled. In light of past literature, this study included the aforementioned covariates in the model.

Figure 1.

Campaign exposure and campaign conversation as dual processes of campaign effects

Method

Data

The hypotheses of this study were tested using the secondary analyses of the Legacy Media Tracking Survey (LMTS) data. LMTS, conducted by the American Legacy Foundation, is a random digit dialing telephone survey of a nationally representative sample of youth ranging from 12 to 24 years (Niederdeppe, Lindsey, Girlando, Ulasevich, & Farrelly, 2003). LMTS was primarily developed to monitor youths’ exposure to and reception of the Truth^SM Campaign. LMTS was conducted three times in total involving different sets of survey respondents at each time: Once prior to the launch of the Truth^SM Campaign and twice during the campaign. This study analyzed the second LMTS data, because it was collected during the first year of the campaign when the campaign was most active. The survey was conducted from September to December in 2000, collecting data from 10,692 youths. Response rate was 52.3% (Niederdeppe et al., 2003). More details about survey respondents are reported elsewhere (Farrelly, Davis, Yarsevich, et al., 2002; Niederdeppe et al., 2003).

Measures

Campaign activity was assessed by the Gross Rating Point (GRP) of the Truth^SM Campaign. GRP is a conventional unit used by advertising practitioners and researchers to measure the availability or intensity of a campaign in the media environment (Levy & Friend, 2000; Southwell, 2005; Southwell et al., 2002). More specifically, GRP is computed by multiplying a campaign’s reach (the percentage of a media’s target audience that is exposed to an ad message at least once during a given time) by its average frequency (the number of times a person or a household is exposed to an ad message). For example, if a campaign reaches 10% of its target audience 10 times, the GRP of this campaign becomes 100. In this study, the Truth^SM GRP was calculated by adding the Truth^SM GRPs of the first three quarters of 2000. This time span was the period that the Truth^SM Campaign was aired prior to the LMTS II survey. The GRP information on the campaign was obtained through staff reports (Jane Allen, American Legacy Foundation).

Campaign exposure was measured using respondents’ aided recall of the Truth^SM campaign messages. More specifically, respondents were given short descriptions of 16 Truth^SM ads and then were asked whether they had seen each ad. Respondents’ recall of each ad (either yes or no) was averaged across the 16 ads. Cronbach’s alpha coefficient for exposure across the 16 ads was .63. Although the reliability score was not high, this study decided to use the measure because the present study tries to assess exposure to various Truth^SM ads rather than exposure to a few well-known Truth^SM ads. With regard to the assessment of campaign conversation, those who answered affirmatively to each of the recall questions were asked whether they had talked to their friends about each of the ads. In other words, respondent’s campaign conversation was measured by whether they had talked to their friends about each of the 16 ads. Responses (either yes or no) were averaged across the ads.

Smoking beliefs were assessed using the following six items: “Young people who smoke cigarettes have more friends,” “Smoking makes people your age look attractive,” “Smoking cigarettes makes people your age look cool or fit in,” “Smoking is a way to show others that you’re not afraid to take risks,” “It is safe to smoke for only a year or two, as long as you quit after that,” and “Smoking cigarettes can help keep your weight down.” For each statement, respondents reported the extent to which they agreed with the statement on a four-point scale ranging from 1 = strongly agree to 4 = strongly disagree. The six items were averaged to measure smoking beliefs because the average of the six belief items was reliable (α = .74).

In addition, background variables, such as age, gender, race, respondents’ smoking status, the extent of peers’ smoking, academic performance, family history of smoking-related death, general television use, and population density were included in the model as covariates.

To assess a person’s smoking status, individuals were categorized into nonsmokers, smokers with intentions to quit smoking, and smokers without intentions to quit smoking. Respondents’ smoking status was measured by whether one had smoked cigarettes during the past 30 days and, if so, whether one wanted to completely stop smoking cigarettes. Those who did not smoke cigarettes during the past 30 days were considered as nonsmokers. Of those who smoked cigarettes during the past 30 days, those who responded that they wanted to completely stop smoking cigarettes were considered as smokers with quitting intentions, whereas those who had no such intentions were categorized as smokers without quitting intentions. Smoking status information was incorporated into the actual analyses using two dummy variables: WillQuit and Won’tQuit. Thus, the reference group in this coding scheme was nonsmokers. The extent of peers’ smoking was assessed by asking respondents to report how many of their four closest friends smoked. Academic performance was measured by respondents’ self-evaluation of how well they had done in school ranging from 1 (much worse than average) to 5 (much better than average). Family history of smoking-related death was measured by whether anyone in respondents’ family, including aunts, uncles, and grandparents, had died from smoking or smoking-related diseases. General television use was measured by asking respondents to report the average amount of hours they watched television during the past 7 days. Finally, with regard to population density, this study obtained the 2000 estimates of population density from DemographicsNow, an online reference source for U.S. demographic data available through many university libraries.

Multilevel Modeling

This study uses multilevel modeling, because it models multilevel relationships across two levels: The media environment level and the individual level. As displayed in Figure 1, campaign activity was measured at the media market level, whereas smoking beliefs were measured at the individual level. A unit of media market used in the analysis was designated market area (DMA), that is, a geographic area corresponding to the broadcast reach of local TV stations in a metropolitan area. There are a total of 210 DMAs in the United States and these nonoverlapping DMAs cover the entire United States (www.nielsenmedia.com).

Individuals who lived in the same DMA were correlated. The nonindependence among the individuals of the same DMA can make error terms correlated and, accordingly, make significance tests incorrect (Osborne, 2000). Multilevel modeling can account for nonindependence via simultaneous estimation of the Level 1 and Level 2 equations (for details, see Hayes, 2006; Raudenbush & Bryk, 2002). More specifically, in multilevel modeling, the Level 1 equation models a relationship between an outcome measure and Level 2 predictors, whereas the Level 2 equations model relationships between Level 1 parameters (i.e., intercept and slopes) and Level 2 predictors.

In the following model, the Level 1 equation models the relationship between campaign exposure and Level 1 predictors. More specifically, exposure is regressed on age, gender, race, smoking status, the extent of peers’ smoking, academic performance, family history of smoking-related death, and general television use. The first Level 2 equation models the relationship between Level 1 intercept and Level 2 predictors (i.e., the Truth^SM GRP and population density), whereas the rest of Level 2 equations predict each of Level 1 slopes without any Level 2 predictors.

Level 1 (individual level) model is as follows:

\begin{array}{l} {(C a m p a i g n E x p o s u r e)}_{i j} = β_{0 j} + β_{1 j} {(A g e)}_{i j} + β_{2 j} {(M a l e)}_{i j} + β_{3 j} {(B l a c k)}_{i j} \\ + β_{4 j} {(H i s p a n i c)}_{i j} + β_{5 j} {(W i l l Q u i t)}_{i j} \\ + β_{6 j} {(W o n ’ t Q u i t)}_{i j} + β_{7 j} (P e e r S m o k i n g) \\ + β_{8 j} (A c a d e m i c) + β_{9 j} (F a m i l y D i e) \\ + β_{10 j} (T V u s e) + e_{i j} \end{array}

Level 2 (DMA level) model is as follows:

\begin{array}{l} β_{o j} = γ_{00} + γ_{01} (G R P) + γ_{02} (P o p_D e n s i t y) + ε_{0 j} \\ β_{1 j} = γ_{00} + ε_{0 j} \\ ⋮ \\ β_{10 j} = γ_{100} + ε_{10 j} \end{array}

There was significant variation in smoking beliefs across DMAs, χ²(132) = 191.78, p < .01. The intraclass correlation of a fully unconditional model (i.e., a model without any predictors) was .01. This suggests that roughly 1% of the total variance in smoking beliefs lies between the Level 2 groups (DMAs). Although the intraclass correlation is relatively small, this study decided to use multilevel modeling because employing multilevel modeling provides benefits, such as avoiding inflated Type I error rates, even when the intraclass correlation is close to zero (Hayes, 2006; Kreft & De Leeuw, 1998). Also, scholars, such as Luke (2004), have noted that whether to use multilevel modeling or not is a theoretical decision. Luke argued that whenever a researcher deals with concepts that function and interact at multiple levels, using multilevel modeling is appropriate. Because this study models relationships between concepts that function at the media market level and the individual level, multilevel modeling seems to be a proper modeling method.

The use of multilevel modeling resulted in a minor loss of data. This study chose to limit DMA groups to those that had at least 10 participants in the data set, because a low number of individuals within a Level 2 group, that is, DMA, can contribute to a potential underestimation of standard errors (Harwell, Post, Cutler, & Maeda, 2005; Raudenbush & Bryk, 2002). As a result, 73 DMAs which had less than 10 survey participants were excluded from analyses. Thus, the final data set included 10,357 individuals from 133 DMAs. The highest number of individuals from a single DMA was 672 with a median of 28 individuals. The HLM software (version 6) was used for the multilevel modeling analyses. Level 1 predictors were group-mean centered, whereas Level 2 predictors were grand-mean centered. The full maximum likelihood method and unstructured covariance matrix were used for parameter estimations.

Tests of Indirect Effects

H1 and H2 hypothesize indirect effects of campaign activity. In order to test the indirect effects conveyed by exposure and conversation, this study used the asymmetric confidence limits approach. Recent multilevel mediation literature (Pituch, Stapleton, & Kang, 2006; Pituch, Whittaker, & Stapleton, 2005) advises researchers to use the asymmetric confidence limits method for a test of the indirect effect because the Sobel method, a popular method used for testing mediation among social scientists, is based on an unrealistic assumption of a normal distribution of an indirect effect.

In a mediation model, researchers estimate (a) the effect of the independent variable on the posited mediator (a) and (b) the effect of the mediator variable on the dependent variable (b) while controlling for the impact of the independent variable. Here, the size of the indirect effect is estimated by the product term, ab. The asymmetric confidence limits approach (Pituch et al., 2005, 2006) tests the presence of the indirect effect by computing confidence interval of the product term, ab. If the interval does not contain zero, the indirect effect is said to be present. The upper and lower limits of the interval are computed by the following equation: ab ± (C.V.)(σ _ab ). Here, σ _ab is the Sobel standard error, that is, $σ_{a b} = \sqrt{a^{2} σ_{b}^{2} + b^{2} σ_{a}^{2}}$ , and the C.V. (critical values) for the upper and lower limits can be obtained by converting a and b to z scores (z_a = a/σ_a, z_b = b/σ _b ) and then finding corresponding critical values using the tables in Meeker, Cornwell, and Aroian (1981; refer to Pituch et al., 2005, 2006, for more detailed information).

Descriptive Statistics

Participants ranged from 12 to 24 years, and the mean age of respondents was 16.96 (SD = 3.57), as presented in Table 1. Approximately 54% of the sample was female. Although approximately 49% of the sample was White, 17% of the sample was African American. Approximately 20% of the sample was Latino (or Latina) or Hispanic. A total of 81% and 4% of the respondents were nonsmokers and smokers without quitting intentions, respectively. Respondents reported that about one of their four closest friends smoked (M = 1.12, SD = 1.41). Self-rated academic performance (M = 3.73, SD = .82) was slightly higher than the mid-point (3). A total of 36% of the sample reported that their family had died from smoking or smoking-related diseases. The average level of TV viewing was 3.01 (SD = 2.91) hours. The average level of exposure to the Truth^SM Campaign was 0.18 (SD = 0.15). This means, on average, a particular Truth^SM ad was viewed by 18% of the respondents. In addition, the average level of conversation about the campaign was 0.21 (SD = 0.33). Mean smoking belief was 3.28 (SD = 0.43), indicating that respondents had moderate antismoking beliefs.

Table 1.

Means and Standard Deviations of Level 1 and Level 2 Variables

Variable	Overall (n = 10,357)*	Nonsmokers (n = 8,379)	Smokers intending to quit (n = 1,433)	Smokers not intending to quit (n = 463)
Level 1 variables
Age	16.96 (3.57)	16.44 (3.50)	19.07 (3.07)	19.40 (2.88)
Gender
Female	54%	55%	52%	47%
Male	46%	45%	48%	53%
Race
Black	17%	18%	13%	9%
Hispanic	20%	21%	18%	16%
Peer smoking	1.12 (1.41)	0.78 (1.18)	2.53 (1.39)	2.70 (1.36)
Academic performance	3.73 (0.82)	3.79 (0.80)	3.46 (0.83)	3.53 (0.93)
Family history	0.36 (0.48)	0.35 (0.48)	0.40 (0.49)	0.31 (0.46)
TV hour	3.01 (2.91)	3.06 (2.92)	2.74 (2.78)	2.90 (3.11)
Exposure	0.18 (0.15)	0.18 (0.15)	0.18 (0.15)	0.17 (0.15)
Conversation	0.21 (0.33)	0.21 (0.34)	0.20 (0.33)	0.16 (0.30)
Smoking belief	3.28 (0.43)	3.33 (0.43)	3.12 (0.40)	3.01 (0.44)
Level 2 variables
Truth^SM GRP	4936.15 (2056.43)	4991.70 (2077.93)	4689.51 (1971.99)	4758.68 (1853.82)
Population density	354.47 (402.31)	362.88 (407.55)	322.34 (386.22)	312.35 (355.81)

Note: Sample sizes of the three smoking status groups do not add up to the total sample size because there were some smoker respondents who did not answer to the question that measured quitting intentions.

With regard to the Level 2 variables, the mean Truth^SM GRPs across individuals and across DMAs were 4,936 (SD = 2,056) and 3,526 (SD = 1,962), respectively. In addition, the mean population densities across individuals and across DMAs were 354 (SD = 402) and 183 (SD = 245), respectively.

Results

Predictors of Exposure, Conversation, and Smoking Beliefs

Table 2 presents that campaign exposure was significantly related to age, gender, race, peer smoking, family history of smoking-related death, general television use, and the Truth^SM GRP. More specifically, younger respondents, γ = −.06, t(132) = −5.52, p < .001; males, γ = .12, t(132) = 12.39, p < .001; African American respondents, γ = .05, t(132) = 2.77, p < .01; and those who reported higher levels of peer smoking, γ = .02, t(132) = 2.05, p < .05; respondents who had a family history of smoking-related death, γ = .05, t(132) = 4.07, p < .001; and respondents who reported a high level of television viewing, γ = .09, t(132) = 7.82, p < .001 were more likely to be exposed to the Truth^SM Campaign. More importantly, those who lived in an area with stronger Truth^SM campaign activities reported higher campaign exposure, γ = .06, t(130) = 3.14, p < .01. R² of this model was .06.

Table 2.

Multilevel Modeling Results Predicting Exposure, Conversation, and Smoking Beliefs

	Coefficient (SE)
Outcome variables	Exposure	Conversation	Smoking beliefs
Fixed effects
Level 1 (n = 10,357)
Intercept	−0.09 (.02)***	−0.02 (.01)	0.03 (.02)
Age	−0.06 (.01)***	−0.02 (.01)	0.06 (.01)***
Male	0.12 (.01)***	−0.04 (.01)***	−0.10 (.01)***
Black	0.05 (.02)**	0.10 (.02)***	−0.03 (.01)
Hispanic	0.002 (.01)	0.05 (.02)**	−0.04 (.01)**
Smoker with quitting intention	0.01 (.01)	−0.01 (.01)	−0.11 (.01)***
Smoker without quitting intention	−0.004 (.01)	−0.03 (.01)*	−0.12 (.01)***
Peer smoking	0.02 (.01)*	0.04 (.01)**	−0.14 (.01)***
Academic performance	0.02 (.01)	0.06 (.01)***	0.09 (.01)***
Family history	0.05 (.01)***	0.06 (.01)***	0.02 (.01)
TV use	0.09 (.01)***	−0.01 (.01)	−0.03(.01)**
Exposure			0.07 (.01)***
Conversation			0.03 (.01)**
Level 2 (133 groups)
Truth^SM GRP	0.06 (.02)**	0.04 (.01)**	0.01 (.02)
Population density	0.03 (.02)	−0.02 (.005)**	−0.02 (.01)
Random effects
τ̂ _intercept	0.02***	0.001	0.01***
τ̂ _age	0.002	0.003	0.002
τ̂ _male	0.001	0.001	0.002
τ̂ _black	0.01**	0.01	0.002
τ̂ _hispanic	0.001	0.004	0.002
τ̂ _willquit	0.001	0.002	0.004
τ̂ _won’tquit	0.002	0.0003	0.002
τ̂ _peersmoke	0.001	0.001	0.002
τ̂ _academic	0.002	0.002	0.003
τ̂ _familydie	0.002	0.001	0.001
τ̂ _TVuse	0.002	0.002	0.001
τ̂ _exposure			0.005
τ̂ _conversation			0.001
σ̂²	0.92	0.97	0.87

Note: Cell entries are multilevel modeling coefficients with standard errors in parentheses. In order to obtain standardized coefficients, values for all variables were standardized using z scores and then were used for the multilevel analyses.

p < .05. **p < .01. ***p < .001.

Campaign conversation was predicted by gender, race, smoking status, peer smoking, academic performance, family history, the Truth^SM GRP, and population density. Females, γ = −.04, t(132) = −3.91, p < .001; African Americans, γ = .10, t(132) = 5.60, p < .001; Hispanics, γ = .05, t(132) = 3.05, p < .01; nonsmokers (vs. smokers without quitting intentions), γ = −.03, t(132) = −2.40, p < .05; and those who reported higher levels of peer smoking, γ = .04, t(132) = 3.12, p < .01; higher academic performance, γ = .06, t(132) = 4.62, p < .001; having a family history of smoking-related death, γ = .06, t(132) = 5.86, p < .001; and those who lived in an area with lower population density, γ = −.02, t(130) = −3.34, p < .01, were more likely to talk about the campaign. More importantly, Truth^SM GRP was a positive predictor of campaign conversation, γ = .04, t(130) = 3.37, p < .01. Those who lived in an area with higher Truth^SM GRPs talked more about the campaign. R² of this model was .03.

A person’s smoking beliefs were associated with age, gender, race, smoking status, peer smoking, academic performance, general television use, exposure, and conversation. Older respondents, γ = .06, t(132) = 4.99, p < .001, and respondents who reported higher academic performance, γ = .09, t(132) = 7.67, p < .001 had stronger antismoking beliefs. On the other hand, males, γ = −.10, t(132) = −10.21, p < .001; Hispanics, γ = −.04, t(132) = −3.32, p < .01; smoker respondents (smokers with quitting intentions: γ = −.11, t(132) = −8.33, p < .001; smokers without quitting intentions: γ = −.12, t(132) = −11.16, p < .001); respondents who reported higher levels of peer smoking, γ = −.14, t(132) = −10.53, p < .001; and respondents who reported greater television use, γ = −.03, t(132) = −2.98, p < .01, had weaker antismoking beliefs. More importantly, respondents who reported higher levels of campaign exposure, γ = .07, t(132) = 4.91, p < .001, and campaign conversation, γ = .03, t(132) = 2.97, p < .01, had stronger antismoking beliefs. However, the Truth^SM GRP was not significantly associated with smoking beliefs, γ = .01, t(130) = 0.84, p = ns. R² of this model was .12.

Hypotheses Testing

H1 predicted that the Truth^SM GRP would have indirect effects on individuals’ smoking beliefs through campaign exposure. As noted above, the Truth^SM GRP was positively associated with exposure (a = 0.06, σ _a = 0.02), when the effects of Level 1 covariates (e.g., age, gender) and Level 2 covariate (i.e., population density) were controlled. In addition, as mentioned earlier, campaign exposure was positively associated with antismoking beliefs (b = 0.07, σ _b = 0.01), when the effects of the Truth^SM GRP, campaign conversation, and covariates (e.g., age, gender, population density) were controlled. Campaign conversation was statistically controlled here to capture the unique effects of campaign exposure that cannot be explained by campaign conversation. To test the presence of the indirect effect, a 95% confidence interval of the product term, ab, was computed. The obtained confidence interval (0.001 to 0.007) shows that the indirect effect through campaign exposure exists because the confidence interval does not contain zero. The existence of the indirect effect suggests that the Truth^SM GRP influenced campaign exposure, which in turn affected smoking beliefs. Thus, H1 was supported.

H2 predicted that the Truth^SM GRP would have indirect effects on smoking beliefs through campaign conversation. As mentioned earlier, there was a positive relationship between the Truth^SM GRP and campaign conversation (a = 0.04, σ _a = 0.01), when the effects of covariates (e.g., age, gender, population density) were controlled. In addition, there was a positive relationship between campaign conversation and smoking beliefs (b = 0.03, σ _b = 0.01), when the effects of the Truth^SM GRP, campaign exposure, and covariates were controlled. Campaign exposure was statistically controlled here to capture the unique effects of campaign conversation that cannot be explained by campaign exposure. A 95% confidence interval of the ab product (0.0004 to 0.002) indicates presence of the indirect effect through campaign conversation. It shows that the Truth^SM GRP affected campaign conversation, which in turn affected smoking beliefs. Thus, H2 was supported.

Finally, H3 predicted that campaign exposure would be positively associated with campaign conversation. Results show that there was a positive relationship between campaign exposure and campaign conversation. Campaign exposure was a positive predictor of campaign conversation, γ = .14, t (132) = 11.19, p < .001, after controlling for the effects of covariates (e.g., age, gender, the Truth^SM GRP, and population density). Thus, H3 was supported.

Discussion

This study provides support for both of the individual exposure model and the social diffusion model. A simultaneous test of both paths shows that each of the two posited processes, that is, campaign exposure and campaign conversation, conveyed the impact of campaign activity on smoking beliefs. The results suggest that campaign effects occur through a person’s direct exposure to a campaign as well as a person’s engagement in campaign-related conversation with others. Thus, this study suggests the utility of the individual exposure model and the social diffusion model by showing that both campaign exposure and campaign conversation can explain the unique variance of campaign effects.

This research provides an empirical test of the interpersonal communication component in the social diffusion model. This study confirms the role of interpersonal communication in the social diffusion model by showing that campaign activity influenced campaign conversation, which in turn affected smoking beliefs. By providing an empirical test of the interpersonal communication component of the model, this research is expected to contribute to the theoretical development of the social diffusion model.

In addition, this study shows that there is a positive relationship between campaign exposure and campaign conversation. Southwell and Yzer’s (2007) review of the processes by which campaigns generate relevant conversation can be particularly helpful in understanding the underlying processes of the relationship. According to Miller (1986), exposure to campaigns can generate pertinent conversation because exposure provides the content of conversation. Also, from a behavioral perspective, campaign exposure can induce conversation because it can change people’s attitudes toward, norms about, and efficacy regarding talking about a particular issue. Southwell and Yzer argued that campaigns can affect individuals’ engagement in conversation behavior by affecting individuals’ overall views regarding talking about an issue, individuals’ perception regarding others’ approval or disapproval in talking about the issue, or individuals’ perceived ability to talk about the issue, based on Fishbein’s integrative model of behavior prediction (Fishbein, 2000; Fishbein & Yzer, 2003).

The role of campaign conversation in the generation and diffusion of campaign effects as shown in the present study has important implications for campaign evaluation. By showing campaign effects can occur not only through exposure but also through conversation, the results of this study suggest that campaign evaluators need to assess not only exposure-based campaign effects but also conversation-based campaign effects, because failing to capture conversation-based campaign effects can result in underestimation of campaign effects.

One of the viable strategies that researchers can use to capture the full impact of a campaign regardless of whether it occurs by direct exposure or campaign conversation is to compare areas with different levels of campaign activity (e.g., campaign vs. no campaign or high vs. low GRPs). Using campaign activity information for campaign evaluation is not new in campaign literature (e.g., Palmgreen, Donohew, Lorch, Hoyle, & Stephenson, 2002; Worden & Flynn, 2002). For example, Worden and Flynn tested the effect of a mass media antismoking campaign by manipulating campaign activity at a community level. More specifically, the researchers selected two matched pairs of communities and then randomly assigned one community within each matched pair to the media-school intervention (a mass media intervention combined with a school program) and the other to the school-only intervention.

This study used a slightly different measure of campaign activity, that is, a campaign GRP. Although a dichotomized existence measure of campaign activity requires a manipulation of campaign activity by campaign practitioners (e.g., campaign vs. no campaign), a GRP measure of campaign activity exploits naturally existing variation in campaign activity across areas (e.g., high vs. low GRPs). Thus, using the GRP measure for a test of campaign effects can save time and efforts. In addition, researchers can assess a dose-response relationship with the GRP measure, because the GRP measure is a continuous measure unlike the dichotomized existence measure. However, using the GRP measure has a problem of potential confounding because, here, campaign activity is not randomly assigned. Thus, more research on how to address the issue of potential confounding (e.g., statistical control) is warranted for the further development of the GRP measure of campaign activity.

The important role that campaign conversation plays in the diffusion of campaign effects has implications for campaign design and planning as well. This research suggests that campaign effects can be amplified by facilitating campaign conversation. Thus, campaign planners need to find a way to spur and generate campaign-related conversation when they design campaign messages. Although little research has been conducted on message characteristics that can spur conversation, there are some findings that can be used to design campaign messages. For example, Phelps, Lewis, Mobilio, Perry, and Raman (2004) reported that email advertising messages that evoke strong emotions are more likely to be forwarded to others. This type of research findings can be incorporated to create more effective campaign messages.

The strategic use of opinion leaders as behavior change endorsers employed by campaign practitioners and researchers (e.g., Kelly, 2004; Kelly et al., 1991, 1992) is worthwhile to note in this sense. The impact of campaign conversation can be maximized when a campaign can encourage opinion leaders to talk about relevant issues. Opinion leaders are believed to be powerful in delivering campaign messages and inducing relevant behavior changes. Atkin (2001), for example, argued that conversations with opinion leaders will be more powerful in inducing behavioral changes among focal audiences than the mass media mediated campaign messages. This is because opinion leaders can provide tailored messages to focal individuals, offer positive and negative reinforcement, and serve as role models. In fact, Kelly actually used opinion leaders as behavior change endorsers for an HIV risk-behavior reduction intervention. Kelly and colleagues recruited and trained popular opinion leaders in the gay community so that the leaders could disseminate the risk reduction recommendation messages to members of their social network. This intervention resulted in a significant reduction in unsafe sexual behavior among the focal audience. Thus, a campaign would be able to enlist the power of opinion leaders by encouraging conversations with opinion leaders.

Despite its efforts to examine the roles of individual exposure and campaign conversation in the generation and diffusion of campaign effects, this study does not provide a complete picture of the processes whereby mass media health campaigns influence individuals’ health perceptions and behaviors. This is because it did not fully examine (a) the potential interaction between individual exposure and campaign conversation and (b) the possibility of other routes of campaign effects.

This study did not explore the potential interaction between individual exposure and campaign conversation. Although the study reported a positive association between campaign exposure and campaign conversation, the relationship between campaign exposure and campaign conversation may be more complicated. For example, campaign conversation could moderate or mediate the impact of campaign exposure (Southwell & Yzer, 2007). Further examination of the relationship between campaign exposure and campaign conversation may reveal a complex nature of the processes of campaign effects and, accordingly, advance our understanding of the processes of campaign effects. Thus, future research is warranted to examine the potentially interesting relationship between the two.

This study proposes campaign exposure and campaign conversation as the dual processes through which a campaign exerts its effects. However, there may exist other important paths that were not examined in this research. For example, Hornik (Hornik, 2002; Hornik & Yanovitzky, 2003) suggested that institutional diffusion is one of the ways that a campaign affects a person’s health perception or behavior. Without a complete examination of other potentially powerful mediators, a comprehensive view of the processes of campaign effects may not be achieved. Thus, future studies are needed to explore and identify important mediators that were not examined in this article.

One of the primary limitations of this study is that conversation was not measured independently from exposure. This is because respondents received the conversation question only when they reported that they watched the ad. In other words, respondents who had not seen the ad were not asked the discussion question. The nonindependence between the two measures makes it difficult to assess the unique effects of conversation, because those who reported having conversation already engaged in exposure as well. The nonindependence issue is particularly problematic when assessing the effects of conversation occurring through the secondary diffusion process. Although it is possible that people who were not directly exposed to a message can be indirectly exposed to the message through conversation and, accordingly, be influenced by the message, such effects of conversation could not be captured well in this study because the conversation measure of this study assumes prior direct exposure to the campaign. In order to address this nonindependence issue, this study assigned a missing value, instead of zero, with regard to the conversation variable for those who did not receive the conversation question. Also, this study tried to tease out the effects of conversation from those of exposure by statistically controlling for the effects of exposure. Using the statistical control, this study was able to capture at least some effects of conversation (possibly occurring through other processes, such as the social influence process), although it may not have properly detected conversation effects occurring via secondary diffusion. Future research would be able to avoid the problem of nonindependence by measuring conversation independently from exposure.

In addition, the relationship reported in this study is based on cross-sectional survey data and, thus, there is a possibility of reverse causation. More specifically, it is possible that those who had greater antismoking beliefs were more likely to be exposed to the campaign and more likely to engage in campaign-related conversation. In this case, we cannot argue that exposure and conversation induced greater antismoking beliefs because here causality works in the opposite direction. Thus, future studies that employ longitudinal data or experimental data are needed to address the issue of potential reverse causality.

A small effect size is another limitation of this study. The effects of interest in the study were indirect effects of campaign activity conveyed through campaign exposure and campaign conversation. As reported in the results section, the size of indirect effects was quite small. However, the small effect size of the present study does not seem to result from the inadequacy of the model of the study. Instead, the small size of indirect effects seems to derive from the small size of total effects of mass media campaigns. According to Snyder and colleagues’ (2004) meta-analysis, the average effect size of mass-mediated antismoking campaigns conducted in the United States is .05, which means on average antismoking campaigns can explain approximately 0.3% of the total variance of a person’s behavior change. Although the meta-analysis did not directly examine effects on individuals’ beliefs about smoking, we can anticipate that the effects on beliefs would not be large. Given the small total effect size, an indirect effect, which is a component of a total effect along with a direct effect, cannot be large. Thus, the small effect size of the present study does not seem to undermine the findings of the study.

The size of indirect effects conveyed by campaign conversation was particularly small compared to the size of indirect effects conveyed through campaign exposure. The small effect size may have been due to the aforementioned nonindependence issue of conversation measurement. Because the conversation question was asked only when respondents reported that they watched the ad, conversation effects that occur without exposure (e.g., the secondary diffusion effects of conversation) were not able to be captured. Therefore, conversation effects may have been underestimated in this study. Future studies that measure the effects of conversation independently from those of exposure may find larger effects of conversation.

Also, the broad measure of smoking beliefs used in this study is another limitation of the study. Future studies can employ a more theoretical measure of smoking beliefs using theories such as the integrative model of behavioral prediction (Fishbein & Yzer, 2003). Finally, because this study is based on a single set of campaign data, the generalizability of the results to other antitobacco campaigns (or to more general mass media health campaigns) may be questionable. Thus, replications in the context of other antitobacco campaigns are strongly encouraged.

Despite these limitations, this study contributes to communication and campaign research by advancing the current understanding of the social diffusion process of mass media health campaign effects and the role of campaign conversation in this process.

Footnotes

Acknowledgements

The author wishes to thank Brian Southwell, Marco Yzer, Ronald Faber, Alexander J. Rothman, and the anonymous reviewers for their helpful comments.

The author declared no potential conflicts of interest with respect to the authorship and/or publication of this article.

The author received no financial support for the research and/or authorship of this article.

References

Atkin

C. K.

(2001). Theory and principles of media health campaigns. In Rice

R. E.

Atkin

C. K.

(Eds.), Public communication campaigns (3rd ed., pp. 49-68). Thousand Oaks, CA: SAGE.

Biglan

Ary

D. V.

Smolkowski

Duncan

Black

(2000). A randomized controlled trial of a community intervention to prevent adolescent tobacco use. Tobacco Control, 9, 24-32.

Boulay

Storey

J. D.

Sood

(2002). Indirect exposure to a family planning mass media campaign in Nepal. Journal of Health Communication, 7, 379-399.

Brownson

R. C.

Jackson-Thompson

Wilkerson

J. C.

Davis

J. R.

Owens

N. W.

Fisher

E. B.

(1992). Demographic and socioeconomic differences in beliefs about the health effects of smoking. American Journal of Public Health, 82, 99-103.

COMMIT Research Group. (1995). Community Intervention Trial for Smoking Cessation (COMMIT): I. Cohort results from a four-year community intervention. American Journal of Public Health, 85, 183-192.

Emery

Gilpin

E. A.

Ake

Farkas

A. J.

Pierce

J. P.

(2000). Characterizing and identifying “hard-core” smokers: Implications for further reducing smoking prevalence. American Journal of Public Health, 90, 387-394.

Ennett

S. T.

Flewelling

R. L.

Lindrooth

R. C.

Norton

E. C.

(1997). School and neighborhood characteristics associated with school rates of alcohol, cigarette, and marijuana use. Journal of Health and Social Behavior, 38, 55-71.

Evans

W. D.

Wasserman

Bertolotti

Martino

(2002). Branding behavior: The strategy behind the Truth^sm campaign. Social Marketing Quarterly, 8, 17-29.

Farrelly

M. C.

Davis

K. C.

Haviland

M. L.

Messeri

Healton

C. G.

(2005). Evidence of a dose-response relationship between “truth” antismoking ads and youth smoking prevalence. American Journal of Public Health, 95, 425-431.

10.

Farrelly

M. C.

Davis

K. C.

Yarsevich

J. M.

Haviland

M. L.

Hersey

J. C.

Girlando

M. E.

Healton

C. G.

(2002). Getting to the truth: Assessing youths’ reactions to the truth and “think. don’t smoke” tobacco countermarketing campaigns (First look report 9). Washington, DC: American Legacy Foundation.

11.

Farrelly

M. C.

Healton

C. G.

Davis

K. C.

Messeri

Hersey

J. C.

Haviland

M. L.

(2002). Getting to the truth: Evaluating national tobacco countermarketing campaigns. American Journal of Public Health, 92, 901-907.

12.

Fishbein

(2000). The role of theory in HIV prevention. AIDS Care, 12, 273-278.

13.

Fishbein

Yzer

M. C.

(2003). Using theory to develop effective health behavior interventions. Communication Theory, 13, 164-183.

14.

Flay

B. R.

(1987). Mass media and smoking cessation: A critical review. American Journal of Public Health, 77, 153-160.

15.

Flynn

B. S.

Worden

J. K.

Secker-Walker

R. H.

Pirie

B. P. L.

Badger

G. J.

Carpenter

J. H.

Geller

B. M.

(1994). Mass media and school interventions for cigarette smoking prevention: Effects 2 years after completion. American Journal of Public Health, 84, 1148-1150.

16.

Harwell

Post

Cutler

Maeda

(2005). A comparative study of standards-based and traditional high school graduates’ achievements in their first year of university mathematics work. Paper presented at the meeting of the American Educational Research Association, Montreal, Quebec, Canada.

17.

Hayes

A. F.

(2006). A primer on multilevel modeling. Human Communication Research, 32, 385-410.

18.

Hindin

M. J.

Kincaid

D. L.

Kumah

O. M.

Morgan

Kim

Y. M.

Ofori

J. K.

(1994). Gender differences in media exposure and action during a family planning campaign in Ghana. Health Communication, 6, 117-135.

19.

Hornik

(2002). Public health communication: Making sense of the evidence. In Hornik

(Ed.), Public health communication: Evidence for behavior change (pp. 1-19). Mahwah, NJ: Lawrence Erlbaum.

20.

Hornik

Yanovitzky

(2003). Using theory to design evaluations of communication campaigns: The case of the National Youth Anti-Drug Media Campaign. Communication Theory, 13, 204-224.

21.

Hunt

Davison

Emslie

Ford

(2000). Are perceptions of a family history of heart disease related to health-related attitudes and behaviours? Health Education Research: Theory, and Practice, 15, 131-143.

22.

Katz

Lazarsfeld

P. F.

(1955). Personal influence. Glencoe, IL: Free Press.

23.

Kelly

J. A.

(2004). Popular opinion leaders and HIV prevention peer education: Resolving discrepant findings, and implications for the development of effective community programmes. AIDS Care, 16, 139-150.

24.

Kelly

J. A.

St. Lawrence

J. S.

Diaz

Y. E.

Stevenson

L. Y.

Hauth

A. C.

Brasfield

T. L.

. . . Andrew

M. E.

(1991). HIV risk behavior reduction following intervention with key opinion leaders of population: An experimental analysis. American Journal of Public Health, 81, 168-171.

25.

Kelly

J. A.

St. Lawrence

J. S.

Stevenson

L. Y.

Hauth

A. C.

Kalichman

S. C.

Diaz

Y. E.

. . . Morgan

M. G.

(1992). Community AIDS/HIV risk reduction: The effects of endorsements by popular people in three cities. American Journal of Public Health, 82, 1483-1489.

26.

Kreft

De Leeuw

(1998). Introducing multilevel modeling. Thousand Oaks, CA: SAGE.

27.

Lapinski

M. K.

Rimal

R. N.

(2005). An explication of social norms. Communication Theory, 15, 127-147.

28.

Levy

D. T.

Friend

(2000). Gauging the effects of mass media policies: What do we need to know? Journal of Public Health Management Practice, 6(3), 95-106.

29.

Luke

D. A.

(2004). Multilevel modeling. Thousand Oaks, CA: SAGE.

30.

Meeker

W. Q.

Cornwell

L. W.

Aroian

L. A.

(1981). Selected tables in mathematical statistics (Vol. 7). Providence, RI: American Mathematical Society.

31.

Miller

G. R.

(1986). A neglected connection: Mass media exposure and interpersonal communicative competency. Gumpert

Cathcart

(Eds.), Inter/media: Interpersonal communication in a media world (pp. 132-139). New York, NY: Oxford University Press.

32.

Mohammed

(2001). Personal communication networks and the effects of an entertainment-education radio soap opera in Tanzania. Journal of Health Communication, 6, 137-154.

33.

Niederdeppe

Lindsey

Girlando

Ulasevich

Farrelly

(2003). Exposure to pro-tobacco messages among teens and young adults (First Look Report 12). American Legacy Foundation, Washington, DC.

34.

Noar

S. M.

(2006). A 10-year retrospective of research in health mass media campaigns: Where do we go from here? Journal of Health Communication, 11, 21-42.

35.

Osborne

J. W.

(2000). Advantages of hierarchical linear modeling. Practical Assessment, Research & Evaluation, 7. Retrieved from http://PAREonline.net/getvn.asp?v=7&n=1

36.

Palmgreen

Donohew

Lorch

E. P.

Hoyle

R. H.

Stephenson

M. T.

(2002). Television campaigns and sensation seeking targeting of adolescent marijuana use: Controlled time-series approach. In Hornik

(Ed.) Public health communication: Evidence for behavior change (pp. 35-56). Mahwah, NJ: Lawrence Erlbaum.

37.

Phelps

J. E.

Lewis

Mobilio

Perry

Raman

(2004). Viral marketing or electronic word-of-mouth advertising: Examining consumer responses and motivations to pass along email. Journal of Advertising Research, 44, 333-348.

38.

Pituch

K. A.

Stapleton

L. M.

Kang

J. Y.

(2006). A comparison of single sample and bootstrap methods to assess mediation in cluster randomized trials. Multivariate Behavioral Research, 41, 67-400.

39.

Pituch

K. A.

Whittaker

T. A.

Stapleton

L. M.

(2005). A comparison of methods to test for mediation in multisite experiments. Multivariate Behavioral Research, 40, 1-23.

40.

Porto

M. P.

(2007). Fighting AIDS among adolescent women: Effects of a public communication campaign in Brazil. Journal of Health Communication, 12, 121-132.

41.

Prochaska

J. O.

DiClemente

C. C.

(1983). Stages and processes of self-change of smoking: Toward an integrative model of change. Journal of Consulting and Clinical Psychology, 51, 390-395.

42.

Prochaska

J. O.

Redding

C. A.

Evers

K. E.

(2008). The transtheoretical model and stages of change. In Glanz

Rimer

B. K.

Viswanath

(Eds.), Health behavior and health education: Theory, research, and practice (pp. 97-121). San Francisco, CA: Jossey-Bass.

43.

Raudenbush

S. W.

Bryk

A. S.

(2002). Hierarchical linear models. Applications and data analysis models (2nd ed.). Thousand Oaks, CA: SAGE.

44.

Real

Rimal

R. N.

(2007). Friends talk to friends about drinking: Exploring the role of peer communication in the theory of normative social behavior. Health Communication, 22, 169-180.

45.

Rimal

R. N.

Flora

J. A.

(1998). Bidirectional familiar influences in dietary behavior: Test of a model of campaign influences. Human Communication Research, 24, 610-637.

46.

Rimal

R. N.

Flora

J. A.

Schooler

(1999). Achieving improvements in overall health orientation. Communication Research, 26, 322-348.

47.

Rimal

R. N.

Real

(2003). Explaining the relationship between norms and behaviors: Implications for campaigns against alcohol abuse by college students. Communication Theory, 13, 184-203.

48.

Rimal

R. N.

Real

(2005). How behaviors are influenced by perceived norms: A test of the theory of normative social behavior. Communication Research, 32, 389-414.

49.

Schuster

D. V.

Valente

T. W.

Skara

S. N.

Wenten

M. R.

Unger

J. B.

Cruz

T. B.

Rohrbach

L. A.

(2006). Intermedia processes in the adoption of tobacco control activities among opinion leaders in California. Communication Theory, 16, 91-117.

50.

Shefner-Rogers

C. L.

Sood

(2004). Involving husbands in safe motherhood: Effects of the SUAMI SIAGA campaign in Indonesia. Journal of Health Communication, 9, 233-258.

51.

Slater

M. D.

(2004). Operationalizing and analyzing exposure: The foundation of media effects research. Journalism and Mass Communication Quarterly, 81, 168-183.

52.

Snyder

L. B.

Hamilton

M. A.

Mitchell

E. W.

Kiwanuka-Tondo

Fleming-Milici

Proctor

(2004). A meta-analysis of the effect of mediated health communication campaigns on behavior change in the United States. Journal of Health Communication, 9, 71-96.

53.

Sood

Nambiar

(2006). Comparative cost-effectiveness of the components of a behavior change communication campaign on HIV/AIDS in North India. Journal of Health Communication, 11, 143-162.

54.

Southwell

B. G.

(2005). Between messages and people: A multilevel model of memory for television content. Communication Research, 32, 112-140.

55.

Southwell

B. G.

Barmada

C. H.

Hornik

R. C.

Maklan

D. M.

(2002). Can we measure encoded exposure? Validation evidence from a national campaign. Journal of Health Communication, 7, 445-453.

56.

Southwell

B. G.

Yzer

M. C.

(2007). The roles of interpersonal communication in mass media campaigns. In Beck

(Ed.), Communication Yearbook 31 (pp. 420-462). New York, NY: Lawrence Erlbaum.

57.

Tucker

J. S.

Ellickson

P. L.

KIein

D. J.

(2003). Predictors of the transition to regular smoking during adolescence and young adulthood. Journal of Adolescent Health, 32, 314-324.

58.

Valente

T. W.

Saba

W. P.

(1998). Mass media and interpersonal influence in a reproductive health communication campaign in Bolivia. Communication Research, 25, 96-124.

59.

Valente

T. W.

Saba

W. P.

(2001). Campaign exposure and interpersonal communication as factors in contraceptive use in Bolivia. Journal of Health Communication, 6, 303-322.

60.

Worden

J. K.

Flynn

B. S.

(2002). Using mass media to prevent cigarette smoking. In Hornik

(Ed.) Public health communication: Evidence for behavior change (pp. 23-34). Mahwah, NJ: Lawrence Erlbaum.

61.

Yanovitzky

(2002). Effects of news coverage on policy attention and actions: A closer look into the media-policy connection. Communication Research, 29, 422-451.

62.

Yanovitzky

Bennett

(1999). Media attention, institutional response, and health behavior change: The case of drunk driving, 1978-1996. Communication Research, 26, 429-453.

63.

Yanovitzky

Stryker

(2001). Mass media, social norms, and health promotion efforts: A longitudinal study of media effects on youth binge drinking. Communication Research, 28, 208-239.