Employing Interpersonal Influence to Promote Multivitamin Use

Running head: INTERPERSONAL INFLUENCE Employing Interpersonal Influence to Promote Multivitamin Use Franklin J. Boster Christopher J. Carpenter Kyle R. Andrews Paul A. Mongeau Abstract Boster et al. (in press) proposed that superdiffusers are well connected, persuasive, and a maven in a content area. They proposed that superdiffusers, if recruited, could promote the adoption of health practices. In this manuscript a model of this process is presented, and an intervention designed to test the efficacy of this influence strategy is introduced. Specifically, superdiffusers were recruited to persuade their peers to take a daily multivitamin. Evidence was found consistent with the intervention’s effectiveness. Employing Interpersonal Influence to Promote Multivitamin Use Scholars examining the diffusion of information and behavior have long examined the influence of media and interpersonal message channels. Recent research on communication campaigns has highlighted the complex interplay between both aspects of influence, and has called attention to the need for additional theoretical research to specify the relationship between communication campaigns and interpersonal talk (Southwell & Yzer, 2007; 2009). The empirical research conducted thus far demonstrates that interpersonal interaction is often related to the primary dependent variables targeted by campaign designers. For example, media exposure on science topics was found to predict perceived topic understanding, which in turn predicted an increased prevalence in relevant conversations (Southwell & Torres, 2006). Such media-motivated interpersonal talk has been found to affect cognitive and behavioral outcomes, such as attempts to quit smoking (Hafstad, Aaro, & Langmark, 1996; Hafstad & Aaro, 1997; Dunlop, Wakefield, & Kashima, 2008), perceptions of personal risk (Morton & Duck, 2006), and political participation and knowledge (Cho, Shah, McLeod, McLeod, Scholl, & Gotlieb, 2009). Of course, it has long been known that some people are more influential within social networks than others (Rogers, 2003). Designing campaigns that can induce influential people to engage in interpersonal talk, then, has the potential to be a particularly promising strategy. Originally, those studying the diffusion of innovations identified the influential members of particular networks after the fact (Katz & Lazarsfeld, 1955). More recently scholars have begun to identify a priori the influential members of a network and seek their assistance in spreading ideas and behavior (Kelley, et al., 1997; Miller, Klotz, & Eckholdt, 1998; Schuster et al., 2006; see Bertrand, 2004 for a review). Interventions such as Kelley et al.’s often rely on insiders nominating opinion leaders of small, well-defined communities. Although these interventions have had some success, applying the opinion leader concept to larger, more heterogeneous communities requires a more systematic method of identifying opinion leaders. Identifying the Extremely Influential Traditional measures of opinion leadership (e.g. Rogers & Cartano, 1962) do not include key aspects of what makes an opinion leader influential (Baumgarten, 1975; Chan & Misra, 1990; Fitzgerald, Ferlie, Wood, & Hawkins, 2002; Tschoertner, Jers, & Schenk, 2006). In order to build upon the traditional concept of an opinion leader, Boster et al. (in press) developed the concept of the superdiffuser, someone who is able to influence many others to accept new ideas or engage in novel action. Based on their review of opinion leadership, they concluded that a superdiffuser must be highly connected, highly persuasive, and highly expert (i.e., a maven) in some domain. According to Boster et al. (in press), someone connected highly must be sociable and enjoy meeting new people. Such persons are often a bridge or a liaison between disparate groups. Their ability to access different groups of people gives connectors the ability to target a large number of people when they choose to attempt to diffuse an idea or practice. A superdiffuser must also be highly persuasive (Boster et al., in press). Such persons are highly skilled at arguing for a position (see Carpenter, Kotowski, Boster, Andrews, Serota, & Shaw, 2009). They are able to frame arguments effectively, tailoring them to have maximum suasory impact on particular people or groups of people. They particularly enjoy argumentation, and will seek out opportunities to persuade others. These superior persuasion skills enable them to convince many of those with whom they interact. Finally, a superdiffuser must be a maven in a particular domain (Boster et al., in press). Although the attributes of connection and persuasion are not domain specific, mavens are extremely knowledgeable about a particular content area such as healthy lifestyles, politics, or popular music. They enjoy giving people advice in their area of expertise and people often seek them out when they need advice in the maven’s domain of expertise. This attribute is particularly pivotal to the diffusion process because mavens’ combination of content specific credibility and knowledge provides the factual foundation that enables superdiffusers to craft effective arguments promoting ideas or action. The original Boster et al. validation study examined mavens within a health context; subsequent research has found evidence for political mavens as well (Serota, Carpenter, Andrews, & Boster, 2009). Boster et al. (in press) developed items to assess each of these three attributes, found them to fit the hypothesized three-factor measurement model, and provided construct validation evidence for each of the three attributes. They defined a superdiffuser as someone high on each of the three attributes, high being given the statistical definition of scoring at or above the 75th percentile. In a subsequent experiment they also found that their measures were able to capture aspects of opinion leadership that previous scales failed to include. For example, Carpenter et al. (2009) demonstrated that superdiffusers produced more complex arguments and were better able to adapt their arguments to the characteristics of their target audience. Boster et al.’s (in press) measures of the attributes of a superdiffuser permit highly influential people in large, diffuse communities to be identified. If those superdiffusers could be recruited to advocate for a change in health behavior, a positive change in the behavior of the members of that community would be predicted to result. The manner in which such an intervention is expected to operate is explicated subsequently. The Dynamics of Superdiffuser Effects Some proportion, α, of a target population, P, may have adopted the advocated health practice prior to the implementation of the intervention. Using this notation the number of pre-experiment adopters is given by αP, where 0 ≤ α ≤ 1. Suppose that prior to the implementation of the intervention a survey designed to identify those high on the connector, persuader, and maven scales is administered to the target population. Furthermore, suppose that all persons high on these three measures are identified, .05 being approximately the proportion of these superdiffusers in the population estimated from prior research (Boster et al., in press). Further suppose that these .05P persons agree both to adopt the advocated health practice and to attempt to diffuse the advocated health practice to persons to whom they are connected. Denoting as A0 those who adopt the advocated health practice prior to the implementation of the intervention, A0 = f(those who have adopted pre-experiment, number of superdiffusers) = those who have adopted + number of superdiffusers = αP + .05P (1) Expression (1) presumes that the set of superdiffusers and the set of early adopters do not overlap. Although this assumption is likely false, α is likely to be so small in applications of interest that violating it will not reduce the predictive power of the model substantially. To the extent that the intervention is effective, superdiffusers would convince other members of the target population to adopt the advocated health practice. After the implementation of the intervention the number of persons in P now practicing the advocated behavior would include those who adopted the practice prior to the intervention plus those who had been convinced during the time that the intervention was implemented. Denoting the number of persons having adopted the advocated health practice post-intervention as A1, A1 = f(those who have adopted pre-experiment, number of superdiffusers, number influenced during the implementation of the intervention) = those who have adopted pre-experiment + number of superdiffusers + those influenced during the intervention = αP + .05P + those influenced during the intervention (2) and, presuming that neither the pre-intervention adopters nor the superdiffusers cease engaging in the advocated health practice, or that the number that does so equals the number of persons who might adopt the advocated health practice in the absence of superdiffuser influence, the change in adoption is given by ΔA = A1 - A0 = those influenced during the intervention. The number of population members who are able to be influenced during the intervention is not equivalent to population size. Instead, because the early adopters and the superdiffusers engage in the advocated health practice prior to the intervention, these sets of persons are no longer able to be convinced to change in the direction of adopting the advocated health practice.1 The reduced population, P΄ = P - (αP + .05P), (3) then denotes the members of P that could possibly change in the manner prescribed by the intervention. If superdiffusers are denoted as I, then each I has a given number of interpersonal contacts that are members of P΄. Denoting the number of these contacts as k, this figure can be expressed as a proportion of P΄. Specifically, k = λP΄, where 0 < λ ≤ 1, (4) where λ is the proportion of P΄ contacted by each superdiffuser. This expression presumes that k is a constant, which is almost certainly false. Nevertheless, if the mean or median (in cases in which the distribution of k is skewed markedly) k, Mk, is substituted, then the estimate of A1 that emerges remains reasonably accurate. This estimate of k also presumes that there is no overlap in the contacts for each I.2 Let φ denote the proportion of those k contacts that each I convinces. Then, if X is the number of persons each I convinces, Xi = φk = φλP΄, where 0 < φ ≤ 1 (5) I I I and Σ Xi = Σ φk = Σ φλP΄ = I φλP΄ is the number of persons convinced. i=1 i=1 i=1 (6) Hence, A1 = f (number convinced, number of pre experiment adopters, I) = f (Xi + A0) I = Σ Xi + αP + .05P = IφλP΄ + αP + .05P i=1 = IφλP΄ + (α + .05) P (7) provides an estimate of the number of persons predicted to adopt the advocated health practice after the duration of the intervention. The parameters in equation 7 must be estimated in order to estimate A1 a priori. This predicted figure could then be compared with an obtained figure generated from an experimental test of the model. The difficulty with this approach is the challenges that arise when attempting to generate estimates of these parameters. An estimate of P might be obtained from population statistics. An empirical estimate of I can be generated by noting the number of superdiffusers participating in the intervention. The parameter α can be estimated by conducting a pretest survey and obtaining an estimate of the proportion of those respondents who report having already adopted the advocated health practice. P΄ can be estimated by assessing in the same pretest survey the number of persons who reported already having adopted the advocated health practice, and subtracting that figure from the total population. Presuming that each superdiffuser contacts each k, the parameter λ can be estimated by asking the participating superdiffusers to report the number of contacts they made during the intervention, calculating the mean of this variable across all superdiffusers, and dividing by the size of the population. Estimating φ presents the greatest challenge. Optimally, each superdiffuser would report the name of each contact, all contacts would be queried as to whether or not they had adopted the advocated health practice, and the ratio of the number of adopters to total contacts would be calculated. This labor intensive method is impractical, likely to result in large amounts of missing data (see Katz & Lazarsfeld, 1955), and likely to produce unstable estimates. Alternatively, given that A1 can be measured, and given that the other parameters can be estimated, an estimate of φ can be obtained by presuming the model to be correct. If accurate, such an estimate is of substantial importance because it would provide a measure of the success of superdiffusers in influencing their peers. An Intervention The empirical test of this intervention proceeded in stages. First, the target community was surveyed using the superdiffuser scales developed by Boster et al. (in press). The health behavior of the target community and a control community were also surveyed prior to the intervention. In the next stage, those whose scores on the superdiffuser scales indicated that they were high on connection, persuasion, and domain expertise (mavens) were contacted and asked to help influence others to adopt a particular health innovation. Finally, after the superdiffusers had been given time to influence their peers,3 the target community and the control community were again surveyed concerning the target health behavior. The behavioral changes in the target community were predicted to exceed those of the control community. Method Locating Superdiffusers Sample. In order to determine which students on the target campus were superdiffusers, each undergraduate student (more than 36,000 students) was emailed a request to complete an online survey. A total of 2,958 undergraduate students (8.1%: 1,925 female, 1,026 male, seven did not indicate their sex) filled out the survey. Their mean age was 20.51, SD = 2.40 and ranged from 18-54. Procedures. Ss were emailed a request to participate in the study in exchange for a chance to win a cash prize. Those who clicked on the link provided were taken to the online consent form. After they indicated their consent, the web page took them to the survey. They were first asked for their age and sex. Then the connector, persuader, and health maven (CPM) scales followed. Finally, they were asked for their names and email addresses so that they could be contacted if they won a prize or if they qualified to participate in a future study. Those who were chosen randomly to win a cash prize were contacted and informed how they could obtain their prize. Instrumentation. The CPM scales were taken from Boster et al. (in press). Responses to the items were made on 5-point Likert response scales anchored by the phrases “disagree strongly” and “agree strongly.” The maven items focused on health so as to identify those knowledgeable about healthy lifestyles. The mean of the responses to each of the three measures were employed to form a connector score, a persuader score, and a health maven score for each participant. There were five connector items. Example items included, “I’m often the link between friends in different groups,” and “I try to bring people I know together when I think they would find each other interesting.” The scores were somewhat skewed negatively with α = .85, M = 3.60, and SD = .76. There were five persuader items. Example items included, “I am good at thinking of multiple ways to explain my position on an issue,” and “More often than not, I am able to convince others of my position during an argument.” The scores were skewed negatively and leptokurtic with α = .86, M = 3.89, and SD = .66. There were five health maven items. Example items included, “When I know something about a healthy lifestyle topic, I feel it is important to share that information with others,” and “Being knowledgeable enough about healthy lifestyles so that I could teach someone else is important to me.” The scores approximated the normal distribution with α = .88, M = 3.51, and SD = .84. Ss were considered a superdiffuser if they scored at or above the 75th percentile on all three CPM scales. In this sample, the cut-off points were at least 4.2 on the connector scale, 4.4 on the persuader scale, and 4.0 on the maven scale. Using this criterion, there were 146 superdiffusers in the sample (4.9%). Recruiting Superdiffusers Ss who met the superdiffuser inclusion criteria received a second email message (see Appendix A). In this message, they were told that their scores on the initial survey made them eligible to participate in the second portion of the study. The payment structure for continued participation was discussed, as were the broad parameters of what they would be asked to do. Next, Ss were asked to attend one of two training sessions. The training sessions began with an overview of the project, how the Ss were selected to be a part of it, and the monetary payments they would receive if they chose to participate. Ss’ role in the study was described as being an advocate of multivitamin use to members of their social network for a period of 17 days. Activities could include promoting the benefits of multivitamins, attempting to persuade others to begin or continue to take multivitamins, or persuading others to advocate the practice of taking multivitamins. Ss were instructed that any communication channel would be acceptable (face-to-face, email, social networking sites, etc.). The particulars of how Ss might diffuse multivitamin information were left unspecified, as it was thought that such guidelines might interfere with the existing behavioral repertoire of someone scoring high on the CPM scales. Ss were, however, given a link to an online resource (Harvard School of Public Health, 2010) that described the benefits of taking a multivitamin. If asked why they were suddenly advocating multivitamins, Ss were told to tell the truth; they were participating in a not-for-profit study organized by a university research team attempting to increase multivitamin usage on campus. After the training session, Ss who were willing to participate in the second stage of the study were given their first monetary payment and emailed a spreadsheet for logging daily interactions regarding multivitamin usage. Information recorded in the logs included the number of people contacted about multivitamins, the communication channels used, the location of the interaction, the length of the interaction, and the names, if remembered, of the interactants. This information was used for descriptive purposes, and to determine which Ss would receive additional prizes for contacting the most people. Initially, 37 (25.3%) superdiffusers agreed to participate. Of these 37, 31 (21.2% of the number of superdiffusers identified; 83.8% of those agreeing to participate) returned their logs at the end of the period. They reported trying to persuade a total of 6,296 students. The mean number of people each superdiffuser tried to persuade was 203.10 (SD = 218.46, Md. = 143), or slightly less than 12 per day. This distribution was skewed positively and leptokurtic. Superdiffusers were also asked to report which communication media they used to persuade on each day of the intervention. Given that 31 superdiffusers completed the activity logs for the 17 days of the campaign, the maximum number of days a medium could be mentioned on the activity logs is 31 x 17 = 527. Face-to-face interaction was reported on the most number of days (214), followed by Facebook (57), instant messaging (37), telephone (31), Twitter (12), email (10), MySpace (4), texting (3), blog (1), and self-made print materials (1). Measuring the Effect of the Intervention Sample. The sample at the target (experimental) school was obtained by emailing 9,974 randomly chosen undergraduate students a request to complete a survey both right away and again several weeks later in exchange for a chance to win a cash prize in a drawing. A total of 783 Ss filled out the pretest survey (257 male, 526 female), and of those, 449 Ss (57.3%: 132 male, 317 female) filled out the posttest. The sample at a comparable control school was obtained by soliciting Ss from the Communication Department participant pool. Students were offered extra credit in exchange for their participation via email. A total of 127 Ss filled out the pretest survey (54 male, 73 female) and of those, 81 Ss (63.8%: 33 male, 48 female) filled out the posttest. Procedures. Ss who chose to participate on the pretest were provided with a link to the online consent form. After indicating their consent the web page took them to the survey. The survey contained the demographic items, items measuring communication about multivitamins, and items measuring vitamin-taking behavior (see Table 1). Finally, they were asked for their names and email addresses so that they could be contacted to participate in the posttest. Those who completed the pretest and provided their contact information were emailed a request to participate in the posttest. The posttest contained the same items as the pretest. Those who were chosen randomly to win a cash prize were contacted and informed how they could obtain their prize. Instrumentation. The Ss were first asked which school they attended (experimental or control). They then responded to several yes-no questions concerning communication about vitamins and their behavior. They were first asked “In the last few months, has anyone tried to persuade you to take multivitamins?” with 244 (26.8%) saying yes and 666 saying no (73.2%) on the pretest and 116 saying yes (21.9%) and 414 saying no (78.1%) on the posttest. They were then asked, “Have you heard of other people beginning to take multivitamins recently?” and 381 said yes (41.9%) and 529 said no (58.1%) on the pretest and 195 said yes (36.8%) and 335 said no (63.2%) on the posttest. The next question asked, “Have you tried to persuade anyone to take multivitamins recently?” and 144 said yes (15.8%) and 766 said no (84.2%) on the pretest and 86 said yes (16.2%) and 444 said no (83.8%) on the posttest. Finally, they were asked if they take a daily multivitamin. On this item 285 said yes (31.3%) and 625 said no (68.7%) on the pretest and 171 said yes (32.3%) and 359 said no (67.7%) on the posttest. On none of the variables measured did subjects who completed both the pretest and posttest differ substantially or statistically significantly from those only completing the pretest. Results The Effect of the Intervention on Change in Communication Patterns There was evidence that the experimental induction affected the change in the frequency of Ss reporting that they had heard recently of other people starting to use multivitamins. As Table 2 indicates, experimental condition Ss were more likely to report a change in the direction of having heard recently of others using multivitamins than were control condition Ss (12.7% v. 3.7%). Moreover, as Table 2 suggests, the experimental group-control group difference in the proportion of Ss who reported not having heard recently of someone starting to take a multivitamin on the posttest, but who had reported having heard of someone starting to take a multivitamin on the pretest, was trivial (17.1% v. 16%). The results of contingency table analysis were consistent with this observation (Χ2(2, N = 530) = 5.92, p = .052). To highlight the nature of this effect the no change and negative change categories were collapsed, and subsequent analyses indicated a substantial impact of the induction (Χ2(1, N = 530) = 5.53, p = .019, Fisher’s Exact Test p = .02, r = .10, OR = 3.78). The character of the observed change was that control group Ss were slightly more likely than experimental group Ss to report having heard of a person who had started using a multivitamin on the pretest (49.4% v. 41.2%), but this difference narrowed substantially on the posttest (37.0% v. 36.7%). There was no indication that the intervention was associated with change in any of the other communication measures. The Effect of the Intervention on Change in Multivitamin Use Table 3 suggests that the intervention had an impact on the change in self-reported multivitamin use, and a subsequent contingency table analysis was consistent with this impression (Χ2(2, N = 530) = 8.06, p = .018). Observing the conditional percentages presented in Table 3 suggests that this effect results from the substantial difference in the proportion of Ss who reported using multivitamins on the pretest but no longer using them on the posttest, and subsequent analyses are consistent with this observation. When the negative change and no change categories are collapsed, there is no longer any evidence of a treatment effect (Χ2(1, N = 530) = 0.77, p = .379, r = -.04, OR = .64). On the other hand, when the positive change and no change categories are collapsed, a substantial difference in those who reported that they stopped using multivitamins on the posttest after reporting using them on the pretest was observed (Χ2(1, N = 530) = 7.06, p = .008, r = .06, OR = 3.17). The nature of this difference is that experimental condition Ss reported that they were less likely to stop using multivitamins on the posttest after reporting having used them on the pretest than were control group Ss (3.3% v. 9.9%). The Effect of Communication on Change in Multivitamin Use Although there was no evidence that reporting having heard of other people taking multivitamins mediated the impact of the intervention on behavioral outcomes, there was evidence that an important aspect of communication behavior affected the change in self-reported multivitamin use. Specifically, when scored dichotomously (positive change v. other), the change in Ss reporting that someone had tried to persuade them to use a multivitamin (positive change) was associated with the change in whether or not Ss reported starting to use a multivitamin, contrasted with those not changing or ceasing use of a multivitamin (no change/stop). These data are presented in Table 4. From this table one may observe that this association was substantial (Χ2(1, N = 530) = 20.37, p < .001, r = .20, OR = 6.96). Those who reported on the posttest that someone had tried to persuade them to use a multivitamin, but who had not heard such a message on the pretest, were much more likely to report having changed in the direction of starting to use a multivitamin (18.9%) than others (3.2%). Performing this analysis for the experimental group Ss only produces a similar effect (Χ2(1, N = 449) = 12.08, p < .001, r = .16, OR = 5.70). A Model of the Social Influence Process It was discovered that changing multivitamin use was also associated with the S trying to persuade others (see Table 5). Collapsing across conditions, those who started multivitamin use were more likely than those who did not start multivitamin use to attempt to persuade others (26.1% v. 4.7%, Χ2(1, N=530)=18.79, p < .001, r = .19, OR = 7.10). Table 6 presents the uncollapsed data. Consequently, analyses were performed to examine the possibility that behavior change mediated the relationship between others attempting to persuade the S, and the S attempting to persuade others. This model is presented in Figure 1. Causal analyses indicated that the data were consistent with this model. The model predicts that the correlation between whether or not someone attempted to persuade the S and whether or not the S attempted to persuade others is the product of the correlation of someone attempting to persuade the S and behavior change (r = .20) and the correlation between behavior change and the S attempting to persuade others (r = .19). This predicted correlation of .04 differed trivially from the obtained value (e = .02, Χ² (1, N = 530) = .13, ns). Estimates of Superdiffuser Influence As indicated previously the variables and parameters in equation 7 may be measured and estimated to produce an estimate of φ, given the assumption that the model is correct. A1 is the estimated number of persons who indicated on the posttest that they used a multivitamin. The undergraduate population of the experimental university is approximately 36,337, an estimated 4% of them were influenced to begin using a multivitamin, and an estimated 30% of them reported using a multivitamin at pretest. Thus, A1 is estimated to be .34 (36,337) = 12,354. Thirty-one influentials (I = 31) participated in the intervention. Dividing k by P’ provides an estimate of λ, so that λ = 203/25,400 = .008 if the mean is employed as the measure of number of contacts per superdiffuser, and λ = 143/25,400 = .006 if the median is employed as the measure of number of contacts per superdiffuser. Because 1,817 superdiffusers would be expected in a population of 36,337, the 31 who agreed to participate in this intervention represent .017P. Thus, equation 7 can be written as A1 = IφλP΄ + (α + .05) P 12,354 = {[(31) φ (.008) (25,400)] + [(.30 + .017) (36,337)]} Hence φ is estimated as .13 if the mean is employed to estimate λ, or .18 if the median is used to estimate λ. Discussion The intervention ran for 17 days and 31 superdiffusers were recruited successfully to help influence others to adopt multivitamin use in a community of more than 36,000 students. Despite the modest scale of this induction, it had an impact on the experimental school. There was a larger increase in Ss hearing about people beginning to take multivitamins at the experimental school than the control school, although, notably, the campaign was not associated with an increase in persuasive attempts. There was also a smaller decrease in Ss quitting multivitamin usage at the experimental school than the control school. Although both of these effects were somewhat modest in size, they are comparable to those often reported in mass media campaigns (Snyder, 2006). Although preventing a negative change in vitamin use was not the objective of the campaign, it was consistent with the goal of harnessing interpersonal networks to improve the health of the target community. These results contribute to an emerging corpus of research indicating that interpersonal communication plays a key role in the success of communication campaigns (Hafstad and Aaro, 1997; Hafstad, Aaro, and Langmark, 1996; Dunlop, Wakefield, and Kashima, 2008; Morton and Duck, 2006; Cho, Shah, McLeod, McLeod, Scholl, and Gotlieb, 2009). This study expands upon previous research on interpersonal talk within campaigns by examining the effect of a particular type of interpersonal talk, that initiated by influential people within social networks. Future research would benefit from examining the implementation of superdiffusers within a traditional mass media campaign; the presence of media messages could serve as a conversation starter for superdiffusers, as well as reinforce their message. In addition to finding evidence consistent with the expected effect of the intervention, evidence was also found consistent with a causal model such that those who someone tried to persuade to take multivitamins were more likely to start taking them and were then in turn more likely to try to persuade others. This outcome suggests that when the superdiffusers persuade their initial targets for vitamin usage, those people then continue to spread multivitamin usage to others, i.e., the presence of modest second order effects. Care must be taken in interpreting the results of the causal analysis, however; an experimental design was not used, and hence the direction of causality is not certain. This intervention also allowed the estimation of the parameters of the proposed diffusion model, given the assumption that the model is correct. Specifically, it was estimated that superdiffusers persuaded approximately one in six of the people they contacted to adopt a particular behavioral change. Given more time, other topics, smaller target populations, or other variations this parameter may increase or decrease, as superdiffusers may have more time or opportunities to use multiple messages and target their messages to specific population members who did not comply with their initial request. Limitations One limitation of this experiment is that the experimental school had a larger and more diverse sample of Ss than the control school. The impact of using the Communication Department subject pool in the control school as opposed to the campus wide survey in the experimental school was likely minimal as there was no reason to believe that Communication students would be more or less likely to change their multivitamin usage than students of any other department. On the other hand, the smaller sample from the control school made the prevalence estimates at the control school less precise. For this reason future research would profit from a larger control sample. Similarly, a higher response rate at the experimental school would have been desirable. It is possible the subjects who did reply to both the pretest and posttest were more attuned to health issues, and thus more able to be influenced on the topic of multivitamin use. This problem is exacerbated by the relatively small number of persons who changed their use of multivitamins. Consequently, future research would also profit by extending the duration, and perhaps increasing the intensity, of the induction. The process of social influence may take more time as repeated attempts by superdiffusers may be required to persuade the more recalcitrant members of their social networks. Furthermore, the current data were consistent with a model such that those who are persuaded to change go on to attempt to persuade others. This diffusion process may also require additional time to unfold. In this way it would be expected that more participants would change their multivitamin use, thus producing more robust results. Alternatively, if this expectation was not met, then the failure of the induction would be clear. Finally, only a quarter of identified superdiffusers agreed to participate in the campaign. With a larger number of superdiffusers operating, the campaign might have produced larger effects. Understanding why these influential people decided not to participate, and devising persuasive strategies to obtain their compliance, are important areas for future research. It is possible further inducements, both monetary and non-monetary, would increase the rate of participation, as would a topic more relevant to the lives of undergraduates. Conclusion This intervention demonstrated that even with a small sample of superdiffusers and a short period of time, recruiting superdiffusers can be an effective way to spread a health innovation across a large, diverse target population. Future research is necessary to make direct comparisons between the effectiveness of a mass media campaign and that of a superdiffuser intervention, as well as to test the efficacy of a superdiffuser intervention in the context of a mass media campaign. Minimally, this intervention demonstrates that identifying and recruiting superdiffusers is a promising tool for health communication scholars and practitioners. References Baumgarten, S. A. (1975). The innovative communicator in the diffusion process. Journal of Marketing Research, 12, 12-18. Bertrand, J. T. (2004). Diffusion of innovations and HIV/ AIDS. Journal of Health Communication, 9, 113-121. Boster, F. J., Andrews, K. R., Kotowski, M. R., & Serota, K. (in press). Identifying Influence: Development and Validation of the Connectivity, Persuasiveness, and Maven Scales. Journal of Communication. Carpenter, C. J., Kotowski, M. R., Boster, F. J., Andrews, K. R., Serota, K., & Shaw, A. S. (2009). Do superdiffusers argue differently? An analysis of argumentation style as a function of diffusion ability. Argumentation and Advocacy, 45, 151-170. Chan, K. K., & Misra, S. (1990). Characteristics of the opinion leader: A new dimension. Journal of Advertising, 19, 53-60. Cho, J., Shah, D. V., McLeod, J. M., McLeod, D. M., Scholl, R. M., & Gotlieb, M. R. (2009). Campaigns, reflection, and deliberation: Advancing an O-S-R-O-R model of communication effects. Communication Theory, 19, 66-88. Dunlop, S. M., Wakefield, M., & Kashima, Y. (2008). The contribution of antismoking advertising to quitting: Intra- and interpersonal processes. Journal of Health Communication, 13, 250-266. Fitzgerald, L., Ferlie, E., Wood, M., Hawkins, C. (2002). Interlocking interactions, the diffusion of innovations in health care. Human Relations, 55, 1429-1449. Hafstad, A., & Aaro, L. E. (1997). Activating interpersonal influence through provocative appeals: Evaluation of a mass media-based antismoking campaign targeting adolescents. Health Communication, 9, 253-272. Hafstad, A., Aaro, L. E., & Langmark, F. (1996). Evaluation of an anti-smoking mass media campaign targeting adolescents: The role of affective responses and interpersonal communication. Health Education Research, 11, 29-38. Harkins, S. G., & Petty, R. E. (1981). Effects of source magnification of cognitive effort on attitudes: An information processing view. Journal of Personality and Social Psychology, 40, 401-413. Harvard School of Public Health. The nutrition source. Retrieved October 29, 2010, from http://www.hsph.harvard.edu/nutritionsource/what-should-you-eat/vitamins/index.html Katz, E., & Lazarsfeld, P. F. (1955). Personal influence: The part played by people in the flow of mass communications. New York: The Free Press. Kelly, J. A., Murphy, D. A., Sikkema, K. J., McAuliffe, T. L., Roffman, R. A., Solomon, L. J., Winett, R. A., & Kalichman, S. C. (1997). Randomized, controlled, community-level HIV-prevention intervention for sexual-risk behaviour among homosexual men in US cities. Lancet, 350, 1500-1505. Miller, R. L., Klotz, D., & Eckholdt, H. M. (1998). HIV prevention with male prostitutes and patrons of hustler bars: Replication of an HIV preventive intervention. American Journal of Community Psychology, 26, 97-131. Morton, T. A., & Duck, J. M. (2006). Enlisting the influence of others: Alternative strategies for persuasive media campaigns. Journal of Applied Social Psychology, 36, 269-296. Rogers, E. (2003). The diffusion of innovations (5th ed.). New York: The Free Press. Rogers, E., & Cartano, D. G. (1962). Methods of measuring opinion leadership. The Public Opinion Quarterly, 26, 435-441. 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 activities among opinion leaders in California. Communication Theory, 16, 91-117. Serota, K. B., Carpenter, C. J., Andrews, K. R., & Boster, F. J. (2009). Influentials in America: Identifying Political Superdiffusers. Paper presented at the annual meeting of the National Communication Association, Chicago, Illinois. Snyder, L. B. (2006). Meta-analyses of mediated health campaigns. In R. W. Preiss, B. M. Gayle, N. Burrel, M. Allen, & J. Bryant (Eds.), Mass media effects research: Advances through meta-analysis. Lawrence Erlbaum Associates: Mahwah, NJ. Southwell, B. G., & Torres, A. (2006). Connecting interpersonal and mass communication: Science news exposure, perceived ability to understand science, and conversation. Communication Monographs, 73, 334-350. Southwell, B. G., & Yzer, M. C. (2007). The roles of interpersonal communication in mass media campaigns. Communication Yearbook, 31, 420-462. Southwell, B. G., & Yzer, M. C. (2009). When (and why) interpersonal talk matters for campaigns. Communication Theory, 19, 1-8. Tschoertner, A. C., Jers, C., Schenk, M. (2006, June). Are all opinion leaders opinion givers? Are all opinion givers opinion leaders?: A clarification of constructs based on empirical data. Paper presented at the International Communication Association Annual Conference, Dresden, Germany. Appendix A: Recruitment email to superdiffusers Hello, You are receiving this email because you completed an online survey recently, and based on your responses have been selected to participate in Stage 2 of our study. If you agree to participate, you will be paid $70, plus will have the chance to win an additional $100. Participation will require you to attend a meeting where we explain the study to you, and describe your role. The goal of the study is to convince the student body to begin taking multivitamins every day. Your job will be to try and convince as many fellow students as possible to take multivitamins. You will receive $35 at the first meeting, and another $35 at the end of the semester. The meeting will be held at this time and date [TBD after IRB approval] in room 154 of the Communication Arts and Sciences Building. Footnotes 1 This point does not discount the possibilities that messages might be sent to them that serve to strengthen their behavioral choices, that might make these choices less vulnerable to counterpersuasion efforts, or that might make them opt to become an advocate of the advocated health practice. 2 This assumption can be relaxed, but at this juncture it can be noted that contacts who receive messages from more than one superdiffuser have a higher likelihood of adopting the recommended health practice (Harkins & Petty, 1981). Thus, when overlap in contacts produces an overestimate of k, the influence parameter, φ, introduced subsequently is underestimated. It is presumed that these errors cancel to within a small margin of error. 3 It might be noticed that there is no model of second order effects, or even higher order effects. That is, the possibility that those influenced by superdiffusers might influence others is not considered. The model was not expanded in this manner for this project because the duration between pretest and posttest was deemed insufficient to produce measurable higher order effects. Table 1 Raw frequencies (percentages) for primary survey questions by condition and time Induction Control Experimental Question Pre-test Post-test Pre-test Post-test Someone tried to persuade S 38 (29.9%) 20 (24.7%) 206 (26.3%) 96 (21.4%) Heard of others taking multivitamins 64 (50.4%) 30 (37.0%) 317 (40.5%) 165 (36.7%) S tried to persuade others 24 (18.9%) 14 (17.3%) 120 (15.3%) 72 (16.0%) S takes a daily multivitamin 49 (38.6%) 28 (34.6%) 236 (30.1%) 143 (31.8%) Note: There were 127 Ss at the control school and 783 Ss at the experimental school. Table 2 Change in Having Heard of Others Starting to Take a Multivitamin as a Function of Condition (N in parentheses) Condition Control Experimental Positive Change 3.7% (3) 12.7% (57) No Change 80.2% (65) 70.2% (315) Negative Change 16.0% (13) 17.1% (77)  Table 3 Change in Multivitamin Use as a Function of Condition (N in parentheses) Location   Control Experimental Began Using 6.2% (5) 4.0% (18) No Change 84.0% (68) 92.7% (416) Stopped Using 9.9% (8) 3.3% (15)  Table 4 Change in Multivitamin Use as a Function of Change in Communication (N in parentheses) Communication No Change/ Stop Positive Began Using 3.2% (16) 18.9% (7) Did Not Begin Using 96.8% (477) 81.1% (30) Note: Χ2(1, N = 530) = 20.37, p < .001, r = .20, OR = 6.96Table 5 Change in Persuasion Attempts as a Function of Change in Multivitamin Use (N in parentheses) Behavior Not Begin Using Begin Using Change / Persuasion Attempt 4.7% (24) 26.17% (6) No Change / No Persuasion Attempt 95.3% (483) 73.9% (17) Note: Χ2(1, N=530)=18.79, p < .001, r = .19, OR = 7.10Table 6 Persuasion Attempts as a Function of Multivitamin Use (N in parentheses) Multivitamin Use Never Took a MV Started Taking a MV Stopped Taking a MV Kept Taking a MV Never Persuaded 305 13 17 73 Started Persuading 11 6 1 12 Stopped Persuading 9 3 2 22 Kept Persuading 11 1 3 41 Figure Caption Figure 1. A causal model of the social influence process. Employing Interpersonal 34