Although proponents argue that DTCA has benefits, such as educating consumers and improving patient-physician interaction, opponents argue that it has many harmful effects, such as misinforming patients and overemphasizing benefits [10]. The United States and New Zealand are the only two developed countries where DTCA is legal [11]. The US Food and Drug Administration (FDA) is responsible for DTCA oversight and has regulated drug marketing that appears through traditional broadcast media, banning misleading statements and creating categories for different types of advertisements. Product claim ads mention a specific drug by name and the ailment it intends to treat. This type of advertisement must follow a “fair balance” rule, meaning that benefits and risks are given equal coverage. Help-seeking ads, which mention an ailment but not the name of a drug, and reminder ads, which mention a drug but not what it treats, are not required to meet the fair balance rule.

The emergence of social media and other interactive platforms has only exacerbated concerns related to DTCA leading prominent scholars and health professionals to wonder “...whether regulatory responses by FDA are responsive and adaptive enough to address the inherent challenges faced by a universe of digital and Internet-based forms of DTCA” (p 271 [12]). Specifically, researchers and health professionals are extremely concerned that companies will market their drugs online in ways that (1) obscure the role companies play in producing drug information, and (2) strategically control user-generated contributions to promote a favorable, one-sided view of a company and its products [5,8,12]. A recent content analysis clearly documents how prominently pharmaceutical companies are using Facebook, YouTube, and Twitter for promotional activities [3]. In order to better understand how these specific concerns about online DTCA might affect the evaluation of drug information appearing on social media, we draw on research that more broadly examines how features of new media affect the evaluation of information.

Researchers have applied warranting theory [13,14] to understand how features of new media affect evaluations of people [15], companies [16,17], products [18], and websites [19]. A central premise of warranting theory is that people trust information more or less depending on its warranting value; the warranting value of information is defined as the degree to which information is controlled or manipulated by the target it describes. The more people perceive information about a target (eg, person, organization, company) to be under the control of the target, the less they trust the information [20].

As in the case of online DTCA, the complexities of new media can make it difficult to know the true identity of an information source or with whom the source is affiliated. Researchers have documented the many ways that online sources try to influence viewers by masking their true identities [21,22]. The prevalence of fake online reviews, commissioned or produced by the target being evaluated [23], provides a prominent example of how source obfuscation occurs online. Warranting theory predicts that information produced by a third party is more influential the less people perceive it to be under the control of the target being evaluated [13]. A recent study supported this prediction by indicating that online reviews were less impactful the more people were uncertain about the true identity of the reviewers [16].

Overall, consumers tend to trust user-generated reviews or word-of-mouth more than traditional advertisements [24,25]. The persuasive value of personal testimonials in health settings is also well understood [6], and is what makes researchers [9] so concerned about online DTCA wherein the affiliations people or companies have with pharmaceutical companies are not disclosed. Consistent with warranting theory, it is expected that people will trust favorable information about a pharmaceutical drug on social media more the less they perceive the drug manufacturer to be affiliated with third-party information sources. This expectation applies to organizations that post drug information to social media as well as people who comment on such posts. Cues that suggest a connection between organizations or commenters and a drug manufacturer should diminish the warranting value of any favorable evaluations they produce, thus (1) making the information less trustworthy, (2) making people less likely to recommend the drug, and (3) making people less likely to share the information with others in their social network.

Even when information sources truly are third parties with no connection to the target they are evaluating, features of new media can still permit targets to exert control over information they produce. Notably, targets can selectively delete some user-generated content to promote a favorable view of themselves or their products. When targets can delete user-generated content, they can exert control over third-party content, not by editing or influencing the content of messages, but by curating the composition of third-party messages that appear online. As such, targets can orchestrate the false appearance that all online commenters or reviewers share the same opinion about some issue or product (eg, they all view a drug favorably). How targets strategically control the dissemination of third-party information is increasingly important to examine because of the enhanced trust people afford user-generated content [26-29]. If no cues exist to suggest that a target is controlling the dissemination of user-generated content, viewers are likely to view the user-generated content as having a high degree of warranting value [13]. However, actions or cues that suggest that a target is controlling the dissemination of user-generated content (eg, deleting comments, restricting access to content) can lower the perceived warranting value of the information, and thus its impact on viewers’ attitudes and behaviors. A recent experiment supports this prediction in an online review setting showing that positive reviews of a company led to more favorable attitudes toward the company, the more viewers believed that the company could not control or influence what reviews were displayed [16]. We argue that the same principle should apply to control over user-generated drug evaluations on social media. Specifically, cues that indicate that user-generated evaluations of a drug have been removed from a post should increase perceptions that the drug manufacturer is controlling the dissemination of the user-generated evaluations, thus (1) lowering trust in any remaining favorable user-generated evaluations, (2) making people less likely to recommend the drug, and (3) making people less likely to share the information with others in their social network.

Participants in this 2 (organization affiliation) × 2 (commenter affiliation) × 2 (comment deletion) between-subjects experiment were randomly assigned to one of eight conditions. Across all conditions, participants viewed an infographic posted on Facebook by a fictitious health organization about a fictitious prescription allergy drug called OpenAir; the post was always accompanied by user-generated comments. After viewing the stimulus materials, participants completed an online questionnaire and were thanked for their participation.

The study was determined category 2 exempt research by the Ohio State University Institutional Review Board in accordance with the U.S. Department of Health & Human Services. It is most appropriately categorized as research that uses common survey procedures. Authors did not register the trial prospectively because they did not regard this as a clinical trial.

The sample consisted of 674 participants from an online panel who received financial compensation in exchange for their participation. Participants ranged in age from 18 to 85 years (mean 52.86, SD 15.00) and identified as Caucasian (n=587), African American (n=44), Asian or Asian American (n=20), Native American (n=3), Hispanic (n=10), and other (n=10). More participants identified as female (n=515) than male (n=159). Participants were recruited by Qualtrics; incentives were handled by Qualtrics and disclosed to participants prior to their participation.

Across all conditions, participants viewed an infographic that the health organization Expert Opinions in Medicine (EOIM) posted to Facebook. The infographic always contained a quote from a medical doctor about a prescription allergy drug called OpenAir made by Darby Pharmaceuticals. The quote indicated that the drug is effective at treating seasonal allergies. The infographic was always accompanied by three comments from users who indicated that the drug is very effective. The health organization EOIM, the company Darby Pharmaceuticals, and the prescription drug OpenAir were all fictitious and were created to resemble actual entities.

For the organization affiliation factor, changes were made to the infographic to induce variability in participants’ awareness of EOIM’s affiliation with Darby Pharmaceuticals—the ostensible manufacturer of the drug OpenAir. In the nonaffiliated condition, the following statement appeared at the bottom of the infographic: “Expert Opinions in Medicine—An Independent Research Organization.” In the affiliated condition, the statement read “Expert Opinions in Medicine—A Research Organization Funded by Darby Pharmaceuticals.” In addition to this difference, a medical doctor who works for EOIM is quoted in the infographic. The nonaffiliated condition indicated the medical doctor is the Executive Director at EOIM, whereas the affiliated condition adds that he is also an OpenAir Senior Research Scientist at Darby Pharmaceuticals.

We manipulated the other two experimental factors in the comment section that accompanied the infographic. Across all conditions, there were three positive comments on the infographic post. For the commenter affiliation induction, “Darby Pharma” appeared next to the commenters’ names in the affiliation condition (eg, Mel L, Stockton-Darby Pharma). In the nonaffiliation condition, no indication was provided that the commenters were associated with Darby Pharmaceuticals. For the dissemination control induction, the caption “We reserve the right to hide or delete comments” appeared above the comments in the deletion condition. In addition, the comment section also indicated that some comments had been hidden. In the nonremoval condition, no statement was provided about the organization’s deletion policy and no cues indicated that deletion had occurred. Sample stimuli are provided in Figures 1 and 2.

Before conducting the primary analyses, we examined how participants responded to the attention check questions. Participants who answered incorrectly to one or more of the attention check items were removed from the analyses (n=265). Participants who did not answer any of the questions incorrectly were retained (n=409). These two groups of participants did not significantly differ in their age, gender, ethnicity, Facebook use, or whether they suffered from seasonal allergies (all P values >.21).

For each experimental factor, we first provide a t test that directly estimated how each induction affected perceptions of the mediating construct it was expected to vary (ie, organization affiliation, dissemination control, or commenter affiliation). Next, the macro PROCESS [31] was used to estimate the indirect effect each induction had on the outcome measures, through the proposed mediator. We first provide unadjusted indirect effect estimates and then provide covariate-adjusted estimates. A zero-order correlation matrix is provided that includes means and standard deviations for all variables in the analyses (Table 1).

An independent samples t test indicated that participants perceived the organization EOIM as more affiliated with Darby Pharmaceuticals when the infographic provided a funding disclosure relative to when it did not (t₄₀₇=12.23, P<.001, η²=.27). Using Model 4 of the macro PROCESS, we estimated the indirect effect the induction had through perceptions of organizational affiliation on each outcome measure; each estimate is provided with its corresponding 95% bias-corrected bootstrap confidence interval based on 10,000 resamples. Significant indirect effects were found on organization trust (point estimate –0.45, 95% CI –0.69 to –0.24), comment trust (point estimate –0.44, 95% CI –0.68 to –0.22), drug recommendation likelihood (point estimate –0.35, 95% CI –0.59 to –0.15), and Facebook endorsement likelihood (point estimate –0.37, 95% CI –0.64 to –0.16). The preceding analyses were reran controlling for perceptions of dissemination control (ie, Darby Pharmaceuticals controlled what comments appeared) and perceptions that the commenters were affiliated with Darby Pharmaceuticals. When including the covariates in the mediation models, all of the indirect effects became nonsignificant.

To further probe the nature of the relationships, we conducted tests of moderated mediation using Model 14 in PROCESS. The significant indirect effects reported previously were not moderated by participants’ perceptions that the commenters were affiliated with Darby Pharmaceuticals. However, perceptions of dissemination control did moderate three of the significant indirect effects. The overall pattern indicated that—when perceptions of dissemination control were low—the organizational affiliation induction did not indirectly affect (1) trust in the organization, (2) trust in the comments, and (3) the likelihood of recommending the drug. However, as perceptions of dissemination control increased, the size of the indirect effects significantly increased to a substantive degree. For each outcome measure, indirect effects were estimated at three levels of the moderating variable (ie, dissemination control): 1 SD below the mean, at the mean, and 1 SD above the mean. The point estimates were as follows for each outcome: organization trust (point estimates –0.09, –0.29, –0.47), comment trust (point estimates –0.02, –0.12, –0.26), and drug recommendation likelihood (point estimates 0.01, –0.14, –0.29). The direct test of moderated mediation for each outcome was as follows: organization trust index (point estimate –0.11, 95% CI –0.21 to –0.02), comment trust index (point estimate –0.09, 95% CI –0.18 to –0.002), and drug recommendation likelihood index (point estimate –0.09, 95% CI –0.18 to –0.004). A similar pattern was found for the Facebook endorsement outcome but the 95% confidence interval narrowly included zero: index (point estimate –0.08, 95% CI –0.19 to 0.01).

An independent samples t test indicated that participants perceived Darby Pharmaceuticals to have more control over what comments appeared on the EOIM post when cues indicated that deletion had occurred (t₄₀₇=4.39, P<.001, η²=.05). Using Model 4 of the macro PROCESS, we estimated the indirect effect the induction had through perceptions of dissemination control on each outcome measure. Significant indirect effects were found on organization trust (point estimate –0.31, 95% CI –0.47 to –0.17), comment trust (point estimate –0.35, 95% CI –0.53 to –0.20), drug recommendation likelihood (point estimate –0.26, 95% CI –0.41 to –0.14), and Facebook endorsement likelihood (point estimate –0.28, 95% CI –0.45 to –0.15). The preceding analyses were reran controlling for perceptions that the organization and the commenters were affiliated with Darby Pharmaceuticals. When including the covariates, all the indirect effects remained significant; the confidence interval of each adjusted estimate overlapped with its respective nonadjusted confidence interval. That is, the estimates did not significantly differ in magnitude.

An independent samples t test indicated that participants perceived the commenters to be more affiliated with Darby Pharmaceuticals when the cue “Darby Pharma” appeared next to their names (t₄₀₇=10.59, P<.001, η²=.22). Using Model 4 of the macro PROCESS, we estimated the indirect effect the induction had through perceptions of commenter affiliation on each outcome measure. Significant indirect effects were found on organization trust (point estimate –0.68, 95% CI –0.90 to –0.49), comment trust (point estimate –0.81, 95% CI –1.04 to –0.59), drug recommendation likelihood (point estimate –0.61, 95% CI –0.82 to –0.43), and Facebook endorsement likelihood (point estimate –0.63, 95% CI –0.87 to –0.43). The preceding analyses were reran controlling for perceptions that the organization was affiliated with Darby Pharmaceuticals and perceptions that Darby Pharmaceuticals controlled the dissemination of the comments. When including the covariates, all the indirect effects remained significant, the confidence interval of each adjusted estimate always overlapped with its respective nonadjusted confidence interval except in one case. The covariate-adjusted indirect effect on comment trust was significantly attenuated (point estimate –0.37, 95% CI –0.54 to –0.23).

As noted previously, participants were removed from the analyses if they incorrectly responded to one of the attention check items. The purpose of removing the participants was to reduce error and provide a clearer test of the hypothesized relationships. However, in real-world settings, people may only provide fleeting attention to social media posts and/or may be unable to accurately recall what they viewed. As such, there is some value in being exhaustive and looking at the estimates for the full sample—even if this includes participants who made no honest attempt to read or respond to the survey items. As indicated in Table 2, all the indirect effects were significant for the full sample. Although the estimates for the full sample are attenuated relative to the reduced sample, they cannot be statistically differentiated because their respective 95% confidence intervals overlap.

The findings from this study illustrate how important it is to better understand the effects of DTCA in a new media environment. A major concern expressed in past research [5,7,9] is that information sources might be “blurred” online, making it difficult to know when a pharmaceutical company is sponsoring or influencing the production of drug information. The results of this study suggest that cues that disclose connections between health organizations and pharmaceutical companies affect how people process drug information posted on social media. Specifically, disclosing an affiliation decreased (1) trust in an organization that posted information about a drug, (2) trust in comments posted by other site users about the drug, (3) the likelihood of recommending the drug to family or friends, and (4) the likelihood of propagating the drug message further throughout their online social network. Illustrating the complexity of new media environments that contain multiple information sources, the results also indicate that these effects are increasingly pronounced when it appears that a website proprietor controls the dissemination of user-generated comments on a webpage.

Beyond moderating the effect perceptions of organizational affiliation had on the outcome measures, perceptions of control over the dissemination of user-generated content independently affected the outcomes. This type of strategic control over user-generated content has been emphasized as a major concern for online DTCA [7,9]. The findings validate these concerns and help estimate how greatly controlling the dissemination of user-generated content can affect people who view health information posted online. Cues that indicated that an organization removed some of the user-generated comments that accompanied their posts increased people’s perceptions that the drug manufacturer was behind the removal. Notably, the more people thought that the drug manufacturer was controlling the dissemination of the user-generated comments, the less people trusted the user comments and the health organization that posted the infographic. Again, the complexities of a new media environment are illustrated. How information posted by website users is perceived to be controlled not only affects how people evaluate remaining user contributions, it also affects how people view the proprietor of the website (eg, the health organization EOIM). In addition to influencing how people trusted the user-generated comments and the health organization EOIM, control over the dissemination of user-generated content affected the likelihood that people would recommend the drug to others, and endorse/share the information with others in their online social network.

The commenter affiliation induction had similar, yet independent, effects on all the outcome measures. The results indicate that a single affiliation cue next to commenters’ names can significantly increase people’s knowledge that the commenters are affiliated with the drug company, which, in turn, can affect trust in a health organization, comment trust, drug recommendation likelihood, and the likelihood of endorsing and sharing the information with others online. Because personal testimonials from average citizens are highly influential, paid representatives or company employees who post information online without disclosing their connection to a pharmaceutical company is thought to be exceedingly troublesome [9]. The findings from this study distinctly illustrate how impactful it can be when company-affiliated individuals masquerade as neutral, third-party contributors online.

From a practical perspective, the results highlight the need for future FDA guidelines to mandate that pharmaceutical companies clearly disclose connections within messages posted to any website or social media platform that they directly fund, control, or support in some manner. Past research suggests companies have minimized or obscured such disclosures [32]. The bigger challenge, however, is whether it is feasible to regulate how pharmaceutical companies control information about their products across new media platforms that they might only indirectly influence or control [7,12]. Obviously pharmaceutical companies should not be expected to police the entire Internet. Nevertheless, regulations that only pertain to content posted to “official” company media might overlook relationships that—according to the results of this study—would greatly influence how people evaluate drug information.

The FDA has issued draft guidance documents that outline (1) when companies are responsible for user-generated content, (2) recommendations for how they should respond to user-generated content for which they are not required to respond, and (3) recommendations for how to convey risk information when online platforms have space or characters limitations [33,34]. Notably, the documents do not directly discuss how companies should respond to online platforms that make it difficult for content contributors to clearly disclose company affiliations and the precise nature of any affiliations. Ironically, the guidance document on user-generated contributions only provides guidance for handling user-generated content that comes from independent, nonaffiliated sources and exists on platforms in which companies have not edited or removed any third-party content. “A firm is thus responsible for communications on the Internet and Internet-based platforms, such as social media, made by its employees or any agents acting on behalf of the firm to promote the firm’s product, and these communications must comply with any applicable regulatory requirements” (pp 3-4 [33]). It is helpful that the guidance document confirms that existing regulations apply to content produced by company-affiliated sources and content that exists on platforms over which companies exert control. However, the problem remains that features of many prominent social media sites, such as Facebook, make it easy to mask the identity of an information source and difficult to ascertain whether companies are removing user-generated contributions.

The findings of this study help extend the explanatory and predictive power of warranting theory by demonstrating how the core theoretical propositions accurately explain and predict phenomena in a new context. Unlike previous tests of warranting theory that have directly examined uncertainty about the true identity of an information source online [16], this study more directly varied and measured the degree to which information sources were affiliated with the target being evaluated. Although the distinction may appear to be minor, uncertainty about the true identity of a third-party source might affect perceptions of warranting value differently relative to perceptions that a third-party source is affiliated with the target of an evaluation. For instance, an unknown source could actually be the target, someone affiliated with the target, or someone unaffiliated with the target. Future research might seek to further explore how these two considerations about the identity of a third-party source relate to one another and affect evaluations of information appearing online.

A notable finding from this work that has novel theoretical implications is that perceptions of organizational affiliation affected the outcomes differently than perceptions of commenter affiliation. It is possible that the differential effects might be attributable to how the constructs were operationalized in this study. However, it is also possible that perceived affiliations between individuals and targets might function differently than perceived affiliations between organizations/companies and targets. If the ceiling for trusting an online commenter is greater than the ceiling for trusting an organization, differential effects might be expected. Future research might seek to explore these possibilities more directly.

Limitations common to experimental research apply to this study. Although the results support theoretically predicted relationships, future research might seek to further the generalizability of the findings. For instance, researchers can seek to examine the effects of online DTCA across different populations, with different drug messages, and on different social media platforms. Researchers might also seek to examine how consumers’ general skepticism toward pharmaceutical marketing can moderate the effects found in this study.

A limitation specific to this study is that the attention check questions may have been overly sensitive. For instance, some participants viewed an infographic that indicated that EOIM was “An Independent Research Organization.” Participants who indicated that EOIM was not an independent organization and instead was funded by Darby Pharmaceuticals were removed. However, in some of the EOIM nonaffiliation conditions, comments were removed and commenters were affiliated with Darby Pharmaceuticals. It is possible that participants interpreted this combination of cues as indicating EOIM was not really independent, despite the claim that was made. We provide results for the full sample to overcome this limitation, but future researchers should consider how multiple cues might operate in conjunction when seeking to include attention check items designed to reduce measurement error.

Pharmaceutical companies will seek to market their drugs through whatever media people regularly consume. In the current media landscape, this means drug marketing will occur through social media and online platforms that are interactive and include information from multiple sources. Any attempt to regulate online DTCA needs to thoroughly consider the unique affordances and characteristics of emerging communication technology. Whether regulations can keep pace with advances in communication technology remains to be seen. However, the results of this study provide clear evidence that obscuring (1) the true identity of an information source, (2) the affiliations of an information source, and (3) control over user-generated content can greatly influence consumer health knowledge and behavior.