We conducted a within- and between-subjects, randomized controlled cross-sectional, web-based study. All data were collected on the internet and no identifying information was collected in order to protect participants’ privacy and confidentiality. The study was reviewed by the institutional review board of Indiana University (protocol# 2004499544) and was determined to be an exempt study. The study was initiated with the institutional review board approving the study data and if participants consented to participate, they were enrolled in the study.

In July, 2020, we recruited 300 participants from Amazon Mechanical Turk in accordance with the usual protocols to ensure data quality [40]. We recruited participants only from the United States who held an Amazon Masters classification and included a captcha to preclude nonhuman responses. Participants were paid US $1.25 and spent an average of about 13 minutes participating in the study (minimum: 3.6 minutes, maximum 11 hours). In total, 8 subjects failed 1 or more of the 3 attention checks (that asked participants to select specific answers), thus yielding 292 participants. All participants received both the control and treatment conditions; hence, demographics are described at the study level. Approximately 49% of participants were female, and 82% were White, 8% were Asian, 7% were Black, and 3% were of other racial backgrounds. The median age of the study participants was 30 years (24%: 18-24 years, 33%: 25-34 years, 19%: 35-44 years, 16%: 45-54 years, 6%: 55-64 years, and 1%: ≥65 years).

At the beginning of the survey, subjects were asked a series of questions to ascertain their identification with Christianity and with the economic health of the country (our selected identities). Participants were provided either the Christian-framed treatment or economy-framed PSAs depending on which identity they identified with more. If subjects identified with both identities equally, they were randomly assigned to 1 of the 2 treatments. The experimental design is presented in Figure 1.

Participants then received two COVID-19 PSAs in random order: 1 advocating wearing a mask and 1 calling on people to stay at home. One was a control PSA with information purportedly from the US Public Health Service and the other was an identity-framed PSA (either Christian-framed or economy-framed). After reading each PSA, participants reported the extent to which they would engage in the advocated behavior. Each subject received 1 identity-framed PSA and 1 control PSA; thus, we could examine the within-person effects of PSA framing and could control for differences in compliance between mask-wearing and staying at home.

The PSAs were of approximately 100 words. The treatments changed the source of the PSA and 1 sentence of their content. The identity-framed PSA contained a single short, substituted phrase designed to appeal either to people who held a Christian social identity (purportedly written by the Chaplain of the US Senate) or to those for whom protecting the country’s economy was a central feature of their personal identity (purportedly written by the US Chamber of Commerce). For example, the control PSA for wearing a mask stated, “You should wear a mask whenever you are in public and see other people,” which was replaced by, “We have a Christian duty to love our neighbors, and wearing a mask whenever you are in public and see other people is a way you can do this” in the Christian-framed PSA and by, “We now know how you can do your part to help us safely reopen our economy: wearing a mask whenever you are in public and see other people” in the economy-framed PSA. All PSAs are provided in Multimedia Appendix 1.

In accordance with previous studies, Christian identity, economically motivated identity, and trust in the 3 PSA sources were measured with single-item 7-point Likert scales [41-43]. Identity-framed PSAs were deemed to be aligned with the participant’s identity when the participant’s identity scores were 6 or 7 and the participant did not distrust the source (ie, trust in the source was ≥4). The likelihood of compliance was measured using 7 items adapted from previous studies [44,45]. The outcome compliance items were measured using scales of 0-100, not the same 7-point scales as those used for the independent variables to classify participants, to reduce the risk of a common method bias [46]. The likelihood of compliance proved reliable (Cronbach α=.94), thus indicating convergent validity. An exploratory factor analysis was performed to assess the discriminant validity among the constructs, which was satisfactory. The results of the items and factor analysis are presented in Multimedia Appendix 1.

A power analysis using G*Power [47] determined that a sample of 300 participants with this design would provide a power of .93 to detect a small effect size (Cohen f=0.10). We analyzed the data using hierarchical linear modeling (HLM) [48] with robust standard errors using HLM for Windows (version 6.00, Scientific Software International, Inc). HLM accounts for the correlation among the repeated within-subject observations and facilitates the assessment of the alignment or nonalignment of the identity-framed treatment condition with the participant’s identity. All tests were 2-tailed with a significance level of α=.05 α=.05. We used the formula of Snijders and Bosker [49] to calculate R². The likelihood of complying was nonnormally distributed. For the PSA advocating mask-wearing, the mean likelihood of compliance was 78.75 (SD 29.79), and the median likelihood was 92.93; for the PAS advocating staying at home, the mean likelihood of compliance was 74.51 (SD 29.67), and the median likelihood was 86.75. HLM is robust to departures from the normality assumption for large samples such as this one [48].

Our study shows that modifying PSAs to leverage social and personal identities can promote increased compliance with public health guidelines for individuals who hold these identities. The Christian-framed PSA increased compliance by approximately 12.74 points (out of 100) when viewed by those with a Christian identity, and the economy-framed PSA increased compliance by approximately 6.77 points when viewed by those with an economically motivated identity. One might conclude that the Christian framing is more powerful, but this only applies to individuals who hold that identity. Hence, we believe that it is better to compare the 2 different identity-framed PSAs to their own controls, not to each other, as the differences in the coefficients between the 2 identity-aligned PSAs may have resulted from various subject-level factors that influence their different identities. For example, those with a Christian identity were slightly less likely to comply (P=.01) with any PSA, possibly because those with conservative beliefs are less likely to believe that COVID-19 is a legitimate threat [1,14].

Based on the social identity theory, we hypothesized that messages designed to activate an identity by using inclusive language and a consistent message source would be more effective in increasing compliance to PSAs rather than those without consistent identity-framing. Identity-framing was intended to emphasize commonalities between the individual and like-group members and to encourage users to act in accordance with those in their group [1]. Identity-framing harnesses the relevant identity that the individual holds and appeals to those relevant traits. This simple act of creating identity-aligned targeted PSAs significantly increased compliance with the behavior advocated in the PSAs. We decided to examine a Christian identity and an economically motivated identity on the basis of excuses that are commonly invoked to justify noncompliance [36-38]. We found both to be effective in increasing compliance. Various other social or personal identities may also be effective in increasing the compliance to PSAs.

The promise of identity-framed PSAs is noteworthy in view of the myriad of rampant rumors, misinformation, and disinformation regarding COVID-19 [12-14]. Rapidly developing situations, uncertainty, and fear foster the spread of false information (created with or without the deliberate intention to mislead people). It is unfortunate that individuals’ responses to COVID-19 have implied that the provision of simple, information-based PSAs sometimes triggered psychological reactance and led to actions that disrupt public health efforts [1,13]. Our results show that designing PSAs to appeal to specific target demographics can increase their effectiveness beyond that of a message that simply provides correct information.

The vast amount of information about individuals available on social media platforms makes it practical to create multiple versions of a PSA and share the most individually relevant version with people, thereby making the message more persuasive [50]. Social media is an attractive channel to rapidly reach many people as it has more than 2 billion active users [50]. Identity-framed PSAs could facilitate public health goals by enabling the PSA to influence people who would otherwise ignore the message, capturing their attention by speaking their language, activating relevant identities, and reducing psychological reactance by framing the actions as being consistent with their identity [25,35]. By encouraging people to view a situation through the lens of a supportive identity, the effects of countervailing identities that dissuade people from the desired outcome can be reduced. Public Health agencies and nonprofits should take advantage of these tools when designing future public awareness campaigns. By leveraging readily available identity information to make minor adjustments to the framing of PSAs, such groups could facilitate higher compliance with public health guidelines, thus enabling better outcomes. With increasing ease of accessing personal data, the small cost of targeting a PSA toward those individuals that would best respond (similar to targeted advertising) has the potential to yield enormous benefits with increased nationwide health outcomes.

We examined 1 social identity (Christian) and 1 personal identity (economically motivated) linked to noncompliance [36-38]. Many other identities may also be leveraged to enhance the effectiveness of PSAs. By reminding people of a role identity as a parent or grandchild, leveraging social group identities such as sport team loyalty, or by appealing to their self-identity as caring individuals, many opportunities are available to use identities to help persuade people to follow public health guidelines. Our use of 2 identities shows that this method can be successful; however, it is unclear which other identities may also be used. Future studies are required to investigate why some identities may be effective when used to increase compliance while others may not be effective. Furthermore, it remains unclear whether compliance is affected more by the relationship between the identity and the target behavior or that between the identity and the individual, or whether both relationships are equally important.

While we control for trust toward the source of the PSA, a potentially interesting question is one regarding the mediating influence of variables (eg, perceived similarity to the identity used in the PSA). While we did not assess perceived similarity, we controlled for the strength of the identity. Since our study suggests that identity-framed PSAs can be used to influence compliance, more studies are needed to investigate the potential mediating and moderating variables that could strengthen or weaken the effectiveness of identity-framed PSAs. For example, our identity-framed PSA contained only 1 modified sentence; however, it remains unclear whether adding more identity-related framing would increase its effectiveness, or whether it is sufficient to simply invoke the identity as in our PSAs. Moreover, it remains unclear whether the source of the message is a critical factor, or whether the message more important than the source.

This study has the usual limitations of randomized controlled cross-sectional study. We assessed self-reported perceptions at one point in time, not actual behavior over several periods of time. Our participants were those who participate in research studies and thus may differ from those who decline to participate in research studies.

Compliance with public health measures designed to mitigate the COVID-19 pandemic has unfortunately become intertwined with identity, and individuals with certain social and personal identities are less likely to comply with the behaviors advocated in the PSAs. Our study shows that identity can also be an effective factor to induce compliance. The development of identity-framed PSAs may be effective in contexts beyond the COVID-19 pandemic. The most important factors to consider when developing effective PSAs are that the identities in question are deeply held and can be associated with recommended actions when coordinated efforts across society are needed.