Role of Social Media in the COVID-19 Pandemic

Social media has played a large role in disseminating pandemic information, but it has also been used to spread misinformation [3,24], such as lack of severity of COVID-19, false virus transmission methods, ineffective prevention and diagnostic methods, unproven/pseudoscience treatments, risks from testing and face masks, and other conspiracy theories [25-28]. Misinformation has also spread about the COVID-19 vaccine, such as claims that vaccine safety was compromised by the rush to market, that the low risk from COVID-19 and effective prevention and treatment make vaccines less necessary, and that variation in the amount and length of effectiveness indicates vaccines are not useful [13,17]. Lower uptake of vaccines in general and lower COVID-19 vaccine intentions have been related to misinformation, unwarranted safety concerns, and conspiracies on social media, as has the practice of NPIs [29,30]. Thus, efforts are needed to promote COVID-19 prevention measures and correct misinformation on social media through fact checking and corrections, counternarratives, peer correction, coherence/credibility appeals, and digital and media literacy [31-38].

Impact of Sources for COVID-19 Information

The Extended Parallel Process Model (EPPM) of risk communication [39], an extension of protection motivation theory (PMT) [40,41], has explained mitigation behaviors in past pandemics, uptake of other vaccines [42-44], and COVID-19 pandemic responses [45]. It holds that the credibility of information sources influences the effectiveness of health messages [46]. High-credibility sources make it difficult for campaign audiences to derogate sources in order to decrease fear from risk information about COVID-19. In this way, messages from high-credibility sources motivate individuals to take actions that reduce risk with NPIs and vaccines.

We experimentally varied 3 types of sources, popular for information about the pandemic [47-49], that can vary in credibility (eg, trustworthiness and accuracy) in the social media posts on COVID-19: government agency, near-peer parents, and news media. Government health authorities are trusted sources of COVID-19 information for many (but not all) people [50,51], with nongovernmental content and unverifiable sources seen as less trustworthy, especially when posted on social media platforms [52,53]. A cross-sectional study of COVID-19 information sources found that attention to government sources is linked to greater COVID-19 knowledge [50]. Content shared on social media from (perceived) knowledgeable peers can have credibility and impact through identification processes based on similarity [54-57]. Peers (eg, friends, family, and work colleagues) have also been an often-used source of information about COVID-19, although they are not always as trusted as government and news media sources [48,51]. Consumers evaluate the credibility of both the source and message content of news media [58]. One study found that exposure to news media reduces conspiracy theories and misinformation beliefs regarding COVID-19 [59], but another reported that COVID-19 knowledge is lower among individuals who have greater trust in these sources [50]. The availability of a variety of information sources can elevate risk perceptions and fear; create information overload, anxiety, stress, and other negative psychological states; and possibly cause people to avoid information [45,47,48,60].

Experimental Design

Mothers were enrolled in a randomized pretest–posttest single-factor-design study with 4 assessments. After completing the baseline survey, mothers were randomly assigned to 1 of 3 experimental conditions that varied in the type of sources in the posts (government health agencies, near-peer parents, or news media) using a routine in Qualtrics survey software. Mothers “friended” the project community manager and were added to a Facebook private group for their assigned condition. As all mothers received experimental social media messages, they were blind to experimental manipulation of the information source. Study staff, other than the community manager and project manager, were blinded, too. The private groups prevented contamination between treatment groups while delivering the social media messages and made it possible to record engagement. Randomization controlled for background secular exposure to information in social media and other sources about COVID-19. Mothers received a series of Facebook posts for 9 weeks starting after randomization from January 25 to March 26, 2021. Each post contained text with a link to related information from 1 of the 3 types of sources. Mothers stayed in the groups for 9 weeks, completing online posttests at 3, 6, and 9 weeks postrandomization. After the intervention, 30 (9.9%) of 303 mothers were randomly selected to participate in focus groups, where they were asked what they liked most and least about the Facebook group and what they learned. A priori statistical power calculations via a Monte Carlo study in Mplus and with the powerlmm package [71] in R software (R Foundation for Statistical Computing) indicated that an initial sample size of 300 mothers (100 per condition) would have 0.90 power to detect a moderate-size rate of increase in vaccine intention (Cohen d=0.50). Retention was achieved by alerting mothers to upcoming posttests and compensating mothers for assessments (US $20 for baseline, US $10 for each posttest). Mothers also received 1 raffle entry for every survey completed in drawings for 20 US $100 gift cards after the final posttest.

Ethical Considerations

Mothers provided informed consent online before completing the baseline survey. The study procedures were approved by the Western Institutional Review Board (1-872442-1).

Intervention

The intervention contained 45 Facebook posts related to COVID-19 (5, 11.1%, per week) designed by the research team based on the EPPM [39] and SCT [65]. Posts addressed 4 topics: the 2 primary outcomes (NPIs and COVID-19 vaccination), digital and media literacy, and mother–daughter communication. These topics were rotated across weekdays by week to ensure that all topics had the same likelihood of being viewed. Posts on digital and media literacy were included to combat misinformation related to NPIs and vaccines by addressing source credibility, fact checking, lateral reading, sharing of posts with family/friends, social media algorithms, rebutting of misinformation, and deep fake videos [72-74]. Posts encouraged mothers to talk with teen daughters about the pandemic and promote prevention behavior [66-68] and sought to improve this communication by teaching skills, such as active listening, self-disclosure, empathy, and conflict management. Across these topics, posts addressed theoretic antecedents, including risk from COVID-19 (ie, severity and susceptibility), self-efficacy and response efficacy of NPIs and vaccination, descriptive norms for NPIs and vaccination, behavioral capability (knowledge of risks of COVID-19 and skills to practice NPIs), and observational learning (stories about dangers of COVID-19 and skills related to NPIs, vaccination, and family communication). To increase mothers’ engagement, posts encouraged mothers to react to (eg, like) and comment on posts, for example, by asking a question to solicit the mothers’ own experience and opinions on a topic. Additionally, 12 posts provided study information or were aimed at engaging mothers with holiday plans, favorite books, family traditions, and recipes.

Each experimental post contained the same content in all 3 groups. The experimental manipulation of information sources was accomplished by linking each message in the posts to additional online content (eg, articles, blog posts, infographics, or videos) from either a government agency (eg, the CDC or the World Health Organization [WHO]), a near-peer parent, or news media. For the near-peer parent group, information was sourced primarily from Twitter, Instagram, Facebook, TikTok, and parenting blog posts or magazines. Near-peer parents were predominantly women. The term “near-peer” was used to reflect that these sources were similar to the participants, being obviously parents (although a few were female journalists, college professors, or nurses), and were selected to be close to the age of the sample (range 28-64 years, mean 42.7, SD 6.7). However, these sources were unlikely to be known personally by participants, as might be a “peer.” For news media, content was sourced from 22 media organizations that focused on delivering news to the general public or a target public. Since individuals can differ in the credibility they assign to various news media, we selected content from news media that ranged from moderately conservative (eg, Fox News and the New York Post) to middle-of-the-road (eg, USA Today and Newsweek) to moderately liberal (eg, Washington Post and ABC), as ranked by All Sides Media Bias [77]. The research team confirmed that all links and content from information sources were accurate. Some of the content from the source was embedded in the experimental post (eg, infographic or screenshot), but a link was always provided to the information source.

Posts were developed by the investigators using an agile development process to reflect the rapidly changing pandemic information environment and ensure content was timely and relevant. Mothers (n=30, 9.9%) participated in virtual focus groups before and during the intervention to review and provide feedback on sample posts. Initially, 2 weeks of posts were prepared, after which new posts were developed weekly. All posts were reviewed by 4 of the investigators (authors DB, BW, WGW, SP), the project manager, and the community manager for readability, theoretical principles, accuracy, and information source prior to posting.

Posts were scheduled by the community manager. They appeared at 10:00 a.m. on Monday, Wednesday, and Friday and 7:00 p.m. on Tuesday and Thursday (1 post per day). Posting times were based on analytics from our prior study regarding the most popular times to view posts [69]. The initial post welcomed participants to the group, invited them to join in discussing the posts, and asked them to be respectful of other group members during discussions and to maintain the privacy of other participants when they communicated about content in the posts with family and friends outside the group. Posts on the 4 topics (NPIs, vaccination, digital and media literacy, and mother–daughter communication) appeared each week (1 post on each of the 3 topics and 2 posts on 1 topic in a week; topics with 2 posts were rotated across the weeks). On Wednesdays, an additional engagement post was published (n=12) to balance the seriousness of the pandemic topics and help keep mothers engaged. The community manager followed a protocol to monitor mothers’ reactions and comments to each post and respond to any uncertainty or misinformation or requests for additional information from mothers. Responses had a respectful, empathy-driven, reflective-listening approach toward the mothers [76] that acknowledged the mothers’ comments, advised them to follow local and national COVID-19 guidelines, and included links to government agencies, professional groups (eg, the American Diabetes Association), and news media.

Measures

All measures were self-reported by mothers and collected using Qualtrics survey software (see Multimedia Appendix 1).

Hypothesis 1 Test: Change in Social Distancing and Vaccine Intentions

At baseline, most mothers reported that they and their daughters were engaging in a moderate to high levels of social distancing (Table 4). Mothers’ reports of social distancing by both themselves and daughters decreased over time when examining all 3 posttests relative to baseline (Table 5), disconfirming H1.

About half of the mothers had high vaccine intentions for themselves and their daughters, but up to one-quarter expressed low vaccine intentions (Table 4). Vaccine intentions for self and daughters increased over time (Table 5), supporting H1. However, vaccine intentions were bimodally distributed, with large groups of mothers consistently indicating low (<20.00 likelihood) and high (80.00 likelihood) intentions across all 4 time points. Thus, baseline vaccine intentions were split into 3 groups (low<20.00, moderate=20.00-79.00, and high80.00 likelihood) and tested as a moderator of change in the 5-level vaccine intention measure in the 3 posttests. There was a statistically significant improvement in vaccine intentions for self (b=0.76, 95% CI 0.31-1.21, P<.01) and daughters (b=0.48, 95% CI 0.06-0.89, P=.02) over time among mothers with moderate intentions at baseline. Likewise, there was a statistically significant increase in vaccine intention for self (b=9.21, 95% CI 6.60-11.82, P<.001) and daughters (b=5.51, 95% CI 3.78-7.23, P<.001) by the 9-week posttest among mothers with high baseline intentions. Mothers with low baseline vaccine intentions reported lower vaccine intention for self (b=–5.99, 95% CI –8.03 to –3.95, P<.001) and daughters (b=–4.83, 95% CI –6.69 to –2.97, P<.01) in the 9-week posttest.

Hypothesis 2 Test: Change in Theoretic Antecedents and Mother–Daughter Communication

Several theoretic antecedents improved over time (Table 4), largely supporting H2. Specifically, self-efficacy for vaccination of self and daughters increased, and response costs for NPIs decreased. There was also some evidence that perceived risk increased over time, particularly with the severity of COVID-19 increasing over time; however, perceived susceptibility declined over time. By contrast, self-efficacy and response efficacy for NPIs did not change, nor did mother–daughter communication (Table 5), contrary to the hypothesis.

Differences Among Information Sources

Social Distancing

The number of views of posts by participants did not influence their reports of social distancing by self or daughters, but reports of social distancing by daughters was higher among mothers who had more reactions and comments (b=0.01, 95% CI 0.01-0.01, P=.04). There was no evidence that engagement moderated differences among information sources (P>.05).

Vaccine Intentions

For views, the increase in vaccine intentions for self over time was attenuated when mothers viewed more posts across all conditions (b=–0.01, 95% CI –0.01 to –0.01, P=.01). This attenuation was stronger in the government agency group (self: b=–0.02, 95% CI –0.04 to 0.00, P<.001; daughters: b=–0.01, 95% CI –0.01 to –0.01, P=.01). By contrast, attenuation of the increase in vaccine intentions was less evident in mothers in the near-peer parents group who had more engagement (self: b=0.02, 95% CI 0.00-0.04, P<.01; daughters: b=0.02, 95% CI 0.00-0.04, P<.001). Engagement measured by reactions and comments did not affect changes in vaccine intentions (P>.05).

Baseline COVID-19 Media Exposure

Baseline exposure to COVID-19 information in news media, averaged across 4 items, was similar across conditions on a 5-point scale (government agency mean 4.11, SD 0.88; near-peer parents mean 4.09, SD 0.91; news media mean 4.01, SD 0.82). Social distancing (self: b=0.46, 95% CI 0.36-0.56, P<.01; daughters: b=0.34, 95% CI 0.24-0.44, P<.01) and vaccine intentions (self: b=3.87, 95% CI 2.62-5.12, P<.001; daughters: b=2.80, 95% CI 1.93-3.66, P<.001) were higher at the 9-week posttest among mothers who reported more media exposure at baseline. However, baseline exposure did not affect differences by information source in either outcome.

Political Leaning

Political leaning was normally distributed among mothers within each condition (government agency: conservative n=25, 25.3%, moderate n=54, 54.6%, liberal n=20, 20.2%; near-peer parents: conservative n=25, 25.8%, moderate n=48, 49.5%, liberal n=24, 24.7%; news media: conservative n=22, 22.2%, moderate n=46, 46.5%, liberal n=31, 31.3%). Mothers reported increased social distancing (self: b=0.40, 95% CI 0.28-0.52, P<.001; daughters: b=0.31, 95% CI 0.19-0.42, P<.001) and vaccine intentions (self: b=3.16, 95% CI 1.49-4.82, P<.001; daughters: b=2.37, 95% CI 1.21-3.53, P<.001) over baseline at the 9-week posttest when they expressed a more liberal than conservative political leaning. Political leaning moderated differences by information source for reports of social distancing by daughters. Mothers who were more liberal and assigned to the near-peer parents group reported greater social distancing by daughters at the final posttest (b=0.19, 95% CI 0.01-0.37, P=.04), while more liberal mothers in the government agency group reported reduced social distancing at the final posttest (b=–0.25, 95% CI –0.43 to –0.07, P<.01). Political leaning did not show any other effects on vaccine intentions for self (near-peer parents: b=0.13, 95% CI –0.18 to 0.44, P=.43; government agency: b=–0.11, 95% CI –0.42 to 0.20, P=.50) or daughters (near-peer parents: b=–0.03, 95% CI –0.30 to 0.24, P=.85; government agency: b=0.20, 95% CI –0.09 to 0.49, P=.18).

Limitations and Strengths

The trial had some limitations. The design lacked a control group that did not receive messages on COVID-19, which made it challenging to determine whether the messages affected social distancing and vaccine intentions irrespective of the information source. The short duration of the intervention may have achieved only small effects. Although the sample was moderate in size and from a number of US states, generalizability was limited by enrolling mothers of teen daughters who may have been more attentive to the social media messages because they had elevated concerns about COVID-19 risks for their families. Whether individuals who are not parents would be affected in the same way is unknown. Mothers had already participated in a trial on other adolescent health topics, so the sample may have been biased to mothers with high interest in adolescent health. Most mothers were originally recruited from the Qualtrics survey panel, which tends to have a relatively high socioeconomic status, and nearly all mothers were non-Hispanic White because of the original trial’s focus on indoor tanning. Although we varied the source of information contained in the posts, all posts were delivered through the Facebook platform, making it the primary source of the intervention and possibly undermining the experimental comparison. The multiple posttest measures may have introduced a testing effect (ie, reactivity) that increased the mothers’ attention to the experimental messages because they knew they would be assessed every 3 weeks. All assessments were self-reporting, although many outcomes were intrapsychic processes (eg, perceptions, opinions, and intentions) measurable only through reports from mothers. We did use published scales, when available.

These limitations were offset somewhat by strengths of the study. Mothers were enrolled and pretested prior to the intervention, allowing for prospective tests of social media’s effects, and were randomly assigned to 3 prominent sources of pandemic information, which improved the validity of these comparisons. A mixed methods approach was used to understand the impact of the social media messages on mothers. Finally, multiple posttests provided information on changes produced by the intervention over time.

The authors thank Christie Idiong and Haley Troy for their assistance with coding the focus group transcripts. This research was funded by the National Cancer Institute (Grant CA192652).