From March 17 to May 7, 2020, participants were recruited from the general community via advertisements placed in Facebook and WhatsApp community groups (eg, residential groups, workplace groups, and university groups), posts on popular web-based forums, and paid Facebook advertisements targeting Singapore-based users. All study activities took place on the web-based survey platform Qualtrics (Qualtrics International Inc), and participants were reimbursed with SGD $5 (US $3.65) upon study completion. The study protocol was approved by the Yale-NUS (National University of Singapore) College Ethics Review Committee (protocol record: 2020-CERC-001) and was preregistered at ClinicalTrials.gov (trial number: NCT04367363).

The participants were 151 adults who met the following inclusion criteria: (1) aged 21 years or older, (2) had lived in Singapore for at least 2 years, and (3) had a WhatsApp account.

Following the provision of informed consent, participants completed (1) a baseline questionnaire, (2) experience sampling responses daily for 7 days, and (3) a final questionnaire (Figure 1).

First, we summarized the data as counts with percentages or as means with SDs, focusing on the following seven quantitative WhatsApp usage variables (Figure 2): the number of (1) COVID-19 messages that participants forwarded, (2) groups to whom messages were forwarded, (3) individuals to whom messages were forwarded, (4) forwarded messages received, (5) individuals from whom messages were received, (6) personal chats involving COVID-19–related conversations, and (7) group chats discussing COVID-19. Multimedia Appendix 1 shows the pattern of correlations across these variables.

Second, to understand WhatsApp user profiles, we performed a latent profile analysis to create a taxonomy of participants based on their WhatsApp usage (R package mclust [18]). Latent profile analysis is a bottom-up statistical clustering method for defining classes of people based on common characteristics. By using all observations of a continuous dependent variable, classes are created such that within each class, indicator variables are statistically uncorrelated [19]. We thus used this technique to cluster participants based on their responses to the seven WhatsApp usage variables; values were obtained by aggregating the reported frequency of each variable over the week. To uncover clusters, we used Gaussian mixture models and assigned participants to clusters by using Bayesian probabilities. The final number of clusters was determined by using the Bayesian information criterion, the integrated completed likelihood criterion, and a bootstrap likelihood test.

Finally, we examined whether day-to-day variations in COVID-19 WhatsApp chatter could be tracked based on variations in COVID-19 concerns. We quantified COVID-19 chatter on personal and group chats, so that such chatter could be used as predictors. For personal chats, the following variables were summed for each day and for each participant: the number of (1) individuals to whom COVID-19 messages were forwarded, (2) individuals from whom forwarded messages were received, and (3) personal conversations discussing COVID-19. For group chats, the following variables were summed: the number of (1) groups participants to whom COVID-19 messages were forwarded and (2) groups where COVID-19 messages were mentioned. Scores were grand-mean centered by subtracting the mean number of chats across subjects and time points from each score (number of chats: mean 2.47 and mean 1.29, respectively). In addition, we created between- and within-subject versions of each predictor [20]. The final analyses involved linear mixed-effects models for each outcome measure (fear of and thoughts about COVID-19). The following were entered as fixed effects: time (centered such that 0 referred to the middle of the week), daily personal chats (between subjects), daily personal chats (within subjects), daily group chats (between subjects), and daily group chats (within subjects). Random intercepts accounted for correlated data resulting from repeated measures.

Across all analyses, the type 1 decision-wise error rate was controlled at an α of .05. All statistical analyses were conducted in R 3.5.0 (R Foundation for Statistical Computing) and SPSS 25 (IBM Corporation).

As shown in Table 1, 68.9% (104/151) of participants were female, and their mean age was 36.35 (SD 14.7) years. Participants were predominantly of Asian ethnicity (Chinese: 140/151, 92.7%) and had at least a postsecondary education (133/151, 88.1%). Further, 39.7% (60/151) of participants were married, and the majority (105/151, 69.5%) belonged to households of at least 4 members.

Participants’ self-reports revealed that WhatsApp was the second most common source for COVID-19 news after news websites or apps (Figure 3). By quantifying this through 1 week of experience sampling, we found that nearly all participants (143/151, 94.7%; 95% CI 90-98%) reported at least 1 COVID-19 related use of WhatsApp. Namely, around 1 in 2 participants (79/151, 52.3%; 95% CI 44%-60%) forwarded at least 1 COVID-19 message (to either individuals or groups), 78.1% (118/151; 95% CI 71%-84%) received at least 1 forwarded message in personal chats, 66.2% (100/151; 95% CI 58%-74%) engaged in personal chat conversations about COVID-19, and 88.1% (133/151; 95% CI 82%-93%) had been in groups where COVID-19 was mentioned.

Figure 2 shows the extent to which participants engaged in each of these activities. On average, participants (1) received 2.3 times more messages than they forwarded and (2) were more likely to forward messages to individuals than to groups (average of 5.3 messages per week vs 2.7 messages per week, respectively). Beyond passive engagement, participants also took part in an average of 3.8 one-to-one conversations about COVID-19 during the week; however, these interactions occurred less frequently than the sending or receiving of forwarded messages in group chats.

The ongoing pandemic has drawn attention to the role of social media in public health. Against this backdrop, we present the first infodemiological study to document the spread of COVID-19 content through WhatsApp. By tracking daily WhatsApp usage for 1 week, we found that (1) nearly every participant engaged in COVID-19 chatter and (2) participants were more likely to share or receive forwarded messages than to engage in conversations about COVID-19.

The volume of forwarded messages that we observed raises concern. On other social media platforms, forwarding behaviors have been linked to the spread of misinformation. For example, a study of 4.5 million Twitter posts found that misinformation was 70% more likely to be shared than posts that were true; correspondingly, any single retweet had a higher probability of containing false news rather than truthful news [15]. Although analogous research has not been conducted on WhatsApp, the app’s developers have likewise deemed forwarded messages as a high-risk source of misinformation [4,5]. 

With regard to the extent that forwarded messages carry misinformation [1,2], our latent profile analyses revealed that around 1 in 10 (21/151, 13.9%) participants were chronic users who received and shared a large volume of such messages. Notably, chronic users disseminated an average of 14 forwarded messages during the week, which is approximately 5 times the number of messages that were sent by all participants in this study. This is reminiscent of research on other social media platforms (eg, Twitter) where a small group of super sharers and super consumers are responsible for the bulk of shared misinformation [22]. Given the potential influence of this group, further research is needed to understand (1) the profile of chronic users, (2) the reasons why they forward messages, and (3) how their forwarding activities may influence outcomes during health crises.

Aside from chronic users, our study also found that around 1 in 3 (47/151, 31.1%) participants were receiving users who had high exposure to forwarded COVID-19 content. Receiving users tended to be older (in line with misinformation studies on Facebook [23]) but were otherwise moderate in terms of their profiles, that is, in terms of COVID-19–related fears or religion (they came from a diverse religious background). Although this group did not spread forwarded messages themselves, their high exposure may nonetheless leave them susceptible to false beliefs. Correspondingly, we urge researchers to conduct further research to understand how WhatsApp use among receiving users influences health behaviors.

Finally, we also found that WhatsApp users who discussed COVID-19 in their personal chats were more likely to think about COVID-19 throughout the day. As similar forms of rumination (involving frequent and persistent thoughts) have been linked to clinical depression [24], this finding may implicate COVID-19 chatter as a risk factor for poorer well-being [25]. Future studies should thus explore this possibility and the potential mechanisms involved.

Taken together, our findings on WhatsApp message transmission have several implications for public health responses during a crisis. First, our taxonomy of user profiles provides a basis for targeted risk communication. Our findings suggest that public health agencies may need to reach out proactively to chronic and receiving users, who handle the bulk of forwarded COVID-19 content on WhatsApp. One possible intervention may be encouraging these users to subscribe to official WhatsApp channels for updates (eg, updates from the World Health Organization) [2] to capitalize on their pre-existing readiness to use the platform.

Tracking WhatsApp chatter may also result in new opportunities for detecting disease outbreaks. The nascent field of digital epidemiology seeks to model how diseases spread by monitoring digital data sources (eg, through Google search data and Twitter posts) [26]. Although the content of private WhatsApp messages is difficult to track, the volume or nature of messages (eg, forwarded messages) may provide information that can be used to support disease surveillance. To this end, further research is needed to explore the predictive utility of message transmission dynamics on WhatsApp.

In reporting these findings, we noted several limitations. First, we opted to study WhatsApp—the most widely used messaging app. At this juncture, it is unclear whether our results generalize to other messaging apps (eg, Facebook Messenger and Telegram).

Second, our recruitment strategy was limited by the nature of the pandemic. Owing to infectious disease protocols and the short time period when the amount of crisis-related communication was high [27], our data collection process was restricted in terms of the recruitment strategy (web-based sampling), sample size (151 participants), and time frame (1 week per participant). Further research is needed to examine whether our findings generalize to the broader population.

Finally, although the experience sampling method captured WhatsApp usage in participants’ naturalistic settings, the method nonetheless required self-reports. By extending our findings, future studies will benefit from having objective metrics of WhatsApp usage.

In conclusion, we used the experience sampling method to capture COVID-19 chatter on WhatsApp for the first time. In total, we tracked 924 days’ worth of chatter in situ, revealing (1) the sheer prevalence of WhatsApp usage, (2) a typology of WhatsApp users, and (3) a link between usage patterns and constant thoughts about the pandemic. These findings have implications for health communication and disease surveillance, bringing the field 1 step closer to characterizing WhatsApp usage and using these data to gain insights on individual and societal concerns.