Scenario-Based Messages on Social Media Motivate COVID-19 Information Seeking

Abstract

Communicating information about health risks empowers individuals to make informed decisions. To identify effective communication strategies, we manipulated the specificity, self-relevance, and emotional framing of messages designed to motivate information seeking about COVID-19 exposure risk. In Study 1 (N=221,829), we conducted a large-scale social media field study. Using Facebook advertisements, we targeted users by age and political attitudes. Episodic specificity drove engagement: Advertisements that contextualized risk in specific scenarios produced the highest click-through rates, across all demographic groups. In Study 2, we replicated and extended our findings in an online experiment (N=4,233). Message specificity (but not self-relevance or emotional valence) drove interest in learning about COVID-19 risks. Across both studies, we found that older adults and liberals were more interested in learning about COVID-19 risks. However, message specificity increased engagement across demographic groups. Overall, evoking specific scenarios motivated information seeking about COVID-19, facilitating risk communication to a broad audience.

Keywords: COVID-19; episodic specificity; information seeking; public health; risk communication; social media.

Figure 1.

Social media ad experiment conditions. Clicking on any ad version redirected users from Facebook to the COVID-19 Event Risk website. Specificity ranged from national (A) to local (B, C) and scenario (D, E) contexts. We also manipulated the emotional valence of the ads by varying the subheadings (Negative: B, C; Positive: D, E).

Figure 2.

A) Average click-through rate (unique link clicks / unique users shown the ad) for the five ad conditions. B) Parameter estimates from a logistic regression model predicting the likelihood of a user clicking on an ad after viewing it. Odds ratios are calculated relative to the following reference groups: Middle-Aged (age), Liberal (politics), National (specificity), and Negative (valence). Error bars indicate 95% confidence intervals. *** p < 0.001

Figure 3.

Comparing user engagement and intervention efficacy across political groups. Error bars indicate 95% confidence intervals, black dotted lines mark zero, and black horizontal bars indicate means. Violin plots depict data distributions (blocked for integer variables, smoothed for continuous variables). A) Average click-through rate by demographic group. Conservatives and older adults were more likely to click on the ads. B) Liberals and conservatives both reported significant decreases in willingness to take risks after viewing the risk map. Negative values on the y-axis indicate being less willing to participate in events that could spread COVID-19. C) Liberals and conservatives both significantly underestimated risk in the risk quiz. Risk estimation error is the average discrepancy between user-estimated exposure risk (“guess”) and prevalence-based exposure risk estimates (“actual risk”). Negative values on the y-axis indicate risk underestimation, positive values indicate risk overestimation, and the dotted line at zero indicates accurate estimation. D) Liberals and conservatives who had been underestimating risk (defined as risk estimation error scores < −10, continuous variable binned for visualization only) reported decreases in willingness after viewing their risk quiz feedback. E) Liberals who were relatively accurate at estimating risk (defined as risk estimation error scores between −10 and +10) still showed a decrease in willingness after receiving feedback, whereas this shift was not significant for conservatives. F) Neither liberals nor conservatives who had been overestimating risk (defined as risk estimation error scores > +10) showed a significant change in willingness after receiving feedback. n.s. = not significant, ~ p < 0.10, ** p < 0.01, *** p < 0.001

Figure 4.

Specificity increased engagement with COVID risk information. Left: In Study 1 (observed naturalistic behavior), participants were substantially more likely to click on more specific ads. Right: In Study 2 (self-reported interest, ranging from 1–5), participants reported greater interest in learning about COVID-19 risks after viewing more specific ads. Note that variance was greater and effect sizes were smaller in Study 2, likely due to the smaller sample, different dependent measure, and changing COVID zeitgeist. Error bars depict 95% confidence intervals.