April 4-6, 2024 • Hyatt Regency • Lexington, KY
Innovations in Health Communication
Innovations in Health Communication
Abstract: Localness V Autonomy: Statistical Methods for Identifying Promising Message Strategies to Reduce Vaccine Hesitancy
Vaccine hesitancy remains a constant problem in the United States, and messaging research may provide promising strategies to increase parent’s comfort in getting their children vaccinated. In this mixed-methods study, we first identified two promising message factors through focus groups with rural parents: (a) levels of “localness” varying in social/geographic scale (provider-patient setting, community-level, and national) and (b) parental autonomy (affirming vs. paternalistic language). Using an online survey experiment, this study tests the effectiveness of these targeting strategies in improving parental flu and COVID-19 vaccine intentions. We first report the interaction effects between these two factors using ordinarily least squares regression, and then we additionally use a Bayesian framework for understanding the overall results.
Methods
Sample and experimental design. A total of 2043 participants (mean age = 39, female = 46%, rural = 42.8%) from seven states with substantial rural population and vaccine hesitancy completed an online survey in Fall 2023. Participants were randomly assigned to one of the 17 conditions in this 2 (vaccine type: influenza vs. COVID-19) x 2 (parental autonomy: confirming vs. paternalistic language) x 4 (“localness” level: clinic vs. community (abstract) vs. community (concrete) vs. national) with a no-message control (Figure 1).
Participants answered demographic and flu-history questions and were shown three different messages randomly selected from a large condition-specific stimuli pool. After seeing all three messages, participants reported their vaccine intentions for their children.
Message stimuli. We created short textual messages promoting pediatric COVID-19 and influenza vaccination. Once a base set of 16 messages was constructed, we used ChatGPT (ver. 4.0, OpenAI 2023) to produce a set of 10 similar messages for each condition. Attached images and logos reinforced signals of localness (e.g., images of a physician for clinic-level messages) consistent with the experimental condition.
Statistical analysis. Focusing on flu vaccination messages, we first used OLS regression with robust standard error (Blair et al., 2023) to predict vaccine intention, vaccine attitude, and overall confidence in attitudes using localness level and parental autonomy as independent variables of interest. We then repeat these analyses under a Bayesian, stan framework using brm (Bürkner, 2017) and uninformative normal priors as a methodological comparison.
Results
We regressed the experimental conditions on three outcomes, compared to the no-message control condition. We found that the autonomous-clinical (b=0.318, p=.011), and paternalistic-concrete-community (b=0.35, p=.009) conditions significantly increased a parent's intention to vaccinate their children. The autonomous-clinical condition also significantly increased flu attitude and general vaccine confidence (Table 1) compared to the no-message control.
With Bayesian regression, we estimate not just a regression coefficient, but a full posterior distribution. The distribution for each estimate is found in Figure 2, and distribution parameters are in Table 2.
Conclusion
In pursuit of understanding ways to influence parental vaccine confidence, we explored a new relevant concept of localness. Contrary to the feedback from focus groups, autonomy-affirming messages did not outperform paternalistic messages. This paper brings innovations in methodology by using Bayesian statistics to study health message effects, and in content by exploring the concepts of localness.
Methods
Sample and experimental design. A total of 2043 participants (mean age = 39, female = 46%, rural = 42.8%) from seven states with substantial rural population and vaccine hesitancy completed an online survey in Fall 2023. Participants were randomly assigned to one of the 17 conditions in this 2 (vaccine type: influenza vs. COVID-19) x 2 (parental autonomy: confirming vs. paternalistic language) x 4 (“localness” level: clinic vs. community (abstract) vs. community (concrete) vs. national) with a no-message control (Figure 1).
Participants answered demographic and flu-history questions and were shown three different messages randomly selected from a large condition-specific stimuli pool. After seeing all three messages, participants reported their vaccine intentions for their children.
Message stimuli. We created short textual messages promoting pediatric COVID-19 and influenza vaccination. Once a base set of 16 messages was constructed, we used ChatGPT (ver. 4.0, OpenAI 2023) to produce a set of 10 similar messages for each condition. Attached images and logos reinforced signals of localness (e.g., images of a physician for clinic-level messages) consistent with the experimental condition.
Statistical analysis. Focusing on flu vaccination messages, we first used OLS regression with robust standard error (Blair et al., 2023) to predict vaccine intention, vaccine attitude, and overall confidence in attitudes using localness level and parental autonomy as independent variables of interest. We then repeat these analyses under a Bayesian, stan framework using brm (Bürkner, 2017) and uninformative normal priors as a methodological comparison.
Results
We regressed the experimental conditions on three outcomes, compared to the no-message control condition. We found that the autonomous-clinical (b=0.318, p=.011), and paternalistic-concrete-community (b=0.35, p=.009) conditions significantly increased a parent's intention to vaccinate their children. The autonomous-clinical condition also significantly increased flu attitude and general vaccine confidence (Table 1) compared to the no-message control.
With Bayesian regression, we estimate not just a regression coefficient, but a full posterior distribution. The distribution for each estimate is found in Figure 2, and distribution parameters are in Table 2.
Conclusion
In pursuit of understanding ways to influence parental vaccine confidence, we explored a new relevant concept of localness. Contrary to the feedback from focus groups, autonomy-affirming messages did not outperform paternalistic messages. This paper brings innovations in methodology by using Bayesian statistics to study health message effects, and in content by exploring the concepts of localness.