Participants

Data was collected from 2,088 mobile survey respondents in Ghana. Respondents were required to provide their consent and be at least 18 years old to participate in the survey. The sample size was determined by comparing 5 experimental treatments to a shared control arm. Using data from similar experiments [38], our computation was based on 80% power to detect a difference in means of 0.4 (M1 = 2.9 M2 = 2.5) for any two arms, assuming a standard deviation (SD) of 1.5 and 2.2 for means respectively on a 5-point Likert scale with a 2-sided significance level of 0.05. Inserting these into STATA’s sample size calculation function (sampsi 2.9 2.5, SD1(1.5) SD2(2.2) power (.80), a sample of 348 for each arm was estimated. The total required sample with six experimental conditions was 2,088.

Data cleaning

In cleaning the data, emphasis was placed on securing high data quality to achieve the goals of the study. A key benchmark was to focus on data from respondents with fully completed survey interviews. Given that with mobile based surveys, there was a possibility of a respondent going through the survey multiple times on different calls, duplicate and incomplete responses from unique respondents were also eliminated. Using this criterion, we excluded data from 594 respondents, resulting in a final sample of 1,494 respondents. All the results reported in this manuscript are from the overall sample with 1,494 respondents or a sub-sample of the same. Of the 1,494 respondents, the majority of the respondents were male (65%) and from rural areas of Ghana (63%). Most of the respondents were fairly young, with 84% between the ages of 18 and 35 years. Most of the respondents were not health-care workers (80%).

Data collection

Data was collected in partnership with Viamo, a digital technology platform that provides interactive and measurable mobile engagement surveys. Viamo conducts mobile surveys and gathers data from hard-to-reach communities. Random Digit Dialing (RDD) was used to recruit study respondents, a method wherein randomly generated mobile numbers are called.

Every member of the target national population had an equal chance of selection. The survey was delivered by Viamo in December 2021 and all data was collected by January 2022. Viamo administered the survey via Interactive Voice Response (IVR), an automated system that delivers pre-recorded voice messages to people over a mobile phone. The IVR system allows voice messages to be sent to users and for users to send messages back to the IVR system. Initial calls were placed between 8am and 8pm local time. Respondents who missed the call or were unable or unwilling to complete the survey at the time of the call could call back using the number recorded in their call logs or by using the redial feature to take the survey at a more convenient time. Each number was dialed only once. Calls were made available in six generally spoken languages in Ghana: English, Hausa, Twi, Ewe, Ga, and Dagbani. Respondents could select the preferred survey language by pressing a number on their telephone keypad. Any person who answered the phone was eligible for survey participation if they were at least 18 years old. The only caveat for mobile surveys was that the survey participant had to be in possession of an active mobile phone. This was mainly possible due to the high mobile phone penetration in the country.

Ethical considerations

The study protocol was approved by the Ethical Board of Ghana Health Services. The study is voluntary and accordingly there is no penalty for refusing to participate. The study design allows participants to skip questions. The study was also confidential. Only a unique subscriber identifier was used for each contact number on the mobile survey platform.

The user contact information was not given out, with a unique identifier generated and used for each participating user. Informed consent was administered before the start of the survey. Upon answering the call, participants listened to a brief description of the study, including the duration of the survey. They were also informed about data confidentiality and were then requested to provide consent if they wished to proceed with the survey by pressing a number on their phones. Only after a participant’s consent and age was recorded, the participant was able to proceed with the survey.

Design

Procedure

Upon receiving and answering the automated call to participate in the survey, respondents (n = 1494) were asked to choose among the six predominantly spoken languages in Ghana: English, Twi, Ga, Ewe, Dagbani and Hausa. Translations for each language were double-checked with key language experts to ascertain and confirm equivalence across languages for the survey. The rest of the survey was presented in the selected language. Respondents were then asked to provide their voluntary consent to participate in the survey and to verify their age. The survey immediately ended for respondents who did not consent to the survey and/or who reported being under 18 years. Throughout the survey, respondents could not go back to previous questions.

In part I of the study, respondents (n = 1494) were first asked about general vaccination behavior and if they ever received a vaccine as a child or an adult. They were then asked if they had received one or more doses of the COVID-19 vaccine. If the respondents (n = 405) reported not receiving one or more doses of the COVID-19 vaccine, they were presented with three questions measuring their intentions to get vaccinated against COVID-19. All respondents (n = 1494) then rated their perceptions on the various drivers of vaccination, namely, risk perceptions, effectiveness, ease of vaccination, and respondents’ own attitudes toward the COVID-19 vaccine as well as their perceptions of their family’s attitudes, community’s attitudes, religious leaders’ attitudes, and healthcare providers’ attitudes toward the COVID-19 vaccine. After rating the prospective drivers of vaccination, respondents (n = 1494) were given a list of possible drivers and asked to indicate what they believe to be the most important and the least important influences on their decision to get vaccinated.

Next, respondents were asked to provide demographic information, including, gender, education, location (urban/rural), whether they had children, whether they were healthcare workers, whether they or their immediate family were infected with COVID-19, religious affiliation, medical status, and medical preference.

Part II of the study was presented after the demographic questions. Respondents were randomly assigned to one of six BI message types (see Appendix A in S1 Appendix). These were presented in the language selected by the participant at the beginning of the survey. After listening, respondents were asked if they would like to hear the BI message one more time and if they understood the message. Next, all respondents (n = 1494), regardless of vaccination status, were asked if they would recommend getting vaccinated to a family member or friend and if they would share the benefits of vaccination with a family member or friend. Unvaccinated respondents (n = 405) were then once again presented with the three questions measuring their intentions to get vaccinated, which were also presented during Part I of the study. Upon completing the survey, respondents were thanked and given an airtime reward of GHS10. This correlated to about USD $1.65 at the time of the study.

Measures

Vaccination behavior was measured by asking respondents to use a binary Yes/No response scale to indicate whether they had received one or more doses of the COVID-19 vaccine. The item included to measure vaccination behavior was, “Have you personally received one or more doses of the COVID-19 vaccine?”

Vaccination intention measured respondents’ intention to get vaccinated against COVID-19 if they indicated not receiving one or more doses of the vaccine. Three items were used to measure vaccination intention. Respondents’ ratings on the three items were averaged to form a composite which was used in all the analyses (α = 0.89). All three items were administered on a five-point Likert scale ranging from 1 (extremely unlikely) to 5 (extremely likely). An example item is, “How likely are you to consider getting vaccinated once the COVID-19 vaccine becomes available?”

Willingness to recommend the COVID-19 vaccine to a family member or friend was measured with a single item, which asked respondents: “When the vaccine becomes available, will you recommend the COVID-19 vaccine to a family member or friend?” This question was presented to respondents after they listened to the BI message (jingle). The three response options were yes, no, and don’t know.

Willingness to share the benefits of vaccination to a family member or friend was measured with a single item, which asked: “Will you try to convince a family member or friend to get vaccinated against COVID-19?” This question was also presented after respondents listened to the BI message. The three response options were yes, no, and don’t know.

Risk perceptions (α = 0.84, ω = 0.84) measured the degree to which respondents sensed a risk of contracting COVID-19. Two items were included to measure risk perceptions using a five-point Likert scale ranging from 1 (extremely unlikely) to 5 (extremely likely). An example item is, “If you do not get the vaccine, how likely do you think it is that you will catch COVID-19?”

Effectiveness (α = 0.79, ω = 0.79), adapted from a previous study [39], measured respondents’ confidence in the COVID-19 vaccine. Two items were included using a five-point Likert scale ranging from 1 (extremely unsafe) to 5 (extremely safe) for the first item and a five-point Likert scale ranging from 1 (extremely ineffective) to 5 (extremely effective) for the second item. An example item is, “How safe do you think the COVID-19 vaccine is?”

Ease of vaccination was measured using two items for those who were vaccinated (α = 0.92, ω = 0.92) and two parallel items (α = 0.97, ω = 0.97) were included for those who were not vaccinated. All items used a five-point Likert scale. The first item asked directly about ease and used a scale ranging from 1 (very difficult) to 5 (very easy). The second item asked about the clarity of the process for getting vaccinated and used a scale ranging from 1 (very unclear) to 5 (very clear). An example item for those who reported being vaccinated is, “How easy was it for you to get the vaccine?” and for those who reported not being vaccinated is, “How easy will it be for you to get the vaccine?”

Own attitudes (α = 0.88, ω = 0.88) were measured using the following two items which assessed respondents’ own attitudes toward the COVID-19 vaccine. The first item, "What is your attitude toward the COVID-19 vaccine generally?" was paired with a five-point Likert-type response scale ranging from 1 (very negative) to 5 (very positive); this response scale was also used to measure perceptions of others’ attitudes (i.e., family, community, religious leaders and healthcare providers) The second item, "How important do you think it is to get vaccinated?” was paired with a five-point Likert-type scale ranging from 1 (not at all important) to 5 (extremely important), which was subsequently used to measure respondents’ perceptions of others’ attitudes toward the COVID-19 vaccine, described next.

Family’s attitudes (α = 0.87, ω = 0.87) were measured using the following two items which assessed respondents’ perceptions of their family’s attitudes toward the COVID-19 vaccine: “What is your family’s attitude toward the COVID-19 vaccine?” and “How important does your family think it is for you to get the vaccine?”

Community’s attitudes (α = 0.86, ω = 0.86) were measured using the following two items which assessed respondents’ perceptions of their community members’ attitudes toward the COVID-19 vaccine: “What is your community’s attitudes toward the COVID-19 vaccine?” and “How important does your community think it is for you to get the vaccine?”

Religious leaders’ attitudes (α = 0.88, ω = 0.88) were measured using the following two items which assessed respondents’ perceptions of religious leaders’ attitudes toward the COVID-19 vaccine: “What are your religious leaders’ attitudes toward the COVID-19 vaccine?” and “How important do your religious leaders think it is for you to get the vaccine?”

Healthcare providers’ attitudes (α = 0.86, ω = 0.86) were measured using the following two items which assessed respondents’ perceptions of healthcare providers’ attitudes toward the COVID-19 vaccine: “What are your healthcare providers’ attitudes toward the COVID-19 vaccine?” and “How important do your healthcare providers think it is for you to get the vaccine?”

Correlations between each of the drivers for both unvaccinated and vaccinated participants are included in Appendices A and B in S1 Appendix.

Self-ranked top influences measured who respondents consciously perceived to have the biggest influence over their decision to get vaccinated. This was assessed using a single item: “Of these choices, who has the biggest influence over your decision about whether to get the COVID-19 vaccine?” Respondents were presented with the following options: self (own attitudes), family, community members, Ghana Health Service, religious leaders, and healthcare providers.

Self-ranked least influences measured who respondents consciously perceived to have the least influence over their decision to get vaccinated. It was measured using a single item: “Who has the least amount of influence over your decision about whether to get the COVID-19 vaccine?” Respondents were presented with the same choices indicated above: self (own attitudes), family, community members, Ghana Health Service, religious leaders, and healthcare providers.

Part I—Vaccination behavior

Self-reported vaccination behavior was examined for the overall sample and disaggregated by sociodemographic characteristics of the sample (Table 2). As can be seen, 73% of the sample reported receiving at least one dose of the COVID-19 vaccine. Subgroup vaccination rates ranged from a low of 64% for women to a high of 86% for healthcare workers.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Descriptive statistics for vaccination behavior.

https://doi.org/10.1371/journal.pone.0292532.t002

Next, the sample was restricted to those (N = 1089) who reported receiving at least one dose of the COVID-19 vaccine. Descriptive statistics were examined for each of the eight potential drivers (Table 3). As can be seen by the average ratings reported in Table 3, respondents perceived themselves and healthcare providers to have relatively positive attitudes toward the COVID-19 vaccine, with average ratings at 4.17 and 4.21 respectively on the 5-point scale. Other perceptions were a bit less favorable. For example, while still above the midpoint, ratings of ease of vaccination were lower, with an average rating of 3.66, suggesting that there was room for improvement with respect to reducing barriers to vaccination at the time of data collection.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Descriptive statistics of potential drivers for vaccination behavior: Vaccinated respondents.

https://doi.org/10.1371/journal.pone.0292532.t003

To test the influence of the eight potential drivers on respondents’ vaccination behavior, a logistic regression analysis was conducted (Table 4). Specifically, we wanted to examine whether the proposed drivers significantly predicted vaccination behavior. The logistic regression model was statistically significant when examined using the Hosmer and Lemeshow goodness of fit test, χ²(11, N = 1494) = 54.369, p < .001. The model explained 7.56% (Cox and Snell Index R²) of the variance in respondents’ vaccination behavior. Of the eight drivers, family’s attitudes and religious leaders’ attitudes significantly predicted vaccination behavior in a positive direction; when respondents perceived family and religious leaders to have favorable attitudes towards the vaccine, they were more likely to get vaccinated. However, perceptions of healthcare providers’ attitudes had a significant negative relationship with respondents’ vaccination behavior. When respondents perceived healthcare providers to have positive attitudes towards the COVID-19 vaccine, they were less likely to get vaccinated. As shown in Table 4, one’s own attitude, risk perception, effectiveness, ease of vaccination, and perceptions of the community’s attitudes toward the COVID-19 vaccine did not significantly predict vaccination behaviors.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Results of the logistic regression analysis and relative weights analysis (RWA): Vaccination behavior.

https://doi.org/10.1371/journal.pone.0292532.t004

An important follow-up question pertains to whether the drivers included in the model varied with respect to their relative influence on vaccination behavior. Simply examining the size of regression coefficients cannot answer this question [40, 41]. Following the logistic regression analysis, a relative weights analysis (RWA) was therefore conducted to examine the relative importance/contribution of the drivers included in the regression model [41]. The relative importance of each predictor variable refers to the proportionate contribution that each predictor makes to the total predicted variance [42]. In addition to relative importance, RWA also provides confidence intervals, which can be used to evaluate both within and between sample differences in predictor weights [43, 44].

The results from any given RWA provide two estimates, the raw importance, and a rescaled importance for each predictor variable. The sum of raw importance of all predictors in a model result in the total model variance, R² [41]. It can be interpreted as the proportion of variance in the criterion variable (vaccination behavior) that is attributed to each of the predictor variables (drivers). The raw importance of predictor variables is each supplemented with confidence intervals, used to explain their precision [41]. Rescaled importance, on the other hand, is calculated by dividing the raw relative weight of each of the predictors in the model by the total model variance, R² [41]. While raw importance uses the metric of relative effect sizes, rescaled importance uses the metric of percentage of the total predicted variance in a model that is attributed to each predictor variable [45].

Table 4 summarizes the RWA conducted to examine the relative importance of the eight drivers in the logistic regression model discussed above, with vaccination behavior as the criterion. Consistent with the regression model, family’s attitudes and religious leaders’ attitudes were amongst the highest raw and rescaled importance. Confidence intervals generated in the RWA output were used to test whether the eight drivers were significantly different from one another with respect to their influence. One’s own attitudes and perceptions of their family’s attitudes were each significantly more influential than risk perception, effectiveness and ease of vaccination. Perceptions of the attitudes held by the community, religious leaders and healthcare providers were also more influential than risk perceptions and ease of vaccination (see Table in Appendix B of S1 Appendix). When interpreting the role of healthcare providers’ attitudes, it is important to keep in mind that this was a negative effect. Perceiving healthcare providers as supportive of the vaccine seemed to discourage vaccination behavior.

Part I—Vaccination intention

Next, we examined what predicted intentions to get vaccinated among the 405 respondents who reported that they had not yet received the COVID-19 vaccine. Table 5 summarizes their vaccination intentions overall, and by sociodemographic group. As can be seen, the overall vaccination intention hovered around the neutral midpoint of 3 and ranged from a low of 2.84 for women and a high of 3.19 for healthcare workers. As indicated by the standard deviations shown in Table 5, there was variability within each group with respect to intentions to get vaccinated. Table 6 shows descriptive statistics for each of the eight drivers as rated by the 405 respondents who reported not receiving the COVID-19 vaccine. As can be seen in Table 6, ratings ranged from a low of 3.09 to a high of 3.64. Specifically, the unvaccinated respondents perceived healthcare providers to have relatively positive attitudes toward the COVID-19 vaccine, with an average rating of 3.64. Respondents rated the ease of vaccination just above the midpoint, with an average of 3.09. Consistent with the average ratings previously reported for the vaccinated subsample, these results suggest that there was some scope for improvement in making it easier to get vaccinated at the time of data collection.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Descriptive statistics for unvaccinated respondents.

https://doi.org/10.1371/journal.pone.0292532.t005

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 6. Descriptive statistics of potential drivers for vaccination intention: unvaccinated respondents.

https://doi.org/10.1371/journal.pone.0292532.t006

To test the influence of the eight drivers on respondents’ vaccination intentions, a linear regression analysis was conducted (Table 7). The overall regression model explained 67.53% of the total variance in respondents’ intention to get vaccinated (R² = .67, F(8, 396) = 106, p < .001). Of the eight drivers, risk perception, effectiveness, ease of vaccination, and religious leaders’ attitudes significantly and positively predicted respondents’ intention to get vaccinated. The influence of healthcare providers was again in the negative direction, but it was not statistically significant in the multiple regression analysis (p = .147). Indeed, none of the four remaining variables—own attitudes, family’s attitudes, community’s attitudes, and healthcare providers’ attitudes—significantly predicted respondents’ intention to get vaccinated.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 7. Results of the linear regression analysis: Vaccination intention.

https://doi.org/10.1371/journal.pone.0292532.t007

The regression analysis was followed by an RWA to determine the degree to which the drivers in the model differentially influenced vaccination intentions. Results are shown in Table 7. Risk perception, ease of vaccination and religious leaders’ attitudes had the highest raw and rescaled relative weights, highlighting their importance when it comes to vaccine intentions. Confidence intervals were used to test the differential importance of the eight drivers. When it comes to forming intentions to get the COVID-19 vaccine in Ghana, perceptions of healthcare practitioners’ attitudes are significantly less important than nearly every other variable in the model: risk perception, ease of vaccination, own attitudes, family’s attitudes, community’s attitudes, and religious leaders’ attitudes. No other significant differences among the drivers were found (see Table in Appendix C of S1 Appendix).

Part I—Self ranking influences

The regressions and RWAs reported above reveal the influence of various factors on decisions and intentions to get vaccinated without asking people about who and what motivates them. This is a way to understand the kinds of unconscious drivers that affect human behavior on a daily basis. It is interesting to compare these results to people’s conscious perceptions of who influences them when it comes to vaccinations. To explore this, we asked both vaccinated (N = 1089) and unvaccinated participants (N = 405) to self-rank which, among six sources (own attitudes, family, community members, Ghana Health Services, religious leaders and healthcare providers), had the biggest influence on their decision to get vaccinated and which among them had the smallest influence on their decision to get vaccinated. Table 8 summarizes the results.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 8. Biggest and smallest self-ranked influence on vaccination intention.

https://doi.org/10.1371/journal.pone.0292532.t008

Among vaccinated respondents, more than a third of the respondents ranked their own attitudes as most influential, indicating that they felt their own attitude toward the COVID-19 vaccine has the biggest effect on their decision to get vaccinated. At the same time, almost a quarter of the sample ranked this variable as having the smallest influence. The community was generally perceived to have a low level of influence, which is consistent with the regression analyses. Interestingly, a quarter of the respondents indicated that the Ghana Health Services has the biggest influence on their intention to get vaccinated and at the same time almost a quarter of the sample also ranked it as having the smallest influence on their intention to get vaccinated. Other self-reports were less consistent with the regression and relative weights analysis. For example, the results in Table 8 suggest that the influence of certain groups, such religious leaders, is largely unrealized by the study sample. While the models tested above (e.g., see Tables 4 and 7) clearly indicate that perceptions of what religious leaders think about the vaccine matter, respondents did not necessarily recognize this when asked to explicitly indicate who influences them.

Among unvaccinated respondents, a third of the sample indicated that family had the biggest influence on their intentions to get vaccinated, followed by a quarter of the sample that indicated that their own attitudes had the biggest influence. Both of these findings were not very consistent with the regression analysis. The community was again perceived to have a low level of influence, consistent with the regression and relative weights analysis. The impact of religious leaders’ influence on respondents’ intention to get vaccinated was quite low based on the self-rankings and inconsistent with the regression and relative weights analysis.

Part II—BI message type (Jingle)

As discussed previously, in part II of the study, respondents were randomly assigned to one of the six BI message types shown in Appendix A in S1 Appendix. After listening to the assigned BI message, they were asked to report their willingness to recommend the COVID-19 vaccine to family members and friends. Table 9 shows the percentage of respondents exposed to each message type who expressed a willingness to recommend the COVID-19 vaccine. As can be seen, a large percentage of respondents, between 85–93%, expressed a willingness to recommend the vaccine to family members and friends with slight variations across BI message types. A chi-square goodness-of-fit test was performed to examine whether all six BI message types were equally effective in this regard. Results from a chi-square goodness-of-fit test suggested differential responses across the message types, χ² (5, N = 1494) = 11.29, p = .046 (Table 9). Pairwise comparisons with a Bonferroni correction were performed to determine whether some BI message types were more effective than the others in encouraging respondents to recommend the vaccine. Pairwise comparisons revealed that no one BI message type was more effective than the others.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 9. Chi-square results for BI message type and willingness to recommend.

https://doi.org/10.1371/journal.pone.0292532.t009

Respondents also reported their willingness to share the benefits of the vaccine after listening to the assigned BI message type. Table 10 summarizes the results. Again, it appears all BI message types were quite effective, with 86–92% of the sample expressing a willingness to spread the word about the vaccine’s benefits. A chi-square goodness-of-fit test was performed to examine whether all six BI message types generated similar responses. No significant differences among message types were indicated in the goodness-of-fit test, χ² (5, N = 1494) = 8.42, p = .135.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 10. Chi-square results for BI message type and willingness to share.

https://doi.org/10.1371/journal.pone.0292532.t010

The effectiveness of the BI message types was similar for both vaccinated and unvaccinated respondents. Among unvaccinated respondents (n = 405), the effectiveness of the BI message types on respondents’ willingness to recommend the vaccine, χ2 (5, N = 405) = 9.35, p = .095, and their willingness to share the benefits of the vaccine, χ2 (5, N = 405) = 8.41, p = .135, were not significantly differently from one another. Similarly, among vaccinated respondents (n = 1089), the effectiveness of the BI message types on respondents’ willingness to recommend the vaccine, χ2 (5, N = 1089) = 5.03, p = .411, and their willingness to share the benefits of the vaccine, χ2 (5, N = 1089) = 5.82, p = .323, were not significantly differently from one another.

A final analysis was conducted on the unvaccinated subsample (N = 405), wherein, respondents once again reported their intentions to get vaccinated using the three items discussed previously. As shown in Table 11, vaccine intentions varied only slightly across message types, ranging between 3.26 and 3.47 on the 5-point Likert-type scale.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 11. Vaccination intentions by BI message type.

https://doi.org/10.1371/journal.pone.0292532.t011

A one-way ANOVA was conducted to determine whether message type affected vaccine intentions. Results indicated no significant differences in respondents’ intentions to get vaccinated based on the BI message they were assigned to, F(5, 399) = 0.34, p = .889 (Table 12).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 12. Effect of BI message type on vaccination intentions.

https://doi.org/10.1371/journal.pone.0292532.t012