This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research, is properly cited. The complete bibliographic information, a link to the original publication on http://www.jmir.org/, as well as this copyright and license information must be included.
Social media utilization is on the rise globally, and the potential of social media for health behavior campaigns is widely recognized. However, as the landscape of social media evolves, so do techniques used to optimize campaign dissemination.
The primary aim of this study was to evaluate the impact of 4 material dissemination paths for a breastfeeding social media marketing campaign in Ghana on exposure and engagement with campaign material.
Campaign materials (n=60) were posted to a Facebook and Twitter campaign page over 12 weeks (ie, baseline). The top 40 performing materials were randomized to 1 of 4 redissemination arms (control simply posted on each platform, key influencers, random influencers, and paid advertisements). Key performance indicator data (ie, exposure and engagement) were extracted from both Facebook and Twitter 2 days after the material was posted. A difference-in-difference model was used to examine the impact of the dissemination paths on performance.
At baseline, campaign materials received an average (SD) exposure of 1178 (670) on Facebook and 1071 (905) on Twitter (n=60). On Facebook, materials posted with paid advertisements had significantly higher exposure and engagement compared with the control arm (
Paid advertisements are an effective mechanism to increase exposure and engagement of campaign posts on Facebook, which was achieved at a low cost.
During social media’s early days in 2005, a mere 5% of Americans utilized it, while more recently in 2018, it was used by 67% and 49% of adults in advanced and emerging economies, respectively [
As social media has evolved, so have the marketing techniques used by businesses to push their products and ideas on the population. Social marketing of public health goods is also finding its place on this environment, with social media becoming widely recognized as “an unprecedented opportunity...to deliver socially influential online behavior change interventions” [
Specifically, social media has been identified as an opportunity for a potentially cost-effective approach to improve breastfeeding outcomes relative to traditional social marketing, yet research for social marketing for breastfeeding promotion is needed to identify best practices and approaches [
In social media marketing, consumer-to-consumer interaction and word-of-mouth dissemination of information is widely used in the form of influencers [
Alternatively, social media platforms have established mechanisms for businesses to pay for advertisements to appear on the news feed of targeted consumer groups. These paid advertisements have been effective for business, especially when paired with creative marketing, and as a result are considered essential in business social media marketing plans [
Between 2008 and 2014, the rate of children under 6 months of age who were exclusively breastfed in Ghana declined from 63% to 52% [
Therefore, the primary aim of this study is to evaluate the impact of different dissemination paths on exposure and engagement with campaign material and to examine the relationship of acceptability of campaign material with material performance.
We implemented a 6-month long Facebook (Menlo Park, CA, US) and Twitter (San Francisco, CA, US) campaign that targeted the protection, promotion, and support of breastfeeding in Ghana, on the basis of evidence from recommendations from the BBF Initiative in Ghana [
Campaign performance for platforms (ie, Facebook and Twitter pages) and individual material performance (ie, Facebook posts and tweets) were monitored using data extracted from Facebook Insights and Twitter Analytics and entered into a Microsoft Access database (Microsoft Corp, Redmond, WA, US). Platform data were collected weekly and included the number of followers and likes on both Facebook and Twitter; and reach, engagements, views, and follower and engagement demographics by age, sex, and country on Facebook. Data on the campaign posts, such as the core campaign materials, were extracted at 3 timepoints: 1 day, 1 week, and 2 weeks after the material was posted. These data included material impressions or reach, likes, share or retweets, and comments or replies. Materials were selected for redissemination in the test phase based on a composite indicator of the amplification and applause rates per 100 followers for each Facebook and Twitter at the 1 day time point. Thus, materials were ranked from highest to lowest performance, by theme, and the top performing 40 campaign materials (13 or 14 per theme) were selected for redissemination across an 8-week period.
Stratified by theme, materials were randomized to 1 of 4 redissemination paths: (1) posted as usual (ie, control group); (2) shared by 6 key influencers on Facebook and 6 key influencers on Twitter after being posted; (3) shared by 6 random influencers on Facebook and 6 random influencers on Twitter after being posted; or (4) paid advertisement with US $6 on each Facebook and Twitter after being posted (
Study design for testing dissemination paths via Facebook and Tweeter. Theme A: Promote correct and complete information about breastfeeding; Theme B: Support women to breastfeeding anytime, anywhere; Theme C: Protect working women’s right to breastfeed.
To select the influencers, publicly available details (ie, number of friends or followers, number of friends or followers in common with the campaign, recent activity, and country of residence) from Facebook and Twitter user profiles were extracted from the respective platform among
The selection of key influencers focused on influencer reach and network ties, and therefore, the previous level of involvement with the campaign was not a selection criterion, although all influencers were at least following the campaign. Those with the highest reach were vetted by the campaign coordinator to make sure that they did not have inappropriate content (eg, pornography, violence, extremist behavior, or other offensive content) on their social media page and were then invited to be a key influencer for the campaign. It was explained to potential influencers that as an influencer they would be asked to share or retweet 1 to 2 campaign posts per week for 8 weeks and would receive a small incentive for their time of 80 Ghana cedis (~ US $20 or US $2.50 per week) for sharing the posts. This was repeated until a total of 6 key influencers on each of Facebook and Twitter were selected. To achieve this sample of 6 key influencers per platform, a total of 19 individuals were approached on Facebook and 9 on Twitter. To select random influencer, the remaining individuals in the sampling frame
Scheduled, targeted advertisements for the selected campaign materials were achieved through business accounts on both Facebook and Twitter, equating to US $6 per platform for each material randomly assigned to the paid advertisement dissemination arm. On Facebook, advertisements targeted 18 to 49 years old women residing in Ghana and were conducted in the form of a post
For Twitter, acquiring a business account in Ghana required going through a third-party company, which required meeting a minimum quarterly advertising budget. Similar to Facebook, advertisements targeted individuals 18 to 49 years old residing in Ghana. For both Facebook and Twitter, the paid advertisement was schedule along with the posts being scheduled, which were generally scheduled at least 1 week before the post date.
Campaign material acceptability was assessed via an online survey conducted over 3 timepoints. Each survey timepoint corresponded with the completion of the initial dissemination of campaign materials from 1 of the 3 campaign themes (ie, dissemination during the initial 12-week campaign period) and asked about the core campaign materials from that particular theme. Survey participants were a convenient sample of women ≥ 18 years of age residing in Ghana. The survey was promoted through an advertisement post on the campaign’s Facebook and Twitter pages and completed through Qualtrics (Provo, UT, USA). Materials presented in the surveys were chosen on the basis of material performance to represent the bottom, median, and top performance within the respective campaign theme, with performance on the basis of the number of
All data were imported into Stata 14.1 (Stata Corp, College Station, TX, USA) for cleaning and analysis.
Key performance indicators (KPIs) for platform and material exposure and engagement were developed from available data and based on Neiger et al 2012 definition of KPIs [
Platform Engagement was defined as the number of followers on each Facebook and Twitter. For this analysis, this KPI was used to adjust material KPIs by follower based at the time the material was posted, to make them comparable across time and platform. As reported elsewhere, the campaign started with 3061 Facebook followers and 27 Twitter followers and ended with 4096 Facebook followers and 736 Twitter followers [
Material Exposure was defined as reach on Facebook, which is the number of unique people who saw the material; and as impressions on Twitter, which is the number of times the material appeared on a Twitter timeline. These values were converted into rates per 100 followers on the respective platform at the time of the material posting to make them comparable across time and platform. These rates were used at the primary KPI for material exposure on each Facebook and Twitter.
Material Engagement comprises 3 subindicators: applause, amplification, and conversation. Applause was defined as the number of likes on each Facebook and Twitter; amplification was defined as the number of shares on Facebook and retweets on Twitter; and conversation was the number of comments on Facebook and replies on Twitter. Each of these subindicators by platform were converted to a rate per 100 followers on the respective platform at the time of the material posting (ie, (subindicator ÷ number of follower) × 100). Applause and amplification rates per 100 followers from all material posts and timepoints (n=300) for each Facebook and Twitter were standardized (μ=0; σ=1), and summed by platform and collectively to generate 3 material engagement scores for each material: total engagement score, Facebook engagement score, and Twitter engagement score.
In the content analysis survey, 13 statements examined the material image, message, and overall acceptability. The 5-point-Likert response options were collapsed to emphasize
To determine the impact of dissemination path on material performance, a series of difference-in-difference models that accounted for material performance at baseline were run for each of the following material KPI: Facebook exposure, Twitter exposure, total engagement score; Facebook engagement score, and Twitter engagement score. Given the difference in the indicator for exposure on Facebook versus Twitter, a total engagement score was not tested.
To determine characteristics of campaign materials that related to higher material performance, content survey data was pooled across the 3 timepoints, and material performance (low, middle, and high) was examined in relation to material acceptability defined as agreement with each of 6 acceptability statements, disagreement with 4 negative statements, and rating the image, message, and material as
This study was approved by the Yale University Institutional review board and the review board for Ghana University hosted by the Noguchi Institute. Influencers and content survey participants provided their electronic consent to participate in the study before their respective participation.
During the 12-week long baseline period, the 60 core campaign materials received an average (SD) exposure of 1178 (670) on Facebook and 1071 (905) on Twitter (
The top 40 performing core campaign materials were selected for the dissemination test period, with baseline characteristics summarized in
KPI for exposure on Facebook and all engagement indicators tending to increase from baseline to the test period, though these increases were only significant in the paid advertisement arm (
All 40 posts were disseminated via scheduled posts on both Facebook and Twitter at baseline and during the test period. Among the 10 materials randomly assigned to the paid advertisement arm, all had the paid advertisements directed at adults (18-49 years) in Ghana scheduled alongside the post schedule, as planned. Among the 20 materials that were randomized to the key influencers (n=10 materials) and random influencer (n=10 materials) arms, influencers were requested to share the material within 48 hours; 2 of the 6 key influencers on Facebook did not share all 10 materials that were requested and 1 of the 6 key influencers on Twitter did not share all the materials requested. All of the random influencers on Facebook (n=6) and Twitter (n=6) shared all 10 materials requested.
Definitions and summary of baseline performance indicators for core campaign materials on Facebook and Twitter, representing all 60 core campaign materials and the subsample of 40. Table is based on data collected 2 weeks after the material was posted.
Performance indicator and platform | Definition | Baseline (n=60) | Baseline (n=40)a | ||||
|
|
Mean (SD) | Mean (SD) per 100 followers | Mean (SD) | Mean (SD) per 100 followers | ||
|
|||||||
|
Reachb | 1178 (670) | 33.12 (19.12) | 1425 (655) | 40.12 (18.84) | ||
|
Impressionsc | 1071 (905) | 375.98 (396.34) | 1345 (951) | 494.13 (433.08) | ||
|
|||||||
|
|
||||||
|
|
Likes | 56.03 (34.01) | 1.59 (0.99) | 67.13 (36.08) | 1.90 (1.06) | |
|
|
Likes | 4.28 (2.99) | 1.63 (2.14) | 5.23 (2.97) | 2.12 (2.46) | |
|
|
||||||
|
|
Shares | 9.37 (5.11) | 0.26 (0.15) | 11.45 (4.77) | 0.32 (0.14) | |
|
|
Retweets | 2.23 (2.07) | 0.81 (1.14) | 2.83 (2.07) | 1.07 (1.29) | |
|
|
||||||
|
|
Comments | 1.33 (2.66) | 0.038 (0.076) | 1.9 (3.08) | 0.054 (0.088) | |
|
|
Replies | 0.02 (0.13) | 0.004 (0.033) | 0.03 (0.16) | 0.006 (0.041) |
aTop and middle performing material based on engagement (sum of applause, amplification, and conversation).
bReach is unique people saw content on Facebook.
cImpressions refers to times it appeared on a Twitter timeline.
Material performance at baseline and repost (test period) by dissemination path arms.
Key performance indicators | Control, mean (SD) | Key influencers, mean (SD) | Random influencers, mean (SD) | Paid advertisements, mean (SD) | |||||
|
RPb | BLc | RP | BL | RP | BL | RP |
|
|
|
|||||||||
|
35.96 (8.64) | 46.51 (31.43) | 50.61 (26.87) | 38.01 (13.36) | 57.19 (35.41) | 39.76 (13.87) | 88.88 (11.79)e | <.001 | |
|
393.57 (312.14) | 446.43 (300.58) | 686.34 (358.34) | 533.02 (444.02) | 502.49 (278.32) | 738.3 (577.61) | 666.85 (548.28) | .58 | |
|
|||||||||
|
Combined | −0.67 (0.59) | −0.05 (1.32) | 0.48 (1.13) | −0.09 (1.62) | 0.63 (1.43) | 1.23 (2.49) | 7.14 (4.18)e | <.001 |
|
−0.47 (0.51) | 0.27 (1.43) | 0.13 (1.02) | −0.19 (0.81) | 0.44 (1.57) | 0.40 (1.11) | 4.21 (1.81)e | <.001 | |
|
−0.20 (0.44) | −0.32 (0.41) | 0.35 (0.51) | 0.10 (1.34) | 0.18 (0.40) | 0.83 (2.54) | 2.93 (5.10) | .36 |
aP value for dissemination arm by repost interaction in the difference-in-difference model.
bRP: repost time point.
cBL: baseline.
dDefined as reach per 100 followers on Facebook and impressions per 100 followers on Twitter.
e
fDefined as the sum of the standardized applause and amplifications rates per 100 followers for both platforms combined, and individually.
Difference-in-difference models for material exposure on Facebook (a) and Twitter (b), and material engagement on Facebook (c) and Twitter (d) across dissemination path arms (95% CIs).
A total of 44 female participants completed the content survey (
Respondents reported the campaign materials to be acceptable, with over 75% agreement (or disagreement to negative statements) for 12 out of 13 acceptability statements across all materials (
Survey participant characteristics (N=44).
Characteristics | Value | |
Age (years), mean (SD) | 28.61 (4.20) | |
Based in Greater Accra, n (%) | 30 (68) | |
Married, n (%) | 19 (43) | |
Employed, n (%) | 34 (77) | |
Education: bachelor’s or higher, n (%) | 38 (86) | |
Had children, n (%) | 18 (41) | |
|
||
|
Internet, data, and Wi-Fi, n (%) | 37 (84) |
|
Facebook, n (%) | 32 (73) |
|
Twitter, n (%) | 14 (32) |
Access to own smartphone, n (%) | 44 (100) |
Material acceptability across material performance levels.
Statements and rating | Prevalence (SE) of |
Odds ratio (SE): measure of material acceptability by material performancea | ||||||
|
All | Low | Middle | High | Low | Middle | High | |
|
||||||||
|
This picture promotes breastfeeding | 79.55 (3.51) | 70.45 (6.88) | 79.55 (6.08) | 88.64 (4.78) | Reference | 1.69 (0.90) | 3.52 (2.08)b |
|
The picture is informative | 78.03 (3.60) | 68.18 (7.02) | 77.27 (6.32) | 88.64 (4.78) | Reference | 1.69 (0.95) | 4.13 (2.31)b |
|
This picture is confusingc | 81.06 (3.41) | 75.00 (6.53) | 75.00 (6.53) | 93.18 (3.80) | Reference | 1 | 4.59 (2.85)b |
|
I like this picture | 64.39 (4.17) | 59.09 (7.41) | 61.36 (7.34) | 72.73 (6.71) | Reference | 1.10 (0.41) | 1.86 (0.72) |
|
This picture is misleading/dishonestc | 87.88 (2.84) | 86.36 (5.17) | 88.64 (4.78) | 88.64 (4.78) | Reference | 1.23 (0.78) | 1.23 (0.69) |
|
||||||||
|
This message promotes breastfeeding | 87.12 (2.92) | 79.55 (6.08) | 95.45 (3.14) | 86.36 (5.17) | Reference | 5.50 (4.79)b | 1.64 (0.92) |
|
The message is informative | 90.91 (2.50) | 81.82 (5.81) | 100.00 (0.0) | 90.91 (4.33) | Reference | —d | 2.28 (1.32) |
|
This message is confusingc | 86.36 (2.99) | 86.36 (5.17) | 88.64 (4.78) | 84.09 (5.51) | Reference | 1.23 (0.87) | 0.83 (0.46) |
|
I like this message | 80.30 (3.46) | 72.73 (6.71) | 88.64 (4.78) | 79.55 (6.08) | Reference | 2.99 (1.54)b | 1.47 (0.50) |
|
This message is misleading/dishonestc | 87.79 (2.86) | 84.09 (5.51) | 90.91 (4.33) | 88.37 (4.89) | Reference | 1.90 (1.09) | 1.44 (0.69) |
|
||||||||
|
Image | 76.52 (3.69) | 61.36 (7.34) | 79.55 (6.08) | 88.64 (4.78) | Reference | 2.50 (1.23)b | 5.07 (2.91)f |
|
Message | 87.88 (2.84) | 79.55 (6.08) | 95.45 (3.14) | 88.64 (4.78) | Reference | 5.54 (4.32)b | 2.03 (1.00) |
|
Material | 78.03 (3.60) | 61.36 (7.34) | 86.36 (5.17) | 86.36 (5.17) | Reference | 4.15 (1.84)f | 4.15 (2.05)f |
aOdds ratio for logistic regression models adjusted for content survey and respondent.
b
cPrevalence represents disagreement with statement.
dCould not calculate Odds ratio because prevalence in middle performing group was 100%.
ePrevalence represents rating as
f
Breastfeed4Ghana core campaign materials achieved higher exposure and engagement on Facebook than on Twitter, and higher exposure and engagement rates per 100 followers on Twitter than on Facebook, because of the relatively small number of followers on Twitter. In this study, paid advertisements significantly increased material exposure and engagement on Facebook. Although neither influencer type had a significant impact on material performance on Facebook, there was a trend in higher engagement and exposure as a result of random influencers, compared with baseline. Conversely, there were no significant differences across the 4 dissemination paths on Twitter (ie, control, key influencers, random influencers, and paid advertisements), and it is important to consider the variations in purposes of different social media platforms and how they are used and how to consider this variation as part of the
A total of US $6 per material per platform was allocated to advertisement for each material in the paid advertisement arm. Accounting for the airtime incentives provided to influencers across the test period, each material in the influencer arm
Word-of-mouth is a recognized marketing approach for the expansion of a product or idea [
There is no prescribed way to recruit, select, and manage influencers, as Keller and Fay describe various case studies in their business marketing report, and influencers generally can impact various outcomes such as message amplification and product sales [
Our key influencers were selected with consideration for social network targeting, and also aimed to examine the difference in selecting key influencers (ie, social network targeting) as is done often with micro- and macroinfluencers versus randomly selected influencers. Similar to findings within a community health program in Honduras, between social network targeting and randomly selecting influencers, there were not significantly different outcomes; however, in Honduras, both social network targeting (akin to macroinfluencers) and random selection yielded significantly higher adoption of the intervention than the control group [
In our examination of domains of material acceptability and material performance, overall acceptability of the image, message, and material were associated with performance in terms of engagement. Most notable was that highest performing materials were those that had images viewed as (1) promoting breastfeeding; (2) informative; and (3)
While this campaign generated 60 core campaign materials that were disseminated 1 to 2 times during the active campaign period, and as a result varied the images so as not to be too repetitive (a feedback from our formative material development work). Despite the variety of images presented in the campaign materials, the highest performing material for each theme was that of a woman breastfeeding. This may suggest that such variety (ie, 60 unique materials) was not necessary, and generating images most aligned with the campaign focus and message will yield the greatest engagement and acceptability among the target population. As such, other social marketing campaigns have generated a small number of messages and materials that are widely and repeatedly disseminated [
This study also reports on KPIs of campaign materials. In public health, return on investment is not always a useful indicator, and Neiger et al summarized KPIs and evaluation metrics for health promotion on social media [
To our knowledge, similar studies have not been published. However, paid advertisements, as well as paid influencers, are widespread social marketing techniques used across businesses and industry. It is important to recognize the limitation in lack of comparable studies and results for us to consider. These results also reflect a short-term study and small number of influencers. Therefore, it is possible that in a study of longer duration with a higher number of influencers could yield different results. The lack of differences across dissemination path arms on Twitter could be the result of a small follower based on that platform, as well as a significant increase in follower (approximately doubled) right before the test period of the study. Finally, we would like to acknowledge that these results come from a breastfeeding social media–based campaign that targeted the population of Ghana. There are variations in how social media is used in different contexts across geographic space, demographics, and time, and therefore, generalizability of these results beyond the context should be done with caution. Similarly, for context, it is important to note the dates of from this campaign (March to September 2018). Social media platforms are continually updating their business platforms, advertising option, and algorithms for what is viewed on user’s news feeds, and therefore, comparability of the findings from this study may be limited based on how the landscape of social media changed and evolves with time.
Paid advertisements are an effective mechanism to increase exposure and engagement of campaign posts on Facebook, achieved at a low cost. Although influencers are used in marketing and are generally considered effective at increasing consumer engagement or sales, microinfluencers were not effective at increasing exposure and engagement in this study. Furthermore, the use of influencers to promote materials required a greater financial cost compared with paid advertisements in our study.
For social marketing, there are challenges with how to compete against product advertisers with bigger budgets and more ways to reach consumers (eg, infant formula companies) [
Materials tested in the content survey.
becoming breastfeeding friendly
key performance indicator
The authors would like to acknowledge Gabriela Buccini for her invaluable advice throughout the campaign on breastfeeding and lactation questions from followers, Amber Hromi-Fiedler for her guidance throughout the campaign, Katherine Doucet and Cara Safron for their input during the campaign development, Katherine Doucet and Abena Engmann for essential logistical support, the survey participants for their time and feedback, and all the Breastfeed4Ghana campaign “friends” and “followers” for their participation in the campaign. The research was funded through the International Society for Research in Human Milk and Lactation Trainee Bridge Fund.
None declared.