Evaluation of a Brief Influencer-Delivered Body Image Video

Background:

Influencers are everyday people who are extraordinarily influential within their social media networks. This project aims to leverage the power of influencers who communicate directly with an engaged audience, who trust and relate to them. This trust is a result of followers having an emotional connection with their preferred influencers, known as a parasocial relationship. Parasocial relationships between followers and influencers have been leveraged in successful public health campaigns, such as with flu vaccinations, safe sex practices, and anti-tobacco messaging. Furthermore, influencers have been effective in reaching underserved populations. Their extensive reach has not yet been harnessed to target people requiring body image support. Given widespread body dissatisfaction among women globally, this study is the first to evaluate the efficacy of a body image intervention created and delivered by an influencer, building upon the investigators' public involvement research around influencers and their followers.

Methods:

One female influencer will create a 60-second video as per an evidence-informed brief co-created with CAR and Dove outlining an exercise intended to support women in challenging beauty standards and improving their body confidence. The influencer will be sourced through a partnership between Dove and an influencer agency. Dove will hold a one-hour online training session as informed by CAR around the brief and guidelines for video creation. Recruitment will be conducted by the influencer whose video will be evaluated. This approach has been successfully used by Twitch influencers in a research study. The influencer will create and disseminate up to three recruitment 30-60 second "stories" (videos that expire in 24 hours). If the desired sample size is not met, additional participants may be recruited through Prolific, a secure online platform for recruiting research participants.

Following, an online survey will be shared with influencers via the influencer agency to capture their perspectives around influencers disseminating body image interventions to their followers.

Aim/objectives:

The overall aim is to determine the feasibility, acceptability, and efficacy of an evidence-informed 60-second video created and delivered by an adult female influencer to her adult female Instagram and TikTok followers.

Research questions (RQs) and hypotheses (Hs):

Primary Research Question:

1. Does watching the intervention video increase state body satisfaction immediately after viewing, relative to a control video? H1. It is anticipated that women who watch the intervention video will have increased state body satisfaction, relative to a control video.

   Secondary RQs:

2. Does watching the intervention video increase state body appreciation immediately after viewing, relative to a control video? H2. It is anticipated that women who watch the intervention video will have increased state body appreciation, relative to a control video.

3. Does watching the intervention video increase state body functionality appreciation immediately after viewing, relative to a control video? H3. It is anticipated that women who watch the intervention video will have increased state body functionality appreciation, relative to a control video.

4. Is the intervention acceptable (i.e., enjoy, like, recommend the video) to participants? H4. It is anticipated that women who watch the intervention video will find it acceptable (i.e., enjoy, like, recommend the video).

5. Does intervention impact depend on baseline body esteem levels? H5: It is anticipated that women who have lower levels of body esteem at baseline will experience greater improvements in state body satisfaction.

Exploratory RQs:

1. Is it feasible for the influencer to recruit the target sample size from the influencer's followers on social media?
2. If recruitment from Prolific is required, are there differential effects between the two sources of recruitment?
3. Are there any moderating effects of age or parasocial relationship?

Sample Size for the Efficacy RCT:

The primary efficacy outcome is state body satisfaction measured immediately pre- and post-exposure. Given the brevity of the intervention, any intervention effects are expected to be small. However, because pre- and post-intervention assessments occur within a very short time interval and measure the same underlying construct, then a relatively strong correlation between measurements is anticipated. The study has been powered using an analysis of covariance (ANCOVA), adjusting for the pre-intervention measure and treatment allocation, with recruitment source included as an additional covariate if Prolific participants are recruited (i.e., Outcome ~ Pre + Treatment, or Outcome ~ Pre + Treatment + Source if Prolific provides participants). The investigators assume that, in the presence of an intervention effect, the corresponding probability is approximately 53%, meaning that a randomly selected intervention participant would be expected to have a more favourable outcome than a randomly selected control participant approximately 53% of the time (irrespective of ties in the data). This corresponds to a standardised mean difference of approximately Cohen's d = 0.11. For the correlation between pre- and post-intervention measurements, the investigators assume r = 0.80. Under these assumptions, a total sample size of approximately N = 1,000 participants (n = 500 per randomised arm) provides 80% power to detect the specified effect (alpha = 0.05, two-sided), the minimum sample size required.

If the planned recruitment cap of N = 1,600 is achieved, then under the same assumptions, the efficacy analysis would have approximately 94% power. The recruitment cap of N = 1,600 therefore provides protection against uncertainty in recruitment rates, uncertainty in the true pre-post correlation, uncertainty in the magnitude of the intervention effect, provides protection against data loss, while also allowing more precise estimation of recruitment outcomes and secondary analyses.

As a sensitivity analysis, the investigators note that if the intervention effect were somewhat smaller (e.g., Cohen's d = 0.10, equivalent to a superiority probability of approximately 52.8%) and the pre-post correlation somewhat weaker (e.g., r = 0.70), then a sample size of N = 1,600 would still provide approximately 80% power for the efficacy analyses.

Analyses of Efficacy Trial:

Analysis of Primary Outcome The primary endpoint is the composite State Body Satisfaction Outcome at T2 defined as the simple average of the three subscales. Non-completion of any item will be classed as missing data on the primary outcome. The baseline composite score will be calculated from the corresponding pre-intervention measures. Cronbach's alpha will be reported descriptively, and the composite will be used regardless of the reliability estimates. The primary estimand is the adjusted mean difference between intervention and control in post-intervention State Body Satisfaction, controlling for baseline State Body Satisfaction and recruitment source.

Baseline adjusted analysis of covariance will be used to assess the primary outcome, with recruitment source as a stratification factor if the recruitment contingency strategy (i.e., Prolific) is employed. Specifically, the primary model is Post=Pre+Treatment+Source.

Mean difference between Intervention and Control will be reported, along with the adjusted 95% Confidence Interval, and standardised effect size (Cohen's d).

Participants with poorer baseline state body satisfaction may have greater scope for improvement and therefore may benefit more from the intervention. For this reason, a key exploratory moderation analysis will be performed defined as Post=Pre+Treatment+Source+Pre×Treatment with an assessment of the interaction term. If the interaction term is statistically significant (p < .05), then a Johnson-Neyman ANCOVA analysis will be undertaken.

Assessment of Assumptions

Statistical assumptions underpinning the analysis of covariance model will be examined. These include:

1. Linearity: The relationship between baseline State Body Satisfaction and post-intervention State Body Satisfaction will be assessed graphically using scatterplots and fitted regression lines.
2. Homogeneity of Regression Slopes: The assumption of homogeneous regression slopes will be assessed through the prespecified Treatment × Baseline interaction model used for the exploratory moderation analysis.
3. Normality of residuals will be assessed graphically for gross departures from normality.
4. Homoscedasticity: Residual plots will be examined to assess whether residual variance is approximately constant across fitted values and treatment groups.
5. Influential Observations: Standardised residuals, leverage statistics and Cook's distance will be examined to identify influential observations.

Given the anticipated large sample size, minor departures from distributional assumptions will not be considered sufficient justification to replace the prespecified primary analysis model. Any substantial violations and any resulting sensitivity analyses will be documented in the final study report.

Reporting of Effects Continuous outcomes will be summarised using mean, standard deviation, median, minimum and maximum by randomised arm, and by randomised arm and source of recruitment.

Treatment effects will be presented as both unadjusted and adjusted mean differences with 95% confidence intervals and two-sided p-values. Primary inference will be based on adjusted analyses.

Standardised effect sizes (Cohen's d) will additionally be reported.

Secondary Outcomes Weight satisfaction, Shape satisfaction, Appearance Satisfaction, State Body Functionality, and State Body Appreciation are all measured at T1 (baseline/immediately pre-assessment) and T2 (immediately post-assessment) and will be analysed using the same modelling approach as described for the primary outcome. Results for secondary outcomes will be exploratory; no adjustment for multiplicity will be applied.

RQ5 is concerned with assessing whether intervention effects vary according to baseline appearance esteem. For this, an ANCOVA model including Treatment, baseline State Body Satisfaction, recruitment source, baseline Appearance Esteem, and a Treatment × Appearance Esteem interaction term will be fitted. Evidence for moderation will be assessed using the interaction term. If the interaction is statistically significant (p < .05), conditional treatment effects will be examined, and Johnson-Neyman regions of significance may be reported.

Exploratory Outcomes ExRQ2 is concerned with answering the question as to whether the impact of the intervention differs between followers of the influencer and non-followers?

To answer this question, this model will be used:

"Post"="Pre"+"Treatment"+"Source"+"Treatment"×"Source" and examine the Treatment by Source interaction effect for statistical significance.

Missing values analysis If more than 5% of randomised participants fail to provide post-intervention data for the primary outcome, a sensitivity analysis using Multiple Imputation by Chained Equations will be undertaken. One hundred imputed datasets will be generated. The imputation model will include treatment allocation, recruitment source, baseline outcome measures, including baseline state body appreciation, baseline state body functionality, baseline state body satisfaction, and age. Imputed datasets will be analysed using the prespecified ANCOVA model and combined using Rubin's Rules. The multiply-imputed analysis will be regarded as a sensitivity analysis to assess the robustness of conclusions from the primary observed-data analysis.