Assessing the Efficacy of a Song to Improve Body Appreciation Among Young Children

Body image develops in early childhood and by age four children report negative feelings towards their body and negative attitudes (e.g., weight-stigmatising attitudes) towards others. Body image concerns are associated with low self-esteem, disordered eating, and academic disengagement as children emerge into adolescence and as such, it is important to intervene and implement age-appropriate prevention strategies as early as possible. One key prevention strategy, endorsed by scholars and body image experts is to foster positive body image among children through the development and evaluation of interventions.

Blippi is a globally loved children's programme with over 18 million subscribers on YouTube. Blippi provides educational learning for children as young as four, with the most popular content via fun and interactive videos on YouTube and Netflix. In 2024, researchers at the Centre for Appearance Research developed a series of songs and associated music videos in collaboration with Moonbug Entertainment (creative partners and creators of children's media programmes, including creators of Blippi) and the Dove Self-Esteem Project (the funders of this project). The investigators ran a randomised controlled trial with 4-6 year olds and found that the music video was protective of children's body appreciation, when compared to an active control condition.

The current study aims to expand on these findings and assess whether the evaluated song can offer protective effects on body appreciation when listened to as audio content only (i.e., without visual content). This is an important question to answer, as within the banquet of content created by Blippi is audio content available via popular streaming platforms such as Apple Music and Spotify. Understanding the isolated impact the audio content is useful for two reasons: to promote viable dissemination strategies for the content (i.e., is promoting the audio content on audio streaming platforms compromising the effectiveness of the content, or enhancing the reach of an effective intervention?), and contributes to the wider literature examining the impact on audio content on children's socio-emotional development.

This study aims to answer three research questions:

RQ1 (Primary): Is body appreciation protected after listening to the song, relative to the control song? Hypothesis 1: The researchers anticipate that body appreciation of children randomised to the intervention condition will be protected immediately after listening to the song, relative to a control song.

RQ2 (Secondary): Is the intervention song acceptable to children? Hypothesis 2: The researchers anticipate that the intervention song will be acceptable to children.

RQ3 (Exploratory): Are observed effects moderated by gender, year group, and/or song comprehension? Hypothesis 3: Moderation for gender and year group are exploratory and thus no formal hypotheses are made. For comprehension, the researchers hypothesise that effects will be stronger for those that comprehend the songs' message, compared to those that do not.

Design: A two-arm (intervention vs active control) RCT will be delivered in-person. Individuals will be randomised to one of the two conditions.

Sample size: Approximately 200 Reception and Year 1 students from two to four primary schools in the Southwest of England. A total of 200 children will be randomised 1:1 (100 per arm) to the Blippi audio intervention or active control.

This sample size was determined a priori using G*Power for an F-test in a repeated-measures ANOVA (within-between interaction) with the following parameters based on effect size f = 0.10 (small), α = 0.05 (two-sided), statistical power = 0.80, for a two-group pre-post design. This power calculation indicates that 100 participants per group would be required. This is a conservative estimate of the power available for the planned primary analysis (baseline-adjusted proportional-odds ordinal logistic regression on the 2-8 body-appreciation score).

The above power calculation approach is a standard, slightly conservative proxy when pre-post correlation is moderate. The target effect size (f = 0.10) is deliberately conservative and reflects the small protective effect observed for the visual version of the same song in the predecessor trial (Craddock et al., 2025; N ≈ 110 per music-video arm).

No formal adjustment was made for the school stratification factor in the power calculation but including school as a fixed effect in the analysis model will increase power by reducing residual variance with only a small loss in degrees of freedom.

Procedure: The investigators will run concurrent one-on-one structured interviews (ratio: one participant to one field researcher) in the school setting. Every child completes baseline, then listens to their randomised song, then completes post measures. Following, to minimise contamination effects, each child will then listen to the non-randomised song.

Primary outcome analysis: The primary outcome is the total score for body appreciation (range 2-8) at post-intervention, derived by summing the two 4-point ordinal items (each scored 1 = No, 2 = A little bit, 3 = A medium bit, 4 = A lot). Post-intervention scores will be summarised by randomised arm.

The primary analysis will use a baseline-adjusted proportional odds (PO) ordinal logistic regression model (cumulative logit model) with the 7-point scale for body appreciation as the outcome variable. Pre-intervention (baseline) versions of the total score will be used as a covariate. These baseline pre-scores will be centred at the sample mean prior to inclusion in the models to aid interpretation of main effects. School is a stratification factor. The model will therefore comprise mean centred pre-intervention score (covariate), school as a stratification factor (fixed effects factor), and randomised arm (dummy variable coded).

The treatment effect will be reported as the common odds ratio (OR) with 95% confidence interval (CI) and associated p-value. This odds ratio represents the odds of being in a higher category on the 2-8 scale for the treatment group versus control.

The above-mentioned model assumes proportional odds. The Brandt test will be used to aid in the assessment of this assumption. If the assumption of proportional odds is not grossly violated, or with no material impact on interpretation, then this proportional odds model will be retained as primary. If the proportional odds assumption is grossly violated, then inferences will be drawn from a series of sensitivity analyses which will be performed irrespective of the proportional odds assumption consideration. The sensitivity analyses will be a series of seven binary logistic regression models on cumulative dichotomisations of the post-total score. Each model will include centred pre-total score, school as a stratification factor, and randomised arm. The results will be presented as a plot of odds ratios (plus 95% confidence intervals) to illustrate any non-proportionality.

As a pre-specified secondary analysis of the primary outcome, the proportional odds model will be extended to additionally include a covariate (baseline covariate) and randomised arm interaction term. If the interaction effect is statistically significant (p < 0.05, two-sided) then treatment effects will be reported at representative baseline values (e.g., low, medium, and high pre-scores) and represented graphically.

Moderation analyses: All moderation tests will be exploratory and will use the same proportional-odds framework as the primary analysis. Each moderator (gender, year group, comprehension) will be examined separately to avoid over-parameterisation. Each moderator is binary in nature, and as such, the same analysis process will apply in each case. The primary model will be the proportional odds ordinal logistic model with post total score as outcome and with the centred pre-total score, school, randomised arm, moderator (M) and an arm by moderator interaction term (arm × M). Of interest is the odds ratio for the arm by M interaction term. If the interaction term is statistically significant (p < 0.05), then the investigators will report the randomised arm odds within each level of M and summarise with a plot of predicted cumulative probabilities across levels of M.

Missing data: Complete-case analysis will be used for the primary model. If the amount of missing data post-intervention is between 5% to 15% then MAR multiple imputation will be used as a sensitivity analysis. If between 15% and 30% of data post-intervention is missing then best-worst, worst-best case NMAR scenarios will be employed. If > 30% of data is missing post-intervention, then results will be given descriptively.