Design: We carried out a split-plot design with four factors unblinded randomized experiment in July 2019. The Ethics, Research and Biosafety committees of the Mexican National Institute of Public Health evaluated and approved this study (approval number: 7-68S4-P62-19).
Recruitment The study was conducted in public elementary schools selected by convenience in the state of Morelos, Mexico. School principals from four selected elementary schools were invited to participate in the study. After explaining study objectives, activities, benefits and potential risks for the children, three of them agreed to participate in the study. Children from 7 to 12 years old from all grades (i.e., from first to sixth grade in the Mexican school system) were invited to participate in the study. Written consent forms were sent home with children and only those with consent to participate by their parents or guardians were included. Before starting study procedures, children were asked to assent to participate in the study.
Participant’s allocation. Participants were randomly assigned to one of four groups: 1) Nutrient Facts Panel (NF), 2) Nutrient Facts Panel with cartoon characters (NF + C), 3) Warning Labels (WL), and 4) Warning Labels with cartoon characters (WL + C). Randomization was done with raffle tickets indicating children´s allocation (i.e., children randomly chose a ticket from a basket). Blinding of participants or researchers was not possible given the nature of the intervention.
Experiment. After allocation, children assigned to both WL groups (WL or WL + C) were randomly required to watch two posters, both displayed together on a desk, (Image 2A) or a video (Image 2B) explaining how to correctly interpret WLs. Children were not given any other explanation on how to interpret WL. Children allocated to other study groups (NF and NF + C) were not provided with any interpreting aids.
The video (Supplementary Video 1) showed examples of products with and without WL (Image 1), visually displaying approval (i.e., a tick) or disapproval (i.e., a cross), respectively, and suggested choosing the healthiest food product (i.e., the one with no WL, with the fewest labels or the smaller number). It also mentioned that minimally processed foods were the healthiest . On the other hand, the posters (50x60 cm)(Image 2) displayed products along with smiling emoji if the product had no WL or dislike emoji if the product had two or more WL, based on previous emoji association with perceptions of healthiness with front of package labelling in children . The key messages of both interpreting aids (i.e., video and posters) were piloted and tested in school-age children to “choose the healthiest” and “it is better if it does not have a warning label”, based on the Chilean government WL campaign [11, 29].
A set of fictitious products were used per study group. Each set contained a total of 42 food products from 6 food groups, with different packaging sizes: 3 regular size groups (beverages in disposable packaging, breakfast cereals, and dairy beverages), and 3 small size products or products with returnable packaging groups (i.e., candies, cookies and beverages in returnable packaging) (Image 4 and 5).
All packaging looked the same across study groups, except for the WL and/or the cartoon characters displayed. Products showed no brands, no other labels, claims or advertising. In order to include a range of products with varying nutritional quality, the nutritional information and ingredients list of similar real products was used. This information was used to evaluate the nutritional quality of food products, according to the Mexican Nutrient Profile (Supplementary Table 1). Additionally, the following variations were considered in the set of products used in each study group:
1) Warning labels: Products used in the NF group only displayed the Nutrient Facts Panel, while for WL groups (WL and WL + C) the products displayed the Nutrient Facts Panel and the corresponding WL, going from zero warnings (the healthiest product) to six warnings (the unhealthiest product). Traditional WL (Image 4) were displayed in regular size food products, whereas numeric WL (Image 5) were displayed in small size products or products with returnable packaging. For regular size food products (> 10 cm2) the Nutrient Facts Table in the back of the package was displayed (Image 4).
2) Cartoon characters: Products used in cartoon characters’ groups (NF + C and WL + C) included one product with a cartoon character. This product displayed 4 WL in the WL + C group or the equivalent product in the NF + C. For candies group, non-nutrition facts table, neither cartoon character was displayed in the product package.
Children were asked to sit in front of a table. Initially, children were shown a set of seven products corresponding to the first food group (e.g., breakfast cereals). For this purpose, children were required to close their eyes while products were set on the table to avoid them from looking at the products before the experiment started. Then they were allowed to open their eyes and were asked: Which food product is the healthiest? Children chose one food product among the seven options. Then, children were asked: Which food product is the least healthy? Again, children chose one food product among the six remaining options. In both occasions, the time required to make a decision (i.e., time in seconds, starting when the child opened his/her eyes, up to when a decision was made) as well as the product chosen by the child were registered. This process was repeated for each of the remaining five food groups.
The order in which food groups were presented was randomly assigned using a total of six possible combinations for the three regular size groups and for the three small size groups. Additionally, the way in which the set of seven products was displayed on the table varied in order to avoid a sequence or order bias. Once their participation was concluded, children were rewarded with stickers and the fruit of their choice (e.g., a banana, apple or an orange). Additionally, the school received one soccer or volleyball ball per classroom for participating in the research project.
The proportion of children correctly identifying the healthiest product (i.e., product with no WL), the least healthy product (i.e., product with six WL) and both products (i.e., healthiest and least healthy product) was considered as the primary outcome. The time required to make a decision among study groups was considered as the secondary outcome.
Covariates. Before allocation, children answered a brief questionnaire that collected socio-demographic characteristics like sex, age, literacy, and habitual food shopping location.
Based on previous studies among Mexican consumers [25, 26, 33], and considering an alpha of 0.05 a power of 80% assuming a usual correlation between them of 0.1 (Cohen, 1992) , we estimated that a total of 70 children were required per study group to detect a difference of proportions of 10 percentage points between the WL (i.e. traditional and numeric) and comparison groups. Considering the four study groups we estimated that a sample size of 280 school children was required. Differences in the characteristics of the four groups of study were explored by using chi-square test.
Comparisons between WL (WL and WL + C) and NF (NF and NF + C) were used to explore the overall objective understanding of WL and the time required to make a decision. To examine the objective understanding of WL, logistic regression models were fitted to estimate the adjusted proportion (by sex, age and school cluster) of children correctly identifying the healthiest option, the least healthy option and both, across these groups. To examine differences in the time required to make a decision, quantile regression models were fitted to estimate the adjusted median time (by sex, age and school cluster) across NF or WL groups.
Differences in the objective understanding of traditional and numeric WL, as well as in the effectiveness of communication strategies to explain the correct use of WL, were explored among children assigned to WL groups (WL and WL + C) only. To explore differences in the objective understanding between traditional and numeric WL, Logistic regression models were fitted to estimate the adjusted proportion (by sex, age and school cluster) of children correctly identifying the healthiest option, the least healthy option and both, across products labeled with traditional and numeric WL. Differences in the time required to make a decision were also explored among these groups using a similar approach as the one described above.
To compare the effectiveness of communication strategies to explain the correct use of WL, the adjusted proportion of children correctly identifying the healthiest product, the least healthy product or both across communication strategies (video or poster) was estimated through logistic regression models. Generalized linear regression models adjusted by sex, age and cluster (school) were used to estimate the difference of median of seconds among communication strategies.
To estimate the impact of cartoon characters in the ability of children to choose correctly, a logistic regression model was fitted to estimate the adjusted proportion (by sex, age and school cluster) of children correctly identifying the healthiest product, the unhealthiest product and both, across cartoon character groups. Stratified proportions for NF and WL groups were displayed.
Stata v14 software was used to develop the statistical analysis.