Acceptability and understanding of front-of-pack nutritional labels: an experimental study in Mexican consumers

From May to June 2018, we developed an online three-arm unblinded randomized experiment. We used an adaptation of the model proposed by Grunert and Wills to study the effects of nutrition labels on consumers [22]. This model states that for nutrition labels to have any effect, consumers must be exposed to them and must perceive them. The effect will then be mediated by consumer understanding, as well as by the acceptability of the label.

This study is part of an online shopping simulation trial testing the effect of FOP labels on shopping intentions. After the shopping simulation, we employed a web-based tool to evaluate the acceptability and objective understanding of the assigned label. This study reports the results of this latter part of the trial. The ethics and research committees of the Mexican National Institute of Public Health approved the study protocol and instruments.

A convenience sample of adults (> 18 years old) was used for this study. Trained undergraduate student research assistants from eight universities across the country recruited the study participants. Two members of our research team (JVM and ACM) trained research assistants on how to approach and recruit participants and obtain informed consent. Research assistants were instructed to recruit at least 20 participants each, in any of the predefined public places. Places were selected by convenience by our research team, based on their use by low- and middle-income groups in Mexico. These places included public schools, public squares, public health centers, as well as supermarket chains and shopping centers located in low-income neighborhoods. At each place, research assistants approached all potential participants and, after explaining study objectives, invited them to be part of the study.

Then, research assistants accessed a unique web address where our web-based tool was hosted using a tablet or laptop with internet-access, and individuals were screened for eligibility using a 3-item screener. Only adults (> 18 years old) consuming at least one of the five food groups included in the shopping site (salty snacks, beverages, dairy products, breakfast cereals and ready-made foods) and who shopped for groceries at least twice per week were eligible. If the potential participant or any of their direct family members worked in the food and beverage industry, they were excluded from the study. The web-based tool automatically informed research assistants if the participant was eligible. In such case, informed written consent was obtained. Then, research assistants handed participants the tablet or laptop, where participants completed a demographic and health survey, and accessed an online shopping site to simulate a shopping situation. After the shopping simulation, participants answered a questionnaire related to the acceptability of the assigned label and then they were required to do some exercises designed to test the objective understanding of the label.

Based on previous studies [30], and considering a significance level of 0.05 and a power of 80%, we estimated that a total of 832 participants were needed in each label group to detect a difference of 6.2 percentage points (the smallest difference reported) in the proportion of participants reporting to like the GDA versus the MTL. Considering three label groups, the ideal total sample size was determined to be of 2496 participants.

Before entering the online shopping site, our web-based tool automatically assigned participants to one of three FOP labels (Fig. 1) using a simple randomization algorithm, blinding the research assistants to the assigned condition:

Blinding of participants was not possible given the nature of the intervention. After completing the demographic and health survey and being assigned to a group, participants were required to view a video explaining how to correctly interpret the assigned label. These videos corresponded to official videos used in each country (i.e. Mexico, Ecuador and Chile) to promote the correct use of the current labelling.

Label acceptability was evaluated using three indicators: likability, attractiveness and perceived cognitive workload. These dimensions were based on the framework of system acceptability developed by Nielsen [32], which have been previously used to explore acceptability of FOP labels in French consumers [30]. Participants were asked to rate the assigned label using 10 statements, adapted from recent FOPL studies in French consumers [30]. To evaluate likability of the label, consumers were asked to rate their agreement with the following statements: 1) “I like this label”, 2) “I want to see this label on the FOP’s”, and 3) This label will help me choose a healthier product”. To evaluate the attractiveness of the label the following statements were used: 1) “This label does not catch my attention”, 2)“This label provides me with the information I need”, 3) “This label is easy to identify”, 4)” This label provides reliable information”. Perceived cognitive workload was evaluated with the following statements: 1) “This label is too complex to understand”, 2) “This label takes too long to understand”, and 3) “This FOPL makes me uncomfortable” [30]. Five response options were included: 1) Strongly agree, 2) Agree, 3) Neither agree nor disagree, 4) Disagree, and 5) Strongly disagree. For each statement, participants choosing options 1) and 2) were classified as agreeing, whereas those choosing options 3), 4) or 5) were classified as disagreeing.

We tested the objective understanding of FOPL with a series of five exercises per participant, corresponding to the five food categories included in the online supermarket (sugary drinks, salted snacks, cereals, dairy products and ready-to-eat foods). Participants were asked to choose the product with the lowest nutritional quality among three products of the same food category.

For this purpose, we used the same set of 60 products used in the online shopping site (12 products for each food category), all of which are normally bought on a weekly basis. These products were selected from a database collected by the Mexican National Institute of Public Health between 2015 and 2016 [33]. We used the Nutrient Profiling Scoring Criterion (NPSC) model to assign all foods a healthfulness score and select a variety of products ranging in nutritional quality [34]. The nutritional quality and nutritional content of the foods included in the virtual supermarket is shown in Additional file 1. Underlying nutrition criteria for each label were used to assign the corresponding label to food products [10, 26, 35]. However, the criteria for classifying nutrient content as high in the MTL and the WL were based on 2016 Chilean regulations [10].

All exercises followed the same dynamic. Participants were shown 3 products ranging in nutritional quality, randomly selected from the list of 12 products of the same food category. The image of each product displayed the label on the FOP. Labels were positioned in the lower right corner of the FOP, covering roughly the same surface area. Additionally, the amplified image of the label was shown on the top of each product. Fig. 2 shows how products were displayed for each of the labelling conditions. Participants were instructed to drag and drop the product with the lowest nutritional quality in a box area in less than 30 s. This procedure was repeated for each one of the five food categories. In order to prevent any ordering effects, we also randomized the order in which the food categories were presented.

The time participants required to identify the product with the lowest nutritional quality was automatically registered by the system.

Information on gender, age, monthly income per household, education level, occupation, body mass index (BMI), presence of chronic conditions, and nutrition knowledge was collected. Only included close-ended questions were used, taken from previously validated surveys with pre-established responses [36].

We tested randomization of observable demographic and health characteristics of participants by comparing variables between experimental groups using Chi-squared tests (for categorical variables) and linear regression models (for continuous variables). Randomization successfully created equivalent groups on all demographic characteristics, so subsequent analyses did not control for these variables. Since time variables were skewed, medians and interquartile ranges were estimated.

To evaluate the acceptability of the labels, we estimated the proportion of participants agreeing to each of the 10 acceptability statements. Differences between proportions were estimated using Chi-squared tests.

To compare the understanding of the assigned labels, we estimated the proportion of participants and odds of correctly identifying the least healthy option in all food categories and across food categories. Differences in objective understanding across demographic characteristics (i.e. gender, age category, household income, education level and nutrition knowledge) were also estimated by introducing an interaction term between the label group and the characteristic of interest (e.g., Label group X age category).

To test differences in the median time required to choose the product with the lowest nutritional quality, we used median regression models introducing the assigned label as the independent variable. Models were run for all food categories and across food categories.

The results of the acceptability of the labels is presented in Table 2.

The proportion of participants correctly identifying the product with the lowest nutritional quality was highest for the MTL and the WL in all food categories and across them (Fig. 3). When considering all food categories, participants had 4.00 (2.86–5.59) times the odds of correctly identifying the product with the lowest nutritional quality when using the MTL label and 4.52 (3.24–6.29) times the odds when using the WL, in comparison to the GDA (p < 0.05). When exploring the objective understanding of the label across food categories, participants in the MTL and the WL were about twice as likely to correctly identify the product with the lowest nutritional quality, compared to the GDA (p < 0.05). No differences were observed between the WL and the MTL, except for non-dairy beverages where the MTL outperformed the WL (p < 0.05).

We did not find differences in label understanding across sociodemographic characteristics (p value for all interaction terms > 0.05) (see Additional file 2). Differences in the proportion of participants correctly identifying the least healthy option in all food categories across labels were similar in the full sample and across groups of participants with similar demographic characteristics.

Fig. 4 presents the median and IQR for the time required to identify the product with the lowest nutritional quality across label groups for all food groups and across food categories. Table 3 presents the regression coefficients for each label group derived from median regression models. Both the MTL and the WL led to shorter median times required to identify the product with the lowest nutritional quality compared to the GDA. Similar results were observed across all food group categories, with the biggest differences for non-dairy beverages.

In terms of label acceptability, results suggest that the MTL was the most accepted among the three labels [38], followed by the WL. Approximately 90% of participants liked the MTL and considered it had a low perceived cognitive workload. These results are in line with previous work among Mexican consumers showing that the MTL are among the most liked label format in this population [20]. Studies among European consumers have also shown high acceptability of MTL [38]. Warning labels were also highly accepted by participants. To date, few studies have evaluated the acceptability of WL among consumers [39]. The successful implementation of any FOP label partially depends on its acceptability by consumers and policy makers [40]. Nonetheless, evidence from warning labels on tobacco products suggest that the effect of these labels is mediated by elicited negative emotional responses [41, 42]. Therefore, in this study we tested a red version of the WL, since previous qualitative research by our team suggested this change would improve label acceptability [31]. Our results are consistent with studies in Chile and Canada supporting the acceptability of these labels [39, 43]. Importantly, the high acceptability of the MTL and the WL confirms results from other studies suggesting that labels providing positive and negative (i.e. MTL) or only negative (i.e. WL) evaluation of foods do not cause any more discomfort to the consumer compared to those providing a neutral evaluation, such as the GDA [30].

Conversely, our study showed that the acceptability of the Mexican GDA was the lowest, with only approximately half of participants liking the label and considering it attractive, and more than half perceiving it required a high cognitive work load. Since this labelling format was introduced in 2012 as a voluntary label, and in 2016 as the mandatory FOP label in Mexico, the acceptability of the label could be expected to be higher due to the familiarity with the format. In fact, previous studies conducted in Canada and France after implementation of the GDA showed that this label format was the preferred among consumers, possibly due to their familiarity with the label [30, 44]. However, our data does not support this hypothesis. Our results, along with previous studies showing difficulties in the understanding of this labelling format [19‐21], confirm that GDA may not be the best labelling format among Mexican consumers.

The understanding of a label is a key factor when processing nutrition labels and making purchasing decisions [22]. Our results showed that the MTL and the WL were the most effective in helping consumers identify the product with the lowest nutritional quality, when considering all food categories, as well as across them. These results are in line with a review showing that color-coded labels and the inclusion of text in a labelling format, as in the case of the MTL and the WL, perform better than GDA to compare the nutritional quality of products [45]. Our results are in line with studies among Chilean consumers showing that WL are understood in any age group [11, 25]. Further, studies suggest that WL have also helped consumers understand the underlying driver addressed by the labels. For example, Correa and collaborators showed that one year after implementing this labelling in Chile, mothers with underage children were aware that WL aimed to reduce obesity [46].

In our study we were also able to explore label understanding across different population groups. Results showed that both the MTL and the WL were more effective than GDA across all subgroups studied, including low income and low education participants, who are the most nutritionally at risk [24].

A novel result of our study was the objective measurement of the time required by participants to identify a product of low nutritional quality when using GDA, WL and MTL. Consumers generally have limited time to process nutritional information provided in nutritional labels. Studies have shown that simple labels, as the WL or the MTL, reduce the cognitive effort and time needed to process the information compared to more detailed labels, like the GDA [45, 47, 48]. In line with these results, in this study participants assigned to the MTL and WL required, on average, approximately 11–12 s to choose the product with the lowest nutritional quality, whereas those in the GDA group required more than 15 s. These results are also in line with our results showing a higher perceived cognitive workload for the GDA, compared to the MTL or the WL. Previous scientific evidence has shown that an effective FOPL requires people take between 12 and 30 s to make an informed food selection [19, 21]. However, analysis from the online shopping task executed by these participants before evaluating the acceptability and subjective understanding of the assigned label showed that purchasing decisions were made in no more than 5 s [49]. Studies among consumers have shown that time required to make decisions can be as short as 0.04 s in simple tasks (e.g. choosing one’s favorite food), but when more complex comparisons are required time can increase up to 18 s [27, 28]. There is also evidence suggesting that consumers only glance the nutrition information and do not process the information further at the point of purchase [22], explaining shorter times involved in making purchasing decisions. In this case, participants were required to execute a specific task, which involved inferring the nutritional quality of the product from the information provided on the FOP. By our results it can be inferred that the MTL and the WL helped consumers identify the product with the lowest nutritional quality more quickly. Previous results by our research team have also shown that these labels help consumers make healthier purchasing intentions more quickly [49]. Effective FOP nutritional labels should help summarize all the information in a simple and easy to understand format to influence the customer’s decision-making process. It could be expected that a FOP label that enables making decisions in a shorter time, as the MTL and the WL, will be more effective in helping consumers make healthier purchasing decisions [22].

To our knowledge, this is the first Mexican experimental study that compares the three FOP labels currently used in Latin-America, among low to middle-income participants, who are the most vulnerable for non-chronic diseases and least nutritionally literate. Strengths of this study include the use of an experimental design, ensuring that the influence of confounding from observed and unobserved factors was minimal. We also demonstrated consistent effects across a variety of study outcomes, using objective measures of label understanding and time required to identify the product with the lowest nutritional quality. Despite the former, this study is subject to a variety of limitations. First, the recruitment process was not intended to provide a representative sample of low- to middle-income adults and we were not able to estimate a response rate, which may have biased our results. Therefore, the representativeness of our results is limited to populations with similar characteristics as our sample. Although the sample approximates the demographic profile of the Mexican population in terms of education level and prevalence of diagnosed diabetes and hypertension [50‐52], and the distribution of household income is similar to that reported by national estimates [37], the reported prevalence of overweight and obesity as well as dyslipidemias was considerably lower [53, 54]. This may be partially explained by the known under-reporting for weight and over-reporting for height; however, it may also mean that our sample had healthier food patterns and therefore may have been more likely to use labels than the general population. On the other hand, it is also possible that given that our sample was comprised of low- and middle-income participants, their self-perception of overweight or awareness of other chronic conditions was lower compared to the general population [52, 55]. Second, we only tested three types of labels and did not include a no-label condition. This decision was made because GDA is the mandatory label in Mexico. Although including a no-label condition would have provided insightful information for other regions where FOP labels have not been introduced, this study design provides key information for decision-makers in Mexico and contributes to labelling efforts in the region by including some of the most relevant labels in Latin America. Third, although we considered real food products and prices, our study was limited in its ability to replicate the real shopping experience. Therefore, our study provides information on the mediators of the effect of FOP labels (i.e., label acceptability and objective understanding); the effect of the labels on real-life purchasing decisions cannot be inferred from these results. Fourth, images of the products only displayed the FOP, and the potential interaction effects with the nutrition fact panel or the list of ingredients was not captured. However, we used real food products, and no nutritional or health declarations on the FOP were removed, allowing participants to make decisions considering these elements too. Finally, our sample size was smaller than the ideal size (i.e. < 832 participants per label group). This may have limited our ability to identify small differences in the acceptability and understanding of across labels (i.e. differences between the MTL and the WL).