**Sugar Content in Processed Foods in Spain and a Comparison of Mandatory Nutrition Labelling and Laboratory Values**

**María José Yusta-Boyo 1,\*, Laura M. Bermejo 2,3, Marta García-Solano 1, Ana M. López-Sobaler 2,3 , Rosa M. Ortega 2,3 , Marta García-Pérez 1, María Ángeles Dal-Re Saavedra <sup>1</sup> and on behalf of the SUCOPROFS Study Researchers** †


Received: 21 February 2020; Accepted: 10 April 2020; Published: 13 April 2020

**Abstract:** To reduce the sugar content of processed foods through reformulation, the first step is to determine the content of the largest sources of sugars in each country's diet. The aim of this work was to describe the sugar content in the most commonly consumed processed foods in Spain and to compare that sugar's labelling and laboratory analysis values (LVs and AVs, respectively) to confirm its adequacy. A sample of the 1173 most commonly consumed processed foods in Spain (28 groups; 77 subcategories) was collected. For each product, the total sugar content was compared according to its AV and LV. The median (25th –75th percentiles, interquartile range) sugar content by group was calculated for the total sample, and the groups were classified as "high sugar content" when this value was above 22.5 g/100g of product. The adequacy of the LV, according to the European Union (EU) tolerance requirements, was then evaluated, and each subcategory median was compared with the AV to determine its appropriateness via a median test for independent samples (*p* < 0.05). In total, 10 out of 28 groups presented high sugar content. Moreover, 98.4% of the products met the EU tolerance ranges. Finally, only one subcategory ("cured ham") presented significant differences between the AV and LV median values (0.4 g vs. 0.1 g sugar/100g, *p* < 0.05). The groups of food products whose sugar content reduction could have the greatest impact on public health were identified. In addition, our study showed the high adequacy of LV with the EU labeling tolerance requirements, as well as the LV's appropriateness as a tool to implement actions aimed at reducing sugar consumption.

**Keywords:** nutrition labeling; food labeling; food processing; nutrition policy; Spain; food analysis; dietary sugars; reformulation

#### **1. Introduction**

The prevalence of overweight, obesity, and related non-communicable diseases (cardiovascular diseases, diabetes, and cancer) remains high in all European countries, including Spain [1].

The impact of dietary risk factors on the mortality and morbidity associated with non-communicable diseases highlights the importance of implementing measures to improve the quality of citizens' diets within national health policies [2].

Diets must meet energy needs and provide a variety of foods of a high nutritional quality that are safe to consume. Moreover, these diets should be sustainable, affordable, accessible, and culturally acceptable [3].

The European Union (EU) has long promoted initiatives to tackle obesity and to improve nutrition in European countries. One of the main initiatives of the European Commission was the adoption of the White Paper of 30 May 2007 entitled 'A Strategy for Europe on Nutrition, Overweight and Obesity related health issues' [4], focusing on actions that can be taken at the local, regional, national, and European levels. One of the initiatives included in this document is for the food industry (including retailers) to reformulate its products, particularly by reducing the content of salt, sugar, and fats.

In this regard, the High Level Group on Nutrition and Physical Activity (HLGNPA), composed of representatives of EU Member States and the European Commission, launched two EU Frameworks for the reformulation of food products: the EU Framework for National Initiatives on salt reductions [5] and the EU Framework for National Initiatives on Selected Nutrients [6] with two annexes: Annex I on saturated fats [7] and Annex II on added sugars [8]. These EU frameworks and annexes establish benchmarks and timelines for nutrient content reduction, focusing their action on certain food categories while taking into account the priorities, health needs, baseline nutrient contents, traditions, and pattern of consumption of each member state.

For more than a decade, the Ministry of Health in Spain, through the NAOS Strategy (Strategy for Nutrition, Physical Activity and the Prevention of Obesity) of the Spanish Agency for Food Safety and Nutrition (Spanish acronym AESAN, formerly Spanish Agency for Consumption, Food Safety and Nutrition, AECOSAN) has promoted reformulation initiatives for food and beverages, following the recommendations outlined in the HLGNPA Frameworks. To establish these initiatives, different studies have been carried out to determine the nutrient consumption and main food sources of the population and to ascertain the nutrient content (mainly fats and salt) of processed products [9–13]. The results of these studies have facilitated measures to reduce fats and salt in the main processed foods in Spain. Among these initiatives are the agreements between AESAN and food sector associations to achieve nutrient reduction targets, which were committed to all companies that belong to the sector association [14]. A successful example of public–private collaboration to achieve the reduction of salt content is the agreement (legal document) between AECOSAN, the Spanish Confederation of Bakers (CEOPAN by its Spanish acronym), and the Spanish Association of Manufacturers of Frozen Dough (ASEMAC by its Spanish acronym) signed in 2004, in which a salt reduction from 22 g NaCl/Kg in bread flour in 2004 to a maximum of 18 g NaCl/kg by 2008 was agreed upon. The average NaCl content measured in 2008 was 16.3 g of NaCl/Kg in bread-making flour [14]. In addition, a new study conducted in 2014 concluded that the salt content in bread in Spain has remained stable since 2008 [15].

However, no study has yet been conducted to ascertain the sugar content in processed products in Spain in order to establish reference values for addressing public health food policies, such as the reformulation or improvement of processed food and beverage composition.

For this reason, at the end of 2016, AESAN conducted the present study to describe the sugar content in the food groups included in the Annex II of added sugar [8] and in the most commonly consumed processed foods in Spain, especially by children and adolescents [16]. In addition, the other secondary objectives were to compare the label value (LV) with the laboratory analysis value (AV) in order to assess the adequacy of the LV based on the EU labelling requirements, and to study the appropriateness of using the label values as reference data to monitor the reformulation of sugar and other nutrients or for strategies such as front-of-pack labelling or marketing restrictions.

#### **2. Materials and Methods**

#### *2.1. Sample Selection*

We selected 28 food and beverage groups of processed foods to perform the present study (Table 1), starting with the 11 groups recommended in the Annex II (on added sugar) in the HLGNPA [8]. Some

of these 11 groups were divided due to the diversity of products included in each of them. For example, "Sugary dairy products and other similar products" was divided into six groups: flavoured milk drinks, drinking yoghurt, yoghurt, dairy-based desserts, cheeses, and soy drinks. Moreover, according to the results observed in the ENALIA study (National Dietary Survey on the Child and Adolescent Population), the groups "milk", "juices", "nectars", and "meat products" were included because these groups represent an important source of energy for children and adolescents in Spain [16].


**Table 1.** Selected food and beverage groups for this study.

The 28 processed food groups selected were classified into 77 subcategories according to their composition and legal denomination. For some analyses, classification by subcategory was used; this type of classification is considered more appropriate than classification by group due to the variability within a group. The number of products to be analyzed in each food subcategory was

established by consensus among researchers. Food products were selected from among those with the greatest presence in the national market at that time (including both brand names and retailer brands), according to data published in the report on the Alimarket study in 2015 [17]. This report provided the most important economic and financial variables by sector in Spain, including information about 9286 companies in the food sector.

#### *2.2. Data Collection*

Once the groups and subcategories to be studied were selected, the total number of processed product samples was 1173. The study was awarded through public tender to "AENOR Laboratorio Alimentacion" (AENOR Food laboratory), which carried out the plans to purchase the products. All samples were acquired in October 2016 and transported under adequate storage conditions to "AENOR Laboratorio Alimentacion" to proceed with the storage and subsequent determination of the total sugar AV of each processed product. In addition, the total sugar LV declared in the mandatory nutrition labelling (MNL) was recorded.

The Luff–Schoorl method was used to measure the total sugar AV. This method is the official method to control sugars intended for human consumption in Spain [18] according to the European legislation First Commission Directive of 26 July 1979, which outlines the community methods for testing certain sugars intended for human consumption (79/786/EEC) [19]. This method involves the elimination of all reducing materials other than sugars present in the sample by drying; subsequently, the sugar content is assessed based on the reducing action in a cupro–alkaline solution.

In order to collect the total sugar LV, we considered the proposed methodology for the monitoring and food reformulation initiatives of the Joint Action on Nutrition and Physical Activity (JANPA) of the European Commission [20]. In this methodology, the MNL was accepted as the data source to collect nutritional information.

#### *2.3. Statistical Analysis*

The data collected were recorded in a database designed ad hoc (for this study). This procedure was carried out in other countries as a tool for the reformulation and monitoring of processed food and beverage composition (Oqali database) [21]. Statistical analyses were performed using the SPSS software (SPSS, version 25.0; SPSS, Chicago, IL, USA).

For a descriptive study of the total sugar AV and LV, the median (25th–75th percentiles, interquartile range) were calculated for each food group, in the total sample (*n* = 1173), and in the two subsamples, according to the presence of nutrition claims about sugar content (light, low sugar content, no added or zero sugar, etc.) on the label (*n* = 64) vs. the absence of a sugar-based nutritional claim (*n* = 1109). Moreover, an adequacy study of the LV based on the EU labelling requirements, using the tolerance of the values declared on the labelling [22], was conducted. The final sample used to study adequacy of the LV included 1074 products (excluding products with nutritional claims and without a LV). A product met the tolerance range ("Meets") when its LV < 10 g/100g of product and its deviation from the AV was ±2 g; when its LV was 10–40 g/100 g and its deviation from the AV was ±20%; or when its LV > 40 g/100 g and deviation from the AV was ±8 g. Any product outside of these tolerance ranges was classified as "Does not meet".

Finally, an appropriateness study of LV as reference data for reformulation, monitoring, and other strategies was conducted. For the sample consisting of 1074 products, the LV and AV medians of each product subcategory were compared using the SPSS Median Test for 2 Independent Medians (*p* < 0.05).

#### **3. Results**

A flowchart with the sample selection and the different analyses carried out in the present study is detailed in Figure 1.

**Figure 1.** Flow diagram. Sample selection and the different analyses carried out.

#### *3.1. Descriptive Study of the Total Sugar AV and LV*

Table 2 provides the results of the AV statistical analysis median (25th–75th percentiles, interquartile range) of the total sample analyzed (*n* = 1173). In addition, the analysis of the subsamples according to the presence or absence of a nutrition claim about sugar content showed that the total sugar AV was higher in products without a nutritional claim (*n* = 1109) than in those with such a claim (*n* = 64).

Figure 2 presents the median content of the total sugar LV for each group of products studied in the total sample. The 25th-75th percentiles and interquartile ranges (IQRs) are also shown.






**Table 2.** *Cont*.

 **Claims**

 **IQR**

 2.4

 1.7

 1.2

 4.2

 15.4

 3.1

 24.1 Including non-dairy desserts (jello, "tocino de cielo" (pudding made with egg yolks and syrup), and chocolate cake) and powders for dessert preparation (flan powder, cake powder, and chocolate cake powder). 2 For groups with 1 or 2 items, the median was not calculated, and the data are presented in the IQR column separated by a comma; IQR = Interquartile range.

*Nutrients* **2020**, *12*, 1078

Moreover, according to data from the AV of products without nutritional claims (*n* = 1109), the food groups were classified as high total sugar content groups (*n* = 10) and no high total sugar content groups (*n* = 18), with the value 22.5 g/100 g as the cut-off point according to the criterion used for the front of the pack in the UK (25% of the recommended intake in Annex XIII of the EU regulation 1169/2011) [23] and in Chile's Law on Food Labelling and Advertising [24]. There were a total of 10 groups with high total sugar content (in descending order): sweets, other sweets, jam, chocolate, confitures, baking and pastries, desserts, breakfast cereals and cereal bars, biscuits, and ice creams. For the dispersion within each group, the estimated interquartile range was greater than 15 g/100 g in three groups (desserts, sauces, and sweets), between 15 and 10 g/100 g in six groups (biscuits, other sweets, jam, breakfast cereals and cereal bars, baking and pastries, and chocolate), and lower than 10 g/100g in 19 groups.

#### *3.2. Adequacy Study of the LV Based on the EU Labelling Requirements for the Tolerance of the Values Declared on the Label*

An adequacy study of the LV compared to the EU labelling requirements was conducted considering the tolerance for the values declared on the label [22]. Tolerance refers to the acceptable difference between the nutritional values declared on the label and those established over the course of official controls in relation to the "nutritional information" or "nutritional labeling" described in Regulation (EU) No. 1169/2011 on the provision of food information provided to consumers [25].

For this study, 64 products with nutritional claims related to sugar content were excluded (Table 2) since their sugar content was reduced to comply with the requirements of Regulation (EC) No. 1924/2006 on the nutritional and health claims made for foods [26], the tolerance requirements for these products are different and are beyond the scope of this study, which focuses instead on products that can be reformulated. In addition, 35 products without LV data (0.09% of the total samples analyzed) were also excluded (11 meat products, five sweets, four ready meals, four from other sweets, three crisps, three savoury snacks, two chocolates, one sauce, one dessert, and one fruit in syrup). Notably, the obligation to show the mandatory nutrition labelling (MNL) (according to Regulation (EU) No. 1169/2011 on the food information provided to the consumer) has been in force since 13 December 2016 [25], but the sampling for this study took place earlier (October 2016). Therefore, the final sample included 1074 products. Therefore, the final sample analyzed was 1074 products.

Of the aforementioned 1074 products, 1057 (98.4%) met the tolerance ranges, while 17 products did not, among which only five (0.45% of the total sample) declared a total sugar LV in the NML that was lower than an AV: one in the "baking and pastries" group (*n* = 60; 1.7% of the group's products), one in the "breakfast cereals and cereal bars" group (n = 106; 0.9%), one in the "special packaged bread" group (*n* = 45; 2.2%), one in the "desserts" group (n = 24; 4.2%), and one in the "dairy-based desserts" group (*n* = 65; 1.5%), while the remaining 12 (1.1% of the total sample) declared a value in their NML greater than the AV (Table 3).

Table 4 shows the 17 products that did not meet the tolerance ranges.


**Table 3.** Products that meet or do not meet the EU tolerance ranges.

**Table 4.** Individual label values (LVs) and analytical values (AVs) of the products that did not meet EU tolerance ranges.



**Table 4.** *Cont*.

*3.3. Appropriateness Study of Using the LV as Reference Data for Reformulation, Monitoring, and other Strategies*

Table 5 shows the medians and the 25th and 75th percentiles of the LV and AV data obtained for each group and subcategory of products. Of the 28 groups studied, only the "meat products" group presented significant differences between both medians, with the LV being greater than the AV (LV: 1.0 (0.0–5.0) g vs. AV: 0.7 (0.1–4.2) g, *p* < 0.05). In addition, in the study by subcategories, within the group of meat products, "cured ham" was the only subcategory that presented significant differences (LV: 0.4 g vs. AV: 0.1 g, *p* < 0.05).

**Table 5.** Comparison of the label value (LV) and analytical value (AV) medians of the total sugar for each group and subcategory of products (appropriateness study).



**Table 5.** *Cont*.


**Table 5.** *Cont*.


**Table 5.** *Cont*.

\* Significant differences between the labelling value (LV) and laboratory analysis value (AV) (median test for unpaired samples); <sup>1</sup> Powder for dessert preparation (flan powder, cake powder, chocolate cake powder, etc.); <sup>2</sup> Sweet gels, liquorice, marshmallow, chewing gum; <sup>3</sup> Wheat rinds, pork rinds, potato sticks, crackers.

#### **4. Discussion**

To study the content of total sugar, we presented the AVs of 28 groups of processed food products that are most frequently consumed by the Spanish population. The groups whose reformulation could have the greatest impact on public health were identified by their high sugar content and dispersion, which indicates that there is room for the reduction of their sugar content and that reformulation is, therefore, possible.

The most commonly consumed groups, especially by children and adolescents, should also be distinguished and prioritized (compared to those consumed more rarely) based on their energy contribution to the diet (cereals and meats and derivatives, among others), added sugar contribution to the diet (sugar sweetened beverages, chocolate, and nectars), or both (dairy products, baking and pastries, and breakfast cereals, among others) [16,27,28]. In Europe, high sugar consumption, especially in children and adolescents [29], has become a major public health concern, which is highlighted by scientific reports and studies associating high sugar consumption with an increased risk of dental caries [30], overweight [31], cardiometabolic risk factors [32], and adult cardiovascular mortality [33].

This study, promoted by the Observatory of Nutrition and Study of Obesity of AESAN in the framework of the NAOS Strategy, answers the call to action of the EU from the Council Conclusions of June 2016 for the improvement of food to develop a national reformulation initiative (the "Collaboration Plan for the improvement of the composition of food and beverages and other measures 2020" [34]), which is in line with the EU Plan of Action against Childhood Obesity 2014–2020 [35] and the WHO European Action Plan for Food and Nutrition 2015–2020 [36].

One of the limitations of this study is its small sample size, which was mainly due to our limited budget. For some subcategories, this reduced sample size was due to other specific reasons. For "pineapple in syrup", the market is dominated by a single brand, and for "beverages with fruits" (included in the group "sugar sweetened beverages"), we decided to create a new subcategory, resulting in only three products; sugar content in fruit is very different from that in the rest of the group, so the groups could not all be considered together. For this reason (sample size) and to prevent the influence of the most distal values, the median was used as a measure of the central tendency. Moreover, to produce a snapshot of sugar availability in our food environment and to estimate the potential impact of reformulation, market shares should have been considered. In our more feasible approach, the products in each subcategory were selected according to the results of a food market study that identified the most commonly consumed food products.

Our results provide novel information on the presence of total sugars in our food environment. This study quantifies the total sugar content in the main groups of processed products on the Spanish market, assesses the adequacy of label values based on the tolerance ranges for sugar according to EU labelling requirements, and assesses the appropriateness of using the LV as reference data. These are relevant aspects for designing and implementing actions aimed at reducing sugar consumption, which will help tackle obesity and its consequences for health.

Reformulation policies aimed at reducing the content of certain nutrients are some of the measures recommended by international institutions and organizations (European Union, WHO, OECD) to improve the quality of diets, reduce the consumption of foods high in salt, fats, and sugars, and prevent obesity and its related non-communicable diseases [37].

The Annex II on added sugars [8] proposes that the Member States should set a general benchmark for a minimum of 10% added sugar reduction in food products against their baseline levels or to move towards a 'best in class' level of sugar content.

In Spain, in order to establish the sugar reduction objectives of the "Collaboration plan for the improvement of the composition of food and beverages and other measures 2020" [34], AESAN opted in 2017 to deploy the first strategy mentioned in Annex II, which entails the reduction of a percentage of sugar from the basal median content, which was considered a viable and realistic method. Thus, all companies in the sector related to each food subcategory made a commitment to follow this plan. The evaluation of this plan, after its completion in 2020, will allow us to determine the degree of compliance with the objectives and to draw pertinent conclusions about the contribution of this initiative and its possible "drag effect" on other subsequent initiatives.

This work shows that most of the analyzed processed products (98.4%) meet the European Union tolerances for nutrient values declared on the labels. A very low percentage of products in our study did not meet the tolerance values because they had a lower LV than AV (0.45%). A study conducted with a sample of products that contribute the most to sodium intake in the United States [38] concluded that the majority of the labeling and analytical values agree with each other; thus, label under-declaration is limited. However, the authors observed that the differences in total sugars were greater and more systematic and that 19% products did not meet tolerance requirements because their labelled total sugars were lower than their analytical data. In our study, each of the five products that did not meet tolerance levels (due to a lower LV than AV) belonged to a different group (baking and pastries, breakfast cereals, special packaged bread, desserts, and dairy-based desserts), which also included a considerable number of other products that did meet the tolerance requirements.

To our best knowledge, this is the first time that the LVs and AVs of the studied processed products groups were compared in Spain, showing that the total sugar values are similar under both methods for most of the products and subcategories, according to the tolerance requirements for the nutritional information established by the European Union. Of the 28 food and beverage groups and the 77 subcategories analyzed, only the "meat products" group and, specifically, the "cured ham" subcategory, showed significant differences between their median AV and LV data. Taking into account the low sugar content of cured ham and the specific characteristics of its manufacturing and maturation process, such as the infiltration variability of additives depending on the part of the product (fat, bone, etc.), these differences are not considered relevant.

In light of these findings, we conclude first that a reduction in sugar content is feasible in a wide range of products. We recommend setting benchmarks at the subcategory level because this level includes similar products to which the same quality standards apply and allows one to compare the nutrient content of each product within a subcategory to its median, thus facilitating the identification of products with the greatest potential for reformulation.

Moreover, our results show a remarkably high compliance with the tolerance requirements and the appropriateness of the declared total sugar content in the MNL for most sold packaged processed products in Spain.

Thus, the MNL provides an accessible and efficient tool for various aims: to inform consumers truthfully, to conduct studies on the sugar content or other nutrients in labeled products, to establish and monitor reformulation initiatives, to implement front-of-pack initiatives, to apply nutritional profiles for different objectives, and for food advertising policies, among others. Regulation (EU) No. 1169/2011 [25] establishes (starting from 13 December 2016) the mandatory obligation to provide a nutritional declaration on the labels of most processed food products. This declaration must include the energy values and six nutrients, one of them being the sugar content.

Briefly, the groups identified to boost reformulation policies based on their sugar content, the differences in sugar content between similar products, and their contributions to energy, added sugars, or both to the diet are the following: sweets, other sweets, jam, chocolate, confitures, desserts, baking and pastries, breakfast cereals and cereals bars, biscuits, ice cream, sauces, meat products, sugar sweetened beverages, nectars, and dairy products [16,27,28].

The results obtained from this research may help make the nutritional composition of food products more visible, to better explore the feasibility of improving nutrition and evaluate related actions.

#### **5. Conclusions**

The results of the present study will help identify the groups of food products whose sugar content reduction could have the greatest impact on public health. In addition, we showed the adequacy of labelling values with the EU labelling tolerance requirements; labelling values are, therefore, an adequate tool to implement and evaluate actions aimed at reducing sugar consumption.

**Author Contributions:** Conceptualization and methodology, M.J.Y.-B., L.M.B., A.M.L.-S., R.M.O., M.Á.D.-R.S., and SUCOPROFS; formal analysis, M.J.Y.-B., L.M.B., M.G.-S., and A.M.L.-S.; original draft preparation, M.J.Y.-B., L.M.B., M.G.-S., and A.M.L.-S.; writing—review and editing, M.J.Y.-B., L.M.B., A.M.L.-S., M.G.-S., R.M.O., M.G.-P., M.Á.D.-R.S., and SUCOPROFS; funding acquisition and contract design, M.J.Y.-B., M.Á.D.-R.S., and SUCOPROFS. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received funding from the Spanish Food Safety and Nutrition Agency.

**Acknowledgments:** Technical support was provided by Estefania Labrado Mendo. Sample collection and analytics were done by the AENOR Laboratory. **SUCOPROFS (Sugar content in processed foods in Spain) Study researchers**. AGENCIA ESPAÑOLA DE SEGURIDAD ALIMENTARIA Y NUTRICIÓN. VOCALÍA ASESORA PARA LA ESTRATEGIA NAOS: Napoleón Pérez Farinós, Sara Santos Sanz, Carmen Villar Villalba, Mª Araceli García López, Teresa Robledo de Dios. UNIVERSIDAD COMPLUTENSE DE MADRID. FACULTAD DE FARMACIA. DEPARTAMENTO DE NUTRICIÓN Y CIENCIA DE LOS ALIMENTOS: Esther Cuadrado Soto, Aránzazu Aparicio Vizuete.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Consumers' Perceptions of the Australian Health Star Rating Labelling Scheme**

**Fiona E. Pelly 1,\* , Libby Swanepoel <sup>1</sup> , Joseph Rinella <sup>1</sup> and Sheri Cooper <sup>2</sup>**


Received: 6 February 2020; Accepted: 4 March 2020; Published: 6 March 2020

**Abstract:** The objective of this study was to explore consumers' use and perception of the Australian Health Star Rating (HSR). A purposive sample of fifteen Australian grocery shoppers was recruited into four focus groups using a supermarket intercept strategy. Focus group discussions were recorded, transcribed and analysed using an iterative approach to thematic analysis. Three key themes emerged from analysis. The HSR was seen as simple, uncluttered, easy to understand and useful for quick comparison across products. The nutrition information was viewed positively; however, there was little confidence in the HSR due to a perceived lack of transparency in the criteria used to determine the number of stars. Highly processed foods were generally seen as having inflated ratings and participants expressed concern that this would increase consumption of these products. Finally, there was a belief that the HSR had a lack of negative imagery limiting the dissuasive impact on consumers when presented with low-rated foods. Consumers saw benefits in the HSR but were sceptical about how the ratings were derived. Transparency about the development and education on the application may assist with consumers' perception of the HSR.

**Keywords:** front-of-pack labelling; health star rating; nutrition labelling; consumer perception; qualitative research

#### **1. Introduction**

Nutrition labelling on food allows consumers to be informed about the nutritional composition of the products they are purchasing [1]. Nutrition labelling on packaging has been demonstrated to improve consumers' ability to assess product healthiness and encourage healthier food choices [2]. The display of nutrition information on packaged goods is mandatory in many countries [3] and typically comes in the form of a nutrition information panel (NIP) and ingredients list. The NIP displays numerical nutrition information on the side or back of a package, which is not always obvious to the consumer. In contrast, front-of-pack nutrition labels (FoPLs) are more likely to facilitate exposure to nutrition information as they are visible at the moment of choice [4].

FoPLs can be categorised according to factual information versus a continuum proposed by Kleef and Dagevos [5]. At one end of this continuum, the 'purely reductive' FoPL presents factual information condensed from the NIP and leaves the evaluation up to the consumer. The impact is therefore likely determined by the consumer's understanding of the facts provided. At the other end of the continuum are the 'purely evaluative' FoPLs that are binary in nature because they depict whether a product meets a particular nutrition standard through the presence or absence of the label, typically with a simple graphic such as a tick or stamp. The impact is reliant upon the consumer's awareness of the meaning of the graphic and the evaluation of the nutrition standard that is being met.

In the middle of the continuum are 'hybrid' FoPLs that present a combination of information from the NIP and an evaluation of that information. A preference for the appearance and use of hybrid compared to purely reductive FoPLs has been found in previous research [2,6,7]. It has been suggested that consumers are generally able to perform tasks related to identifying healthier food products when using a hybrid FoPL compared to a purely reductive version [8,9].

In 2014, the Health Star Rating (HSR) [10] was introduced for use on packaged foods in Australia. The hybrid scheme displays an evaluative component based on an algorithm-derived star rating from half a star (least healthy) to five stars (most healthy). Foods high in energy, saturated fat, sodium or total sugar are assigned lower star ratings than similar foods with fewer of these components. The star rating is increased based on the amount of fruits, nuts, vegetables, legumes, and in some cases protein and dietary fibre in the food. The HSR also presents a reductive component, that is, numeric nutrient information per 100 g or 100 mL for energy, sugar, saturated fat and sodium and one additional positive nutrient [10]. The HSR is a voluntary scheme that has recently undergone a formal review after a five-year implementation period. The recommendations from the report suggest some changes should be made to the appearance and calculation of the HSR, but with continued support of the system [11].

The aim of this study was to undertake an in-depth exploration of consumers' perceptions of the HSR considering the visual layout, nutrient information provided and application to a select number of food products. A secondary aim was to explore how consumers' use nutrition labelling to inform decisions around the healthiness of packaged food.

#### **2. Materials and Methods**

This study was underpinned by descriptive phenomenology given that the underlying aim was to explore and describe consumers' experiences of the HSR. A qualitative approach with focus group discussions was used to gain a rich understanding and comparison of consumers' perceptions of the HSR. Participants were recruited by the third author (male) using supermarket intercept convenience sampling from two major supermarket chains in regional locations in Queensland, Australia. Two additional participants were recruited through snowball sampling. Participants were considered eligible for the study if they were over 18 years and reported to do at least half of the shopping for themselves or their household. Those with nutrition education were excluded from the study. No prior relationship existed between the researchers and participants. All participants gave their informed consent prior to participation. This study was approved by the Human Research Ethics Committee of the University of the Sunshine Coast (S/14/709).

Four semi-structured focus group discussions involving 15 participants were conducted and moderated by the third author who was previously trained in focus group facilitation. Focus group discussions were guided by an interview protocol designed by the research team based on inquiry logic that was informed by the literature and the study aims (Table 1). A pilot focus group was conducted, feedback sought and the order and wording of some questions were modified accordingly. Participants were asked if they consider the healthiness of food when grocery shopping, and if so, how this is determined. All other questions were delivered following the presentation of props, beginning with four A4 pages (297 × 210 mm) showing different examples of the full HSR format which includes the HSR, and the energy and nutrient icons [12]. This approach invited participants' initial impression of the visual layout and the nutrient information provided by the HSR outside of the context of a food package. Five pairs of nutritionally equivalent packaged products (breakfast cereal, artificially sweetened carbonated beverage, cordial, crackers and sweet biscuits) that differed in their FoPL scheme were used as food props. First, products labelled with the Daily Intake Guide (%DI) were presented, and this was followed by those labelled with the HSR. The %DI is a reductive scheme that represents the energy or nutrient content per serve as a percentage of a standard reference value [13]. This allowed for comparisons between labelling formats and also facilitated discussion about perceptions of the HSR within the context of food products. The food props were purchased from a major supermarket in regional Australia and were selected based on availability of HSR-labelled

products prior to commencement of the first focus group. Focus group discussions lasted between 35 and 65 min and were audio recorded and transcribed with permission from the participants.

**Table 1.** Questions from the focus group questionnaire on the Health Star Rating.


Focus group transcripts were thematically analysed by two members of the research team (LS and JR) using an iterative approach as described by Srivastava and Hopwood [14]. Analysis followed the process described by Green et al. [15], where researchers initially immersed themselves in the data, then conceptualised the parts of the data that addressed the research questions into codes. Similar codes were then grouped into broader categories and connections between categories were examined. Categories that emerged were considered important based on length, depth of discussion, order of emergence as well as tendency to appear in more than one focus group. Explanations and interpretations of categories as themes were discussed and agreed upon by the research team in a process of peer debriefing [16] in order to increase the trustworthiness of findings. As all researchers had expertise in nutrition and health, bracketing was employed to further improve the trustworthiness of the data, whereby the researchers attempted to suspend their own perspectives and biases in order to focus on the participants' descriptions of their experience during the focus groups [17]. Focus group recruitment ceased when no new relevant information emerged.

#### **3. Results**

Three men and twelve women participated in the focus groups. Nine were over the age of 50, three were aged between 35 and 49 and three were aged between 25 and 34; four participants reported to be educated at a postgraduate level, four at a bachelor level and the remainder had high school, diploma or trade training. Participants' purchasing behaviours were influenced by various factors including individual health conditions, personal nutritional priorities, allergies, food safety, weight control, taste and price. The NIP was used by most participants to help determine the healthiness of a product, rather than use of any particular FoPL. Sugar, fat, saturated fat and food additives were the nutrients of most interest.

#### *3.1. Themes*

Three key themes relating to participants' perceptions of the HSR emerged in the focus group discussions: (1) Practicality of the HSR; (2) Lack of confidence in the HSR; and (3) Lack of dissuasive impact of the HSR, as described below.

#### 3.1.1. Theme One: Practicality of the HSR

Participants' first impression of the HSR was that it appeared simple, uncluttered and easy to understand. The stars were a commonly recognized symbol that most participants understood to relate to the healthfulness of the product on a scale of half a star to five.

*"I think it's really straightforward, really obvious, really easy to get a quick glance at something of how many stars it is that's a really common visual reference for many people of one out of five stars or five out of five stars being good or bad so it's quite easy to read." (Participant 9)*

In each of the four focus groups, participants related the HSR to the energy rating used on electrical appliances in Australia [18]. This familiarity made it easier to understand how to utilize the label without guidance.

*"I think the stars are good in that people are already familiar with appliances, so you don't have to fully educate them on what the concept is."*

#### *(Participant 4)*

The explanatory text was also pointed out as being a useful visual aid as it provided context to the numerical information on the HSR. Most participants believed that the HSR would facilitate comparisons between similar products at a glance. In this way, participants felt that the HSR would influence them to purchase higher-rated products.

*"Yeah if I went in and saw my regular chips that I was going to get and they were like 1 star and these ones were right next to them and they're 3 stars, yeah sure I'd try them."*

#### *(Participant 10)*

Most participants felt that higher-rated products still required supporting information and verification that was only available by checking the NIP and ingredients list. Most participants preferred the NIP and the ingredients when making purchasing decisions as they were viewed to be more transparent, thorough and credible sources of nutrition information.

*"I think I'd still look at the table on the back and then make my own assessment from that. There's no spin or there's no kind of magic numbers."*

*(Participant 5)*

The HSR was seen by most participants to be aimed towards people who were time poor.

*"A busy mum with three kids who's doing a weekly shop isn't necessarily going to have time to thoroughly read the label."*

*(Participant 6)*

#### 3.1.2. Theme Two: Lack of Confidence in the HSR

There was a general lack of confidence in the HSR. Many participants viewed there to be a lack of transparency in the process used to determine the ratings. Participants felt there was an incongruence between their perceptions of the healthiness of a food product and the respective star rating. A number of participants were sceptical of the food industry, with concerns that food companies would change the nutritional makeup of their products to increase their HSR. These participants felt that rather than making the food products healthier, the reformulation would be superficial and exploitative.

*"...companies will just manipulate it. They will make subtle changes. Add things, take things out for their product to exploit the algorithm." (Participant 4)*

Participants were interested in the governing body behind the HSR and the nutrition science used to develop the algorithm. Although most participants trusted the HSR governance, there was suspicion around the evidence base underpinning the algorithm, with some participants voicing concern around the food industries involvement in boosting ratings.

*"The sceptic in me says that those scientists are lobbied by the food industry to present things that will be favourable towards agricultural, you know, whatever." (Participant 2)*

Other participants felt that consumers needed to be aware of how the algorithm works because an informed consumer is less able to be manipulated.

*"I support any measure that helps people make more informed choices on the nutritional value of food absolutely, I just want to be able to trust that that is, that people understand what's behind, how these things are rated."*

*(Participant 5)*

Participants voiced more concern around the ratings of foods towards the lower end of the scale. They felt that ratings seemed to be inflated for some foods they deemed to be 'junk foods' with little or no health value. These foods were products with ratings ranging from one and a half to two stars and included the cordial, sweet biscuits and the artificially sweetened carbonated beverages. It was generally the high proportions of sugar in the first two products and the additives in the third that led to dissonance with the ratings.

*"I still think that that rating is very high with all that sugar in it. I've got a problem with how they've come up with this rating, this number."*

*(Participant 15)*

Participants appeared to have more confidence in the HSR when rating foods at the higher end of the scale. This view is shown through the following quote relating to the five-star rating of a breakfast cereal.

*"Well five stars is pretty unequivocal, it's pretty clear. You're not going to get away with claiming 5 stars unless you can back it up."*

*(Participant 4)*

3.1.3. Theme Three: Lack of Dissuasive Impact of the HSR

Many participants expressed concerns that the HSR did not appropriately dissuade consumers from purchasing lower-rated products. This view is illustrated in the following quote relating to a product with a rating of two stars:

*"Ok that's less than 50% so logic would tell you that it's not so healthy but even so, 2 is sounding reasonable."*

*(Participant 15)*

It was suggested that low ratings could be accompanied by a symbol reinforcing the negative. Incorporating traffic light colouring into the HSR was also suggested in two of the four focus groups to address this perceived limitation.

*"Even like a colour scheme wouldn't be a bad thing if that one over there had a green star and this one's got a red star you're like "Woah, that's bad.""*

*(Participant 10)*

The framing of the label as a 'Health Star Rating' was also viewed as potentially confusing, as participants felt that this indicates that there is an absolute health value to any food with a rating.

#### **4. Discussion**

This study explored consumers' perceptions of the HSR. Three key themes emerged from the results, namely the practicality of the HSR, the lack of confidence in the HSR and the lack of dissuasive impact of the labelling scheme. The HSR was considered useful for quick comparisons across similar products at a glance due to its summary indicator, which was predicted to be particularly useful in the grocery shopping environment. The nutrient information provided on the HSR was considered to be important and useful to the participants in this study. Participants responded positively to the simplistic visual design of the HSR, contrary to a global study that found the HSR was perceived to have little visual appeal and 'not stand out' [19]. Elements that were identified as simple by the participants were as follows: (1) the explanatory text, (2) the uncluttered design, (3) the picture-based interpretation and (4) the familiarity of the stars. A study by Talati et al. [20] found similar results from focus group discussions, with the HSR being preferred over both the %DI and, to a lesser extent, the traffic light system due to the speed with which an evaluation could be made from the summary indicator saving time and effort while shopping. Similarly, simple FoPLs [21] or graphically representative FoPLs [22] have been shown to be liked by consumers and are considered easier to understand than labels with a lot of numbers and words. Many consumers evaluate as little information as possible to make their purchase decisions [23], and although nutrition content may be considered, it is often a lower priority when grocery shopping than factors such as price, taste and food safety [21,24].

Lack of confidence in the HSR was a key theme that emerged from this study. The provision of information on governance was considered essential to our participants. Transparency in the organisation behind the FoPL is known to increase credibility and trust [21], with well-known and trusted organisations being most credible for consumers [25]. The perception by some participants that labelling schemes are not backed by credible organisations may have influenced their confidence in the HSR by association. This finding suggests that consumers need to be educated about the governing body of the HSR as well as how its application onto products is regulated and monitored. The review of the HSR has suggested that there has been improved transparency in the information provided to consumers through the HSR system website, but greater confidence would be apparent if transferred to Food Standards Australia New Zealand [11].

Participants reacted differently to the idea that the implementation of the HSR may influence the food industry in reformulating their products. Some were concerned that reformulation efforts would be superficial and provide limited health benefits to consumers, whereas others felt that these changes would tangibly improve the nutritional profile of food products. Product reformulation of foods following the implementation of an FoPL has previously been successful [26,27]. The implementation of the HSR has the potential of encouraging food companies to reformulate their products by reducing levels of sugar, sodium and saturated fat, and also by increasing nutrients such as dietary fibre. This has been recently demonstrated in the reformulation of children's packaged foods [28].

Participants in this study also wanted to know how the star ratings were calculated before they could have confidence in the rating. Scepticism of the algorithm was highest when participants were presented with discretionary foods (artificially sweetened carbonated beverage, cordial and sweet biscuits), which they felt were rated too highly. Participants felt that, in these cases, the mechanics of the algorithm may be flawed, which is concurrent with previous studies indicating transparent labelling criteria to instil consumer trust [2,5,21]. This supports the recommendation to better align the HSR with the Australian Dietary Guidelines through changes to the calculator [11]. Consumers use food labels to make judgements about the food product as well as the food supply system behind the product [29]. Health and nutrition content claims on the front of the pack are often viewed by consumers as 'just an advertising tool' [30]. This was the case in our study, where participants felt distrust for the HSR, linking this to the food supply system by making judgements about the credibility of the food manufacturer. On the other hand, the NIP and ingredients list on the side or back of the pack (BoP) were viewed by our participants as highly trustworthy sources of information, which may reflect their confidence in how this BoP nutrient information is determined.

The participants in this study felt that the HSR had a limited ability to dissuade consumers to purchase lower-rated products due to its positively framed imagery. All foods eligible for the HSR scheme obtain, at worst, a half star rating, with the only negative communication being the optional explanatory text indicating a 'high' level of one of the key nutrients. It was suggested across several of our focus groups that incorporating some negative framing such as red colouring could help to address this. Similarly, traffic light colouring has been proposed to reduce the complexity of numerical information presented on the HSR [20], and interpretive aids such as colour are viewed favourably by consumers [19]. There is preliminary research to suggest that the inclusion of the traffic light colouring would be an effective way of modifying the HSR [31]. However, the recent review of the HSR did not support the use of the traffic light system [11].

While our recruitment strategy and methodology allowed for a rich understanding into consumers' perceptions of the HSR, it also led to some limitations in our findings. We conducted four focus groups at which point we reached data saturation, whereby no new themes emerged from the data, giving a high indication of the trustworthiness of our findings [32]. Data saturation is the optimal guide for sample size in qualitative research; however, it is known that a sample size of two or three focus groups will likely capture at least 80% of themes on a topic [33]. The qualitative exploratory nature of our study allowed participants to discuss issues around health, nutrition and labelling with high levels of passion, confidence and articulation, which suggests the findings from this study provide authentic insight into consumers' perception of the HSR, but they may not be generalisable to the wider Australian population. Our participant demographics also showed some variance when compared with the general grocery shopper population in Australia. Eight out of 15 (53%) participants had a bachelor or postgraduate education, compared to approximately 33% of Australian adults in the general population [34]. A further limitation of this study was the small range of HSR-labelled food products that were available for use as food props in the focus groups at the time of this study. An increasing number of products currently display the HSR, and it is recommended that future studies select foods that reflect the full range of products that consumers are exposed to in the supermarket environment.

This study provides an insight into consumers' perceptions of the HSR. The HSR was perceived as simple and easy to understand, and as being most useful when comparing across similar products at a glance. It was perceived as less useful for analysing single products in isolation and particularly for lower-rated products due to its positively framed communication design. While our results cannot be generalised to the wider Australian grocery shopper population, this sample of participants indicated that there was a lack of confidence in the HSR. This was due to a variety of reasons including a lack of familiarity, a lack of transparency on how the ratings are calculated and a disagreement with the product ratings.

#### **5. Conclusions**

There is a need to improve consumer confidence in the HSR to ensure accurate guidance when navigating the modern food environment. Campaigns to promote the use of the HSR should focus on improving consumer understanding and the evidence base underpinning the HSR.

**Author Contributions:** All authors were involved in the study design. Authors are listed in order of contribution. F.E.P. was the researcher leader on this study. J.R. transcribed the audio recordings following focus group training, and J.R. and L.S. conducted the thematic analysis. F.E.P., L.S., S.C., and J.R. contributed to the interpretation of the results and research write-up. The researcher held no assumptions and did not know the participants before the focus groups. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **How Much Sugar is in My Drink? The Power of Visual Cues**

#### **Bethany D. Merillat \* and Claudia González-Vallejo**

Ohio University, Athens, OH 45701, USA; gonzalez@ohio.edu

**\*** Correspondence: bl872911@ohio.edu; Tel.: (440) 829-7634

Received: 29 December 2019; Accepted: 30 January 2020; Published: 2 February 2020

**Abstract:** Despite widespread attempts to educate consumers about the dangers of sugar, as well as the advent of nutritional labeling, individuals still struggle to make educated decisions about the foods they eat, and/or to use the Nutrition Facts Panel. This study examined the effect of visual aids on judgments of sugar quantity in popular drinks, and choices. 261 volunteers at four different locations evaluated 11 common beverages. Key measures were estimates of sugar in the drinks, nutrition knowledge, and desire to consume them. In the experimental condition, participants viewed beverages along with test tubes filled with the total amount of sugar in each drink; the control condition had no sugar display. Both groups were encouraged to examine the Nutrition Facts Panel when making their evaluations. Correlational analyses revealed that consumers exposed to the visual aid overestimated sugar content and the length of time needed to exercise to burn off the calories; they also had lower intentions to consume any of the beverages. Individuals asserting to use the Nutrition Facts Panel (NFP) in general were also less likely to admit using it in this particular study (*r* = −2, *p* = 0.001). This study suggests that a simple visual aid intervention affected judgments and choices towards curtailing sugar intake. This has implications for labeling format implementation.

**Keywords:** nutrition facts panel; food label; consumer behavior; food decision making; food packaging; food choice; nutrition and health claims; food label; sugar; nutrition

#### **1. Introduction**

" ... sugar is cheap, sugar tastes good and sugar sells, so companies have little incentive to change [1] (p.29)."

A U.S. citizen consumes an average of 216 L of soda per year, of which 58% contains sugar [1]. The obesity problem in the U.S. is well documented, with obesity rates as high as 25% in 41 states, and above 20% in every state [2]. Importantly, child obesity has increased in the 1999–2016 period as a recent report shows [3]. Researchers have attributed the obesity problem, in part, to consumption of nutrient poor processed foods, containing high amounts of sugar, sodium, and saturated fats [4,5].

Added sugars alone are linked to a wide range of non-communicable diseases, including tooth decay, gout, heart disease, diabetes, obesity, and metabolic syndrome which is characterized by higher blood pressure, blood sugar, and triglycerides, and lower "good" cholesterol [6]. The health care costs associated with metabolic syndrome is estimated to be \$150 billion annually [1]. Furthermore, the United Nations World Health Organization places non-communicable diseases as the leading cause of deaths globally, responsible for about 68% of deaths worldwide in 2012 [7]. Given the relationship between added sugar consumption and metabolic syndrome, researchers have gone as far as to call sugar a toxin, and have proposed stricter regulations for added sugars similar to those controlling alcohol [1].

While the American Heart Association recommends that the population limit added sugar to six teaspoons a day for women and nine for men [6,7], the average sugar consumption in the

form of fructose is at an all-time high with current estimates at around 57.7g/day (or approximately 14.42 teaspoons-4g per teaspoon) accounting for 10.2% of total caloric intake. A study also placed sugar-sweetened beverages as the most important contributor of fructose intake (30.1%) [8].

#### *1.1. Strategies to Promote Healthier Eating*

Increasing the availability of healthy foods, and/or restricting specific types of foods, such as soft drinks in school settings, have proven effective methods to curtail poor nutritional consumption by both the Food and Agricultural Organization of the United Nations (FAO) [9,10] and the Centers for Disease control [11] (see also [12–14]. Curbing choices via pricing has also proven effective [15] (see the Chilean experience in Jacobs [16]). The FAO further notes that education-only interventions appeared less successful than those including environmental changes [10]. This poses a challenge for promoting better food choices outside school settings because price adjusting and/or limiting the supply of certain foods is more difficult to implement [16].

In a positive light, a number of prominent health campaigns are targeting the consumption of sugary drinks in many states (e.g., California's Kick the Can campaign; the Kansas' Just Add Water! public health intervention). However, a report from the Food and Agriculture Organization of the United Nations [10] concludes that public awareness campaigns, which take many different forms all over the world, have received mixed support regarding their effectiveness.

Another strategy to encourage healthy eating is by means of nutrition labelling. To address the issue of unhealthy eating, the US Nutritional Labeling and Education Act of 1990 [17] mandated the use of a standardized nutrition label (the Nutrition Facts Panel, NFP). The aim of the law was to provide consumers with nutritional information that was accurate and easy to read and encourage healthier food choices [18]. Studies by the US Agriculture Department found that the percentage of adults who reported using the NFP 'always or most of the time' went from 34% in 2007–08 to 42% in 2009–10 [19] and 77% in 2014 [20].

However, the assumption that the NFP indeed helps consumers to judge the nutritional quality of the foods and to make better decisions is debatable. In the 2014 survey, half of those who reported rarely or never using the NFP said they did not feel they needed to use the label [20], and several studies indicate that there has been no aggregate improvement of American nutrient consumption since the implementation of the NFP [21,22]. People may think that they do not need to use the label, but their health may be suffering because with the myriad of products in the market today, understanding of, and the ability to use the NFP can make a significant difference in one's ability to judge the healthfulness of food and drink options.

Current studies find positive and significant correlation between judgments of nutrition quality of foods based on the NFP, and a nutrition quality expert standard, but the levels of agreement are low [23,24]. More broadly, these studies and others investigating a host of other ecological factors, including those related to dietary choices, necessitate viewing the role of the NFP through the context of a multi-factored public health issue.

The FAO's 2013 report reveals a greater understanding of nutritional information form label usage, but not necessarily improvements in consumption [10]. Additionally, nutrient lists, which is the format used by the U.S. Nutrition Facts Panel (NFP), are often found to be confusing, and may disproportionally affect individuals having lower knowledge about nutrition and health. Why? While a number of factors certainly contribute to the problem, research has found that both those of lower socio-economic status and individuals with lower knowledge concerning nutrition and health are less likely to use such labels [25]. While more generally, research shows that the process by which food marketing affects food decisions is not well understood [26], and although a number of suggestions on how to improve consumer choice have been proposed, few are supported with empirical research [27]. Further investigation of packaged label use is still needed to determine whether they have a positive effect on nutritional understanding and decision making [28].

However, there are a number of other interventions which have both real-world applicability and have been proven to improve consumer choices. For example, research by Donnelly et al. found that evocative, graphic warning labels, as compared to text warning labels (calorie labels and no labels) significantly reduced the share of sugary drinks purchased in a cafeteria [29]. These graphic labels also served to heighten negative affect (toward unhealthy options) while promoting deeper thought concerning the health consequences of consuming sugary options [29]. We point this out to highlight that for an intervention to have success in changing consumer behavior, it must be both effective in research studies, and have the capability of being implemented and accepted in the broader consumer market (including both consumers and retailers).

While these studies have certainly played an important role in our understanding of the power of visual aids, it is important to consider that they may have limited real-world applicability because of the difficulty of adopting them in the market, and making them visible/available to consumers on a wide-scale.

#### *1.2. The Present Study*

The central and negative role of sugar in human health has been identified in numerous sources (e.g., Williams and Nestle [30]) and in conjunction with the current debates on designing successful interventions via NFP changes [31], a direct examination of judgments of nutrients from label information is in demand. In particular, the current study contrasts perceptions of sugar content in beverages when consumers use the current NFP vs. using the NFP with the addition of a simple visual aid.

A study by Viskaal-van Dongen, de Graaf, Siebelink, and Kok [32] exemplified the importance of visualizing nutrition content in order to properly judge it. In that study, participants consumed either a meal with visibly fatty food, (e.g., bread with butter on top), or invisible fat (e.g., bread baked with extra oil). Unbeknownst to the participants, both meals contained the exact same amount of energy, fat, carbohydrates and proteins, but participants ate 9% more calories when the fat was hidden than when it was visible. Hence, judging the hidden nutritional make up of food is not simple and can lead to overconsumption. More generally, psychological research has shown that judgments are fallible in many domains [33,34].

Following the work of Viskaal-van et al. [32] we coin the term the hidden sugar hypothesis to propose that individuals are not able to make accurate judgments of sugar content in beverages because the solid sugar, like many other nutrients, is invisible or abstract, even when numerical information is available via the NFP, without any visible cues. We hypothesize that participants underestimate the amount of sugar and the number of calories in a drink when the sugar is hidden. In contrast, we expect different and more accurate perceptions when the amount of sugar is explicitly present. Better perceptions of amount would also lead to better judgments of other related variables, such as the amount of time needed to walk to burn off the calories in the drink. We also assumed that variability in judgments would relate to consumption intentions.

One aspect pertaining to the effectiveness of interventions on food consumption is nutrition knowledge. Studies have shown that greater nutrition knowledge is associated with increased intake of fruits and vegetables and greater adherence to recommendations on fat intake [35]. These researchers developed the Nutrition Knowledge Questionnaire (NKQ) and found that a lack of nutritional knowledge impacted the relationship between diet and disease (e.g., between high fat and salt intake and cardiovascular disease) [35]. Similarly, individuals high in motivation and obesity knowledge, termed the 'nutrition elite' were found to have appropriate evaluations of nutrient claims that impacted consumption intentions [36].

We thus measured participants on several individual-level measures including the participant's health, nutrition knowledge, education, and other demographic information that could impact their judgments and choices. We predicted that higher scores on the NKQ, indicating greater nutrition knowledge, would be associated with more accurate ratings of the healthfulness of beverages, and more accurate estimates of the amount of sugar contained in the drinks and walking estimates. Additionally, we predicted that for individuals with diabetes, accuracy of evaluations would be greatest irrespective of the display manipulation. Following past research as reviewed above, we also expected relations of self-report of NFP usage with variables such as nutrition knowledge, education, and income.

#### **2. Materials and Methods**

#### *2.1. Participants*

Participants (*n* = 261) were volunteers who came to shop at four different locales in Ohio. They were predominantly female (54.8%), and not currently dieting (75.1%), with 43.7% reporting that they were employed, and 6.9% were on disability. While the sample was largely Caucasian (87.70%), it also included African Americans (3.80%), Hispanics (1.10%), and Asians (2.70%). The mean age was 45.80 (SD = 17.02), and the average BMI for this group was 28.41 (SD = 6.89), which is considered overweight (BMIs 25–29.9) by the U.S. Department of Agriculture and U.S. Department of Health and Human Services Report (2010). Seventy percent of the sample reported an annual income of less than \$49,000 per year. With relation to health, 37.5 % reported having health issues or other dietary restrictions that influenced their food choices. 52% reported eating out at least once a week or more, and over half the sample (58.5%) said they used the NFP 70% of the time.

Of the participants, 146 completed the survey in the sugar/tube and NPF condition (referred to as the Sugar group in what follows); these participants saw test tubes filled with the exact amount of sugar in each drink attached to each of the beverages, with the NFP visible as well. 115 participants completed the survey in the NFP alone condition (referred to as the No-Sugar group or control group). Participants in the No-Sugar condition only saw the beverage—no test tubes were attached but the NFP was visible. There were no significant differences between the two conditions for gender, dieting, employment, race, age, BMI, income, health issues, eating out or use of the NFP.

#### *2.2. Procedure*

For the study, the researchers set up a small folding station at five different locations in Ohio: Lottridge Ridge Food Pantry, Save-A-Lot, the Athens Farmers Market, and the Solon Community Center (Table 1). Participants were randomly assigned to the control (No-Sugar) and experimental (Sugar) groups. The stand was set on different days with randomized days to conditions so as to obtain approximately equal number of participants from each location in each experimental condition.


**Table 1.** Sample Size in Different Locations by Experimental Condition.

On the table, there were 11 popular beverages (see Table 2). In the No-Sugar condition, the beverages were presented alone; in the experimental Sugar condition, sugar bottles (test tubes) filled with the exact grams of sugar contained in the entire beverage were attached to the drink with rubber bands. Figure 1 shows one such display.



#### **Table 2.** Study Drinks with Key Information.

tsp =

teaspoons.

**Figure 1.** A drink with its sugar content displayed.

Participants were solicited for the study as they walked by the booth. They were asked if they would be willing to complete a short survey in exchange for the chance to win a \$50 Visa gift card. If they agreed, they were read the informed consent statement and signed that they understood and were willing to participate. Participants were given a survey packet and writing utensil and completed the survey as they stood by the table which contained unopened bottles of all 11 drinks. They were told that they should answer the questions to the best of their knowledge, and were encouraged to use the NFP and all other information about the drinks to help answer the questions.

Participants, after answering a series of questions for each drink, completed a demographic questionnaire, a nutrition quiz, and follow-up questions concerning their affective state and experience participating in the study. All measures are found in Supplementary Materials Figures S1–S4. After completing the survey, they were thanked for their time and debriefed.

#### *2.3. Measures*

Expert Nutrition Quality Scores: NuVal® is a Nutrition Scoring System developed by medical and nutritional experts which summarizes the overall nutrition of a food on a scale from 1 to 100 (with higher scores indicating more nutritious food) [37]. It utilizes an Overall Nutritional Quality Index (ONQI) algorithm to convert the complex nutritional information from the Nutrition Facts Panel into a single score. For this study, NuVal scores were used as the "gold standard" with which to determine how accurately participants could judge the healthfulness of a food. NuVal ratings for the beverages used in this study can be found in Table 2.

Beverage Questions: For each of the 11 beverages in the study, participants were asked to, "Please answer the following questions based on what you observed today at the nutrition and sugar display." On each page of the packet, a picture of the beverage from the display was shown, along with the drink's name, and the participant was asked to answer seven questions concerning the drink: "If you consume (drink name) how many times a week do you drink it (put 0 if you never consume it or don't like it)?"; "What proportion of this beverage is sugar (e.g., if a drink contains 1/2 a cup of sugar, and 1/2 cup of milk, the beverage would be 1/2 or 50% sugar)?"; "How HEALTHY is this beverage?" (on a scale from 0 to 100, with 100 being the healthiest); "How well does the beverage meet nutritional requirements/how NUTRITIOUS is the drink?" (on a scale from 0 to 100, with 100 "meets them extremely well"); "How many teaspoons of sugar are in this drink?"; "How many minutes of brisk walking (3.5 mph) would it take to burn off the calories from consuming this drink (assume you are drinking the ENTIRE bottle, which may contain more than one serving size)?", and "How confident are you in your answer?" (for the walking estimate). They completed all of these questions for each of the 11 drinks found in Table 2.

Demographic Questions: Participants answered a comprehensive set of questions concerning their height, weight, age, gender highest level of education and employment. Information was also gathered concerning their eating habits, medical history, and use of packaging/NFP when making purchases. Participants also rated the importance of the various nutrients in the NFP in general, and in relation to their use in this study. Finally, they were asked qualitative questions concerning their participation in the study, and factors they believed would influence their choice of healthy vs. unhealthy foods, as well as their knowledge of health guidelines.

Nutrition Knowledge Questionnaire (NKQ): The NKQ [38] was designed to provide a comprehensive measure of nutritional knowledge in adult populations. The scale consists of items concerning dietary advice, dieting and disease in five main areas: understanding of health terminology (e.g., fiber and cholesterol); awareness of dietary recommendations; knowledge of food sources related to the recommendations (e.g., which foods contain which nutrients); using dietary information to make dietary choices, and awareness of the association between diet and disease. For the present study, a modified 12-item survey was created using items from the original scale. Participants were asked to decide whether or not they believed a health statement was true or false (e.g., "Butter is higher in calories than regular margarine"). Higher scores or more correct answers reflect better nutrition knowledge.

Choices. Participants were asked which beverages they would consume right now, given the choice, and how thirsty they were at the present time.

#### *2.4. Analyses*

Data was analyzed using SPSS. A Chi-squared test, correlational statistics, and a MANOVA were run.

#### **3. Results**

#### *3.1. NFP Usage*

Self-report of NFP usage was assessed in three different questions. Individuals could check yes or no to viewing the NFP in food packages in general, when they shop for food. Results showed that 61.1% (*n* = 159 of 260 participants who provided answers) affirmed using the NFP. In contrast, participants were less likely to report using the NFP in the current study (*n* = 100 of 258, 38.76%) (It is important to note that there are slight discrepancies in the total respondents for several questions, as not all participants answered all of the questions. Therefore the *n* value will vary slightly). This relatively low rate of NFP usage in the present study is surprising given that participants were encouraged to do so. Nevertheless, individuals in the No-Sugar condition reported using the NFP to evaluate the drinks at a higher rate (49.12%, *n* = 56 out of 114 individuals with no missing values) than those in the Sugar

condition (30.5%, *n* = 44 of 144 individuals) (X2 (1) = 9.24, *p* = 0.002). This is expected because the only way to judge content accurately would be from the label, but again we note the rates are not high. This result contrasts to no significance difference in reporting general NFP usage between the two groups (X2 (1) = 0.184, *p* = 0.668).

The third assessment of NFP usage pertained to self-report frequency, or how often the participant states using the NFP when considering to purchase or consume a food item. A total of 151 participants (58.52%, *n* = 258), across both experimental conditions reported using the NFP 70% of the time or more. There were no differences in the pattern of responses to this question between the Sugar and No-Sugar individuals (X<sup>2</sup> (13) = 16, *p* = 0.249).

In terms of predicted relations of self-report of NFP usage with demographics, we found positive and significant correlations (all *p*s < 0.0001) with: nutrition knowledge (*r* = 0.325), education (*r* = 0.361), self-report of being healthy (*r* = 0.22), income (*r* = 0.25), self-report of eating healthy (*r* =0.5), eating regular meals (*r* = 0.46), being concerned with healthy eating (*r* = 0.56). Surprisingly, individuals asserting to use the NFP in general were less likely to admit using it in this particular study (*r* = −2, *p* = 0.001).

#### *3.2. Accuracy of Sugar Estimates*

The relationship (correlation) between the subjective and the objective amount of sugar in the beverages was examined in order to assess the degree to which individuals discriminated high versus low sugary drinks. This achievement measure derives from judgment analysis [39], a theory and methodology based on Brunswik's lens model [40]. Using the number of teaspoons of sugar as the unit, Pearson's correlation coefficient was computed for the judged and actual number of teaspoons of sugar across the 11 drinks for each person who had judged at least six drinks. As expected the judgment achievement of the Sugar group was higher (a median correlation equal to 0.6, which is strong and positive) than that of the No-sugar group with a median equal to 0.44. In other words, one half of the participants in the Sugar group had correlation equal to 0.6 or higher. Contrasting the mean correlation of the groups (mean *r* = 0.55 for the Sugar group, and mean *r* = 0.42 for the No-Sugar group), they are significantly different (using Fisher z transformation and unequal variance correction, t (244.36) = 3.48, *p* < 0.01). Thus, a simple display that makes the hidden sugar explicit allowed consumers to give estimates that more closely related to the actual amounts of sugar across the drinks.

In terms of raw estimates of the number of sugar teaspoons, the proportion of sugar in each drink, and the amount of walking needed to burn the calories in the drink, the Sugar group tended to produce greater overestimation (greater error) in all cases. A MANOVA using the mean absolute error, computed as a difference between subjective and objective quantities (computed for each person) revealed a main effect of condition (F (3,251) = 2.72, Wilks' Lambda = 0.97, *p* = 0.045) with larger means for the Sugar group with means equal to: 9.95 (SE = 1.89) for the teaspoon judgment; 40.04 min (SE = 3.47) for the walking judgment, and 28.58% for the proportion of sugar in the drink judgment (SE = 1.1). The corresponding means for the No-Sugar group were: 9.50 (SE = 2.15) for the teaspoon sugar judgment; 25.62 min for the walking estimate (SE = 3.95), and 26.68% for the proportion (SE = 1.26).

We note that the differences were first computed; the sign of the average differences were positive for both groups but greater for the Sugar group (raw mean differences equal to 4.3 for the Sugar group and 2.06 for the No-sugar participants; the mean of the Sugar group is significantly larger (t (252) = 1.86, *p* = 0.03) (three outliers with means 2 standard deviations above the mean were removed for this test). The proportion of participants with positive means (displaying overestimation) was greater in the Sugar than the No-Sugar group (81 of 144, 56.25%, participants in the Sugar condition; 48 of 113, 42.47%, in the No-sugar condition), X<sup>2</sup> (1) = 4.8, *p* = 0.028. In combination, using either the absolute or the raw differences, results point to greater overestimation by the Sugar than the No-Sugar group.

Comparing the judgments of healthiness of the beverages with the beverages' NuVal showed similarity of the two groups. The median correlation for the Sugar group was 0.63 and that of the No-sugar group was equal to 0.61. Comparing the mean correlations resulted in no significant mean difference (mean correlations equal to 0.58 and 0.55, for the Sugar and No-sugar groups, respectively; *p* = 0.34). Thus, the sugar visualization did not affect judgments of nutrition quality of the drinks. Focusing on judgments errors with regards to judging healthiness, the groups did not differ either. However, absolute judgment errors in this variable tended to be smaller for individuals with higher nutrition knowledge (*r* = −124, *p* = 0.023) and education (*r* = −27, *p* = 0.00). Furthermore, individuals reporting higher nutrition knowledge also reported having better health (*r* = 0.134, *p* = 0.015) (all *p*s one-tail tests).

#### *3.3. Participants with Diabetes*

Focusing on participants who reported having diabetes (*n* = 43, 16.4%), this group had higher BMI (28.71) and lower income (median \$20k and below annually) when compared to the rest of the participants (BMI = 26.24, median income \$20–\$29k annually). They were also older (median age 54; median age 46 for others). Results showed greater overestimation of sugar content by these individuals. The mean absolute error overestimating teaspoons of sugar was equal to 15.42 (with raw mean difference equal to 11.01). Without including the three outliers (one who was in the diabetics group), the mean of the absolute error describing overestimation for the diabetic group (mean = 9.9) is significantly larger than that of the rest of the participants (mean = 6.7), t (74.37) = 1.68, *p* = 0.049.

#### *3.4. Person Level Factors that Relate to Judgment Accuracy*

Regression analysis was employed to predict the accuracy measures from person-level characteristics. In particular, we hypothesized that nutrition knowledge and concern for healthy eating would result in greater accuracy. Because of the special health concern of diabetics, we also expected greater accuracy for this sub-group.

With regards to the correlation between the judged vs. objective total number of teaspoons in the drinks results showed that indeed the availability of sugar affected discrimination accuracy in the expected direction (β = 0.203, *p* < 0.01), but additionally individuals with higher levels of education and higher BMI had greater accuracy (β = 0.28, *p* < 0.01, for Education; β = 0.16, *p* < 0.01, for BMI; F (3, 242) = 12.62, *p* < 0.0001, adj R<sup>2</sup> = 0.12). Surprisingly, higher nutrition knowledge, or higher concern for healthy eating did not predict this accuracy criterion. Of great interest is that individuals reporting having diabetes had no greater accuracy in judging relative sugar content than did individuals not having such a health issue. No other individual level variables were significant predictors of the relationship between subjective and objective amounts of sugar.

In terms of the average difference between judged and objective amounts of sugar (both in terms of proportion and of number of teaspoons), we found that nutrition knowledge was not predictive of these variables. In terms of the accuracy of judging amount of walking to be done to burn the calories, a model with condition, nutrition knowledge, diabetes and income as predictors resulted in, F (4, 240) = 2.72, *p* = 0.03, adj R<sup>2</sup> = 0.027, but with significant beta weights for only the condition experimental manipulation (β = 0.17) with greater overestimation for those viewing the sugar display (i.e., the Sugar group).

Individuals with diabetes had greater overestimation of the amount of sugar present in the drinks as earlier stated. Additionally, the group of diabetics gave greater importance to sugar when judging the overall nutrition of foods (means = 51.63 and 42.83 for diabetics and controls, respectively). A MANOVA with both measures as dependent variables and group (diabetes vs. control) as independent variable revealed a significant group effect (F (2, 237) = 3.77, *p* = 0.024, Wilks' Lambda = 0.97) (this analysis does not include the three outliers who produced very large estimates; results do not change when included).

#### *3.5. Choice of Drink as a Function of the Visual Aid*

The great majority of participants stated not wanting to consume any of the sugary drinks being judged at the moment (88.4% response rate towards not wanting to consume across participants and across the 10 drinks containing sugar). The condition manipulation, nevertheless, lowered the intentions of consuming any of the drinks; the mean number of drinks individuals felt like consuming was equal to 1.165 drinks for the No-Sugar group and equal to 0.89 for the Sugar group and this difference was statistically significant, t (255) = 2.077, *p* = 0.02 (one-tail). Another way to look at this is that in the No-Sugar group, across all participants and drinks, the average selection of sugary drinks was 14.9%, and this proportion was equal to only 8.8% in the Sugar group—a 40.93% decrease. These proportions are significantly different by z-test (*z* = 4.8, *p* < 0.0001).

We must note that on average people stated not being very thirsty with means equal to 39.33 and 36.89 for the Sugar and No-sugar groups, respectively, using the 0–100 scale with 100 denoting maximum thirst (these means are not statistically significantly different). Additionally, the open-ended question about drinks showed that the choices available in the study were not common drink options for participants. Orange juice was the most selected drink and this was stated by only 22 participants in the entire sample (8.4%). The next most popular drink was coke, but with only 5 selections. Besides these, water, coffee, and milk were more commonly listed as beverages consumers drink. Thus, the effect of the manipulation is likely to be stronger than observed if the individuals were thirstier and the sugary drinks were their habitual choices.

The results of this study also suggested that self-report of NFP usage both in general and for this study, as well as knowledge of nutrition and concern for healthy eating, did not play a significant role in predicting the total number of sugary drinks that participants reported they would hypothetically consume. However, it is important to note that these factors could be linked to habitual choices which would remain regardless of the intervention, and may be difficult to change. Other variables predictive of the choices, once the effect of the experimental manipulation was accounted for, were degree to which the person eats healthy (β = −18, *p* = 0.009), and education (β = −18, *p* = 0.006), F (5, 247) = 4.33, *p* < 0.0001, adj. R<sup>2</sup> = 0.138.

Finally, we also found that in terms of estimating sugar in drinks, individuals were generally off, overestimating sugar by three spoons or more. This was true even for individuals with higher levels of education, nutrition knowledge, and concern for healthy eating. The simplest explanation is that, in general, people have no concept of the correlation between grams (the measure on the NFP) and teaspoons. Grams, for US participants, is also a more abstract concept. This would suggest that the NFP is of little benefit whether or not it is used, in helping to determine overall quantities of sugar. Strong positive correlations among self-report of NFP usage in daily life with concerns for eating healthy, education, income, nutrition knowledge suggests a "wealthier get wealthier" scenario, in that those who are aware of the value of, and are concerned with, health knowledge, are better prepared to make nutritional judgments than those who are not. This will be further elaborated on in the discussion.

#### **4. Discussion**

The evidence from medical and health care research is mounting to support the link between sugar consumption and cardiovascular disease and mortality [41]. The politics behind the high availability of sugary drinks and food products containing added sugars is complex [42]. In the center of these realities lies the psychological machinery that reacts positively to sugar and does not perceive the world in a purely objective way. It is the judgment and decision-making processes that ultimately determine the degree to which consumers are able to judge information effectively and use it to make smart food selections.

Our work focuses on understanding the psychological judgment processes with the hope that interventions, other than those based on pricing and/or availability of products, can be developed to support effective decision making. How well can individuals judge how much of a nutrient is present in a food product? What factors contribute to accurate perceptions and cognitions? Answers to these questions, we believe, are essential in determining support systems that result in calibrated perceptions and more optimal food choices.

The study was conducted mostly in rural Appalachia, but it also had individuals from a community center in Cleveland which allows our results to generalize to a range of income and education. We tested a simple intervention designed to make sugar explicit when considering amounts of sugar in a set of popular drinks. Two important findings from this study are: 1) estimation of nutrient content was difficult even when sugar amounts were made obvious via the test-tube sugar displays for each drink, and 2) both judgments and choices were influenced by the intervention. With a few exceptions, other person level characteristics, such as nutrition knowledge and concern for healthy eating, did not influence judgment accuracy.

In terms of estimating sugar content, such as number of teaspoons of sugar, individuals were better able to discriminate among drinks when sugar was made explicit. Additionally, higher levels of education and higher BMI related to higher accuracy, but contrary to our expectation, no relationship was found with nutrition knowledge and concern for healthy eating. Exact estimates, on the other hand, were no more accurate, but tended to move in the direction of overestimation. The overestimation also occurred with regards to amount of time walking needed to burn the calories.

From the perspective of the helpfulness of the NFP information we note that sugar amounts, as described in the label, did not translate into common units such as teaspoons, and individuals were generally off, overestimating sugar by three spoons or more. This was true even for individuals with higher levels of education, nutrition knowledge, and concern for healthy eating. Of great consequence is the fact that we found strong positive correlations among self-report of NFP usage in daily life with concerns for eating healthy, education, income, nutrition knowledge. However, NFP frequency related negatively to using the NFP in the current study and it did not predict judgment accuracy of any type, nor did it predict choice. Thus, our results cast doubts on the meaning and validity of high levels of NFP usage derived from self-report.

On the positive side, whatever information was used from the label, or from past experience, the judgments about the overall nutritional quality of the drinks produced relatively high discriminations as measured by the correlation between NuVal (the objective nutrition scores) and the subjective impressions. Accuracy with respect to NuVal also depended on nutrition knowledge and education.

In terms of drink selection, we found a low rate of preference for the options the study provided, yet the visual aid manipulation influenced choice. Using the total number of sugary-drinks a participant may drink as a measure of consumption intention, we found that the visual displayed produced lower rates of consumption. Beyond this manipulation effect, self-report of eating healthy and education were the only other predictors of choices. Interestingly, the status of being diabetic, having concern for healthy eating, or identifying sugar as an important nutrient did not predict choice.

Focusing on the group of 43 diabetics across locations, we noted that they reported giving great importance to sugar when judging nutrition as would be expected. In addition, their estimates of sugar content were greater than the rest of the participants by an average of approximately three teaspoons. But the group did not differ in terms of drink selections, as previously mentioned, which highlights the possible disconnect between beliefs and actions.

Finally, another interesting finding was that individuals who reported that they regularly used the NFP were less likely to admit to using it in the present study. As this did not vary as a function of condition (e.g., they were not more or less likely to use the NFP if the sugar tubes were present), other factors could be at work. More research is needed to determine if this was simply a function of being part of a research study, or their normal habitual behavior.

It is also possible that these individuals in general tended to adhere to the social desirability bias, and thus wanted to report that using the NFP was a regular habit, as it was the center of the study and known to be beneficial. They may also want to use it, but in general tend to forget, or get distracted when they do.

#### **5. Conclusions**

In the sugar debate, the psychology of sugar needs greater attention with emphasis on the perceptual and cognitive processes that determine judgments and choices. The human perception system is not a purely bottom up information processor reflecting objective quantities, and judgments are influenced as much from expectations and suggestions as they are from the sensory processes from which those judgments come from [43]. Greater attention to these psychological underpinnings is in demand in order to progress towards creating environments that support effective choices. Such environments may need to go beyond placing limits on food availability via pricing, or the lowering of supply, which present implementation challenges. Our findings demonstrate that nutrient visualization can support judgments and decisions and thus may be a viable tool for curtailing consumption of undesirable nutrients. Perhaps, labels that more obviously convey information, such as providing the exact number of teaspoons of sugar in the product, are in demand.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2072-6643/12/2/394/s1, Figure S1: Questions About Drinks; Figure S2: demographic questionnaire; Figure S3: Nutrition Questions; Figure S4: Overall Questions

**Author Contributions:** Conceptualization, B.D.M. and C.G.-V.; methodology, B.D.M. and C.G.-V.; validation, B.D.M. and C.G.-V.; formal analysis, B.D.M. and C.G.-V.; investigation, B.D.M. and C.G.-V.; resources, B.D.M. and C.G.-V.; data curation, B.D.M. and C.G.-V.; writing—original draft preparation, B.D.M. and C.G.-V.; writing—review and editing, B.D.M. and C.G.-V.; visualization, B.D.M. and C.G.-V.; supervision, C.G.-V.; project administration, B.D.M. and C.G.-V.; funding acquisition, B.D.M. and C.G.-V. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the Diabetes Institute at Ohio University via the Student Research Award.

**Acknowledgments:** The authors would like to thank the Diabetes Institute at Ohio University for support to this research via The Student Research Award, and NuVal® for providing the nutritional ratings for the food items used in this study.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

*Article*

### **The Color Nutrition Information Paradox: E**ff**ects of Suggested Sugar Content on Food Cue Reactivity in Healthy Young Women**

#### **Jonas Pottho**ff **\* , Annalisa La Face and Anne Schienle**

Institute of Psychology, University of Graz, Universitaetsplatz 2, 8010 Graz, Austria;

annalisa.la-face@uni-graz.at (A.L.F.); anne.schienle@uni-graz.at (A.S.)

**\*** Correspondence: jonas.potthoff@uni-graz.at; Tel.: +43-316-380-3883

Received: 12 December 2019; Accepted: 21 January 2020; Published: 24 January 2020

**Abstract:** Color nutrition information (CNI) based on a traffic light system conveys information about food quality with a glance. The color red typically indicates detrimental food characteristics (e.g., very high sugar content) and aims at inhibiting food shopping and consumption. Red may, however, also elicit cross-modal associations with sweet taste, which is a preferable food characteristic. We conducted two experiments. An eye-tracking study investigated whether CNI has an effect on cue reactivity (dwell time, saccadic latency, wanting/liking) for sweet foods. The participants were presented with images depicting sweets (e.g., cake). Each image was preceded by a colored circle that informed about the sugar content of the food (red = high, green = low, gray = unknown). It was tested whether the red circle would help the participants to direct their gaze away from the 'high sugar' item. A second experiment investigated whether colored prime circles (red, green, gray) without nutrition information would influence the assumed sweetness of a food. In Experiment 1, CNI had the opposite of the intended effect. Dwell time and saccadic latency were higher for food items preceded by a red compared to a green circle. This unintended response was positively associated with participants' liking of sweet foods. CNI did not change the wanting/liking of the displayed foods. In Experiment 2, we found no evidence for color priming on the assumed sweetness of food. Our results question whether CNI is helpful to influence initial cue reactivity toward sweet foods.

**Keywords:** nutrition facts; food cue reactivity; sugar; eye tracking; priming; color

#### **1. Introduction**

Food is a primary reinforcer that automatically captures visual attention. This evolutionary-based mechanism assists with the localization of food sources within the environment and, in turn, enables sufficient caloric uptake by the individual [1]. Studies utilizing neurophysiological measures and eye-tracking have shown that the human attention system very quickly identifies visual food cues and differentiates them from non-food objects [2–4]. Additionally, high-calorie food captures more attention than low-calorie food [5,6].

The increased attention to cues of high-calorie food has become problematic in Western countries because the exposure to such stimuli triggers the urge to eat [7]. Food cues and (high-calorie) foods are almost omnipresent in our everyday lives. Therefore, a link between individual food cue reactivity (FCR), overeating, and weight gain is not surprising [7].

In order to reduce the shopping and consumption of high-calorie food, effective interventions that are able to reduce FCR are urgently needed. It has already been demonstrated that nutritional knowledge is able to influence FCR [8]. A number of studies has found a positive correlation between nutritional knowledge and healthy dietary habits [9–14]. The knowledge transfer about the sugar content of food seems to be a promising starting point for such interventions because large proportions of calories are consumed in the form of sugar [15]. Moreover, the excessive consumption of sugary food is associated with an increased risk of cardiovascular disease, cancer, and diabetes [16]. However, findings regarding the relationship between individual knowledge about the sugar content of specific foods and actual consumption are heterogeneous [17–19]. Therefore, it seems likely that knowledge about the sugar content of food cannot always be accessed easily and quickly enough [20–23].

Therefore, color nutrition information (CNI) based on a traffic light system seems to be an efficient method to convey information about food quality. This system is already used in front of pack food labels [24]. The color red (as a stop signal) typically indicates detrimental food characteristics (e.g., very high sugar content), whereas green signals positive features [25–28].

However, even though the traffic light system is widely used, little is known about how CNI influences initial food cue reactivity. Furthermore, little is known about possible unintended effects of the commonly used colors (red, green). The color red may elicit cross-modal associations with sweet taste, which is a preferable food characteristic [29,30]. For example, cider was perceived as sweeter when served in a bottle with a red label compared to a green label [31]. The red-sweetness association seems to be stronger for drinks compared to solid foods. Lemos et al. [32] presented colored prime stimuli (red, green, amber cycles) that were followed by an image of a salty or sweet food item. The seven sweet food items used in this experiment were, on average, rated as more positive (hedonic valence) after the presentation of a red cycle compared to a green cycle. This effect was most pronounced for the only liquid (a carbonated soft drink) used as stimulus material. However, for half of the solid sweet foods, the hedonic valence was actually lower after the presentation of a red cycle compared to a green one. Based on this previous research, it remains unclear whether red color used in food labels as 'warning signals' implicitly primes sweet taste associations.

The aim of the present investigation was twofold. We investigated effects of colored nutrition information (traffic light symbols indicating the sugar content) on initial food cue reactivity (Experiment 1). In a second experiment, we investigated priming effects of the colors red and green on assumed sugar content/sweet taste (Experiment 2).

#### **2. Materials and Methods**

#### *2.1. Sample*

Experiments 1 and 2 were conducted following the rules of the Declaration of Helsinki of 1975, revised in 2013. The experiments were approved by the ethics committee of the University of Graz (ethical approval code: 39/31/63 ex 2018/19).

#### 2.1.1. Sample Experiment 1

Fifty-one women (mean age: 22.0 years, SD = 2.99; range 18–33) with a body mass index (BMI) of *M* = 22.5 (SD = 3.85) took part in this study. We selected women because previous research has suggested that the use and understanding of nutrition information is related to demographic characteristics, notably social grade, age, and gender [33]. Participants had normal or corrected-to-normal vision and did not report any current medication or mental disorder. Forty-nine participants were university students, and the other were white-collar workers. Participants were recruited via email lists and postings at the university campus as well as dormitories. Psychology students (*N* = 32) received course credits for their participation. Sample characteristics are displayed in Table 1.


**Table 1.** Sample characteristics and rating data.

#### 2.1.2. Sample Experiment 2

A total of 99 participants (age: *M* = 25.03 years, SD = 6.17 years; BMI: *M* = 22.61 kg/m2, SD = 2.81 kg/m2) completed an online experiment. Of the participants, 55 had a high school diploma, 44 participants graduated from college. The majority of participants was female (female: *N* = 74, male: *N* = 25).

#### *2.2. Stimuli and Design Experiment 1*

We presented color nutrition information (CNI) that reflected the sugar content of a specific food item (green: Low sugar content, red: High sugar content, gray: Unknown sugar content; diameter: 354 pixels) and 48 pictures of sweet foods (e.g., cakes, ice cream, candies from the FoodPics database [34]). Each picture had a size of 600 × 450 pixels. Food images and CNI were presented on a white background on an LCD screen. We selected food products of which low sugar versions are commonly available on the market. We assigned 16 images to each category (low/high/unknown sugar content) and created three parallel versions of the experiment. Due to the parallel versions, each image was suggested to have a low, high, or unknown sugar-content for one-third of the participants. Participants were randomly assigned to one of the three parallel versions.

At the beginning of each trial, a circle was presented on either the center of the left or the right half of the screen. As soon as participants were gazing at it steadily for 1000 ms, the CNI disappeared and the allocated food image was presented for 1500 ms. Each pair of CNI and food image was shown in two trials: Once the food image appeared in the same location as the CNI (current gaze location: Figure 1), the other time the food image was presented on the opposite side of the screen (peripheral location), resulting in 96 trials (16 per suggested sugar content: Low, high, unknown; and position: Current gaze location, peripheral). Trials were followed by an intertrial interval of 200 ms. The trial order was randomized.

The participants were instructed to inspect the circles. Throughout the paradigm, two food items of each category were presented in the center of the screen. Participants were asked to rate these food items regarding their specific appetite ("How much would you like to taste this food right now?" 0: "Not at all", 6: "Very much") and general liking ("How much do you like this food in general?" 0: "Not at all", 6: "Very much").

**Figure 1.** Example trials for (from left to right) low sugar label, high sugar label, and unknown sugar label. Each food image was presented twice: Once in the same location as the label (example: low sugar & unknown sugar) and once in the peripheral location (example: high sugar label). In 50% of trials, the label was presented on the left side of the screen. In the other 50%, the label was presented on the right side of the screen (not displayed here).

#### *2.3. Procedure Experiment 1*

After providing written informed consent, participants read a short info sheet about color-coded nutrition facts (high sugar/red symbol: Above 12.5 g sugar per 100 g food, low sugar/green symbol: Below 5 g sugar per 100 g food). Subsequently, participants rated their general appetite and hunger on a seven-point scale (appetite: 0: "I have no appetite at all.", 6: "I have an extreme urge to eat something right now."; hunger: 0: "I have no hunger at all.", 6: "I am extremely hungry."). Furthermore, participants rated their preference for sweet food ("How much do you like sweet food in general?", 0: "Not at all.", 4: "Very much."). Subsequently, the eye-tracking paradigm described above was conducted.

Following the eye-tracking paradigm, participants conducted a survey about their demographics and the following questionnaires.

#### *2.4. Questionnaires Experiment 1*

The participants completed the Eating Disorder Examination-Questionnaire (EDE-Q; [35]) and the Impulsivity Short Scale (I-8; [36]). The EDE-Q consists of 41 items (e.g., "Were you afraid to lose control over your eating?") that are answered on seven-point scales (0: "Not at all", 6: "Very much") and are concerned with the previous four weeks. Furthermore, the EDE-Q inquires weight and size (i.e., BMI). In the present sample, Cronbach's alpha for the EDE-Q was α = 92. The I-8 consists of eight items (e.g., "I usually think carefully before I act."), which are answered on five-point scales (1: "Doesn't apply at all", 5: "Applies completely"; Cronbach's α = 75 for the I-8).

The questionnaires were selected because disordered eating and impulsivity have been associated with elevated food cue reactivity in previous research [37].

#### *2.5. Eye Movement Recording and Analysis Experiment 1*

Two-dimensional eye movements were recorded using an SMI RED250mobile eye-tracker with a sampling rate of 250 Hz. Head movements were minimized by a chin rest. We calibrated both eyes and analyzed data from the eye that produced the better spatial resolution (typically more accurate than a 0.35◦ visual angle). Stimuli were presented on a white background on a 24-inch screen with a resolution of 1920 × 1080 pixels. The viewing distance was 60 cm, resulting in a size of 15.6◦ × 11.7◦ viewing angle for food images and a diameter of 9.2◦ viewing angle for CNIs. The experiment was controlled using the SMI Experiment Center (Version 3.6.53, SensoMotoric Instruments, Teltow,

Germany). For event detection, standard thresholds of the SMI BeGaze Software (Version 3.6.52, SensoMotoric Instruments, Teltow, Germany) for high speed eye-tracking data (recommended for sampling rate > 200 Hz) were used to detect saccades (velocity threshold: 40◦/s). Data were exported using SMI BeGaze and customized Python scripts. Within BeGaze, we defined the food images as areas of interest (AOI). We conducted gaze data analysis exclusively for the food AOI of each trial. We defined the dependent variable, 'saccadic latency', as the time from stimulus onset to the start of the first saccade that ended outside of the food AOI. Saccadic latency was calculated only for trials with the participants' gaze position within the AOI at stimulus onset (CNI was presented in the same position as the subsequently presented food). Saccadic latency therefore measured how long it took participants to actively relocate their gaze away from a food item.

The second dependent variable, 'dwell time', was defined as the sum of fixation durations within the AOI. Other than saccadic latency, we computed dwell time for all trials (trials in which the food appeared at gaze location, as well as trials in which food appeared in the peripheral location).

#### *2.6. Stimuli and Design Experiment 2*

Thirty pictures of sweet food from Experiment 1 (size: 600 × 450 pixels) were presented in the center of the computer screen for 1500 ms each. Prior to the picture presentation, one of three colored circles (red, green, gray) was shown. The circles (diameter: 354 pixels) were displayed centrally on a white background for 1000 ms. The circles did not contain any text and were presented without any further instructions. We created three subsets of prime-stimulus combinations to ensure that each picture was preceded by a red, green, or gray circle. The participants were randomly assigned to one of the three color-food combinations (combination 1: *N* = 28, combination 2: *N* = 38, combination 3: *N* = 33). There was no significant difference between groups in mean age (*F*(2,96) = 0.17, *p* = 0.84, η2*p* = 0.004), BMI (*F*(2,67) = 1.64, *p* = 0.20, η2*p* = 0.047), hunger level (*F*(2,96) = 0.83, *p* = 0.44, η2*p* = 0.02), or gender distribution (Chi<sup>2</sup> (2, *N* = 99) = 1.52, *p* = 0.47).

After the presentation of each food image, the participants rated the assumed sweetness of the food on a scale from 0% ("not sweet at all") to 100% ("extremely sweet"). Additionally, the valence of two food images per color was rated (0%: "Extremely unpleasant", 100%: "Extremely pleasant"). The trials were presented in random order.

#### *2.7. Procedure Experiment 2*

Participants were asked to conduct the experiment at home without distraction on a computer with a (hardware) keyboard and mouse. After giving informed consent, participants provided demographic data (age, education, gender). They reported their current hunger level ("How hungry are you right now?" 0: "Not hungry at all", 6: "Extremely hungry"), weight, and height. Subsequently, the participants were presented with 30 images of sweet food in randomized order. The experiment was conducted using Pavlovia and was programmed in Python using PsychoPy 3.2.2 [38].

#### *2.8. Statistical Analysis*

Repeated measures analyses of variance (ANOVAs) were computed to test the effect of CNI (low, high, unknown sugar content) on specific appetite, general liking of the displayed food items, and dwell time spent on food images, as well as saccadic latency away from food. For trials in which the food image was presented in the periphery of the current gaze, the repeated measures ANOVA was conducted only for dwell time (Experiment 1). In Experiment 2, ANOVAs were conducted to test the effect of color. If sphericity was violated (Mauchly's Test of Sphericity), Greenhouse–Geisser correction was applied. We reported the effect size as η2*p* (partial eta squared) and Holm adjusted p-values. The *p*-values smaller than 0.05 were considered statistically significant. Data are available online at OSF (OSF Project DOI: 10.17605/OSF.IO/FJ3UZ, Center for Open Science, Charlottesville, VA): www.osf.io/g4d7s/

#### **3. Results**

#### *3.1. Results Experiment 1*

#### 3.1.1. Questionnaire Data

Participants obtained an average EDE-Q score of *M* = 1.33 (SD = 0.96), which did not differ significantly from the mean (*M* = 1.44) of the healthy norm sample (individuals without any current diagnosis of an eating disorder, *N* = 409, [35]), *t*(50) = 0.80, *p* = 0.43, *d* = 0.11. The mean I-8 score of the present sample of *M* = 2.63 (SD = 0.62) did not differ significantly from the average impulsivity of the female norm sample aged between 18 and 35 years (*M* = 2.62), *t*(50) = 0.09, *p* = 0.93, *d* = 0.01.

#### 3.1.2. Appetite and General Liking of Presented Food Images

CNI had no statistically significant effect on reported appetite (*F*(2,100) = 0.38, *p* = 0.68, η2*p* = 0.008) and general liking of the depicted food items (*F*(2,100) = 0.58, *p* = 0.56, η2*p* = 0.01; see Table 1).

#### 3.1.3. Eye Movements

Saccadic Latency*:* For gaze relocation (same position), the repeated measures ANOVA revealed a significant main effect of CNI on saccadic latency (*F*(1.71,85.55) = 4.98, *p* = 0.012, η2*p* = 0.091). The saccadic latency was significantly lower for food with a low sugar content compared to food with a high sugar content (*t*(50) = 2.35, *p* = 0.045, *d* = 0.33) and unknown sugar content (*t*(50) = 2.63, *p* = 0.034, *d* = 0.37). The saccadic latency did not differ between unknown and high sugar content (*t*(50) = 0.24, *p* = 0.82, *d* = 0.03; see Figure 2).

Dwell Time: The ANOVA revealed a significant main effect of CNI on the dwell time spent on the food images (*F*(1.61,80.48) = 5.61, *p* = 0.009, η2*p* = 0.10). The dwell time was shorter for food with a low sugar content compared to food with a high sugar content (*t*(50) = 2.37, *p* = 0.043, *d* = 0.33) and unknown sugar content (*t*(50) = 2.83, *p* = 0.020, *d* = 0.40). Dwell time did not differ between unknown and high sugar content (*t*(50) = 0.44, *p* = 0.66, *d* = 0.06; see Figure 2 and Table 2).



In Experiment 1, green indicated low sugar, red indicated high sugar, and gray indicated no specific sugar content. For gaze data (Experiment 1), mean durations in milliseconds are given. Sweetness and valence were rated from 0% to 100%. Asterisks indicate significant main effects.

For gaze avoidance (peripheral position), there was no significant effect of CNI on dwell time (*F*(2,100) = 1.70, *p* = 0.19, η2*p* = 0.03). The dwell time did not differ significantly between low sugar content (*M* = 71.5, SD = 61.1), high sugar content (*M* = 58.4, SD = 52.1), and unknown sugar content (*M* = 65.0, SD = 49.1).

**Figure 2.** Mean saccadic latency and dwell time for trials in which food appeared in the current gaze location for three CNI conditions: Low (green CNI/low sugar), high (red CNI/high sugar), and unknown (gray CNI/unknown sugar). Whiskers indicate standard errors. Asterisks indicate Holm-adjusted *p* < 0.05.

#### 3.1.4. Exploratory Analysis

To analyze if the general preference for sweet foods was correlated with CNI, we calculated Pearson correlations between liking of sweet foods and (1) the difference in saccadic latency between high and low sugar content (high sugar saccadic latency minus low sugar saccadic latency) and with (2) the difference in dwell time (high sugar dwell time minus low sugar dwell time). On average, the reported liking was *M* = 2.86 (SD = 1.02). We found positive correlations between liking and difference in saccadic latency (*r* = 0.298, *p* = 0.034) and dwell time (*r* = 0.369, *p* = 0.008).

The difference in saccadic latency was not correlated with the I-8 score (*r* = 0.022, *p* = 0.879), the EDE-Q score (*r* = 0.217, *p* = 0.126), the BMI (*r* = 0.01, *p* = 0.95), hunger (*r* = 0.169, *p* = 0.236), or appetite (*r* = 0.055, *p* = 0.700). Also, the difference in dwell time was not correlated with the I-8 score (*r* = −0.081, *p* = 0.570), the EDE-Q (*r* = 0.247, *p* = 0.081), the BMI (*r* = 0.08, *p* = 0.58), hunger (*r* = 0.103, *p* = 0.474), or appetite (*r* = −0.003, *p* = 0.982).

#### *3.2. Results Experiment 2*

We calculated two ANOVAs to test the effects of color (red, green, gray circles) on estimated sweetness and valence of the food stimuli. We found no significant color effect for sweetness *F*(2,196) = 0.22, *p* = 0.81, η2*p* = 0.002 (Table 2). The effect for valence was significant, *F*(2,196) = 3.16, *p* = 0.045, η*2p* = 0.031 (Table 2). The post-hoc pairwise comparisons were not significant (all *p* > 0.08). Food items preceded by a gray circle received marginally higher valence ratings compared to red circles.

#### **4. Discussion**

The shopping of food, including high-calorie sweet snack foods, is often impulsive. In order to influence this spontaneous shopping behavior, simple interventions are needed that are able to interrupt this process. The current eye-tracking study investigated the influence of provided information about a product's sugar content on visual food cue reactivity. It was tested whether a red circle that indicated a high sugar content of a product would be able to help the participants to direct their gaze away from the displayed food item. The results showed that the intervention had the opposite of the intended effect. The dwell time and the saccadic latency were lower for food items preceded by a green circle compared to a red and gray circle. Obviously, it was easier for the participants to ignore food cues if low sugar content was assumed relative to high or unknown sugar content. Thus, the participants showed a paradox reaction.

Similar paradox effects have been reported in studies that attempted to influence knowledge and beliefs about food [39,40]. A study by Berry et al. [39] examined how calorie information on menus in chain restaurants affected the food choice. The results indicated that calorie labeling even increased the calories ordered if the consumers were taste-oriented rather than health-oriented. Similarly, Provencher et al. [40] found that participants ate 30% more of the same cookies when labeled as healthy.

Whereas the current study contributes to the existing evidence that nutrition facts may be ineffective [39–42], other findings have indicated that nutrition fact information provided via food labels is a useful tool to target food cue reactivity and food choices [28,43,44]. Further research is needed to evaluate in which cases unintended effects of CNI on food cue reactivity might occur. Our exploratory analysis indicated that it was more difficult for participants with a high compared to a low preference for sweet foods to avoid 'high sugar' foods. Thus, individual preferences might overrule CNI [41].

Additionally, previous research has indicated that the color-coding itself may elicit unintended effects on FCR. Cross-modal associations between the color red and sweet taste have been reported in many studies. Cross-modal associations were observed primarily for fluids [30] and not for solid foods [45]. The present study (with exclusively solid foods) found no evidence for priming effects of red on estimated sweetness and pleasantness of the depicted food products. The food items even received marginally higher valence ratings after the presentation of a gray circle compared to a red circle. Thus, it is unlikely that the results of Experiment 1 were caused by cross-modal associations between priming color and visual food perception.

We need to mention the following limitations of the present study. In Experiment 1, we only studied female participants. The majority of the women were university students. Therefore, our findings cannot be generalized to other samples. However, it is important to note that we used an innovative gaze performance task to evaluate visual food cue reactivity without the possible effects of self-monitored gaze direction or social desirability (as opposed to free exploration paradigms and self-reports). The task was very easy and therefore should have been accomplished by this group of highly educated women. Nevertheless, to determine if the basic findings of the present study can be applied to other participants and circumstances (e.g., male and/or less-educated participants), a replication study is highly recommended. Experiment 2 was not conducted in the lab, but at home. Thus, we were not able to control unintended distractions during participation.

**Author Contributions:** Conceptualization, J.P., A.L.F. and A.S.; methodology, J.P., A.L.F. and A.S..; software, J.P.; validation, J.P. and A.S.; formal analysis, J.P. and A.S.; investigation, A.L.F. and J.P.; writing—original draft preparation, J.P., A.L.F. and A.S.; writing—review and editing, J.P. and A.S.; visualization, J.P.; supervision, A.S.; project administration, A.S. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Acknowledgments:** Open Access Funding by the University of Graz. The authors acknowledge the financial support by the University of Graz.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

*Article*
