Archives of Design Research

Visualization of complex data allows for an intuitive understanding of the data, as diagrams elucidate the conceptual structures of information (Oxman & Planning, 1997). So do FBDG designs from countries; they reflect the agricultural production system, culinary tradition, and lifestyle-induced health issues specific to each country. A global review of food-based dietary guidelines (Appendix A) sheds light on the differences in format and contents.

Format: 67% of the 34 European FBDGs are in the pyramid format. The pie chart is the second popular choice, while the column chart and other unique designs are occasionally used (Montagnese et al., 2015). Perelman (2011) reported a controversy caused by the widely used pyramid format in the U.S.; the top section food can be either the highest in nutrition or the smallest to be consumed. Recently developed German 3D Food Pyramid (Oberritter et al., 2013) clarified it with red-green spectrums that indicate less healthy items were placed higher. Japanese cleverly dodged this issue with an inversely conical pyramid where the smallest space was placed at the bottom (“Japanese Food Guide Spinning Top,” n.d.) The pyramid format, however, suggests fixed proportionality—when all tier heights are equal, the top three tier area ratio is fixed to 1:3:5—thus designing FBDGs in the pyramid format requires more caution. Pie charts were reported to be preferred to the pyramid format in other studies. While the British FBDG was developed, Hunt, Gatenby, and Rayner (1995) tested ten combinations of design variables: formats (pyramid, flat pie chart, or tilted pie chart), color schemes (multi-color or single-color), and presentation styles (color photos or line illustrations). The study concluded that while the most preferred design was the tilted pie chart with photo icons, design was not significantly correlated to communication performance, as much as the study participants' gender, age, or socio-economic status did.

Moderation/proportionality: The U.K. Eatwell Guide (Public Health England, 2018) addresses the gender differences in recommended daily calories, and Japanese Food Guide Spinning Top (Japanese Ministry of Health, Labour and Welfare, 2010) spells out the food group proportionality, recommended number of servings, and serving sizes.

Diversity: FBDGs present food groups and food items in the groups on different abstraction levels. American MyPlate (U.S. Department of Health and Human Services and U.S Department of Agriculture, 2015) and Hungarian House of Healthy Nutrition (Rodler, 2004) only present food group labels, while German 3D Food Pyramid (Oberritter et al., 2013), Australian Guide to Healthy Eating (2013), and the U.K. Eatwell Guide (Public Health England, 2018) show tangible examples in each group. The Finnish Plate Model (Fogelholm et al., 2014) more specifically shows a typical public lunch catering meal of potatoes, fish, and vegetables served on a plate to contextualize the information better. The food examples are presented in either illustrations or photos. Hunt et al. (1995) tested two presentation styles (color photos or line illustrations) and found photo icons presented in the tilted pie chart were preferred the most. Talati et al. (2017) also confirmed Australian consumers' preference for the circular pie chart and photo icons to other designs. Subjects in various age groups preferred realistic photo images of food to cartoon-style drawings, because "you can actually tell what they are" (p. 173).

Hydration: The German 3D Food Pyramid (Oberritter et al., 2013), Japanese Food Guide Spinning Top (Japanese Ministry of Health, Labour and Welfare, 2010), the U.K. Eatwell Guide (Public Health England, 2018), Australian Guide to Healthy Eating (2013), and Finnish Plate Model (Fogelholm et al., 2014) mention hydration as an important part of healthy eating.

The review of FBDGs inspired the authors to explore various frequency visualization formats other than the pyramid or pie chart, as well as photo icons in place of illustrations for accurate descriptions. A detailed review of communication problems in the current Food Balance Wheel design, including the misleading proportionality visualization of the rear wheel, continues in the next section.

Functionally, the Korean Food Balance Wheel (Fig. 1b) is two pie charts that illustrate the concepts of diversity, proportionality, and hydration. In Hong (2013)’s study with elementary school students, and the authors’ unpublished qualitative study with college students, the current design failed to clarify the following information.

(a) Personalization-moderation: the personalized recommendation of daily calorie intake. Being a static image, the Food Balance Wheel fails to individuate 37 dietary patterns (from 1000 kcal to 2700 kcal). Servings in each food group are written in range values (Fig. 1b, “Eat 2-4 times a day from the Grains group”), which can be ambiguous and arbitrarily interpreted. From the information, some female elementary school students thought they needed three Grains servings daily (Hong, 2013), but KDRIs recommends only 2 or 2.5 Grains servings for them.

(b) Proportionality: the recommended food intake from each food group. The current rear wheel design is misleading in two aspects.

• What slices are proportional to is not clarified: the number of servings, weight of food, volume of food, or calories. Nonetheless, the slices appear to indicate how much to eat from each group. The Vegetables slice is only slightly larger than the Meat slice. In fact, the college students participated in the authors' unpublished study have mistaken it as the recommended vegetable–meat consumption ratio. They also thought protein should take up about 30% of the daily calorie intake, while the expert recommendation is between 7-20% (Jung et al., 2015, p. 927).
• Arranging order: In the rear wheel, slices are arranged in the order of Grains, Meat, Vegetables, Fruits, and Dairy. The Meat slice comes before Vegetables, and it led some male elementary students thought they should eat more meat than vegetables following the order (Hong, 2013).

(c) Hydration: recommended daily fluid intake. The Food Balance Wheel does not clarify how much water to drink and whether the amount refers to fluid from all food and beverages or just plain water. Recommended fluid intake is 1900 ml for girls and 2100 ml for boys (Jung et al., 2015, p. 213), while Hong (2013) 's study participants thought to drink 500-1000 ml of water daily. The two wheels in the Food Balance Wheel cause unnecessary misunderstanding. Some college students participated in the authors' unpublished qualitative study interpreted that "The 12 front-wheel slices are the cups of water you need to drink", "The rear-wheel slices are the cups of water you need to drink, one per each food group", or "The front and rear wheel diameters ratio is the water to food ratio." They also shared a common belief of "2000 ml of water daily" that they learned from various news media.

Though widely used for frequency data visualization, empirical evidence speaks against using the pie chart for the quantity/proportion comparison task that requires a series of cognitive processes. A pie chart is complicated as a shape thanks to its curved and straight lines; the different orientations of pie slices add more cognitive difficulties of mentally rotating them for comparison (Cleveland and McGill, 1984, as cited in Huestegge & Pötzsch, 2018). Simkin and Hastie (1987) compared the pie chart that requires angle assessment, the simple bar chart that requires position comparison along a common scale, and the divided bar chart that requires length comparison. For the proportion-of-the-whole estimation task, the pie chart was as accurate as the simple and divided bar charts. For the task of comparing two parts of the same graph, however, the pie chart was the least accurate among them. The results were attributed to the position information in the simple bar chart that facilitated the anchoring cognitive process better than length or angle. According to Mackinlay (1986, as cited in Kilb & Jansen, 2016), position and length encode quantitative values more accurately than angle and area. As a pie chart visualizes information with angle and area, it is less effective than alternatives that use length or position, such as bar charts (Kilb & Jansen, 2016, pp. 196-198). Siirtola (2019, pp. 151-156) concluded that the pie chart was slower and less accurate than the stacked bar chart, especially when the difference between the elements was small.

Contrarily, Huestegge and Pötzsch (2018) reported the superiority of the CCW treemap to the pie chart in proportion and comparison judgment tasks. The CCW treemap combines the benefits of the column, pie and treemap charts altogether: (1) the ease of one-dimensional length comparison of a column chart, (2) visual cues such as a quarter, a half in a pie chart (p. 213) and (3) the high space efficiency of a treemap that allows for labelling inside the chart (Fig. 2).

Figure 2 A comparison of the pie chart, column chart, and CCW treemap

Such findings inspired us to design the new FBDG in the CCW treemap format (Fig. 3). The slices were made proportional to the weights of five food group representative items (a rice bowl for the Grains group, e.g.) for the following reasons:

Figure 3 The Food Balance Lunchbox with labelled photo icons

• Recommended dietary patterns are written in the number of servings, but servings in food groups differ in calories: a Grains serving is 300 kcal, a Meat serving is 100 kcal, while a Vegetables serving is 15 kcal. For consumers who do not know this, visualizing proportionality based on the number of servings can be misleading.
• The proportionality between food groups is best calculated and visualized in calories, but Zhou, Bell, Nusrat, Hingle, Surdeanu, and Kobourov (2018) reported that the consumer’s calorie estimation success rate is as low as 5 out of 20 correct guesses. Weight or volume information, meanwhile, is a more tangible and memorable cue for serving size estimation.
• Accurate volume measurement, comparison, or calculation, however, are difficult for irregular forms of food, e.g., ribs or salad, while weighing is feasible for most food items.
• Representative items in five food groups were selected because servings in a food group differ in both weight and volume: 100 g of tomatoes or 30 g of mushroom, e.g.

CCW treemap slices in the Food Balance Lunchbox are scaled up or down, to visualize quantitative differences between 37 calorie cases. In Fig. 3, B1200, B1700, and B2500 cases were drawn in different sizes.

The authors hypothesized that repeated exposure to accurate serving sizes enables people to estimate better in the eating context. Robinson et al. (2016) reported that repeated exposure to a large portion (not serving) sizes affect what consumers think a normal-sized food portion. Spanos, Kenda, and Vartanian (2015) concluded that informing consumers on how many servings were in the food portion reduced their food consumption. The Food Balance Lunchbox design accurately portrays serving sizes with photos to the last step of classification (rice, brown rice, or multigrain steamed rice, e.g.) Labels (“Tomato, 100 g”) clarify the serving size in weight, while photos illustrate it in volume. Miscategorizable items were clarified; for example, walnuts rich in both fat and protein were intentionally listed up in the Meat group.

The confusion caused by the water cup and two wheels in the Food Balance Wheel was resolved with a text description of recommended daily fluid intake—the amount including indirect water intake from food and beverages—added on the right side.

A quantitative study of comparing the Food Balance Wheel (“Wheel”) and the Food Balance Lunchbox (“Lunchbox”) was conducted with two research questions.

• (1) How does the Lunchbox design compare to the current Wheel design in communicating moderation, diversity, proportionality, and hydration information, according to Korean males in 30s?
• (2) How does the Lunchbox design compare to the current Wheel design in interestingness, aesthetics, and proportionality visualization, according to Korean males in 30s?

The first question investigated the quantified performance of the Lunchbox design in communicating four areas of food knowledge. Research hypotheses regarding the first question are:

• H1a: [Moderation] Serving size scores will be different between designs.
• H1b: [Diversity] Food item recognition and categorization scores will be different between designs.
• H1c: [Proportionality] Food groups and macronutrients percentages scores will be different between designs.
• H1d: [Hydration] Hydration scores will be different between designs.

The second question, meanwhile, asked participants’ subjective perception on the design; if the design appears to be interesting, looking good, or suitable for proportionality visualization. Research hypotheses are:

• H2a: Design has a significant influence on Interestingness ratings.
• H2b: Design has a significant influence on Aesthetics ratings.
• H2c: Design has a significant influence on Proportionality Visualization ratings.

Design evaluation data were collected with an online survey questionnaire until 200 valid datasets were obtained (Power = 0.8, α = 0.05, effect size = 0.05). Male consumers in 30s were recruited as study participants, because the obesity rate (46.7%, KOSIS, 2019) and the overeating rate, i.e., eating more than 125% of recommended daily calorie intake, are the highest (13.8%, KCDC, 2018, pp. 202-203) in this age-gender group. Preventive measures targeted to this group, meanwhile, are expected to be the most effective; they have various motivations to stay in a good shape.

Part I of the online questionnaire tested H1a, H1b, H1c, and H1d. Participants were randomly assigned to either Q_Wheel group or Q_Lunchbox group where the corresponding FBDG was given. Participants studied the information in each FBDG and finished four groups of tasks exemplified in Appendix B.

(1) Moderation tasks:

• (1.a) Serving sizes of given food items in weight. The Q_Wheel group saw a squid, potatoes, a watermelon piece illustrated in the Wheel design, while the Q_Lunchbox group saw a chicken drumstick, potatoes, and a watermelon piece included to the Lunchbox design.
• (1.b) Serving sizes in Grains, Meat, and Fruits groups and where participants located relevant information in the FBDG.

(2) Diversity tasks:

• (2.a) Identification of bean sprouts, corns, and lettuce in illustrations and photos.
• (2.b) Categorization of mushroom, orange juice, walnuts, and coffee mix into five food groups.

(3) Proportionality tasks and where participants located relevant information in the FBDG

• (3.a) Recommended food groups proportionality in percentage. The Q_Wheel group reported proportionality calculated with calories, while the Q_Lunchbox group reported proportionality calculated with weight.
• (3.b) Recommended macronutrients proportionality in percentage.

(4) Hydration task: recommended fluid intake in cups.

In Part II of the online questionnaire, H2a, H2b, and H2c were tested with participants’ ratings and comments on both Wheel and Lunchbox designs. Ratings on 10-point scales were given for Interestingness, Aesthetics, and Proportionality Visualization.

The Part 1 communication performance scores data were tested for normality and the results were all significant at p < 0.01. Non-parametric Mann-Whitney Wilcoxon U test was performed using the R software. The analyses of scores were summarized between designs (Table 2) and between tasks (Table 3).

Overall, Q_Lunchbox group scores (Mdn = 19) were higher than the Q_Wheel group scores (Mdn = 5) at p < 0.001. Moderation task scores were higher in Q_Lunchbox (Mdn = 5) than in Q_Wheel (Mdn = 1) at p < 0.001; H1a is accepted. Higher scores from the Q_Lunchbox group were foreseeable, considering relevant information was completely missing in the Wheel design. Diversity task scores were higher in Q_Lunchbox (Mdn = 6) than in Q_Wheel (Mdn = 3) at p < 0.001; H1b is accepted. The low recognition rates of bean sprouts (17%) and lettuce (48%) illustrations and low categorization scores of walnuts (10%) and coffee mix (28%) were much improved with the Lunchbox design. Proportionality tasks scores were higher in Q_Lunchbox (Mdn = 8) than in Q_Wheel (Mdn = 1) at p < 0.001; H1c is accepted. The Q_Wheel group responses to the Proportionality tasks showed clear discrepancies between expert recommendations and the popular beliefs of low-carb, high protein diet; participants thought that a third of their calorie intake should come from protein. Hydration task scores were higher in Q_Lunchbox (Mdn = 1) than in Q_Wheel (Mdn = 0) at p < 0.001; H1d is accepted.

Subsets of ratings data were tested for normality and the results were all significant at p < 0.001. Mann-Whitney U tests were conducted for comparison (Table 4). Interestingness data analysis confirmed that Wheel (Mdn = 6.00) was considered more interesting than Lunchbox (Mdn = 5.00) at p = 0.002. H2a is accepted. The Q_Lunchbox group rated Wheel (Mdn = 6.00) significantly higher than Lunchbox (Mdn = 5.00) at p = 0.007. Aesthetics ratings data analysis also confirmed that Wheel (Mdn = 6.00) was considered more beautiful than Lunchbox (Mdn = 5.00) at p = 0.002. H2b is accepted. The Q_Lunchbox group also rated Wheel higher (Mdn = 6.00) than Lunchbox (Mdn = 5.00) at p < 0.001, while the Q_Wheel group ratings were not different between designs. Proportionality Visualization ratings, however, was not significant between the designs, so H2c is rejected.

An analysis of participants’ comments was summarized in Table 5. On Wheel, its design and ease of understanding were positively mentioned; “interesting”, “original”, “organized”, or “easy to grasp at once.” The participants, however, also noticed that the Wheel did not provide critical information for practicing KDRIs guidelines, and the overall design was considered “difficult to understand the message”, “too abstract”, “lacks specific information on serving sizes or calories”, “did not know that it is visualizing certain information”, or “could not read proportionality from the wheel.” Some reported the bicycle metaphor “being irrelevant to the content.”

Lunchbox was reported to be more informative. Participants positively mentioned the serving size and proportionality information. The design was “easy to grasp with useful information”, “intuitive”, “much better than the other design on serving size information”, “allows for easier comparison”, and “looking organized.” The same design, meanwhile, was reputed to be “boring”, “not organized”, “not pretty”, and “difficult to understand” due to its “too much information.”