Nudging the cart in the supermarket: How much is enough information for food shoppers? Peter M. Todd Yvonne Rogers Stephen J. Payne Indiana University The Open University University of Bath Bloomington Milton Keynes Bath IN 47406, USA MK7 6AA, UK BA2 7AY, UK 001 812 855-3914 011 44 1908 652346 011 44 1225 384085 pmtodd@indiana.edu y.rogers@open.ac.uk s.j.payne@bath.ac.uk ABSTRACT about? Technology pundits and researchers are beginning to The amount of information available to help decide what foods to promote ‘augmented reality’ that uses Smartphones and other buy and eat is increasing rapidly with the advent of concerns ubiquitous technologies as the latest solution to this problem. about, and data on, health impacts, environmental effects, and Kuang [7], for example, marvels at the possibility: “What if all economic consequences. But this glut of information can be the food in your grocery store was marked with a QR code — you distracting or overwhelming when presented within the context of could compare the carbon footprints of two batches of produce… a high time-pressure, low involvement activity such as without having to spend any time or effort looking it up…” He supermarket shopping. How can we nudge people’s food continues by claiming it is “The best chance we have to speed shopping behavior in desired directions through targeted delivery crucial information about our world to the people living in it”. of appropriate information? We are investigating whether This vision, however, begs the research questions: Will people be augmented reality can deliver relevant 'instant information', that able to read and act upon such ‘instant information’? Will just can be interpreted and acted upon in situ, enabling people to make throwing more information at people have the desired galvanizing more informed choices. The challenge is to balance the need to effect of encouraging and empowering people to act upon various simplify and streamline the information presented with the need social causes (e.g., reducing carbon emissions) or improve their to provide enough information that shoppers can adjust their well-being (e.g., changing their diet)? Or do we need to tailor behavior toward meeting their goals. that information glut into simple nudges that make behavior change easy to achieve? And if so, what kind of nudges will work? Categories and Subject Descriptors Having instant information at one’s fingertips is certainly a promising technological approach but for it to succeed in H.5.2 [User Interfaces]: Evaluation/methodology. changing people’s behavior we need to understand how new forms of augmented reality are interpreted and used, especially General Terms when in situ. While the capabilities of the emerging technologies Design, Experimentation, Human Factors. are impressive in how they can project contextualised information, there is a paucity of research into whether people can Keywords process and exploit that extra information profitably. While it is Food information displays, supermarket shopping, ambient easy to imagine soda drinkers enjoying the surprise of being information interfaces, simple heuristics presented with a new branded game or a funny website on their mobile phone it is less clear whether people will make greener and healthier choices whilst managing their weekly budget when 1. INTRODUCTION presented with extra information of one form or another in the Increasingly we are told about the risks, costs, and benefits of middle of their busy shopping trip. Thus, research is needed, particular food choices. In response, a flood of information is firstly, to determine whether instant information will enable becoming available, online, on food labels, in information leaflets people to make better-informed choices when shopping and and books, from a variety of sources, aimed at informing the secondly, to ascertain whether and how such information is able consumer so that better decisions can be made while shopping. to change people’s behavior in the longer term. But all this information risks overwhelming and overloading the Technology for ubiquitous information delivery must balance shopper trying to navigate the complex store environment in a giving people enough new information to improve their decisions hurry, leading to the opposite outcome—poor decisions made against overwhelming them with new things to consider. Ambient without the proper input. How can all this information be information displays, as already used in homes and offices to consolidated, pruned down, and presented to supermarket provide feedback about energy consumption and nudge users shoppers in an easy to understand and meaningful form that will toward greater conservation, may strike the right balance in food actually help them make better choices about values they care purchase and consumption as well. However, as we discuss Copyright is held by the author/owner(s). MobileHCI 2010 September 7-10, 2010, Lisboa, Portugal. below, moving beyond momentary nudges toward long-term ACM 978-1-60558-835-3/10/09. behavior change requires providing detailed-enough feedback to More recently, Bargrams have been developed for e-commerce enable learning what to do in the future, for instance on the next applications. For example, EZChooser helps consumers choose shopping trip. We argue that we must improve our (currently one item from many (e.g., cars) through selecting attributes that limited) understanding of whether and how people attend to and are visualized as parallel horizontal interactive histograms along a learn from visualizations of multi-dimensional information while number of dimensions [16]. engaged in an ongoing activity such as food shopping, using But even though these kinds of visualizations are mostly targeted cognitive science models of decision-making and learning at non-expert users, they are essentially visual query languages together with design principles for information visualization and that require considerable cognitive effort to interpret. Can relevant interaction design. dimensions of products such as food be represented in simple ways that can be glanced at and perceived rapidly to guide 2. BACKGROUND shopping decisions in situ? Rational theories of decision-making [e.g., 15] posit that making a choice involves weighing up the costs and benefits of different 3. DISPLAYING NUDGES courses of action. When alternatives are ordered on more than one We propose that rather than providing ever more information to relative dimension, this involves compensatory strategies where enable consumers to compare products in minute detail when information is processed exhaustively and trade-offs made making a choice, a better strategy is to design technological between features. Such strategies are very costly in computational interventions that provide just enough information and in the right and informational terms – not least because they require the form to facilitate good choices. One solution is to exploit new decision-maker to find a way to compare apples and oranges. forms of augmented reality technology that enable ‘information- Non-compensatory strategies may be used instead as a form of frugal’ decision-making, in the context of an intensive activity bounded rationality where not all of the available information is replete with distractions (i.e., shopping in a supermarket or used and trade-offs can be ignored [10]. Furthermore, recent deciding at the kitchen table what to have for dinner). research in cognitive psychology has shown people tend to use An important consideration when representing multiple simple heuristics of this sort when making decisions [6]. A dimensions that can be glanced at and perceived rapidly is to theoretical explanation is that human minds have evolved to act enable comparisons to be made and cumulative information quickly, making ‘just good enough’ decisions by using fast and inferred in situ. For example, simple contrasting icons (e.g., frugal heuristics. We typically ignore most of the available thermometer icons, percentage bars, balls that change in color) information and rely only on a few important cues. In the can be presented which increase or decrease in amount in relation supermarket, shoppers make snap judgments based on a paucity to the values being represented. Another approach is to fuse of information, such as buying brands they recognize, are low- relative measures on different dimensions (e.g., greenness, price, priced, or have attractive packaging [12] – seldom reading other fat level) into singular displays where shape carries the salient package information. information, such as a rectangle that gets taller to convey a At the same time, recent consumer surveys reveal that shoppers nutritional dimension that is general (healthiness) or specific (e.g., are demanding more information about the products they buy and salt content) and wider to convey price. A third dimension, such are becoming increasingly aware of the global consequences of as ‘greenness’, could be added by filling in the rectangle with a the decisions they make [4]. This raises the question of whether it shade from red to green to show the amount of carbon emissions is possible to encourage people to pay attention to more for that product. Similar to the idea behind Chernoff faces, the information, such as nutritional, ethical, and environmental visualizations will be placed side by side to enable quick features, when making their food purchases and subsequently comparisons. deciding how to use what they have bought to make healthy meals Another important question is whether to use ‘emotive’ that have a low carbon footprint. visualizations that can persuade people to select food items they However, there is a scarcity of research on how people use multi- might not otherwise choose. Various persuasive technologies have dimensional information under time pressure and the extent to recently been developed to encourage people to take more which it effects rapid decision-making [5]. Visualization research exercise. Examples include Fish‘n’Steps [8]; Chick Clique ([13] has tended to adopt an unbounded rationality perspective, and UbiFit [2] where various types of graphic representations assuming that people have the time and cognitive capacity to pull (e.g., butterflies, flowers, bar charts) are used to represent amount out and use whatever information the displays provide. Within the of exercise type performed, e.g., cardio, strength training, and field of Information Visualization there have been a number of walking. Findings from a three-month field trial of UbiFit showed tools that have been developed specifically to represent that these display systems can be motivating, encouraging multidimensional data that allow for comparisons [1]. Other participants to maintain fitness levels that were significantly simple canonical forms such as tables and trend graphs have been higher than for a control group without the visualizations [3]. developed for web-based decision-making activities, including More dramatically, Shultz et al. [11] have shown how emoticons online shopping, making investments, choosing insurance policies can have a powerful effect on changing behavior for energy or buying a house. An innovative approach has been to develop consumption. In their study, a number of householders were told interactive visualizations that show some aspects of the exactly how much energy they had used and the average performance of objects for a range of different parameter values. consumption of energy by others in their neighborhood. The An early example was the Influence Explorer [14] that allowed a above-average energy users then significantly decreased their user to compare how products (e.g., a light bulb) perform on core energy use while the below-average energy users significantly values (e.g., brightness and working life) when varying multiple increased theirs (presumably because they felt they had more parameters (e.g., diameter, length, material and number of coils). room to increase their consumption). But then the researchers them feel good in front of other shoppers [11]? Would the tested the effect of instead giving householders who consumed prospect of others seeing just how much butter and cheese they more than average an unhappy smiley icon – suggesting it was are buying make shoppers think about buying less, or just socially disapproved – and those who consumed less than the thinking about shopping elsewhere? norm a happy smiley icon – suggesting their energy consumption Assuming such an ambient information display Cumulative Tool was socially approved. The impact of providing these two achieves the desired features of providing some feedback without visualizations was dramatic: The big energy users showed an even overloading the decision maker, without undesired effects of larger decrease in their energy use while the below-average users scaring shoppers off or making them “boomerang” and offset their did not change their energy consumption upward (presumably good behavior with poorer choices, the question remains whether because the addition of the happy emoticon suggested they were this kind of simple display provides enough feedback to allow the doing just fine). shopper to adjust behavior in the desired direction, e.g. reduced sodium or enhanced green-ness. Seeing that one’s entire cart is 4. LEARNING FROM NUDGES red-lining above the goal level may motivate behavior, but it does What then is a good way to provide appropriate information not directly indicate what to do to bring the level back down. quickly and simply to shoppers in order to aid their decision- Thus, we must develop and test methods for ensuring that the making during the hectic, distracting setting of a trip to the (minimal) information delivered is actually actionable and supermarket? Here we assume the shoppers have selected a conducive to behavior change. particular dimension that they care about and want to change in terms of their buying behavior—for instance, choosing products There are at least three approaches that can be taken to solving that are lower fat, or more sustainably grown. To inform shoppers this problem, which is essentially one of allocating global about how they are doing in achieving this particular goal during feedback appropriately to individual choices of products (akin to their shopping expedition, cumulative values of the dimensions of the “credit assignment” problem in machine learning). First, we interest across all products chosen so far could be summed up and could leave it all up to the users, and assume (or hope) that when displayed in an ambient manner as the current ongoing overall they end their shop with a “green” cart, they will buy more things score “projected” onto the handle of the shopping cart as a color. like those the next time around, and when they get a “red” cart, For example, a green handle could signify that the shopper has they will buy different things next time. This leverages the obtained a ‘carbon footprint’ or ‘fat content’ score below their human shopper’s intelligent ability to learn from diffuse target (or below some population average), while a red handle reinforcement over time, but it will probably be slow, requiring would indicate that the cart’s contents are above the desired level, many shopping outings before reliable change occurs. Second, to with intermediate levels indicated by intermediate colors (see speed up this process, we could provide more specific feedback Figure 1). about each product that goes into the cart, for instance momentarily flashing the ambient display with a color corresponding to the box of sugar-frosted chocolate bombs or bag of figs being chosen. This will allow shoppers to make more targeted decisions about each product, provided they remember that individual feedback. Third, to remove the need for such memory, a further interface can be developed to let shoppers query how they should adjust their purchases to come closer to their goal. This could take two main forms. A Comparative Tool could run as a ‘private’ mobile application on a smartphone or PDA and be displayed on the device or somewhere in the environment, such as the shopper’s hand or the product package itself. After identifying the product via a photo or code scanner, the tool will show the product values on the dimensions of interest, and indicate whether this product Figure 1: Two hypothetical shopping carts with (a) red and helps or hinders the achievement of the current shopping goal. (b) green glowing handles, indicating aggregate ‘healthiness’ This interface could also be used in a comparative manner, of products selected relative to the average for a weekly shop scanning two or more products while they are still on the shelf for a family of four and then showing at a glance which product is best based on the selected dimensions. Such an ambient and publicly visible display must first be studied As a second ‘off-line’ form of providing more explicit feedback, a to see if it fits with how people want to shop, or engenders Collaborative Tool running on a home computer or surface unexpected side-effects. Will people be more or less likely to display would allow shoppers to find out further information change their behavior when information about the contents of about the products they have bought once they get them home, their shopping cart is publicly visible for all to see rather than along with input from their families. Multiple users could reflect being privately displayed? Would shoppers try to fill their cart and discuss together the decisions behind their food purchases with healthy and green foods and on finding they were under the with a view to attaining their goals at their next weekly shop, average then treat themselves to luxury goods high in fat and food exploiting collaborative planning and social pressures that take miles? Would having their shopping cart glow green at the check- place in a family setting. An interactive planner application would out, indicating the contents were well below the average, make enable family members to find out more about particular dimensions (e.g., nutritional values) on a product, meal, or [4] EDS IDG Shopping Report (2007) Shopping Choices: weekly-shop basis, and provide recipe-specific visualizations attraction or distraction? Downloaded 28/08/09 enabling items to be swapped. For example, a suggestion by dad http://www.eds.com/industries/cir/downloads/EDSIDGRepor to cook coq-au-vin for dinner will show it is low on ‘greenness’ t_aw_final.pdf (because of a large carbon footprint). This is a dimension the son [5] Feunekes G., Gortemaker, I.. Willems, A., et al (2008) Front- has selected as an informational layer. Alternative items can be of-pack nutrition labelling: Testing effectiveness of different swapped with the chicken, such as tofu, which may then be shown nutrition labelling formats front-of-pack in 4 European by the application to have a higher greenness value (i.e., smaller countries, Appetite, 50, 57-70 carbon footprint). Finally, specific shopping lists could be generated that would achieve the goals set by the shopper and [6] Gigerenzer, G., Todd, P.M., and the ABC Research Group others involved. (1999) Simple heuristics that make us smart. New York: To test whether any of these approaches succeeds in nudging Oxford University Press. shoppers’ behaviour in specific directions within a reasonable [7] Kuang, C. (2009) Better Choices through technology. time-span, both lab-based experiments and field studies are Downloaded http://www.good.is/post/better-choices- needed. One line of investigation must assess how the different through-technology/ 28/8/09 information displays for the tools described above affect user decision-making strategy, focusing on when and how the [8] Lin, J.J. Mamykina, L., Lindtner, S., Delajoux, G. and Strub, interactive display of information enables fast and frugal H. (2006) Fish ‘n’ Steps: Encouraging Physical Activity with decisions. This must then be tested further in supermarket an Interactive Computer Game. Proceedings of UbiComp. studies, using techniques such as mobile eye tracking, observation 261-278. and talk aloud methods to determine what people look at and how [9] Rogers, Y., Lim, Y. Hazlewood, W. and Marshall, P. (2009) they use the comparative and cumulative tools. Longitudinal Equal opportunities: Do shareable interfaces promote more studies are also needed to determine whether the tools proposed group participation than single users displays? Human- have long-term impact on behavior, and how quickly such change Computer Interaction, 24 (2), 79-116. occurs. Various kinds of households (e.g., family, young people, retired single) should be compared in terms of whether and how [10] Rothrock, L & Yin, J. (2008) Integrating compensatory and their shopping patterns and meal planning behavior change when non-compensatory decision making strategies in dynamic using the tools—different groups of people may be more or less task environments. In T Kugler et al. (eds) Decision influenced by different types of nudges, and we cannot assume a Modeling and Behavior in Complex Environments. NY: one-size-fits-all approach. Springer Whether these various kinds of information delivery can help [11] Shultz, W., Nolan, J., Cialdini, R., Goldstein, N., and move people in the direction of better decisions—in the food Griskevicius, V. (2007) The constructive, destructive and shopping domain, or in other applications—remains to be seen. reconstructive power of social norms. Psychological Science, Emerging research suggests that simple visualizations can be 18, 429-34. designed to be information-frugal and emotive – encouraging people to change their behavior at the point of decision-making. [12] Todd, P.M. (2007) How much information do we need? But the trick will be balancing frugality and simplicity with European Journal of Operational Research, 177, 1317-1332. enough feedback detail to allow people to change their choices at [13] Toscos, T., Faber, A.M., An, S. and Gandhi, M. (2006) a pace that is sufficiently rapid and noticeable to be rewarding Chick clique: persuasive technology to motivate teenage and motivating for long-term behavior change. girls to exercise. Proc. CHI Extended Abstracts, 1873-1878. 5. ACKNOWLEDGEMENTS [14] Tweedie, L., Spence, B., Williams, D., Bhogal, R. (1994) Thanks to Ricky Morris for creating Figure 1. The attribute explorer. CHI Companion Proceedings, 435- 436. 6. REFERENCES [15] von Neumann, J. and Morgenstern, O. (1944) Theory of [1] Card, S., Mackinlay, J., Shneidermann, B. (1999) Readings Games and Economic Behavior. Princeton University Press. in Information Visualization. Academic Press. [16] Wittenburg, K., Lanning, T., Heinrichs, M. and Stanton, M. [2] Consolvo, S., Klasnja, P., McDonald, D. W., et al. (2008) (2001) Parallel bargrams for consumer-based information Flowers or a robot army?: encouraging awareness & activity exploration and choice. UIST Proceedings, 51-60. with personal, mobile displays. Proc. UbiComp'08, 54-63. [3] Consolvo, S., McDonald, D.W., and Landay, J.A. (2009) Theory-driven design strategies for technologies that support behavior change in everyday life. Proc. CHI '09, 405-414.