EALTHY ROCERY HOPPING H G S

H

EALTHY

G

ROCERY

S

HOPPING

How real-time nutritional feedback affects the healthiness of grocery shopping

baskets.

BY

H

EALTHY

G

ROCERY

S

HOPPING

How real-time nutritional feedback affects the healthiness of the grocery

shopping baskets

-RESEARCH MASTER THESIS-

Author : Wieteke de Vries

Address : Gedempte Kattendiep 100 | 9711 PT | Groningen

Phone Number : +31 (0) 642634896

E-mail address : w.c.de.vries@rug.nl

Student Number : S3199053

Department : Faculty of Economics and Business

Study : MSc Research Master

First Supervisor : Prof. Dr. Ir. Koert van Ittersum

Second Supervisor : Prof. Dr. Jenny van Doorn

ABSTRACT

PREFACE

When I enrolled for the marketing pre-master, I never could have imagined to be at the point where I am today. I initially started with the marketing pre-master to find new challenges for myself; this journey has certainly not let me down in that regard. Courses from mathematics (that was especially a though one) till consumer psychology, I was intrigued and enjoyed every bit of it. During my four years as a student at the University of Groningen, I have come to learn that the field of marketing and consumer behaviour offers endless research opportunities, there is still so much more to uncover. Moreover, I believe that good can be achieved when we use marketing to improve consumer and societal well-being. I count myself lucky to be able to continue working on my research in a field that has sparked my interest from the start.

This research is the cherry on top of my four-year academic cake. But I could never have done any of this without the great support of family, friends, teachers and fellow students. A special thanks goes out to the support of my parents Sipke and Lucie de Vries (jullie steunen en motiveren mij altijd ook al is het soms lastig te begrijpen waar ik nou precies mee bezig ben. Dat vraag ik mijzelf ook weleens af trouwens), my beloved boyfriend Ward Brassé, and the great guidance of my supervisors Koert van Ittersum and Jenny van Doorn. I look back with pleasure on a valuable learning experience, and I am excited for what the future holds.

T

C

1 INTRODUCTION 6

2 THEORETICAL FRAMEWORK 9

2.1 THE DYNAMIC DECISION MAKING ENVIRONMENT OF GROCERY STORES: BALANCING BEHAVIOUR 9 2.2 DEVELOPING EFFECTIVE IN-STORE INTERVENTIONS: REAL-TIME NUTRITIONAL FEEDBACK 10 2.3 BOUNDARY CONDITIONS OF REAL-TIME NUTRITIONAL FEEDBACK: FEEDBACK FORMATS 13 2.4 BOUNDARY CONDITIONS OF REAL-TIME NUTRITIONAL FEEDBACK: INTEREST IN HEALTH 15

3 METHODS 17

3.1 EXPERIMENT 1 17

3.1.1 PROCEDURE 18

3.1.2 FEEDBACK MANIPULATION 19

3.1.3 SHOPPING EXPERIENCE AND SHOPPING SATISFACTION 20

3.1.4 ATTENTION PAID TO ATTRIBUTES 20

3.1.5 INTEREST IN HEALTHY EATING 20

3.1.6 CONTROL VARIABLES 21 3.2 RESULTS 21 3.2.1 FINAL SAMPLE 21 3.2.2 DESCRIPTIVES AND FREQUENCIES OF PRODUCT CHOICES 22 3.2.3 SHOPPING SATISFACTION AND ATTENTION PAID TO ATTRIBUTES 23 3.2.4 RELIABILITY OF THE MEASUREMENT SCALES 24 3.2.5 NEW VARIABLES 25 3.2.6 CORRELATIONS 25

3.2.7 LINEAR REGRESSION ANALYSIS 26

3.2.8 DISCUSSION 28

3.3 EXPERIMENT 2 29

3.3.1 PROCEDURE 30

3.3.2 FEEDBACK MANIPULATION 31

3.3.3 SHOPPING EXPERIENCE & SHOPPING SATISFACTION 32

3.3.4 ATTENTION PAID TO ATTRIBUTES 33

3.3.5 INTEREST IN HEALTHY EATING 33

3.3.6 CONTROL VARIABLES 33

3.4.1 FINAL SAMPLE 35

3.4.2 DESCRIPTIVES AND FREQUENCIES OF PRODUCT CHOICES 35

3.4.3 SHOPPING SATISFACTION AND ATTENTION PAID TO ATTRIBUTES 37

3.4.4 RELIABILITY OF MEASUREMENT SCALES 38

3.4.5 NEW VARIABLES 39

3.4.6 CORRELATIONS 39

3.4.7 LINEAR REGRESSION ANALYSIS 40

3.4.8 DISCUSSION 46

4 GENERAL DISCUSSION 48

4.1 CONTRIBUTIONS AND IMPLICATIONS 51

4.2 LIMITATIONS AND FUTURE RESEARCH: 52

4.3 CONCLUSION 54

1 Introduction

Obesity is a global health concern with great negative consequences for the quality of life; it has a severe impact on individual and societal well-being (WHO, 2020). For individuals, obesity is a major risk factor for noncommunicable diseases such as diabetes or cardiovascular diseases (e.g., heart diseases, stroke), thereby putting pressure on societies and the health care systems (WHO, 2020). With over 1.9 billion adults suffering from being overweight, and over 650 million adults suffering from obesity globally, the obesity epidemic has become one of the grand challenges societies are currently facing.

consumers while shopping for groceries, from which in turn effective health interventions can be built that may reduce overconsumption tendencies and that help to combat the obesity epidemic.

Current in-store interventions mostly promote healthy purchases within product categories, such as front-of-package labels (FOP labels) that provide at-a-glance nutritional information for a food-item to assist consumers in making accurate and healthy choices at point-of-purchase (Cowburn & Stockley, 2005; Lowe, Fraser, & Souza-Monteiro, 2015; Temple, 2020). Such in-store interventions are based on food-related research that has sought to understand consumers’ purchase decisions for individual products, rather than baskets of products (Lowe et al., 2015). As a result, in-store (within-product category) interventions fail to take the interdependence of food purchase decisions into account (Gilbride, Inman, & Stilley, 2015), thereby implicitly and erroneously assuming that a healthy choice triggered by the intervention does not affect subsequent food purchases. However, consumers have the tendency to compensate the healthiness of one purchase with unhealthy, but tastier, purchases (i.e., balancing) (Ma, Ailawadi, & Grewal, 2013; Raghunathan, Naylor, & Hoyer, 2006; Trivedi, 2016). Additionally, consumers are limited in their ability to accurately evaluate nutritional information for a range of foods (Cowburn & Stockley, 2005), which often results in an underestimation of the nutritional value that (un)healthy foods contain (Chandon & Wansink, 2007; Chernev & Gal, 2010). Consequently, while within-product category interventions can effectively stimulate one healthy purchase, they may fail to improve, or can even reduce the healthiness of the end-of-trip basket (e.g., Waterlander et al., 2012). Therefore, we propose to shift focus from understanding individual food decision-making to understanding how consumers make multiple, interdependent food purchase decisions. This shift of focus allowed us to take into account the interdependency of food purchases throughout the grocery shopping trip. Moreover, the perspective of grocery shopping trips as a sequence of interdependent food purchases (as opposed to individual food purchases) allowed us to develop and test interventions for improving the healthiness of the end-of-trip basket, which most in-store interventions currently have not been able to do. Specifically, we examined the effect of real-time nutritional feedback – the provision of aggregated nutritional information updated as consumers select or remove items from their shopping basket - on the healthiness of end-of-trip baskets.

grocery stores. Yet, research suggests that providing feedback may yield different positive effects (e.g., Fischer, 2008; Hutton, Mauser, Filiatrault, & Ahtola, 1986; Kluger & DeNisi, 1996; van Ittersum, Wansink, Pennings, & Sheehan, 2013). For instance, providing consumers with real-time spending feedback can stimulate consumers to spend more of their budget, decrease mental stress, and increase customer satisfaction (van Ittersum et al., 2013). Building on this literature, this research examined the extent to which real-time nutritional feedback affects the healthiness of food purchase decisions made throughout a simulated shopping trip. Additionally, we explored the boundary conditions of real-time nutritional feedback by not only taking into account consumers’ interest in healthy eating, but also to examine the differential effect of the feedback format, as research shows that the format in which feedback is presented can affect how individuals process and use feedback for subsequent decision making (Häubl & Trifts, 2000; Lurie & Swaminathan, 2009). We first examined the main effect of real-time nutritional feedback on the relative healthiness of end-of-trip baskets. The results of this study demonstrated a positive main effect of real-time nutritional feedback: feedback regarding the overall (relative) healthiness of the shopping basket resulted in a decrease in the number of calories in the end-of-trip basket. Secondly, we examined the differential effect of two feedback formats that differed in compatibility on healthiness of the end-of-trip basket, as well explored possible moderating effects of an interest in healthy eating. The results of this study showed that providing real-time nutritional feedback (for the entire shopping basket) as the cumulative sum of food purchases significantly and systematically improved the healthiness of the end-of-trip basket. Providing real-time nutritional feedback (for the entire shopping basket) as the average nutritional value of food purchases however was only effective for participants who indicated to have a great interest in healthy eating. Overall, the results of this research suggest that providing consumers with real-time nutritional feedback for the shopping basket can effectively facilitate healthy purchases. Moreover, the findings suggest that real-time nutritional feedback can improve the overall healthiness of the end-of-trip basket, as opposed one individual purchase decision.

tendency to compensate for (un)healthy food purchases (i.e., balancing behaviour) (Trivedi, 2016). With this research, we aim to contribute to the complex and increasing body of food-related literature on in-store decision making by uncovering relevant insights regarding consumer purchase decisions in response to real-time nutritional feedback. As current in-store interventions neglect the interdependence of food purchase decisions, and since consumers balance their (un)healthy food purchases between-product categories, most (within-product category) in-store interventions currently are limited in improving the healthiness of end-of-trip baskets. With this research, we aim to enhance scientific understanding of the dynamic decision making process in grocery stores by examining how real-time nutritional feedback affects the healthiness of all food purchases throughout the shopping trip (i.e., end-of-tip basket), as opposed to individual food purchases. Additionally, the feedback literature tends to focus on long-term feedback effects on decision-making (Atkins, Wood, & Rutgers, 2002; Diehl & Sterman, 1995). However, while shopping, consumers make many purchase decisions within a short period of time. By examining feedback in a grocery shopping context, we also aim to increase scientific understanding of short-term feedback effects on consumer purchase decisions. Moreover, as the results suggest that real-time nutritional feedback can significantly improve the healthiness of food purchases within one shopping trip, this research holds important implications for the development of effective in-store interventions that go beyond the one-choice context.

2 Theoretical Framework

2.1 The dynamic decision making environment of grocery stores: Balancing behaviour

to accomplish this. For instance, when processing resources are constrained, consumers are more likely to engage in automatic processing resulting in affect-laden purchase decisions (i.e., unhealthy, tasty foods) (Shiv & Fedorikhin, 1999).

Besides the limitations of cognitive processing resources, consumers are also prone to psychological processes of which they are often not aware, yet these processes influence the healthiness of their purchases decisions (Wansink, 2004; Wansink, Kent, & Hoch, 1998). In particular, consumers are prone to balancing behaviour; consumers tend to select healthy foods from one product category to compensate for unhealthy, but tastier, food purchases from other product categories (Hui, Bradlow, & Fader, 2009; Trivedi, 2016). However, as consistently making accurate healthy food decisions is cognitively demanding (Crosetto, Muller, & Ruffieux, 2016; Shiv & Fedorikhin, 1999), and since most consumers are cognitively limited in their ability to accurately evaluate nutritional information for a range of foods (Cowburn & Stockley, 2005), balancing behaviour can lead to the underestimation of nutrients (e.g., sugar, fat) or micro-nutrients (e.g., vitamins) that are purchased (Chernev & Gal, 2010). In-store interventions promoting healthy food purchases within a specific product category may therefore have a limited impact on the overall healthiness of the end-of-trip basket. For instance, research shows that while price reductions can effectively stimulate a healthy purchase, they may fail to improve, or can even reduce the healthiness of the end-of-trip basket (Waterlander et al., 2012). Hence, instead of promoting healthy within-product category purchases, in-store interventions may lead to healthier end-of-trip baskets when the intervention is developed for promoting healthy between-product category purchases (Cawley et al., 2015).

In sum, during grocery shopping, consumers (often unconsciously) balance their (un)healthy food purchases and inaccurately evaluate the healthiness of those purchases, leading to a gap between the perceived and the actual healthiness of end-of-trip baskets. This has important implications for in-store interventions, as promoting individual healthy purchases may have a limited, or even the opposite effect of what the intervention is intended for (i.e., improving consumption patterns and diet quality).

2.2 Developing effective in-store interventions: Real-time nutritional feedback

patterns and diet quality shows that the provision of health information can effectively promote healthy purchases when the presented information makes the health trade-off salient: by framing information to draw attention to trade-offs (Downs, Wisdom, & Loewenstein, 2015), by encouraging consumers to explicitly consider their health (Hare, Malmaud, & Rangel, 2011), or by using salient nudges to increase attention to healthier alternatives (Wilson, Buckley, Buckley, & Bogomolova, 2016; Wisdom, Downs, & Loewenstein, 2010). To facilitate healthier end-of-trip baskets, we propose to offer consumers real-time nutritional feedback, which we define as the provision of aggregated nutritional information updated as consumers select or remove items from their shopping basket (altered from real-time calorie feedback defition of Gustafson and Zeballos, 2019). In particular, real-time nutritional feedback may increase the salience of the health trade-off for all purchase decisions made by consumers, thereby promoting healthy purchases throughout the entire shopping trip. Research shows that highlighting nutritional value can promote healthier purchases (Cawley, 2015; Kiesel & Villas-Boas, 2013), particularly when consumers face time or attentional constraints while making their purchase decisions (Crosetto et al., 2016). Moreover, as purchase decisions made in grocery stores are often low in involvement (Dickson & Sawyer, 1990.; Hoyer, 1984), consumers tend to devote attentional selection to only the most salient product attributes, which in turn has a large influence on purchase decisions (Bordalo, Gennaioli, & Shleifer, 2013). Not only do such findings suggest that consumers do not actively process all the nutritional information available, it also highlights how attentional focus can easily shift away from the healthiness of food purchase decisions (let alone the healthiness of the end-of-trip basket). Hence, the health trade-off may not be regularly and consciously salient in consumers’ minds. By providing real-time nutritional feedback however, we may shift consumers’ attention towards the (actual) nutritional value of the end-of-trip basket, as opposed to one salient attribute of one individual purchase decision. Consequently, real-time nutritional feedback may increase the salience of the health trade-off for all purchase decisions.

errors as the feedback provides an accurate overview of the nutritional value in the shopping basket. Moreover, real-time nutritional feedback can decrease cognitive costs of monitoring the healthiness of the shopping basket. Since cognitive capacity limitations constrain the processing of information (Marois & Ivanoff, 2005), and since consumers are more likely to purchase unhealthy, but tasty, foods when processing resources are constrained (Shiv & Fedorikhin, 1999), the real-time nutritional feedback could alleviate cognitive capacity limitations, thereby reducing the likelihood of affect-laden (unhealthy) purchases over healthier purchase decisions. Additionally, scholars have argued that the ability to monitor food purchases is important for improving diet quality (Bublitz et al., 2010; Glanz & Bishop, 2010; Hollywood et al., 2013). As such, the presence of real-time nutritional feedback can facilitate a more accurate evaluation of nutritional value, which in turn may results in healthier end-of-trip baskets.

If real-time nutritional feedback can effectively reduce random errors, it may also correct for the systematic under-counting of the nutritional value of food purchases that may stem from motivated reasoning (Gustafson & Zeballos, 2019). Individuals often (unconsciously) find ways to interpret negative information as irrelevant or biased when this information contradicts their own perceptions (Alloy & Abramson, 1979; Bénabou & Tirole, 2016). Hence, to justify unhealthy food purchases, consumers tend disregard or underestimate the nutritional information (Scharff, 2009), interpreting it as “not that bad” - even when the nutritional information highlights the unhealthiness of the food purchase (e.g., by a front-of-package label). Consumers can also choose to justify (un)healthy purchases by compensating with a subsequent (un)healthy food purchase (i.e., balancing) (Trivedi, 2016). However, as the real-time nutritional feedback makes the health trade-off of all food purchases salient, it leaves less room for motivated consumers to systematically disregard or underestimate the nutritional value of food purchases. As such, real-time nutritional feedback may also increase consumer awareness of balancing tendencies.

believe that real-time nutritional feedback can effectively promote a healthier end-of-trip basket by 1) reducing random errors, and 2) by reducing the tendency to disregard or underestimate the nutritional value of food purchases. This leads to the following hypothesis:

H1: Real-time nutritional feedback positively affects the healthiness of the end-of-trip basket.

2.3 Boundary conditions of real-time nutritional feedback: Feedback formats

The format in which the real-time nutritional feedback is presented may determine the extent to which consumers use the feedback for healthier purchase decisions. Research shows that the format in which information is presented (i.e., feedback format) affects how individuals use the provided information for subsequent decision-making (Häubl & Trifts, 2000). In particular, feedback formats affect the ease of processing (fluency) which in turn affects evaluations and decision making (Reber, Schwarz, & Winkielman, 2004). In complex decision-making environments, consumers can greatly benefit from feedback in a format that is compatible with the decision-making environment (Häubl & Trifts, 2000), which can also be referred to as the proximity compatibility principle (PCP). The PCP is a general formulation of research findings demonstrating that people make effective use of feedback when this is presented in a format (perceptual proximity) that is compatible with the demands of the decision-making task (process proximity) (Boles & Wickens, 1987; Wickens & Andre, 1990; Wickens & Carswell, 1995). In particular, PCP assumes that compatible feedback is beneficial in dynamic decision-making environments because it facilitates a more fluent processing of all the available information, thereby alleviating limitations of cognitive processing (Atkins et al., 2002). Hence, the compatibility between the real-time nutritional feedback format (perceptual proximity) and the dynamic decision-making environment of grocery stores (processing proximity) may affect the extent to which consumers use the feedback for healthier purchase decisions.

is high when the feedback format combines multiple pieces of information (e.g., averages, simplified nutritional overviews such as front-of-package labels) and is low when the feedback format presents multiple pieces of information separately (e.g., nutrition labels that present individual numbers such as calories, fat, sugars). Following PCP logic, consumers will benefit most from real-time nutritional feedback presented in a high perceptual proximity format, as this is compatible with the (high processing proximity) decision-making environment of the grocery store. Hence, to unsure an effective use of real-time nutritional feedback by consumers (i.e., healthier purchase decisions), the feedback should present an integrated overview of the nutritional information available, as this would facilitate a more fluent processing of the information (e.g., by presenting the average relative healthiness of the shopping basket).

The integration of all the available nutritional information presented by the real-time nutritional feedback should facilitate fluent processing and in turn positively affect purchase decisions, yet providing feedback that is too aggregated may lose its significant salience. As most purchase decisions are low involvement (Dickson & Sawyer, 1990; Hoyer, 1984; Zaichkowsky, 1985), and since consumers devote their attention to the most salient product attributes (Bordalo et al., 2013), real-time nutritional feedback needs to systematically enhance the salience of the health trade-off (i.e., throughout the entire shopping trip). However, when the presented nutritional information is too aggregated, consumers may not (be able to) notice any changes in the real-time nutritional feedback after each purchase decision. In particular, changes in the relative health of the shopping basket - as presented by real-time nutritional feedback that is too aggregated - could fall below the “just noticeable difference” threshold, which is a general threshold of consumers’ ability to detect changes in stimuli in the environment (i.e., changes in the real-time nutritional feedback) (Britt & Nelson, 1976)1. Consequently, the impact of real-time nutritional feedback may be limited, leaving room for random errors and motivated consumers to disregard or discount the nutritional information (i.e., limited use of information for making healthier purchase decisions). On the other hand, providing detailed nutritional information about food purchases may be detrimental for improving the healthiness of the end-of-trip basket as this would increase the already copious number of informational cues that consumer have to cognitively process (e.g., nutritional information on labels, price, promotions, brands, etc.). Moreover, research shows that people who experience difficulties in processing feedback are less willing to use that information for subsequent decisions (Song & Schwarz,

2008). Thus, for an effective use of real-time nutritional feedback, the compatibility between the feedback format and the decision making environment is crucial; the feedback should facilitate fluent processing to ensure effective use of the feedback, as well to meet the threshold of just noticeable difference to ensure that attention selection is directed towards the health trade-off (i.e., increase in salience of health trade-off).

In sum, the format in which the real-time nutritional feedback is presented may affect the extent to which feedback is used for healthier purchase decisions. Following the proximity compatibility principle, high perceptual proximity feedback formats are compatible with the dynamic decision-making environment of grocery stores, thereby supporting cognitive processing recourses of consumers for a better evaluation of the nutritional information. However, the feedback should move beyond the just noticeable difference threshold, as a failure to meet this threshold may result in consumers disregarding or discounting the nutritional feedback. This leads to the following hypothesis:

H2: The positive effect of real-time nutritional feedback is moderated by the format in which the feedback is presented, whereby a compatible feedback format strengthens the

relationship.

2.4 Boundary conditions of real-time nutritional feedback: Interest in health

interest in healthy eating on the other hand, tend to base their purchase decisions on taste (Mai & Hoffmann, 2012) and devote less attention to nutrition labels, although this does not necessarily affect subsequent food purchases (Clement, Kristensen, & Grønhaug, 2013; Fenko, Nicolaas, & Galetzka, 2018).

As we expect that the real-time nutritional feedback enhances the salience of the health trade-off, the attentional selection of consumers with a lower interest in healthy eating may shift more towards the nutritional value of their food purchases, thereby facilitating healthier purchase decisions. Indeed, research shows that the provision of simplified nutritional information is especially beneficial for less knowledgeable and motivated consumers (i.e., consumers with a lower interest in healthy eating) (Andrews, Netemeyer, & Burton, 2009; Lowe, de Souza-Monteiro, & Fraser, 2013). However, consumers with an interest in healthy eating – being consciously aware of health and placing high value on health – are more likely to base their purchase decisions on nutritional information (Mai & Hoffmann, 2012). As the ability to monitor food purchases is crucial for improving diet quality (Bublitz et al., 2010; Glanz & Bishop, 2010; Hollywood et al., 2013), and since the real-time nutritional feedback allows for an accurate evaluation of the nutritional value contained by foods, these consumers may perceive the real-time nutritional feedback as useful guidance throughout their shopping trip, thereby making more effective use of the real-time nutritional feedback.

In sum, consumers’ interest in healthy eating may affect the extent to which real-time nutritional feedback is used for healthier purchase decisions. On the one hand, the real-time nutritional feedback may shift attentional selection towards the health trade-off for consumers with a lower interest in health. On the other hand, consumers with a higher interest in health are more likely to use nutritional information when making purchase decisions and thus may make more effective use of the real-time nutritional feedback. Hence, real-time nutritional feedback can especially be beneficial for those consumers who have an interest in healthy eating and may therefore positively strengthen the relationship of real-time nutritional feedback and the healthiness of the end-of-trip basket. Specifically, we propose that:

H3: The positive effect of real-time nutritional feedback on the healthiness of the shopping basket is moderated by the interest in healthy eating, whereby the effect is stronger for

3 Methods

The aim of our research was to gain a better understanding of the effect of real-time nutritional feedback on the healthiness of food purchases, as well to determine its boundary conditions. We tested our model (see figure 3-A) and the corresponding hypotheses in two experimental studies. We first examined the main effect of real-time nutritional feedback and interest in healthy eating on healthiness of the end-of-trip basket in study 1 (hypothesis 1 & 3). In study 2, we examined how different formats of real-time nutritional feedback (average format vs. cumulative format) affects the healthiness of the end-of-trip basket and whether an interest in healthy eating moderates this relationship (all hypotheses).

Figure 3-A: Conceptual model depicting the hypotheses and the direction of the relationships.

3.1 Experiment 1

The objective of study 1 was to get a better understanding on the main effect of real-time feedback and the moderating effect of interest in healthy eating; whether nutritional information regarding the relative healthiness of the shopping basket influences purchase decisions and if this relationship is shaped by an interest in healthy eating. In particular, we examined the effect of real-time nutritional feedback by benchmarking feedback regarding the relative “health” of the shopping basket not only against not giving feedback, but also against giving feedback regarding the “tastiness” of the shopping basket. Research shows that health information influences purchase decisions more slowly compared to information regarding the “tastiness” of foods (Sullivan, Hutcherson, Harris, & Rangel, 2015). As consumers are generally more

tempted by unhealthy, but tasty, foods (i.e., hedonic foods) (e.g., Talukdar & Lindsey, 2013; Wertenbroch, 1998; Yan, Tian, Heravi, & Morgan, 2017), they may also be more receptive to information regarding the tastiness of foods, especially those consumers who have a low interest in healthy eating. Hence, exploring how health and hedonic information provision affects purchase decisions allows us a to better understand the boundary conditions of real-time nutritional feedback and the extent to which it affects purchase decisions.

We examined the effect of real-time nutritional feedback with a between-subjects design in which participants were randomly assigned to the conditions (health feedback, hedonic feedback, control). The experimental study consisted out of three parts. First, participants completed a simulated shopping task in which they were asked to select foods from 14 product categories. Depending on the condition, participants received either feedback about the relative healthiness of their purchase decisions, the hedonic nature of their purchase decisions, or no feedback regarding their purchase decisions was given (control group). The second part of the study consisted out of questions regarding the shopping trip and their interest in healthy eating. The last part of study consisted out of measurements scales of other individual characteristics that may have affected their purchase decisions (control variables, dietary restrain, self-control) and demographics (see appendix K for full survey).

3.1.1 Procedure

For each decision, a picture of the products, price, and the number of calories was shown to the participants (see figure 3.1-A below). After participants made a choice, the next set of products appeared on the screen. To avoid response order biases (Krosnick & Presser, 2010), the order of the product categories, as well the order of products presented per product category were fully randomized.

Figure 3.1-A Examples of the simulated shopping task. Product categories ice cream and soup.

3.1.2 Feedback Manipulation

3.1.3 Shopping experience and Shopping Satisfaction

To establish the reliability of participants’ purchase decisions, we measured how realistically the participants engaged in the decision making process with a 3-item shopping experience Likert scale of van Ittersum et al. (2013) (e.g., “I was familiar with the products that were presented to me during the grocery shopping trip”; 1-7 agree/disagree).

As constantly making accurate healthy purchase decisions is cognitively demanding, participants may experience fatigue or negative emotions (Shiv & Fedorikhin, 1998). Real-time nutritional feedback should ease the constrains on cognitive processing recourses and thereby should facilitate accurate decision making. Therefore, we expected that the provision of real-time nutritional feedback would positively affect participants’ shopping satisfaction. We measured shopping satisfaction by asking participants the extent to which they liked the grocery shopping trip and the extent to which they appreciated the presentation of nutritional information (e.g., “I liked this grocery trip”, 1-7 agree/disagree).

3.1.4 Attention Paid to Attributes

The real-time nutritional feedback should make the health trade-off more salient for the participants. Therefore, to get a better understanding on the salience and importance of the different attributes, we asked participants to indicate how much attention they paid to the product specifications, price of the product, feedback (when applicable), and calories (1-7 no attention/ full attention). We expected that there would be a difference in attention paid to attributes between the conditions. Specifically, we expected that if the real-time nutritional feedback increases the salience of the health trade-off, participants would indicate to direct their attentional selection more towards the calories of products.

3.1.5 Interest in Healthy Eating

capture specific aspects of (un)healthy eating that were not as relevant for the purpose of this study.

3.1.6 Control Variables

To control for any other individual characteristics that may explain the relative healthiness of the end-of-trip basket, we measured dietary restrain and self-control abilities in the last part of the study. Consumers with dietary restrain apply and submit to their own individual diet goals, such as restricting selected foods or eating certain amounts of foods (Herman & Mack, 1975). Research shows that restrained eaters generally possess more nutritional knowledge and pay more attention to food cues (Bublitz et al., 2010). Hence, restrained eaters may be more receptive to the real-time nutritional feedback. Dietary restrain was measured with a 10-item Restrain Scale of Herman and Polivy (1978), of which scores ranged between 0-4 and 0-3 (e.g., “how aware are you of that what you eat?” 1-4 not aware at all/extremely aware).

The last scale measured participants’ level of self-control. Research shows that higher levels of self-control positively affects health behaviours (Will Crescioni et al., 2011), and can reduce the consumption of unhealthy foods (Redden & Haws, 2013). Therefore, we measured self-control with the 13-item Self-Control Scale of Tangney, Baumeister, and Boone (2004) as this may allow for a comprehensive understanding of the individual characteristics and the effect of real-time nutritional feedback. Participants indicated on a 1-5 Likert scale (1= not at all like me, 5= very much like me), to what extent 13 statements reflected their behaviour (e.g., “I refuse things that are bad for me”). An overview of the measurement scales used in study 1 can be found in the appendix B.

3.2 Results

3.2.1 Final Sample

purchase behaviour (M= 5.40, SD=1.62) and participants were familiar with products presented during the grocery shopping trip (M=5.61, SD=1.23).

3.2.2 Descriptives and frequencies of product choices

Figure 3.4-A below depicts the frequencies of product choices per product category, of which all respondents who completed the shopping task were taken into account. Two products were chosen frequently by participants in the product category milk (N=80) and peanut butter (N=92).

Figure 3.2—A Frequencies of product choices per product category.

The average number of calories in participants’ end-of-trip basket was 3587 (M= 3587.92, SD= 349.17), which was 756 calories above the minimum amount, and 774 calories below the maximum amount. As shown in figure 3.2-B below, participants who received feedback regarding the relative health of their shopping basket had, on average, the lowest number of calories in the end-trip-basket (M=3565, SD=335), followed by participants who did not receive any feedback (M=3634, SD=295). Participants who received hedonic feedback had, on average, the highest number of calories in the end-of-trip basket (M=3712, SD=588).

Figure 3.2—A Average number of calories in the end-of-trip basket per condition. 0 20 40 60 80 100 120 140 Healthy In Between Unhealthy 3000 3200 3400 3600 3800 4000

3.2.3 Shopping satisfaction and attention paid to attributes

Research shows that providing real-time feedback can decrease mental stress and increase customer satisfaction (van Ittersum et al., 2013). As such, we expected that participants who received feedback would report a higher shopping satisfaction. We examined if any differences between conditions existed by employing a one-way ANOVA for shopping satisfaction. As the internal consistency of the scale was too low for creating a new variable (α= .447), we examined the statements individually. Levene’s test of homogeneity of variances was non-significant for the two statements (p= .337, p= .132). The results revealed no significant differences between the conditions for the extent to which participants liked the grocery shopping trip (F(2,115)=1.398, p= .251) and the extent to which participants liked the way the nutritional information was presented to them (F(2,116)= .364, p= .696). These results do not confirm our expectations of a higher shopping satisfaction for those participants who received feedback. On average, participants liked the grocery shopping trip (M=5.35, SD=1.04) and liked the way the nutritional information was presented (M=5.21, SD=1.41).

differences between the conditions. As such, the results of the one-way ANOVA did not provide sufficient evidence to support our theoretical arguments of the underlying mechanism (i.e., increase in salience of the health trade-off). Overall, participants indicated that attentional selection was the highest for product specifications (M=3.22, SD=1.21), followed by calories (M=3.01, SD=1.22), and price (M=2.06, SD=1.11).

We ran some additional analyses to examine whether there were any differences in shopping satisfaction and attention paid to attributes existed depending on the level of interest in healthy eating. Specifically, we ran several one-way ANOVA models with general health interest as factor variable. Following Roininen et al. (1999), participants were categorized as either having a low (N= 36), medium (N=15), or high (N=68) interest in healthy eating by taking the 33rd (< 31) and 66th (> 36) percentile. The results revealed a significant difference between the groups in how much attention was paid to calories F(2,114)=12.55, p <.001. On average, attention devoted to calories was highest for participants with a high interest in healthy eating (M=3.45, SD=1.22), followed by participants with a moderate interest in healthy eating (M=2.54, SD=1.20). Attention devoted to calories was the lowest for the participants with a low interest in healthy eating (M=2.34, SD=1.14), which aligns with findings from previous research (e.g., Cooke & Papadaki, 2014). Additionally, a post-hoc Tukey HSD revealed a significant difference in how much attention was paid to calories between participants with a low and high interest in healthy eating (-1.11, p <.001), and between participants with a high and moderate interest in healthy eating (-.909, p= .022). Overall, the findings align with our expectations regarding interest in healthy eating: participants with an interest in healthy eating paid more attention to the amount of calories of products (i.e., increased salience of health trade-off), which in turn may positively affect their food purchase decisions.

3.2.4 Reliability of the measurement scales

α M SD

Item Scale Item Scale

General Health Interest (8 items) .74 3.71 29.69 2.8 6.08

Dietary Restrain (10 items) .64 2.20 22 1.17 4.70

Self-Control (13 items) .79 3.12 40.49 1.06 7.48

Table 3.2-A Results of the reliability analyses.

3.2.5 New Variables

As the internal consistency was established, we created new variables for general health interest, dietary restrain, and self-control. We used these new variables for testing the hypotheses. The new variables for general health interest and self-control were created by mean centring the total scores (sum of all items) of the scales. For dietary restrain, we created a new categorical variable. Following the classification method of Herman and Polivy (1978), a new dietary restrain variable was created by classifying respondents either as restrained eaters (N= 63), or non-restrained eaters (N=47), based on the median (21) split of scores.

After creating the new variables, we examined whether differences between the conditions existed for any of the variables that could have affected the results (e.g., biased estimates, confounds) by employing a one-way ANOVA with the condition as factor (see table 3.2-B below). Levene’s test of homogeneity of variances was non-significant (p > 0.1) for all variables and no significant differences between conditions were found for any of the variables.

Variables Levine’s Test

Significance F Significance

General Health Interest .814 1.57 .213

Dietary Restrain .406 .260 .722

Self-Control .550 .887 .415

Age .161 .903 .408

Gender .094 .414 .662

Table 3.2-B Results of one-way ANOVA with conditions as factor. No significant differences were found between the conditions for any of the variables.

3.2.6 Correlations

self-control did not significantly correlate with the amount of calories in the end-of-trip basket. However, these two variables did correlate with general health interest.

General Health Interest Dietary Restrain Self-Control Total Calories SB -.302** -.099 -.132 General Health Interest .231* -332** Dietary Restrain -.052

Table 3.2-C Correlation matrix of the independent and dependent variables. ** Significant at .01 * Significant at .05

3.2.7 Linear Regression Analysis

We tested whether the real-time nutritional feedback significantly affected total calories in the end-of-trip basket by running several linear regression models. In model 1, we tested the effect of providing real-time nutritional feedback (feedback yes/no) and the effect of interest in healthy eating (general health interest). In model 2, we benchmarked the effect of real-time health and hedonic feedback against the control group (2 dummy variables) as well to test the effect of interest in healthy eating (general health interest). We subsequently added interaction terms in model 3 to examine the moderating effect of interest in healthy eating. Lastly, as a robustness check, we added the two control variables (self-control, dietary restrain) to the model.

Checking Assumptions

with large values for Cook’s distance that may have been problematic. Therefore, we tested the model again excluding these cases (N=110). The model-fit increased (R2=.093 vs. R2=.122) sufficiently, and there were no substantial differences in the regression coefficients. The scatterplots of the standardized residuals and predicted values did not show an obvious pattern, which suggested that the homoscedasticity assumption is met (see appendix D). The results of the regression models are shown in table 3.2-D below.

Model 1 (N=110) Model 2 (N=115) Model 3 (N=115) Model 4 (N=105) Intercept 3604*** 3615*** 3773*** 3860*** Feedback: Yes -28.03 (-.042) General Health Interest -14.18***

(-.346) -10.62** (-.253) -6.03 (-.144) -8.88 (-.203) Feedback: Hedonic 70.78 (.100) 53.95 (.075) 49.88 (.069) Feedback: Health -155.58* (-.214) -150.85* (-.144) -155.15† (-.208) Feedback Health* General Health

Interest.

-7.30 (-.092)

-2.41 (-.029) Feedback Hedonic* General Health

Interest -8.68 (-.112) -6.24 (-.084) Self-Control -.645 (-.015) Dietary Restrain -8.34 (-.012) R2 _{F f} 2 Model 1 .122 7.53*** .14 Model 2 .160 7.09*** .19 Model 3 .168 4.54*** .20 Model 4 .163 2.72** .19

Table 3.2-D Results of linear regression models

*** Significant at .001 ** Significant at .01 * Significant at .05 † Significant .05 - .10 Cohens’ f 2 for the model effect size is calculated as: f 2= R2/(1-R2)

Effects of General Health Interest

calories in the end-of-trip basket for participants with an average interest in health. Feedback did not significantly affect the number of calories in the end-of-trip basket (β=-28.03, p= .641), but general health interest did have a significant effect (β=-14.18, p <.001). These findings suggest that having an interest in health can positively affect the healthiness of the end-of-trip basket.

Effects of feedback

The results of the first regression model revealed a non-significant effect of feedback, suggesting that the feedback given participants did not affect product choices. However, as participants received either health or hedonic feedback, we suspected that the non-significant effect may be driven by one of the two conditions. We therefore employed another linear regression analysis to explore the different feedback types more in depth. We estimated the main effects of the different types of feedback (hedonic vs. healthy) and interest in healthy eating. The results revealed a significant effect of the health feedback condition (β=-155.58 p= .033), a significant effect of general health interest (β=-10.62, p= .005), but no significant effects were found for the hedonic feedback condition (β=61.21, p= .392). Subsequently, we added an interaction term for the feedback variables and general health interest to examine whether the effect of feedback on total calories in the end-of-trip basket is moderated by interest in healthy eating. However, the result revealed only a significant effect of the health feedback condition (β=-150.85, p= .042). Lastly, we added the control variables dietary restrain and self-control to check whether this would result in different outcomes. The results of this model did showed a marginally significant effect of the health feedback condition (β=-155.15, p= .071).

3.2.8 Discussion

decreased the number of calories in the end-of-trip basket (model 2-4). This suggests that real-time nutritional feedback can promote healthier purchase decisions. Hedonic feedback did not significantly affect the number of calories in the end-of-trip basket. A possible explanation for this could stem from the strong intuitive tendency to link concepts as “unhealthy” to “tasty” (Raghunathan et al., 2006), and a stronger impulse to consume unhealthy foods compared to healthy foods (Talukdar & Lindsey, 2013). Specifically, participants may already have had a strong preference for certain foods that they consider to be tasty. To provide hedonic feedback based on what “others” consider to be tasty (i.e., subjective information) may therefore not affect these already strong preferences. Overall, the findings suggest real-time nutritional feedback positively affected the purchase decisions of those participants who received the feedback. Thus, we found support for the first hypothesis; real-time nutritional feedback positively affects the healthiness of the end-of-trip basket.

3.3 Experiment 2

The results of study 1 tentatively demonstrated the positive main effect of real-time health feedback on the relative healthiness of the end-of-trip basket. Moreover, the results suggest that the type of information given to participants yields differential results. That is, whereas time health feedback positively affected the healthiness of purchase decisions, providing real-time hedonic feedback did not significantly affect purchase decisions (compared to no feedback given). The aim of study 2 was to find more support for the positive effect of real-time nutritional feedback on the healthiness of the end-of-trip basket (hypothesis 1), as well to determine its boundary conditions by examining the differential effect of the feedback formats (hypothesis 2), and the moderating role of an interest in healthy eating (hypothesis 3). In an experimental design that resembles study 1, we asked participants (N=256) to make purchase decisions from 16 common product categories, providing feedback accordingly. Additionally, we wanted to examine the role of consumers’ interest in healthy eating.

After the shopping task, we measured participants shopping experience, shopping satisfaction, attentional selection to product attributes, general health interest, other individual characteristics that may explain purchase decisions (control variables, nutritional literacy, dietary restrain, price consciousness, self-control), and demographics (see appendix J for full survey).

3.3.1 Procedure

We recruited Dutch participants from ProLific on the 6th and 7th of June who participated for a fee. To control for any confounds (i.e., significant differences due to over/under representation of certain type of participants within each condition, instead of differences due to the treatment), participants were randomly assigned to one of the three conditions (average feedback, cumulative feedback, no feedback) and started with a shopping task that resembled the shopping task in study 1. Participants were asked to make food selections from 16 common product categories, having 5 products to choose from per product category (80 products to choose from in total). The products and product categories were carefully chosen, ensuring that enough variation in terms of nutritional value within- and between- product category, while also taken into account other product specifications such as price and flavour. At least three different Nutri-Scores were included within one product category. Between product categories, we balanced categories that consisted predominantly out of unhealthy products with categories who predominantly consisted out of healthy products (e.g., butter ranged from C – E and salads ranged from A – C). This resulted in a slightly altered product category set compared to study 1 (see appendix B). For instance, the product category Milk was left out, as most milk products have a Nutri-Score of A (only whole milk has a Nutri-Score of B).

Figure 3.3—A Example of product categories in the study.

3.3.2 Feedback Manipulation

Depending on the condition, participants were presented with the nutritional value of their shopping baskets (i.e., real-time nutritional feedback). We used two different formats to visualize the real-time nutritional feedback that both were high in perceptual proximity: feedback was presented either as the average Nutri-Score of the shopping basket, or as the cumulative sum of Nutri-Scores in the shopping basket (see figure 3.3-B below). Although both feedback formats are high in perceptual proximity and should be compatible to the shopping task (following PCP logic), there was a difference between the formats in terms of the intensity of changes in the format as a response to participants’ purchase decisions. Consumers perceive changes in stimuli with diminishing sensitivity when the changes become too small, or when changes for all the attributes are similar (Bordalo et al., 2013; Britt & Nelson, 1976). As such, we expected that the feedback formats could still produce differential results (i.e., differences in the extent to which the feedback is used for purchase decisions).

Figure 3.3—B Real-time nutritional feedback formats. Left: average nutritional value. Right: cumulative sum of nutritional value.

Nutri-Score of the shopping basket, we recoded the Nutri-Nutri-Scores (A=1 – E=5) and subsequently took the average of the sum of Nutri-Score values in the shopping basket. After each purchase decision, the real-time nutritional feedback was updated and presented to the participants in the top right corner. In the average feedback format condition, an arrow below the Nutri-Score indicated the average Nutri-Score of the shopping basket (i.e., average Nutri-Score of purchase decisions thus far) and moved either left or right depending participants’ purchase decisions. As we presented the average Nutri-Score of the shopping basket, the intensity of changes in the real-time nutritional feedback decreased as participants proceeded on their shopping trip (i.e., the movements of the arrow, either to the left or right of the scale, became smaller). As such, we expected that participants may not notice changes in the average feedback format at a certain point during the shopping trip (i.e., not meet the threshold of just noticeable difference). In the “cumulative” condition, each purchase decision was added as a building block, creating a graphical distribution of Nutri-Scores in the shopping basket. In this condition, the intensity by which the feedback changed should remain at the same level throughout the entire shopping trip. Moreover, the changes in the feedback – visualized as an increase in sum of Nutri-Scores in the shopping basket – would differ after each purchase decision (in colour and shape). Hence, we expected that participants would be able to notice changes in the cumulative feedback format throughout the entire shopping trip (i.e., systematically meet the threshold of just noticeable difference).

3.3.3 Shopping Experience & Shopping Satisfaction

To establish the reliability of participants’ purchase decisions, we measured how realistically the participants engaged in the decision making process with a 3-item, Likert scale of van Ittersum et al. (2013) that was translated to Dutch (e.g., “the choices I made accurately reflect what I would do in my regular grocery store”; 1-7 agree/disagree; see full survey in appendix I for Dutch version).

3.3.4 Attention Paid to Attributes

The results of study 1 provided limited support for an increased salience of the health trade-off as a result of the real-time nutritional feedback. To affirm the findings, we included the measurement scale of attention paid to attributes. We asked participants to indicate how much attention they paid to the product specifications, price of the product, the Nutri-Score of the product and, when applicable, the Nutri-Score of shopping basket (1-7 no attention at all/full attention). We expected that participants who received real-time nutritional feedback would devote more attention to the Nutri-Score than participants who did not receive feedback. As a manipulation check, we asked participants to indicate which Nutri-Scores were presented to them during the shopping task. Participants could select multiple answers (No Nutri-Score, Nutri-Score of shopping basket, Nutri-Score of product, Nutri-Score of product category).

3.3.5 Interest in Healthy Eating

We measured interest in healthy eating in a similar manner as we did in study 1 (GHIS scale). The results of study 1 revealed that the GHIS had a good internal consistency and thus was a reliable measurement of interest in healthy eating. Moreover, the results revealed a positive main effect of interest in healthy eating on the (relative) healthiness of the end-of-trip basket (decrease in calories). Hence, the general health interest scale of Roininen, Lähteenmäki, and Tuorila (1999) was an appropriate proxy for measuring interest in healthy eating.

3.3.6 Control Variables

The last part of study consisted out of measurement scales of the control variables and demographics. Similar to study 1, we measured dietary restrain with the Restrain Scale of Herman and Polivy (1978) and measured self-control with the Self-Control Scale of Tangeny, Baumeister, and Boone (2004). Additionally, we included two other measurement scales that measured individual characteristics which (potentially) could explain the healthiness of the end-of-trip basket: price consciousness and nutritional literacy.

consciousness and price consciousness scale of van Ittersum et al. (2013) (e.g., “I am price conscious during grocery shopping”;1-7 agree/disagree).

Nutritional literacy is centred around an individual’s cognitive capabilities to understand and use information about nutrition, which in turn is a crucial for (improving) diet quality (Blitstein & Evans, 2006; Silk et al., 2008; Vaitkeviciute, Ball, & Harris, 2015; Velardo, 2015). We measured participants’ nutritional literacy level with two scales that measured nutritional knowledge, as well as the ability to understand and use nutritional information: The Nutritional Literacy Scale (NLS) of Diamond (2007) and the Newest Vital Sign (NSV) of Weiss et al. (2005). The NLS consisted out of 23 Cloze procedure sentences in which one or more words were removed from a sentence (28 words in total). Each sentence included several different options and participants were asked to pick the word that would be most appropriate in the sentence. For instance, the sentence “Eating a ____ of foods ensures you get all the nutrients needed for good health” was followed with four choices (lot, many, variety, pound) that participants could select from. The Newest Vital Sign (NVS), measured participants’ ability to use nutritional information for food decision making. Participants were presented with a nutrition label of an ice cream container and subsequently had to answer six questions about scenarios related to eating from the container of ice cream (e.g., “If you usually eat 2500 calories in a day, what percentage of your daily value of calories will you be eating if you eat one serving?”). As both NLS and NSV are English scales, we translated the scales to Dutch using a backward translation method. First, two people individually translate the NLS to Dutch and were subsequently compared and combined into one scale. Second, one person translated the Dutch NLS back in English to check whether any misinterpretation had occurred or whether different meanings were given to words. 2

The final part of the study consisted out of demographics (age, gender) and participants were given the opportunity to leave a comment when they had any remarks related to the study. An overview of the scales can be found in the appendix B.

2

3.4 Results

3.4.1 Final Sample

In total, 258 participants participated in the study (gave consent), of which 59% were male (N=152). The average age of the participants was 27 years old (M=27.47, SD=8.35). Before analysing the data, we checked whether biased responses existed among the observations (reliability of purchase decisions, extreme values, manipulation check). First, we examined the extent to which participants’ purchase decisions accurately reflected their normal shopping behaviour (shopping experience, three statements). The internal consistency of the measurement scale was sufficient (α = .744), with no large changes in alpha when item deleted. On average, the purchase decisions of participants sufficiently reflected their normal shopping behaviour (M=5.71, SD=1.01). Second, we examined whether the manipulation of the conditions was successful (Manipulation check “What kind of Nutri-Score(s) did you see during the shopping trip?”). Six participants indicated that they either did not see any Nutri-Scores during their shopping trip, or that they were presented with Nutri-Score for the entire product category (both options were not possible). To determine the reliability of these participants, we included and excluded their responses in our analyses and subsequently examined whether this lead to large differences in outcomes. No large differences were detected. Therefore, we decided to include these observations in our analyses. Of all participants, 86 were presented with the average Nutri-Score of their shopping basket (average condition) and 85 participants were presented with the cumulative sum of all Nutri-Scores in their shopping basket (cumulative condition).

3.4.2 Descriptives and Frequencies of Product Choices

Figure 3.4—A Frequencies of product choices per product category.

The average sum of Nutri-Scores in the shopping baskets was 44.59 (SD=7.09), which translates to an overall Score of C (3). Three participants had the lowest sum of Nutri-Scores in the end-of-trip basket (Nutri-Score B)3, while the highest score (N=2) was 58 (Nutri-Score D). Figure 3.4-B below visualizes the average total sum of Nutri-(Nutri-Scores in the end-of-trip basket of participants. Participants in the cumulative feedback conditions had, on average, the lowest total number of Nutri-Scores in the end-of-trip basket (M=43.48, SD=7.50), followed by participants in the average feedback condition (M=44.68, SD=7.50). Participants who were not presented with feedback had, on average, the highest number of Nutri-Scores in the end-of-trip basket (M=45.61, SD=6.12).

Figure 3.4—B Average sum of Nutri-Scores in end-of-trip basket

3_{Two participants did not select a product from each category and had a very low sum of Nutri-Scores (3, 9) in the end-of-trip basket.} 0 50 100 150 200 250 A B C D E 35 38 41 44 47 50 53

No Feedback Average Feedback Cumulative Feedback

3.4.3 Shopping satisfaction and attention paid to attributes

We expected that participants who received feedback would pay more attention to the Nutri-Score compared to participants who did not (as the health trade-off should be more salient). Additionally, we expected that participants who received real-time nutritional feedback would report a higher level of shopping satisfaction. The shopping satisfaction scale (3 statements) was subjected to a reliability analysis and revealed a Cronbach’s alpha of .815 with no large differences in alpha when deleting an item. As the internal consistency is established, we created a new variable that represented the overall shopping satisfaction (the average of the three statements). Subsequently, we subjected differences in shopping satisfaction to a one-way ANOVA with the conditions as factor. Levene’s test of homogeneity of variances was significant (p= .03) and thus every finding needed to be interpreted with care. However, no significant differences existed between the conditions in terms of shopping satisfaction (F (2,256)=0.16, p= .856). On average, participants reported a moderately high level of shopping satisfaction (M=5.36, SD=.072).

Differences in how much attention was paid to the attributes was subjected to a one-way ANOVA with the conditions as factor. Levene’s test of homogeneity of variances was not significant for the variables Nutri-Score (p= .693), price (p= .317), and product specifications (p= .089), thus equal variances were assumed. The results revealed no significant differences in how much attention was paid to the Nutri-Score (F(2,256)=1.157, p= .316), the product specifications (F(2,256)=.304, p= .738), or the price of a products (F(2,256)=1.542, p= .216). On average, participants indicated that they paid most attention to product specifications (M= 5.06, SD=1.24), followed by price (M=4.53, SD=1.63), and the Nutri-Score (M=4.32, SD=1.43).4

As research shows that consumers with an interest in healthy eating are more likely to base their purchase decisions on nutritional value (Mai & Hoffmann, 2012), we ran some additional analyses to examine whether there were any differences in shopping satisfaction, shopping experience, and attention paid to attributes depending on the level of interest in healthy eating. Specifically, we ran several one-way ANOVA models with general health interest as factor variable. Following Roininen et al. (1999), participants were categorized as either having a low

4

(N=87), medium (N=83), or high (N=83) interest in health by taking the 33rd (< 31) and 66th (> 36) percentile. The results revealed a significant difference between the groups in how much attention was paid to the Nutri-Score F(2,256)=13.74, p <.001. Additionally, a post-hoc Tukey HSD revealed a significant difference in how much attention was paid to the Nutri-Score between participants with a low and high interest in healthy eating (-1.19, p <.001), and between participants with a low and average interest in healthy eating (-.874, p= .001). These findings align with the results found in study 1, as well do they support previous research (e.g., Cooke & Papadaki, 2014): those who have an interest in healthy eating devote more attention to the health trade-off, which in turn can positively affect the healthiness of food purchases.

3.4.4 Reliability of measurement scales

To examine the internal consistency of the scales, all measurement scales were subjected to a reliability analysis. The results of the reliability analysis and the average scores on the scale are shown in table 3.4-A below.

α M SD

Item Scale Item Scale

Price Consciousness (4 items) .65 4.02 16.06 2.99 3.52

Newest Vital Sign (NSV) (6 items) .52 .860 5.16 .124 1.69

General Health Interest (9 items) .80 4.19 33.53 1.37 7.09

Dietary Restrain (10 items) .74 2.35 13.49 1.44 5.29

Self-Control (13 items) .81 3.08 40.04 1.11 7.59

Nutritional Literacy Scale (NLS) (28 items) .71 .889 24.88 .667 2.72

NLS & NSV (34 items) .79 .885 30.09 .685 3.54

Table 3—A Results of the reliability analyses.

consciousness is used as a control variable, the reliability of the scale was considered as sufficient.

3.4.5 New Variables

After establishing internal consistency, we created new variables for the measurements scales general health interest, self-control, and price consciousness by mean centring the total scores of the scales. For dietary restrain, we created a new categorical variable. Following the classification method of Herman and Polivy (1978), the new dietary restrain variable was created by classifying respondents either as restrained eaters (N=144), or non-restrained eaters (N=113), based on the median (24) split of scores.

After creating the new variables, we examined whether differences between the conditions existed for any of the variables, as a difference between the conditions could lead to biased estimates (i.e., any significant effect on the dependent variable that is not attributed to the treatment effect, but rather to the differences between conditions for that variable). All measurements scales were subjected to a one-way ANOVA with the three conditions as factor. As shown in table 3.4-B below, there were no significant differences between the conditions for any of the variables.

Variables Levine’s Test

Significance F Significance

General Health Interest .908 .499 .608

Nutrition Literacy .994 .043 .958 Price Consciousness .751 .917 .401 Self-Control .873 .112 .895 Dietary Restrain .751 .884 .414 Age .199 .299 .742 Gender .032 1.62 .200

Table 3—B Results of one-way ANOVA with the conditions as factor variable.

3.4.6 Correlations

not significantly correlate to the Nutri-score in the end-of-trip basket. Dietary restrain did correlate with general health interest and self-control.

General Health Interest Nutritional Literacy Price Consciousness Self-Control Dietary Restrain Nutri-Score SB -.377** -.007 .131* -.280** -.017 General Health Interest .056 -.076 .322** .274** Nutritional Literacy -.035 .006 -.102 Price Consciousness .034 -.069 Self-Control -.193**

Table 3—C Correlation matrix of the independent and dependent variables.

3.4.7 Linear Regression Analysis

We examined whether real-time nutritional feedback significantly affected the sum of Nutri-Scores in the end-of-trip basket by running several linear regression models. In model 1, we examined the effect of providing feedback (feedback yes/no) and the effect of general health interest. In model 2, we added an interaction term for the feedback and general health interest. We tested the effect of the different feedback formats (average vs. cumulative) and the effect of general health interest in model 3, and subsequently added an interaction term for the two feedback formats and general health interest in model 4. As a robustness check, we included all the control variables in the last model (5).

Checking assumptions